Introduction
The Vlasov–Navier–Stokes system
We consider the Vlasov–Navier–Stokes system (VNS) on T d × R d \mathbb{T}^{d} \times\mathbb{R}^{d} T d × R d , d ≥ 2 d \ge2 d ≥ 2 ,
{ ∂ t f + v ⋅ ∇ x f + div v [ ( u − v ) f ] = 0 , ∂ t u − Δ u + div ( u ⊗ u ) + ∇ p = ∫ R d ( v − u ) f d v , div u = 0. (1) \begin{cases}
\partial_t f + v \cdot\nabla_x f + \operatorname{div}_v\left[(u-v)f\right] = 0,\\
\partial_t u - \Delta u + \operatorname{div}(u \otimes u) + \nabla p = \int_{\mathbb{R}^{d}} (v-u)f\,\mathrm{d}v,\qquad\operatorname{div} u = 0. \tag*{(1)}
\end{cases} { ∂ t f + v ⋅ ∇ x f + div v [ ( u − v ) f ] = 0 , ∂ t u − Δ u + div ( u ⊗ u ) + ∇ p = ∫ R d ( v − u ) f d v , div u = 0. (1)
It describes a dispersed phase of small particles, with distribution function f ( t , x , v ) ≥ 0 f(t,x,v) \ge0 f ( t , x , v ) ≥ 0 , transported by an incompressible viscous fluid with velocity u ( t , x ) u(t,x) u ( t , x ) . The phases interact through the drag (Brinkman) force: the particles are accelerated by u − v u-v u − v , and the fluid feels the opposite force ∫ ( v − u ) f d v = j f − ρ f u \int(v-u)f\,\mathrm{d}v = j_f-\rho_f u ∫ ( v − u ) f d v = j f − ρ f u , where ρ f = ∫ f d v \rho_f = \int f\,\mathrm{d}v ρ f = ∫ f d v and j f = ∫ v f d v j_f = \int vf\,\mathrm{d}v j f = ∫ v f d v . All physical constants are set to one. Smooth solutions satisfy the energy identity
d d t ( 1 2 ∥ u ∥ L 2 2 + 1 2 ∬ ∣ v ∣ 2 f d x d v ) + ∥ ∇ u ∥ L 2 2 + ∬ ∣ v − u ∣ 2 f d x d v = 0 , \frac{\mathrm{d}}{\mathrm{d}t}\left(\frac{1}{2}\lVert u\rVert_{L^{2}}^{2}+\frac{1}{2}\iint|v|^{2}f\,\mathrm{d}x\,\mathrm{d}v\right)+\lVert\nabla u\rVert_{L^{2}}^{2}+\iint|v-u|^{2}f\,\mathrm{d}x\,\mathrm{d}v=0, d t d ( 2 1 ∥ u ∥ L 2 2 + 2 1 ∬ ∣ v ∣ 2 f d x d v ) + ∥ ∇ u ∥ L 2 2 + ∬ ∣ v − u ∣ 2 f d x d v = 0 ,
and conserve the total momentum ∫ u d x + ∬ v f d x d v \int u\,\mathrm{d}x+\iint vf\,\mathrm{d}x\,\mathrm{d}v ∫ u d x + ∬ v f d x d v ; the spatial mean of u u u alone is not conserved.
Global weak solutions satisfying the energy inequality exist in various settings [2 , 6 , 30 , 29 , 7 ] . In two dimensions, Han-Kwan, Miot, Moussa and Moyano proved that weak solutions in the energy class are unique, for kinetic data satisfying a mild decay condition in velocity [20 ] . The long-time behavior [19 ] and hydrodynamic limits [18 ] have also been studied extensively. In three dimensions, uniqueness of energy solutions is open, as for the Navier–Stokes equations; a weak–strong uniqueness principle beyond the Osgood class was obtained in [21 ] . In view of the nonuniqueness of Leray solutions of the forced Navier–Stokes equations [1 ] , uniqueness is not to be expected in general. To our knowledge, nonuniqueness of weak solutions of (1.1) has not been established. We prove it in a class that is sharp with respect to the Ladyzhenskaya–Prodi–Serrin scaling, first for smooth data and then for rough data.
Convex integration and the memory of the particles
Since the work of De Lellis and Székelyhidi [13 , 14 ] , convex integration has produced nonunique weak solutions of many equations of fluid dynamics; see Isett’s resolution of Onsager’s conjecture [22 ] . For the Navier–Stokes equations, Buckmaster and Vicol [8 ] constructed nonunique weak solutions in C t L 2 C_tL^2 C t L 2 , and Buckmaster, Colombo and Vicol [5 ] obtained wild solutions whose singular set in time has Hausdorff dimension less than one. For the transport equation, see [26 , 11 ] . Cheskidov and Luo Cheskidov and Luo [12 ] proved that weak solutions of the Navier–Stokes equations are nonunique in L t p L ∞ ∩ L t 1 W 1 , q L_t^pL^\infty\cap L_t^1W^{1,q} L t p L ∞ ∩ L t 1 W 1 , q for every p < 2 p<2 p < 2 , q < ∞ q<\infty q < ∞ , while solutions in L t 2 L ∞ L_t^2L^\infty L t 2 L ∞ are unique, a classical fact (see [12 Theorem 1.3 and Appendix A] ). Their wild solutions are smooth outside a closed set of times of arbitrarily small Hausdorff dimension; this rests on a concentration step that glues local exact solutions onto the approximate one [12 Section 3] . They also obtain arbitrary L 2 L^2 L 2 data for 2 ≤ d ≤ 4 2\le d\le4 2 ≤ d ≤ 4 , by modifying a Leray–Hopf solution after a time at which it is regular [12 Theorem 1.5 and Remark 1.7(3)] .
The drag force makes (1.1) a Navier–Stokes system with a self-consistent forcing, and in our construction the nonuniqueness comes from the fluid. The new feature is that this forcing has memory . The distribution function at time t t t is the push-forward of f 0 f_0 f 0 along the phase-space flow of u u u on [ 0 , t ] [0,t] [ 0 , t ] , so it depends on the whole past of the velocity. A modification of u u u on a short time interval changes the particle state, hence the drag force, at all later times. This causes no difficulty for nonuniqueness alone, where every perturbation is small globally in time, but it affects all arguments that are local in time: in [12 Section 3] the approximate solution is replaced by local exact solutions on good intervals and left untouched elsewhere, and for (1.1) a local exact solution computed with the old particle state is no longer a solution once anything earlier has changed. The main technical contribution of this paper is a causal version of the gluing, carried out from left to right in time, in which every good interval is re-solved with the particle state produced by everything done before it. It gives intervals of regularity (Theorem 1.3 ). Causality enters in a different form in the extension to rough data (Theorem 1.4 ): there the background is mollified causally, using only its past, so that the particle state on an initial interval is left untouched.
Main results
Weak solutions are defined in Definition 2.1. In particular, u ∈ L t , x 2 ∩ L t 1 L ∞ u\in L_{t,x}^2\cap L_t^1L^\infty u ∈ L t , x 2 ∩ L t 1 L ∞ , and f ∈ L ∞ f\in L^\infty f ∈ L ∞ is nonnegative and compactly supported in v v v . Solutions of Leray–Hopf type are those satisfying the energy inequality (Definition 2.2).
Theorem 1.1 (Nonuniqueness). Let d ≥ 2 d\ge2 d ≥ 2 , 1 ≤ p < 2 1\le p<2 1 ≤ p < 2 and 1 ≤ q < ∞ 1\le q<\infty 1 ≤ q < ∞ . Let u i n ∈ C ∞ ( T d ) u_{\mathrm{in}}\in C^\infty(\mathbb{T}^d) u in ∈ C ∞ ( T d ) be divergence-free and let f 0 ∈ C c ∞ ( T d × R d ) f_0\in C^\infty_c(\mathbb{T}^d\times\mathbb{R}^d) f 0 ∈ C c ∞ ( T d × R d ) be nonnegative. Assume that (1.1) has a smooth solution on some interval [ 0 , t 1 ] [0,t_1] [ 0 , t 1 ] with data ( u i n , f 0 ) (u_{\mathrm{in}},f_0) ( u in , f 0 ) . Then there are infinitely many weak solutions ( u , f ) (u,f) ( u , f ) of (1.1) on [ 0 , 1 ] [0,1] [ 0 , 1 ] with these data, all coinciding with the smooth solution on [ 0 , t 1 / 4 ] [0,t_1/4] [ 0 , t 1 /4 ] , such that
(i) u ∈ L t , x 2 ∩ L t p L ∞ ∩ L t 1 W 1 , q u\in L_{t,x}^2\cap L_t^pL^\infty\cap L_t^1W^{1,q} u ∈ L t , x 2 ∩ L t p L ∞ ∩ L t 1 W 1 , q ;
(ii) f f f is bounded, has uniformly bounded velocity support, belongs to C t L s C_tL^s C t L s for every s < ∞ s<\infty s < ∞ , and is the unique bounded (hence renormalized) solution of the kinetic equation with velocity field u u u ;
(iii) the total momentum is conserved.
Infinitely many of these solutions are not of Leray–Hopf type.
Theorem 1.1 follows from Theorem 2.3 and Corollary 2.4. Theorem 2.3 is a flexibility statement: modulo spatial means, the velocity can be prescribed up to an arbitrarily small error in L t p L ∞ ∩ L t 1 W 1 , q L_t^pL^\infty\cap L_t^1W^{1,q} L t p L ∞ ∩ L t 1 W 1 , q . The endpoint p = 2 p=2 p = 2 cannot be reached.
Theorem 1.2 (Uniqueness in the critical class). Let d ≥ 2 d\ge2 d ≥ 2 and T ∈ ( 0 , 1 ] T\in(0,1] T ∈ ( 0 , 1 ] . Let u i n ∈ C ∞ ( T d ) u_{\mathrm{in}}\in C^\infty(\mathbb{T}^d) u in ∈ C ∞ ( T d ) be divergence-free and let f 0 ∈ L ∞ f_0\in L^\infty f 0 ∈ L ∞ be nonnegative and compactly supported in v v v . Then there is at most one weak solution of (1.1) on [ 0 , T ] [0,T] [ 0 , T ] with data ( u i n , f 0 ) (u_{\mathrm{in}},f_0) ( u in , f 0 ) and u ∈ L 2 ( 0 , T ; L ∞ ) u\in L^2(0,T;L^\infty) u ∈ L 2 ( 0 , T ; L ∞ ) . Such a solution satisfies ∇ u ∈ L 2 ( 0 , T ; L ∞ ) \nabla u\in L^2(0,T;L^\infty) ∇ u ∈ L 2 ( 0 , T ; L ∞ ) , its distribution function is transported by the classical phase-space flow, and it satisfies the energy equality. In particular, on any interval on which a smooth solution exists, the smooth solution is the only weak solution with u ∈ L t 2 L ∞ u\in L_t^2L^\infty u ∈ L t 2 L ∞ , while for every p < 2 p<2 p < 2 there are infinitely many weak solutions with u ∈ L t p L ∞ u\in L_t^pL^\infty u ∈ L t p L ∞ .
Theorem 1.2 is Theorem 5.1 together with Corollary 5.2; Corollary 5.11 extends it to u i n ∈ B q , 2 1 u_{\mathrm{in}}\in B^1_{q,2} u in ∈ B q , 2 1 , q > d q>d q > d . The causal gluing gives the analogue of the intervals of regularity of [12 ] .
Theorem 1.3 (Intervals of regularity). Under the assumptions of Theorem 1.1, and for every ε > 0 \varepsilon>0 ε > 0 , the solutions can be chosen so that, in addition, there is a closed set S ⊂ ( 0 , 1 ) S\subset(0,1) S ⊂ ( 0 , 1 ) with dim H S ≤ ε \dim_{\mathrm{H}}S\le\varepsilon dim H S ≤ ε such that, on every compact interval K ⊂ ( 0 , 1 ] ∖ S K\subset(0,1]\setminus S K ⊂ ( 0 , 1 ] ∖ S , u ∈ L 2 ( K ; W 1 , ∞ ) ∩ C ( K ; W 1 , s ) ∩ L 2 ( K ; W 2 , s ) u\in L^2(K;W^{1,\infty})\cap C(K;W^{1,s})\cap L^2(K;W^{2,s}) u ∈ L 2 ( K ; W 1 , ∞ ) ∩ C ( K ; W 1 , s ) ∩ L 2 ( K ; W 2 , s ) for every s < ∞ . O n s u c h s<\infty. On such s < ∞. O n s u c h K$ the solution is the only weak solution with velocity in L 2 ( K ; L ∞ ) L^2(K;L^\infty) L 2 ( K ; L ∞ ) starting from its own state at inf K \inf K inf K .
Theorem 1.3 is Theorem 6.2 . Off the singular set the solutions are strong, but in general not C ∞ C^\infty C ∞ : the distribution function inherits the limited regularity produced by earlier perturbations, and the drag force passes it on to the velocity. This is a direct effect of the memory, and it is why uniqueness on K \mathcal{K} K is stated from the state at inf K \inf\mathcal{K} inf K rather than from the initial data.
The next result removes the smoothness assumptions on the data. The kinetic datum only needs to be bounded, and the velocity datum only needs to generate a weak solution that becomes Lipschitz after some time.
Theorem 1.4 (Rough data). Let d ≥ 2 d \ge2 d ≥ 2 , 1 ≤ p < 2 1 \le p < 2 1 ≤ p < 2 , 1 ≤ q < ∞ 1 \le q < \infty 1 ≤ q < ∞ . Let f 0 ∈ L ∞ ( T d × R d ) f_0 \in L^\infty(\mathbb{T}^d \times\mathbb{R}^d) f 0 ∈ L ∞ ( T d × R d ) be nonnegative and vanish for large ∣ v ∣ |v| ∣ v ∣ , and let u i n u_{\mathrm{in}} u in be divergence-free with
(a) u i n ∈ B q 0 , 2 1 u_{\mathrm{in}} \in B^{1}_{q_0,2} u in ∈ B q 0 , 2 1 for some q 0 > d q_0 > d q 0 > d (for instance u i n ∈ H s u_{\mathrm{in}} \in H^s u in ∈ H s , s > d 2 s > \frac{d}{2} s > 2 d ), if d ≥ 2 d \ge2 d ≥ 2 ; or
(b) u i n ∈ H 1 u_{\mathrm{in}} \in H^1 u in ∈ H 1 , if d ∈ { 2 , 3 } d \in\{2,3\} d ∈ { 2 , 3 } ; or
(c) u i n ∈ L 2 u_{\mathrm{in}} \in L^2 u in ∈ L 2 , if d ∈ { 2 , 3 } d \in\{2,3\} d ∈ { 2 , 3 } .
Then there are infinitely many weak solutions of (1.1) on [ 0 , 1 ] [0,1] [ 0 , 1 ] with data ( u i n , f 0 ) (u_{\mathrm{in}},f_0) ( u in , f 0 ) such that u ∈ L t p L x 2 ∩ L t 1 L ∞ u \in L^p_t L^2_x \cap L^1_t L^\infty u ∈ L t p L x 2 ∩ L t 1 L ∞ ; f f f is bounded, has bounded velocity support, belongs to C t s L x 1 C^s_t L^1_x C t s L x 1 for every s < ∞ s < \infty s < ∞ , and is the unique bounded (hence renormalized) solution of the kinetic equation with velocity field u u u ; and the total momentum is conserved. Moreover u ∈ L p ( t 1 , 1 ; L ∞ ) ∩ L 1 ( t 1 , 1 ; W 1 , q ) u \in L^p(t_1,1;L^\infty) \cap L^1(t_1,1;W^{1,q}) u ∈ L p ( t 1 , 1 ; L ∞ ) ∩ L 1 ( t 1 , 1 ; W 1 , q ) for some t 1 ∈ [ 0 , 1 2 ) t_1 \in[0,\frac{1}{2}) t 1 ∈ [ 0 , 2 1 ) , with t 1 = 0 t_1 = 0 t 1 = 0 in cases (a), (b), and u ∈ L p ( 0 , 1 ; L ∞ ) u \in L^p(0,1;L^\infty) u ∈ L p ( 0 , 1 ; L ∞ ) unless d = 3 d = 3 d = 3 in case (c). Infinitely many of them are not of Leray–Hopf type. For d = 2 d = 2 d = 2 in case (c), at most one of them belongs to L ∞ ( 0 , 1 ; L 2 ) ∩ L 2 ( 0 , 1 ; H 1 ) L^\infty(0,1;L^2) \cap L^2(0,1;H^1) L ∞ ( 0 , 1 ; L 2 ) ∩ L 2 ( 0 , 1 ; H 1 ) ; in particular at most one is of Leray–Hopf type.
Theorem 1.4 follows from Theorems 7.2 and 7.3 and Remark 7.4. For d = 2 d = 2 d = 2 in case (c), together with the uniqueness theorem of [20 ] , whose decay condition on f 0 f_0 f 0 holds here, it shows that the energy class is exactly the borderline of uniqueness for every finite-energy datum: uniqueness holds in L t ∞ L 2 ∩ L t 2 H 1 L^\infty_t L^2 \cap L^2_t H^1 L t ∞ L 2 ∩ L t 2 H 1 , even without the energy inequality, and fails as soon as one leaves it, even for velocities in L t p L ∞ L^p_t L^\infty L t p L ∞ , p < 2 p < 2 p < 2 , and bounded particle densities.
The tools developed for rough data also give a Leray-type structure theorem in three dimensions.
Theorem 1.5 (Structure of Leray solutions, d = 3 d = 3 d = 3 ). Let u i n ∈ L 2 ( T 3 ) u_{\mathrm{in}} \in L^2(\mathbb{T}^3) u in ∈ L 2 ( T 3 ) be divergence-free, and let f 0 ∈ L ∞ f_0 \in L^\infty f 0 ∈ L ∞ be nonnegative and vanish for large ∣ v ∣ |v| ∣ v ∣ . Every Leray solution of (1.1) satisfying the strong energy inequality is strong outside a relatively closed set of times S ⊂ ( 0 , 1 ) S \subset(0,1) S ⊂ ( 0 , 1 ) with H 1 / 2 ( S ) = 0 H^{1/2}(S) = 0 H 1/2 ( S ) = 0 . More precisely, the conclusions of Theorem 10.1 hold. Such solutions exist for every such datum (Theorem 11.4 ).
Together with Theorem 1.4 (c), this gives the following picture for d = 3 d = 3 d = 3 and every u i n ∈ L 2 u_{\mathrm{in}} \in L^2 u in ∈ L 2 : Leray solutions with the strong energy inequality exist and are strong on an open set of full measure, while there are infinitely many weak solutions with velocity in L t 1 L ∞ L^1_tL^\infty L t 1 L ∞ that coincide with one of them on an initial interval and fail the energy inequality.
Main ideas
Perturbing the fluid only. One could try to let the particles produce the nonuniqueness, as the density does for the inhomogeneous Navier–Stokes equations [23 ] . There, small density weights a k a_k a k are moved from the fluid velocity u u u to concentrated profiles transported with large velocities V k V_k V k , the fluid being set equal to V k V_k V k on their support; the mismatch a k ( V k − u ) a_k(V_k-u) a k ( V k − u ) and a k ( V k ⊗ V k − u ⊗ u ) a_k(V_k \otimes V_k-u \otimes u) a k ( V k ⊗ V k − u ⊗ u ) of the mass and momentum fluxes cancels both defects, because the density enters the single momentum equation through ρ u ⊗ u \rho u \otimes u ρ u ⊗ u . This fails for VNS for two structural reasons. Particles moving with the local fluid velocity exert no drag, and the momentum flux of particles that solve the kinetic equation is balanced by their own momentum, so they cannot cancel a Reynolds stress of the fluid. And in phase space the flux of the particles at a given velocity is that velocity times their mass: a flux defect at bounded velocities can be changed only by changing the distribution function there by a comparable amount, while the force, the only field that can be perturbed, produces a flux in v v v whose oscillatory part is not balanced by the transport of the profiles. We therefore perturb the fluid as in [12 ] and solve the kinetic equation exactly at every step: f n = [ f ] n f_n = [f]_n f n = [ f ] n is the push-forward of f 0 f_0 f 0 along the phase-space flow of u n u_n u n , and the approximate solutions solve a VNS–Reynolds system (Definition 2.6) in which only the fluid equation carries a defect. Since u ∈ L t 1 W 1 , q u \in L^1_tW^{1,q} u ∈ L t 1 W 1 , q , the DiPerna–Lions theory [15 , 4 ] applies to ( v , u − v ) (v,u-v) ( v , u − v ) , and the distribution function is the renormalized solution determined by u u u . The situation is thus opposite to that of [23 ] , where the density is not renormalized and the velocity is not bounded.
Adding a fluid perturbation w w w changes f [ u ] f[u] f [ u ] to f [ u + w ] f[u+w] f [ u + w ] . The main estimate is a dual stability bound (Lemma 3.2): for every test function ψ \psi ψ ,
∣ ∬ ( f [ u + w ] − f [ u ] ) ( t ) ψ ∣ ≤ e 1 + ∥ ∇ u ∥ L t 1 L x ∞ [ ψ ] C 1 M 0 ∥ w ∥ L t 1 L ∞ . \left|\iint(f[u+w]-f[u])(t)\,\psi\right| \le e^{1+\|\nabla u\|_{L^{1}_{t}L^{\infty}_{x}}}[\psi]_{C^{1}}M_{0}\|w\|_{L^{1}_{t}L^{\infty}}. ∬ ( f [ u + w ] − f [ u ]) ( t ) ψ ≤ e 1 + ∥∇ u ∥ L t 1 L x ∞ [ ψ ] C 1 M 0 ∥ w ∥ L t 1 L ∞ .
Only the Lipschitz norm of the old field enters; the gradient of w w w , which is very large, is not used. The constant depends only on the previous step, so it can be compensated by choosing δ n \delta_n δ n after u n u_n u n , while the temporal intermittency of [12 ] makes ∥ w ∥ L t 1 L ∞ \|w\|_{L^{1}_{t}L^{\infty}} ∥ w ∥ L t 1 L ∞ arbitrarily small. The change in the drag force is converted into a small Reynolds stress by an antidivergence estimate in a weak norm (Lemma 3.3), and a spatially constant correction, obtained by a Volterra fixed point, restores the exact momentum balance (Lemma 3.8). This reduces the iteration to a Navier–Stokes step (Proposition 2.8), which is the step of [12 ] with minor adjustments (§ 4 ; full proof in Appendices § 14 and § 16 ).
Memory and causal gluing. For Theorem 1.3 , the stress of an approximate solution is to be pushed onto few short intervals, the bad set, and the solution on the long good intervals between them is to be replaced by an exact solution. By the memory effect described in § 1.2 , the replacement on a good interval [ a , b ] [a,b] [ a , b ] depends on every modification made before a a a , including the convex integration perturbations on earlier bad intervals, so the construction must be causal (§ 6.6 and Figure 1 ). It maintains a current global solution and sweeps [ 0 , 1 ] [0,1] [ 0 , 1 ] from left to right in two passes. In the first pass, a segment corrector (Proposition 6.8 ) re-solves each good interval with the current particle state and glues the result in thin layers, so that the stress is concentrated on the bad set. In the second pass, a localized convex integration step is applied at each junction, and the next good interval is re-solved with the particle state that this step has produced. The two passes cannot be merged (see § 6.6 ). In each corrector, the drag difference created by all earlier modifications enters as a forcing, not as an error to be absorbed later.
Figure 1. (a) A perturbation w w w of the velocity on a short interval [ a , b ] [a,b] [ a , b ] changes the particle state, and hence the drag force, at all later times. (b) The causal construction of § 6.6 . In pass 1, each good interval G i G_i G i is re-solved with the particle state produced by everything before it, and the stress is concentrated on the junctions J i J_i J i . In pass 2, a localized convex integration step (CI) is applied at each junction, and the next good interval is re-solved again. The corrector on G 1 G_1 G 1 depends, through the particles, on the CI perturbation at J 1 J_1 J 1 , which is built to cancel the stress left by pass 1, so the two passes cannot be merged.
The scheme closes because of the following four points.
A linear kinetic duality in ( W 1 , 1 ) ∗ (W^{1,1})^* ( W 1 , 1 ) ∗ , which measures the perturbation in L t 1 L x q L^{1}_{t}L^{q}_{x} L t 1 L x q rather than L t 1 L x ∞ L^{1}_{t}L^{\infty}_{x} L t 1 L x ∞ (Lemma 6.5 ). The instantaneous part of the drag difference, which is not controlled by the past of the perturbation, is kept separate (Remark 6.7).
A two-norm fixed point for the corrector: bounded in maximal regularity, small in a space with one derivative less, and hence, by interpolation, small in L t 2 W 1 , ∞ L^{2}_{t}W^{1,\infty} L t 2 W 1 , ∞ .
Constants fixed before the scales. Kinetic comparisons are started at the beginning a a a of each segment from the common particle state, so the size of the base is measured only after a a a . Earlier convex integration perturbations, which are small in L t 1 W 1 , q L^{1}_{t}W^{1,q} L t 1 W 1 , q but large in L t ∞ W 1 , q L^{\infty}_{t}W^{1,q} L t ∞ W 1 , q , vanish there and do not enter. This allows us to fix all upper bounds in two levels, one per pass, before the scales are chosen.
Mean-preserving gluing (Proposition 6.9). The spatial mean of the fluid is driven by the momentum exchange with the particles, and so it too has memory. Cutting off the mean of a corrector would create a force with nonzero mean, which is not the divergence of a stress. We cut off only the mean-zero part; the mean is carried through the gluing layers and follows the exact momentum balance.
Rough data. The memory also governs the extension to rough data. The kinetic estimates depend on the velocity only through its Lipschitz norm integrated in time, and extend to fields in L t 1 W 1 , ∞ L^{1}_{t}W^{1,\infty} L t 1 W 1 , ∞ and bounded kinetic data via the Carathéodory flow of ( v , u − v ) (v,u-v) ( v , u − v ) (Lemma 8.1). The convex integration step needs a smooth background; we run it on a causal space–time mollification of the background, which only looks at the past, and put the mollification error into the new stress (Lemma 8.4 ). Causality keeps the stress zero on an initial interval and leaves the particle state there untouched. The iteration starts from a weak solution that is Lipschitz only after some time t 1 t_1 t 1 (Definition 7.1 ); all kinetic comparisons start at t 1 t_1 t 1 from the common particle state (Lemma 8.6 ), so before t 1 t_1 t 1 only u ∈ L t 1 W x 1 , 1 ∩ L t 1 L ∞ u\in L^{1}_{t}W^{1,1}_{x}\cap L^{1}_{t}L^{\infty} u ∈ L t 1 W x 1 , 1 ∩ L t 1 L ∞ is used. Such strong continuations are obtained by maximal regularity (Proposition 9.1), by an H 1 H^1 H 1 theory for d ≤ 3 d\le3 d ≤ 3 whose existence time depends only on the H 1 H^1 H 1 norm and on bounds for f f f (Proposition 9.2), and, for L 2 L^2 L 2 data, from an energy solution after a time at which it is in H 1 H^1 H 1 . Before that time the velocity support of f f f must stay bounded, which requires u ∈ L t 1 L ∞ u\in L^{1}_{t}L^{\infty} u ∈ L t 1 L ∞ : for d = 2 d=2 d = 2 we prove a periodic version of [20 Lemma 3] for energy solutions (Lemma 11.5), and for d = 3 d=3 d = 3 we propagate the third and sixth velocity moments using only the energy inequality (Lemma 11.2). In both dimensions the energy solution is the one of Theorem 11.4 , so the only result of [20 ] used in this paper is their uniqueness theorem [20 Theorem 1] , applied on [ 0 , 1 ] [0,1] [ 0 , 1 ] ; its proof is local in time.
Sharpness. For Theorem 1.2 , a duality argument as in [12 Appendix A] , after [25 ] , identifies any L t , x 2 L^2_{t,x} L t , x 2 weak solution of the Oseen problem with drift in L t 2 L ∞ L^2_tL^\infty L t 2 L ∞ with its strong solution. Since the drag force of a critical solution is only bounded in x x x , maximal L t 2 L q L^2_tL^q L t 2 L q regularity is used to obtain ∇ u ∈ L t 2 L ∞ \nabla u \in L^2_tL^\infty ∇ u ∈ L t 2 L ∞ , which makes the phase-space flow Lipschitz. Uniqueness then follows from a Gronwall estimate for
∥ u 1 − u 2 ∥ L 2 2 + ∫ f 0 ( z ) ∣ Z 1 ( t , z ) − Z 2 ( t , z ) ∣ 2 d z , \lVert u_1-u_2\rVert_{L^2}^2+\int f_0(z)\lvert Z_1(t,z)-Z_2(t,z)\rvert^2\,\mathrm{d}z, ∥ u 1 − u 2 ∥ L 2 2 + ∫ f 0 ( z ) ∣ Z 1 ( t , z ) − Z 2 ( t , z ) ∣ 2 d z ,
where Z i Z_i Z i are the phase-space flows, following Loeper’s functional [24 ] as used in [20 ] ; a maximal-function inequality replaces the C 1 C^1 C 1 norm of the test function.
Structure theorem. The same Lagrangian argument gives weak–strong uniqueness against Lipschitz strong solutions (Lemma 11.3 ). Leray solutions with the strong energy inequality are constructed by a Galerkin scheme for the fluid in which the kinetic equation is again solved exactly (Theorem 11.4 ); the approximate solutions then satisfy the energy equality on every interval. With the velocity bound and the H 1 H^1 H 1 theory, Leray’s covering argument gives Theorem 1.5 .
The inviscid case. The convex integration step does not use the Laplacian (Remark B.9), and the kinetic estimates of § 3 are insensitive to viscosity; we expect the nonuniqueness part of Theorem 2.3 to extend to the Vlasov–Euler system, but have not checked the details. The results of § 5 and § 6 use the viscosity in an essential way.
Further questions
We list some problems suggested by this work.
Particle-driven nonuniqueness. Can the particles, rather than the fluid, carry the nonuniqueness? This would yield distribution functions that are not renormalized, and velocities with more Sobolev regularity in space and continuity in time, as in [23 ] , where the velocity is in C t H s C_tH^s C t H s , s < 3 2 s<\frac{3}{2} s < 2 3 , but is not bounded. The obstructions described above suggest that a new building block is needed.
Kinetic transport with rough forces. For the linear equation ∂ t f + v ⋅ ∇ x f + div v ( E f ) = 0 \partial_t f+v\cdot\nabla_x f+\operatorname{div}_v(Ef)=0 ∂ t f + v ⋅ ∇ x f + div v ( E f ) = 0 , Champagnat and Jabin proved well-posedness of the Lagrangian flow for E ∈ H 3 / 4 E\in H^{3/4} E ∈ H 3/4 [10 ] . Whether non-Lagrangian solutions exist for Sobolev force fields below this threshold is, as far as we know, open. Convex integration in phase space is constrained by the fact that the x x x -component of the field is v v v and cannot be perturbed.
C ∞ C^\infty C ∞ regularity of the singular set in Theorem 1.3 , and nonuniqueness for arbitrary L 2 L^2 L 2 data when d ≥ 4 d\ge4 d ≥ 4 (the cases d ≤ 3 d\le3 d ≤ 3 are covered by Theorem 1.4 ). Already for the heat flow of a general L 2 L^2 L 2 datum, ∥ e t Δ u i n ∥ L ∞ ∼ t − d / 4 \lVert e^{t\Delta}u_{\mathrm{in}}\rVert_{L^\infty}\sim t^{-d/4} ∥ e t Δ u in ∥ L ∞ ∼ t − d /4 is not integrable near t = 0 t=0 t = 0 when d ≥ 4 d\ge4 d ≥ 4 .
Critical data such as u i n ∈ H 1 / 2 ( T 3 ) u_{\mathrm{in}}\in H^{1/2}(\mathbb{T}^3) u in ∈ H 1/2 ( T 3 ) ; a Fujita–Kato type solution with bounded drag force should provide a strong continuation.
The strong energy inequality. Theorem 11.4 provides Leray solutions satisfying it. Whether every energy solution satisfies it, for instance those constructed in [6 ] , is open, as for the Navier–Stokes equations.
Other critical spaces X p , q X^{p,q} X p , q , 2 p + d q = 1 \frac{2}{p}+\frac{d}{q}=1 p 2 + q d = 1 , q < ∞ q<\infty q < ∞ , in Theorem 1.2 , and kinetic models with diffusion in velocity (Vlasov–Fokker–Planck–Navier–Stokes).
Structure of the paper
Section 2 introduces weak solutions and the VNS–Reynolds system, states the main theorem in its precise form, and reduces it to the Navier–Stokes step of Proposition 2.8 . Section 3 contains the kinetic lemmas: characteristics, dual stability, the coupling error, the kinetic limit and the mean-momentum correction. Section 4 recalls the Navier–Stokes step of [12 ] and lists the differences; its full proof is in Appendices A and B. Section 5 proves the uniqueness in the critical class. Section 6 proves the intervals of regularity. Section 7 defines strong continuations and states the results for rough data; Section 8 extends the construction to Lipschitz backgrounds and proves Theorem 7.2, and Section 9 constructs strong continuations. Section 10 proves the structure theorem, and Section 11 contains the results on energy solutions used there: a velocity bound, weak–strong uniqueness, and the existence of Leray solutions with the strong energy inequality.
Weak solutions, main results and the iteration
Notation
Throughout, d ≥ 2 d \ge2 d ≥ 2 , T d = R d / Z d \mathbb{T}^{d} = \mathbb{R}^{d}/\mathbb{Z}^{d} T d = R d / Z d with ∣ T d ∣ = 1 |\mathbb{T}^{d}| = 1 ∣ T d ∣ = 1 , and the time interval is [ 0 , 1 ] [0,1] [ 0 , 1 ] . We write z = ( x , v ) ∈ T d × R d z = (x,v) \in\mathbb{T}^{d} \times\mathbb{R}^{d} z = ( x , v ) ∈ T d × R d and f T d g d x f_{\mathbb{T}^{d}} g\,dx f T d g d x . Norms such as L t p L x q L_{t}^{p}L_{x}^{q} L t p L x q refer to L p ( 0 , 1 ; L q ( T d ) ) L^{p}(0,1;L^{q}(\mathbb{T}^{d})) L p ( 0 , 1 ; L q ( T d )) . For a , b ∈ R d a,b \in\mathbb{R}^{d} a , b ∈ R d let a ⊗ ∘ b = a ⊗ b − 1 d ( a ⋅ b ) Id a \mathbin{\mathop{\otimes}\limits^{\circ}} b = a \otimes b - \frac{1}{d}(a \cdot b)\operatorname{Id} a ⊗ ∘ b = a ⊗ b − d 1 ( a ⋅ b ) Id , and let S 0 d × d S_{0}^{d \times d} S 0 d × d be the space of symmetric traceless d × d d \times d d × d matrices.
For a smooth vector field g g g on T d \mathbb{T}^{d} T d we use the projection onto mean-zero divergence-free fields, P g = g − ∇ Δ − 1 div g − ∫ T d g Pg = g - \nabla\Delta^{-1}\operatorname{div}g - \int_{\mathbb{T}^{d}} g P g = g − ∇ Δ − 1 div g − ∫ T d g , and the antidivergence R \mathcal{R} R of [12 Definition B.2] . We recall that R g \mathcal{R}g R g is S 0 d × d S_{0}^{d \times d} S 0 d × d -valued, that
div R g = g − ∫ T d g , R ( c o n s t ) = 0 , ∥ R g ∥ L q ≤ C d , q ∥ g ∥ L q ( 1 ≤ q ≤ ∞ ) , (2) \operatorname{div}\mathcal{R}g = g - \int_{\mathbb{T}^{d}} g,\qquad\mathcal{R}(\mathrm{const}) = 0,\qquad\|\mathcal{R}g\|_{L^{q}} \le C_{d,q}\|g\|_{L^{q}}\quad(1 \le q \le\infty), \tag*{(2)} div R g = g − ∫ T d g , R ( const ) = 0 , ∥ R g ∥ L q ≤ C d , q ∥ g ∥ L q ( 1 ≤ q ≤ ∞ ) , (2)
see [12 Theorem B.3] , and that R \mathcal{R} R acts in space only, so it preserves smoothness in time and vanishing on time intervals. Moreover R g = K ∗ g \mathcal{R}g = K * g R g = K ∗ g for a matrix-valued kernel K K K that is smooth on T d ∖ { 0 } \mathbb{T}^{d} \setminus\{0\} T d ∖ { 0 } and satisfies
∣ K ( z ) ∣ ≤ C d dist ( z , Z d ) 1 − d , ∣ ∇ K ( z ) ∣ ≤ C d dist ( z , Z d ) − d . (3) |K(z)| \le C_{d}\operatorname{dist}(z,\mathbb{Z}^{d})^{1-d},\qquad|\nabla K(z)| \le C_{d}\operatorname{dist}(z,\mathbb{Z}^{d})^{-d}. \tag*{(3)} ∣ K ( z ) ∣ ≤ C d dist ( z , Z d ) 1 − d , ∣∇ K ( z ) ∣ ≤ C d dist ( z , Z d ) − d . (3)
Indeed, the entries of K K K are linear combinations of first derivatives of the periodic Green function G G G of Δ \Delta Δ and third derivatives of G 2 = G ∗ G G_{2}=G*G G 2 = G ∗ G ; near the origin G G G and G 2 G_{2} G 2 differ from the Euclidean fundamental solutions of Δ \Delta Δ and Δ 2 \Delta^{2} Δ 2 by smooth functions, and the logarithms occurring for d ∈ { 2 , 4 } d \in\{2,4\} d ∈ { 2 , 4 } disappear after the indicated number of derivatives.
Kinetic notation. The kinetic datum f 0 ∈ C c ∞ ( T d × R d ) f_{0} \in C_{c}^{\infty}(\mathbb{T}^{d} \times\mathbb{R}^{d}) f 0 ∈ C c ∞ ( T d × R d ) is fixed and nonnegative. We set
supp f 0 ⊂ T d × B R v 0 , M 0 = ∥ f 0 ∥ L 1 , M ∞ = ∥ f 0 ∥ L ∞ , J 0 = ∬ v f 0 d z . \operatorname{supp} f_{0} \subset\mathbb{T}^{d} \times B_{R_{v}^{0}},\qquad M_{0}=\|f_{0}\|_{L^{1}},\qquad M_{\infty}=\|f_{0}\|_{L^{\infty}},\qquad J_{0}=\iint vf_{0}\,dz. supp f 0 ⊂ T d × B R v 0 , M 0 = ∥ f 0 ∥ L 1 , M ∞ = ∥ f 0 ∥ L ∞ , J 0 = ∬ v f 0 d z .
Radii in velocity are denoted by R v R^{v} R v , with subscripts ( R 0 v , R 1 v , R ∞ v , … ) (R_{0}^{v},R_{1}^{v},R_{\infty}^{v},\ldots) ( R 0 v , R 1 v , R ∞ v , … ) , to distinguish them from Reynolds stresses, which are always denoted by R R R . Let V = C ( [ 0 , 1 ] ; C 1 ( T d ; R d ) ) \mathcal{V}=C([0,1];C^{1}(\mathbb{T}^{d};\mathbb{R}^{d})) V = C ([ 0 , 1 ] ; C 1 ( T d ; R d )) . For U ∈ V U \in\mathcal{V} U ∈ V we denote by Z U ( t ; s , z ) Z_{U}(t;s,z) Z U ( t ; s , z ) the flow of X ˙ = V \dot{X}=V X ˙ = V , V ˙ = U ( t , X ) − V \dot{V}=U(t,X)-V V ˙ = U ( t , X ) − V , and by f [ U ] f[U] f [ U ] the solution of
∂ t f + v ⋅ ∇ x f + div v [ ( U − v ) f ] = 0 , f ( 0 ) = f 0 . (4) \partial_{t}f+v\cdot\nabla_{x}f+\operatorname{div}_{v}\bigl[(U-v)f\bigr]=0,\qquad f(0)=f_{0}. \tag*{(4)} ∂ t f + v ⋅ ∇ x f + div v [ ( U − v ) f ] = 0 , f ( 0 ) = f 0 . (4)
We write
ρ [ U ] = ∫ f [ U ] d v , j [ U ] = ∫ v f [ U ] d v , J [ U ] ( t ) = ∬ v f [ U ] ( t ) d z , \rho[U]=\int f[U]\,dv,\qquad j[U]=\int vf[U]\,dv,\qquad J[U](t)=\iint vf[U](t)\,dz, ρ [ U ] = ∫ f [ U ] d v , j [ U ] = ∫ v f [ U ] d v , J [ U ] ( t ) = ∬ v f [ U ] ( t ) d z ,
and, for a distribution function f f f and a velocity field U U U , the drag force
F [ f , U ] ( t , x ) = ∫ R d ( v − U ( t , x ) ) f ( t , x , v ) d v . F[f,U](t,x)=\int_{\mathbb{R}^{d}}(v-U(t,x))f(t,x,v)\,dv. F [ f , U ] ( t , x ) = ∫ R d ( v − U ( t , x )) f ( t , x , v ) d v .
Basic properties of f [ U ] f[U] f [ U ] are collected in Lemma 3.1.
Weak solutions
The Vlasov–Navier–Stokes (VNS) system reads
{ ∂ t f + v ⋅ ∇ x f + div v [ ( u − v ) f ] = 0 , ∂ t u − Δ u + div ( u ⊗ u ) + ∇ p = F [ f , u ] , div u = 0 , ( f , u ) ( 0 ) = ( f 0 , u i n ) . (5) \left\{
\begin{aligned}
\partial_{t}f+v\cdot\nabla_{x}f+\operatorname{div}_{v}\bigl[(u-v)f\bigr] &= 0,\\
\partial_{t}u-\Delta u+\operatorname{div}(u\otimes u)+\nabla p &= F[f,u],\qquad\operatorname{div}u=0,
\end{aligned}
\right.
\qquad(f,u)(0)=(f_{0},u_{\mathrm{in}}). \tag*{(5)} { ∂ t f + v ⋅ ∇ x f + div v [ ( u − v ) f ] ∂ t u − Δ u + div ( u ⊗ u ) + ∇ p = 0 , = F [ f , u ] , div u = 0 , ( f , u ) ( 0 ) = ( f 0 , u in ) . (5)
Definition 2.1 (Weak solutions). Let u i n ∈ L 2 ( T d ) u_{\mathrm{in}} \in L^{2}(\mathbb{T}^{d}) u in ∈ L 2 ( T d ) be weakly divergence-free. A pair ( u , f ) (u,f) ( u , f ) is a weak solution of (2.4) on [ 0 , 1 ] [0,1] [ 0 , 1 ] with data ( u i n , f 0 ) (u_{\mathrm{in}},f_{0}) ( u in , f 0 ) if:
(i) u ∈ L 2 ( ( 0 , 1 ) × T d ) ∩ L 1 ( 0 , 1 ; L ∞ ( T d ) ) u \in L^{2}((0,1) \times\mathbb{T}^{d}) \cap L^{1}(0,1; L^{\infty}(\mathbb{T}^{d})) u ∈ L 2 (( 0 , 1 ) × T d ) ∩ L 1 ( 0 , 1 ; L ∞ ( T d )) and u ( t ) u(t) u ( t ) is weakly divergence-free for a.e. t t t ;
(ii) f ∈ L ∞ ( ( 0 , 1 ) × T d × R d ) f \in L^{\infty}((0,1) \times\mathbb{T}^{d} \times\mathbb{R}^{d}) f ∈ L ∞ (( 0 , 1 ) × T d × R d ) , f ≥ 0 f \ge0 f ≥ 0 , and f = 0 f=0 f = 0 for ∣ v ∣ ≥ R v |v| \ge R_{v} ∣ v ∣ ≥ R v , for some R v < ∞ R_{v}<\infty R v < ∞ ;
(iii) for every φ ∈ C c ∞ ( [ 0 , 1 ) × T d × R d ) \varphi\in C_{c}^{\infty}([0,1) \times\mathbb{T}^{d} \times\mathbb{R}^{d}) φ ∈ C c ∞ ([ 0 , 1 ) × T d × R d ) ,
∬ f 0 φ ( 0 ) d z + ∫ 0 1 ∬ f ( ∂ t φ + v ⋅ ∇ x φ + ( u − v ) ⋅ ∇ v φ ) d z d t = 0 ; \iint f_{0}\varphi(0)\,\mathrm{d}z+\int_{0}^{1}\iint f\left(\partial_{t}\varphi+v\cdot\nabla_{x}\varphi+(u-v)\cdot\nabla_{v}\varphi\right)\,\mathrm{d}z\,\mathrm{d}t=0; ∬ f 0 φ ( 0 ) d z + ∫ 0 1 ∬ f ( ∂ t φ + v ⋅ ∇ x φ + ( u − v ) ⋅ ∇ v φ ) d z d t = 0 ;
(iv) for every divergence-free φ ∈ C ∞ ( [ 0 , 1 ] × T d ; R d ) \varphi\in C^{\infty}([0,1] \times\mathbb{T}^{d};\mathbb{R}^{d}) φ ∈ C ∞ ([ 0 , 1 ] × T d ; R d ) vanishing near t = 1 t=1 t = 1 ,
∫ u i n ⋅ φ ( 0 ) d x + ∫ 0 1 ∫ ( u ⋅ ( ∂ t φ + Δ φ ) + u ⊗ u : ∇ φ + F [ f , u ] ⋅ φ ) d x d t = 0. \int u_{\mathrm{in}}\cdot\varphi(0)\,\mathrm{d}x+\int_{0}^{1}\int\left(u\cdot(\partial_{t}\varphi+\Delta\varphi)+u\otimes u:\nabla\varphi+F[f,u]\cdot\varphi\right)\,\mathrm{d}x\,\mathrm{d}t=0. ∫ u in ⋅ φ ( 0 ) d x + ∫ 0 1 ∫ ( u ⋅ ( ∂ t φ + Δ φ ) + u ⊗ u : ∇ φ + F [ f , u ] ⋅ φ ) d x d t = 0.
By (i)–(ii), F [ f , u ] = j f − ρ f u ∈ L 1 ( ( 0 , 1 ) × T d ) F[f,u]=j_{f}-\rho_{f}u\in L^{1}((0,1)\times\mathbb{T}^{d}) F [ f , u ] = j f − ρ f u ∈ L 1 (( 0 , 1 ) × T d ) , so (iv) makes sense. Constant test fields are admissible in (iv), so the definition contains the equation d d t ∫ f u = ∫ f F [ f , u ] \frac{\mathrm{d}}{\mathrm{d}t}\int f u=\int fF[f,u] d t d ∫ f u = ∫ f F [ f , u ] in D ′ ( 0 , 1 ) \mathcal{D}'(0,1) D ′ ( 0 , 1 ) . The spatial mean of u u u is not conserved; the conserved quantity is the total momentum ∫ f u + ∬ v f d z \int f u+\iint vf\,\mathrm{d}z ∫ f u + ∬ v f d z .
Definition 2.2 (Leray–Hopf solutions). A weak solution is of Leray–Hopf type if u ∈ L ∞ ( 0 , 1 ; L 2 ) ∩ L 2 ( 0 , 1 ; H 1 ) u\in L^{\infty}(0,1;L^{2})\cap L^{2}(0,1;H^{1}) u ∈ L ∞ ( 0 , 1 ; L 2 ) ∩ L 2 ( 0 , 1 ; H 1 ) and, for a.e. t ∈ [ 0 , 1 ] t\in[0,1] t ∈ [ 0 , 1 ] ,
E ( t ) + ∫ 0 t ( ∥ ∇ u ∥ L 2 2 + ∬ ∣ v − u ∣ 2 f d z ) d s ≤ E ( 0 ) , E ( t ) = 1 2 ∥ u ( t ) ∥ L 2 2 + 1 2 ∬ ∣ v ∣ 2 f ( t ) d z . \mathcal{E}(t)+\int_{0}^{t}\left(\|\nabla u\|_{L^{2}}^{2}+\iint|v-u|^{2}f\,\mathrm{d}z\right)\,\mathrm{d}s\le\mathcal{E}(0),\qquad\mathcal{E}(t)=\frac{1}{2}\|u(t)\|_{L^{2}}^{2}+\frac{1}{2}\iint|v|^{2}f(t)\,\mathrm{d}z. E ( t ) + ∫ 0 t ( ∥∇ u ∥ L 2 2 + ∬ ∣ v − u ∣ 2 f d z ) d s ≤ E ( 0 ) , E ( t ) = 2 1 ∥ u ( t ) ∥ L 2 2 + 2 1 ∬ ∣ v ∣ 2 f ( t ) d z .
Main results
Theorem 2.3 (Main theorem). Let d ≥ 2 d\ge2 d ≥ 2 , 1 ≤ p < 2 1\le p<2 1 ≤ p < 2 , 1 ≤ q < ∞ 1\le q<\infty 1 ≤ q < ∞ and ε ∈ ( 0 , 1 ) \varepsilon\in(0,1) ε ∈ ( 0 , 1 ) . Let t 0 ∈ ( 0 , 1 ] t_{0}\in(0,1] t 0 ∈ ( 0 , 1 ] and let ( u ˉ , f ˉ ) (\bar{u},\bar{f}) ( u ˉ , f ˉ ) be a smooth solution of (2.4) on [ 0 , t 0 ] [0,t_{0}] [ 0 , t 0 ] with data ( u i n , f 0 ) (u_{\mathrm{in}},f_{0}) ( u in , f 0 ) . Let u ~ ∈ C ∞ ( [ 0 , 1 ] × T d ; R d ) \tilde{u}\in C^{\infty}([0,1]\times\mathbb{T}^{d};\mathbb{R}^{d}) u ~ ∈ C ∞ ([ 0 , 1 ] × T d ; R d ) be divergence-free with u ~ = u ˉ \tilde{u}=\bar{u} u ~ = u ˉ on [ 0 , t 0 ] [0,t_{0}] [ 0 , t 0 ] . Then there is a weak solution ( u , f ) (u,f) ( u , f ) of (2.4) on [ 0 , 1 ] [0,1] [ 0 , 1 ] with data ( u i n , f 0 ) (u_{\mathrm{in}},f_{0}) ( u in , f 0 ) such that:
(1) u ∈ L t , x 2 ∩ L t p L ∞ ∩ L t 1 W 1 , q u\in L_{t,x}^{2}\cap L_{t}^{p}L^{\infty}\cap L_{t}^{1}W^{1,q} u ∈ L t , x 2 ∩ L t p L ∞ ∩ L t 1 W 1 , q ;
(2) ( u , f ) = ( u ˉ , f ˉ ) (u,f)=(\bar{u},\bar{f}) ( u , f ) = ( u ˉ , f ˉ ) on [ 0 , t 0 / 2 ] [0,t_{0}/2] [ 0 , t 0 /2 ] ;
(3) ∥ ( u − f u ) − ( u ~ − f u ~ ) ∥ L t p L ∞ ∩ L t 1 W 1 , q ≤ ε \|(u-fu)-(\tilde{u}-f\tilde{u})\|_{L_{t}^{p}L^{\infty}\cap L_{t}^{1}W^{1,q}}\le\varepsilon ∥ ( u − f u ) − ( u ~ − f u ~ ) ∥ L t p L ∞ ∩ L t 1 W 1 , q ≤ ε ;
(4) 0 ≤ f ≤ e d M ∞ 0\le f\le e^{d}M_{\infty} 0 ≤ f ≤ e d M ∞ , f f f vanishes for ∣ v ∣ ≥ R v ∞ |v|\ge R_{v}^{\infty} ∣ v ∣ ≥ R v ∞ for some R v ∞ < ∞ R_{v}^{\infty}<\infty R v ∞ < ∞ , f ∈ C ( [ 0 , 1 ] ; L s ( T d × R d ) ) f\in C([0,1];L^{s}(\mathbb{T}^{d}\times\mathbb{R}^{d})) f ∈ C ([ 0 , 1 ] ; L s ( T d × R d )) for every s < ∞ s<\infty s < ∞ , and f f f is the unique bounded weak solution of the kinetic equation in Definition 2.1(iii) with velocity field u u u ; in particular f f f is a renormalized solution;
(5) the total momentum is conserved: ∫ f u ( t ) + ∬ v f ( t ) d z = ∫ f u i n + J 0 \int fu(t)+\iint vf(t)\,\mathrm{d}z=\int fu_{\mathrm{in}}+J_{0} ∫ f u ( t ) + ∬ v f ( t ) d z = ∫ f u in + J 0 for a.e. t t t .
The norm in (3) is taken modulo spatial means because the fluid mean is slaved to the particle momentum by (5); it is corrected at each step of the construction (Lemma 3.8).
Corollary 2.4 (Nonuniqueness of the Cauchy problem). Let d ≥ 2 d\ge2 d ≥ 2 , 1 ≤ p < 2 1\le p<2 1 ≤ p < 2 , 1 ≤ q < ∞ 1\le q<\infty 1 ≤ q < ∞ , and let ( u i n , f 0 ) (u_{\mathrm{in}},f_{0}) ( u in , f 0 ) be smooth data for which (2.4) has a smooth solution on [ 0 , t 1 ] [0,t_{1}] [ 0 , t 1 ] for some t 1 ∈ ( 0 , 1 ] t_{1}\in(0,1] t 1 ∈ ( 0 , 1 ] . Then there are infinitely many distinct weak solutions of (2.4) on [ 0 , 1 ] [0,1] [ 0 , 1 ] with data ( u i n , f 0 ) (u_{\mathrm{in}},f_{0}) ( u in , f 0 ) , all in L t , x 2 ∩ L t p L ∞ ∩ L t 1 W 1 , q L_{t,x}^{2}\cap L_{t}^{p}L^{\infty}\cap L_{t}^{1}W^{1,q} L t , x 2 ∩ L t p L ∞ ∩ L t 1 W 1 , q , all coinciding with the smooth solution on [ 0 , t 1 / 4 ] [0,t_{1}/4] [ 0 , t 1 /4 ] . Infinitely many of them are not of Leray–Hopf type.
Remark 2.5. For smooth data with f 0 f_{0} f 0 compactly supported, smooth solutions of (2.4) exist on a short time interval; this is classical and we take it as given. The hypothesis of Corollary 2.4 is thus satisfied by all such data.
The VNS–Reynolds system
Definition 2.6. A smooth solution of the VNS–Reynolds system is a triple ( u , p , R ) (u,p,R) ( u , p , R ) of smooth functions on [ 0 , 1 ] × T d [0,1]\times\mathbb{T}^{d} [ 0 , 1 ] × T d , with u u u vector-valued, div u = 0 \operatorname{div}u=0 div u = 0 , p p p scalar and R R R taking values in S 0 d × d S_{0}^{d\times d} S 0 d × d , such that
∂ t u − Δ u + div ( u ⊗ u ) + ∇ p = F [ f [ u ] , u ] + div R . (6) \partial_{t}u-\Delta u+\operatorname{div}(u\otimes u)+\nabla p=F[f[u],u]+\operatorname{div}R. \tag*{(6)} ∂ t u − Δ u + div ( u ⊗ u ) + ∇ p = F [ f [ u ] , u ] + div R . (6)
The kinetic equation is solved exactly, through f [ u ] f[u] f [ u ] , and the only defect is the stress R R R in the fluid equation. Since div R \operatorname{div} R div R , ∇ p \nabla p ∇ p and div ( u ⊗ u ) \operatorname{div}(u \otimes u) div ( u ⊗ u ) have zero mean, (6) contains the exact mean equation d d t ∫ f = ∫ F [ f [ u ] , u ] \frac{d}{dt}\int f=\int F[f[u],u] d t d ∫ f = ∫ F [ f [ u ] , u ] . By Lemma 3.1(d) this is equivalent to
∫ u ( t ) + J [ u ] ( t ) = ∫ u ( 0 ) + J 0 ( 0 ≤ t ≤ 1 ) . (7) \int u(t)+J[u](t)=\int u(0)+J_0 \qquad(0\le t\le1). \tag*{(7)} ∫ u ( t ) + J [ u ] ( t ) = ∫ u ( 0 ) + J 0 ( 0 ≤ t ≤ 1 ) . (7)
Lemma 2.7 (Initial approximate solution). Under the hypotheses of Theorem 2.3, there is a function m ∈ C ∞ ( [ 0 , 1 ] ; R d ) m\in C^\infty([0,1];\mathbb{R}^d) m ∈ C ∞ ([ 0 , 1 ] ; R d ) with m = 0 m=0 m = 0 on [ 0 , t 0 ] [0,t_0] [ 0 , t 0 ] such that u 0 : = u ~ + m u_0:=\tilde u+m u 0 := u ~ + m , together with suitable p 0 , R 0 p_0,R_0 p 0 , R 0 , is a smooth solution of the VNS–Reynolds system, and R 0 = 0 R_0=0 R 0 = 0 on [ 0 , t 0 ] [0,t_0] [ 0 , t 0 ] .
Proof. Apply Lemma 3.8(a) to U = u ~ U=\tilde u U = u ~ , and let m m m be the resulting correction. On [ 0 , t 0 ] [0,t_0] [ 0 , t 0 ] the pair ( u ~ , f [ u ~ ] ) = ( u ˉ , f ˉ ) (\tilde u,f[\tilde u])=(\bar u,\bar f) ( u ~ , f [ u ~ ]) = ( u ˉ , f ˉ ) solves (2.4), so (7) holds there and Lemma 3.8(b) gives m = 0 m=0 m = 0 on [ 0 , t 0 ] [0,t_0] [ 0 , t 0 ] . Set u 0 = u ~ + m u_0=\tilde u+m u 0 = u ~ + m , which is smooth and divergence-free, and
E : = ∂ t u 0 − Δ u 0 + div ( u 0 ⊗ u 0 ) − F [ f [ u 0 ] , u 0 ] . E:=\partial_tu_0-\Delta u_0+\operatorname{div}(u_0\otimes u_0)-F[f[u_0],u_0]. E := ∂ t u 0 − Δ u 0 + div ( u 0 ⊗ u 0 ) − F [ f [ u 0 ] , u 0 ] .
By Lemma 3.8(a), ∫ E = d d t ∫ u 0 − ∫ F [ f [ u 0 ] , u 0 ] = 0 \int E=\frac{d}{dt}\int u_0-\int F[f[u_0],u_0]=0 ∫ E = d t d ∫ u 0 − ∫ F [ f [ u 0 ] , u 0 ] = 0 . Put p 0 : = − Δ − 1 div E p_0:=-\Delta^{-1}\operatorname{div}E p 0 := − Δ − 1 div E and R 0 : = R P E R_0:=\mathcal{R}PE R 0 := R P E . Since ∫ E = 0 \int E=0 ∫ E = 0 and ∫ P E = 0 \int PE=0 ∫ P E = 0 ,
E + ∇ p 0 = E − ∇ Δ − 1 div E = P E = div R 0 , E+\nabla p_0=E-\nabla\Delta^{-1}\operatorname{div}E=PE=\operatorname{div}R_0, E + ∇ p 0 = E − ∇ Δ − 1 div E = P E = div R 0 ,
which is (6) . On [ 0 , t 0 ] [0,t_0] [ 0 , t 0 ] we have u 0 = u ˉ u_0=\bar u u 0 = u ˉ and f [ u 0 ] = f ˉ f[u_0]=\bar f f [ u 0 ] = f ˉ by causality (Lemma 3.1(b)), hence E = − ∇ p ˉ E=-\nabla\bar p E = − ∇ p ˉ is a gradient, P E = 0 PE=0 P E = 0 and R 0 = 0 R_0=0 R 0 = 0 .
The iteration
The convex integration step is a Navier–Stokes statement:
Proposition 2.8 (Navier–Stokes step, after [12 ] ). Let d ≥ 2 d\ge2 d ≥ 2 , 1 ≤ p < 2 1\le p<2 1 ≤ p < 2 , 1 ≤ q < ∞ 1\le q<\infty 1 ≤ q < ∞ . There are M ≥ 1 M\ge1 M ≥ 1 depending only on d d d , and r ∈ ( 1 , d d − 1 ) r\in\left(1,\frac{d}{d-1}\right) r ∈ ( 1 , d − 1 d ) depending only on d , p , q d,p,q d , p , q , with the following property. Let u ∈ C ∞ ( [ 0 , 1 ] × T d ; R d ) u\in C^\infty([0,1]\times\mathbb{T}^d;\mathbb{R}^d) u ∈ C ∞ ([ 0 , 1 ] × T d ; R d ) be divergence-free, R ∈ C ∞ ( [ 0 , 1 ] × T d ; S 0 d × d ) R\in C^\infty([0,1]\times\mathbb{T}^d;\mathcal{S}^{d\times d}_0) R ∈ C ∞ ([ 0 , 1 ] × T d ; S 0 d × d ) , and 0 < s < t ∗ ≤ 1 0<s<t_*\le1 0 < s < t ∗ ≤ 1 with R = 0 R=0 R = 0 on [ 0 , t ∗ ] [0,t_*] [ 0 , t ∗ ] . For every δ > 0 \delta>0 δ > 0 there are smooth w , P , R C L w,P,R^{\mathrm{CL}} w , P , R CL , with div w = 0 \operatorname{div}w=0 div w = 0 and R C L R^{\mathrm{CL}} R CL taking values in S 0 d × d \mathcal{S}^{d\times d}_0 S 0 d × d , such that
∂ t w − Δ w + div ( u ⊗ w + w ⊗ u + w ⊗ w ) + div R + ∇ P = div R C L \partial_tw-\Delta w+\operatorname{div}(u\otimes w+w\otimes u+w\otimes w)+\operatorname{div}R+\nabla P=\operatorname{div}R^{\mathrm{CL}} ∂ t w − Δ w + div ( u ⊗ w + w ⊗ u + w ⊗ w ) + div R + ∇ P = div R CL ;
∫ w ( t ) = 0 \int w(t)=0 ∫ w ( t ) = 0 for all t t t , and w = 0 w=0 w = 0 , R C L = 0 R^{\mathrm{CL}}=0 R CL = 0 on [ 0 , s ] [0,s] [ 0 , s ] ;
∥ w ∥ L t , x 2 ≤ M ∥ R ∥ L t , x 1 1 / 2 \|w\|_{L^2_{t,x}}\le M\|R\|_{L^1_{t,x}}^{1/2} ∥ w ∥ L t , x 2 ≤ M ∥ R ∥ L t , x 1 1/2 ;
∥ w ∥ L t p L ∞ + ∥ w ∥ L t 1 W 1 , q ≤ δ \|w\|_{L^p_tL^\infty}+\|w\|_{L^1_tW^{1,q}}\le\delta ∥ w ∥ L t p L ∞ + ∥ w ∥ L t 1 W 1 , q ≤ δ ;
∥ R C L ∥ L 1 ( 0 , 1 ; L r ) ≤ δ \|R^{\mathrm{CL}}\|_{L^1(0,1;L^r)}\le\delta ∥ R CL ∥ L 1 ( 0 , 1 ; L r ) ≤ δ .
:::
Proposition 2.10 (VNS iteration step). Let d , p , q d,p,q d , p , q and M , r M,r M , r be as in Proposition 2.8. Let ( u , p u , R ) (u,p_u,R) ( u , p u , R ) be a smooth solution of the VNS–Reynolds system and 0 < s < t ∗ ≤ 1 0<s<t_*\le1 0 < s < t ∗ ≤ 1 with R = 0 R=0 R = 0 on [ 0 , t ∗ ] [0,t_*] [ 0 , t ∗ ] . For every δ ∈ ( 0 , 1 ] \delta\in(0,1] δ ∈ ( 0 , 1 ] there is a smooth solution ( u 1 , p 1 , R 1 ) (u_1,p_1,R_1) ( u 1 , p 1 , R 1 ) of the VNS–Reynolds system such that
u 1 = u u_1=u u 1 = u and R 1 = 0 R_1=0 R 1 = 0 on [ 0 , s ] [0,s] [ 0 , s ] ;
∥ u 1 − u ∥ L t , x 2 ≤ M ∥ R ∥ L t , x 1 1 / 2 + δ \|u_1-u\|_{L^2_{t,x}}\le M\|R\|_{L^1_{t,x}}^{1/2}+\delta ∥ u 1 − u ∥ L t , x 2 ≤ M ∥ R ∥ L t , x 1 1/2 + δ ;
∥ u 1 − u ∥ L t p L ∞ + ∥ u 1 − u ∥ L t 1 W 1 , q ≤ δ \|u_1-u\|_{L^p_tL^\infty}+\|u_1-u\|_{L^1_tW^{1,q}}\le\delta ∥ u 1 − u ∥ L t p L ∞ + ∥ u 1 − u ∥ L t 1 W 1 , q ≤ δ ;
∥ R 1 ∥ L 1 ( 0 , 1 ; L r ) ≤ δ \|R_1\|_{L^1(0,1;L^r)}\le\delta ∥ R 1 ∥ L 1 ( 0 , 1 ; L r ) ≤ δ .
Proposition 2.10 is proved in Section 3.6, assuming Proposition 2.8.
Proof of Theorem 2.3
Step 1: construction. Let ( u 0 , p 0 , R 0 ) (u_0,p_0,R_0) ( u 0 , p 0 , R 0 ) be given by Lemma 2.7, and set s n = t 0 ( 1 2 + 2 − n − 1 ) s_n=t_0\left(\frac{1}{2}+2^{-n-1}\right) s n = t 0 ( 2 1 + 2 − n − 1 ) for n ≥ 0 n\geq0 n ≥ 0 , so that s 0 = t 0 s_0=t_0 s 0 = t 0 and s n ↓ t 0 / 2 s_n\downarrow t_0/2 s n ↓ t 0 /2 . Suppose ( u n , p n , R n ) (u_n,p_n,R_n) ( u n , p n , R n ) is a smooth solution of the VNS–Reynolds system with R n = 0 R_n=0 R n = 0 on [ 0 , s n ] [0,s_n] [ 0 , s n ] . Put Λ n = ∥ ∇ u n ∥ L t 1 L ∞ < ∞ \Lambda_n=\|\nabla u_n\|_{L^1_tL^\infty}<\infty Λ n = ∥∇ u n ∥ L t 1 L ∞ < ∞ and
δ n = min { ε 4 − n − 1 , 4 − n , 2 − n e − 1 − Λ n 1 + M 0 } . \delta_n=\min\left\{\varepsilon4^{-n-1},\ 4^{-n},\ \frac{2^{-n}e^{-1-\Lambda_n}}{1+M_0}\right\}. δ n = min { ε 4 − n − 1 , 4 − n , 1 + M 0 2 − n e − 1 − Λ n } .
Proposition 2.10, applied with ( t ∗ , s ) = ( s n , s n + 1 ) (t_*,s)=(s_n,s_{n+1}) ( t ∗ , s ) = ( s n , s n + 1 ) and δ = δ n \delta=\delta_n δ = δ n , gives ( u n + 1 , p n + 1 , R n + 1 ) (u_{n+1},p_{n+1},R_{n+1}) ( u n + 1 , p n + 1 , R n + 1 ) with R n + 1 = 0 R_{n+1}=0 R n + 1 = 0 on [ 0 , s n + 1 ] [0,s_{n+1}] [ 0 , s n + 1 ] . By induction, for all n ≥ 0 n\geq0 n ≥ 0 :
(P1) u n + 1 = u n u_{n+1}=u_n u n + 1 = u n on [ 0 , s n + 1 ] [0,s_{n+1}] [ 0 , s n + 1 ] , hence u n = u 0 = u ˉ u_n=u_0=\bar u u n = u 0 = u ˉ on [ 0 , t 0 / 2 ] [0,t_0/2] [ 0 , t 0 /2 ] ;
(P2) ∥ R n + 1 ∥ L t , x 1 ≤ ∥ R n + 1 ∥ L t 1 L x r ≤ δ n \|R_{n+1}\|_{L^1_{t,x}}\leq\|R_{n+1}\|_{L^1_tL^r_x}\leq\delta_n ∥ R n + 1 ∥ L t , x 1 ≤ ∥ R n + 1 ∥ L t 1 L x r ≤ δ n ;
(P3) ∥ u n + 1 − u n ∥ L t p L ∞ + ∥ u n + 1 − u n ∥ L t 1 W 1 , q ≤ δ n \|u_{n+1}-u_n\|_{L^p_tL^\infty}+\|u_{n+1}-u_n\|_{L^1_tW^{1,q}}\leq\delta_n ∥ u n + 1 − u n ∥ L t p L ∞ + ∥ u n + 1 − u n ∥ L t 1 W 1 , q ≤ δ n ;
(P4) ∥ u n + 1 − u n ∥ L t , x 2 ≤ M ∥ R n ∥ L t , x 1 1 / 2 + δ n ≤ M δ n − 1 1 / 2 + δ n ≤ M 2 1 − n + 4 − n \|u_{n+1}-u_n\|_{L^2_{t,x}}\leq M\|R_n\|_{L^1_{t,x}}^{1/2}+\delta_n\leq M\delta_{n-1}^{1/2}+\delta_n\leq M2^{1-n}+4^{-n} ∥ u n + 1 − u n ∥ L t , x 2 ≤ M ∥ R n ∥ L t , x 1 1/2 + δ n ≤ M δ n − 1 1/2 + δ n ≤ M 2 1 − n + 4 − n for n ≥ 1 n\geq1 n ≥ 1 .
Step 2: the velocity. By (P3)–(P4), ( u n ) (u_n) ( u n ) is Cauchy in X : = L t , x 2 ∩ L t p L ∞ ∩ L t 1 W 1 , q X:=L^2_{t,x}\cap L^p_tL^\infty\cap L^1_tW^{1,q} X := L t , x 2 ∩ L t p L ∞ ∩ L t 1 W 1 , q ; let u u u be its limit. Then u ( t ) u(t) u ( t ) is weakly divergence-free for a.e. t t t , and (1) holds. By (P3), ∥ u − u 0 ∥ L t p L ∞ ∩ L t 1 W 1 , q ≤ ∑ n δ n ≤ ε / 3 \|u-u_0\|_{L^p_tL^\infty\cap L^1_tW^{1,q}}\leq\sum_n\delta_n\leq\varepsilon/3 ∥ u − u 0 ∥ L t p L ∞ ∩ L t 1 W 1 , q ≤ ∑ n δ n ≤ ε /3 . Since u 0 = u ~ + m u_0=\widetilde u+m u 0 = u + m with m m m constant in space, ( u − u ˉ ) − ( u ~ − u ~ ˉ ) = ( u − u 0 ) − ( u − u 0 ) ‾ (u-\bar u)-(\widetilde u-\bar{\widetilde u})=(u-u_0)-\overline{(u-u_0)} ( u − u ˉ ) − ( u − u ˉ ) = ( u − u 0 ) − ( u − u 0 ) . Subtracting the mean at most doubles the norms of L ∞ L^\infty L ∞ and W 1 , q W^{1,q} W 1 , q , which gives (3). Since p ≥ 1 p\geq1 p ≥ 1 ,
sup n ∥ u n ∥ L t 1 L ∞ ≤ ∥ u 0 ∥ L t 1 L ∞ + 1 = : K , ∑ n e 1 + Λ n ∥ u n + 1 − u n ∥ L t 1 L ∞ ≤ ∑ n 2 − n < ∞ . (8) \sup_n\|u_n\|_{L^1_tL^\infty}\leq\|u_0\|_{L^1_tL^\infty}+1=:K,\qquad\sum_n e^{1+\Lambda_n}\|u_{n+1}-u_n\|_{L^1_tL^\infty}\leq\sum_n2^{-n}<\infty. \tag*{(8)} n sup ∥ u n ∥ L t 1 L ∞ ≤ ∥ u 0 ∥ L t 1 L ∞ + 1 =: K , n ∑ e 1 + Λ n ∥ u n + 1 − u n ∥ L t 1 L ∞ ≤ n ∑ 2 − n < ∞. (8)
Step 3: the distribution function. By (2.7) and u n → u u_n\to u u n → u in L t 1 L ∞ L^1_tL^\infty L t 1 L ∞ , Lemma 3.5 applies to f n : = f [ u n ] f_n:=f[u_n] f n := f [ u n ] . It yields a limit f f f with 0 ≤ f ≤ e d M ∞ 0\leq f\leq e^{dM_\infty} 0 ≤ f ≤ e d M ∞ , f = 0 f=0 f = 0 for ∣ v ∣ ≥ R v ∞ : = R v 0 + K |v|\geq R_v^\infty:=R_v^0+K ∣ v ∣ ≥ R v ∞ := R v 0 + K , which solves the weak kinetic equation of Definition 2.1(iii) with field u u u ; moreover F [ f n , u n ] → F [ f , u ] F[f_n,u_n]\to F[f,u] F [ f n , u n ] → F [ f , u ] in the sense of Lemma 3.5(d), and J [ u n ] ( t ) → ∬ v f ( t ) d z J[u_n](t)\to\iint vf(t)\,dz J [ u n ] ( t ) → ∬ v f ( t ) d z for every t t t .
Step 4: the fluid equation. Let φ \varphi φ be an admissible test field in Definition 2.1(iv). Testing (2.5) for u n u_n u n with φ \varphi φ and integrating by parts, using u n ( 0 ) = u i n u_n(0)=u_{\mathrm{in}} u n ( 0 ) = u in by (P1),
∫ u i n ⋅ φ ( 0 ) + ∫ 0 1 ∫ ( u n ⋅ ( ∂ t φ + Δ φ ) + u n ⊗ u n : ∇ φ + F [ f n , u n ] ⋅ φ − R n : ∇ φ ) = 0. \int u_{\mathrm{in}}\cdot\varphi(0)+\int_0^1\int\left(u_n\cdot(\partial_t\varphi+\Delta\varphi)+u_n\otimes u_n:\nabla\varphi+F[f_n,u_n]\cdot\varphi-R_n:\nabla\varphi\right)=0. ∫ u in ⋅ φ ( 0 ) + ∫ 0 1 ∫ ( u n ⋅ ( ∂ t φ + Δ φ ) + u n ⊗ u n : ∇ φ + F [ f n , u n ] ⋅ φ − R n : ∇ φ ) = 0.
As n → ∞ n\to\infty n → ∞ : u n → u u_n\to u u n → u in L t , x 2 L^2_{t,x} L t , x 2 gives convergence of the first two terms; R n → 0 R_n\to0 R n → 0 in L t , x 1 L^1_{t,x} L t , x 1 by (P2); and ∬ F [ f n , u n ] ⋅ φ → ∬ F [ f , u ] ⋅ φ \iint F[f_n,u_n]\cdot\varphi\to\iint F[f,u]\cdot\varphi ∬ F [ f n , u n ] ⋅ φ → ∬ F [ f , u ] ⋅ φ by Lemma 3.5(d). Hence (iv) holds, and ( u , f ) (u,f) ( u , f ) is a weak solution.
Step 5: properties (2), (4), (5). By (P1), u = u ˉ u=\bar u u = u ˉ on [ 0 , t 0 / 2 ] [0,t_0/2] [ 0 , t 0 /2 ] , and f n = f [ u ˉ ] = f ˉ f_n=f[\bar u]=\bar f f n = f [ u ˉ ] = f ˉ on [ 0 , t 0 / 2 ] [0,t_0/2] [ 0 , t 0 /2 ] by causality, hence f = f ˉ f=\bar f f = f ˉ there; this is (2). Since u ∈ L t 1 W 1 , q ∩ L t 1 L ∞ u\in L^1_tW^{1,q}\cap L^1_tL^\infty u ∈ L t 1 W 1 , q ∩ L t 1 L ∞ , Lemma 3.6 gives the uniqueness, renormalization and continuity statements in (4). Finally, each u n u_n u n satisfies (2.6), which reads ∫ u n ( t ) + J [ u n ] ( t ) = ∫ u i n + J 0 \int u_n(t)+J[u_n](t)=\int u_{\mathrm{in}}+J_0 ∫ u n ( t ) + J [ u n ] ( t ) = ∫ u in + J 0 . Along a subsequence ∫ u n ( t ) → ∫ u ( t ) \int u_n(t)\to\int u(t) ∫ u n ( t ) → ∫ u ( t ) for a.e. t t t , while J [ u n ] ( t ) → ∬ v f ( t ) d z J[u_n](t)\to\iint vf(t)\,dz J [ u n ] ( t ) → ∬ v f ( t ) d z for every t t t ; this gives (5). ∎
Remark 2.11 (Exponents uniform in p p p and q q q ). Theorem 2.3 is stated for fixed p < 2 p<2 p < 2 and q < ∞ q<\infty q < ∞ . A single solution lying in L t p L ∞ ∩ L t 1 W 1 , q L^p_tL^\infty\cap L^1_tW^{1,q} L t p L ∞ ∩ L t 1 W 1 , q for all p < 2 p<2 p < 2 and q < ∞ q<\infty q < ∞ simultaneously can be obtained by a diagonal argument: choose p n ↑ 2 p_n\uparrow2 p n ↑ 2 and q n ↑ ∞ q_n\uparrow\infty q n ↑ ∞ , and apply Proposition 2.10 at step n n n with the exponents ( p n , q n ) (p_n,q_n) ( p n , q n ) , the corresponding r n r_n r n , and δ n \delta_n δ n as above. Since every constant at step n n n is chosen after u n u_n u n is fixed, the construction is unchanged. For fixed m m m and n ≥ m n\geq m n ≥ m , the bound ∥ u n + 1 − u n ∥ L t p m L ∞ ∩ L t 1 W 1 , q m ≤ δ n \|u_{n+1}-u_n\|_{L^{p_m}_tL^\infty\cap L^1_tW^{1,q_m}}\leq\delta_n ∥ u n + 1 − u n ∥ L t p m L ∞ ∩ L t 1 W 1 , q m ≤ δ n implies the same bound in L t p m L ∞ ∩ L t 1 W 1 , q m L^{p_m}_tL^\infty\cap L^1_tW^{1,q_m} L t p m L ∞ ∩ L t 1 W 1 , q m , since ∣ T d ∣ = 1 |\mathbb T^d|=1 ∣ T d ∣ = 1 and the time interval has length one. Hence ( u n ) (u_n) ( u n ) is Cauchy in every L t p m L ∞ ∩ L t 1 W 1 , q m L^{p_m}_tL^\infty\cap L^1_tW^{1,q_m} L t p m L ∞ ∩ L t 1 W 1 , q m , and the limit lies in all of them. The convergence of the stresses, ∥ R n ∥ L t , x 1 ≤ ∥ R n ∥ L 1 L r n ≤ δ n − 1 \|R_n\|_{L^1_{t,x}}\leq\|R_n\|_{L^1L^{r_n}}\leq\delta_{n-1} ∥ R n ∥ L t , x 1 ≤ ∥ R n ∥ L 1 L r n ≤ δ n − 1 , is unaffected. We do not use this refinement elsewhere, and all statements below are to be read with fixed exponents.
Proof of Corollary 2.4
Let ( u ˉ , f ˉ ) (\bar u,\bar f) ( u ˉ , f ˉ ) be the smooth solution on [ 0 , t 1 ] [0,t_1] [ 0 , t 1 ] . Choose χ ∈ C ∞ ( R ; [ 0 , 1 ] ) \chi\in C^\infty(\mathbb R;[0,1]) χ ∈ C ∞ ( R ; [ 0 , 1 ]) with χ = 1 \chi=1 χ = 1 on ( − ∞ , t 1 / 2 ) (-\infty,t_1/2) ( − ∞ , t 1 /2 ) and χ = 0 \chi=0 χ = 0 on [ 3 t 1 / 4 , ∞ ) [3t_1/4,\infty) [ 3 t 1 /4 , ∞ ) , and choose a smooth divergence-free Ψ \Psi Ψ on [ 0 , 1 ] × T d [0,1]\times\mathbb T^d [ 0 , 1 ] × T d with ∫ Ψ ( t ) = 0 \int\Psi(t)=0 ∫ Ψ ( t ) = 0 , Ψ = 0 \Psi=0 Ψ = 0 for t ≤ t 1 t \le t_1 t ≤ t 1 , and ∥ Ψ ∥ L p ( 0 , 1 ; L 2 ) ≥ 1 \|\Psi\|_{L^p(0,1;L^2)} \ge1 ∥Ψ ∥ L p ( 0 , 1 ; L 2 ) ≥ 1 . For instance, Ψ = c η ( t ) ( sin 2 π x 2 , 0 , … , 0 ) \Psi=c\eta(t)(\sin2\pi x_2,0,\ldots,0) Ψ = cη ( t ) ( sin 2 π x 2 , 0 , … , 0 ) with η \eta η supported in ( t 1 , 1 ] (t_1,1] ( t 1 , 1 ] , if t 1 < 1 t_1<1 t 1 < 1 ; if t 1 = 1 t_1=1 t 1 = 1 , replace t 1 t_1 t 1 by 1 / 2 1/2 1/2 throughout. For k ∈ N k\in\mathbb{N} k ∈ N put
u ~ k = χ u ˉ + k Ψ , \widetilde{u}_k=\chi\bar{u}+k\Psi, u k = χ u ˉ + k Ψ ,
which is smooth, divergence-free and equal to u ˉ \bar{u} u ˉ on [ 0 , t 1 / 2 ] [0,t_1/2] [ 0 , t 1 /2 ] . Theorem 2.3 with t 0 = t 1 / 2 t_0=t_1/2 t 0 = t 1 /2 and ε = 1 4 \varepsilon=\frac14 ε = 4 1 gives weak solutions ( u k , f k ) (u_k,f_k) ( u k , f k ) with data ( u i n , f 0 ) (u_{\mathrm{in}},f_0) ( u in , f 0 ) , coinciding with ( u ˉ , f ˉ ) (\bar{u},\bar{f}) ( u ˉ , f ˉ ) on [ 0 , t 1 / 4 ] [0,t_1/4] [ 0 , t 1 /4 ] .
Distinctness. For k ≠ l k\ne l k = l , ( u ~ k − f u ~ k ) − ( u ~ l − f u ~ l ) = ( k − l ) Ψ (\widetilde{u}_k-f\widetilde{u}_k)-(\widetilde{u}_l-f\widetilde{u}_l)=(k-l)\Psi ( u k − f u k ) − ( u l − f u l ) = ( k − l ) Ψ , so by Theorem 2.3(3) and ∥ Ψ ∥ L t p L ∞ ≥ ∥ Ψ ∥ L t p L 2 ≥ 1 \|\Psi\|_{L^p_tL^\infty}\ge\|\Psi\|_{L^p_tL^2}\ge1 ∥Ψ ∥ L t p L ∞ ≥ ∥Ψ ∥ L t p L 2 ≥ 1 ,
∥ ( u k − f u k ) − ( u l − f u l ) ∥ L t p L ∞ ≥ ∣ k − l ∣ − 2 ε ≥ 1 2 . \|(u_k-fu_k)-(u_l-fu_l)\|_{L^p_tL^\infty}\ge|k-l|-2\varepsilon\ge\frac12. ∥ ( u k − f u k ) − ( u l − f u l ) ∥ L t p L ∞ ≥ ∣ k − l ∣ − 2 ε ≥ 2 1 .
Failure of the energy inequality. If ( u k , f k ) (u_k,f_k) ( u k , f k ) were of Leray–Hopf type, then ∥ u k ( t ) ∥ L 2 2 ≤ 2 E ( 0 ) \|u_k(t)\|_{L^2}^2\le2\mathcal{E}(0) ∥ u k ( t ) ∥ L 2 2 ≤ 2 E ( 0 ) for a.e. t t t , since f k ≥ 0 f_k\ge0 f k ≥ 0 , and hence ∥ u k ∥ L t p L 2 ≤ ( 2 E ( 0 ) ) 1 / 2 \|u_k\|_{L^p_tL^2}\le(2\mathcal{E}(0))^{1/2} ∥ u k ∥ L t p L 2 ≤ ( 2 E ( 0 ) ) 1/2 . On the other hand, by orthogonality of the mean and by Theorem 2.3(3),
∥ u k ∥ L t p L 2 ≥ ∥ u k − f u k ∥ L t p L 2 ≥ k ∥ Ψ ∥ L t p L 2 − ∥ χ u ˉ − f χ u ˉ ∥ L t p L 2 − 2 ε → k → ∞ ∞ . \|u_k\|_{L^p_tL^2}\ge\|u_k-fu_k\|_{L^p_tL^2}\ge k\|\Psi\|_{L^p_tL^2}-\|\chi\bar{u}-f\chi\bar{u}\|_{L^p_tL^2}-2\varepsilon\xrightarrow{k\to\infty}\infty. ∥ u k ∥ L t p L 2 ≥ ∥ u k − f u k ∥ L t p L 2 ≥ k ∥Ψ ∥ L t p L 2 − ∥ χ u ˉ − f χ u ˉ ∥ L t p L 2 − 2 ε k → ∞ ∞.
Hence ( u k , f k ) (u_k,f_k) ( u k , f k ) is not of Leray–Hopf type for all large k k k .
Kinetic lemmas
This section collects the kinetic ingredients and proves Proposition 2.10. The central estimate is the dual stability Lemma 3.2: the distance between f [ u ] f[u] f [ u ] and f [ u + w ~ ] f[u+\widetilde{w}] f [ u + w ] is controlled by ∥ w ~ ∥ L t 1 L ∞ \|\widetilde{w}\|_{L^1_tL^\infty} ∥ w ∥ L t 1 L ∞ times a constant that depends only on the Lipschitz norm of the old field u u u . In the convex integration step ∇ ( u + w ~ ) \nabla(u+\widetilde{w}) ∇ ( u + w ) is of size σ μ \sigma\mu σ μ and cannot be allowed to enter. The constants depending on u u u are compensated by choosing δ \delta δ after u u u .
For a function ψ ∈ C 1 ( T d × R d ) \psi\in C^1(\mathbb{T}^d\times\mathbb{R}^d) ψ ∈ C 1 ( T d × R d ) and R > 0 R>0 R > 0 we write
[ ψ ] R = sup T d × B R max { ∣ ∇ x ψ ∣ , ∣ ∇ v ψ ∣ } , ∥ ψ ∥ C b 1 = ∥ ψ ∥ L ∞ + ∥ ∇ x ψ ∥ L ∞ + ∥ ∇ v ψ ∥ L ∞ , [\psi]_R=\sup_{\mathbb{T}^d\times B_R}\max\{|\nabla_x\psi|,|\nabla_v\psi|\},\qquad
\|\psi\|_{C^1_b}=\|\psi\|_{L^\infty}+\|\nabla_x\psi\|_{L^\infty}+\|\nabla_v\psi\|_{L^\infty}, [ ψ ] R = T d × B R sup max { ∣ ∇ x ψ ∣ , ∣ ∇ v ψ ∣ } , ∥ ψ ∥ C b 1 = ∥ ψ ∥ L ∞ + ∥ ∇ x ψ ∥ L ∞ + ∥ ∇ v ψ ∥ L ∞ ,
and for a vector field h h h on T d \mathbb{T}^d T d ,
∥ h ∥ ( C 1 ) ∗ = sup { ∣ ∫ h ⋅ ϕ d x ∣ : ϕ ∈ C 1 ( T d ; R d ) , ∥ ϕ ∥ L ∞ + ∥ ∇ ϕ ∥ L ∞ ≤ 1 } . \|h\|_{(C^1)^*}=\sup\left\{\left|\int h\cdot\phi\,dx\right|:\phi\in C^1(\mathbb{T}^d;\mathbb{R}^d),\ \|\phi\|_{L^\infty}+\|\nabla\phi\|_{L^\infty}\le1\right\}. ∥ h ∥ ( C 1 ) ∗ = sup { ∫ h ⋅ ϕ d x : ϕ ∈ C 1 ( T d ; R d ) , ∥ ϕ ∥ L ∞ + ∥∇ ϕ ∥ L ∞ ≤ 1 } .
Characteristics
Lemma 3.1. Let U ∈ V U\in\mathcal{V} U ∈ V and f = f [ U ] f=f[U] f = f [ U ] .
(a) The flow Z U Z_U Z U is C 1 C^1 C 1 in all variables, det D z Z U ( t ; s , z ) = e − d ( t − s ) \det D_zZ_U(t;s,z)=e^{-d(t-s)} det D z Z U ( t ; s , z ) = e − d ( t − s ) , and
f ( t , Z U ( t ; 0 , z ) ) = e d t f 0 ( z ) , f(t,Z_U(t;0,z))=e^{dt}f_0(z), f ( t , Z U ( t ; 0 , z )) = e d t f 0 ( z ) ,
∬ f ( t ) ψ d z = ∬ f 0 ( z ) ψ ( Z U ( t ; 0 , z ) ) d z . (9) \iint f(t)\psi\,dz=\iint f_0(z)\psi(Z_U(t;0,z))\,dz. \tag*{(9)} ∬ f ( t ) ψ d z = ∬ f 0 ( z ) ψ ( Z U ( t ; 0 , z )) d z . (9)
In particular f ∈ C 1 f\in C^1 f ∈ C 1 solves (2.3) classically, 0 ≤ f ≤ e d M ∞ 0\le f\le e^dM_\infty 0 ≤ f ≤ e d M ∞ , and ∥ f ( t ) ∥ L 1 = M 0 \|f(t)\|_{L^1}=M_0 ∥ f ( t ) ∥ L 1 = M 0 . If U U U is smooth on [ 0 , 1 ] × T d [0,1]\times\mathbb{T}^d [ 0 , 1 ] × T d , so is f f f .
(b) (Causality.) If U 1 = U 2 U_1=U_2 U 1 = U 2 on [ 0 , t 0 ] [0,t_0] [ 0 , t 0 ] , then f [ U 1 ] = f [ U 2 ] f[U_1]=f[U_2] f [ U 1 ] = f [ U 2 ] on [ 0 , t 0 ] [0,t_0] [ 0 , t 0 ] .
(c) (Velocity support.) For t ≥ s t\ge s t ≥ s , ∣ V ( t ; s , z ) ∣ ≤ ∣ v ∣ + ∫ s t ∥ U ( τ ) ∥ L ∞ d τ |V(t;s,z)|\le|v|+\int_s^t\|U(\tau)\|_{L^\infty}\,d\tau ∣ V ( t ; s , z ) ∣ ≤ ∣ v ∣ + ∫ s t ∥ U ( τ ) ∥ L ∞ d τ . Consequently
supp f ( t ) ⊂ T d × B R U V ( t ) , R U V ( t ) : = R 0 V + ∫ 0 t ∥ U ∥ L ∞ d τ , ∥ ρ [ U ] ( t ) ∥ L ∞ ≤ e d M ∞ ∣ B 1 ∣ ( R U V ( t ) ) d . \operatorname{supp}f(t)\subset\mathbb{T}^d\times B_{R_U^V(t)},\qquad
R_U^V(t):=R_0^V+\int_0^t\|U\|_{L^\infty}\,d\tau,\qquad
\|\rho[U](t)\|_{L^\infty}\le e^dM_\infty|B_1|(R_U^V(t))^d. supp f ( t ) ⊂ T d × B R U V ( t ) , R U V ( t ) := R 0 V + ∫ 0 t ∥ U ∥ L ∞ d τ , ∥ ρ [ U ] ( t ) ∥ L ∞ ≤ e d M ∞ ∣ B 1 ∣ ( R U V ( t ) ) d .
(d) (Momentum.) J [ U ] ∈ C 1 ( [ 0 , 1 ] ; R d ) J[U]\in C^1([0,1];\mathbb{R}^d) J [ U ] ∈ C 1 ([ 0 , 1 ] ; R d ) , J [ U ] ( 0 ) = J 0 J[U](0)=J_0 J [ U ] ( 0 ) = J 0 , and
d d t J [ U ] = ∬ ( U − v ) f d z = − ∫ F [ f , U ] . (10) \frac{d}{dt}J[U]=\iint(U-v)f\,dz=-\int F[f,U]. \tag*{(10)} d t d J [ U ] = ∬ ( U − v ) f d z = − ∫ F [ f , U ] . (10)
Proof. (a) The field b U ( t , z ) = ( v , U ( t , x ) − v ) b_U(t,z)=(v,U(t,x)-v) b U ( t , z ) = ( v , U ( t , x ) − v ) is continuous in t t t and C 1 C^1 C 1 in z z z , with globally bounded z z z -derivative, so the flow is global and C 1 C^1 C 1 in ( t , s , z ) (t,s,z) ( t , s , z ) . Liouville’s formula and div z b U = − d \operatorname{div}_z b_U=-d div z b U = − d give the Jacobian, and (3.1) is the push-forward of f 0 f_0 f 0 written in Eulerian and Lagrangian form. The remaining statements follow from (3.1) and the smooth dependence of the flow on parameters.
(b) The flows coincide on [ 0 , t 0 ] [0,t_0] [ 0 , t 0 ] .
(c) Where V ≠ 0 V\ne0 V = 0 , d d τ ∣ V ∣ = V ∣ V ∣ ⋅ ( U ( τ , X ) − V ) ≤ ∥ U ( τ ) ∥ L ∞ \frac{\mathrm{d}}{\mathrm{d}\tau}|V|=\frac{V}{|V|}\cdot(U(\tau,X)-V)\le\|U(\tau)\|_{L^\infty} d τ d ∣ V ∣ = ∣ V ∣ V ⋅ ( U ( τ , X ) − V ) ≤ ∥ U ( τ ) ∥ L ∞ ; at zeros of V V V use ( ∣ V ∣ 2 + η 2 ) 1 / 2 (|V|^2+\eta^2)^{1/2} ( ∣ V ∣ 2 + η 2 ) 1/2 and let η → 0 \eta\to0 η → 0 . The support bound follows from (3.1), and the density bound from f ≤ e d M ∞ f\le e^dM_\infty f ≤ e d M ∞ .
(d) By (c), f ( t ) f(t) f ( t ) has compact support in v v v , uniformly in t t t . Multiplying (2.3) by v v v and integrating by parts, the term v ⋅ ∇ x f v\cdot\nabla_xf v ⋅ ∇ x f integrates to zero on T d \mathbb{T}^d T d , and − ∬ v div v [ ( U − v ) f ] = ∬ ( U − v ) f -\iint v\operatorname{div}_v[(U-v)f]=\iint(U-v)f − ∬ v div v [( U − v ) f ] = ∬ ( U − v ) f . The last identity in (3.2) is the definition of F F F .
Dual stability
Lemma 3.2 (Dual stability). Let u , w ~ ∈ V u,\widetilde w\in\mathcal{V} u , w ∈ V , f = f [ u ] f=f[u] f = f [ u ] , f + = f [ u + w ~ ] f^+=f[u+\widetilde w] f + = f [ u + w ] , ρ + = ∫ f + d v \rho^+=\int f^+\,\mathrm{d}v ρ + = ∫ f + d v , and
Λ : = ∫ 0 1 ∥ ∇ u ( τ ) ∥ L ∞ d τ , R 1 V : = R 0 V + ∫ 0 1 ( 2 ∥ u ∥ L ∞ + ∥ w ~ ∥ L ∞ ) d τ . \Lambda:=\int_0^1\|\nabla u(\tau)\|_{L^\infty}\,\mathrm{d}\tau,\qquad R^V_1:=R^V_0+\int_0^1(2\|u\|_{L^\infty}+\|\widetilde w\|_{L^\infty})\,\mathrm{d}\tau. Λ := ∫ 0 1 ∥∇ u ( τ ) ∥ L ∞ d τ , R 1 V := R 0 V + ∫ 0 1 ( 2∥ u ∥ L ∞ + ∥ w ∥ L ∞ ) d τ .
Then for every t ∈ [ 0 , 1 ] t\in[0,1] t ∈ [ 0 , 1 ] and ψ ∈ C 1 ( T d × R d ) \psi\in C^1(\mathbb{T}^d\times\mathbb{R}^d) ψ ∈ C 1 ( T d × R d ) ,
∣ ∬ ( f + − f ) ( t ) ψ d z ∣ ≤ e 1 + Λ [ ψ ] R 1 V ∫ 0 t ∫ T d ∣ w ~ ∣ ρ + d x d s ≤ e 1 + Λ [ ψ ] R 1 V M 0 ∥ w ~ ∥ L 1 ( 0 , t ; L ∞ ) . (11) \left|\iint(f^+-f)(t)\psi\,\mathrm{d}z\right|\le e^{1+\Lambda[\psi]R^V_1}\int_0^t\int_{\mathbb{T}^d}|\widetilde w|\rho^+\,\mathrm{d}x\,\mathrm{d}s\le e^{1+\Lambda[\psi]R^V_1}M_0\|\widetilde w\|_{L^1(0,t;L^\infty)}. \tag*{(11)} ∬ ( f + − f ) ( t ) ψ d z ≤ e 1 + Λ [ ψ ] R 1 V ∫ 0 t ∫ T d ∣ w ∣ ρ + d x d s ≤ e 1 + Λ [ ψ ] R 1 V M 0 ∥ w ∥ L 1 ( 0 , t ; L ∞ ) . (11)
Proof. Step 1: equation for the difference. Write b = b u b=b_u b = b u . Since b u + w ~ = b + ( 0 , w ~ ) b_{u+\widetilde w}=b+(0,\widetilde w) b u + w = b + ( 0 , w ) , the function g = f + − f ∈ C 1 g=f^+-f\in C^1 g = f + − f ∈ C 1 satisfies
∂ t g + div z ( b g ) = − div v ( w ~ f + ) , g ( 0 ) = 0 , (12) \partial_tg+\operatorname{div}_z(bg)=-\operatorname{div}_v(\widetilde wf^+),\qquad g(0)=0, \tag*{(12)} ∂ t g + div z ( b g ) = − div v ( w f + ) , g ( 0 ) = 0 , (12)
and, by Lemma 3.1(c), f ( s ) f(s) f ( s ) and f + ( s ) f^+(s) f + ( s ) are supported in T d × B R 1 V \mathbb{T}^d\times B_{R^V_1} T d × B R 1 V for all s s s .
Step 2: duality. Fix t t t , ψ \psi ψ and set Φ ( s , z ) = ψ ( Z u ( t ; s , z ) ) \Phi(s,z)=\psi(Z_u(t;s,z)) Φ ( s , z ) = ψ ( Z u ( t ; s , z )) for s ∈ [ 0 , t ] s\in[0,t] s ∈ [ 0 , t ] . Then Φ ∈ C 1 \Phi\in C^1 Φ ∈ C 1 and ∂ s Φ + b ⋅ ∇ z Φ = 0 \partial_s\Phi+b\cdot\nabla_z\Phi=0 ∂ s Φ + b ⋅ ∇ z Φ = 0 , Φ ( t ) = ψ \Phi(t)=\psi Φ ( t ) = ψ . Since g g g and f + f^+ f + have compact support in v v v , integration by parts in z z z gives
d d s ∬ g Φ d z = ∬ ( b g ⋅ ∇ z Φ + f + w ~ ⋅ ∇ v Φ ) d z − ∬ g b ⋅ ∇ z Φ d z = ∬ f + w ~ ⋅ ∇ v Φ d z , \frac{\mathrm{d}}{\mathrm{d}s}\iint g\Phi\,\mathrm{d}z=\iint(bg\cdot\nabla_z\Phi+f^+\widetilde w\cdot\nabla_v\Phi)\,\mathrm{d}z-\iint gb\cdot\nabla_z\Phi\,\mathrm{d}z=\iint f^+\widetilde w\cdot\nabla_v\Phi\,\mathrm{d}z, d s d ∬ g Φ d z = ∬ ( b g ⋅ ∇ z Φ + f + w ⋅ ∇ v Φ ) d z − ∬ g b ⋅ ∇ z Φ d z = ∬ f + w ⋅ ∇ v Φ d z ,
and integrating over [ 0 , t ] [0,t] [ 0 , t ] ,
∬ g ( t ) ψ d z = ∫ 0 t ∬ f + ( s , z ) w ~ ( s , x ) ⋅ ∇ v Φ ( s , z ) d z d s . (13) \iint g(t)\psi\,\mathrm{d}z=\int_0^t\iint f^+(s,z)\widetilde w(s,x)\cdot\nabla_v\Phi(s,z)\,\mathrm{d}z\,\mathrm{d}s. \tag*{(13)} ∬ g ( t ) ψ d z = ∫ 0 t ∬ f + ( s , z ) w ( s , x ) ⋅ ∇ v Φ ( s , z ) d z d s . (13)
Step 3: the linearized flow. For s ≤ τ ≤ t s\le\tau\le t s ≤ τ ≤ t and ξ = ( ξ x , ξ v ) \xi=(\xi_x,\xi_v) ξ = ( ξ x , ξ v ) , let ( δ X , δ V ) ( τ ) = D z Z u ( τ ; s , z ) ξ (\delta X,\delta V)(\tau)=D_zZ_u(\tau;s,z)\xi ( δ X , δ V ) ( τ ) = D z Z u ( τ ; s , z ) ξ . Then
δ X ˙ = δ V and δ V ˙ = ∇ u ( τ , X ) δ X − δ V . \dot{\delta X}=\delta V\quad\text{and}\quad\dot{\delta V}=\nabla u(\tau,X)\delta X-\delta V. δ X ˙ = δ V and δ V ˙ = ∇ u ( τ , X ) δ X − δ V .
With a = ∣ δ X ∣ a=|\delta X| a = ∣ δ X ∣ , b = ∣ δ V ∣ b=|\delta V| b = ∣ δ V ∣ we have, for a.e. τ \tau τ , a ′ ≤ b a'\le b a ′ ≤ b and b ′ ≤ ∥ ∇ u ( τ ) ∥ L ∞ a b'\le\|\nabla u(\tau)\|_{L^\infty}a b ′ ≤ ∥∇ u ( τ ) ∥ L ∞ a , hence ( a + b ) ′ ≤ ( 1 + ∥ ∇ u ( τ ) ∥ L ∞ ) ( a + b ) (a+b)'\le(1+\|\nabla u(\tau)\|_{L^\infty})(a+b) ( a + b ) ′ ≤ ( 1 + ∥∇ u ( τ ) ∥ L ∞ ) ( a + b ) and
∣ δ X ( t ) ∣ + ∣ δ V ( t ) ∣ ≤ e 1 + Λ ( ∣ ξ x ∣ + ∣ ξ v ∣ ) . (14) |\delta X(t)|+|\delta V(t)|\le e^{1+\Lambda}(|\xi_x|+|\xi_v|). \tag*{(14)} ∣ δ X ( t ) ∣ + ∣ δ V ( t ) ∣ ≤ e 1 + Λ ( ∣ ξ x ∣ + ∣ ξ v ∣ ) . (14)
Step 4: conclusion. For η ∈ R d \eta\in\mathbb{R}^d η ∈ R d , ∇ v Φ ( s , z ) ⋅ η = ∇ z ψ ( Z u ( t ; s , z ) ) ⋅ D z Z u ( t ; s , z ) ( 0 , η ) \nabla_v\Phi(s,z)\cdot\eta=\nabla_z\psi(Z_u(t;s,z))\cdot D_zZ_u(t;s,z)(0,\eta) ∇ v Φ ( s , z ) ⋅ η = ∇ z ψ ( Z u ( t ; s , z )) ⋅ D z Z u ( t ; s , z ) ( 0 , η ) . If z ∈ supp f + ( s ) z\in\operatorname{supp}f^+(s) z ∈ supp f + ( s ) , then ∣ v ∣ ≤ R 0 V + ∫ 0 s ∥ u + w ~ ∥ L ∞ |v|\le R^V_0+\int_0^s\|u+\widetilde w\|_{L^\infty} ∣ v ∣ ≤ R 0 V + ∫ 0 s ∥ u + w ∥ L ∞ by Lemma 3.1(c) for the field u + w ~ u+\widetilde w u + w , and the same lemma for the field u u u on [ s , t ] [s,t] [ s , t ] shows Z u ( t ; s , z ) ∈ T d × B R 1 V Z_u(t;s,z)\in\mathbb{T}^d\times B_{R^V_1} Z u ( t ; s , z ) ∈ T d × B R 1 V . Therefore (3.6) gives ∣ ∇ v Φ ( s , z ) ∣ ≤ e 1 + Λ [ ψ ] R 1 V |\nabla_v\Phi(s,z)|\le e^{1+\Lambda[\psi]R^V_1} ∣ ∇ v Φ ( s , z ) ∣ ≤ e 1 + Λ [ ψ ] R 1 V on supp f + ( s ) \operatorname{supp}f^+(s) supp f + ( s ) , and (3.5) yields the first inequality in (3.3). The second follows from ∫ ρ + ( s ) d x = M 0 \int\rho^+(s)\,\mathrm{d}x=M_0 ∫ ρ + ( s ) d x = M 0 .
The coupling error
Lemma 3.3 (Antidivergence in a weak norm). Let 1 < r < d d − 1 1<r<\frac{d}{d-1} 1 < r < d − 1 d , a : = d r ( d − 1 ) ∈ ( 0 , 1 ) a:=\frac{d}{r}(d-1)\in(0,1) a := r d ( d − 1 ) ∈ ( 0 , 1 ) and θ : = a a + d + 1 \theta:=\frac{a}{a+d+1} θ := a + d + 1 a . There is C = C ( d , r ) C=C(d,r) C = C ( d , r ) such that for every smooth vector field h h h on T d \mathbb{T}^d T d ,
∥ R h ∥ L r ≤ C ( ∥ h ∥ L 1 1 − θ ∥ h ∥ ( C 1 ) ∗ θ + ∥ h ∥ ( C 1 ) ∗ ) . (15) \|\mathcal{R}h\|_{L^r}\le C\left(\|h\|_{L^1}^{1-\theta}\|h\|_{(C^1)^*}^{\theta}+\|h\|_{(C^1)^*}\right). \tag*{(15)} ∥ R h ∥ L r ≤ C ( ∥ h ∥ L 1 1 − θ ∥ h ∥ ( C 1 ) ∗ θ + ∥ h ∥ ( C 1 ) ∗ ) . (15)
Proof. Let η \eta η be a standard mollifier, η ε = ε − d η ( ⋅ / ε ) \eta_\varepsilon=\varepsilon^{-d}\eta(\cdot/\varepsilon) η ε = ε − d η ( ⋅ / ε ) , ε ∈ ( 0 , 1 ] \varepsilon\in(0,1] ε ∈ ( 0 , 1 ] , and write R h = ( K − K ∗ η ε ) ∗ h + ( K ∗ η ε ) ∗ h \mathcal{R}h=(K-K*\eta_\varepsilon)*h+(K*\eta_\varepsilon)*h R h = ( K − K ∗ η ε ) ∗ h + ( K ∗ η ε ) ∗ h . Let δ ( z ) = dist ( z , Z d ) \delta(z)=\operatorname{dist}(z,\mathbb{Z}^d) δ ( z ) = dist ( z , Z d ) .
For the first term, by Young’s inequality it suffices to bound ∥ K − K ∗ η ε ∥ L r \|K-K*\eta_\varepsilon\|_{L^r} ∥ K − K ∗ η ε ∥ L r . On { δ ≤ 2 ε } \{\delta\le2\varepsilon\} { δ ≤ 2 ε } we have ∣ K ∗ η ε ∣ ≤ ( ∣ K ∣ 1 { δ ≤ 3 ε } ) ∗ η ε |K*\eta_\varepsilon|\le(|K|\mathbf{1}_{\{\delta\le3\varepsilon\}})*\eta_\varepsilon ∣ K ∗ η ε ∣ ≤ ( ∣ K ∣ 1 { δ ≤ 3 ε } ) ∗ η ε , so by (3)
∥ K − K ∗ η ε ∥ L r ( { δ ≤ 2 ε } ) ≤ 2 ∥ K ∥ L r ( { δ ≤ 3 ε } ) ≤ C ( ∫ 0 3 ε ϱ ( 1 − d ) r + d − 1 d ϱ ) 1 / r = C ε a . \|K-K*\eta_\varepsilon\|_{L^r(\{\delta\le2\varepsilon\})}\le2\|K\|_{L^r(\{\delta\le3\varepsilon\})}\le C\left(\int_0^{3\varepsilon}\varrho^{(1-d)r+d-1}\,\mathrm{d}\varrho\right)^{1/r}=C\varepsilon^a. ∥ K − K ∗ η ε ∥ L r ({ δ ≤ 2 ε }) ≤ 2∥ K ∥ L r ({ δ ≤ 3 ε }) ≤ C ( ∫ 0 3 ε ϱ ( 1 − d ) r + d − 1 d ϱ ) 1/ r = C ε a .
using r < d d − 1 r<\frac{d}{d-1} r < d − 1 d . On { δ ≥ 2 ε } \{\delta\ge2\varepsilon\} { δ ≥ 2 ε } , all points of B ( z , ε ) B(z,\varepsilon) B ( z , ε ) are at distance ≥ δ ( z ) / 2 \ge\delta(z)/2 ≥ δ ( z ) /2 from Z d \mathbb{Z}^d Z d , so ∣ K − K ∗ η ε ∣ ( z ) ≤ ε sup B ( z , ε ) ∣ ∇ K ∣ ≤ C ε δ ( z ) − d |K-K*\eta_\varepsilon|(z)\le\varepsilon\sup_{B(z,\varepsilon)}|\nabla K|\le C\varepsilon\delta(z)^{-d} ∣ K − K ∗ η ε ∣ ( z ) ≤ ε sup B ( z , ε ) ∣∇ K ∣ ≤ C ε δ ( z ) − d and, using r > 1 r>1 r > 1 ,
∥ K − K ∗ η ε ∥ L r ( { δ ≥ 2 ε } ) ≤ C ε ( ∫ 2 ε ∞ ϱ − d r + d − 1 d ϱ ) 1 / r = C ε a . \|K-K*\eta_\varepsilon\|_{L^r(\{\delta\ge2\varepsilon\})}\le C\varepsilon\left(\int_{2\varepsilon}^{\infty}\varrho^{-dr+d-1}\,\mathrm{d}\varrho\right)^{1/r}=C\varepsilon^a. ∥ K − K ∗ η ε ∥ L r ({ δ ≥ 2 ε }) ≤ C ε ( ∫ 2 ε ∞ ϱ − d r + d − 1 d ϱ ) 1/ r = C ε a .
For the second term, each entry of ( K ∗ η ε ) ∗ h (K*\eta_\varepsilon)*h ( K ∗ η ε ) ∗ h at x x x equals ∫ h ⋅ ϕ x d y \int h\cdot\phi_x\,\mathrm{d}y ∫ h ⋅ ϕ x d y with ϕ x ( y ) = ( K i j k ∗ η ε ) ( x − y ) \phi_x(y)=(K_{ijk}*\eta_\varepsilon)(x-y) ϕ x ( y ) = ( K ij k ∗ η ε ) ( x − y ) , and ∥ ϕ x ∥ L ∞ + ∥ ∇ ϕ x ∥ L ∞ ≤ ∥ K ∥ L 1 ( ∥ η ε ∥ L ∞ + ∥ ∇ η ε ∥ L ∞ ) ≤ C ε − d − 1 \|\phi_x\|_{L^\infty}+\|\nabla\phi_x\|_{L^\infty}\le\|K\|_{L^1}(\|\eta_\varepsilon\|_{L^\infty}+\|\nabla\eta_\varepsilon\|_{L^\infty})\le C\varepsilon^{-d-1} ∥ ϕ x ∥ L ∞ + ∥∇ ϕ x ∥ L ∞ ≤ ∥ K ∥ L 1 ( ∥ η ε ∥ L ∞ + ∥∇ η ε ∥ L ∞ ) ≤ C ε − d − 1 . Hence ∥ ( K ∗ η ε ) ∗ h ∥ L r ≤ ∥ ( K ∗ η ε ) ∗ h ∥ L ∞ ≤ C ε − d − 1 ∥ h ∥ ( C 1 ) ∗ \|(K*\eta_\varepsilon)*h\|_{L^r}\le\|(K*\eta_\varepsilon)*h\|_{L^\infty}\le C\varepsilon^{-d-1}\|h\|_{(C^1)^*} ∥ ( K ∗ η ε ) ∗ h ∥ L r ≤ ∥ ( K ∗ η ε ) ∗ h ∥ L ∞ ≤ C ε − d − 1 ∥ h ∥ ( C 1 ) ∗ .
