Review Article Volume 1 Issue 2
Department of Mathematics and Research Institute of Science and Engineering, Waseda University, Japan
Correspondence: Yoshihiro Shibata, Department of Mathematics and Research Institute of Science and Engineering, Waseda University, Ohkubo 3-4-1, Shinjuku-ku, Tokyo 169-8555, Japan
Received: May 29, 2017 | Published: October 18, 2017
Citation: Shibata Y. Global wellposedness of a free boundary problem for the navier-stokes equations in an exterior domain. Fluid Mech Res Int. 2017;1(2):56-72. DOI: 10.15406/fmrij.2017.01.00008
In this paper, we prove a unique existence theorem of globally in time strong solutions to free boundary problem for the Navier-Stokes equations in an exterior domain in the case that initial data are small enough. The key step is to prove decay properties of locally in time solutions, which is derived by combination of maximal Lp-Lq regularity with Lp-Lq decay estimates for solutions of slightly perturbed Stokes equations with free boundary condition in an exterior domain.
Keywords: navier-stokes equations, free boundary problem, global well-posedness, exterior domain, polynomial decay, maximal lp-lq regularity
This paper deals with global well-posedness of the following free boundary problem for the Navier-Stokes equations. Let Ω be an exterior domain in the N dimensional Eucledian space RN occupied by an incompressible viscous fluid. Let Γ be the boundary of Ω that is a C2 compact hyper surface with the unit outer normal n. Let Ωt be the evolution of Ω at time t. Let Γt be the boundary of Ωt with the unit outer normal nt. Let ρ and µ be positive numbers denoting the mass density and the viscosity coefficient, respectively. Let u =⊤(u1, . . . , uN) be an N-vector of functions describing the velocity field, where ⊤M denotes the the transposed M, and let p be a scalar function describing the pressure field. We consider the initial boundary value problem for the Navier-Stokes equations in Ωt given by
ρ(∂tu+u·∇u)−Div(µD(u)−ΡI)=0, div u=0 in ∪0<t<TΩt×{t}, (µD(u)−ΡI)nt =0, VΓt =nt.uon ∪0<t<TΓt×{t},u|t=0 =u0, Ωt|t=0 =Ω (1)
Here, D(u) =∇u +⊤∇u denotes the doubled deformation tensor, I the N × N identity matrix, and VΓt the evolution speed of the surface Γt in the nt direction. Moreover, for any matrix field K with (i, j) componentKij, the quantity Div K is an N vector of functions whose ith component is N∑j=1∂jKij , ∂j=∂/∂xj and for any N vector of function w=(w1,...,wN),divw=N∑j=1∂jwi and the quantity w·∇w is an N -vector of functions whose ith component is N∑j=1∂jwi
One phase problem for the Navier-Stokes equations formulated in (1) with (µD(u)− ΡI)nt=cσHnt in place of (µD(u)− ΡI)nt=0 has been received wide attention for many years, where H is the doubledmean curvature of Γt and cσ is a non-negative constant describing the coefficient of surface tension. Inparticular, the following two cases have been studied by many mathematicians: (1) the motion of an isolated liquid mass and (2) the motion of a viscous incompressible fluid contained in an ocean of infinite extent. In case (1), the initial domain Ω0 is bounded and local well posedness in the case that cσ>0 was proved by Solonnikov1-4 in the L2 Sobolev-Slobodetskii space, by Schweizer5 in the semi group setting, and by Moglievskii and Solonnikov6 in the Holder spaces. And, in the case that cσ=0 , local wellposedness was proved by Solonnikov,7 Mucha and Zaja¸czkowski8,9 in the LpSobolev-Slobodetskii space, and by Shibata and Shimizu10,11 in the Lpin time and Lqin space setting. Global wellposedness in the case that cσ=0 for small initial data by Solonnikov4,7 in the LpSobolev-Slobodetskii space and by Shibata12 in the Lpin time and Lqin space setting. Global wellposedness in the case that cσ>0 was proved under the assumption that the initial domain Ω0 is sufficiently close to a ball and initial data are very small by Solonnikov13 in the L2 Sobolev-Slobodetskii space, by Padula and Solonnikov14 in the Holder spaces, and by Shibata15 in the Lpin time and Lqin space setting.
