Annihilation of slowly-decaying terms of Navier-Stokes flows by external forcing
aa r X i v : . [ m a t h . A P ] J a n Annihilation of slowly-decaying terms of Navier-Stokes flowsby external forcing
Lorenzo Brandolese
Institut Camille JordanUniversit´e Lyon 1E-mail: [email protected]
Takahiro Okabe
Graduate School of Engineering ScienceOsaka UniversityE-mail: [email protected]
Abstract
The goal of this paper is to provide an algorithm that, for any sufficiently localised, divergence-free small initial data, explicitly constructs a localised external force leading to a rapidly dissipativesolutions of the Navier–Stokes equations R n : namely, the energy decay rate of the flow will beforced to satisfy k u ( t ) k = o ( t − ( n +2) / ) as t → ∞ , which is beyond the usual optimal rate. Animportant feature of our construction is that this force can always be taken compactly supportedin space-time, and its profile arbitrarily prescribed up to a spatial rescaling. Since the forcing termvanishes after a finite time interval, our result suggests that nontrivial interactions between the linearand nonlinear parts occur, annihilating all the slowly decaying terms contained in Miyakawa andSchonbek’s asymptotic profiles. Key word:
Navier-Stokes equations, Energy decay, Asymptotic profiles.
MSC(2010):
Let n ≥
2. We consider the incompressible Navier-Stokes equations in R n :(N-S) ∂ t u − ∆ u + u · ∇ u + ∇ π = ∇ · f in R n × (0 , ∞ ) , div u = 0 in R n × (0 , ∞ ) ,u ( · ,
0) = a in R n , where u = u ( x, t ) = (cid:0) u ( x, t ) , . . . , u n ( x, t ) (cid:1) and π = π ( x, t ) denote the unknown velocity and the pressureof the fluid at ( x, t ) ∈ R n × (0 , ∞ ), respectively, while, f = f ( x, t ) = (cid:0) f kℓ ( x, t ) (cid:1) k,ℓ =1 ,...,n denotes theexternal forcing tensor and a = a ( x ) = (cid:0) a ( x ) , . . . , a n ( x ) (cid:1) denotes the given initial data.Starting with the celebrated work of Leray [9], the time decay problem has been a major issue in themathematical study of fluid flows. Masuda [10], Schonbek [15], Kajikiya and Miyakawa [7] and Wiegner[17], for instance, obtained pioneering contributions in this direction. Their results imply that, in theabsence of external forcing, the optimal decay rate for a weak solution is(1.1) k u ( t ) k ≤ C (1 + t ) − n +24 , t > , for initial data a in L σ ( R n ) under suitable additional conditions. To this purpose, the moment condition R R n (1 + | x | ) | a ( x ) | d x < ∞ , for example, would be enough. Subsequently, Fujigaki-Miyakawa [5] clarifiedthat the decay rate in the right-hand side of (1.1) actually describes the decay rate of the nonlinear terms.Indeed, they derived the asymptotic expansion of the linear part and of the nonlinear part as followslim t →∞ t + n (1 − q ) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) u j ( t ) + n X k =1 ( ∂ k E t )( · ) Z R n y k a j ( y ) dy + n X ℓ,k =1 F ℓk,j ( · , t ) Z ∞ Z R n ( u ℓ u k )( y, s ) d y d s (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) q = 0 , j = 1 , . . . , n and for all 1 ≤ q ≤ ∞ , where E t ( x ) = (4 πt ) − n/ exp (cid:0) − | x | t (cid:1) is the heat kernel and F ℓk,j ( x, t ) = ∂ ℓ E t ( x ) δ jk + Z ∞ t ∂ ℓ ∂ k ∂ j E s ( x ) d s, where δ jk is the Kronecker symbol.Miyakawa and Schonbek [13] deduced from the above asymptotic profile necessary and sufficientconditions ensuring that the flow is rapidly dissipative , in the sense thatlim t →∞ t + n (1 − q ) k u ( t ) k q = 0 , ≤ q ≤ ∞ . For initial data with finite moments up to the first order, these conditions read(1.2a) Z R n y k a j ( y ) dy = 0 for j, k = 1 , . . . , n, and(1.2b) ∃ c ∈ R such that Z ∞ Z R n ( u k u ℓ )( y, s ) d y d s = c δ kℓ for k, ℓ = 1 , . . . , n. Condition (1.2b) is difficult to check, as it requires information of the (unknown) flow over wholespace-time region. For this reason, it is usually not possible to predict whether or not a given flowis rapidly dissipative. To overcome this difficulty, and to make evidence that condition (1.2b) is notreduced to the identically zero solution, one could restrict the problem to flows invariant under the actionof suitable symmetry groups. For example, in 2D one could consider flows with radial vorticity, but thisidea is no longer effective in 3D because of topological obstructions. The first author, to circumvent thisdifficulty, considered to this purpose cyclically symmetry of the flow, i.e.,(a) u j ( · , t ) is odd in x j and even in each other variables,(b) u ( x , . . . , x n , t ) = u ( x n , x , . . . , x n − , t ) = · · · = u n ( x , . . . , x n , x , t ) . These conditions are easier to check, as they are preserved during the evolution (for the unique strongsolution, if this is known to exist, but also for any Leray solutions constructed so far) if they are satisfiedfor the initial data. In the class of flows satisfying (a) and (b), the first author [1], Miyakawa [11, 12]made evidence of the existence of rapidly dissipative solutions. They also observed that the faster decayof these solutions agrees with the rate of the second-order, or of the third order terms, in the asymptoticexpansion of the flow. Later, the second author and Tsutsui [14] gave a generalization of [2], [11, 12] withweighted Hardy spaces. More general group actions were discussed by the first author [3].However, symmetry conditions, like (a) and (b), look somewhat artificial. In particular, these sym-metric flows are non-stable in the class of generic flows: to make them physically realistic, an additionalcontrol f ( x, t ) acting in the whole time interval (0 , + ∞ ) should be required: it is therefore a naturalproblem to see how a generic flow (featuring no special symmetry) could evolve into a rapidly dissipativeflow during the evolution, thanks to some other process, implying the annihilation of the slowly decayingterms of its asymptotic profiles. The purpose of this paper is to address this issue.As mentioned above, the essential difficulty will be the verification of (1.2b). Instead of the cyclicsymmetry, for any initial velocity which is small in a suitable sense, we construct an associated externalforce and a rapidly dissipative solution of the forced Navier-Stokes equations.This approach not only looks mathematically natural, but also realistic in physics or engineering, asthe flow will be forced to slow down in the large time at faster rates, through the introduction of a adhoc forcing term, depending on the given initial state, that will act only on a bounded region and over afinite time interval, that can be taken arbitrarily short.For a given initial velocity field a , our strategy will be to provide an algorithm, leading in the limitto the construction of a force of divergence form ∇ · f , compactly supported in space-time, so that, for j = 1 , . . . , n ,(1.3) Z t [ e ( t − s )∆ P ∇ · f ] j ( s ) d s ∼ n X k,ℓ =1 F ℓk,j ( · , t ) Z ∞ Z R n ( u ℓ u k )( y, s ) d y d s for large t > . P is the Leray-Hopf (also called after Weyl-Helmholtz or Fujita-Kato) projectiononto solenoidal vectors. For the realization of (1.3), we introduce the following computable procedure:(1.4) u ( m ) ( t ) = e t ∆ a + Z t e ( t − s )∆ P ∇ · f ( m ) ( s ) d s − Z t e ( t − s )∆ P ∇ · ( u ( m ) ⊗ u ( m ) )( s ) d s m = 1 , , . . . . Here, the forcing tensor f ( m ) = ( f ( m ) kℓ ) is given by f (0) ≡ f ( m ) kℓ ( x, t ) = ( c ( m − kℓ φ ( x, t ) , k = ℓ, ( c ( m − kk − ¯ c ( m − ) φ ( x, t ) , k = ℓ, for some function φ ∈ C ∞ ( R n × [0 , ∞ )), where c ( m ) kℓ = R ∞ R R n ( u ( m ) k u ( m ) ℓ )( y, s ) d y d s and ¯ c ( m ) = c ( m )11 + · · · + c ( m ) nn . We note that since we are able to take φ compactly supported in both space and time, inorder to control the flow, it is enough that the force is applied to