Existence of weak solutions for unsteady motions of generalized Newtonian fluids
LARSDIENING, MICHAELR ˚UZIˇ CKA ANDˇ J ¨ORGWOLF
Abstract. We prove the existence of weak solutionsu:QT →Rnof the equa- tions of unsteady motion of an incompressible fluid with shear-dependent viscos- ity in a cylinderQT =×(0,T), where⊂Rn denotes a bounded domain.
Under the assumption that the extra stress tensorSpossesses aq-structure with q > n2n+2, we are able to construct a weak solutionu ∈ Lq(0,T;W01,q())∩ Cw([0,T];L2())with divu=0. Our approach is based on the Lipschitz trun- cation method, which is new in this context.
Mathematics Subject Classification (2000):76D03(primary);35D05,46,34A34 (secondary).
1.
Introduction. Statement of the main result
Let ⊂
R
n, n ≥ 2, be a bounded domain. For 0 < T < ∞we set QT :=×(0,T). The isothermal motion of a homogeneous, incompressible fluid through is governed by the balance equations for linear momentum and mass, which read1
ρ ∂tu+ρdiv(u⊗u)−divS+ ∇p=ρf in QT, (1.1) divu=0 in QT, (1.2) where u = (u1, . . . ,un)is the velocity, p the pressure,S = (Si j)ni,j=1 the extra stress tensor,f =(f1, . . . ,fn)the external body force andρthe constant density.
In the following we divide the equation (1.1) by the constant densityρand relabel S/ρand p/ρagain asSandp, respectively. Moreover, for mathematical simplicity we assume thatf=divF, whereF=(Fi j)ni,j=1is a given tensor. The above system (1.1), (1.2) has to be completed by boundary and initial conditions and by consti- tutive assumptions for the extra stress tensor. Concerning the former we assume at
1Here, divv:=
i∂xivi, where∂xi = ∂∂xi(i=1, . . . ,n). Received June 3, 2008; accepted in revised form March 10, 2009.
the initial timet =0
u(0)=u0 in , (1.3)
whereu0denotes the initial velocity, which is a given vector field with divu0 =0.
On the boundary∂we assume the following condition of adherence
u=0 on ∂×(0,T). (1.4)
If one assumes a linear dependence of the extra stress tensorSon the symmetric part of the velocity gradient D = D(u) := 12(∇u+ ∇u ), then system (1.1)- (1.2) is just the classical Navier–Stokes equations. However, there exists many homogeneous, incompressible fluids that cannot be adequately described by such a simple constitutive relation. Such fluids are usually called non-Newtonian fluids.
There are many ways in which a non–Newtonian behaviour can manifest itself and we refer the reader to [4] and [26] for a general continuum mechanical background and to [3, 6, 23, 44, 50], and [35] for a detailed discussion of non-Newtonian fluids.
There is a large class of non-Newtonian fluids for which the dominant departure from a Newtonian behaviour is that in a simple shear flow the viscosity and the shear rate are not proportional. Such fluids are called fluids withshear-dependent viscosityorgeneralized Newtonian fluidsand are often modeled by the constitutive law
S=µ(DI I)D, (1.5)
whereDI I =D:Dis the second invariant2ofDandµis the generalized viscosity function. The model (1.5) includes all power-law and Carreau-type models, which are quite popular among rheologists. Such models are used in many areas of en- gineering sciences such as chemical engineering, colloidal mechanics, glaciology, geology, and blood rheology (cf.[36] for a discussion of such models and further references). Typical examples for the constitutive relation (1.5) are
S=µ0
δ+ |D|q−2
D, S=µ0
δ2+ |D|2q−2
2 D, (1.6)
with 1<q<∞,δ≥0 andµ0>0.
Motivated by the above discussion we impose the following conditions onS.
We say that the extra stress tensor S possesses a q-structure if there exist q ∈(1,∞)andδ≥0, such that
(I) S: QT ×
M
nsym→M
nsym is a Carath´eodory function3;2For two matricesA,B∈Rn2byA:Bwe denote the sum
i j Ai jBi j.
3 Mnsymis the vector space of all symmetricn×nmatricesξ = (ξi j)ni,j=1. We equipMnsym with the scalar productξ:ηand the norm|ξ| :=(ξ:ξ)1/2.Bya·bwe denote the usual scalar product inRnand by|a|we denote the Euclidean norm.
