E quations aux
D ´e r i v ´e e s
P a r t i e l l e s
2007-2008
Taoufik Hmidi
On the global well-posedness of the Boussinesq system with zero viscosity Séminaire É. D. P.(2007-2008), Exposé noXXIV, 15 p.
<http://sedp.cedram.org/item?id=SEDP_2007-2008____A24_0>
U.M.R. 7640 du C.N.R.S.
F-91128 PALAISEAU CEDEX Fax : 33 (0)1 69 33 49 49 Tél : 33 (0)1 69 33 49 99
ON THE GLOBAL WELL-POSEDNESS OF THE BOUSSINESQ SYSTEM WITH ZERO VISCOSITY
TAOUFIK HMIDI
(JOINT WORK WITH S. KERAANI)
ABSTRACT. In this paper we prove the global well-posedness of the two-dimensional Boussinesq system with zero viscosity for rough initial data.
1. INTRODUCTION
This paper deals with the global well-posedness for the two-dimensional Boussinesq sys- tem,
(Bν,κ)
∂tv+v· ∇v−ν∆v+∇π=θe2,
∂tθ+v· ∇θ−κ∆θ=0, divv=0,
v|t=0 =v0, θ|t=0 =θ0.
Here,e2 denotes the vector(0, 1),v = (v1,v2)is the velocity field,πthe scalar pressure andθthe temperature. The coefficientsνandκare assumed to be positive;νis called the kinematic viscosity andκthe molecular conductivity.
In the case of strictly positive coefficients ν and κ both velocity and temperature have sufficiently smoothing effects leading to the global well-posedness of smooth solutions.
This was proved by numerous authors in various function spaces (see [4, 9, 15] and the references therein).
Forν>0 andκ=0 the problem of global well-posedness is well understood. In [5], Chae proved global well-posedness for initial data (v0,θ0)lying in Sobolev spaces Hs×Hs, withs>2 ( see also [14]). This result has been recently improved by Keraani and the author [12] to initial data inHs×Hs, withs >0. However we give only a global existence result without uniqueness in the energy spaceL2×L2. In a joint work with Abidi [1], we prove the uniqueness for data belonging toL2∩B∞−1,1×B2,10 . More recently Danchin and Paicu [8] have proved the uniqueness in the energy space.
Our goal here is to study the global well-posedness of the system(B0,κ), withκ>0. First of all, let us recall that the two-dimensional incompressible Euler system, corresponding to θ0 = 0, is globally well-posed in the Sobolev spaceHs, with s > 2. This is due to the advection of the vorticity by the flow: there is no accumulation of the vorticity and thus there is no finite-time singularities according to B-K-M criterion [3]. In critical spaces likeBp,12p+1the situation is more complicate because we do not know if the B-K-M criterion works or not. In [16], Vishik proved that Euler system is globally well-posed in these critical Besov spaces. He used for the proof a new logarithmic estimate taking advantage of the particular structure of the vorticity equation in dimension two. For the Boussinesq
system(B0,κ), Chae proved in [5] the global well-posedness for initial data v0,θ0 lying in Soboloev spaceHs, withs>2. We intend here to improve this result for rough initial data. Our results reads as follows.
Theorem 1.1. Let p ∈ [1,∞[, v0 ∈ B1+
2 p
p,1 be a divergence-free vector field of R2 andθ0 ∈ B−p,11+2p ∩Lr,with2 < r < ∞. There exists a unique global solution(v,θ)to the Boussinesq system(B0,κ),κ>0such that
v∈ C(R+;B1+
2 p
p,1 ) and θ ∈ L∞loc(R+;B−1+
2 p
p,1 ∩Lr)∩eL1loc(R+;B1+
2 p
p,1 ∩B2r,∞). The situation in the case p = +∞ is more subtle since Leray’s projector is not continu- ous on L∞ and we overcome this by working in homogeneous Besov spaces leading to more technical difficulties. Before stating our result we introduce the following sub-space ofL∞ :
u∈B∞ ⇔ kukB∞ :=kukL∞+k∆−1ukB˙0
∞,1 <∞.