finite time and bounded space region.Remarkably, as we will see, the profile φ can be prescribed in an essentially arbitrary way.For our approach to be effective, it will be crucial to derive a bound K , independent of m , such that(1.6) | c ( m ) kℓ | = (cid:12)(cid:12)(cid:12)(cid:12)Z ∞ Z R n ( u ( m ) k u ( m ) ℓ )( y, s ) d y d s (cid:12)(cid:12)(cid:12)(cid:12) ≤ Z ∞ k u ( m ) ( s ) k d s ≤ K. Condition (1.6) would follow from suitable L -decay estimates. A natural idea would consist in relyingon a Fourier splitting technique to get estimates for the L -norm of u ( m ) , with the needed rate. See[15, 7, 17]. But in our context, the Fourier splitting method seems to have a slight drawback: in decayestimates like (1.1), the constant C appearing in the right-hand side depends not only on the size of thedata, but also, in a quite complicated way, on the shape of the data and of the force. This is a seriousissue, as in our case we have to deal with a recursively defined sequence of forces: all the constants in L decay estimates a priori depend on m and this makes estimate (1.6) not so straightforward.On the other hand, one could also apply a Fujita-Kato method, see Fujita and Kato [6] and Kato[8]. In this case the difficulty is that the size of f ( m ) , i.e., of the coefficients c ( m − kℓ has a priori aninfluence on the lifetime of the mild solutions u ( m ) , so an important technical issue arises: to prove thatthe recursively defined solutions u ( m ) are indeed global in time, for all natural integer m .Due to the above difficulties, we need to develop an alternative approach and to establish time decayestimates, as in (1.1), by carefully making explicit how the constants depend on the given data a , on φ and on the dimension n . For this purpose, weighted Hardy spaces would be an effective tool. Indeed,the second author and Tsutsui [14] introduced weighted Hardy spaces to derive higher order asymptoticexpansions: weighted Hardy spaces enable us to deal with higher order weights and to obtain more rapiddecay compared with the weighted Lebesgue spaces. With the aid of the weighted Hardy norm, wecould make a specific refinement of the Fujita-Kato iteration scheme, first giving bounds as in (1.6), nextensuring the convergence of our procedure (1.4) toward an external force f and to a rapidly dissipativesolution u .In this paper, however, we will adopt an alternative strategy: this will consist in the systematicderivation of quantified scale-invariant decay estimates for the L -norm of the solution. Let us illustratewhat we mean by a specific example: in the case n = 3, for well localised initial data a such that k a k << f ≡
0, we will derive the estimate(1.7) k u ( t ) k ≤ min n k a k , cK ( a ) t − / o where c is an absolute constant and K ( a ) = R R | x | | a ( x ) | d x + k a k / k a k (cid:16)R R | x | | a ( x ) | d x (cid:17) / + k a k / k a k . Estimate (1.7) is what we call the “quantified scale-invariant” version of classical Wiegner’s estimate, k u ( t ) k ≤ C (1 + t ) − / . Its crucial advantage is that the dependence of the constant C on the initial data3s made completely explicit. Notice that the three terms defining K ( a ) rescale in the same way, so (1.7)is indeed scale-invariant. Uniform bounds of the form (1.6) will rely on such type of estimates.This paper is organised as follows. We state our main result, Theorem 2.3, in Section 2. In Section 3we recall some classical estimates. Section 4 is entirely devoted to the proof of our main result. We beginin Subsection 4.1 with considering the case of a vanishing force f (0) = 0, and we present a slight refinementof Fujita-Kato scheme, with the purpose of putting in evidence the dependance of all constants appearingin the fixed point argument. Here we also establish a class of quantified scale invariant estimates, inthe same spirit as (1.7). In Subsection 4.2, we introduce the sequence of external tensors { f ( m ) } andwe construct the associated solutions u ( m ) ( t ) of (N-S), m = 0 , , , . . . , of (1.4). The main step of thissection is the conclusion that these solutions u ( m ) are all global in time, under appropriate smallnessconditions on the data and that all the relevant estimates are independent on m . Finally, we completethe proof of Theorem 2.3 by proving the convergence of u ( m ) and of f ( m ) and the decay estimate for thelimit solution. Our starting point will be the following result of Fujigaki and Miyakawa [5].
Theorem 2.1.
Let a ∈ L ( R n ) ∩ L nσ ( R n ) with R R n | x || a ( x ) | dx < ∞ and f ∈ C ∞ c (cid:0) R n × [0 , ∞ ) (cid:1) . Suppose u ∈ BC (cid:0) [0 , ∞ ); L nσ ( R n ) (cid:1) is a global mild solution of (N-S). If n ≥ , assume also that lim inf t →∞ k u ( t ) k n issufficiently small. Then it holds that lim t →∞ t + n (1 − q ) (cid:13)(cid:13)(cid:13) u j ( t ) + n X k =1 ( ∂ k E t )( · ) Z R n y k a j ( y ) dy − n X k,ℓ =1 F ℓk,j ( · , t ) Z ∞ Z R n f kℓ ( y, s ) d y d s + n X k,ℓ =1 F ℓk,j ( · , t ) Z ∞ Z R n ( u ℓ u k )( y, s ) d y d s (cid:13)(cid:13)(cid:13) q = 0 for ≤ q ≤ ∞ . Though [5] dealt with only the case f ≡
0, the proof is essentially same. The derivation of the leadingorder term for the Duhamel term of f is just analogue to that for the nonlinear term. We omit the proof.As an immediate consequence from Miyakawa and Schonbek [13], the condition associated with (1.2b) ismodified as follows: Corollary 2.2.
Let a , f and u be as in Theorem 2.1. It holds that, lim t →∞ t + n (1 − q ) k u ( t ) − e t ∆ a k q = 0 , (1 ≤ q ≤ ∞ ) if and only if there exists c ∈ R such that (2.1) Z ∞ Z R n f kℓ ( y, s ) d y d s − Z ∞ Z R n ( u ℓ u k )( y, s ) d y d s = c δ kℓ ( k, ℓ = 1 , . . . , n ) . Moreover, lim t →∞ t + n (1 − q ) k u ( t ) k q = 0 (1 ≤ q ≤ ∞ ) if and only if condition (2.1) holds and also R R n y k a ( y ) d y = 0 for all k = 1 , . . . , n .Furthermore, if (2.1) does not hold, or if at least one of the first-order moments of a does not vanish,then lim inf t →∞ t + n (1 − q ) k u ( t ) k q > for ≤ q ≤ ∞ . Remark 2.1.
We note that if Z ∞ Z R n f kℓ ( y, s ) d y d s = Z ∞ Z R n f ℓk ( y, s ) d y d s for some k and ℓ, f is not symmetric in the above sense, onecannot expect rapid time decay, no matter how fast f decays at spatial infinity and time infinity. In thiscase, by the last assertion of the corollary, we havelim inf t →∞ t + n (1 − q ) k u ( t ) k q > , ≤ q ≤ ∞ . Corollary 2.2 is the natural generalisation of [13], when an external force acts on the flow. It will beour tool to show how to control the large time decay of the flow (obtaining rapidly dissipative solutions)by forcing the fluid only during a short time interval. We now state our main result.
Theorem 2.3. (i) Let n ≥ . There exists δ = δ ( n ) > with the following property. If a ∈ L nσ ( R n ) satisfies R R n (1 + | x | ) | a ( x ) | d x < ∞ and the smallness condition (S) k a k n ≤ δ, then there exists a forcing term of divergence form ∇ · f , with compact support in space-time, f ∈ C ∞ c (cid:0) R n × [0 , ∞ ) (cid:1) , such that the unique global solution of the Navier–Stokes equations (N-S)satisfies lim t →∞ t + n (1 − q ) (cid:13)(cid:13) u ( t ) − e t ∆ a (cid:13)(cid:13) q = 0 , ≤ q ≤ ∞ . In particular, if the n first-order moments of a all vanish, then u is rapidly dissipative: (2.2) lim t →∞ t + n (1 − q ) k u ( t ) k q = 0 , ≤ q ≤ ∞ . (ii) Moreover, the shape of the above forcing term can be arbitrarily prescribed , in the following sense:for any compactly supported scalar function Ψ ∈ L ∞ c ( R n × R + ) , such that R ∞ R R n Ψ( y, s ) d y d s = 0 ,there exist R > and coefficients λ ℓk ∈ R such that, in Item (i), the forcing tensor f = ( f ℓk ) canbe taken of the form f ℓk = λ ℓk Ψ( · /R, t ) , ( ℓ, k = 1 , . . . , n ) . Remark 2.2.