Finally, with A = ∥ h ∥ L 1 A=\|h\|_{L^1} A = ∥ h ∥ L 1 and B = ∥ h ∥ ( C 1 ) ∗ B=\|h\|_{(C^1)^*} B = ∥ h ∥ ( C 1 ) ∗ we have ∥ R h ∥ L r ≤ C ( ε a A + ε − d − 1 B ) \|\mathcal{R}h\|_{L^r}\le C(\varepsilon^aA+\varepsilon^{-d-1}B) ∥ R h ∥ L r ≤ C ( ε a A + ε − d − 1 B ) for all ε ∈ ( 0 , 1 ] \varepsilon\in(0,1] ε ∈ ( 0 , 1 ] . If B < A B<A B < A , take ε = ( B / A ) 1 / ( a + d + 1 ) \varepsilon=(B/A)^{1/(a+d+1)} ε = ( B / A ) 1/ ( a + d + 1 ) ; both terms equal A 1 − θ B θ A^{1-\theta}B^\theta A 1 − θ B θ . If B ≥ A B\ge A B ≥ A , take ε = 1 \varepsilon=1 ε = 1 .
Proposition 3.4 (Coupling error). Let u , w ~ ∈ V u,\widetilde w\in\mathcal{V} u , w ∈ V with W : = ∥ w ~ ∥ L t 1 L ∞ ≤ 1 W:=\|\widetilde w\|_{L^1_tL^\infty}\le1 W := ∥ w ∥ L t 1 L ∞ ≤ 1 , and let 1 < r < d d − 1 1<r<\frac{d}{d-1} 1 < r < d − 1 d , θ \theta θ as in Lemma 3.3. Put u + = u + w ~ u^+=u+\widetilde w u + = u + w and
h : = F [ f [ u + ] , u + ] − F [ f [ u ] , u ] . h:=F[f[u^+],u^+]-F[f[u],u]. h := F [ f [ u + ] , u + ] − F [ f [ u ] , u ] .
If h h h is smooth (e.g. if u , w ~ u,\widetilde w u , w are smooth), then
∥ R h ∥ L 1 ( 0 , 1 ; L r ) ≤ C ⋆ ( W θ + W ) , (16) \|\mathcal{R}h\|_{L^1(0,1;L^r)}\le C_\star(W^\theta+W), \tag*{(16)} ∥ R h ∥ L 1 ( 0 , 1 ; L r ) ≤ C ⋆ ( W θ + W ) , (16)
where C ⋆ C_\star C ⋆ depends only on d , r , M 0 , M ∞ , R 0 v d,r,M_0,M_\infty,R_0^v d , r , M 0 , M ∞ , R 0 v and L : = ∥ u ∥ L ∞ ( 0 , 1 ; W 1 , ∞ ) L:=\|u\|_{L^\infty(0,1;W^{1,\infty})} L := ∥ u ∥ L ∞ ( 0 , 1 ; W 1 , ∞ ) .
Proof. Write f = f [ u ] f=f[u] f = f [ u ] , f + = f [ u + ] f^+=f[u^+] f + = f [ u + ] . Since F [ f + , u + ] = ∫ ( v − u ) f + d v − ρ + w ~ F[f^+,u^+]=\int(v-u)f^+\,\mathrm{d}v-\rho^+\widetilde w F [ f + , u + ] = ∫ ( v − u ) f + d v − ρ + w , we have h = h 1 − ρ + w ~ h=h_1-\rho^+\widetilde w h = h 1 − ρ + w with h 1 = ∫ ( v − u ) ( f + − f ) d v h_1=\int(v-u)(f^+-f)\,\mathrm{d}v h 1 = ∫ ( v − u ) ( f + − f ) d v . As W ≤ 1 W\le1 W ≤ 1 , the radius in Lemma 3.2 satisfies R 1 v ≤ R 0 v + 2 L + 1 R_1^v\le R_0^v+2L+1 R 1 v ≤ R 0 v + 2 L + 1 , and Λ ≤ L \Lambda\le L Λ ≤ L .
To bound h 1 h_1 h 1 , fix t t t and ϕ ∈ C 1 ( T d ; R d ) \phi\in C^1(\mathbb{T}^d;\mathbb{R}^d) ϕ ∈ C 1 ( T d ; R d ) with ∥ ϕ ∥ L ∞ + ∥ ∇ ϕ ∥ L ∞ ≤ 1 \|\phi\|_{L^\infty}+\|\nabla\phi\|_{L^\infty}\le1 ∥ ϕ ∥ L ∞ + ∥∇ ϕ ∥ L ∞ ≤ 1 , and take ψ ( x , v ) = ( v − u ( t , x ) ) ⋅ ϕ ( x ) \psi(x,v)=(v-u(t,x))\cdot\phi(x) ψ ( x , v ) = ( v − u ( t , x )) ⋅ ϕ ( x ) . Then ∇ v ψ = ϕ \nabla_v\psi=\phi ∇ v ψ = ϕ and ∇ x ψ = ( ∇ ϕ ) T ( v − u ) − ( ∇ u ) T ϕ \nabla_x\psi=(\nabla\phi)^T(v-u)-(\nabla u)^T\phi ∇ x ψ = ( ∇ ϕ ) T ( v − u ) − ( ∇ u ) T ϕ , so [ ψ ] R 1 v ≤ 1 + R 1 v + 2 L [\psi]_{R_1^v}\le1+R_1^v+2L [ ψ ] R 1 v ≤ 1 + R 1 v + 2 L , and Lemma 3.2 gives ∥ h 1 ( t ) ∥ ( C 1 ) ∗ ≤ B 0 W \|h_1(t)\|_{(C^1)^*}\le B_0W ∥ h 1 ( t ) ∥ ( C 1 ) ∗ ≤ B 0 W with B 0 = e 1 + L ( 1 + R 1 v + 2 L ) M 0 B_0=e^{1+L}(1+R_1^v+2L)M_0 B 0 = e 1 + L ( 1 + R 1 v + 2 L ) M 0 . By Lemma 3.1(c), ∥ h 1 ( t ) ∥ L 1 ≤ ( R 1 v + L ) ∬ ( f + + f ) d z = 2 ( R 1 v + L ) M 0 = : A 0 \|h_1(t)\|_{L^1}\le(R_1^v+L)\iint(f^++f)\,\mathrm{d}z=2(R_1^v+L)M_0=:A_0 ∥ h 1 ( t ) ∥ L 1 ≤ ( R 1 v + L ) ∬ ( f + + f ) d z = 2 ( R 1 v + L ) M 0 =: A 0 . Lemma 3.3 gives ∥ R h 1 ( t ) ∥ L r ≤ C ( A 0 1 − θ B 0 θ W θ + B 0 W ) \|\mathcal{R}h_1(t)\|_{L^r}\le C(A_0^{1-\theta}B_0^\theta W^\theta+B_0W) ∥ R h 1 ( t ) ∥ L r ≤ C ( A 0 1 − θ B 0 θ W θ + B 0 W ) for every t t t .
For the second term, by (2) and Lemma 3.1(c),
∥ R ( ρ + w ~ ) ( t ) ∥ L r ≤ C ∥ ρ + ( t ) ∥ L ∞ ∥ w ~ ( t ) ∥ L ∞ ≤ C e d M ∞ ∣ B 1 ∣ R 1 v d ∥ w ~ ( t ) ∥ L ∞ . \|\mathcal{R}(\rho^+\widetilde w)(t)\|_{L^r}\le C\|\rho^+(t)\|_{L^\infty}\|\widetilde w(t)\|_{L^\infty}\le Ce^dM_\infty|B_1|R_1^{v\,d}\|\widetilde w(t)\|_{L^\infty}. ∥ R ( ρ + w ) ( t ) ∥ L r ≤ C ∥ ρ + ( t ) ∥ L ∞ ∥ w ( t ) ∥ L ∞ ≤ C e d M ∞ ∣ B 1 ∣ R 1 v d ∥ w ( t ) ∥ L ∞ .
Integrating both bounds in t t t proves (3.8).
The kinetic limit
Lemma 3.5 (Convergence of distribution functions). Let u n ∈ C ∞ ( [ 0 , 1 ] × T d ; R d ) u_n\in C^\infty([0,1]\times\mathbb{T}^d;\mathbb{R}^d) u n ∈ C ∞ ([ 0 , 1 ] × T d ; R d ) , n ≥ 0 n\ge0 n ≥ 0 , and u ∈ L 1 ( 0 , 1 ; L ∞ ( T d ) ) u\in L^1(0,1;L^\infty(\mathbb{T}^d)) u ∈ L 1 ( 0 , 1 ; L ∞ ( T d )) satisfy
K : = sup n ∥ u n ∥ L t 1 L ∞ < ∞ , ∑ n e 1 + Λ n ∥ u n + 1 − u n ∥ L t 1 L ∞ < ∞ , u n → u in L t 1 L ∞ , K:=\sup_n\|u_n\|_{L^1_tL^\infty}<\infty,\qquad\sum_n e^{1+\Lambda_n}\|u_{n+1}-u_n\|_{L^1_tL^\infty}<\infty,\qquad u_n\to u\text{ in }L^1_tL^\infty, K := n sup ∥ u n ∥ L t 1 L ∞ < ∞ , n ∑ e 1 + Λ n ∥ u n + 1 − u n ∥ L t 1 L ∞ < ∞ , u n → u in L t 1 L ∞ ,
where Λ n = ∥ ∇ u n ∥ L t 1 L ∞ \Lambda_n=\|\nabla u_n\|_{L^1_tL^\infty} Λ n = ∥∇ u n ∥ L t 1 L ∞ . Let f n = f [ u n ] f_n=f[u_n] f n = f [ u n ] and R ∞ v = R 0 v + K R_\infty^v=R_0^v+K R ∞ v = R 0 v + K . Then:
(a) 0 ≤ f n ≤ e d M ∞ 0\le f_n\le e^dM_\infty 0 ≤ f n ≤ e d M ∞ and f n = 0 f_n=0 f n = 0 for ∣ v ∣ ≥ R ∞ v |v|\ge R_\infty^v ∣ v ∣ ≥ R ∞ v ;
(b) there is f ∈ L ∞ ( ( 0 , 1 ) × T d × R d ) f\in L^\infty((0,1)\times\mathbb{T}^d\times\mathbb{R}^d) f ∈ L ∞ (( 0 , 1 ) × T d × R d ) , with a representative t ↦ f ( t ) t\mapsto f(t) t ↦ f ( t ) defined for every t t t , such that sup t sup ∥ ψ ∥ C 1 ≤ 1 ∣ ∬ ( f n − f ) ( t ) ψ ∣ → 0 \sup_t\sup_{\|\psi\|_{C^1}\le1}\left|\iint(f_n-f)(t)\psi\right|\to0 sup t sup ∥ ψ ∥ C 1 ≤ 1 ∬ ( f n − f ) ( t ) ψ → 0 ; f n ( t ) ⇀ ∗ f ( t ) f_n(t)\rightharpoonup^*f(t) f n ( t ) ⇀ ∗ f ( t ) in L ∞ L^\infty L ∞ for every t t t ; f n ⇀ ∗ f f_n\rightharpoonup^*f f n ⇀ ∗ f in L ∞ ( ( 0 , 1 ) × T d × R d ) L^\infty((0,1)\times\mathbb{T}^d\times\mathbb{R}^d) L ∞ (( 0 , 1 ) × T d × R d ) ; and f f f satisfies the bounds of (a), with ∬ f ( t ) d z = M 0 \iint f(t)\,\mathrm{d}z=M_0 ∬ f ( t ) d z = M 0 and f ( 0 ) = f 0 f(0)=f_0 f ( 0 ) = f 0 ;
(c) f f f satisfies Definition 2.1(iii) with velocity field u u u ; (d) for every Φ ∈ C ( [ 0 , 1 ] × T d ; R d ) \Phi\in C([0,1] \times\mathbb{T}^{d};\mathbb{R}^{d}) Φ ∈ C ([ 0 , 1 ] × T d ; R d ) , ∬ F [ f n , u n ] ⋅ Φ d x d t → ∬ F [ f , u ] ⋅ Φ d x d t \iint F[f_n,u_n]\cdot\Phi\,dx\,dt \to\iint F[f,u]\cdot\Phi\,dx\,dt ∬ F [ f n , u n ] ⋅ Φ d x d t → ∬ F [ f , u ] ⋅ Φ d x d t , and J [ u n ] ( t ) → ∬ v f ( t ) d z J[u_n](t) \to\iint v f(t)\,dz J [ u n ] ( t ) → ∬ v f ( t ) d z for every t t t .
Proof. (a) is Lemma 3.1(a),(c).
(b) Since [ ψ ] R ≤ ∥ ψ ∥ C b 1 [\psi]_{\mathbb{R}} \leq\|\psi\|_{C_b^1} [ ψ ] R ≤ ∥ ψ ∥ C b 1 , Lemma 3.2 with base field u k u_k u k and perturbation u k + 1 − u k u_{k+1}-u_k u k + 1 − u k gives, for m > n m>n m > n and all t t t ,
sup ∥ ψ ∥ C b 1 ≤ 1 ∣ ∬ ( f m − f n ) ( t ) ψ ∣ ≤ M 0 ∑ k ≥ n e 1 + Λ k ∥ u k + 1 − u k ∥ L t 1 L ∞ ⟶ n → ∞ 0. \sup_{\|\psi\|_{C_b^1}\leq1}\left|\iint(f_m-f_n)(t)\psi\right|
\leq M_0\sum_{k\geq n}e^{1+\Lambda_k}\|u_{k+1}-u_k\|_{L_t^1L^\infty}
\longrightarrow_{n\to\infty}0. ∥ ψ ∥ C b 1 ≤ 1 sup ∬ ( f m − f n ) ( t ) ψ ≤ M 0 k ≥ n ∑ e 1 + Λ k ∥ u k + 1 − u k ∥ L t 1 L ∞ ⟶ n → ∞ 0.
Fix t t t . By (a), ( f n ( t ) ) (f_n(t)) ( f n ( t )) is bounded in L ∞ L^\infty L ∞ and supported in a fixed compact set. Any two weak-* cluster points coincide on the dense subspace C c 1 C_c^1 C c 1 of L 1 L^1 L 1 , by the Cauchy property, so f n ( t ) ⇀ ∗ f ( t ) f_n(t)\rightharpoonup^* f(t) f n ( t ) ⇀ ∗ f ( t ) for a unique f ( t ) f(t) f ( t ) , which inherits the bounds of (a). The convergence is uniform in t t t in the dual C b 1 C_b^1 C b 1 norm. Testing with ψ = χ ( v ) \psi=\chi(v) ψ = χ ( v ) , χ ∈ C c 1 \chi\in C_c^1 χ ∈ C c 1 equal to 1 on B R v ∞ B_{R_v^\infty} B R v ∞ , gives the mass identity; f ( 0 ) = f 0 f(0)=f_0 f ( 0 ) = f 0 because f n ( 0 ) = f 0 f_n(0)=f_0 f n ( 0 ) = f 0 . For G ∈ L 1 ( ( 0 , 1 ) × T d × R d ) G\in L^1((0,1)\times\mathbb{T}^d\times\mathbb{R}^d) G ∈ L 1 (( 0 , 1 ) × T d × R d ) , the functions t ↦ ∫ f n ( t ) G ( t ) t\mapsto\int f_n(t)G(t) t ↦ ∫ f n ( t ) G ( t ) are measurable, bounded by e d M ∞ ∥ G ( t ) ∥ L 1 e^dM_\infty\|G(t)\|_{L^1} e d M ∞ ∥ G ( t ) ∥ L 1 and converge pointwise to ∫ f ( t ) G ( t ) \int f(t)G(t) ∫ f ( t ) G ( t ) ; by dominated convergence, ∭ f n G → ∫ 0 1 ∫ f ( t ) G ( t ) d t \iiint f_nG\to\int_0^1\int f(t)G(t)\,dt ∭ f n G → ∫ 0 1 ∫ f ( t ) G ( t ) d t . Hence every weak-* cluster point of ( f n ) (f_n) ( f n ) in L ∞ ( ( 0 , 1 ) × T d × R d ) L^\infty((0,1)\times\mathbb{T}^d\times\mathbb{R}^d) L ∞ (( 0 , 1 ) × T d × R d ) is represented by t ↦ f ( t ) t\mapsto f(t) t ↦ f ( t ) , and the whole sequence converges.
(c) Each f n f_n f n is a classical, hence weak, solution with field u n u_n u n and datum f 0 f_0 f 0 . For φ ∈ C ∞ ( [ 0 , 1 ) × T d × R d ) \varphi\in C^\infty([0,1)\times\mathbb{T}^d\times\mathbb{R}^d) φ ∈ C ∞ ([ 0 , 1 ) × T d × R d ) , the terms ∭ f n ( ∂ t φ + v ⋅ ∇ x φ − v ⋅ ∇ v φ ) \iiint f_n(\partial_t\varphi+v\cdot\nabla_x\varphi-v\cdot\nabla_v\varphi) ∭ f n ( ∂ t φ + v ⋅ ∇ x φ − v ⋅ ∇ v φ ) and ∭ f n u ⋅ ∇ v φ \iiint f_nu\cdot\nabla_v\varphi ∭ f n u ⋅ ∇ v φ converge by (b), since the multipliers belong to L 1 L^1 L 1 . The remaining term is bounded by
∣ ∭ f n ( u n − u ) ⋅ ∇ v φ ∣ ≤ e d M ∞ ∫ 0 1 ∥ u n − u ∥ L ∞ ∥ ∇ v φ ( t ) ∥ L 1 d t ≤ C φ ∥ u n − u ∥ L t 1 L ∞ → 0. \left|\iiint f_n(u_n-u)\cdot\nabla_v\varphi\right|
\leq e^dM_\infty\int_0^1\|u_n-u\|_{L^\infty}\|\nabla_v\varphi(t)\|_{L^1}\,dt
\leq C_\varphi\|u_n-u\|_{L_t^1L^\infty}\to0. ∭ f n ( u n − u ) ⋅ ∇ v φ ≤ e d M ∞ ∫ 0 1 ∥ u n − u ∥ L ∞ ∥ ∇ v φ ( t ) ∥ L 1 d t ≤ C φ ∥ u n − u ∥ L t 1 L ∞ → 0.
(d) Let χ \chi χ be as in (b). Then ∬ F [ f n , u n ] ⋅ Φ = ∭ f n χ ( v ) v ⋅ Φ − ∭ f n χ ( v ) u n ⋅ Φ \iint F[f_n,u_n]\cdot\Phi=\iiint f_n\chi(v)v\cdot\Phi-\iiint f_n\chi(v)u_n\cdot\Phi ∬ F [ f n , u n ] ⋅ Φ = ∭ f n χ ( v ) v ⋅ Φ − ∭ f n χ ( v ) u n ⋅ Φ . The first term and ∭ f n χ u ⋅ Φ \iiint f_n\chi u\cdot\Phi ∭ f n χ u ⋅ Φ converge by (b), since χ v ⋅ Φ \chi v\cdot\Phi χ v ⋅ Φ and χ u ⋅ Φ \chi u\cdot\Phi χ u ⋅ Φ are integrable; the difference ∭ f n χ ( u n − u ) ⋅ Φ \iiint f_n\chi(u_n-u)\cdot\Phi ∭ f n χ ( u n − u ) ⋅ Φ is bounded by e d M ∞ ∣ B R v ∞ ∣ ∥ Φ ∥ L ∞ ∥ u n − u ∥ L t 1 L ∞ e^dM_\infty|B_{R_v^\infty}|\|\Phi\|_{L^\infty}\|u_n-u\|_{L_t^1L^\infty} e d M ∞ ∣ B R v ∞ ∣∥Φ ∥ L ∞ ∥ u n − u ∥ L t 1 L ∞ . The convergence of J [ u n ] ( t ) J[u_n](t) J [ u n ] ( t ) follows from (b) with ψ = χ ( v ) v i \psi=\chi(v)v_i ψ = χ ( v ) v i .
Lemma 3.6 (Identification of the limit)
Let u ∈ L 1 ( 0 , 1 ; W 1 , 1 ( T d ) ) ∩ L 1 ( 0 , 1 ; L ∞ ( T d ) ) u\in L^1(0,1;W^{1,1}(\mathbb{T}^d))\cap L^1(0,1;L^\infty(\mathbb{T}^d)) u ∈ L 1 ( 0 , 1 ; W 1 , 1 ( T d )) ∩ L 1 ( 0 , 1 ; L ∞ ( T d )) . Then the kinetic equation of Definition 2.1 (iii) with velocity field u u u and datum f 0 f_0 f 0 has at most one weak solution in L ∞ ( ( 0 , 1 ) × T d × R d ) L^\infty((0,1)\times\mathbb{T}^d\times\mathbb{R}^d) L ∞ (( 0 , 1 ) × T d × R d ) . The same holds for any bounded datum vanishing for large ∣ v ∣ |v| ∣ v ∣ , in place of f 0 f_0 f 0 ; the smoothness of f 0 f_0 f 0 is not used, and this general form is the one applied in § 5 and § 8 –§ 11 . This solution, when it exists, is renormalized, i.e.
∂ t β ( f ) + div z ( b β ( f ) ) − d ( f β ′ ( f ) − β ( f ) ) = 0 in D ′ \partial_t\beta(f)+\operatorname{div}_z\bigl(b\beta(f)\bigr)-d\bigl(f\beta'(f)-\beta(f)\bigr)=0\quad\text{in }\mathcal{D}' ∂ t β ( f ) + div z ( b β ( f ) ) − d ( f β ′ ( f ) − β ( f ) ) = 0 in D ′
for every β ∈ C 1 ( R ) \beta\in C^1(\mathbb{R}) β ∈ C 1 ( R ) , where b = ( v , u − v ) b=(v,u-v) b = ( v , u − v ) . If it vanishes for ∣ v ∣ ≥ R v |v|\geq R_v ∣ v ∣ ≥ R v , it belongs to C ( [ 0 , 1 ] ; L s ( T d × R d ) ) C([0,1];L^s(\mathbb{T}^d\times\mathbb{R}^d)) C ([ 0 , 1 ] ; L s ( T d × R d )) for every s < ∞ s<\infty s < ∞ .
Proof. The equation is the continuity equation ∂ t f + div z ( b f ) = 0 \partial_tf+\operatorname{div}_z(bf)=0 ∂ t f + div z ( b f ) = 0 on T d × R d \mathbb{T}^d\times\mathbb{R}^d T d × R d with b ( t , x , v ) = ( v , u ( t , x ) − v ) b(t,x,v)=(v,u(t,x)-v) b ( t , x , v ) = ( v , u ( t , x ) − v ) . We verify the hypotheses of the DiPerna–Lions theory [15 Section II] (see also [4 ] ) for bounded solutions, in the form used in [21 Section 1.3] for the same phase-space flow: ∇ x b = 0 \nabla_xb=0 ∇ x b = 0 , ∇ v b = ( I d , − I d ) \nabla_vb=(\mathrm{Id},-\mathrm{Id}) ∇ v b = ( Id , − Id ) , so b ∈ L 1 ( 0 , 1 ; W 1 , 1 ) b\in L^1(0,1;W^{1,1}) b ∈ L 1 ( 0 , 1 ; W 1 , 1 ) ; div z b = − d \operatorname{div}_zb=-d div z b = − d is bounded; and ∣ b ( t , z ) ∣ / ( 1 + ∣ z ∣ ) ≤ 2 + ∥ u ( t ) ∥ L ∞ ∈ L 1 ( 0 , 1 ) |b(t,z)|/(1+|z|)\leq2+\|u(t)\|_{L^\infty}\in L^1(0,1) ∣ b ( t , z ) ∣/ ( 1 + ∣ z ∣ ) ≤ 2 + ∥ u ( t ) ∥ L ∞ ∈ L 1 ( 0 , 1 ) , since ∣ b ∣ ≤ 2 ∣ v ∣ + ∣ u ∣ |b|\leq2|v|+|u| ∣ b ∣ ≤ 2∣ v ∣ + ∣ u ∣ . The commutator lemma for b ∈ L t 1 W 1 , 1 b\in L_t^1W^{1,1} b ∈ L t 1 W 1 , 1 and f ∈ L ∞ f\in L^\infty f ∈ L ∞ yields renormalization; uniqueness follows from [15 Theorem II.2] and strong continuity in time as in [15 Section II] . The Lagrangian flow is given by [15 Theorem III.2] . The arguments are local in z z z and apply to bounded solutions, so periodicity in x x x plays no role.
Remark 3.7. Lemma 3.6 shows that, in the solutions of Theorem 2.3 , the distribution function is determined by the velocity, and the nonuniqueness comes from the fluid. This is in contrast with nonuniqueness mechanisms driven by the transported quantity, where the transported density fails to be renormalized.
Mean momentum correction
The antidivergence R \mathcal{R} R inverts the divergence only on mean-zero fields, while the drag force has a time-dependent mean. The following lemma restores the exact mean equation by a spatially constant correction c ( t ) c(t) c ( t ) . Since c c c changes the kinetic solution, it is found by a fixed point.
Lemma 3.8 (Mean momentum correction). Let U ∈ C ∞ ( [ 0 , 1 ] × T d ; R d ) U\in C^\infty([0,1]\times\mathbb{T}^d;\mathbb{R}^d) U ∈ C ∞ ([ 0 , 1 ] × T d ; R d ) be divergence-free. (a) There is a unique c ∈ C ( [ 0 , 1 ] ; R d ) c \in C([0,1];\mathbb{R}^{d}) c ∈ C ([ 0 , 1 ] ; R d ) such that
c(t)=\fint U(0)-\fint U(t)+J_{0}-J[U+c](t) \qquad(0\leq t\leq1). \tag*{(17)}
Moreover c ∈ C ∞ ( [ 0 , 1 ] ; R d ) c\in C^{\infty}([0,1];\mathbb{R}^{d}) c ∈ C ∞ ([ 0 , 1 ] ; R d ) , c ( 0 ) = 0 c(0)=0 c ( 0 ) = 0 , and U ♯ : = U + c U^{\sharp}:=U+c U ♯ := U + c satisfies (2.6) and \frac{d}{dt}\fint U^{\sharp}=\fint F[f[U^{\sharp}],U^{\sharp}] .
(b) If U U U satisfies (2.6) on [ 0 , t 0 ] [0,t_{0}] [ 0 , t 0 ] , then c = 0 c=0 c = 0 on [ 0 , t 0 ] [0,t_{0}] [ 0 , t 0 ] .
(c) Suppose U = u + w U=u+w U = u + w , where u u u is smooth, divergence-free and satisfies (2.6) on [ 0 , 1 ] [0,1] [ 0 , 1 ] , and w w w is smooth, divergence-free with \fint w(t)=0 for all t t t . Then
∣ c ( t ) ∣ ≤ A e A t ∥ w ∥ L 1 ( 0 , t ; L ∞ ) , A : = d e 1 + Λ M 0 , Λ = ∥ ∇ u ∥ L t 1 L ∞ . (18) |c(t)|\leq Ae^{At}\|w\|_{L^{1}(0,t;L^{\infty})},\qquad A:=\sqrt{d}e^{1+\Lambda}M_{0},\qquad\Lambda=\|\nabla u\|_{L^{1}_{t}L^{\infty}}. \tag*{(18)} ∣ c ( t ) ∣ ≤ A e A t ∥ w ∥ L 1 ( 0 , t ; L ∞ ) , A := d e 1 + Λ M 0 , Λ = ∥∇ u ∥ L t 1 L ∞ . (18)
Proof. Step 1: the map. For c ∈ C ( [ 0 , 1 ] ; R d ) c\in C([0,1];\mathbb{R}^{d}) c ∈ C ([ 0 , 1 ] ; R d ) we have U + c ∈ V U+c\in\mathcal{V} U + c ∈ V , so J [ U + c ] J[U+c] J [ U + c ] is defined and continuous by Lemma 3.1. Let T ( c ) \mathcal{T}(c) T ( c ) denote the right-hand side of (3.9); then T ( c ) ∈ C ( [ 0 , 1 ] ; R d ) \mathcal{T}(c)\in C([0,1];\mathbb{R}^{d}) T ( c ) ∈ C ([ 0 , 1 ] ; R d ) , T ( c ) ( 0 ) = 0 \mathcal{T}(c)(0)=0 T ( c ) ( 0 ) = 0 , and T \mathcal{T} T is causal: T ( c ) ( t ) \mathcal{T}(c)(t) T ( c ) ( t ) depends only on c ∣ [ 0 , t ] c|_{[0,t]} c ∣ [ 0 , t ] .
Step 2: existence and uniqueness. Let c 1 , c 2 ∈ C ( [ 0 , 1 ] ; R d ) c_{1},c_{2}\in C([0,1];\mathbb{R}^{d}) c 1 , c 2 ∈ C ([ 0 , 1 ] ; R d ) . Lemma 3.2 with base field U + c 2 ∈ V U+c_{2}\in\mathcal{V} U + c 2 ∈ V (whose spatial gradient is ∇ U \nabla U ∇ U ), perturbation c 1 − c 2 ∈ V c_{1}-c_{2}\in\mathcal{V} c 1 − c 2 ∈ V , and test function ψ = v i \psi=v_{i} ψ = v i (for which [ ψ ] R = 1 [\psi]_{\mathbb{R}}=1 [ ψ ] R = 1 ) gives
∣ T ( c 1 ) ( t ) − T ( c 2 ) ( t ) ∣ ≤ β U ∫ 0 t ∣ c 1 − c 2 ∣ d s , β U : = d e 1 + ∥ ∇ U ∥ L t 1 L ∞ M 0 . |\mathcal{T}(c_{1})(t)-\mathcal{T}(c_{2})(t)|\leq\beta_{U}\int_{0}^{t}|c_{1}-c_{2}|\,ds,\qquad\beta_{U}:=\sqrt{d}e^{1+\|\nabla U\|_{L^{1}_{t}L^{\infty}}}M_{0}. ∣ T ( c 1 ) ( t ) − T ( c 2 ) ( t ) ∣ ≤ β U ∫ 0 t ∣ c 1 − c 2 ∣ d s , β U := d e 1 + ∥∇ U ∥ L t 1 L ∞ M 0 .
In the norm ∥ c ∥ ∗ = sup t e − 2 β U t ∣ c ( t ) ∣ \|c\|_{*}=\sup_{t}e^{-2\beta_{U}t}|c(t)| ∥ c ∥ ∗ = sup t e − 2 β U t ∣ c ( t ) ∣ , T \mathcal{T} T is a contraction with constant 1 2 \frac{1}{2} 2 1 , and has a unique fixed point. Applying the same argument on [ 0 , t 0 ] [0,t_{0}] [ 0 , t 0 ] shows that the fixed point of the restricted problem is the restriction of c c c .
Step 3: regularity and the mean equation. If c ∈ C k ( [ 0 , 1 ] ) c\in C^{k}([0,1]) c ∈ C k ([ 0 , 1 ]) , then U + c U+c U + c is C k C^{k} C k in time with values in C ∞ C^{\infty} C ∞ , so the flow, f [ U + c ] f[U+c] f [ U + c ] and F [ f [ U + c ] , U + c ] F[f[U+c],U+c] F [ f [ U + c ] , U + c ] are C k C^{k} C k in time. By (3.2), J [ U + c ] ∈ C k + 1 J[U+c]\in C^{k+1} J [ U + c ] ∈ C k + 1 , hence c = T ( c ) ∈ C k + 1 c=\mathcal{T}(c)\in C^{k+1} c = T ( c ) ∈ C k + 1 . By induction c ∈ C ∞ c\in C^{\infty} c ∈ C ∞ , and c ( 0 ) = 0 c(0)=0 c ( 0 ) = 0 . Since \fint U^{\sharp}=\fint U+c , (3.9) is (2.6) for U ♯ U^{\sharp} U ♯ ; differentiating and using (3.2) gives the mean equation.
Step 4: proof of (b). If U U U satisfies (2.6) on [ 0 , t 0 ] [0,t_{0}] [ 0 , t 0 ] , then T ( 0 ) = 0 \mathcal{T}(0)=0 T ( 0 ) = 0 on [ 0 , t 0 ] [0,t_{0}] [ 0 , t 0 ] , so 0 0 0 is the fixed point of the problem restricted to [ 0 , t 0 ] [0,t_{0}] [ 0 , t 0 ] , and c = 0 c=0 c = 0 there by Step 2.
Step 5: proof of (c). Since \fint w=0 , we have \fint U=\fint u , and (2.6) for u u u turns (3.9) into
c ( t ) = J [ u ] ( t ) − J [ u + w + c ] ( t ) . c(t)=J[u](t)-J[u+w+c](t). c ( t ) = J [ u ] ( t ) − J [ u + w + c ] ( t ) .
Lemma 3.2 with base field u u u (not U + c U+c U + c ), perturbation w + c w+c w + c , and ψ = v i \psi=v_{i} ψ = v i gives
∣ c ( t ) ∣ ≤ A ∫ 0 t ( ∥ w ( s ) ∥ L ∞ + ∣ c ( s ) ∣ ) d s , |c(t)|\leq A\int_{0}^{t}\left(\|w(s)\|_{L^{\infty}}+|c(s)|\right)\,ds, ∣ c ( t ) ∣ ≤ A ∫ 0 t ( ∥ w ( s ) ∥ L ∞ + ∣ c ( s ) ∣ ) d s ,
and Gronwall’s inequality yields (3.10). \hfill □ \square □
Remark 3.9. The correction c c c restores the conservation of total momentum. It is not compactly supported in time: after the support of w w w , the particles carry a slightly different momentum, which the fluid mean has to compensate. This does not affect the construction, since only c = 0 c=0 c = 0 before the support of w w w is needed. For the intervals of regularity, where the concentration step of [12 Section 3] has to be carried out causally in time, the mean is handled by the gluing of Proposition 6.9; see § 6 .
Proof of Proposition 2.10
Let ( u , p u , R ) (u,p_{u},R) ( u , p u , R ) , 0 < s < t ∗ ≤ 1 0<s<t_{*}\leq1 0 < s < t ∗ ≤ 1 and δ ∈ ( 0 , 1 ] \delta\in(0,1] δ ∈ ( 0 , 1 ] be as in Proposition 2.10. Set
L = ∥ u ∥ L ∞ ( 0 , 1 ; W 1 , ∞ ) , A = d e 1 + Λ M 0 , A ‾ = 1 + A e A , L=\|u\|_{L^{\infty}(0,1;W^{1,\infty})},\qquad A=\sqrt{d}e^{1+\Lambda}M_{0},\qquad\overline{A}=1+Ae^{A}, L = ∥ u ∥ L ∞ ( 0 , 1 ; W 1 , ∞ ) , A = d e 1 + Λ M 0 , A = 1 + A e A ,
let C ⋆ C_{\star} C ⋆ be the constant of Proposition 3.4 for this L L L and the exponent r r r of Proposition 2.8, and choose δ ′ ∈ ( 0 , δ ] \delta'\in(0,\delta] δ ′ ∈ ( 0 , δ ] so small that
A ‾ δ ′ ≤ 1 , 2 A ‾ δ ′ ≤ δ , δ ′ + C ⋆ ( ( A ‾ δ ′ ) θ + A ‾ δ ′ ) + 4 A e A ( L + 1 ) δ ′ ≤ δ . (19) \overline{A}\delta'\leq1,\qquad2\overline{A}\delta'\leq\delta,\qquad\delta'+C_{\star}\left((\overline{A}\delta')^{\theta}+\overline{A}\delta'\right)+4Ae^{A}(L+1)\delta'\leq\delta. \tag*{(19)} A δ ′ ≤ 1 , 2 A δ ′ ≤ δ , δ ′ + C ⋆ ( ( A δ ′ ) θ + A δ ′ ) + 4 A e A ( L + 1 ) δ ′ ≤ δ . (19)
The choice depends only on u u u , f 0 f_{0} f 0 , d d d , p p p , q q q and δ \delta δ . Step 1: the Navier–Stokes step. Apply Proposition 2.8 to ( u , R ) (u,R) ( u , R ) with s s s , t ∗ t_\ast t ∗ and δ ′ \delta' δ ′ , obtaining w w w , P P P , R C L R^{\mathrm{CL}} R CL . Since p ≥ 1 p \ge1 p ≥ 1 ,
W : = ∥ w ∥ L t 1 L x ∞ ≤ ∥ w ∥ L t p L x ∞ ≤ δ ′ . W := \|w\|_{L^1_tL^\infty_x} \le\|w\|_{L^p_tL^\infty_x} \le\delta'. W := ∥ w ∥ L t 1 L x ∞ ≤ ∥ w ∥ L t p L x ∞ ≤ δ ′ .
Step 2: the mean correction. Since ( u , p u , R ) (u,p_u,R) ( u , p u , R ) solves the VNS–Reynolds system, u u u satisfies (7) . Lemma 3.8 with U = u + w U=u+w U = u + w gives c ∈ C ∞ ( [ 0 , 1 ] ; R d ) c \in C^\infty([0,1];\mathbb{R}^d) c ∈ C ∞ ([ 0 , 1 ] ; R d ) with ∣ c ( t ) ∣ ≤ A e A W ≤ A e A δ ′ |c(t)| \le Ae^{AW} \le Ae^{A\delta'} ∣ c ( t ) ∣ ≤ A e A W ≤ A e A δ ′ , and c = 0 c=0 c = 0 on [ 0 , s ] [0,s] [ 0 , s ] by (18) , since w = 0 w=0 w = 0 there. Define
u 1 = u + w + c , h = F [ f [ u 1 ] , u 1 ] − F [ f [ u ] , u ] . u_1=u+w+c,\qquad h=F[f[u_1],u_1]-F[f[u],u]. u 1 = u + w + c , h = F [ f [ u 1 ] , u 1 ] − F [ f [ u ] , u ] .
By Lemma 3.8(a), d d t f u 1 = f F [ f [ u 1 ] , u 1 ] \frac{d}{dt}f\,u_1=fF[f[u_1],u_1] d t d f u 1 = f F [ f [ u 1 ] , u 1 ] , while d d t f u = f F [ f [ u ] , u ] \frac{d}{dt}f\,u=fF[f[u],u] d t d f u = f F [ f [ u ] , u ] . Since f u 1 = f u + c f\,u_1=f\,u+c f u 1 = f u + c , this gives
c ′ = ∫ f h . (20) c'=\int f\,h. \tag*{(20)} c ′ = ∫ f h . (20)
Step 3: the new stress. Define
R 1 = R C L + c ⊗ ∘ ( u + w ) + ( u + w ) ⊗ ∘ c − R h , p 1 = p u + P − 2 d c ⋅ ( u + w ) . R_1=R^{\mathrm{CL}}+c\mathbin{\mathop{\otimes}^{\circ}}(u+w)+(u+w)\mathbin{\mathop{\otimes}^{\circ}}c-\mathcal{R}h,\qquad p_1=p_u+P-\frac{2}{d}c\cdot(u+w). R 1 = R CL + c ⊗ ∘ ( u + w ) + ( u + w ) ⊗ ∘ c − R h , p 1 = p u + P − d 2 c ⋅ ( u + w ) .
These are smooth, R 1 R_1 R 1 is S 0 d × d S^{d\times d}_0 S 0 d × d -valued, and div u 1 = 0 \operatorname{div}u_1=0 div u 1 = 0 . Since c c c depends only on t t t and div ( u + w ) = 0 \operatorname{div}(u+w)=0 div ( u + w ) = 0 , we have div ( c ⊗ c ) = 0 \operatorname{div}(c\otimes c)=0 div ( c ⊗ c ) = 0 and
div ( ( u + w ) ⊗ c + c ⊗ ( u + w ) ) = div ( c ⊗ ∘ ( u + w ) + ( u + w ) ⊗ ∘ c ) + 2 d ∇ ( c ⋅ ( u + w ) ) . \operatorname{div}\bigl((u+w)\otimes c+c\otimes(u+w)\bigr)
=\operatorname{div}\bigl(c\mathbin{\mathop{\otimes}^{\circ}}(u+w)+(u+w)\mathbin{\mathop{\otimes}^{\circ}}c\bigr)+\frac{2}{d}\nabla(c\cdot(u+w)). div ( ( u + w ) ⊗ c + c ⊗ ( u + w ) ) = div ( c ⊗ ∘ ( u + w ) + ( u + w ) ⊗ ∘ c ) + d 2 ∇ ( c ⋅ ( u + w )) .
Adding (6) for u u u and Proposition 2.8(i), and using ∂ t u 1 = ∂ t u + ∂ t w + c ′ \partial_tu_1=\partial_tu+\partial_tw+c' ∂ t u 1 = ∂ t u + ∂ t w + c ′ ,
∂ t u 1 − Δ u 1 + div ( u 1 ⊗ u 1 ) + ∇ ( p u + P ) = F [ f [ u ] , u ] + div R C L + div ( c ⊗ ∘ ( u + w ) + ( u + w ) ⊗ ∘ c ) + 2 d ∇ ( c ⋅ ( u + w ) ) + c ′ . \partial_tu_1-\Delta u_1+\operatorname{div}(u_1\otimes u_1)+\nabla(p_u+P)
=F[f[u],u]+\operatorname{div}R^{\mathrm{CL}}+\operatorname{div}\bigl(c\mathbin{\mathop{\otimes}^{\circ}}(u+w)+(u+w)\mathbin{\mathop{\otimes}^{\circ}}c\bigr)
+\frac{2}{d}\nabla(c\cdot(u+w))+c'. ∂ t u 1 − Δ u 1 + div ( u 1 ⊗ u 1 ) + ∇ ( p u + P ) = F [ f [ u ] , u ] + div R CL + div ( c ⊗ ∘ ( u + w ) + ( u + w ) ⊗ ∘ c ) + d 2 ∇ ( c ⋅ ( u + w )) + c ′ .
Moving the gradient term to the left-hand side gives ∇ p 1 \nabla p_1 ∇ p 1 with the minus sign in the definition of p 1 p_1 p 1 . By (20) and (2) , c ′ = ∫ f h = h − div R h c'=\int f\,h=h-\operatorname{div}\mathcal{R}h c ′ = ∫ f h = h − div R h . The right-hand side therefore equals F [ f [ u 1 ] , u 1 ] + div R 1 F[f[u_1],u_1]+\operatorname{div}R_1 F [ f [ u 1 ] , u 1 ] + div R 1 , and ( u 1 , p 1 , R 1 ) (u_1,p_1,R_1) ( u 1 , p 1 , R 1 ) is a smooth solution of the VNS–Reynolds system.
Step 4: the estimates. (i) On [ 0 , s ] [0,s] [ 0 , s ] we have w = 0 w=0 w = 0 and c = 0 c=0 c = 0 , so u 1 = u u_1=u u 1 = u , f [ u 1 ] = f [ u ] f[u_1]=f[u] f [ u 1 ] = f [ u ] by Lemma 3.1(b), and h = 0 h=0 h = 0 . Together with R C L = 0 R^{\mathrm{CL}}=0 R CL = 0 on [ 0 , s ] [0,s] [ 0 , s ] , this gives R 1 = 0 R_1=0 R 1 = 0 there.
(ii)–(iii) Since c c c is constant in space and ∣ T d ∣ = 1 |T^d|=1 ∣ T d ∣ = 1 , each of the norms ∥ c ∥ L t , x 2 \|c\|_{L^2_{t,x}} ∥ c ∥ L t , x 2 , ∥ c ∥ L t p L x ∞ \|c\|_{L^p_tL^\infty_x} ∥ c ∥ L t p L x ∞ , ∥ c ∥ L t 1 W 1 , q \|c\|_{L^1_tW^{1,q}} ∥ c ∥ L t 1 W 1 , q is at most sup t ∣ c ( t ) ∣ ≤ A e A δ ′ \sup_t|c(t)|\le Ae^{A\delta'} sup t ∣ c ( t ) ∣ ≤ A e A δ ′ . With Proposition 2.8(iii),(iv) and (19) this gives (ii) and (iii).
(iv) The field w ~ = w + c \widetilde w=w+c w = w + c is smooth with ∥ w ~ ∥ L t 1 L x ∞ ≤ A δ ‾ ′ ≤ 1 \|\widetilde w\|_{L^1_tL^\infty_x}\le A\overline{\delta}'\le1 ∥ w ∥ L t 1 L x ∞ ≤ A δ ′ ≤ 1 , so Proposition 3.4 gives ∥ R h ∥ L t 1 L r ≤ C ⋆ ( ( A δ ‾ ′ ) θ + A δ ‾ ′ ) \|\mathcal{R}h\|_{L^1_tL^r}\le C_\star((A\overline{\delta}')^\theta+A\overline{\delta}') ∥ R h ∥ L t 1 L r ≤ C ⋆ (( A δ ′ ) θ + A δ ′ ) . Moreover
∥ c ⊗ ∘ ( u + w ) + ( u + w ) ⊗ ∘ c ∥ L t 1 L r ≤ 4 sup t ∣ c ( t ) ∣ ( ∥ u ∥ L t 1 L ∞ + ∥ w ∥ L t 1 L ∞ ) ≤ 4 A e A ( L + 1 ) δ ′ , \left\|c\mathbin{\mathop{\otimes}^{\circ}}(u+w)+(u+w)\mathbin{\mathop{\otimes}^{\circ}}c\right\|_{L^1_tL^r}
\le4\sup_t|c(t)|\left(\|u\|_{L^1_tL^\infty}+\|w\|_{L^1_tL^\infty}\right)
\le4Ae^A(L+1)\delta', ∥ c ⊗ ∘ ( u + w ) + ( u + w ) ⊗ ∘ c ∥ L t 1 L r ≤ 4 t sup ∣ c ( t ) ∣ ( ∥ u ∥ L t 1 L ∞ + ∥ w ∥ L t 1 L ∞ ) ≤ 4 A e A ( L + 1 ) δ ′ ,
and ∥ R C L ∥ L t 1 L r ≤ δ ′ \|R^{\mathrm{CL}}\|_{L^1_tL^r}\le\delta' ∥ R CL ∥ L t 1 L r ≤ δ ′ . By (19) , ∥ R 1 ∥ L t 1 L r ≤ δ \|R_1\|_{L^1_tL^r}\le\delta ∥ R 1 ∥ L t 1 L r ≤ δ .
Remark 3.10 (Dependence of the constants). The constants A A A and C ⋆ C_\star C ⋆ depend on e L e^L e L , i.e. on the Lipschitz norm of the current velocity, which grows along the iteration. They do not multiply a quantity that depends on the new perturbation through anything other than ∥ w ∥ L t 1 L ∞ \|w\|_{L^1_tL^\infty} ∥ w ∥ L t 1 L ∞ , and (19) fixes δ ′ \delta' δ ′ after u u u is known. In the proof of Theorem 2.3 the same mechanism appears in the choice of δ n \delta_n δ n , which contains e − Λ n e^{-\Lambda_n} e − Λ n so that the distribution functions converge (Lemma 3.5).
The Navier–Stokes step: differences from CL22
Proposition 2.8 is the convex integration step of Cheskidov and Luo [12 Sections 4–5] , without their concentration step. Neither the drag force nor the distribution function enters it. In this section we recall the construction and list the changes. For completeness, a full proof in the notation used here is given in Appendices A and B.
The construction. If R ≡ 0 R \equiv0 R ≡ 0 we take w = 0 w = 0 w = 0 . Otherwise the perturbation depends on four parameters: ν \nu ν (temporal oscillation), κ \kappa κ (temporal concentration), σ \sigma σ (spatial oscillation) and μ \mu μ (spatial concentration). All four are powers of a single frequency λ \lambda λ , see (B.3), with ν ≪ μ ≪ σ ≪ κ 1 / 2 \nu\ll\mu\ll\sigma\ll\kappa^{1/2} ν ≪ μ ≪ σ ≪ κ 1/2 .
Spatial building blocks. We use the stationary Mikado flows W k ( σ x ) W_k(\sigma x) W k ( σ x ) , k ∈ Ξ k \in\Xi k ∈ Ξ , of [16 , 26 ] , where Ξ \Xi Ξ is given by the geometric lemma. Each W k W_k W k is concentrated in a tube of radius μ − 1 \mu^{-1} μ − 1 around a periodic line in the direction k k k , is normalized in L 2 L^2 L 2 , and satisfies div ( W k ⊗ W k ) = 0 \operatorname{div}(W_k \otimes W_k) = 0 div ( W k ⊗ W k ) = 0 and W k = div Ω k W_k = \operatorname{div} \Omega_k W k = div Ω k with Ω k \Omega_k Ω k skew-symmetric (Lemma A.3).
Temporal intermittency. Let g κ g_\kappa g κ be the 1-periodic extension of κ 1 / 2 g ( κ t ) \kappa^{1/2}g(\kappa t) κ 1/2 g ( κ t ) . Then ∫ 0 1 g κ 2 = 1 \int_0^1 g_\kappa^2 = 1 ∫ 0 1 g κ 2 = 1 , while g κ g_\kappa g κ is supported on a set of measure κ − 1 \kappa^{-1} κ − 1 per period. This makes ∥ w ∥ L t p L x ∞ \|w\|_{L^p_tL^\infty_x} ∥ w ∥ L t p L x ∞ small for p < 2 p < 2 p < 2 while keeping ∥ w ∥ L t , x 2 ≲ ∥ R ∥ L t , x 1 1 / 2 \|w\|_{L^2_{t,x}} \lesssim\|R\|_{L^1_{t,x}}^{1/2} ∥ w ∥ L t , x 2 ≲ ∥ R ∥ L t , x 1 1/2 .
Amplitudes and perturbation. We set a k = θ ( t ) g κ ( ν t ) ρ 1 / 2 Γ k ( Id − R / ρ ) a_k = \theta(t)g_\kappa(\nu t)\rho^{1/2}\Gamma_k(\operatorname{Id}-R/\rho) a k = θ ( t ) g κ ( ν t ) ρ 1/2 Γ k ( Id − R / ρ ) , where ρ \rho ρ is a regularization of 4 ∣ R ∣ 4|R| 4∣ R ∣ (A.8) and θ \theta θ is a cutoff in time (A.9). The perturbation is w = w ( p ) + w ( c ) + w ( t ) w = w^{(p)} + w^{(c)} + w^{(t)} w = w ( p ) + w ( c ) + w ( t ) :
the principal part w ( p ) = ∑ k a k W k ( σ x ) w^{(p)} = \sum_k a_k W_k(\sigma x) w ( p ) = ∑ k a k W k ( σ x ) ;
a divergence corrector w ( c ) w^{(c)} w ( c ) , chosen so that w ( p ) + w ( c ) w^{(p)} + w^{(c)} w ( p ) + w ( c ) is the divergence of a skew-symmetric field (A.13);
the temporal corrector w ( t ) = ν − 1 h κ ( ν t ) ( div R − ∇ Δ − 1 div div R ) w^{(t)} = \nu^{-1}h_\kappa(\nu t)\left(\operatorname{div}R - \nabla\Delta^{-1}\operatorname{div}\operatorname{div}R\right) w ( t ) = ν − 1 h κ ( ν t ) ( div R − ∇ Δ − 1 div div R ) (A.14), which absorbs the part of g κ 2 ( ν t ) R g_\kappa^2(\nu t)R g κ 2 ( ν t ) R that oscillates in time.
Differences from [12 ] .
The concentration step of [12 Section 3] is omitted. The two-sided cutoff [12 (4.12)] is replaced by a one-sided cutoff θ \theta θ with θ = 0 \theta= 0 θ = 0 on ( − ∞ , s ] (-\infty,s] ( − ∞ , s ] and θ = 1 \theta= 1 θ = 1 on [ t ∗ , ∞ ) [t_*,\infty) [ t ∗ , ∞ ) . Since R = 0 R = 0 R = 0 on [ 0 , t ∗ ] [0,t_*] [ 0 , t ∗ ] , we still have θ 2 R = R \theta^2R = R θ 2 R = R (A.10), which is all that the identity of [12 Lemma 4.4] uses. At t = 1 t = 1 t = 1 no cutoff is needed. In [12 ] the stress vanishes near t = 0 t = 0 t = 0 and t = 1 t = 1 t = 1 , so the functions of time to which [12 Lemma B.1] is applied are periodic. Here R R R need not vanish near t = 1 t = 1 t = 1 , and the use of [12 Lemma B.1] in time is replaced by an elementary decorrelation estimate (Lemma A.4(c)).
As in [12 ] , the perturbation has zero mean: w ( p ) + w ( c ) w^{(p)} + w^{(c)} w ( p ) + w ( c ) is the divergence of a skew-symmetric field, and w ( t ) w^{(t)} w ( t ) is built from div R \operatorname{div}R div R and a gradient. We record it in (ii) because here the background u u u need not have zero mean (item 4 below), and the spatial mean of the velocity is changed only by the constant correction of Lemma 3.8.
Every term of R C L R^{\mathrm{CL}} R CL contains a k a_k a k , w w w or R R R , or one of their time derivatives, and all of these vanish on [ 0 , s ] [0,s] [ 0 , s ] ; this gives (ii).
Nothing in the construction uses ∫ u = 0 \int u = 0 ∫ u = 0 . The identity (i) is algebraic and does not require ( u , R ) (u,R) ( u , R ) to solve any equation.
The L 2 L^2 L 2 bound (iii) follows from [12 Propositions 5.3–5.5] and the proof of [12 Lemma 5.7] , for large λ \lambda λ . The statements of [12 Propositions 2.2(2) and 4.1(2)] are written with ∥ R ∥ L 1 \|R\|_{L^1} ∥ R ∥ L 1 ; the proofs there give the bound with ∥ R ∥ L 1 1 / 2 \|R\|_{L^1}^{1/2} ∥ R ∥ L 1 1/2 , and only the summability of the L 2 L^2 L 2 norms of the perturbations is used in [12 ] .
The exponent r r r of [12 (5.2)] satisfies d − d / r ≤ γ < 1 d - d/r \le\gamma< 1 d − d / r ≤ γ < 1 , hence r < d / ( d − 1 ) r < d/(d-1) r < d / ( d − 1 ) .
Three minor points of [12 Section 4] are adjusted:
the directions are normalized to primitive vectors, Ξ \Xi Ξ is treated as an indexed family in which two indices may carry the same vector, and the base points of the Mikado lines are chosen so that parallel lines are disjoint (A.2);
the far-field interaction is replaced by its traceless part, the trace going into the pressure;
the last term of the pressure appears with the opposite sign in [12 (4.25)] ; this does not affect the estimates (Lemma A.7).
Use in Section 6. Outside Section 3.6, Proposition 2.8 is used as a black box, except in the localized step of Section 6 (Lemma 6.13). That step uses one feature of the proof: the algebra of Appendix A only uses θ 2 R = R \theta^{2}R=R θ 2 R = R , and the estimates of Appendix B only use the smoothness of θ \theta θ and 0 ≤ θ ≤ 1 0\leq\theta\leq1 0 ≤ θ ≤ 1 . Moreover, the time supports of a k a_k a k , w ( p ) w^{(p)} w ( p ) and w ( c ) w^{(c)} w ( c ) lie in supp θ \operatorname{supp}\theta supp θ , and those of w ( t ) w^{(t)} w ( t ) and of the terms of R C L R^{\mathrm{CL}} R CL not containing θ \theta θ lie in supp t R ⊂ { θ = 1 } \operatorname{supp}_{t}R\subset\{\theta=1\} supp t R ⊂ { θ = 1 } , by θ 2 R = R \theta^{2}R=R θ 2 R = R ; hence supp t w , supp t R C L ⊂ supp θ \operatorname{supp}_{t}w,\operatorname{supp}_{t}R^{\mathrm{CL}}\subset\operatorname{supp}\theta supp t w , supp t R CL ⊂ supp θ . So θ \theta θ may be replaced by a two-sided cutoff. The part of the stress that is not cancelled passes unchanged into the new stress by the linearity of the identity (i) in the added term div S s \operatorname{div}S_s div S s . Elsewhere in Section 6, the exponent r r r of Appendix B.1 and the relations (B.1)–(B.2) are used (see the remark after Lemma 6.5). The Laplacian plays no role in the construction (Remark B.9).
Sharpness: uniqueness in the critical class
The solutions of Theorem 2.3 have u ∈ L t p L ∞ u\in L_t^pL^\infty u ∈ L t p L ∞ for every p < 2 p<2 p < 2 . In this section we show that the endpoint p = 2 p=2 p = 2 cannot be reached: weak solutions with u ∈ L t 2 L ∞ u\in L_t^2L^\infty u ∈ L t 2 L ∞ are unique, regular and of Leray–Hopf type. This is the analogue for VNS of the Ladyzhenskaya–Prodi–Serrin endpoint in [12 Theorem 1.3 and Appendix A] . The proof has three steps.
Duality. Any L t , x 2 L_{t,x}^{2} L t , x 2 weak solution of the Oseen problem with an L t 2 L ∞ L_t^2L^\infty L t 2 L ∞ drift coincides with its strong solution (Section 5.3). This is the argument of [12 Appendix A] , after [25 ] .
Regularity. A critical VNS solution solves an Oseen problem with drift u u u and force F [ f , u ] ∈ L t 2 L ∞ F[f,u]\in L_t^2L^\infty F [ f , u ] ∈ L t 2 L ∞ . Maximal L t 2 L q L_t^2L^q L t 2 L q regularity then gives ∇ u ∈ L t 2 L ∞ \nabla u\in L_t^2L^\infty ∇ u ∈ L t 2 L ∞ and the energy equality (Section 5.4).
Lagrangian stability. The difference of two solutions is controlled by the fluid energy of u 1 − u 2 u_1-u_2 u 1 − u 2 together with an L 2 ( f 0 ) L^2(f_0) L 2 ( f 0 ) distance between the two phase-space flows (Section 5.5). This is a Lagrangian version of the dual stability Lemma 3.2. A maximal-function estimate replaces the C 1 C^1 C 1 norm of the test function there, which is not available here. This step follows the method of Han-Kwan, Miot, Moussa and Moyano [20 ] , who combined Loeper’s optimal-transport functional [24 ] with the maximal function to prove uniqueness of energy solutions for d = 2 d=2 d = 2 ; see also [21 Section 2] for the three-dimensional weak–strong setting.
Conventions. Throughout this section T ∈ ( 0 , 1 ] T\in(0,1] T ∈ ( 0 , 1 ] . Definitions 2.1 and 2.2 are used on [ 0 , T ] [0,T] [ 0 , T ] in place of [ 0 , 1 ] [0,1] [ 0 , 1 ] , with test functions vanishing near t = T t=T t = T . Let Q g : = g − ∇ Δ − 1 div g Qg:=g-\nabla\Delta^{-1}\operatorname{div}g Q g := g − ∇ Δ − 1 div g be the Leray projector. Unlike P P P , it keeps the spatial mean. It is self-adjoint on L 2 L^2 L 2 , bounded on L q L^q L q and W k , q W^{k,q} W k , q for 1 < q < ∞ 1<q<\infty 1 < q < ∞ , commutes with Δ \Delta Δ , satisfies Q g = g Qg=g Q g = g if div g = 0 \operatorname{div}g=0 div g = 0 , and g − Q g g-Qg g − Q g is a gradient. We write D T \mathcal{D}_T D T for the set of divergence-free φ ∈ C ∞ ( [ 0 , T ] × T d ; R d ) \varphi\in C^\infty([0,T]\times\mathbb{T}^d;\mathbb{R}^d) φ ∈ C ∞ ([ 0 , T ] × T d ; R d ) vanishing near t = T t=T t = T .
Main result
Theorem 5.1 (Uniqueness and regularity in the critical class). Let d ≥ 2 d\geq2 d ≥ 2 and T ∈ ( 0 , 1 ] T\in(0,1] T ∈ ( 0 , 1 ] . Let u i n ∈ C ∞ ( T d ; R d ) u_{\mathrm{in}}\in C^\infty(\mathbb{T}^d;\mathbb{R}^d) u in ∈ C ∞ ( T d ; R d ) be divergence-free, and let f 0 ∈ L ∞ ( T d × R d ) f_0\in L^\infty(\mathbb{T}^d\times\mathbb{R}^d) f 0 ∈ L ∞ ( T d × R d ) be nonnegative and vanish for ∣ v ∣ ≥ R 0 |v|\geq R_0 ∣ v ∣ ≥ R 0 . Then:
(a) every weak solution ( u , f ) (u,f) ( u , f ) of (2.4) on [ 0 , T ] [0,T] [ 0 , T ] with data ( u i n , f 0 ) (u_{\mathrm{in}},f_0) ( u in , f 0 ) and
u ∈ L 2 ( 0 , T ; L ∞ ( T d ) ) u\in L^2(0,T;L^\infty(\mathbb{T}^d)) u ∈ L 2 ( 0 , T ; L ∞ ( T d )) satisfies ∂ t u ∈ L 2 ( 0 , T ; L q ) \partial_tu\in L^2(0,T;L^q) ∂ t u ∈ L 2 ( 0 , T ; L q ) and u ∈ L 2 ( 0 , T ; W 2 , q ) ∩ C ( [ 0 , T ] ; W 1 , q ) u\in L^2(0,T;W^{2,q})\cap C([0,T];W^{1,q}) u ∈ L 2 ( 0 , T ; W 2 , q ) ∩ C ([ 0 , T ] ; W 1 , q ) for every q < ∞ q<\infty q < ∞ ; in particular ∇ u ∈ L 2 ( 0 , T ; L ∞ ) \nabla u\in L^2(0,T;L^\infty) ∇ u ∈ L 2 ( 0 , T ; L ∞ ) . Moreover f f f is the push-forward of f 0 f_0 f 0 under the classical phase-space flow of u u u , and ( u , f ) (u,f) ( u , f ) satisfies the energy equality
E ( t ) + ∫ 0 t ( ∥ ∇ u ∥ L 2 2 + ∬ ∣ v − u ∣ 2 f d z ) d s = E ( 0 ) ( 0 ≤ t ≤ T ) ; (21) \mathcal{E}(t)+\int_0^t\left(\|\nabla u\|_{L^2}^2+\iint|v-u|^2f\,\mathrm{d}z\right)\,\mathrm{d}s=\mathcal{E}(0)\qquad(0\leq t\leq T); \tag*{(21)} E ( t ) + ∫ 0 t ( ∥∇ u ∥ L 2 2 + ∬ ∣ v − u ∣ 2 f d z ) d s = E ( 0 ) ( 0 ≤ t ≤ T ) ; (21) in particular ( u , f ) (u,f) ( u , f ) is of Leray–Hopf type;
(b) there is at most one such weak solution.
Corollary 5.2 (Sharp dichotomy). Let d ≥ 2 d\geq2 d ≥ 2 , 1 ≤ p < 2 1\leq p<2 1 ≤ p < 2 , 1 ≤ q < ∞ 1\leq q<\infty 1 ≤ q < ∞ . Let ( u i n , f 0 ) (u_{\mathrm{in}},f_0) ( u in , f 0 ) be smooth data, with f 0 ≥ 0 f_0\geq0 f 0 ≥ 0 compactly supported, and let ( u ˉ , f ˉ ) (\bar{u},\bar{f}) ( u ˉ , f ˉ ) be a smooth solution of (2.4) on [ 0 , t 1 ] [0,t_1] [ 0 , t 1 ] with these data. Let T ˉ ∈ ( 0 , t 1 ] \bar{T}\in(0,t_1] T ˉ ∈ ( 0 , t 1 ] . Then:
(i) ( u ˉ , f ˉ ) (\bar{u},\bar{f}) ( u ˉ , f ˉ ) is the only weak solution on [ 0 , T ] [0,T] [ 0 , T ] with u ∈ L 2 ( 0 , T ; L ∞ ) u \in L^{2}(0,T;L^{\infty}) u ∈ L 2 ( 0 , T ; L ∞ ) ;
(ii) there are infinitely many weak solutions on [ 0 , T ] [0,T] [ 0 , T ] with u ∈ L p ( 0 , T ; L ∞ ) ∩ L 1 ( 0 , T ; W 1 , q ) u \in L^{p}(0,T;L^{\infty})\cap L^{1}(0,T;W^{1,q}) u ∈ L p ( 0 , T ; L ∞ ) ∩ L 1 ( 0 , T ; W 1 , q ) . All of them except possibly one satisfy ∥ u ∥ L 2 ( 0 , T ; L ∞ ) = ∞ \lVert u\rVert_{L^{2}(0,T;L^{\infty})}=\infty ∥ u ∥ L 2 ( 0 , T ; L ∞ ) = ∞ , and infinitely many of them are not of Leray–Hopf type.
Proof. (i) follows from Theorem 5.1 (b), since u ˉ ∈ L 2 ( 0 , T ; L ∞ ) \bar{u}\in L^{2}(0,T;L^{\infty}) u ˉ ∈ L 2 ( 0 , T ; L ∞ ) .
(ii) Argue as in the proof of Corollary 2.4, with t 1 t_{1} t 1 replaced by T T T . Choose χ ∈ C ∞ ( R ; [ 0 , 1 ] ) \chi\in C^{\infty}(\mathbb{R};[0,1]) χ ∈ C ∞ ( R ; [ 0 , 1 ]) with χ = 1 \chi=1 χ = 1 on [ 0 , T / 2 ] [0,T/2] [ 0 , T /2 ] and χ = 0 \chi=0 χ = 0 on [ 3 T / 4 , ∞ ) [3T/4,\infty) [ 3 T /4 , ∞ ) . Choose a smooth divergence-free Ψ \Psi Ψ with ∫ T d Ψ = 0 \int_{\mathbb{T}^{d}}\Psi=0 ∫ T d Ψ = 0 , supported in ( T / 2 , T ) × T d (T/2,T)\times\mathbb{T}^{d} ( T /2 , T ) × T d , and with ∥ Ψ ∥ L p ( 0 , T ; L 2 ) ≥ 1 \lVert\Psi\rVert_{L^{p}(0,T;L^{2})}\ge1 ∥ Ψ ∥ L p ( 0 , T ; L 2 ) ≥ 1 . Apply Theorem 2.3 with t 0 = T / 2 t_{0}=T/2 t 0 = T /2 , ε = 1 4 \varepsilon=\frac14 ε = 4 1 and u ~ k = χ u ˉ + k Ψ \widetilde{u}_{k}=\chi\bar{u}+k\Psi u k = χ u ˉ + k Ψ , and restrict the resulting solutions to [ 0 , T ] [0,T] [ 0 , T ] . A weak solution on [ 0 , 1 ] [0,1] [ 0 , 1 ] restricts to a weak solution on [ 0 , T ] [0,T] [ 0 , T ] , because a test function in D T \mathcal{D}_{T} D T extends by zero to an admissible test function on [ 0 , 1 ] [0,1] [ 0 , 1 ] . Since the error terms in Theorem 2.3 (3) are measured on [ 0 , 1 ] ⊃ [ 0 , T ] [0,1]\supset[0,T] [ 0 , 1 ] ⊃ [ 0 , T ] ,
∥ ( u k − u ˉ k ) − ( u l − u ˉ l ) ∥ L p ( 0 , T ; L ∞ ) ≥ ∣ k − l ∣ ∥ Ψ ∥ L p ( 0 , T ; L 2 ) − 2 ε ≥ 1 2 ( k ≠ l ) . \lVert(u_{k}-\bar{u}_{k})-(u_{l}-\bar{u}_{l})\rVert_{L^{p}(0,T;L^{\infty})}\ge|k-l|\lVert\Psi\rVert_{L^{p}(0,T;L^{2})}-2\varepsilon\ge\frac12\qquad(k\ne l). ∥( u k − u ˉ k ) − ( u l − u ˉ l ) ∥ L p ( 0 , T ; L ∞ ) ≥ ∣ k − l ∣ ∥ Ψ ∥ L p ( 0 , T ; L 2 ) − 2 ε ≥ 2 1 ( k = l ) .
So the restrictions are pairwise distinct, and by (i) at most one of them lies in L 2 ( 0 , T ; L ∞ ) L^{2}(0,T;L^{\infty}) L 2 ( 0 , T ; L ∞ ) . The failure of the Leray–Hopf property for all large k k k follows as in Corollary 2.4. Indeed, a Leray–Hopf solution satisfies ∥ u k ∥ L p ( 0 , T ; L 2 ) ≤ ( 2 E ( 0 ) ) 1 / 2 \lVert u_{k}\rVert_{L^{p}(0,T;L^{2})}\le(2\mathcal{E}(0))^{1/2} ∥ u k ∥ L p ( 0 , T ; L 2 ) ≤ ( 2 E ( 0 ) ) 1/2 . On the other hand, ∥ u k ∥ L p ( 0 , T ; L 2 ) ≥ k ∥ Ψ ∥ L p ( 0 , T ; L 2 ) − ∥ χ u ˉ − χ u ˉ ∥ L p ( 0 , T ; L 2 ) − 2 ε → ∞ \lVert u_{k}\rVert_{L^{p}(0,T;L^{2})}\ge k\lVert\Psi\rVert_{L^{p}(0,T;L^{2})}-\lVert\chi\bar{u}-\chi\bar{u}\rVert_{L^{p}(0,T;L^{2})}-2\varepsilon\to\infty ∥ u k ∥ L p ( 0 , T ; L 2 ) ≥ k ∥ Ψ ∥ L p ( 0 , T ; L 2 ) − ∥ χ u ˉ − χ u ˉ ∥ L p ( 0 , T ; L 2 ) − 2 ε → ∞ .
Remark 5.3. 1. In Theorem 5.1 the smoothness of u i n u_{\mathrm{in}} u in is used only through u i n ∈ B q , 2 1 ( T d ) u_{\mathrm{in}}\in B^{1}_{q,2}(\mathbb{T}^{d}) u in ∈ B q , 2 1 ( T d ) for one q > d q>d q > d ; see Corollary 5.11 for the precise statement. No regularity of f 0 f_{0} f 0 is needed.
The class L t 2 L ∞ L^{2}_{t}L^{\infty} L t 2 L ∞ is the endpoint q = ∞ q=\infty q = ∞ of the scaling-critical family X p , q X^{p,q} X p , q , 2 p + d q = 1 \frac{2}{p}+\frac{d}{q}=1 p 2 + q d = 1 . The other critical spaces could be treated with the interpolation inequality [12 Proposition A.4] in Lemma 5.7 . We do not pursue this, since L t 2 L ∞ L^{2}_{t}L^{\infty} L t 2 L ∞ is the endpoint matching Theorem 2.3 .
Since the force F [ f , u ] = j f − ρ f u F[f,u]=jf-\rho_{f}u F [ f , u ] = j f − ρ f u is only bounded in x x x , one cannot expect more than ∇ 2 u ∈ L t q L q \nabla^{2}u\in L^{q}_{t}L^{q} ∇ 2 u ∈ L t q L q , q < ∞ q<\infty q < ∞ , from the fluid equation. Energy estimates alone would give ∇ 2 u ∈ L t 2 L 2 \nabla^{2}u\in L^{2}_{t}L^{2} ∇ 2 u ∈ L t 2 L 2 , which does not make the phase-space flow Lipschitz in d ≥ 2 d\ge2 d ≥ 2 . For this reason we use L q L^{q} L q -maximal regularity, which yields ∇ u ∈ L t 2 L ∞ \nabla u\in L^{2}_{t}L^{\infty} ∇ u ∈ L t 2 L ∞ .
Linear tools
For 2 ≤ q < ∞ 2\le q<\infty 2 ≤ q < ∞ let X q : = B q , 2 1 ( T d ) X_{q}:=B^{1}_{q,2}(\mathbb{T}^{d}) X q := B q , 2 1 ( T d ) . For an interval I = [ a , b ] ⊂ [ 0 , 1 ] I=[a,b]\subset[0,1] I = [ a , b ] ⊂ [ 0 , 1 ] let
M R q ( I ) : = { v ∈ L 2 ( I ; W 2 , q ( T d ) ) : ∂ t v ∈ L 2 ( I ; L q ( T d ) ) } MR_{q}(I):=\{v\in L^{2}(I;W^{2,q}(\mathbb{T}^{d})):\partial_{t}v\in L^{2}(I;L^{q}(\mathbb{T}^{d}))\} M R q ( I ) := { v ∈ L 2 ( I ; W 2 , q ( T d )) : ∂ t v ∈ L 2 ( I ; L q ( T d ))} , ∥ v ∥ M R q ( I ) : = ∥ v ∥ L 2 ( I ; W 2 , q ) + ∥ ∂ t v ∥ L 2 ( I ; L q ) \lVert v\rVert_{MR_{q}(I)}:=\lVert v\rVert_{L^{2}(I;W^{2,q})}+\lVert\partial_{t}v\rVert_{L^{2}(I;L^{q})} ∥ v ∥ M R q ( I ) := ∥ v ∥ L 2 ( I ; W 2 , q ) + ∥ ∂ t v ∥ L 2 ( I ; L q ) .
Lemma 5.4 (Maximal regularity). Let 2 ≤ q < ∞ 2\le q<\infty 2 ≤ q < ∞ . There is C q ≥ 1 C_{q}\ge1 C q ≥ 1 , independent of the interval I = [ a , b ] ⊂ [ 0 , 1 ] I=[a,b]\subset[0,1] I = [ a , b ] ⊂ [ 0 , 1 ] , such that:
(a) for v 0 ∈ X q v_{0}\in X_{q} v 0 ∈ X q and g ∈ L 2 ( I ; L q ) g\in L^{2}(I;L^{q}) g ∈ L 2 ( I ; L q ) there is a unique v ∈ M R q ( I ) v\in MR_{q}(I) v ∈ M R q ( I ) with ∂ t v − Δ v = g \partial_{t}v-\Delta v=g ∂ t v − Δ v = g a.e. and v ( a ) = v 0 v(a)=v_{0} v ( a ) = v 0 , and ∥ v ∥ M R q ( I ) ≤ C q ( ∥ v 0 ∥ X q + ∥ g ∥ L 2 ( I ; L q ) ) \lVert v\rVert_{MR_{q}(I)}\le C_{q}(\lVert v_{0}\rVert_{X_{q}}+\lVert g\rVert_{L^{2}(I;L^{q})}) ∥ v ∥ M R q ( I ) ≤ C q (∥ v 0 ∥ X q + ∥ g ∥ L 2 ( I ; L q ) ) ;
(b) every v ∈ M R q ( I ) v\in MR_{q}(I) v ∈ M R q ( I ) belongs to C ( I ; X q ) C(I;X_{q}) C ( I ; X q ) , with
sup I ∥ v ∥ X q ≤ C q ( ∥ v ( a ) ∥ X q + ∥ v ∥ M R q ( I ) ) , \sup_{I}\lVert v\rVert_{X_{q}}\le C_{q}(\lVert v(a)\rVert_{X_{q}}+\lVert v\rVert_{MR_{q}(I)}), I sup ∥ v ∥ X q ≤ C q (∥ v ( a ) ∥ X q + ∥ v ∥ M R q ( I ) ) ,
and sup I ∥ v ∥ X q ≤ C q ∥ v ∥ M R q ( I ) \sup_{I}\lVert v\rVert_{X_{q}}\le C_{q}\lVert v\rVert_{MR_{q}(I)} sup I ∥ v ∥ X q ≤ C q ∥ v ∥ M R q ( I ) if v ( a ) = 0 v(a)=0 v ( a ) = 0 ;
(c) C ∞ ( T d ) ⊂ X q ↪ W 1 , q C^{\infty}(\mathbb{T}^{d})\subset X_{q}\hookrightarrow W^{1,q} C ∞ ( T d ) ⊂ X q ↪ W 1 , q , M R q ( I ) ↪ M R 2 ( I ) MR_{q}(I)\hookrightarrow MR_{2}(I) M R q ( I ) ↪ M R 2 ( I ) , and M R 2 ( I ) ↪ C ( I ; H 1 ) MR_{2}(I)\hookrightarrow C(I;H^{1}) M R 2 ( I ) ↪ C ( I ; H 1 ) ;
(d) if v 0 v_{0} v 0 and g g g are divergence-free, so is v v v .
References. On L q ( T d ) L^{q}(\mathbb{T}^{d}) L q ( T d ) , 1 < q < ∞ 1<q<\infty 1 < q < ∞ , the Laplacian has maximal L 2 L^{2} L 2 -regularity on bounded intervals. The trace space of M R q MR_{q} M R q is the real interpolation space ( L q , W 2 , q ) 1 / 2 , 2 = B q , 2 1 (L^{q},W^{2,q})_{1/2,2}=B^{1}_{q,2} ( L q , W 2 , q ) 1/2 , 2 = B q , 2 1 ; see [3 Chapter III] and [27 Chapter 3] . In particular this yields (a) and (b) on [ 0 , 1 ] [0,1] [ 0 , 1 ] . The constants are uniform in I I I by the usual extensions: in (a), translate to a = 0 a=0 a = 0 and extend g g g by zero to [ 0 , 1 ] [0,1] [ 0 , 1 ] ; in (b) with v ( a ) = 0 v(a)=0 v ( a ) = 0 , extend v v v by zero to [ a − 1 , a ] [a-1,a] [ a − 1 , a ] and use the same embedding on intervals of length in [ 1 , 2 ] [1,2] [ 1 , 2 ] . The general case of (b) follows by subtracting e ( t − a ) Δ v ( a ) e^{(t-a)\Delta}v(a) e ( t − a ) Δ v ( a ) . For (c): B q , 2 1 ↪ F q , 2 1 = W 1 , q B^{1}_{q,2}\hookrightarrow F^{1}_{q,2}=W^{1,q} B q , 2 1 ↪ F q , 2 1 = W 1 , q for q ≥ 2 q\ge2 q ≥ 2 [28 ] ; the other embeddings hold since ∣ T d ∣ = 1 |\mathbb{T}^{d}|=1 ∣ T d ∣ = 1 , and the last one is the Lions–Magenes lemma. For (d): Q \mathbb{Q} Q commutes with the heat flow, so Q v \mathbb{Q}v Q v solves the same problem as v v v .
Lemma 5.5 (Maximal function). Let g ∈ H 1 ( T d ) g \in H^1(\mathbb{T}^d) g ∈ H 1 ( T d ) , extended periodically to R d \mathbb{R}^d R d , and let M \mathcal{M} M be the Hardy–Littlewood maximal operator on R d \mathbb{R}^d R d . Then M ∣ ∇ g ∣ \mathcal{M}|\nabla g| M ∣∇ g ∣ is periodic,
∥ M ∣ ∇ g ∣ ∥ L 2 ( T d ) ≤ C M ∥ ∇ g ∥ L 2 ( T d ) , (22) \|\mathcal{M}|\nabla g|\|_{L^2(\mathbb{T}^d)} \le C_M\|\nabla g\|_{L^2(\mathbb{T}^d)}, \tag*{(22)} ∥ M ∣∇ g ∣ ∥ L 2 ( T d ) ≤ C M ∥∇ g ∥ L 2 ( T d ) , (22) and there is a null set N ⊂ R d N \subset\mathbb{R}^d N ⊂ R d such that
∣ g ( x ) − g ( y ) ∣ ≤ C d ∣ x − y ∣ ( M ∣ ∇ g ∣ ( x ) + M ∣ ∇ g ∣ ( y ) ) for all x , y ∈ R d ∖ N . (23) |g(x)-g(y)| \le C_d|x-y|(\mathcal{M}|\nabla g|(x)+\mathcal{M}|\nabla g|(y)) \qquad\text{for all }x,y\in\mathbb{R}^d\setminus N. \tag*{(23)} ∣ g ( x ) − g ( y ) ∣ ≤ C d ∣ x − y ∣ ( M ∣∇ g ∣ ( x ) + M ∣∇ g ∣ ( y )) for all x , y ∈ R d ∖ N . (23) The constants C M C_M C M and C d C_d C d depend only on d d d .
Proof. (5.3) is the classical estimate for W l o c 1 , 1 W^{1,1}_{\mathrm{loc}} W loc 1 , 1 functions, with N N N the complement of the Lebesgue points of g g g ; see [9 Lemma A.3] . It is stated there with the maximal function truncated at a radius R ≥ ∣ x − y ∣ R \ge|x-y| R ≥ ∣ x − y ∣ , which is bounded by M \mathcal{M} M . For (5.2), let h = ∣ ∇ g ∣ h=|\nabla g| h = ∣∇ g ∣ and x ∈ [ 0 , 1 ] d x\in[0,1]^d x ∈ [ 0 , 1 ] d . Balls of radius r < 1 r<1 r < 1 around x x x lie in B : = [ − 1 , 2 ] d B:=[-1,2]^d B := [ − 1 , 2 ] d , so the supremum over r < 1 r<1 r < 1 is at most M ( h 1 B ) ( x ) \mathcal{M}(h1_B)(x) M ( h 1 B ) ( x ) . For r ≥ 1 r\ge1 r ≥ 1 , the ball B ( x , r ) B(x,r) B ( x , r ) meets at most C ( 1 + r ) d C(1+r)^d C ( 1 + r ) d unit cells, so \fint_{B(x,r)}h\le C\|h\|_{L^1(\mathbb{T}^d)} . Hence ∥ M h ∥ L 2 ( [ 0 , 1 ] d ) ≤ ∥ M ( h 1 B ) ∥ L 2 ( R d ) + C ∥ h ∥ L 1 ( T d ) ≤ C ∥ h ∥ L 2 ( T d ) \|\mathcal{M}h\|_{L^2([0,1]^d)}\le\|\mathcal{M}(h1_B)\|_{L^2(\mathbb{R}^d)}+C\|h\|_{L^1(\mathbb{T}^d)}\le C\|h\|_{L^2(\mathbb{T}^d)} ∥ M h ∥ L 2 ([ 0 , 1 ] d ) ≤ ∥ M ( h 1 B ) ∥ L 2 ( R d ) + C ∥ h ∥ L 1 ( T d ) ≤ C ∥ h ∥ L 2 ( T d ) .
The Oseen problem: strong solutions and duality
Let b ∈ L 2 ( 0 , T ; L ∞ ( T d ; R d ) ) b\in L^2(0,T;L^\infty(\mathbb{T}^d;\mathbb{R}^d)) b ∈ L 2 ( 0 , T ; L ∞ ( T d ; R d )) be weakly divergence-free for a.e. t t t . For G ∈ L 1 ( ( 0 , T ) × T d ) G\in L^1((0,T)\times\mathbb{T}^d) G ∈ L 1 (( 0 , T ) × T d ) and v 0 ∈ L 2 v_0\in L^2 v 0 ∈ L 2 , we call v ∈ L 2 ( ( 0 , T ) × T d ) v\in L^2((0,T)\times\mathbb{T}^d) v ∈ L 2 (( 0 , T ) × T d ) , weakly divergence-free for a.e. t t t , a weak solution of the Oseen problem
∂ t v − Δ v + Q ( b ⋅ ∇ v ) = Q G , div v = 0 , v ( 0 ) = v 0 , (24) \partial_t v-\Delta v+\mathbb{Q}(b\cdot\nabla v)=\mathbb{Q}G,\qquad\operatorname{div}v=0,\qquad v(0)=v_0, \tag*{(24)} ∂ t v − Δ v + Q ( b ⋅ ∇ v ) = Q G , div v = 0 , v ( 0 ) = v 0 , (24)
if for every φ ∈ D T \varphi\in\mathcal{D}_T φ ∈ D T
∫ v 0 ⋅ φ ( 0 ) d x + ∫ 0 T ∫ ( v ⋅ ( ∂ t φ + Δ φ + b ⋅ ∇ φ ) + G ⋅ φ ) d x d t = 0. (25) \int v_0\cdot\varphi(0)\,dx+\int_0^T\int\left(v\cdot(\partial_t\varphi+\Delta\varphi+b\cdot\nabla\varphi)+G\cdot\varphi\right)\,dx\,dt=0. \tag*{(25)} ∫ v 0 ⋅ φ ( 0 ) d x + ∫ 0 T ∫ ( v ⋅ ( ∂ t φ + Δ φ + b ⋅ ∇ φ ) + G ⋅ φ ) d x d t = 0. (25)
Here ( b ⋅ ∇ φ ) i = b j ∂ j φ i (b\cdot\nabla\varphi)_i=b_j\partial_j\varphi_i ( b ⋅ ∇ φ ) i = b j ∂ j φ i . This formulation is meaningful because b ⋅ ∇ φ ∈ L t , x 2 b\cdot\nabla\varphi\in L^2_{t,x} b ⋅ ∇ φ ∈ L t , x 2 .
Proposition 5.6 (Strong solutions). Let 2 ≤ q < ∞ 2\le q<\infty 2 ≤ q < ∞ , G ∈ L 2 ( 0 , T ; L q ) G\in L^2(0,T;L^q) G ∈ L 2 ( 0 , T ; L q ) , and let v 0 ∈ X q v_0\in X_q v 0 ∈ X q be divergence-free. There is a unique divergence-free v ∈ M R q ( [ 0 , T ] ) v\in\mathrm{MR}_q([0,T]) v ∈ MR q ([ 0 , T ]) solving (5.4) a.e. in ( 0 , T ) × T d (0,T)\times\mathbb{T}^d ( 0 , T ) × T d . It is a weak solution in the sense of (5.5). If the hypotheses hold for several q q q , the solutions coincide.
Proof. For v ∈ M R q ( I ) v\in\mathrm{MR}_q(I) v ∈ MR q ( I ) , Lemma 5.4(b),(c) gives ∇ v ∈ L ∞ ( I ; L q ) \nabla v\in L^\infty(I;L^q) ∇ v ∈ L ∞ ( I ; L q ) , hence
∥ Q ( b ⋅ ∇ v ) ∥ L 2 ( I ; L q ) ≤ C ∥ b ∥ L 2 ( I ; L ∞ ) sup t ∈ I ∥ ∇ v ( t ) ∥ L q < ∞ . (26) \|\mathbb{Q}(b\cdot\nabla v)\|_{L^2(I;L^q)}\le C\|b\|_{L^2(I;L^\infty)}\sup_{t\in I}\|\nabla v(t)\|_{L^q}<\infty. \tag*{(26)} ∥ Q ( b ⋅ ∇ v ) ∥ L 2 ( I ; L q ) ≤ C ∥ b ∥ L 2 ( I ; L ∞ ) t ∈ I sup ∥∇ v ( t ) ∥ L q < ∞. (26)
Since ∥ b ( ⋅ ) ∥ L ∞ 2 ∈ L 1 ( 0 , T ) \|b(\cdot)\|_{L^\infty}^2\in L^1(0,T) ∥ b ( ⋅ ) ∥ L ∞ 2 ∈ L 1 ( 0 , T ) , there is a partition 0 = τ 0 < τ 1 < ⋯ < τ n = T 0=\tau_0<\tau_1<\cdots<\tau_n=T 0 = τ 0 < τ 1 < ⋯ < τ n = T with C C q 2 ∥ b ∥ L 2 ( I j ; L ∞ ) ≤ 1 2 CC_q^2\|b\|_{L^2(I_j;L^\infty)}\le\frac12 C C q 2 ∥ b ∥ L 2 ( I j ; L ∞ ) ≤ 2 1 on I j = [ τ j − 1 , τ j ] I_j=[\tau_{j-1},\tau_j] I j = [ τ j − 1 , τ j ] , where C C C is the constant in (5.6).
We construct v v v interval by interval. Suppose v ( τ j − 1 ) = : v j − 1 ∈ X q v(\tau_{j-1})=:v_{j-1}\in X_q v ( τ j − 1 ) =: v j − 1 ∈ X q is divergence-free. On X j : = { v ∈ M R q ( I j ) : v ( τ j − 1 ) = v j − 1 } X_j:=\{v\in\mathrm{MR}_q(I_j):v(\tau_{j-1})=v_{j-1}\} X j := { v ∈ MR q ( I j ) : v ( τ j − 1 ) = v j − 1 } , let Θ ( v ) \Theta(v) Θ ( v ) be the solution of Lemma 5.4(a) with g = Q G − Q ( b ⋅ ∇ v ) g=\mathbb{Q}G-\mathbb{Q}(b\cdot\nabla v) g = Q G − Q ( b ⋅ ∇ v ) . For v , v ′ ∈ X j v,v'\in X_j v , v ′ ∈ X j the difference Θ ( v ) − Θ ( v ′ ) \Theta(v)-\Theta(v') Θ ( v ) − Θ ( v ′ ) has zero initial value, so by Lemma 5.4(a),(b) and (5.6),
∥ Θ ( v ) − Θ ( v ′ ) ∥ M R q ( I j ) ≤ C q C ∥ b ∥ L 2 ( I j ; L ∞ ) C q ∥ v − v ′ ∥ M R q ( I j ) ≤ 1 2 ∥ v − v ′ ∥ M R q ( I j ) . \|\Theta(v)-\Theta(v')\|_{\mathrm{MR}_q(I_j)}
\le C_qC\|b\|_{L^2(I_j;L^\infty)}C_q\|v-v'\|_{\mathrm{MR}_q(I_j)}
\le\frac12\|v-v'\|_{\mathrm{MR}_q(I_j)}. ∥Θ ( v ) − Θ ( v ′ ) ∥ MR q ( I j ) ≤ C q C ∥ b ∥ L 2 ( I j ; L ∞ ) C q ∥ v − v ′ ∥ MR q ( I j ) ≤ 2 1 ∥ v − v ′ ∥ MR q ( I j ) .
The unique fixed point solves (5.4) on I j I_j I j . It is divergence-free by Lemma 5.4(d), and v ( τ j ) ∈ X q v(\tau_j)\in X_q v ( τ j ) ∈ X q by Lemma 5.4(b). The concatenation is continuous at the τ j \tau_j τ j , hence lies in M R q ( [ 0 , T ] ) \mathrm{MR}_q([0,T]) MR q ([ 0 , T ]) . Uniqueness follows from the contraction on each I j I_j I j .
The solution is weak: test (5.4) with φ ∈ D T \varphi\in\mathcal{D}_T φ ∈ D T and integrate by parts in t t t and x x x . Since div φ = 0 \operatorname{div}\varphi=0 div φ = 0 , ∫ Q g ⋅ φ = ∫ g ⋅ φ \int\mathbb{Q}g\cdot\varphi=\int g\cdot\varphi ∫ Q g ⋅ φ = ∫ g ⋅ φ . Moreover, ∫ ( b ⋅ ∇ v ) ⋅ φ = − ∫ v ⋅ ( b ⋅ ∇ φ ) \int(b\cdot\nabla v)\cdot\varphi=-\int v\cdot(b\cdot\nabla\varphi) ∫ ( b ⋅ ∇ v ) ⋅ φ = − ∫ v ⋅ ( b ⋅ ∇ φ ) because ∫ b ⋅ ∇ ( v ⋅ φ ) = 0 \int b\cdot\nabla(v\cdot\varphi)=0 ∫ b ⋅ ∇ ( v ⋅ φ ) = 0 for the weakly divergence-free b ∈ L ∞ b\in L^\infty b ∈ L ∞ and v ⋅ φ ∈ W 1 , 1 v\cdot\varphi\in W^{1,1} v ⋅ φ ∈ W 1 , 1 .
For q ≥ q ′ ≥ 2 q\ge q'\ge2 q ≥ q ′ ≥ 2 we have M R q ⊂ M R q ′ \mathrm{MR}_q\subset\mathrm{MR}_{q'} MR q ⊂ MR q ′ and X q ⊂ X q ′ X_q\subset X_{q'} X q ⊂ X q ′ . Both solutions solve the problem in M R q ′ \mathrm{MR}_{q'} MR q ′ , where the solution is unique.
Lemma 5.7 (Adjoint problem). Let H ∈ C ∞ ( [ 0 , T ] × T d ; R d ) H\in C^\infty([0,T]\times\mathbb{T}^d;\mathbb{R}^d) H ∈ C ∞ ([ 0 , T ] × T d ; R d ) . There is a divergence-free Φ \Phi Φ with
Φ ∈ L 2 ( 0 , T ; H 2 ) ∩ C ( [ 0 , T ] ; H 1 ) , ∂ t Φ ∈ L t , x 2 , b ⋅ ∇ Φ ∈ L t , x 2 , Φ ( T ) = 0 , \Phi\in L^2(0,T;H^2)\cap C([0,T];H^1),\qquad\partial_t\Phi\in L^2_{t,x},\qquad b\cdot\nabla\Phi\in L^2_{t,x},\qquad\Phi(T)=0, Φ ∈ L 2 ( 0 , T ; H 2 ) ∩ C ([ 0 , T ] ; H 1 ) , ∂ t Φ ∈ L t , x 2 , b ⋅ ∇Φ ∈ L t , x 2 , Φ ( T ) = 0 , solving − ∂ t Φ − Δ Φ − Q ( b ⋅ ∇ Φ ) = Q H -\partial_t\Phi-\Delta\Phi-\mathbb{Q}(b\cdot\nabla\Phi)=\mathbb{Q}H − ∂ t Φ − ΔΦ − Q ( b ⋅ ∇Φ ) = Q H a.e. in ( 0 , T ) × T d (0,T)\times\mathbb{T}^d ( 0 , T ) × T d .
Proof. Galerkin scheme. Let V n V_n V n be the space of real divergence-free trigonometric polynomials of degree at most n n n , and Π n \Pi_n Π n the L 2 L^2 L 2 -orthogonal projection onto V n V_n V n . Note that Π n = Q \Pi_n=Q Π n = Q followed by truncation of Fourier modes, so it commutes with Δ \Delta Δ . Let Φ n ∈ W 1 , 1 ( 0 , T ; V n ) \Phi_n\in W^{1,1}(0,T;V_n) Φ n ∈ W 1 , 1 ( 0 , T ; V n ) solve the linear backward ODE
− ∂ t Φ n − Δ Φ n − Π n ( b ⋅ ∇ Φ n ) = Π n H , Φ n ( T ) = 0. -\partial_t\Phi_n-\Delta\Phi_n-\Pi_n(b\cdot\nabla\Phi_n)=\Pi_nH,\qquad\Phi_n(T)=0. − ∂ t Φ n − Δ Φ n − Π n ( b ⋅ ∇ Φ n ) = Π n H , Φ n ( T ) = 0.
Its coefficients are integrable in time, since b ∈ L 1 ( 0 , T ; L ∞ ) b\in L^1(0,T;L^\infty) b ∈ L 1 ( 0 , T ; L ∞ ) , so the solution exists and is unique.
Energy estimates. Since div b = 0 \operatorname{div}b=0 div b = 0 , ∫ ( b ⋅ ∇ Φ n ) ⋅ Φ n = 0 \int(b\cdot\nabla\Phi_n)\cdot\Phi_n=0 ∫ ( b ⋅ ∇ Φ n ) ⋅ Φ n = 0 , so
− 1 2 d d t ∥ Φ n ∥ 2 2 + ∥ ∇ Φ n ∥ 2 2 = ∫ H ⋅ Φ n . -\frac{1}{2}\frac{\mathrm{d}}{\mathrm{d}t}\|\Phi_n\|_2^2+\|\nabla\Phi_n\|_2^2=\int H\cdot\Phi_n. − 2 1 d t d ∥ Φ n ∥ 2 2 + ∥∇ Φ n ∥ 2 2 = ∫ H ⋅ Φ n .
Testing with − Δ Φ n ∈ V n -\Delta\Phi_n\in V_n − Δ Φ n ∈ V n ,
− 1 2 d d t ∥ ∇ Φ n ∥ 2 2 + ∥ Δ Φ n ∥ 2 2 = − ∫ H ⋅ Δ Φ n − ∫ ( b ⋅ ∇ Φ n ) ⋅ Δ Φ n ≤ 1 2 ∥ Δ Φ n ∥ 2 2 + ∥ H ∥ 2 2 + ∥ b ∥ L ∞ 2 ∥ ∇ Φ n ∥ 2 2 . -\frac{1}{2}\frac{\mathrm{d}}{\mathrm{d}t}\|\nabla\Phi_n\|_2^2+\|\Delta\Phi_n\|_2^2=-\int H\cdot\Delta\Phi_n-\int(b\cdot\nabla\Phi_n)\cdot\Delta\Phi_n\leq\frac{1}{2}\|\Delta\Phi_n\|_2^2+\|H\|_2^2+\|b\|_{L^\infty}^2\|\nabla\Phi_n\|_2^2. − 2 1 d t d ∥∇ Φ n ∥ 2 2 + ∥Δ Φ n ∥ 2 2 = − ∫ H ⋅ Δ Φ n − ∫ ( b ⋅ ∇ Φ n ) ⋅ Δ Φ n ≤ 2 1 ∥Δ Φ n ∥ 2 2 + ∥ H ∥ 2 2 + ∥ b ∥ L ∞ 2 ∥∇ Φ n ∥ 2 2 .
Gronwall’s inequality backward from t = T t=T t = T gives, uniformly in n n n ,
sup t ( ∥ Φ n ( t ) ∥ 2 2 + ∥ ∇ Φ n ( t ) ∥ 2 2 ) + ∫ 0 T ∥ Δ Φ n ∥ 2 2 d t ≤ C ∥ H ∥ L t , x 2 2 exp ( 2 ∥ b ∥ L 2 ( 0 , T ; L ∞ ) 2 ) . \sup_t\left(\|\Phi_n(t)\|_2^2+\|\nabla\Phi_n(t)\|_2^2\right)+\int_0^T\|\Delta\Phi_n\|_2^2\,\mathrm{d}t\leq C\|H\|_{L^2_{t,x}}^2\exp\left(2\|b\|_{L^2(0,T;L^\infty)}^2\right). t sup ( ∥ Φ n ( t ) ∥ 2 2 + ∥∇ Φ n ( t ) ∥ 2 2 ) + ∫ 0 T ∥Δ Φ n ∥ 2 2 d t ≤ C ∥ H ∥ L t , x 2 2 exp ( 2∥ b ∥ L 2 ( 0 , T ; L ∞ ) 2 ) .
Hence ∥ b ⋅ ∇ Φ n ∥ L t , x 2 ≤ ∥ b ∥ L t 2 L ∞ ∥ ∇ Φ n ∥ L t ∞ L 2 \|b\cdot\nabla\Phi_n\|_{L^2_{t,x}}\leq\|b\|_{L^2_tL^\infty}\|\nabla\Phi_n\|_{L^\infty_tL^2} ∥ b ⋅ ∇ Φ n ∥ L t , x 2 ≤ ∥ b ∥ L t 2 L ∞ ∥∇ Φ n ∥ L t ∞ L 2 and, from the equation, ∥ ∂ t Φ n ∥ L t , x 2 \|\partial_t\Phi_n\|_{L^2_{t,x}} ∥ ∂ t Φ n ∥ L t , x 2 are bounded uniformly in n n n .
Limit. Along a subsequence, Φ n ⇀ ∗ Φ \Phi_n\rightharpoonup^*\Phi Φ n ⇀ ∗ Φ in L ∞ ( 0 , T ; H 1 ) L^\infty(0,T;H^1) L ∞ ( 0 , T ; H 1 ) and weakly in L 2 ( 0 , T ; H 2 ) L^2(0,T;H^2) L 2 ( 0 , T ; H 2 ) , and ∂ t Φ n ⇀ ∂ t Φ \partial_t\Phi_n\rightharpoonup\partial_t\Phi ∂ t Φ n ⇀ ∂ t Φ in L t , x 2 L^2_{t,x} L t , x 2 . For g ∈ L t , x 2 g\in L^2_{t,x} g ∈ L t , x 2 we have ∬ ( b ⋅ ∇ Φ n ) ⋅ g = ∬ ∇ Φ n : ( g ⊗ b ) \iint(b\cdot\nabla\Phi_n)\cdot g=\iint\nabla\Phi_n:(g\otimes b) ∬ ( b ⋅ ∇ Φ n ) ⋅ g = ∬ ∇ Φ n : ( g ⊗ b ) with g ⊗ b ∈ L 1 ( 0 , T ; L 2 ) g\otimes b\in L^1(0,T;L^2) g ⊗ b ∈ L 1 ( 0 , T ; L 2 ) , so b ⋅ ∇ Φ n ⇀ b ⋅ ∇ Φ b\cdot\nabla\Phi_n\rightharpoonup b\cdot\nabla\Phi b ⋅ ∇ Φ n ⇀ b ⋅ ∇Φ in L t , x 2 L^2_{t,x} L t , x 2 . Moreover ∬ Π n ( b ⋅ ∇ Φ n ) ⋅ g = ∬ ( b ⋅ ∇ Φ n ) ⋅ Π n g \iint\Pi_n(b\cdot\nabla\Phi_n)\cdot g=\iint(b\cdot\nabla\Phi_n)\cdot\Pi_ng ∬ Π n ( b ⋅ ∇ Φ n ) ⋅ g = ∬ ( b ⋅ ∇ Φ n ) ⋅ Π n g , and Π n g → Q g \Pi_ng\to Qg Π n g → Q g strongly in L t , x 2 L^2_{t,x} L t , x 2 . Hence the equation passes to the limit. Φ ( T ) = 0 \Phi(T)=0 Φ ( T ) = 0 because Φ n ( T ) ⇀ Φ ( T ) \Phi_n(T)\rightharpoonup\Phi(T) Φ n ( T ) ⇀ Φ ( T ) , Φ \Phi Φ is divergence-free, and Φ ∈ C ( [ 0 , T ] ; H 1 ) \Phi\in C([0,T];H^1) Φ ∈ C ([ 0 , T ] ; H 1 ) by Lemma 5.4(c).
Lemma 5.8 (Admissibility). Let Φ \Phi Φ be as in Lemma 5.7. There are φ k ∈ D T \varphi_k\in D_T φ k ∈ D T with ∂ t φ k + Δ φ k + b ⋅ ∇ φ k → ∂ t Φ + Δ Φ + b ⋅ ∇ Φ \partial_t\varphi_k+\Delta\varphi_k+b\cdot\nabla\varphi_k\to\partial_t\Phi+\Delta\Phi+b\cdot\nabla\Phi ∂ t φ k + Δ φ k + b ⋅ ∇ φ k → ∂ t Φ + ΔΦ + b ⋅ ∇Φ in L 2 ( ( 0 , T ) × T d ) L^2((0,T)\times\mathbb{T}^d) L 2 (( 0 , T ) × T d ) .
Proof. Extend Φ \Phi Φ by 0 0 0 to ( T , 2 T ] (T,2T] ( T , 2 T ] . Since Φ ( T ) = 0 \Phi(T)=0 Φ ( T ) = 0 and Φ ∈ C ( [ 0 , T ] ; H 1 ) \Phi\in C([0,T];H^1) Φ ∈ C ([ 0 , T ] ; H 1 ) , the extension Φ ~ \widetilde{\Phi} Φ lies in H 1 ( 0 , 2 T ; L 2 ) ∩ L 2 ( 0 , 2 T ; H 2 ) H^1(0,2T;L^2)\cap L^2(0,2T;H^2) H 1 ( 0 , 2 T ; L 2 ) ∩ L 2 ( 0 , 2 T ; H 2 ) . For small h > 0 h>0 h > 0 let φ h : = η h ∗ t , x Φ ~ ( ⋅ + 2 h ) \varphi_h:=\eta_{h}*_{t,x}\widetilde{\Phi}(\,\cdot+2h) φ h := η h ∗ t , x Φ ( ⋅ + 2 h ) , restricted to [ 0 , T ] [0,T] [ 0 , T ] , where η h \eta_h η h is a space–time mollifier supported in { ∣ t ∣ , ∣ x ∣ ≤ h } \{|t|,|x|\leq h\} { ∣ t ∣ , ∣ x ∣ ≤ h } . Then φ h ( t ) \varphi_h(t) φ h ( t ) only involves Φ ~ \widetilde{\Phi} Φ on [ t + h , t + 3 h ] ⊂ ( 0 , 2 T ] [t+h,t+3h]\subset(0,2T] [ t + h , t + 3 h ] ⊂ ( 0 , 2 T ] ; it is smooth and divergence-free, and it vanishes for t ≥ T − h t\geq T-h t ≥ T − h . By continuity of translations and mollification,
φ h → Φ in H 1 ( 0 , T ; L 2 ) ∩ L 2 ( 0 , T ; H 2 ) , hence in C ( [ 0 , T ] ; H 1 ) . \varphi_h\to\Phi\quad\text{in }H^1(0,T;L^2)\cap L^2(0,T;H^2),
\qquad\text{hence in }C([0,T];H^1). φ h → Φ in H 1 ( 0 , T ; L 2 ) ∩ L 2 ( 0 , T ; H 2 ) , hence in C ([ 0 , T ] ; H 1 ) .
Therefore ∂ t φ h + Δ φ h → ∂ t Φ + Δ Φ \partial_t\varphi_h+\Delta\varphi_h\to\partial_t\Phi+\Delta\Phi ∂ t φ h + Δ φ h → ∂ t Φ + ΔΦ in L t , x 2 L^2_{t,x} L t , x 2 , and ∥ b ⋅ ∇ ( φ h − Φ ) ∥ L t , x 2 ≤ ∥ b ∥ L t 2 L ∞ ∥ ∇ ( φ h − Φ ) ∥ L t ∞ L 2 → 0 \|b\cdot\nabla(\varphi_h-\Phi)\|_{L^2_{t,x}}\leq\|b\|_{L^2_tL^\infty}\|\nabla(\varphi_h-\Phi)\|_{L^\infty_tL^2}\to0 ∥ b ⋅ ∇ ( φ h − Φ ) ∥ L t , x 2 ≤ ∥ b ∥ L t 2 L ∞ ∥∇ ( φ h − Φ ) ∥ L t ∞ L 2 → 0 .
Proposition 5.9 (Uniqueness by duality). Let w ∈ L 2 ( ( 0 , T ) × T d ) w\in L^2((0,T)\times\mathbb{T}^d) w ∈ L 2 (( 0 , T ) × T d ) be weakly divergence-free for a.e. t t t and satisfy
∫ 0 T ∫ w ⋅ ( ∂ t φ + Δ φ + b ⋅ ∇ φ ) d x d t = 0 for all φ ∈ D T . \int_0^T\int w\cdot(\partial_t\varphi+\Delta\varphi+b\cdot\nabla\varphi)\,\mathrm{d}x\,\mathrm{d}t=0\qquad\text{for all }\varphi\in D_T. ∫ 0 T ∫ w ⋅ ( ∂ t φ + Δ φ + b ⋅ ∇ φ ) d x d t = 0 for all φ ∈ D T .
Then w = 0 w=0 w = 0 .
Proof. Let H H H be smooth and let Φ \Phi Φ be given by Lemma 5.7. By Lemma 5.8, the identity extends to φ = Φ \varphi=\Phi φ = Φ . By the equation of Φ \Phi Φ ,
∂ t Φ + Δ Φ + b ⋅ ∇ Φ = − H + ( Id − Q ) H + ( Id − Q ) ( b ⋅ ∇ Φ ) , \partial_t\Phi+\Delta\Phi+b\cdot\nabla\Phi=-H+(\operatorname{Id}-Q)H+(\operatorname{Id}-Q)(b\cdot\nabla\Phi), ∂ t Φ + ΔΦ + b ⋅ ∇Φ = − H + ( Id − Q ) H + ( Id − Q ) ( b ⋅ ∇Φ ) ,
and the last two terms are gradients of H 1 H^1 H 1 functions for a.e. t t t , which are orthogonal to w ( t ) w(t) w ( t ) . Hence ∫ w ⋅ H = 0 \int w\cdot H=0 ∫ w ⋅ H = 0 for all smooth H H H , so w = 0 w=0 w = 0 .
Regularity of critical solutions
Proposition 5.10. Under the hypotheses of Theorem 5.1 , let ( u , f ) (u,f) ( u , f ) be a weak solution on [ 0 , T ] [0,T] [ 0 , T ] with u ∈ L 2 ( 0 , T ; L ∞ ) u\in L^2(0,T;L^\infty) u ∈ L 2 ( 0 , T ; L ∞ ) . Let R ∗ v : = R 0 v + ∥ u ∥ L 1 ( 0 , T ; L ∞ ) R^v_*:=R^v_0+\|u\|_{L^1(0,T;L^\infty)} R ∗ v := R 0 v + ∥ u ∥ L 1 ( 0 , T ; L ∞ ) . Then:
(a) F : = F [ f , u ] ∈ L 2 ( 0 , T ; L ∞ ) F:=F[f,u]\in L^2(0,T;L^\infty) F := F [ f , u ] ∈ L 2 ( 0 , T ; L ∞ ) ;
(b) u ∈ M R q ( [ 0 , T ] ) u\in MR_q([0,T]) u ∈ M R q ([ 0 , T ]) for every q ∈ [ 2 , ∞ ) q\in[2,\infty) q ∈ [ 2 , ∞ ) , and ∂ t u − Δ u + Q ( u ⋅ ∇ u ) = Q F \partial_tu-\Delta u+Q(u\cdot\nabla u)=QF ∂ t u − Δ u + Q ( u ⋅ ∇ u ) = QF a.e.; in particular ∇ u ∈ L 2 ( 0 , T ; L ∞ ) \nabla u\in L^2(0,T;L^\infty) ∇ u ∈ L 2 ( 0 , T ; L ∞ ) ; (c) the field b u ( t , x , v ) = ( v , u ( t , x ) − v ) b_u(t,x,v)=(v,u(t,x)-v) b u ( t , x , v ) = ( v , u ( t , x ) − v ) has a unique flow Z u ( t ; s , z ) Z_u(t;s,z) Z u ( t ; s , z ) in the Carathéodory sense. Each Z u ( t ; s , ⋅ ) Z_u(t;s,\cdot) Z u ( t ; s , ⋅ ) is bi-Lipschitz with det D z Z u ( t ; s , z ) = e − d ( t − s ) \det D_zZ_u(t;s,z)=e^{-d(t-s)} det D z Z u ( t ; s , z ) = e − d ( t − s ) . Moreover f ( t ) = Z u ( t ; 0 , ⋅ ) # f 0 f(t)=Z_u(t;0,\cdot)_{\#}f_0 f ( t ) = Z u ( t ; 0 , ⋅ ) # f 0 for every t t t (after modification on a null set of times), 0 ≤ f ≤ e d ∥ f 0 ∥ L ∞ 0\le f\le e^d\|f_0\|_{L^\infty} 0 ≤ f ≤ e d ∥ f 0 ∥ L ∞ , and f ( t ) = 0 f(t)=0 f ( t ) = 0 for ∣ v ∣ ≥ R v ∗ |v|\ge R_v^* ∣ v ∣ ≥ R v ∗ ;
(d) the energy equality (21) holds.
Proof. (a) Let R v R_v R v be such that f = 0 f=0 f = 0 for ∣ v ∣ ≥ R v |v|\ge R_v ∣ v ∣ ≥ R v (Definition 2.1(ii)). Then ∥ ρ f ∥ L ∞ + ∥ j f ∥ L ∞ ≤ ( 1 + R v ) ∣ B R v ∣ ∥ f ∥ L ∞ \|\rho_f\|_{L^\infty}+\|j_f\|_{L^\infty}\le(1+R_v)|B_{R_v}|\|f\|_{L^\infty} ∥ ρ f ∥ L ∞ + ∥ j f ∥ L ∞ ≤ ( 1 + R v ) ∣ B R v ∣∥ f ∥ L ∞ , and F = j f − ρ f u F=j_f-\rho_fu F = j f − ρ f u .