We notice thatB∞is a Banach space and independent of the choice of the dyadic partition of unity. For the definition of Besov spaces and the frequency localization operator∆−1
we can see next section. Our second main result is the following:
Theorem 1.2. Let v0 ∈ B∞,11 ,with zero divergence andθ0 ∈ B∞ . Then there exists a unique global solution(v,θ)to the Boussinesq system(B0,κ),κ>0such that
v∈C(R+;B1∞,1) and θ ∈ L∞loc(R+;B∞)∩eL1loc(R+;B2∞,∞). For the proofs it suffices to estimate the quantities k∇v(t)kL∞ and k∇θkL1
tL∞. This will be done by using some logarithmic estimates combined with some smoothing effects.
Theorem 3.1 is very crucial and it describes new smoothing effects for the transport- diffusion equation governed by a vector field which is not necessary Lipschitz but only quasi-lipschitz.
The rest of this paper is organized as follows. In section 2, we recall some preliminary results on Besov spaces. Section 3 is devoted to the proof of smoothing effects. In section 4 and 5 we give respectively the proof of Theorem 1.1 and 1.2. We give in the appendix a commutator lemma.
2. NOTATION AND PRELIMINARIES
Throughout this paper we shall denote by Csome real positive constant which may be different in each occurrence and by C0 a real positive constant depending on the norms of the initial data.
Let us introduce the so-called Littlewood-Paley decomposition and the correspond- ing cut-off operators. There exist two radial positive functions χ ∈ D(Rd) and ϕ ∈ D(Rd\{0})such that
i) χ(ξ) +
∑
q≥0
ϕ(2−qξ) =1; ∀ q≥1, suppχ∩supp ϕ(2−q) =∅ ii) suppϕ(2−p·)∩supp ϕ(2−q·) =∅, if|p−q| ≥2.
For everyv∈S0(Rd)we set
∆−1v=χ(D)v;∀q∈N, ∆qv= ϕ(2−qD)v and Sq=
∑
−1≤p≤q−1
∆p. The homogeneous operators are defined by
∆˙qv= ϕ(2−qD)v, S˙qv=
∑
j≤q−1
∆˙jv, ∀q∈Z.
¿From [2] we split the productuvinto three parts:
uv= Tuv+Tvu+R(u,v), with
Tuv=
∑
q
Sq−1u∆qv and R(u,v) =
∑
|q0−q|≤1
∆qu∆q0v.
Let us now define inhomogeneous and homogeneous Besov spaces. For(p,r)∈[1,+∞]2 ands ∈ Rwe define the inhomogeneous Besov spaceBsp,ras the set of tempered distri- butionsusuch that
kukBsp,r :=2qsk∆qukLp
`r <+∞.
The homogeneous Besov space ˙Bsp,ris defined as the set ofu∈S0(Rd)up to polynomials such that
kukB˙s
p,r :=2qsk∆˙qukLp
`r(Z) <+∞.
LetT>0 andρ ≥1, we denote byLρTBsp,rthe space of distributionsusuch that kukLρ
TBsp,r :=2qsk∆qukLp
`r
LρT <+∞. We say thatubelongs to the spaceeLρTBsp,rif
kuk
eLρTBsp,r :=2qsk∆qukLρ
TLp
`r <+∞.
The following result is due to Vishik [16].
Lemma 2.1. Let d ≥ 2, there exists a positive constant C such that for any smooth function f and for any diffeomorphismψofRd preserving Lebesgue measure, we have for all p ∈ [1,+∞] and for all j,q∈Z,
k∆˙j(∆˙qf ◦ψ)kLp ≤ C2−|j−q|k∇ψη(j,q)kL∞k∆˙qfkLp, with
η(j,q) =sign(j−q). Let us now recall the following result proven in [8, 10].
Proposition 2.2. Letν ≥ 0,(p,r) ∈ [1,∞]2, s ∈]−1, 1[, v ∈ L1loc(R+; Lip(Rd))with zero divergence and f be a smooth function. Let a be any smooth solution of the transport-diffusion equation
∂ta+v· ∇a−ν∆a= f.