From the second assertion, it follows that one can force the flow to have a fast decay inlarge time, by acting with an external force in a bounded region, during a time interval that can be takenarbitrarily short.Let us discuss the case of an initial data with first order vanishing moments, and let us apply aforcing term to the flow, as described by our theorem. We thus get a rapidly dissipative flow. Moreover,in Theorem 2.3, since f identically vanishes after a finite time interval, the solution eventually behaveslike a non-forced Navier–Stokes flow. More precisely, after some time t > t , the flow is governed by u ( t ) = e ( t − t )∆ u ( t ) − Z tt e ( t − s )∆ P ∇ · ( u ⊗ u )( s ) d s, t > t . This does not mean that the effect of the force disappears for t > t , as the new initial data u ( t ) doesdepend on the past action of f during the interval [0 , t ].So, due to the result of [13], we have two possibilities.i) The first possible scenario is that the moment condition for u ( t ) breaks down at some times t ≥ t ,i.e., R R n (1 + | x | ) | u ( x, t ) | d x = ∞ . At those time instants t , the linear part t e t ∆ u ( t ) and thenonlinear part t R tt e ( t − s )∆ P ∇ · ( u ⊗ u ) d s , individually, may both decay slowly, but they featurea nontrivial interaction annihilating the slowly decay terms, thus leading to a fast decay of solutionof the free Navier–Stokes equations starting from u ( t ).5i) The second possible scenario is that the finiteness of first order moment of u ( t ) is preserved, i.e., R R n (1 + | x | ) | u ( x, t ) | d x < ∞ for all t ≥ t . In this case, a much stronger condition than (1.2b) mustbe true, namely: Z R n ( u ℓ u k )( x, t ) d x = 0 for k = ℓ, and Z R n u ( x, t ) d x = . . . = Z R n u n ( x, t ) d x, for all t ≥ t . (See [1]). This second scenario looks non-generic as it contrasts with the general spatial spreadingphenomenon of the velocity field: the only known examples of flows satisfying the above orthogo-nality relations are those constructed putting symmetries as in [1, 2, 12].Let us stress the fact that, if one allows non-compactly supported external forces (not necessarily ofdivergence form), then it would be a trivial task to achieve the goal of forcing the flow to be rapidlydissipative, as in (2.2) (or even to bring it to rest in finite time). Indeed one could just first define anydivergence-free vector field v ( x, t ) which is equal to a ( x ) as t = 0 and equal to 0 for t ≥ t , next definethe force to be the residual of the Navier–Stokes operator. But of course, such a force would be spreadout in the whole space because of the nonlocal nature of the pressure. Let us introduce some notations and function spaces. Let C ∞ c (Ω) denote the set of all C ∞ -functions (orvectors) with compact support in a connected set Ω. Let C ∞ c,σ ( R n ) denote the set of all C ∞ -solenoidalvectors ϕ with compact support in R n , i.e., div ϕ = 0 in R n . The space L rσ ( R n ) is the closure of C ∞ c,σ ( R n )with respect to the L r -norm k · k r , 1 < r < ∞ ; L r ( R n ) denote the usual (vector-valued) Lebesguespace over R n . Moreover, C ( I ; X ), BC ( I ; X ) and L r ( I ; X ) denote the X -valued continuous and boundedcontinuous functions over the interval I ⊂ R , and X -valued L r -functions, respectively.In the estimates of this paper, we will mainly compare functions of the time variable. When we write A ( t ) . B ( t )we mean that there exists a constant c > only depending on the space dimension n , such that A ( t ) ≤ cB ( t ) for all t . In particular, when the functions A and B depend on other parameters (such as the initialdata a , the recursive parameter m , etc.), the constant c will be independent on these parameters.When the constants in our estimates depend on parameters other than the space dimension (as in (3.1)below), we will indicate this fact explicitly in our notations.We start recalling some well-known L p - L q estimates, that play an important role through this paper. Proposition 3.1.
Let ≤ p ≤ q ≤ ∞ . Then there exists a constant C q,p > such that k e t ∆ a k q ≤ C q,p t − n ( p − q ) k a k p , t > , (3.1) k∇ e t ∆ a k q ≤ C q,p t − n ( p − q ) − k a k p , t > for a function, velocity vector or tensor a ∈ L p ( R n ) . The proof of (3.1) and (3.2) are immediately derived from the Young inequality for the heat kerneland a . For the Stokes semigroup on L pσ ( R n ), we exclude the case ( p, q ) = ( ∞ , ∞ ) and ( p, q ) = (1 , L -norm: Lemma 3.1. If a ∈ L ( R n ) with R R n | x | | a ( x ) | d x < ∞ and div a = 0 , then k e t ∆ a k . (cid:16)R R n | x || a ( x ) | d x (cid:17) t − ( n +2) / . For the proof, see [4]. We give a proof below for reader’s convenience:6 roof.
Since a ∈ L ( R n ) and div a = 0, we note that R R n a ( y ) d y = 0. Hence, we easily obtain e t ∆ a ( x ) = Z R n E t ( x − y ) a ( y ) d y = Z R n (cid:0) E t ( x − y ) − E t ( x ) (cid:1) a ( y ) d y = − Z R n Z ∇ E t ( x − θy ) · ya ( y ) d θ d y. Then the Minkovski inequality (for integrals) implies that k e t ∆ a k ≤ Z R n Z | y | | a ( y ) | (cid:20)Z R n |∇ E t ( x − θy ) | d x (cid:21) / d θ d y = k∇ E t k Z R n | y | | a ( y ) | d y. Since k∇ E t k = t − n +24 k∇ E k , the proof is completed.For 1 < r < ∞ , the Leray projection P : L r ( R n ) → L rσ ( R n ) satisfies k P u k r ≤ A r k u k r for all u ∈ L r ( R n ) with some constant A r > u ( t ) = e t ∆ a + Z t e ( t − s )∆ P ∇ · f ( s ) d s + G ( u, u )( s ) , ∇ · a = 0 , where(3.4) G ( u, v ) = − Z t e ( t − s )∆ P ∇ · ( u ⊗ v )( s ) d s. We will systematically glue together the heat kernel, the Leray projector and the divergence operator,obtaining in this way the convolution operator e t ∆ P div. The kernel of this operator is denoted by F . Itscomponents are given by F ℓk,j ( x, t ) = ∂ ℓ E t ( x ) δ jk + R ∞ t ∂ ℓ ∂ k ∂ j E s ( x ) d s , for ℓ, k, j = 1 , . . . , n . Such kernelsatisfies F ∈ C ∞ ( R n × (0 , ∞ )) and the scaling relations F ( x, t ) = t − ( n +1) / F ( x/ √ t, t > k F ( · , t ) k p = c p t − ( n +1) / n/ (2 p ) , ≤ p ≤ ∞ , for some constant c p > n and p . See [5]. Construction of the solution of the free Navier–Stokes equations.
In this paragraph we quicklypresent a sightly simplified version of Kato’s method for the construction of mild solutions in L n ( R n ).Let us make use of Kato’s space, defined for n ≤ p ≤ ∞ by, X p = (cid:8) v ∈ L ∞ loc (cid:0) R + ; L p ( R n ) (cid:1) : k v k X p = ess sup t> t − n p k v ( t ) k p < ∞ (cid:9) . From now on we will abusively write sup t> instead of ess sup t> to simplify our notations. Notice that X n = L ∞ ( R + ; L n ( R n )). By the standard heat-kernel estimate (3.1),(4.1) k e t ∆ a k X p ≤ C p,n k a k n n ≤ p ≤ ∞ . It easily follows from (3.5) and the usual H¨older and Young estimates that(4.2) k G ( u, v ) k X r ≤ κ r,p k u k X p k v k X p , r ≤ p < n + r , n < p < ∞ , n ≤ r ≤ ∞ , for some constant κ r,p depending only on p, r and n . In particular, choosing, e.g., p = r = 2 n in (4.2),(4.3) k G ( u, v ) k X n ≤ κ n, n k u k X n k v k X n .