(b) Fix q ∈ ( d , ∞ ) q\in(d,\infty) q ∈ ( d , ∞ ) . Let v ∈ M R q ( [ 0 , T ] ) v\in\mathrm{MR}_q([0,T]) v ∈ MR q ([ 0 , T ]) be the strong solution of Proposition 5.6 with b : = u b:=u b := u , G : = F G:=F G := F and v 0 : = u i n v_0:=u_{\mathrm{in}} v 0 := u in ; it is a weak solution in the sense of (5.5). Since u ⊗ u : ∇ φ = u ⋅ ( u ⋅ ∇ φ ) u\otimes u:\nabla\varphi=u\cdot(u\cdot\nabla\varphi) u ⊗ u : ∇ φ = u ⋅ ( u ⋅ ∇ φ ) , Definition 2.1(iv) says that u u u is also a weak solution of the same Oseen problem. The difference w = u − v ∈ L t , x 2 w=u-v\in L^2_{t,x} w = u − v ∈ L t , x 2 is weakly divergence-free and satisfies the homogeneous identity of Proposition 5.9, so u = v u=v u = v . Hence u ∈ M R q u\in\mathrm{MR}_q u ∈ MR q for all q ∈ [ 2 , ∞ ) q\in[2,\infty) q ∈ [ 2 , ∞ ) by Proposition 5.6, and ∇ u ∈ L 2 ( 0 , T ; W 1 , q ) ⊂ L 2 ( 0 , T ; L ∞ ) \nabla u\in L^2(0,T;W^{1,q})\subset L^2(0,T;L^\infty) ∇ u ∈ L 2 ( 0 , T ; W 1 , q ) ⊂ L 2 ( 0 , T ; L ∞ ) for q > d q>d q > d .
(c) By (b), b u b_u b u is measurable in t t t , Lipschitz in z z z with constant 1 + ∥ ∇ u ( t ) ∥ L ∞ ∈ L 1 ( 0 , T ) 1+\|\nabla u(t)\|_{L^\infty}\in L^1(0,T) 1 + ∥∇ u ( t ) ∥ L ∞ ∈ L 1 ( 0 , T ) , and grows at most linearly. Carathéodory theory gives a unique flow, bi-Lipschitz in z z z . The Jacobian identity follows from Liouville’s formula and div z b u = − d \operatorname{div}_z b_u=-d div z b u = − d . The proof of Lemma 3.1(a),(c) applies verbatim, so the push-forward f ~ ( t ) = Z u ( t ; 0 , ⋅ ) # f 0 \widetilde f(t)=Z_u(t;0,\cdot)_{\#}f_0 f ( t ) = Z u ( t ; 0 , ⋅ ) # f 0 is a bounded weak solution with the stated bounds. Since u ∈ L 1 ( 0 , T ; W 1 , 1 ) ∩ L 1 ( 0 , T ; L ∞ ) u\in L^1(0,T;W^{1,1})\cap L^1(0,T;L^\infty) u ∈ L 1 ( 0 , T ; W 1 , 1 ) ∩ L 1 ( 0 , T ; L ∞ ) , Lemma 3.6 gives f = f ~ f=\widetilde f f = f .
(d) By (b) and Lemma 5.4(c), u ∈ C ( [ 0 , T ] ; H 1 ) u\in C([0,T];H^1) u ∈ C ([ 0 , T ] ; H 1 ) and ∂ t u ∈ L t , x 2 \partial_tu\in L^2_{t,x} ∂ t u ∈ L t , x 2 , so t ↦ 1 2 ∥ u ( t ) ∥ 2 t\mapsto\frac12\|u(t)\|^2 t ↦ 2 1 ∥ u ( t ) ∥ 2 is absolutely continuous with derivative ∫ ∂ t u ⋅ u \int\partial_tu\cdot u ∫ ∂ t u ⋅ u . Since Q Q Q is self-adjoint and Q u = u Qu=u Q u = u , and since ∫ ( u ⋅ ∇ u ) ⋅ u = ∫ u ⋅ ∇ ∣ u ∣ 2 2 = 0 \int(u\cdot\nabla u)\cdot u=\int u\cdot\nabla\frac{|u|^2}{2}=0 ∫ ( u ⋅ ∇ u ) ⋅ u = ∫ u ⋅ ∇ 2 ∣ u ∣ 2 = 0 (as ∣ u ∣ 2 ∈ W 1 , 1 |u|^2\in W^{1,1} ∣ u ∣ 2 ∈ W 1 , 1 for a.e. t t t ),
d d t 1 2 ∥ u ∥ 2 + ∥ ∇ u ∥ 2 = ∫ F ⋅ u = ∫ j f ⋅ u − ∬ ρ f ∣ u ∣ 2 . \frac{d}{dt}\frac12\|u\|^2+\|\nabla u\|^2=\int F\cdot u=\int j_f\cdot u-\iint\rho_f|u|^2. d t d 2 1 ∥ u ∥ 2 + ∥∇ u ∥ 2 = ∫ F ⋅ u = ∫ j f ⋅ u − ∬ ρ f ∣ u ∣ 2 .
For the particles, by (c), 1 2 ∬ ∣ v ∣ 2 f ( t ) = 1 2 ∫ f 0 ( z ) ∣ V ( t ; 0 , z ) ∣ 2 d z \frac12\iint|v|^2f(t)=\frac12\int f_0(z)|V(t;0,z)|^2\,\mathrm{d}z 2 1 ∬ ∣ v ∣ 2 f ( t ) = 2 1 ∫ f 0 ( z ) ∣ V ( t ; 0 , z ) ∣ 2 d z . The integrand is absolutely continuous in t t t with derivative V ⋅ ( u ( t , X ) − V ) V\cdot(u(t,X)-V) V ⋅ ( u ( t , X ) − V ) , bounded by an integrable function on supp f 0 \operatorname{supp}f_0 supp f 0 . Hence
d d t 1 2 ∬ ∣ v ∣ 2 f d z = ∫ f 0 V ⋅ ( u ( t , X ) − V ) d z = ∫ j f ⋅ u d x − ∬ ∣ v ∣ 2 f . \frac{d}{dt}\frac12\iint|v|^2f\,\mathrm{d}z=\int f_0\,V\cdot(u(t,X)-V)\,\mathrm{d}z=\int j_f\cdot u\,\mathrm{d}x-\iint|v|^2f. d t d 2 1 ∬ ∣ v ∣ 2 f d z = ∫ f 0 V ⋅ ( u ( t , X ) − V ) d z = ∫ j f ⋅ u d x − ∬ ∣ v ∣ 2 f .
Adding the two identities and using ∣ v − u ∣ 2 = ∣ v ∣ 2 − 2 v ⋅ u + ∣ u ∣ 2 |v-u|^2=|v|^2-2v\cdot u+|u|^2 ∣ v − u ∣ 2 = ∣ v ∣ 2 − 2 v ⋅ u + ∣ u ∣ 2 gives
d d t E + ∥ ∇ u ∥ 2 + ∬ ∣ v − u ∣ 2 f = 0 , \frac{\mathrm{d}}{\mathrm{d}t}\mathcal{E}+\|\nabla u\|^2+\iint|v-u|^2f=0, d t d E + ∥∇ u ∥ 2 + ∬ ∣ v − u ∣ 2 f = 0 ,
which integrates to (21) .
Corollary 5.11 (Rough initial velocity). Let d ≥ 2 d\ge2 d ≥ 2 , T ∈ ( 0 , 1 ] T\in(0,1] T ∈ ( 0 , 1 ] , q 0 ∈ ( d , ∞ ) q_0\in(d,\infty) q 0 ∈ ( d , ∞ ) , let u i n ∈ B q 0 , 2 1 ( T d ) u_{\mathrm{in}}\in B^1_{q_0,2}(\mathbb{T}^d) u in ∈ B q 0 , 2 1 ( T d ) be divergence-free, and let f 0 ∈ L ∞ f_0\in L^\infty f 0 ∈ L ∞ be nonnegative and vanish for ∣ v ∣ ≥ R v 0 |v|\ge R_v^0 ∣ v ∣ ≥ R v 0 . Then every weak solution ( u , f ) (u,f) ( u , f ) on [ 0 , T ] [0,T] [ 0 , T ] with these data and u ∈ L 2 ( 0 , T ; L ∞ ) u\in L^2(0,T;L^\infty) u ∈ L 2 ( 0 , T ; L ∞ ) satisfies:
(a) u ∈ M R q 0 ( [ 0 , T ] ) u\in\mathrm{MR}_{q_0}([0,T]) u ∈ MR q 0 ([ 0 , T ]) , hence ∇ u ∈ L 2 ( 0 , T ; L ∞ ) \nabla u\in L^2(0,T;L^\infty) ∇ u ∈ L 2 ( 0 , T ; L ∞ ) , and u ∈ M R s ( [ t ∗ , T ] ) u\in\mathrm{MR}_s([t_*,T]) u ∈ MR s ([ t ∗ , T ]) for every s < ∞ s<\infty s < ∞ and t ∗ ∈ ( 0 , T ) t_*\in(0,T) t ∗ ∈ ( 0 , T ) ;
(b) f f f is the push-forward of f 0 f_0 f 0 under the Lipschitz phase-space flow of u u u , and the energy equality (21) holds;
(c) there is at most one such weak solution.
Proof. The proof of Proposition 5.10(b) applies with q = q 0 q=q_0 q = q 0 , since Proposition 5.6 only needs v 0 ∈ X q 0 v_0\in X_{q_0} v 0 ∈ X q 0 . This gives u ∈ M R q 0 u\in\mathrm{MR}_{q_0} u ∈ MR q 0 and, as q 0 > d q_0>d q 0 > d , ∇ u ∈ L 2 ( 0 , T ; L ∞ ) \nabla u\in L^2(0,T;L^\infty) ∇ u ∈ L 2 ( 0 , T ; L ∞ ) . Parts (c) and (d) of that proposition then follow verbatim, which gives (b). Let η ∈ C ∞ ( R ; [ 0 , 1 ] ) \eta\in C^\infty(\mathbb{R};[0,1]) η ∈ C ∞ ( R ; [ 0 , 1 ]) vanish near 0 0 0 and equal 1 1 1 on [ t ∗ , ∞ ) [t_*,\infty) [ t ∗ , ∞ ) . Then v : = η u v:=\eta u v := η u is a weak solution of the Oseen problem (5.4) with drift u u u , zero initial value and forcing η F + η ′ u ∈ L 2 ( 0 , T ; L s ) \eta F+\eta'u\in L^2(0,T;L^s) η F + η ′ u ∈ L 2 ( 0 , T ; L s ) for every s < ∞ s<\infty s < ∞ , since F F F is bounded and u ∈ L 2 ( 0 , T ; L ∞ ) u\in L^2(0,T;L^\infty) u ∈ L 2 ( 0 , T ; L ∞ ) . By Propositions 5.6 and 5.9, v v v is the strong solution, so v ∈ M R s ( [ 0 , T ] ) v\in\mathrm{MR}_s([0,T]) v ∈ MR s ([ 0 , T ]) and u ∈ M R s ( [ t ∗ , T ] ) u\in\mathrm{MR}_s([t_*,T]) u ∈ MR s ([ t ∗ , T ]) . Finally, the proof of Theorem 5.1(b) in Section 5.5 uses only (a) with q = q 0 q=q_0 q = q 0 and (b), which gives (c).
Lagrangian stability and proof of Theorem 5.1
Part (a) of Theorem 5.1 is Proposition 5.10. Part (b) adapts [20 Section 4.3] to dimension d d d and to the class L t 2 L ∞ L^2_tL^\infty L t 2 L ∞ . In two dimensions, the bound u ∈ L t 2 L ∞ u\in L^2_tL^\infty u ∈ L t 2 L ∞ holds for every weak solution in the energy class [20 Lemma 3] , as a consequence of the energy identity [20 Lemma 2] , and the velocity is log-Lipschitz, so an Osgood argument is used there. In our setting ∇ u i ∈ L t 2 L ∞ \nabla u_i\in L^2_tL^\infty ∇ u i ∈ L t 2 L ∞ by Proposition 5.10, and Gronwall’s lemma suffices. For (b), let ( u 1 , f 1 ) (u_1,f_1) ( u 1 , f 1 ) and ( u 2 , f 2 ) (u_2,f_2) ( u 2 , f 2 ) be two weak solutions as in the theorem, and let Z i = ( X i , V i ) Z_i=(X_i,V_i) Z i = ( X i , V i ) be the flows of Proposition 5.10(c), lifted to R d × R d \mathbb{R}^{d} \times\mathbb{R}^{d} R d × R d (with u i u_i u i extended periodically) and started at time 0 0 0 . Set
w : = u 1 − u 2 , δ Z : = ( δ X , δ V ) : = Z 1 − Z 2 , D ( t ) 2 : = ∫ f 0 ( z ) ∣ δ Z ( t , z ) ∣ 2 d z , Y : = ∥ w ∥ L 2 2 + D 2 . w := u_1-u_2,\qquad\delta Z := (\delta X,\delta V) := Z_1-Z_2,\qquad D(t)^2 := \int f_0(z)|\delta Z(t,z)|^2\,\mathrm{d}z,\qquad Y := \|w\|_{L^2}^2+D^2. w := u 1 − u 2 , δ Z := ( δ X , δ V ) := Z 1 − Z 2 , D ( t ) 2 := ∫ f 0 ( z ) ∣ δ Z ( t , z ) ∣ 2 d z , Y := ∥ w ∥ L 2 2 + D 2 .
Let R ∗ v R_*^v R ∗ v bound the velocity supports of f 1 , f 2 f_1,f_2 f 1 , f 2 , and let ϱ ∗ : = max i ∥ ρ f i ∥ L ∞ ≤ e d ∥ f 0 ∥ L ∞ ∣ B R ∗ v ∣ \varrho_* := \max_i\|\rho_{f_i}\|_{L^\infty} \leq e^d\|f_0\|_{L^\infty}|B_{R_*^v}| ϱ ∗ := max i ∥ ρ f i ∥ L ∞ ≤ e d ∥ f 0 ∥ L ∞ ∣ B R ∗ v ∣ . By Proposition 5.10(c), for every periodic measurable h ≥ 0 h \geq0 h ≥ 0 ,
∫ f 0 ( z ) h ( X i ( t , z ) ) d z = ∫ T d ρ f i ( t , x ) h ( x ) d x ≤ ϱ ∗ ∥ h ∥ L 1 ( T d ) . (27) \int f_0(z)h(X_i(t,z))\,\mathrm{d}z=\int_{\mathbb{T}^d}\rho_{f_i}(t,x)h(x)\,\mathrm{d}x\leq\varrho_*\|h\|_{L^1(\mathbb{T}^d)}. \tag*{(27)} ∫ f 0 ( z ) h ( X i ( t , z )) d z = ∫ T d ρ f i ( t , x ) h ( x ) d x ≤ ϱ ∗ ∥ h ∥ L 1 ( T d ) . (27)
Step 1: the flows. For f 0 f_0 f 0 -a.e. z z z , the map t ↦ δ Z ( t , z ) t\mapsto\delta Z(t,z) t ↦ δ Z ( t , z ) is absolutely continuous with δ Z ( 0 , z ) = 0 \delta Z(0,z)=0 δ Z ( 0 , z ) = 0 , d d t δ X = δ V \frac{\mathrm{d}}{\mathrm{d}t}\delta X=\delta V d t d δ X = δ V and
d d t δ V = ( u 1 ( t , X 1 ) − u 1 ( t , X 2 ) ) + w ( t , X 2 ) − δ V . \frac{\mathrm{d}}{\mathrm{d}t}\delta V=(u_1(t,X_1)-u_1(t,X_2))+w(t,X_2)-\delta V. d t d δ V = ( u 1 ( t , X 1 ) − u 1 ( t , X 2 )) + w ( t , X 2 ) − δ V .
Since ∣ u 1 ( t , X 1 ) − u 1 ( t , X 2 ) ∣ ≤ ∥ ∇ u 1 ( t ) ∥ L ∞ ∣ δ X ∣ |u_1(t,X_1)-u_1(t,X_2)|\leq\|\nabla u_1(t)\|_{L^\infty}|\delta X| ∣ u 1 ( t , X 1 ) − u 1 ( t , X 2 ) ∣ ≤ ∥∇ u 1 ( t ) ∥ L ∞ ∣ δ X ∣ , Young’s inequality gives
d d t ∣ δ Z ∣ 2 ≤ 2 ( 1 + ∥ ∇ u 1 ( t ) ∥ L ∞ ) ∣ δ Z ∣ 2 + ∣ w ( t , X 2 ) ∣ 2 . \frac{\mathrm{d}}{\mathrm{d}t}|\delta Z|^2\leq2(1+\|\nabla u_1(t)\|_{L^\infty})|\delta Z|^2+|w(t,X_2)|^2. d t d ∣ δ Z ∣ 2 ≤ 2 ( 1 + ∥∇ u 1 ( t ) ∥ L ∞ ) ∣ δ Z ∣ 2 + ∣ w ( t , X 2 ) ∣ 2 .
Integrating against f 0 f_0 f 0 and using (5.7) with h = ∣ w ( t ) ∣ 2 h=|w(t)|^2 h = ∣ w ( t ) ∣ 2 ,
d d t D 2 ≤ 2 ( 1 + ∥ ∇ u 1 ( t ) ∥ L ∞ ) D 2 + ϱ ∗ ∥ w ( t ) ∥ L 2 2 . (28) \frac{\mathrm{d}}{\mathrm{d}t}D^2\leq2(1+\|\nabla u_1(t)\|_{L^\infty})D^2+\varrho_*\|w(t)\|_{L^2}^2. \tag*{(28)} d t d D 2 ≤ 2 ( 1 + ∥∇ u 1 ( t ) ∥ L ∞ ) D 2 + ϱ ∗ ∥ w ( t ) ∥ L 2 2 . (28)
This is justified by Fubini, since ∣ δ Z ∣ |\delta Z| ∣ δ Z ∣ is bounded on [ 0 , T ] × supp f 0 [0,T]\times\operatorname{supp}f_0 [ 0 , T ] × supp f 0 .
Step 2: the fluid. By Proposition 5.10(b), w ∈ M R 2 w\in M_{R^2} w ∈ M R 2 solves ∂ t w − Δ w + Q ( u 1 ⋅ ∇ w + w ⋅ ∇ u 2 ) = Q ( F 1 − F 2 ) \partial_t w-\Delta w+\mathbb{Q}(u_1\cdot\nabla w+w\cdot\nabla u_2)=\mathbb{Q}(F_1-F_2) ∂ t w − Δ w + Q ( u 1 ⋅ ∇ w + w ⋅ ∇ u 2 ) = Q ( F 1 − F 2 ) with F i = F [ f i , u i ] F_i=F[f_i,u_i] F i = F [ f i , u i ] . Testing with w w w , we use ∫ ( u 1 ⋅ ∇ w ) ⋅ w = 0 \int(u_1\cdot\nabla w)\cdot w=0 ∫ ( u 1 ⋅ ∇ w ) ⋅ w = 0 and ∫ ( w ⋅ ∇ u 2 ) ⋅ w = − ∫ ( w ⋅ ∇ w ) ⋅ u 2 \int(w\cdot\nabla u_2)\cdot w=-\int(w\cdot\nabla w)\cdot u_2 ∫ ( w ⋅ ∇ u 2 ) ⋅ w = − ∫ ( w ⋅ ∇ w ) ⋅ u 2 . Since F 1 − F 2 = ∫ ( v − u 1 ) ( f 1 − f 2 ) d v − ρ f 2 w F_1-F_2=\int(v-u_1)(f_1-f_2)\,\mathrm{d}v-\rho_{f_2}w F 1 − F 2 = ∫ ( v − u 1 ) ( f 1 − f 2 ) d v − ρ f 2 w ,
1 2 d d t ∥ w ∥ 2 2 + ∥ ∇ w ∥ 2 2 ≤ ∥ u 2 ∥ L ∞ ∥ w ∥ 2 ∥ ∇ w ∥ 2 + I ( t ) , I ( t ) : = ∬ ( f 1 − f 2 ) ( t ) ψ t d z , (29) \frac{1}{2}\frac{\mathrm{d}}{\mathrm{d}t}\|w\|_2^2+\|\nabla w\|_2^2\leq\|u_2\|_{L^\infty}\|w\|_2\|\nabla w\|_2+I(t),\qquad I(t):=\iint(f_1-f_2)(t)\psi_t\,\mathrm{d}z, \tag*{(29)} 2 1 d t d ∥ w ∥ 2 2 + ∥∇ w ∥ 2 2 ≤ ∥ u 2 ∥ L ∞ ∥ w ∥ 2 ∥∇ w ∥ 2 + I ( t ) , I ( t ) := ∬ ( f 1 − f 2 ) ( t ) ψ t d z , (29)
where ψ t ( x , v ) = ( v − u 1 ( t , x ) ) ⋅ w ( t , x ) \psi_t(x,v)=(v-u_1(t,x))\cdot w(t,x) ψ t ( x , v ) = ( v − u 1 ( t , x )) ⋅ w ( t , x ) and we dropped − ∫ ρ f 2 ∣ w ∣ 2 ≤ 0 -\int\rho_{f_2}|w|^2\leq0 − ∫ ρ f 2 ∣ w ∣ 2 ≤ 0 .
Step 3: the coupling term. By Proposition 5.10(c), I ( t ) = ∫ f 0 ( z ) ( ψ t ( Z 1 ) − ψ t ( Z 2 ) ) d z I(t)=\int f_0(z)(\psi_t(Z_1)-\psi_t(Z_2))\,\mathrm{d}z I ( t ) = ∫ f 0 ( z ) ( ψ t ( Z 1 ) − ψ t ( Z 2 )) d z , and
ψ t ( Z 1 ) − ψ t ( Z 2 ) = δ V ⋅ w ( X 1 ) − ( u 1 ( X 1 ) − u 1 ( X 2 ) ) ⋅ w ( X 1 ) + ( V 2 − u 1 ( X 2 ) ) ⋅ ( w ( X 1 ) − w ( X 2 ) ) , \psi_t(Z_1)-\psi_t(Z_2)=\delta V\cdot w(X_1)-(u_1(X_1)-u_1(X_2))\cdot w(X_1)+(V_2-u_1(X_2))\cdot(w(X_1)-w(X_2)), ψ t ( Z 1 ) − ψ t ( Z 2 ) = δ V ⋅ w ( X 1 ) − ( u 1 ( X 1 ) − u 1 ( X 2 )) ⋅ w ( X 1 ) + ( V 2 − u 1 ( X 2 )) ⋅ ( w ( X 1 ) − w ( X 2 )) ,
with all functions evaluated at time t t t . We bound the three terms after integration against f 0 f_0 f 0 .
For the third, note that ∣ V 2 − u 1 ( X 2 ) ∣ ≤ R ∗ v + ∥ u 1 ( t ) ∥ L ∞ |V_2-u_1(X_2)|\leq R_*^v+\|u_1(t)\|_{L^\infty} ∣ V 2 − u 1 ( X 2 ) ∣ ≤ R ∗ v + ∥ u 1 ( t ) ∥ L ∞ on supp f 0 \operatorname{supp}f_0 supp f 0 . For a.e. t t t we have w ( t ) ∈ H 1 w(t)\in H^1 w ( t ) ∈ H 1 . Let N t N_t N t be the null set of Lemma 5.5 for g = w ( t ) g=w(t) g = w ( t ) . The sets { z : X i ( t , z ) ∈ N t } \{z:X_i(t,z)\in N_t\} { z : X i ( t , z ) ∈ N t } are null, because Z i ( t ; 0 , ⋅ ) Z_i(t;0,\cdot) Z i ( t ; 0 , ⋅ ) is bi-Lipschitz. By (5.3), Cauchy–Schwarz, (5.7) and (5.2), the third term is at most
( R ∗ v + ∥ u 1 ∥ L ∞ ) C d ∑ i = 1 , 2 ∫ f 0 ∣ δ X ∣ M ∣ ∇ w ∣ ( X i ) d z ≤ 2 C d C M ϱ ∗ 1 / 2 ( R ∗ v + ∥ u 1 ∥ L ∞ ) D ∥ ∇ w ∥ 2 . (R_*^v+\|u_1\|_{L^\infty})C_d\sum_{i=1,2}\int f_0|\delta X|M|\nabla w|(X_i)\,\mathrm{d}z\leq2C_dC_M\varrho_*^{1/2}(R_*^v+\|u_1\|_{L^\infty})D\|\nabla w\|_2. ( R ∗ v + ∥ u 1 ∥ L ∞ ) C d i = 1 , 2 ∑ ∫ f 0 ∣ δ X ∣ M ∣∇ w ∣ ( X i ) d z ≤ 2 C d C M ϱ ∗ 1/2 ( R ∗ v + ∥ u 1 ∥ L ∞ ) D ∥∇ w ∥ 2 .
Step 4: Gronwall. Insert Step 3 into (5.9) and absorb ∥ ∇ w ∥ 2 \|\nabla w\|_2 ∥∇ w ∥ 2 by Young’s inequality. Combined with (5.8), this gives, for a.e. t t t ,
d d t Y ≤ C a ( t ) Y , a ( t ) : = 1 + ∥ u 1 ( t ) ∥ L ∞ 2 + ∥ u 2 ( t ) ∥ L ∞ 2 + ∥ ∇ u 1 ( t ) ∥ L ∞ , \frac{\mathrm{d}}{\mathrm{d}t}Y\leq Ca(t)Y,\qquad a(t):=1+\|u_1(t)\|_{L^\infty}^2+\|u_2(t)\|_{L^\infty}^2+\|\nabla u_1(t)\|_{L^\infty}, d t d Y ≤ C a ( t ) Y , a ( t ) := 1 + ∥ u 1 ( t ) ∥ L ∞ 2 + ∥ u 2 ( t ) ∥ L ∞ 2 + ∥∇ u 1 ( t ) ∥ L ∞ ,
with C C C depending only on d d d , R ∗ v R_*^v R ∗ v and ϱ ∗ \varrho_* ϱ ∗ . Here a ∈ L 1 ( 0 , T ) a\in L^1(0,T) a ∈ L 1 ( 0 , T ) by the hypothesis u i ∈ L t 2 L ∞ u_i\in L_t^2L^\infty u i ∈ L t 2 L ∞ and by Proposition 5.10(b). Since Y Y Y is absolutely continuous and Y ( 0 ) = 0 Y(0)=0 Y ( 0 ) = 0 , we get Y ≡ 0 Y\equiv0 Y ≡ 0 . Thus u 1 = u 2 u_1=u_2 u 1 = u 2 , and then f 1 = f 2 f_1=f_2 f 1 = f 2 by Lemma 3.6, or directly since δ Z = 0 \delta Z=0 δ Z = 0 on supp f 0 \operatorname{supp}f_0 supp f 0 . This proves Theorem 5.1(b). □ \square □
Remark 5.12. Step 4 uses the Lipschitz bound only for u 1 u_1 u 1 . The argument is therefore a weak–strong uniqueness statement for the flows, combined with the fact that L t 2 L ∞ L_t^2L^\infty L t 2 L ∞ solutions are automatically strong (Proposition 5.10). The endpoint enters in two places: in the integrability of ∥ u i ∥ L ∞ 2 \|u_i\|_{L^\infty}^2 ∥ u i ∥ L ∞ 2 in a ( t ) a(t) a ( t ) , and in the absorption of the drift term in the maximal regularity argument (Proposition 5.6). Both fail for u ∈ L t p L ∞ u\in L_t^pL^\infty u ∈ L t p L ∞ with p < 2 p<2 p < 2 , in agreement with Theorem 2.3.
Intervals of regularity
Cheskidov and Luo show that their wild solutions are smooth outside a closed set of times of Hausdorff dimension at most ε \varepsilon ε [12 Theorem 1.8] . The tool is a concentration step that glues local exact solutions, so that the stress of each approximate solution lives on few short intervals [12 Section 3] . In this section we prove the analogue for VNS.
Because of the memory of the particles (§ 1.2 ), the solution can no longer be left untouched outside the bad set: on good intervals it must be re-solved with the new kinetic state, and every gluing must be causal , that is, carried out from left to right in time. The regularity obtained on good intervals is that of a strong solution, in the sense of § 5 , and not C ∞ C^\infty C ∞ in general (see the comment after Theorem 1.3 ). Figure 1 illustrates the memory effect and the resulting two-pass construction.
Statement
Since ∥ ⋅ ∥ W 1 , q ≤ ∥ ⋅ ∥ W 1 , q ′ \|\cdot\|_{W^{1,q}}\leq\|\cdot\|_{W^{1,q'}} ∥ ⋅ ∥ W 1 , q ≤ ∥ ⋅ ∥ W 1 , q ′ on T d \mathbb{T}^d T d for q ≤ q ′ q\leq q' q ≤ q ′ , it suffices to prove Theorem 6.2 below for large q q q . Throughout this section, 1 ≤ p < 2 1\leq p<2 1 ≤ p < 2 , 2 ≤ q < ∞ 2\leq q<\infty 2 ≤ q < ∞ with q > d q>d q > d , β ∈ ( 0 , 1 − d q ) \beta\in(0,1-\frac{d}{q}) β ∈ ( 0 , 1 − q d ) , ε ∈ ( 0 , 1 ) \varepsilon\in(0,1) ε ∈ ( 0 , 1 ) , and r = r ( d , p , q ) r=r(d,p,q) r = r ( d , p , q ) is the exponent of Appendix § 16 .§ 1 . All VNS–Reynolds solutions (Definition 2.6) carry the kinetic state f [ U ] f[U] f [ U ] , the push-forward of f 0 f_0 f 0 along the whole history of U U U .
Notation. In this section ς \varsigma ς denotes the size of a base stress, β ∈ ( 0 , 1 − d q ) \beta\in(0,1-\frac{d}{q}) β ∈ ( 0 , 1 − q d ) an interpolation exponent, and N c N_c N c the number of components of I ~ \tilde{I} I ~ . As in § 3 , radii in velocity are denoted by R ∗ v R_\ast^v R ∗ v , R 1 v R_1^v R 1 v , and R R R denotes a Reynolds stress.
Definition 6.1 (Well-prepared). A smooth VNS–Reynolds solution ( U , P , S ) (U,P,S) ( U , P , S ) on [ 0 , 1 ] [0,1] [ 0 , 1 ] is well-prepared for a set I I I and a scale τ > 0 \tau>0 τ > 0 if:
I I I is a union of at most τ − ε \tau^{-\varepsilon} τ − ε pairwise disjoint closed intervals of length 5 τ 5\tau 5 τ , with 0 , 1 ∉ I 0,1\notin I 0 , 1 ∈ / I ;
S ( t ) = 0 S(t)=0 S ( t ) = 0 whenever dist ( t , I c ) ≤ τ \operatorname{dist}(t,I^{c})\le\tau dist ( t , I c ) ≤ τ .
Theorem 6.2 (Intervals of regularity). Let d ≥ 2 d\ge2 d ≥ 2 , 1 ≤ p < 2 1\le p<2 1 ≤ p < 2 , q < ∞ q<\infty q < ∞ and ε > 0 \varepsilon>0 ε > 0 . Let ( u i n , f 0 ) (u_{\mathrm{in}},f_0) ( u in , f 0 ) be smooth data, with f 0 ≥ 0 f_0\ge0 f 0 ≥ 0 compactly supported, admitting a smooth VNS solution on some [ 0 , t 1 ] [0,t_1] [ 0 , t 1 ] . Then there are infinitely many weak solutions with these data, all coinciding with the smooth solution near t = 0 t=0 t = 0 , such that:
u ∈ L t , x 2 ∩ L t p L ∞ ∩ L t 1 W 1 , q u\in L^2_{t,x}\cap L^p_tL^\infty\cap L^1_tW^{1,q} u ∈ L t , x 2 ∩ L t p L ∞ ∩ L t 1 W 1 , q ;
there is a closed set S ⊂ ( 0 , 1 ) S\subset(0,1) S ⊂ ( 0 , 1 ) with dim H S ≤ ε \dim_{\mathrm H}S\le\varepsilon dim H S ≤ ε such that, on every compact interval K ⊂ ( 0 , 1 ] ∖ S K\subset(0,1]\setminus S K ⊂ ( 0 , 1 ] ∖ S and for every s < ∞ s<\infty s < ∞ , u ∈ L 2 ( K ; W 1 , ∞ ) ∩ L 2 ( K ; W 2 , s ) ∩ C ( K ; W 1 , s ) u\in L^2(K;W^{1,\infty})\cap L^2(K;W^{2,s})\cap C(K;W^{1,s}) u ∈ L 2 ( K ; W 1 , ∞ ) ∩ L 2 ( K ; W 2 , s ) ∩ C ( K ; W 1 , s ) , and f f f is transported on K K K by a Lipschitz phase-space flow. In particular, by Corollary 5.11, the solution is the unique weak solution on K K K with velocity in L 2 ( K ; L ∞ ) L^2(K;L^\infty) L 2 ( K ; L ∞ ) and initial state ( u , f ) ( inf K ) (u,f)(\inf K) ( u , f ) ( inf K ) .
The theorem follows from iterating the following proposition.
Proposition 6.3 (Well-prepared iteration). There are M W M_W M W and C r C_r C r , depending only on d d d and r r r , with the following property. Let ( u , p u , R ) (u,p_u,R) ( u , p u , R ) be well-prepared for ( I ~ , τ ~ ) (\widetilde I,\widetilde\tau) ( I , τ ) , and let δ > 0 \delta>0 δ > 0 . Then there is ( u 1 , p 1 , R 1 ) (u_1,p_1,R_1) ( u 1 , p 1 , R 1 ) , well-prepared for some ( I , τ ) (I,\tau) ( I , τ ) with I ⊂ I ~ I\subset\widetilde I I ⊂ I and τ ≤ τ ~ / 2 \tau\le\widetilde\tau/2 τ ≤ τ /2 , such that:
(W1) ∥ R 1 ∥ L 1 ( 0 , 1 ; L r ) ≤ δ \|R_1\|_{L^1(0,1;L^r)}\le\delta ∥ R 1 ∥ L 1 ( 0 , 1 ; L r ) ≤ δ ;
(W2) ∥ u 1 − u ∥ L t , x 2 ≤ M W ∥ R ∥ L 1 ( 0 , 1 ; L r ) 1 / 2 + δ \|u_1-u\|_{L^2_{t,x}}\le M_W\|R\|_{L^1(0,1;L^r)}^{1/2}+\delta ∥ u 1 − u ∥ L t , x 2 ≤ M W ∥ R ∥ L 1 ( 0 , 1 ; L r ) 1/2 + δ ;
(W3) ∥ u 1 − u ∥ L p L ∞ + ∥ u 1 − u ∥ L 1 W 1 , q ≤ δ \|u_1-u\|_{L^pL^\infty}+\|u_1-u\|_{L^1W^{1,q}}\le\delta ∥ u 1 − u ∥ L p L ∞ + ∥ u 1 − u ∥ L 1 W 1 , q ≤ δ ;
(W4) u 1 = u u_1=u u 1 = u on [ 0 , inf I ~ ] [0,\inf\widetilde I] [ 0 , inf I ] ;
(W5) ∥ u 1 − u ∥ L 2 ( I ~ c ; W 1 , ∞ ) ≤ δ \|u_1-u\|_{L^2(\widetilde I^c;W^{1,\infty})}\le\delta ∥ u 1 − u ∥ L 2 ( I c ; W 1 , ∞ ) ≤ δ .
(W1)–(W4) are, up to the square root in (W2) (see § 4 ) and the weaker support condition in (W4), contained in [12 Proposition 2.2] . (W5) is new. Since the kinetic state on I ~ c \widetilde I^c I c depends on the modifications made in I ~ \widetilde I I , the velocity on I ~ c \widetilde I^c I c must be corrected. (W5) states that this correction is small in the Lipschitz norm, which is what makes the limit strong off S S S .
Two analytic tools
For s ∈ R s\in\mathbb{R} s ∈ R let J s = ( 1 − Δ ) s / 2 J^s=(1-\Delta)^{s/2} J s = ( 1 − Δ ) s /2 and H s , q = J − s L q ( T d ) H^{s,q}=J^{-s}L^q(\mathbb{T}^d) H s , q = J − s L q ( T d ) ; recall that H 1 , q = W 1 , q H^{1,q}=W^{1,q} H 1 , q = W 1 , q for 1 < q < ∞ 1<q<\infty 1 < q < ∞ . Let X q = B q , 2 1 X_q=B^1_{q,2} X q = B q , 2 1 and let M R q ( I ) MR_q(I) M R q ( I ) be as in Lemma 5.4. Set
Y 0 ( I ) : = M R q ( I ) , Y 1 ( I ) : = J M R q ( I ) = L 2 ( I ; H 1 , q ) ∩ H 1 ( I ; H − 1 , q ) , ∥ v ∥ Y 1 ( I ) : = ∥ J − 1 v ∥ M R q ( I ) . Y_0(I):=MR_q(I), \qquad Y_1(I):=JMR_q(I)=L^2(I;H^{1,q})\cap H^1(I;H^{-1,q}), \qquad\|v\|_{Y_1(I)}:=\|J^{-1}v\|_{MR_q(I)}. Y 0 ( I ) := M R q ( I ) , Y 1 ( I ) := J M R q ( I ) = L 2 ( I ; H 1 , q ) ∩ H 1 ( I ; H − 1 , q ) , ∥ v ∥ Y 1 ( I ) := ∥ J − 1 v ∥ M R q ( I ) .
Lemma 6.4 (Shifted maximal regularity). There is C q ≥ 1 C_q\ge1 C q ≥ 1 , independent of I ⊂ [ 0 , 1 ] I\subset[0,1] I ⊂ [ 0 , 1 ] , such that:
(a) Lemma 5.4(a), (b) hold in Y 1 Y_1 Y 1 , with data in J X q = B q , 2 0 JX_q=B^0_{q,2} J X q = B q , 2 0 and forcing in L 2 ( I ; H − 1 , q ) L^2(I;H^{-1,q}) L 2 ( I ; H − 1 , q ) ;
(b) B q , 2 0 ↪ L q B^0_{q,2}\hookrightarrow L^q B q , 2 0 ↪ L q , X q ↪ W 1 , q ↪ L ∞ X_q\hookrightarrow W^{1,q}\hookrightarrow L^\infty X q ↪ W 1 , q ↪ L ∞ , Y 1 ↪ L 2 L ∞ Y_1\hookrightarrow L^2L^\infty Y 1 ↪ L 2 L ∞ and Y 0 ↪ L 2 W 1 , ∞ Y_0\hookrightarrow L^2W^{1,\infty} Y 0 ↪ L 2 W 1 , ∞ ;
(c) ∥ v ∥ L 2 H 2 − β , q ≤ C ∥ v ∥ L 2 H 2 , q 1 − β ∥ v ∥ L 2 H 1 , q β \|v\|_{L^2H^{2-\beta,q}}\le C\|v\|_{L^2H^{2,q}}^{1-\beta}\|v\|_{L^2H^{1,q}}^\beta ∥ v ∥ L 2 H 2 − β , q ≤ C ∥ v ∥ L 2 H 2 , q 1 − β ∥ v ∥ L 2 H 1 , q β and H 2 − β , q ↪ W 1 , ∞ H^{2-\beta,q}\hookrightarrow W^{1,\infty} H 2 − β , q ↪ W 1 , ∞ ;
(d) ∥ v ( t ) ∥ B q , 2 1 − β ≤ C ∥ v ( t ) ∥ X q 1 − β ∥ v ( t ) ∥ B q , 2 0 β \|v(t)\|_{B^{1-\beta}_{q,2}}\le C\|v(t)\|_{X_q}^{1-\beta}\|v(t)\|_{B^0_{q,2}}^\beta ∥ v ( t ) ∥ B q , 2 1 − β ≤ C ∥ v ( t ) ∥ X q 1 − β ∥ v ( t ) ∥ B q , 2 0 β and B q , 2 1 − β ↪ L ∞ B^{1-\beta}_{q,2}\hookrightarrow L^\infty B q , 2 1 − β ↪ L ∞ ;
(e) for 1 < s < ∞ 1<s<\infty 1 < s < ∞ , the antidivergence satisfies ∥ R g ∥ L s ≤ C s ∥ g ∥ H − 1 , s \|\mathcal{R}g\|_{L^s}\le C_s\|g\|_{H^{-1,s}} ∥ R g ∥ L s ≤ C s ∥ g ∥ H − 1 , s and ∥ R g ∥ W 1 , s ≤ C s ∥ g ∥ L s \|\mathcal{R}g\|_{W^{1,s}}\le C_s\|g\|_{L^s} ∥ R g ∥ W 1 , s ≤ C s ∥ g ∥ L s .
Proof. (a): conjugate Lemma 5.4 by J − 1 J^{-1} J − 1 , which commutes with ∂ t \partial_t ∂ t , Δ \Delta Δ and the heat semigroup. (b), (c), (d): Sobolev and Besov embeddings and real or complex interpolation [28 ] , using q > d q>d q > d and 1 − β > d q 1-\beta>\frac{d}{q} 1 − β > q d . (e): by [12 Definition B.2] , R \mathcal{R} R is a linear combination of ∇ Δ − 1 \nabla\Delta^{-1} ∇ Δ − 1 and ∇ 3 Δ − 2 \nabla^3\Delta^{-2} ∇ 3 Δ − 2 , applied to mean-free functions. These are Fourier multipliers of order − 1 -1 − 1 , so ∇ R \nabla\mathcal{R} ∇ R is a Calderón–Zygmund operator and R J \mathcal{R}J R J is bounded on L s L^s L s .
The next lemma sharpens Lemma 3.2: the perturbation is measured in L t 1 L x s L^1_tL^s_x L t 1 L x s for any s ∈ [ 1 , ∞ ] s \in[1,\infty] s ∈ [ 1 , ∞ ] instead of L t 1 L x ∞ L^1_tL^\infty_x L t 1 L x ∞ , and the bound is linear in the perturbation.
Lemma 6.5 (Kinetic duality in ( W 1 , 1 ) ∗ (W^{1,1})^* ( W 1 , 1 ) ∗ ). Let 0 ≤ t 0 < t ≤ 1 0 \le t_0 < t \le1 0 ≤ t 0 < t ≤ 1 . Let b , U b,U b , U be fields on [ t 0 , 1 ] × T d [t_0,1]\times\mathbb{T}^d [ t 0 , 1 ] × T d , Lipschitz in x x x for a.e. time, with ∇ b , ∇ U ∈ L 1 ( t 0 , 1 ; L ∞ ) \nabla b,\nabla U \in L^1(t_0,1;L^\infty) ∇ b , ∇ U ∈ L 1 ( t 0 , 1 ; L ∞ ) and b , U ∈ L 1 ( t 0 , 1 ; L ∞ ) b,U \in L^1(t_0,1;L^\infty) b , U ∈ L 1 ( t 0 , 1 ; L ∞ ) . Set Λ b : = ∫ t 0 1 ∥ ∇ b ∥ ∞ \Lambda_b := \int_{t_0}^1 \|\nabla b\|_\infty Λ b := ∫ t 0 1 ∥∇ b ∥ ∞ . Let f ∗ ≥ 0 f_* \ge0 f ∗ ≥ 0 be bounded and vanish for ∣ v ∣ > R ∗ v |v|>R_*^v ∣ v ∣ > R ∗ v , and let f U , f b f^U,f^b f U , f b be its push-forwards from t 0 t_0 t 0 along b U b_U b U and b b b_b b b . Put
R 1 v : = R ∗ v + ∥ U ∥ L 1 L ∞ + ∥ b ∥ L 1 L ∞ , K : = e 1 + Λ b + d ∥ f ∗ ∥ ∞ ∣ B R 1 v ∣ . R_1^v := R_*^v+\|U\|_{L^1L^\infty}+\|b\|_{L^1L^\infty},\qquad K:=e^{1+\Lambda_b+d}\|f_*\|_\infty|B_{R_1^v}|. R 1 v := R ∗ v + ∥ U ∥ L 1 L ∞ + ∥ b ∥ L 1 L ∞ , K := e 1 + Λ b + d ∥ f ∗ ∥ ∞ ∣ B R 1 v ∣. (a) Let 1 ≤ s ≤ ∞ 1\le s\le\infty 1 ≤ s ≤ ∞ , ψ ∈ C 1 ( T d × R d ) \psi\in C^1(\mathbb{T}^d\times\mathbb{R}^d) ψ ∈ C 1 ( T d × R d ) and 0 ≤ k ∈ L s ′ ( T d ) 0\le k\in L^{s'}(\mathbb{T}^d) 0 ≤ k ∈ L s ′ ( T d ) with max { ∣ ∇ x ψ ( x , v ) ∣ , ∣ ∇ v ψ ( x , v ) ∣ } ≤ k ( x ) \max\{|\nabla_x\psi(x,v)|,|\nabla_v\psi(x,v)|\}\le k(x) max { ∣ ∇ x ψ ( x , v ) ∣ , ∣ ∇ v ψ ( x , v ) ∣ } ≤ k ( x ) for a.e. x x x and all ∣ v ∣ ≤ R 1 v |v|\le R_1^v ∣ v ∣ ≤ R 1 v . Then
∣ ∬ ( f U − f b ) ( t ) ψ ∣ ≤ K ∥ k ∥ L s ′ ( T d ) ∫ t 0 t ∥ U − b ∥ L s d τ . \left|\iint(f^U-f^b)(t)\psi\right|\le K\|k\|_{L^{s'}(\mathbb{T}^d)}\int_{t_0}^t\|U-b\|_{L^s}\,d\tau. ∬ ( f U − f b ) ( t ) ψ ≤ K ∥ k ∥ L s ′ ( T d ) ∫ t 0 t ∥ U − b ∥ L s d τ . (b) Let d < q < ∞ d<q<\infty d < q < ∞ , q ′ < s ≤ q q'<s\le q q ′ < s ≤ q , W ∈ W 1 , q ( T d ) W\in W^{1,q}(\mathbb{T}^d) W ∈ W 1 , q ( T d ) , and h : = ∫ ( v − W ) ( f U − f b ) ( t ) d v h:=\int(v-W)(f^U-f^b)(t)\,dv h := ∫ ( v − W ) ( f U − f b ) ( t ) d v . Then
∥ h ∥ H − 1 , s ≤ C s , q K ( 1 + R 1 v + ∥ W ∥ W 1 , q ) ∫ t 0 t ∥ U − b ∥ L s d τ . \|h\|_{H^{-1,s}}\le C_{s,q}K(1+R_1^v+\|W\|_{W^{1,q}})\int_{t_0}^t\|U-b\|_{L^s}\,d\tau. ∥ h ∥ H − 1 , s ≤ C s , q K ( 1 + R 1 v + ∥ W ∥ W 1 , q ) ∫ t 0 t ∥ U − b ∥ L s d τ .
Proof. Step 1: smooth fields. Assume first that b b b and U U U are smooth on [ t 0 , 1 ] × T d [t_0,1]\times\mathbb{T}^d [ t 0 , 1 ] × T d ; f ∗ f_* f ∗ is only bounded, so we argue in Lagrangian form. Let Φ ( τ , y ) : = ψ ( Z b ( t ; τ , y ) ) \Phi(\tau,y):=\psi(Z_b(t;\tau,y)) Φ ( τ , y ) := ψ ( Z b ( t ; τ , y )) , which is C 1 C^1 C 1 and satisfies ∂ τ Φ + b b ⋅ ∇ y Φ = 0 \partial_\tau\Phi+b_b\cdot\nabla_y\Phi=0 ∂ τ Φ + b b ⋅ ∇ y Φ = 0 . For fixed ζ \zeta ζ the function γ ( τ ) : = Φ ( τ , Z U ( τ ; t 0 , ζ ) ) \gamma(\tau):=\Phi(\tau,Z_U(\tau;t_0,\zeta)) γ ( τ ) := Φ ( τ , Z U ( τ ; t 0 , ζ )) is C 1 C^1 C 1 on [ t 0 , t ] [t_0,t] [ t 0 , t ] , with γ ( t 0 ) = ψ ( Z b ( t ; t 0 , ζ ) ) \gamma(t_0)=\psi(Z_b(t;t_0,\zeta)) γ ( t 0 ) = ψ ( Z b ( t ; t 0 , ζ )) , γ ( t ) = ψ ( Z U ( t ; t 0 , ζ ) ) \gamma(t)=\psi(Z_U(t;t_0,\zeta)) γ ( t ) = ψ ( Z U ( t ; t 0 , ζ )) and
γ ′ ( τ ) = ( ∂ τ Φ + b U ⋅ ∇ y Φ ) ( τ , Z U ) = ( ( b U − b b ) ⋅ ∇ y Φ ) ( τ , Z U ) = ( U − b ) ( τ , X U ) ⋅ ∇ v Φ ( τ , Z U ) . \gamma'(\tau)=(\partial_\tau\Phi+b_U\cdot\nabla_y\Phi)(\tau,Z_U)=((b_U-b_b)\cdot\nabla_y\Phi)(\tau,Z_U)=(U-b)(\tau,X_U)\cdot\nabla_v\Phi(\tau,Z_U). γ ′ ( τ ) = ( ∂ τ Φ + b U ⋅ ∇ y Φ ) ( τ , Z U ) = (( b U − b b ) ⋅ ∇ y Φ ) ( τ , Z U ) = ( U − b ) ( τ , X U ) ⋅ ∇ v Φ ( τ , Z U ) .
since b U − b b = ( 0 , U − b ) b_U-b_b=(0,U-b) b U − b b = ( 0 , U − b ) . Integrating in τ \tau τ , multiplying by f ∗ ( ζ ) f_*(\zeta) f ∗ ( ζ ) , integrating in ζ \zeta ζ , and using the push-forward identities ∬ f U ( t ) ψ = ∫ f ∗ ψ ( Z U ( t ; t 0 , ⋅ ) ) \iint f^U(t)\psi=\int f_*\psi(Z_U(t;t_0,\cdot)) ∬ f U ( t ) ψ = ∫ f ∗ ψ ( Z U ( t ; t 0 , ⋅ )) , ∬ f b ( t ) ψ = ∫ f ∗ ψ ( Z b ( t ; t 0 , ⋅ ) ) \iint f^b(t)\psi=\int f_*\psi(Z_b(t;t_0,\cdot)) ∬ f b ( t ) ψ = ∫ f ∗ ψ ( Z b ( t ; t 0 , ⋅ )) and f U ( τ ) = Z U ( τ ; t 0 , ⋅ ) # f ∗ f^U(\tau)=Z_U(\tau;t_0,\cdot)_\#f_* f U ( τ ) = Z U ( τ ; t 0 , ⋅ ) # f ∗ , we get
∬ ( f U − f b ) ( t ) ψ = ∫ t 0 t ∬ f U ( τ , y ) ( U − b ) ( τ , x ) ⋅ ∇ v Φ ( τ , y ) d y d τ , y = ( x , v ) . (30) \iint(f^U-f^b)(t)\psi=\int_{t_0}^t\iint f^U(\tau,y)(U-b)(\tau,x)\cdot\nabla_v\Phi(\tau,y)\,dy\,d\tau,\qquad y=(x,v). \tag*{(30)} ∬ ( f U − f b ) ( t ) ψ = ∫ t 0 t ∬ f U ( τ , y ) ( U − b ) ( τ , x ) ⋅ ∇ v Φ ( τ , y ) d y d τ , y = ( x , v ) . (30)
No regularity of f ∗ f_* f ∗ is used. If y ∈ supp f U ( τ ) y\in\operatorname{supp}f^U(\tau) y ∈ supp f U ( τ ) , its velocity is at most R ∗ v + ∥ U ∥ L 1 L ∞ R_*^v+\|U\|_{L^1L^\infty} R ∗ v + ∥ U ∥ L 1 L ∞ , and it grows by at most ∥ b ∥ L 1 L ∞ \|b\|_{L^1L^\infty} ∥ b ∥ L 1 L ∞ along Z b Z_b Z b , so Z b ( t ; τ , y ) ∈ { ∣ v ∣ ≤ R 1 v } Z_b(t;\tau,y)\in\{|v|\le R_1^v\} Z b ( t ; τ , y ) ∈ { ∣ v ∣ ≤ R 1 v } . Step 3 of the proof of Lemma 3.2 then gives ∣ ∇ v Φ ( τ , y ) ∣ ≤ e 1 + Λ b k ( X b ( t ; τ , y ) ) |\nabla_v\Phi(\tau,y)|\le e^{1+\Lambda_b}k(X_b(t;\tau,y)) ∣ ∇ v Φ ( τ , y ) ∣ ≤ e 1 + Λ b k ( X b ( t ; τ , y )) on supp f U ( τ ) \operatorname{supp}f^U(\tau) supp f U ( τ ) . Let ϱ : = e d ∥ f ∗ ∥ ∞ ∣ B R 1 v ∣ \varrho:=e^d\|f_*\|_\infty|B_{R_1^v}| ϱ := e d ∥ f ∗ ∥ ∞ ∣ B R 1 v ∣ . Then ρ U : = ∫ f U d v ≤ ϱ \rho^U:=\int f^U\,dv\le\varrho ρ U := ∫ f U d v ≤ ϱ . The push-forward of the measure f U ( τ ) d y f^U(\tau)\,dy f U ( τ ) d y by Z b ( t ; τ , ⋅ ) Z_b(t;\tau,\cdot) Z b ( t ; τ , ⋅ ) has density e d ( t − τ ) f U ( τ , Z b ( τ ; t , ⋅ ) ) ≤ e d ∥ f ∗ ∥ ∞ e^{d(t-\tau)}f^U(\tau,Z_b(\tau;t,\cdot))\le e^d\|f_*\|_\infty e d ( t − τ ) f U ( τ , Z b ( τ ; t , ⋅ )) ≤ e d ∥ f ∗ ∥ ∞ and is supported in { ∣ v ∣ ≤ R 1 v } \{|v|\le R_1^v\} { ∣ v ∣ ≤ R 1 v } . By Hölder’s inequality in y y y ,
∬ f U ∣ U − b ∣ ( x ) k ( X b ( t ; τ , y ) ) d y ≤ ( ∬ f U ∣ U − b ∣ s ) 1 / s ( ∬ f U k ( X b ) s ′ ) 1 / s ′ ≤ ϱ 1 / s ∥ ( U − b ) ( τ ) ∥ L s ⋅ ϱ 1 / s ′ ∥ k ∥ L s ′ , \begin{aligned}
\iint f^U|U-b|(x)k(X_b(t;\tau,y))\,dy
\le\left(\iint f^U|U-b|^s\right)^{1/s}\left(\iint f^Uk(X_b)^{s'}\right)^{1/s'} \\
&\le\varrho^{1/s}\|(U-b)(\tau)\|_{L^s}\cdot\varrho^{1/s'}\|k\|_{L^{s'}},
\end{aligned} ∬ f U ∣ U − b ∣ ( x ) k ( X b ( t ; τ , y )) d y ≤ ( ∬ f U ∣ U − b ∣ s ) 1/ s ( ∬ f U k ( X b ) s ′ ) 1/ s ′ ≤ ϱ 1/ s ∥ ( U − b ) ( τ ) ∥ L s ⋅ ϱ 1/ s ′ ∥ k ∥ L s ′ ,
with the obvious changes for s ∈ { 1 , ∞ } s\in\{1,\infty\} s ∈ { 1 , ∞ } . This proves (a) for smooth fields.
Step 2: general fields. The chain rule along τ ↦ Z b ( t ; τ , Z U ( τ ; t 0 , ζ ) ) \tau\mapsto Z_b(t;\tau,Z_U(\tau;t_0,\zeta)) τ ↦ Z b ( t ; τ , Z U ( τ ; t 0 , ζ )) is not available directly for Carathéodory flows, so we pass to the limit in the inequality rather than in (6.1). Extend b , U b,U b , U by 0 0 0 to τ < t 0 \tau<t_0 τ < t 0 , and let b ε , U ε b_\varepsilon,U_\varepsilon b ε , U ε be their mollifications in x x x and, causally, in time (averages over ( τ − ε , τ ) ) (\tau-\varepsilon,\tau)) ( τ − ε , τ )) . Then ∫ t 0 1 ∥ ∇ b ε ∥ ∞ ≤ Λ b \int_{t_0}^1\|\nabla b_\varepsilon\|_\infty\le\Lambda_b ∫ t 0 1 ∥∇ b ε ∥ ∞ ≤ Λ b , ∥ b ε ∥ L 1 L ∞ ≤ ∥ b ∥ L 1 L ∞ \|b_\varepsilon\|_{L^1L^\infty}\le\|b\|_{L^1L^\infty} ∥ b ε ∥ L 1 L ∞ ≤ ∥ b ∥ L 1 L ∞ , the same holds for U ε U_\varepsilon U ε , and ∫ t 0 t ∥ U ε − b ε ∥ L s d τ ≤ ∫ t 0 t ∥ U − b ∥ L s d τ \int_{t_0}^t\|U_\varepsilon-b_\varepsilon\|_{L^s}\,d\tau\le\int_{t_0}^t\|U-b\|_{L^s}\,d\tau ∫ t 0 t ∥ U ε − b ε ∥ L s d τ ≤ ∫ t 0 t ∥ U − b ∥ L s d τ by Minkowski’s inequality. Hence Step 1 applies to ( b ε , U ε ) (b_\varepsilon,U_\varepsilon) ( b ε , U ε ) with constants R 1 , ε v ≤ R 1 v R_{1,\varepsilon}^v\le R_1^v R 1 , ε v ≤ R 1 v and K ε ≤ K K_\varepsilon\le K K ε ≤ K . Since b ( τ ) b(\tau) b ( τ ) is Lipschitz, b ε ∈ L 1 ( t 0 , 1 ; C ( T d ) ) b_\varepsilon\in L^1(t_0,1;C(\mathbb{T}^d)) b ε ∈ L 1 ( t 0 , 1 ; C ( T d )) as a Bochner function, so b ε → b b_\varepsilon\to b b ε → b in L 1 ( t 0 , 1 ; L ∞ ) L^1(t_0,1;L^\infty) L 1 ( t 0 , 1 ; L ∞ ) , and likewise for U U U . Since the phase-space fields are Lipschitz in z z z with constant 1 + ∥ ∇ b ∥ L ∞ 1+\|\nabla b\|_{L^\infty} 1 + ∥∇ b ∥ L ∞ , Grönwall’s inequality gives ∣ Z b ε ( t ; t 0 , ζ ) − Z b ( t ; t 0 , ζ ) ∣ ≤ e 1 + Λ b ∫ t 0 t ∥ b ε − b ∥ ∞ |Z_{b_\varepsilon}(t;t_0,\zeta)-Z_b(t;t_0,\zeta)|\le e^{1+\Lambda_b}\int_{t_0}^t\|b_\varepsilon-b\|_\infty ∣ Z b ε ( t ; t 0 , ζ ) − Z b ( t ; t 0 , ζ ) ∣ ≤ e 1 + Λ b ∫ t 0 t ∥ b ε − b ∥ ∞ , and likewise for U U U ; hence the flows from t 0 t_0 t 0 converge uniformly on supp f ∗ \operatorname{supp}f_* supp f ∗ . Therefore ∬ ( f U ε − f b ε ) ( t ) ψ = ∬ f ∗ ( ζ ) [ ψ ( Z U ε ( t ; t 0 , ζ ) ) − ψ ( Z b ε ( t ; t 0 , ζ ) ) ] d ζ \iint(f^{U_\varepsilon}-f^{b_\varepsilon})(t)\psi=\iint f_*(\zeta)[\psi(Z_{U_\varepsilon}(t;t_0,\zeta))-\psi(Z_{b_\varepsilon}(t;t_0,\zeta))]\,d\zeta ∬ ( f U ε − f b ε ) ( t ) ψ = ∬ f ∗ ( ζ ) [ ψ ( Z U ε ( t ; t 0 , ζ )) − ψ ( Z b ε ( t ; t 0 , ζ ))] d ζ converges to the corresponding expression for ( b , U ) (b,U) ( b , U ) by dominated convergence. This proves (a). Step 3: part (b). First let W W W be smooth. Take ψ = ( v − W ) ⋅ φ \psi=(v-W)\cdot\varphi ψ = ( v − W ) ⋅ φ with φ ∈ C 1 \varphi\in C^1 φ ∈ C 1 . Then ∇ v ψ = φ \nabla_v\psi=\varphi ∇ v ψ = φ and ∇ x ψ = ( ∇ φ ) T ( v − W ) − ( ∇ W ) T φ \nabla_x\psi=(\nabla\varphi)^{\mathsf T}(v-W)-(\nabla W)^{\mathsf T}\varphi ∇ x ψ = ( ∇ φ ) T ( v − W ) − ( ∇ W ) T φ , so on { ∣ v ∣ ≤ R 1 v } \{|v|\le R^v_1\} { ∣ v ∣ ≤ R 1 v } we may take k = ( 1 + R 1 v + ∥ W ∥ ∞ ) ( ∣ φ ∣ + ∣ ∇ φ ∣ ) + ∣ ∇ W ∣ ∣ φ ∣ k=(1+R^v_1+\|W\|_\infty)(|\varphi|+|\nabla\varphi|)+|\nabla W||\varphi| k = ( 1 + R 1 v + ∥ W ∥ ∞ ) ( ∣ φ ∣ + ∣∇ φ ∣ ) + ∣∇ W ∣∣ φ ∣ . Since q ′ < s ≤ q q'<s\le q q ′ < s ≤ q , we have s ′ < q s'<q s ′ < q ; define p ∗ p_* p ∗ by 1 p ∗ = 1 s ′ − 1 q ∈ ( 0 , 1 ) \frac{1}{p_*}=\frac{1}{s'}-\frac{1}{q}\in(0,1) p ∗ 1 = s ′ 1 − q 1 ∈ ( 0 , 1 ) . As q > d q>d q > d , 1 p ∗ > 1 s ′ − 1 d \frac{1}{p_*}>\frac{1}{s'}-\frac{1}{d} p ∗ 1 > s ′ 1 − d 1 , so W 1 , s ′ ↪ L p ∗ W^{1,s'}\hookrightarrow L^{p_*} W 1 , s ′ ↪ L p ∗ . Hence, using W 1 , q ↪ L ∞ W^{1,q}\hookrightarrow L^\infty W 1 , q ↪ L ∞ ,
∥ k ∥ L s ′ ≤ ( 1 + R 1 v + ∥ W ∥ ∞ ) ∥ φ ∥ W 1 , s ′ + ∥ ∇ W ∥ L q ∥ φ ∥ L p ∗ ≤ C s , q ( 1 + R 1 v + ∥ W ∥ W 1 , q ) ∥ φ ∥ W 1 , s ′ . \|k\|_{L^{s'}}\le(1+R^v_1+\|W\|_\infty)\|\varphi\|_{W^{1,s'}}+\|\nabla W\|_{L^q}\|\varphi\|_{L^{p_*}}\le C_{s,q}(1+R^v_1+\|W\|_{W^{1,q}})\|\varphi\|_{W^{1,s'}}. ∥ k ∥ L s ′ ≤ ( 1 + R 1 v + ∥ W ∥ ∞ ) ∥ φ ∥ W 1 , s ′ + ∥∇ W ∥ L q ∥ φ ∥ L p ∗ ≤ C s , q ( 1 + R 1 v + ∥ W ∥ W 1 , q ) ∥ φ ∥ W 1 , s ′ .
Apply (a). For general W ∈ W 1 , q W\in W^{1,q} W ∈ W 1 , q , take smooth W n → W W_n\to W W n → W in W 1 , q W^{1,q} W 1 , q , hence uniformly; since f U − f b f^U-f^b f U − f b is bounded with compact velocity support, ∫ h n ⋅ φ → ∫ h ⋅ φ \int h_n\cdot\varphi\to\int h\cdot\varphi ∫ h n ⋅ φ → ∫ h ⋅ φ . Conclude by density, since H − 1 , s = ( W 1 , s ′ ) ∗ H^{-1,s}=(W^{1,s'})^* H − 1 , s = ( W 1 , s ′ ) ∗ .
Remark 6.7 (Starting at t 0 > 0 t_0>0 t 0 > 0 ; the full drag difference). Lemmas 3.2 and 3.8 and Proposition 3.4 hold when started at a time t 0 > 0 t_0>0 t 0 > 0 from a common kinetic state f ∗ f_* f ∗ . More precisely, the statements hold on [ t 0 , 1 ] [t_0,1] [ t 0 , 1 ] , which has length at most one, with the following replacements: f 0 f_0 f 0 by f ∗ f_* f ∗ ; M 0 M_0 M 0 by ∥ f ∗ ∥ L 1 = M 0 \|f_*\|_{L^1}=M_0 ∥ f ∗ ∥ L 1 = M 0 ; M ∞ M_\infty M ∞ by ∥ f ∗ ∥ ∞ ≤ e d M ∞ \|f_*\|_\infty\le e^dM_\infty ∥ f ∗ ∥ ∞ ≤ e d M ∞ ; R 0 v R^v_0 R 0 v by the radius R 0 v + ∫ 0 t 0 ∥ U ∥ L ∞ R^v_0+\int_0^{t_0}\|U\|_{L^\infty} R 0 v + ∫ 0 t 0 ∥ U ∥ L ∞ of the velocity support of f ∗ f_* f ∗ ; the integrals ∫ 0 t \int_0^t ∫ 0 t and the norms L t 1 L ∞ L^1_tL^\infty L t 1 L ∞ , L t 1 W 1 , ∞ L^1_tW^{1,\infty} L t 1 W 1 , ∞ by the same quantities on [ t 0 , t ] [t_0,t] [ t 0 , t ] and [ t 0 , 1 ] [t_0,1] [ t 0 , 1 ] ; and, in Lemma 3.8, (17) by c ( t ) = ∫ U ( t 0 ) − U ( t ) + J ( t 0 ) − J [ U + c ] ( t ) c(t)=\int U(t_0)-U(t)+J(t_0)-J[U+c](t) c ( t ) = ∫ U ( t 0 ) − U ( t ) + J ( t 0 ) − J [ U + c ] ( t ) for t ≥ t 0 t\ge t_0 t ≥ t 0 , with c ( t 0 ) = 0 c(t_0)=0 c ( t 0 ) = 0 , where J ( t 0 ) J(t_0) J ( t 0 ) is the common momentum of the particles at t 0 t_0 t 0 . In this section f ∗ = f [ U ] ( t 0 ) f_*=f[U](t_0) f ∗ = f [ U ] ( t 0 ) for a smooth field U U U and smooth f 0 f_0 f 0 , so f ∗ f_* f ∗ is smooth and the proofs apply verbatim after a translation in time. For a merely bounded f ∗ f_* f ∗ the Eulerian computation in Step 2 of the proof of Lemma 3.2 is not available as written; one uses instead the Lagrangian identity (30) , which only involves the flows, or an approximation of f ∗ f_* f ∗ in L 1 L^1 L 1 by smooth functions.
For two fields U , b U,b U , b as in Lemma 6.5, write a : = U − b a:=U-b a := U − b and
h : = F [ f U , U ] − F [ f b , b ] = ∫ ( v − b ) ( f U − f b ) d v − ρ U a . h:=F[f^U,U]-F[f^b,b]=\int(v-b)(f^U-f^b)\,dv-\rho^U a. h := F [ f U , U ] − F [ f b , b ] = ∫ ( v − b ) ( f U − f b ) d v − ρ U a .
The first term is a history term , controlled by Lemma 6.5(b). The second is instantaneous : it is of the size of a ( t ) a(t) a ( t ) itself, and is not controlled by the past of a a a . By Lemma 6.4(e), for d < q < ∞ d<q<\infty d < q < ∞ ,
∥ R h ( t ) ∥ L q ≤ C q K ( 1 + R 1 v + ∥ b ( t ) ∥ W 1 , q ) ∫ t 0 t ∥ a ∥ L q d τ + C q ρ ∥ a ( t ) ∥ L q , ρ = e d ∥ f ∗ ∥ ∞ ∣ B R 1 v ∣ . (31) \|\mathcal{R}h(t)\|_{L^q}\le C_qK(1+R^v_1+\|b(t)\|_{W^{1,q}})\int_{t_0}^t\|a\|_{L^q}\,d\tau+C_q\rho\|a(t)\|_{L^q},\qquad\rho=e^d\|f_*\|_\infty|B_{R^v_1}|. \tag*{(31)} ∥ R h ( t ) ∥ L q ≤ C q K ( 1 + R 1 v + ∥ b ( t ) ∥ W 1 , q ) ∫ t 0 t ∥ a ∥ L q d τ + C q ρ ∥ a ( t ) ∥ L q , ρ = e d ∥ f ∗ ∥ ∞ ∣ B R 1 v ∣. (31)
This replaces the Hölder bound of Proposition 3.4. In the applications below, the instantaneous term is controlled for two reasons. On the intervals where a a a is spatially constant, it is bounded by the momentum defect, which is a history term. Where a a a contains a convex integration perturbation w w w , the term R ( ρ U w ) \mathcal{R}(\rho^U w) R ( ρ U w ) is small in L t 1 L q L^1_tL^q L t 1 L q , since ∥ w ∥ L t 1 L ∞ \|w\|_{L^1_tL^\infty} ∥ w ∥ L t 1 L ∞ is small, and it only enters the new stress on the support of w w w .
Segment correctors and mean-preserving gluing
Let ( U , P , S ) (U,P,S) ( U , P , S ) be a smooth VNS–Reynolds solution on [ 0 , 1 ] [0,1] [ 0 , 1 ] , called the base , and let a ∈ [ 0 , 1 ] a\in[0,1] a ∈ [ 0 , 1 ] be the time at which a segment (or, in Lemma 6.13, a convex integration zone) starts. The size of the base is measured after a a a :
B U : = sup t ∈ [ a , 1 ] ∥ U ( t ) ∥ W 1 , q , A U : = ∫ a 1 ∥ ∇ U ∥ L ∞ d t , L U : = ∫ 0 a ∥ U ∥ L ∞ d t . (32) B_U:=\sup_{t\in[a,1]}\|U(t)\|_{W^{1,q}},\qquad A_U:=\int_a^1\|\nabla U\|_{L^\infty}\,dt,\qquad L_U:=\int_0^a\|U\|_{L^\infty}\,dt. \tag*{(32)} B U := t ∈ [ a , 1 ] sup ∥ U ( t ) ∥ W 1 , q , A U := ∫ a 1 ∥∇ U ∥ L ∞ d t , L U := ∫ 0 a ∥ U ∥ L ∞ d t . (32)
Note that ∥ U ∥ L ∞ ( ( a , 1 ) × T d ) ≤ C E B U \|U\|_{L^\infty((a,1)\times\mathbb{T}^d)}\le C_EB_U ∥ U ∥ L ∞ (( a , 1 ) × T d ) ≤ C E B U by Lemma 6.4(b). The Lipschitz norm of U U U enters only through A U A_U A U , an L 1 L^1 L 1 norm in time. Only the base on [ a , 1 ] [a,1] [ a , 1 ] enters the kinetic comparisons, since they are started at a a a from the common kinetic state f [ U ] ( a ) f[U](a) f [ U ] ( a ) , whose velocity support has radius at most R 0 v + L U R^v_0+L_U R 0 v + L U by Lemma 3.1(c). This is used in the second pass of Section 6.6, where the base contains earlier convex integration perturbations w i w_i w i , which are small in L t 1 W 1 , q L^1_tW^{1,q} L t 1 W 1 , q but whose sup t ∥ w i ( t ) ∥ W 1 , q \sup_t\|w_i(t)\|_{W^{1,q}} sup t ∥ w i ( t ) ∥ W 1 , q depends on their frequency; they vanish on [ a , 1 ] [a,1] [ a , 1 ] .
A segment is an interval Σ = [ a , b ] ⊂ [ 0 , 1 ] \Sigma=[a,b]\subset[0,1] Σ = [ a , b ] ⊂ [ 0 , 1 ] with a cutoff χ ∈ C ∞ ( R ; [ 0 , 1 ] ) \chi\in C^\infty(\mathbb{R};[0,1]) χ ∈ C ∞ ( R ; [ 0 , 1 ]) that vanishes near a a a and either vanishes near b b b (two-sided ) or equals 1 1 1 near b = 1 b=1 b = 1 (one-sided ). For a field z z z on Σ \Sigma Σ , write c z : = ∫ z c_z:=\int z c z := ∫ z , z ~ : = z − c z \tilde z:=z-c_z z ~ := z − c z , and define the glued field and its kinetic state
U ^ z : = U + χ z ~ + c z on Σ , f ^ z : = f [ U ∣ [ 0 , a ) ∪ U ^ z ] . \widehat U_z:=U+\chi\tilde z+c_z\quad\text{on }\Sigma,\qquad\widehat f_z:=f[U|_{[0,a)}\cup\widehat U_z]. U z := U + χ z ~ + c z on Σ , f z := f [ U ∣ [ 0 , a ) ∪ U z ] .
The cutoff acts only on the mean-zero part. The segment corrector is the solution of
∂ t z − Δ z + Q div ( U ⊗ z + z ⊗ U + z ⊗ z ) = Q [ − div S + g z ] , z ( a ) = 0 , g z : = F [ f ^ z , U ^ z ] − F [ f [ U ] , U ] . (33) \partial_t z-\Delta z+\mathbb{Q}\operatorname{div}(U\otimes z+z\otimes U+z\otimes z)=\mathbb{Q}[-\operatorname{div}S+g_z],\qquad z(a)=0,\qquad g_z:=F[\widehat f_z,\widehat U_z]-F[f[U],U]. \tag*{(33)} ∂ t z − Δ z + Q div ( U ⊗ z + z ⊗ U + z ⊗ z ) = Q [ − div S + g z ] , z ( a ) = 0 , g z := F [ f z , U z ] − F [ f [ U ] , U ] . (33)
Where χ = 1 \chi=1 χ = 1 we have U ^ z = U + z \widehat U_z=U+z U z = U + z , and (6.4) says that ( U + z , f ^ z ) (U+z,\widehat f_z) ( U + z , f z ) solves VNS exactly.
Put ς : = ∥ S ∥ L ∞ ( Σ ; L q ) \varsigma:=\lVert S\rVert_{L^\infty(\Sigma;L^q)} ς := ∥ S ∥ L ∞ ( Σ ; L q ) and N S : = ∥ S ∥ L ∞ ( Σ ; W 1 , q ) N_S:=\lVert S\rVert_{L^\infty(\Sigma;W^{1,q})} N S := ∥ S ∥ L ∞ ( Σ ; W 1 , q ) . Let M ∞ = ∥ f 0 ∥ ∞ M_\infty=\lVert f_0\rVert_\infty M ∞ = ∥ f 0 ∥ ∞ , let R 0 v R_0^v R 0 v bound the velocity support of f 0 f_0 f 0 , and set
R 1 v : = R 0 v + L U + 2 C E B U + 8 , ϱ : = e d M ∞ ∣ B R 1 v ∣ , C h : = 2 ( R 1 v + C E B U ) ϱ . R_1^v:=R_0^v+L_U+2C_EB_U+8,\qquad\varrho:=e^{dM_\infty}|B_{R_1^v}|,\qquad C_h:=2(R_1^v+C_EB_U)\varrho. R 1 v := R 0 v + L U + 2 C E B U + 8 , ϱ := e d M ∞ ∣ B R 1 v ∣ , C h := 2 ( R 1 v + C E B U ) ϱ .
The term L U L_U L U accounts for the velocity that the particles gain before the segment starts. Every glued field U ^ \widehat U U considered below satisfies ∥ U ^ − U ∥ L 1 ( a , 1 ; L ∞ ) ≤ 4 \lVert\widehat U-U\rVert_{L^1(a,1;L^\infty)}\leq4 ∥ U − U ∥ L 1 ( a , 1 ; L ∞ ) ≤ 4 : the contribution of Σ \Sigma Σ is at most 3 3 3 , by (6.7) in (a) and by (6.10) in (b), and after an exit by the mean the contribution of [ b , 1 ] [b,1] [ b , 1 ] is at most 1 1 1 , by the continuity argument in the proof of (c). Hence ∥ U ^ ∥ L 1 ( a , 1 ; L ∞ ) ≤ C E B U + 4 \lVert\widehat U\rVert_{L^1(a,1;L^\infty)}\leq C_EB_U+4 ∥ U ∥ L 1 ( a , 1 ; L ∞ ) ≤ C E B U + 4 . By Lemma 3.1(c), all kinetic states are supported in { ∣ v ∣ ≤ R 0 v + L U + C E B U + 4 } ⊂ { ∣ v ∣ ≤ R 1 v } \{|v|\leq R_0^v+L_U+C_EB_U+4\}\subset\{|v|\leq R_1^v\} { ∣ v ∣ ≤ R 0 v + L U + C E B U + 4 } ⊂ { ∣ v ∣ ≤ R 1 v } , and the radius of Lemma 6.5, applied with two such fields (or with one such field and U U U ) in the roles of U U U and b b b , is at most R 0 v + L U + 2 ( C E B U + 4 ) = R 1 v R_0^v+L_U+2(C_EB_U+4)=R_1^v R 0 v + L U + 2 ( C E B U + 4 ) = R 1 v .
Proposition 6.8 (Segment correctors). In this proposition, the constants C C C and K K K depend only on d , q , r , β , R 0 v , M ∞ d,q,r,\beta,R_0^v,M_\infty d , q , r , β , R 0 v , M ∞ and on upper bounds for B U B_U B U , Λ U \Lambda_U Λ U and L U L_U L U (see (6.3)). Primed constants, C ′ C' C ′ and ℓ 0 ′ \ell_0' ℓ 0 ′ and c ∗ ′ c_*' c ∗ ′ , depend in addition on an upper bound for N S N_S N S .
(a) (Short segments.) If ∣ Σ ∣ ≤ ℓ 0 ′ |\Sigma|\leq\ell_0' ∣Σ∣ ≤ ℓ 0 ′ , then (6.4) has a solution z ∈ Y 0 ( Σ ) z\in\mathcal{Y}_0(\Sigma) z ∈ Y 0 ( Σ ) , unique among solutions with ∥ z ∥ Y 0 ( Σ ) ≤ 1 \lVert z\rVert_{\mathcal{Y}_0(\Sigma)}\leq1 ∥ z ∥ Y 0 ( Σ ) ≤ 1 , and for t ∈ Σ t\in\Sigma t ∈ Σ :
(i) ∥ z ∥ Y 0 ( Σ ) + sup Σ ∥ z ∥ X q ≤ C ′ ∣ Σ ∣ 1 / 2 \lVert z\rVert_{\mathcal{Y}_0(\Sigma)}+\sup_\Sigma\lVert z\rVert_{X_q}\leq C'|\Sigma|^{1/2} ∥ z ∥ Y 0 ( Σ ) + sup Σ ∥ z ∥ X q ≤ C ′ ∣Σ ∣ 1/2 ;
(ii) ∥ z ∥ Y 1 ( Σ ) + sup Σ ∥ z ∥ L q ≤ C ∣ Σ ∣ 1 / 2 ς \lVert z\rVert_{\mathcal{Y}_1(\Sigma)}+\sup_\Sigma\lVert z\rVert_{L^q}\leq C|\Sigma|^{1/2}\varsigma ∥ z ∥ Y 1 ( Σ ) + sup Σ ∥ z ∥ L q ≤ C ∣Σ ∣ 1/2 ς ;
(iii) ∥ z ∥ L 2 ( Σ ; W 1 , ∞ ) + sup Σ ∥ z ∥ L ∞ ≤ C ′ ∣ Σ ∣ 1 / 2 ς β \lVert z\rVert_{L^2(\Sigma;W^{1,\infty})}+\sup_\Sigma\lVert z\rVert_{L^\infty}\leq C'|\Sigma|^{1/2}\varsigma^\beta ∥ z ∥ L 2 ( Σ ; W 1 , ∞ ) + sup Σ ∥ z ∥ L ∞ ≤ C ′ ∣Σ ∣ 1/2 ς β ;
(iv) ∥ R z ( t ) ∥ L r ≤ C r ∫ a t ∥ S ∥ L r + C ′ ∣ Σ ∣ 3 / 2 ς \lVert Rz(t)\rVert_{L^r}\leq C_r\int_a^t\lVert S\rVert_{L^r}+C'|\Sigma|^{3/2}\varsigma ∥ R z ( t ) ∥ L r ≤ C r ∫ a t ∥ S ∥ L r + C ′ ∣Σ ∣ 3/2 ς .
(b) (Long segments.) There are ε ∗ ′ > 0 \varepsilon_*'>0 ε ∗ ′ > 0 and c ∗ ′ ∈ ( 0 , ε ∗ ′ ] c_*'\in(0,\varepsilon_*'] c ∗ ′ ∈ ( 0 , ε ∗ ′ ] such that, if ς ≤ c ∗ ′ \varsigma\leq c_*' ς ≤ c ∗ ′ , then (6.4) has a solution z ∈ Y 0 ( Σ ) z\in\mathcal{Y}_0(\Sigma) z ∈ Y 0 ( Σ ) , unique among solutions with ∥ z ∥ Y 1 ( Σ ) ≤ ε ∗ ′ \lVert z\rVert_{\mathcal{Y}_1(\Sigma)}\leq\varepsilon_*' ∥ z ∥ Y 1 ( Σ ) ≤ ε ∗ ′ and ∥ z ∥ L 1 ( Σ ; W 1 , ∞ ) ≤ 1 \lVert z\rVert_{L^1(\Sigma;W^{1,\infty})}\leq1 ∥ z ∥ L 1 ( Σ ; W 1 , ∞ ) ≤ 1 , and (i)–(iv) hold with ∣ Σ ∣ 1 / 2 |\Sigma|^{1/2} ∣Σ ∣ 1/2 and ∣ Σ ∣ 3 / 2 |\Sigma|^{3/2} ∣Σ ∣ 3/2 replaced by 1 1 1 ; the constants, ε ∗ ′ \varepsilon_*' ε ∗ ′ and c ∗ ′ c_*' c ∗ ′ do not depend on ∣ Σ ∣ ≤ 1 |\Sigma|\leq1 ∣Σ∣ ≤ 1 .
(c) (Momentum and mean.) In both cases, c z ′ = ∫ g z c'_z=\int g_z c z ′ = ∫ g z and c z = J [ f [ U ] ] − J [ f ^ z ] c_z=J[f[U]]-J[\widehat f_z] c z = J [ f [ U ]] − J [ f z ] on Σ \Sigma Σ . Write a z : = U ^ z − U = χ z ˚ + c z a_z:=\widehat U_z-U=\chi\mathring z+c_z a z := U z − U = χ z ˚ + c z . For t ∈ Σ t\in\Sigma t ∈ Σ ,
∣ c z ( t ) ∣ ≤ K ∫ a t ∥ a z ∥ L q , ∥ R g z ( t ) ∥ L q ≤ K ( ∫ a t ∥ a z ∥ L q + ∥ a z ( t ) ∥ L q ) , (34) |c_z(t)|\leq K\int_a^t\lVert a_z\rVert_{L^q},\qquad\lVert Rg_z(t)\rVert_{L^q}\leq K\left(\int_a^t\lVert a_z\rVert_{L^q}+\lVert a_z(t)\rVert_{L^q}\right), \tag*{(34)} ∣ c z ( t ) ∣ ≤ K ∫ a t ∥ a z ∥ L q , ∥ R g z ( t ) ∥ L q ≤ K ( ∫ a t ∥ a z ∥ L q + ∥ a z ( t ) ∥ L q ) , (34) so that sup Σ ( ∣ c z ∣ + ∥ R g z ∥ L q ) ≤ 9 K sup Σ ∥ z ∥ L q \sup_\Sigma(|c_z|+\lVert Rg_z\rVert_{L^q})\leq9K\sup_\Sigma\lVert z\rVert_{L^q} sup Σ ( ∣ c z ∣ + ∥ R g z ∥ L q ) ≤ 9 K sup Σ ∥ z ∥ L q . If χ \chi χ is two-sided and the glued field exits by the mean (Proposition 6.9), then a z = c z a_z=c_z a z = c z on [ b , 1 ] [b,1] [ b , 1 ] , (6.5) holds for t ∈ [ b , 1 ] t\in[b,1] t ∈ [ b , 1 ] , and
sup [ b , 1 ] ( ∣ c z ∣ + ∥ R g z ∥ L q ) ≤ C ∣ Σ ∣ sup Σ ∥ z ∥ L q . (35) \sup_{[b,1]}(|c_z|+\lVert Rg_z\rVert_{L^q})\leq C|\Sigma|\sup_\Sigma\lVert z\rVert_{L^q}. \tag*{(35)} [ b , 1 ] sup ( ∣ c z ∣ + ∥ R g z ∥ L q ) ≤ C ∣Σ∣ Σ sup ∥ z ∥ L q . (35) (d) (Persistent stress.) In the situation of (6.6), the stress Π z : = U ⊗ ˚ c z + c z ⊗ ˚ U − R g z \Pi_z:=U\mathring\otimes c_z+c_z\mathring\otimes U-Rg_z Π z := U ⊗ ˚ c z + c z ⊗ ˚ U − R g z satisfies
sup [ b , 1 ] ∥ Π z ∥ L q ≤ C ∣ Σ ∣ sup Σ ∥ z ∥ L q , sup [ b , 1 ] ∥ Π z ∥ W 1 , q ≤ C . \sup_{[b,1]}\lVert\Pi_z\rVert_{L^q}\leq C|\Sigma|\sup_\Sigma\lVert z\rVert_{L^q},\qquad\sup_{[b,1]}\lVert\Pi_z\rVert_{W^{1,q}}\leq C. [ b , 1 ] sup ∥ Π z ∥ L q ≤ C ∣Σ∣ Σ sup ∥ z ∥ L q , [ b , 1 ] sup ∥ Π z ∥ W 1 , q ≤ C . In both cases z z z is smooth on [ a , b ] × T d [a,b]\times\mathbb{T}^d [ a , b ] × T d , with one-sided derivatives at t = a t=a t = a and t = b t=b t = b .
Estimate (ii) gains a factor ς \varsigma ς in the weaker norm Y 1 \mathcal{Y}_1 Y 1 ; the factor ∣ Σ ∣ 1 / 2 |\Sigma|^{1/2} ∣Σ ∣ 1/2 in (i)–(iv) makes the bounds summable over many short segments. The persistent stress in (d) is bounded in W 1 , q W^{1,q} W 1 , q by a constant because R R R gains one derivative; no derivative of the kinetic state is needed.
Proof. Kinetic bounds. For ζ ∈ Y 0 ( Σ ) \zeta\in\mathcal{Y}_0(\Sigma) ζ ∈ Y 0 ( Σ ) with ζ ( a ) = 0 \zeta(a)=0 ζ ( a ) = 0 and ∥ ζ ∥ Y 0 ≤ 1 \lVert\zeta\rVert_{\mathcal{Y}_0}\leq1 ∥ ζ ∥ Y 0 ≤ 1 , the glued field U ^ ζ \widehat U_\zeta U ζ is Lipschitz in x x x with an integrable constant, since Y 0 ↪ L 2 W 1 , ∞ \mathcal{Y}_0\hookrightarrow L^2W^{1,\infty} Y 0 ↪ L 2 W 1 , ∞ . Its kinetic state is defined by the Carathéodory flow, as in Proposition 5.10(c). We always take ℓ 0 ′ ≤ C E − 2 \ell_0'\leq C_E^{-2} ℓ 0 ′ ≤ C E − 2 , so that for ∣ Σ ∣ ≤ ℓ 0 ′ |\Sigma|\leq\ell_0' ∣Σ∣ ≤ ℓ 0 ′ ,
∥ ζ ∥ L 1 ( Σ ; W 1 , ∞ ) ≤ ∣ Σ ∣ 1 / 2 ∥ ζ ∥ L 2 W 1 , ∞ ≤ C E ∣ Σ ∣ 1 / 2 ∥ ζ ∥ Y 0 ≤ 1. (36) \lVert\zeta\rVert_{L^1(\Sigma;W^{1,\infty})}\leq|\Sigma|^{1/2}\lVert\zeta\rVert_{L^2W^{1,\infty}}\leq C_E|\Sigma|^{1/2}\lVert\zeta\rVert_{\mathcal{Y}_0}\leq1. \tag*{(36)} ∥ ζ ∥ L 1 ( Σ ; W 1 , ∞ ) ≤ ∣Σ ∣ 1/2 ∥ ζ ∥ L 2 W 1 , ∞ ≤ C E ∣Σ ∣ 1/2 ∥ ζ ∥ Y 0 ≤ 1. (36) In (b) the bound ∥ ζ ∥ L 1 W 1 , ∞ ≤ 1 \lVert\zeta\rVert_{L^1W^{1,\infty}}\leq1 ∥ ζ ∥ L 1 W 1 , ∞ ≤ 1 is part of the admissibility condition (6.10) below and is verified there. Hence ∥ χ ζ ˚ + c ζ ∥ L 1 L ∞ ≤ 3 ∥ ζ ∥ L 1 L ∞ ≤ 3 \lVert\chi\mathring\zeta+c_\zeta\rVert_{L^1L^\infty}\leq3\lVert\zeta\rVert_{L^1L^\infty}\leq3 ∥ χ ζ ˚ + c ζ ∥ L 1 L ∞ ≤ 3 ∥ ζ ∥ L 1 L ∞ ≤ 3 , so by Lemma 3.1(c), f ^ ζ ≤ e d M ∞ \widehat f_\zeta\leq e^{dM_\infty} f ζ ≤ e d M ∞ is supported in { ∣ v ∣ ≤ R 1 v } \{|v|\leq R_1^v\} { ∣ v ∣ ≤ R 1 v } and ρ f ^ ζ ≤ ϱ \rho_{\widehat f_\zeta}\leq\varrho ρ f ζ ≤ ϱ . Write
g ζ = h ζ − ρ f ^ ζ a ζ , h ζ : = ∫ ( v − U ) ( f ^ ζ − f [ U ] ) d v , a ζ : = χ ζ ˚ + c ζ . g_\zeta=h_\zeta-\rho_{\widehat f_\zeta}a_\zeta,\qquad h_\zeta:=\int(v-U)(\widehat f_\zeta-f[U])\,dv,\qquad a_\zeta:=\chi\mathring\zeta+c_\zeta. g ζ = h ζ − ρ f ζ a ζ , h ζ := ∫ ( v − U ) ( f ζ − f [ U ]) d v , a ζ := χ ζ ˚ + c ζ .
Then ∣ h ζ ∣ ≤ C h |h_\zeta| \le C_h ∣ h ζ ∣ ≤ C h and ∥ a ζ ( t ) ∥ L s ≤ 3 ∥ ζ ( t ) ∥ L s \|a_\zeta(t)\|_{L^s} \le3\|\zeta(t)\|_{L^s} ∥ a ζ ( t ) ∥ L s ≤ 3∥ ζ ( t ) ∥ L s for every s s s . The kinetic states agree at a a a . Lemma 6.5(b), with t 0 = a t_0=a t 0 = a , base U U U , W = U ( t ) W=U(t) W = U ( t ) and s = q s=q s = q , gives
∥ h ζ ( t ) ∥ H − 1 , q ≤ K 1 ∫ a t ∥ a ζ ∥ L q ≤ 3 K 1 ∫ a t ∥ ζ ∥ L q . (37) \|h_\zeta(t)\|_{H^{-1,q}} \le K_1 \int_a^t \|a_\zeta\|_{L^q} \le3K_1 \int_a^t \|\zeta\|_{L^q}. \tag*{(37)} ∥ h ζ ( t ) ∥ H − 1 , q ≤ K 1 ∫ a t ∥ a ζ ∥ L q ≤ 3 K 1 ∫ a t ∥ ζ ∥ L q . (37)
Here K 1 K_1 K 1 depends on B U B_U B U through R ν 1 R^1_\nu R ν 1 and ∥ U ( t ) ∥ W 1 , q \|U(t)\|_{W^{1,q}} ∥ U ( t ) ∥ W 1 , q , and on Λ U \Lambda_U Λ U . For two candidates ζ , ζ ′ \zeta,\zeta' ζ , ζ ′ that coincide up to a time s 0 s_0 s 0 ,
g ζ − g ζ ′ = ∫ ( v − U ^ ζ ′ ) ( f ^ ζ − f ^ ζ ′ ) d v − ρ f ^ ( a ζ − a ζ ′ ) . g_\zeta-g_{\zeta'}=\int(v-\widehat{U}_{\zeta'})(\widehat{f}_\zeta-\widehat{f}_{\zeta'})\,\mathrm{d}v-\rho_{\hat f}(a_\zeta-a_{\zeta'}). g ζ − g ζ ′ = ∫ ( v − U ζ ′ ) ( f ζ − f ζ ′ ) d v − ρ f ^ ( a ζ − a ζ ′ ) .
Lemma 6.5(b), with t 0 = s 0 t_0=s_0 t 0 = s 0 , base U ^ ζ ′ \widehat{U}_{\zeta'} U ζ ′ , whose Lipschitz norm is at most Λ U + 3 \Lambda_U+3 Λ U + 3 in L t 1 L^1_t L t 1 by (6.7), gives
∥ ( g ζ − g ζ ′ ) ( t ) ∥ H − 1 , q ≤ K 2 ( 1 + ∥ ζ ′ ( t ) ∥ W 1 , q ) ∫ s 0 t ∥ ζ − ζ ′ ∥ L q + 3 ϱ ∥ ( ζ − ζ ′ ) ( t ) ∥ L q . (38) \|(g_\zeta-g_{\zeta'})(t)\|_{H^{-1,q}} \le K_2\left(1+\|\zeta'(t)\|_{W^{1,q}}\right)\int_{s_0}^t\|\zeta-\zeta'\|_{L^q}+3\varrho\|(\zeta-\zeta')(t)\|_{L^q}. \tag*{(38)} ∥ ( g ζ − g ζ ′ ) ( t ) ∥ H − 1 , q ≤ K 2 ( 1 + ∥ ζ ′ ( t ) ∥ W 1 , q ) ∫ s 0 t ∥ ζ − ζ ′ ∥ L q + 3 ϱ ∥ ( ζ − ζ ′ ) ( t ) ∥ L q . (38)
The map. For an interval I = [ s 0 , s 1 ] ⊂ Σ I=[s_0,s_1]\subset\Sigma I = [ s 0 , s 1 ] ⊂ Σ and a given solution on [ a , s 0 ] [a,s_0] [ a , s 0 ] , let Θ ζ \Theta\zeta Θ ζ solve, on I I I , ∂ t ϑ − Δ ϑ = Q [ − div S + g ζ − div ( U ⊗ ζ + ζ ⊗ U + ζ ⊗ ζ ) ] \partial_t\vartheta-\Delta\vartheta=Q[-\operatorname{div}S+g_\zeta-\operatorname{div}(U\otimes\zeta+\zeta\otimes U+\zeta\otimes\zeta)] ∂ t ϑ − Δ ϑ = Q [ − div S + g ζ − div ( U ⊗ ζ + ζ ⊗ U + ζ ⊗ ζ )] , with the given value at s 0 s_0 s 0 . The forcing lies in L 2 ( I ; L q ) L^2(I;L^q) L 2 ( I ; L q ) , by the bound ∣ h ∣ ≤ C h |h|\le C_h ∣ h ∣ ≤ C h . Let C T C_T C T and C E C_E C E denote the trace and embedding constants of Lemma 6.4. We use repeatedly the trace estimates of Lemmas 5.4(b) and 6.4(a) on I = [ s 0 , s 1 ] I=[s_0,s_1] I = [ s 0 , s 1 ] ,
sup I ∥ ζ ∥ X q ≤ C T ( ∥ ζ ( s 0 ) ∥ X q + ∥ ζ ∥ Y 0 ( I ) ) , sup I ∥ ζ ∥ B q , 2 0 ≤ C T ( ∥ ζ ( s 0 ) ∥ B q , 2 0 + ∥ ζ ∥ Y 1 ( I ) ) , \sup_I\|\zeta\|_{X_q}\le C_T\left(\|\zeta(s_0)\|_{X_q}+\|\zeta\|_{Y_0(I)}\right),\qquad
\sup_I\|\zeta\|_{B^0_{q,2}}\le C_T\left(\|\zeta(s_0)\|_{B^0_{q,2}}+\|\zeta\|_{Y_1(I)}\right), I sup ∥ ζ ∥ X q ≤ C T ( ∥ ζ ( s 0 ) ∥ X q + ∥ ζ ∥ Y 0 ( I ) ) , I sup ∥ ζ ∥ B q , 2 0 ≤ C T ( ∥ ζ ( s 0 ) ∥ B q , 2 0 + ∥ ζ ∥ Y 1 ( I ) ) ,
and ∥ ⋅ ∥ L q ≤ C ∥ ⋅ ∥ B q , 2 0 \|\cdot\|_{L^q}\le C\|\cdot\|_{B^0_{q,2}} ∥ ⋅ ∥ L q ≤ C ∥ ⋅ ∥ B q , 2 0 , and
∥ div ( U ⊗ ζ + ζ ⊗ U ) ∥ L q ≤ C E B U ∥ ζ ∥ W 1 , q + ∥ ∇ U ∥ L q ∥ ζ ∥ L ∞ ≤ 2 C E B U ∥ ζ ∥ X q . \|\operatorname{div}(U\otimes\zeta+\zeta\otimes U)\|_{L^q}\le C_EB_U\|\zeta\|_{W^{1,q}}+\|\nabla U\|_{L^q}\|\zeta\|_{L^\infty}\le2C_EB_U\|\zeta\|_{X_q}. ∥ div ( U ⊗ ζ + ζ ⊗ U ) ∥ L q ≤ C E B U ∥ ζ ∥ W 1 , q + ∥∇ U ∥ L q ∥ ζ ∥ L ∞ ≤ 2 C E B U ∥ ζ ∥ X q .
The initial traces cannot be dropped when s 0 > a s_0>a s 0 > a , even though ζ ( a ) = 0 \zeta(a)=0 ζ ( a ) = 0 . They vanish in case (a), where I = Σ I=\Sigma I = Σ and s 0 = a s_0=a s 0 = a , and for the difference of two candidates with the same value at s 0 s_0 s 0 ; in case (b) they are bounded by the a priori bounds (6.11) and (6.13) at $s_0.
(a) Short segments. Take I = Σ I=\Sigma I = Σ and E : = { ζ ∈ Y 0 ( Σ ) : ζ ( a ) = 0 , ∥ ζ ∥ Y 0 ≤ A ′ ∣ Σ ∣ 1 / 2 } E:=\{\zeta\in Y_0(\Sigma):\zeta(a)=0,\ \|\zeta\|_{Y_0}\le A'|\Sigma|^{1/2}\} E := { ζ ∈ Y 0 ( Σ ) : ζ ( a ) = 0 , ∥ ζ ∥ Y 0 ≤ A ′ ∣Σ ∣ 1/2 } , with A ′ : = 2 C q ( C N S + C h ) A':=2C_q(C_{NS}+C_h) A ′ := 2 C q ( C N S + C h ) and the metric of Y 1 Y_1 Y 1 . By Lemma 5.4,
∥ Θ ζ ∥ Y 0 ≤ C q ( ∣ Σ ∣ 1 / 2 ( C N S + C h ) + ∣ Σ ∣ 1 / 2 ( 3 ϱ + 2 C E B U ) C T A ′ ∣ Σ ∣ 1 / 2 + ∣ Σ ∣ 1 / 2 C E C T 2 ( A ′ ∣ Σ ∣ 1 / 2 ) 2 ) ≤ A ′ ∣ Σ ∣ 1 / 2 \|\Theta\zeta\|_{Y_0}\le C_q\left(|\Sigma|^{1/2}(C_{NS}+C_h)+|\Sigma|^{1/2}(3\varrho+2C_EB_U)C_TA'|\Sigma|^{1/2}+|\Sigma|^{1/2}C_EC_T^2(A'|\Sigma|^{1/2})^2\right)\le A'|\Sigma|^{1/2} ∥Θ ζ ∥ Y 0 ≤ C q ( ∣Σ ∣ 1/2 ( C N S + C h ) + ∣Σ ∣ 1/2 ( 3 ϱ + 2 C E B U ) C T A ′ ∣Σ ∣ 1/2 + ∣Σ ∣ 1/2 C E C T 2 ( A ′ ∣Σ ∣ 1/2 ) 2 ) ≤ A ′ ∣Σ ∣ 1/2
if ∣ Σ ∣ ≤ ℓ 0 ′ |\Sigma|\le\ell'_0 ∣Σ∣ ≤ ℓ 0 ′ is small. By Lemma 6.4(a), (6.9) and ∥ div ( ζ ⊗ ζ − ζ ′ ⊗ ζ ′ ) ∥ H − 1 , q ≤ C ( ∥ ζ ∥ ∞ + ∥ ζ ′ ∥ ∞ ) ∥ ζ − ζ ′ ∥ L q \|\operatorname{div}(\zeta\otimes\zeta-\zeta'\otimes\zeta')\|_{H^{-1,q}}\le C(\|\zeta\|_\infty+\|\zeta'\|_\infty)\|\zeta-\zeta'\|_{L^q} ∥ div ( ζ ⊗ ζ − ζ ′ ⊗ ζ ′ ) ∥ H − 1 , q ≤ C ( ∥ ζ ∥ ∞ + ∥ ζ ′ ∥ ∞ ) ∥ ζ − ζ ′ ∥ L q ,
∥ Θ ζ − Θ ζ ′ ∥ Y 1 ≤ C q ( K 2 ( 1 + C T A ′ ) ∣ Σ ∣ 3 / 2 C T + ( 3 ϱ + 2 C E B U ) ∣ Σ ∣ 1 / 2 C T + 2 C C E C T 2 A ′ ∣ Σ ∣ ) ∥ ζ − ζ ′ ∥ Y 1 ≤ 1 2 ∥ ζ − ζ ′ ∥ Y 1 \begin{aligned}
\|\Theta\zeta-\Theta\zeta'\|_{Y_1}\le C_q\left(K_2(1+C_TA')|\Sigma|^{3/2}C_T+(3\varrho+2C_EB_U)|\Sigma|^{1/2}C_T+2CC_EC_T^2A'|\Sigma|\right)\|\zeta-\zeta'\|_{Y_1} \\
&\le\frac12\|\zeta-\zeta'\|_{Y_1}
\end{aligned} ∥Θ ζ − Θ ζ ′ ∥ Y 1 ≤ C q ( K 2 ( 1 + C T A ′ ) ∣Σ ∣ 3/2 C T + ( 3 ϱ + 2 C E B U ) ∣Σ ∣ 1/2 C T + 2 C C E C T 2 A ′ ∣Σ∣ ) ∥ ζ − ζ ′ ∥ Y 1 ≤ 2 1 ∥ ζ − ζ ′ ∥ Y 1
for ∣ Σ ∣ ≤ ℓ 0 ′ |\Sigma|\le\ell'_0 ∣Σ∣ ≤ ℓ 0 ′ . The set E E E is complete for the Y 1 Y_1 Y 1 metric: a Y 1 Y_1 Y 1 -limit of elements of E E E is a weak limit in the reflexive space Y 0 Y_0 Y 0 , so the bound and the initial value pass to the limit. The fixed point z z z exists and is unique in E E E . For uniqueness on the larger set { ∥ ζ ∥ Y 0 ( Σ ) ≤ 1 } \{\|\zeta\|_{Y_0(\Sigma)}\le1\} { ∥ ζ ∥ Y 0 ( Σ ) ≤ 1 } , let z ~ \widetilde z z be a second solution with ∥ z ~ ∥ Y 0 ( Σ ) ≤ 1 \|\widetilde z\|_{Y_0(\Sigma)}\le1 ∥ z ∥ Y 0 ( Σ ) ≤ 1 . The kinetic bounds above and (6.9) hold for z ~ \widetilde z z by (6.7). In the contraction estimate the bound ∥ ζ ∥ Y 0 ≤ A ′ ∣ Σ ∣ 1 / 2 \|\zeta\|_{Y_0}\le A'|\Sigma|^{1/2} ∥ ζ ∥ Y 0 ≤ A ′ ∣Σ ∣ 1/2 entered in two places. In the kinetic term it entered only through 1 + ∥ ζ ′ ( t ) ∥ W 1 , q ≤ 1 + C T A ′ 1+\|\zeta'(t)\|_{W^{1,q}}\le1+C_TA' 1 + ∥ ζ ′ ( t ) ∥ W 1 , q ≤ 1 + C T A ′ , and 1 + C T 1+C_T 1 + C T suffices there, since the factor ∣ Σ ∣ 3 / 2 |\Sigma|^{3/2} ∣Σ ∣ 3/2 is explicit. In the quadratic term it entered through ∥ ζ ∥ L 2 ( Σ ; L ∞ ) \|\zeta\|_{L^2(\Sigma;L^\infty)} ∥ ζ ∥ L 2 ( Σ ; L ∞ ) , and merely replacing A ′ ∣ Σ ∣ 1 / 2 A'|\Sigma|^{1/2} A ′ ∣Σ ∣ 1/2 by 1 1 1 would lose the small factor. Instead, since ζ ( a ) = 0 \zeta(a)=0 ζ ( a ) = 0 , Lemma 5.4(b) and X q ↪ L ∞ X_q\hookrightarrow L^\infty X q ↪ L ∞ give
∥ ζ ∥ L 2 ( Σ ; L ∞ ) ≤ ∣ Σ ∣ 1 / 2 sup Σ ∥ ζ ∥ L ∞ ≤ C E C T ∣ Σ ∣ 1 / 2 ∥ ζ ∥ Y 0 ( Σ ) ≤ C E C T ∣ Σ ∣ 1 / 2 , \|\zeta\|_{L^2(\Sigma;L^\infty)}\le|\Sigma|^{1/2}\sup_\Sigma\|\zeta\|_{L^\infty}\le C_EC_T|\Sigma|^{1/2}\|\zeta\|_{Y_0(\Sigma)}\le C_EC_T|\Sigma|^{1/2}, ∥ ζ ∥ L 2 ( Σ ; L ∞ ) ≤ ∣Σ ∣ 1/2 Σ sup ∥ ζ ∥ L ∞ ≤ C E C T ∣Σ ∣ 1/2 ∥ ζ ∥ Y 0 ( Σ ) ≤ C E C T ∣Σ ∣ 1/2 ,
and the same for ζ ′ \zeta' ζ ′ . Hence, for z = Θ z z=\Theta z z = Θ z and z ~ = Θ z ~ \widetilde z=\Theta\widetilde z z = Θ z ,
∥ z − z ~ ∥ Y 1 ≤ C q ( K 2 ( 1 + C T ) ∣ Σ ∣ 3 / 2 C T + ( 3 ϱ + 2 C E B U ) ∣ Σ ∣ 1 / 2 C T + 2 C C E C T 2 ∣ Σ ∣ 1 / 2 ) ∥ z − z ~ ∥ Y 1 ≤ 1 2 ∥ z − z ~ ∥ Y 1 \|z-\widetilde z\|_{Y_1}\le C_q\left(K_2(1+C_T)|\Sigma|^{3/2}C_T+(3\varrho+2C_EB_U)|\Sigma|^{1/2}C_T+2CC_EC_T^2|\Sigma|^{1/2}\right)\|z-\widetilde z\|_{Y_1}\le\frac12\|z-\widetilde z\|_{Y_1} ∥ z − z ∥ Y 1 ≤ C q ( K 2 ( 1 + C T ) ∣Σ ∣ 3/2 C T + ( 3 ϱ + 2 C E B U ) ∣Σ ∣ 1/2 C T + 2 C C E C T 2 ∣Σ ∣ 1/2 ) ∥ z − z ∥ Y 1 ≤ 2 1 ∥ z − z ∥ Y 1
after decreasing ℓ 0 ′ \ell'_0 ℓ 0 ′ , so z ~ = z \widetilde z=z z = z . This proves (i). The gain ∣ Σ ∣ 1 / 2 |\Sigma|^{1/2} ∣Σ ∣ 1/2 does not require ζ \zeta ζ or N S N_S N S to be small: every forcing term is bounded in L ∞ ( Σ ; L q ) L^\infty(\Sigma;L^q) L ∞ ( Σ ; L q ) , and the factor comes from the L 2 ( Σ ) L^2(\Sigma) L 2 ( Σ ) norm in time.
For (ii), estimate z = Θ z z=\Theta z z = Θ z in Y 1 Y_1 Y 1 with zero data. By Lemma 6.4(a), (6.8) with s = q s=q s = q , and ∥ div S ∥ H − 1 , q ≤ C ∥ S ∥ L q \|\operatorname{div} S\|_{H^{-1,q}}\le C\|S\|_{L^q} ∥ div S ∥ H − 1 , q ≤ C ∥ S ∥ L q ,
∥ z ∥ Y 1 ≤ C q ( C ∣ Σ ∣ 1 / 2 ς + 3 K 1 ∣ Σ ∣ 3 / 2 C T ∥ z ∥ Y 1 + ( 3 ϱ + 2 C E B U ) ∣ Σ ∣ 1 / 2 C T ∥ z ∥ Y 1 + C E C T A ′ 1 / 2 C T ∥ z ∥ Y 1 ) . \|z\|_{Y_1}\le C_q\left(C|\Sigma|^{1/2}\varsigma+3K_1|\Sigma|^{3/2}C_T\|z\|_{Y_1}+(3\varrho+2C_EB_U)|\Sigma|^{1/2}C_T\|z\|_{Y_1}+C_EC_TA'^{1/2}C_T\|z\|_{Y_1}\right). ∥ z ∥ Y 1 ≤ C q ( C ∣Σ ∣ 1/2 ς + 3 K 1 ∣Σ ∣ 3/2 C T ∥ z ∥ Y 1 + ( 3 ϱ + 2 C E B U ) ∣Σ ∣ 1/2 C T ∥ z ∥ Y 1 + C E C T A ′ 1/2 C T ∥ z ∥ Y 1 ) .
where the last term comes from ∥ div ( z ⊗ z ) ∥ H − 1 , q ≤ ∥ z ∥ ∞ ∥ z ∥ L q \|\operatorname{div}(z\otimes z)\|_{H^{-1,q}}\le\|z\|_{\infty}\|z\|_{L^q} ∥ div ( z ⊗ z ) ∥ H − 1 , q ≤ ∥ z ∥ ∞ ∥ z ∥ L q . For ∣ Σ ∣ ≤ ℓ 0 ′ |\Sigma|\le\ell'_0 ∣Σ∣ ≤ ℓ 0 ′ the terms containing ∥ z ∥ Y 1 \|z\|_{Y_1} ∥ z ∥ Y 1 are absorbed, and ∥ z ∥ Y 1 ≤ 2 C q C ∣ Σ ∣ 1 / 2 ς \|z\|_{Y_1}\le2C_qC|\Sigma|^{1/2}\varsigma ∥ z ∥ Y 1 ≤ 2 C q C ∣Σ ∣ 1/2 ς .