Then there is a constant C :=C(s,d)such that for every t∈R+
kak
eL∞tBsp,r+ν
1
mka−∆−1ak
eLmtBsp,r+m2
≤CeCV(t) ka0kBs
p,r +
Z t
0
kf(τ)kBs
p,rdτ
, where V(t):=
Z t
0
k∇v(τ)kL∞dτ.
We shall now give a logarithmic estimate which is an extension of Vishik’s one [16]. For the proof we refer to [13].
Proposition 2.3. Let p,r ∈ [1,+∞], v be a divergence-free vector field belonging to the space L1loc(R+; Lip(Rd))and let a be a smooth solution of the following transport-diffusion equation (withν≥0),
∂ta+v· ∇a−ν∆a= f a|t=0 =a0.
If the initial data a0∈ B0p,r,then we have for all t∈R+
kak
eL∞t B0p,r ≤C ka0kB0
p,r +kfk
eL1tB0p,r
1+
Z t
0
k∇v(τ)kL∞dτ , where C depends only on the dimension d but not on the viscosityν.
3. SMOOTHING EFFECTS
This section is devoted to the proof of a smoothing effect for a transport-diffusion equa- tion with respect to a vector field which is not necessary Lipschitz. This problem was studied by the author [11] in the case of singular vortex patches for two dimensional Navier-Stokes equations. The estimate given below is more precise than [11].
Theorem 3.1. Let v be a smooth divergence-free vector field ofRdwith vorticityω := curlv.
Let a be a smooth solution of the transport-diffusion equation
∂ta+v· ∇a−∆a=0; a|t=0= a0. Then we have for q∈N∪ {−1}and t ≥0
22q Z t
0
k∆qa(τ)kL∞dτ . ka0kL∞
1+t+ (q+2)kωkL1
tL∞+k∇∆−1vkL1 tL∞
. Remark 1. In [10], the author proved in the case of Lipschitz velocity the following esti- mate
(1) 22q
Z t
0
k∆qa(τ)kL∞dτ.ka0kL∞
1+t+
Z t
0
k∇v(τ)kL∞dτ .
But this is not useful in our case. We emphasize that the above theorem is also true when we changeL∞ byLp, withp∈ [1,∞].
Proof. The idea of the proof is the same as in [10]. We use Lagrangian formulation com- bined with intensive use of paradifferential calculus.
Letq∈N∗, then the Fourier localized functionaq:=∆qasatisfies
(2) ∂taq+Sq−1v· ∇aq−∆aq= (Sq−1−Id)v· ∇aq−[∆q,v· ∇]a := gq. Letψqdenote the flow of the regularized velocitySq−1v:
ψq(t,x) =x+
Z t
0 Sq−1v τ,ψq(τ,x)dτ.
We set
¯
aq(t,x) =aq(t,ψq(t,x)) and g¯q(t,x) =gq(t,ψq(t,x)).
¿From Leibniz formula we deduce the following identity
∆a¯q(t,x) =
∑
d i=1D
Hq·(∂iψq)(t,x),(∂iψq)(t,x)E+ (∇aq)(t,ψq(t,x))·∆ψq(t,x), whereHq(t,x):= (∇2aq)(t,ψq(t,x))is the Hessian matrix.
Straightforward computations based on the definition of the flow and Gronwall’s in- equality yield
∂iψq(t,x) =ei+hiq(t,x),
where(ei)id=1is the canonical basis ofRdand the functionhiqis estimated as follows (3) khiq(t)kL∞ .Vq(t)eCVq(t), with Vq(t):=
Z t
0
k∇Sq−1v(τ)kL∞dτ.
Applying Leibniz formula and Bernstein inequality we find (4) k∆ψq(t)kL∞ .2qVq(t)eCVq(t). The outcome is
(5) ∆a¯q(t,x) = (∆aq)(t,ψq(t,x))− Rq(t,x), with
kRq(t)kL∞ . k∇aq(t)kL∞k∆ψq(t)kL∞
+ k∇2aq(t)kL∞sup
i
khiq(t)kL∞+khiq(t)k2L∞ . 22qVq(t)eCVq(t)kaq(t)kL∞.