7n the other hand, by the usual heat-kernel estimate k e t ∆ a k X n ≤ C n,n k a k n . In this subsection we just consider (N-S) in the case f ≡
0. In this case, by an appropriate choice of δ >
0, namely, choosing(D1) 0 < δ < / (4 C n,n κ n, n ) , from the smallness assumption (S) we can ensure that(4.4) k a k n < / (4 C n,n κ n, n ) . So the usual fixed point Lemma, applied to the equation u = e t ∆ a + G ( u, u ) in the space X n , impliesthat a mild solution u ∈ X n to (N-S) (with identically zero external force) does exist. Moreover, k u k X n ≤ C n,n k a k n . k a k n and this condition uniquely defines u . Such solution is obtained as the limit, in the X n -norm, of thesequence of approximate solutions(4.5) u k +1 = e t ∆ a + G ( u k , u k ) , k ∈ N , with u = e t ∆ a. Here are some further estimates on u , that directly follow from the equation u = e t ∆ a + G ( u, u )and the application of (4.1) and (4.2) with different choice of the parameters (we also use (4.4) for theright inequalities below): k u k X n ≤ C n,n k a k n + κ n, n k u k X n . k a k n , k u k X n ≤ C n,n k a k n + κ n, n k u k X n . k a k n , k u k X ∞ ≤ C ∞ ,n k a k n + κ ∞ , n k u k X n . k a k n . (4.6)In particular, as all previous norms are finite, u ∈ X n ∩ X ∞ . First quantified L -decay estimates. The goal of this and next paragraph is to provide some quan-tified versions of L -decay rate estimates. At the end of next paragraph we will be able to quantifyWiegner’s fundamental estimate (1.1). More precisely, our goal will be to make explicit how the constant C , in the right-hand side of the optimal decay result k u ( t ) k ≤ C (1 + t ) − ( n +2) / , depends on a . In fact, we will do more than this. A drawback of the above estimate is that it is notscale-invariant under the usual scaling a λa ( λ · ) and u λu ( λ · , λ · ). A better way of estimating the L -norm, respecting the natural scaling of the Navier–Stokes equations, would be to look for an estimateof the form(4.7) k u ( t ) k ≤ k a k ∧ K ( a ) t − ( n +2) / (the wedge symbol stands for the minimum), for some functional K = K ( a ), independent on t , andsatisfying the scaling relations(4.8) K ( a ) = λ n K (cid:0) λ a ( λ · ) (cid:1) , for all λ > . We will achieve this at the end of next paragraph, by making explicit the expression of K ( a ).8e start with estimating the L -norm of the approximate solutions u k defined in (4.5). We have k u ( t ) k ≤ k a k . Next, applying (3.5) with p = 1, and using that a ∈ L ( R n ) ∩ L n ( R n ) ⊂ L ( R n ), weget, for k ∈ N , k u k +1 ( t ) k ≤ k a k + c Z t ( t − s ) − / k u k ( s ) k k u k ( s ) k ∞ d s ≤ k a k + πc k u k k X ∞ sup s> k u k ( s ) k . But the approximate solutions u k satisfy the same estimates as (4.6), uniformly with respect to k ∈ N .In particular, k u k k X ∞ . k a k n . Hence, by our smallness condition (S) and appropriate choice of δ > ,we can ensure that πc k u k k X ∞ ≤ / < t> k u k ( t ) k ≤ k a k for all k ∈ N . So, by Fatou’s Lemma,(4.9) sup t> k u ( t ) k ≤ k a k . (Notice that this argument does not require the use of the energy inequality. One could remove thefactor 2 by using the energy inequality. In the sequel, throughout the paper, we will avoid the use of theenergy inequality to make evidence that Theorem 2.3 remains true for simpler toy models that share thesame scaling as Navier–Stokes.)Let us go further with L -decay estimates. To start with, we begin by obtaining the quantifiedversion of the (non-optimal) decay k u ( t ) k = O ( t − / ). We have, by interpolation, a ∈ L n/ (3+2 n ) ( R n ) ⊂ L ( R n ) ∩ L ( R n ). Moreover, k u ( t ) k . k a k n/ (3+2 n ) t − / . For k ∈ N we have, k u k +1 ( t ) k ≤ k u ( t ) k + c Z t ( t − s ) − / k u k ( s ) k k u k ( s ) k ∞ d s . k u ( t ) k + c Z t ( t − s ) − / s − / d s k u k k X ∞ sup s> s / k u k ( s ) k . t − / (cid:16) k a k n/ (3+2 n ) + k a k n sup s> s / k u k ( s ) k (cid:17) . (4.10)With an appropriate choice of δ >
0, (D3), by (S), iterating the above estimate we obtain(4.11) sup k ∈ N ,t> t / k u k ( t ) k . k a k n/ (3+2 n ) . Hence, sup t> t / k u ( t ) k . k a k n/ (3+2 n ) . (4.12)Next we establish the quantified version of the improved, but still non-optimal, decay k u ( t ) k = O ( t − ( n +1) / ). Combining the standard estimate k e t ∆ a k . t − n/ k a k with Lemma 3.1, we get theestimate(4.13) sup t> t ( n +1) / k e t ∆ a k . k a k / (cid:16)R R n | x | | a ( x ) | d x (cid:17) / . This requires to replace our previous condition (D1) on δ by a more restrictive condition of the form(D2) 0 < δ < δ , where δ is not made explicit for sake of conciseness. In fact, in the subsequent steps of the proof, we will need to furtherstrengthen condition (D2) by more restrictive ones:0 < δ < δ (cf. (4.11) below) , (D3) . . . < δ < δ (cf. the end of the proof of Lemma 4.1) . (D9)All these constants δ , . . . , δ just depend on the space dimension.
9n the other hand, applying (3.5), (4.12), (4.13), and that k u k k X ∞ . k a k n (see (4.6)), we get k u k +1 ( t ) k ≤ k u ( t ) k + c Z t/ ( t − s ) − ( n +2) / k u k ( s ) k d s + c Z tt/ ( t − s ) − / k u k ( s ) k k u k ( s ) k ∞ d s . k u ( t ) k + k a k n/ (3+2 n ) Z t/ ( t − s ) − ( n +2) / s − / d s + k u k k X ∞ Z tt/ ( t − s ) − / s − / k u k ( s ) k d s . t − ( n +1) / (cid:18) k a k / (cid:16)R R n | x | | a ( x ) | d x (cid:17) / + k a k n/ (3+2 n ) + k a k n sup s> s ( n +1) / k u k ( s ) k (cid:19) . (4.14)With an appropriate choice of δ >