Growth condition:
(II) |S(x,t,ξ)| ≤ c0(δ+ |ξ|)q−2|ξ| +κ1(x,t) for all ξ ∈
M
nsym, almost all (x,t)∈QT (c0>0, κ1 ∈Lq(QT), κ1≥0);Coercivity:
(III) S(x,t,ξ):ξ ≥ν0(δ+ |ξ|)q−2|ξ|2−κ2(x,t) for all ξ ∈
M
nsym,almost all (x,t)∈QT (ν0>0, κ2∈L1(QT), κ2≥0);Strict monotonicity:
(IV)
(S(x,t,ξ)−S(x,t,η)):(ξ−η) >0
∀ξ,η∈
M
nsym(ξ=η), almost all(x,t)∈ QT.Before we introduce the notion of a weak solution of the system (1.1)-(1.4) let us provide some notation and function spaces which will be used throughout the paper. As usual letC0∞()denote the space of all smooth functions having compact support in . By Wk,q(), k ∈
N
,1 ≤ q ≤ ∞, we denote the usual Sobolev spaces. We define W0k,q()as the closure of C0∞()in Wk,q(). We will not distinguish between scalar-valued, vector-valued or tension-valued versions of these function spaces. The Lebesgue measure inR
n+1will be denoted byL
n+1(·). For A⊂R
n+1with 0<L
n+1(A) <∞andg∈L1loc(R
n+1)we denote bygAthe mean value ofgover the setA,i.e.vA :=
A
vdX = 1
L
n+1(A)A
vdX.
Let (X, · X)be a normed space. By Lq(0,T;X)we denote the space of all Bochner measurable functionsϕ:(0,T)→ X, such that
ϕLq(0,T;X):= T
0
ϕ(t)qXdt q1
<+∞ if 1≤q <+∞, ϕL∞(0,T;X):=ess sup
t∈(0,T)ϕ(t)X <+∞ if q= +∞.
We set
Hq = Hq():= {ϕ∈C0∞()| divϕ=0}·Lq(), Vq =Vq():= {ϕ∈C0∞()| divϕ=0}·Vq, where
ϕVq := D(ϕ)Lq().
In the case thatq =2 we omit the subscriptqand denoteH :=H2andV :=V2. It immediately follows from our definition ofW01,q()and [39, 18] that there exists a positive constantγ0, such that the following Korn’s inequality holds
∇vLq()≤γ0D(v)Lq() ∀v∈W01,q(). (1.7) Definition 1.1. Assume thatSsatisfies (I) and (II) withq ∈ [n2n+2,∞). LetF ∈ L1(QT)andu0 ∈ H. A vector-valued functionu∈ Lq(0,T;Vq)∩L∞(0,T;H) is called aweak solutionto (1.1)-(1.4) if the following identity
−
QT
u·∂tϕdxdt +
QT
S(x,t,D(u)):D(ϕ)dxdt
−
QT(u⊗u):D(ϕ)dxdt
=
QT
F: ∇ϕdxdt+
u0·ϕ(0)dx
(1.8)
holds for allϕ∈C∞(QT)with divϕ=0 and supp(ϕ)⊂⊂× [0,T)4.
Remark 1.2. By virtue of Sobolev’s embedding theorem and H¨older’s inequality one easily verifies
Lq(0,T;Vq)∩L∞(0,T;H) → Lqn+2n (QT). (1.9) The aim of the present paper is to prove the existence of a weak solution to the system (1.1)-(1.4) for suchqthat div(u⊗u)merely belongs to the spaceL1(0,T; (W01,∞())∗). Our main result is the following:
Theorem 1.3 (Main theorem). Let ⊂
R
n, n ≥2, be a bounded open set, and 0<T <∞. Assume thatSsatisfies(I), (II), (III), and(IV)for some q with2n
n+2 <q <∞ (1.10)
andδ ≥0. Suppose thatu0 ∈ H andF∈ Lq(QT)are given. Then there exists a weak solutionu∈Lq(0,T;Vq)∩Cw([0,T];H)to(1.1)-(1.4).
Remark 1.4.