(6)
In the last line we have used Bernstein inequality.
¿From (2) and (5) we see that ¯aqsatisfies
(∂t−∆)a¯q(t,x) =Rq(t,x) +g¯q(t,x).
Now, we will again localize in frequency this equation through the operator∆j. So we write from Duhamel formula,
∆ja¯q(t,x) = et∆∆jaq(0) +
Z t
0 e(t−τ)∆∆jRq(τ,x)dτ +
Z t
0 e(t−τ)∆∆jg¯q(τ,x)dτ.
(7)
At this stage we need the following lemma (see for instance [7]).
Lemma 3.2. For u∈ L∞and j∈N,
(8) ket∆∆jukL∞ ≤Ce−ct22jk∆jukL∞, where the constants C and c depend only on the dimension d.
Combined with (6) this lemma yields, for everyj∈N,
(9) ke(t−τ)∆∆jRq(τ)kL∞ .22qVq(τ)eCVq(τ)e−c(t−τ)22jkaq(τ)kL∞. Since the flow is an homeomorphism then we get again in view of Lemma 8
ke(t−τ)∆∆jg¯q(τ)kL∞ . e−c(t−τ)22jk[∆q,v.∇]a(τ)kL∞
+ k(Sq−1v−v)· ∇aqkL∞
. (10)
From Proposition 6.1 we have
k[∆q,v· ∇]a(t)kL∞ . ka(t)kL∞
k∇∆−1v(t)kL∞+ (q+2)kω(t)kL∞
. ka0kL∞
k∇∆−1v(t)kL∞+ (q+2)kω(t)kL∞
. (11)
We have used in the last line the maximum principle: ka(t)kL∞ ≤ ka0kL∞. On the other hand sinceq∈N∗, we can easily obtain
k(Sq−1v−v)· ∇aqkL∞ . kaqkL∞2q
∑
j≥q−1
2−jk∆jωkL∞
. ka0kL∞kωkL∞. (12)
Putting together (7), (9), (10), (11) and (12) we find k∆ja¯q(t)kL∞ . e−ct22jk∆ja0qkL∞
+ Vq(t)eCVq(t)22q Z t
0 e−c(t−τ)22jkaq(τ)kL∞dτ + (q+2)ka0kL∞
Z t
0 e−c(t−τ)22jkω(τ)kL∞dτ + ka0kL∞
Z t
0 e−c(t−τ)22jk∇∆−1v(τ)kL∞dτ.
Integrating in time and using Young inequalities, we obtain for allj∈N k∆ja¯qkL1
tL∞ . (22j)−1k∆ja0qkL∞+ (q+2)ka0kL∞kωkL1
tL∞+
ka0kL∞k∇∆−1vkL1 tL∞
+Vq(t)eCVq(t)22(q−j)kaqkL1 tL∞.
Let Nbe a large integer that will be chosen later. Since the flow is an homeomorphism, then we can write
22qkaqkL1
tL∞ = 22qka¯qkL1 tL∞
≤ 22q
|j−
∑
q|<Nk∆ja¯qkL1
tL∞+
∑
|j−q|≥N
k∆ja¯qkL1 tL∞
.
Hence, for allq> N, one has 22qkaqkL1
tL∞ . ka0kL∞+22Nka0kL∞
(q+2)kωkL1
tL∞+k∇∆−1vkL1 tL∞
+ Vq(t)eCVq(t)22N22qkaqkL1
tL∞+22q
∑
|j−q|≥N
k∆ja¯qkL1 tL∞. According to Lemma 2.1, we have
k∆ja¯q(t)kL∞ .2−|q−j|eCVq(t)kaq(t)kL∞. Thus, we infer
22qkaqkL1
tL∞ . ka0kL∞+22Nka0kL∞
(q+2)kωkL1
tL∞+k∇∆−1vkL1 tL∞
+ Vq(t)eCVq(t)22N22qkaqkL1
tL∞+2−NeCVq(t)22qkaqkL1 tL∞. For low frequencies,q≤ N, we write
22qkaqkL1
tL∞.22NkakL1 tL∞. Therefore we get forq∈N∪ {−1},
22qkaqkL1
tL∞ . ka0kL∞+22NkakL1
tL∞
+ 22Nka0kL∞
(q+2)kωkL1
tL∞+k∇∆−1vkL1 tL∞
+ Vq(t)eCVq(t)22N+2−NeCVq(t)
22qkaqkL1 tL∞. ChoosingNandtsuch that
Vq(t)eCVq(t)22N+eCVq(t)2−N .e, wheree<<1. This is possible for small timetsuch that
Vq(t)≤C1, whereC1is a small absolute constant.