(iii) follows from (i), (ii) and Lemma 6.4(c), (d):
∥ z ∥ L 2 W 1 , ∞ ≤ C ∥ z ∥ Y 0 1 − β ∥ z ∥ Y 1 β \|z\|_{L^2W^{1,\infty}}\le C\|z\|_{Y_0}^{1-\beta}\|z\|_{Y_1}^{\beta} ∥ z ∥ L 2 W 1 , ∞ ≤ C ∥ z ∥ Y 0 1 − β ∥ z ∥ Y 1 β
and
∥ z ( t ) ∥ L ∞ ≤ C ∥ z ( t ) ∥ X q 1 − β ∥ z ( t ) ∥ B q 2 β . \|z(t)\|_{L^\infty}\le C\|z(t)\|_{X_q}^{1-\beta}\|z(t)\|_{B_q^2}^{\beta}. ∥ z ( t ) ∥ L ∞ ≤ C ∥ z ( t ) ∥ X q 1 − β ∥ z ( t ) ∥ B q 2 β .
For (iv): since R \mathcal{R} R and Q \mathcal{Q} Q are Fourier multipliers and the heat semigroup is an L r L^r L r contraction,
∥ R z ( t ) ∥ L r ≤ ∫ a t ∥ R Q ( ⋯ ) ∥ L r . \|\mathcal{R}z(t)\|_{L^r}\le\int_a^t\|\mathcal{R}\mathcal{Q}(\cdots)\|_{L^r}. ∥ R z ( t ) ∥ L r ≤ ∫ a t ∥ R Q ( ⋯ ) ∥ L r .
The integrand is bounded as follows, using Lemma 6.4(e) and r ≤ 2 ≤ q r\le2\le q r ≤ 2 ≤ q :
∥ R Q div S ∥ L r ≤ C r ∥ S ∥ L r \|\mathcal{R}\mathcal{Q}\operatorname{div}S\|_{L^r}\le C_r\|S\|_{L^r} ∥ R Q div S ∥ L r ≤ C r ∥ S ∥ L r ;
∥ R Q h z ∥ L r ≤ ∥ R Q h z ∥ L q ≤ C ∥ h z ∥ H − 1 , q ≤ 3 C K 1 ∣ Σ ∣ sup Σ ∥ z ∥ L q \|\mathcal{R}\mathcal{Q}h_z\|_{L^r}\le\|\mathcal{R}\mathcal{Q}h_z\|_{L^q}\le C\|h_z\|_{H^{-1,q}}\le3CK_1|\Sigma|\sup_{\Sigma}\|z\|_{L^q} ∥ R Q h z ∥ L r ≤ ∥ R Q h z ∥ L q ≤ C ∥ h z ∥ H − 1 , q ≤ 3 C K 1 ∣Σ∣ sup Σ ∥ z ∥ L q , by ∣ T d ∣ = 1 |T^d|=1 ∣ T d ∣ = 1 , r ≤ q r\le q r ≤ q , Lemma 6.4(e) and (6.8) (Lemma 6.5(b) is not available with s = r s=r s = r );
∥ R Q ( ϱ a z ) ∥ L r ≤ 3 C ϱ sup Σ ∥ z ∥ L q \|\mathcal{R}\mathcal{Q}(\varrho_a z)\|_{L^r}\le3C\varrho\sup_{\Sigma}\|z\|_{L^q} ∥ R Q ( ϱ a z ) ∥ L r ≤ 3 C ϱ sup Σ ∥ z ∥ L q ;
the terms with U ⊗ z U\otimes z U ⊗ z and z ⊗ z z\otimes z z ⊗ z are at most C ( C E B U + sup ∥ z ∥ ∞ ) sup Σ ∥ z ∥ L q C(C_EB_U+\sup\|z\|_\infty)\sup_{\Sigma}\|z\|_{L^q} C ( C E B U + sup ∥ z ∥ ∞ ) sup Σ ∥ z ∥ L q .
Integrating over [ a , t ] [a,t] [ a , t ] and using (ii) gives (iv). The constant is primed because of the factor sup ∥ z ∥ ∞ ≤ C ′ ∣ Σ ∣ 1 / 2 ς β \sup\|z\|_\infty\le C'|\Sigma|^{1/2}\varsigma^\beta sup ∥ z ∥ ∞ ≤ C ′ ∣Σ ∣ 1/2 ς β from (iii).
(b) Long segments. Here ∣ Σ ∣ |\Sigma| ∣Σ∣ is not small, so the forcing h h h is not small in L 2 L q L^2L^q L 2 L q and the nonlinear term is controlled by the smallness of ς \varsigma ς . The argument is the one of (a) on pieces, with a priori bounds between the pieces. Call a solution z z z of (6.4) on [ a , t ′ ] ⊂ Σ [a,t']\subset\Sigma [ a , t ′ ] ⊂ Σ admissible if
∥ z ∥ Y 1 ( a , t ′ ) ≤ ε 0 , ∥ z ∥ L 1 ( a , t ′ ; W 1 , ∞ ) ≤ 1. (39) \|z\|_{Y_1(a,t')}\le\varepsilon_0,\qquad\|z\|_{L^1(a,t';W^{1,\infty})}\le1. \tag*{(39)} ∥ z ∥ Y 1 ( a , t ′ ) ≤ ε 0 , ∥ z ∥ L 1 ( a , t ′ ; W 1 , ∞ ) ≤ 1. (39)
where ε 0 ∈ ( 0 , 1 ] \varepsilon_0\in(0,1] ε 0 ∈ ( 0 , 1 ] is fixed below. For admissible z z z the kinetic bounds of the first paragraph hold on [ a , t ′ ] [a,t'] [ a , t ′ ] : the glued field has L 1 L ∞ L^1L^\infty L 1 L ∞ norm at most C E B U + 3 C_EB_U+3 C E B U + 3 and Lipschitz norm at most Λ U + 3 \Lambda_U+3 Λ U + 3 in L 1 L^1 L 1 . So (6.8) and (6.9) hold with the constants K 1 , K 2 K_1,K_2 K 1 , K 2 of (a), and ∣ h z ∣ ≤ C h |h_z|\le C h ∣ h z ∣ ≤ C h . The constants are chosen in the following order; each depends only on the quantities listed in the statement and on those chosen before it:
ε 0 , ℓ 1 ⟶ C B ⟶ A , A 0 ⟶ A 2 , ℓ 2 ⟶ c ∗ ≤ ε ∗ : = ε 0 . \varepsilon_0,\ \ell_1\longrightarrow C_B\longrightarrow A,\ A_0\longrightarrow A_2,\ \ell_2\longrightarrow c_* \le\varepsilon_*:=\varepsilon_0. ε 0 , ℓ 1 ⟶ C B ⟶ A , A 0 ⟶ A 2 , ℓ 2 ⟶ c ∗ ≤ ε ∗ := ε 0 .
The parameters ε 0 \varepsilon_0 ε 0 and ℓ 1 \ell_1 ℓ 1 are fixed once, at the beginning, so that both absorption conditions below (in the a priori bounds in Y 1 Y_1 Y 1 and in Y 0 Y_0 Y 0 ) hold; both involve only unprimed constants. In particular ℓ 1 \ell_1 ℓ 1 , N 1 N_1 N 1 and C B C_B C B are unprimed and are not changed afterwards; N S N_S N S enters from A A A on.
A priori bound in Y 1 Y_1 Y 1 . Let z z z be admissible on [ a , t ′ ] [a,t'] [ a , t ′ ] , and cover [ a , t ′ ] [a,t'] [ a , t ′ ] by consecutive intervals I j = [ s j , s j + 1 ] I_j=[s_j,s_{j+1}] I j = [ s j , s j + 1 ] , 0 ≤ j < N 1 0\le j<N_1 0 ≤ j < N 1 , of length at most ℓ 1 \ell_1 ℓ 1 , with N 1 ≤ 1 / ℓ 1 + 1 N_1\le1/\ell_1+1 N 1 ≤ 1/ ℓ 1 + 1 . On I j I_j I j we use Lemma 6.4(a) with data z ( s j ) z(s_j) z ( s j ) . Since z ( a ) = 0 z(a)=0 z ( a ) = 0 , sup [ a , s ] ∥ z ∥ L q ≤ C T ∥ z ∥ Y 1 ( a , s ) \sup_{[a,s]}\|z\|_{L^q}\le C_T\|z\|_{Y_1(a,s)} sup [ a , s ] ∥ z ∥ L q ≤ C T ∥ z ∥ Y 1 ( a , s ) ; on I j I_j I j , sup I j ∥ z ∥ L q ≤ C T ( ∥ z ( s j ) ∥ B q , 2 0 + ∥ z ∥ Y 1 ( I j ) ) \sup_{I_j}\|z\|_{L^q}\le C_T(\|z(s_j)\|_{B_{q,2}^0}+\|z\|_{Y_1(I_j)}) sup I j ∥ z ∥ L q ≤ C T ( ∥ z ( s j ) ∥ B q , 2 0 + ∥ z ∥ Y 1 ( I j ) ) ; and ∥ div ( z ⊗ z ) ∥ L 2 ( I j ; H − 1 , q ) ≤ C ∥ z ∥ L 2 ( I j ; L ∞ ) sup I j ∥ z ∥ L q ≤ C C E ε 0 sup I j ∥ z ∥ L q \|\operatorname{div}(z\otimes z)\|_{L^2(I_j;H^{-1,q})}\le C\|z\|_{L^2(I_j;L^\infty)}\sup_{I_j}\|z\|_{L^q}\le CC_E\varepsilon_0\sup_{I_j}\|z\|_{L^q} ∥ div ( z ⊗ z ) ∥ L 2 ( I j ; H − 1 , q ) ≤ C ∥ z ∥ L 2 ( I j ; L ∞ ) sup I j ∥ z ∥ L q ≤ C C E ε 0 sup I j ∥ z ∥ L q . Hence
∥ z ∥ Y 1 ( I j ) ≤ C q ( ∥ z ( s j ) ∥ B q , 2 0 + ℓ 1 1 / 2 ( C S + 3 K 1 C T ∥ z ∥ Y 1 ( a , s j + 1 ) ) \|z\|_{Y_1(I_j)}\le C_q\left(\|z(s_j)\|_{B_{q,2}^0}+\ell_1^{1/2}(C_S+3K_1C_T\|z\|_{Y_1(a,s_{j+1})})\right. ∥ z ∥ Y 1 ( I j ) ≤ C q ( ∥ z ( s j ) ∥ B q , 2 0 + ℓ 1 1/2 ( C S + 3 K 1 C T ∥ z ∥ Y 1 ( a , s j + 1 ) )
+ ( ( 3 ϱ + 2 C E B U ) ℓ 1 1 / 2 + C C E ε 0 ) C T ( ∥ z ( s j ) ∥ B q , 2 0 + ∥ z ∥ Y 1 ( I j ) ) ) . \left.+((3\varrho+2C_EB_U)\ell_1^{1/2}+CC_E\varepsilon_0)C_T(\|z(s_j)\|_{B_{q,2}^0}+\|z\|_{Y_1(I_j)})\right). + (( 3 ϱ + 2 C E B U ) ℓ 1 1/2 + C C E ε 0 ) C T ( ∥ z ( s j ) ∥ B q , 2 0 + ∥ z ∥ Y 1 ( I j ) ) ) .
We choose ε 0 \varepsilon_0 ε 0 and then ℓ 1 \ell_1 ℓ 1 so small that the coefficients of ∥ z ∥ Y 1 ( I j ) \|z\|_{Y_1(I_j)} ∥ z ∥ Y 1 ( I j ) on the right are at most 1 4 \frac14 4 1 in total, and that C q ℓ 1 1 / 2 ( 3 ϱ + 2 C E B U ) C T ≤ 1 4 C_q\ell_1^{1/2}(3\varrho+2C_EB_U)C_T\le\frac14 C q ℓ 1 1/2 ( 3 ϱ + 2 C E B U ) C T ≤ 4 1 , which is the absorption condition used in the a priori bound in Y 0 Y_0 Y 0 below. Since S 0 = 0 S_0=0 S 0 = 0 , and ∥ z ( s j + 1 ) ∥ B q , 2 0 ≤ C T ( ∥ z ( s j ) ∥ B q , 2 0 + ∥ z ∥ Y 1 ( I j ) ) \|z(s_{j+1})\|_{B_{q,2}^0}\le C_T(\|z(s_j)\|_{B_{q,2}^0}+\|z\|_{Y_1(I_j)}) ∥ z ( s j + 1 ) ∥ B q , 2 0 ≤ C T ( ∥ z ( s j ) ∥ B q , 2 0 + ∥ z ∥ Y 1 ( I j ) ) , this gives S j + 1 ≤ C ′ ′ ( S j + ς ) S_{j+1}\le C''(S_j+\varsigma) S j + 1 ≤ C ′′ ( S j + ς ) , with C ′ ′ C'' C ′′ unprimed. Hence
∥ z ∥ Y 1 ( a , t ′ ) + sup [ a , t ′ ] ∥ z ∥ B q , 2 0 ≤ C B ς , C B : = ( 2 C ′ ′ ) N 1 . (40) \|z\|_{Y_1(a,t')}+\sup_{[a,t']}\|z\|_{B_{q,2}^0}\le C_B\varsigma,\qquad C_B:=(2C'')^{N_1}. \tag*{(40)} ∥ z ∥ Y 1 ( a , t ′ ) + [ a , t ′ ] sup ∥ z ∥ B q , 2 0 ≤ C B ς , C B := ( 2 C ′′ ) N 1 . (40)
A priori bound in Y 0 Y_0 Y 0 . Let z z z be admissible on [ a , t ′ ] [a,t'] [ a , t ′ ] , and assume in addition
sup [ a , t ′ ] ∥ z ∥ L ∞ ≤ 1 8 C q . (41) \sup_{[a,t']}\|z\|_{L^\infty}\le\frac{1}{8C_q}. \tag*{(41)} [ a , t ′ ] sup ∥ z ∥ L ∞ ≤ 8 C q 1 . (41)
On I j I_j I j , Lemma 5.4(a), (b), the bounds ∥ div S ∥ L q ≤ C N S \lVert\operatorname{div} S\rVert_{L^q} \le C N_S ∥ div S ∥ L q ≤ C N S , ∥ g z ∥ L q ≤ C h + 3 ϱ ∥ z ∥ L q \lVert g z\rVert_{L^q} \le C_h+3\varrho\lVert z\rVert_{L^q} ∥ g z ∥ L q ≤ C h + 3 ϱ ∥ z ∥ L q , ∥ div ( U ⊗ z + z ⊗ U ) ∥ L q ≤ 2 C E B U ∥ z ∥ X q \lVert\operatorname{div}(U\otimes z+z\otimes U)\rVert_{L^q} \le2C_{E}B_U\lVert z\rVert_{X_q} ∥ div ( U ⊗ z + z ⊗ U ) ∥ L q ≤ 2 C E B U ∥ z ∥ X q and ∥ div ( z ⊗ z ) ∥ L q ≤ 2 ∥ z ∥ ∞ ∥ ∇ z ∥ L q \lVert\operatorname{div}(z\otimes z)\rVert_{L^q} \le2\lVert z\rVert_{\infty}\lVert\nabla z\rVert_{L^q} ∥ div ( z ⊗ z ) ∥ L q ≤ 2 ∥ z ∥ ∞ ∥ ∇ z ∥ L q give
∥ z ∥ Y 0 ( I j ) ≤ C q ( ∥ z ( s j ) ∥ X q + ℓ 1 1 / 2 ( C N S + C h + ( 3 ϱ + 2 C E B U ) C T ( ∥ z ( s j ) ∥ X q + ∥ z ∥ Y 0 ( I j ) ) ) ) + 1 4 ∥ z ∥ Y 0 ( I j ) . \lVert z\rVert_{Y_0(I_j)} \le C_q\left(\lVert z(s_j)\rVert_{X_q}+\ell_1^{1/2}\left(CN_S+C_h+(3\varrho+2C_EB_U)C_T\left(\lVert z(s_j)\rVert_{X_q}+\lVert z\rVert_{Y_0(I_j)}\right)\right)\right)+\frac{1}{4}\lVert z\rVert_{Y_0(I_j)}. ∥ z ∥ Y 0 ( I j ) ≤ C q ( ∥ z ( s j ) ∥ X q + ℓ 1 1/2 ( C N S + C h + ( 3 ϱ + 2 C E B U ) C T ( ∥ z ( s j ) ∥ X q + ∥ z ∥ Y 0 ( I j ) ) ) ) + 4 1 ∥ z ∥ Y 0 ( I j ) .
By the choice of ℓ 1 \ell_1 ℓ 1 , the coefficient of ∥ z ∥ Y 0 ( I j ) \lVert z\rVert_{Y_0(I_j)} ∥ z ∥ Y 0 ( I j ) coming from the linear terms is at most 1 4 \frac{1}{4} 4 1 , so it is absorbed together with the last term, and X j + 1 : = sup [ a , s j + 1 ] ∥ z ∥ X q X_{j+1}:=\sup_{[a,s_{j+1}]}\lVert z\rVert_{X_q} X j + 1 := sup [ a , s j + 1 ] ∥ z ∥ X q satisfies X j + 1 ≤ C ′ ′ ′ ( X j + 1 ) X_{j+1}\le C'''(X_j+1) X j + 1 ≤ C ′′′ ( X j + 1 ) , X 0 = 0 X_0=0 X 0 = 0 , where C ′ ′ ′ C''' C ′′′ depends on N S N_S N S . Hence, with A : = ( 2 C ′ ′ ′ ) N 1 A:=(2C''')^{N_1} A := ( 2 C ′′′ ) N 1 and A 0 : = 2 C q N 1 ( C T A + C N S + C h + 1 ) A_0:=2C_qN_1(C_TA+CN_S+C_h+1) A 0 := 2 C q N 1 ( C T A + C N S + C h + 1 ) ,
sup [ a , t ′ ] ∥ z ∥ X q ≤ A , ∥ z ∥ Y 0 ( a , t ′ ) ≤ A 0 . (42) \sup_{[a,t']}\lVert z\rVert_{X_q}\le A,\qquad\lVert z\rVert_{Y_0(a,t')}\le A_0. \tag*{(42)} [ a , t ′ ] sup ∥ z ∥ X q ≤ A , ∥ z ∥ Y 0 ( a , t ′ ) ≤ A 0 . (42)
By Lemma 6.4(c), (d), applied on each I j I_j I j , and by (6.11)–(6.13),
sup [ a , t ′ ] ∥ z ∥ L ∞ ≤ C A 1 − β ( C B ς ) β , ∥ z ∥ L 1 ( a , t ′ ; W 1 , ∞ ) ≤ ∥ z ∥ L 2 ( a , t ′ ; W 1 , ∞ ) ≤ C N 1 1 / 2 A 0 1 − β ( C B ς ) β . (43) \sup_{[a,t']}\lVert z\rVert_{L^\infty}\le CA^{1-\beta}(C_B\varsigma)^\beta,\qquad\lVert z\rVert_{L^1(a,t';W^{1,\infty})}\le\lVert z\rVert_{L^2(a,t';W^{1,\infty})}\le CN_1^{1/2}A_0^{1-\beta}(C_B\varsigma)^\beta. \tag*{(43)} [ a , t ′ ] sup ∥ z ∥ L ∞ ≤ C A 1 − β ( C B ς ) β , ∥ z ∥ L 1 ( a , t ′ ; W 1 , ∞ ) ≤ ∥ z ∥ L 2 ( a , t ′ ; W 1 , ∞ ) ≤ C N 1 1/2 A 0 1 − β ( C B ς ) β . (43)
We require of c ∗ ′ c_*' c ∗ ′ that, for ς ≤ c ∗ ′ \varsigma\le c_*' ς ≤ c ∗ ′ , the right-hand sides of (6.14) are at most 1 16 C q \frac{1}{16C_q} 16 C q 1 and 1 2 \frac{1}{2} 2 1 , and that C q C B ς ≤ ε 0 / 4 C_qC_B\varsigma\le\varepsilon_0/4 C q C B ς ≤ ε 0 /4 ; in particular C B ς ≤ ε 0 / 2 C_B\varsigma\le\varepsilon_0/2 C B ς ≤ ε 0 /2 , since C q ≥ 1 C_q\ge1 C q ≥ 1 . Then (6.12) is not needed as a hypothesis: t ↦ sup [ a , t ] ∥ z ∥ ∞ t\mapsto\sup_{[a,t]}\lVert z\rVert_\infty t ↦ sup [ a , t ] ∥ z ∥ ∞ is continuous, since z ∈ C ( [ a , t ′ ] ; X q ) z\in C([a,t'];X_q) z ∈ C ([ a , t ′ ] ; X q ) and X q ↪ L ∞ X_q\hookrightarrow L^\infty X q ↪ L ∞ , and it vanishes at t = a t=a t = a ; by (6.14) applied on [ a , t ] [a,t] [ a , t ] it can never reach 1 8 C q \frac{1}{8C_q} 8 C q 1 . Consequently, for ς ≤ c ∗ ′ \varsigma\le c_*' ς ≤ c ∗ ′ , every admissible solution on [ a , t ′ ] [a,t'] [ a , t ′ ] satisfies the strict bounds
∥ z ∥ Y 1 ( a , t ′ ) ≤ ε 0 2 , ∥ z ∥ L 1 ( a , t ′ ; W 1 , ∞ ) ≤ 1 2 , sup [ a , t ′ ] ∥ z ∥ X q ≤ A , ∥ z ∥ Y 0 ( a , t ′ ) ≤ A 0 . (44) \lVert z\rVert_{Y_1(a,t')}\le\frac{\varepsilon_0}{2},\qquad\lVert z\rVert_{L^1(a,t';W^{1,\infty})}\le\frac{1}{2},\qquad\sup_{[a,t']}\lVert z\rVert_{X_q}\le A,\qquad\lVert z\rVert_{Y_0(a,t')}\le A_0. \tag*{(44)} ∥ z ∥ Y 1 ( a , t ′ ) ≤ 2 ε 0 , ∥ z ∥ L 1 ( a , t ′ ; W 1 , ∞ ) ≤ 2 1 , [ a , t ′ ] sup ∥ z ∥ X q ≤ A , ∥ z ∥ Y 0 ( a , t ′ ) ≤ A 0 . (44)
Existence. Let A 2 : = 2 C q ( A + 1 ) + A 0 A_2:=2C_q(A+1)+A_0 A 2 := 2 C q ( A + 1 ) + A 0 , and fix a partition a = s 0 < s 1 < ⋯ < s N 2 = b a=s_0<s_1<\cdots<s_{N_2}=b a = s 0 < s 1 < ⋯ < s N 2 = b of mesh ℓ 2 ≤ ℓ 1 \ell_2\le\ell_1 ℓ 2 ≤ ℓ 1 . Suppose an admissible solution z z z is constructed on [ a , s i ] [a,s_i] [ a , s i ] (for i = 0 i=0 i = 0 , z ( a ) = 0 z(a)=0 z ( a ) = 0 ). On I = [ s i , s i + 1 ] I=[s_i,s_{i+1}] I = [ s i , s i + 1 ] consider
E I : = { ζ ∈ Y 0 ( I ) : ζ ( s i ) = z ( s i ) , ∥ ζ ∥ Y 0 ( I ) ≤ A 2 , ∥ ζ ∥ Y 1 ( I ) ≤ ε 0 2 } , E_I:=\left\{\zeta\in Y_0(I):\ \zeta(s_i)=z(s_i),\ \lVert\zeta\rVert_{Y_0(I)}\le A_2,\ \lVert\zeta\rVert_{Y_1(I)}\le\frac{\varepsilon_0}{2}\right\}, E I := { ζ ∈ Y 0 ( I ) : ζ ( s i ) = z ( s i ) , ∥ ζ ∥ Y 0 ( I ) ≤ A 2 , ∥ ζ ∥ Y 1 ( I ) ≤ 2 ε 0 } ,
with the metric of Y 1 ( I ) Y_1(I) Y 1 ( I ) , each ζ \zeta ζ being extended by z z z on [ a , s i ] [a,s_i] [ a , s i ] . For ζ ∈ E I \zeta\in E_I ζ ∈ E I the extension is admissible on [ a , s i + 1 ] [a,s_{i+1}] [ a , s i + 1 ] , because (6.15) on [ a , s i ] [a,s_i] [ a , s i ] and ∥ ζ ∥ L 1 ( I ; W 1 , ∞ ) ≤ ℓ 2 1 / 2 C E A 2 ≤ 1 2 \lVert\zeta\rVert_{L^1(I;W^{1,\infty})}\le\ell_2^{1/2}C_EA_2\le\frac{1}{2} ∥ ζ ∥ L 1 ( I ; W 1 , ∞ ) ≤ ℓ 2 1/2 C E A 2 ≤ 2 1 for ℓ 2 \ell_2 ℓ 2 small; so the kinetic bounds are available. Let Θ \Theta Θ be the map of (a) on I I I with data z ( s i ) z(s_i) z ( s i ) . For ζ ∈ E I \zeta\in E_I ζ ∈ E I the trace estimates include the initial value: by (6.15) at s i s_i s i , sup I ∥ ζ ∥ X q ≤ C T ( A + A 2 ) \sup_I\lVert\zeta\rVert_{X_q}\le C_T(A+A_2) sup I ∥ ζ ∥ X q ≤ C T ( A + A 2 ) and sup I ∥ ζ ∥ L q ≤ C ( C B ς + ε 0 ) \sup_I\lVert\zeta\rVert_{L^q}\le C(C_B\varsigma+\varepsilon_0) sup I ∥ ζ ∥ L q ≤ C ( C B ς + ε 0 ) , and these bounds enter the constant C ( A 2 ) C(A_2) C ( A 2 ) below, which therefore also depends on A A A ; this is admissible, since A A A is fixed before A 2 A_2 A 2 . In the contraction estimate the difference of two candidates has zero value at s i s_i s i , so the zero-trace form applies to it, while the factor 1 + ∥ ζ ( t ) ∥ W 1 , q 1+\lVert\zeta(t)\rVert_{W^{1,q}} 1 + ∥ ζ ( t ) ∥ W 1 , q in (6.9) is bounded by 1 + C T ( A + A 2 ) 1+C_T(A+A_2) 1 + C T ( A + A 2 ) .
As in (a), ∥ Θ ζ ∥ Y 0 ( I ) ≤ C q ( ∥ z ( s i ) ∥ X q + ℓ 2 1 / 2 C ( A 2 ) ) ≤ A 2 \lVert\Theta\zeta\rVert_{Y_0(I)}\le C_q(\lVert z(s_i)\rVert_{X_q}+\ell_2^{1/2}C(A_2))\le A_2 ∥ Θ ζ ∥ Y 0 ( I ) ≤ C q (∥ z ( s i ) ∥ X q + ℓ 2 1/2 C ( A 2 )) ≤ A 2 and ∥ Θ ζ ∥ Y 1 ( I ) ≤ C q ( ∥ z ( s i ) ∥ B q , 2 0 + ℓ 2 1 / 2 C ( A 2 ) ) ≤ C q C B ς + C q ℓ 2 1 / 2 C ( A 2 ) ≤ ε 0 2 \lVert\Theta\zeta\rVert_{Y_1(I)}\le C_q(\lVert z(s_i)\rVert_{B^0_{q,2}}+\ell_2^{1/2}C(A_2))\le C_qC_B\varsigma+C_q\ell_2^{1/2}C(A_2)\le\frac{\varepsilon_0}{2} ∥ Θ ζ ∥ Y 1 ( I ) ≤ C q (∥ z ( s i ) ∥ B q , 2 0 + ℓ 2 1/2 C ( A 2 )) ≤ C q C B ς + C q ℓ 2 1/2 C ( A 2 ) ≤ 2 ε 0 , since C q C B ς ≤ ε 0 4 C_qC_B\varsigma\le\frac{\varepsilon_0}{4} C q C B ς ≤ 4 ε 0 and after decreasing ℓ 2 \ell_2 ℓ 2 so that C q ℓ 2 1 / 2 C ( A 2 ) ≤ ε 0 4 C_q\ell_2^{1/2}C(A_2)\le\frac{\varepsilon_0}{4} C q ℓ 2 1/2 C ( A 2 ) ≤ 4 ε 0 . The contraction estimate of (a), in which the history term (6.9) is supported in [ a , s i ] [a,s_i] [ a , s i ] , has factor at most 1 2 \frac{1}{2} 2 1 for ℓ 2 \ell_2 ℓ 2 small. E I E_I E I is complete as in (a). The fixed point extends z z z to an admissible solution on [ a , s i + 1 ] [a,s_{i+1}] [ a , s i + 1 ] , which satisfies (6.15) again. After N 2 N_2 N 2 steps z z z is defined on Σ \Sigma Σ .
Uniqueness. Let z , z ~ z,\widetilde z z , z be solutions on Σ \Sigma Σ with ∥ ⋅ ∥ Y 1 ( Σ ) ≤ ε ∗ ′ = ε 0 \lVert\cdot\rVert_{Y_1(\Sigma)}\le\varepsilon_*'=\varepsilon_0 ∥ ⋅ ∥ Y 1 ( Σ ) ≤ ε ∗ ′ = ε 0 and ∥ ⋅ ∥ L 1 ( Σ ; W 1 , ∞ ) ≤ 1 \lVert\cdot\rVert_{L^1(\Sigma;W^{1,\infty})}\le1 ∥ ⋅ ∥ L 1 ( Σ ; W 1 , ∞ ) ≤ 1 . Both are admissible, hence satisfy (6.15). Let s ∗ : = sup { s : z = z ~ on [ a , s ] } s_*:=\sup\{s:\ z=\widetilde z\text{ on }[a,s]\} s ∗ := sup { s : z = z on [ a , s ]} and suppose s ∗ < b s_*<b s ∗ < b . On I = [ s ∗ , min { s ∗ + ℓ 2 , b } ] I=[s_*,\min\{s_*+\ell_2,b\}] I = [ s ∗ , min { s ∗ + ℓ 2 , b }] both norms have Y 0 ( I ) Y_0(I) Y 0 ( I ) norm at most A 2 A_2 A 2 and Y 1 ( I ) Y_1(I) Y 1 ( I ) norm at most ε 0 2 \frac{\varepsilon_0}{2} 2 ε 0 , and the contraction estimate gives ∥ z − z ~ ∥ Y 1 ( I ) ≤ 1 2 ∥ z − z ~ ∥ Y 1 ( I ) \lVert z-\widetilde z\rVert_{Y_1(I)}\le\frac{1}{2}\lVert z-\widetilde z\rVert_{Y_1(I)} ∥ z − z ∥ Y 1 ( I ) ≤ 2 1 ∥ z − z ∥ Y 1 ( I ) , so z = z ~ z=\widetilde z z = z on I I I , a contradiction.
Bounds (i)–(iv). (i) and (ii) are (6.13) and (6.11), (iii) is (6.14), and (iv) follows as in (a), since every integrand there is bounded by a constant times sup Σ ∥ z ∥ L q ≤ C B ς \sup_\Sigma\lVert z\rVert_{L^q}\le C_B\varsigma sup Σ ∥ z ∥ L q ≤ C B ς , or by C r ∥ S ∥ L r C_r\lVert S\rVert_{L^r} C r ∥ S ∥ L r , and $|\Sigma|\le1.
(c) Momentum and mean. Take the mean of (6.4): c z ′ = ∫ g z c'_z=\int g z c z ′ = ∫ g z . By (3.2), which holds a.e. for fields in L 1 W 1 , ∞ L^1W^{1,\infty} L 1 W 1 , ∞ , by Lemma 8.1(a), d d t ( ∫ [ f z ] − ∫ [ f [ U ] ] ) = − ∫ f g z \frac{d}{dt}\left(\int[f_z]-\int[f[U]]\right)=-\int f g_z d t d ( ∫ [ f z ] − ∫ [ f [ U ]] ) = − ∫ f g z , since the kinetic fields are U ^ z \widehat U_z U z and U U U . Both quantities vanish at which a a a , which gives, along with the bound on c z c_z c z is Lemma 6.5(a) with ψ v \psi_v ψ v , v = 1 v=1 v = 1 and s = q s=q s = q . The bound on R g z R g_z R g z is (6.2), i.e. (6.8) for the history term and Lemma 6.4(e) for the instantaneous term ρ z \rho_z ρ z . Since ∥ a z ∥ L q ≤ 3 ∥ z ∥ L q \lVert a_z\rVert_{L^q}\le3\lVert z\rVert_{L^q} ∥ a z ∥ L q ≤ 3 ∥ z ∥ L q and ∣ Σ ∣ ≤ 1 |\Sigma|\le1 ∣Σ∣ ≤ 1 , the supremum bound follows.
After an exit by the mean, the glued field on [ b , 1 ] [b,1] [ b , 1 ] is U + c z U+c_z U + c z , where c z c_z c z is continued by (6.18); this is the equation of Lemma 3.8 started at b b b , so c z c_z c z is defined and continuous on all of [ b , 1 ] [b,1] [ b , 1 ] . Thus a z = c z a_z=c_z a z = c z there, and the kinetic states still agree at a a a . The constants of Lemma 6.5 with radius R V 1 R^{V_1} R V 1 are available on [ a , t ] [a,t] [ a , t ] as long as ∫ b t ∣ c z ∣ ≤ 1 \int_b^t|c_z|\le1 ∫ b t ∣ c z ∣ ≤ 1 ; we verify this by a continuity argument. Let t ∗ : = sup { t ∈ [ b , 1 ] : ∫ b t ∣ c z ∣ ≤ 1 } t^*:=\sup\{t\in[b,1]:\int_b^t|c_z|\le1\} t ∗ := sup { t ∈ [ b , 1 ] : ∫ b t ∣ c z ∣ ≤ 1 } . For t ≤ t ∗ t\le t^* t ≤ t ∗ , Lemma 6.5 applies on [ a , t ] [a,t] [ a , t ] with the base U U U , and the first bound in (6.5) gives
∣ c z ( t ) ∣ ≤ K ( 3 ∣ Σ ∣ sup Σ ∥ z ∥ L q + ∫ b t ∣ c z ∣ ) . \lvert c_z(t)\rvert\le K\left(3|\Sigma|\sup_{\Sigma}\lVert z\rVert_{L^q}+\int_b^t\lvert c_z\rvert\right). ∣ c z ( t )∣ ≤ K ( 3∣Σ∣ Σ sup ∥ z ∥ L q + ∫ b t ∣ c z ∣ ) .
so Gronwall’s inequality yields sup [ b , t ∗ ] ∣ c z ∣ ≤ 3 K e K ∣ Σ ∣ sup Σ ∥ z ∥ L q \sup_{[b,t_*]} |c_z| \le3Ke^K|\Sigma|\sup_{\Sigma}\|z\|_{L^q} sup [ b , t ∗ ] ∣ c z ∣ ≤ 3 K e K ∣Σ∣ sup Σ ∥ z ∥ L q . We require
3 K e K ∣ Σ ∣ sup Σ ∥ z ∥ L q ≤ 1 2 . (45) 3Ke^K|\Sigma|\sup_{\Sigma}\|z\|_{L^q} \le\frac{1}{2}. \tag*{(45)} 3 K e K ∣Σ∣ Σ sup ∥ z ∥ L q ≤ 2 1 . (45)
In case (a), sup Σ ∥ z ∥ L q ≤ C ∣ Σ ∣ 1 / 2 ς \sup_{\Sigma}\|z\|_{L^q} \le C|\Sigma|^{1/2}\varsigma sup Σ ∥ z ∥ L q ≤ C ∣Σ ∣ 1/2 ς by (ii) and ς ≤ C N S \varsigma\le CN_S ς ≤ C N S , so (6.16) holds after decreasing ℓ 0 ′ \ell'_0 ℓ 0 ′ . In case (b), sup Σ ∥ z ∥ L q ≤ C B ς \sup_{\Sigma}\|z\|_{L^q} \le C B\varsigma sup Σ ∥ z ∥ L q ≤ C B ς by (6.11) and ∣ Σ ∣ ≤ 1 |\Sigma| \le1 ∣Σ∣ ≤ 1 , so (6.16) holds after decreasing c ∗ ′ c'_* c ∗ ′ , which is chosen last. Then ∫ b t ∗ ∣ c z ∣ ≤ 1 2 \int_b^{t^*}|c_z| \le\frac{1}{2} ∫ b t ∗ ∣ c z ∣ ≤ 2 1 , so t ∗ = 1 t^*=1 t ∗ = 1 by continuity, and sup [ b , 1 ] ∣ c z ∣ ≤ 3 K e K ∣ Σ ∣ sup Σ ∥ z ∥ L q \sup_{[b,1]}|c_z| \le3Ke^K|\Sigma|\sup_{\Sigma}\|z\|_{L^q} sup [ b , 1 ] ∣ c z ∣ ≤ 3 K e K ∣Σ∣ sup Σ ∥ z ∥ L q . In particular the contribution of [ b , 1 ] [b,1] [ b , 1 ] to ∥ U ^ − U ∥ L 1 ( a , 1 ; L ∞ ) \|\widehat U-U\|_{L^1(a,1;L^\infty)} ∥ U − U ∥ L 1 ( a , 1 ; L ∞ ) is at most 1 2 ≤ 1 \frac{1}{2}\le1 2 1 ≤ 1 , as used in the definition of R v 1 R_{v1} R v 1 . Inserting this bound in the second bound of (6.5) gives (6.6).
(d) Persistent stress. By (6.6), ∥ U ⊗ ∘ c z + c z ⊗ ∘ U ∥ W 1 , q ≤ 2 ∥ U ∥ W 1 , q ∣ c z ∣ ≤ 2 B U ∣ c z ∣ \|U\mathbin{\mathop{\otimes}^{\circ}} c_z+c_z\mathbin{\mathop{\otimes}^{\circ}}U\|_{W^{1,q}}\le2\|U\|_{W^{1,q}}|c_z|\le2B_U|c_z| ∥ U ⊗ ∘ c z + c z ⊗ ∘ U ∥ W 1 , q ≤ 2∥ U ∥ W 1 , q ∣ c z ∣ ≤ 2 B U ∣ c z ∣ . By Lemma 6.4(e), ∥ R g z ∥ W 1 , q ≤ C ∥ g z ∥ L q ≤ C ( C h + ϱ ∣ c z ∣ ) \|R_{g_z}\|_{W^{1,q}}\le C\|g_z\|_{L^q}\le C(C_h+\varrho|c_z|) ∥ R g z ∥ W 1 , q ≤ C ∥ g z ∥ L q ≤ C ( C h + ϱ ∣ c z ∣ ) , since ∣ h z ∣ ≤ C h |h_z|\le C_h ∣ h z ∣ ≤ C h and ∣ ρ a z ∣ ≤ ϱ ∣ c z ∣ |\rho_{a z}|\le\varrho|c_z| ∣ ρ a z ∣ ≤ ϱ ∣ c z ∣ on [ b , 1 ] [b,1] [ b , 1 ] . The L q L^q L q bound follows from (6.6) and ∥ U ∥ ∞ ≤ C E B U \|U\|_\infty\le C_EB_U ∥ U ∥ ∞ ≤ C E B U .
Smoothness. Recall that in this section f 0 f_0 f 0 and the base are smooth, so f Σ : = f [ U ] ( a ) f_\Sigma:=f[U](a) f Σ := f [ U ] ( a ) is smooth with compact velocity support, and f ^ z ( t ) = e d ( t − a ) f Σ ∘ Z U ^ z ( a ; t , ⋅ ) \widehat f_z(t)=e^{d(t-a)}f_\Sigma\circ Z_{\widehat U_z}(a;t,\cdot) f z ( t ) = e d ( t − a ) f Σ ∘ Z U z ( a ; t , ⋅ ) for t ∈ Σ t\in\Sigma t ∈ Σ . We show that z ∈ C ∞ ( Σ × T d ) z\in C^\infty(\Sigma\times\mathbb T^d) z ∈ C ∞ ( Σ × T d ) , with one-sided derivatives at t = a t=a t = a and t = b t=b t = b . Then c z c_z c z , U ^ z \widehat U_z U z and f ^ z \widehat f_z f z are smooth on Σ \Sigma Σ as well. (The extension of z z z by zero to t < a t<a t < a is in general not smooth, since ∂ t z ( a ) = − Q div S ( a ) \partial_tz(a)=-\mathbb Q\operatorname{div}S(a) ∂ t z ( a ) = − Q div S ( a ) ; this is why the cutoff χ \chi χ vanishes near a a a .)
Space. We prove by induction on k ≥ 0 k\ge0 k ≥ 0 that J k z ∈ M R q ( Σ ) J^kz\in\mathrm{MR}_q(\Sigma) J k z ∈ MR q ( Σ ) , that is, z ∈ L 2 ( Σ ; H k + 2 , q ) z\in L^2(\Sigma;H^{k+2,q}) z ∈ L 2 ( Σ ; H k + 2 , q ) and ∂ t z ∈ L 2 ( Σ ; H k , q ) \partial_tz\in L^2(\Sigma;H^{k,q}) ∂ t z ∈ L 2 ( Σ ; H k , q ) . By Lemma 5.4(b),(c), applied to J k z J^kz J k z , this implies z ∈ C ( Σ ; W k + 1 , q ) z\in C(\Sigma;W^{k+1,q}) z ∈ C ( Σ ; W k + 1 , q ) . The case k = 0 k=0 k = 0 is z ∈ Y 0 ( Σ ) z\in Y_0(\Sigma) z ∈ Y 0 ( Σ ) . Assume the claim for some k k k , and let G G G be the right-hand side of the heat equation solved by z z z ,
∂ t z − Δ z = G : = Q [ − div S + g z − div ( U ⊗ z + z ⊗ U + z ⊗ z ) ] . \partial_tz-\Delta z=G:=\mathbb Q\left[-\operatorname{div}S+g_z-\operatorname{div}(U\mathbin{\mathop{\otimes}}z+z\mathbin{\mathop{\otimes}}U+z\mathbin{\mathop{\otimes}}z)\right]. ∂ t z − Δ z = G := Q [ − div S + g z − div ( U ⊗ z + z ⊗ U + z ⊗ z ) ] .
Since H k + 2 , q ↪ C k + 1 H^{k+2,q}\hookrightarrow C^{k+1} H k + 2 , q ↪ C k + 1 for q > d q>d q > d , the derivatives in y = ( x , v ) y=(x,v) y = ( x , v ) of order 1 , … , k + 1 1,\ldots,k+1 1 , … , k + 1 of the phase-space field of U ^ z = U + χ ⊗ ∘ z + c z \widehat U_z=U+\chi\mathbin{\mathop{\otimes}^{\circ}}z+c_z U z = U + χ ⊗ ∘ z + c z are bounded by a function in L ∞ ( Σ ) L^\infty(\Sigma) L ∞ ( Σ ) . By differentiating the equation of the flow, the maps Z U ^ z ( a ; t , ⋅ ) Z_{\widehat U_z}(a;t,\cdot) Z U z ( a ; t , ⋅ ) are of class C k + 1 C^{k+1} C k + 1 , with derivatives bounded on compact sets uniformly in t ∈ Σ t\in\Sigma t ∈ Σ . Hence f ^ z ( t ) \widehat f_z(t) f z ( t ) is bounded in C k + 1 C^{k+1} C k + 1 , uniformly in t t t , with velocity support in a fixed ball, and so are its density and momentum in C k + 1 ( T d ) C^{k+1}(\mathbb T^d) C k + 1 ( T d ) . Since U ^ z ∈ L ∞ ( Σ ; W k + 1 , q ) \widehat U_z\in L^\infty(\Sigma;W^{k+1,q}) U z ∈ L ∞ ( Σ ; W k + 1 , q ) , we get g z ∈ L ∞ ( Σ ; W k + 1 , q ) g_z\in L^\infty(\Sigma;W^{k+1,q}) g z ∈ L ∞ ( Σ ; W k + 1 , q ) . Moreover H k + 1 , q H^{k+1,q} H k + 1 , q is an algebra, because q > d q>d q > d , so div ( z ⊗ z ) = z ⋅ ∇ z ∈ L 2 ( Σ ; H k + 1 , q ) \operatorname{div}(z\mathbin{\mathop{\otimes}}z)=z\cdot\nabla z\in L^2(\Sigma;H^{k+1,q}) div ( z ⊗ z ) = z ⋅ ∇ z ∈ L 2 ( Σ ; H k + 1 , q ) ; the terms containing U U U and S S S belong to the same space. Thus G ∈ L 2 ( Σ ; H k + 1 , q ) G\in L^2(\Sigma;H^{k+1,q}) G ∈ L 2 ( Σ ; H k + 1 , q ) . By Lemma 5.4(a), the heat equation with forcing J k + 1 G J^{k+1}G J k + 1 G and zero datum at a a a has a solution ζ ∈ M R q ( Σ ) \zeta\in\mathrm{MR}_q(\Sigma) ζ ∈ MR q ( Σ ) , and J − ( k + 1 ) ζ = z J^{-(k+1)}\zeta=z J − ( k + 1 ) ζ = z by uniqueness. This is the claim for k + 1 k+1 k + 1 .
Time. By the first part, z z z , and hence U ^ z \widehat U_z U z , belong to C ( Σ ; W m , q ) C(\Sigma;W^{m,q}) C ( Σ ; W m , q ) for every m m m . Therefore the flow of U ^ z \widehat U_z U z and all its derivatives in y y y are continuous in ( t , y ) (t,y) ( t , y ) , and f ^ z ∈ C ( Σ ; C m ) \widehat f_z\in C(\Sigma;C^m) f z ∈ C ( Σ ; C m ) for every m m m . We prove by induction on j ≥ 0 j\ge0 j ≥ 0 that z ∈ C j ( Σ ; W m , q ) z\in C^j(\Sigma;W^{m,q}) z ∈ C j ( Σ ; W m , q ) for every m m m . If this holds for some j j j , it follows that f ^ z ∈ C j ( Σ ; C m ) \widehat f_z\in C^j(\Sigma;C^m) f z ∈ C j ( Σ ; C m ) for every m m m . Since ∂ t f ^ z = − div v ( b ^ z f ^ z ) \partial_t\widehat f_z=-\operatorname{div}_v(\widehat b_z\widehat f_z) ∂ t f z = − div v ( b z f z ) , a second induction shows that f ^ z ∈ C j + 1 ( Σ ; C m ) \widehat f_z\in C^{j+1}(\Sigma;C^m) f z ∈ C j + 1 ( Σ ; C m ) for every m m m . Hence g z ∈ C j ( Σ ; W m , q ) g_z\in C^j(\Sigma;W^{m,q}) g z ∈ C j ( Σ ; W m , q ) , so G ∈ C j ( Σ ; W m , q ) G\in C^j(\Sigma;W^{m,q}) G ∈ C j ( Σ ; W m , q ) , and ∂ t z = Δ z + G ∈ C j ( Σ ; W m , q ) \partial_tz=\Delta z+G\in C^j(\Sigma;W^{m,q}) ∂ t z = Δ z + G ∈ C j ( Σ ; W m , q ) for every m m m . This is the claim for j + 1 j+1 j + 1 .
Proposition 6.9 (Mean-preserving gluing). Let z z z be a segment corrector, c = c z c=c_z c = c z , U ^ = U ^ z \widehat U=\widehat U_z U = U z , f ^ = f ^ z \widehat f=\widehat f_z f = f z , and let q z : = Δ − 1 div ( − div S + g z − div ( U ⊗ z + z ⊗ U + z ⊗ z ) ) q_z:=\Delta^{-1}\operatorname{div}(-\operatorname{div}S+g_z-\operatorname{div}(U\mathbin{\mathop{\otimes}}z+z\mathbin{\mathop{\otimes}}U+z\mathbin{\mathop{\otimes}}z)) q z := Δ − 1 div ( − div S + g z − div ( U ⊗ z + z ⊗ U + z ⊗ z )) , so that ∂ t z − Δ z + div ( U ⊗ z + z ⊗ U + z ⊗ z ) + ∇ q z = − div S + g z \partial_tz-\Delta z+\operatorname{div}(U\mathbin{\mathop{\otimes}}z+z\mathbin{\mathop{\otimes}}U+z\mathbin{\mathop{\otimes}}z)+\nabla q_z=-\operatorname{div}S+g_z ∂ t z − Δ z + div ( U ⊗ z + z ⊗ U + z ⊗ z ) + ∇ q z = − div S + g z exactly; q z q_z q z is smooth on Σ \Sigma Σ . On Σ \Sigma Σ ,
∂ t U ^ − Δ U ^ + div ( U ^ ⊗ U ^ ) + ∇ ( P + χ q z ) − F [ f ^ , U ^ ] = χ ′ z + ( χ 2 − χ ) div ( z ⊗ z ) + ( 1 − χ ) [ div S + c ⋅ ∇ U + c ′ − g z ] . (46) \partial_t\widehat U-\Delta\widehat U+\operatorname{div}(\widehat U\mathbin{\mathop{\otimes}}\widehat U)+\nabla(P+\chi q_z)-F[\widehat f,\widehat U]=\chi' z+(\chi^2-\chi)\operatorname{div}(z\mathbin{\mathop{\otimes}}z)+(1-\chi)[\operatorname{div}S+c\cdot\nabla U+c'-g_z]. \tag*{(46)} ∂ t U − Δ U + div ( U ⊗ U ) + ∇ ( P + χ q z ) − F [ f , U ] = χ ′ z + ( χ 2 − χ ) div ( z ⊗ z ) + ( 1 − χ ) [ div S + c ⋅ ∇ U + c ′ − g z ] . (46) Since c ′ = ∫ g z c'= \int g_z c ′ = ∫ g z , the right-hand side has zero mean. Hence ( U ^ , f ^ ) (\widehat U,\widehat f) ( U , f ) is a smooth VNS–Reynolds solution on Σ \Sigma Σ , with stress
S ^ = χ ′ R z + ( χ 2 − χ ) ⊗ ∘ z ⊗ ∘ z + ( 1 − χ ) ( S + U ⊗ ∘ c + c ⊗ ∘ U − R g z ) , \widehat S=\chi'Rz+(\chi^2-\chi)\mathbin{\mathop{\otimes}^{\circ}}z\mathbin{\mathop{\otimes}^{\circ}}z+(1-\chi)(S+U\mathbin{\mathop{\otimes}^{\circ}}c+c\mathbin{\mathop{\otimes}^{\circ}}U-Rg_z), S = χ ′ R z + ( χ 2 − χ ) ⊗ ∘ z ⊗ ∘ z + ( 1 − χ ) ( S + U ⊗ ∘ c + c ⊗ ∘ U − R g z ) , and pressure P + χ q z − 1 d ( χ 2 − χ ) ∣ z ∣ 2 − 2 d ( 1 − χ ) U ⋅ c P+\chi q_z-\frac{1}{d}(\chi^2-\chi)|z|^2-\frac{2}{d}(1-\chi)U\cdot c P + χ q z − d 1 ( χ 2 − χ ) ∣ z ∣ 2 − d 2 ( 1 − χ ) U ⋅ c . In particular S ^ = 0 \widehat S=0 S = 0 where χ = 1 \chi=1 χ = 1 .
If χ \chi χ is two-sided, define the exit by the mean: after b b b , set U ~ : = U + c ( t ) \widetilde U:=U+c(t) U := U + c ( t ) , where c c c solves the Volterra equation
c ′ = ∫ ( F [ f [ U ] [ 0 , a ] ⊕ U ^ , U ^ ] − F [ f [ U ] , U ] ) , (47) c'=\int\left(F[f[U]_{[0,a]}\oplus\widehat U,\widehat U]-F[f[U],U]\right), \tag*{(47)} c ′ = ∫ ( F [ f [ U ] [ 0 , a ] ⊕ U , U ] − F [ f [ U ] , U ] ) , (47) continuing c z c_z c z ; here U ^ = U ^ z \widehat U=\widehat U_z U = U z on Σ \Sigma Σ and U ~ = U + c \widetilde U=U+c U = U + c on [ b , 1 ] [b,1] [ b , 1 ] , so the kinetic state is f ^ z \widehat f_z f z continued. Then (6.17) holds with χ = 0 \chi=0 χ = 0 on [ b , 1 ] [b,1] [ b , 1 ] , where g z g_z g z is understood as the continued drag difference F [ f ^ , U ^ ] − F [ f [ U ] , U ] F[\widehat f,\widehat U]-F[f[U],U] F [ f , U ] − F [ f [ U ] , U ] , and the glued field is a smooth VNS–Reynolds solution on [ 0 , 1 ] [0,1] [ 0 , 1 ] .
Proof. Write U ^ = U + a \widehat U=U+a U = U + a with a = χ z ˚ + c a=\chi\mathring z+c a = χ z ˚ + c , and ( b ⋅ ∇ g ) i = b j ∂ j g i (b\cdot\nabla g)_i=b_j\partial_jg_i ( b ⋅ ∇ g ) i = b j ∂ j g i . Since div U = div z ˚ = 0 \operatorname{div}U=\operatorname{div}\mathring z=0 div U = div z ˚ = 0 and c c c depends only on t t t ,
div ( U ⊗ a + a ⊗ U ) = χ ( z ˚ ⋅ ∇ U + U ⋅ ∇ z ˚ ) + c ⋅ ∇ U , div ( a ⊗ a ) = χ 2 z ˚ ⋅ ∇ z ˚ + χ c ⋅ ∇ z ˚ . \operatorname{div}(U\otimes a+a\otimes U)=\chi(\mathring z\cdot\nabla U+U\cdot\nabla\mathring z)+c\cdot\nabla U,\qquad\operatorname{div}(a\otimes a)=\chi^2\mathring z\cdot\nabla\mathring z+\chi c\cdot\nabla\mathring z. div ( U ⊗ a + a ⊗ U ) = χ ( z ˚ ⋅ ∇ U + U ⋅ ∇ z ˚ ) + c ⋅ ∇ U , div ( a ⊗ a ) = χ 2 z ˚ ⋅ ∇ z ˚ + χc ⋅ ∇ z ˚ .
Hence, by the equation of the base,
∂ t U ^ − Δ U ^ + div ( U ^ ⊗ U ^ ) + ∇ P − F [ f [ U ] , U ] − div S = c ′ z ˚ + c ′ + c ⋅ ∇ U + χ [ ∂ t z ˚ − Δ z ˚ + z ˚ ⋅ ∇ U + U ⋅ ∇ z ˚ + c ⋅ ∇ z ˚ ] + χ 2 z ˚ ⋅ ∇ z ˚ . \begin{aligned}
\partial_t\widehat U-\Delta\widehat U+\operatorname{div}(\widehat U\otimes\widehat U)+\nabla P-F[f[U],U]-\operatorname{div}S
={}&c'\mathring z+c'+c\cdot\nabla U\\
&+\chi\left[\partial_t\mathring z-\Delta\mathring z+\mathring z\cdot\nabla U+U\cdot\nabla\mathring z+c\cdot\nabla\mathring z\right]+\chi^2\mathring z\cdot\nabla\mathring z.
\end{aligned} ∂ t U − Δ U + div ( U ⊗ U ) + ∇ P − F [ f [ U ] , U ] − div S = c ′ z ˚ + c ′ + c ⋅ ∇ U + χ [ ∂ t z ˚ − Δ z ˚ + z ˚ ⋅ ∇ U + U ⋅ ∇ z ˚ + c ⋅ ∇ z ˚ ] + χ 2 z ˚ ⋅ ∇ z ˚ .
On the other hand, (6.4), written with its pressure q z q_z q z and with z = z ˚ + c z=\mathring z+c z = z ˚ + c , reads
∂ t z ˚ − Δ z ˚ + z ˚ ⋅ ∇ U + U ⋅ ∇ z ˚ + c ⋅ ∇ z ˚ + ∇ q z = − div S + g z − c ′ − c ⋅ ∇ U − z ˚ ⋅ ∇ z ˚ . \partial_t\mathring z-\Delta\mathring z+\mathring z\cdot\nabla U+U\cdot\nabla\mathring z+c\cdot\nabla\mathring z+\nabla q_z=-\operatorname{div}S+g_z-c'-c\cdot\nabla U-\mathring z\cdot\nabla\mathring z. ∂ t z ˚ − Δ z ˚ + z ˚ ⋅ ∇ U + U ⋅ ∇ z ˚ + c ⋅ ∇ z ˚ + ∇ q z = − div S + g z − c ′ − c ⋅ ∇ U − z ˚ ⋅ ∇ z ˚ .
Inserting this into the bracket, and using F [ f ^ , U ^ ] = F [ f [ U ] , U ] + g z F[\widehat f,\widehat U]=F[f[U],U]+g_z F [ f , U ] = F [ f [ U ] , U ] + g z and z ˚ ⋅ ∇ z ˚ = div ( z ˚ ⊗ z ˚ ) \mathring z\cdot\nabla\mathring z=\operatorname{div}(\mathring z\otimes\mathring z) z ˚ ⋅ ∇ z ˚ = div ( z ˚ ⊗ z ˚ ) , we obtain (46) : the terms c ′ + c ⋅ ∇ U c'+c\cdot\nabla U c ′ + c ⋅ ∇ U , div S \operatorname{div}S div S and g z g_z g z appear with the factor 1 − χ 1-\chi 1 − χ , and the quadratic term with the factor χ 2 − χ \chi^2-\chi χ 2 − χ .
The stress form follows from χ ′ z ˚ = div ( χ ′ R z ) \chi'\mathring z=\operatorname{div}(\chi'\mathcal Rz) χ ′ z ˚ = div ( χ ′ R z ) , div ( z ˚ ⊗ z ˚ ) = div ( z ˚ ⊗ z ˚ ) + 1 d ∇ ∣ z ˚ ∣ 2 \operatorname{div}(\mathring z\otimes\mathring z)=\operatorname{div}(\mathring z\otimes\mathring z)+\frac{1}{d}\nabla|\mathring z|^2 div ( z ˚ ⊗ z ˚ ) = div ( z ˚ ⊗ z ˚ ) + d 1 ∇∣ z ˚ ∣ 2 , c ⋅ ∇ U = div ( U ⊗ c + c ⊗ U ) + 2 d ∇ ( U ⋅ c ) c\cdot\nabla U=\operatorname{div}(U\otimes c+c\otimes U)+\frac{2}{d}\nabla(U\cdot c) c ⋅ ∇ U = div ( U ⊗ c + c ⊗ U ) + d 2 ∇ ( U ⋅ c ) , and c ′ − g z = − div R g z c'-g_z=-\operatorname{div}\mathcal Rg_z c ′ − g z = − div R g z . The gradient terms are moved to the left-hand side, which produces the minus signs in the pressure.
Near a a a , where χ = 0 \chi=0 χ = 0 , we have U ^ z = U + c z \widehat U_z=U+c_z U z = U + c z , so c z c_z c z solves the Volterra equation c ′ = f ( F [ f [ U ∣ [ 0 , a ] ⊕ ( U + c ) ] , U + c ] − F [ f [ U ] , U ] ) c'=f(F[f[U|_{[0,a]}\oplus(U+c)],U+c]-F[f[U],U]) c ′ = f ( F [ f [ U ∣ [ 0 , a ] ⊕ ( U + c )] , U + c ] − F [ f [ U ] , U ]) with c ( a ) = 0 c(a)=0 c ( a ) = 0 . The function c ≡ 0 c\equiv0 c ≡ 0 solves it, so c z = 0 c_z=0 c z = 0 there by uniqueness (Lemma 3.8, Step 2). Hence U ^ = U \widehat U=U U = U , f ^ = f [ U ] \widehat f=f[U] f = f [ U ] and S ^ = S \widehat S=S S = S near a a a , and the glued solution is smooth across a a a .
After b b b , (6.18) is the equation of Lemma 3.8 started at b b b , written for c c c alone, and it has a unique smooth solution. On the zone where χ = 0 \chi=0 χ = 0 near b b b , U ^ \widehat U U depends on z z z only through c z c_z c z , which satisfies the same equation. So the continuation is smooth.
Lemma 6.10 (Junctions). Suppose a segment exits by the mean. Suppose a later segment or convex integration step (Lemma 6.13), whose modifications vanish near its start, begins in the zone where the first cutoff vanishes. Then, on the overlap, the glued velocity is B + c t o t ( t ) B+c_{\mathrm{tot}}(t) B + c tot ( t ) . Here B B B is the base before the first segment, and c t o t c_{\mathrm{tot}} c tot is the unique solution of (6.18) relative to B B B . In particular, the glued solution is smooth across the junction.
Proof. After the second object starts, its mean satisfies c n e w ′ = f ( F [ f ^ , U + c n e w ] − F [ f [ U ] , U ] ) c'_{\mathrm{new}}=f(F[\widehat f,U+c_{\mathrm{new}}]-F[f[U],U]) c new ′ = f ( F [ f , U + c new ] − F [ f [ U ] , U ]) , where U = B + c p r e v U=B+c_{\mathrm{prev}} U = B + c prev . Adding the equation of c p r e v c_{\mathrm{prev}} c prev , the middle terms cancel, so c p r e v + c n e w c_{\mathrm{prev}}+c_{\mathrm{new}} c prev + c new solves (6.18) relative to B B B with the true kinetic state. Before the start, c t o t = c p r e v c_{\mathrm{tot}}=c_{\mathrm{prev}} c tot = c prev solves the same equation. Conclude by uniqueness (Lemma 3.8, Step 2).
Corollary 6.11 (Layer stress). Let χ \chi χ be monotone on each transition layer, with total transition length λ χ \lambda_\chi λ χ , so that ∥ χ ′ ∥ L 1 ≤ 2 \|\chi'\|_{L^1}\le2 ∥ χ ′ ∥ L 1 ≤ 2 and ∥ χ − χ 2 ∥ L 1 + ∥ 1 − χ ∥ L 1 ( Σ ) ≤ λ χ \|\chi-\chi^2\|_{L^1}+\|1-\chi\|_{L^1(\Sigma)}\le\lambda_\chi ∥ χ − χ 2 ∥ L 1 + ∥1 − χ ∥ L 1 ( Σ ) ≤ λ χ on the transition and zero zones inside Σ \Sigma Σ . Then
∥ S ^ ∥ L 1 ( Σ ; L r ) ≤ 2 sup Σ ∥ R z ∥ L r + λ χ ( sup Σ ∥ S ∥ L r + sup Σ ∥ z ˚ ∥ L ∞ 2 + 2 C E B U sup Σ ∣ c ∣ + sup Σ ∥ R g z ∥ L q ) . \|\widehat S\|_{L^1(\Sigma;L^r)}\le2\sup_{\Sigma}\|\mathcal Rz\|_{L^r}+\lambda_\chi\left(\sup_{\Sigma}\|S\|_{L^r}+\sup_{\Sigma}\|\mathring z\|_{L^\infty}^2+2C_EB_U\sup_{\Sigma}|c|+\sup_{\Sigma}\|\mathcal Rg_z\|_{L^q}\right). ∥ S ∥ L 1 ( Σ ; L r ) ≤ 2 Σ sup ∥ R z ∥ L r + λ χ ( Σ sup ∥ S ∥ L r + Σ sup ∥ z ˚ ∥ L ∞ 2 + 2 C E B U Σ sup ∣ c ∣ + Σ sup ∥ R g z ∥ L q ) .
Proof. Bound each term of S ^ \widehat S S by the L 1 L^1 L 1 norm in time of its cutoff factor times the supremum of the remaining factor.
The persistent stress of a single segment is bounded in W 1 , q W^{1,q} W 1 , q by a constant, but not by a small one: the drag difference g z g_z g z is bounded pointwise, while the kinetic states differ by an amount that is small only in a weak norm. These bounds are not summable over many segments. Instead we use that the drag differences telescope:
Lemma 6.12 (Accumulated persistent stress). Let ( U 0 , P 0 , S 0 ) (U_0,P_0,S_0) ( U 0 , P 0 , S 0 ) be a smooth VNS–Reynolds solution, and let ( U l , P l , S l ) (U_l,P_l,S_l) ( U l , P l , S l ) , l = 1 , … , n l=1,\ldots,n l = 1 , … , n , be obtained by successive applications of Propositions 6.8 and 6.9 with exit by the mean, or of Lemma 6.13, each starting after the previous one has exited (after sup Z1 for Lemma 6.13). Let t t t be a time after the last exit, and let Π l \Pi_l Π l be the persistent stress of the l l l -th operation (Proposition 6.8(d), respectively c ⊗ U + U ⊗ c − R h c\otimes U+U\otimes c-\mathcal Rh c ⊗ U + U ⊗ c − R h in Lemma 6.13). Then U n ( t ) = U 0 ( t ) + C n ( t ) U_n(t)=U_0(t)+C_n(t) U n ( t ) = U 0 ( t ) + C n ( t ) with C n = ∑ l ≤ n c l C_n=\sum_{l\le n}c_l C n = ∑ l ≤ n c l spatially constant, and
∑ l ≤ n Π l ( t ) = U 0 ⊗ C n + C n ⊗ U 0 − R ( F [ f [ U n ] , U n ] − F [ f [ U 0 ] , U 0 ] ) ( t ) + M n ( t ) , \sum_{l\le n}\Pi_l(t)=U_0\otimes C_n+C_n\otimes U_0-\mathcal R\bigl(F[f[U_n],U_n]-F[f[U_0],U_0]\bigr)(t)+M_n(t), l ≤ n ∑ Π l ( t ) = U 0 ⊗ C n + C n ⊗ U 0 − R ( F [ f [ U n ] , U n ] − F [ f [ U 0 ] , U 0 ] ) ( t ) + M n ( t ) ,
where M n ( t ) M_n(t) M n ( t ) is a spatially constant traceless matrix with ∣ M n ( t ) ∣ ≤ 2 max l ≤ n ∣ C l ( t ) ∣ ∑ l ≤ n ∣ c l ( t ) ∣ |M_n(t)| \le2\max_{l\le n}|C_l(t)|\sum_{l\le n}|c_l(t)| ∣ M n ( t ) ∣ ≤ 2 max l ≤ n ∣ C l ( t ) ∣ ∑ l ≤ n ∣ c l ( t ) ∣ . In particular, if R 1 v R^v_1 R 1 v bounds the velocity supports of all kinetic states involved and ϱ , C h \varrho, C_h ϱ , C h are as in Section 6.3, then
∥ ∑ l ≤ n Π l ( t ) ∥ W 1 , q ≤ 2 ∥ U 0 ( t ) ∥ W 1 , q ∣ C n ( t ) ∣ + C ( C h + ϱ ∣ C n ( t ) ∣ ) + ∣ M n ( t ) ∣ . \left\|\sum_{l\le n}\Pi_l(t)\right\|_{W^{1,q}} \le2\|U_0(t)\|_{W^{1,q}}|C_n(t)|+C(C_h+\varrho|C_n(t)|)+|M_n(t)|. l ≤ n ∑ Π l ( t ) W 1 , q ≤ 2∥ U 0 ( t ) ∥ W 1 , q ∣ C n ( t ) ∣ + C ( C h + ϱ ∣ C n ( t ) ∣ ) + ∣ M n ( t ) ∣.
with C C C depending only on d d d , q q q , and independent of n n n . The stress is not modified: M n M_n M n is kept, and since div M n = 0 \operatorname{div}M_n=0 div M n = 0 it plays no role in the Reynolds equation; it only enters norms through the last term.
Proof. After the l l l -th exit the base of the ( l + 1 ) (l+1) ( l + 1 ) -st operation is the glued field U l = U l − 1 + c l U_l=U_{l-1}+c_l U l = U l − 1 + c l with its kinetic state f [ U l ] f[U_l] f [ U l ] . Hence the drag difference of the l l l -th operation is g l = F [ f [ U l ] , U l ] − F [ f [ U l − 1 ] , U l − 1 ] g_l=F[f[U_l],U_l]-F[f[U_{l-1}],U_{l-1}] g l = F [ f [ U l ] , U l ] − F [ f [ U l − 1 ] , U l − 1 ] at time t t t , and ∑ l g l \sum_l g_l ∑ l g l telescopes. For the cross terms, U l − 1 = U 0 + C l − 1 U_{l-1}=U_0+C_{l-1} U l − 1 = U 0 + C l − 1 , so ∑ l ( U l − 1 ⊗ ˚ c l + c l ⊗ ˚ U l − 1 ) = U 0 ⊗ ˚ C n + C n ⊗ ˚ U 0 + M n \sum_l(U_{l-1}\mathring{\otimes}c_l+c_l\mathring{\otimes}U_{l-1})=U_0\mathring{\otimes}C_n+C_n\mathring{\otimes}U_0+M_n ∑ l ( U l − 1 ⊗ ˚ c l + c l ⊗ ˚ U l − 1 ) = U 0 ⊗ ˚ C n + C n ⊗ ˚ U 0 + M n with M n = ∑ l ( C l − 1 ⊗ ˚ c l + c l ⊗ ˚ C l − 1 ) M_n=\sum_l(C_{l-1}\mathring{\otimes}c_l+c_l\mathring{\otimes}C_{l-1}) M n = ∑ l ( C l − 1 ⊗ ˚ c l + c l ⊗ ˚ C l − 1 ) , and ∣ a ⊗ ˚ b ∣ ≤ ∣ a ∣ ∣ b ∣ |a\mathring{\otimes}b|\le|a||b| ∣ a ⊗ ˚ b ∣ ≤ ∣ a ∣∣ b ∣ gives the bound on ∣ M n ∣ |M_n| ∣ M n ∣ . The bound follows from Lemma 6.4(e), ∥ R g ∥ W 1 , q ≤ C ∥ g ∥ L q \|\mathcal{R}g\|_{W^{1,q}}\le C\|g\|_{L^q} ∥ R g ∥ W 1 , q ≤ C ∥ g ∥ L q , and the pointwise bound ∣ F [ f [ U n ] , U n ] − F [ f [ U 0 ] , U 0 ] ∣ ≤ C h + ϱ ∣ C n ∣ |F[f[U_n],U_n]-F[f[U_0],U_0]|\le C_h+\varrho|C_n| ∣ F [ f [ U n ] , U n ] − F [ f [ U 0 ] , U 0 ] ∣ ≤ C h + ϱ ∣ C n ∣ , as in Proposition 6.8(d).
The localized convex integration step
Lemma 6.13 (Localized CI step). Let ( U , P , S ) (U,P,S) ( U , P , S ) be a smooth VNS–Reynolds solution with B U , Λ U < ∞ B_U,\Lambda_U<\infty B U , Λ U < ∞ . Let Z 0 ⊂ Z 1 ⊂ Z ‾ 1 ⊂ [ 0 , 1 ] Z_0\subset Z_1\subset\overline{Z}_1\subset[0,1] Z 0 ⊂ Z 1 ⊂ Z 1 ⊂ [ 0 , 1 ] be intervals, let θ 0 ∈ C c ∞ ( Z 1 ; [ 0 , 1 ] ) \theta_0\in C^\infty_c(Z_1;[0,1]) θ 0 ∈ C c ∞ ( Z 1 ; [ 0 , 1 ]) equal 1 on Z 0 Z_0 Z 0 , and let θ 1 ∈ C c ∞ ( Z 1 ; [ 0 , 1 ] ) \theta_1\in C^\infty_c(Z_1;[0,1]) θ 1 ∈ C c ∞ ( Z 1 ; [ 0 , 1 ]) equal 1 on Z ‾ 1 \overline{Z}_1 Z 1 . Set S c : = θ 0 S S_c:=\theta_0S S c := θ 0 S and S s : = ( 1 − θ 0 ) S S_s:=(1-\theta_0)S S s := ( 1 − θ 0 ) S . For every δ ′ > 0 \delta'>0 δ ′ > 0 there are smooth w w w and c c c with the following properties:
U ♯ = U + w + c U^\sharp=U+w+c U ♯ = U + w + c is a smooth VNS–Reynolds solution with stress S ♯ = S C L + S s + c ⊗ ˚ ( U + w ) + ( U + w ) ⊗ ˚ c − R h S^\sharp=S^{\mathrm{CL}}+S_s+c\mathring{\otimes}(U+w)+(U+w)\mathring{\otimes}c-\mathcal{R}h S ♯ = S CL + S s + c ⊗ ˚ ( U + w ) + ( U + w ) ⊗ ˚ c − R h , h = F [ f [ U ♯ ] , U ♯ ] − F [ f [ U ] , U ] h=F[f[U^\sharp],U^\sharp]-F[f[U],U] h = F [ f [ U ♯ ] , U ♯ ] − F [ f [ U ] , U ] ;
supp t S C L ⊂ Z 1 \operatorname{supp}_t S^{\mathrm{CL}}\subset Z_1 supp t S CL ⊂ Z 1 , ∥ S C L ∥ L 1 L r ≤ δ ′ \|S^{\mathrm{CL}}\|_{L^1L^r}\le\delta' ∥ S CL ∥ L 1 L r ≤ δ ′ , ∥ w ∥ L p L ∞ ∩ L 1 W 1 , q ≤ δ ′ \|w\|_{L^pL^\infty\cap L^1W^{1,q}}\le\delta' ∥ w ∥ L p L ∞ ∩ L 1 W 1 , q ≤ δ ′ , and ∥ w ∥ L 2 ≤ M ∥ S c ∥ L t , x 1 1 / 2 \|w\|_{L^2}\le M\|S_c\|_{L^1_{t,x}}^{1/2} ∥ w ∥ L 2 ≤ M ∥ S c ∥ L t , x 1 1/2 ;
sup t ∣ c ( t ) ∣ ≤ K ♯ δ ′ \sup_t|c(t)|\le K^\sharp\delta' sup t ∣ c ( t ) ∣ ≤ K ♯ δ ′ , ∥ R h ∥ L 1 ( 0 , 1 ; L q ) ≤ K ♯ δ ′ \|\mathcal{R}h\|_{L^1(0,1;L^q)}\le K^\sharp\delta' ∥ R h ∥ L 1 ( 0 , 1 ; L q ) ≤ K ♯ δ ′ ; and for t ∉ Z 1 t\notin Z_1 t ∈ / Z 1 , ∥ R h ( t ) ∥ L q ≤ K ♯ ′ \|\mathcal{R}h(t)\|_{L^q}\le K^{\sharp\prime} ∥ R h ( t ) ∥ L q ≤ K ♯′ and ∥ c ⊗ ˚ U + U ⊗ ˚ c − R h ∥ W 1 , q ≤ C ♯ \|c\mathring{\otimes}U+U\mathring{\otimes}c-\mathcal{R}h\|_{W^{1,q}}\le C^\sharp ∥ c ⊗ ˚ U + U ⊗ ˚ c − R h ∥ W 1 , q ≤ C ♯ . The constants K ♯ , C ♯ K^\sharp,C^\sharp K ♯ , C ♯ depend only on d , q , R 0 v , M ∞ d,q,R^v_0,M_\infty d , q , R 0 v , M ∞ and upper bounds for B U , Λ U , L U B_U,\Lambda_U,L_U B U , Λ U , L U , defined by (6.3) with a = inf Z 1 a=\inf Z_1 a = inf Z 1 .
Proof. Apply Proposition 2.8 to ( U , S c ) (U,S_c) ( U , S c ) , with the two-sided cutoff θ 1 \theta_1 θ 1 in place of (A.9). The algebra of Appendix A uses only θ 1 S c = S c \theta_1S_c=S_c θ 1 S c = S c , see (A.10), which holds since θ 1 = 1 \theta_1=1 θ 1 = 1 on supp S c \operatorname{supp}S_c supp S c . The estimates of Appendix B use only the smoothness of θ 1 \theta_1 θ 1 and 0 ≤ θ 1 ≤ 1 0\le\theta_1\le1 0 ≤ θ 1 ≤ 1 ; the constants C u C_u C u now depend on θ 1 \theta_1 θ 1 , while M M M does not. Since ∥ ∂ t k θ 1 ∥ ∞ ≲ k τ − k \|\partial_t^k\theta_1\|_\infty\lesssim_k\tau^{-k} ∥ ∂ t k θ 1 ∥ ∞ ≲ k τ − k , these constants depend on τ \tau τ ; this is admissible, since the frequency λ \lambda λ of the step is chosen after τ \tau τ (see the order of choices at the end of the proof of Proposition 6.3). The stress S s S_s S s is carried along unchanged: Proposition 2.8(i) holds with S c S_c S c in place of R R R , and adding div S s \operatorname{div}S_s div S s to both sides shows that the new stress is S C L + S s S^{\mathrm{CL}}+S_s S CL + S s , up to the kinetic terms treated below. Then proceed as in Section 3.6, starting Lemma 3.8 at t 0 : = inf Z 1 t_0:=\inf Z_1 t 0 := inf Z 1 from the kinetic state f [ U ] ( t 0 ) f[U](t_0) f [ U ] ( t 0 ) .
For the kinetic bounds, apply Remark 6.7 with b = U b=U b = U , a = w + c a=w+c a = w + c . The momentum identity and Lemma 6.5(a) with ψ = v i \psi=v_i ψ = v i give ∣ c ( t ) ∣ ≤ K ( ∥ w ∥ L 1 ( 0 , t ; L q ) + ∫ t 0 t ∣ c ∣ ) |c(t)|\le K(\|w\|_{L^1(0,t;L^q)}+\int_{t_0}^t|c|) ∣ c ( t ) ∣ ≤ K ( ∥ w ∥ L 1 ( 0 , t ; L q ) + ∫ t 0 t ∣ c ∣ ) , hence sup ∣ c ∣ ≤ K e K δ ′ \sup|c|\le K e^{K\delta'} sup ∣ c ∣ ≤ K e K δ ′ by Gronwall, using ∥ w ∥ L 1 L q ≤ ∥ w ∥ L p L ∞ ≤ δ ′ \|w\|_{L^1L^q}\le\|w\|_{L^pL^\infty}\le\delta' ∥ w ∥ L 1 L q ≤ ∥ w ∥ L p L ∞ ≤ δ ′ . Here K K K is the constant of Lemma 6.5 for the radius R 1 v = R 0 v + L U + 2 ( C E B U + 2 ) R^v_1=R^v_0+L_U+2(C_EB_U+2) R 1 v = R 0 v + L U + 2 ( C E B U + 2 ) , which is admissible as long as ∥ w ∥ L 1 L ∞ + ∫ t 0 t ∣ c ∣ ≤ 2 \|w\|_{L^1L^\infty}+\int_{t_0}^t|c|\le2 ∥ w ∥ L 1 L ∞ + ∫ t 0 t ∣ c ∣ ≤ 2 . As in the proof of Proposition 6.8(c), a continuity argument in t t t shows that this holds on [ t 0 , 1 ] [t_0,1] [ t 0 , 1 ] if δ ′ ≤ 1 \delta'\le1 δ ′ ≤ 1 and K e K δ ′ ≤ 1 Ke^{K\delta'}\le1 K e K δ ′ ≤ 1 , which we assume. By (6.2),
∥ R h ( t ) ∥ L q ≤ C q K ( 1 + R 1 v + ∥ U ( t ) ∥ W 1 , q ) ( ∫ t 0 t ∥ w ∥ L q + ∫ t 0 t ∣ c ∣ ) + C q ϱ ( ∥ w ( t ) ∥ L q + ∣ c ( t ) ∣ ) . \|\mathcal{R}h(t)\|_{L^q}\le C_qK(1+R^v_1+\|U(t)\|_{W^{1,q}})\left(\int_{t_0}^t\|w\|_{L^q}+\int_{t_0}^t|c|\right)+C_q\varrho(\|w(t)\|_{L^q}+|c(t)|). ∥ R h ( t ) ∥ L q ≤ C q K ( 1 + R 1 v + ∥ U ( t ) ∥ W 1 , q ) ( ∫ t 0 t ∥ w ∥ L q + ∫ t 0 t ∣ c ∣ ) + C q ϱ ( ∥ w ( t ) ∥ L q + ∣ c ( t ) ∣ ) .
The instantaneous term ϱ ∥ w ( t ) ∥ L q \varrho\|w(t)\|_{L^q} ϱ ∥ w ( t ) ∥ L q is small only after integration in time, and it vanishes for t ∉ Z 1 t\notin Z_1 t ∈ / Z 1 . This gives the stated bounds. The W 1 , q W^{1,q} W 1 , q bound off Z 1 Z_1 Z 1 follows as in Proposition 6.8(d), since ∣ h ∣ ≤ C h + ϱ ∣ c ∣ |h|\le C_h+\varrho|c| ∣ h ∣ ≤ C h + ϱ ∣ c ∣ there. ▫
Time geometry
Let I ~ = ⋃ k = 1 N c [ α k , α k + 5 τ ~ ] \widetilde{I}=\bigcup_{k=1}^{N_c}[\alpha_k,\alpha_k+5\widetilde{\tau}] I = ⋃ k = 1 N c [ α k , α k + 5 τ ] , with α 1 < ⋯ < α N c \alpha_1<\cdots<\alpha_{N_c} α 1 < ⋯ < α N c . Since these intervals are pairwise disjoint, α k + 1 > α k + 5 τ ~ \alpha_{k+1}>\alpha_k+5\widetilde{\tau} α k + 1 > α k + 5 τ . By well-preparedness, R = 0 R=0 R = 0 on [ α k , α k + τ ~ ] ∪ [ α k + 4 τ ~ , α k + 5 τ ~ ] [\alpha_k,\alpha_k+\widetilde{\tau}]\cup[\alpha_k+4\widetilde{\tau},\alpha_k+5\widetilde{\tau}] [ α k , α k + τ ] ∪ [ α k + 4 τ , α k + 5 τ ] . Choose τ ≤ τ ~ / 2 \tau\leq\widetilde{\tau}/2 τ ≤ τ /2 and T = τ ε T=\tau^\varepsilon T = τ ε with
T < τ ~ 2 , 5 τ < T , 5 2 τ < τ ~ 2 . T<\frac{\widetilde{\tau}}{2},\qquad5\tau<T,\qquad\frac{5}{2}\tau<\frac{\widetilde{\tau}}{2}. T < 2 τ , 5 τ < T , 2 5 τ < 2 τ .
As in [12 Section 3.1] , junctions lie on the global grid T Z T\mathbb{Z} T Z . The chain of component k k k consists of the grid points t k , 0 < ⋯ < t k , N k t_{k,0}<\cdots<t_{k,N_k} t k , 0 < ⋯ < t k , N k in [ α k + τ ~ / 2 , α k + 9 τ ~ / 2 ] [\alpha_k+\widetilde{\tau}/2,\alpha_k+9\widetilde{\tau}/2] [ α k + τ /2 , α k + 9 τ /2 ] . Then t k , 0 ≤ α k + τ ~ t_{k,0}\leq\alpha_k+\widetilde{\tau} t k , 0 ≤ α k + τ and t k , N k ≥ α k + 4 τ ~ t_{k,N_k}\geq\alpha_k+4\widetilde{\tau} t k , N k ≥ α k + 4 τ , so the windows [ t k , i − 1 , t k , i ] [t_{k,i-1},t_{k,i}] [ t k , i − 1 , t k , i ] cover the old stress support. Around a junction t t t define
Z 0 = [ t ± τ 2 ] , Z 1 ′ = [ t ± 3 τ 4 ] , Z 1 = [ t ± τ ] , Z 2 = [ t ± 5 τ 4 ] , W = [ t ± 5 τ 2 ] , Z_0=[t\pm\tfrac{\tau}{2}],\qquad Z'_1=[t\pm\tfrac{3\tau}{4}],\qquad Z_1=[t\pm\tau],\qquad Z_2=[t\pm\tfrac{5\tau}{4}],\qquad W=[t\pm\tfrac{5\tau}{2}], Z 0 = [ t ± 2 τ ] , Z 1 ′ = [ t ± 4 3 τ ] , Z 1 = [ t ± τ ] , Z 2 = [ t ± 4 5 τ ] , W = [ t ± 2 5 τ ] ,
where [ t ± η ] : = [ t − η , t + η ] [t\pm\eta]:=[t-\eta,t+\eta] [ t ± η ] := [ t − η , t + η ] , and set I : = ⋃ W ( t k , i ) I:=\bigcup W(t_{k,i}) I := ⋃ W ( t k , i ) .
Lemma 6.14 (Geometry). (a) I ⊂ I ~ I\subset\widetilde{I} I ⊂ I .
(b) The windows W W W are pairwise disjoint, and 0 , 1 ∉ I 0,1\notin I 0 , 1 ∈ / I .
(c) Z 2 ( t ) ⊂ { s : dist ( s , I c ) ≥ τ } Z_2(t)\subset\{s:\operatorname{dist}(s,I^c)\geq\tau\} Z 2 ( t ) ⊂ { s : dist ( s , I c ) ≥ τ } .
(d) There are fewer than τ − ε \tau^{-\varepsilon} τ − ε windows.
(e) Consecutive junctions are more than 5 τ 5\tau 5 τ apart.
Proof. Interval arithmetic. For (b) and (e), junctions of the same chain are distinct points of T Z T\mathbb{Z} T Z , hence at distance at least T > 5 τ T>5\tau T > 5 τ . For junctions of different chains k < k ′ k<k' k < k ′ , the disjointness of the components gives α k ′ > α k + 5 τ ~ \alpha_{k'}>\alpha_k+5\widetilde{\tau} α k ′ > α k + 5 τ , hence
t k ′ , 0 − t k , N k ≥ ( α k ′ + τ ~ 2 ) − ( α k + 9 τ ~ 2 ) > τ ~ > 5 τ , t_{k',0}-t_{k,N_k}\geq\left(\alpha_{k'}+\frac{\widetilde{\tau}}{2}\right)-\left(\alpha_k+\frac{9\widetilde{\tau}}{2}\right)>\widetilde{\tau}>5\tau, t k ′ , 0 − t k , N k ≥ ( α k ′ + 2 τ ) − ( α k + 2 9 τ ) > τ > 5 τ ,
using 5 2 τ < τ ~ 2 \frac{5}{2}\tau<\frac{\widetilde{\tau}}{2} 2 5 τ < 2 τ . Since the windows have length 5 τ 5\tau 5 τ , they are pairwise disjoint; in particular the new set I I I is again a union of pairwise disjoint intervals, as required in Definition 6.1. For (d), distinct junctions are distinct grid points in ( 0 , 1 ) (0,1) ( 0 , 1 ) , and there are fewer than T − 1 T^{-1} T − 1 of these.
Proof of Proposition 6.3
The construction runs from left to right in time and maintains a current global solution U U U : a smooth VNS–Reynolds solution on [ 0 , 1 ] [0,1] [ 0 , 1 ] with stress S U S_U S U . Initially ( U , S U ) = ( u , R ) (U,S_U)=(u,R) ( U , S U ) = ( u , R ) . It has two passes. The second pass cannot be merged with the first. The entry layer of a corrector to the right of a junction depends on the kinetic state, and therefore on the convex integration perturbation at that junction, which is built to cancel the stress there.
Uniformity of constants. The constants of Proposition 6.8 and Lemma 6.13 depend on the base only through upper bounds for B U , Λ U , L U B_U,\Lambda_U,L_U B U , Λ U , L U of (6.3) and for N S N_S N S . We fix these upper bounds before T T T , in two levels, and then check, for T T T , τ \tau τ and δ ⋆ \delta_\star δ ⋆ small, that every base used in the construction satisfies them. The bounds of level 2 depend on the constants of level 1, but not conversely, because a base of pass 1 never contains a corrector of pass 1 on [ a , 1 ] [a,1] [ a , 1 ] , and a base of pass 2 never contains a corrector of pass 2 on [ a , 1 ] [a,1] [ a , 1 ] .
Level 1 (pass 1). Put B ‾ 1 : = B u + 1 \overline{B}_1:=B_u+1 B 1 := B u + 1 , Λ ‾ 1 : = Λ u + 1 \overline{\Lambda}_1:=\Lambda_u+1 Λ 1 := Λ u + 1 , L ‾ 1 : = ∥ u ∥ L 1 L ∞ + 1 \overline{L}_1:=\|u\|_{L^1L^\infty}+1 L 1 := ∥ u ∥ L 1 L ∞ + 1 , where B u , Λ u B_u,\Lambda_u B u , Λ u are taken with a = 0 a=0 a = 0 , and
N ‾ 1 : = ∥ R ∥ L ∞ W 1 , q + 2 B ‾ 1 + C d , q ( C h ( B ‾ 1 , L ‾ 1 ) + ϱ ( B ‾ 1 , L ‾ 1 ) ) + 2. \overline{N}_1:=\|R\|_{L^\infty W^{1,q}}+2\overline{B}_1+C_{d,q}\bigl(C_h(\overline{B}_1,\overline{L}_1)+\varrho(\overline{B}_1,\overline{L}_1)\bigr)+2. N 1 := ∥ R ∥ L ∞ W 1 , q + 2 B 1 + C d , q ( C h ( B 1 , L 1 ) + ϱ ( B 1 , L 1 ) ) + 2.
Let C 1 ′ , ℓ 0 ′ , 1 , C ∗ , 1 ′ C'_1,\ell'_0,1,C'_{*,1} C 1 ′ , ℓ 0 ′ , 1 , C ∗ , 1 ′ and K 1 , C 1 K_1,C_1 K 1 , C 1 be the constants of Proposition 6.8 for these bounds. Every operation of pass 1 starts after all earlier operations of pass 1 have exited, and its segment and cutoff vanish near its start. Hence on [ a , 1 ] [a,1] [ a , 1 ] its base is u + C t o t u+C_{\mathrm{tot}} u + C tot with C t o t C_{\mathrm{tot}} C tot spatially constant, so B U ≤ B u + ∣ C t o t ∣ B_U\leq B_u+|C_{\mathrm{tot}}| B U ≤ B u + ∣ C tot ∣ and Λ U ≤ Λ u \Lambda_U\leq\Lambda_u Λ U ≤ Λ u . Moreover L U ≤ ∥ u ∥ L 1 L ∞ + ∑ i ∥ z i ∥ L 1 L ∞ + ∣ C t o t ∣ L_U\leq\|u\|_{L^1L^\infty}+\sum_i\|z_i\|_{L^1L^\infty}+|C_{\mathrm{tot}}| L U ≤ ∥ u ∥ L 1 L ∞ + ∑ i ∥ z i ∥ L 1 L ∞ + ∣ C tot ∣ , where the window correctors contribute at most T − 1 ⋅ T ⋅ C 1 ′ T 1 / 2 T^{-1}\cdot T\cdot C'_1T^{1/2} T − 1 ⋅ T ⋅ C 1 ′ T 1/2 and the long correctors at most N c C 1 ′ ( C T 1 / 2 ) β N_cC'_1(CT^{1/2})^\beta N c C 1 ′ ( C T 1/2 ) β , by Proposition 6.8(a)(i), (b)(iii) and the bound ς ≤ C T 1 / 2 \varsigma\leq CT^{1/2} ς ≤ C T 1/2 on good segments shown below. The base stress is R R R plus the accumulated persistent stress, whose W 1 , q W^{1,q} W 1 , q norm is bounded by Lemma 6.12 with U 0 = u U_0=u U 0 = u . So all level-1 bounds hold as soon as ∣ C t o t ∣ ≤ 1 2 |C_{\mathrm{tot}}|\leq\frac12 ∣ C tot ∣ ≤ 2 1 , ∑ i sup ∣ c i ∣ ≤ 1 2 \sum_i\sup|c_i|\leq\frac12 ∑ i sup ∣ c i ∣ ≤ 2 1 and T T T is small.
Level 2 (pass 2). Let u ˉ \bar{u} u ˉ be the result of pass 1. The pass-1 correctors have disjoint supports in time, so sup t ∥ u ˉ − u − C t o t ∥ W 1 , q ≤ C max ( sup ∥ z i ∥ X q ) ≤ C C 1 ′ T 1 / 2 \sup_t \|\bar{u}-u-C_{\mathrm{tot}}\|_{W^{1,q}} \le C\max(\sup\|z_i\|_{X_q}) \le CC'_1T^{1/2} sup t ∥ u ˉ − u − C tot ∥ W 1 , q ≤ C max ( sup ∥ z i ∥ X q ) ≤ C C 1 ′ T 1/2 . Also ∥ ∇ ( u ˉ − u ) ∥ L 1 L ∞ ≤ T − 1 ⋅ T 1 / 2 \|\nabla(\bar{u}-u)\|_{L^1L^\infty} \le T^{-1}\cdot T^{1/2} ∥∇ ( u ˉ − u ) ∥ L 1 L ∞ ≤ T − 1 ⋅ T 1/2 . C 1 ′ T 1 / 2 ς ˉ 1 + N c C'_1T^{1/2}\bar{\varsigma}_1+N_c C 1 ′ T 1/2 ς ˉ 1 + N c , with ς ˉ 1 : = ∥ R ∥ L ∞ L q + C \bar{\varsigma}_1:=\|R\|_{L^\infty L^q}+C ς ˉ 1 := ∥ R ∥ L ∞ L q + C the size of the base stresses of the windows. Put
B ˉ 2 : = B u + C C 1 ′ + 1 , Λ ˉ 2 : = Λ u + C 1 ′ ς ˉ 1 β + N c + 1 , L ˉ 2 : = L ˉ 1 + 1 , \bar{B}_2:=B_u+CC'_1+1,\qquad\bar{\Lambda}_2:=\Lambda_u+C'_1\bar{\varsigma}_1^\beta+N_c+1,\qquad\bar{L}_2:=\bar{L}_1+1, B ˉ 2 := B u + C C 1 ′ + 1 , Λ ˉ 2 := Λ u + C 1 ′ ς ˉ 1 β + N c + 1 , L ˉ 2 := L ˉ 1 + 1 ,
N ˉ 2 : = N ˉ 1 + 2 B ˉ 2 + C d , q ( C h ( B ˉ 2 , L ˉ 2 ) + ϱ ( B ˉ 2 , L ˉ 2 ) ) + 2 , \bar{N}_2:=\bar{N}_1+2\bar{B}_2+C_{d,q}\bigl(C_h(\bar{B}_2,\bar{L}_2)+\varrho(\bar{B}_2,\bar{L}_2)\bigr)+2, N ˉ 2 := N ˉ 1 + 2 B ˉ 2 + C d , q ( C h ( B ˉ 2 , L ˉ 2 ) + ϱ ( B ˉ 2 , L ˉ 2 ) ) + 2 ,
and let C 2 ′ , ℓ 0 ′ , c ∗ ′ , ε ∗ ′ , K 2 ♯ , C 2 ♯ C'_2,\ell'_0,c'_*,\varepsilon'_*,K^\sharp_2,C^\sharp_2 C 2 ′ , ℓ 0 ′ , c ∗ ′ , ε ∗ ′ , K 2 ♯ , C 2 ♯ be the constants of Proposition 6.8 and Lemma 6.13 for these bounds. Every operation of pass 2 starts after all earlier operations of pass 2 have exited, and every w j w_j w j vanishes after sup Z 1 ( t j ) \sup Z_1(t_j) sup Z 1 ( t j ) , which precedes the start of every later operation (Lemma 6.14 (e)). Hence on [ a , 1 ] [a,1] [ a , 1 ] the base of a pass-2 operation is u ˉ + C t o t \bar{u}+C_{\mathrm{tot}} u ˉ + C tot , and the w j w_j w j enter only through L U L_U L U , with
∑ j ∥ w j ∥ L 1 L ∞ ≤ N J δ ⋆ ≤ 1 2 . \sum_j\|w_j\|_{L^1L^\infty}\le N_J\delta_\star\le\frac{1}{2}. j ∑ ∥ w j ∥ L 1 L ∞ ≤ N J δ ⋆ ≤ 2 1 .
The base stress on a re-solved segment is R ˉ \bar{R} R ˉ plus the accumulated persistent stresses of pass 2; the former vanishes there, since it is supported in ⋃ Z 0 ( t j ) \bigcup Z_0(t_j) ⋃ Z 0 ( t j ) (see the paragraph following the description of pass 1), and the latter is bounded by Lemma 6.12 with U 0 = u ˉ U_0=\bar{u} U 0 = u ˉ . So all level-2 bounds hold as soon as ∣ C t o t ∣ ≤ 1 2 |C_{\mathrm{tot}}|\le\frac{1}{2} ∣ C tot ∣ ≤ 2 1 , ∑ l sup ∣ c l ∣ ≤ 1 2 \sum_l\sup|c_l|\le\frac{1}{2} ∑ l sup ∣ c l ∣ ≤ 2 1 and N J δ ⋆ ≤ 1 2 N_J\delta_\star\le\frac{1}{2} N J δ ⋆ ≤ 2 1 .
Summing the bounds of Proposition 6.8 (d) termwise, instead of using Lemma 6.12 , would only give N S ≤ C T − 1 N_S\le CT^{-1} N S ≤ C T − 1 , which would make ℓ 0 ′ \ell'_0 ℓ 0 ′ depend on T T T . From now on C ′ C' C ′ ,ℓ 0 ′ \ell'_0 ℓ 0 ′ ,c ∗ ′ c'_* c ∗ ′ ,ε ∗ ′ \varepsilon'_* ε ∗ ′ denote the larger, respectively smaller, of the two levels, and C C C denotes a constant depending only on d , p , q , r , β , ε , f 0 d,p,q,r,\beta,\varepsilon,f_0 d , p , q , r , β , ε , f 0 , on ( u , R , I ~ , τ ~ ) (u,R,\widetilde{I},\widetilde{\tau}) ( u , R , I , τ ) and on C 1 ′ , C 2 ′ C'_1,C'_2 C 1 ′ , C 2 ′ ; it does not depend on T , τ T,\tau T , τ or the parameters δ j ′ \delta'_j δ j ′ . We take T ≤ ℓ 0 ′ T\le\ell'_0 T ≤ ℓ 0 ′ . The conditions ∣ C t o t ∣ ≤ 1 2 |C_{\mathrm{tot}}|\le\frac{1}{2} ∣ C tot ∣ ≤ 2 1 and ∑ l sup ∣ c l ∣ ≤ 1 2 \sum_l\sup|c_l|\le\frac{1}{2} ∑ l sup ∣ c l ∣ ≤ 2 1 are part of (6.21): by Proposition 6.8 (c) and Lemma 6.13, ∑ l sup ∣ c l ∣ ≤ C T 1 / 2 + C N J δ ⋆ + C ( N c + 1 ) ς ˉ \sum_l\sup|c_l|\le CT^{1/2}+CN_J\delta_\star+C(N_c+1)\bar{\varsigma} ∑ l sup ∣ c l ∣ ≤ C T 1/2 + C N J δ ⋆ + C ( N c + 1 ) ς ˉ .
Pass 1: causal concentration. On [ 0 , t 1 , 0 ] [0,t_{1,0}] [ 0 , t 1 , 0 ] nothing is done. For each component k k k , in order:
for i = 1 , … , N k i=1,\ldots,N_k i = 1 , … , N k , apply Proposition 6.8 (a) on the window Σ k , i = [ t k , i − 1 , t k , i ] \Sigma_{k,i}=[t_{k,i-1},t_{k,i}] Σ k , i = [ t k , i − 1 , t k , i ] with base the current U U U . Use a two-sided cutoff vanishing on [ t k , i − 1 , t k , i − 1 + τ 4 ] ∪ [ t k , i − τ 4 , t k , i ] [t_{k,i-1},t_{k,i-1}+\frac{\tau}{4}]\cup[t_{k,i}-\frac{\tau}{4},t_{k,i}] [ t k , i − 1 , t k , i − 1 + 4 τ ] ∪ [ t k , i − 4 τ , t k , i ] and equal to 1 on [ t k , i − 1 + τ 4 , t k , i − τ 4 ] [t_{k,i-1}+\frac{\tau}{4},t_{k,i}-\frac{\tau}{4}] [ t k , i − 1 + 4 τ , t k , i − 4 τ ] . Glue by Proposition 6.9, exiting by the mean;
then apply Proposition 6.8 (b) on the good segment [ t k , N k , t k + 1 , 0 ] [t_{k,N_k},t_{k+1,0}] [ t k , N k , t k + 1 , 0 ] , or [ t N c , N N c , 1 ] [t_{N_c,N_{N_c}},1] [ t N c , N N c , 1 ] with a one-sided cutoff, with layers of length τ 4 \frac{\tau}{4} 4 τ inside the Z 0 Z_0 Z 0 of the endpoints.
The base stress on a window is R R R plus the accumulated persistent stress of earlier windows, so ς ≤ ∥ R ∥ L ∞ L q + C \varsigma\le\|R\|_{L^\infty L^q}+C ς ≤ ∥ R ∥ L ∞ L q + C and, by Lemma 6.12 , N S ≤ C N_S\le C N S ≤ C ; it need not be small, and (a) does not require it. After a window with corrector z i z_i z i exits, the base stress for all later times acquires the persistent term Π i \Pi_i Π i . By Proposition 6.8 (a)(ii) and (d), with ∣ Σ k , i ∣ = T |\Sigma_{k,i}|=T ∣ Σ k , i ∣ = T ,
sup t ≥ t k , i ∥ Π i ( t ) ∥ L q ≤ C T sup Σ k , i ∥ z i ∥ L q ≤ C T 3 / 2 . \sup_{t\ge t_{k,i}}\|\Pi_i(t)\|_{L^q}\le CT\sup_{\Sigma_{k,i}}\|z_i\|_{L^q}\le CT^{3/2}. t ≥ t k , i sup ∥ Π i ( t ) ∥ L q ≤ C T Σ k , i sup ∥ z i ∥ L q ≤ C T 3/2 .
Since there are fewer than T − 1 T^{-1} T − 1 windows, ∑ i sup ∥ Π i ∥ L q ≤ C T 1 / 2 \sum_i\sup\|\Pi_i\|_{L^q}\le CT^{1/2} ∑ i sup ∥ Π i ∥ L q ≤ C T 1/2 . On the good segments R = 0 R=0 R = 0 , so the base stress there consists of the persistent terms of all earlier windows and of all earlier good segments. Let s k s_k s k be its size on the k k k -th good segment. By (6.6) and Proposition 6.8 (b)(ii), the k ′ k' k ′ -th good segment leaves a persistent term of size at most C C B s k ′ CC_Bs_{k'} C C B s k ′ . Hence s k ≤ C T 1 / 2 + C ∑ k ′ < k s k ′ s_k\le CT^{1/2}+C\sum_{k'<k}s_{k'} s k ≤ C T 1/2 + C ∑ k ′ < k s k ′ , and by induction s k ≤ ( 1 + C ) N c C T 1 / 2 ≤ c ∗ ′ s_k\le(1+C)^{N_c}CT^{1/2}\le c'_* s k ≤ ( 1 + C ) N c C T 1/2 ≤ c ∗ ′ for T T T small, as required by (b). The factor ( 1 + C ) N c (1+C)^{N_c} ( 1 + C ) N c depends only on I ~ \widetilde{I} I , which is fixed before T T T . The same persistent terms enter the base stress of the later windows; they are bounded by C T 1 / 2 CT^{1/2} C T 1/2 in L q L^q L q and, by Lemma 6.12 , by the same bounds in W 1 , q W^{1,q} W 1 , q , so the bounds ς ≤ ∥ R ∥ L ∞ L q + C \varsigma\le\|R\|_{L^\infty L^q}+C ς ≤ ∥ R ∥ L ∞ L q + C and N S ≤ C N_S\le C N S ≤ C for the windows are unaffected. In what follows, constants C C C in pass 1 may depend on N c N_c N c .
Let ( u ˉ , p ˉ , R ˉ ) (\bar{u},\bar{p},\bar{R}) ( u ˉ , p ˉ , R ˉ ) be the result. The persistent terms do not survive outside the layers. Indeed, the corrector of each segment solves (6.4), whose forcing contains − div S -\operatorname{div}S − div S for the full base stress S S S , which includes all persistent terms created earlier; hence, by Proposition 6.9, the glued stress vanishes wherever the cutoff of that segment equals 1. After the exit of a segment the base stress, together with the new persistent term, reappears, but only until the cutoff of the next segment reaches 1. The windows and good segments cover [ t 1 , 0 , 1 ] [t_{1,0},1] [ t 1 , 0 , 1 ] , their cutoffs equal 1 outside ⋃ Z 0 ( t k , i ) \bigcup Z_0(t_{k,i}) ⋃ Z 0 ( t k , i ) , the last cutoff equals 1 near t = 1 t=1 t = 1 , and R = 0 R=0 R = 0 on [ 0 , t 1 , 0 ] [0,t_{1,0}] [ 0 , t 1 , 0 ] . Therefore R ˉ \bar{R} R ˉ is supported in ⋃ Z 0 ( t k , i ) \bigcup Z_0(t_{k,i}) ⋃ Z 0 ( t k , i ) , where it consists of R R R , the layer terms and the persistent terms. By Corollary 6.11 and Proposition 6.8 (a), the layer stress of the window Σ k , i \Sigma_{k,i} Σ k , i is at most
2 C Γ ∫ Σ k , i ∥ R ∥ L r + C T 3 / 2 + C τ ( ∥ R ∥ L ∞ L r + C + T ) , 2C_\Gamma\int_{\Sigma_{k,i}}\|R\|_{L^r}+CT^{3/2}+C\tau\bigl(\|R\|_{L^\infty L^r}+C+T\bigr), 2 C Γ ∫ Σ k , i ∥ R ∥ L r + C T 3/2 + C τ ( ∥ R ∥ L ∞ L r + C + T ) ,
and summing over windows, as in [12 Proposition 3.3] ,
∥ R ˉ ∥ L 1 L r ≤ 2 C Γ ∥ R ∥ L 1 L r + C ( T 1 / 2 + τ 1 − ε ) . (48) \|\bar{R}\|_{L^1L^r}\le2C_\Gamma\|R\|_{L^1L^r}+C\bigl(T^{1/2}+\tau^{1-\varepsilon}\bigr). \tag*{(48)} ∥ R ˉ ∥ L 1 L r ≤ 2 C Γ ∥ R ∥ L 1 L r + C ( T 1/2 + τ 1 − ε ) . (48)
Moreover, by Proposition 6.8 (a)(i) on windows (as in (W3) below), and (b)(i), (iii) on good segments:
∥ u ˉ − u ∥ L ∞ L ∞ ∩ L 1 W 1 , q ≤ C T 1 / 2 + C ( T 1 / 2 ) β \|\bar u-u\|_{L^\infty L^\infty\cap L^1W^{1,q}}\le CT^{1/2}+C(T^{1/2})^\beta ∥ u ˉ − u ∥ L ∞ L ∞ ∩ L 1 W 1 , q ≤ C T 1/2 + C ( T 1/2 ) β ;
∥ u ˉ − u ∥ L 2 ( I ~ c ; W 1 , ∞ ) ≤ C ( T 1 / 2 ) β \|\bar u-u\|_{L^2(\widetilde I^c;W^{1,\infty})}\le C(T^{1/2})^\beta ∥ u ˉ − u ∥ L 2 ( I c ; W 1 , ∞ ) ≤ C ( T 1/2 ) β , since I ~ c \widetilde I^c I c meets only the good segments;
u ˉ = u \bar u=u u ˉ = u on [ 0 , t 1 , 0 ] [0,t_{1,0}] [ 0 , t 1 , 0 ] .
Pass 2: convex integration and re-solving. Enumerate the junctions t 1 < ⋯ < t N J t_1<\cdots<t_{N_J} t 1 < ⋯ < t N J , with N J < τ − ε N_J<\tau^{-\varepsilon} N J < τ − ε . Start from U = u ˉ U=\bar u U = u ˉ . For j = 1 , … , N J j=1,\ldots,N_J j = 1 , … , N J :
(CI) apply Lemma 6.13 at t j t_j t j with the zones Z 0 , Z 1 ′ , Z 1 Z_0,Z'_1,Z_1 Z 0 , Z 1 ′ , Z 1 of t j t_j t j , the current stress S S S , and a parameter δ j ′ \delta'_j δ j ′ . The cancelled part S c = θ 0 S S_c=\theta_0S S c = θ 0 S contains R ˉ \bar R R ˉ on Z 0 ( t j ) Z_0(t_j) Z 0 ( t j ) ;
(re-solve) on Σ j : = [ t j + τ , t j + 1 − 9 τ 8 ] \Sigma_j:=[t_j+\tau,t_{j+1}-\frac{9\tau}{8}] Σ j := [ t j + τ , t j + 1 − 8 9 τ ] , or on Σ N J : = [ t N J + τ , 1 ] \Sigma_{N_J}:=[t_{N_J}+\tau,1] Σ N J := [ t N J + τ , 1 ] with a one-sided cutoff, with base the current U U U : apply Proposition 6.8(a) if t j t_j t j and t j + 1 t_{j+1} t j + 1 belong to the same chain, and Proposition 6.8(b) otherwise, i.e. if t j = t k , N k t_j=t_{k,N_k} t j = t k , N k is the last junction of a chain or j = N J j=N_J j = N J . In the latter case ∣ Σ j ∣ |\Sigma_j| ∣ Σ j ∣ is comparable to the gap between two components of I ~ \widetilde I I and need not be smaller than ℓ 0 \ell_0 ℓ 0 . Let L \mathcal L L be the set of these indices; # L ≤ N c \#\mathcal L\le N_c # L ≤ N c . The layers are [ t j + τ , t j + 9 τ 8 ] [t_j+\tau,t_j+\frac{9\tau}{8}] [ t j + τ , t j + 8 9 τ ] and [ t j + 1 − 5 τ 4 , t j + 1 − 9 τ 8 ] [t_{j+1}-\frac{5\tau}{4},t_{j+1}-\frac{9\tau}{8}] [ t j + 1 − 4 5 τ , t j + 1 − 8 9 τ ] , and the segment starts after supp w j \operatorname{supp} w_j supp w j .
On the re-solved segment R ˉ \bar R R ˉ vanishes, so the base stress consists of the persistent terms of pass 2: those of the convex integration steps i ≤ j i\le j i ≤ j (the terms c i ⊗ ˚ U + U ⊗ ˚ c i − R h i c_i\mathbin{\mathring\otimes}U+U\mathbin{\mathring\otimes}c_i-Rh_i c i ⊗ ˚ U + U ⊗ ˚ c i − R h i off Z 1 ( t i ) Z_1(t_i) Z 1 ( t i ) ), and those of the exits of earlier re-solved segments. They are the mismatch between the pass-1 solution u ˉ \bar u u ˉ , which is exact there, and the kinetic state modified by pass 2.
Supports. After pass 2, the stress vanishes on the bulks [ t j + 5 τ 4 , t j + 1 − 5 τ 4 ] [t_j+\frac{5\tau}{4},t_{j+1}-\frac{5\tau}{4}] [ t j + 4 5 τ , t j + 1 − 4 5 τ ] . In Z 2 ( t j ) Z_2(t_j) Z 2 ( t j ) it consists of:
S j C L S_j^{\mathrm{CL}} S j CL , the cross terms c j ⊗ ˚ ( U + w j ) + ( U + w j ) ⊗ ˚ c j c_j\mathbin{\mathring\otimes}(U+w_j)+(U+w_j)\mathbin{\mathring\otimes}c_j c j ⊗ ˚ ( U + w j ) + ( U + w j ) ⊗ ˚ c j and − R h j -Rh_j − R h j , on Z 1 ( t j ) Z_1(t_j) Z 1 ( t j ) ;
the layers of the two adjacent segments;
the persistent terms, alive between the exit of one segment and the entry of the next.
Before t 1 − 5 τ 4 t_1-\frac{5\tau}{4} t 1 − 4 5 τ nothing has changed. By Lemma 6.14(b)–(d), u 1 u_1 u 1 is well-prepared for ( I , τ ) (I,\tau) ( I , τ ) , and I ⊂ I ~ I\subset\widetilde I I ⊂ I by Lemma 6.14(a).