1. The mathematical analysis of the the system (1.1)-(1.3) was initiated by La- dyzhenskaya (cf.[25, 27, 28]). She proved the existence of weak solutions in the case of Dirichlet boundary conditions (1.4) forq≥1+n2n+2using the monotone operator theory combined with compactness arguments. This weak solution is unique ifq ≥ n+22 andu0∈ H. The same results have been proved in [30] for the extra stress tensorSdepending on the full velocity gradient∇urather than onD(u)only.
4HereA⊂⊂BmeansA,Bare open subsets ofRn,Ais bounded andA¯ ⊂B.
2. The special caseq =2, for which (1.1)-(1.2) are just the classical Navier-Stokes equations, cannot be treated with the theory of monotone operators ifn ≥ 3.
However, the existence of weak solutions in the case of Dirichlet boundary con- ditions (1.4) is well known (cf.[29, 22]).
3. The existence of measure-valued solutions for the system(1.1)-(1.3) with Dirich- let boundary conditions (1.4) was proved forq > n2n+2 in [41, 21] (cf.[33, 32]) using Ball’s theorem on Young measures. The existence of solutions satisfying a variational inequality instead of (1.8) was proved for the system (1.1)-(1.3) with Dirichlet boundary conditions (1.4) forq≥2,n≥2 in [8].
4. The lower bound for the existence of weak solutions of the system (1.1)-(1.3) with space periodic boundary conditions was extended toq> n3n+2by Neˇcas and his collaborators (cf.[33, 5, 32]). The results have been obtained by using−u as a test function and deriving higher fractional differentiability in space. At the same time this technique gives the existence of a more regular solution, namely u ∈ L2(0,T;W2,2())∩ L∞(0,T;Vq)∩Lq(0,T;Vqn
n−2),∂tu ∈ L2(QT)for q ≥1+ n2n+2. This solution is unique in the class of weak solutions, provided u0 ∈ V2. The question of the uniqueness of weak solutions for values of p below 1+n2n+2is still an open problem, which is not adressed in the paper. Note that the extra stress tensorSmust satisfy slightly more restrictive assumptions than (I)-(IV), namely instead of (IV) one needs thatSis uniformly monotone, i.e.∂S∂(ξ)ξ :η⊗η≥c(1+|ξ|)q−2|η|2holds for allξ,η∈
M
nsym(cf.[45,46]). The method was extended in [34] to Dirichlet boundary condition proving for 2 ≤ q < 3,n = 3, the existence of weak solutions and some additional regularity properties for 94 ≤ q < 3,n = 3 (cf.[10, 11] for some improvements). The special caseq = 3,n =2,3, which corresponds to the Smagorinsky model in turbulence, was treated in [42], proving the existence of a unique weak solution which additionally satisfies∂tu∈ L2(0,T;V2)∩L∞(0,T;H), providedu0 ∈ H∩W2,2(). The existence of local in time strong solutions of the system (1.1)- (1.3) with space periodic boundary conditions was proved in [32] forq > 53, n =3. This result was improved in [14] toq > 75,n=3, also providing better regularity properties of the solution. The case of Dirichlet boundary conditions was considered in [7]. In that paper Sis assumed to be uniformly monotone withq ≥1. The case of small data and global in time strong solutions has been treated in [2].5. The lower bound for the existence of weak solutions of the system (1.1)-(1.3) with space periodic boundary conditions was further extended to q > 2(nn++21) in [16] usingL∞-test functions. The same result was obtained in [52] in the case of Dirichlet boundary conditions usingL∞-test functions and the local pressure method.
6. The result in Theorem 1.3 is optimal in the following sense: Forq < n2n+2 the
term
QT(u⊗u):D(ϕ)dxdt, ϕ∈L∞(0,T;W01,∞())
is not well defined using the parabolic embedding (1.9) sinceqn+n2 < 2. Nev- ertheless it is still possible to give sense to the above integral using the informa- tionu∈L∞(0,T;L2())coming from the time derivative only. Moreover, for q < n2n+2 the space W01,q()does not embed intoL2(). Thus we have that Lq(0,T;W01,q()) → Lq(0,T;(W01,q()∗
). Consequently, there are prob- lems to give sense to the distributional time derivative.