Under this assumption, one obtains forq≥ −1 22qkaqkL1
tL∞ . kakL1
tL∞+ka0kL∞
1+ (q+2)kωkL1
tL∞+k∇∆−1vkL1 tL∞
.
Let us now see how to extend this for arbitrarly large time T. We take a partition(Ti)Mi=1 of[0,T]such that
Z Ti+1
Ti
k∇Sq−1v(t)kL∞dt'C1.
Reproducing the same arguments as above we find in view ofka(Ti)kL∞ ≤ ka0kL∞,
22q Z Ti+1
Ti
kaq(t)kL∞dt .
Z Ti+1
Ti
ka(t)kL∞dt+ka0kL∞
+ ka0kL∞
(q+2)
Z Ti+1
Ti
kω(t)kL∞dt
+
Z Ti+1
Ti
k∇∆−1v(t)kL∞dt . Summing these estimates we get
22qkaqkL1
TL∞ . kakL1
TL∞+ (M+1)ka0kL∞+
+ ka0kL∞
(q+2)kωkL1
TL∞+k∇∆−1vkL1 TL∞
. AsM≈Vq(T), then
22qkaqkL1
TL∞ . kakL1
TL∞+ (Vq(T) +1)ka0kL∞+ + ka0kL∞
(q+2)kωkL1
TL∞+k∇∆−1vkL1 TL∞
. Since
k∇Sq−1vkL∞ ≤ k∇∆−1vkL∞+ (q+2)kωkL∞, then inserting this estimate into the previous one
22qkaqkL1
TL∞ . ka0kL∞
(1+T) + (q+2)kωkL1
TL∞+k∇∆−1vkL1 TL∞
.
This is the desired result.
4. PROOF OFTHEOREM1.1
We restrict ourselves to thea prioriestimates. The existence and uniqueness parts can be done in classical manner.
Proposition 4.1. For v0∈ B1+
2p
p,1 andθ0 ∈ Lr,with2<r< ∞,we have for t∈R+ 1)
kθ(t)kLr ≤ kθ0kLr. 2)
kθk
L1tBr,11+2r +kω(t)kL∞+kωk
eL∞tB∞,10 ≤C0eeC0t. 3)
kθk
eL1tB2r,∞+kθk
L1tB1p,1+2p +kvk
eL∞t B1p,1+2p ≤C0eeeC0
t
, where the constant C0depends on the initial data.
Proof. The first estimate can be easily obtained from Lrenergy estimate. For the second one, we recall that the vorticityω =∂1v2−∂2v1satisifies the equation
(13) ∂tω+v· ∇ω =∂1θ.
Taking theL∞ norm we get
(14) kω(t)kL∞ ≤ kω0kL∞+k∇θkL1 tL∞. Using the embeddingB1r,1+2r ,→Lip(R2)we obtain
kω(t)kL∞ . kω0kL∞+kθk
L1tBr,11+2r. (15)
¿From Theorem 3.1 applied to the temperature equation and by Bernstein inequalities we deduce fore>0,
kθkL1
tB2r,∞−e . kθ0kLr1+t+kωkL1
tL∞+k∆−1∇vkL1 tL∞
. kθ0kLr1+t+kωkL1
tL∞+k∇vkL1 tLp¯
, with ¯p:=max{p,r}. This leads forr>2 to the inequality
kθk
L1tB1r,∞+2r .kθ0kLr1+t+kωkL1
tL∞+k∇vkL1 tLp¯
.