In case (2), the initial domain
Ω0
is a perturbed layer like:
Ω0 ={x ∈ℝ3 | −b < x3 <η(x′),x′ = (x1,x2)∈ℝ2}
and local wellposed was proved by Allain,16 Beale17 and Tani18 in the L2 Sobolev-Slobodetskii space when
cσ>0
and by Abels19 in the LpSobolev-Slobodetskii space when
cσ=0
.
Global wellposedness for small initial velocity was proved in the L2 Sobolev-Slobodetskii space by Beale20 and Tani and Tanaka21 in the case that
cσ>0
and by Sylvester22 in the case that
cσ=0
. The
decay rate was studied by Beale and Nishida,23 Sylvestre,24 Hataya25 and Hataya and Kawashima.26 In the case of the Ocean problem without bottom,
Ω0 = {x ∈ℝ3 |x3 <η(x′),x′=(x1,x2) ∈ ℝ2}
.
In this case, global well posedness for small initial data and the decay properties of solutions have been studied by Saito and Shibata.28,29 Recently, local well-posedness for the one phase problem of the Navie-Stokes equations was proved in the general unbounded domain case by Shibata12 in the cσ=0 case and by Shibata29,30 in the case cσ>0 .
We remark that two phase problem of the Navier-Stokes equations has been studied by many math- ematicians,31-46 and references therein. Although many papers dealt with global well-posend, as far as the author knows, global well-posedness of free boundary problem for the Navier-Stokes equations in an exterior domain has never be treated, and the purpose of this paper is to prove global well-posedness of problem (1) in the Lpin time and Lqin space setting. Since only polynomial decays are obtained for solutions of Stokes equations with free boundary conditions in the exterior domain case;47,48 it is necessary to choose a large exponent p to guarantees Lpintegrability of solutions, so that the maximal Lp- Lqregularity for the Stokes equations with free boundary consition proved in Shibata30,49,50 and also in Pruess and Simonett40 in the dfferent p and q case are one of essential tools.
Now we consider the transformation that transforms Ωt to a fixed domain. If Ω is a bounded domain, then we have the exponential stability of the corresponding Stokes equations with free boundary conditions in some quotient space, so that we can use the Lagrange transformation to transform Ωt to Ω.7,12 But, is now an exterior domain, so that solutions of the Stokes equations with free boundary conditions decay polynomially as mentioned above. Thus, the Lagrange transformation is not available, because the polynomial decay does not seem to be enough to control the term ∫t0∇u ds . Another known transformation is the Hanzawa one. But, this transformation requires at least the W3−1/qq(N< q) regularity of the height function representing Γ , and such regularity is usually derived from surface tension. In our case, surface tension is not taken into account, so that such regularity is unable to be obtained. To overcome such difficulty, our idea is to use the Lagrange transformation only near Γ .
Let R be a positive number such that Ο=ℝN\Ω⊂BR/2 , where BL ={x∈ℝN||x|<L} , and let κ be a C∞ function such that κ(ξ)=1 for |ξ| ≤ R and κ(ξ)=0 for |ξ| ≥ 2R . Let v(ξ,t) and q(ξ, t) be the velocity field and the pressure field in Lagrange coordinates {ξ} . Let us consider a transformation,
x=L(ξ,t):=ξ+∫t0κ(ξ)v(ξ,s)ds (2)
Let δ be a positive number such that the transformation: x=L(ξ, t) is bijective from Ω on to Ωt={x=L(ξ, t) | ξ ∈ Ω} for each t ∈(0, T) provided that
∫T0‖∇(κ(·)v(·,s))∥L∞(Ω) ds≤δ (3)
Since δ will be chosen as a small positive number eventually, we may assume that 0< δ ≤1 . Let
lij (t)=δij +∫t0∂∂ξj(κ(ξ)vi(ξ,s)) ds (v=⊤(v1=,...,vN)), A(t) (lij(t))−1=(aij(t)) (4)
where
δij
are the Kronecker delta symbols, that is
δii =1
and
δij =0
for
i ≠j
. Here and hereafter, a function
a=a(ξ, t)
is written simply by a(t) and (aij) denotes the
N × N
matrix whose (i, j) component is aij, unless confusion may occur. For a while, we assume that the
N × N
matrix
(lij(t))
is invertible.