The paper is organized as follows. In Section 2 we use the local pressure method to introduce the pressure for the problem (1.1), (1.2) and decompose it similar to the procedure in [52]. However, here we use only standard results for the Stokes system and the divergence equation. In Section 3 we establish the Lipschitz trunca- tion method for unsteady problems. For that we use a Whitney type decomposition to extend a function in which irregular regions have been cut off before. We also prove that this extension belongs to the spaceL∞(0,T;W1,∞())and that a cer- tain partial integration formula with respect to time holds true. Similar results have been proved in [24]. Here it is to the knowledge of the authors the first time that the Lipschitz truncation method is used to prove an existence result for unsteady prob- lems. For steady problems the Lipschitz truncation method was introduced in [1].
In Section 4 we approximate the problem (1.1), (1.2) appropriately. The approxi- mate problem possesses weak solutions (cf.[52]) for which we justify the limiting process using essentially the properties of the Lipschitz truncation established be- fore. This gives a proof of the main Theorem 1.3. The steady version of the problem (1.1), (1.2) was treated in [17, 13] with the help of the Lipschitz truncation method.
2.
Pressure representation
The aim of this section is to introduce the pressure for the problem (1.1)-(1.2).
We decompose the pressure corresponding to the respective terms appearing in the equation. This forms the basis for studying the local behaviour of the velocity and the pressure.
Throughout this section letG ⊂
R
nbe a bounded domain with∂G∈C2. Lemma 2.1. Let 1 < s < ∞. Letv∗ ∈ (W01,s(G))∗ with v∗,v = 0 for all v∈Vs(G). Then there exists a unique p∈Ls(G)with pG =0, such thatv∗,v =
G
p divvdx ∀v∈W01,s(G).
Furthermore there holds
pLs(G)≤cv∗(W1,s
0 (G))∗. (2.1)
Proof. [19, Theorems III.3.1 and III.5.2].
Arguing similar as in [52] we will introduce the pressure.
Theorem 2.2. Let u ∈ Cw([0,T];H(G)) and let Hi ∈ Lsi(0,T;Lsi(G)), 1<si <∞, i =1,2. Suppose that
− T
0
G
u·∂tϕdxdt = T
0
G(H1+H2): ∇ϕdxdt (2.2) holds for allϕ∈C0∞(G×(0,T))withdivϕ=0. Then there exist unique functions
pi ∈Lsi(0,T;Lsi(G)),i =1,2, and ph∈Cw([0,T];W1,2(G)), such that
− T
0
G
u·∂tϕdxdt= T
0
G
(H1+H2): ∇ϕdxdt +
T
0
G
(p1+p2)divϕ+
T
0
G
∇ph·∂tϕdxdt +
G
u(0)·ϕ(0)dx
(2.3)
for allϕ ∈ C∞(G×(0,T))withsupp(ϕ) ⊂⊂ G × [0,T). In addition, we have
−ph=0, ph(0)=0and the apriori estimates
piLsi(G×(0,T))≤ciHiLsi(G×(0,T)), i =1,2, (2.4) ph(t)W1,2(G) ≤chu(t)−u(0)L2(G), t ∈ [0,T], (2.5) with constants ci depending only on n,G, and si,i = 1,2, and a constant ch depending only on n and G.
Proof. Arguing as in the proof of [52, Theorem 2.6] one gets the existence of a pressure p˜ ∈Cw([0,T];L1(G)), such that
G
(u(t)−u(0))ψdx =
G
(H1(t)+H2(t)): ∇ψdx +
Gp˜(t)divψdx, ∀ψ ∈W01,2(G),
(2.6)
for all 0≤t ≤T, where Hi(t)=
t
0
Hi(τ)dτ, 0<t <T, i =1,2.
Without loss of generality we may assume that p˜(t)G =0 for all 0≤t ≤T. Note that in this case the pressure p˜is unique.
Next, letvi(t)∈Vsi(G),0≤t ≤T, denote the unique solution to the Stokes problem
−vi(t)+ ∇πi(t)= −divHi(t) in G, divvi(t)=0 in G, vi(t)=0 on ∂G,
(2.7)
fori =1,2. According to Lemma 2.1 we haveπi(t)∈ Lsi(G)with(πi(t))G =0.