On the other hand we have the classical resultk∇vkLp¯ ≈ kωkLp¯. Thus we get
(16) kθk
L1tBr,11+2r .kθ0kLr1+t+kωkL1
tL∞+kωkL1 tLp¯
. The estimate of theLp¯norm of the vorticity can be done as itsL∞ norm
kω(t)kLp¯ ≤ kω0kLp¯ +k∇θkL1 tLp¯. . kω0kLp¯ +kθk
L1tB1r,1+2r. (17)
Set f(t):= kθk
L1tB1r,1+2r, then combining (15), (16) and (17) leads f(t).kθ0kLr(1+t+tkω0kL∞∩Lp¯) +kθ0kLr
Z t
0 f(τ)dτ.
According to Gronwall’s inequality, one has
(18) kθk
L1tBr,11+2r ≤C0eC0t,
whereC0is a constant depending on the initial data. From (15) and (16) we deduce kω0kL∞∩Lp¯ ≤C0eC0t.
Let us now turn to the estimate ofkω(t)kB0
∞,1. Applying Proposition 2.3 to the vorticity equation and using Besov embeddings,
kωk
eL∞tB∞,10 . kω0kL∞+kθkL1 tB∞,11
1+k∇vkL1 tL∞
. kω0kL∞+kθk
L1tBr,11+2r
1+k∇vkL1 tL∞
. (19)
On the other hand we have
k∇v(t)kL∞ ≤ k∇∆−1v(t)kL∞+
∑
q∈N
k∆q∇v(t)kL∞
. k∇∆−1v(t)kLp¯+kω(t)kB0
∞,1
. kω(t)kLp¯ +kωk
eL∞t B0∞,1. (20)
Putting together (18), (19) and (20) and using Gronwall’s inequality gives
(21) k∇vkL∞+kωk
eL∞t B0∞,1 ≤C0eeC0t.
It remains to prove the third point of the proposition. The first smoothing effect onθis a direct consequence of (1) and the above inequality,
kθk
eL1tBr,∞2 ≤C0eeC0t.
For the second one we apply Proposition 2.2 to the temperature equation . To establish the velocity estimate we write
kvk
eL∞t B1p,1+2p .kvkL∞
t Lp+kωk
eL∞t Bp,12p . Using the velocity equation, we obatin
kv(t)kLp .kv0kLp+tkθ0kLp+
Z t
0
kP(v· ∇v)(τ)kLpdτ.
wherePdenotes Leray projector. It is well-known that Riesz transforms act continuously onLp, with 1< p<∞, which yields
kP(v· ∇v)kLp .kv· ∇vkLp . kvkLpk∇vkL∞. Thus we get in view of Gronwall’s inequality and (21)
(22) kvkL∞
t Lp ≤C0eeeC0
t
. It remains to estimatekω(t)k
Bp,12p
. We apply Proposition 2.2 to the vorticity equation com- bined with the temperature smoothing, (18) and (21),
kωk
eL∞t Bp,12p . eCV(t)(kω0k
Bp,12p +kθk
L1tB1p,1+2p).
5. PROOF OFTHEOREM1.2
The case p = +∞ is more subtle and the difficulty comes from the term k∇∆−1vkL∞, since Riesz transforms do not mapL∞to itself. To avoid this problem we use a frequency interpolation method. The main result is the following:
Proposition 5.1. There exists a constant C0 depending onkv0kB0
∞,1 andkθ0kB∞ such that for t ∈[0,∞[
kθ(t)kL∞ ≤ kθ0kL∞; kθkL1
tB∞,11 ≤ C0eC0t3 and
kvk
eL∞t B1∞,1+kθk
eL1tB∞,∞2 +kθ(t)kB∞ ≤ C0eeC0t3.
Proof. TheL∞-bound of the temperature can be easily obtained from the maximum prin- ciple. To give the other bounds we start with the following estimate for the vorticity, which is again a direct consequence of the maximum principle,
(23) kω(t)kL∞ ≤ kω0kL∞+k∇θkL1
tL∞ ≤ kω0kL∞+kθkL1 tB∞,11 .