Smallness. Let ς j \varsigma_j ς j be the size, in L ∞ L q L^\infty L^q L ∞ L q , of the base stress on the j j j -th re-solved segment Σ j \Sigma_j Σ j . By the bounds above, Lemma 6.13 and (6.6),
ς j ≤ C T 1 / 2 + C ∑ i ≤ j δ i ′ + C ∑ i < j ∣ Σ i ∣ sup Σ i ∥ z i ∥ L q ≤ C T 1 / 2 + C ∑ i ≤ j δ i ′ + C ∑ i < j , i ∉ L ∣ Σ i ∣ ς i + C ∑ i < j , i ∈ L ς i . \varsigma_j\le CT^{1/2}+C\sum_{i\le j}\delta'_i+C\sum_{i<j}|\Sigma_i|\sup_{\Sigma_i}\|z_i\|_{L^q}\le CT^{1/2}+C\sum_{i\le j}\delta'_i+C\sum_{i<j,\ i\notin\mathcal L}|\Sigma_i|\varsigma_i+C\sum_{i<j,\ i\in\mathcal L}\varsigma_i. ς j ≤ C T 1/2 + C i ≤ j ∑ δ i ′ + C i < j ∑ ∣ Σ i ∣ Σ i sup ∥ z i ∥ L q ≤ C T 1/2 + C i ≤ j ∑ δ i ′ + C i < j , i ∈ / L ∑ ∣ Σ i ∣ ς i + C i < j , i ∈ L ∑ ς i .
using Proposition 6.8(a)(ii) for i ∉ L i\notin\mathcal L i ∈ / L and (b)(ii) for i ∈ L i\in\mathcal L i ∈ L . By induction on j j j ,
ς j ≤ A ∏ i < j , i ∉ L ( 1 + C ∣ Σ i ∣ ) ∏ i < j , i ∈ L ( 1 + C ) \varsigma_j\le A\prod_{i<j,\ i\notin\mathcal L}(1+C|\Sigma_i|)\prod_{i<j,\ i\in\mathcal L}(1+C) ς j ≤ A i < j , i ∈ / L ∏ ( 1 + C ∣ Σ i ∣ ) i < j , i ∈ L ∏ ( 1 + C )
with A = C T 1 / 2 + C N J max j δ j ′ A=CT^{1/2}+CN_J\max_j\delta'_j A = C T 1/2 + C N J max j δ j ′ , and since ∑ i ∣ Σ i ∣ ≤ 1 \sum_i|\Sigma_i|\le1 ∑ i ∣ Σ i ∣ ≤ 1 and # L ≤ N c \#\mathcal L\le N_c # L ≤ N c ,
max j ς j ≤ ς ˉ : = e C ( 1 + C ) N c ( C T 1 / 2 + C N J max j δ j ′ ) . (49) \max_j\varsigma_j\le\bar\varsigma:=e^C(1+C)^{N_c}\left(CT^{1/2}+CN_J\max_j\delta'_j\right). \tag*{(49)} j max ς j ≤ ς ˉ := e C ( 1 + C ) N c ( C T 1/2 + C N J j max δ j ′ ) . (49)
The application of (b) on Σ i \Sigma_i Σ i , i ∈ L i\in\mathcal L i ∈ L , requires ς i ≤ c ∗ ′ \varsigma_i\le c_*' ς i ≤ c ∗ ′ , which is part of the induction and follows from ς ˉ ≤ c ∗ ′ \bar\varsigma\le c_*' ς ˉ ≤ c ∗ ′ . All bounds in this recursion are linear in the stress, because the persistent terms are controlled by the time integral of the perturbation, through (6.6); no factor exponential in N J N_J N J appears. The factor ( 1 + C ) N c (1+C)^{N_c} ( 1 + C ) N c depends only on I ˉ \bar I I ˉ , which is fixed before T T T . The Hölder power β \beta β appears only in the outputs below. We choose δ j ′ ≤ δ ∗ \delta'_j\le\delta_* δ j ′ ≤ δ ∗ , with δ ∗ \delta_* δ ∗ and T T T so small that
ς ˉ ≤ c ∗ ′ , C ( N c + 1 ) ς ˉ + C N J δ ∗ ≤ δ 10 , ( N c + 1 ) 1 / 2 C ′ ς ˉ β ≤ δ 10 , \bar\varsigma\le c_*',\qquad C(N_c+1)\bar\varsigma+CN_J\delta_*\le\frac{\delta}{10},\qquad(N_c+1)^{1/2}C'\bar\varsigma^\beta\le\frac{\delta}{10}, ς ˉ ≤ c ∗ ′ , C ( N c + 1 ) ς ˉ + C N J δ ∗ ≤ 10 δ , ( N c + 1 ) 1/2 C ′ ς ˉ β ≤ 10 δ ,
C T 1 / 2 + C N J δ ∗ + C ( N c + 1 ) ς ˉ ≤ 1 2 , N J δ ∗ ≤ 1 2 . (50) CT^{1/2}+CN_J\delta_*+C(N_c+1)\bar\varsigma\le\frac12,\qquad N_J\delta_*\le\frac12. \tag*{(50)} C T 1/2 + C N J δ ∗ + C ( N c + 1 ) ς ˉ ≤ 2 1 , N J δ ∗ ≤ 2 1 . (50)
where C ′ C' C ′ is the larger of the constants of Proposition 6.8(a)(iii) and (b)(iii). These conditions are met by choosing first T T T (hence τ \tau τ ) small and then δ ∗ \delta_* δ ∗ small in terms of N J N_J N J . Note that N J ≤ τ − ε N_J\le\tau^{-\varepsilon} N J ≤ τ − ε is fixed before δ ∗ \delta_* δ ∗ .
Conclusion.
(W1): on Z 1 ( t j ) Z_1(t_j) Z 1 ( t j ) , ∥ S j C ∥ L 1 L r + ∥ c j ⊗ ˙ ( U + w j ) + ( U + w j ) ⊗ ˙ c j ∥ L 1 L r + ∥ R h j ∥ L 1 L q ≤ C ς j \|S_j^{\mathcal{C}}\|_{L^1L^r}+\|c_j\mathbin{\dot\otimes}(U+w_j)+(U+w_j)\mathbin{\dot\otimes}c_j\|_{L^1L^r}+\|Rh_j\|_{L^1L^q}\leq C\varsigma_j ∥ S j C ∥ L 1 L r + ∥ c j ⊗ ˙ ( U + w j ) + ( U + w j ) ⊗ ˙ c j ∥ L 1 L r + ∥ R h j ∥ L 1 L q ≤ C ς j by Lemma 6.13. The layer stress of Σ ~ j \widetilde{\Sigma}_j Σ j is bounded by Corollary 6.11 and Proposition 6.8(a)(ii)–(iv):
2 ( C r ∣ Σ j ∣ ς j + C ∣ Σ j ∣ 3 / 2 ς j ) + C τ ( ς j + C r / 2 ∣ Σ j ∣ ς j β + ∣ Σ j ∣ 3 / 2 ς j + ∣ Σ j ∣ 1 / 2 ς j ) ≤ C ∣ Σ j ∣ ς ˉ + C τ ς ˉ β 2(C_r|\Sigma_j|\varsigma_j+C|\Sigma_j|^{3/2}\varsigma_j)+C\tau(\varsigma_j+C^{r/2}|\Sigma_j|\varsigma_j^\beta+|\Sigma_j|^{3/2}\varsigma_j+|\Sigma_j|^{1/2}\varsigma_j)\leq C|\Sigma_j|\bar{\varsigma}+C\tau\bar{\varsigma}^{\beta} 2 ( C r ∣ Σ j ∣ ς j + C ∣ Σ j ∣ 3/2 ς j ) + C τ ( ς j + C r /2 ∣ Σ j ∣ ς j β + ∣ Σ j ∣ 3/2 ς j + ∣ Σ j ∣ 1/2 ς j ) ≤ C ∣ Σ j ∣ ς ˉ + C τ ς ˉ β
for j ∉ L j\notin\mathcal{L} j ∈ / L , and by (b)(ii)–(iv) at most C ς ˉ C\bar{\varsigma} C ς ˉ for j ∈ L j\in\mathcal{L} j ∈ L . In the bracket, ∣ Σ j ∣ 3 / 2 ς j |\Sigma_j|^{3/2}\varsigma_j ∣ Σ j ∣ 3/2 ς j bounds 2 C E B U sup ∣ c ∣ 2C E_{BU}\sup|c| 2 C E B U sup ∣ c ∣ , and ∣ Σ j ∣ 1 / 2 ς j |\Sigma_j|^{1/2}\varsigma_j ∣ Σ j ∣ 1/2 ς j bounds sup Σ j ∥ R q z ∥ L q \sup_{\Sigma_j}\|R q_z\|_{L^q} sup Σ j ∥ R q z ∥ L q , whose instantaneous part is of the size of sup Σ j ∥ z ∥ L q ≤ C ∣ Σ j ∣ 1 / 2 ς j \sup_{\Sigma_j}\|z\|_{L^q}\leq C|\Sigma_j|^{1/2}\varsigma_j sup Σ j ∥ z ∥ L q ≤ C ∣ Σ j ∣ 1/2 ς j ; we used ς j ≤ ς ˉ ≤ 1 \varsigma_j\leq\bar{\varsigma}\leq1 ς j ≤ ς ˉ ≤ 1 . The persistent terms on Z 2 ( t j ) ∖ Z 0 ( t j ) Z_2(t_j)\setminus Z_0(t_j) Z 2 ( t j ) ∖ Z 0 ( t j ) , which include the part S s = ( 1 − θ 0 ) S S_s=(1-\theta_0)S S s = ( 1 − θ 0 ) S carried along in Lemma 6.13, contribute at most C τ ς ˉ C\tau\bar{\varsigma} C τ ς ˉ . Summing over j ≤ N J < τ − ϵ j\leq N_J<\tau^{-\epsilon} j ≤ N J < τ − ϵ and using ∑ ∣ Σ j ∣ ≤ 1 \sum|\Sigma_j|\leq1 ∑ ∣ Σ j ∣ ≤ 1 and # L ≤ N c \#\mathcal{L}\leq N_c # L ≤ N c ,
∥ R 1 ∥ L 1 L r ≤ C N J δ ⋆ + C ( N c + 1 ) ς ˉ + C τ 1 − ϵ ς ˉ β ≤ δ \|R_1\|_{L^1L^r}\leq CN_J\delta_\star+C(N_c+1)\bar{\varsigma}+C\tau^{1-\epsilon}\bar{\varsigma}^{\beta}\leq\delta ∥ R 1 ∥ L 1 L r ≤ C N J δ ⋆ + C ( N c + 1 ) ς ˉ + C τ 1 − ϵ ς ˉ β ≤ δ
by (6.21), after decreasing τ \tau τ .
(W2): the w j w_j w j have disjoint supports, and the S c , j = θ 0 S j S_{c,j}=\theta_0S_j S c , j = θ 0 S j are supported in the disjoint intervals Z 1 ′ ( t j ) Z'_1(t_j) Z 1 ′ ( t j ) . On Z 1 ′ ( t j ) Z'_1(t_j) Z 1 ′ ( t j ) the current stress S j S_j S j is R ‾ \overline{R} R plus the accumulated persistent stresses of pass 2, and the latter are bounded in L ∞ L q L^\infty L^q L ∞ L q by ς ˉ \bar{\varsigma} ς ˉ , as in the recursion for ς j \varsigma_j ς j . Since ∣ Z 1 ′ ( t j ) ∣ = 3 τ 2 |Z'_1(t_j)|=\frac{3\tau}{2} ∣ Z 1 ′ ( t j ) ∣ = 2 3 τ and N J < τ − ϵ N_J<\tau^{-\epsilon} N J < τ − ϵ ,
∥ ∑ j w j ∥ L 2 2 ≤ M 2 ∑ j ∥ S c , j ∥ L t 1 , x ≤ M 2 ( ∥ R ‾ ∥ L 1 L r + 3 2 τ 1 − ϵ ς ˉ ) ≤ M 2 ( 2 C r ∥ R ∥ L 1 L r + o ( 1 ) ) , \left\|\sum_j w_j\right\|_{L^2}^2\leq M^2\sum_j\|S_{c,j}\|_{L^1_t,x}\leq M^2\left(\|\overline{R}\|_{L^1L^r}+\frac{3}{2}\tau^{1-\epsilon}\bar{\varsigma}\right)\leq M^2(2C_r\|R\|_{L^1L^r}+o(1)), j ∑ w j L 2 2 ≤ M 2 j ∑ ∥ S c , j ∥ L t 1 , x ≤ M 2 ( ∥ R ∥ L 1 L r + 2 3 τ 1 − ϵ ς ˉ ) ≤ M 2 ( 2 C r ∥ R ∥ L 1 L r + o ( 1 )) ,
by (6.19), where o ( 1 ) → 0 o(1)\to0 o ( 1 ) → 0 as T , τ , δ ⋆ → 0 T,\tau,\delta_\star\to0 T , τ , δ ⋆ → 0 . All other contributions are small. The cut-off window correctors χ i z ˚ i \chi_i\mathring{z}_i χ i z ˚ i of both passes have time supports in the windows, and at any time at most two of them (one from each pass) are nonzero. By Proposition 6.8(a)(i), sup ∥ z i ∥ L 2 ≤ sup ∥ z i ∥ X q ≤ C ′ T 1 / 2 \sup\|z_i\|_{L^2}\leq\sup\|z_i\|_{X_q}\leq C'T^{1/2} sup ∥ z i ∥ L 2 ≤ sup ∥ z i ∥ X q ≤ C ′ T 1/2 , so, adding squares over fewer than 2 T − 1 2T^{-1} 2 T − 1 windows of length at most T T T ,
∥ ∑ i χ i z ˚ i ∥ L t , x 2 2 ≤ 2 ∑ i ∣ Σ i ∣ sup Σ i ∥ z i ∥ L 2 2 ≤ 2 ⋅ 2 T − 1 ⋅ T ⋅ C ′ 2 T = 4 C ′ 2 T . \left\|\sum_i\chi_i\mathring{z}_i\right\|_{L^2_{t,x}}^2\leq2\sum_i|\Sigma_i|\sup_{\Sigma_i}\|z_i\|_{L^2}^2\leq2\cdot2T^{-1}\cdot T\cdot C'^2T=4C'^2T. i ∑ χ i z ˚ i L t , x 2 2 ≤ 2 i ∑ ∣ Σ i ∣ Σ i sup ∥ z i ∥ L 2 2 ≤ 2 ⋅ 2 T − 1 ⋅ T ⋅ C ′2 T = 4 C ′2 T .
Here the disjointness of the supports is needed: the sum of the norms, ∑ i ∣ Σ i ∣ 1 / 2 sup ∥ z i ∥ L 2 \sum_i|\Sigma_i|^{1/2}\sup\|z_i\|_{L^2} ∑ i ∣ Σ i ∣ 1/2 sup ∥ z i ∥ L 2 , is only bounded by C ′ C' C ′ . The long segment correctors contribute C ( N c + 1 ) ς ˉ C(N_c+1)\bar{\varsigma} C ( N c + 1 ) ς ˉ , and the means C N J δ ⋆ + C ς ˉ CN_J\delta_\star+C\bar{\varsigma} C N J δ ⋆ + C ς ˉ . Hence (W2) holds with M W : = M 2 C r M_W:=M\sqrt{2C_r} M W := M 2 C r .
(W3): the window correctors contribute at most C T 1 / 2 CT^{1/2} C T 1/2 . Here (a)(iii) is not enough: it gives ∥ z i ∥ L 1 ( Σ i ; W 1 , ∞ ) ≤ T 1 / 2 ∥ z i ∥ L 2 ( Σ i ; W 1 , ∞ ) ≤ C ′ T \|z_i\|_{L^1(\Sigma_i;W^{1,\infty})}\leq T^{1/2}\|z_i\|_{L^2(\Sigma_i;W^{1,\infty})}\leq C'T ∥ z i ∥ L 1 ( Σ i ; W 1 , ∞ ) ≤ T 1/2 ∥ z i ∥ L 2 ( Σ i ; W 1 , ∞ ) ≤ C ′ T , and the sum over fewer than 2 T − 1 2T^{-1} 2 T − 1 windows is only bounded. Instead we use (a)(i): sup Σ i ∥ z i ∥ W 1 , q + sup Σ i ∥ z i ∥ L ∞ ≤ C sup Σ i ∥ z i ∥ X q ≤ C ′ T 1 / 2 \sup_{\Sigma_i}\|z_i\|_{W^{1,q}}+\sup_{\Sigma_i}\|z_i\|_{L^\infty}\leq C\sup_{\Sigma_i}\|z_i\|_{X_q}\leq C'T^{1/2} sup Σ i ∥ z i ∥ W 1 , q + sup Σ i ∥ z i ∥ L ∞ ≤ C sup Σ i ∥ z i ∥ X q ≤ C ′ T 1/2 . At any time at most two cut-off window correctors are nonzero, and ∥ χ i z ˚ i ∥ ≤ 2 ∥ z i ∥ \|\chi_i\mathring{z}_i\|\leq2\|z_i\| ∥ χ i z ˚ i ∥ ≤ 2∥ z i ∥ in these norms, so
∥ ∑ i χ i z ˚ i ∥ L ∞ ( 0 , 1 ; L ∞ ∩ W 1 , q ) ≤ 4 C ′ T 1 / 2 , \left\|\sum_i\chi_i\mathring{z}_i\right\|_{L^\infty(0,1;L^\infty\cap W^{1,q})}\leq4C'T^{1/2}, i ∑ χ i z ˚ i L ∞ ( 0 , 1 ; L ∞ ∩ W 1 , q ) ≤ 4 C ′ T 1/2 ,
which bounds the L p L ∞ L^pL^\infty L p L ∞ and L 1 W 1 , q L^1W^{1,q} L 1 W 1 , q norms; the re-solved segments contribute at most C ′ ς ˉ β C'\bar{\varsigma}^{\beta} C ′ ς ˉ β for the short ones, for the same reason, and N c C ′ ς ˉ β N_cC'\bar{\varsigma}^{\beta} N c C ′ ς ˉ β for the long ones; the w j w_j w j contribute N J δ ⋆ N_J\delta_\star N J δ ⋆ ; the means contribute C N J δ ⋆ + C ς ˉ CN_J\delta_\star+C\bar{\varsigma} C N J δ ⋆ + C ς ˉ .
(W5): I ~ c \widetilde{I}^{c} I c meets only the good segments of pass 1 and the re-solved segments. The former contribute C ( T 1 / 2 ) β C(T^{1/2})^\beta C ( T 1/2 ) β . For the latter, by (a)(iii), (b)(iii) and since L 2 L^2 L 2 norms over disjoint intervals add in squares,
∥ ∑ j z j ∥ L 2 ( I ~ c ; W 1 , ∞ ) 2 ≤ ∑ j ∉ L ( C ′ ∣ Σ j ∣ 1 / 2 ς j β ) 2 + ∑ j ∈ L ( C ′ ς j β ) 2 ≤ ( N c + 1 ) ( C ′ ς ˉ β ) 2 . \left\|\sum_{j}z_j\right\|_{L^2(\widetilde{I}^{c};W^{1,\infty})}^2\leq\sum_{j\notin\mathcal{L}}(C'|\Sigma_j|^{1/2}\varsigma_j^\beta)^2+\sum_{j\in\mathcal{L}}(C'\varsigma_j^\beta)^2\leq(N_c+1)(C'\bar{\varsigma}^{\beta})^2. j ∑ z j L 2 ( I c ; W 1 , ∞ ) 2 ≤ j ∈ / L ∑ ( C ′ ∣ Σ j ∣ 1/2 ς j β ) 2 + j ∈ L ∑ ( C ′ ς j β ) 2 ≤ ( N c + 1 ) ( C ′ ς ˉ β ) 2 .
Together with the means, this is at most δ \delta δ by (6.21).
(W4) holds since nothing changes before t 1 , 0 − 5 τ 4 > inf I ~ t_{1,0}-\frac{5\tau}{4}>\inf\widetilde{I} t 1 , 0 − 4 5 τ > inf I .
The order of choices is: u , R , I ~ , τ ~ , δ u,R,\widetilde{I},\widetilde{\tau},\delta u , R , I , τ , δ (hence N c N_c N c ); then the level-1 bounds and constants C 1 ′ , ℓ 0 , 1 ′ , c ⋆ , 1 ′ C'_1,\ell'_{0,1},c'_{\star,1} C 1 ′ , ℓ 0 , 1 ′ , c ⋆ , 1 ′ ; then the level-2 bounds and constants C 2 ′ , ℓ 0 , 2 ′ , c ⋆ , 2 ′ , K 2 ′ , C 2 ′ C'_2,\ell'_{0,2},c'_{\star,2},K'_2,C'_2 C 2 ′ , ℓ 0 , 2 ′ , c ⋆ , 2 ′ , K 2 ′ , C 2 ′ ; then T T T and τ \tau τ ; then pass 1, which fixes u ~ \widetilde{u} u and R ‾ \overline{R} R ; then N J N_J N J and δ ⋆ \delta_\star δ ⋆ ; then the frequencies of the convex integration steps, in time order. Every constant depends only on objects fixed earlier. The large gradients of the convex integration perturbations do not enter, since all kinetic estimates use only the flow of the base and the L t 1 L q L^1_tL^q L t 1 L q norm of the perturbation. □
The first step
The initial approximate solution of Lemma 2.7 is exact on [ 0 , t 0 ] [0,t_0] [ 0 , t 0 ] but its stress may reach t = 1 t=1 t = 1 , so it is not well-prepared. The following variant of Proposition 6.3 produces a well-prepared solution.
Proposition 6.15 (Preparation). Let ( u 0 , p 0 , R 0 ) (u_0,p_0,R_0) ( u 0 , p 0 , R 0 ) be a smooth VNS–Reynolds solution on [ 0 , 1 ] [0,1] [ 0 , 1 ] with R 0 = 0 R_0=0 R 0 = 0 on [ 0 , t 0 ] [0,t_0] [ 0 , t 0 ] , t 0 ∈ ( 0 , 1 ) t_0\in(0,1) t 0 ∈ ( 0 , 1 ) , and let δ > 0 \delta>0 δ > 0 . Then there is a smooth VNS–Reynolds solution ( u 1 , p 1 , R 1 ) (u_1,p_1,R_1) ( u 1 , p 1 , R 1 ) , well-prepared for some ( I 1 , τ 1 ) (I_1,\tau_1) ( I 1 , τ 1 ) with I 1 ⊂ ( t 0 2 , 1 ) I_1\subset\left(\frac{t_0}{2},1\right) I 1 ⊂ ( 2 t 0 , 1 ) , such that (W1)–(W3) of Proposition 6.3 hold with ( u , R ) = ( u 0 , R 0 ) (u,R)=(u_0,R_0) ( u , R ) = ( u 0 , R 0 ) , and u 1 = u 0 u_1=u_0 u 1 = u 0 on [ 0 , t 0 2 ] \left[0,\frac{t_0}{2}\right] [ 0 , 2 t 0 ] .
Proof. We run the proof of Proposition 6.3 with a single chain and no good segments. Only the changes are listed.
Geometry. Choose τ \tau τ and T = τ ε T=\tau^\varepsilon T = τ ε with 5 τ < T < t 0 4 5\tau<T<\frac{t_0}{4} 5 τ < T < 4 t 0 and 2 T ≤ ℓ 0 2T\le\ell_0 2 T ≤ ℓ 0 , where ℓ 0 \ell_0 ℓ 0 is fixed as in the paragraph “Uniformity of constants” with ( u , R ) = ( u 0 , R 0 ) (u,R)=(u_0,R_0) ( u , R ) = ( u 0 , R 0 ) and N c : = 0 N_c:=0 N c := 0 . Let t 1 < ⋯ < t N t_1<\cdots<t_N t 1 < ⋯ < t N be the points of T Z T\mathbb{Z} T Z in [ t 0 2 + T , 1 − T ] \left[\frac{t_0}{2}+T,1-T\right] [ 2 t 0 + T , 1 − T ] . Then t 1 ≤ t 0 2 + 2 T < t 0 t_1\le\frac{t_0}{2}+2T<t_0 t 1 ≤ 2 t 0 + 2 T < t 0 and t N > 1 − 2 T t_N>1-2T t N > 1 − 2 T . The windows are Σ i = [ t i − 1 , t i ] \Sigma_i=[t_{i-1},t_i] Σ i = [ t i − 1 , t i ] , 2 ≤ i ≤ N 2\le i\le N 2 ≤ i ≤ N , and the last window Σ N + 1 = [ t N , 1 ] \Sigma_{N+1}=[t_N,1] Σ N + 1 = [ t N , 1 ] , of length less than 2 T ≤ ℓ 0 2T\le\ell_0 2 T ≤ ℓ 0 ; together they cover [ t 1 , 1 ] ⊃ supp R 0 [t_1,1]\supset\operatorname{supp}R_0 [ t 1 , 1 ] ⊃ supp R 0 . With the zones of Section 6.5, put I 1 : = ⋃ i W ( t i ) I_1:=\bigcup_i W(t_i) I 1 := ⋃ i W ( t i ) and τ 1 : = τ \tau_1:=\tau τ 1 := τ . Since t 1 − 5 τ 2 > t 0 2 t_1-\frac{5\tau}{2}>\frac{t_0}{2} t 1 − 2 5 τ > 2 t 0 and t N + 5 τ 2 < 1 t_N+\frac{5\tau}{2}<1 t N + 2 5 τ < 1 , we have I 1 ⊂ ( t 0 2 , 1 ) I_1\subset\left(\frac{t_0}{2},1\right) I 1 ⊂ ( 2 t 0 , 1 ) , and Lemma 6.14(b)–(e) hold with the same proofs.
Pass 1. For i = 2 , … , N i=2,\ldots,N i = 2 , … , N apply Proposition 6.8(a) on Σ i \Sigma_i Σ i with the two-sided cutoff of Proposition 6.3, and exit by the mean. On Σ N + 1 \Sigma_{N+1} Σ N + 1 apply Proposition 6.8(a) with a one-sided cutoff vanishing on [ t N , t N + τ 4 ] \left[t_N,t_N+\frac{\tau}{4}\right] [ t N , t N + 4 τ ] and equal to 1 1 1 on [ t N + τ 2 , 1 ] \left[t_N+\frac{\tau}{2},1\right] [ t N + 2 τ , 1 ] ; there is no exit. The base stress on each window is R 0 R_0 R 0 plus the accumulated persistent stress, so ς ≤ ∥ R 0 ∥ L ∞ L q + C \varsigma\le\lVert R_0\rVert_{L^\infty L^q}+C ς ≤ ∥ R 0 ∥ L ∞ L q + C and N S ≤ C N_S\le C N S ≤ C by Lemma 6.12, as in pass 1 of Proposition 6.3. The result ( u ˉ , R ˉ ) (\bar u,\bar R) ( u ˉ , R ˉ ) has R ˉ \bar R R ˉ supported in ⋃ i Z 0 ( t i ) \bigcup_i Z_0(t_i) ⋃ i Z 0 ( t i ) , since the stress vanishes where the cutoffs equal 1 1 1 , in particular on [ t N + τ 2 , 1 ] \left[t_N+\frac{\tau}{2},1\right] [ t N + 2 τ , 1 ] , and (48) holds with R = R 0 R=R_0 R = R 0 , by the same summation over fewer than T − 1 + 1 T^{-1}+1 T − 1 + 1 windows. Nothing is changed before t 1 t_1 t 1 .
Pass 2. For j = 1 , … , N j=1,\ldots,N j = 1 , … , N : apply Lemma 6.13 at t j t_j t j , and then Proposition 6.8(a) on Σ j ′ : = [ t j + τ , t j + 1 − 9 τ 8 ] \Sigma'_j:=[t_j+\tau,t_{j+1}-\frac{9\tau}{8}] Σ j ′ := [ t j + τ , t j + 1 − 8 9 τ ] for j < N j<N j < N , and on Σ N ′ : = [ t N + τ , 1 ] \Sigma'_N:=[t_N+\tau,1] Σ N ′ := [ t N + τ , 1 ] with a one-sided cutoff. All these segments have length at most 2 T ≤ ℓ 0 2T\le\ell_0 2 T ≤ ℓ 0 , so (b) is never used and L = ∅ \mathcal{L}=\varnothing L = ∅ . The paragraphs “Supports”, “Smallness” and “Conclusion” of the proof of Proposition 6.3 apply verbatim with N c = 0 N_c=0 N c = 0 and give (W1)–(W3). (W5) is not claimed.
Well-preparedness and (W4). After pass 2 the stress vanishes on the bulks of the re-solved segments, on [ t N + 9 τ 8 , 1 ] \left[t_N+\frac{9\tau}{8},1\right] [ t N + 8 9 τ , 1 ] (where the last cutoff equals 1 1 1 ), and on [ 0 , t 1 − 5 τ 4 ] \left[0,t_1-\frac{5\tau}{4}\right] [ 0 , t 1 − 4 5 τ ] , where nothing has changed and R 0 = 0 R_0=0 R 0 = 0 because t 1 < t 0 t_1<t_0 t 1 < t 0 . Elsewhere it is supported in ⋃ j Z 2 ( t j ) \bigcup_j Z_2(t_j) ⋃ j Z 2 ( t j ) , so ( u 1 , p 1 , R 1 ) (u_1,p_1,R_1) ( u 1 , p 1 , R 1 ) is well-prepared for ( I 1 , τ 1 ) (I_1,\tau_1) ( I 1 , τ 1 ) by Lemma 6.14(b), (c). The number of intervals in I 1 I_1 I 1 is N < T − 1 = τ 1 − ε N<T^{-1}=\tau_1^{-\varepsilon} N < T − 1 = τ 1 − ε , and 0 , 1 ∉ I 1 0,1\notin I_1 0 , 1 ∈ / I 1 . Finally u 1 = u 0 u_1=u_0 u 1 = u 0 on [ 0 , t 1 − 5 τ 4 ] ⊃ [ 0 , t 0 2 ] \left[0,t_1-\frac{5\tau}{4}\right]\supset\left[0,\frac{t_0}{2}\right] [ 0 , t 1 − 4 5 τ ] ⊃ [ 0 , 2 t 0 ] .
Proof of Theorem 6.2
First step. Let ( u 0 , p 0 , R 0 ) (u_0,p_0,R_0) ( u 0 , p 0 , R 0 ) be the initial approximate solution of Lemma 2.7, exact on [ 0 , t 0 ] [0,t_0] [ 0 , t 0 ] . Proposition 6.15 with δ = δ 0 \delta=\delta_0 δ = δ 0 gives a solution well-prepared for some ( I 1 , τ 1 ) (I_1,\tau_1) ( I 1 , τ 1 ) with I 1 ⊂ ( t 0 2 , 1 ) I_1\subset\left(\frac{t_0}{2},1\right) I 1 ⊂ ( 2 t 0 , 1 ) , equal to u 0 u_0 u 0 on [ 0 , t 0 2 ] \left[0,\frac{t_0}{2}\right] [ 0 , 2 t 0 ] .
Iteration. Apply Proposition 6.3 with δ n \delta_n δ n chosen as in the proof of Theorem 2.3, including the factor e − Λ n e^{-\Lambda_n} e − Λ n needed for Lemma 3.5. The limit is a weak solution, as in Section 2. Let S : = ⋂ n I n \mathcal{S}:=\bigcap_n I_n S := ⋂ n I n . Each I n I_n I n is covered by τ n − ε \tau_n^{-\varepsilon} τ n − ε intervals of length 5 τ n 5\tau_n 5 τ n , with τ n → 0 \tau_n\to0 τ n → 0 , so dim H S ≤ ε \dim_{\mathcal{H}}\mathcal{S}\le\varepsilon dim H S ≤ ε [12, Section 2.6].
Regularity off S \mathcal{S} S . Let K ⊂ ( 0 , 1 ] ∖ S K\subset(0,1]\setminus\mathcal{S} K ⊂ ( 0 , 1 ] ∖ S be a compact interval. Since S \mathcal{S} S is closed and inf K > 0 \inf K>0 inf K > 0 , there is η > 0 \eta>0 η > 0 such that K ′ : = [ inf K − η , sup K ] ⊂ ( 0 , 1 ] ∖ S K':=[\inf K-\eta,\sup K]\subset(0,1]\setminus\mathcal{S} K ′ := [ inf K − η , sup K ] ⊂ ( 0 , 1 ] ∖ S . Then K ′ ⊂ I m c K'\subset I_m^c K ′ ⊂ I m c for all large m m m . By (W5), ∑ m ∥ u m + 1 − u m ∥ L 2 ( K ′ ; W 1 , ∞ ) < ∞ \sum_m\lVert u_{m+1}-u_m\rVert_{L^2(K';W^{1,\infty})}<\infty ∑ m ∥ u m + 1 − u m ∥ L 2 ( K ′ ; W 1 , ∞ ) < ∞ , so u ∈ L 2 ( K ′ ; W 1 , ∞ ) u\in L^2(K';W^{1,\infty}) u ∈ L 2 ( K ′ ; W 1 , ∞ ) and the phase-space flow is Lipschitz on K ′ K' K ′ ; f f f is its push-forward by Lemma 3.6.
Trace. Let η 1 ∈ C ∞ ( R ) \eta_1\in C^\infty(\mathbb{R}) η 1 ∈ C ∞ ( R ) vanish on ( − ∞ , inf K − η / 2 ] (-\infty,\inf K-\eta/2] ( − ∞ , inf K − η /2 ] and equal 1 1 1 on [ inf K , ∞ ) [\inf K,\infty) [ inf K , ∞ ) , and set v : = η 1 u v:=\eta_1u v := η 1 u on K ′ K' K ′ . Since u u u is a weak solution, v v v is a weak solution of the Oseen problem (5.4) on K ′ K' K ′ , with drift b = u ∈ L 2 ( K ′ ; L ∞ ) b=u\in L^2(K';L^\infty) b = u ∈ L 2 ( K ′ ; L ∞ ) , zero initial value, and forcing
G = η 1 F [ f , u ] + η 1 ′ u ∈ L 2 ( K ′ ; L s ) for every s < ∞ , G=\eta_1F[f,u]+\eta'_1u\in L^2(K';L^s)\qquad\text{for every }s<\infty, G = η 1 F [ f , u ] + η 1 ′ u ∈ L 2 ( K ′ ; L s ) for every s < ∞ ,
because F [ f , u ] F[f,u] F [ f , u ] is bounded and u ∈ L 2 ( K ′ ; L ∞ ) u\in L^2(K';L^\infty) u ∈ L 2 ( K ′ ; L ∞ ) . By Propositions 5.6 and 5.9, v v v is the strong solution, so v ∈ M R ∞ ( K ′ ) v\in MR_\infty(K') v ∈ M R ∞ ( K ′ ) , and by Lemma 5.4(b), u = v ∈ C ( K ′ ; X s ) ∩ L 2 ( K ′ ; W 2 , s ) u=v\in C(K';X_s)\cap L^2(K';W^{2,s}) u = v ∈ C ( K ′ ; X s ) ∩ L 2 ( K ′ ; W 2 , s ) for every s < ∞ s<\infty s < ∞ . In particular u ( inf K ) ∈ X s u(\inf K)\in X_s u ( inf K ) ∈ X s for every s s s .
Uniqueness. Apply Corollary 5.11 on K K K , with the initial state ( u ( inf K ) , f ( inf K ) ) (u(\inf K),f(\inf K)) ( u ( inf K ) , f ( inf K )) ; the velocity datum lies in X s X_s X s for some s > d s>d s > d , and the kinetic datum is bounded and vanishes for ∣ v ∣ |v| ∣ v ∣ large. This gives the uniqueness statement. Infinitely many distinct solutions are obtained as in Corollary 2.4.
Strong continuations and precise statements
Theorem 2.3 assumes smooth data and a smooth local solution. Here we remove both assumptions. The kinetic datum is only bounded, and the velocity datum only needs to generate a solution that becomes strong after some time. Only Lipschitz bounds on the velocity are ever used:
the kinetic estimates of Section 3 depend on the base field through its Lipschitz norm;
the convex integration step can be run on a mollified background, at the price of a small error in the stress.
From now on, the kinetic datum f 0 ∈ L ∞ ( T d × R d ) f_0 \in L^\infty(\mathbb{T}^d \times\mathbb{R}^d) f 0 ∈ L ∞ ( T d × R d ) is only assumed nonnegative and to vanish for ∣ v ∣ ≥ R 0 v |v| \ge R_0^v ∣ v ∣ ≥ R 0 v ; it need not be smooth. The quantities M 0 M_0 M 0 , M ∞ M_\infty M ∞ , J 0 J_0 J 0 are defined as in Section 2.1.
Definition 7.1 (Strong continuation). Let u i n ∈ L 2 ( T d ) u_{\mathrm{in}} \in L^2(\mathbb{T}^d) u in ∈ L 2 ( T d ) be weakly divergence-free. A strong continuation of the data ( u i n , f 0 ) (u_{\mathrm{in}}, f_0) ( u in , f 0 ) is a weak solution ( u ˉ , f ˉ ) (\bar{u}, \bar{f}) ( u ˉ , f ˉ ) of (2.4) on [ 0 , T 0 ] [0,T_0] [ 0 , T 0 ] , 0 < T 0 ≤ 1 0 < T_0 \le1 0 < T 0 ≤ 1 , together with times 0 ≤ t 1 ≤ a < T 0 0 \le t_1 \le a < T_0 0 ≤ t 1 ≤ a < T 0 and an exponent q 0 ∈ ( d , ∞ ) ∩ [ 2 , ∞ ) q_0 \in(d,\infty) \cap[2,\infty) q 0 ∈ ( d , ∞ ) ∩ [ 2 , ∞ ) , such that:
(S1) u ˉ ∈ C w ( [ 0 , T 0 ] ; L 2 ) ∩ L 1 ( 0 , T 0 ; W 1 , 1 ) ∩ L 1 ( 0 , T 0 ; L ∞ ) \bar{u} \in C_{\mathrm{w}}([0,T_0];L^2) \cap L^1(0,T_0;W^{1,1}) \cap L^1(0,T_0;L^\infty) u ˉ ∈ C w ([ 0 , T 0 ] ; L 2 ) ∩ L 1 ( 0 , T 0 ; W 1 , 1 ) ∩ L 1 ( 0 , T 0 ; L ∞ ) , and u ˉ ∈ C ( [ t 1 , T 0 ] ; L 2 ) ∩ L 2 ( t 1 , T 0 ; L ∞ ) \bar{u} \in C([t_1,T_0];L^2) \cap L^2(t_1,T_0;L^\infty) u ˉ ∈ C ([ t 1 , T 0 ] ; L 2 ) ∩ L 2 ( t 1 , T 0 ; L ∞ ) ;
(S2) ∇ u ˉ ∈ L 1 ( t 1 , T 0 ; L ∞ ) \nabla\bar{u} \in L^1(t_1,T_0;L^\infty) ∇ u ˉ ∈ L 1 ( t 1 , T 0 ; L ∞ ) ;
(S3) u ˉ ∈ M R q 0 ( [ a , T 0 ] ) \bar{u} \in MR_{q_0}([a,T_0]) u ˉ ∈ M R q 0 ([ a , T 0 ]) .
Data admitting a strong continuation are called admissible .
Condition (S2) asks for a Lipschitz velocity only after time t 1 t_1 t 1 . Before t 1 t_1 t 1 the solution may be rough. By (S1) and Lemma 3.6, f ˉ \bar{f} f ˉ is the unique bounded solution of the kinetic equation with field u ˉ \bar{u} u ˉ , and it is continuous with values in L 1 L^1 L 1 .
Theorem 7.2 (Nonuniqueness for admissible data). Let d ≥ 2 d \ge2 d ≥ 2 , 1 ≤ p < 2 1 \le p < 2 1 ≤ p < 2 , 1 ≤ q < ∞ 1 \le q < \infty 1 ≤ q < ∞ . Let ( u i n , f 0 ) (u_{\mathrm{in}}, f_0) ( u in , f 0 ) be admissible , with strong continuation ( u ˉ , f ˉ , t 1 , a , T 0 , q 0 ) (\bar{u}, \bar{f}, t_1, a, T_0, q_0) ( u ˉ , f ˉ , t 1 , a , T 0 , q 0 ) . Then there are infinitely many distinct weak solutions ( u , f ) (u,f) ( u , f ) of (2.4) on [ 0 , 1 ] [0,1] [ 0 , 1 ] with data ( u i n , f 0 ) (u_{\mathrm{in}},f_0) ( u in , f 0 ) such that:
(1) u ∈ L t , x 2 ∩ L 1 ( 0 , 1 ; L ∞ ) u \in L^2_{t,x} \cap L^1(0,1;L^\infty) u ∈ L t , x 2 ∩ L 1 ( 0 , 1 ; L ∞ ) and u ∈ L p ( t 1 , 1 ; L ∞ ) ∩ L 1 ( t 1 , 1 ; W 1 , q ) u \in L^p(t_1,1;L^\infty) \cap L^1(t_1,1;W^{1,q}) u ∈ L p ( t 1 , 1 ; L ∞ ) ∩ L 1 ( t 1 , 1 ; W 1 , q ) ; if moreover u ˉ ∈ L p ( 0 , t 1 ; L ∞ ) \bar{u} \in L^p(0,t_1;L^\infty) u ˉ ∈ L p ( 0 , t 1 ; L ∞ ) , then u ∈ L p ( 0 , 1 ; L ∞ ) u \in L^p(0,1;L^\infty) u ∈ L p ( 0 , 1 ; L ∞ ) ;
(2) ( u , f ) = ( u ˉ , f ˉ ) (u,f)=(\bar{u},\bar{f}) ( u , f ) = ( u ˉ , f ˉ ) on [ 0 , t 1 + τ ] [0,t_1+\tau] [ 0 , t 1 + τ ] , for some τ > 0 \tau>0 τ > 0 that does not depend on the solution;
(3) f f f is bounded, vanishes for ∣ v ∣ |v| ∣ v ∣ large, belongs to C ( [ 0 , 1 ] ; L s ) C([0,1];L^s) C ([ 0 , 1 ] ; L s ) for every s < ∞ s<\infty s < ∞ , and is the unique bounded (hence renormalized) solution of the kinetic equation with field u u u ;
(4) the total momentum is conserved.
Infinitely many of these solutions are not of Leray–Hopf type.
Theorem 7.3 (Admissible data). Let f 0 ∈ L ∞ ( T d × R d ) f_0 \in L^\infty(\mathbb{T}^d \times\mathbb{R}^d) f 0 ∈ L ∞ ( T d × R d ) be nonnegative and vanish for ∣ v ∣ ≥ R 0 v |v| \ge R_0^v ∣ v ∣ ≥ R 0 v , and let u i n u_{\mathrm{in}} u in be divergence-free. The data ( u i n , f 0 ) (u_{\mathrm{in}},f_0) ( u in , f 0 ) are admissible in each of the following cases:
(a) d ≥ 2 d \ge2 d ≥ 2 and u i n ∈ B q 0 , 2 1 ( T d ) u_{\mathrm{in}} \in B^{1}_{q_0,2}(\mathbb{T}^d) u in ∈ B q 0 , 2 1 ( T d ) for some q 0 ∈ ( d , ∞ ) ∩ [ 2 , ∞ ) q_0 \in(d,\infty) \cap[2,\infty) q 0 ∈ ( d , ∞ ) ∩ [ 2 , ∞ ) . This includes H s H^s H s for s > d 2 s > \frac{d}{2} s > 2 d and C 1 , α C^{1,\alpha} C 1 , α for α > 0 \alpha> 0 α > 0 .
(b) d ∈ { 2 , 3 } d \in\{2,3\} d ∈ { 2 , 3 } and u i n ∈ H 1 ( T d ) u_{\mathrm{in}} \in H^1(\mathbb{T}^d) u in ∈ H 1 ( T d ) .
(c) d = 2 d=2 d = 2 and u i n ∈ L 2 ( T 2 ) u_{\mathrm{in}} \in L^2(\mathbb{T}^2) u in ∈ L 2 ( T 2 ) ;
(d) d = 3 d=3 d = 3 and u i n ∈ L 2 ( T 3 ) u_{\mathrm{in}} \in L^2(\mathbb{T}^3) u in ∈ L 2 ( T 3 ) .
In cases (a) and (b) one can take t 1 = 0 t_1=0 t 1 = 0 . In cases (a)–(c), u ˉ ∈ L 2 ( 0 , T 0 ; L ∞ ) \bar{u} \in L^2(0,T_0;L^\infty) u ˉ ∈ L 2 ( 0 , T 0 ; L ∞ ) .
Remark 7.4. (i) In case (c), Theorem 7.2 gives nonuniqueness for every finite-energy velocity datum in two dimensions. By [20 Theorem 1] , stated there for R 2 \mathbb{R}^2 R 2 and T 2 \mathbb{T}^2 T 2 , there is at most one weak solution in the sense of [20 Definition 1] ; the proof gives uniqueness on every interval [ 0 , T ] [0,T] [ 0 , T ] . The admissibility condition (4) of [20 Definition 2] holds: if f 0 f_0 f 0 is bounded and supported for ∣ v ∣ ≥ R 0 v |v| \ge R_0^v ∣ v ∣ ≥ R 0 v , then the supremum of f 0 f_0 f 0 over T 2 × B ( e t v , R ) \mathbb{T}^2 \times B(e^t v,R) T 2 × B ( e t v , R ) is at most ∥ f 0 ∥ L ∞ 1 { e t ∣ v ∣ ≤ R + R 0 v } \|f_0\|_{L^\infty}\mathbf{1}_{\{e^t|v|\le R+R_0^v\}} ∥ f 0 ∥ L ∞ 1 { e t ∣ v ∣ ≤ R + R 0 v } , so ( 1 + ∣ v ∣ 2 ) sup T 2 × B ( e t v , R ) f 0 (1+|v|^2)\sup_{\mathbb{T}^2 \times B(e^t v,R)}f_0 ( 1 + ∣ v ∣ 2 ) sup T 2 × B ( e t v , R ) f 0 is bounded in L 1 ( R v 2 ) L^1(\mathbb{R}^2_v) L 1 ( R v 2 ) uniformly in t ≥ 0 t \ge0 t ≥ 0 . Now let ( u , f ) (u,f) ( u , f ) be a weak solution in the sense of Definition 2.1 with u ∈ L ∞ ( 0 , 1 ; L 2 ) ∩ L 2 ( 0 , 1 ; H 1 ) u \in L^\infty(0,1;L^2) \cap L^2(0,1;H^1) u ∈ L ∞ ( 0 , 1 ; L 2 ) ∩ L 2 ( 0 , 1 ; H 1 ) . Then ρ f \rho_f ρ f and j f j_f j f are bounded, so F [ f , u ] ∈ L 2 ( 0 , 1 ; L 2 ) F[f,u] \in L^2(0,1;L^2) F [ f , u ] ∈ L 2 ( 0 , 1 ; L 2 ) , and u ⊗ u ∈ L 2 ( 0 , 1 ; L 2 ) u \otimes u \in L^2(0,1;L^2) u ⊗ u ∈ L 2 ( 0 , 1 ; L 2 ) by Ladyzhenskaya’s inequality. From Definition 2.1(iv), ∂ t u ∈ L 2 ( 0 , 1 ; H d i v − 1 ) \partial_t u \in L^2(0,1;H^{-1}_{\mathrm{div}}) ∂ t u ∈ L 2 ( 0 , 1 ; H div − 1 ) , so u ∈ C ( [ 0 , 1 ] ; H d i v − 1 ) u \in C([0,1];H^{-1}_{\mathrm{div}}) u ∈ C ([ 0 , 1 ] ; H div − 1 ) and the weak formulation holds for every t t t in the form required by [20 Definition 1] ; moreover f ∈ L ∞ ( 0 , 1 ; L 1 ∩ L ∞ ) f \in L^\infty(0,1;L^1 \cap L^\infty) f ∈ L ∞ ( 0 , 1 ; L 1 ∩ L ∞ ) . Hence ( u , f ) (u,f) ( u , f ) is a solution on [ 0 , 1 ] [0,1] [ 0 , 1 ] in the sense of [20 Definition 1] , that is, it satisfies the requirements of that definition, which is stated for global solutions, on [ 0 , 1 ] [0,1] [ 0 , 1 ] . The proof of [20 Theorem 1] is local in time, as noted above, and applies to such solutions. No energy inequality is needed. Consequently at most one of the solutions of Theorem 7.2 lies in L ∞ ( 0 , 1 ; L 2 ) ∩ L 2 ( 0 , 1 ; H 1 ) L^\infty(0,1;L^2)\cap L^2(0,1;H^1) L ∞ ( 0 , 1 ; L 2 ) ∩ L 2 ( 0 , 1 ; H 1 ) , in particular at most one is of Leray–Hopf type.
(ii) Condition (S2) only concerns times after t 1 t_1 t 1 . For d = 3 d=3 d = 3 we expect critical data, such as u i n ∈ H 1 / 2 u_{\mathrm{in}}\in H^{1/2} u in ∈ H 1/2 , to be admissible: a Fujita–Kato type solution with bounded drag force is smooth for t > 0 t>0 t > 0 . We do not pursue this.
(iii) Case (d) rests on the bound u ∈ L l o c 1 L ∞ u\in L^1_{\mathrm{loc}}L^\infty u ∈ L loc 1 L ∞ for Leray solutions of the three-dimensional system, due to Han-Kwan, Miot, Moussa and Moyano [21 Lemma B.1] ; Lemma 11.2 gives a self-contained proof on T 3 \mathbb{T}^3 T 3 . It keeps the velocity support of f f f bounded before t 1 t_1 t 1 . For d ≥ 4 d\geq4 d ≥ 4 this bound already fails for the heat flow of a general L 2 L^2 L 2 datum, since ∥ e t Δ u i n ∥ L ∞ ∼ t − d / 4 \|e^{t\Delta}u_{\mathrm{in}}\|_{L^\infty}\sim t^{-d/4} ∥ e t Δ u in ∥ L ∞ ∼ t − d /4 . The H 1 H^1 H 1 theory of Proposition 9.2 is also restricted to d ≤ 3 d\leq3 d ≤ 3 . Arbitrary L 2 L^2 L 2 data for d ≥ 4 d\geq4 d ≥ 4 remain open.
(iv) In case (d) the solutions of Theorem 7.2 only satisfy u ∈ L t 1 L ∞ u\in L^1_tL^\infty u ∈ L t 1 L ∞ near t = 0 t=0 t = 0 . This is the integrability of the Leray solution they coincide with initially.
The construction on Lipschitz backgrounds
Kinetic lemmas for Lipschitz fields
Let V s \mathcal{V}_s V s be the space of measurable U : ( 0 , 1 ) × T d → R d U:(0,1)\times\mathbb{T}^d\to\mathbb{R}^d U : ( 0 , 1 ) × T d → R d such that U ( t ) U(t) U ( t ) is Lipschitz for a.e. t t t and
∥ U ∥ L 1 W 1 , ∞ : = ∫ 0 1 ∥ U ( t ) ∥ W 1 , ∞ d t < ∞ ; \|U\|_{L^1W^{1,\infty}}:=\int_0^1\|U(t)\|_{W^{1,\infty}}\,dt<\infty; ∥ U ∥ L 1 W 1 , ∞ := ∫ 0 1 ∥ U ( t ) ∥ W 1 , ∞ d t < ∞ ;
we write V s = L 1 ( 0 , 1 ; W 1 , ∞ ) \mathcal{V}_s=L^1(0,1;W^{1,\infty}) V s = L 1 ( 0 , 1 ; W 1 , ∞ ) with this meaning. Such a U U U is a Carathéodory function, so t ↦ U ( t ) ∈ C ( T d ) t\mapsto U(t)\in C(\mathbb{T}^d) t ↦ U ( t ) ∈ C ( T d ) is weakly measurable; since C ( T d ) C(\mathbb{T}^d) C ( T d ) is separable, Pettis’ theorem shows that it is strongly measurable, and U ∈ L 1 ( 0 , 1 ; C ( T d ) ) U\in L^1(0,1;C(\mathbb{T}^d)) U ∈ L 1 ( 0 , 1 ; C ( T d )) as a Bochner function. For U ∈ V s U\in\mathcal{V}_s U ∈ V s the field b U = ( v , U − v ) b_U=(v,U-v) b U = ( v , U − v ) is Lipschitz in z z z , with constant 1 + ∥ ∇ U ( t ) ∥ L ∞ ∈ L 1 ( 0 , 1 ) 1+\|\nabla U(t)\|_{L^\infty}\in L^1(0,1) 1 + ∥∇ U ( t ) ∥ L ∞ ∈ L 1 ( 0 , 1 ) , and grows linearly. So it has a unique Carathéodory flow Z U Z_U Z U , bi-Lipschitz in z z z , with det D z Z U ( t ; s , ⋅ ) = e − d ( t − s ) \det D_zZ_U(t;s,\cdot)=e^{-d(t-s)} det D z Z U ( t ; s , ⋅ ) = e − d ( t − s ) ; see the proof of Proposition 5.10(c). We set f [ U ] ( t ) : = Z U ( t ; 0 , ⋅ ) # f 0 f[U](t):=Z_U(t;0,\cdot)_\#f_0 f [ U ] ( t ) := Z U ( t ; 0 , ⋅ ) # f 0 , and keep the notation ρ [ U ] , j [ U ] , J [ U ] \rho[U],j[U],J[U] ρ [ U ] , j [ U ] , J [ U ] of § 2.1 .
Lemma 8.1. The following hold.
(a) Lemma 3.1 holds for U ∈ V s U\in\mathcal{V}_s U ∈ V s , except for the statements about C 1 C^1 C 1 and smooth regularity of f f f . In their place: f [ U ] f[U] f [ U ] is the unique bounded weak solution of (2.3), it belongs to C ( [ 0 , 1 ] ; L 1 ) C([0,1];L^1) C ([ 0 , 1 ] ; L 1 ) , and J [ U ] ∈ W 1 , 1 ( 0 , 1 ) J[U]\in W^{1,1}(0,1) J [ U ] ∈ W 1 , 1 ( 0 , 1 ) satisfies (3.2) a.e.
(b) The second inequality in (3.3) holds for u , w ~ ∈ V s u,\widetilde w\in\mathcal{V}_s u , w ∈ V s , and for every ψ \psi ψ that is continuous on T d × R d \mathbb{T}^d\times\mathbb{R}^d T d × R d and Lipschitz on T d × B R v + 1 v \mathbb{T}^d\times B_{R^{v}+1}^{v} T d × B R v + 1 v , with [ ψ ] R v [\psi]_{R^v} [ ψ ] R v replaced by [ ψ ] R v + 1 [\psi]_{R^{v}+1} [ ψ ] R v + 1 , defined through the essential suprema of ∣ ∇ x ψ ∣ |\nabla_x\psi| ∣ ∇ x ψ ∣ and ∣ ∇ v ψ ∣ |\nabla_v\psi| ∣ ∇ v ψ ∣ on T d × B R v + 1 v \mathbb{T}^d\times B_{R^{v}+1}^{v} T d × B R v + 1 v .
(c) Lemma 3.3 holds for every h ∈ L 1 ( T d ; R d ) h\in L^1(\mathbb{T}^d;\mathbb{R}^d) h ∈ L 1 ( T d ; R d ) , with R h : = K ∗ h ∈ L r \mathcal{R}h:=K*h\in L^r R h := K ∗ h ∈ L r .
(d) Proposition 3.4 holds for u , w ~ ∈ V s u,\widetilde w\in\mathcal{V}_s u , w ∈ V s , without assuming that h h h is smooth, with L : = ∥ u ∥ L 1 ( 0 , 1 ; W 1 , ∞ ) L:=\|u\|_{L^1(0,1;W^{1,\infty})} L := ∥ u ∥ L 1 ( 0 , 1 ; W 1 , ∞ ) .
(e) Lemma 3.5 holds for u n ∈ V s u_n\in\mathcal{V}_s u n ∈ V s .
(f) Let U ∈ V s ∩ C ( [ 0 , 1 ] ; L 2 ) U\in\mathcal{V}_s\cap C([0,1];L^2) U ∈ V s ∩ C ([ 0 , 1 ] ; L 2 ) be divergence-free. Then Lemma 3.8 holds with the following change. In (a), c ∈ C ( [ 0 , 1 ] ; R d ) c\in C([0,1];\mathbb{R}^d) c ∈ C ([ 0 , 1 ] ; R d ) , f ( U + c ) ∈ W 1 , 1 ( 0 , 1 ) f(U+c)\in W^{1,1}(0,1) f ( U + c ) ∈ W 1 , 1 ( 0 , 1 ) , and the mean equation holds a.e. If moreover f U ∈ W 1 , 1 ( 0 , 1 ) fU\in W^{1,1}(0,1) f U ∈ W 1 , 1 ( 0 , 1 ) , then c ∈ W 1 , 1 ( 0 , 1 ) c\in W^{1,1}(0,1) c ∈ W 1 , 1 ( 0 , 1 ) . In (c), u , w ∈ V s ∩ C ( [ 0 , 1 ] ; L 2 ) u,w\in\mathcal{V}_s\cap C([0,1];L^2) u , w ∈ V s ∩ C ([ 0 , 1 ] ; L 2 ) .
Remark 8.2 (Bounded data at a later time). Lemmas 3.6 and 8.1(a),(b) also hold, with the same proofs and with constants computed from the norms of the datum, for every bounded nonnegative kinetic datum vanishing for ∣ v ∣ ≥ R v |v|\geq R^v ∣ v ∣ ≥ R v , prescribed at any time t 0 ∈ [ 0 , 1 ) t_0\in[0,1) t 0 ∈ [ 0 , 1 ) , on any interval [ t 0 , T ] ⊂ [ 0 , 1 ] [t_0,T]\subset[0,1] [ t 0 , T ] ⊂ [ 0 , 1 ] ; only the C 1 C^1 C 1 statements are lost. Indeed, the push-forward argument does not use the smoothness of f 0 f_0 f 0 , and for a fixed flow both sides of the bound in (b) are linear in the datum and continuous with respect to its L 1 L^1 L 1 norm, so (b) follows by approximating the datum by smooth ones. Translation in time and restriction to subintervals are used below without comment.
Proof. Approximation. Extend U ∈ V s U \in V_s U ∈ V s by zero outside [ 0 , 1 ] [0,1] [ 0 , 1 ] , and let U ε U_\varepsilon U ε be its mollification in ( t , x ) (t,x) ( t , x ) . Since U ∈ L 1 ( 0 , 1 ; C ( T d ) ) U \in L^1(0,1;C(\mathbb{T}^d)) U ∈ L 1 ( 0 , 1 ; C ( T d )) as a Bochner function, U ε → U U_\varepsilon\to U U ε → U in L 1 ( 0 , 1 ; L ∞ ) L^1(0,1;L^\infty) L 1 ( 0 , 1 ; L ∞ ) . Moreover, since ∇ U ε = ( ∇ U ) ε \nabla U_\varepsilon=(\nabla U)_\varepsilon ∇ U ε = ( ∇ U ) ε and by Minkowski’s inequality,
∥ U ε ∥ L 1 L ∞ ≤ ∥ U ∥ L 1 L ∞ , ∥ ∇ U ε ∥ L 1 L ∞ ≤ ∥ ∇ U ∥ L 1 L ∞ = : Λ . \|U_\varepsilon\|_{L^1L^\infty}\leq\|U\|_{L^1L^\infty},\qquad\|\nabla U_\varepsilon\|_{L^1L^\infty}\leq\|\nabla U\|_{L^1L^\infty}=:\Lambda. ∥ U ε ∥ L 1 L ∞ ≤ ∥ U ∥ L 1 L ∞ , ∥∇ U ε ∥ L 1 L ∞ ≤ ∥∇ U ∥ L 1 L ∞ =: Λ.
By Gronwall’s inequality,
∣ Z U ε ( t ; s , z ) − Z U ( t ; s , z ) ∣ ≤ e 1 + Λ ∫ s t ∥ U ε − U ∥ L ∞ → 0 , |Z_{U_\varepsilon}(t;s,z)-Z_U(t;s,z)|\leq e^{1+\Lambda}\int_s^t\|U_\varepsilon-U\|_{L^\infty}\to0, ∣ Z U ε ( t ; s , z ) − Z U ( t ; s , z ) ∣ ≤ e 1 + Λ ∫ s t ∥ U ε − U ∥ L ∞ → 0 ,
uniformly in ( t , s , z ) (t,s,z) ( t , s , z ) . Also let f 0 η ∈ C ∞ f_0^\eta\in C^\infty f 0 η ∈ C ∞ satisfy 0 ≤ f 0 η ≤ M 0 ∞ 0\leq f_0^\eta\leq M_0^\infty 0 ≤ f 0 η ≤ M 0 ∞ , f 0 η → f 0 f_0^\eta\to f_0 f 0 η → f 0 in L 1 L^1 L 1 , and f 0 η = 0 f_0^\eta=0 f 0 η = 0 for ∣ v ∣ ≥ R 0 v + η |v|\geq R_0^v+\eta ∣ v ∣ ≥ R 0 v + η .
(a). The push-forward identities, the Jacobian and the support bound pass to the limit from ( U ε , f 0 η ) (U_\varepsilon,f_0^\eta) ( U ε , f 0 η ) ; alternatively, they follow directly from the Carathéodory flow, as in Proposition 5.10(c). The push-forward is a weak solution: for φ ∈ C c ∞ ( [ 0 , 1 ) × T d × R d ) \varphi\in C_c^\infty([0,1)\times\mathbb{T}^d\times\mathbb{R}^d) φ ∈ C c ∞ ([ 0 , 1 ) × T d × R d ) and every z z z , the function t ↦ φ ( t , Z U ( t ; 0 , z ) ) t\mapsto\varphi(t,Z_U(t;0,z)) t ↦ φ ( t , Z U ( t ; 0 , z )) is absolutely continuous with derivative ( ∂ t φ + v ⋅ ∇ x φ + ( U − v ) ⋅ ∇ v φ ) ( t , Z U ( t ; 0 , z ) ) (\partial_t\varphi+v\cdot\nabla_x\varphi+(U-v)\cdot\nabla_v\varphi)(t,Z_U(t;0,z)) ( ∂ t φ + v ⋅ ∇ x φ + ( U − v ) ⋅ ∇ v φ ) ( t , Z U ( t ; 0 , z )) for a.e. t t t , by the chain rule for the Carathéodory solution Z U Z_U Z U ; integrating in t t t and against f 0 f_0 f 0 , and using the push-forward identity, gives Definition 2.1(iii). Uniqueness and C ( [ 0 , 1 ] ; L 1 ) C([0,1];L^1) C ([ 0 , 1 ] ; L 1 ) follow from Lemma 3.6, since U ∈ L 1 W 1 , 1 ∩ L 1 L ∞ U\in L^1W^{1,1}\cap L^1L^\infty U ∈ L 1 W 1 , 1 ∩ L 1 L ∞ . For the momentum, J [ U ] ( t ) = ∫ f 0 ( z ) V ( t ; 0 , z ) d z J[U](t)=\int f_0(z)V(t;0,z)\,dz J [ U ] ( t ) = ∫ f 0 ( z ) V ( t ; 0 , z ) d z , where V ( ⋅ ; 0 , z ) V(\cdot;0,z) V ( ⋅ ; 0 , z ) is absolutely continuous with V ˙ = U ( t , X ) − V \dot V=U(t,X)-V V ˙ = U ( t , X ) − V , bounded by an integrable function on supp f 0 \operatorname{supp}f_0 supp f 0 . Hence J [ U ] ∈ W 1 , 1 J[U]\in W^{1,1} J [ U ] ∈ W 1 , 1 and
J [ U ] ′ = ∫ f 0 ( U ( t , X ) − V ) = ∬ ( U − v ) f [ U ] . J[U]'=\int f_0(U(t,X)-V)=\iint(U-v)f[U]. J [ U ] ′ = ∫ f 0 ( U ( t , X ) − V ) = ∬ ( U − v ) f [ U ] .
(b). First let ψ ∈ C 1 \psi\in C^1 ψ ∈ C 1 . Apply Lemma 3.2 to ( u ε , w ~ ε ) (u_\varepsilon,\widetilde w_\varepsilon) ( u ε , w ε ) with datum f 0 η f_0^\eta f 0 η . The left side equals ∫ f 0 η ( ψ ( Z u ε + w ~ ε ) − ψ ( Z u ε ) ) \int f_0^\eta(\psi(Z_{u_\varepsilon+\widetilde w_\varepsilon})-\psi(Z_{u_\varepsilon})) ∫ f 0 η ( ψ ( Z u ε + w ε ) − ψ ( Z u ε )) , which converges as ε , η → 0 \varepsilon,\eta\to0 ε , η → 0 . On the right side, Λ ε ≤ Λ \Lambda_\varepsilon\leq\Lambda Λ ε ≤ Λ , M 0 η → M 0 M_0^\eta\to M_0 M 0 η → M 0 , and R 1 , ε v ≤ R 1 v + η R_{1,\varepsilon}^v\leq R_1^v+\eta R 1 , ε v ≤ R 1 v + η . Moreover [ ψ ] R 1 v + η → [ ψ ] R 1 v [\psi]_{R_1^v+\eta}\to[\psi]_{R_1^v} [ ψ ] R 1 v + η → [ ψ ] R 1 v , since ∇ ψ \nabla\psi ∇ ψ is continuous.
Now let ψ \psi ψ be continuous, and Lipschitz on T d × B R 1 v + 1 \mathbb{T}^d\times B_{R_1^v+1} T d × B R 1 v + 1 . Only the values of ψ \psi ψ on T d × B R 1 v \mathbb{T}^d\times B_{R_1^v} T d × B R 1 v enter the left side, since all kinetic states involved are supported there. Let ψ κ : = ψ ∗ η κ \psi_\kappa:=\psi*\eta_\kappa ψ κ := ψ ∗ η κ be a mollification in z z z , κ ∈ ( 0 , 1 ) \kappa\in(0,1) κ ∈ ( 0 , 1 ) . Then ψ κ ∈ C 1 \psi_\kappa\in C^1 ψ κ ∈ C 1 , ψ κ → ψ \psi_\kappa\to\psi ψ κ → ψ uniformly on T d × B R 1 v \mathbb{T}^d\times B_{R_1^v} T d × B R 1 v , and [ ψ κ ] R 1 v ≤ [ ψ ] R 1 v + κ ≤ [ ψ ] R 1 v + 1 [\psi_\kappa]_{R_1^v}\leq[\psi]_{R_1^v+\kappa}\leq[\psi]_{R_1^v+1} [ ψ κ ] R 1 v ≤ [ ψ ] R 1 v + κ ≤ [ ψ ] R 1 v + 1 . Applying the C 1 C^1 C 1 case to ψ κ \psi_\kappa ψ κ and letting κ → 0 \kappa\to0 κ → 0 gives the claim.
(c). K ∈ L r K\in L^r K ∈ L r by (2.2) and r < d d − 1 r<\frac{d}{d-1} r < d − 1 d , so R h = K ∗ h ∈ L r R_h=K*h\in L^r R h = K ∗ h ∈ L r by Young’s inequality. Let h ε = h ∗ η ε h_\varepsilon=h*\eta_\varepsilon h ε = h ∗ η ε . Then ∥ h ε ∥ L 1 ≤ ∥ h ∥ L 1 \|h_\varepsilon\|_{L^1}\leq\|h\|_{L^1} ∥ h ε ∥ L 1 ≤ ∥ h ∥ L 1 and ∥ h ε ∥ ( C 1 ) ∗ ≤ ∥ h ∥ ( C 1 ) ∗ \|h_\varepsilon\|_{(C^1)^*}\leq\|h\|_{(C^1)^*} ∥ h ε ∥ ( C 1 ) ∗ ≤ ∥ h ∥ ( C 1 ) ∗ , because φ ↦ φ ∗ η ˇ ε \varphi\mapsto\varphi*\check\eta_\varepsilon φ ↦ φ ∗ η ˇ ε preserves the C 1 C^1 C 1 norm. Also R h ε = ( R h ) ∗ η ε → R h Rh_\varepsilon=(Rh)*\eta_\varepsilon\to Rh R h ε = ( R h ) ∗ η ε → R h in L r L^r L r . Apply (3.7) to h ε h_\varepsilon h ε and let ε → 0 \varepsilon\to0 ε → 0 .
(d). In the proof of Proposition 3.4, use (b) and (c) instead of Lemmas 3.2 and 3.3, and keep the time dependence of the constants. Fix t t t outside the null set where u ( t ) u(t) u ( t ) fails to be Lipschitz; the bounds below then hold for a.e. t t t , which suffices since they are integrated in time. The test function ψ ( x , v ) = ( v − u ( t , x ) ) ⋅ ϕ ( x ) \psi(x,v)=(v-u(t,x))\cdot\phi(x) ψ ( x , v ) = ( v − u ( t , x )) ⋅ ϕ ( x ) is only Lipschitz, since u ( t ) ∈ W 1 , ∞ u(t)\in W^{1,\infty} u ( t ) ∈ W 1 , ∞ ; it is admissible in (b), and its gradient bounds below hold for a.e. x x x and all ∣ v ∣ ≤ R 1 v + 1 |v|\leq R_1^v+1 ∣ v ∣ ≤ R 1 v + 1 , so that [ ψ ] R 1 v + 1 ≤ R 1 v + 1 + 2 ∥ u ( t ) ∥ W 1 , ∞ [\psi]_{R_1^v+1}\leq R_1^v+1+2\|u(t)\|_{W^{1,\infty}} [ ψ ] R 1 v + 1 ≤ R 1 v + 1 + 2∥ u ( t ) ∥ W 1 , ∞ ; we absorb the + 1 +1 + 1 into R 1 v R_1^v R 1 v . At time t t t :
[ ψ ] R 1 v ≤ R 1 v + 2 ∥ u ( t ) ∥ W 1 , ∞ [\psi]_{R_1^v}\leq R_1^v+2\|u(t)\|_{W^{1,\infty}} [ ψ ] R 1 v ≤ R 1 v + 2∥ u ( t ) ∥ W 1 , ∞ ;
∥ h 1 ( t ) ∥ ( C 1 ) ∗ ≤ e 1 + L ( R 1 v + 2 ∥ u ( t ) ∥ W 1 , ∞ ) M 0 W \|h_1(t)\|_{(C^1)^*}\leq e^{1+L}(R_1^v+2\|u(t)\|_{W^{1,\infty}})M_0W ∥ h 1 ( t ) ∥ ( C 1 ) ∗ ≤ e 1 + L ( R 1 v + 2∥ u ( t ) ∥ W 1 , ∞ ) M 0 W ;
∥ h 1 ( t ) ∥ L 1 ≤ 2 ( R 1 v + ∥ u ( t ) ∥ L ∞ ) M 0 \|h_1(t)\|_{L^1}\leq2(R_1^v+\|u(t)\|_{L^\infty})M_0 ∥ h 1 ( t ) ∥ L 1 ≤ 2 ( R 1 v + ∥ u ( t ) ∥ L ∞ ) M 0 .
Integrate in t t t , using Hölder’s inequality in t t t for the product A 0 ( t ) 1 − θ B 0 ( t ) θ A_0(t)^{1-\theta}B_0(t)^\theta A 0 ( t ) 1 − θ B 0 ( t ) θ . This gives (3.8) with a constant depending on L = ∥ u ∥ L 1 W 1 , ∞ L=\|u\|_{L^1W^{1,\infty}} L = ∥ u ∥ L 1 W 1 , ∞ . Note that h ∈ L 1 ( 0 , 1 ; L ∞ ) h\in L^1(0,1;L^\infty) h ∈ L 1 ( 0 , 1 ; L ∞ ) .
(e). The proof of Lemma 3.5 uses only (a), (b) and the weak formulation.
(f). Steps 1, 2, 4 and 5 of the proof of Lemma 3.8 use only (a) and (b). Continuity of T ( c ) \mathcal{T}(c) T ( c ) uses that f U ∈ C ( [ 0 , 1 ] ; L 2 ) f_U\in C([0,1];L^2) f U ∈ C ([ 0 , 1 ] ; L 2 ) , which holds since U ∈ C ( [ 0 , 1 ] ; L 2 ) U\in C([0,1];L^2) U ∈ C ([ 0 , 1 ] ; L 2 ) . In Step 3, the fixed-point identity gives f ( U + c ) = f ( U ) ( 0 ) + J 0 − J [ U + c ] ∈ W 1 , ∞ f(U+c)=f(U)(0)+J_0-J[U+c]\in W^{1,\infty} f ( U + c ) = f ( U ) ( 0 ) + J 0 − J [ U + c ] ∈ W 1 , ∞ . Differentiating with (a) gives the mean equation a.e.
Definition 8.3 (Strong VNS–Reynolds solutions). A strong solution of the VNS–Reynolds system on [ 0 , 1 ] [0,1] [ 0 , 1 ] is a pair ( u , R ) (u,R) ( u , R ) such that:
u ∈ C ( [ 0 , 1 ] ; L 2 ) ∩ V s u\in C([0,1];L^2)\cap V_s u ∈ C ([ 0 , 1 ] ; L 2 ) ∩ V s is divergence-free;
R ∈ L 1 ( 0 , 1 ; L 1 ) R\in L^1(0,1;L^1) R ∈ L 1 ( 0 , 1 ; L 1 ) takes values in S 0 d × d S_0^{d\times d} S 0 d × d ;
for every divergence-free φ ∈ C ∞ ( [ 0 , 1 ] × T d ; R d ) \varphi\in C^\infty([0,1]\times\mathbb{T}^d;\mathbb{R}^d) φ ∈ C ∞ ([ 0 , 1 ] × T d ; R d ) vanishing near t = 1 t=1 t = 1 ,
∫ u ( 0 ) ⋅ φ ( 0 ) d x + ∫ 0 1 ∫ ( u ⋅ ( ∂ t φ + Δ φ ) + u ⊗ u : ∇ φ + F [ f [ u ] , u ] ⋅ φ − R : ∇ φ ) d x d t = 0. (51) \int u(0)\cdot\varphi(0)\,\mathrm{d}x+\int_0^1\int\left(u\cdot(\partial_t\varphi+\Delta\varphi)+u\otimes u:\nabla\varphi+F[f[u],u]\cdot\varphi-R:\nabla\varphi\right)\,\mathrm{d}x\,\mathrm{d}t=0. \tag*{(51)} ∫ u ( 0 ) ⋅ φ ( 0 ) d x + ∫ 0 1 ∫ ( u ⋅ ( ∂ t φ + Δ φ ) + u ⊗ u : ∇ φ + F [ f [ u ] , u ] ⋅ φ − R : ∇ φ ) d x d t = 0. (51)
Every smooth solution of Definition 2.6 is a strong solution. Constant test fields give the mean equation. By Lemma 8.1(a), f u ∈ W 1 , 1 f u\in W^{1,1} f u ∈ W 1 , 1 and (2.6) holds.
The convex integration step on a rough background
Lemma 8.4 (Navier–Stokes step, rough background). Let d , p , q , M , r d,p,q,M,r d , p , q , M , r be as in Proposition 2.8. Let u ∈ C ( [ 0 , 1 ] ; L 2 ) u\in C([0,1];L^2) u ∈ C ([ 0 , 1 ] ; L 2 ) be divergence-free, let R ∈ L 1 ( 0 , 1 ; L r ) R\in L^1(0,1;L^r) R ∈ L 1 ( 0 , 1 ; L r ) take values in S 0 d × d \mathcal{S}^{d\times d}_0 S 0 d × d , and let 0 < s < t ∗ ≤ 1 0<s<t_*\leq1 0 < s < t ∗ ≤ 1 with R = 0 R=0 R = 0 on [ 0 , t ∗ ] [0,t_*] [ 0 , t ∗ ] . For every δ > 0 \delta>0 δ > 0 there are smooth w w w , a distribution P P P and R ~ ∈ L 1 ( 0 , 1 ; L r ) \widetilde R\in L^1(0,1;L^r) R ∈ L 1 ( 0 , 1 ; L r ) with values in S 0 d × d \mathcal{S}^{d\times d}_0 S 0 d × d , such that:
(i) ∂ t w − Δ w + div ( u ⊗ w + w ⊗ u + w ⊗ w ) + div R + ∇ P = div R ~ \partial_t w-\Delta w+\operatorname{div}(u\otimes w+w\otimes u+w\otimes w)+\operatorname{div}R+\nabla P=\operatorname{div}\widetilde R ∂ t w − Δ w + div ( u ⊗ w + w ⊗ u + w ⊗ w ) + div R + ∇ P = div R in D ′ ( ( 0 , 1 ) × T d ) \mathcal{D}'((0,1)\times\mathbb{T}^d) D ′ (( 0 , 1 ) × T d ) ;
(ii) div w = 0 \operatorname{div}w=0 div w = 0 , f w ( t ) = 0 f_w(t)=0 f w ( t ) = 0 for all t t t , and w = 0 w=0 w = 0 , R ~ = 0 \widetilde R=0 R = 0 on [ 0 , s ] [0,s] [ 0 , s ] ;
(iii) ∥ w ∥ L t , x 2 ≤ M ∥ R ∥ L t , x 1 1 / 2 \|w\|_{L^2_{t,x}}\leq M\|R\|_{L^1_{t,x}}^{1/2} ∥ w ∥ L t , x 2 ≤ M ∥ R ∥ L t , x 1 1/2 ;
(iv) ∥ w ∥ L t p L ∞ + ∥ w ∥ L t 1 W 1 , q ≤ δ \|w\|_{L^p_tL^\infty}+\|w\|_{L^1_tW^{1,q}}\leq\delta ∥ w ∥ L t p L ∞ + ∥ w ∥ L t 1 W 1 , q ≤ δ ;
(v) ∥ R ~ ∥ L 1 ( 0 , 1 ; L r ) ≤ δ \|\widetilde R\|_{L^1(0,1;L^r)}\leq\delta ∥ R ∥ L 1 ( 0 , 1 ; L r ) ≤ δ .
Proof. Let η ∈ C c ∞ ( ( 0 , 1 ) × B 1 ) \eta\in C^\infty_c((0,1)\times B_1) η ∈ C c ∞ (( 0 , 1 ) × B 1 ) be nonnegative with ∫ η = 1 \int\eta=1 ∫ η = 1 , and η ℓ ( τ , y ) = ℓ − 1 − d η ( τ / ℓ , y / ℓ ) \eta_\ell(\tau,y)=\ell^{-1-d}\eta(\tau/\ell,y/\ell) η ℓ ( τ , y ) = ℓ − 1 − d η ( τ / ℓ , y / ℓ ) . Extend u u u and R R R by zero to t < 0 t<0 t < 0 , and define the causal mollifications
u ℓ ( t , x ) = ∬ u ( t − τ , x − y ) η ℓ ( τ , y ) d τ d y , R ℓ ( t , x ) = ∬ R ( t − τ , x − y ) η ℓ ( τ , y ) d τ d y . u_\ell(t,x)=\iint u(t-\tau,x-y)\eta_\ell(\tau,y)\,\mathrm{d}\tau\,\mathrm{d}y,\qquad R_\ell(t,x)=\iint R(t-\tau,x-y)\eta_\ell(\tau,y)\,\mathrm{d}\tau\,\mathrm{d}y. u ℓ ( t , x ) = ∬ u ( t − τ , x − y ) η ℓ ( τ , y ) d τ d y , R ℓ ( t , x ) = ∬ R ( t − τ , x − y ) η ℓ ( τ , y ) d τ d y .
Since η ℓ \eta_\ell η ℓ is supported in τ ∈ ( 0 , ℓ ) \tau\in(0,\ell) τ ∈ ( 0 , ℓ ) , R ℓ ( t ) R_\ell(t) R ℓ ( t ) depends only on R R R on ( t − ℓ , t ) (t-\ell,t) ( t − ℓ , t ) . Hence R ℓ = 0 R_\ell=0 R ℓ = 0 on [ 0 , t ∗ ] [0,t_*] [ 0 , t ∗ ] and R ℓ R_\ell R ℓ is smooth with values in S 0 d × d \mathcal{S}^{d\times d}_0 S 0 d × d . Moreover u ℓ u_\ell u ℓ is smooth and divergence-free, with sup t ∥ u ℓ ( t ) ∥ L 2 ≤ sup t ∥ u ( t ) ∥ L 2 = : N \sup_t\|u_\ell(t)\|_{L^2}\leq\sup_t\|u(t)\|_{L^2}=:N sup t ∥ u ℓ ( t ) ∥ L 2 ≤ sup t ∥ u ( t ) ∥ L 2 =: N , and ∥ R ℓ ∥ L t , x 1 ≤ ∥ R ∥ L t , x 1 \|R_\ell\|_{L^1_{t,x}}\leq\|R\|_{L^1_{t,x}} ∥ R ℓ ∥ L t , x 1 ≤ ∥ R ∥ L t , x 1 , R ℓ → R R_\ell\to R R ℓ → R in L 1 ( 0 , 1 ; L r ) L^1(0,1;L^r) L 1 ( 0 , 1 ; L r ) . Fix ℓ \ell ℓ with ∥ R − R ℓ ∥ L 1 L r ≤ δ / 3 \|R-R_\ell\|_{L^1L^r}\leq\delta/3 ∥ R − R ℓ ∥ L 1 L r ≤ δ /3 . Apply Proposition 2.8 to ( u ℓ , R ℓ ) (u_\ell,R_\ell) ( u ℓ , R ℓ ) with s , t ∗ s,t_* s , t ∗ and δ ′ ′ : = min { δ , δ / ( 24 ( N + 1 ) ) } \delta^{\prime\prime}:=\min\{\delta,\delta/(24(N+1))\} δ ′′ := min { δ , δ / ( 24 ( N + 1 ))} , obtaining w , P ℓ , R C L w,P_\ell,R^{\mathrm{CL}} w , P ℓ , R CL . Set
R ~ : = R C L + ( R − R ℓ ) + ( u − u ℓ ) ⊗ ∘ w + w ⊗ ∘ ( u − u ℓ ) , P : = P ℓ − d 2 ( u − u ℓ ) ⋅ w . \widetilde R:=R^{\mathrm{CL}}+(R-R_\ell)+(u-u_\ell)\mathbin{\mathop{\otimes}\limits^\circ}w+w\mathbin{\mathop{\otimes}\limits^\circ}(u-u_\ell),\qquad P:=P_\ell-\frac{d}{2}(u-u_\ell)\cdot w. R := R CL + ( R − R ℓ ) + ( u − u ℓ ) ⊗ ∘ w + w ⊗ ∘ ( u − u ℓ ) , P := P ℓ − 2 d ( u − u ℓ ) ⋅ w .
Since a ⊗ b + b ⊗ a = a ⊗ ∘ b + b ⊗ ∘ a + 2 d ( a ⋅ b ) I d a\otimes b+b\otimes a=a\mathbin{\mathop{\otimes}\limits^\circ}b+b\mathbin{\mathop{\otimes}\limits^\circ}a+\frac{2}{d}(a\cdot b)\mathrm{Id} a ⊗ b + b ⊗ a = a ⊗ ∘ b + b ⊗ ∘ a + d 2 ( a ⋅ b ) Id , the trace part of ( u − u ℓ ) ⊗ w + w ⊗ ( u − u ℓ ) (u-u_\ell)\otimes w+w\otimes(u-u_\ell) ( u − u ℓ ) ⊗ w + w ⊗ ( u − u ℓ ) is a gradient, which is moved to the left-hand side; identity (i) then follows from Proposition 2.8(i). The matrix R ~ \widetilde R R is symmetric and traceless.
(ii) and (iv) are inherited, since R − R ℓ = 0 R-R_\ell=0 R − R ℓ = 0 on [ 0 , t ∗ ] [0,t_*] [ 0 , t ∗ ] . (iii) follows from ∥ R ℓ ∥ L t , x 1 ≤ ∥ R ∥ L t , x 1 \|R_\ell\|_{L^1_{t,x}}\leq\|R\|_{L^1_{t,x}} ∥ R ℓ ∥ L t , x 1 ≤ ∥ R ∥ L t , x 1 . For (v), since r ≤ 2 r\leq2 r ≤ 2 and ∣ T d ∣ = 1 |\mathbb{T}^d|=1 ∣ T d ∣ = 1 ,
∥ ( u − u ℓ ) ⊗ ∘ w + w ⊗ ∘ ( u − u ℓ ) ∥ L 1 L r ≤ 4 sup t ∥ u − u ℓ ∥ L 2 ∥ w ∥ L 1 L ∞ ≤ 8 N δ ′ ′ ≤ δ / 3. \|(u-u_\ell)\mathbin{\mathop{\otimes}\limits^\circ}w+w\mathbin{\mathop{\otimes}\limits^\circ}(u-u_\ell)\|_{L^1L^r}\leq4\sup_t\|u-u_\ell\|_{L^2}\|w\|_{L^1L^\infty}\leq8N\delta^{\prime\prime}\leq\delta/3. ∥ ( u − u ℓ ) ⊗ ∘ w + w ⊗ ∘ ( u − u ℓ ) ∥ L 1 L r ≤ 4 t sup ∥ u − u ℓ ∥ L 2 ∥ w ∥ L 1 L ∞ ≤ 8 N δ ′′ ≤ δ /3.
Together with ∥ R C L ∥ L 1 L r ≤ δ ′ ′ ≤ δ / 3 \|R^{\mathrm{CL}}\|_{L^1L^r}\leq\delta^{\prime\prime}\leq\delta/3 ∥ R CL ∥ L 1 L r ≤ δ ′′ ≤ δ /3 , this gives (v). □ \square □
Proposition 8.5 (VNS step, strong solutions). Let d , p , q , M , r d,p,q,M,r d , p , q , M , r be as in Proposition 2.8. Let ( u , R ) (u,R) ( u , R ) be a strong solution of the VNS–Reynolds system with R ∈ L 1 ( 0 , 1 ; L r ) R\in L^1(0,1;L^r) R ∈ L 1 ( 0 , 1 ; L r ) , and let 0 < s < t ∗ ≤ 1 0<s<t_*\leq1 0 < s < t ∗ ≤ 1 with R = 0 R=0 R = 0 on [ 0 , t ∗ ] [0,t_*] [ 0 , t ∗ ] . For every δ ∈ ( 0 , 1 ] \delta\in(0,1] δ ∈ ( 0 , 1 ] there is a strong solution ( u 1 , R 1 ) (u_1,R_1) ( u 1 , R 1 ) with R 1 ∈ L 1 ( 0 , 1 ; L r ) R_1\in L^1(0,1;L^r) R 1 ∈ L 1 ( 0 , 1 ; L r ) satisfying (i)–(iv) of Proposition 2.10.
Proof. Follow Section 3.6 with these replacements:
L : = ∥ u ∥ L 1 ( 0 , 1 ; W 1 , ∞ ) L:=\|u\|_{L^1(0,1;W^{1,\infty})} L := ∥ u ∥ L 1 ( 0 , 1 ; W 1 , ∞ ) ;
Lemma 8.4 instead of Proposition 2.8, with R C L R^{\mathrm{CL}} R CL replaced by R ~ \widetilde R R ;
Lemma 8.1(a), (b), (d), (f) instead of Lemmas 3.1, 3.2, 3.8 and Proposition 3.4.
The mean correction c c c lies in W 1 , 1 ( 0 , 1 ) W^{1,1}(0,1) W 1 , 1 ( 0 , 1 ) , since f ( u + w ) = f u ∈ W 1 , 1 f(u+w)=f u\in W^{1,1} f ( u + w ) = f u ∈ W 1 , 1 . Hence u 1 = u + w + c ∈ C ( [ 0 , 1 ] ; L 2 ) ∩ V s u_1=u+w+c\in C([0,1];L^2)\cap\mathcal{V}_s u 1 = u + w + c ∈ C ([ 0 , 1 ] ; L 2 ) ∩ V s , and (3.12) holds a.e. Step 3 is an identity between distributions. Tested against divergence-free φ \varphi φ , it gives (8.1) for ( u 1 , R 1 ) (u_1,R_1) ( u 1 , R 1 ) . Here we use that ∫ c ⊗ c : ∇ φ d x = 0 \int c\otimes c:\nabla\varphi\,\mathrm{d}x=0 ∫ c ⊗ c : ∇ φ d x = 0 , that the trace part of ( u + w ) ⊗ c + c ⊗ ( u + w ) (u+w)\otimes c+c\otimes(u+w) ( u + w ) ⊗ c + c ⊗ ( u + w ) is a gradient, and that ∫ 0 1 c ⋅ ∂ t φ = − ∫ 0 1 c ′ ⋅ φ \int_0^1 c\cdot\partial_t\varphi=-\int_0^1 c'\cdot\varphi ∫ 0 1 c ⋅ ∂ t φ = − ∫ 0 1 c ′ ⋅ φ , since c ( 0 ) = 0 c(0)=0 c ( 0 ) = 0 . The initial values agree since w = c = 0 w=c=0 w = c = 0 near t = 0 t=0 t = 0 . The estimates of Step 4 are unchanged, with
∥ c ⊗ ∘ ( u + w ) + ( u + w ) ⊗ ∘ c ∥ L 1 L r ≤ 4 sup ∣ c ∣ ( L + 1 ) . \|c\mathbin{\mathop{\otimes}\limits^\circ}(u+w)+(u+w)\mathbin{\mathop{\otimes}\limits^\circ}c\|_{L^1L^r}\leq4\sup|c|(L+1). ∥ c ⊗ ∘ ( u + w ) + ( u + w ) ⊗ ∘ c ∥ L 1 L r ≤ 4 sup ∣ c ∣ ( L + 1 ) .
□ \square □
Concatenation
Lemma 8.6. Let 0 < t 1 < T ≤ 1 0 < t_1 < T \le1 0 < t 1 < T ≤ 1 . Let ( u 1 , f 1 ) (u^1,f^1) ( u 1 , f 1 ) be a weak solution on [ 0 , t 1 ] [0,t_1] [ 0 , t 1 ] with data ( u i n , f 0 ) (u_{\mathrm{in}},f_0) ( u in , f 0 ) , such that u 1 ∈ C w ( [ 0 , t 1 ] ; L 2 ) ∩ L 1 ( 0 , t 1 ; W 1 , 1 ) u^1 \in C_{\mathrm{w}}([0,t_1];L^2) \cap L^1(0,t_1;W^{1,1}) u 1 ∈ C w ([ 0 , t 1 ] ; L 2 ) ∩ L 1 ( 0 , t 1 ; W 1 , 1 ) . Let ( u 2 , f 2 ) (u^2,f^2) ( u 2 , f 2 ) be a weak solution on [ t 1 , T ] [t_1,T] [ t 1 , T ] with data ( u 1 ( t 1 ) , f 1 ( t 1 ) ) (u^1(t_1),f^1(t_1)) ( u 1 ( t 1 ) , f 1 ( t 1 )) , in the sense of Definition 2.1 translated in time. Then the concatenation is a weak solution on [ 0 , T ] [0,T] [ 0 , T ] with data ( u i n , f 0 ) (u_{\mathrm{in}},f_0) ( u in , f 0 ) .
Proof. The integrability requirements of Definition 2.1 hold on each piece. By Lemma 3.6, the function f 1 f^1 f 1 is continuous from [ 0 , t 1 ] [0,t_1] [ 0 , t 1 ] to L 1 L^1 L 1 . Let φ \varphi φ be a test function on [ 0 , T ] [0,T] [ 0 , T ] . Let ζ n ∈ C ∞ ( R ; [ 0 , 1 ] ) \zeta_n \in C^\infty(\mathbb{R};[0,1]) ζ n ∈ C ∞ ( R ; [ 0 , 1 ]) be nonincreasing, equal to 1 on ( − ∞ , t 1 − 2 n ] (-\infty,t_1-\frac{2}{n}] ( − ∞ , t 1 − n 2 ] and to 0 on [ t 1 − 1 n , ∞ ) [t_1-\frac{1}{n},\infty) [ t 1 − n 1 , ∞ ) . Test the equations of the first piece with ζ n φ \zeta_n\varphi ζ n φ , which is admissible and divergence-free if φ \varphi φ is. The terms with ζ n ′ \zeta'_n ζ n ′ are
∬ ζ n ′ f 1 φ d z d t → − ∫ f 1 ( t 1 ) φ ( t 1 ) d z , ∬ ζ n ′ u 1 ⋅ φ d x d t → − ∫ u 1 ( t 1 ) ⋅ φ ( t 1 ) d x . \iint\zeta'_n f^1\varphi\, dz\,dt \to-\int f^1(t_1)\varphi(t_1)\,dz,\qquad\iint\zeta'_n u^1\cdot\varphi\, dx\,dt \to-\int u^1(t_1)\cdot\varphi(t_1)\,dx. ∬ ζ n ′ f 1 φ d z d t → − ∫ f 1 ( t 1 ) φ ( t 1 ) d z , ∬ ζ n ′ u 1 ⋅ φ d x d t → − ∫ u 1 ( t 1 ) ⋅ φ ( t 1 ) d x .
by continuity of f 1 f^1 f 1 and weak continuity of u 1 u^1 u 1 at t 1 t_1 t 1 ; for the latter, write u 1 ( t ) ⋅ φ ( t ) − u 1 ( t 1 ) ⋅ φ ( t 1 ) = u 1 ( t ) ⋅ ( φ ( t ) − φ ( t 1 ) ) + ( u 1 ( t ) − u 1 ( t 1 ) ) ⋅ φ ( t 1 ) u^1(t)\cdot\varphi(t)-u^1(t_1)\cdot\varphi(t_1)=u^1(t)\cdot(\varphi(t)-\varphi(t_1))+(u^1(t)-u^1(t_1))\cdot\varphi(t_1) u 1 ( t ) ⋅ φ ( t ) − u 1 ( t 1 ) ⋅ φ ( t 1 ) = u 1 ( t ) ⋅ ( φ ( t ) − φ ( t 1 )) + ( u 1 ( t ) − u 1 ( t 1 )) ⋅ φ ( t 1 ) and use sup t ∥ u 1 ( t ) ∥ L 2 < ∞ \sup_t\|u^1(t)\|_{L^2}<\infty sup t ∥ u 1 ( t ) ∥ L 2 < ∞ . All other terms converge by dominated convergence. The resulting identities on [ 0 , t 1 ] [0,t_1] [ 0 , t 1 ] contain the boundary terms at t 1 t_1 t 1 with a minus sign. The identities on [ t 1 , T ] [t_1,T] [ t 1 , T ] contain the same terms as initial data. Adding the two gives Definition 2.1(iii) and (iv) on [ 0 , T ] [0,T] [ 0 , T ] .
Proof of Theorem 7.2
Step 0: reduction to t 1 = 0 t_1=0 t 1 = 0 . Let ( u ˉ , f ˉ , t 1 , a , T 0 , q 0 ) (\bar{u},\bar{f},t_1,a,T_0,q_0) ( u ˉ , f ˉ , t 1 , a , T 0 , q 0 ) be the strong continuation. By (S2), the definition of weak solutions and Lemma 8.1(a), on [ t 1 , T 0 ] [t_1,T_0] [ t 1 , T 0 ] the function f ˉ \bar{f} f ˉ is the push-forward of f ˉ ( t 1 ) \bar{f}(t_1) f ˉ ( t 1 ) along the Carathéodory flow of u ˉ \bar{u} u ˉ . Moreover f ˉ ( t 1 ) ∈ L ∞ \bar{f}(t_1)\in L^\infty f ˉ ( t 1 ) ∈ L ∞ is nonnegative and vanishes for ∣ v ∣ |v| ∣ v ∣ large. Translate time by t 1 t_1 t 1 : the data become ( u ˉ ( t 1 ) , f ˉ ( t 1 ) ) (\bar{u}(t_1),\bar{f}(t_1)) ( u ˉ ( t 1 ) , f ˉ ( t 1 )) , and the translated solution lives on [ 0 , T ∗ ] [0,T_*] [ 0 , T ∗ ] with T ∗ : = T 0 − t 1 ≤ 1 − t 1 T_*:=T_0-t_1\le1-t_1 T ∗ := T 0 − t 1 ≤ 1 − t 1 .
All results of Sections 2–3 and of Section 8 hold with the kinetic datum f ˉ ( t 1 ) \bar{f}(t_1) f ˉ ( t 1 ) , by Lemma 8.1. We construct solutions on [ 0 , 1 ] [0,1] [ 0 , 1 ] , restrict them to [ 0 , 1 − t 1 ] [0,1-t_1] [ 0 , 1 − t 1 ] , translate back, and concatenate them with ( u ˉ , f ˉ ) ∣ [ 0 , t 1 ] (\bar{u},\bar{f})|_{[0,t_1]} ( u ˉ , f ˉ ) ∣ [ 0 , t 1 ] by Lemma 8.6. When t 1 > 0 t_1>0 t 1 > 0 , the hypotheses of Lemma 8.6 are (S1).
We may therefore assume t 1 = 0 t_1=0 t 1 = 0 , and write ( u ˉ , f ˉ ) (\bar{u},\bar{f}) ( u ˉ , f ˉ ) on [ 0 , T ∗ ] [0,T_*] [ 0 , T ∗ ] with
u ˉ ∈ C ( [ 0 , T ∗ ] ; L 2 ) ∩ L 2 ( 0 , T ∗ ; L ∞ ) ∩ L 1 ( 0 , T ∗ ; W 1 , ∞ ) ∩ M R q 0 ( [ a , T ∗ ] ) , f ˉ = f [ u ˉ ] . \bar{u}\in C([0,T_*];L^2)\cap L^2(0,T_*;L^\infty)\cap L^1(0,T_*;W^{1,\infty})\cap MR_{q_0}([a,T_*]),\qquad\bar{f}=f[\bar{u}]. u ˉ ∈ C ([ 0 , T ∗ ] ; L 2 ) ∩ L 2 ( 0 , T ∗ ; L ∞ ) ∩ L 1 ( 0 , T ∗ ; W 1 , ∞ ) ∩ M R q 0 ([ a , T ∗ ]) , f ˉ = f [ u ˉ ] .
Step 1: initial approximate solutions. Fix a < a ′ < t 2 ′ < t 2 < t 3 < T ∗ a<a'<t'_2<t_2<t_3<T_* a < a ′ < t 2 ′ < t 2 < t 3 < T ∗ . Choose χ ∈ C ∞ ( R ; [ 0 , 1 ] ) \chi\in C^\infty(\mathbb{R};[0,1]) χ ∈ C ∞ ( R ; [ 0 , 1 ]) with χ = 1 \chi=1 χ = 1 on ( − ∞ , t 2 ] (-\infty,t_2] ( − ∞ , t 2 ] and χ = 0 \chi=0 χ = 0 on [ t 3 , ∞ ) [t_3,\infty) [ t 3 , ∞ ) . Choose Ψ \Psi Ψ smooth, divergence-free, with ∬ Ψ = 0 \iint\Psi=0 ∬ Ψ = 0 , supported in t ∈ ( t 3 , 1 − t 1 ) t\in(t_3,1-t_1) t ∈ ( t 3 , 1 − t 1 ) , and with ∥ Ψ ∥ L p ( 0 , 1 − t 1 ; L 2 ) ≥ 1 \|\Psi\|_{L^p(0,1-t_1;L^2)}\ge1 ∥Ψ ∥ L p ( 0 , 1 − t 1 ; L 2 ) ≥ 1 . For k ∈ N k\in\mathbb{N} k ∈ N put u ~ k : = χ u ˉ + k Ψ ∈ C ( [ 0 , 1 ] ; L 2 ) ∩ V s \widetilde{u}_k:=\chi\bar{u}+k\Psi\in C([0,1];L^2)\cap V_s u k := χ u ˉ + k Ψ ∈ C ([ 0 , 1 ] ; L 2 ) ∩ V s .
Since u ˉ \bar{u} u ˉ is a weak solution, it satisfies (2.6) on [ 0 , T ∗ ] [0,T_*] [ 0 , T ∗ ] , and f ˉ u ˉ ∈ W 1 , 1 \bar{f}\,\bar{u}\in W^{1,1} f ˉ u ˉ ∈ W 1 , 1 . Lemma 8.1(f) gives a correction m k ∈ W 1 , 1 m_k\in W^{1,1} m k ∈ W 1 , 1 with m k = 0 m_k=0 m k = 0 on [ 0 , t 2 ] [0,t_2] [ 0 , t 2 ] . Set u 0 , k : = u ~ k + m k u_{0,k}:=\widetilde{u}_k+m_k u 0 , k := u k + m k . We claim that there is R 0 , k ∈ L 1 ( 0 , 1 ; L q 0 ) R_{0,k}\in L^1(0,1;L^{q_0}) R 0 , k ∈ L 1 ( 0 , 1 ; L q 0 ) , vanishing on [ 0 , t 2 ] [0,t_2] [ 0 , t 2 ] , such that ( u 0 , k , R 0 , k ) (u_{0,k},R_{0,k}) ( u 0 , k , R 0 , k ) is a strong solution of the VNS–Reynolds system.
On ( a , 1 ) (a,1) ( a , 1 ) define the function
G : = ∂ t u 0 , k − Δ u 0 , k + u 0 , k ⋅ ∇ u 0 , k − F [ f [ u 0 , k ] , u 0 , k ] ∈ L 1 ( a , 1 ; L q 0 ) . G:=\partial_tu_{0,k}-\Delta u_{0,k}+u_{0,k}\cdot\nabla u_{0,k}-F[f[u_{0,k}],u_{0,k}]\in L^1(a,1;L^{q_0}). G := ∂ t u 0 , k − Δ u 0 , k + u 0 , k ⋅ ∇ u 0 , k − F [ f [ u 0 , k ] , u 0 , k ] ∈ L 1 ( a , 1 ; L q 0 ) .
It is integrable: on ( a , T ∗ ) (a,T_*) ( a , T ∗ ) we have u ˉ ∈ M R q 0 ⊂ C ( X q 0 ) ⊂ L t , x ∞ \bar{u}\in MR_{q_0}\subset C(X_{q_0})\subset L^\infty_{t,x} u ˉ ∈ M R q 0 ⊂ C ( X q 0 ) ⊂ L t , x ∞ , so ∂ t u ˉ , Δ u ˉ , u ˉ ⋅ ∇ u ˉ ∈ L 2 L q 0 \partial_t\bar{u},\Delta\bar{u},\bar{u}\cdot\nabla\bar{u}\in L^2L^{q_0} ∂ t u ˉ , Δ u ˉ , u ˉ ⋅ ∇ u ˉ ∈ L 2 L q 0 ; moreover m k ′ ∈ L 1 m'_k\in L^1 m k ′ ∈ L 1 , Ψ \Psi Ψ is smooth, and F ∈ L 1 L ∞ F\in L^1L^\infty F ∈ L 1 L ∞ by Lemma 8.1(a). By causality, u 0 , k = u ˉ u_{0,k}=\bar{u} u 0 , k = u ˉ and f [ u 0 , k ] = f ˉ f[u_{0,k}]=\bar{f} f [ u 0 , k ] = f ˉ on [ 0 , t 2 ] [0,t_2] [ 0 , t 2 ] . Since u ˉ \bar{u} u ˉ is a weak solution, ∬ G ⋅ ψ = 0 \iint G\cdot\psi=0 ∬ G ⋅ ψ = 0 for every divergence-free ψ \psi ψ supported in ( a , t 2 ) (a,t_2) ( a , t 2 ) , so P G = 0 \mathbb{P}G=0 P G = 0 a.e. on ( a , t 2 ) (a,t_2) ( a , t 2 ) . Moreover f G = 0 fG=0 f G = 0 a.e. on $(a,1) by Lemma 8.1(f).
Let ζ ∈ C ∞ ( R ; [ 0 , 1 ] ) \zeta\in C^\infty(\mathbb{R};[0,1]) ζ ∈ C ∞ ( R ; [ 0 , 1 ]) with ζ = 1 \zeta=1 ζ = 1 on ( − ∞ , a ′ ] (-\infty,a'] ( − ∞ , a ′ ] and ζ = 0 \zeta=0 ζ = 0 on [ t 2 , ∞ ) [t_2,\infty) [ t 2 , ∞ ) , and set R 0 , k : = ( 1 − ζ ) P G R_{0,k}:=(1-\zeta)\mathbb{P}G R 0 , k := ( 1 − ζ ) P G . Then R 0 , k = 0 R_{0,k}=0 R 0 , k = 0 on [ 0 , t 2 ] [0,t_2] [ 0 , t 2 ] . Let φ \varphi φ be a divergence-free test field vanishing near t = 1 t=1 t = 1 , and split φ = ζ φ + ( 1 − ζ ) φ \varphi=\zeta\varphi+(1-\zeta)\varphi φ = ζ φ + ( 1 − ζ ) φ . Both parts are divergence-free.
For ζ φ \zeta\varphi ζ φ , which vanishes after t 2 ′ t'_2 t 2 ′ , the terms of (8.1) without the stress vanish, since u 0 , k = u ˉ u_{0,k}=\bar{u} u 0 , k = u ˉ and f [ u 0 , k ] = f ˉ f[u_{0,k}]=\bar{f} f [ u 0 , k ] = f ˉ on [ 0 , t 2 ] [0,t_2] [ 0 , t 2 ] , and u ˉ \bar{u} u ˉ is a weak solution on [ 0 , T ∗ ] [0,T_*] [ 0 , T ∗ ] .
For ψ : = ( 1 − ζ ) φ \psi:=(1-\zeta)\varphi ψ := ( 1 − ζ ) φ , which vanishes near [ 0 , a ′ ] [0,a'] [ 0 , a ′ ] , integration by parts turns these terms into − ∬ G ⋅ ψ -\iint G\cdot\psi − ∬ G ⋅ ψ . Since ψ \psi ψ is divergence-free and f G = 0 fG=0 f G = 0 , this equals − ∬ P G ⋅ ψ = ∬ P G : ∇ ψ = ∬ R 0 , k : ∇ φ -\iint\mathbb{P}G\cdot\psi=\iint\mathbb{P}G:\nabla\psi=\iint R_{0,k}:\nabla\varphi − ∬ P G ⋅ ψ = ∬ P G : ∇ ψ = ∬ R 0 , k : ∇ φ , because ζ \zeta ζ depends only on t t t .
These terms are linear in the test field, so for φ \varphi φ they add up to ∬ R 0 , k : ∇ φ \iint R_{0,k}:\nabla\varphi ∬ R 0 , k : ∇ φ , and (8.1) holds with R = R 0 , k R=R_{0,k} R = R 0 , k .
Step 2: iteration. Run the proof of Theorem 2.3 with:
s n : = t 2 ( 1 2 + 2 − n − 1 ) s_n := t_2\left(\frac{1}{2}+2^{-n-1}\right) s n := t 2 ( 2 1 + 2 − n − 1 ) ;
Proposition 8.5 instead of Proposition 2.10;
Lemma 8.1(e) instead of Lemma 3.5.
Every iterate is a strong solution, so Λ n < ∞ \Lambda_n<\infty Λ n < ∞ . Step 4 of that proof uses (8.1) instead of (2.5). The result is a weak solution ( u k , f k ) (u_k,f_k) ( u k , f k ) on [ 0 , 1 ] [0,1] [ 0 , 1 ] with data ( u ˉ ( 0 ) , f ˉ ( 0 ) ) (\bar{u}(0),\bar{f}(0)) ( u ˉ ( 0 ) , f ˉ ( 0 )) . It satisfies properties (1), (3), (4), (5) of Theorem 2.3, with u ~ = u ~ k \tilde{u}=\tilde{u}_k u ~ = u ~ k and ε = 1 4 \varepsilon=\frac{1}{4} ε = 4 1 , and coincides with ( u ˉ , f ˉ ) (\bar{u},\bar{f}) ( u ˉ , f ˉ ) on [ 0 , t 2 / 2 ] [0,t_2/2] [ 0 , t 2 /2 ] . Property (1) uses u 0 , k ∈ L 2 ( 0 , 1 ; L ∞ ) ∩ L 1 ( 0 , 1 ; W 1 , ∞ ) u_{0,k}\in L^2(0,1;L^\infty)\cap L^1(0,1;W^{1,\infty}) u 0 , k ∈ L 2 ( 0 , 1 ; L ∞ ) ∩ L 1 ( 0 , 1 ; W 1 , ∞ ) .
Step 3: conclusion. Restrict to [ 0 , 1 − t 1 ] [0,1-t_1] [ 0 , 1 − t 1 ] , translate back, and concatenate with ( u ˉ , f ˉ ) ∣ [ 0 , t 1 ] (\bar{u},\bar{f})|_{[0,t_1]} ( u ˉ , f ˉ ) ∣ [ 0 , t 1 ] by Lemma 8.6. This gives (2) with τ = t 2 / 2 \tau=t_2/2 τ = t 2 /2 . Property (1) on [ 0 , t 1 ] [0,t_1] [ 0 , t 1 ] follows from (S1). For (3), u ∈ L 1 ( 0 , 1 ; W 1 , 1 ) ∩ L 1 ( 0 , 1 ; L ∞ ) u\in L^1(0,1;W^{1,1})\cap L^1(0,1;L^\infty) u ∈ L 1 ( 0 , 1 ; W 1 , 1 ) ∩ L 1 ( 0 , 1 ; L ∞ ) , so Lemma 3.6 applies on [ 0 , 1 ] [0,1] [ 0 , 1 ] . Momentum is conserved on each piece.
Distinctness and the failure of the energy inequality follow as in the proof of Corollary 2.4. The fluctuations of u k − u l u_k-u_l u k − u l are within 1 2 \frac{1}{2} 2 1 of ( k − l ) Ψ (k-l)\Psi ( k − l ) Ψ in L p ( 0 , 1 − t 1 ; L ∞ ) L^p(0,1-t_1;L^\infty) L p ( 0 , 1 − t 1 ; L ∞ ) , after translation. A solution of Leray–Hopf type satisfies ∥ u k ∥ L t p L 2 ≤ ( 2 E ( 0 ) ) 1 / 2 \|u_k\|_{L^p_tL^2}\le(2\mathcal{E}(0))^{1/2} ∥ u k ∥ L t p L 2 ≤ ( 2 E ( 0 ) ) 1/2 .
Local strong solutions: proof of Theorem 7.3
Proposition 9.1 (Data in B q 0 , 2 1 B^1_{q_0,2} B q 0 , 2 1 ). Let d ≥ 2 d\ge2 d ≥ 2 and q 0 ∈ ( d , ∞ ) ∩ [ 2 , ∞ ) q_0\in(d,\infty)\cap[2,\infty) q 0 ∈ ( d , ∞ ) ∩ [ 2 , ∞ ) . Let u ∗ ∈ X q 0 u_*\in X_{q_0} u ∗ ∈ X q 0 be divergence-free, and let f ∗ ∈ L ∞ f_*\in L^\infty f ∗ ∈ L ∞ be nonnegative and vanish for ∣ v ∣ ≥ R ∗ v |v|\ge R^v_* ∣ v ∣ ≥ R ∗ v . There is T > 0 T>0 T > 0 , depending only on d d d , q 0 q_0 q 0 and upper bounds for ∥ u ∗ ∥ X q 0 \|u_*\|_{X_{q_0}} ∥ u ∗ ∥ X q 0 , R ∗ v R^v_* R ∗ v and ∥ f ∗ ∥ L ∞ \|f_*\|_{L^\infty} ∥ f ∗ ∥ L ∞ , and there is u ∈ M R q 0 ( [ 0 , T ] ) u\in MR_{q_0}([0,T]) u ∈ M R q 0 ([ 0 , T ]) with div u = 0 \operatorname{div}u=0 div u = 0 , such that
∂ t u − Δ u + Q ( u ⋅ ∇ u ) = Q F [ f [ u ] , u ] a.e. , u ( 0 ) = u ∗ . (52) \partial_tu-\Delta u+\mathbb{Q}(u\cdot\nabla u)=\mathbb{Q}F[f[u],u]\quad\text{a.e.},\qquad u(0)=u_*. \tag*{(52)} ∂ t u − Δ u + Q ( u ⋅ ∇ u ) = Q F [ f [ u ] , u ] a.e. , u ( 0 ) = u ∗ . (52)
Here f [ u ] f[u] f [ u ] is the push-forward of f ∗ f_* f ∗ along b u b_u b u . The pair ( u , f [ u ] ) (u,f[u]) ( u , f [ u ]) is a weak solution on [ 0 , T ] [0,T] [ 0 , T ] and a strong continuation of ( u ∗ , f ∗ ) (u_*,f_*) ( u ∗ , f ∗ ) with t 1 = a = 0 t_1=a=0 t 1 = a = 0 .