Replacing in (2.7) t byt +h,0 < h ≤ T −h, and taking the difference of both equations we get in view of Lemma 2.1 the estimate
πi(t+h)−πi(t)Lsi(G)≤c
∇(vi(t+h)−vi(t))Lsi(G)+Hi(t+h)−Hi(t)Lsi(G) . From [20, Theorem 2.1], and [19, Theorem IV.6.1] we obtain
∇(vi(t +h)−vi(t))Lsi(G) ≤cHi(t +h)−Hi(t)Lsi(G), which leads to
πi(t +h)−πi(t)sLisi(G)≤c t+h
t
Hi(τ)dτ si
Lsi(G).
Dividing both sides byhsi, integrating over the interval(0,T−h), applying H¨older’s inequality and Fubini’s theorem yields
T−h
0
πi(t+h)−πi(t)sLisi(G)
hsi dt ≤cHisLisi(G×(0,T)).
This shows thatπi ∈ W1,si(0,T;Lsi(G)). Set pi := ∂tπi. In the estimate above passing to the limit ash→0 implies (2.4)
Next, letvh(t)∈V2(G)∩W2,2(G),0≤t ≤T, denote the unique solution to the Stokes problem
−vh(t)+ ∇ph(t)= −u(t)+u(0) in G, divvh(t)=0, in G, vh(t)=0 on ∂G.
(2.8)
Appealing to [19, Theorem IV.6.1] one finds ph(t)∈W1,2(G)with(ph(t))G =0.
Moreover, we have
ph(t)W1,2(G) ≤cu(t)−u(0)L2(G). Whence, (2.5).
Definev(t):=v1(t)+v2(t)+vh(t). From (2.6) follows
−divH2(t)−divH2(t)−u(t)+u(0)= ∇ ˜p(t).
Thus, taking the sum of (2.7) and (2.8) leads to
−v(t)+ ∇(π1(t)+π2(t)+ ph(t)− ˜p(t))=0 in G, divv(t)=0, in G, v(t)=0 on ∂G,
(2.9) which impliesv=0and p˜(t)=π1(t)+π2(t)+ ph(t). Thus, insertingψ(·,t)=
∂tϕ(·,t),t ∈(0,T), into (2.6), integrating this identity over(0,T), and applying integration by parts yields (2.3).
3.
Parabolic Lipschitz truncation
In this section we will establish a Lipschitz truncation method for unsteady prob- lems. The idea is to regularize a given function by cutting off regions of irregularity and then to extend this restricted function by a Whitney type extension to the whole domain again. In a different context this method has been used in [24]. The ba- sis of this approach is a covering lemma of Whitney type, which is well-known (cf.[48, Chapter 6], [12, Chapter1], [9, Theorem 3.2]). Here we use these general results for the Euclidean space
R
n+1with an anisotropic parabolic metric. The typ- ical points(x,t)and(y,s)ofR
n+1are shortly denoted by X andY, respectively.Givenα >0 we equip
R
n+1with the following anisotropic (scaled parabolic) met- ricdα(X,Y):=max{|x−y|,|α−1(s−t)|1/2}, X,Y ∈
R
n+1.ForX0∈
R
n+1andr >0 we define ballsQαr(X0):= {X∈R
n+1|dα(X,X0) <r}, which are equal to the cylindrical sets Br(x0)×(t0−αr2,t0+αr2). One easily checks that for allX ∈R
n+1, 0<r <∞andα >0 holdsL
n+1(Qαr(X))≤2n+2L
n+1(Qαr 2(X)).This implies that(
R
n+1,dα)satisfies the following homogeneity property:There exists a number Nˆ ∈
N
, depending only on n, such that, each cylinder Qr(X0)contains no more thanN pointsˆ {Xi}with dα(Xi,Xj)≥ r4 for i = j . Thus we have (cf.[48, Chapter 6], [12, Chapter 1], [9, 47, Theorem 3.2]) the fol- lowing result:Lemma 3.1 (Whitney type covering). Let E be a non-empty, open, bounded set of
R
n+1equipped with the metric dα,α >0. Then there exists a covering{Qαi} = {Qαrαi(Xiα)}of E, such that5
(W1)
i1
2Qαi = E,
(W2) for all i ∈
N
we have8Qαi ⊂ E and16Qiα∩(R
n+1\E)= ∅, (W3) if Qαi ∩Qαj = ∅then 12rαj ≤riα<2rαj,(W4) each X ∈ E belongs to at most120n+2of the sets4Qαi. We have denoted byγQiα(γ > 0)the ball Qαγrα
i(Xiα). Note that this scaling de- pends onα.