0, (D4), by (S), iterating the above estimate and taking k → ∞ weobtain sup t> t ( n +1) / k u ( t ) k . k a k / (cid:16)R R n | x | | a ( x ) | d x (cid:17) / + k a k n/ (3+2 n ) =: J ( a ) . (4.15)Of course we also have, because of (4.9), k u ( t ) k . k a k ∧ J ( a ) t − ( n +1) / , (4.16)where the notation α ∧ β stands for min { α, β } . The two terms defining J ( a ) rescale in the same way:if λ > a λ = λ a ( λ · ), then we see that λ n − / J ( a λ ) = J ( a ). This makes (4.16) a scale invariantestimate.Estimate (4.16) implies that u ∈ L (cid:0) R + ; L ( R n ) (cid:1) . More precisely, for any τ > Z ∞ k u ( s ) k d s . Z ∞ (cid:0) k a k ∧ J ( a ) s − ( n +1) / (cid:1) d s ≤ Z τ k a k d s + Z ∞ τ J ( a ) s − ( n +1) / d s . τ k a k + J ( a ) τ − ( n − / . Equalising the two last terms in the right-hand side leads to the choice τ = ( J ( a ) / k a k ) / ( n +1) . Then weget the quantified scale-invariant L (cid:0) R + ; L ( R n ) (cid:1) estimate(4.17) Z ∞ k u ( s ) k d s . J ( a ) / ( n +1) k a k n − / ( n +1)2 . The optimal, scale invariant and quantified L -decay estimate. We are now in the position ofobtaining the quantified version of the optimal decay k u ( t ) k = O ( t − ( n +2) / ). Indeed, applying Lemma 3.1we get k u ( t ) k . t − ( n +2) / R R n | x | | a ( x ) | d x . Moreover, by (4.16), (4.17), that by construction hold truefor the approximate solutions u k , uniformly with respect to k ∈ N , we obtain, k u k +1 ( t ) k ≤ k u ( t ) k + c Z t/ ( t − s ) − ( n +2) / k u k ( s ) k d s + c Z tt/ ( t − s ) − / k u k ( s ) k k u k ( s ) k ∞ d s . k u ( t ) k + t − ( n +2) / Z ∞ k u k ( s ) k d s + k u k k X ∞ Z tt/ ( t − s ) − / s − / k u k ( s ) k d s . t − ( n +2) / (cid:16)R R n | x | | a ( x ) | d x + J ( a ) / ( n +1) k a k n − / ( n +1)2 + k a k n sup s> s ( n +2) / k u k ( s ) k (cid:17) . (4.18)Once more, provided we make an appropriate choice of δ >
0, from the smallness condition (S) we obtain,by iteration, sup t> t ( n +2) / k u ( t ) k . R R n | x | | a ( x ) | d x + J ( a ) / ( n +1) k a k n − / ( n +1)2 =: K ( a ) . k u ( t ) k . k a k ∧ K ( a ) t − ( n +2) / , where K ( a ) = R R n | x | | a ( x ) | d x + J ( a ) / ( n +1) k a k n − / ( n +1)2 ≃ R R n | x | | a ( x ) | d x + k a k / ( n +1)1 k a k n − / ( n +1)2 (cid:16)R R n | x | | a ( x ) | d x (cid:17) / ( n +1) + k a k / ( n +1)4 n/ (3+2 n ) k a k n − / ( n +1)2 (4.20)Observe that the functional K satisfies the scaling property (4.8). In 2D, K ( a ) can be rewritten as: K ( a ) ≃ R R n | x | | a ( x ) | d x + k a k / (cid:18) k a k (cid:0)R R n | x | | a ( x ) | d x (cid:1) + k a k / (cid:19) / ( n = 2) . { f ( m ) } m ∈ N and { u ( m ) } m ∈ N . We now generalise the procedure of the previous subsection, by first recursively generating a sequence ofexternal forces f ( m ) , next constructing the corresponding solutions u ( m ) of the Navier-Stokes equations:(N-S m ) u ( m ) ( t ) = e t ∆ a + Z t e ( t − s )∆ P ∇ · f ( m ) ( s ) d s − Z t e ( t − s )∆ P ∇ · ( u ( m ) ⊗ u ( m ) )( s ) d s, for m ∈ N by Fujita–Kato’s method.First of all we set f (0) ≡ , so that u (0) is just the Navier-Stokes flow with vanishing external force constructed before. Next, let φ be any compactly supported measurable function in R n × [0 , ∞ ), such that(4.21) φ ∈ L ∞ c ( R n × R + ) and Z ∞ Z R n φ ( x, t ) d x d t = 1 . Moreover, we set, for m ∈ N , c ( m ) kℓ = Z ∞ Z R n u ( m ) k u ( m ) ℓ d y d s, k, ℓ = 1 , . . . , n, ¯ c ( m ) = c ( m )11 + · · · + c ( m ) nn , (4.22)where u ( m ) = ( u ( m )1 , . . . , u ( m ) n ), and, for m = 1 , . . . ,(4.23) f ( m ) kℓ ( x, t ) = ( c ( m − kℓ φ ( x, t ) k = ℓ, ( c ( m − kk − ¯ c ( m − ) φ ( x, t ) k = ℓ. Moreover, we denote(4.24) I ( m ) = Z ∞ Z R n | u ( m ) ( x, s ) | d x d s. At this stage, the definition of f ( m ) , u ( m ) and I ( m ) , for m ≥
1, is only formal. For the solutions u ( m ) tobe well-defined and global-in-time (it is a priori not obvious that the lifetime of the solution of (N-S m )is indeed infinite), we need to ensure that the external force f ( m ) does satisfy an appropriate smallnesscondition. We will specify this condition below, by prescribing an additional smallness condition on thefunction φ . To make our definitions of f ( m ) , u ( m ) and I ( m ) rigorous for all natural number m , we willproceed by induction. 11otice first that our estimates in the previous subsection imply u (0) ∈ X n and u (0) ∈ L (cid:0) R + ; L ( R n ) (cid:1) . The latter ensures that I (0) < ∞ , hence c (0) kℓ , ¯ c (0) , and so f (1) , are all well defined for k, ℓ ∈ { , . . . , n } .More precisely (see (4.17)):(4.25) I (0) . J ( a ) / ( n +1) k a k n − / ( n +1)2 , where J ( a ) was defined in (4.15).Let m ∈ { , , . . . } . Let us make the inductive assumptions (IA1)-(IA3) that a global mild solution u ( m − of (N-S m − ) does exist, such that u ( m − ∈ X n , (IA1) u ( m − ∈ L (cid:0) R + ; L ( R n ) (cid:1) (IA2)and that(IA3) I ( m − ≤ L ( a ) . Here,(4.27) L ( a ) := γ n J ( a ) / ( n +1) k a k n − / ( n +1)2 , where the constant γ n > γ n is the compatibility with (4.25), which is of course possible.By (IA3), I ( m − < ∞ and so f ( m ) is well defined by (4.23). Our next goal is to show that a globalsolution u ( m ) of (N-S m ) does exist, satisfying (IA1)–(IA3) with the integer m instead of m − (cid:13)(cid:13)(cid:13)(cid:13)Z t e ( t − s )∆ P ∇ · f ( m ) ( s ) d s (cid:13)(cid:13)(cid:13)(cid:13) n . Z t ( t − s ) − / s − / d s sup s> s / k f ( m ) ( s ) k n . t − / sup s> s / k f ( m ) ( s ) k n . Moreover, by the definition of f ( m ) , we see that k f ( m ) ( s ) k p . I ( m − k φ ( s ) k p , for 1 ≤ p ≤ ∞ . So we obtain that (cid:13)(cid:13)(cid:13)(cid:13) e t ∆ a + Z t e ( t − s )∆ P ∇ · f ( m ) ( s ) d s (cid:13)(cid:13)(cid:13)(cid:13) X n . k a k n + I ( m − sup u ( m ) of (N-S m ) does exist, such that u ( m ) ∈ X n and k u ( m ) k X n . k a k n . u ( m ) can be obtained as the limit for k → ∞ , in the X n -norm, of the sequence ofapproximate solutions u ( m ) k ( k = 0 , , . . . )(4.28) u ( m ) k +1 ( t ) = e t ∆ a + Z t e ( t − s )∆ P ∇ · f ( m ) ( s ) d s + G ( u ( m ) k , u ( m ) k )( t ) , with u ( m )0 ( t ) = e t ∆ a + Z t e ( t − s )∆ P ∇ · f ( m ) ( s ) d s. Similarly as we did in subsection 4.1, we now verify that u ( m ) ∈ X n ∩ X ∞ . For this, we recall that(N-S m ) reads u ( m ) ( t ) = e t ∆ a + R t e ( t − s )∆ P ∇ · f ( s ) d s + G ( u ( m ) , u ( m ) ). Recalling the arguments in (4.6)we only have to check that the forcing term R t e ( t − s )∆ P ∇ · f ( m ) ( s ) d s does belong to X n ∩ X ∞ . But thisis immediate, as one can see applying twice (3.5) with p = 1 and p = (2 n ) ′ , to get (cid:13)(cid:13)(cid:13)(cid:13)Z t e ( t − s )∆ P ∇ · f ( m ) ( s ) d s (cid:13)(cid:13)(cid:13)(cid:13) X n . sup s> s / k f ( m ) ( s ) k n . L ( a ) sup s> s / k φ ( s ) k n . and (cid:13)(cid:13)(cid:13)(cid:13)Z t e ( t − s )∆ P ∇ · f ( m ) ( s ) d s (cid:13)(cid:13)(cid:13)(cid:13) X ∞ . sup s> s / k f ( m ) ( s ) k n . L ( a ) sup s> s / k φ ( s ) k n . We now choose φ such that, in addition to (A1), it satisfies(A2) L ( a ) sup s> s / k φ ( s ) k n ≤ k a k n . Then, just like we did in (4.6), we get k u ( m ) k X n . k a k n (4.29) k u ( m ) k X ∞ . k a k n . (4.30)Let us go further with the relevant L -estimates for u : observe that, from (3.5), (cid:13)(cid:13)(cid:13)(cid:13)Z t e ( t − s )∆ P ∇ · f ( m ) ( s ) d s (cid:13)(cid:13)(cid:13)(cid:13) ≤ πc sup s> s / k f ( m ) ( s ) k ≤ πc I ( m − sup s> s / k φ ( s ) k ≤ πc L ( a ) sup s> s / k φ ( s ) k . (4.31)We now choose φ in a such way that it satisfies also(A3) πc L ( a ) sup s> s / k φ ( s ) k ≤ k a k . Then, estimating the approximate solutions u ( m ) k for k ∈ N , in the same way as we argued to obtain (4.9),from (4.28), we obtain k u ( m ) k +1 ( t ) k ≤ k a k + c Z t ( t − s ) − / k u ( m ) k ( s ) k k u ( m ) k ( s ) k ∞ d s ≤ k a k + πc k u ( m ) k k X ∞ sup s> k u ( m ) k ( s ) k . If necessary, we should put the vector symbol ~u ( m ) k just because we already used with a different meaning u ( m ) k (=the k component of the vector u ( m ) , in (4.22). For the simplicity, we omit the vector symbol since we can distinguish themfrom the context. πc k u ( m ) k k X ∞ . k a k n , from (S), by an appropriate choice of δ >
0, (D6), we can ensure that πc k u ( m ) k k X ∞ ≤ / < t> k u ( m ) ( t ) k ≤ k a k . (4.32)Next step consists in proving decay estimates for k u ( m ) ( t ) k . As in the previous section, we beginestablishing a (uniform-in- m ) quantified version of the decay estimate k u ( m ) ( t ) k = O ( t − / ). Let usrecall that k e t ∆ a k . k a k n/ (3+2 n ) t − / . Concerning the forcing term, we have (cid:13)(cid:13)(cid:13)(cid:13)Z t e ( t − s )∆ P ∇ · f ( m ) ( s ) d s (cid:13)(cid:13)(cid:13)(cid:13) . (cid:16)Z t ( t − s ) − / s − / d s (cid:17) sup s> s / k f ( m ) ( s ) k . t − / L ( a ) sup s> s / k φ ( s ) k . (4.33)This leads us to add another condition to φ , namely(A4) L ( a ) sup s> s / k φ ( s ) k ≤ k a k n/ (3+2 n ) . Therefore(4.34) sup t> t / (cid:13)(cid:13)(cid:13)(cid:13) e t ∆ a + Z t e ( t − s )∆ P ∇ · f ( m ) ( s ) d s (cid:13)(cid:13)(cid:13)(cid:13) . k a k n/ (3+2 n ) . We can now reproduce the same calculations as in (4.10)-(4.12) with an appropriate choice of δ > t> t / k u ( m ) ( t ) k . k a k n/ (3+2 n ) . Next, let us prescribe the conditions on φ in order to obtain a (uniform-in- m ) quantified version ofthe estimate k u ( m ) ( t ) k = O ( t − ( n +1) / ). First of all, let us recall estimate (4.13), that reads k e t ∆ a k . (cid:16)Z R n | x | | a ( x ) | d x (cid:17) / k a k / t − ( n +1) / . To get the same decay rate for the forcing term, we set q = 1 − n . Using (3.5) with p = 2 n/ ( n + 1), weobtain (cid:13)(cid:13)(cid:13)(cid:13)Z t e ( t − s )∆ P ∇ · f ( m ) ( s ) d s (cid:13)(cid:13)(cid:13)(cid:13) . t − ( n +1) / Z t/ k f ( m ) ( s ) k q d s + Z tt/ ( t − s ) − / k f ( m ) ( s ) k d s . t − ( n +1) / L ( a ) (cid:16)Z ∞ k φ ( s ) k q d s + sup s> s ( n +3) / k φ ( s ) k (cid:17) . We then choose φ in such a way that(A5) L ( a ) (cid:16)Z ∞ k φ ( s ) k n/ (2 n − d s + sup s> s ( n +3) / k φ ( s ) k (cid:17) ≤ (cid:16)Z R n | x | | a ( x ) | d x (cid:17) / k a k / . Then we get(4.36) sup t> t ( n +1) / (cid:13)(cid:13)(cid:13)(cid:13) e t ∆ a + Z t e ( t − s )∆ P ∇ · f ( m ) ( s ) d s (cid:13)(cid:13)(cid:13)(cid:13) . (cid:16)Z R n | x | | a ( x ) | d x (cid:17) / k a k / . δ >
0, (D8),by (S), we get sup t> t ( n +1) / k u ( m ) ( t ) k . k a k / (cid:16)R R n | x | | a ( x ) | d x (cid:17) / + k a k n/ (3+2 n ) = J ( a ) . (4.37)¿From the latter estimate, and the already established uniform bound k u ( m ) ( t ) k . k a k , arguing as wedid before in (4.17), we get(4.38) Z ∞ k u ( m ) ( s ) k d s . J ( a ) / ( n +1) k a k n − / ( n +1)2 . Now, let us be more explicit with the constants that appear in two of the estimates obtained before:From (4.25), we infer that there exists K n > a ) such that(4.39) I (0) ≤ K n J ( a ) / ( n +1) k a k n − / ( n +1)2 . From (4.38), we deduce the existence of K ′ n > m and a ) such that(4.40) I ( m ) ≤ K ′ n J ( a ) / ( n +1) k a k n − / ( n +1)2 ( m = 1 , , . . . ) . This leads us to choose, in (4.27), γ n = max { K n , K ′ n } . It then follows that I (0) ≤ L ( a ) and that I ( m ) ≤ L ( a ). This allows to close the inductive argument (IA3).Summarising, we proved the existence of δ > k a k n ≤ δ and the smallnessconditions (A1)–(A5) on φ , there exists a sequence of global solutions u ( m ) ( m = 0 , , , . . . ) of (N-S m ),which is bounded in X n , and also bounded in X n ∩ X ∞ , in L ∞ (cid:0) R + ; L ( R n ) (cid:1) , and in L (cid:0) R + ; L ( R n ) (cid:1) .Strengthening a little bit the smallness conditions on φ , it would be possible in the same way to get alsothe boundedness of u ( m ) under the stronger norm v sup t> (1 + t ) ( n +2) / k v ( t ) k . { u ( m ) } m ∈ N . In this section we study the convergence of u ( m ) in the space(4.41) Y = { v ∈ L ∞ (cid:0) R + ; L ( R n ) (cid:1) : k v k Y := sup s> (1 + s ) ( n +1) / k v ( s ) k < ∞} . This is not a scale invariant space: what it does matter here is that Y is imbedded in L (cid:0) R + ; L ( R n ) (cid:1) .This non-invariance explains why an additional artificial smallness assumption on a appears in the nextLemma. In any case, this artificial assumption will be removed at the end of the proof of our theorem. Lemma 4.1.
Let u ( m ) be the of solution of (N-S m ) constructed in the previous section. There exist twoconstants δ, δ ′ > , depending only on the space dimension, such that if k a k n ≤ δ and (S ′ ) k a k /nn ( J ( a ) − /n + k a k − /n ) ≤ δ ′ , where J ( a ) was defined in (4.15) , then, for m ∈ N , (4.42) k u ( m +1) − u ( m ) k Y . (cid:13)(cid:13)(cid:13)Z t e ( t − s )∆ P ∇ · [ f ( m +1) − f ( m ) ]( s ) d s (cid:13)(cid:13)(cid:13) Y . Proof.
We note that u ( m +1) ( t ) − u ( m ) ( t ) = Z t e ( t − s )∆ P ∇ · [ f ( m +1) − f ( m ) ]( s ) d s + Z t e ( t − s )∆ P ∇ · (cid:2) ( u ( m +1) − u ( m ) ) ⊗ u ( m +1) (cid:3) ( s ) d s + Z t e ( t − s )∆ P ∇ · (cid:2) u ( m ) ⊗ ( u ( m +1) − u ( m ) ) (cid:3) ( s ) d s =: I ( m )1 ( t ) + I ( m )2 ( t ) + I ( m )3 ( t ) .