Proof. Write q = q 0 q=q_0 q = q 0 , and let C q , C T , C E C_q,C_T,C_E C q , C T , C E be the constants of Lemmas 5.4 and 6.4. For T ≤ 1 T\le1 T ≤ 1 and A ≥ 1 A\ge1 A ≥ 1 let
E T : = { u ∈ M R q ( [ 0 , T ] ) : div u = 0 , u ( 0 ) = u ∗ , ∥ u ∥ M R q ≤ A } , E_T:=\{u\in MR_q([0,T]):\operatorname{div}u=0,\ u(0)=u_*,\ \|u\|_{MR_q}\le A\}, E T := { u ∈ M R q ([ 0 , T ]) : div u = 0 , u ( 0 ) = u ∗ , ∥ u ∥ M R q ≤ A } ,
with the metric ∥ u − u ′ ∥ Y 1 ( 0 , T ) \|u-u'\|_{Y_1(0,T)} ∥ u − u ′ ∥ Y 1 ( 0 , T ) . The set E T E_T E T is complete, as in the proof of Proposition 6.8(a).
Bounds on E T E_T E T . Let u ∈ E T u\in E_T u ∈ E T . By Lemma 5.4(b) and Lemma 6.4(b),
sup t ∥ u ( t ) ∥ X q ≤ A 1 : = C q ( ∥ u ∗ ∥ X q + A ) , ∥ u ∥ L t , x ∞ ≤ C E A 1 , ∫ 0 T ∥ ∇ u ∥ L ∞ ≤ C E T 1 / 2 A . \sup_t\|u(t)\|_{X_q}\le A_1:=C_q(\|u_*\|_{X_q}+A),\qquad\|u\|_{L^\infty_{t,x}}\le C_EA_1,\qquad\int_0^T\|\nabla u\|_{L^\infty}\le C_ET^{1/2}A. t sup ∥ u ( t ) ∥ X q ≤ A 1 := C q ( ∥ u ∗ ∥ X q + A ) , ∥ u ∥ L t , x ∞ ≤ C E A 1 , ∫ 0 T ∥∇ u ∥ L ∞ ≤ C E T 1/2 A .
Take T ≤ T 1 T\le T_1 T ≤ T 1 , with T 1 T_1 T 1 such that T 1 C E A 1 ≤ 1 T_1C_EA_1\le1 T 1 C E A 1 ≤ 1 and C E T 1 1 / 2 A ≤ 1 C_ET_1^{1/2}A\le1 C E T 1 1/2 A ≤ 1 . Then, by Lemma 8.1(a), f [ u ] f[u] f [ u ] vanishes for ∣ v ∣ ≥ R ∗ v + 1 |v|\ge R^v_*+1 ∣ v ∣ ≥ R ∗ v + 1 , and ρ [ u ] ≤ ϱ : = e d ∥ f ∗ ∥ ∞ ∣ B R ∗ v + 1 ∣ \rho[u]\le\varrho:=e^d\|f_*\|_\infty|B_{R^v_*+1}| ρ [ u ] ≤ ϱ := e d ∥ f ∗ ∥ ∞ ∣ B R ∗ v + 1 ∣ . So ∣ F [ f [ u ] , u ] ∣ ≤ ϱ ( R ∗ v + 1 + C E A 1 ) |F[f[u],u]|\le\varrho(R^v_*+1+C_EA_1) ∣ F [ f [ u ] , u ] ∣ ≤ ϱ ( R ∗ v + 1 + C E A 1 ) .
Invariance. Let Θ u \Theta u Θ u be the solution of ∂ t ϑ − Δ ϑ = Q ( F [ f [ u ] , u ] − u ⋅ ∇ u ) \partial_t\vartheta-\Delta\vartheta=\mathbb{Q}\big(F[f[u],u]-u\cdot\nabla u\big) ∂ t ϑ − Δ ϑ = Q ( F [ f [ u ] , u ] − u ⋅ ∇ u ) with ϑ ( 0 ) = u ∗ \vartheta(0)=u_* ϑ ( 0 ) = u ∗ . It is divergence-free by Lemma 5.4(d). Since ∥ u ⋅ ∇ u ∥ L q ≤ ∥ u ∥ ∞ ∥ ∇ u ∥ L q ≤ C E 2 A 1 2 \|u\cdot\nabla u\|_{L^q}\le\|u\|_\infty\|\nabla u\|_{L^q}\le C_E^2A_1^2 ∥ u ⋅ ∇ u ∥ L q ≤ ∥ u ∥ ∞ ∥∇ u ∥ L q ≤ C E 2 A 1 2 ,
∥ Θ u ∥ M R q ≤ C q ∥ u ∗ ∥ X q + C q T 1 / 2 ( C E 2 A 1 2 + ϱ ( R ∗ v + 1 + C E A 1 ) ) . \|\Theta u\|_{MR_q}\le C_q\|u_*\|_{X_q}+C_qT^{1/2}\left(C_E^2A_1^2+\varrho(R^v_*+1+C_EA_1)\right). ∥Θ u ∥ M R q ≤ C q ∥ u ∗ ∥ X q + C q T 1/2 ( C E 2 A 1 2 + ϱ ( R ∗ v + 1 + C E A 1 ) ) .
Choose A : = 2 C q ∥ u ∗ ∥ X q + 1 A:=2C_q\|u_*\|_{X_q}+1 A := 2 C q ∥ u ∗ ∥ X q + 1 , and then T T T small; this gives Θ ( E T ) ⊂ E T \Theta(E_T)\subset E_T Θ ( E T ) ⊂ E T .
Contraction. Let u , u ′ ∈ E T u,u'\in E_T u , u ′ ∈ E T and δ u = u − u ′ \delta u=u-u' δ u = u − u ′ , so δ u ( 0 ) = 0 \delta u(0)=0 δ u ( 0 ) = 0 . Write
F [ f [ u ] , u ] − F [ f [ u ′ ] , u ′ ] = ∫ ( v − u ′ ) ( f [ u ] − f [ u ′ ] ) d v − ρ [ u ] δ u . F[f[u],u]-F[f[u'],u']=\int(v-u')(f[u]-f[u'])\,\mathrm{d}v-\rho[u]\delta u. F [ f [ u ] , u ] − F [ f [ u ′ ] , u ′ ] = ∫ ( v − u ′ ) ( f [ u ] − f [ u ′ ]) d v − ρ [ u ] δ u .
Lemma 6.5(b), with t 0 = 0 t_0=0 t 0 = 0 , base b = u ′ b=u' b = u ′ , U = u U=u U = u and W = u ′ W=u' W = u ′ , gives the following bound. Here K ′ K' K ′ depends only on R ∗ v R^v_* R ∗ v , ∥ f ∗ ∥ ∞ \|f_*\|_\infty ∥ f ∗ ∥ ∞ and A 1 A_1 A 1 , since Λ u ′ ≤ 1 \Lambda_{u'}\le1 Λ u ′ ≤ 1 , R v 1 ≤ R ∗ v + 2 R_{v_1}\le R^v_*+2 R v 1 ≤ R ∗ v + 2 and sup t ∥ u ′ ( t ) ∥ W 1 , q ≤ sup t ∥ u ′ ( t ) ∥ X q ≤ A 1 \sup_t\|u'(t)\|_{W^{1,q}}\le\sup_t\|u'(t)\|_{X_q}\le A_1 sup t ∥ u ′ ( t ) ∥ W 1 , q ≤ sup t ∥ u ′ ( t ) ∥ X q ≤ A 1 :
∥ ∥ ∫ ( v − u ′ ) ( f [ u ] − f [ u ′ ] ) ( t ) d v ∥ H − 1 , q ∥ ≤ K ′ ∥ δ u ∥ L 1 ( 0 , t ; L q ) ≤ K ′ ∥ δ u ∥ L 1 ( 0 , t ; L ∞ ) . \left\|\left\|\int(v-u')(f[u]-f[u'])(t)\,\mathrm{d}v\right\|_{H^{-1,q}}\right\|\le K'\|\delta u\|_{L^1(0,t;L^q)}\le K'\|\delta u\|_{L^1(0,t;L^\infty)}. ∫ ( v − u ′ ) ( f [ u ] − f [ u ′ ]) ( t ) d v H − 1 , q ≤ K ′ ∥ δ u ∥ L 1 ( 0 , t ; L q ) ≤ K ′ ∥ δ u ∥ L 1 ( 0 , t ; L ∞ ) .
By Lemma 6.4,
∥ δ u ∥ L 1 ( 0 , T ; L ∞ ) ≤ C E T 1 / 2 ∥ δ u ∥ Y 1 , sup t ∥ δ u ∥ L q ≤ C T ∥ δ u ∥ Y 1 . \|\delta u\|_{L^1(0,T;L^\infty)}\le C_ET^{1/2}\|\delta u\|_{Y_1},\qquad\sup_t\|\delta u\|_{L^q}\le C_T\|\delta u\|_{Y_1}. ∥ δ u ∥ L 1 ( 0 , T ; L ∞ ) ≤ C E T 1/2 ∥ δ u ∥ Y 1 , t sup ∥ δ u ∥ L q ≤ C T ∥ δ u ∥ Y 1 .
Using ∥ div ( u ⊗ u − u ′ ⊗ u ′ ) ∥ H − 1 , q ≤ C ∥ u ⊗ u − u ′ ⊗ u ′ ∥ L q \lVert\operatorname{div}(u \otimes u-u' \otimes u')\rVert_{H^{-1,q}} \le C\lVert u \otimes u-u' \otimes u'\rVert_{L^q} ∥ div ( u ⊗ u − u ′ ⊗ u ′ ) ∥ H − 1 , q ≤ C ∥ u ⊗ u − u ′ ⊗ u ′ ∥ L q and Lemma 6.4(a),
∥ Θ u − Θ u ′ ∥ Y 1 ≤ C q ( T 1 / 2 C ( C E A 1 + ϱ ) C T + K ′ C E T ) ∥ δ u ∥ Y 1 ≤ 1 2 ∥ δ u ∥ Y 1 \lVert\Theta u-\Theta u'\rVert_{Y_1} \le C_q\left(T^{1/2}C(C_EA_1+\varrho)C_T+K'C_ET\right)\lVert\delta u\rVert_{Y_1}\le\frac{1}{2}\lVert\delta u\rVert_{Y_1} ∥ Θ u − Θ u ′ ∥ Y 1 ≤ C q ( T 1/2 C ( C E A 1 + ϱ ) C T + K ′ C E T ) ∥ δ u ∥ Y 1 ≤ 2 1 ∥ δ u ∥ Y 1
for T T T small. The resulting T T T depends only on the quantities listed.
Conclusion. The fixed point u u u solves (52) . Since Q Q Q is self-adjoint and Q φ = φ Q\varphi=\varphi Qφ = φ for divergence-free φ \varphi φ , testing (9.1) gives Definition 2.1(iv). Lemma 8.1(a) gives (iii). Properties (S1)–(S3) hold:
M R q ⊂ C ( X q ) ∩ L 2 ( 0 , T ; W 1 , ∞ ) , X q ⊂ W 1 , q ⊂ L ∞ . MR_q\subset C(X_q)\cap L^2(0,T;W^{1,\infty}),\qquad X_q\subset W^{1,q}\subset L^\infty. M R q ⊂ C ( X q ) ∩ L 2 ( 0 , T ; W 1 , ∞ ) , X q ⊂ W 1 , q ⊂ L ∞ .
□
Proposition 9.2 (Data in H 1 H^1 H 1 , d ≤ 3 d\le3 d ≤ 3 ). Let d ∈ { 2 , 3 } d\in\{2,3\} d ∈ { 2 , 3 } . Let u ∗ ∈ H 1 u_*\in H^1 u ∗ ∈ H 1 be divergence-free, and let f ∗ ∈ L ∞ f_*\in L^\infty f ∗ ∈ L ∞ be nonnegative and vanish for ∣ v ∣ ≥ R ∗ v \lvert v\rvert\ge R_*^v ∣ v ∣ ≥ R ∗ v . There are T > 0 T>0 T > 0 and a weak solution ( u , f [ u ] ) (u,f[u]) ( u , f [ u ]) on [ 0 , T ] [0,T] [ 0 , T ] with data ( u ∗ , f ∗ ) (u_*,f_*) ( u ∗ , f ∗ ) , such that
u ∈ C ( [ 0 , T ] ; L 2 ) ∩ L ∞ ( 0 , T ; H 1 ) ∩ L 2 ( 0 , T ; H 2 ) , ∂ t u ∈ L 2 ( 0 , T ; L 2 ) , ∇ u ∈ L 1 ( 0 , T ; L ∞ ) . u\in C([0,T];L^2)\cap L^\infty(0,T;H^1)\cap L^2(0,T;H^2),\qquad\partial_tu\in L^2(0,T;L^2),\qquad\nabla u\in L^1(0,T;L^\infty). u ∈ C ([ 0 , T ] ; L 2 ) ∩ L ∞ ( 0 , T ; H 1 ) ∩ L 2 ( 0 , T ; H 2 ) , ∂ t u ∈ L 2 ( 0 , T ; L 2 ) , ∇ u ∈ L 1 ( 0 , T ; L ∞ ) .
Moreover u ( t ) ∈ X 4 u(t)\in X_4 u ( t ) ∈ X 4 for a.e. t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) , and ( u , f [ u ] ) (u,f[u]) ( u , f [ u ]) satisfies the energy inequality of Definition 2.2 on [ 0 , T ] [0,T] [ 0 , T ] . One can take
T = c ∗ ( 1 + ∥ u ∗ ∥ H 1 2 ) − 2 , (53) T=c_*\left(1+\lVert u_*\rVert_{H^1}^2\right)^{-2}, \tag*{(53)} T = c ∗ ( 1 + ∥ u ∗ ∥ H 1 2 ) − 2 , (53)
where c ∗ > 0 c_*>0 c ∗ > 0 depends only on d d d and on upper bounds for ∥ u ∗ ∥ L 2 \lVert u_*\rVert_{L^2} ∥ u ∗ ∥ L 2 , R ∗ v R_*^v R ∗ v , ∥ f ∗ ∥ L ∞ \lVert f_*\rVert_{L^\infty} ∥ f ∗ ∥ L ∞ and ∥ f ∗ ∥ L 1 \lVert f_*\rVert_{L^1} ∥ f ∗ ∥ L 1 ; in particular c ∗ c_* c ∗ can be taken nonincreasing in each of these quantities.
Proof. Constants C C C depend only on d d d , ∥ u ∗ ∥ H 1 \lVert u_*\rVert_{H^1} ∥ u ∗ ∥ H 1 , R ∗ v R_*^v R ∗ v , ∥ f ∗ ∥ L ∞ \lVert f_*\rVert_{L^\infty} ∥ f ∗ ∥ L ∞ and ∥ f ∗ ∥ L 1 \lVert f_*\rVert_{L^1} ∥ f ∗ ∥ L 1 , unless indicated otherwise.
Step 1: approximation. Let u ∗ ε u_*^\varepsilon u ∗ ε be the Fourier truncations of u ∗ u_* u ∗ . They are smooth, divergence-free, satisfy ∥ u ∗ ε ∥ H 1 ≤ ∥ u ∗ ∥ H 1 \lVert u_*^\varepsilon\rVert_{H^1}\le\lVert u_*\rVert_{H^1} ∥ u ∗ ε ∥ H 1 ≤ ∥ u ∗ ∥ H 1 , and converge to u ∗ u_* u ∗ in H 1 H^1 H 1 . Since 4 > d 4>d 4 > d , Proposition 9.1 with q 0 = 4 q_0=4 q 0 = 4 gives solutions u ε ∈ M R 4 u^\varepsilon\in MR_4 u ε ∈ M R 4 of (52) with data ( u ∗ ε , f ∗ ) (u_*^\varepsilon,f_*) ( u ∗ ε , f ∗ ) on some interval. We show that they exist on an interval [ 0 , T ] [0,T] [ 0 , T ] independent of ε \varepsilon ε , with uniform bounds.
For a solution on [ 0 , t ] [0,t] [ 0 , t ] write
R v ε ( t ) : = R ∗ v + ∫ 0 t ∥ u ε ∥ L ∞ , ϱ ( t ) : = e d ∥ f ∗ ∥ ∞ ∣ B R v ε ( t ) ∣ . R^{v^\varepsilon}(t):=R_*^v+\int_0^t\lVert u^\varepsilon\rVert_{L^\infty},\qquad\varrho(t):=e^d\lVert f_*\rVert_\infty\lvert B_{R^{v^\varepsilon}(t)}\rvert. R v ε ( t ) := R ∗ v + ∫ 0 t ∥ u ε ∥ L ∞ , ϱ ( t ) := e d ∥ f ∗ ∥ ∞ ∣ B R v ε ( t ) ∣ .
Then f [ u ε ] ( t ) f[u^\varepsilon](t) f [ u ε ] ( t ) vanishes for ∣ v ∣ ≥ R v ε ( t ) \lvert v\rvert\ge R^{v^\varepsilon}(t) ∣ v ∣ ≥ R v ε ( t ) , ρ [ u ε ] ( t ) ≤ ϱ ( t ) \rho[u^\varepsilon](t)\le\varrho(t) ρ [ u ε ] ( t ) ≤ ϱ ( t ) , and ∥ f [ u ε ] ( t ) ∥ ∞ ≤ e d ∥ f ∗ ∥ ∞ \lVert f[u^\varepsilon](t)\rVert_\infty\le e^d\lVert f_*\rVert_\infty ∥ f [ u ε ] ( t ) ∥ ∞ ≤ e d ∥ f ∗ ∥ ∞ .
Step 2: uniform a priori bounds. Let u = u ε u=u^\varepsilon u = u ε be an M R 4 MR_4 M R 4 solution on [ 0 , t ] [0,t] [ 0 , t ] with t ≤ T t\le T t ≤ T and R v ε ( t ) ≤ R ∗ v + 1 R^{v^\varepsilon}(t)\le R_*^v+1 R v ε ( t ) ≤ R ∗ v + 1 . Write ϱ 1 : = ϱ ∣ R v = R ∗ v + 1 \varrho_1:=\varrho\vert_{R^v=R_*^v+1} ϱ 1 := ϱ ∣ R v = R ∗ v + 1 .
H 1 H^1 H 1 bound. Test (52) with − Δ u -\Delta u − Δ u , which is divergence-free. By the Gagliardo–Nirenberg inequalities in d ≤ 3 d\le3 d ≤ 3 , ∣ ∫ ( u ⋅ ∇ u ) ⋅ Δ u ∣ ≤ 1 4 ∥ Δ u ∥ L 2 2 + C Y 3 \lvert\int(u\cdot\nabla u)\cdot\Delta u\rvert\le\frac14\lVert\Delta u\rVert_{L^2}^2+CY^3 ∣ ∫ ( u ⋅ ∇ u ) ⋅ Δ u ∣ ≤ 4 1 ∥ Δ u ∥ L 2 2 + C Y 3 , where Y : = 1 + ∥ u ∥ H 1 2 Y:=1+\lVert u\rVert_{H^1}^2 Y := 1 + ∥ u ∥ H 1 2 . Also ∥ F ∥ L 2 ≤ ϱ 1 ( R ∗ v + 1 + ∥ u ∥ L 2 ) ≤ C \lVert F\rVert_{L^2}\le\varrho_1(R_*^v+1+\lVert u\rVert_{L^2})\le C ∥ F ∥ L 2 ≤ ϱ 1 ( R ∗ v + 1 + ∥ u ∥ L 2 ) ≤ C . Hence
Y ′ + ∥ Δ u ∥ L 2 2 ≤ C Y 3 + C . Y'+\lVert\Delta u\rVert_{L^2}^2\le CY^3+C. Y ′ + ∥ Δ u ∥ L 2 2 ≤ C Y 3 + C .
The constant C C C here depends only on ∥ u ∗ ∥ L 2 \lVert u_*\rVert_{L^2} ∥ u ∗ ∥ L 2 , R ∗ v R_*^v R ∗ v , ∥ f ∗ ∥ L ∞ \lVert f_*\rVert_{L^\infty} ∥ f ∗ ∥ L ∞ and ∥ f ∗ ∥ L 1 \lVert f_*\rVert_{L^1} ∥ f ∗ ∥ L 1 , not on ∥ u ∗ ∥ H 1 \lVert u_*\rVert_{H^1} ∥ u ∗ ∥ H 1 . Comparison with the Riccati equation gives Y ≤ 2 Y ( 0 ) Y\le2Y(0) Y ≤ 2 Y ( 0 ) for s ≤ T 2 : = c Y ( 0 ) − 2 s\le T_2:=cY(0)^{-2} s ≤ T 2 := c Y ( 0 ) − 2 . Integrating, ∫ 0 s ∥ Δ u ∥ L 2 2 ≤ Y ( 0 ) + C s Y ( 0 ) 3 ≤ 2 Y ( 0 ) \int_0^s\lVert\Delta u\rVert_{L^2}^2\le Y(0)+CsY(0)^3\le2Y(0) ∫ 0 s ∥ Δ u ∥ L 2 2 ≤ Y ( 0 ) + C s Y ( 0 ) 3 ≤ 2 Y ( 0 ) for s ≤ T 2 s\le T_2 s ≤ T 2 , after decreasing c c c .
Velocity support. H 2 ⊂ L ∞ H^2\subset L^\infty H 2 ⊂ L ∞ for d ≤ 3 d\le3 d ≤ 3 , so ∫ 0 s ∥ u ∥ L ∞ ≤ C s 1 / 2 ( ∫ 0 s ∥ u ∥ H 2 2 ) 1 / 2 ≤ C s 1 / 2 Y ( 0 ) 1 / 2 \int_0^s\lVert u\rVert_{L^\infty}\le Cs^{1/2}\left(\int_0^s\lVert u\rVert_{H^2}^2\right)^{1/2}\le Cs^{1/2}Y(0)^{1/2} ∫ 0 s ∥ u ∥ L ∞ ≤ C s 1/2 ( ∫ 0 s ∥ u ∥ H 2 2 ) 1/2 ≤ C s 1/2 Y ( 0 ) 1/2 . Hence R v ( s ) ≤ R ∗ v + 1 2 R^v(s)\le R_*^v+\frac12 R v ( s ) ≤ R ∗ v + 2 1 for s ≤ T 3 : = c Y ( 0 ) − 1 s\le T_3:=cY(0)^{-1} s ≤ T 3 := c Y ( 0 ) − 1 .
Set T : = min { T 2 , T 3 , 1 } T:=\min\{T_2,T_3,1\} T := min { T 2 , T 3 , 1 } . Thus every M R 4 MR_4 M R 4 solution on [ 0 , t ] [0,t] [ 0 , t ] with t ≤ T t\le T t ≤ T and R v ( t ) ≤ R ∗ v + 1 R^v(t)\le R_*^v+1 R v ( t ) ≤ R ∗ v + 1 satisfies the bounds above, and in particular R v ( t ) ≤ R ∗ v + 1 2 R^v(t)\le R_*^v+\frac12 R v ( t ) ≤ R ∗ v + 2 1 . In addition:
Lipschitz bound. Write G : = F − u ⋅ ∇ u G:=F-u\cdot\nabla u G := F − u ⋅ ∇ u . Then ∥ G ∥ L 6 ≤ ∥ u ∥ ∞ ∥ ∇ u ∥ L 6 + ∥ F ∥ L 6 ≤ C ∥ u ∥ H 2 2 + C ( 1 + ∥ u ∥ H 2 ) \lVert G\rVert_{L^6}\le\lVert u\rVert_\infty\lVert\nabla u\rVert_{L^6}+\lVert F\rVert_{L^6}\le C\lVert u\rVert_{H^2}^2+C(1+\lVert u\rVert_{H^2}) ∥ G ∥ L 6 ≤ ∥ u ∥ ∞ ∥ ∇ u ∥ L 6 + ∥ F ∥ L 6 ≤ C ∥ u ∥ H 2 2 + C ( 1 + ∥ u ∥ H 2 ) , so ∥ G ∥ L 1 ( 0 , t ; L 6 ) ≤ C \lVert G\rVert_{L^1(0,t;L^6)}\le C ∥ G ∥ L 1 ( 0 , t ; L 6 ) ≤ C . On T d \mathbb{T}^d T d , for 0 < s ≤ 1 0<s\le1 0 < s ≤ 1 ,
∥ ∇ e s Δ g ∥ ∞ ≤ C s − d / 4 ∥ ∇ g ∥ L 2 , ∥ ∇ e s Δ Q g ∥ ∞ ≤ C s − 1 2 − d 12 ∥ g ∥ L 6 . \lVert\nabla e^{s\Delta}g\rVert_\infty\le Cs^{-d/4}\lVert\nabla g\rVert_{L^2},\qquad\lVert\nabla e^{s\Delta}Qg\rVert_\infty\le Cs^{-\frac12-\frac d{12}}\lVert g\rVert_{L^6}. ∥ ∇ e s Δ g ∥ ∞ ≤ C s − d /4 ∥ ∇ g ∥ L 2 , ∥ ∇ e s Δ Q g ∥ ∞ ≤ C s − 2 1 − 12 d ∥ g ∥ L 6 .
The exponents d 4 \frac d4 4 d and 1 2 + d 12 \frac12+\frac d{12} 2 1 + 12 d are less than 1 1 1 . From the Duhamel formula, ∫ 0 t ∥ ∇ u ∥ L ∞ ≤ C \int_0^t\lVert\nabla u\rVert_{L^\infty}\le C ∫ 0 t ∥ ∇ u ∥ L ∞ ≤ C .
X 4 X_4 X 4 bound (non-uniform). Let y ( s ) : = sup [ 0 , s ] ∥ u ∥ X 4 y(s) := \sup_{[0,s]}\|u\|_{X_4} y ( s ) := sup [ 0 , s ] ∥ u ∥ X 4 . By Lemma 5.4(a), (b) and ∥ u ⋅ ∇ u ∥ L 4 ≤ ∥ u ∥ ∞ ∥ ∇ u ∥ L 4 ≤ C y 1 / 2 ∥ u ∥ H 2 \|u\cdot\nabla u\|_{L^4} \le\|u\|_\infty\|\nabla u\|_{L^4} \le Cy^{1/2}\|u\|_{H^2} ∥ u ⋅ ∇ u ∥ L 4 ≤ ∥ u ∥ ∞ ∥∇ u ∥ L 4 ≤ C y 1/2 ∥ u ∥ H 2 ,
y ( s ) ≤ C ( ∥ u ∗ ε ∥ X 4 2 + 1 ) + C ∫ 0 s y ( σ ) ∥ u ( σ ) ∥ H 2 2 d σ . y(s) \le C(\|u^\varepsilon_*\|_{X_4}^2+1)+C\int_0^s y(\sigma)\|u(\sigma)\|_{H^2}^2\,\mathrm{d}\sigma. y ( s ) ≤ C ( ∥ u ∗ ε ∥ X 4 2 + 1 ) + C ∫ 0 s y ( σ ) ∥ u ( σ ) ∥ H 2 2 d σ .
Gronwall’s inequality gives y ≤ B ε < ∞ y\le B_\varepsilon<\infty y ≤ B ε < ∞ on [ 0 , t ] [0,t] [ 0 , t ] .
Step 3: existence on [ 0 , T ] [0,T] [ 0 , T ] . Fix ε \varepsilon ε . Let T \mathcal T T be the set of t ∈ ( 0 , T ] t\in(0,T] t ∈ ( 0 , T ] such that (9.1) with data ( u ∗ ε , f ∗ ) (u^\varepsilon_*,f_*) ( u ∗ ε , f ∗ ) has a solution u ∈ M R 4 ( [ 0 , t ] ) u\in\mathrm{MR}_4([0,t]) u ∈ MR 4 ([ 0 , t ]) with R v ( t ) ≤ R ∗ v + 1 R^v(t)\le R^v_*+1 R v ( t ) ≤ R ∗ v + 1 . Two such solutions coincide on their common interval: both are weak solutions with velocity in L 2 L ∞ L^2L^\infty L 2 L ∞ and datum u ∗ ε ∈ X 4 u^\varepsilon_*\in X_4 u ∗ ε ∈ X 4 , 4 > d 4>d 4 > d , so Corollary 5.11(c) applies. Hence the solutions for t ∈ T t\in\mathcal T t ∈ T are restrictions of a single function. By Proposition 9.1, T ≠ ∅ \mathcal T\ne\varnothing T = ∅ , since R v R^v R v is continuous with R v ( 0 ) = R ∗ v R^v(0)=R^v_* R v ( 0 ) = R ∗ v . For t 0 ∈ T t_0\in\mathcal T t 0 ∈ T , Step 2 gives R v ( t 0 ) ≤ R ∗ v + 1 2 R^v(t_0)\le R^v_*+\frac12 R v ( t 0 ) ≤ R ∗ v + 2 1 and ∥ u ( t 0 ) ∥ X 4 ≤ B ε \|u(t_0)\|_{X_4}\le B_\varepsilon ∥ u ( t 0 ) ∥ X 4 ≤ B ε , where B ε B_\varepsilon B ε does not depend on t 0 t_0 t 0 . Proposition 9.1, applied from t 0 t_0 t 0 with data ( u ( t 0 ) , f [ u ] ( t 0 ) ) (u(t_0),f[u](t_0)) ( u ( t 0 ) , f [ u ] ( t 0 )) , extends the solution to [ t 0 , t 0 + τ ε ] [t_0,t_0+\tau_\varepsilon] [ t 0 , t 0 + τ ε ] . Here ∥ f [ u ] ( t 0 ) ∥ ∞ ≤ e d ∥ f ∗ ∥ ∞ \|f[u](t_0)\|_\infty\le e^d\|f_*\|_\infty ∥ f [ u ] ( t 0 ) ∥ ∞ ≤ e d ∥ f ∗ ∥ ∞ , the velocity support of f [ u ] ( t 0 ) f[u](t_0) f [ u ] ( t 0 ) has radius at most R ∗ v + 1 R^v_*+1 R ∗ v + 1 , and τ ε > 0 \tau_\varepsilon>0 τ ε > 0 depends only on B ε B_\varepsilon B ε , R ∗ v R^v_* R ∗ v and ∥ f ∗ ∥ ∞ \|f_*\|_\infty ∥ f ∗ ∥ ∞ . The kinetic state of the extension is the push-forward of f ∗ f_* f ∗ from time 0, by the flow property, and the concatenation is in M R 4 \mathrm{MR}_4 MR 4 , since the traces match (Lemma 5.4(b)). On the extension R v ≤ R ∗ v + 1 R^v\le R^v_*+1 R v ≤ R ∗ v + 1 persists up to min { t 0 + τ ε , T } \min\{t_0+\tau_\varepsilon,T\} min { t 0 + τ ε , T } . Otherwise there is a first time t ♯ t_\sharp t ♯ with R v ( t ♯ ) = R ∗ v + 1 R^v(t_\sharp)=R^v_*+1 R v ( t ♯ ) = R ∗ v + 1 ; then t ♯ ∈ T t_\sharp\in\mathcal T t ♯ ∈ T , and Step 2 gives R v ( t ♯ ) ≤ R ∗ v + 1 2 R^v(t_\sharp)\le R^v_*+\frac12 R v ( t ♯ ) ≤ R ∗ v + 2 1 , a contradiction. So min { t 0 + τ ε , T } ∈ T \min\{t_0+\tau_\varepsilon,T\}\in\mathcal T min { t 0 + τ ε , T } ∈ T for every t 0 ∈ T t_0\in\mathcal T t 0 ∈ T . Since τ ε \tau_\varepsilon τ ε does not depend on t 0 t_0 t 0 , this gives T ∈ T T\in\mathcal T T ∈ T .
Step 4: limit. By Step 2 and (9.1), ( u ε ) (u^\varepsilon) ( u ε ) is bounded in L ∞ H 1 ∩ L 2 H 2 L^\infty H^1\cap L^2H^2 L ∞ H 1 ∩ L 2 H 2 and ∂ t u ε \partial_tu^\varepsilon ∂ t u ε is bounded in L 2 L 2 L^2L^2 L 2 L 2 . By the Aubin–Lions lemma, a subsequence converges in L 2 ( 0 , T ; H 7 / 4 ) ∩ C ( [ 0 , T ] ; L 2 ) L^2(0,T;H^{7/4})\cap C([0,T];L^2) L 2 ( 0 , T ; H 7/4 ) ∩ C ([ 0 , T ] ; L 2 ) to some u u u with the stated regularity. Since H 7 / 4 ⊂ L ∞ H^{7/4}\subset L^\infty H 7/4 ⊂ L ∞ , u ε → u u^\varepsilon\to u u ε → u in L 1 L ∞ L^1L^\infty L 1 L ∞ . Along a further subsequence, ∇ u ε ( t ) → ∇ u ( t ) \nabla u^\varepsilon(t)\to\nabla u(t) ∇ u ε ( t ) → ∇ u ( t ) in L 2 L^2 L 2 for a.e. t t t . By lower semicontinuity of ∥ ⋅ ∥ L ∞ \|\cdot\|_{L^\infty} ∥ ⋅ ∥ L ∞ under this convergence and Fatou’s lemma, ∫ 0 T ∥ ∇ u ∥ ∞ ≤ C \int_0^T\|\nabla u\|_\infty\le C ∫ 0 T ∥∇ u ∥ ∞ ≤ C . Hence u ∈ V s u\in V_s u ∈ V s on [ 0 , T ] [0,T] [ 0 , T ] , and Lemma 8.1(b), with base u u u , gives
sup t sup ∥ ψ ∥ C b 1 ≤ 1 ∣ ∬ ( f [ u ε ] − f [ u ] ) ( t ) ψ ∣ ≤ e 1 + C M 0 ∥ u ε − u ∥ L 1 L ∞ → 0. \sup_t\sup_{\|\psi\|_{C_b^1}\le1}\left|\iint(f[u^\varepsilon]-f[u])(t)\psi\right|
\le e^{1+C}M_0\|u^\varepsilon-u\|_{L^1L^\infty}\to0. t sup ∥ ψ ∥ C b 1 ≤ 1 sup ∬ ( f [ u ε ] − f [ u ]) ( t ) ψ ≤ e 1 + C M 0 ∥ u ε − u ∥ L 1 L ∞ → 0.
The velocity supports are uniformly bounded. As in Lemma 3.5(d), the drag forces converge against continuous test functions. Since u ε → u u^\varepsilon\to u u ε → u in L t , x 2 L^2_{t,x} L t , x 2 , we may pass to the limit in the weak formulation of (9.1), so ( u , f [ u ] ) (u,f[u]) ( u , f [ u ]) is a weak solution.
Finally, ( u , f [ u ] ) (u,f[u]) ( u , f [ u ]) satisfies the energy equality, hence the energy inequality. The computation in the proof of Proposition 5.10(d) applies: it only uses ∂ t u ∈ L 2 L 2 \partial_tu\in L^2L^2 ∂ t u ∈ L 2 L 2 , u ∈ L 2 H 2 u\in L^2H^2 u ∈ L 2 H 2 and the Lipschitz phase-space flow.
Step 5: X 4 X_4 X 4 at a.e. time. u ( t ) ∈ H 2 u(t)\in H^2 u ( t ) ∈ H 2 for a.e. t t t . Moreover H 2 = B 2 , 2 2 ↪ B 4 , 2 2 − d / 4 ↪ B 4 , 2 1 = X 4 H^2=B^2_{2,2}\hookrightarrow B^{2-d/4}_{4,2}\hookrightarrow B^1_{4,2}=X_4 H 2 = B 2 , 2 2 ↪ B 4 , 2 2 − d /4 ↪ B 4 , 2 1 = X 4 , since d ≤ 3 d\le3 d ≤ 3 [28 ] .
Proof of Theorem 7.3. (a) This is Proposition 9.1, with u ∗ = u i n u_*=u_{\mathrm{in}} u ∗ = u in and f ∗ = f 0 f_*=f_0 f ∗ = f 0 . The examples embed into X q 0 X_{q_0} X q 0 for a suitable q 0 > d q_0>d q 0 > d : H s = B 2 , 2 s ↪ B q 0 , 2 s − d / 2 + d / q 0 ↪ B q 0 , 2 1 H^s=B^s_{2,2}\hookrightarrow B^{s-d/2+d/q_0}_{q_0,2}\hookrightarrow B^1_{q_0,2} H s = B 2 , 2 s ↪ B q 0 , 2 s − d /2 + d / q 0 ↪ B q 0 , 2 1 if s ≥ 1 + d 2 − d q 0 s\ge1+\frac d2-\frac d{q_0} s ≥ 1 + 2 d − q 0 d , which can be arranged for s > d 2 s>\frac d2 s > 2 d with q 0 q_0 q 0 close to d d d , and C 1 , α ↪ B q 0 , 2 1 + α ↪ B q 0 , 2 1 C^{1,\alpha}\hookrightarrow B^{1+\alpha}_{q_0,2}\hookrightarrow B^1_{q_0,2} C 1 , α ↪ B q 0 , 2 1 + α ↪ B q 0 , 2 1 .
(b) Let ( u , f ) (u,f) ( u , f ) be given by Proposition 9.2 on [ 0 , T ] [0,T] [ 0 , T ] ; replacing T T T by min { T , 1 2 } \min\{T,\frac12\} min { T , 2 1 } , we may assume T ≤ 1 2 T\le\frac12 T ≤ 2 1 . Choose a ∈ ( 0 , T ) a\in(0,T) a ∈ ( 0 , T ) with u ( a ) ∈ X 4 u(a)\in X_4 u ( a ) ∈ X 4 . Note that f ( a ) f(a) f ( a ) is bounded and vanishes for ∣ v ∣ |v| ∣ v ∣ large. Proposition 9.1, applied from time a a a with data ( u ( a ) , f ( a ) ) (u(a),f(a)) ( u ( a ) , f ( a )) , gives an M R 4 \mathrm{MR}_4 MR 4 solution on [ a , a + T ′ ] [a,a+T'] [ a , a + T ′ ] . Concatenate by Lemma 8.6, using u ∈ C ( [ 0 , a ] ; L 2 ) ∩ L 1 W 1 , 1 u\in C([0,a];L^2)\cap L^1W^{1,1} u ∈ C ([ 0 , a ] ; L 2 ) ∩ L 1 W 1 , 1 . The concatenation is a strong continuation with t 0 = 0 t_0=0 t 0 = 0 , q 0 = 4 q_0=4 q 0 = 4 and T 0 = min { a + T ′ , T } T_0=\min\{a+T',T\} T 0 = min { a + T ′ , T } . Restricting to a shorter final time preserves (S1)–(S3), with T 0 ≤ 1 2 T_0\le\frac12 T 0 ≤ 2 1 . Indeed, (S1) holds since L 2 H 2 ⊂ L 2 L ∞ L^2H^2\subset L^2L^\infty L 2 H 2 ⊂ L 2 L ∞ for $d\le3 and the pieces match continuously in L 2 L^2 L 2 at a a a ; (S2) holds on both pieces; and (S3) holds on [ a , T 0 ] [a,T_0] [ a , T 0 ] .
(c) and (d). Let d ∈ { 2 , 3 } d\in\{2,3\} d ∈ { 2 , 3 } . By Theorem 11.4, there is an energy solution ( u , f ) (u,f) ( u , f ) on [ 0 , 1 ] [0,1] [ 0 , 1 ] with data ( u i n , f 0 ) (u_{\mathrm{in}},f_0) ( u in , f 0 ) which satisfies the strong energy inequality (10.1). By Lemma 11.2 if d = 2 d=2 d = 2 , and by Lemma 11.5 if d = 2 d=2 d = 2 , u ∈ L 1 ( 0 , 1 ; L ∞ ) u\in L^1(0,1;L^\infty) u ∈ L 1 ( 0 , 1 ; L ∞ ) , f f f is bounded and vanishes for ∣ v ∣ ≥ R 0 v + ∥ u ∥ L 1 L ∞ |v|\ge R^v_0+\|u\|_{L^1L^\infty} ∣ v ∣ ≥ R 0 v + ∥ u ∥ L 1 L ∞ , and ( u , f ) (u,f) ( u , f ) is a weak solution in the sense of Definition 2.1; if d = 2 d=2 d = 2 , moreover u ∈ L 2 ( 0 , 1 ; L ∞ ) u\in L^2(0,1;L^\infty) u ∈ L 2 ( 0 , 1 ; L ∞ ) . Since u ∈ C w ( [ 0 , 1 ] ; L 2 ) ∩ L 2 ( 0 , 1 ; H 1 ) u\in C_w([0,1];L^2)\cap L^2(0,1;H^1) u ∈ C w ([ 0 , 1 ] ; L 2 ) ∩ L 2 ( 0 , 1 ; H 1 ) , (S1) holds on [ 0 , t 1 ] [0,t_1] [ 0 , t 1 ] for every t 1 t_1 t 1 .
Choose t 1 ∈ ( 0 , 1 2 ) t_1\in(0,\frac12) t 1 ∈ ( 0 , 2 1 ) such that u ( t 1 ) ∈ H 1 u(t_1)\in H^1 u ( t 1 ) ∈ H 1 and (10.1) holds with s = t 1 s=t_1 s = t 1 ; almost every t 1 t_1 t 1 has these properties. Note that f ( t 1 ) f(t_1) f ( t 1 ) is bounded and vanishes for ∣ v ∣ |v| ∣ v ∣ large. Apply (b) from time t 1 t_1 t 1 to the data ( u ( t 1 ) , f ( t 1 ) ) (u(t_1),f(t_1)) ( u ( t 1 ) , f ( t 1 )) ; it gives a solution on [ t 1 , t 1 + T 0 ′ ] [t_1,t_1+T'_0] [ t 1 , t 1 + T 0 ′ ] with T 0 ′ ≤ 1 2 T'_0\le\frac12 T 0 ′ ≤ 2 1 (see the proof of (b)), so that T 0 : = t 1 + T 0 ′ < 1 T_0:=t_1+T'_0<1 T 0 := t 1 + T 0 ′ < 1 . Concatenate by Lemma 8.6, which needs u ∈ C w ( [ 0 , t 1 ] ; L 2 ) ∩ L 1 ( 0 , t 1 ; W 1 , 1 ) u\in C_w([0,t_1];L^2)\cap L^1(0,t_1;W^{1,1}) u ∈ C w ([ 0 , t 1 ] ; L 2 ) ∩ L 1 ( 0 , t 1 ; W 1 , 1 ) ; this holds by (E1). The result is a strong continuation ( u ˉ , f ˉ ) (\bar u,\bar f) ( u ˉ , f ˉ ) on [ 0 , T 0 ] [0,T_0] [ 0 , T 0 ] with this t 1 t_1 t 1 . It satisfies u ˉ ∈ C ( [ t 1 , T 0 ] ; L 2 ) ∩ L 2 ( t 1 , T 0 ; L ∞ ) \bar u\in C([t_1,T_0];L^2)\cap L^2(t_1,T_0;L^\infty) u ˉ ∈ C ([ t 1 , T 0 ] ; L 2 ) ∩ L 2 ( t 1 , T 0 ; L ∞ ) by the construction of (b): Proposition 9.2 on the first piece and Proposition 9.1 on the second. If d = 2 d=2 d = 2 , then also u ˉ = u ∈ L 2 ( 0 , t 1 ; L ∞ ) \bar{u}=u\in L^{2}(0,t_{1};L^{\infty}) u ˉ = u ∈ L 2 ( 0 , t 1 ; L ∞ ) , so u ˉ ∈ L 2 ( 0 , T 0 ; L ∞ ) \bar{u}\in L^{2}(0,T_{0};L^{\infty}) u ˉ ∈ L 2 ( 0 , T 0 ; L ∞ ) , as claimed for case (c).
In fact ( u ˉ , f ˉ ) = ( u , f ) (\bar{u},\bar{f})=(u,f) ( u ˉ , f ˉ ) = ( u , f ) on [ 0 , T 0 ] [0,T_{0}] [ 0 , T 0 ] , so the strong continuation is a restriction of the Leray solution. Indeed, on [ t 1 , T 0 ] [t_{1},T_{0}] [ t 1 , T 0 ] the velocity u ˉ \bar{u} u ˉ satisfies hypothesis (a) of Lemma 11.3: on the first piece of the construction of (b) by Proposition 9.2, on the second piece because M R 4 ⊂ M R 2 ∩ L 2 W 1 , ∞ \mathrm{MR}_{4}\subset\mathrm{MR}_{2}\cap L^{2}W^{1,\infty} MR 4 ⊂ MR 2 ∩ L 2 W 1 , ∞ , and the two pieces match in H 1 H^{1} H 1 . The restriction of ( u , f ) (u,f) ( u , f ) to [ t 1 , T 0 ] [t_{1},T_{0}] [ t 1 , T 0 ] is a weak solution with data ( u ( t 1 ) , f ( t 1 ) ) (u(t_{1}),f(t_{1})) ( u ( t 1 ) , f ( t 1 )) , as in the proof of Lemma 8.6, and it satisfies hypothesis (b) of Lemma 11.3 by the choice of t 1 t_{1} t 1 .
Remark 9.3. As explained in Section 1.4, the kinetic comparisons start at t 1 t_{1} t 1 from the common kinetic state f ˉ ( t 1 ) \bar{f}(t_{1}) f ˉ ( t 1 ) , so that no regularity of f 0 f_{0} f 0 and of u i n u_{\mathrm{in}} u in beyond (S1) is needed.
A structure theorem in three dimensions
The same ingredients give a Leray-type structure theorem for energy solutions of the three-dimensional system. Let d ∈ { 2 , 3 } d\in\{2,3\} d ∈ { 2 , 3 } . We say that ( u , f ) (u,f) ( u , f ) is a Leray solution with the strong energy inequality if it is an energy solution on [ 0 , 1 ] [0,1] [ 0 , 1 ] in the sense of Definition 11.1 and, for s = 0 s=0 s = 0 and almost every s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) , and for all t ∈ [ s , 1 ] t\in[s,1] t ∈ [ s , 1 ] ,
E ( t ) + ∫ s t ( ∥ ∇ u ∥ L 2 2 + ∬ ∣ v − u ∣ 2 f d z ) d σ ≤ E ( s ) . (54) \mathcal{E}(t)+\int_{s}^{t}\left(\|\nabla u\|_{L^{2}}^{2}+\iint|v-u|^{2}f\,\mathrm{d}z\right)\,\mathrm{d}\sigma\leq\mathcal{E}(s). \tag*{(54)} E ( t ) + ∫ s t ( ∥∇ u ∥ L 2 2 + ∬ ∣ v − u ∣ 2 f d z ) d σ ≤ E ( s ) . (54)
Since (10.1) is required for every t t t , we fix representatives: u u u is the representative in C w ( [ 0 , 1 ] ; L 2 ) C_{w}([0,1];L^{2}) C w ([ 0 , 1 ] ; L 2 ) , given by (E1), and f f f is the representative in C ( [ 0 , 1 ] ; L 1 ) C([0,1];L^{1}) C ([ 0 , 1 ] ; L 1 ) given by Lemma 11.2(c) or Lemma 11.5(c). By Lemma 11.2 if d = 3 d=3 d = 3 , and by Lemma 11.5 if d = 2 d=2 d = 2 , such a solution is a weak solution in the sense of Definition 2.1: u ∈ L 2 ( 0 , 1 ; L ∞ ) u\in L^{2}(0,1;L^{\infty}) u ∈ L 2 ( 0 , 1 ; L ∞ ) , f f f has bounded velocity support, and f ∈ C ( [ 0 , 1 ] ; L 1 ) f\in C([0,1];L^{1}) f ∈ C ([ 0 , 1 ] ; L 1 ) is the unique bounded, hence renormalized, solution of the kinetic equation with field u u u . By Theorem 11.4, Leray solutions with the strong energy inequality exist for all data considered here.
Theorem 10.1 (Structure of Leray solutions, d = 3 d=3 d = 3 ). Let d = 3 d=3 d = 3 , let u i n ∈ L 2 ( T 3 ) u_{\mathrm{in}}\in L^{2}(\mathbb{T}^{3}) u in ∈ L 2 ( T 3 ) be divergence-free, and let f 0 ∈ L ∞ f_{0}\in L^{\infty} f 0 ∈ L ∞ be nonnegative and vanish for ∣ v ∣ ≥ R 0 v |v|\geq R_{0}^{v} ∣ v ∣ ≥ R 0 v . Let ( u , f ) (u,f) ( u , f ) be a Leray solution with the strong energy inequality and data ( u i n , f 0 ) (u_{\mathrm{in}},f_{0}) ( u in , f 0 ) . Then there is an open set O ⊂ ( 0 , 1 ) \mathcal{O}\subset(0,1) O ⊂ ( 0 , 1 ) such that:
(a) the singular set S : = ( 0 , 1 ) ∖ O \mathcal{S}:=(0,1)\setminus\mathcal{O} S := ( 0 , 1 ) ∖ O satisfies H 1 / 2 ( S ) = 0 \mathcal{H}^{1/2}(\mathcal{S})=0 H 1/2 ( S ) = 0 ; in particular ∣ S ∣ = 0 |\mathcal{S}|=0 ∣ S ∣ = 0 ;
(b) on every compact interval K ⊂ O K\subset\mathcal{O} K ⊂ O , u ∈ L ∞ ( K ; H 1 ) ∩ L 2 ( K ; H 2 ) u\in L^{\infty}(K;H^{1})\cap L^{2}(K;H^{2}) u ∈ L ∞ ( K ; H 1 ) ∩ L 2 ( K ; H 2 ) and ∇ u ∈ L 1 ( K ; L ∞ ) \nabla u\in L^{1}(K;L^{\infty}) ∇ u ∈ L 1 ( K ; L ∞ ) ; f f f is transported on K K K by a bi-Lipschitz phase-space flow;
(c) for almost every t 0 ∈ ( 0 , 1 ) t_{0}\in(0,1) t 0 ∈ ( 0 , 1 ) , the interval ( t 0 , t 0 + T ( t 0 ) ) ∩ ( 0 , 1 ) (t_{0},t_{0}+T(t_{0}))\cap(0,1) ( t 0 , t 0 + T ( t 0 )) ∩ ( 0 , 1 ) is contained in O \mathcal{O} O , where T ( t 0 ) = c ( 1 + ∥ u ( t 0 ) ∥ H 1 2 ) − 1 T(t_{0})=c(1+\|u(t_{0})\|_{H^{1}}^{2})^{-1} T ( t 0 ) = c ( 1 + ∥ u ( t 0 ) ∥ H 1 2 ) − 1 and c > 0 c>0 c > 0 depends only on the data.
Proof. Step 1: uniform kinetic bounds. By Lemma 11.2, K 1 : = ∥ u ∥ L 1 ( 0 , 1 ; L ∞ ) < ∞ K_{1}:=\|u\|_{L^{1}(0,1;L^{\infty})}<\infty K 1 := ∥ u ∥ L 1 ( 0 , 1 ; L ∞ ) < ∞ . For every t t t , f ( t ) f(t) f ( t ) vanishes for ∣ v ∣ ≥ R 1 v : = R 0 v + K 1 |v|\geq R_{1}^{v}:=R_{0}^{v}+K_{1} ∣ v ∣ ≥ R 1 v := R 0 v + K 1 , ∥ f ( t ) ∥ L ∞ ≤ e 3 ∥ f 0 ∥ L ∞ \|f(t)\|_{L^{\infty}}\leq e^{3}\|f_{0}\|_{L^{\infty}} ∥ f ( t ) ∥ L ∞ ≤ e 3 ∥ f 0 ∥ L ∞ , ∥ f ( t ) ∥ L 1 = ∥ f 0 ∥ L 1 \|f(t)\|_{L^{1}}=\|f_{0}\|_{L^{1}} ∥ f ( t ) ∥ L 1 = ∥ f 0 ∥ L 1 , and ∥ u ( t ) ∥ L 2 2 ≤ 2 E ( 0 ) \|u(t)\|_{L^{2}}^{2}\leq2\mathcal{E}(0) ∥ u ( t ) ∥ L 2 2 ≤ 2 E ( 0 ) by (10.1) with s = 0 s=0 s = 0 . Let c > 0 c>0 c > 0 be the constant c ∗ c_{*} c ∗ of (9.2), computed with the upper bounds ( 2 E ( 0 ) ) 1 / 2 (2\mathcal{E}(0))^{1/2} ( 2 E ( 0 ) ) 1/2 for ∥ u ( t ) ∥ L 2 \|u(t)\|_{L^{2}} ∥ u ( t ) ∥ L 2 , R 1 v R_{1}^{v} R 1 v , e 3 ∥ f 0 ∥ ∞ e^{3}\|f_{0}\|_{\infty} e 3 ∥ f 0 ∥ ∞ and ∥ f 0 ∥ L 1 \|f_{0}\|_{L^{1}} ∥ f 0 ∥ L 1 ; since c ∗ c_{*} c ∗ is nonincreasing in these bounds, it applies at every t 0 ∈ G t_{0}\in G t 0 ∈ G . It depends only on the data.
Step 2: local identification. Let G ⊂ ( 0 , 1 ) G\subset(0,1) G ⊂ ( 0 , 1 ) be the set of times t 0 t_{0} t 0 such that u ( t 0 ) ∈ H 1 u(t_{0})\in H^{1} u ( t 0 ) ∈ H 1 and (10.1) holds with s = t 0 s=t_{0} s = t 0 . It has full measure. Fix t 0 ∈ G t_{0}\in G t 0 ∈ G and put T ( t 0 ) = c ( 1 + ∥ u ( t 0 ) ∥ H 1 2 ) − 1 T(t_{0})=c(1+\|u(t_{0})\|_{H^{1}}^{2})^{-1} T ( t 0 ) = c ( 1 + ∥ u ( t 0 ) ∥ H 1 2 ) − 1 . Proposition 9.2, applied from time t 0 t_{0} t 0 with data ( u ( t 0 ) , f ( t 0 ) ) (u(t_{0}),f(t_{0})) ( u ( t 0 ) , f ( t 0 )) , gives a solution ( u ~ , f ~ ) (\tilde{u},\tilde{f}) ( u ~ , f ~ ) on J : = [ t 0 , t 0 + T ( t 0 ) ) ∩ [ 0 , 1 ] J:=[t_{0},t_{0}+T(t_{0}))\cap[0,1] J := [ t 0 , t 0 + T ( t 0 )) ∩ [ 0 , 1 ] . It satisfies u ~ ∈ L ∞ H 1 ∩ L 2 H 2 \tilde{u}\in L^{\infty}H^{1}\cap L^{2}H^{2} u ~ ∈ L ∞ H 1 ∩ L 2 H 2 , ∂ t u ~ ∈ L 2 L 2 \partial_{t}\tilde{u}\in L^{2}L^{2} ∂ t u ~ ∈ L 2 L 2 and ∇ u ~ ∈ L 1 L ∞ \nabla\tilde{u}\in L^{1}L^{\infty} ∇ u ~ ∈ L 1 L ∞ , hence u ~ ∈ L 2 ( J ; L ∞ ) ∩ L 1 ( J ; W 1 , ∞ ) \tilde{u}\in L^{2}(J;L^{\infty})\cap L^{1}(J;W^{1,\infty}) u ~ ∈ L 2 ( J ; L ∞ ) ∩ L 1 ( J ; W 1 , ∞ ) , and the energy inequality from t 0 t_{0} t 0 .
In particular ( u ~ , f ~ ) (\tilde{u},\tilde{f}) ( u ~ , f ~ ) satisfies hypothesis (a) of Lemma 11.3 on J J J . The restriction ( u , f ) ∣ J (u,f)|_{J} ( u , f ) ∣ J is a weak solution on J J J with data ( u ( t 0 ) , f ( t 0 ) ) (u(t_{0}),f(t_{0})) ( u ( t 0 ) , f ( t 0 )) , in the sense of Definition 2.1 translated in time: this follows from the weak formulations on [ 0 , 1 ] [0,1] [ 0 , 1 ] by a cutoff in time, using the weak continuity of u u u and the continuity of f f f in L 1 L^{1} L 1 , as in the proof of Lemma 8.6. Since t 0 ∈ G t_{0}\in G t 0 ∈ G , it satisfies hypothesis (b) of Lemma 11.3, with u ∗ = u ( t 0 ) u_{*}=u(t_{0}) u ∗ = u ( t 0 ) and f ∗ = f ( t 0 ) f_{*}=f(t_{0}) f ∗ = f ( t 0 ) , which is bounded and vanishes for ∣ v ∣ ≥ R 1 v |v|\geq R_{1}^{v} ∣ v ∣ ≥ R 1 v . By Lemma 11.3, ( u , f ) = ( u ~ , f ~ ) (u,f)=(\tilde{u},\tilde{f}) ( u , f ) = ( u ~ , f ~ ) on J J J .
Step 3: the open set. Let O : = ⋃ t 0 ∈ G ( t 0 , t 0 + T ( t 0 ) ) ∩ ( 0 , 1 ) \mathcal{O}:=\bigcup_{t_{0}\in G}(t_{0},t_{0}+T(t_{0}))\cap(0,1) O := ⋃ t 0 ∈ G ( t 0 , t 0 + T ( t 0 )) ∩ ( 0 , 1 ) . It is open, and (c) holds by construction. A compact interval K ⊂ O K\subset\mathcal{O} K ⊂ O is covered by finitely many of these intervals, on each of which u = u ~ u=\tilde{u} u = u ~ has the regularity of Step 2. Hence (b) holds, with the flow of Lemma 8.1(a) started at inf K \inf K inf K ; this flow exists because ∇ u ∈ L 1 ( K ; L ∞ ) \nabla u\in L^{1}(K;L^{\infty}) ∇ u ∈ L 1 ( K ; L ∞ ) on the whole interval K K K , and f f f is transported by it because, by the uniqueness part of Lemma 8.1(a) (see Remark 8.2), f f f is the only bounded weak solution on K \mathcal{K} K with datum f ( inf K ) f(\inf\mathcal{K}) f ( inf K ) . For a compact set K \mathcal{K} K that is not an interval, the regularity of u u u still holds on K \mathcal{K} K , but [ inf K , sup K ] [\inf\mathcal{K},\sup\mathcal{K}] [ inf K , sup K ] may meet S S S .
Step 4: the singular set. Let t ∈ S t \in S t ∈ S . For every s ∈ G ∩ ( 0 , t ) s \in G \cap(0,t) s ∈ G ∩ ( 0 , t ) we have t ∉ ( s , s + T ( s ) ) t \notin(s,s+T(s)) t ∈ / ( s , s + T ( s )) , so T ( s ) ≤ t − s T(s) \le t-s T ( s ) ≤ t − s , that is,
∥ ∇ u ( s ) ∥ L 2 2 ≥ c 1 / 2 ( t − s ) − 1 / 2 − 1 − 2 E ( 0 ) for a.e. s ∈ ( 0 , t ) . \|\nabla u(s)\|_{L^2}^2 \ge c^{1/2}(t-s)^{-1/2}-1-2\mathcal{E}(0) \qquad\text{for a.e. }s \in(0,t). ∥∇ u ( s ) ∥ L 2 2 ≥ c 1/2 ( t − s ) − 1/2 − 1 − 2 E ( 0 ) for a.e. s ∈ ( 0 , t ) .
Hence there is ε 0 > 0 \varepsilon_0>0 ε 0 > 0 , depending only on c c c and E ( 0 ) \mathcal{E}(0) E ( 0 ) , such that for ε ≤ min { ε 0 , t } \varepsilon\le\min\{\varepsilon_0,t\} ε ≤ min { ε 0 , t } ,
∫ t − ε t ∥ ∇ u ∥ L 2 2 d s ≥ 2 c 1 / 2 ε 1 / 2 − ( 1 + 2 E ( 0 ) ) ε ≥ c 1 / 2 ε 1 / 2 . (55) \int_{t-\varepsilon}^{t}\|\nabla u\|_{L^2}^2\,ds \ge2c^{1/2}\varepsilon^{1/2}-(1+2\mathcal{E}(0))\varepsilon\ge c^{1/2}\varepsilon^{1/2}. \tag*{(55)} ∫ t − ε t ∥∇ u ∥ L 2 2 d s ≥ 2 c 1/2 ε 1/2 − ( 1 + 2 E ( 0 )) ε ≥ c 1/2 ε 1/2 . (55)
Fix δ ∈ ( 0 , ε 0 ) \delta\in(0,\varepsilon_0) δ ∈ ( 0 , ε 0 ) , and let S δ S_\delta S δ be the δ \delta δ -neighborhood of S ∩ [ δ , 1 ) S \cap[\delta,1) S ∩ [ δ , 1 ) . Consider the closed intervals [ t − ε , t ] [t-\varepsilon,t] [ t − ε , t ] with t ∈ S ∩ [ δ , 1 ) t \in S\cap[\delta,1) t ∈ S ∩ [ δ , 1 ) and ε < δ \varepsilon<\delta ε < δ ; each point of S ∩ [ δ , 1 ) S\cap[\delta,1) S ∩ [ δ , 1 ) belongs to one of them (as its right endpoint), and their lengths are bounded. By the Vitali covering lemma there are pairwise disjoint intervals I k = [ t k − ε k , t k ] I_k=[t_k-\varepsilon_k,t_k] I k = [ t k − ε k , t k ] of this family, at most countably many, such that the closed intervals 5 I k 5I_k 5 I k , with the same centers and five times the lengths, cover S ∩ [ δ , 1 ) S\cap[\delta,1) S ∩ [ δ , 1 ) . The bound (10.2) holds for the closed intervals I k I_k I k as well, since endpoints have measure zero. Then, by (10.2),
H 10 δ 1 / 2 ( S ∩ [ δ , 1 ) ) ≤ ∑ k ( 5 ε k ) 1 / 2 ≤ 5 1 / 2 c − 1 / 2 ∑ k ∫ I k ∥ ∇ u ∥ L 2 2 ≤ C ∫ S δ ∥ ∇ u ∥ L 2 2 . \mathcal{H}^{1/2}_{10\delta}(S\cap[\delta,1))\le\sum_k(5\varepsilon_k)^{1/2}\le5^{1/2}c^{-1/2}\sum_k\int_{I_k}\|\nabla u\|_{L^2}^2\le C\int_{S_\delta}\|\nabla u\|_{L^2}^2. H 10 δ 1/2 ( S ∩ [ δ , 1 )) ≤ k ∑ ( 5 ε k ) 1/2 ≤ 5 1/2 c − 1/2 k ∑ ∫ I k ∥∇ u ∥ L 2 2 ≤ C ∫ S δ ∥∇ u ∥ L 2 2 .
The right side is at most C ∫ 0 1 ∥ ∇ u ∥ L 2 2 < ∞ C\int_0^1\|\nabla u\|_{L^2}^2<\infty C ∫ 0 1 ∥∇ u ∥ L 2 2 < ∞ , so H 1 / 2 ( S ∩ [ δ , 1 ) ) < ∞ \mathcal{H}^{1/2}(S\cap[\delta,1))<\infty H 1/2 ( S ∩ [ δ , 1 )) < ∞ for every δ \delta δ , and hence ∣ S ∣ = 0 |S|=0 ∣ S ∣ = 0 . Every point of S δ S_\delta S δ lies within δ \delta δ of S S S , so lim sup δ → 0 1 S δ ≤ 1 S \limsup_{\delta\to0}\mathbf{1}_{S_\delta}\le\mathbf{1}_S lim sup δ → 0 1 S δ ≤ 1 S pointwise, and ∣ S δ ∣ → ∣ S ∣ = 0 |S_\delta|\to|S|=0 ∣ S δ ∣ → ∣ S ∣ = 0 since S S S is relatively closed in ( 0 , 1 ) (0,1) ( 0 , 1 ) . By dominated convergence, ∫ S δ ∥ ∇ u ∥ L 2 2 → 0 \int_{S_\delta}\|\nabla u\|_{L^2}^2\to0 ∫ S δ ∥∇ u ∥ L 2 2 → 0 . Now fix δ ′ > 0 \delta'>0 δ ′ > 0 . For δ ≤ δ ′ \delta\le\delta' δ ≤ δ ′ ,
H 10 δ ′ 1 / 2 ( S ∩ [ δ ′ , 1 ) ) ≤ H 10 δ 1 / 2 ( S ∩ [ δ , 1 ) ) ≤ C ∫ S δ ∥ ∇ u ∥ L 2 2 , \mathcal{H}^{1/2}_{10\delta'}(S\cap[\delta',1))\le\mathcal{H}^{1/2}_{10\delta}(S\cap[\delta,1))\le C\int_{S_\delta}\|\nabla u\|_{L^2}^2, H 10 δ ′ 1/2 ( S ∩ [ δ ′ , 1 )) ≤ H 10 δ 1/2 ( S ∩ [ δ , 1 )) ≤ C ∫ S δ ∥∇ u ∥ L 2 2 ,
and letting δ → 0 \delta\to0 δ → 0 gives H 1 / 2 ( S ∩ [ δ ′ , 1 ) ) = 0 \mathcal{H}^{1/2}(S\cap[\delta',1))=0 H 1/2 ( S ∩ [ δ ′ , 1 )) = 0 . Taking the union over δ ′ = 1 / m \delta'=1/m δ ′ = 1/ m , m ∈ N m\in\mathbb{N} m ∈ N , gives H 1 / 2 ( S ) = 0 \mathcal{H}^{1/2}(S)=0 H 1/2 ( S ) = 0 .
Remark 10.2. (i) Theorem 10.1 is the analogue of Leray’s structure theorem for the Navier–Stokes equations. The coupling enters at two points. The local existence time of Proposition 9.2 must be controlled uniformly in time. This requires the bound u ∈ L t 1 L ∞ u\in L_t^1L^\infty u ∈ L t 1 L ∞ of Lemma 11.2, after [21 Lemma B.1] , which keeps the velocity support, the density and the drag force bounded along the solution. And the identification needs the weak–strong uniqueness of Lemma 11.3 ; see [21 Theorem 1.2] for strong solutions beyond the Lipschitz class.
(ii) The strong energy inequality (10.1) is needed to restart the comparison at a.e. t 0 t_0 t 0 . The solutions of Theorem 11.4 satisfy it, because the approximate solutions of its Galerkin scheme satisfy the energy equality on every interval [ s , t ] [s,t] [ s , t ] . We do not know whether every energy solution satisfies it.
(iii) For the consequences in d = 3 d=3 d = 3 in combination with Theorem 7.3 (d), see the discussion after Theorem 1.5 .
Energy solutions on the torus
This section collects the facts on energy solutions used in § 9 and § 10 : a velocity bound (Lemma 11.2), weak–strong uniqueness for Lipschitz strong solutions (Lemma 11.3 ), the existence of Leray solutions satisfying the strong energy inequality (Theorem 11.4 ), and a two-dimensional velocity bound (Lemma 11.5). The first and the last are periodic versions, for energy solutions, of [21 Lemma B.1] and [20 Lemma 3] . The proofs are self-contained, apart from standard tools (maximal regularity of the heat equation, the Aubin–Lions lemma, the maximal-function estimate of Lemma 5.5 ) and the uniqueness of bounded solutions of the kinetic equation for velocities in L t 1 W 1 , 1 ∩ L t 1 L ∞ L_t^1W^{1,1}\cap L_t^1L^\infty L t 1 W 1 , 1 ∩ L t 1 L ∞ (Lemma 3.6). The kinetic lemmas of § 3 and § 8 are used for bounded data and on subintervals [ t 0 , T ] [t_0,T] [ t 0 , T ] , as explained in Remark 8.2. Throughout, d ∈ { 2 , 3 } d\in\{2,3\} d ∈ { 2 , 3 } and f 0 f_0 f 0 is bounded, nonnegative and vanishes for ∣ v ∣ ≥ R v 0 |v|\ge R_v^0 ∣ v ∣ ≥ R v 0 ; Lemma 11.2 is stated for d = 3 d=3 d = 3 , and the last paragraph treats d = 2 d=2 d = 2 . Let Δ j \Delta_j Δ j , j ≥ 0 j\ge0 j ≥ 0 , be the periodic Littlewood–Paley blocks, localizing to ∣ η ∣ ∼ 2 j |\eta|\sim2^j ∣ η ∣ ∼ 2 j , and let P 0 P_0 P 0 be the projection on the zero mode.
Velocity moments. For a nonnegative function g g g on T d × R d \mathbb{T}^{d} \times\mathbb{R}^{d} T d × R d we write
m k g ( x ) : = ∫ R d g ( x , v ) ∣ v ∣ k d v , M k ( g ) : = ∬ g ∣ v ∣ k d z . m_k g(x) := \int_{\mathbb{R}^{d}} g(x,v)|v|^k\,dv,\qquad M_k(g) := \iint g|v|^k\,dz. m k g ( x ) := ∫ R d g ( x , v ) ∣ v ∣ k d v , M k ( g ) := ∬ g ∣ v ∣ k d z .
If 0 ≤ g ≤ F 0 \leq g \leq F 0 ≤ g ≤ F on R v d \mathbb{R}^{d}_v R v d and 0 ≤ k < m 0 \leq k < m 0 ≤ k < m , splitting the integral at ∣ v ∣ = r |v|=r ∣ v ∣ = r and optimizing in r r r gives
∫ g ∣ v ∣ k d v ≤ C F m − k m + d ( ∫ g ∣ v ∣ m d v ) k + d m + d . \int g|v|^k\,dv \leq C F^{\frac{m-k}{m+d}}\left(\int g|v|^m\,dv\right)^{\frac{k+d}{m+d}}. ∫ g ∣ v ∣ k d v ≤ C F m + d m − k ( ∫ g ∣ v ∣ m d v ) m + d k + d .
Hence, if 0 ≤ f ≤ F ∞ 0 \leq f \leq F_\infty 0 ≤ f ≤ F ∞ on T d × R d \mathbb{T}^{d} \times\mathbb{R}^{d} T d × R d ,
∥ m k f ∥ L m + d k + d ( T d ) ≤ C F ∞ m − k m + d M m ( f ) k + d m + d ( 0 ≤ k < m ) . (56) \|m_k f\|_{L^{\frac{m+d}{k+d}}(\mathbb{T}^{d})} \leq C F_\infty^{\frac{m-k}{m+d}} M_m(f)^{\frac{k+d}{m+d}}\qquad(0 \leq k < m). \tag*{(56)} ∥ m k f ∥ L k + d m + d ( T d ) ≤ C F ∞ m + d m − k M m ( f ) m + d k + d ( 0 ≤ k < m ) . (56)
Since ∣ T d ∣ = 1 |\mathbb{T}^{d}|=1 ∣ T d ∣ = 1 , we also have ∥ g ∥ L r ≤ ∥ g ∥ L s \|g\|_{L^r} \leq\|g\|_{L^s} ∥ g ∥ L r ≤ ∥ g ∥ L s for r ≤ s r \leq s r ≤ s , and H 1 ( T d ) ↪ L 6 H^1(\mathbb{T}^{d}) \hookrightarrow L^6 H 1 ( T d ) ↪ L 6 for d ≤ 3 d \leq3 d ≤ 3 .
Definition 11.1 (Energy solutions). Let d ∈ { 2 , 3 } d \in\{2,3\} d ∈ { 2 , 3 } , 0 < T ≤ 1 0<T\leq1 0 < T ≤ 1 , and let u i n ∈ L 2 ( T d ) u_{\mathrm{in}} \in L^2(\mathbb{T}^{d}) u in ∈ L 2 ( T d ) be weakly divergence-free. An energy solution on [ 0 , T ] [0,T] [ 0 , T ] with data ( u i n , f 0 ) (u_{\mathrm{in}},f_0) ( u in , f 0 ) is a pair ( u , f ) (u,f) ( u , f ) such that:
(E1) u ∈ C w ( [ 0 , T ] ; L 2 ) ∩ L 2 ( 0 , T ; H 1 ) u \in C_{\mathrm{w}}([0,T];L^2) \cap L^2(0,T;H^1) u ∈ C w ([ 0 , T ] ; L 2 ) ∩ L 2 ( 0 , T ; H 1 ) , and u ( t ) u(t) u ( t ) is weakly divergence-free;
(E2) f ∈ L ∞ ( ( 0 , T ) × T d × R d ) ∩ L ∞ ( 0 , T ; L 1 ( T d × R d ) ) f \in L^\infty((0,T)\times\mathbb{T}^{d}\times\mathbb{R}^{d})\cap L^\infty(0,T;L^1(\mathbb{T}^{d}\times\mathbb{R}^{d})) f ∈ L ∞ (( 0 , T ) × T d × R d ) ∩ L ∞ ( 0 , T ; L 1 ( T d × R d )) , f ≥ 0 f\geq0 f ≥ 0 , and t ↦ M 2 ( f ( t ) ) t\mapsto M_2(f(t)) t ↦ M 2 ( f ( t )) is essentially bounded;
(E3) the identities of Definition 2.1 (iii) and (iv) hold on [ 0 , T ] [0,T] [ 0 , T ] , with test functions vanishing near t = T t=T t = T ;
(E4) the energy inequality of Definition 2.2 holds for a.e. t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] , with E ( 0 ) : = 1 2 ∥ u i n ∥ L 2 2 + 1 2 M 2 ( f 0 ) \mathcal{E}(0):=\frac{1}{2}\|u_{\mathrm{in}}\|_{L^2}^2+\frac{1}{2}M_2(f_0) E ( 0 ) := 2 1 ∥ u in ∥ L 2 2 + 2 1 M 2 ( f 0 ) .
In contrast with Definition 2.1 (i)–(ii), neither u ∈ L t 1 L ∞ u\in L^1_tL^\infty u ∈ L t 1 L ∞ nor a bound on the velocity support of f f f is assumed; both are consequences (Lemmas 11.2 and 11.5). No renormalization property is assumed either. The identities in (E3) are meaningful: by (E2) and (11.1) with m = 2 m=2 m = 2 , ρ f ∈ L ∞ ( 0 , T ; L ( d + 2 ) / d ) \rho_f\in L^\infty(0,T;L^{(d+2)/d}) ρ f ∈ L ∞ ( 0 , T ; L ( d + 2 ) / d ) and j f ∈ L ∞ ( 0 , T ; L ( d + 2 ) / ( d + 1 ) ) j_f\in L^\infty(0,T;L^{(d+2)/(d+1)}) j f ∈ L ∞ ( 0 , T ; L ( d + 2 ) / ( d + 1 ) ) , while u ∈ L 2 ( 0 , T ; L 6 ) u\in L^2(0,T;L^6) u ∈ L 2 ( 0 , T ; L 6 ) by (E1); since d + 2 d ≥ 6 5 \frac{d+2}{d}\geq\frac{6}{5} d d + 2 ≥ 5 6 , F [ f , u ] = j f − ρ f u ∈ L 1 ( ( 0 , T ) × T d ) F[f,u]=j_f-\rho_fu\in L^1((0,T)\times\mathbb{T}^{d}) F [ f , u ] = j f − ρ f u ∈ L 1 (( 0 , T ) × T d ) , and f ∣ u ∣ f|u| f ∣ u ∣ is locally integrable.
The velocity bound
Lemma 11.2 (Velocity bound for energy solutions). Let d = 3 d=3 d = 3 , and let ( u , f ) (u,f) ( u , f ) be an energy solution on [ 0 , T ] [0,T] [ 0 , T ] , T ≤ 1 T\leq1 T ≤ 1 , with data ( u i n , f 0 ) (u_{\mathrm{in}},f_0) ( u in , f 0 ) . Then:
(a) u ∈ L 1 ( 0 , T ; L ∞ ) u\in L^1(0,T;L^\infty) u ∈ L 1 ( 0 , T ; L ∞ ) , with a bound depending only on T T T , E ( 0 ) \mathcal{E}(0) E ( 0 ) , ∥ f 0 ∥ L 1 \|f_0\|_{L^1} ∥ f 0 ∥ L 1 , ∥ f 0 ∥ L ∞ \|f_0\|_{L^\infty} ∥ f 0 ∥ L ∞ and R v 0 R_v^0 R v 0 ;
(b) 0 ≤ f ≤ e 3 ∥ f 0 ∥ L ∞ 0\leq f\leq e^3\|f_0\|_{L^\infty} 0 ≤ f ≤ e 3 ∥ f 0 ∥ L ∞ , and f ( t ) f(t) f ( t ) vanishes for ∣ v ∣ ≥ R v 0 + ∥ u ∥ L 1 ( 0 , T ; L ∞ ) |v|\geq R_v^0+\|u\|_{L^1(0,T;L^\infty)} ∣ v ∣ ≥ R v 0 + ∥ u ∥ L 1 ( 0 , T ; L ∞ ) ;
(c) ( u , f ) (u,f) ( u , f ) is a weak solution on [ 0 , T ] [0,T] [ 0 , T ] in the sense of Definition 2.1 ; f f f is the unique bounded (hence renormalized) solution of the kinetic equation with field u u u , it belongs to C ( [ 0 , T ] ; L 1 ) C([0,T];L^1) C ([ 0 , T ] ; L 1 ) , and ∬ f ( t ) d z = ∥ f 0 ∥ L 1 \iint f(t)\,dz=\|f_0\|_{L^1} ∬ f ( t ) d z = ∥ f 0 ∥ L 1 for every t t t .
Proof. Let F ∞ : = ∥ f ∥ L ∞ F_\infty:=\|f\|_{L^\infty} F ∞ := ∥ f ∥ L ∞ . Until Step 8, constants C C C depend only on T T T , E ( 0 ) \mathcal{E}(0) E ( 0 ) , ∥ f 0 ∥ L 1 \|f_0\|_{L^1} ∥ f 0 ∥ L 1 , R v 0 R_v^0 R v 0 and F ∞ F_\infty F ∞ . In Step 8 we show that F ∞ ≤ e 3 ∥ f 0 ∥ L ∞ F_\infty\leq e^3\|f_0\|_{L^\infty} F ∞ ≤ e 3 ∥ f 0 ∥ L ∞ , so that in the end they depend only on the quantities listed in (a).
By (E1) and (E4), ∥ u ∥ L ∞ L 2 + ∥ u ∥ L 2 H 1 ≤ C \|u\|_{L^\infty L^2}+\|u\|_{L^2H^1}\leq C ∥ u ∥ L ∞ L 2 + ∥ u ∥ L 2 H 1 ≤ C , hence ∥ u ∥ L 2 ( 0 , T ; L 6 ) ≤ C \|u\|_{L^2(0,T;L^6)}\leq C ∥ u ∥ L 2 ( 0 , T ; L 6 ) ≤ C , and M 2 ( f ( t ) ) ≤ 2 E ( 0 ) M_2(f(t))\leq2\mathcal{E}(0) M 2 ( f ( t )) ≤ 2 E ( 0 ) for a.e. t t t . By (11.1) with m = 2 m=2 m = 2 and ∣ T 3 ∣ = 1 |\mathbb{T}^3|=1 ∣ T 3 ∣ = 1 ,
∥ ρ f ∥ L ∞ ( 0 , T ; L 6 / 5 ) ≤ ∥ ρ f ∥ L ∞ ( 0 , T ; L 5 / 3 ) ≤ C . (57) \|\rho_f\|_{L^\infty(0,T;L^{6/5})}\leq\|\rho_f\|_{L^\infty(0,T;L^{5/3})}\leq C. \tag*{(57)} ∥ ρ f ∥ L ∞ ( 0 , T ; L 6/5 ) ≤ ∥ ρ f ∥ L ∞ ( 0 , T ; L 5/3 ) ≤ C . (57)
The main difficulty is the drag force: the pointwise bound ∣ F ∣ 2 ≤ ρ f ∫ ∣ v − u ∣ 2 f d v |F|^2\leq\rho_f\int|v-u|^2f\,dv ∣ F ∣ 2 ≤ ρ f ∫ ∣ v − u ∣ 2 f d v would require ρ f ∈ L ∞ \rho_f\in L^\infty ρ f ∈ L ∞ , that is, a bound on the velocity support, which is what we want to prove. We therefore propagate velocity moments, first with the energy alone and then with an improved integrability of u u u . This replaces [21 Lemma B.1] , where the bound F ∈ L l o c 2 L 2 F\in L^2_{\mathrm{loc}}L^2 F ∈ L loc 2 L 2 is obtained by referring to [17 Lemma 4.7] . The moments are propagated directly from the weak formulation, with truncated weights, so that no uniqueness or stability theorem for the kinetic equation with an unbounded velocity field is needed.
Step 1: test functions of v v v . Let Φ ∈ C 1 ( R 3 ) \Phi\in C^1(\mathbb{R}^3) Φ ∈ C 1 ( R 3 ) be bounded, with ∇ Φ \nabla\Phi ∇Φ compactly supported. We claim that N Φ ( t ) : = ∬ f ( t ) Φ ( v ) d z N_\Phi(t):=\iint f(t)\Phi(v)\,dz N Φ ( t ) := ∬ f ( t ) Φ ( v ) d z has an absolutely continuous representative on [ 0 , T ] [0,T] [ 0 , T ] with
N Φ ( 0 ) = ∬ f 0 Φ d z , N Φ ′ ( t ) = ∬ f ( t ) ( u − v ) ⋅ ∇ v Φ d z for a.e. t . (58) N_\Phi(0)=\iint f_0\Phi\,dz,\qquad N_\Phi'(t)=\iint f(t)(u-v)\cdot\nabla_v\Phi\,dz\quad\text{for a.e. }t. \tag*{(58)} N Φ ( 0 ) = ∬ f 0 Φ d z , N Φ ′ ( t ) = ∬ f ( t ) ( u − v ) ⋅ ∇ v Φ d z for a.e. t . (58)
By density, the identity of Definition 2.1 (iii) holds for test functions in C c ∞ ( [ 0 , T ) × T 3 × R 3 ) C^\infty_c([0,T)\times\mathbb{T}^3\times\mathbb{R}^3) C c ∞ ([ 0 , T ) × T 3 × R 3 ) . Let χ ∈ C c ∞ ( B 2 ) \chi\in C^\infty_c(B_2) χ ∈ C c ∞ ( B 2 ) with χ = 1 \chi=1 χ = 1 on B 1 B_1 B 1 , let χ L : = χ ( ⋅ / L ) \chi_L:=\chi(\cdot/L) χ L := χ ( ⋅ / L ) for L ≥ 1 L\geq1 L ≥ 1 , and let η ∈ C c ∞ ( [ 0 , T ) ) \eta\in C^\infty_c([0,T)) η ∈ C c ∞ ([ 0 , T )) . Testing with φ = η ( t ) Φ ( v ) χ L ( v ) \varphi=\eta(t)\Phi(v)\chi_L(v) φ = η ( t ) Φ ( v ) χ L ( v ) , which does not depend on x x x , gives
η ( 0 ) ∬ f 0 Φ χ L d z + ∫ 0 T ∬ f ( η ′ Φ χ L + η χ L ( u − v ) ⋅ ∇ v Φ + η Φ ( u − v ) ⋅ ∇ v χ L ) d z d t = 0. \eta(0)\iint f_0\Phi\chi_L\,\mathrm{d}z+\int_0^T\iint f\left(\eta'\Phi\chi_L+\eta\chi_L(u-v)\cdot\nabla_v\Phi+\eta\Phi(u-v)\cdot\nabla_v\chi_L\right)\,\mathrm{d}z\,\mathrm{d}t=0. η ( 0 ) ∬ f 0 Φ χ L d z + ∫ 0 T ∬ f ( η ′ Φ χ L + η χ L ( u − v ) ⋅ ∇ v Φ + η Φ ( u − v ) ⋅ ∇ v χ L ) d z d t = 0.
As L → ∞ L\to\infty L → ∞ , the first two terms converge by dominated convergence, since f ∈ L ∞ ( 0 , T ; L 1 ) f\in L^\infty(0,T;L^1) f ∈ L ∞ ( 0 , T ; L 1 ) , and χ L ∇ Φ = ∇ Φ \chi_L\nabla\Phi=\nabla\Phi χ L ∇Φ = ∇Φ for large L L L . Since ( ∣ u ∣ + ∣ v ∣ ) ∣ ∇ χ L ∣ ≤ C ( ∣ u ∣ L − 1 + 1 ) 1 { ∣ v ∣ ≥ L } (|u|+|v|)|\nabla\chi_L|\leq C(|u|L^{-1}+1)\mathbf{1}_{\{|v|\geq L\}} ( ∣ u ∣ + ∣ v ∣ ) ∣∇ χ L ∣ ≤ C ( ∣ u ∣ L − 1 + 1 ) 1 { ∣ v ∣ ≥ L } , the last term is bounded by
C ∥ η ∥ L ∞ ∥ Φ ∥ L ∞ ( L − 1 ∫ 0 T ∥ u ∥ L 6 ∥ ρ f ∥ L 6 / 5 d t + L − 2 ∫ 0 T M 2 ( f ( t ) ) d t ) ⟶ 0 , C\|\eta\|_{L^\infty}\|\Phi\|_{L^\infty}\left(L^{-1}\int_0^T\|u\|_{L^6}\|\rho_f\|_{L^{6/5}}\,\mathrm{d}t+L^{-2}\int_0^T M_2(f(t))\,\mathrm{d}t\right)\longrightarrow0, C ∥ η ∥ L ∞ ∥Φ ∥ L ∞ ( L − 1 ∫ 0 T ∥ u ∥ L 6 ∥ ρ f ∥ L 6/5 d t + L − 2 ∫ 0 T M 2 ( f ( t )) d t ) ⟶ 0 ,
by (57) . Hence, for every η ∈ C c ∞ ( [ 0 , T ) ) \eta\in C^\infty_c([0,T)) η ∈ C c ∞ ([ 0 , T )) ,
− ∫ 0 T η ′ N Φ d t = η ( 0 ) ∬ f 0 Φ d z + ∫ 0 T η G d t , G ( t ) : = ∬ f ( t ) ( u − v ) ⋅ ∇ v Φ d z , -\int_0^T\eta'N_\Phi\,\mathrm{d}t=\eta(0)\iint f_0\Phi\,\mathrm{d}z+\int_0^T\eta G\,\mathrm{d}t,\qquad G(t):=\iint f(t)(u-v)\cdot\nabla_v\Phi\,\mathrm{d}z, − ∫ 0 T η ′ N Φ d t = η ( 0 ) ∬ f 0 Φ d z + ∫ 0 T η G d t , G ( t ) := ∬ f ( t ) ( u − v ) ⋅ ∇ v Φ d z ,
and G ∈ L 1 ( 0 , T ) G\in L^1(0,T) G ∈ L 1 ( 0 , T ) because ∇ Φ \nabla\Phi ∇Φ is bounded with compact support. This is (58) . With Φ ≡ 1 \Phi\equiv1 Φ ≡ 1 we obtain ∬ f ( t ) d z = ∥ f 0 ∥ L 1 \iint f(t)\,\mathrm{d}z=\|f_0\|_{L^1} ∬ f ( t ) d z = ∥ f 0 ∥ L 1 for a.e. t t t .
Step 2: truncated moments. Let k ∈ { 3 , 6 } k\in\{3,6\} k ∈ { 3 , 6 } and p : = k + 3 p:=k+3 p := k + 3 . Fix a nondecreasing θ ∈ C ∞ ( [ 0 , ∞ ) ) \theta\in C^\infty([0,\infty)) θ ∈ C ∞ ([ 0 , ∞ )) with θ ( s ) = s k \theta(s)=s^k θ ( s ) = s k for s ≤ 2 s\leq2 s ≤ 2 and θ \theta θ constant on [ 3 , ∞ ) [3,\infty) [ 3 , ∞ ) , and let Φ L ( v ) : = L k θ ( ∣ v ∣ / L ) \Phi_L(v):=L^k\theta(|v|/L) Φ L ( v ) := L k θ ( ∣ v ∣/ L ) for L ≥ R v 0 L\geq R_v^0 L ≥ R v 0 . Then Φ L ∈ C 1 ( R 3 ) \Phi_L\in C^1(\mathbb{R}^3) Φ L ∈ C 1 ( R 3 ) is bounded and radially nondecreasing, Φ L ( v ) = ∣ v ∣ k \Phi_L(v)=|v|^k Φ L ( v ) = ∣ v ∣ k for ∣ v ∣ ≤ 2 L |v|\leq2L ∣ v ∣ ≤ 2 L , and, with constants depending only on k k k and θ \theta θ ,
c k ∣ v ∣ k 1 { ∣ v ∣ ≤ 3 L } ≤ Φ L ( v ) ≤ C k ∣ v ∣ k , ∣ ∇ Φ L ( v ) ∣ ≤ C k ∣ v ∣ k − 1 1 { ∣ v ∣ ≤ 3 L } . c_k|v|^k\mathbf{1}_{\{|v|\leq3L\}}\leq\Phi_L(v)\leq C_k|v|^k,\qquad|\nabla\Phi_L(v)|\leq C_k|v|^{k-1}\mathbf{1}_{\{|v|\leq3L\}}. c k ∣ v ∣ k 1 { ∣ v ∣ ≤ 3 L } ≤ Φ L ( v ) ≤ C k ∣ v ∣ k , ∣∇ Φ L ( v ) ∣ ≤ C k ∣ v ∣ k − 1 1 { ∣ v ∣ ≤ 3 L } .
Since ∇ Φ L ( v ) \nabla\Phi_L(v) ∇ Φ L ( v ) is a nonnegative multiple of v v v , we have ( u − v ) ⋅ ∇ v Φ L ≤ ∣ u ∣ ∣ ∇ Φ L ∣ (u-v)\cdot\nabla_v\Phi_L\leq|u||\nabla\Phi_L| ( u − v ) ⋅ ∇ v Φ L ≤ ∣ u ∣∣∇ Φ L ∣ . By Step 1, N L ( t ) : = ∬ f ( t ) Φ L d z N_L(t):=\iint f(t)\Phi_L\,\mathrm{d}z N L ( t ) := ∬ f ( t ) Φ L d z is absolutely continuous and
N L ′ ≤ C k ∫ T 3 ∣ u ∣ m L d x , m L ( t , x ) : = ∫ { ∣ v ∣ ≤ 3 L } f ∣ v ∣ k − 1 d v . N'_L\leq C_k\int_{\mathbb{T}^3}|u|m^L\,\mathrm{d}x,\qquad m^L(t,x):=\int_{\{|v|\leq3L\}}f|v|^{k-1}\,\mathrm{d}v. N L ′ ≤ C k ∫ T 3 ∣ u ∣ m L d x , m L ( t , x ) := ∫ { ∣ v ∣ ≤ 3 L } f ∣ v ∣ k − 1 d v .
Splitting the last integral at ∣ v ∣ = r |v|=r ∣ v ∣ = r and using ∣ v ∣ k 1 { ∣ v ∣ ≤ 3 L } ≤ c k − 1 Φ L |v|^k\mathbf{1}_{\{|v|\leq3L\}}\leq c_k^{-1}\Phi_L ∣ v ∣ k 1 { ∣ v ∣ ≤ 3 L } ≤ c k − 1 Φ L , we get m L ≤ C ( r k + 2 + r − 1 ∫ f Φ L d v ) m^L\leq C(r^{k+2}+r^{-1}\int f\Phi_L\,\mathrm{d}v) m L ≤ C ( r k + 2 + r − 1 ∫ f Φ L d v ) for every r > 0 r>0 r > 0 , and after optimizing in r r r ,
m L ≤ C F ∞ 1 / p ( ∫ f Φ L d v ) ( p − 1 ) / p , hence ∥ m L ( t ) ∥ L p / ( p − 1 ) ( T 3 ) ≤ C F ∞ 1 / p N L ( t ) ( p − 1 ) / p . m^L\leq CF_\infty^{1/p}\left(\int f\Phi_L\,\mathrm{d}v\right)^{(p-1)/p},\qquad\text{hence}\qquad\|m^L(t)\|_{L^{p/(p-1)}(\mathbb{T}^3)}\leq CF_\infty^{1/p}N_L(t)^{(p-1)/p}. m L ≤ C F ∞ 1/ p ( ∫ f Φ L d v ) ( p − 1 ) / p , hence ∥ m L ( t ) ∥ L p / ( p − 1 ) ( T 3 ) ≤ C F ∞ 1/ p N L ( t ) ( p − 1 ) / p .
By Hölder’s inequality, N L ′ ≤ C ∥ u ( t ) ∥ L p N L 1 − 1 / p N'_L\leq C\|u(t)\|_{L^p}N_L^{1-1/p} N L ′ ≤ C ∥ u ( t ) ∥ L p N L 1 − 1/ p for a.e. t t t . Thus, for every ϵ > 0 \epsilon>0 ϵ > 0 , the function ( N L + ϵ ) 1 / p (N_L+\epsilon)^{1/p} ( N L + ϵ ) 1/ p is absolutely continuous with derivative at most C ∥ u ( t ) ∥ L p C\|u(t)\|_{L^p} C ∥ u ( t ) ∥ L p . Since N L ( 0 ) ≤ C k M k ( f 0 ) ≤ C k R v k ∥ f 0 ∥ L 1 N_L(0)\leq C_kM_k(f_0)\leq C_kR_v^k\|f_0\|_{L^1} N L ( 0 ) ≤ C k M k ( f 0 ) ≤ C k R v k ∥ f 0 ∥ L 1 , we get N L ( t ) 1 / p ≤ C ( 1 + ∥ u ∥ L 1 ( 0 , T ; L p ) ) N_L(t)^{1/p}\leq C(1+\|u\|_{L^1(0,T;L^p)}) N L ( t ) 1/ p ≤ C ( 1 + ∥ u ∥ L 1 ( 0 , T ; L p ) ) . Since Φ L ≥ 0 \Phi_L\geq0 Φ L ≥ 0 and Φ L → ∣ v ∣ k \Phi_L\to|v|^k Φ L → ∣ v ∣ k pointwise as L → ∞ L\to\infty L → ∞ , Fatou’s lemma, applied along the integers outside a common null set of times, gives
M k ( f ( t ) ) ≤ C ( 1 + ∥ u ∥ L 1 ( 0 , T ; L k + 3 ) ) k + 3 for a.e. t , k ∈ { 3 , 6 } , (59) M_k(f(t))\leq C(1+\|u\|_{L^1(0,T;L^{k+3})})^{k+3}\qquad\text{for a.e. }t,\quad k\in\{3,6\}, \tag*{(59)} M k ( f ( t )) ≤ C ( 1 + ∥ u ∥ L 1 ( 0 , T ; L k + 3 ) ) k + 3 for a.e. t , k ∈ { 3 , 6 } , (59)
whenever the right-hand side is finite.
Step 3: the third moment. Since ∥ u ∥ L 1 ( 0 , T ; L 6 ) ≤ C \|u\|_{L^1(0,T;L^6)}\leq C ∥ u ∥ L 1 ( 0 , T ; L 6 ) ≤ C , (59) with k = 3 k=3 k = 3 gives M 3 ( f ( t ) ) ≤ C M_3(f(t))\leq C M 3 ( f ( t )) ≤ C for a.e. t t t . By (56) with m = 3 m=3 m = 3 and k = 0 , 1 k=0,1 k = 0 , 1 ,
∥ ρ f ∥ L ∞ ( 0 , T ; L 2 ) + ∥ j f ∥ L ∞ ( 0 , T ; L 3 / 2 ) ≤ C , hence ∥ F ∥ L 2 ( 0 , T ; L 3 / 2 ) ≤ C + ∥ ρ f ∥ L ∞ L 2 ∥ u ∥ L 2 L 6 ≤ C , \|\rho_f\|_{L^\infty(0,T;L^2)}+\|j_f\|_{L^\infty(0,T;L^{3/2})}\leq C,\qquad\text{hence}\qquad\|F\|_{L^2(0,T;L^{3/2})}\leq C+\|\rho_f\|_{L^\infty L^2}\|u\|_{L^2L^6}\leq C, ∥ ρ f ∥ L ∞ ( 0 , T ; L 2 ) + ∥ j f ∥ L ∞ ( 0 , T ; L 3/2 ) ≤ C , hence ∥ F ∥ L 2 ( 0 , T ; L 3/2 ) ≤ C + ∥ ρ f ∥ L ∞ L 2 ∥ u ∥ L 2 L 6 ≤ C ,
where F : = F [ f , u ] F:=F[f,u] F := F [ f , u ] .
Step 4: an L 3 L^3 L 3 bound for u u u . The mean P 0 u P_0u P 0 u is bounded by ∥ u ∥ L ∞ L 2 \|u\|_{L^\infty L^2} ∥ u ∥ L ∞ L 2 . The mean-free part satisfies the heat equation with forcing Q ( F − div ( u ⊗ u ) ) Q(F-\operatorname{div}(u\otimes u)) Q ( F − div ( u ⊗ u )) in the weak sense of Definition 2.1 (iv), with u ∈ L ∞ L 2 ∩ L 2 H 1 u\in L^\infty L^2\cap L^2 H^1 u ∈ L ∞ L 2 ∩ L 2 H 1 . Comparing Fourier coefficients as in the Decomposition step of the proof of Lemma 11.5, with a range over an orthonormal basis of n ⊥ n^\perp n ⊥ , we obtain the Duhamel formula for a.e. t ∈ [ 0 , T ) t\in[0,T) t ∈ [ 0 , T ) (the integral B ( t ) B(t) B ( t ) below converges for a.e. t t t , which suffices for the L t 1 L^1_t L t 1 bounds that follow):
( 1 − P 0 ) u ( t ) = e t Δ ( 1 − P 0 ) u i n + ∫ 0 t e ( t − s ) Δ ( 1 − P 0 ) Q F d s − ∫ 0 t e ( t − s ) Δ ( 1 − P 0 ) Q div ( u ⊗ u ) d s = : A + D + B . (60) (1-P_0)u(t)=e^{t\Delta}(1-P_0)u_{\mathrm{in}}+\int_0^t e^{(t-s)\Delta}(1-P_0)QF\,\mathrm{d}s-\int_0^t e^{(t-s)\Delta}(1-P_0)Q\operatorname{div}(u\otimes u)\,\mathrm{d}s=:A+D+B. \tag*{(60)} ( 1 − P 0 ) u ( t ) = e t Δ ( 1 − P 0 ) u in + ∫ 0 t e ( t − s ) Δ ( 1 − P 0 ) QF d s − ∫ 0 t e ( t − s ) Δ ( 1 − P 0 ) Q div ( u ⊗ u ) d s =: A + D + B . (60)
On mean-free functions on T 3 \mathbb{T}^{3} T 3 , for 0 < τ ≤ 1 0<\tau\leq1 0 < τ ≤ 1 , 1 < p < ∞ 1<p<\infty 1 < p < ∞ and p ≤ s ≤ ∞ p\leq s\leq\infty p ≤ s ≤ ∞ , the heat semigroup satisfies
∥ e τ Δ Q g ∥ L s ≤ C τ − 3 2 ( 1 p − 1 s ) ∥ g ∥ L p , ∥ e τ Δ Q div g ∥ L s ≤ C τ − 1 2 − 3 2 ( 1 p − 1 s ) ∥ g ∥ L p , \|e^{\tau\Delta}Qg\|_{L^{s}}\leq C\tau^{-\frac{3}{2}\left(\frac{1}{p}-\frac{1}{s}\right)}\|g\|_{L^{p}},\qquad\|e^{\tau\Delta}Q\operatorname{div}g\|_{L^{s}}\leq C\tau^{-\frac{1}{2}-\frac{3}{2}\left(\frac{1}{p}-\frac{1}{s}\right)}\|g\|_{L^{p}}, ∥ e τ Δ Q g ∥ L s ≤ C τ − 2 3 ( p 1 − s 1 ) ∥ g ∥ L p , ∥ e τ Δ Q div g ∥ L s ≤ C τ − 2 1 − 2 3 ( p 1 − s 1 ) ∥ g ∥ L p ,
and the first bound also holds for p = 2 p=2 p = 2 without Q Q Q . The restriction p < ∞ p<\infty p < ∞ is needed: Q Q Q is not bounded on L ∞ L^{\infty} L ∞ , so the first bound fails for p = s = ∞ p=s=\infty p = s = ∞ as τ → 0 \tau\to0 τ → 0 . All applications below use finite p p p . With s = 9 s=9 s = 9 , and with p = 2 p=2 p = 2 , p = 3 2 p=\frac{3}{2} p = 2 3 and p = 3 p=3 p = 3 respectively,
∥ A ( t ) ∥ L 9 ≤ C t − 7 / 12 ∥ u i n ∥ L 2 , \|A(t)\|_{L^{9}}\leq Ct^{-7/12}\|u_{\mathrm{in}}\|_{L^{2}}, ∥ A ( t ) ∥ L 9 ≤ C t − 7/12 ∥ u in ∥ L 2 ,
∥ D ( t ) ∥ L 9 ≤ C ∫ 0 t ( t − s ) − 5 / 6 ∥ F ( s ) ∥ L 3 / 2 d s , ∥ B ( t ) ∥ L 9 ≤ C ∫ 0 t ( t − s ) − 5 / 6 ∥ u ( s ) ∥ L 6 2 d s . \|D(t)\|_{L^{9}}\leq C\int_{0}^{t}(t-s)^{-5/6}\|F(s)\|_{L^{3/2}}\,ds,\qquad\|B(t)\|_{L^{9}}\leq C\int_{0}^{t}(t-s)^{-5/6}\|u(s)\|_{L^{6}}^{2}\,ds. ∥ D ( t ) ∥ L 9 ≤ C ∫ 0 t ( t − s ) − 5/6 ∥ F ( s ) ∥ L 3/2 d s , ∥ B ( t ) ∥ L 9 ≤ C ∫ 0 t ( t − s ) − 5/6 ∥ u ( s ) ∥ L 6 2 d s .
The kernels are integrable, so Young’s inequality in time and Step 3 give ∥ u ∥ L 1 ( 0 , T ; L 9 ) ≤ C \|u\|_{L^{1}(0,T;L^{9})}\leq C ∥ u ∥ L 1 ( 0 , T ; L 9 ) ≤ C .
Step 5: the sixth moment and the drag force. By Step 4 and (11.4) with $k=6,
M 6 ( f ( t ) ) ≤ C for a.e. t ∈ [ 0 , T ] . (61) M_{6}(f(t))\leq C\qquad\text{for a.e. }t\in[0,T]. \tag*{(61)} M 6 ( f ( t )) ≤ C for a.e. t ∈ [ 0 , T ] . (61)
By (11.1) with m = 6 m=6 m = 6 and k = 0 , 1 k=0,1 k = 0 , 1 ,
∥ ρ f ( t ) ∥ L 3 ≤ C F ∞ 2 / 3 M 6 ( f ( t ) ) 1 / 3 , ∥ j f ( t ) ∥ L 9 / 4 ≤ C F ∞ 5 / 9 M 6 ( f ( t ) ) 4 / 9 . \|\rho_{f}(t)\|_{L^{3}}\leq CF_{\infty}^{2/3}M_{6}(f(t))^{1/3},\qquad\|j_{f}(t)\|_{L^{9/4}}\leq CF_{\infty}^{5/9}M_{6}(f(t))^{4/9}. ∥ ρ f ( t ) ∥ L 3 ≤ C F ∞ 2/3 M 6 ( f ( t ) ) 1/3 , ∥ j f ( t ) ∥ L 9/4 ≤ C F ∞ 5/9 M 6 ( f ( t ) ) 4/9 .
Since ∣ T 3 ∣ = 1 |\mathbb{T}^{3}|=1 ∣ T 3 ∣ = 1 and H 1 ( T 3 ) ↪ L 6 H^{1}(\mathbb{T}^{3})\hookrightarrow L^{6} H 1 ( T 3 ) ↪ L 6 ,
∥ F ( t ) ∥ L 2 ≤ ∥ j f ( t ) ∥ L 9 / 4 + ∥ ρ f ( t ) ∥ L 3 ∥ u ( t ) ∥ L 6 ≤ C ( 1 + ∥ u ( t ) ∥ H 1 ) , \|F(t)\|_{L^{2}}\leq\|j_{f}(t)\|_{L^{9/4}}+\|\rho_{f}(t)\|_{L^{3}}\|u(t)\|_{L^{6}}\leq C(1+\|u(t)\|_{H^{1}}), ∥ F ( t ) ∥ L 2 ≤ ∥ j f ( t ) ∥ L 9/4 + ∥ ρ f ( t ) ∥ L 3 ∥ u ( t ) ∥ L 6 ≤ C ( 1 + ∥ u ( t ) ∥ H 1 ) ,
and F ∈ L 2 ( 0 , T ; L 2 ) F\in L^{2}(0,T;L^{2}) F ∈ L 2 ( 0 , T ; L 2 ) follows from u ∈ L 2 ( 0 , T ; H 1 ) u\in L^{2}(0,T;H^{1}) u ∈ L 2 ( 0 , T ; H 1 ) .
Step 6: the linear terms in L 1 L ∞ L^{1}L^{\infty} L 1 L ∞ . We return to (11.5). By the bounds of Step 4 with s = ∞ s=\infty s = ∞ and p = 2 p=2 p = 2 ,
∥ A ( t ) ∥ L ∞ ≤ C t − 3 / 4 ∥ u i n ∥ L 2 , ∫ 0 T ∥ D ∥ L ∞ ≤ C ∫ 0 T ∫ 0 t ( t − s ) − 3 / 4 ∥ F ( s ) ∥ L 2 d s d t ≤ C T 1 / 4 ∥ F ∥ L 1 L 2 . \|A(t)\|_{L^{\infty}}\leq Ct^{-3/4}\|u_{\mathrm{in}}\|_{L^{2}},\qquad\int_{0}^{T}\|D\|_{L^{\infty}}\leq C\int_{0}^{T}\int_{0}^{t}(t-s)^{-3/4}\|F(s)\|_{L^{2}}\,ds\,dt\leq CT^{1/4}\|F\|_{L^{1}L^{2}}. ∥ A ( t ) ∥ L ∞ ≤ C t − 3/4 ∥ u in ∥ L 2 , ∫ 0 T ∥ D ∥ L ∞ ≤ C ∫ 0 T ∫ 0 t ( t − s ) − 3/4 ∥ F ( s ) ∥ L 2 d s d t ≤ C T 1/4 ∥ F ∥ L 1 L 2 .
Step 7: the bilinear term in L 1 L ∞ L^{1}L^{\infty} L 1 L ∞ . This replaces the Besov lemma of Chemin used in [21 ] . We first claim that, for a , b ∈ H 1 ( T 3 ) a,b\in H^{1}(\mathbb{T}^{3}) a , b ∈ H 1 ( T 3 ) ,
∑ j ≥ 0 2 j / 2 ∥ Δ j ( a b ) ∥ L 2 ≤ C ( ∥ ∇ a ∥ L 2 + ∣ P 0 a ∣ ) ( ∥ ∇ b ∥ L 2 + ∣ P 0 b ∣ ) . (62) \sum_{j\geq0}2^{j/2}\|\Delta_{j}(ab)\|_{L^{2}}\leq C(\|\nabla a\|_{L^{2}}+|P_{0}a|)(\|\nabla b\|_{L^{2}}+|P_{0}b|). \tag*{(62)} j ≥ 0 ∑ 2 j /2 ∥ Δ j ( ab ) ∥ L 2 ≤ C ( ∥∇ a ∥ L 2 + ∣ P 0 a ∣ ) ( ∥∇ b ∥ L 2 + ∣ P 0 b ∣ ) . (62)
Write a = P 0 a + a ˚ a=P_{0}a+\mathring{a} a = P 0 a + a ˚ . Products involving P 0 a P_{0}a P 0 a or P 0 b P_{0}b P 0 b are bounded by ∣ P 0 a ∣ ∑ j 2 j / 2 ∥ Δ j b ˚ ∥ L 2 ≤ C ∣ P 0 a ∣ ∥ ∇ b ∥ L 2 |P_{0}a|\sum_{j}2^{j/2}\|\Delta_{j}\mathring{b}\|_{L^{2}}\leq C|P_{0}a|\|\nabla b\|_{L^{2}} ∣ P 0 a ∣ ∑ j 2 j /2 ∥ Δ j b ˚ ∥ L 2 ≤ C ∣ P 0 a ∣∥∇ b ∥ L 2 , using 2 j / 2 = 2 − j / 2 2 j 2^{j/2}=2^{-j/2}2^{j} 2 j /2 = 2 − j /2 2 j and Cauchy–Schwarz. For a ˚ b ˚ \mathring{a}\mathring{b} a ˚ b ˚ , use Bony’s decomposition and set a k = 2 k ∥ Δ k a ˚ ∥ L 2 a_{k}=2^{k}\|\Delta_{k}\mathring{a}\|_{L^{2}} a k = 2 k ∥ Δ k a ˚ ∥ L 2 , b k = 2 k ∥ Δ k b ˚ ∥ L 2 b_{k}=2^{k}\|\Delta_{k}\mathring{b}\|_{L^{2}} b k = 2 k ∥ Δ k b ˚ ∥ L 2 , so that ∥ ( a k ) ∥ ℓ 2 ≃ ∥ ∇ a ∥ L 2 \|(a_{k})\|_{\ell^{2}}\simeq\|\nabla a\|_{L^{2}} ∥ ( a k ) ∥ ℓ 2 ≃ ∥∇ a ∥ L 2 .
∑ j 2 j / 2 ∥ Δ j ( T a ˚ b ˚ ) ∥ L 2 ≤ C ∑ k b k ∑ k ′ < k 2 ( k ′ − k ) / 2 a k ′ ≤ C ∥ ( a k ) ∥ ℓ 2 ∥ ( b k ) ∥ ℓ 2 , \sum_{j}2^{j/2}\|\Delta_{j}(T_{\mathring{a}}\mathring{b})\|_{L^{2}}\leq C\sum_{k}b_{k}\sum_{k'<k}2^{(k'-k)/2}a_{k'}\leq C\|(a_{k})\|_{\ell^{2}}\|(b_{k})\|_{\ell^{2}}, j ∑ 2 j /2 ∥ Δ j ( T a ˚ b ˚ ) ∥ L 2 ≤ C k ∑ b k k ′ < k ∑ 2 ( k ′ − k ) /2 a k ′ ≤ C ∥ ( a k ) ∥ ℓ 2 ∥ ( b k ) ∥ ℓ 2 ,
by Young’s inequality for sequences; here we used that Δ j ( S k − 1 a ˚ Δ k b ˚ ) = 0 \Delta_{j}(S_{k-1}\mathring{a}\,\Delta_{k}\mathring{b})=0 Δ j ( S k − 1 a ˚ Δ k b ˚ ) = 0 unless ∣ j − k ∣ ≤ 2 |j-k|\leq2 ∣ j − k ∣ ≤ 2 . The paraproduct T b ˚ a ˚ T_{\mathring{b}}\mathring{a} T b ˚ a ˚ is bounded in the same way.