For the readers convenience an elementary proof can be found in the Appendix C of this paper.
5 ByγQαi (γ >0)we denote the ballQαγrα
i (Xαi).
Remark 3.2.
1. The covering{Qαi}ofEhas the following additional properties:
(W5)
j
L
n+1(4Qαj)≤120n+2L
n+1(E).Property (W4) of the above lemma can be written as
iχQαi ≤120n+2, which implies
j
L
n+1(4Qαj)=j
E
χ4QαjdX ≤ 120n+2
L
n+1(E).2. Define Ai :=
j ∈
N
23Qαj ∩ 23Qαi = ∅(i = 1,2, . . .). Note that from Lemma C.1 follows that #Ai ≤120n+2. Moreover, we have
(W6) Qαj ⊂4Qαi ⊂ E for all j ∈ Ai.
Indeed, if 23Qαj ∩ 23Qαi = ∅, then we haverαj ≤2riα due to (W3). ForY ∈ Qαj we estimate
dα(Xiα,Y)≤dα(Xαi,X)+dα(X,Xαj)+dα(Xαj,Y)
< 2 3riα+4
3riα+2riα=4riα. Consequently,Y ∈4Qαi.
Next, let us introduce the notion of Lipschitz continuous functions with respect to the metricdα.
Definition 3.3. For a given Lebesgue measurable set A ⊂
R
n+1 byCd0α,1(A)we denote the space of all essentially bounded functionsu: A→R
, satisfyingLipdα;A(u) <∞, where
Lipdα;A(u):= sup
X,Y∈A X=Y
|u(X)−u(Y)|
dα(X,Y) .
Partition of unity. LetEbe a non-empty, bounded, open subset of
R
n+1equipped with the metricdα. Let{Qαi}denote a Whitney type covering ofEfrom Lemma 3.1.Letηi ∈C0∞(
R
n+1)denote cut-off functions with 0 ≤ηi ≤1 inR
n+1,ηi = 0 inR
n+1\23Qαi,ηi =1 on 12Qαi , such thatLipdα(ηi)+ |α ∂tηi|12 ≤ cn(riα)−1 in
R
n+1 (i ∈N
). (3.1)Recalling the definition of Ai (cf.Remark 3.2) we have
j
ηj =
j∈Ai
ηj in Qαi, and
jηj ∈C∞(
R
n+1). Next, we set ψi(X):=ηi(X)j
ηj(X), X ∈E (i ∈
N
), which is well defined, since by (W1) there holdsjηj ≥1 inE. Clearly, we have thatψi ∈C∞(
R
n+1)satisfyψi ≡0 inR
n+1\23Qαi . Moreover, we have
j
ψj =
j∈Ai
ψj =1 inQαi. (3.2)
Properties (W1), (W2), (W3), (W4), and (3.1) yield
|ψi(X)−ψi(Y)| ≤ |ηi(X)−ηi(Y)| +
j∈Ai
|ηj(X)−ηj(Y)|
≤cn
(riα)−1+
j∈Ai
(rαj)−1
dα(X,Y) (3.3)
≤cn(1+2·120n+2)(riα)−1dα(X,Y).
Similarly, one proves that
|∂tψi| ≤ |∂tηi| +
j∈Ai
|∂tηj| ≤c2n(1+4·120n+2) α−1(riα)−2 in
R
n+1. Thus,Lipdα(ψi)+ |α ∂tψi|12 ≤ cn(1+4·120n+2) (riα)−1 in
R
n+1. (3.4) Definition of the truncation operator. LetG ⊂R
nbe a non-empty, open, bounded set and 0 < T <∞. For a non-empty, open set E ⊂ G×(0,T)let{Qαi}be the corresponding Whitney covering from Lemma 3.1 for the metric dα, and{ψi}the associated partition of unity. Foru∈ L1loc(G×(0,T))we define(
T
Eαu)(X) :=
u(X) if X ∈(G×(0,T))\E, ∞
i=1
ψi(X)uQα
i if X ∈E.