15e make two separate estimates for I ( m )2 . The first one will be useful for 0 ≤ t ≤ kI ( t ) k ≤ c Z t ( t − s ) − / k u ( m +1) ( s ) − u ( m ) ( s ) k k u ( m +1) ( s ) k ∞ d s . k u ( m +1) − u ( m ) k Y k u ( m +1) k X ∞ Z t ( t − s ) − / (1 + s ) − ( n +1) / s − / d s . k u ( m +1) − u ( m ) k Y k a k n . The second estimate of I ( m )2 ( t ) will be useful for t ≥
1. It is obtained using (3.5) in the first inequality,interpolation in the second, (4.30), (4.32) and (4.37) in the third inequality below: kI ( t ) k ≤ c n/ ( n +1) Z t/ ( t − s ) − ( n +1) / k u ( m +1) ( s ) − u ( m ) ( s ) k k u ( m +1) ( s ) k n/ ( n − d s + c Z tt/ ( t − s ) − / k u ( m +1) ( s ) − u ( m ) ( s ) k k u ( m +1) ( s ) k ∞ d s . t − ( n +1) / k u ( m +1) − u ( m ) k Y Z t/ (1 + s ) − ( n +1) / k u ( m +1) ( s ) k − /n k u ( m +1) ( s ) k /n ∞ d s + (1 + t ) − ( n +1) / k u ( m +1) − u ( m ) k Y k u ( m ) k X ∞ . t − ( n +1) / k a k /nn k u ( m +1) − u ( m ) k Y (cid:16)Z t/ (1 + s ) − ( n +1) / (cid:0) J ( a ) s − ( n +1) / ∧ k a k (cid:1) (1 − /n ) s − / (2 n ) d s + k a k − /nn (cid:17) . t − ( n +1) / k a k /nn k u ( m +1) − u ( m ) k Y (cid:16) J ( a ) − /n + k a k − /n + k a k − /nn (cid:17) . Combining the two previous estimates for I ( m )2 ( t ), we deduce(4.43) kI ( m )2 k Y . k a k /nn (cid:16) J ( a ) − /n + k a k − /n + k a k − /nn (cid:17) k u ( m +1) − u ( m ) k Y . In the same manner, we have(4.44) kI ( m )3 k Y . k a k /nn (cid:16) J ( a ) − /n + k a k − /n + k a k − /nn (cid:17) k u ( m +1) − u ( m ) k Y . Now, if δ >
0, is small enough, (D9), and δ ′ > a fulfils the smallness condition (S ′ ),along with (S). For the convergence of { u ( m ) } , we choose δ > I ( m )1 ( t ): for 0 < t ≤
1, we have kI ( m )1 ( t ) k ≤ c Z t ( t − s ) − / k f ( m +1) ( s ) − f ( m ) ( s ) k d s . sup s> s / k f ( m +1) ( s ) − f ( m ) ( s ) k . But k f ( m +1) ( s ) − f ( m ) ( s ) k ≤ n X k,ℓ =1 k f ( m +1) kℓ ( s ) − f ( m ) kℓ ( s ) k = X k = ℓ (cid:12)(cid:12) c ( m ) kℓ − c ( m − kℓ (cid:12)(cid:12) k φ ( s ) k + n X k =1 (cid:12)(cid:12) ( c ( m ) kk − ¯ c ( m ) ) − ( c ( m − kk − ¯ c ( m − ) (cid:12)(cid:12) k φ ( s ) k k = ℓ , applying (4.35) in the last inequality, and the fact that R ∞ (1+ s ) − ( n +1) / s − / d s < ∞ ,we have: (cid:12)(cid:12) c ( m ) kℓ − c ( m − kℓ (cid:12)(cid:12) = (cid:12)(cid:12)(cid:12)(cid:12)Z ∞ Z R n u ( m ) k u ( m ) ℓ − u ( m − k u ( m − ℓ d y d s (cid:12)(cid:12)(cid:12)(cid:12) . Z ∞ k u ( m ) ( s ) − u ( m − ( s ) k (cid:0) k u ( m ) ( s ) k + k u ( m − k (cid:1) d s . k u ( m ) − u ( m − k Y k a k n/ (3+2 n ) . For the case k = ℓ we have (cid:12)(cid:12) ( c ( m ) kk − ¯ c ( m ) ) − ( c ( m − kk − ¯ c ( m − ) (cid:12)(cid:12) = X i = k (cid:12)(cid:12) ( c ( m ) ii − c ( m − ii ) (cid:12)(cid:12) , so that the same estimates as before applies. Therefore, k f ( m +1) ( s ) − f ( m ) ( s ) k . k φ ( s ) k k a k n/ (3+2 n ) k u ( m ) − u ( m − k Y . Therefore we obtain that for 0 < t < kI ( m )1 ( t ) k . (cid:16) sup s< s / k φ ( s ) k (cid:17) k a k n/ (3+2 n ) k u ( m ) − u ( m − k Y . For t >
1, by similar way, we have with q = 1 − n kI ( m )1 ( t ) k . Z t/ ( t − s ) − n +14 k f ( m +1) ( s ) − f ( m ) ( s ) k q d s + Z tt/ ( t − s ) − / k f ( m +1) ( s ) − f ( m ) ( s ) k d s . t − n +14 (cid:16)Z ∞ k φ ( s ) k q d s + sup s> s n +34 k φ ( s ) k (cid:17) k a k n/ (3+2 n ) k u ( m ) − u ( m − k Y . Then we obtain kI ( m )1 k Y . (cid:16) sup s< s / k φ ( s ) k + Z ∞ k φ ( s ) k n/ (2 n − d s + sup s> s n +34 k φ ( s ) k (cid:17) k a k n/ (3+2 n ) k u ( m ) − u ( m − k Y . By Lemma 4.1, there exists a constant δ ′′ >
0, only depending on the space dimension, such that if(A6) (cid:16) sup s< s / k φ ( s ) k + Z ∞ k φ ( s ) k n/ (2 n − d s + sup s> s n +34 k φ ( s ) k (cid:17) k a k n/ (3+2 n ) ≤ δ ′′ , then k u ( m +1) − u ( m ) k Y ≤ k u ( m ) − u ( m − k Y . This implies that u ( m ) converges in the Y -norm to some limit v ∈ Y .The convergence u ( m ) → v in Y implies that u ( m ) → v in L (cid:0) R + ; L ( R n ) (cid:1) . The fact that v ∈ L (cid:0) R + ; L ( R n ) (cid:1) allows us to define f ( ∞ ) kℓ ( x, t ) = ( c ( ∞ ) kℓ φ ( x, t ) k = ℓ, ( c ( ∞ ) kk − ¯ c ( ∞ ) ) φ ( x, t ) k = ℓ,c ( ∞ ) kℓ = Z ∞ Z R n v k v ℓ d y d s, k, ℓ = 1 , . . . , n, ¯ c ( ∞ ) = c ( ∞ )11 + · · · + c ( ∞ ) nn . In particular, we see that c ( m ) kℓ → c ( ∞ ) kℓ and ¯ c ( m ) → ¯ c ( ∞ ) as m → ∞ . Moreover, it holds that k f ( m ) kℓ ( t ) − f ( ∞ ) kℓ ( t ) k p → m → ∞ for all t ≥ ≤ p ≤ ∞ .17ence, R t e ( t − s )∆ P ∇ · f ( m ) ( s ) d s → R t e ( t − s )∆ P ∇ · f ( ∞ ) ( s ) d s for all t >
0. The strong convergence of u ( m ) to v in Y allows to pass to the limit in the nonlinear term of (N-S m ). This implies that the limit v satisfies the Navier–Stokes equations, with force ∇ · f ( ∞ ) , and initial data a , in its integral form: v ( t ) = e t ∆ a + Z t e ( t − s )∆ P ∇ · f ( ∞ ) ( s ) d s − Z t e ( t − s )∆ P ∇ · ( v ⊗ v )( s ) d s, for all t > v and f ( ∞ ) are well defined, one can easily prove, in a similar way, that u ( m ) → v in X n , then, in other spaces, e.g., in BC (cid:0) R + ; L n ( R n ) (cid:1) by (S), (S ′ ) and (A1)–(A6). Theconvergence u ( m ) → v could be proved to be true also in L with the optimal time-weight, i.e., sup t> (1 + t ) ( n +2) / k u ( m ) ( t ) − v ( t ) k →
0, if we suitable strengthen the smallness conditions, but this will not neededto prove our theorem. v ( t ) as t → ∞ As before, we work in this section under the smallness assumptions on a (S)-(S ′ ) and all the smallnessconditions on φ previously specified.In this subsection, we discuss the large time decay rate of the Lebesgue norms of v ( t ), by applyingthe asymptotic profiles of Miyakawa and Schonbek [13]. We put β k,ℓ = Z ∞ Z R n v k v ℓ d y d s − Z ∞ Z R n f ( ∞ ) kℓ d y d s Here, when k = ℓ , we see that β k,ℓ = Z ∞ Z R n v k v ℓ d y d s − c ( ∞ ) kℓ Z ∞ Z R n φ ( y, s ) d y d s = 0 . When k = ℓ , we see that β k,k = Z ∞ Z R n v k d y d s − ( c ( ∞ ) kk − ¯ c ( ∞ ) ) Z ∞ Z R n φ ( y, s ) d y d s = ¯ c ( ∞ ) . Therefore, β k,ℓ = cδ kℓ for some c ∈ R , and k, ℓ = 1 , . . . , n. Now, the application of Corollary 2.2 gives the desired conclusions for v , namely:lim t →∞ t + n (1 − q ) (cid:13)(cid:13) v ( t ) − e t ∆ a (cid:13)(cid:13) q = 0 , ≤ q ≤ ∞ and, under