2 j / 2 ∥ Δ j ( Δ k a ˚ Δ k b ˚ ~ ) ∥ L 2 ≤ C 2 2 j ∥ Δ k a ˚ ∥ L 2 ∥ Δ k b ˚ ~ ∥ L 2 = C 2 2 ( j − k ) a k b ~ k . 2^{j/2}\|\Delta_{j}(\Delta_{k}\mathring{a}\,\widetilde{\Delta_{k}\mathring{b}})\|_{L^{2}}\leq C2^{2j}\|\Delta_{k}\mathring{a}\|_{L^{2}}\|\widetilde{\Delta_{k}\mathring{b}}\|_{L^{2}}=C2^{2(j-k)}a_{k}\widetilde{b}_{k}. 2 j /2 ∥ Δ j ( Δ k a ˚ Δ k b ˚ ) ∥ L 2 ≤ C 2 2 j ∥ Δ k a ˚ ∥ L 2 ∥ Δ k b ˚ ∥ L 2 = C 2 2 ( j − k ) a k b k .
Summing over j ≤ k + 3 j\leq k+3 j ≤ k + 3 , then over k k k , gives C ∑ k a k b ~ k C\sum_{k}a_{k}\widetilde{b}_{k} C ∑ k a k b k .
This proves (11.7). For j ≥ 0 j\geq0 j ≥ 0 , ∥ Δ j e τ Δ g ∥ L 2 ≤ e − c τ 4 j ∥ Δ j g ∥ L 2 \|\Delta_{j}e^{\tau\Delta}g\|_{L^{2}}\leq e^{-c\tau4^{j}}\|\Delta_{j}g\|_{L^{2}} ∥ Δ j e τ Δ g ∥ L 2 ≤ e − c τ 4 j ∥ Δ j g ∥ L 2 and ∥ Δ j Q div g ∥ L 2 ≤ C 2 j ∥ Δ j g ∥ L 2 \|\Delta_{j}Q\operatorname{div}g\|_{L^{2}}\leq C2^{j}\|\Delta_{j}g\|_{L^{2}} ∥ Δ j Q div g ∥ L 2 ≤ C 2 j ∥ Δ j g ∥ L 2 . Hence
∫ 0 T ∥ Δ j B ( t ) ∥ L 2 d t ≤ C 2 j ∫ 0 T ∫ 0 t e − c ( t − s ) 4 j ∥ Δ j ( u ⊗ u ) ( s ) ∥ L 2 d s d t ≤ C 2 − j ∫ 0 T ∥ Δ j ( u ⊗ u ) ∥ L 2 d s . \int_{0}^{T}\|\Delta_{j}B(t)\|_{L^{2}}\,dt\leq C2^{j}\int_{0}^{T}\int_{0}^{t}e^{-c(t-s)4^{j}}\|\Delta_{j}(u\otimes u)(s)\|_{L^{2}}\,ds\,dt\leq C2^{-j}\int_{0}^{T}\|\Delta_{j}(u\otimes u)\|_{L^{2}}\,ds. ∫ 0 T ∥ Δ j B ( t ) ∥ L 2 d t ≤ C 2 j ∫ 0 T ∫ 0 t e − c ( t − s ) 4 j ∥ Δ j ( u ⊗ u ) ( s ) ∥ L 2 d s d t ≤ C 2 − j ∫ 0 T ∥ Δ j ( u ⊗ u ) ∥ L 2 d s .
By Bernstein, ∥ Δ j B ∥ L ∞ ≤ C 2 3 j / 2 ∥ Δ j B ∥ L 2 \|\Delta_j B\|_{L^\infty} \le C 2^{3j/2}\|\Delta_j B\|_{L^2} ∥ Δ j B ∥ L ∞ ≤ C 2 3 j /2 ∥ Δ j B ∥ L 2 . Summing over j j j and using (62) ,
∫ 0 T ∥ B ∥ L ∞ ≤ C ∫ 0 T ∑ j 2 j / 2 ∥ Δ j ( u ⊗ u ) ∥ L 2 ≤ C ∫ 0 T ( ∥ ∇ u ∥ L 2 + ∥ u ∥ L 2 ) 2 < ∞ . \int_0^T \|B\|_{L^\infty} \le C\int_0^T \sum_j 2^{j/2}\|\Delta_j(u\otimes u)\|_{L^2} \le C\int_0^T \left(\|\nabla u\|_{L^2}+\|u\|_{L^2}\right)^2<\infty. ∫ 0 T ∥ B ∥ L ∞ ≤ C ∫ 0 T j ∑ 2 j /2 ∥ Δ j ( u ⊗ u ) ∥ L 2 ≤ C ∫ 0 T ( ∥∇ u ∥ L 2 + ∥ u ∥ L 2 ) 2 < ∞.
Together with ∣ P 0 u ∣ ≤ ∥ u ∥ L ∞ L 2 |P_0u|\le\|u\|_{L^\infty L^2} ∣ P 0 u ∣ ≤ ∥ u ∥ L ∞ L 2 and Step 6, this gives ∥ u ∥ L 1 ( 0 , T ; L ∞ ) ≤ C \|u\|_{L^1(0,T;L^\infty)}\le C ∥ u ∥ L 1 ( 0 , T ; L ∞ ) ≤ C .
Step 8: velocity support, the bound on f f f , and uniqueness. By Step 7 and (E1), u ∈ L 1 ( 0 , T ; L ∞ ) ∩ L 1 ( 0 , T ; W 1 , 1 ) u\in L^1(0,T;L^\infty)\cap L^1(0,T;W^{1,1}) u ∈ L 1 ( 0 , T ; L ∞ ) ∩ L 1 ( 0 , T ; W 1 , 1 ) . Extend u u u by zero outside [ 0 , T ] [0,T] [ 0 , T ] and let u ε u_\varepsilon u ε be its mollification in ( t , x ) (t,x) ( t , x ) . Then u ε u_\varepsilon u ε is smooth, ∥ u ε ∥ L 1 ( 0 , T ; L ∞ ) ≤ ∥ u ∥ L 1 ( 0 , T ; L ∞ ) \|u_\varepsilon\|_{L^1(0,T;L^\infty)}\le\|u\|_{L^1(0,T;L^\infty)} ∥ u ε ∥ L 1 ( 0 , T ; L ∞ ) ≤ ∥ u ∥ L 1 ( 0 , T ; L ∞ ) by Minkowski’s inequality, and u ε → u u_\varepsilon\to u u ε → u in L 1 ( ( 0 , T ) × T 3 ) L^1((0,T)\times\mathbb{T}^3) L 1 (( 0 , T ) × T 3 ) . Let f ε : = f [ u ε ] f_\varepsilon:=f[u_\varepsilon] f ε := f [ u ε ] be the push-forward of f 0 f_0 f 0 along the flow of ( v , u ε − v ) (v,u_\varepsilon-v) ( v , u ε − v ) . By Lemma 8.1(a), f ε f_\varepsilon f ε is a weak solution of the kinetic equation with field u ε u_\varepsilon u ε and datum f 0 f_0 f 0 ,
0 ≤ f ε ≤ e 3 ∥ f 0 ∥ L ∞ , f ε = 0 for ∣ v ∣ ≥ R 0 v + ∥ u ∥ L 1 ( 0 , T ; L ∞ ) . 0\le f_\varepsilon\le e^3\|f_0\|_{L^\infty},\qquad f_\varepsilon=0\quad\text{for }|v|\ge R_0^v+\|u\|_{L^1(0,T;L^\infty)}. 0 ≤ f ε ≤ e 3 ∥ f 0 ∥ L ∞ , f ε = 0 for ∣ v ∣ ≥ R 0 v + ∥ u ∥ L 1 ( 0 , T ; L ∞ ) .
Along a subsequence, f ε ⇀ ∗ g f_\varepsilon\rightharpoonup^*g f ε ⇀ ∗ g in L ∞ ( ( 0 , T ) × T 3 × R 3 ) L^\infty((0,T)\times\mathbb{T}^3\times\mathbb{R}^3) L ∞ (( 0 , T ) × T 3 × R 3 ) , and g g g satisfies the same two bounds. For a test function φ \varphi φ ,
∣ ∭ f ε ( u ε − u ) ⋅ ∇ v φ ∣ ≤ e 3 ∥ f 0 ∥ L ∞ C φ ∥ u ε − u ∥ L 1 ( ( 0 , T ) × T 3 ) → 0 , \left|\iiint f_\varepsilon(u_\varepsilon-u)\cdot\nabla_v\varphi\right|\le e^3\|f_0\|_{L^\infty}C_\varphi\|u_\varepsilon-u\|_{L^1((0,T)\times\mathbb{T}^3)}\to0, ∭ f ε ( u ε − u ) ⋅ ∇ v φ ≤ e 3 ∥ f 0 ∥ L ∞ C φ ∥ u ε − u ∥ L 1 (( 0 , T ) × T 3 ) → 0 ,
while the other terms of the weak formulation pass to the limit because their multipliers are integrable. Hence g g g is a bounded weak solution of the kinetic equation with field u u u and datum f 0 f_0 f 0 , and so is f f f , by (E2)–(E3). By Lemma 3.6, applied on [ 0 , T ] [0,T] [ 0 , T ] , f = g f=g f = g . This proves (b), in particular F ∞ ≤ e 3 ∥ f 0 ∥ L ∞ F_\infty\le e^3\|f_0\|_{L^\infty} F ∞ ≤ e 3 ∥ f 0 ∥ L ∞ , and it shows that ( u , f ) (u,f) ( u , f ) satisfies Definition 2.1(i)–(ii). The remaining statements of (c) follow from Lemma 3.6 and from the conservation of mass of Step 1, which extends to every t t t by the continuity of f f f in L 1 L^1 L 1 .
Weak–strong uniqueness
The next lemma compares an energy solution with a strong solution whose velocity is Lipschitz in space. It is a special case of the situation treated in [21 Theorem 1.2] , where the strong solution may be much rougher. In the Lipschitz case the Lagrangian argument of § 5.5 applies directly, on the torus, and we give the proof.
Lemma 11.3 (Weak–strong uniqueness). Let d ∈ { 2 , 3 } d\in\{2,3\} d ∈ { 2 , 3 } and 0 ≤ t 0 < T ≤ 1 0\le t_0<T\le1 0 ≤ t 0 < T ≤ 1 . Let u ∗ ∈ L 2 ( T d ) u_*\in L^2(\mathbb{T}^d) u ∗ ∈ L 2 ( T d ) be divergence-free, and let f ∗ ∈ L ∞ ( T d × R d ) f_*\in L^\infty(\mathbb{T}^d\times\mathbb{R}^d) f ∗ ∈ L ∞ ( T d × R d ) be nonnegative and vanish for ∣ v ∣ ≥ R ∗ v |v|\ge R_*^v ∣ v ∣ ≥ R ∗ v . Let ( u 1 , f 1 ) (u_1,f_1) ( u 1 , f 1 ) and ( u 2 , f 2 ) (u_2,f_2) ( u 2 , f 2 ) be weak solutions of (2.4) on [ t 0 , T ] [t_0,T] [ t 0 , T ] with data ( u ∗ , f ∗ ) (u_*,f_*) ( u ∗ , f ∗ ) at time t 0 t_0 t 0 , in the sense of Definition 2.1 translated in time. Assume that:
(a) u 1 ∈ L ∞ ( t 0 , T ; H 1 ) ∩ L 2 ( t 0 , T ; H 2 ) u_1\in L^\infty(t_0,T;H^1)\cap L^2(t_0,T;H^2) u 1 ∈ L ∞ ( t 0 , T ; H 1 ) ∩ L 2 ( t 0 , T ; H 2 ) , ∂ t u 1 ∈ L 2 ( t 0 , T ; L 2 ) \partial_tu_1\in L^2(t_0,T;L^2) ∂ t u 1 ∈ L 2 ( t 0 , T ; L 2 ) and ∇ u 1 ∈ L 1 ( t 0 , T ; L ∞ ) \nabla u_1\in L^1(t_0,T;L^\infty) ∇ u 1 ∈ L 1 ( t 0 , T ; L ∞ ) ;
(b) u 2 ∈ C w ( [ t 0 , T ] ; L 2 ) ∩ L 2 ( t 0 , T ; H 1 ) u_2\in C_w([t_0,T];L^2)\cap L^2(t_0,T;H^1) u 2 ∈ C w ([ t 0 , T ] ; L 2 ) ∩ L 2 ( t 0 , T ; H 1 ) and, for every t ∈ [ t 0 , T ] t\in[t_0,T] t ∈ [ t 0 , T ] ,
E 2 ( t ) + ∫ t 0 t ( ∥ ∇ u 2 ∥ L 2 2 + ∭ ∣ v − u 2 ∣ 2 f 2 d z ) d σ ≤ 1 2 ∥ u ∗ ∥ L 2 2 + 1 2 M 2 ( f ∗ ) , \mathcal{E}_2(t)+\int_{t_0}^t\left(\|\nabla u_2\|_{L^2}^2+\iiint|v-u_2|^2f_2\,\mathrm{d}z\right)\,\mathrm{d}\sigma\le\frac{1}{2}\|u_*\|_{L^2}^2+\frac{1}{2}M_2(f_*), E 2 ( t ) + ∫ t 0 t ( ∥∇ u 2 ∥ L 2 2 + ∭ ∣ v − u 2 ∣ 2 f 2 d z ) d σ ≤ 2 1 ∥ u ∗ ∥ L 2 2 + 2 1 M 2 ( f ∗ ) , where E 2 ( t ) = 1 2 ∥ u 2 ( t ) ∥ L 2 2 + 1 2 M 2 ( f 2 ( t ) ) \mathcal{E}_2(t)=\frac{1}{2}\|u_2(t)\|_{L^2}^2+\frac{1}{2}M_2(f_2(t)) E 2 ( t ) = 2 1 ∥ u 2 ( t ) ∥ L 2 2 + 2 1 M 2 ( f 2 ( t )) .
Then ( u 1 , f 1 ) = ( u 2 , f 2 ) (u_1,f_1)=(u_2,f_2) ( u 1 , f 1 ) = ( u 2 , f 2 ) on [ t 0 , T ] [t_0,T] [ t 0 , T ] .
Proof. We follow § 5.5 . There both velocities are Lipschitz; here only u 1 u_1 u 1 is, and the flow of u 2 u_2 u 2 is replaced by the flows of mollified fields. Let w : = u 1 − u 2 w:=u_1-u_2 w := u 1 − u 2 , F i : = F [ f i , u i ] F_i:=F[f_i,u_i] F i := F [ f i , u i ] ,
R ♯ v : = R ∗ v + ∥ u 1 ∥ L 1 ( t 0 , T ; L ∞ ) + ∥ u 2 ∥ L 1 ( t 0 , T ; L ∞ ) , ρ ♯ : = e d ∥ f ∗ ∥ L ∞ ∣ B R ♯ v ∣ , R_\sharp^v:=R_*^v+\|u_1\|_{L^1(t_0,T;L^\infty)}+\|u_2\|_{L^1(t_0,T;L^\infty)},\qquad\rho_\sharp:=e^d\|f_*\|_{L^\infty}|B_{R_\sharp^v}|, R ♯ v := R ∗ v + ∥ u 1 ∥ L 1 ( t 0 , T ; L ∞ ) + ∥ u 2 ∥ L 1 ( t 0 , T ; L ∞ ) , ρ ♯ := e d ∥ f ∗ ∥ L ∞ ∣ B R ♯ v ∣ ,
a ( t ) : = 1 + ∥ u 1 ( t ) ∥ L ∞ + ∥ ∇ u 1 ( t ) ∥ L ∞ . a(t):=1+\|u_1(t)\|_{L^\infty}+\|\nabla u_1(t)\|_{L^\infty}. a ( t ) := 1 + ∥ u 1 ( t ) ∥ L ∞ + ∥∇ u 1 ( t ) ∥ L ∞ .
By (a) and H 2 ( T d ) ↪ L ∞ H^2(\mathbb{T}^d)\hookrightarrow L^\infty H 2 ( T d ) ↪ L ∞ (d ≤ 3 d\le3 d ≤ 3 ), u 1 ∈ L 2 ( t 0 , T ; L ∞ ) u_1\in L^2(t_0,T;L^\infty) u 1 ∈ L 2 ( t 0 , T ; L ∞ ) ; hence a ∈ L 1 ( t 0 , T ) a\in L^1(t_0,T) a ∈ L 1 ( t 0 , T ) and u 1 ∈ L 1 ( t 0 , T ; W 1 , ∞ ) u_1\in L^1(t_0,T;W^{1,\infty}) u 1 ∈ L 1 ( t 0 , T ; W 1 , ∞ ) . Moreover u 1 ∈ C ( [ t 0 , T ] ; H 1 ) u_1\in C([t_0,T];H^1) u 1 ∈ C ([ t 0 , T ] ; H 1 ) by Lemma 5.4(c).
Step 1: the kinetic states. Both velocities belong to L 1 ( t 0 , T ; W 1 , 1 ) ∩ L 1 ( t 0 , T ; L ∞ ) L^1(t_0,T;W^{1,1})\cap L^1(t_0,T;L^\infty) L 1 ( t 0 , T ; W 1 , 1 ) ∩ L 1 ( t 0 , T ; L ∞ ) , so by Lemma 3.6 each f i f_i f i is the unique bounded solution of its kinetic equation and belongs to C ( [ t 0 , T ] ; L 1 ) C([t_0,T];L^1) C ([ t 0 , T ] ; L 1 ) . By Lemma 8.1(a), started at t 0 t_0 t 0 from f ∗ f_* f ∗ , f 1 ( t ) f_1(t) f 1 ( t ) is the push-forward of f ∗ f_* f ∗ under the Carathéodory flow Z 1 ( t , ⋅ ) : = Z u 1 ( t ; t 0 , ⋅ ) Z_1(t,\cdot):=Z_{u_1}(t;t_0,\cdot) Z 1 ( t , ⋅ ) := Z u 1 ( t ; t 0 , ⋅ ) . For f 2 f_2 f 2 we argue as in Step 8 of the proof of Lemma 11.2: extend u 2 u_2 u 2 by zero outside [ t 0 , T ] [t_0,T] [ t 0 , T ] , let u 2 ε u_2^\varepsilon u 2 ε be its mollification in ( t , x ) (t,x) ( t , x ) , let Z 2 ε ( t , ⋅ ) Z_2^\varepsilon(t,\cdot) Z 2 ε ( t , ⋅ ) be the flow of ( v , u 2 ε − v ) (v,u_2^\varepsilon-v) ( v , u 2 ε − v ) started at t 0 t_0 t 0 , and let f 2 ε ( t ) f_2^\varepsilon(t) f 2 ε ( t ) be the push-forward of f ∗ f_* f ∗ under Z 2 ε ( t , ⋅ ) Z_2^\varepsilon(t,\cdot) Z 2 ε ( t , ⋅ ) . Then ∥ u 2 ε ∥ L 1 ( t 0 , T ; L ∞ ) ≤ ∥ u 2 ∥ L 1 ( t 0 , T ; L ∞ ) \|u_2^\varepsilon\|_{L^1(t_0,T;L^\infty)}\leq\|u_2\|_{L^1(t_0,T;L^\infty)} ∥ u 2 ε ∥ L 1 ( t 0 , T ; L ∞ ) ≤ ∥ u 2 ∥ L 1 ( t 0 , T ; L ∞ ) , u 2 ε → u 2 u_2^\varepsilon\to u_2 u 2 ε → u 2 in L 2 ( ( t 0 , T ) × T d ) L^2((t_0,T)\times\mathbb{T}^d) L 2 (( t 0 , T ) × T d ) , and
f 2 ε ⇀ ∗ f 2 in L ∞ ( ( t 0 , T ) × T d × R d ) ( ε → 0 ) , (63) f_2^\varepsilon\rightharpoonup^* f_2\quad\text{in }L^\infty((t_0,T)\times\mathbb{T}^d\times\mathbb{R}^d)\qquad(\varepsilon\to0), \tag*{(63)} f 2 ε ⇀ ∗ f 2 in L ∞ (( t 0 , T ) × T d × R d ) ( ε → 0 ) , (63)
for the whole family, since every weak-* cluster point is a bounded weak solution with field u 2 u_2 u 2 and datum f ∗ f_* f ∗ , and hence equals f 2 f_2 f 2 by Lemma 3.6. The functions f 1 f_1 f 1 , f 2 f_2 f 2 , f 2 ε f_2^\varepsilon f 2 ε are bounded by e d ∥ f ∗ ∥ L ∞ e^d\|f_*\|_{L^\infty} e d ∥ f ∗ ∥ L ∞ and vanish for ∣ v ∣ ≥ R v ♯ |v|\geq R_v^\sharp ∣ v ∣ ≥ R v ♯ , so their densities are bounded by ϱ ♯ \varrho_\sharp ϱ ♯ . Consequently, for every t t t and every periodic measurable h ≥ 0 h\geq0 h ≥ 0 ,
∫ f ∗ ( z ) h ( X 1 ( t , z ) ) d z = ∫ T d ρ f 1 ( t ) h d x ≤ ϱ ♯ ∥ h ∥ L 1 ( T d ) , ∫ f ∗ ( z ) h ( X 2 ε ( t , z ) ) d z ≤ ϱ ♯ ∥ h ∥ L 1 ( T d ) , (64) \int f_*(z)h(X_1(t,z))\,dz=\int_{\mathbb{T}^d}\rho_{f_1}(t)h\,dx\leq\varrho_\sharp\|h\|_{L^1(\mathbb{T}^d)},\qquad\int f_*(z)h(X_2^\varepsilon(t,z))\,dz\leq\varrho_\sharp\|h\|_{L^1(\mathbb{T}^d)}, \tag*{(64)} ∫ f ∗ ( z ) h ( X 1 ( t , z )) d z = ∫ T d ρ f 1 ( t ) h d x ≤ ϱ ♯ ∥ h ∥ L 1 ( T d ) , ∫ f ∗ ( z ) h ( X 2 ε ( t , z )) d z ≤ ϱ ♯ ∥ h ∥ L 1 ( T d ) , (64)
where Z 1 = ( X 1 , V 1 ) Z_1=(X_1,V_1) Z 1 = ( X 1 , V 1 ) and Z 2 ε = ( X 2 ε , V 2 ε ) Z_2^\varepsilon=(X_2^\varepsilon,V_2^\varepsilon) Z 2 ε = ( X 2 ε , V 2 ε ) . In particular, compositions of almost everywhere defined functions of x x x with X 1 ( t , ⋅ ) X_1(t,\cdot) X 1 ( t , ⋅ ) or X 2 ε ( t , ⋅ ) X_2^\varepsilon(t,\cdot) X 2 ε ( t , ⋅ ) are well defined f ∗ f_* f ∗ -almost everywhere.
Step 2: the energy inequality of the fluid. Let χ ∈ C ∞ ( R d ) \chi\in C^\infty(\mathbb{R}^d) χ ∈ C ∞ ( R d ) be equal to 1 1 1 on B R v ♯ B_{R_v^\sharp} B R v ♯ . Testing Definition 2.1 (iii) for f 2 f_2 f 2 with φ = η ( t ) 1 2 ∣ v ∣ 2 χ ( v ) \varphi=\eta(t)\frac{1}{2}|v|^2\chi(v) φ = η ( t ) 2 1 ∣ v ∣ 2 χ ( v ) shows that d d t 1 2 M 2 ( f 2 ) = ∫ j f 2 ⋅ u 2 d x − M 2 ( f 2 ) \frac{d}{dt}\frac{1}{2}M_2(f_2)=\int j_{f_2}\cdot u_2\,dx-M_2(f_2) d t d 2 1 M 2 ( f 2 ) = ∫ j f 2 ⋅ u 2 d x − M 2 ( f 2 ) in the sense of distributions, with initial value 1 2 M 2 ( f ∗ ) \frac{1}{2}M_2(f_*) 2 1 M 2 ( f ∗ ) . Since t ↦ M 2 ( f 2 ( t ) ) t\mapsto M_2(f_2(t)) t ↦ M 2 ( f 2 ( t )) is continuous, by Step 1,
1 2 M 2 ( f 2 ( t ) ) + ∫ t 0 t M 2 ( f 2 ) d σ = 1 2 M 2 ( f ∗ ) + ∫ t 0 t ∫ j f 2 ⋅ u 2 d x d σ ( t 0 ≤ t ≤ T ) . \frac{1}{2}M_2(f_2(t))+\int_{t_0}^t M_2(f_2)\,d\sigma=\frac{1}{2}M_2(f_*)+\int_{t_0}^t\int j_{f_2}\cdot u_2\,dx\,d\sigma\qquad(t_0\leq t\leq T). 2 1 M 2 ( f 2 ( t )) + ∫ t 0 t M 2 ( f 2 ) d σ = 2 1 M 2 ( f ∗ ) + ∫ t 0 t ∫ j f 2 ⋅ u 2 d x d σ ( t 0 ≤ t ≤ T ) .
Since ∬ ∣ v − u 2 ∣ 2 f 2 d z = M 2 ( f 2 ) − 2 ∫ j f 2 ⋅ u 2 d x + ∫ ρ f 2 ∣ u 2 ∣ 2 d x \iint|v-u_2|^2f_2\,dz=M_2(f_2)-2\int j_{f_2}\cdot u_2\,dx+\int\rho_{f_2}|u_2|^2\,dx ∬ ∣ v − u 2 ∣ 2 f 2 d z = M 2 ( f 2 ) − 2 ∫ j f 2 ⋅ u 2 d x + ∫ ρ f 2 ∣ u 2 ∣ 2 d x and F 2 ⋅ u 2 = j f 2 ⋅ u 2 − ρ f 2 ∣ u 2 ∣ 2 F_2\cdot u_2=j_{f_2}\cdot u_2-\rho_{f_2}|u_2|^2 F 2 ⋅ u 2 = j f 2 ⋅ u 2 − ρ f 2 ∣ u 2 ∣ 2 , subtracting this identity from (b) gives
1 2 ∥ u 2 ( t ) ∥ L 2 2 + ∫ t 0 t ∥ ∇ u 2 ∥ L 2 2 d σ ≤ 1 2 ∥ u ∗ ∥ L 2 2 + ∫ t 0 t ∫ F 2 ⋅ u 2 d x d σ ( t 0 ≤ t ≤ T ) . (65) \frac{1}{2}\|u_2(t)\|_{L^2}^2+\int_{t_0}^t\|\nabla u_2\|_{L^2}^2\,d\sigma\leq\frac{1}{2}\|u_*\|_{L^2}^2+\int_{t_0}^t\int F_2\cdot u_2\,dx\,d\sigma\qquad(t_0\leq t\leq T). \tag*{(65)} 2 1 ∥ u 2 ( t ) ∥ L 2 2 + ∫ t 0 t ∥∇ u 2 ∥ L 2 2 d σ ≤ 2 1 ∥ u ∗ ∥ L 2 2 + ∫ t 0 t ∫ F 2 ⋅ u 2 d x d σ ( t 0 ≤ t ≤ T ) . (65)
Step 3: the relative energy. We claim that, for every t ∈ [ t 0 , T ] t\in[t_0,T] t ∈ [ t 0 , T ] ,
1 2 ∥ w ( t ) ∥ L 2 2 + ∫ t 0 t ∥ ∇ w ∥ L 2 2 d σ ≤ ∫ t 0 t ( ∥ ∇ u 1 ∥ L ∞ ∥ w ∥ L 2 2 + I ( σ ) ) d σ , I ( σ ) : = ∬ ( f 1 − f 2 ) ( σ ) ψ σ d z , (66) \frac{1}{2}\|w(t)\|_{L^2}^2+\int_{t_0}^t\|\nabla w\|_{L^2}^2\,d\sigma\leq\int_{t_0}^t\left(\|\nabla u_1\|_{L^\infty}\|w\|_{L^2}^2+I(\sigma)\right)\,d\sigma,\qquad I(\sigma):=\iint(f_1-f_2)(\sigma)\psi_\sigma\,dz, \tag*{(66)} 2 1 ∥ w ( t ) ∥ L 2 2 + ∫ t 0 t ∥∇ w ∥ L 2 2 d σ ≤ ∫ t 0 t ( ∥∇ u 1 ∥ L ∞ ∥ w ∥ L 2 2 + I ( σ ) ) d σ , I ( σ ) := ∬ ( f 1 − f 2 ) ( σ ) ψ σ d z , (66)
where ψ σ ( x , v ) : = ( v − u 1 ( σ , x ) ) ⋅ w ( σ , x ) \psi_\sigma(x,v):=(v-u_1(\sigma,x))\cdot w(\sigma,x) ψ σ ( x , v ) := ( v − u 1 ( σ , x )) ⋅ w ( σ , x ) .
By (a) and Step 1, u 1 ⋅ ∇ u 1 u_1\cdot\nabla u_1 u 1 ⋅ ∇ u 1 and F 1 F_1 F 1 belong to L 2 ( t 0 , T ; L 2 ) L^2(t_0,T;L^2) L 2 ( t 0 , T ; L 2 ) . Hence Definition 2.1 (iv) gives ∂ t u 1 − Δ u 1 + Q ( u 1 ⋅ ∇ u 1 ) = Q F 1 \partial_tu_1-\Delta u_1+\mathbb{Q}(u_1\cdot\nabla u_1)=\mathbb{Q}F_1 ∂ t u 1 − Δ u 1 + Q ( u 1 ⋅ ∇ u 1 ) = Q F 1 a.e., u 1 ( t 0 ) = u ∗ u_1(t_0)=u_* u 1 ( t 0 ) = u ∗ , and, as in the proof of Proposition 5.10(d),
1 2 ∥ u 1 ( t ) ∥ L 2 2 + ∫ t 0 t ∥ ∇ u 1 ∥ L 2 2 d σ = 1 2 ∥ u ∗ ∥ L 2 2 + ∫ t 0 t ∫ F 1 ⋅ u 1 d x d σ . \frac{1}{2}\|u_1(t)\|_{L^2}^2+\int_{t_0}^t\|\nabla u_1\|_{L^2}^2\,d\sigma=\frac{1}{2}\|u_*\|_{L^2}^2+\int_{t_0}^t\int F_1\cdot u_1\,dx\,d\sigma. 2 1 ∥ u 1 ( t ) ∥ L 2 2 + ∫ t 0 t ∥∇ u 1 ∥ L 2 2 d σ = 2 1 ∥ u ∗ ∥ L 2 2 + ∫ t 0 t ∫ F 1 ⋅ u 1 d x d σ .
Next, as in the proof of Lemma 8.6 , Definition 2.1 (iv) for u 2 u_2 u 2 and the weak continuity of u 2 u_2 u 2 give, for every t ∈ [ t 0 , T ] t\in[t_0,T] t ∈ [ t 0 , T ] and every smooth divergence-free φ \varphi φ on [ t 0 , T ] × T d [t_0,T]\times\mathbb{T}^d [ t 0 , T ] × T d ,
∫ u 2 ( t ) ⋅ φ ( t ) d x − ∫ u ∗ ⋅ φ ( t 0 ) d x = ∫ t 0 t ∫ ( u 2 ⋅ ( ∂ t φ + Δ φ ) + u 2 ⊗ u 2 : ∇ φ + F 2 ⋅ φ ) d x d σ . \int u_2(t)\cdot\varphi(t)\,dx-\int u_*\cdot\varphi(t_0)\,dx=\int_{t_0}^t\int\left(u_2\cdot(\partial_t\varphi+\Delta\varphi)+u_2\otimes u_2:\nabla\varphi+F_2\cdot\varphi\right)\,dx\,d\sigma. ∫ u 2 ( t ) ⋅ φ ( t ) d x − ∫ u ∗ ⋅ φ ( t 0 ) d x = ∫ t 0 t ∫ ( u 2 ⋅ ( ∂ t φ + Δ φ ) + u 2 ⊗ u 2 : ∇ φ + F 2 ⋅ φ ) d x d σ .
Smooth divergence-free fields are dense in the space of divergence-free fields in H 1 ( t 0 , T ; L 2 ) ∩ L 2 ( t 0 , T ; H 2 ) H^1(t_0,T;L^2)\cap L^2(t_0,T;H^2) H 1 ( t 0 , T ; L 2 ) ∩ L 2 ( t 0 , T ; H 2 ) (reflect in time and mollify), which is embedded in C ( [ t 0 , T ] ; H 1 ) C([t_0,T];H^1) C ([ t 0 , T ] ; H 1 ) . Every term of the identity is continuous in this topology, because u 2 ∈ L t , x 2 u_2\in L^2_{t,x} u 2 ∈ L t , x 2 , u 2 ⊗ u 2 ∈ L 1 ( t 0 , T ; L 3 ) u_2\otimes u_2\in L^1(t_0,T;L^3) u 2 ⊗ u 2 ∈ L 1 ( t 0 , T ; L 3 ) while ∇ φ \nabla\varphi ∇ φ is controlled in C ( [ t 0 , T ] ; L 3 / 2 ) C([t_0,T];L^{3/2}) C ([ t 0 , T ] ; L 3/2 ) , and F 2 ∈ L t , x 2 F_2\in L^2_{t,x} F 2 ∈ L t , x 2 . Hence the identity holds with φ = u 1 \varphi=u_1 φ = u 1 . Inserting the equation of u 1 u_1 u 1 , and using ∫ u 2 ⋅ Q g = ∫ u 2 ⋅ g \int u_2\cdot\mathbb{Q}g=\int u_2\cdot g ∫ u 2 ⋅ Q g = ∫ u 2 ⋅ g and u 2 ⊗ u 2 : ∇ u 1 = ( u 2 ⋅ ∇ u 1 ) ⋅ u 2 u_2\otimes u_2:\nabla u_1=(u_2\cdot\nabla u_1)\cdot u_2 u 2 ⊗ u 2 : ∇ u 1 = ( u 2 ⋅ ∇ u 1 ) ⋅ u 2 ,
∫ u 1 ( t ) ⋅ u 2 ( t ) d x = ∥ u ∗ ∥ L 2 2 + ∫ t 0 t ∫ ( − 2 ∇ u 1 : ∇ u 2 + ( u 2 ⋅ ∇ u 1 ) ⋅ u 2 − ( u 1 ⋅ ∇ u 1 ) ⋅ u 2 + F 1 ⋅ u 2 + F 2 ⋅ u 1 ) d x d σ . \int u_1(t)\cdot u_2(t)\,dx=\|u_*\|_{L^2}^2+\int_{t_0}^t\int\left(-2\nabla u_1:\nabla u_2+(u_2\cdot\nabla u_1)\cdot u_2-(u_1\cdot\nabla u_1)\cdot u_2+F_1\cdot u_2+F_2\cdot u_1\right)\,dx\,d\sigma. ∫ u 1 ( t ) ⋅ u 2 ( t ) d x = ∥ u ∗ ∥ L 2 2 + ∫ t 0 t ∫ ( − 2∇ u 1 : ∇ u 2 + ( u 2 ⋅ ∇ u 1 ) ⋅ u 2 − ( u 1 ⋅ ∇ u 1 ) ⋅ u 2 + F 1 ⋅ u 2 + F 2 ⋅ u 1 ) d x d σ .
The cubic terms equal − ( w ⋅ ∇ u 1 ) ⋅ u 2 = ( w ⋅ ∇ u 1 ) ⋅ w − w ⋅ ∇ ∣ u 1 ∣ 2 2 -(w\cdot\nabla u_1)\cdot u_2=(w\cdot\nabla u_1)\cdot w-w\cdot\nabla\frac{|u_1|^2}{2} − ( w ⋅ ∇ u 1 ) ⋅ u 2 = ( w ⋅ ∇ u 1 ) ⋅ w − w ⋅ ∇ 2 ∣ u 1 ∣ 2 , and the last term integrates to zero because w w w is divergence-free. Adding the energy equality of u 1 u_1 u 1 and (65) , and subtracting the last identity, we obtain
1 2 ∥ w ( t ) ∥ L 2 2 + ∫ t 0 t ∥ ∇ w ∥ L 2 2 d σ ≤ ∫ t 0 t ∫ ( ( F 1 − F 2 ) ⋅ w − ( w ⋅ ∇ u 1 ) ⋅ w ) d x d σ . \frac{1}{2}\|w(t)\|_{L^2}^2+\int_{t_0}^t\|\nabla w\|_{L^2}^2\,d\sigma\leq\int_{t_0}^t\int\left((F_1-F_2)\cdot w-(w\cdot\nabla u_1)\cdot w\right)\,dx\,d\sigma. 2 1 ∥ w ( t ) ∥ L 2 2 + ∫ t 0 t ∥∇ w ∥ L 2 2 d σ ≤ ∫ t 0 t ∫ ( ( F 1 − F 2 ) ⋅ w − ( w ⋅ ∇ u 1 ) ⋅ w ) d x d σ .
Since F 1 − F 2 = ∫ ( v − u 1 ) ( f 1 − f 2 ) d v − ρ f 2 w F_1-F_2=\int(v-u_1)(f_1-f_2)\,\mathrm{d}v-\rho f_2w F 1 − F 2 = ∫ ( v − u 1 ) ( f 1 − f 2 ) d v − ρ f 2 w and ρ f 2 ≥ 0 \rho f_2\ge0 ρ f 2 ≥ 0 , this implies (11.11).
Step 4: the coupling term. Lift Z 1 Z_1 Z 1 and Z 2 ε Z_2^\varepsilon Z 2 ε to R d × R d \mathbb{R}^d\times\mathbb{R}^d R d × R d , with the fields extended periodically, started at t 0 t_0 t 0 from the same point, and set
δ Z ε : = ( δ X ε , δ V ε ) : = Z 1 − Z 2 ε , D ε ( t ) 2 : = ∫ f ∗ ( z ) ∣ δ Z ε ( t , z ) ∣ 2 d z , w ε : = u 1 − u 2 ε . \delta Z^\varepsilon:=(\delta X^\varepsilon,\delta V^\varepsilon):=Z_1-Z_2^\varepsilon,\qquad D_\varepsilon(t)^2:=\int f_*(z)|\delta Z^\varepsilon(t,z)|^2\,\mathrm{d}z,\qquad w_\varepsilon:=u_1-u_2^\varepsilon. δ Z ε := ( δ X ε , δ V ε ) := Z 1 − Z 2 ε , D ε ( t ) 2 := ∫ f ∗ ( z ) ∣ δ Z ε ( t , z ) ∣ 2 d z , w ε := u 1 − u 2 ε .
As in Step 1 of Section 5.5, d d t ∣ δ Z ε ∣ 2 ≤ 2 ( 1 + ∥ ∇ u 1 ( t ) ∥ L ∞ ) ∣ δ Z ε ∣ 2 + ∣ w ε ( t , X 2 ε ) ∣ 2 \frac{\mathrm{d}}{\mathrm{d}t}|\delta Z^\varepsilon|^2\le2(1+\|\nabla u_1(t)\|_{L^\infty})|\delta Z^\varepsilon|^2+|w_\varepsilon(t,X_2^\varepsilon)|^2 d t d ∣ δ Z ε ∣ 2 ≤ 2 ( 1 + ∥∇ u 1 ( t ) ∥ L ∞ ) ∣ δ Z ε ∣ 2 + ∣ w ε ( t , X 2 ε ) ∣ 2 . Integrating against f ∗ f_* f ∗ , and using (11.9) and Gronwall’s inequality,
D ε ( t ) 2 ≤ C 1 ∫ t 0 t ∥ w ε ( σ ) ∥ L 2 2 d σ , C 1 : = ϱ ♯ exp ( 2 + 2 ∥ ∇ u 1 ∥ L 1 ( t 0 , T ; L ∞ ) ) . D_\varepsilon(t)^2\le C_1\int_{t_0}^{t}\|w_\varepsilon(\sigma)\|_{L^2}^2\,\mathrm{d}\sigma,\qquad C_1:=\varrho_\sharp\exp\left(2+2\|\nabla u_1\|_{L^1(t_0,T;L^\infty)}\right). D ε ( t ) 2 ≤ C 1 ∫ t 0 t ∥ w ε ( σ ) ∥ L 2 2 d σ , C 1 := ϱ ♯ exp ( 2 + 2∥∇ u 1 ∥ L 1 ( t 0 , T ; L ∞ ) ) .
Since w ε → w w_\varepsilon\to w w ε → w in L 2 ( ( t 0 , T ) × T d ) L^2((t_0,T)\times\mathbb{T}^d) L 2 (( t 0 , T ) × T d ) , the functions D ε D_\varepsilon D ε are bounded uniformly in ε \varepsilon ε and t t t , and
lim sup ε → 0 D ε ( t ) 2 ≤ C 1 ∫ t 0 t ∥ w ( σ ) ∥ L 2 2 d σ ( t 0 ≤ t ≤ T ) . (67) \limsup_{\varepsilon\to0}D_\varepsilon(t)^2\le C_1\int_{t_0}^{t}\|w(\sigma)\|_{L^2}^2\,\mathrm{d}\sigma\qquad(t_0\le t\le T). \tag*{(67)} ε → 0 lim sup D ε ( t ) 2 ≤ C 1 ∫ t 0 t ∥ w ( σ ) ∥ L 2 2 d σ ( t 0 ≤ t ≤ T ) . (67)
Let I ε ( σ ) : = ∬ ( f 1 − f 2 ε ) ( σ ) ψ σ d z I_\varepsilon(\sigma):=\iint(f_1-f_2^\varepsilon)(\sigma)\psi_\sigma\,\mathrm{d}z I ε ( σ ) := ∬ ( f 1 − f 2 ε ) ( σ ) ψ σ d z . For a.e. σ \sigma σ we have u 1 ( σ ) ∈ W 1 , ∞ u_1(\sigma)\in W^{1,\infty} u 1 ( σ ) ∈ W 1 , ∞ and w ( σ ) ∈ H 1 w(\sigma)\in H^1 w ( σ ) ∈ H 1 , and
I ε = ∫ f ∗ ( δ V ε ⋅ w ( X 1 ) − ( u 1 ( X 1 ) − u 1 ( X 2 ε ) ) ⋅ w ( X 1 ) + ( V 2 ε − u 1 ( X 2 ε ) ) ⋅ ( w ( X 1 ) − w ( X 2 ε ) ) ) d z , I_\varepsilon=\int f_*\left(\delta V^\varepsilon\cdot w(X_1)-(u_1(X_1)-u_1(X_2^\varepsilon))\cdot w(X_1)+(V_2^\varepsilon-u_1(X_2^\varepsilon))\cdot(w(X_1)-w(X_2^\varepsilon))\right)\,\mathrm{d}z, I ε = ∫ f ∗ ( δ V ε ⋅ w ( X 1 ) − ( u 1 ( X 1 ) − u 1 ( X 2 ε )) ⋅ w ( X 1 ) + ( V 2 ε − u 1 ( X 2 ε )) ⋅ ( w ( X 1 ) − w ( X 2 ε )) ) d z ,
with all functions evaluated at time σ \sigma σ . The three terms are bounded as in Step 3 of Section 5.5. By the Cauchy–Schwarz inequality and (11.9), the first two are at most ( 1 + ∥ ∇ u 1 ∥ L ∞ ) ϱ ♯ 1 / 2 D ε ∥ w ∥ L 2 (1+\|\nabla u_1\|_{L^\infty})\varrho_\sharp^{1/2}D_\varepsilon\|w\|_{L^2} ( 1 + ∥∇ u 1 ∥ L ∞ ) ϱ ♯ 1/2 D ε ∥ w ∥ L 2 . For the third, ∣ V 2 ε − u 1 ( X 2 ε ) ∣ ≤ R ♯ v + ∥ u 1 ∥ L ∞ |V_2^\varepsilon-u_1(X_2^\varepsilon)|\le R_\sharp^v+\|u_1\|_{L^\infty} ∣ V 2 ε − u 1 ( X 2 ε ) ∣ ≤ R ♯ v + ∥ u 1 ∥ L ∞ on the support of f ∗ f_* f ∗ . Let N σ N_\sigma N σ be the null set of Lemma 5.5 for g = w ( σ ) g=w(\sigma) g = w ( σ ) ; by (11.9) with h = 1 N σ h=\mathbf{1}_{N_\sigma} h = 1 N σ , the sets { X 1 ( σ , ⋅ ) ∈ N σ } \{X_1(\sigma,\cdot)\in N_\sigma\} { X 1 ( σ , ⋅ ) ∈ N σ } and { X 2 ε ( σ , ⋅ ) ∈ N σ } \{X_2^\varepsilon(\sigma,\cdot)\in N_\sigma\} { X 2 ε ( σ , ⋅ ) ∈ N σ } are f ∗ f_* f ∗ -null; here (11.9) applies to h = 1 N σ h=\mathbf{1}_{N_\sigma} h = 1 N σ because N σ N_\sigma N σ , the complement of the Lebesgue points of the periodic function w ( σ ) w(\sigma) w ( σ ) , is periodic. By (5.3), (11.9) and (5.2), the third term is at most 2 C d C M ϱ ♯ 1 / 2 ( R ♯ v + ∥ u 1 ∥ L ∞ ) D ε ∥ ∇ w ∥ L 2 2C_dC_M\varrho_\sharp^{1/2}(R_\sharp^v+\|u_1\|_{L^\infty})D_\varepsilon\|\nabla w\|_{L^2} 2 C d C M ϱ ♯ 1/2 ( R ♯ v + ∥ u 1 ∥ L ∞ ) D ε ∥∇ w ∥ L 2 . By Young’s inequality,
∣ I ε ( σ ) ∣ ≤ 1 2 ∥ ∇ w ( σ ) ∥ L 2 2 + C 2 a ( σ ) ( D ε ( σ ) 2 + ∥ w ( σ ) ∥ L 2 2 ) for a.e. σ , (68) |I_\varepsilon(\sigma)|\le\frac{1}{2}\|\nabla w(\sigma)\|_{L^2}^2+C_2a(\sigma)\left(D_\varepsilon(\sigma)^2+\|w(\sigma)\|_{L^2}^2\right)\qquad\text{for a.e. }\sigma, \tag*{(68)} ∣ I ε ( σ ) ∣ ≤ 2 1 ∥∇ w ( σ ) ∥ L 2 2 + C 2 a ( σ ) ( D ε ( σ ) 2 + ∥ w ( σ ) ∥ L 2 2 ) for a.e. σ , (68)
with C 2 C_2 C 2 depending only on d d d , R ♯ v R_\sharp^v R ♯ v and ϱ ♯ \varrho_\sharp ϱ ♯ .
Finally, ∣ ψ σ ∣ ≤ ( R ♯ v + ∣ u 1 ∣ ) ∣ w ∣ |\psi_\sigma|\le(R_\sharp^v+|u_1|)|w| ∣ ψ σ ∣ ≤ ( R ♯ v + ∣ u 1 ∣ ) ∣ w ∣ for ∣ v ∣ ≤ R ♯ v |v|\le R_\sharp^v ∣ v ∣ ≤ R ♯ v , which is integrable on ( t 0 , T ) × T d × B R ♯ v (t_0,T)\times\mathbb{T}^d\times B_{R_\sharp^v} ( t 0 , T ) × T d × B R ♯ v . By (11.8), ∫ t 0 t I d σ = lim k → ∞ ∫ t 0 t I ε k d σ \int_{t_0}^{t}I\,\mathrm{d}\sigma=\lim_{k\to\infty}\int_{t_0}^{t}I_{\varepsilon_k}\,\mathrm{d}\sigma ∫ t 0 t I d σ = lim k → ∞ ∫ t 0 t I ε k d σ for every t t t and every sequence ε k → 0 \varepsilon_k\to0 ε k → 0 . Fix such a sequence. By (11.13), Fatou’s lemma applied to the nonnegative functions a ( sup ε , t D ε 2 − D ε k 2 ) a(\sup_{\varepsilon,t}D_\varepsilon^2-D_{\varepsilon_k}^2) a ( sup ε , t D ε 2 − D ε k 2 ) , and (11.12) for the measurable function lim sup k D ε k ( σ ) 2 \limsup_kD_{\varepsilon_k}(\sigma)^2 lim sup k D ε k ( σ ) 2 ,
∫ t 0 t I d σ ≤ 1 2 ∫ t 0 t ∥ ∇ w ∥ L 2 2 d σ + C 2 ∫ t 0 t a ( σ ) ( C 1 ∫ t 0 σ ∥ w ∥ L 2 2 d τ + ∥ w ( σ ) ∥ L 2 2 ) d σ . (69) \int_{t_0}^{t}I\,\mathrm{d}\sigma\le\frac{1}{2}\int_{t_0}^{t}\|\nabla w\|_{L^2}^2\,\mathrm{d}\sigma+C_2\int_{t_0}^{t}a(\sigma)\left(C_1\int_{t_0}^{\sigma}\|w\|_{L^2}^2\,\mathrm{d}\tau+\|w(\sigma)\|_{L^2}^2\right)\,\mathrm{d}\sigma. \tag*{(69)} ∫ t 0 t I d σ ≤ 2 1 ∫ t 0 t ∥∇ w ∥ L 2 2 d σ + C 2 ∫ t 0 t a ( σ ) ( C 1 ∫ t 0 σ ∥ w ∥ L 2 2 d τ + ∥ w ( σ ) ∥ L 2 2 ) d σ . (69)
Step 5: Gronwall. Let y ( t ) : = sup t 0 ≤ σ ≤ t ∥ w ( σ ) ∥ L 2 2 y(t):=\sup_{t_0\le\sigma\le t}\|w(\sigma)\|_{L^2}^2 y ( t ) := sup t 0 ≤ σ ≤ t ∥ w ( σ ) ∥ L 2 2 , which is finite and nondecreasing. Inserting (11.14) into (11.11), and using T − t 0 ≤ 1 T-t_0\le1 T − t 0 ≤ 1 and ∥ ∇ u 1 ∥ L ∞ ≤ a \|\nabla u_1\|_{L^\infty}\le a ∥∇ u 1 ∥ L ∞ ≤ a , we get 1 2 ∥ w ( t ) ∥ L 2 2 ≤ C 3 ∫ t 0 t a ( σ ) y ( σ ) d σ \frac{1}{2}\|w(t)\|_{L^2}^2\le C_3\int_{t_0}^{t}a(\sigma)y(\sigma)\,\mathrm{d}\sigma 2 1 ∥ w ( t ) ∥ L 2 2 ≤ C 3 ∫ t 0 t a ( σ ) y ( σ ) d σ for every t t t , with C 3 : = 1 + C 2 ( 1 + C 1 ) C_3:=1+C_2(1+C_1) C 3 := 1 + C 2 ( 1 + C 1 ) . The right-hand side is nondecreasing in t t t , so y ( t ) ≤ 2 C 3 ∫ t 0 t a y d σ y(t)\le2C_3\int_{t_0}^{t}a\,y\,\mathrm{d}\sigma y ( t ) ≤ 2 C 3 ∫ t 0 t a y d σ , and Gronwall’s inequality gives y ≡ 0 y\equiv0 y ≡ 0 . Thus u 1 = u 2 u_1=u_2 u 1 = u 2 , and then f 1 = f 2 f_1=f_2 f 1 = f 2 by Lemma 3.6.
Leray solutions with the strong energy inequality
Theorem 11.4 (Existence). Let d ∈ { 2 , 3 } d\in\{2,3\} d ∈ { 2 , 3 } , let u i n ∈ L 2 ( T d ) u_{\mathrm{in}}\in L^2(\mathbb{T}^d) u in ∈ L 2 ( T d ) be divergence-free, and let f 0 ∈ L ∞ ( T d × R d ) f_0\in L^\infty(\mathbb{T}^d\times\mathbb{R}^d) f 0 ∈ L ∞ ( T d × R d ) be nonnegative and vanish for ∣ v ∣ ≥ R 0 v |v|\ge R_0^v ∣ v ∣ ≥ R 0 v . There is an energy solution ( u , f ) (u,f) ( u , f ) on [ 0 , 1 ] [0,1] [ 0 , 1 ] with data ( u i n , f 0 ) (u_{\mathrm{in}},f_0) ( u in , f 0 ) which satisfies the strong energy inequality: for s = 0 s=0 s = 0 and for almost every s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) , and for all t ∈ [ s , 1 ] t\in[s,1] t ∈ [ s , 1 ] , (10.1) holds. In particular, by Lemma 11.2 if d = 3 d=3 d = 3 and by Lemma 11.5 if d = 2 d=2 d = 2 , ( u , f ) (u,f) ( u , f ) is a weak solution in the sense of Definition 2.1, and a Leray solution with the strong energy inequality in the sense of Section 10.
The proof uses a Galerkin approximation of the fluid equation in which the kinetic equation is solved exactly, as in the rest of the paper. The approximate solutions then satisfy the energy equality on every time interval, and the strong energy inequality follows from the strong convergence of the velocities in L t , x 2 L^2_{t,x} L t , x 2 and the convergence of the kinetic energy at every time.
Proof. Let E ( 0 ) = 1 2 ∥ u i n ∥ L 2 2 + 1 2 M 2 ( f 0 ) \mathcal{E}(0)=\frac{1}{2}\|u_{\mathrm{in}}\|_{L^2}^2+\frac{1}{2}M_2(f_0) E ( 0 ) = 2 1 ∥ u in ∥ L 2 2 + 2 1 M 2 ( f 0 ) and F ∞ : = e d ∥ f 0 ∥ L ∞ F_\infty:=e^{d\|f_0\|_{L^\infty}} F ∞ := e d ∥ f 0 ∥ L ∞ . Constants C C C depend only on E ( 0 ) \mathcal{E}(0) E ( 0 ) , ∥ f 0 ∥ L 1 \|f_0\|_{L^1} ∥ f 0 ∥ L 1 , ∥ f 0 ∥ L ∞ \|f_0\|_{L^\infty} ∥ f 0 ∥ L ∞ and R 0 v R^v_0 R 0 v .
Step 1: Galerkin approximation. Let V n V_n V n and Π n \Pi_n Π n be as in the proof of Lemma 5.7 . For U ∈ C ( [ 0 , 1 ] ; V n ) U\in C([0,1];V_n) U ∈ C ([ 0 , 1 ] ; V n ) , the flow of ( v , U − v ) (v,U-v) ( v , U − v ) is classical, and f [ U ] f[U] f [ U ] denotes the push-forward of f 0 f_0 f 0 along it, as in § 8 . We look for u n ∈ C 1 ( [ 0 , 1 ] ; V n ) u_n\in C^1([0,1];V_n) u n ∈ C 1 ([ 0 , 1 ] ; V n ) such that
u n ′ = Δ u n − Π n ( u n ⋅ ∇ u n ) + Π n F [ f n , u n ] , u n ( 0 ) = Π n u i n , f n : = f [ u n ] . (70) u_n'=\Delta u_n-\Pi_n(u_n\cdot\nabla u_n)+\Pi_nF[f_n,u_n],\qquad u_n(0)=\Pi_nu_{\mathrm{in}},\qquad f_n:=f[u_n]. \tag*{(70)} u n ′ = Δ u n − Π n ( u n ⋅ ∇ u n ) + Π n F [ f n , u n ] , u n ( 0 ) = Π n u in , f n := f [ u n ] . (70)
Since V n V_n V n is finite-dimensional, there is C n C_n C n with ∥ ϕ ∥ W 1 , ∞ + ∥ Δ ϕ ∥ L 2 ≤ C n ∥ ϕ ∥ L 2 \|\phi\|_{W^{1,\infty}}+\|\Delta\phi\|_{L^2}\leq C_n\|\phi\|_{L^2} ∥ ϕ ∥ W 1 , ∞ + ∥Δ ϕ ∥ L 2 ≤ C n ∥ ϕ ∥ L 2 for ϕ ∈ V n \phi\in V_n ϕ ∈ V n . Let A : = ( 2 E ( 0 ) ) 1 / 2 + 1 A:=(2\mathcal{E}(0))^{1/2}+1 A := ( 2 E ( 0 ) ) 1/2 + 1 , and let P A ( ϕ ) : = ϕ min { 1 , A / ∥ ϕ ∥ L 2 } P_A(\phi):=\phi\min\{1,A/\|\phi\|_{L^2}\} P A ( ϕ ) := ϕ min { 1 , A /∥ ϕ ∥ L 2 } be the retraction of V n V_n V n onto its ball of radius A A A , which is 1-Lipschitz in L 2 L^2 L 2 . For U ∈ C ( [ 0 , 1 ] ; V n ) U\in C([0,1];V_n) U ∈ C ([ 0 , 1 ] ; V n ) let U ~ ( t ) : = P A ( U ( t ) ) \widetilde U(t):=P_A(U(t)) U ( t ) := P A ( U ( t )) and
G ( U ) ( t ) : = ( Δ U ~ − Π n ( U ~ ⋅ ∇ U ~ ) + Π n F [ f [ U ~ ] , U ~ ] ) ( t ) ∈ V n . \mathcal{G}(U)(t):=\left(\Delta\widetilde U-\Pi_n(\widetilde U\cdot\nabla\widetilde U)+\Pi_nF[f[\widetilde U],\widetilde U]\right)(t)\in V_n. G ( U ) ( t ) := ( Δ U − Π n ( U ⋅ ∇ U ) + Π n F [ f [ U ] , U ] ) ( t ) ∈ V n .
Since ∥ U ~ ∥ L ∞ ( 0 , 1 ; W 1 , ∞ ) ≤ C n A \|\widetilde U\|_{L^\infty(0,1;W^{1,\infty})}\leq C_nA ∥ U ∥ L ∞ ( 0 , 1 ; W 1 , ∞ ) ≤ C n A , Lemma 8.1(a) shows that f [ U ~ ] f[\widetilde U] f [ U ] vanishes for ∣ v ∣ ≥ R 0 v + C n A |v|\geq R^v_0+C_nA ∣ v ∣ ≥ R 0 v + C n A and that ρ [ U ~ ] \rho[\widetilde U] ρ [ U ] is bounded by a constant depending on n n n . For ϕ ∈ V n \phi\in V_n ϕ ∈ V n , the function t ↦ ∫ F [ f [ U ~ ] , U ~ ] ( t ) ⋅ ϕ d x = ∬ f 0 ( z ) ( V ( t ) − U ~ ( t , X ( t ) ) ) ⋅ ϕ ( X ( t ) ) d z t\mapsto\int F[f[\widetilde U],\widetilde U](t)\cdot\phi\,dx=\iint f_0(z)(V(t)-\widetilde U(t,X(t)))\cdot\phi(X(t))\,dz t ↦ ∫ F [ f [ U ] , U ] ( t ) ⋅ ϕ d x = ∬ f 0 ( z ) ( V ( t ) − U ( t , X ( t ))) ⋅ ϕ ( X ( t )) d z , where ( X , V ) ( t ) = Z U ~ ( t ; 0 , z ) (X,V)(t)=Z_{\widetilde U}(t;0,z) ( X , V ) ( t ) = Z U ( t ; 0 , z ) , is continuous; hence G ( U ) ∈ C ( [ 0 , 1 ] ; V n ) \mathcal{G}(U)\in C([0,1];V_n) G ( U ) ∈ C ([ 0 , 1 ] ; V n ) .
Let U , U ′ ∈ C ( [ 0 , 1 ] ; V n ) U,U'\in C([0,1];V_n) U , U ′ ∈ C ([ 0 , 1 ] ; V n ) . For ϕ ∈ V n \phi\in V_n ϕ ∈ V n with ∥ ϕ ∥ L 2 ≤ 1 \|\phi\|_{L^2}\leq1 ∥ ϕ ∥ L 2 ≤ 1 and ψ ( x , v ) : = ( v − U ~ ′ ( t , x ) ) ⋅ ϕ ( x ) \psi(x,v):=(v-\widetilde U'(t,x))\cdot\phi(x) ψ ( x , v ) := ( v − U ′ ( t , x )) ⋅ ϕ ( x ) ,
∫ ( F [ f [ U ~ ] , U ~ ] − F [ f [ U ~ ′ ] , U ~ ′ ] ) ( t ) ⋅ ϕ d x = ∬ ( f [ U ~ ] − f [ U ~ ′ ] ) ( t ) ψ d z − ∫ ρ [ U ~ ] ( U ~ − U ~ ′ ) ( t ) ⋅ ϕ d x . \int\left(F[f[\widetilde U],\widetilde U]-F[f[\widetilde U'],\widetilde U']\right)(t)\cdot\phi\,dx
=\iint(f[\widetilde U]-f[\widetilde U'])(t)\psi\,dz-\int\rho[\widetilde U](\widetilde U-\widetilde U')(t)\cdot\phi\,dx. ∫ ( F [ f [ U ] , U ] − F [ f [ U ′ ] , U ′ ] ) ( t ) ⋅ ϕ d x = ∬ ( f [ U ] − f [ U ′ ]) ( t ) ψ d z − ∫ ρ [ U ] ( U − U ′ ) ( t ) ⋅ ϕ d x .
By Lemma 8.1(b), with base field U ~ ′ \widetilde U' U ′ and perturbation U ~ − U ~ ′ \widetilde U-\widetilde U' U − U ′ , the first term on the right is bounded by a constant depending on n n n times ∫ 0 t ∥ ( U ~ − U ~ ′ ) ( s ) ∥ L ∞ d s \int_0^t\|(\widetilde U-\widetilde U')(s)\|_{L^\infty}\,ds ∫ 0 t ∥ ( U − U ′ ) ( s ) ∥ L ∞ d s . Together with the elementary bounds for the linear and the quadratic terms on V n V_n V n , this gives
∥ G ( U ) ( t ) − G ( U ′ ) ( t ) ∥ L 2 ≤ L n sup 0 ≤ s ≤ t ∥ U ( s ) − U ′ ( s ) ∥ L 2 , \|\mathcal{G}(U)(t)-\mathcal{G}(U')(t)\|_{L^2}\leq L_n\sup_{0\leq s\leq t}\|U(s)-U'(s)\|_{L^2}, ∥ G ( U ) ( t ) − G ( U ′ ) ( t ) ∥ L 2 ≤ L n 0 ≤ s ≤ t sup ∥ U ( s ) − U ′ ( s ) ∥ L 2 ,
with L n L_n L n depending on n n n , A A A and f 0 f_0 f 0 . As in Step 2 of the proof of Lemma 3.8, the map U ↦ Π n u i n + ∫ 0 t G ( U ) d s U\mapsto\Pi_nu_{\mathrm{in}}+\int_0^t\mathcal{G}(U)\,ds U ↦ Π n u in + ∫ 0 t G ( U ) d s is a contraction of C ( [ 0 , 1 ] ; V n ) C([0,1];V_n) C ([ 0 , 1 ] ; V n ) for the norm sup t e − 2 L n t ∥ U ( t ) ∥ L 2 \sup_t e^{-2L_nt}\|U(t)\|_{L^2} sup t e − 2 L n t ∥ U ( t ) ∥ L 2 . Its fixed point u n u_n u n belongs to C 1 ( [ 0 , 1 ] ; V n ) C^1([0,1];V_n) C 1 ([ 0 , 1 ] ; V n ) and solves u n ′ = G ( u n ) u_n'=\mathcal{G}(u_n) u n ′ = G ( u n ) , u n ( 0 ) = Π n u i n u_n(0)=\Pi_nu_{\mathrm{in}} u n ( 0 ) = Π n u in .
It remains to remove the retraction. Let T n : = sup { t ∈ [ 0 , 1 ] : ∥ u n ∥ L 2 < A on [ 0 , t ] } T_n:=\sup\{t\in[0,1]:\|u_n\|_{L^2}<A\text{ on }[0,t]\} T n := sup { t ∈ [ 0 , 1 ] : ∥ u n ∥ L 2 < A on [ 0 , t ]} , which is positive because ∥ Π n u i n ∥ L 2 ≤ ∥ u i n ∥ L 2 < A \|\Pi_nu_{\mathrm{in}}\|_{L^2}\leq\|u_{\mathrm{in}}\|_{L^2}<A ∥ Π n u in ∥ L 2 ≤ ∥ u in ∥ L 2 < A . On [ 0 , T n ) [0,T_n) [ 0 , T n ) we have u ~ n = u n \widetilde u_n=u_n u n = u n , so (70) holds there. Testing with u n u_n u n , and using ∫ Π n ( u n ⋅ ∇ u n ) ⋅ u n = ∫ ( u n ⋅ ∇ u n ) ⋅ u n = 0 \int\Pi_n(u_n\cdot\nabla u_n)\cdot u_n=\int(u_n\cdot\nabla u_n)\cdot u_n=0 ∫ Π n ( u n ⋅ ∇ u n ) ⋅ u n = ∫ ( u n ⋅ ∇ u n ) ⋅ u n = 0 ,
d d t 1 2 ∥ u n ∥ L 2 2 + ∥ ∇ u n ∥ L 2 2 = ∫ F [ f n , u n ] ⋅ u n d x = ∫ j n ⋅ u n d x − ∫ ρ n ∣ u n ∣ 2 d x , \frac{d}{dt}\frac{1}{2}\|u_n\|_{L^2}^2+\|\nabla u_n\|_{L^2}^2=\int F[f_n,u_n]\cdot u_n\,dx=\int j_n\cdot u_n\,dx-\int\rho_n|u_n|^2\,dx, d t d 2 1 ∥ u n ∥ L 2 2 + ∥∇ u n ∥ L 2 2 = ∫ F [ f n , u n ] ⋅ u n d x = ∫ j n ⋅ u n d x − ∫ ρ n ∣ u n ∣ 2 d x ,
where ρ n : = ρ [ u n ] \rho_n:=\rho[u_n] ρ n := ρ [ u n ] and j n : = j [ u n ] j_n:=j[u_n] j n := j [ u n ] . For the particles, with ( X ( n ) , V ( n ) ) ( t , z ) : = Z u n ( t ; 0 , z ) (X^{(n)},V^{(n)})(t,z):=Z_{u_n}(t;0,z) ( X ( n ) , V ( n ) ) ( t , z ) := Z u n ( t ; 0 , z ) ,
d d t 1 2 M 2 ( f n ) = d d t 1 2 ∫ f 0 ∣ V ( n ) ∣ 2 d z = ∫ f 0 V ( n ) ⋅ ( u n ( t , X ( n ) ) − V ( n ) ) d z = ∫ j n ⋅ u n d x − M 2 ( f n ) . \frac{d}{dt}\frac{1}{2}M_2(f_n)=\frac{d}{dt}\frac{1}{2}\int f_0|V^{(n)}|^2\,dz=\int f_0V^{(n)}\cdot(u_n(t,X^{(n)})-V^{(n)})\,dz=\int j_n\cdot u_n\,dx-M_2(f_n). d t d 2 1 M 2 ( f n ) = d t d 2 1 ∫ f 0 ∣ V ( n ) ∣ 2 d z = ∫ f 0 V ( n ) ⋅ ( u n ( t , X ( n ) ) − V ( n ) ) d z = ∫ j n ⋅ u n d x − M 2 ( f n ) .
Adding the two identities, with E n ( t ) : = 1 2 ∥ u n ( t ) ∥ L 2 2 + 1 2 M 2 ( f n ( t ) ) \mathcal{E}_n(t):=\frac{1}{2}\|u_n(t)\|_{L^2}^2+\frac{1}{2}M_2(f_n(t)) E n ( t ) := 2 1 ∥ u n ( t ) ∥ L 2 2 + 2 1 M 2 ( f n ( t )) ,
E n ( t ) + ∫ s t ( ∥ ∇ u n ∥ L 2 2 + ∬ ∣ v − u n ∣ 2 f n d z ) d σ = E n ( s ) ( 0 ≤ s ≤ t < T n ) . (71) \mathcal{E}_n(t)+\int_s^t\left(\|\nabla u_n\|_{L^2}^2+\iint|v-u_n|^2f_n\,dz\right)\,d\sigma=\mathcal{E}_n(s)\qquad(0\leq s\leq t<T_n). \tag*{(71)} E n ( t ) + ∫ s t ( ∥∇ u n ∥ L 2 2 + ∬ ∣ v − u n ∣ 2 f n d z ) d σ = E n ( s ) ( 0 ≤ s ≤ t < T n ) . (71)
In particular ∥ u n ( t ) ∥ L 2 2 ≤ 2 E n ( 0 ) ≤ 2 E ( 0 ) = ( A − 1 ) 2 \|u_n(t)\|_{L^2}^2\leq2\mathcal{E}_n(0)\leq2\mathcal{E}(0)=(A-1)^2 ∥ u n ( t ) ∥ L 2 2 ≤ 2 E n ( 0 ) ≤ 2 E ( 0 ) = ( A − 1 ) 2 on [ 0 , T n ) [0,T_n) [ 0 , T n ) . By continuity, T n = 1 T_n=1 T n = 1 , and (70) and (71) hold on [ 0 , 1 ] [0,1] [ 0 , 1 ] .
Step 2: uniform bounds. By (71) and Lemma 8.1(a), for all n n n and t t t ,
∥ u n ( t ) ∥ L 2 2 + M 2 ( f n ( t ) ) ≤ 2 E ( 0 ) , ∫ 0 1 ∥ ∇ u n ∥ L 2 2 d t ≤ E ( 0 ) , 0 ≤ f n ≤ F ∞ , ∥ f n ( t ) ∥ L 1 = ∥ f 0 ∥ L 1 . (72) \|u_n(t)\|_{L^2}^2+M_2(f_n(t))\leq2\mathcal{E}(0),\qquad\int_0^1\|\nabla u_n\|_{L^2}^2\,dt\leq\mathcal{E}(0),\qquad0\leq f_n\leq F_\infty,\qquad\|f_n(t)\|_{L^1}=\|f_0\|_{L^1}. \tag*{(72)} ∥ u n ( t ) ∥ L 2 2 + M 2 ( f n ( t )) ≤ 2 E ( 0 ) , ∫ 0 1 ∥∇ u n ∥ L 2 2 d t ≤ E ( 0 ) , 0 ≤ f n ≤ F ∞ , ∥ f n ( t ) ∥ L 1 = ∥ f 0 ∥ L 1 . (72)
In particular ∥ u n ∥ L 2 ( 0 , 1 ; L 6 ) ≤ C \|u_n\|_{L^2(0,1;L^6)}\leq C ∥ u n ∥ L 2 ( 0 , 1 ; L 6 ) ≤ C . The function M 3 ( f n ( t ) ) = ∫ f 0 ∣ V ( n ) ( t , z ) ∣ 3 d z M_3(f_n(t))=\int f_0|V^{(n)}(t,z)|^3\,dz M 3 ( f n ( t )) = ∫ f 0 ∣ V ( n ) ( t , z ) ∣ 3 d z is of class C 1 C^1 C 1 , and by (56) with ( k , m ) = ( 2 , 3 ) (k,m)=(2,3) ( k , m ) = ( 2 , 3 ) , together with d + 3 d + 2 ≥ 6 5 \frac{d+3}{d+2}\geq\frac{6}{5} d + 2 d + 3 ≥ 5 6 for d ≤ 3 d\leq3 d ≤ 3 ,
d d t M 3 ( f n ) = 3 ∫ f 0 ∣ V ( n ) ∣ V ( n ) ⋅ ( u n ( t , X ( n ) ) − V ( n ) ) d z ≤ 3 ∥ u n ∥ L 6 ∥ m 2 f n ∥ L d + 3 d + 2 ≤ C ∥ u n ∥ L 6 M 3 ( f n ) d + 2 d + 3 . \frac{d}{dt}M_3(f_n)=3\int f_0|V^{(n)}|V^{(n)}\cdot(u_n(t,X^{(n)})-V^{(n)})\,dz\leq3\|u_n\|_{L^6}\|m_2f_n\|_{L^{\frac{d+3}{d+2}}}\leq C\|u_n\|_{L^6}M_3(f_n)^{\frac{d+2}{d+3}}. d t d M 3 ( f n ) = 3 ∫ f 0 ∣ V ( n ) ∣ V ( n ) ⋅ ( u n ( t , X ( n ) ) − V ( n ) ) d z ≤ 3∥ u n ∥ L 6 ∥ m 2 f n ∥ L d + 2 d + 3 ≤ C ∥ u n ∥ L 6 M 3 ( f n ) d + 3 d + 2 .
Hence M 3 ( f n ( t ) ) 1 d + 3 ≤ ( R v 0 3 ∥ f 0 ∥ L 1 ) 1 d + 3 + C ∥ u n ∥ L 1 L 6 ≤ C M_{3}(f_{n}(t))^{\frac{1}{d+3}} \leq\left(R v_{0}^{3}\|f_{0}\|_{L^{1}}\right)^{\frac{1}{d+3}} + C\|u_{n}\|_{L^{1}L^{6}} \leq C M 3 ( f n ( t ) ) d + 3 1 ≤ ( R v 0 3 ∥ f 0 ∥ L 1 ) d + 3 1 + C ∥ u n ∥ L 1 L 6 ≤ C , and by (56) with m = 3 m=3 m = 3 , which gives ρ n ∈ L ∞ L ( d + 3 ) / d ⊂ L ∞ L 2 \rho_{n}\in L^{\infty}L^{(d+3)/d}\subset L^{\infty}L^{2} ρ n ∈ L ∞ L ( d + 3 ) / d ⊂ L ∞ L 2 and j n ∈ L ∞ L ( d + 3 ) / ( d + 1 ) ⊂ L ∞ L 3 / 2 j_{n}\in L^{\infty}L^{(d+3)/(d+1)}\subset L^{\infty}L^{3/2} j n ∈ L ∞ L ( d + 3 ) / ( d + 1 ) ⊂ L ∞ L 3/2 for d ≤ 3 d\leq3 d ≤ 3 ,
sup t M 3 ( f n ( t ) ) + ∥ ρ n ∥ L ∞ ( 0 , 1 ; L 2 ) + ∥ j n ∥ L ∞ ( 0 , 1 ; L 3 / 2 ) + ∥ F n ∥ L 2 ( 0 , 1 ; L 3 / 2 ) ≤ C , F n : = F [ f n , u n ] = j n − ρ n u n . (73) \sup_{t}M_{3}(f_{n}(t))+\|\rho_{n}\|_{L^{\infty}(0,1;L^{2})}+\|j_{n}\|_{L^{\infty}(0,1;L^{3/2})}+\|F_{n}\|_{L^{2}(0,1;L^{3/2})}\leq C,\qquad F_{n}:=F[f_{n},u_{n}]=j_{n}-\rho_{n}u_{n}. \tag*{(73)} t sup M 3 ( f n ( t )) + ∥ ρ n ∥ L ∞ ( 0 , 1 ; L 2 ) + ∥ j n ∥ L ∞ ( 0 , 1 ; L 3/2 ) + ∥ F n ∥ L 2 ( 0 , 1 ; L 3/2 ) ≤ C , F n := F [ f n , u n ] = j n − ρ n u n . (73)
For ϕ ∈ H 3 ( T d ; R d ) \phi\in H^{3}(\mathbb{T}^{d};\mathbb{R}^{d}) ϕ ∈ H 3 ( T d ; R d ) , since u n ′ ∈ V n u_{n}'\in V_{n} u n ′ ∈ V n and Π n \Pi_{n} Π n is self-adjoint and commutes with derivatives,
∫ u n ′ ⋅ ϕ d x = ∫ ( − ∇ u n : ∇ Π n ϕ + u n ⊗ u n : ∇ Π n ϕ + F n ⋅ Π n ϕ ) d x , \int u_{n}'\cdot\phi\,dx=\int\left(-\nabla u_{n}:\nabla\Pi_{n}\phi+u_{n}\otimes u_{n}:\nabla\Pi_{n}\phi+F_{n}\cdot\Pi_{n}\phi\right)\,dx, ∫ u n ′ ⋅ ϕ d x = ∫ ( − ∇ u n : ∇ Π n ϕ + u n ⊗ u n : ∇ Π n ϕ + F n ⋅ Π n ϕ ) d x ,
and ∥ Π n ϕ ∥ H 3 ≤ ∥ ϕ ∥ H 3 \|\Pi_{n}\phi\|_{H^{3}}\leq\|\phi\|_{H^{3}} ∥ Π n ϕ ∥ H 3 ≤ ∥ ϕ ∥ H 3 , H 3 ↪ W 1 , ∞ H^{3}\hookrightarrow W^{1,\infty} H 3 ↪ W 1 , ∞ for d ≤ 3 d\leq3 d ≤ 3 . Hence ∥ u n ′ ∥ H − 3 ≤ C ( ∥ u n ∥ H 1 + ∥ u n ∥ L 2 2 + ∥ F n ∥ L 3 / 2 ) \|u_{n}'\|_{H^{-3}}\leq C(\|u_{n}\|_{H^{1}}+\|u_{n}\|_{L^{2}}^{2}+\|F_{n}\|_{L^{3/2}}) ∥ u n ′ ∥ H − 3 ≤ C ( ∥ u n ∥ H 1 + ∥ u n ∥ L 2 2 + ∥ F n ∥ L 3/2 ) , which is bounded in L 2 ( 0 , 1 ) L^{2}(0,1) L 2 ( 0 , 1 ) .
Step 3: compactness. By (72) , the bound on u n ′ u_{n}' u n ′ , the Aubin–Lions lemma and the Arzelà–Ascoli theorem, there are a subsequence, not relabeled, and a weakly divergence-free u ∈ C w ( [ 0 , 1 ] ; L 2 ) ∩ L 2 ( 0 , 1 ; H 1 ) u\in C_{w}([0,1];L^{2})\cap L^{2}(0,1;H^{1}) u ∈ C w ([ 0 , 1 ] ; L 2 ) ∩ L 2 ( 0 , 1 ; H 1 ) such that
u n → u in L 2 ( ( 0 , 1 ) × T d ) , u n ( t ) ⇀ u ( t ) in L 2 for every t , u n ( s ) → u ( s ) in L 2 for a.e. s , u_{n}\to u\quad\text{in }L^{2}((0,1)\times\mathbb{T}^{d}),\qquad u_{n}(t)\rightharpoonup u(t)\quad\text{in }L^{2}\text{ for every }t,\qquad u_{n}(s)\to u(s)\quad\text{in }L^{2}\text{ for a.e. }s, u n → u in L 2 (( 0 , 1 ) × T d ) , u n ( t ) ⇀ u ( t ) in L 2 for every t , u n ( s ) → u ( s ) in L 2 for a.e. s ,
and ∇ u n ⇀ ∇ u \nabla u_{n}\rightharpoonup\nabla u ∇ u n ⇀ ∇ u in L t , x 2 L^{2}_{t,x} L t , x 2 . Interpolating with the bound in L 2 ( 0 , 1 ; L 6 ) L^{2}(0,1;L^{6}) L 2 ( 0 , 1 ; L 6 ) , u n → u u_{n}\to u u n → u in L 2 ( 0 , 1 ; L p ) L^{2}(0,1;L^{p}) L 2 ( 0 , 1 ; L p ) for every p < 6 p<6 p < 6 .
For ψ ∈ C 1 ( T d × R d ) \psi\in C^{1}(\mathbb{T}^{d}\times\mathbb{R}^{d}) ψ ∈ C 1 ( T d × R d ) , the function t ↦ ∬ f n ( t ) ψ d z = ∬ f 0 ψ ( Z u n ( t ; 0 , z ) ) d z t\mapsto\iint f_{n}(t)\psi\,dz=\iint f_{0}\psi(Z_{u_{n}}(t;0,z))\,dz t ↦ ∬ f n ( t ) ψ d z = ∬ f 0 ψ ( Z u n ( t ; 0 , z )) d z has derivative ∬ f n ( v ⋅ ∇ x ψ + ( u n − v ) ⋅ ∇ v ψ ) d z \iint f_{n}(v\cdot\nabla_{x}\psi+(u_{n}-v)\cdot\nabla_{v}\psi)\,dz ∬ f n ( v ⋅ ∇ x ψ + ( u n − v ) ⋅ ∇ v ψ ) d z , which is bounded by C ψ ( 1 + ∥ u n ( t ) ∥ L 2 ) ≤ C ψ C_{\psi}(1+\|u_{n}(t)\|_{L^{2}})\leq C_{\psi} C ψ ( 1 + ∥ u n ( t ) ∥ L 2 ) ≤ C ψ , uniformly in n n n and t t t . By the Arzelà–Ascoli theorem, a diagonal argument over a countable subset of C 1 C^{1} C 1 that is dense in L 1 L^{1} L 1 , and the bound 0 ≤ f n ≤ F ∞ 0\leq f_{n}\leq F_{\infty} 0 ≤ f n ≤ F ∞ , a further subsequence satisfies
f n ( t ) ⇀ ∗ f ( t ) in L ∞ ( T d × R d ) for every t ∈ [ 0 , 1 ] , f_{n}(t)\rightharpoonup^{*}f(t)\quad\text{in }L^{\infty}(\mathbb{T}^{d}\times\mathbb{R}^{d})\quad\text{for every }t\in[0,1], f n ( t ) ⇀ ∗ f ( t ) in L ∞ ( T d × R d ) for every t ∈ [ 0 , 1 ] ,
where 0 ≤ f ( t ) ≤ F ∞ 0\leq f(t)\leq F_{\infty} 0 ≤ f ( t ) ≤ F ∞ and t ↦ ∬ f ( t ) ψ d z t\mapsto\iint f(t)\psi\,dz t ↦ ∬ f ( t ) ψ d z is continuous. As in the proof of Lemma 3.5(b), f n ⇀ ∗ f f_{n}\rightharpoonup^{*}f f n ⇀ ∗ f in L ∞ ( ( 0 , 1 ) × T d × R d ) L^{\infty}((0,1)\times\mathbb{T}^{d}\times\mathbb{R}^{d}) L ∞ (( 0 , 1 ) × T d × R d ) . By (73) ,
∬ { ∣ v ∣ ≥ L } f n ( t ) ( 1 + ∣ v ∣ 2 ) d z ≤ C L − 1 for L ≥ 1 , \iint_{\{|v|\geq L\}}f_{n}(t)(1+|v|^{2})\,dz\leq CL^{-1}\quad\text{for }L\geq1, ∬ { ∣ v ∣ ≥ L } f n ( t ) ( 1 + ∣ v ∣ 2 ) d z ≤ C L − 1 for L ≥ 1 ,
uniformly in n n n and t t t , and the same holds for f ( t ) f(t) f ( t ) by lower semicontinuity. Hence
∬ f n ( t ) Ψ d z → ∬ f ( t ) Ψ d z \iint f_{n}(t)\Psi\,dz\to\iint f(t)\Psi\,dz ∬ f n ( t ) Ψ d z → ∬ f ( t ) Ψ d z
for every t t t and every Ψ ∈ C ( T d × R d ) \Psi\in C(\mathbb{T}^{d}\times\mathbb{R}^{d}) Ψ ∈ C ( T d × R d ) with ∣ Ψ ∣ ≤ C ( 1 + ∣ v ∣ 2 ) |\Psi|\leq C(1+|v|^{2}) ∣Ψ∣ ≤ C ( 1 + ∣ v ∣ 2 ) . (11.19)
In particular M 2 ( f n ( t ) ) → M 2 ( f ( t ) ) M_{2}(f_{n}(t))\to M_{2}(f(t)) M 2 ( f n ( t )) → M 2 ( f ( t )) and ∬ f ( t ) d z = ∥ f 0 ∥ L 1 \iint f(t)\,dz=\|f_{0}\|_{L^{1}} ∬ f ( t ) d z = ∥ f 0 ∥ L 1 for every t t t . By (11.19), dominated convergence in time and (73) ,
ρ n ⇀ ∗ ρ f in L ∞ ( 0 , 1 ; L 2 ) , j n ⇀ ∗ j f in L ∞ ( 0 , 1 ; L 3 / 2 ) . (74) \rho_{n}\rightharpoonup^{*}\rho_{f}\quad\text{in }L^{\infty}(0,1;L^{2}),\qquad j_{n}\rightharpoonup^{*}j_{f}\quad\text{in }L^{\infty}(0,1;L^{3/2}). \tag*{(74)} ρ n ⇀ ∗ ρ f in L ∞ ( 0 , 1 ; L 2 ) , j n ⇀ ∗ j f in L ∞ ( 0 , 1 ; L 3/2 ) . (74)
Step 4: the limit equations. Each f n f_{n} f n satisfies Definition 2.1 (iii) with field u n u_{n} u n , by Lemma 8.1(a). For a test function ϕ \phi ϕ ,
∣ ∭ f n ( u n − u ) ⋅ ∇ v ϕ ∣ ≤ F ∞ C ϕ ∥ u n − u ∥ L t , x 1 → 0 , \left|\iiint f_{n}(u_{n}-u)\cdot\nabla_{v}\phi\right|\leq F_{\infty}C_{\phi}\|u_{n}-u\|_{L^{1}_{t,x}}\to0, ∭ f n ( u n − u ) ⋅ ∇ v ϕ ≤ F ∞ C ϕ ∥ u n − u ∥ L t , x 1 → 0 ,
and the remaining terms pass to the limit because their multipliers are integrable. Hence f f f satisfies Definition 2.1 (iii) with field u u u .
Let φ \varphi φ be a smooth divergence-free field vanishing near t = 1 t=1 t = 1 . Testing (70) with φ \varphi φ and integrating by parts,
∫ Π n u i n ⋅ φ ( 0 ) d x + ∫ 0 1 ∫ ( u n ⋅ ( ∂ t φ + Δ φ ) + u n ⊗ u n : ∇ Π n φ + F n ⋅ Π n φ ) d x d t = 0. \int\Pi_{n}u_{\mathrm{in}}\cdot\varphi(0)\,dx+\int_{0}^{1}\int\left(u_{n}\cdot(\partial_{t}\varphi+\Delta\varphi)+u_{n}\otimes u_{n}:\nabla\Pi_{n}\varphi+F_{n}\cdot\Pi_{n}\varphi\right)\,dx\,dt=0. ∫ Π n u in ⋅ φ ( 0 ) d x + ∫ 0 1 ∫ ( u n ⋅ ( ∂ t φ + Δ φ ) + u n ⊗ u n : ∇ Π n φ + F n ⋅ Π n φ ) d x d t = 0.
Since φ \varphi φ is smooth, Π n φ → φ \Pi_{n}\varphi\to\varphi Π n φ → φ in C ( [ 0 , 1 ] ; W 1 , ∞ ) C([0,1];W^{1,\infty}) C ([ 0 , 1 ] ; W 1 , ∞ ) . The first three terms converge because u n → u u_{n}\to u u n → u in L t , x 2 L^{2}_{t,x} L t , x 2 . Moreover ∬ j n ⋅ Π n φ → ∬ j f ⋅ φ \iint j_{n}\cdot\Pi_{n}\varphi\to\iint j_{f}\cdot\varphi ∬ j n ⋅ Π n φ → ∬ j f ⋅ φ by (74) , and ∬ ρ n u n ⋅ Π n φ → ∬ ρ f u ⋅ φ \iint\rho_{n}u_{n}\cdot\Pi_{n}\varphi\to\iint\rho_{f}u\cdot\varphi ∬ ρ n u n ⋅ Π n φ → ∬ ρ f u ⋅ φ by (74) and the strong convergence of u n ⋅ Π n φ u_{n}\cdot\Pi_{n}\varphi u n ⋅ Π n φ in L 1 ( 0 , 1 ; L 2 ) L^{1}(0,1;L^{2}) L 1 ( 0 , 1 ; L 2 ) . Hence Definition 2.1 (iv) holds for ( u , f ) (u,f) ( u , f ) .
Step 5: the energy inequalities. Fix 0 ≤ s ≤ t ≤ 1 0\leq s\leq t\leq1 0 ≤ s ≤ t ≤ 1 and pass to the limit in (71) .
Drag dissipation. We have ∬ ∣ v − u n ∣ 2 f n d z = M 2 ( f n ) − 2 ∫ j n ⋅ u n d x + ∫ ρ n ∣ u n ∣ 2 d x \iint|v-u_{n}|^{2}f_{n}\,dz=M_{2}(f_{n})-2\int j_{n}\cdot u_{n}\,dx+\int\rho_{n}|u_{n}|^{2}\,dx ∬ ∣ v − u n ∣ 2 f n d z = M 2 ( f n ) − 2 ∫ j n ⋅ u n d x + ∫ ρ n ∣ u n ∣ 2 d x . The first term is bounded and converges at every time. Since u n → u u_{n}\to u u n → u in L 1 ( 0 , 1 ; L 3 ) L^{1}(0,1;L^{3}) L 1 ( 0 , 1 ; L 3 ) and ∣ u n ∣ 2 → ∣ u ∣ 2 |u_{n}|^{2}\to|u|^{2} ∣ u n ∣ 2 → ∣ u ∣ 2 in L 1 ( 0 , 1 ; L 2 ) L^{1}(0,1;L^{2}) L 1 ( 0 , 1 ; L 2 ) , the other two converge after integration in time, by (74) . Hence ∫ s t ∬ ∣ v − u n ∣ 2 f n → ∫ s t ∬ ∣ v − u ∣ 2 f \int_{s}^{t}\iint|v-u_{n}|^{2}f_{n}\to\int_{s}^{t}\iint|v-u|^{2}f ∫ s t ∬ ∣ v − u n ∣ 2 f n → ∫ s t ∬ ∣ v − u ∣ 2 f .
Therefore (10.1) holds for s = 0 s = 0 s = 0 and a.e. s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) , and all t ∈ [ s , 1 ] t \in[s,1] t ∈ [ s , 1 ] .
Step 6: conclusion. By Steps 3–5, ( u , f ) (u,f) ( u , f ) satisfies (E1)–(E4) of Definition 11.1: (E2) follows from 0 ≤ f ≤ F ∞ 0 \le f \le F_\infty 0 ≤ f ≤ F ∞ , the conservation of mass and M 2 ( f ( t ) ) ≤ 2 E ( 0 ) M_2(f(t)) \le2\mathcal{E}(0) M 2 ( f ( t )) ≤ 2 E ( 0 ) , and (E4) is (10.1) with s = 0 s = 0 s = 0 . The last statement follows from Lemma 11.2 or Lemma 11.5.
Finally, (10.1) is an every-time statement in t t t , so we fix the representative. In Steps 3–5, f ( t ) f(t) f ( t ) is the weak-∗ * ∗ limit of f n ( t ) f_n(t) f n ( t ) for every t t t , and t ↦ ∬ f ( t ) ψ d z t \mapsto\iint f(t)\psi\, dz t ↦ ∬ f ( t ) ψ d z is continuous for ψ ∈ C c 1 \psi\in C^1_c ψ ∈ C c 1 . The function f f f has a representative in C ( [ 0 , 1 ] ; L 1 ) C([0,1];L^1) C ([ 0 , 1 ] ; L 1 ) , by Lemma 11.2(c) or Lemma 11.5(c). The two agree for a.e. t t t ; both are weakly continuous in time against C c 1 C^1_c C c 1 test functions, so they agree for every t t t . In particular, M 2 ( f ( t ) ) M_2(f(t)) M 2 ( f ( t )) in (10.1) is the same for both, and the statement holds for the representative in C ( [ 0 , 1 ] ; L 1 ) C([0,1];L^1) C ([ 0 , 1 ] ; L 1 ) used in Sections 9 and 10.
The two-dimensional torus
In this paragraph d = 2 d = 2 d = 2 . The proof of the L t 2 L ∞ L^2_tL^\infty L t 2 L ∞ bound of [20 Lemma 3] is written on R 2 \mathbb{R}^2 R 2 , for solutions in the sense of [20 Definition 1] , and it uses the energy identity of [20 Lemma 2] . Its whole-space ingredients are maximal regularity for the Stokes problem, a Gagliardo–Nirenberg inequality, and the estimate of the free heat flow through the embedding L 2 ( R 2 ) ↪ B ∞ , 2 − 1 L^2(\mathbb{R}^2) \hookrightarrow B^{-1}_{\infty,2} L 2 ( R 2 ) ↪ B ∞ , 2 − 1 . We prove the bound on T 2 \mathbb{T}^2 T 2 for the energy solutions of Definition 11.1, with periodic versions of these three ingredients. The energy inequality (E4) replaces [20 Lemmas 1 and 2] , so that no result of [20 ] is used.
Lemma 11.5 (Velocity bound in two dimensions; after [20 Lemma 3] ). Let d = 2 d = 2 d = 2 , and let ( u , f ) (u,f) ( u , f ) be an energy solution on [ 0 , T ] [0,T] [ 0 , T ] , T ≤ 1 T \le1 T ≤ 1 , with data ( u i n , f 0 ) (u_{\mathrm{in}}, f_0) ( u in , f 0 ) , in the sense of Definition 11.1. Then:
(a) u ∈ L 2 ( 0 , T ; L ∞ ) u \in L^2(0,T;L^\infty) u ∈ L 2 ( 0 , T ; L ∞ ) , with a bound depending only on T T T , E ( 0 ) \mathcal{E}(0) E ( 0 ) , ∥ f 0 ∥ L 1 \lVert f_0\rVert_{L^1} ∥ f 0 ∥ L 1 , ∥ f 0 ∥ L ∞ \lVert f_0\rVert_{L^\infty} ∥ f 0 ∥ L ∞ and R 0 v R_0^v R 0 v ;
(b) 0 ≤ f ≤ e 2 ∥ f 0 ∥ L ∞ 0 \le f \le e^2\lVert f_0\rVert_{L^\infty} 0 ≤ f ≤ e 2 ∥ f 0 ∥ L ∞ , and f ( t ) f(t) f ( t ) vanishes for ∣ v ∣ ≥ R 0 v + ∥ u ∥ L 1 ( 0 , T ; L ∞ ) |v| \ge R_0^v + \lVert u\rVert_{L^1(0,T;L^\infty)} ∣ v ∣ ≥ R 0 v + ∥ u ∥ L 1 ( 0 , T ; L ∞ ) ;
(c) ( u , f ) (u,f) ( u , f ) is a weak solution on [ 0 , T ] [0,T] [ 0 , T ] in the sense of Definition 2.1; f f f is the unique bounded (hence renormalized) solution of the kinetic equation with field u u u , it belongs to C ( [ 0 , T ] ; L 1 ) C([0,T];L^1) C ([ 0 , T ] ; L 1 ) , and ∬ f ( t ) d z = ∥ f 0 ∥ L 1 \iint f(t) \, dz = \lVert f_0\rVert_{L^1} ∬ f ( t ) d z = ∥ f 0 ∥ L 1 for every t t t .
Proof. Let F ∞ : = ∥ f ∥ L ∞ F_\infty:= \lVert f\rVert_{L^\infty} F ∞ := ∥ f ∥ L ∞ . Until the last step, constants C C C depend on the listed quantities and on F ∞ F_\infty F ∞ ; in the last step we show that F ∞ ≤ e 2 ∥ f 0 ∥ L ∞ F_\infty\le e^2\lVert f_0\rVert_{L^\infty} F ∞ ≤ e 2 ∥ f 0 ∥ L ∞ ; that F ∞ < ∞ F_\infty< \infty F ∞ < ∞ is part of (E2). By (E1) and (E4), ∥ u ∥ L ∞ L 2 2 + ∥ u ∥ L 2 H 1 2 ≤ C \lVert u\rVert^2_{L^\infty L^2} + \lVert u\rVert^2_{L^2 H^1} \le C ∥ u ∥ L ∞ L 2 2 + ∥ u ∥ L 2 H 1 2 ≤ C and M 2 ( f ( t ) ) ≤ 2 E ( 0 ) M_2(f(t)) \le2\mathcal{E}(0) M 2 ( f ( t )) ≤ 2 E ( 0 ) for a.e. t t t .
The forcing. By (11.1) with d = 2 d = 2 d = 2 , m = 2 m = 2 m = 2 and k = 0 , 1 k = 0,1 k = 0 , 1 , we have ρ f ∈ L ∞ ( 0 , T ; L 2 ) \rho_f \in L^\infty(0,T;L^2) ρ f ∈ L ∞ ( 0 , T ; L 2 ) and j f ∈ L ∞ ( 0 , T ; L 4 / 3 ) j_f \in L^\infty(0,T;L^{4/3}) j f ∈ L ∞ ( 0 , T ; L 4/3 ) . By Ladyzhenskaya’s inequality ∥ u ∥ L 4 2 ≤ C ∥ u ∥ L 2 ∥ u ∥ H 1 \lVert u\rVert^2_{L^4} \le C\lVert u\rVert_{L^2}\lVert u\rVert_{H^1} ∥ u ∥ L 4 2 ≤ C ∥ u ∥ L 2 ∥ u ∥ H 1 , so u ∈ L 4 ( 0 , T ; L 4 ) u \in L^4(0,T;L^4) u ∈ L 4 ( 0 , T ; L 4 ) . Hence
G : = j f − ρ f u − u ⋅ ∇ u ∈ L 4 / 3 ( 0 , T ; L 4 / 3 ) , G := j_f - \rho_f u - u \cdot\nabla u \in L^{4/3}(0,T;L^{4/3}), G := j f − ρ f u − u ⋅ ∇ u ∈ L 4/3 ( 0 , T ; L 4/3 ) ,
since ∥ ρ f u ∥ L 4 / 3 ≤ ∥ ρ f ∥ L 2 ∥ u ∥ L 4 \lVert\rho_f u\rVert_{L^{4/3}} \le\lVert\rho_f\rVert_{L^2}\lVert u\rVert_{L^4} ∥ ρ f u ∥ L 4/3 ≤ ∥ ρ f ∥ L 2 ∥ u ∥ L 4 and ∥ u ⋅ ∇ u ∥ L 4 / 3 ≤ ∥ u ∥ L 4 ∥ ∇ u ∥ L 2 \lVert u \cdot\nabla u\rVert_{L^{4/3}} \le\lVert u\rVert_{L^4}\lVert\nabla u\rVert_{L^2} ∥ u ⋅ ∇ u ∥ L 4/3 ≤ ∥ u ∥ L 4 ∥ ∇ u ∥ L 2 .
Decomposition. The mean P 0 u P_0u P 0 u is bounded by ∥ u ∥ L ∞ L 2 \lVert u\rVert_{L^\infty L^2} ∥ u ∥ L ∞ L 2 . Let W W W be the mean-free solution of ∂ t W − Δ W = ( 1 − P 0 ) Q G \partial_t W-\Delta W=(1-P_0)\mathbb{Q}G ∂ t W − Δ W = ( 1 − P 0 ) Q G with W ( 0 ) = 0 W(0)=0 W ( 0 ) = 0 given by maximal L 4 / 3 L^{4/3} L 4/3 -regularity of the heat equation on T 2 \mathbb{T}^2 T 2 [3 Chapter III] , so that ∥ ∇ 2 W ∥ L 4 / 3 ( 0 , T ; L 4 / 3 ) ≤ C ∥ G ∥ L 4 / 3 L 4 / 3 \lVert\nabla^2 W\rVert_{L^{4/3}(0,T;L^{4/3})} \le C\lVert G\rVert_{L^{4/3}L^{4/3}} ∥ ∇ 2 W ∥ L 4/3 ( 0 , T ; L 4/3 ) ≤ C ∥ G ∥ L 4/3 L 4/3 . We claim that
( 1 − P 0 ) u = e t Δ ( 1 − P 0 ) u i n + W . (1-P_0)u=e^{t\Delta}(1-P_0)u_{\mathrm{in}}+W. ( 1 − P 0 ) u = e t Δ ( 1 − P 0 ) u in + W .
We argue on Fourier coefficients, which avoids having to identify a function space for ∂ t u \partial_tu ∂ t u ; we write g ^ n : = ∫ T 2 g e − 2 π i n ⋅ x d x \widehat{g}_n:=\int_{\mathbb{T}^2}g e^{-2\pi i n\cdot x}\,dx g n := ∫ T 2 g e − 2 π in ⋅ x d x . Fix n ∈ Z 2 ∖ { 0 } n\in\mathbb{Z}^2\setminus\{0\} n ∈ Z 2 ∖ { 0 } and a unit vector e ⊥ n e\perp n e ⊥ n . Testing Definition 2.1(iv) with the real and imaginary parts of φ = η ( t ) e − 2 π i n ⋅ x \varphi=\eta(t)e^{-2\pi i n\cdot x} φ = η ( t ) e − 2 π in ⋅ x , η ∈ C c ∞ ( [ 0 , T ) ) \eta\in C^\infty_c([0,T)) η ∈ C c ∞ ([ 0 , T )) , which are smooth and divergence-free since e ⋅ n = 0 e\cdot n=0 e ⋅ n = 0 , shows that y ( t ) : = e ⋅ u ^ n ( t ) y(t):=e\cdot\widehat{u}_n(t) y ( t ) := e ⋅ u n ( t ) satisfies y ′ + 4 π 2 ∣ n ∣ 2 y = e ⋅ G ^ n y'+4\pi^2|n|^2y=e\cdot\widehat{G}_n y ′ + 4 π 2 ∣ n ∣ 2 y = e ⋅ G n in D ′ ( 0 , T ) \mathcal{D}'(0,T) D ′ ( 0 , T ) , with y ∈ L ∞ ( 0 , T ) y\in L^\infty(0,T) y ∈ L ∞ ( 0 , T ) and y ( 0 ) = e ⋅ ( u ^ i n ) n y(0)=e\cdot(\widehat{u}_{\mathrm{in}})_n y ( 0 ) = e ⋅ ( u in ) n in the sense of the weak formulation; here we used Q div ( u ⊗ u ) = Q ( u ⋅ ∇ u ) \mathbb{Q}\operatorname{div}(u\otimes u)=\mathbb{Q}(u\cdot\nabla u) Q div ( u ⊗ u ) = Q ( u ⋅ ∇ u ) and e ⋅ ( Q g ) n = e ⋅ g ^ n e\cdot(\mathbb{Q}g)_n=e\cdot\widehat{g}_n e ⋅ ( Q g ) n = e ⋅ g n . The right-hand side of the claim has coefficients solving the same problem, and this scalar problem has a unique solution in L 1 ( 0 , T ) L^1(0,T) L 1 ( 0 , T ) : the difference of two solutions is c e − 4 π 2 ∣ n ∣ 2 t ce^{-4\pi^2|n|^2t} c e − 4 π 2 ∣ n ∣ 2 t , and testing with η ( 0 ) = 1 \eta(0)=1 η ( 0 ) = 1 gives c = 0 c=0 c = 0 . The n n n -th coefficients of both sides of the claim are therefore equal for n n n ; for the left side because u u u is divergence-free; for the right side because u i n u_{\mathrm{in}} u in is divergence-free and n ⋅ W ^ n n\cdot\widehat{W}_n n ⋅ W n solves the same scalar problem with zero datum and zero forcing, since the forcing ( 1 − P 0 ) Q G (1-P_0)\mathbb{Q}G ( 1 − P 0 ) Q G has coefficients orthogonal to n n n . Hence the two sides agree for a.e. t t t . The left side is weakly continuous with values in L 2 L^2 L 2 and the right side is continuous with values in L 4 / 3 L^{4/3} L 4/3 , so they agree for every t t t , and the claim follows. In particular W = ( 1 − P 0 ) u − e t Δ ( 1 − P 0 ) u i n ∈ L ∞ ( 0 , T ; L 2 ) W=(1-P_0)u-e^{t\Delta}(1-P_0)u_{\mathrm{in}}\in L^\infty(0,T;L^2) W = ( 1 − P 0 ) u − e t Δ ( 1 − P 0 ) u in ∈ L ∞ ( 0 , T ; L 2 ) . The Gagliardo–Nirenberg inequality for mean-free functions on T 2 \mathbb{T}^2 T 2 , ∥ ϕ ∥ L ∞ ≤ C ∥ ϕ ∥ L 2 1 / 3 ∥ ∇ 2 ϕ ∥ L 4 / 3 2 / 3 \|\phi\|_{L^\infty}\leq C\|\phi\|_{L^2}^{1/3}\|\nabla^2\phi\|_{L^{4/3}}^{2/3} ∥ ϕ ∥ L ∞ ≤ C ∥ ϕ ∥ L 2 1/3 ∥ ∇ 2 ϕ ∥ L 4/3 2/3 , gives ∥ W ∥ L ∞ ≤ C ∥ W ∥ L 2 1 / 3 ∥ ∇ 2 W ∥ L 4 / 3 2 / 3 ∈ L 1 ( 0 , T ) \|W\|_{L^\infty}\leq C\|W\|_{L^2}^{1/3}\|\nabla^2W\|_{L^{4/3}}^{2/3}\in L^1(0,T) ∥ W ∥ L ∞ ≤ C ∥ W ∥ L 2 1/3 ∥ ∇ 2 W ∥ L 4/3 2/3 ∈ L 1 ( 0 , T ) .
The free heat flow. Let g = ( 1 − P 0 ) u i n g=(1-P_0)u_{\mathrm{in}} g = ( 1 − P 0 ) u in and a j = ∥ Δ j g ∥ L 2 a_j=\|\Delta_jg\|_{L^2} a j = ∥ Δ j g ∥ L 2 , j ≥ 0 j\geq0 j ≥ 0 . By Bernstein’s inequality and the decay of the heat semigroup on dyadic annuli, ∥ e t Δ Δ j g ∥ L ∞ ≤ C 2 j e − c t 4 j a j \|e^{t\Delta}\Delta_jg\|_{L^\infty}\leq C2^je^{-ct4^j}a_j ∥ e t Δ Δ j g ∥ L ∞ ≤ C 2 j e − c t 4 j a j . Hence
∫ 0 T ∥ e t Δ g ∥ L ∞ 2 d t ≤ C ∑ j , k ≥ 0 2 j + k a j a k ∫ 0 ∞ e − c t ( 4 j + 4 k ) d t ≤ C ∑ j , k ≥ 0 2 − ∣ j − k ∣ a j a k ≤ C ∑ j a j 2 ≤ C ∥ u i n ∥ L 2 2 , \int_0^T\|e^{t\Delta}g\|_{L^\infty}^2\,dt\leq C\sum_{j,k\geq0}2^{j+k}a_ja_k\int_0^\infty e^{-ct(4^j+4^k)}\,dt\leq C\sum_{j,k\geq0}2^{-|j-k|}a_ja_k\leq C\sum_j a_j^2\leq C\|u_{\mathrm{in}}\|_{L^2}^2, ∫ 0 T ∥ e t Δ g ∥ L ∞ 2 d t ≤ C j , k ≥ 0 ∑ 2 j + k a j a k ∫ 0 ∞ e − c t ( 4 j + 4 k ) d t ≤ C j , k ≥ 0 ∑ 2 − ∣ j − k ∣ a j a k ≤ C j ∑ a j 2 ≤ C ∥ u in ∥ L 2 2 ,
by Schur’s test. Together with the previous step, u ∈ L 2 ( 0 , T ; L ∞ ) u\in L^2(0,T;L^\infty) u ∈ L 2 ( 0 , T ; L ∞ ) .
Velocity support and the bound on f f f . We now know that u ∈ L 1 ( 0 , T ; W 1 , 1 ) ∩ L 1 ( 0 , T ; L ∞ ) u\in L^1(0,T;W^{1,1})\cap L^1(0,T;L^\infty) u ∈ L 1 ( 0 , T ; W 1 , 1 ) ∩ L 1 ( 0 , T ; L ∞ ) , and f f f is a bounded weak solution of the kinetic equation with field u u u . Step 8 of the proof of Lemma 11.2, which uses only these two facts, Lemma 8.1(a) and Lemma 3.6, applies in dimension two with e 2 e^2 e 2 in place of e 3 e^3 e 3 : f f f is the weak-* limit of the push-forwards f [ u ε ] f[u_\varepsilon] f [ u ε ] of f 0 f_0 f 0 . This proves (b) and shows that ( u , f ) (u,f) ( u , f ) satisfies Definition 2.1(i)–(ii). The remaining statements of (c) follow from Lemma 3.6 and from the conservation of mass, which is obtained as in Step 1 of the proof of Lemma 11.2. In fact the removal of the cutoff is now trivial: by (b), for a.e. t t t the function f ( t ) f(t) f ( t ) vanishes for ∣ v ∣ ≥ R 0 v + ∥ u ∥ L 1 L ∞ |v|\geq R_0^v+\|u\|_{L^1L^\infty} ∣ v ∣ ≥ R 0 v + ∥ u ∥ L 1 L ∞ , so the term containing ∇ v χ L \nabla_v\chi_L ∇ v χ L is zero for L L L large.
Acknowledgements
The author thanks Xiaoyutao Luo for carefully reading an earlier version of the manuscript and for pointing out several errors in Section 6. The author is supported by the PRIN 2022 project “Turbulent effects vs Stability in Equations from Oceanography” (TESEO), project number 2022HSSYPN.
The author used an AI language model (Claude, Anthropic) to check parts of the proofs and of the text, and to harmonize language and notation. The author verified all mathematical content and takes full responsibility for it.
The convex integration step
In this appendix and the next we prove Proposition 2.8, following [12 Sections 4–5] ; Section 4 summarizes the construction and the differences from [12 ] . In this section we construct w w w , P P P and R C L R^{\mathrm{CL}} R CL and verify the algebraic properties (i) and (ii) of Proposition 2.8 (Proposition A.9). The parameters are fixed and the quantitative bounds (iii)–(v) are proved in Appendix B.
Standing assumptions. Throughout this appendix, d ≥ 2 d\geq2 d ≥ 2 , 1 ≤ p < 2 1\leq p<2 1 ≤ p < 2 , 1 ≤ q < ∞ 1\leq q<\infty 1 ≤ q < ∞ and δ > 0 \delta>0 δ > 0 are fixed, and ( u , R ) (u,R) ( u , R ) , 0 < s < t ∗ ≤ 1 0<s<t_*\leq1 0 < s < t ∗ ≤ 1 are as in Proposition 2.8:
u ∈ C ∞ ( [ 0 , 1 ] × T d ; R d ) , div u = 0 , R ∈ C ∞ ( [ 0 , 1 ] × T d ; S 0 d × d ) , R = 0 on [ 0 , t ∗ ] × T d . u\in C^\infty([0,1]\times\mathbb{T}^d;\mathbb{R}^d),\qquad\operatorname{div}u=0,\qquad R\in C^\infty([0,1]\times\mathbb{T}^d;S_0^{d\times d}),\qquad R=0\text{ on }[0,t_*]\times\mathbb{T}^d. u ∈ C ∞ ([ 0 , 1 ] × T d ; R d ) , div u = 0 , R ∈ C ∞ ([ 0 , 1 ] × T d ; S 0 d × d ) , R = 0 on [ 0 , t ∗ ] × T d .
If R ≡ 0 R\equiv0 R ≡ 0 , Proposition 2.8 holds with w = 0 w=0 w = 0 , P = 0 P=0 P = 0 , R C L = 0 R^{\mathrm{CL}}=0 R CL = 0 . We therefore assume
∥ R ∥ 1 : = ∥ R ∥ L 1 ( [ 0 , 1 ] × T d ) > 0. \|R\|_1:=\|R\|_{L^1([0,1]\times\mathbb{T}^d)}>0. ∥ R ∥ 1 := ∥ R ∥ L 1 ([ 0 , 1 ] × T d ) > 0.
The perturbation depends on four parameters,
ν ∈ N (temporal oscillation) , κ ≥ 1 (temporal concentration) , \nu\in\mathbb{N}\ \text{(temporal oscillation)},\qquad\kappa\geq1\ \text{(temporal concentration)}, ν ∈ N (temporal oscillation) , κ ≥ 1 (temporal concentration) ,
σ ∈ N (spatial oscillation) , μ ≥ 1 (spatial concentration) , \sigma\in\mathbb{N}\ \text{(spatial oscillation)},\qquad\mu\geq1\ \text{(spatial concentration)}, σ ∈ N (spatial oscillation) , μ ≥ 1 (spatial concentration) ,
which are taken large and are fixed in Appendix B as powers of a single frequency λ \lambda λ . All objects in this section are well defined for every admissible choice of these parameters with μ ≥ μ 0 \mu\geq\mu_0 μ ≥ μ 0 , where μ 0 \mu_0 μ 0 is the geometric constant of Section A.2.