LetN∈ N∗, then we get by definition of Besov spaces and the maximum principle kθkL1
tB1∞,1 =
∑
q≤N−1
2qk∆qθkL1
tL∞+
∑
q≥N
2qk∆qθkL1 tL∞
. 2Ntkθ0kL∞+
∑
q≥N
2qk∆qθkL1 tL∞. By virtue of Theorem 3.1 one has
kθkL1
tB1∞,1 . 2Ntkθ0kL∞+2−Nkθ0kL∞
1+t+k∇∆−1vkL1
tL∞+NkωkL1 tL∞
. 2Ntkθ0kL∞+2−Nkθ0kL∞
1+t+k∇∆−1vkL1 tL∞
+kωkL1 tL∞. Choosing judiciously Nwe get
(24) kθkL1
tB1∞,1.kωkL1
tL∞+t12kθ0kL∞
1+t+k∇∆−1vkL1 tL∞
12
The following lemma gives an estimate of the low frequency of the velocity.
Lemma 5.2. For all t≥0,wed have
k∇∆−1v(t)kL∞ .1+C0 1+t kωkL∞
t L∞+tkωk2L∞ t L∞, with C0a constant depending on the norms of the initial data.
Proof. FixN∈N∗. Since∆−1 =∆−1(S˙−N+∑0q=−N∆˙q)then we have k∇∆−1vkL∞ . k∇S˙−NvkL∞+
∑
0 q=−Nk∇∆˙qvkL∞
. 2−NkvkL∞+
∑
0−N
k∆˙qωkL∞
. 2−NkvkL∞+NkωkL∞. TakingN≈log(e+kvkL∞)we get
(25) k∇∆−1vkL∞ .1+kωkL∞log(e+kvkL∞).
It remains to estimatekvkL∞. Let M∈Nthen we have kvkL∞ .kS˙−MvkL∞+2MkωkL∞. Now using the equation of the velocity we get
kS˙−Mv(t)kL∞ ≤ kS˙−Mv0kL∞+kPS˙−MθkL1 tL∞
+
Z t
0
kS˙−MdivP(v⊗v)(τ)kL∞dτ . kv0kL∞+k∆−1θkL1
tB˙0∞,1+2−M Z t
0
kv(τ)k2L∞dτ.
We have used the following inequality based on the uniform continuity with respect toq of the operator ˙∆qP : L∞ →L∞ :
kS˙−MdivP(v⊗v)kL∞ ≤
∑
q≤−M−1
k∆˙qdivP(v⊗v)kL∞
.
∑
q≤−M−1
2qkv⊗vkL∞. Thus we obtain
kvkL∞.kv0kL∞+k∆−1θkL1
tB˙0∞,1+2−M
Z t
0
kv(τ)k2L∞dτ+2Mkω(t)kL∞. To estimatek∆−1θkL1
tB˙∞,10 we use the temperature equation, k∆˙qθ(t)kL∞ ≤ k∆˙qθ0kL∞+k∆˙q(v· ∇θ)kL1
tL∞+k∆˙q∆θkL1 tL∞
. k∆˙qθ0kL∞+2qkvθkL1
tL∞+22qkθkL1 tL∞
. k∆˙qθ0kL∞+2qkθ0kL∞kvkL1
tL∞+22qtkθ0kL∞. Therefore we get
q
∑
≤0k∆˙qθ(t)kL∞ .
∑
q≤0
k∆˙qθ0kL∞+tkθ0kL∞+kθ0kL∞
Z t
0
kv(τ)kL∞dτ.
TakingMsuch that
22M ≈ Rt
0kvk2L∞dτ kωkL∞ , we find
kvkL∞ .C0(1+t) +kθ0kL∞
Z t
0
kv(τ)kL∞dτ+kω(t)kL12∞
Z t
0
kv(τ)k2L∞dτ12 . According to Gronwall’s inequality we get
(26) kvkL∞ ≤C0eC0teCtkωkL∞t L∞.
Inserting this estimate into (25) we find the desired inequality.