From the construction of
T
Eαu it is clear that the restriction ofT
Eαu upon E is smooth.Next, we prove several properties of the operator
T
Eαu, which will be used in what follows.Lemma 3.4. Let u∈L1loc(G×(0,T)). Then there holds
|
T
Eαu| ≤10n+2M
∗(|u|) a.e. in G×(0,T), (3.5) where c=const>0.Proof. If X ∈(G ×(0,T))\E, then|TEαu(X)| = |u(X)| ≤
M
∗(|u|)(X)by the properties ofM
∗ (cf. Appendix A). Now, let X ∈ E. Then by (W1) we haveX∈Qαi for somei. By the definition of
T
Eαuand of the setAi it follows that (T
Eαu)(X)=j∈Ai
uQα
jψj(X).
Taking into account (W6) and (W3) implies
|
T
Eαu(X)| ≤j∈Ai
|uQα
j|ψj(X)≤8n+2
4Qαi
|u|dY. (3.6)
Since 4Qαi ⊂Qα5rα
i(X)we obtain
|
T
Eαu(X)| ≤10n+2Qα5rα
i(X)|u|dY ≤10n+2
M
∗(|u|)(X).The proof of the lemma is completed.
Lemma 3.5. For every1≤ p≤ ∞there exists a constant c such that
T
EαuLp(G×(0,T))≤cuLp(G×(0,T)) ∀u∈Lp(G×(0,T)), (3.7) where c depends only n.Proof. By the definition of
T
Eα and the properties (W1), (W6), and (W5) of the Whitney covering we have
G×(0,T)|(
T
Eαu)(X)|dX =G×(0,T)\E|u(X)|dX+
E|(
T
Eu)(X)|dX≤ uL1(G×(0,T))+
i
Qiα
|(
T
Eu)(X)|dX≤ uL1(G×(0,T))+
i
Qiα
j∈Ai
Qj|u(Y)|dY
dX
≤ uL1(G×(0,T))+2n+2
i
j∈Ai
4Qαi
|u(X)|dX
≤
1+240n+2
uL1(G×(0,T)).
From the definition of
T
EαfollowsT
EαuL∞(G×(0,T))≤ uL∞(G×(0,T)). The assertion thus follows from interpolation also for 1< p<∞.Lemma 3.6. Let u∈L1loc(G×(0,T)). Then we have for all Y,Z ∈Qαi, i ∈
N
|
T
Eαu(Y)−T
Eαu(Z)| ≤c(riα)−1dα(Y,Z)4Qαi
|u−u4Qα
i|dX, (3.8) where c depends only on n.
Proof. By the definition of
T
Eαone easily calculates for allY,Z ∈QαiT
Eαu(Y)−T
Eαu(Z)=j∈Ai
uQα
j(ψj(Y)−ψj(Z))
=
j∈Ai
(uQα j −uQα
i)(ψj(Y)−ψj(Z)),
where we used (3.2). Using Jensen’s inequality along with property (W6) one gets
|uQα j −uQα
i| ≤4n+2·2
4Qαi
|u−u4Qα
i|dX ∀j ∈ Ai. (3.9) The last two estimates together with the Lipschitz bound (3.4) and the properties (W3) and (W4) of the Whitney covering imply (3.8).
Lemma 3.7. Let u∈L1loc(G×(0,T)). Then for all Y ∈Qiα, i ∈
N
|
T
Eαu(Y)−u(Y)| ≤c4Qαi
|u(X)−u(Y)|dX, (3.10) where c depends only on n.
Proof. LetY ∈Qαi,i ∈
N
. The definition ofT
Eαand (W6) yield|
T
Eαu(Y)−u(Y)| ≤j∈Ai
ψj(Y)|uQα
j −u(Y)| ≤c
4Qαi |u(X)−u(Y)|dX. Whence (3.10).
Note that from Lemma 3.7 one obtains the following estimate fori ∈
N
Qαi
|
T
Eαu−u|dY ≤c4Qαi
|u−u4Qα
i|dY, (3.11)
wherecdepends only onn.