Conventions. For matrices we use the Frobenius norm ∣ A ∣ = ( A : A ) 1 / 2 |A|=(A:A)^{1/2} ∣ A ∣ = ( A : A ) 1/2 . For a vector v v v and a matrix A A A we write ( v ⋅ A ) i = v j A i j (v\cdot A)_i=v_jA_{ij} ( v ⋅ A ) i = v j A ij , so that div ( f A ) = ∇ f ⋅ A + f div A \operatorname{div}(fA)=\nabla f\cdot A+f\operatorname{div}A div ( f A ) = ∇ f ⋅ A + f div A with ( div A ) i = ∂ j A i j (\operatorname{div}A)_i=\partial_jA_{ij} ( div A ) i = ∂ j A ij . For f f f on T d \mathbb{T}^d T d and σ ∈ N \sigma\in\mathbb{N} σ ∈ N , f ( σ ⋅ ) f(\sigma\cdot) f ( σ ⋅ ) denotes the σ − 1 \sigma^{-1} σ − 1 -periodic function x ↦ f ( σ x ) x\mapsto f(\sigma x) x ↦ f ( σ x ) ; recall
∥ f ( σ ⋅ ) ∥ L p ( T d ) = ∥ f ∥ L p ( T d ) , ∇ m ( f ( σ ⋅ ) ) = σ m ( ∇ m f ) ( σ ⋅ ) , ∫ f ( σ ⋅ ) = ∫ f . (75) \|f(\sigma\cdot)\|_{L^p(\mathbb{T}^d)}=\|f\|_{L^p(\mathbb{T}^d)},\qquad\nabla^m(f(\sigma\cdot))=\sigma^m(\nabla^m f)(\sigma\cdot),\qquad\int f(\sigma\cdot)=\int f. \tag*{(75)} ∥ f ( σ ⋅ ) ∥ L p ( T d ) = ∥ f ∥ L p ( T d ) , ∇ m ( f ( σ ⋅ )) = σ m ( ∇ m f ) ( σ ⋅ ) , ∫ f ( σ ⋅ ) = ∫ f . (75)
Antidivergence operators
Besides the operator R \mathcal{R} R of Section 2.1, we use the bilinear antidivergence of [12 Appendix B.3] ,
B ( v , A ) : = v ⋅ R A − R ( ∇ v ⋅ R A ) , \mathcal{B}(v,A):=v\cdot\mathcal{R}A-\mathcal{R}(\nabla v\cdot\mathcal{R}A), B ( v , A ) := v ⋅ R A − R ( ∇ v ⋅ R A ) ,
defined for v ∈ C ∞ ( T d ; R d ) v\in C^\infty(\mathbb{T}^d;\mathbb{R}^d) v ∈ C ∞ ( T d ; R d ) and A ∈ C ∞ ( T d ; R d × d ) A\in C^\infty(\mathbb{T}^d;\mathbb{R}^{d\times d}) A ∈ C ∞ ( T d ; R d × d ) with ∫ A = 0 \int A=0 ∫ A = 0 . Here R \mathcal{R} R acts on A A A row by row, and the contractions are those of [12 Appendix B.3] . We only use the following facts.
Lemma A.1 ((a), (c): [12 Theorems B.3, B.4] ). (a) R \mathcal{R} R maps smooth vector fields to smooth S 0 d × d \mathcal{S}^{d\times d}_0 S 0 d × d -valued fields, div R g = g − ∫ g \operatorname{div}\mathcal{R}g=g-\int g div R g = g − ∫ g , and ∥ R g ∥ L p ≲ ∥ g ∥ L p \|\mathcal{R}g\|_{L^p}\lesssim\|g\|_{L^p} ∥ R g ∥ L p ≲ ∥ g ∥ L p for 1 ≤ p ≤ ∞ 1\le p\le\infty 1 ≤ p ≤ ∞ . If ∫ g = 0 \int g=0 ∫ g = 0 and σ ∈ N \sigma\in\mathbb{N} σ ∈ N , then ∥ R ( g ( σ ⋅ ) ) ∥ L p ≲ σ − 1 ∥ g ∥ L p \|\mathcal{R}(g(\sigma\cdot))\|_{L^p}\lesssim\sigma^{-1}\|g\|_{L^p} ∥ R ( g ( σ ⋅ )) ∥ L p ≲ σ − 1 ∥ g ∥ L p .
(b) For 1 < p < ∞ 1<p<\infty 1 < p < ∞ , the operator R div \mathcal{R}\operatorname{div} R div is bounded on L p ( T d ; R d × d ) L^p(\mathbb{T}^d;\mathbb{R}^{d\times d}) L p ( T d ; R d × d ) .
(c) Let A A A be smooth, symmetric and of zero mean. Then B ( v , A ) \mathcal{B}(v,A) B ( v , A ) is smooth and S 0 d × d \mathcal{S}^{d\times d}_0 S 0 d × d -valued,
div B ( v , A ) = v ⋅ A − ∫ v ⋅ A , ∥ B ( v , A ) ∥ L p ≲ ∥ v ∥ C 1 ∥ R A ∥ L p ( 1 ≤ p ≤ ∞ ) . \operatorname{div}\mathcal{B}(v,A)=v\cdot A-\int v\cdot A,\qquad\|\mathcal{B}(v,A)\|_{L^p}\lesssim\|v\|_{C^1}\|\mathcal{R}A\|_{L^p}\quad(1\le p\le\infty). div B ( v , A ) = v ⋅ A − ∫ v ⋅ A , ∥ B ( v , A ) ∥ L p ≲ ∥ v ∥ C 1 ∥ R A ∥ L p ( 1 ≤ p ≤ ∞ ) .
Part (b) holds because R div \mathcal{R}\operatorname{div} R div is a matrix of zeroth-order Fourier multipliers of Calderón–Zygmund type (compositions of Riesz transforms), smooth away from the origin.
Stationary Mikado flows
The building blocks are the stationary Mikado flows of Daneri and Székelyhidi [16 ] , with spatial concentration as in [26 , 12 ] . The following geometric lemma goes back to Nash; see [14 Lemma 3.2] .
Lemma A.2 (Geometric lemma). Let N : = { A ∈ S d × d : ∣ A − Id ∣ ≤ 1 2 } \mathcal{N}:=\{A\in\mathcal{S}^{d\times d}:|A-\operatorname{Id}|\le\frac12\} N := { A ∈ S d × d : ∣ A − Id ∣ ≤ 2 1 } , a compact subset of the positive definite symmetric matrices. There are a finite set Ξ ⊂ Z d ∖ { 0 } \Xi\subset\mathbb{Z}^d\setminus\{0\} Ξ ⊂ Z d ∖ { 0 } and smooth functions Γ k ∈ C ∞ ( N ; R ) \Gamma_k\in C^\infty(\mathcal{N};\mathbb{R}) Γ k ∈ C ∞ ( N ; R ) , k ∈ Ξ k\in\Xi k ∈ Ξ , such that
A = ∑ k ∈ Ξ Γ k 2 ( A ) e k ⊗ e k for all A ∈ N , e k : = k ∣ k ∣ . A=\sum_{k\in\Xi}\Gamma_k^2(A)e_k\otimes e_k\qquad\text{for all }A\in\mathcal{N},\qquad e_k:=\frac{k}{|k|}. A = k ∈ Ξ ∑ Γ k 2 ( A ) e k ⊗ e k for all A ∈ N , e k := ∣ k ∣ k .
Since only the directions e k e_k e k enter, we may replace each k ∈ Ξ k\in\Xi k ∈ Ξ by k / gcd ( k 1 , … , k d ) k/\gcd(k_1,\ldots,k_d) k / g cd( k 1 , … , k d ) ; we therefore assume that every k ∈ Ξ k\in\Xi k ∈ Ξ is primitive. After this replacement two elements of Ξ \Xi Ξ may coincide (for instance k k k and 2 k 2k 2 k ). We do not merge them, since ( Γ k 2 + Γ k ′ 2 ) 1 / 2 (\Gamma_k^2+\Gamma_{k'}^2)^{1/2} ( Γ k 2 + Γ k ′ 2 ) 1/2 need not be smooth where both functions vanish. Instead we regard Ξ \Xi Ξ as a finite indexed family: in what follows, k ≠ k ′ k\ne k' k = k ′ refers to distinct indices, which may carry the same vector, and each index has its own function Γ k \Gamma_k Γ k , its own base point p k p_k p k and its own Mikado flow. The family Ξ \Xi Ξ and the functions Γ k \Gamma_k Γ k are fixed once and for all, and depend only on d d d .
Periodic lines. For k ∈ Ξ k\in\Xi k ∈ Ξ and p k ∈ T d p_k\in\mathbb{T}^d p k ∈ T d let l k = { p k + t k : t ∈ R } ⊂ T d l_k=\{p_k+tk:t\in\mathbb{R}\}\subset\mathbb{T}^d l k = { p k + t k : t ∈ R } ⊂ T d . Since k k k is primitive, it is a closed geodesic of length ∣ k ∣ |k| ∣ k ∣ , traversed once as t t t runs over [ 0 , 1 ] [0,1] [ 0 , 1 ] . Two periodic lines with parallel directions either coincide or are disjoint, and two lines with non-parallel directions meet in finitely many points. We choose the points p k p_k p k , k ∈ Ξ k\in\Xi k ∈ Ξ , one after the other, so that
l k ∩ l k ′ = ∅ whenever k ≠ k ′ and k ∥ k ′ . (76) l_k\cap l_{k'}=\varnothing\qquad\text{whenever }k\ne k'\text{ and }k\parallel k'. \tag*{(76)} l k ∩ l k ′ = ∅ whenever k = k ′ and k ∥ k ′ . (76)
This is possible because, for each k k k , the forbidden set of p k p_k p k is a finite union of periodic lines. We make this choice for definiteness; it also covers parallel directions such as k ′ = − k k'=-k k ′ = − k , and two indices carrying the same vector. By compactness and transversality, there are μ 0 ≥ 1 \mu_0\ge1 μ 0 ≥ 1 and M Ξ ≥ 1 M_\Xi\ge1 M Ξ ≥ 1 , depending only on Ξ \Xi Ξ and the points p k p_k p k , such that for μ ≥ μ 0 \mu\ge\mu_0 μ ≥ μ 0 the function x ↦ dist ( x , l k ) x\mapsto\operatorname{dist}(x,l_k) x ↦ dist ( x , l k ) is smooth on { 0 < dist ( x , l k ) ≤ μ − 1 } \{0<\operatorname{dist}(x,l_k)\le\mu^{-1}\} { 0 < dist ( x , l k ) ≤ μ − 1 } , and for k ≠ k ′ k\ne k' k = k ′ ,
{ dist ( x , l k ) ≤ μ − 1 } ∩ { dist ( x , l k ′ ) ≤ μ − 1 } ⊂ ⋃ y ∈ l k ∩ l k ′ B ( y , M Ξ μ − 1 ) , (77) \{\operatorname{dist}(x,l_k)\le\mu^{-1}\}\cap\{\operatorname{dist}(x,l_{k'})\le\mu^{-1}\}\subset\bigcup_{y\in l_k\cap l_{k'}}B(y,M_\Xi\mu^{-1}), \tag*{(77)} { dist ( x , l k ) ≤ μ − 1 } ∩ { dist ( x , l k ′ ) ≤ μ − 1 } ⊂ y ∈ l k ∩ l k ′ ⋃ B ( y , M Ξ μ − 1 ) , (77)
where the right-hand side is empty if k ∥ k ′ k\parallel k' k ∥ k ′ . Profiles. Fix a radial function ϕ ∈ C c ∞ ( R d − 1 ) \phi\in C^\infty_c(\mathbb{R}^{d-1}) ϕ ∈ C c ∞ ( R d − 1 ) , not identically zero, supported in the annulus { 1 2 ≤ ∣ y ∣ ≤ 1 } \left\{\frac{1}{2} \le|y| \le1\right\} { 2 1 ≤ ∣ y ∣ ≤ 1 } , and let ψ = Δ R d − 1 ϕ \psi= \Delta_{\mathbb{R}^{d-1}}\phi ψ = Δ R d − 1 ϕ , which is radial, supported in the same annulus and not identically zero. Write ϕ ( ∣ y ∣ ) , ψ ( ∣ y ∣ ) \phi(|y|),\psi(|y|) ϕ ( ∣ y ∣ ) , ψ ( ∣ y ∣ ) for their radial profiles. For μ ≥ μ 0 \mu\ge\mu_0 μ ≥ μ 0 and k ∈ Ξ k \in\Xi k ∈ Ξ set
ψ k ( x ) : = c k μ d − 1 2 ψ ( μ dist ( x , l k ) ) , ϕ k ( x ) : = c k μ d − 1 2 − 2 ϕ ( μ dist ( x , l k ) ) , (78) \psi_k(x) := c_k\mu^{\frac{d-1}{2}}\psi\bigl(\mu\operatorname{dist}(x,l_k)\bigr),\qquad\phi_k(x) := c_k\mu^{\frac{d-1}{2}-2}\phi\bigl(\mu\operatorname{dist}(x,l_k)\bigr), \tag*{(78)} ψ k ( x ) := c k μ 2 d − 1 ψ ( μ dist ( x , l k ) ) , ϕ k ( x ) := c k μ 2 d − 1 − 2 ϕ ( μ dist ( x , l k ) ) , (78)
where c k = ( ∣ k ∣ ∫ R d − 1 ψ 2 ) − 1 / 2 c_k = \left(|k|\int_{\mathbb{R}^{d-1}}\psi^2\right)^{-1/2} c k = ( ∣ k ∣ ∫ R d − 1 ψ 2 ) − 1/2 . In coordinates adapted to l k l_k l k (arclength along l k l_k l k and Euclidean coordinates y y y in the orthogonal ( d − 1 ) (d-1) ( d − 1 ) -plane), ψ k \psi_k ψ k and ϕ k \phi_k ϕ k depend only on y y y , so
Δ ϕ k = ψ k , e k ⋅ ∇ ψ k = e k ⋅ ∇ ϕ k = 0 , ∫ T d ψ k 2 d x = 1. (79) \Delta\phi_k=\psi_k,\qquad e_k\cdot\nabla\psi_k=e_k\cdot\nabla\phi_k=0,\qquad\int_{\mathbb{T}^d}\psi_k^2\,\mathrm{d}x=1. \tag*{(79)} Δ ϕ k = ψ k , e k ⋅ ∇ ψ k = e k ⋅ ∇ ϕ k = 0 , ∫ T d ψ k 2 d x = 1. (79)
The Mikado flows and their potentials are
W k : = ψ k e k , Ω k : = e k ⊗ ∇ ϕ k − ∇ ϕ k ⊗ e k . (80) W_k:=\psi_k e_k,\qquad\Omega_k:=e_k\otimes\nabla\phi_k-\nabla\phi_k\otimes e_k. \tag*{(80)} W k := ψ k e k , Ω k := e k ⊗ ∇ ϕ k − ∇ ϕ k ⊗ e k . (80)
Lemma A.3 (Properties of Mikado flows). For μ ≥ μ 0 \mu\ge\mu_0 μ ≥ μ 0 and k , k ′ ∈ Ξ k,k' \in\Xi k , k ′ ∈ Ξ :
(a) W k , Ω k ∈ C ∞ ( T d ) W_k,\Omega_k\in C^\infty(\mathbb{T}^d) W k , Ω k ∈ C ∞ ( T d ) , Ω k \Omega_k Ω k is skew-symmetric, and
div Ω k = W k , div W k = 0 , div ( W k ⊗ W k ) = 0. \operatorname{div}\Omega_k=W_k,\qquad\operatorname{div}W_k=0,\qquad\operatorname{div}(W_k\otimes W_k)=0. div Ω k = W k , div W k = 0 , div ( W k ⊗ W k ) = 0. (b) For m ≥ 0 m\ge0 m ≥ 0 and 1 ≤ p ≤ ∞ 1\le p\le\infty 1 ≤ p ≤ ∞ , uniformly in μ \mu μ ,
∥ ∇ m W k ∥ L p ≲ m μ m + d − 1 2 − d − 1 p , ∥ ∇ m Ω k ∥ L p ≲ m μ m − 1 + d − 1 2 − d − 1 p . \|\nabla^mW_k\|_{L^p}\lesssim_m\mu^{m+\frac{d-1}{2}-\frac{d-1}{p}},\qquad\|\nabla^m\Omega_k\|_{L^p}\lesssim_m\mu^{m-1+\frac{d-1}{2}-\frac{d-1}{p}}. ∥ ∇ m W k ∥ L p ≲ m μ m + 2 d − 1 − p d − 1 , ∥ ∇ m Ω k ∥ L p ≲ m μ m − 1 + 2 d − 1 − p d − 1 . (c) ∫ W k ⊗ W k = e k ⊗ e k \int W_k\otimes W_k=e_k\otimes e_k ∫ W k ⊗ W k = e k ⊗ e k , ∫ W k = 0 \int W_k=0 ∫ W k = 0 , and, for k ≠ k ′ k\ne k' k = k ′ and 1 ≤ p ≤ ∞ 1\le p\le\infty 1 ≤ p ≤ ∞ , ∥ W k ⊗ W k ′ ∥ L p ≲ μ d − 1 − d p \|W_k\otimes W_{k'}\|_{L^p}\lesssim\mu^{d-1-\frac{d}{p}} ∥ W k ⊗ W k ′ ∥ L p ≲ μ d − 1 − p d .
Proof. (a) Smoothness follows from the smoothness of dist ( ⋅ , l k ) \operatorname{dist}(\cdot,l_k) dist ( ⋅ , l k ) on the support of the profiles. Ω k \Omega_k Ω k is skew by definition, and by (A.5)
( div Ω k ) i = ∂ j ( e k , i ∂ j ϕ k − ∂ i ϕ k e k , j ) = e k , i Δ ϕ k − ∂ i ( e k ⋅ ∇ ϕ k ) = ψ k e k , i . (\operatorname{div}\Omega_k)_i=\partial_j(e_{k,i}\partial_j\phi_k-\partial_i\phi_k e_{k,j})=e_{k,i}\Delta\phi_k-\partial_i(e_k\cdot\nabla\phi_k)=\psi_k e_{k,i}. ( div Ω k ) i = ∂ j ( e k , i ∂ j ϕ k − ∂ i ϕ k e k , j ) = e k , i Δ ϕ k − ∂ i ( e k ⋅ ∇ ϕ k ) = ψ k e k , i .
Hence div W k = div div Ω k = 0 \operatorname{div}W_k=\operatorname{div}\operatorname{div}\Omega_k=0 div W k = div div Ω k = 0 , since Ω k \Omega_k Ω k is skew. Finally, ( div ( W k ⊗ W k ) ) i = e k , i e k ⋅ ∇ ( ψ k 2 ) = 0 \bigl(\operatorname{div}(W_k\otimes W_k)\bigr)_i=e_{k,i}e_k\cdot\nabla(\psi_k^2)=0 ( div ( W k ⊗ W k ) ) i = e k , i e k ⋅ ∇ ( ψ k 2 ) = 0 .
(b) In adapted coordinates, ∇ m \nabla^m ∇ m falls on ψ ( μ ∣ y ∣ ) \psi(\mu|y|) ψ ( μ ∣ y ∣ ) and produces μ m \mu^m μ m , while the L p L^p L p norm of μ d − 1 2 ψ ( μ ∣ y ∣ ) \mu^{\frac{d-1}{2}}\psi(\mu|y|) μ 2 d − 1 ψ ( μ ∣ y ∣ ) over a tube of length ∣ k ∣ |k| ∣ k ∣ is ≃ μ d − 1 2 − d − 1 p \simeq\mu^{\frac{d-1}{2}-\frac{d-1}{p}} ≃ μ 2 d − 1 − p d − 1 . The same applies to ∇ ϕ k \nabla\phi_k ∇ ϕ k , which carries the factor μ d − 1 2 − 2 ⋅ μ \mu^{\frac{d-1}{2}-2}\cdot\mu μ 2 d − 1 − 2 ⋅ μ .
(c) By (A.5) and ∣ T d ∣ = 1 |\mathbb{T}^d|=1 ∣ T d ∣ = 1 , ∫ W k ⊗ W k = e k ⊗ e k ∫ ψ k 2 = e k ⊗ e k \int W_k\otimes W_k=e_k\otimes e_k\int\psi_k^2=e_k\otimes e_k ∫ W k ⊗ W k = e k ⊗ e k ∫ ψ k 2 = e k ⊗ e k . Also ∫ W k = ∫ div Ω k = 0 \int W_k=\int\operatorname{div}\Omega_k=0 ∫ W k = ∫ div Ω k = 0 . For k ≠ k ′ k\ne k' k = k ′ , W k ⊗ W k ′ W_k\otimes W_{k'} W k ⊗ W k ′ is supported in the set (A.3), which has measure ≲ μ − d \lesssim\mu^{-d} ≲ μ − d and is empty if k ∥ k ′ k\parallel k' k ∥ k ′ . Since ∥ W k ∥ L ∞ ≲ μ d − 1 2 \|W_k\|_{L^\infty}\lesssim\mu^{\frac{d-1}{2}} ∥ W k ∥ L ∞ ≲ μ 2 d − 1 , ∥ W k ⊗ W k ′ ∥ L p ≲ μ d − 1 − d p \|W_k\otimes W_{k'}\|_{L^p}\lesssim\mu^{d-1-\frac{d}{p}} ∥ W k ⊗ W k ′ ∥ L p ≲ μ d − 1 − p d .
Temporal intermittency
Fix a nonnegative g ∈ C ∞ ( ( 0 , 1 ) ) g\in C^\infty((0,1)) g ∈ C ∞ (( 0 , 1 )) with ∫ 0 1 g 2 = 1 \int_0^1 g^2=1 ∫ 0 1 g 2 = 1 . For κ ≥ 1 \kappa\ge1 κ ≥ 1 let g κ g_\kappa g κ be the 1 1 1 -periodic extension of t ↦ κ 1 / 2 g ( κ t ) t\mapsto\kappa^{1/2}g(\kappa t) t ↦ κ 1/2 g ( κ t ) from [ 0 , 1 ) [0,1) [ 0 , 1 ) to R \mathbb{R} R , and set
h κ ( t ) : = ∫ 0 t ( g κ 2 ( τ ) − 1 ) d τ . h_\kappa(t):=\int_0^t\bigl(g_\kappa^2(\tau)-1\bigr)\,\mathrm{d}\tau. h κ ( t ) := ∫ 0 t ( g κ 2 ( τ ) − 1 ) d τ .
Lemma A.4. For κ ≥ 1 \kappa\ge1 κ ≥ 1 and ν ∈ N \nu\in\mathbb{N} ν ∈ N :
(a) g κ g_\kappa g κ is smooth and 1 1 1 -periodic, and for n ≥ 0 n\ge0 n ≥ 0 , 1 ≤ p ≤ ∞ 1\le p\le\infty 1 ≤ p ≤ ∞ ,
∥ ∂ t n ( g κ ( ν ⋅ ) ) ∥ L p ( 0 , 1 ) ≲ n ( ν κ ) n κ 1 2 − 1 p . \|\partial_t^n(g_\kappa(\nu\cdot))\|_{L^p(0,1)}\lesssim_n(\nu\kappa)^n\kappa^{\frac{1}{2}-\frac{1}{p}}. ∥ ∂ t n ( g κ ( ν ⋅ )) ∥ L p ( 0 , 1 ) ≲ n ( ν κ ) n κ 2 1 − p 1 . (b) h κ h_\kappa h κ is smooth, 1 1 1 -periodic, ∥ h κ ∥ L ∞ ≤ 1 \|h_\kappa\|_{L^\infty}\le1 ∥ h κ ∥ L ∞ ≤ 1 , and
∂ t ( ν − 1 h κ ( ν t ) ) = g κ 2 ( ν t ) − 1. (81) \partial_t\bigl(\nu^{-1}h_\kappa(\nu t)\bigr)=g_\kappa^2(\nu t)-1. \tag*{(81)} ∂ t ( ν − 1 h κ ( ν t ) ) = g κ 2 ( ν t ) − 1. (81) (c) (Temporal decorrelation). For every F ∈ C 1 ( [ 0 , 1 ] ) F\in C^1([0,1]) F ∈ C 1 ([ 0 , 1 ]) ,
∣ ∫ 0 1 g κ 2 ( ν t ) F ( t ) d t − ∫ 0 1 F ( t ) d t ∣ ≤ 2 ν − 1 ∥ F ′ ∥ L ∞ ( 0 , 1 ) . \left|\int_0^1g_\kappa^2(\nu t)F(t)\,\mathrm{d}t-\int_0^1F(t)\,\mathrm{d}t\right|\le2\nu^{-1}\|F'\|_{L^\infty(0,1)}. ∫ 0 1 g κ 2 ( ν t ) F ( t ) d t − ∫ 0 1 F ( t ) d t ≤ 2 ν − 1 ∥ F ′ ∥ L ∞ ( 0 , 1 ) .
Proof. (a) κ 1 / 2 g ( κ ⋅ ) \kappa^{1/2}g(\kappa\cdot) κ 1/2 g ( κ ⋅ ) is supported in ( 0 , κ − 1 ) ⊂ ( 0 , 1 ) (0,\kappa^{-1})\subset(0,1) ( 0 , κ − 1 ) ⊂ ( 0 , 1 ) , so its periodic extension is smooth. On [ 0 , 1 ] [0,1] [ 0 , 1 ] the function g κ ( ν ⋅ ) g_{\kappa}(\nu\cdot) g κ ( ν ⋅ ) consists of ν \nu ν translated copies of κ 1 / 2 g ( ν κ ⋅ ) \kappa^{1/2}g(\nu\kappa\cdot) κ 1/2 g ( ν κ ⋅ ) , which gives
∥ ∂ t n ( g κ ( ν ⋅ ) ) ∥ L p ( 0 , 1 ) = ( ν κ ) n + 1 2 − 1 p ∥ g ( n ) ∥ L p . \|\partial_t^n(g_{\kappa}(\nu\cdot))\|_{L^p(0,1)}=(\nu\kappa)^{n+\frac12-\frac1p}\|g^{(n)}\|_{L^p}. ∥ ∂ t n ( g κ ( ν ⋅ )) ∥ L p ( 0 , 1 ) = ( ν κ ) n + 2 1 − p 1 ∥ g ( n ) ∥ L p .
(b) Since ∫ 0 1 ( g κ 2 − 1 ) = 0 \int_0^1(g_{\kappa}^2-1)=0 ∫ 0 1 ( g κ 2 − 1 ) = 0 , h κ h_{\kappa} h κ is 1 1 1 -periodic. For t ∈ [ 0 , 1 ] t\in[0,1] t ∈ [ 0 , 1 ] , h κ ( t ) = ∫ 0 t g κ 2 − t h_{\kappa}(t)=\int_0^t g_{\kappa}^2-t h κ ( t ) = ∫ 0 t g κ 2 − t with both terms in [ 0 , 1 ] [0,1] [ 0 , 1 ] , so ∣ h κ ( t ) ∣ ≤ 1 |h_{\kappa}(t)|\leq1 ∣ h κ ( t ) ∣ ≤ 1 . The identity (A.7) is the chain rule.
(c) Let J j = [ j / ν , ( j + 1 ) / ν ] J_j=[j/\nu,(j+1)/\nu] J j = [ j / ν , ( j + 1 ) / ν ] , 0 ≤ j < ν 0\leq j<\nu 0 ≤ j < ν . By periodicity,
∫ J j ( g κ 2 ( ν t ) − 1 ) d t = ν − 1 ∫ 0 1 ( g κ 2 − 1 ) = 0 , \int_{J_j}(g_{\kappa}^2(\nu t)-1)\,dt=\nu^{-1}\int_0^1(g_{\kappa}^2-1)=0, ∫ J j ( g κ 2 ( ν t ) − 1 ) d t = ν − 1 ∫ 0 1 ( g κ 2 − 1 ) = 0 ,
hence
∣ ∫ J j ( g κ 2 ( ν t ) − 1 ) F ( t ) d t ∣ = ∣ ∫ J j ( g κ 2 ( ν t ) − 1 ) ( F ( t ) − F ( j / ν ) ) d t ∣ ≤ ν − 1 ∥ F ′ ∥ L ∞ ∫ J j ( g κ 2 ( ν t ) + 1 ) d t = 2 ν − 2 ∥ F ′ ∥ L ∞ . \begin{aligned}
\left|\int_{J_j}(g_{\kappa}^2(\nu t)-1)F(t)\,dt\right|=\left|\int_{J_j}(g_{\kappa}^2(\nu t)-1)(F(t)-F(j/\nu))\,dt\right| \\
&\leq\nu^{-1}\|F'\|_{L^\infty}\int_{J_j}(g_{\kappa}^2(\nu t)+1)\,dt=2\nu^{-2}\|F'\|_{L^\infty}.
\end{aligned} ∫ J j ( g κ 2 ( ν t ) − 1 ) F ( t ) d t = ∫ J j ( g κ 2 ( ν t ) − 1 ) ( F ( t ) − F ( j / ν )) d t ≤ ν − 1 ∥ F ′ ∥ L ∞ ∫ J j ( g κ 2 ( ν t ) + 1 ) d t = 2 ν − 2 ∥ F ′ ∥ L ∞ .
Summing over j j j gives the claim.
Part (c) replaces the use of [12 Lemma B.1] in time. That lemma is stated for periodic functions. In [12 ] the stress vanishes near t = 0 t=0 t = 0 and t = 1 t=1 t = 1 , so it applies to t ↦ ∫ ρ d x t\mapsto\int\rho\,dx t ↦ ∫ ρ d x ; here R R R need not vanish near t = 1 t=1 t = 1 , and the corresponding function of time is not periodic.
Cutoffs
Stress cutoff. Let χ ~ ∈ C ∞ ( [ 0 , ∞ ) ) \widetilde{\chi}\in C^\infty([0,\infty)) χ ∈ C ∞ ([ 0 , ∞ )) be nondecreasing with
χ ~ ( r ) = 4 ∥ R ∥ 1 ( 0 ≤ r ≤ ∥ R ∥ 1 ) , χ ~ ( r ) = 4 r ( r ≥ 2 ∥ R ∥ 1 ) . \widetilde{\chi}(r)=4\|R\|_1\quad(0\leq r\leq\|R\|_1),\qquad\widetilde{\chi}(r)=4r\quad(r\geq2\|R\|_1). χ ( r ) = 4∥ R ∥ 1 ( 0 ≤ r ≤ ∥ R ∥ 1 ) , χ ( r ) = 4 r ( r ≥ 2∥ R ∥ 1 ) .
Such a function exists since 4 ∥ R ∥ 1 < 8 ∥ R ∥ 1 = 4 ⋅ 2 ∥ R ∥ 1 4\|R\|_1<8\|R\|_1=4\cdot2\|R\|_1 4∥ R ∥ 1 < 8∥ R ∥ 1 = 4 ⋅ 2∥ R ∥ 1 . Set χ ( A ) : = χ ~ ( ∣ A ∣ ) \chi(A):=\widetilde{\chi}(|A|) χ ( A ) := χ ( ∣ A ∣ ) for symmetric A A A , which is smooth because χ ~ \widetilde{\chi} χ is constant near 0 0 0 , and
ρ : = χ ( R ) ∈ C ∞ ( [ 0 , 1 ] × T d ) . (82) \rho:=\chi(R)\in C^\infty([0,1]\times\mathbb{T}^d). \tag*{(82)} ρ := χ ( R ) ∈ C ∞ ([ 0 , 1 ] × T d ) . (82)
Lemma A.5. On [ 0 , 1 ] × T d [0,1]\times\mathbb{T}^d [ 0 , 1 ] × T d ,
4 ∥ R ∥ 1 ≤ ρ ≤ 8 ∥ R ∥ 1 + 4 ∣ R ∣ , Id − R ρ ∈ N . 4\|R\|_1\leq\rho\leq8\|R\|_1+4|R|,\qquad\operatorname{Id}-\frac{R}{\rho}\in\mathcal{N}. 4∥ R ∥ 1 ≤ ρ ≤ 8∥ R ∥ 1 + 4∣ R ∣ , Id − ρ R ∈ N . In particular ρ 1 / 2 \rho^{1/2} ρ 1/2 and Γ k ( Id − R / ρ ) \Gamma_k(\operatorname{Id}-R/\rho) Γ k ( Id − R / ρ ) are smooth, and ∫ 0 1 ∫ T d ρ d x d t ≤ 12 ∥ R ∥ 1 \int_0^1\int_{\mathbb{T}^d}\rho\,dx\,dt\leq12\|R\|_1 ∫ 0 1 ∫ T d ρ d x d t ≤ 12∥ R ∥ 1 .
Proof. The bounds on ρ \rho ρ follow from the monotonicity of χ ~ \widetilde{\chi} χ . If ∣ R ∣ ≤ 2 ∥ R ∥ 1 |R|\leq2\|R\|_1 ∣ R ∣ ≤ 2∥ R ∥ 1 , then ∣ R ∣ / ρ ≤ 2 ∥ R ∥ 1 / ( 4 ∥ R ∥ 1 ) = 1 2 |R|/\rho\leq2\|R\|_1/(4\|R\|_1)=\frac12 ∣ R ∣/ ρ ≤ 2∥ R ∥ 1 / ( 4∥ R ∥ 1 ) = 2 1 ; if ∣ R ∣ ≥ 2 ∥ R ∥ 1 |R|\geq2\|R\|_1 ∣ R ∣ ≥ 2∥ R ∥ 1 , then ∣ R ∣ / ρ = 1 4 |R|/\rho=\frac14 ∣ R ∣/ ρ = 4 1 . Hence ∣ ( Id − R / ρ ) − Id ∣ ≤ 1 2 |(\operatorname{Id}-R/\rho)-\operatorname{Id}|\leq\frac12 ∣ ( Id − R / ρ ) − Id ∣ ≤ 2 1 , and Id − R / ρ \operatorname{Id}-R/\rho Id − R / ρ is symmetric.
Temporal cutoff. Fix θ ∈ C ∞ ( R ; [ 0 , 1 ] ) \theta\in C^\infty(\mathbb{R};[0,1]) θ ∈ C ∞ ( R ; [ 0 , 1 ]) with
θ = 0 on ( − ∞ , s ] , θ = 1 on [ t ∗ , ∞ ) . (83) \theta=0\text{ on }(-\infty,s],\qquad\theta=1\text{ on }[t_*,\infty). \tag*{(83)} θ = 0 on ( − ∞ , s ] , θ = 1 on [ t ∗ , ∞ ) . (83)
Since R = 0 R=0 R = 0 on [ 0 , t ∗ ] [0,t_*] [ 0 , t ∗ ] ,
θ R = θ 2 R = R on [ 0 , 1 ] × T d . (84) \theta R=\theta^2R=R\qquad\text{on }[0,1]\times\mathbb{T}^d. \tag*{(84)} θ R = θ 2 R = R on [ 0 , 1 ] × T d . (84)
This is the only property of θ \theta θ used in the algebra below. In [12 ] the cutoff (4.12) is two-sided and adapted to the well-prepared set I I I ; we do not use the concentration step, and a one-sided cutoff suffices.
The velocity perturbation
Amplitudes. For k ∈ Ξ k\in\Xi k ∈ Ξ let
a k ( t , x ) : = θ ( t ) g κ ( ν t ) ρ 1 / 2 ( t , x ) Γ k ( Id − R ( t , x ) ρ ( t , x ) ) . (85) a_k(t,x):=\theta(t)g_{\kappa}(\nu t)\rho^{1/2}(t,x)\Gamma_k\left(\operatorname{Id}-\frac{R(t,x)}{\rho(t,x)}\right). \tag*{(85)} a k ( t , x ) := θ ( t ) g κ ( ν t ) ρ 1/2 ( t , x ) Γ k ( Id − ρ ( t , x ) R ( t , x ) ) . (85)
By Lemma A.5, a k ∈ C ∞ ( [ 0 , 1 ] × T d ) a_k\in C^\infty([0,1]\times\mathbb{T}^d) a k ∈ C ∞ ([ 0 , 1 ] × T d ) .
Principal part and divergence corrector.
w ( p ) : = ∑ k ∈ Ξ a k W k ( σ x ) , w ( c ) : = σ − 1 ∑ k ∈ Ξ ∇ a k ⋅ Ω k ( σ x ) . (86) w^{(p)}:=\sum_{k\in\Xi}a_kW_k(\sigma x),\qquad w^{(c)}:=\sigma^{-1}\sum_{k\in\Xi}\nabla a_k\cdot\Omega_k(\sigma x). \tag*{(86)} w ( p ) := k ∈ Ξ ∑ a k W k ( σ x ) , w ( c ) := σ − 1 k ∈ Ξ ∑ ∇ a k ⋅ Ω k ( σ x ) . (86)
where ( ∇ a ⋅ Ω ) i = ∂ j a Ω i j (\nabla a\cdot\Omega)_i=\partial_j a\,\Omega_{ij} ( ∇ a ⋅ Ω ) i = ∂ j a Ω ij . Since div ( Ω k ( σ x ) ) = σ W k ( σ x ) \operatorname{div}(\Omega_k(\sigma x))=\sigma W_k(\sigma x) div ( Ω k ( σ x )) = σ W k ( σ x ) ,
w ( p ) + w ( c ) = σ − 1 div ( ∑ k ∈ Ξ a k Ω k ( σ x ) ) . (87) w^{(p)}+w^{(c)}=\sigma^{-1}\operatorname{div}\left(\sum_{k\in\Xi}a_k\Omega_k(\sigma x)\right). \tag*{(87)} w ( p ) + w ( c ) = σ − 1 div ( k ∈ Ξ ∑ a k Ω k ( σ x ) ) . (87)
Temporal corrector.
w ( t ) : = ν − 1 h κ ( ν t ) ( div R − ∇ Δ − 1 div div R ) , (88) w^{(t)}:=\nu^{-1}h_\kappa(\nu t)\left(\operatorname{div}R-\nabla\Delta^{-1}\operatorname{div}\operatorname{div}R\right), \tag*{(88)} w ( t ) := ν − 1 h κ ( ν t ) ( div R − ∇ Δ − 1 div div R ) , (88)
which is well defined since div div R = ∂ i ∂ j R i j \operatorname{div}\operatorname{div}R=\partial_i\partial_jR_{ij} div div R = ∂ i ∂ j R ij has zero mean. Finally,
w : = w ( p ) + w ( c ) + w ( t ) . (89) w:=w^{(p)}+w^{(c)}+w^{(t)}. \tag*{(89)} w := w ( p ) + w ( c ) + w ( t ) . (89)
Lemma A.6. (a) w ∈ C ∞ ( [ 0 , 1 ] × T d ; R d ) w\in C^\infty([0,1]\times\mathbb{T}^d;\mathbb{R}^d) w ∈ C ∞ ([ 0 , 1 ] × T d ; R d ) , div w = 0 \operatorname{div}w=0 div w = 0 , and ∫ w ( t ) = 0 \int w^{(t)}=0 ∫ w ( t ) = 0 for all t t t .
(b) a k = 0 a_k=0 a k = 0 on [ 0 , s ] × T d [0,s]\times\mathbb{T}^d [ 0 , s ] × T d and w ( t ) = 0 w^{(t)}=0 w ( t ) = 0 on [ 0 , t ∗ ] × T d [0,t_*]\times\mathbb{T}^d [ 0 , t ∗ ] × T d . In particular w = 0 w=0 w = 0 on [ 0 , s ] × T d [0,s]\times\mathbb{T}^d [ 0 , s ] × T d .
(c) On [ 0 , 1 ] × T d [0,1]\times\mathbb{T}^d [ 0 , 1 ] × T d ,
∑ k ∈ Ξ a k 2 ∫ T d W k ⊗ W k d x = θ 2 g k 2 ( ν t ) ρ I d − g k 2 ( ν t ) R . (90) \sum_{k\in\Xi}a_k^2\int_{\mathbb{T}^d}W_k\otimes W_k\,\mathrm{d}x=\theta^2g_k^2(\nu t)\rho\,\mathrm{Id}-g_k^2(\nu t)R. \tag*{(90)} k ∈ Ξ ∑ a k 2 ∫ T d W k ⊗ W k d x = θ 2 g k 2 ( ν t ) ρ Id − g k 2 ( ν t ) R . (90)
Proof. (a) Smoothness is clear. By (A.13), w ( p ) + w ( c ) w^{(p)}+w^{(c)} w ( p ) + w ( c ) is the divergence of a skew-symmetric matrix field, hence divergence-free and of zero mean. For (A.14), div w ( t ) = ν − 1 h κ ( ν t ) ( div div R − Δ Δ − 1 div div R ) = 0 \operatorname{div}w^{(t)}=\nu^{-1}h_\kappa(\nu t)(\operatorname{div}\operatorname{div}R-\Delta\Delta^{-1}\operatorname{div}\operatorname{div}R)=0 div w ( t ) = ν − 1 h κ ( ν t ) ( div div R − Δ Δ − 1 div div R ) = 0 , since div div R \operatorname{div}\operatorname{div}R div div R has zero mean, and w ( t ) w^{(t)} w ( t ) has zero mean as the sum of a divergence and a gradient.
(b) θ = 0 \theta=0 θ = 0 on [ 0 , s ] [0,s] [ 0 , s ] gives a k = 0 a_k=0 a k = 0 , hence w ( p ) = w ( c ) = 0 w^{(p)}=w^{(c)}=0 w ( p ) = w ( c ) = 0 there. R = 0 R=0 R = 0 on [ 0 , t ∗ ] [0,t_*] [ 0 , t ∗ ] gives w ( t ) = 0 w^{(t)}=0 w ( t ) = 0 there.
(c) By Lemma A.3(c), Lemma A.2 and Lemma A.5,
∑ k a k 2 e k ⊗ e k = θ 2 g k 2 ( ν t ) ρ ∑ k Γ k 2 ( I d − R ρ ) e k ⊗ e k = θ 2 g k 2 ( ν t ) ρ I d − θ 2 g k 2 ( ν t ) R , \sum_k a_k^2e_k\otimes e_k=\theta^2g_k^2(\nu t)\rho\sum_k\Gamma_k^2\left(\mathrm{Id}-\frac{R}{\rho}\right)e_k\otimes e_k=\theta^2g_k^2(\nu t)\rho\,\mathrm{Id}-\theta^2g_k^2(\nu t)R, k ∑ a k 2 e k ⊗ e k = θ 2 g k 2 ( ν t ) ρ k ∑ Γ k 2 ( Id − ρ R ) e k ⊗ e k = θ 2 g k 2 ( ν t ) ρ Id − θ 2 g k 2 ( ν t ) R ,
and θ 2 R = R \theta^2R=R θ 2 R = R by (A.10).
The new stress
Oscillation. Let
T k : = W k ⊗ W k − ∫ W k ⊗ W k , T_k:=W_k\otimes W_k-\int W_k\otimes W_k, T k := W k ⊗ W k − ∫ W k ⊗ W k ,
which is smooth, symmetric, of zero mean, and satisfies div T k = 0 \operatorname{div}T_k=0 div T k = 0 by Lemma A.3(a). Expanding the square and separating the diagonal terms,
w ( p ) ⊗ w ( p ) = ∑ k a k 2 T k ( σ x ) + ∑ k a k 2 ∫ W k ⊗ W k + R f a r f u l l , R f a r f u l l : = ∑ k ≠ k ′ a k a k ′ W k ( σ x ) ⊗ W k ′ ( σ x ) . (91) w^{(p)}\otimes w^{(p)}=\sum_k a_k^2T_k(\sigma x)+\sum_k a_k^2\int W_k\otimes W_k+R_{\mathrm{far}}^{\mathrm{full}},\qquad R_{\mathrm{far}}^{\mathrm{full}}:=\sum_{k\ne k'}a_ka_{k'}W_k(\sigma x)\otimes W_{k'}(\sigma x). \tag*{(91)} w ( p ) ⊗ w ( p ) = k ∑ a k 2 T k ( σ x ) + k ∑ a k 2 ∫ W k ⊗ W k + R far full , R far full := k = k ′ ∑ a k a k ′ W k ( σ x ) ⊗ W k ′ ( σ x ) . (91)
The sum in R f a r f u l l R_{\mathrm{far}}^{\mathrm{full}} R far full runs over ordered pairs, so R f a r f u l l R_{\mathrm{far}}^{\mathrm{full}} R far full is symmetric; it is not traceless. Let
R f a r : = R f a r f u l l − 1 d ( tr R f a r f u l l ) I d ∈ S 0 d × d . R_{\mathrm{far}}:=R_{\mathrm{far}}^{\mathrm{full}}-\frac{1}{d}\left(\operatorname{tr}R_{\mathrm{far}}^{\mathrm{full}}\right)\mathrm{Id}\in\mathcal{S}_0^{d\times d}. R far := R far full − d 1 ( tr R far full ) Id ∈ S 0 d × d .
Since div ( T k ( σ ⋅ ) ) = σ ( div T k ) ( σ ⋅ ) = 0 \operatorname{div}(T_k(\sigma\cdot))=\sigma(\operatorname{div}T_k)(\sigma\cdot)=0 div ( T k ( σ ⋅ )) = σ ( div T k ) ( σ ⋅ ) = 0 , we have
div ( ∑ k a k 2 T k ( σ x ) ) = ∑ k ∇ ( a k 2 ) ⋅ T k ( σ x ) , \operatorname{div}\left(\sum_k a_k^2T_k(\sigma x)\right)=\sum_k\nabla(a_k^2)\cdot T_k(\sigma x), div ( k ∑ a k 2 T k ( σ x ) ) = k ∑ ∇ ( a k 2 ) ⋅ T k ( σ x ) ,
which has zero mean, being a divergence. Define
R o s c , x : = ∑ k ∈ Ξ B ( ∇ ( a k 2 ) , T k ( σ ⋅ ) ) , R o s c , t : = ν − 1 h κ ( ν t ) ∂ t R . (92) R_{\mathrm{osc},x}:=\sum_{k\in\Xi}\mathcal{B}(\nabla(a_k^2),T_k(\sigma\cdot)),\qquad R_{\mathrm{osc},t}:=\nu^{-1}h_\kappa(\nu t)\partial_tR. \tag*{(92)} R osc , x := k ∈ Ξ ∑ B ( ∇ ( a k 2 ) , T k ( σ ⋅ )) , R osc , t := ν − 1 h κ ( ν t ) ∂ t R . (92)
By Lemma A.1(c), R o s c , x R_{\mathrm{osc},x} R osc , x is S 0 d × d \mathcal{S}_0^{d\times d} S 0 d × d -valued and div R o s c , x = ∑ k ∇ ( a k 2 ) ⋅ T k ( σ x ) \operatorname{div}R_{\mathrm{osc},x}=\sum_k\nabla(a_k^2)\cdot T_k(\sigma x) div R osc , x = ∑ k ∇ ( a k 2 ) ⋅ T k ( σ x ) .
Lemma A.7 (Oscillation identity). Let R o s c : = R o s c , x + R o s c , t + R f a r R_{\mathrm{osc}} := R_{\mathrm{osc},x} + R_{\mathrm{osc},t} + R_{\mathrm{far}} R osc := R osc , x + R osc , t + R far and
P : = − θ 2 g κ 2 ( ν t ) ρ − 1 d tr R f a r f u l l + ( g κ 2 ( ν t ) − 1 ) Δ − 1 div div R + ν − 1 h κ ( ν t ) Δ − 1 div div ∂ t R . (93) P := -\theta^2 g_\kappa^2(\nu t)\rho- \frac{1}{d}\operatorname{tr} R_{\mathrm{far}}^{\mathrm{full}} + \left(g_\kappa^2(\nu t)-1\right)\Delta^{-1}\operatorname{div}\operatorname{div} R + \nu^{-1}h_\kappa(\nu t)\Delta^{-1}\operatorname{div}\operatorname{div}\partial_t R. \tag*{(93)} P := − θ 2 g κ 2 ( ν t ) ρ − d 1 tr R far full + ( g κ 2 ( ν t ) − 1 ) Δ − 1 div div R + ν − 1 h κ ( ν t ) Δ − 1 div div ∂ t R . (93) Then
∂ t w ( t ) + div ( w ( p ) ⊗ w ( p ) + R ) + ∇ P = div R o s c . (94) \partial_t w^{(t)}+\operatorname{div}\left(w^{(p)}\otimes w^{(p)}+R\right)+\nabla P=\operatorname{div}R_{\mathrm{osc}}. \tag*{(94)} ∂ t w ( t ) + div ( w ( p ) ⊗ w ( p ) + R ) + ∇ P = div R osc . (94) Proof. By (A.17) and (A.16),
div ( w ( p ) ⊗ w ( p ) + R ) = div R o s c , x + div R f a r + 1 d ∇ tr R f a r f u l l + ∇ ( θ 2 g κ 2 ( ν t ) ρ ) + ( 1 − g κ 2 ( ν t ) ) div R , \operatorname{div}\left(w^{(p)}\otimes w^{(p)}+R\right)=\operatorname{div}R_{\mathrm{osc},x}+\operatorname{div}R_{\mathrm{far}}+\frac{1}{d}\nabla\operatorname{tr}R_{\mathrm{far}}^{\mathrm{full}}+\nabla\left(\theta^2g_\kappa^2(\nu t)\rho\right)+\left(1-g_\kappa^2(\nu t)\right)\operatorname{div}R, div ( w ( p ) ⊗ w ( p ) + R ) = div R osc , x + div R far + d 1 ∇ tr R far full + ∇ ( θ 2 g κ 2 ( ν t ) ρ ) + ( 1 − g κ 2 ( ν t ) ) div R , where we used that g κ ( ν t ) g_\kappa(\nu t) g κ ( ν t ) does not depend on x x x . By (A.7),
∂ t w ( t ) = ( g κ 2 ( ν t ) − 1 ) ( div R − ∇ Δ − 1 div div R ) + ν − 1 h κ ( ν t ) ( div ∂ t R − ∇ Δ − 1 div div ∂ t R ) . \partial_t w^{(t)}=\left(g_\kappa^2(\nu t)-1\right)\left(\operatorname{div}R-\nabla\Delta^{-1}\operatorname{div}\operatorname{div}R\right)+\nu^{-1}h_\kappa(\nu t)\left(\operatorname{div}\partial_tR-\nabla\Delta^{-1}\operatorname{div}\operatorname{div}\partial_tR\right). ∂ t w ( t ) = ( g κ 2 ( ν t ) − 1 ) ( div R − ∇ Δ − 1 div div R ) + ν − 1 h κ ( ν t ) ( div ∂ t R − ∇ Δ − 1 div div ∂ t R ) . The terms ± ( g κ 2 ( ν t ) − 1 ) div R \pm\left(g_\kappa^2(\nu t)-1\right)\operatorname{div}R ± ( g κ 2 ( ν t ) − 1 ) div R cancel, and ν − 1 h κ ( ν t ) div ∂ t R = div R o s c , t \nu^{-1}h_\kappa(\nu t)\operatorname{div}\partial_tR=\operatorname{div}R_{\mathrm{osc},t} ν − 1 h κ ( ν t ) div ∂ t R = div R osc , t . All remaining terms are gradients, and they are cancelled by ∇ P \nabla P ∇ P .
Remark A.8. By (A.7), ν − 1 ∂ t ( R h κ ( ν t ) ) = ( g κ 2 ( ν t ) − 1 ) R + ν − 1 h κ ( ν t ) ∂ t R \nu^{-1}\partial_t\left(Rh_\kappa(\nu t)\right)=\left(g_\kappa^2(\nu t)-1\right)R+\nu^{-1}h_\kappa(\nu t)\partial_tR ν − 1 ∂ t ( R h κ ( ν t ) ) = ( g κ 2 ( ν t ) − 1 ) R + ν − 1 h κ ( ν t ) ∂ t R , so
P = − θ 2 g κ 2 ( ν t ) ρ − 1 d tr R f a r f u l l + ν − 1 Δ − 1 div div ∂ t ( R h κ ( ν t ) ) . P=-\theta^2g_\kappa^2(\nu t)\rho-\frac{1}{d}\operatorname{tr}R_{\mathrm{far}}^{\mathrm{full}}+\nu^{-1}\Delta^{-1}\operatorname{div}\operatorname{div}\partial_t\left(Rh_\kappa(\nu t)\right). P = − θ 2 g κ 2 ( ν t ) ρ − d 1 tr R far full + ν − 1 Δ − 1 div div ∂ t ( R h κ ( ν t ) ) .
Thus (A.19) differs from [12 (4.25)] only by the trace part of the far-field interaction and by the sign with which the last term appears. These terms do not affect any estimate of the stress. Including the trace in the pressure makes the new stress traceless, as required by the definition of the Reynolds system.
Linear and correction errors. As in [12 (4.26)–(4.27)] , define
R l i n : = R ( ∂ t ( w ( p ) + w ( c ) ) − Δ w + div ( u ⊗ w + w ⊗ u ) ) , (95) R_{\mathrm{lin}}:=\mathcal{R}\left(\partial_t\left(w^{(p)}+w^{(c)}\right)-\Delta w+\operatorname{div}(u\otimes w+w\otimes u)\right), \tag*{(95)} R lin := R ( ∂ t ( w ( p ) + w ( c ) ) − Δ w + div ( u ⊗ w + w ⊗ u ) ) , (95)
R c o r : = R div ( ( w ( c ) + w ( t ) ) ⊗ w + w ( p ) ⊗ ( w ( c ) + w ( t ) ) ) . (96) R_{\mathrm{cor}}:=\mathcal{R}\operatorname{div}\left(\left(w^{(c)}+w^{(t)}\right)\otimes w+w^{(p)}\otimes\left(w^{(c)}+w^{(t)}\right)\right). \tag*{(96)} R cor := R div ( ( w ( c ) + w ( t ) ) ⊗ w + w ( p ) ⊗ ( w ( c ) + w ( t ) ) ) . (96)
The arguments of R \mathcal{R} R have zero mean: ∂ t ( w ( p ) + w ( c ) ) \partial_t\left(w^{(p)}+w^{(c)}\right) ∂ t ( w ( p ) + w ( c ) ) by (A.13), Δ w \Delta w Δ w as a Laplacian, and the other terms as divergences. Hence
div R l i n = ∂ t ( w ( p ) + w ( c ) ) − Δ w + div ( u ⊗ w + w ⊗ u ) , \operatorname{div}R_{\mathrm{lin}}=\partial_t\left(w^{(p)}+w^{(c)}\right)-\Delta w+\operatorname{div}(u\otimes w+w\otimes u), div R lin = ∂ t ( w ( p ) + w ( c ) ) − Δ w + div ( u ⊗ w + w ⊗ u ) ,
div R c o r = div ( ( w ( c ) + w ( t ) ) ⊗ w + w ( p ) ⊗ ( w ( c ) + w ( t ) ) ) . (97) \operatorname{div}R_{\mathrm{cor}}=\operatorname{div}\left(\left(w^{(c)}+w^{(t)}\right)\otimes w+w^{(p)}\otimes\left(w^{(c)}+w^{(t)}\right)\right). \tag*{(97)} div R cor = div ( ( w ( c ) + w ( t ) ) ⊗ w + w ( p ) ⊗ ( w ( c ) + w ( t ) ) ) . (97)
Proposition A.9 (The algebraic part of Proposition 2.8). Let w w w be given by (A.15), P P P by (A.19), and
R C L : = R l i n + R c o r + R o s c = R l i n + R c o r + R o s c , x + R o s c , t + R f a r . (98) R^{\mathrm{CL}}:=R_{\mathrm{lin}}+R_{\mathrm{cor}}+R_{\mathrm{osc}}=R_{\mathrm{lin}}+R_{\mathrm{cor}}+R_{\mathrm{osc},x}+R_{\mathrm{osc},t}+R_{\mathrm{far}}. \tag*{(98)} R CL := R lin + R cor + R osc = R lin + R cor + R osc , x + R osc , t + R far . (98) Then w w w , P P P and R C L R^{\mathrm{CL}} R CL are smooth on [ 0 , 1 ] × T d [0,1]\times\mathbb{T}^d [ 0 , 1 ] × T d , div w = 0 \operatorname{div}w=0 div w = 0 , R C L R^{\mathrm{CL}} R CL takes values in S 0 d × d \mathcal{S}_0^{d\times d} S 0 d × d , and properties (i) and (ii) of Proposition 2.8 hold:
(i) ∂ t w − Δ w + div ( u ⊗ w + w ⊗ u + w ⊗ w ) + div R + ∇ P = div R C L \partial_t w-\Delta w+\operatorname{div}(u\otimes w+w\otimes u+w\otimes w)+\operatorname{div}R+\nabla P=\operatorname{div}R^{\mathrm{CL}} ∂ t w − Δ w + div ( u ⊗ w + w ⊗ u + w ⊗ w ) + div R + ∇ P = div R CL ;
(ii) ∫ w ( t ) = 0 \int w(t)=0 ∫ w ( t ) = 0 for all t t t , and w = 0 w=0 w = 0 , R C L = 0 R^{\mathrm{CL}}=0 R CL = 0 on [ 0 , s ] × T d [0,s]\times\mathbb{T}^d [ 0 , s ] × T d .
This holds for every ν , σ ∈ N \nu,\sigma\in\mathbb{N} ν , σ ∈ N , κ ≥ 1 \kappa\ge1 κ ≥ 1 and μ ≥ μ 0 \mu\ge\mu_0 μ ≥ μ 0 .
Proof. All ingredients are smooth in ( t , x ) (t,x) ( t , x ) . R l i n R_{\mathrm{lin}} R lin , R c o r R_{\mathrm{cor}} R cor are S 0 d × d \mathcal{S}_0^{d\times d} S 0 d × d -valued by Lemma A.1(a), R o s c , x R_{\mathrm{osc},x} R osc , x by Lemma A.1(c), R o s c , t R_{\mathrm{osc},t} R osc , t because R R R is, and R f a r R_{\mathrm{far}} R far by construction.
For the identity (i), since w = w ( p ) + ( w ( c ) + w ( t ) ) w=w^{(p)}+\left(w^{(c)}+w^{(t)}\right) w = w ( p ) + ( w ( c ) + w ( t ) ) ,
w ⊗ w = w ( p ) ⊗ w ( p ) + ( w ( c ) + w ( t ) ) ⊗ w + w ( p ) ⊗ ( w ( c ) + w ( t ) ) . w\otimes w=w^{(p)}\otimes w^{(p)}+\left(w^{(c)}+w^{(t)}\right)\otimes w+w^{(p)}\otimes\left(w^{(c)}+w^{(t)}\right). w ⊗ w = w ( p ) ⊗ w ( p ) + ( w ( c ) + w ( t ) ) ⊗ w + w ( p ) ⊗ ( w ( c ) + w ( t ) ) .
Hence, using ∂ t w = ∂ t ( w ( p ) + w ( c ) ) + ∂ t w ( t ) \partial_t w=\partial_t\left(w^{(p)}+w^{(c)}\right)+\partial_t w^{(t)} ∂ t w = ∂ t ( w ( p ) + w ( c ) ) + ∂ t w ( t ) ,
∂ t w − Δ w + div ( u ⊗ w + w ⊗ u + w ⊗ w ) + div R + ∇ P = [ ∂ t ( w ( p ) + w ( c ) ) − Δ w + div ( u ⊗ w + w ⊗ u ) ] + div [ ( w ( c ) + w ( t ) ) ⊗ w + w ( p ) ⊗ ( w ( c ) + w ( t ) ) ] + [ ∂ t w ( t ) + div ( w ( p ) ⊗ w ( p ) + R ) + ∇ P ] = div R l i n + div R c o r + div R o s c , \begin{aligned}
\partial_t w-\Delta w+\operatorname{div}(u\otimes w+w\otimes u+w\otimes w)+\operatorname{div}R+\nabla P
={}&\left[\partial_t\left(w^{(p)}+w^{(c)}\right)-\Delta w+\operatorname{div}(u\otimes w+w\otimes u)\right]\\
&+\operatorname{div}\left[\left(w^{(c)}+w^{(t)}\right)\otimes w+w^{(p)}\otimes\left(w^{(c)}+w^{(t)}\right)\right]\\
&+\left[\partial_t w^{(t)}+\operatorname{div}\left(w^{(p)}\otimes w^{(p)}+R\right)+\nabla P\right]\\
={}&\operatorname{div}R_{\mathrm{lin}}+\operatorname{div}R_{\mathrm{cor}}+\operatorname{div}R_{\mathrm{osc}},
\end{aligned} ∂ t w − Δ w + div ( u ⊗ w + w ⊗ u + w ⊗ w ) + div R + ∇ P = = [ ∂ t ( w ( p ) + w ( c ) ) − Δ w + div ( u ⊗ w + w ⊗ u ) ] + div [ ( w ( c ) + w ( t ) ) ⊗ w + w ( p ) ⊗ ( w ( c ) + w ( t ) ) ] + [ ∂ t w ( t ) + div ( w ( p ) ⊗ w ( p ) + R ) + ∇ P ] div R lin + div R cor + div R osc ,
by (A.23) and Lemma A.7.
For (ii), the mean and the vanishing of w w w on [ 0 , s ] [0,s] [ 0 , s ] are Lemma A.6(a),(b). Since w w w and the a k a_k a k vanish identically on [ 0 , s ] × T d [0,s] \times\mathbb{T}^d [ 0 , s ] × T d and are smooth, all their derivatives vanish there as well (one-sided at t = s t=s t = s ). Hence the arguments of R \mathcal{R} R in (A.21)–(A.22) vanish on [ 0 , s ] [0,s] [ 0 , s ] , and so do R o s c , x \mathcal{R}_{\mathrm{osc},x} R osc , x (which is bilinear in ∇ ( a k 2 ) \nabla(a_k^2) ∇ ( a k 2 ) and acts in space only) and R f a r \mathcal{R}_{\mathrm{far}} R far (which contains a k a k ′ a_k a_{k'} a k a k ′ ). Finally, R = 0 \mathcal{R}=0 R = 0 on [ 0 , t ∗ ] [0,t_*] [ 0 , t ∗ ] gives ∂ t R = 0 \partial_t\mathcal{R}=0 ∂ t R = 0 there, so R o s c , t = 0 \mathcal{R}_{\mathrm{osc},t}=0 R osc , t = 0 on [ 0 , s ] [0,s] [ 0 , s ] .
Remark A.10. Proposition A.9 contains no smallness. In Appendix B the parameters are fixed as powers of a single frequency λ \lambda λ , and the bounds (iii)–(v) of Proposition 2.8 are proved for λ \lambda λ large.
Estimates and proof of Proposition 2.8
In this appendix we fix the parameters of Appendix A and prove the bounds (iii)–(v) of Proposition 2.8 for the perturbation and stress of Proposition A.9. We follow [12 Section 5] . The standing assumptions of Appendix A are in force; in particular ∥ R ∥ 1 > 0 \|\mathcal{R}\|_1>0 ∥ R ∥ 1 > 0 .
Constants. We write A ≲ B A\lesssim B A ≲ B if A ≤ C B A\le CB A ≤ C B with C C C depending only on d d d , p p p , q q q and the fixed choices Ξ \Xi Ξ , Γ k \Gamma_k Γ k , p k p_k p k , φ \varphi φ , g g g , which depend only on d d d . We write C u C_u C u for constants that may in addition depend on u u u , R \mathcal{R} R , s s s , t ∗ t_* t ∗ and θ \theta θ . Neither kind of constant depends on λ \lambda λ , and both may change from line to line. On [ 0 , 1 ] [0,1] [ 0 , 1 ] and on T d \mathbb{T}^d T d , which have unit measure, L p L^p L p norms are nondecreasing in p p p ; we use this without comment.
Reduction. Since ∥ ⋅ ∥ W 1 , q ≤ ∥ ⋅ ∥ W 1 , q ′ \|\,\cdot\,\|_{W^{1,q}}\le\|\,\cdot\,\|_{W^{1,q'}} ∥ ⋅ ∥ W 1 , q ≤ ∥ ⋅ ∥ W 1 , q ′ for q ≤ q ′ q\le q' q ≤ q ′ , it suffices to prove Proposition 2.8 with q q q replaced by max { q , 2 } \max\{q,2\} max { q , 2 } . We therefore assume 2 ≤ q < ∞ 2\le q<\infty 2 ≤ q < ∞ .
Choice of parameters
Fix m ∈ N m\in\mathbb{N} m ∈ N so large that γ : = 1 / m \gamma:=1/m γ := 1/ m satisfies
10 d γ ≤ min { 1 p − 1 2 , 1 q } . (99) 10d\gamma\le\min\left\{\frac{1}{p}-\frac{1}{2},\frac{1}{q}\right\}. \tag*{(99)} 10 d γ ≤ min { p 1 − 2 1 , q 1 } . (99)
and set
r : = d d − γ , r:=\frac{d}{d-\gamma}, r := d − γ d ,
so that
d − d r = γ , 1 < r < d d − 1 ≤ 2 ≤ q . (100) d-\frac{d}{r}=\gamma,\qquad1<r<\frac{d}{d-1}\le2\le q. \tag*{(100)} d − r d = γ , 1 < r < d − 1 d ≤ 2 ≤ q . (100)
Both γ \gamma γ and r r r depend only on d , p , q d,p,q d , p , q . Since q ≥ 2 q\ge2 q ≥ 2 , (B.1) gives γ ≤ 1 20 d \gamma\le\frac{1}{20d} γ ≤ 20 d 1 . For N ∈ N N\in\mathbb{N} N ∈ N let λ : = N m \lambda:=N^m λ := N m and
ν : = λ γ = N , σ : = λ 1 / γ = N m 2 , μ : = λ , κ : = λ K , K : = 2 γ + d + 1 − 6 γ . (101) \nu:=\lambda^\gamma=N,\qquad\sigma:=\lambda^{1/\gamma}=N^{m^2},\qquad\mu:=\lambda,\qquad\kappa:=\lambda^K,\qquad K:=\frac{2}{\gamma}+d+1-6\gamma. \tag*{(101)} ν := λ γ = N , σ := λ 1/ γ = N m 2 , μ := λ , κ := λ K , K := γ 2 + d + 1 − 6 γ . (101)
Then ν , σ , μ ∈ N \nu,\sigma,\mu\in\mathbb{N} ν , σ , μ ∈ N , κ ≥ 1 \kappa\ge1 κ ≥ 1 , and, since K / 2 − 1 / γ = d + 1 2 − 3 γ > 0 K/2-1/\gamma=\frac{d+1}{2}-3\gamma>0 K /2 − 1/ γ = 2 d + 1 − 3 γ > 0 ,
ν ≪ μ ≪ σ ≪ κ 1 / 2 . \nu\ll\mu\ll\sigma\ll\kappa^{1/2}. ν ≪ μ ≪ σ ≪ κ 1/2 .
In the end N N N , and hence λ \lambda λ , is chosen large depending on u u u , R \mathcal{R} R , s s s , t ∗ t_* t ∗ , δ \delta δ . In particular we always assume μ = λ ≥ μ 0 \mu=\lambda\ge\mu_0 μ = λ ≥ μ 0 .
Lemma B.1 (Scaling inequalities). For every λ ≥ 1 \lambda\ge1 λ ≥ 1 :
ν κ 1 / 2 σ − 1 μ − 1 + d − 1 2 − d − 1 r ≤ λ − γ , (102) \nu\kappa^{1/2}\sigma^{-1}\mu^{-1+\frac{d-1}{2}-\frac{d-1}{r}}\le\lambda^{-\gamma}, \tag*{(102)} ν κ 1/2 σ − 1 μ − 1 + 2 d − 1 − r d − 1 ≤ λ − γ , (102) κ 1 2 − 1 p μ d − 1 2 ≤ λ − γ , (103) \kappa^{\frac{1}{2}-\frac{1}{p}}\mu^{\frac{d-1}{2}}\le\lambda^{-\gamma}, \tag*{(103)} κ 2 1 − p 1 μ 2 d − 1 ≤ λ − γ , (103) κ − 1 2 σ μ 1 + d − 1 2 − d − 1 q ≤ λ − γ , (104) \kappa^{-\frac{1}{2}}\sigma\mu^{1+\frac{d-1}{2}-\frac{d-1}{q}}\le\lambda^{-\gamma}, \tag*{(104)} κ − 2 1 σ μ 1 + 2 d − 1 − q d − 1 ≤ λ − γ , (104) σ − 1 μ − 1 + d − 1 2 ≤ λ − γ , σ − 1 μ ( d − 1 ) ( 1 − 1 r ) ≤ λ − γ , μ d − 1 − d r ≤ λ − γ , (105) \sigma^{-1}\mu^{-1+\frac{d-1}{2}}\le\lambda^{-\gamma},\qquad
\sigma^{-1}\mu^{(d-1)(1-\frac{1}{r})}\le\lambda^{-\gamma},\qquad
\mu^{d-1-\frac{d}{r}}\le\lambda^{-\gamma}, \tag*{(105)} σ − 1 μ − 1 + 2 d − 1 ≤ λ − γ , σ − 1 μ ( d − 1 ) ( 1 − r 1 ) ≤ λ − γ , μ d − 1 − r d ≤ λ − γ , (105) ν − 1 = λ − γ , σ − 1 / 2 ≤ λ − γ , ν − 1 / 2 = λ − γ / 2 . (106) \nu^{-1}=\lambda^{-\gamma},\qquad
\sigma^{-1/2}\le\lambda^{-\gamma},\qquad
\nu^{-1/2}=\lambda^{-\gamma/2}. \tag*{(106)} ν − 1 = λ − γ , σ − 1/2 ≤ λ − γ , ν − 1/2 = λ − γ /2 . (106) Proof. We compare exponents of λ \lambda λ , using (B.1), (B.2) and γ ≤ 1 20 d \gamma\le\frac{1}{20d} γ ≤ 20 d 1 . (B.4): the exponent is γ + ( 1 γ + d + 1 2 − 3 γ ) − 1 γ − 1 + d − 1 2 − d − 1 r = − 2 γ + ( d − 1 ) ( 1 − 1 r ) \gamma+\left(\frac{1}{\gamma}+\frac{d+1}{2}-3\gamma\right)-\frac{1}{\gamma}-1+\frac{d-1}{2}-\frac{d-1}{r}=-2\gamma+(d-1)\left(1-\frac{1}{r}\right) γ + ( γ 1 + 2 d + 1 − 3 γ ) − γ 1 − 1 + 2 d − 1 − r d − 1 = − 2 γ + ( d − 1 ) ( 1 − r 1 ) , and ( d − 1 ) ( 1 − 1 r ) ≤ d − d r = γ (d-1)\left(1-\frac{1}{r}\right)\le d-\frac{d}{r}=\gamma ( d − 1 ) ( 1 − r 1 ) ≤ d − r d = γ .
(B.5): since 1 2 − 1 p ≤ − 10 d γ < 0 \frac{1}{2}-\frac{1}{p}\leq-10d\gamma<0 2 1 − p 1 ≤ − 10 d γ < 0 and K ≥ 2 γ K\geq\frac{2}{\gamma} K ≥ γ 2 , the exponent is at most − 10 d γ ⋅ 2 γ + d − 1 2 = − 20 d + d − 1 2 ≤ − γ -10d\gamma\cdot\frac{2}{\gamma}+\frac{d-1}{2}=-20d+\frac{d-1}{2}\leq-\gamma − 10 d γ ⋅ γ 2 + 2 d − 1 = − 20 d + 2 d − 1 ≤ − γ .
(B.6): the exponent is − 1 γ − d + 1 2 + 3 γ + 1 γ + 1 + d − 1 q − d − 1 q = 3 γ − d − 1 q -\frac{1}{\gamma}-\frac{d+1}{2}+3\gamma+\frac{1}{\gamma}+1+\frac{d-1}{q}-\frac{d-1}{q}=3\gamma-\frac{d-1}{q} − γ 1 − 2 d + 1 + 3 γ + γ 1 + 1 + q d − 1 − q d − 1 = 3 γ − q d − 1 , and d − 1 q ≥ 10 d ( d − 1 ) γ ≥ 20 γ \frac{d-1}{q}\geq10d(d-1)\gamma\geq20\gamma q d − 1 ≥ 10 d ( d − 1 ) γ ≥ 20 γ .
(B.7): the exponents are − 1 γ − 1 + d − 1 2 ≤ − 20 d + d − 1 2 ≤ − γ -\frac{1}{\gamma}-1+\frac{d-1}{2}\leq-20d+\frac{d-1}{2}\leq-\gamma − γ 1 − 1 + 2 d − 1 ≤ − 20 d + 2 d − 1 ≤ − γ ; then − 1 γ + ( d − 1 ) ( 1 − 1 r ) ≤ − 1 γ + γ ≤ − γ -\frac{1}{\gamma}+(d-1)(1-\frac{1}{r})\leq-\frac{1}{\gamma}+\gamma\leq-\gamma − γ 1 + ( d − 1 ) ( 1 − r 1 ) ≤ − γ 1 + γ ≤ − γ ; and d − 1 − d r = γ − 1 ≤ − γ d-1-\frac{d}{r}=\gamma-1\leq-\gamma d − 1 − r d = γ − 1 ≤ − γ .
(B.8) is immediate from 1 2 γ ≥ γ \frac{1}{2\gamma}\geq\gamma 2 γ 1 ≥ γ .
Amplitudes
Write a k = θ ( t ) g κ ( ν t ) b k ( t , x ) a_k=\theta(t)g_\kappa(\nu t)b_k(t,x) a k = θ ( t ) g κ ( ν t ) b k ( t , x ) with b k : = ρ 1 / 2 Γ k ( Id − R / ρ ) b_k:=\rho^{1/2}\Gamma_k(\operatorname{Id}-R/\rho) b k := ρ 1/2 Γ k ( Id − R / ρ ) . By Lemma A.5, b k b_k b k is smooth on [ 0 , 1 ] × T d [0,1]\times\mathbb{T}^d [ 0 , 1 ] × T d , and it does not depend on λ \lambda λ . Hence ∥ ∂ t i ∇ m ( θ b k ) ∥ L t , x ∞ ≤ C u , m , n \|\partial_t^i\nabla^m(\theta b_k)\|_{L^\infty_{t,x}}\leq C_{u,m,n} ∥ ∂ t i ∇ m ( θ b k ) ∥ L t , x ∞ ≤ C u , m , n .
Lemma B.2 (Amplitudes). For k ∈ Ξ k\in\Xi k ∈ Ξ , m , n ≥ 0 m,n\geq0 m , n ≥ 0 and 1 ≤ p ′ ≤ ∞ 1\leq p'\leq\infty 1 ≤ p ′ ≤ ∞ ,
∥ ∂ t i ∇ m a k ∥ L p ′ ( 0 , 1 ; L ∞ ( T d ) ) ≤ C u , m , n ( ν κ ) n ( 1 2 − 1 p ′ ) , ∥ ∇ m ( a k 2 ) ∥ L 1 ( 0 , 1 ; L ∞ ( T d ) ) ≤ C u , m . (107) \|\partial_t^i\nabla^m a_k\|_{L^{p'}(0,1;L^\infty(\mathbb{T}^d))}\leq C_{u,m,n}(\nu\kappa)^{n\left(\frac{1}{2}-\frac{1}{p'}\right)},\qquad\|\nabla^m(a_k^2)\|_{L^1(0,1;L^\infty(\mathbb{T}^d))}\leq C_{u,m}. \tag*{(107)} ∥ ∂ t i ∇ m a k ∥ L p ′ ( 0 , 1 ; L ∞ ( T d )) ≤ C u , m , n ( ν κ ) n ( 2 1 − p ′ 1 ) , ∥ ∇ m ( a k 2 ) ∥ L 1 ( 0 , 1 ; L ∞ ( T d )) ≤ C u , m . (107)
Moreover, for every t ∈ [ 0 , 1 ] t\in[0,1] t ∈ [ 0 , 1 ] ,
∥ a k ( t ) ∥ L 2 ( T d ) 2 ≤ C Γ 2 θ 2 ( t ) g κ 2 ( ν t ) ∫ T d ρ ( t , x ) d x , C Γ : = max k ∈ Ξ sup N ∣ Γ k ∣ . (108) \|a_k(t)\|_{L^2(\mathbb{T}^d)}^2\leq C_\Gamma^2\theta^2(t)g_\kappa^2(\nu t)\int_{\mathbb{T}^d}\rho(t,x)\,dx,\qquad C_\Gamma:=\max_{k\in\Xi}\sup_{\mathcal{N}}|\Gamma_k|. \tag*{(108)} ∥ a k ( t ) ∥ L 2 ( T d ) 2 ≤ C Γ 2 θ 2 ( t ) g κ 2 ( ν t ) ∫ T d ρ ( t , x ) d x , C Γ := k ∈ Ξ max N sup ∣ Γ k ∣. (108)
Proof. By the Leibniz rule, ∂ t i ∇ m a k = ∑ i ≤ n ( n i ) ∂ t i ( g κ ( ν t ) ) ∂ t n − i ∇ m ( θ b k ) \partial_t^i\nabla^m a_k=\sum_{i\leq n}\binom{n}{i}\partial_t^i(g_\kappa(\nu t))\partial_t^{n-i}\nabla^m(\theta b_k) ∂ t i ∇ m a k = ∑ i ≤ n ( i n ) ∂ t i ( g κ ( ν t )) ∂ t n − i ∇ m ( θ b k ) . By Lemma A.4(a) and ν κ ≥ 1 \nu\kappa\geq1 ν κ ≥ 1 ,
∥ ∂ t i ( g κ ( ν ⋅ ) ) ∥ L p ′ ( 0 , 1 ) ≲ ( ν κ ) n ( 1 2 − 1 p ′ ) \|\partial_t^i(g_\kappa(\nu\cdot))\|_{L^{p'}(0,1)}\lesssim(\nu\kappa)^{n\left(\frac{1}{2}-\frac{1}{p'}\right)} ∥ ∂ t i ( g κ ( ν ⋅ )) ∥ L p ′ ( 0 , 1 ) ≲ ( ν κ ) n ( 2 1 − p ′ 1 )
for i ≤ n i\leq n i ≤ n , which gives the first bound. For the second, a k 2 = g κ 2 ( ν t ) θ 2 b k 2 a_k^2=g_\kappa^2(\nu t)\theta^2b_k^2 a k 2 = g κ 2 ( ν t ) θ 2 b k 2 and ∥ g κ 2 ( ν ⋅ ) ∥ L 1 ( 0 , 1 ) = 1 \|g_\kappa^2(\nu\cdot)\|_{L^1(0,1)}=1 ∥ g κ 2 ( ν ⋅ ) ∥ L 1 ( 0 , 1 ) = 1 . The last bound follows from a k 2 = θ 2 g κ 2 ( ν t ) ρ Γ k 2 ( Id − R / ρ ) a_k^2=\theta^2g_\kappa^2(\nu t)\rho\Gamma_k^2(\operatorname{Id}-R/\rho) a k 2 = θ 2 g κ 2 ( ν t ) ρ Γ k 2 ( Id − R / ρ ) and Lemma A.5.
The velocity perturbation
We use the improved Hölder inequality of Modena and Székelyhidi [26 Lemma 2.1] , see also [12 Lemma B.1] : for 1 ≤ p ′ ≤ ∞ 1\leq p'\leq\infty 1 ≤ p ′ ≤ ∞ , smooth a , f a,f a , f on T d \mathbb{T}^d T d and σ ∈ N \sigma\in\mathbb{N} σ ∈ N ,
∥ a f ( σ ⋅ ) ∥ L p ′ − ∥ a ∥ L p ′ ∥ f ∥ L p ′ ≲ σ − 1 / p ′ ∥ a ∥ C 1 ∥ f ∥ L p ′ . (109) \|af(\sigma\cdot)\|_{L^{p'}}-\|a\|_{L^{p'}}\|f\|_{L^{p'}}\lesssim\sigma^{-1/p'}\|a\|_{C^1}\|f\|_{L^{p'}}. \tag*{(109)} ∥ a f ( σ ⋅ ) ∥ L p ′ − ∥ a ∥ L p ′ ∥ f ∥ L p ′ ≲ σ − 1/ p ′ ∥ a ∥ C 1 ∥ f ∥ L p ′ . (109)
Proposition B.3 (Principal part). There is C 0 ≥ 1 C_0\geq1 C 0 ≥ 1 depending only on d d d such that
∥ w ( p ) ∥ L 2 ( [ 0 , 1 ] × T d ) ≤ C 0 ∥ R ∥ 1 1 / 2 + C u λ − γ / 2 , (110) \|w^{(p)}\|_{L^2([0,1]\times\mathbb{T}^d)}\leq C_0\|R\|_1^{1/2}+C_u\lambda^{-\gamma/2}, \tag*{(110)} ∥ w ( p ) ∥ L 2 ([ 0 , 1 ] × T d ) ≤ C 0 ∥ R ∥ 1 1/2 + C u λ − γ /2 , (110)
∥ w ( p ) ∥ L p ( 0 , 1 ; L ∞ ) + ∥ w ( p ) ∥ L 1 ( 0 , 1 ; W 1 , q ) ≤ C u λ − γ . (111) \|w^{(p)}\|_{L^p(0,1;L^\infty)}+\|w^{(p)}\|_{L^1(0,1;W^{1,q})}\leq C_u\lambda^{-\gamma}. \tag*{(111)} ∥ w ( p ) ∥ L p ( 0 , 1 ; L ∞ ) + ∥ w ( p ) ∥ L 1 ( 0 , 1 ; W 1 , q ) ≤ C u λ − γ . (111)
Proof. L 2 L^2 L 2 bound. By (B.11) with p ′ = 2 p'=2 p ′ = 2 and ∥ W k ∥ L 2 = 1 \|W_k\|_{L^2}=1 ∥ W k ∥ L 2 = 1 (Lemma A.3(c)), for every t t t ,
∥ w ( p ) ( t ) ∥ L 2 ≤ ∑ k ∈ Ξ ( ∥ a k ( t ) ∥ L 2 + C σ − 1 / 2 ∥ a k ( t ) ∥ C 1 ) . \|w^{(p)}(t)\|_{L^2}\leq\sum_{k\in\Xi}\left(\|a_k(t)\|_{L^2}+C\sigma^{-1/2}\|a_k(t)\|_{C^1}\right). ∥ w ( p ) ( t ) ∥ L 2 ≤ k ∈ Ξ ∑ ( ∥ a k ( t ) ∥ L 2 + C σ − 1/2 ∥ a k ( t ) ∥ C 1 ) .
Taking the L 2 L^2 L 2 norm in time, by (B.10) and (B.9) (with p ′ = 2 p'=2 p ′ = 2 ),
∥ w ( p ) ∥ L t , x 2 ≤ ∣ Ξ ∣ C Γ ( ∫ 0 1 g κ 2 ( ν t ) F ( t ) d t ) 1 / 2 + C u σ − 1 / 2 , F ( t ) : = θ 2 ( t ) ∫ T d ρ ( t , x ) d x . \|w^{(p)}\|_{L^2_{t,x}}\leq|\Xi|C_\Gamma\left(\int_0^1g_\kappa^2(\nu t)F(t)\,dt\right)^{1/2}+C_u\sigma^{-1/2},\qquad F(t):=\theta^2(t)\int_{\mathbb{T}^d}\rho(t,x)\,dx. ∥ w ( p ) ∥ L t , x 2 ≤ ∣Ξ∣ C Γ ( ∫ 0 1 g κ 2 ( ν t ) F ( t ) d t ) 1/2 + C u σ − 1/2 , F ( t ) := θ 2 ( t ) ∫ T d ρ ( t , x ) d x .
F F F is smooth and independent of λ \lambda λ , so ∥ F ′ ∥ L ∞ ≤ C u \|F'\|_{L^\infty}\leq C_u ∥ F ′ ∥ L ∞ ≤ C u . By Lemma A.4(c) and Lemma A.5,
∫ 0 1 g κ 2 ( ν t ) F ( t ) d t ≤ ∫ 0 1 F ( t ) d t + 2 ν − 1 C u ≤ 12 ∥ R ∥ 1 + C u ν − 1 . \int_0^1g_\kappa^2(\nu t)F(t)\,dt\leq\int_0^1F(t)\,dt+2\nu^{-1}C_u\leq12\|R\|_1+C_u\nu^{-1}. ∫ 0 1 g κ 2 ( ν t ) F ( t ) d t ≤ ∫ 0 1 F ( t ) d t + 2 ν − 1 C u ≤ 12∥ R ∥ 1 + C u ν − 1 .
Using A + B ≤ A + B \sqrt{A+B}\leq\sqrt{A}+\sqrt{B} A + B ≤ A + B and (B.8), we obtain (B.12) with C 0 : = 12 ∣ Ξ ∣ C Γ C_0:=\sqrt{12}\,|\Xi|C_\Gamma C 0 := 12 ∣Ξ∣ C Γ .
L t p L ∞ L_t^pL^\infty L t p L ∞ bound. By Lemma A.3(b), (B.9) with p ′ = p p'=p p ′ = p , and (B.5),
∥ w ( p ) ∥ L t p L ∞ ≤ ∑ k ∥ a k ∥ L t p L ∞ ∥ W k ∥ L ∞ ≤ C u κ 1 2 − 1 p μ d − 1 2 ≤ C u λ − γ . \|w^{(p)}\|_{L_t^pL^\infty}\leq\sum_k\|a_k\|_{L_t^pL^\infty}\|W_k\|_{L^\infty}\leq C_u\kappa^{\frac{1}{2}-\frac{1}{p}}\mu^{\frac{d-1}{2}}\leq C_u\lambda^{-\gamma}. ∥ w ( p ) ∥ L t p L ∞ ≤ k ∑ ∥ a k ∥ L t p L ∞ ∥ W k ∥ L ∞ ≤ C u κ 2 1 − p 1 μ 2 d − 1 ≤ C u λ − γ .
L t 1 W 1 , q L_t^1W^{1,q} L t 1 W 1 , q bound. By (A.1) and Lemma A.3(b), using σ μ ≥ 1 \sigma\mu\geq1 σ μ ≥ 1 ,
∥ W k ( σ ⋅ ) ∥ W 1 , q = ∥ W k ∥ L q + σ ∥ ∇ W k ∥ L q ≲ σ μ d − 1 2 − d − 1 q . \|W_k(\sigma\cdot)\|_{W^{1,q}}=\|W_k\|_{L^q}+\sigma\|\nabla W_k\|_{L^q}\lesssim\sigma\mu^{\frac{d-1}{2}-\frac{d-1}{q}}. ∥ W k ( σ ⋅ ) ∥ W 1 , q = ∥ W k ∥ L q + σ ∥∇ W k ∥ L q ≲ σ μ 2 d − 1 − q d − 1 .
Since ∥ a W k ( σ ⋅ ) ∥ W 1 , q ≲ ∥ a ∥ C 1 ∥ W k ( σ ⋅ ) ∥ W 1 , q \|aW_k(\sigma\cdot)\|_{W^{1,q}}\lesssim\|a\|_{C^1}\|W_k(\sigma\cdot)\|_{W^{1,q}} ∥ a W k ( σ ⋅ ) ∥ W 1 , q ≲ ∥ a ∥ C 1 ∥ W k ( σ ⋅ ) ∥ W 1 , q , (B.9) with p ′ = 1 p'=1 p ′ = 1 and (B.6) give
∥ w ( p ) ∥ L t 1 W 1 , q ≤ ∑ k ∥ a k ∥ L t 1 C 1 ∥ W k ( σ ⋅ ) ∥ W 1 , q ≤ C u κ − 1 / 2 σ μ 1 + d − 1 2 − d − 1 q ≤ C u λ − γ . \|w^{(p)}\|_{L_t^1W^{1,q}}\leq\sum_k\|a_k\|_{L_t^1C^1}\|W_k(\sigma\cdot)\|_{W^{1,q}}\leq C_u\kappa^{-1/2}\sigma\mu^{1+\frac{d-1}{2}-\frac{d-1}{q}}\leq C_u\lambda^{-\gamma}. ∥ w ( p ) ∥ L t 1 W 1 , q ≤ k ∑ ∥ a k ∥ L t 1 C 1 ∥ W k ( σ ⋅ ) ∥ W 1 , q ≤ C u κ − 1/2 σ μ 1 + 2 d − 1 − q d − 1 ≤ C u λ − γ .
Proposition B.4 (Correctors).
∥ w ( c ) ∥ L 2 ( 0 , 1 ; L ∞ ) + ∥ w ( c ) ∥ L 1 ( 0 , 1 ; W 1 , q ) ≤ C u λ − γ , (112) \lVert w^{(c)}\rVert_{L^2(0,1;L^\infty)}+\lVert w^{(c)}\rVert_{L^1(0,1;W^{1,q})}\le C_u\lambda^{-\gamma}, \tag*{(112)} ∥ w ( c ) ∥ L 2 ( 0 , 1 ; L ∞ ) + ∥ w ( c ) ∥ L 1 ( 0 , 1 ; W 1 , q ) ≤ C u λ − γ , (112) ∥ w ( t ) ∥ L ∞ ( 0 , 1 ; W 1 , ∞ ) ≤ C u λ − γ . (113) \lVert w^{(t)}\rVert_{L^\infty(0,1;W^{1,\infty})}\le C_u\lambda^{-\gamma}. \tag*{(113)} ∥ w ( t ) ∥ L ∞ ( 0 , 1 ; W 1 , ∞ ) ≤ C u λ − γ . (113) Proof. For (B.14): by Lemma A.3 (b), (B.9) with p ′ = 2 p'=2 p ′ = 2 , and (B.7),
∥ w ( c ) ∥ L t 2 L x ∞ ≲ σ − 1 ∑ k ∥ ∇ a k ∥ L t 2 L ∞ ∥ Ω k ∥ L ∞ ≤ C u σ − 1 μ − 1 + d − 1 2 ≤ C u λ − γ . \lVert w^{(c)}\rVert_{L^2_tL^\infty_x}\lesssim\sigma^{-1}\sum_k\lVert\nabla a_k\rVert_{L^2_tL^\infty}\lVert\Omega_k\rVert_{L^\infty}\le C_u\sigma^{-1}\mu^{-1+\frac{d-1}{2}}\le C_u\lambda^{-\gamma}. ∥ w ( c ) ∥ L t 2 L x ∞ ≲ σ − 1 k ∑ ∥ ∇ a k ∥ L t 2 L ∞ ∥ Ω k ∥ L ∞ ≤ C u σ − 1 μ − 1 + 2 d − 1 ≤ C u λ − γ . Similarly, ∥ Ω k ( σ ⋅ ) ∥ W 1 , q ≲ σ μ μ − 1 + d − 1 2 − d − 1 q \lVert\Omega_k(\sigma\cdot)\rVert_{W^{1,q}}\lesssim\sigma\mu\mu^{-1+\frac{d-1}{2}-\frac{d-1}{q}} ∥ Ω k ( σ ⋅ ) ∥ W 1 , q ≲ σ μ μ − 1 + 2 d − 1 − q d − 1 , so by (B.9) with p ′ = 1 p'=1 p ′ = 1 ,
∥ w ( c ) ∥ L t 1 W 1 , q ≲ σ − 1 ∑ k ∥ a k ∥ L t 1 C 2 ∥ Ω k ( σ ⋅ ) ∥ W 1 , q ≤ C u κ − 1 2 μ d − 1 2 − d − 1 q = C u ( σ μ ) − 1 κ − 1 2 σ μ 1 + d − 1 2 − d − 1 q , \lVert w^{(c)}\rVert_{L^1_tW^{1,q}}\lesssim\sigma^{-1}\sum_k\lVert a_k\rVert_{L^1_tC^2}\lVert\Omega_k(\sigma\cdot)\rVert_{W^{1,q}}\le C_u\kappa^{-\frac{1}{2}}\mu^{\frac{d-1}{2}-\frac{d-1}{q}}=C_u(\sigma\mu)^{-1}\kappa^{-\frac{1}{2}}\sigma\mu^{1+\frac{d-1}{2}-\frac{d-1}{q}}, ∥ w ( c ) ∥ L t 1 W 1 , q ≲ σ − 1 k ∑ ∥ a k ∥ L t 1 C 2 ∥ Ω k ( σ ⋅ ) ∥ W 1 , q ≤ C u κ − 2 1 μ 2 d − 1 − q d − 1 = C u ( σ μ ) − 1 κ − 2 1 σ μ 1 + 2 d − 1 − q d − 1 , which is at most C u λ − γ C_u\lambda^{-\gamma} C u λ − γ by (B.6).
For (B.15): since ∣ h κ ∣ ≤ 1 |h_\kappa|\le1 ∣ h κ ∣ ≤ 1 (Lemma A.4 (b)), and since div R − ∇ Δ − 1 div div R \operatorname{div}R-\nabla\Delta^{-1}\operatorname{div}\operatorname{div}R div R − ∇ Δ − 1 div div R is smooth and independent of λ \lambda λ , we get ∥ w ( t ) ∥ L t ∞ W 1 , ∞ ≤ C u ν − 1 = C u λ − γ \lVert w^{(t)}\rVert_{L^\infty_tW^{1,\infty}}\le C_u\nu^{-1}=C_u\lambda^{-\gamma} ∥ w ( t ) ∥ L t ∞ W 1 , ∞ ≤ C u ν − 1 = C u λ − γ .
Corollary B.5 (The perturbation). Let M : = max { 1 , 2 C 0 } M:=\max\{1,2C_0\} M := max { 1 , 2 C 0 } , which depends only on d d d . There is λ 1 \lambda_1 λ 1 , depending on u u u , R R R , s s s , t ∗ t_* t ∗ , δ \delta δ , such that, for λ ≥ λ 1 \lambda\ge\lambda_1 λ ≥ λ 1 ,
∥ w ∥ L t , x 2 ≤ M ∥ R ∥ 1 1 / 2 , ∥ w ∥ L t p L ∞ + ∥ w ∥ L t 1 W 1 , q ≤ C u λ − γ ≤ δ . \lVert w\rVert_{L^2_{t,x}}\le M\lVert R\rVert_1^{1/2},\qquad\lVert w\rVert_{L^p_tL^\infty}+\lVert w\rVert_{L^1_tW^{1,q}}\le C_u\lambda^{-\gamma}\le\delta. ∥ w ∥ L t , x 2 ≤ M ∥ R ∥ 1 1/2 , ∥ w ∥ L t p L ∞ + ∥ w ∥ L t 1 W 1 , q ≤ C u λ − γ ≤ δ . Moreover ∥ w ( c ) ∥ L t 2 L ∞ + ∥ w ( t ) ∥ L t 2 L ∞ ≤ C u λ − γ \lVert w^{(c)}\rVert_{L^2_tL^\infty}+\lVert w^{(t)}\rVert_{L^2_tL^\infty}\le C_u\lambda^{-\gamma} ∥ w ( c ) ∥ L t 2 L ∞ + ∥ w ( t ) ∥ L t 2 L ∞ ≤ C u λ − γ and ∥ w ∥ L t , x 2 + ∥ w ( p ) ∥ L t , x 2 ≤ C u \lVert w\rVert_{L^2_{t,x}}+\lVert w^{(p)}\rVert_{L^2_{t,x}}\le C_u ∥ w ∥ L t , x 2 + ∥ w ( p ) ∥ L t , x 2 ≤ C u .
Proof. Add Propositions B.3 and B.4, using ∥ ⋅ ∥ L t , x 2 ≤ ∥ ⋅ ∥ L t 2 L ∞ \lVert\cdot\rVert_{L^2_{t,x}}\le\lVert\cdot\rVert_{L^2_tL^\infty} ∥ ⋅ ∥ L t , x 2 ≤ ∥ ⋅ ∥ L t 2 L ∞ , ∥ ⋅ ∥ L t p L ∞ ≤ ∥ ⋅ ∥ L t 2 L ∞ \lVert\cdot\rVert_{L^p_tL^\infty}\le\lVert\cdot\rVert_{L^2_tL^\infty} ∥ ⋅ ∥ L t p L ∞ ≤ ∥ ⋅ ∥ L t 2 L ∞ (since p < 2 p<2 p < 2 ) and ∥ ⋅ ∥ L t 1 W 1 , q ≤ ∥ ⋅ ∥ L t ∞ W 1 , ∞ \lVert\cdot\rVert_{L^1_tW^{1,q}}\le\lVert\cdot\rVert_{L^\infty_tW^{1,\infty}} ∥ ⋅ ∥ L t 1 W 1 , q ≤ ∥ ⋅ ∥ L t ∞ W 1 , ∞ . This gives ∥ w ∥ L t , x 2 ≤ C 0 ∥ R ∥ 1 1 / 2 + C u λ − γ / 2 \lVert w\rVert_{L^2_{t,x}}\le C_0\lVert R\rVert_1^{1/2}+C_u\lambda^{-\gamma/2} ∥ w ∥ L t , x 2 ≤ C 0 ∥ R ∥ 1 1/2 + C u λ − γ /2 . Since ∥ R ∥ 1 > 0 \lVert R\rVert_1>0 ∥ R ∥ 1 > 0 , the last term is at most C 0 ∥ R ∥ 1 1 / 2 C_0\lVert R\rVert_1^{1/2} C 0 ∥ R ∥ 1 1/2 for λ \lambda λ large.
The new stress
Recall from Lemma A.1 that R R R is bounded on L r L^r L r , that R div R\operatorname{div} R div is bounded on L r L^r L r since 1 < r < ∞ 1<r<\infty 1 < r < ∞ , and that ∥ R ( g ( σ ⋅ ) ) ∥ L r ≲ σ − 1 ∥ g ∥ L r \lVert R(g(\sigma\cdot))\rVert_{L^r}\lesssim\sigma^{-1}\lVert g\rVert_{L^r} ∥ R ( g ( σ ⋅ )) ∥ L r ≲ σ − 1 ∥ g ∥ L r for g g g of zero mean.
Lemma B.6 (Linear error). ∥ R l i n ∥ L 1 ( 0 , 1 ; L r ) ≤ C u λ − γ \lVert R_{\mathrm{lin}}\rVert_{L^1(0,1;L^r)}\le C_u\lambda^{-\gamma} ∥ R lin ∥ L 1 ( 0 , 1 ; L r ) ≤ C u λ − γ .
Proof. We treat the three terms of (A.21) separately.
Dissipative term. R Δ w = R div ( ∇ w ) R\Delta w=R\operatorname{div}(\nabla w) R Δ w = R div ( ∇ w ) , so ∥ R Δ w ∥ L t 1 L r ≲ ∥ ∇ w ∥ L t 1 L r ≤ ∥ w ∥ L t 1 W 1 , q ≤ C u λ − γ \lVert R\Delta w\rVert_{L^1_tL^r}\lesssim\lVert\nabla w\rVert_{L^1_tL^r}\le\lVert w\rVert_{L^1_tW^{1,q}}\le C_u\lambda^{-\gamma} ∥ R Δ w ∥ L t 1 L r ≲ ∥ ∇ w ∥ L t 1 L r ≤ ∥ w ∥ L t 1 W 1 , q ≤ C u λ − γ , since r < q r<q r < q .
Time derivative. By (A.13), ∂ t ( w ( p ) + w ( c ) ) = σ − 1 div ( ∑ k ∂ t a k Ω k ( σ x ) ) \partial_t(w^{(p)}+w^{(c)})=\sigma^{-1}\operatorname{div}\left(\sum_k\partial_ta_k\Omega_k(\sigma x)\right) ∂ t ( w ( p ) + w ( c ) ) = σ − 1 div ( ∑ k ∂ t a k Ω k ( σ x ) ) . Hence, by Lemma A.3 (b), (B.9) with n = 1 n=1 n = 1 , p ′ = 1 p'=1 p ′ = 1 , and (B.4),
∥ R ∂ t ( w ( p ) + w ( c ) ) ∥ L t 1 L r ≲ σ − 1 ∑ k ∥ ∂ t a k ∥ L t 1 L ∞ ∥ Ω k ∥ L r ≤ C u σ − 1 ν κ 1 2 μ − 1 + d − 1 2 − d − 1 r ≤ C u λ − γ . \lVert R\partial_t(w^{(p)}+w^{(c)})\rVert_{L^1_tL^r}\lesssim\sigma^{-1}\sum_k\lVert\partial_ta_k\rVert_{L^1_tL^\infty}\lVert\Omega_k\rVert_{L^r}\le C_u\sigma^{-1}\nu\kappa^{\frac{1}{2}}\mu^{-1+\frac{d-1}{2}-\frac{d-1}{r}}\le C_u\lambda^{-\gamma}. ∥ R ∂ t ( w ( p ) + w ( c ) ) ∥ L t 1 L r ≲ σ − 1 k ∑ ∥ ∂ t a k ∥ L t 1 L ∞ ∥ Ω k ∥ L r ≤ C u σ − 1 ν κ 2 1 μ − 1 + 2 d − 1 − r d − 1 ≤ C u λ − γ . Transport term. ∥ R div ( u ⊗ w + w ⊗ u ) ∥ L t 1 L r ≲ ∥ u ∥ L t , x ∞ ∥ w ∥ L t 1 L r ≤ C u ∥ w ∥ L t p L ∞ ≤ C u λ − γ \lVert R\operatorname{div}(u\otimes w+w\otimes u)\rVert_{L^1_tL^r}\lesssim\lVert u\rVert_{L^\infty_{t,x}}\lVert w\rVert_{L^1_tL^r}\le C_u\lVert w\rVert_{L^p_tL^\infty}\le C_u\lambda^{-\gamma} ∥ R div ( u ⊗ w + w ⊗ u ) ∥ L t 1 L r ≲ ∥ u ∥ L t , x ∞ ∥ w ∥ L t 1 L r ≤ C u ∥ w ∥ L t p L ∞ ≤ C u λ − γ .
Lemma B.7 (Correction error). ∥ R c o r ∥ L 1 ( 0 , 1 ; L r ) ≤ C u λ − γ \lVert R_{\mathrm{cor}}\rVert_{L^1(0,1;L^r)}\le C_u\lambda^{-\gamma} ∥ R cor ∥ L 1 ( 0 , 1 ; L r ) ≤ C u λ − γ .
Proof. By the boundedness of R div R\operatorname{div} R div on L r L^r L r and Hölder’s inequality in t t t and x x x , using r < 2 r<2 r < 2 ,
∥ R c o r ∥ L t 1 L r ≲ ( ∥ w ( c ) ∥ L t 2 L ∞ + ∥ w ( t ) ∥ L t 2 L ∞ ) ( ∥ w ∥ L t , x 2 + ∥ w ( p ) ∥ L t , x 2 ) ≤ C u λ − γ , \lVert R_{\mathrm{cor}}\rVert_{L^1_tL^r}\lesssim\left(\lVert w^{(c)}\rVert_{L^2_tL^\infty}+\lVert w^{(t)}\rVert_{L^2_tL^\infty}\right)\left(\lVert w\rVert_{L^2_{t,x}}+\lVert w^{(p)}\rVert_{L^2_{t,x}}\right)\le C_u\lambda^{-\gamma}, ∥ R cor ∥ L t 1 L r ≲ ( ∥ w ( c ) ∥ L t 2 L ∞ + ∥ w ( t ) ∥ L t 2 L ∞ ) ( ∥ w ∥ L t , x 2 + ∥ w ( p ) ∥ L t , x 2 ) ≤ C u λ − γ , by Corollary B.5.
Lemma B.8 (Oscillation error). ∥ R o s c ∥ L 1 ( 0 , 1 ; L r ) ≤ C u λ − γ \lVert R_{\mathrm{osc}}\rVert_{L^1(0,1;L^r)}\le C_u\lambda^{-\gamma} ∥ R osc ∥ L 1 ( 0 , 1 ; L r ) ≤ C u λ − γ .
Proof. Spatial oscillation. T k T_k T k has zero mean, so by Lemma A.1(a),(c),
∥ B ( ∇ ( a k 2 ) , T k ( σ ⋅ ) ) ∥ L r ≲ ∥ ∇ ( a k 2 ) ∥ C 1 ∥ R ( T k ( σ ⋅ ) ) ∥ L r ≲ σ − 1 ∥ a k 2 ∥ C 2 ∥ T k ∥ L r . \|\mathcal{B}(\nabla(a_k^2),T_k(\sigma\cdot))\|_{L^r} \lesssim\|\nabla(a_k^2)\|_{C^1}\|\mathcal{R}(T_k(\sigma\cdot))\|_{L^r} \lesssim\sigma^{-1}\|a_k^2\|_{C^2}\|T_k\|_{L^r}. ∥ B ( ∇ ( a k 2 ) , T k ( σ ⋅ )) ∥ L r ≲ ∥∇ ( a k 2 ) ∥ C 1 ∥ R ( T k ( σ ⋅ )) ∥ L r ≲ σ − 1 ∥ a k 2 ∥ C 2 ∥ T k ∥ L r .
By Lemma A.3 (b),(c), ∥ T k ∥ L r ≤ ∥ W k ∥ L 2 r 2 + 1 ≲ μ ( d − 1 ) ( 1 − 1 r ) \|T_k\|_{L^r}\leq\|W_k\|_{L^{2r}}^2+1\lesssim\mu^{(d-1)(1-\frac{1}{r})} ∥ T k ∥ L r ≤ ∥ W k ∥ L 2 r 2 + 1 ≲ μ ( d − 1 ) ( 1 − r 1 ) . With (B.9) and (B.7), ∥ R o s c , x ∥ L t 1 L r ≤ C u σ − 1 μ ( d − 1 ) ( 1 − 1 r ) ≤ C u λ − γ \|R_{\mathrm{osc},x}\|_{L_t^1L^r}\leq C_u\sigma^{-1}\mu^{(d-1)(1-\frac{1}{r})}\leq C_u\lambda^{-\gamma} ∥ R osc , x ∥ L t 1 L r ≤ C u σ − 1 μ ( d − 1 ) ( 1 − r 1 ) ≤ C u λ − γ .
Temporal oscillation. Since ∣ h κ ∣ ≤ 1 |h_\kappa|\leq1 ∣ h κ ∣ ≤ 1 , ∥ R o s c , t ∥ L t 1 L r ≤ ν − 1 ∥ ∂ t R ∥ L t 1 L r ≤ C u ν − 1 = C u λ − γ \|R_{\mathrm{osc},t}\|_{L_t^1L^r}\leq\nu^{-1}\|\partial_tR\|_{L_t^1L^r}\leq C_u\nu^{-1}=C_u\lambda^{-\gamma} ∥ R osc , t ∥ L t 1 L r ≤ ν − 1 ∥ ∂ t R ∥ L t 1 L r ≤ C u ν − 1 = C u λ − γ .
Far-field interaction. The traceless part is an orthogonal projection, so ∣ R f a r ∣ ≤ ∣ R f a r f u l l ∣ |R_{\mathrm{far}}|\leq|R_{\mathrm{far}}^{\mathrm{full}}| ∣ R far ∣ ≤ ∣ R far full ∣ . By (A.1), W k ( σ ⋅ ) ⊗ W k ′ ( σ ⋅ ) = ( W k ⊗ W k ′ ) ( σ ⋅ ) W_k(\sigma\cdot)\otimes W_{k'}(\sigma\cdot)=(W_k\otimes W_{k'})(\sigma\cdot) W k ( σ ⋅ ) ⊗ W k ′ ( σ ⋅ ) = ( W k ⊗ W k ′ ) ( σ ⋅ ) has the same L r L^r L r norm as W k ⊗ W k ′ W_k\otimes W_{k'} W k ⊗ W k ′ . Hence, by Lemma A.3 (c), (B.9) with p ′ = 2 p'=2 p ′ = 2 , and (B.7),
∥ R f a r ∥ L t 1 L r ≤ ∑ k ≠ k ′ ∥ a k ∥ L t 2 L ∞ ∥ a k ′ ∥ L t 2 L ∞ ∥ W k ⊗ W k ′ ∥ L r ≤ C u μ d − 1 − d r ≤ C u λ − γ . \|R_{\mathrm{far}}\|_{L_t^1L^r}\leq\sum_{k\neq k'}\|a_k\|_{L_t^2L^\infty}\|a_{k'}\|_{L_t^2L^\infty}\|W_k\otimes W_{k'}\|_{L^r}\leq C_u\mu^{d-1-\frac{d}{r}}\leq C_u\lambda^{-\gamma}. ∥ R far ∥ L t 1 L r ≤ k = k ′ ∑ ∥ a k ∥ L t 2 L ∞ ∥ a k ′ ∥ L t 2 L ∞ ∥ W k ⊗ W k ′ ∥ L r ≤ C u μ d − 1 − r d ≤ C u λ − γ .
Proof of Proposition 2.8
Let d , p , q d,p,q d , p , q be given, and assume q ≥ 2 q\geq2 q ≥ 2 as explained at the beginning of this section. Let r = r ( d , p , q ) r=r(d,p,q) r = r ( d , p , q ) be defined by (B.2); then 1 < r < d d − 1 1<r<\frac{d}{d-1} 1 < r < d − 1 d . Let M : = max { 1 , 2 C 0 } M:=\max\{1,2C_0\} M := max { 1 , 2 C 0 } , which depends only on d d d . Let ( u , R ) (u,R) ( u , R ) , 0 < s < t ∗ ≤ 1 0<s<t_*\leq1 0 < s < t ∗ ≤ 1 and δ > 0 \delta>0 δ > 0 be given. If R ≡ 0 R\equiv0 R ≡ 0 , take w = 0 w=0 w = 0 , P = 0 P=0 P = 0 , R C L = 0 R^{\mathrm{CL}}=0 R CL = 0 . Otherwise, let w , P , R C L w,P,R^{\mathrm{CL}} w , P , R CL be given by Proposition A.9 with the parameters (B.3). Then:
smoothness, div w = 0 \operatorname{div}w=0 div w = 0 , the S 0 d × d S_0^{d\times d} S 0 d × d -valuedness of R C L R^{\mathrm{CL}} R CL , and properties (i) and (ii) hold for every λ \lambda λ , by Proposition A.9;
properties (iii) and (iv) hold for λ ≥ λ 1 \lambda\geq\lambda_1 λ ≥ λ 1 , by Corollary B.5;
by (A.24) and Lemmas B.6 –B.8 , ∥ R C L ∥ L t 1 L r ≤ C u λ − γ \|R^{\mathrm{CL}}\|_{L_t^1L^r}\leq C_u\lambda^{-\gamma} ∥ R CL ∥ L t 1 L r ≤ C u λ − γ , which is at most δ \delta δ for λ \lambda λ large; this is (v).
Choosing N N N , and hence λ = N m \lambda=N^m λ = N m , large enough (also so that λ ≥ μ 0 \lambda\geq\mu_0 λ ≥ μ 0 ) completes the proof.
Remark B.9. 1. In particular ∥ w ∥ L t 1 L ∞ ≤ ∥ w ∥ L t p L ∞ ≤ C u λ − γ \|w\|_{L_t^1L^\infty}\leq\|w\|_{L_t^{p}L^\infty}\leq C_u\lambda^{-\gamma} ∥ w ∥ L t 1 L ∞ ≤ ∥ w ∥ L t p L ∞ ≤ C u λ − γ . This is the only norm of w w w that enters the kinetic estimates of Section 3.
The Laplacian plays no role in the construction. Proposition 2.8 therefore holds without the term − Δ w -\Delta w − Δ w , the dissipative term in Lemma B.6 being simply absent. This is the starting point for the inviscid Vlasov–Euler system.