the additional moment condition R R n y k a j ( y ) d y = 0 for all j, k = 1 , . . . , n ,lim t →∞ t + n (1 − q ) k v ( t ) k q = 0 , ≤ q ≤ ∞ . Let us first complete the proof under the additional artificial smallness assumption (S ′ ) on a . To finishthe proof in this case, we only have to collect all the previous needed conditions on φ (namely, (A1),(A2), (A3), (A4), (A5) and (A6)) and to ensure their compatibility.Let us recall that J ( a ) and L ( a ) were defined by (4.15) and (4.27): J ( a ) = k a k / (cid:16)R R n | x | | a ( x ) | d x (cid:17) / + k a k n/ (3+2 n ) and L ( a ) = γ n J ( a ) / ( n +1) k a k n − / ( n +1)2 , γ n only depends on n . So, the conditions to be imposed on φ are: φ ∈ L ∞ c ( R n × R + ) , Z ∞ Z R n φ ( x, t ) d x d t = 1 , and L ( a ) sup s / k φ ( s ) k ≤ k a k ,L ( a ) sup s> s / k φ ( s ) k ≤ k a k n/ (3+2 n ) L ( a ) (cid:16)R ∞ k φ ( s ) k n/ ( n − d s + sup s> s ( n +3) / k φ ( s ) k (cid:17) ≤ (cid:16)R R n | x | | a ( x ) | d x (cid:17) / k a k / , (cid:16) sup s< s / k φ ( s ) k + Z ∞ k φ ( s ) k n/ (2 n − d s + sup s> s n +34 k φ ( s ) k (cid:17) k a k n/ (3+2 n ) ≤ δ ′′ , (A 1–6)Recall that δ ′′ > n . In fact, nowhere we really needed that φ ∈ L ∞ c ( R n × R + ): the more general condition φ ∈ L ( R n × R + ), together with (A 1–6) would havebeen enough. In any case, the most obvious way to construct such a function φ is to start from acompacly supported and essentially bounded function Φ ∈ L ∞ c ( R n × R + ), supported in a cube K =[ − M, M ] n × [0 , T ] for some M > T >
0, such that R ∞ R R n Φ( x, t ) d x d t = 1. But, for any1 < p ≤ ∞ , we have k R − n Φ( · /R, t ) k p → R → + ∞ . Therefore, there exists R = R (cid:0) n, M, T , k a k , k a k n , R R n | x | | a ( x ) | d x (cid:1) , (with R depending only on the above mentioned parameters), such that, if we take R ≥ R and φ ( x, t ) = R − n Φ( x/R, t ) , then the function φ satisfies all the required conditions (A 1–6).This completes the proof of Theorem 2.3, at least under the additional smallness condition (S ′ ) for a ,that was needed in Lemma 4.1. So, it now remains to show that this smallness condition (S ′ ) can beremoved.This can be done via a standard scaling argument. Indeed, assuming (according with the assumptionsof Theorem 2.3) that R | a ( x ) | (1 + | x | ) d x < ∞ , and that k a k n < δ , we consider the rescaled data a λ = λa ( λ · ). We have k a λ k n = k a k n , k a λ k = λ − n/ k a k and J ( a λ ) = λ − ( n − / J ( a ).So condition (S ′ ) is fulfilled for the rescaled initial data a λ , provided we choose λ ≥ λ for some λ = λ ( n ) depending only on the space-dimension. This is true also when n = 2, as in this case k a k isitself assumed to be small. By what we proved so far, we can construct an external force ∇ · f λ as above(with f λ with compact support in space-time) and a solution u λ of the Navier–Stokes equation arisingfrom a λ with force ∇ · f λ , that satisfies k u λ ( t ) − e t ∆ a λ k q = o ( t − − n (1 − q ) ) as t → ∞ for all 1 ≤ q ≤ ∞ .Now, let us set u ( x, t ) = λ − u λ ( λ − x, λ − t ) and f ( x, t ) = λ − f λ ( λ − x, λ − t ). Then f is compactlysupported in space-time and u solves (N-S). Moreover, k u ( t ) − e t ∆ a k q = o ( t − − n (1 − q ) ) as t → ∞ .Moreover, a has vanishing first-order moments if and only if the same is true for a λ . Therefore, weget under this moment condition the rapid dissipation property k u ( t ) k q = o ( t − − n (1 − q ) ) as t → ∞ .This fully establishes Theorem 2.3. Acknowledgement.
The authors are grateful to Professor Taku Yanagisawa for his comments andto the referees for their careful reading. The second author is partially supported by JSPS Grant-in-Aidfor Young Scientists (B) 17K14215. 19 eferences [1] L. Brandolese,
On the localization of symmetric and asymmetric solutions of the Navier-Stokes equa-tions in R n , C. R. Acad. Sci. Paris S´er. I Math., (2001) 125–130.[2] L. Brandolese, Asymptotic behavior of the energy and pointwise estimates for solutions to the Navier-Stokes equations , Rev. Mat. Iberoamericana, (2004), 225–258.[3] L. Brandolese, Space-time decay of Navier-Stokes flows invariant under rotations , Math. Ann., (2004), 685–706.[4] J. Duoandikoetxea, E. Zuazua,
Moments, masses de Dirac et d´ecompositions de fonctions , C.R. Aca.Sci. Paris, t. 315 , s´erie I, (1992), 693–698.[5] Y. Fujigaki and T. Miyakawa,
Asymptotic profiles of nonstationary incompressible Navier-Stokesflows in the whole space , SIAM J. Math. Anal., (2001), 523–544.[6] H. Fujita and T. Kato, On the Navier-Stokes initial value problem. I
Arch. Rational Mech. Anal., (1964), 269–315.[7] R. Kajikiya and T. Miyakawa, On L decay of weak solutions of the Navier-Stokes equations in R n ,Math. Z., (1986), 135–148.[8] T. Kato, Strong L p -solutions of the Navier-Stokes equation in R m , with applications to weak solu-tions , Math. Z., (1984), 471–480.[9] J. Leray, Sur le mouvement d’un liquide visqueux emplissant l’espace , Acta Math., (1934), 193–248.[10] K. Masuda, Weak solutions of Navier-Stokes equations , Tohoku Math. J. (2), (1984), 623–646.[11] T. Miyakawa, Notes on space-time decay properties of nonstationary incompressible Navier-Stokesflows in R n , Funkcial. Ekvac., (2002), 271–289.[12] T. Miyakawa, On upper and lower bounds of rates of decay for nonstationary Navier-Stokes flows inthe whole space , Hiroshima Math. J., (2002), 431–462.[13] T. Miyakawa and M. E. Schonbek, On optimal decay rates for weak solutions to the Navier-Stokesequations in R n , in “Proceedings of Partial Differential Equations and Applications (Olomouc,1999),” Math. Bohem., (2001), 443–455.[14] T. Okabe and Y Tsutsui, Navier-Stokes flow in the weighted Hardy space with applications to timedecay problem , J. Differential Equations, (2016), 1712–1755.[15] M. E. Schonbek, L decay for weak solutions of the Navier-Stokes equations , Arch. Rational Mech.Anal., (1985), 209–222.[16] Y. Tsutsui, An application of weighted Hardy spaces to the Navier-Stokes equations , J. Funct. Anal., (2014), 1395–1420.[17] M. Wiegner,
Decay results for weak solutions of the Navier-Stokes equations on R n , J. London Math.Soc. (2), (1987), 303–313.[18] M. Wiegner, Decay of the L ∞ -norm of solutions of Navier-Stokes equations in unbounded domains ,Acta Appl. Math.,37