Let
Ωt={x=L (ξ, t) | ξ∈Ω}, Γt={x=L(ξ, t)|ξ ∈Γ},
u(x, t) =v(L−1(x, t), t) and p(x, t) =q(L−1(x, t), t) in Eq. (1), and then, v and q satisfy the following equations:
{∂tv−Div(µD(v)−qI)=f(v), in Ω×(0,T), div v=g(v) =div g(v) in Ω×(0,T), (µD(v)−qI)n=h(v) on Γ×(0,T), v|t=0 =u0 in Ω. (5)
Here, f (v) is consisting of some linear combinations of nonlinear functions of the forms
V0(∫t0∇(κv)ds)∇2v, V1(∫t0∇(κv)ds)∂tv, V2(∫t0∇(κv)ds)(∫t0∇2(κv)ds)∇v, (1− κ)(⊤A−1){v·(⊤A∇v)}; (6)and g(v) and g(v) and h(v) are nonlinear functions of the forms:
g(v) =V3(∫to∇(κv)ds)∇v; g(v) =V4(∫t0∇(κv)ds)v; h(v) =V5(∫t0∇(κv)ds)∇v (7)
with some nonlinear functions Visuch that Vi(0) = 0 except for i = 2.
The main result of this paper is the following theorem that shows the unique existence theorem of global in time solutions of Eq. (5) and asymptotics as t → ∞ .
Let N ≥ 3 and let q1 and q2 be exponents such that max(N, 2NN−2) <q2<∞ and 1/q1=1/q2+1/N . Let b and p be numbers defined by
b=3N2q2+12, p=2q2(1+σ)q2−N (8)
with some very small positive number σ . Then, there exists an ∈> 0 such that if initial data u0∈B2(1−1/p)q2,p(Ω)N ∩B2(1−1/p)q1/2,p(Ω)N satisfies the compatibility condition:
div u0 = 0 in Ω, D(u0)n−<D(u0)n,n>n=0 on Γ (9)
and the smallness condition: with I = ∥u0∥B2(1−1/p)q2,p + ∥u0∥B2(1−1/p)q1/2,p , then Eq. (5) admits unique solutions v and q with
v∈Lp((0,∞),H2q2 (Ω) N)∩H1p ((0,∞),Lq2 (Ω) N), q∈Lp((0,∞),H1q2(Ω)+ˆH1q2,0(Ω)) ,
possessing the estimate [v]∞ ≤ Cϵ with
[v]T={∫T0(< s >b∥v(·,s)∥H1∞ (Ω))pds +∫T0(< s >(b−N2q1)∥v(·,s)∥H1q1(Ω))pds+(sup0<s<T <s>N2q1 ∥v(·,s)∥Lq1(Ω))p +∫T0(< s >(b−N2q2)∥v(·,s)∥H2q2(Ω)+∥∂tv(·,s)∥Lq2(Ω))pds}1p.
Here, < s >= (1 + s2)1/2 and C is a constant that is independent of ∈ .
Remark 2
Let p′ = p/(p−1), that is 1/p′ = 1−1/p. And then,
1p′=(1+2σ)q2+N2q2(1+σ)
We choose σ>0 small enough in such a way that the following relations hold,
1<q1<2,Nq1>b>1p′, (Nq1−b) p>1, (b−N2q2)p>1, b≥N2q1,b≥Nq2, (N2q2+12)p′<1, (b−N2q2)p′>1, Nq2+2p<1. (11)
Remark 3
The exponent q2 is used to control the nonlinear terms, so that q2 is chosen as N<q2<∞.
Let
1q1=1N+1q2, 1q3=1q1+1q2. (12)
And the condition: q2>2NN−2 implies that q1 > 2 and q3 > 1 which is necessary to prove Theorem 1.
Thus, we assume that
max(N,2NN−2)<q2<∞ .
Remark 4
We can choose δ > 0 so small that x = L(ξ, t) is a diffeomorphism with suitable regularity from Ω onto Ωt , so that the original problem (1) is globally well-posed.
Finally, we explain several symbols used in this paper. We use bold small letters to denote N -vectors of functions and bold capital letters to denote N × N matrix of functions. For a scalor function h = h(x), ∇h is an N vector whose ith component is ∂ih = ∂h/∂xi . For v(x) = (v1(x), . . . , vN (x)), ∇v is an N × N matrix of functions whose (i, j) component is ∂ivj . Given exponent 1< q< ∞ , let q′ = q/(q − 1).
Let Lq (Ω),Hmq(Ω)and Bsq,p (Ω) be the standard Lebesgue, Sobolev, and Besov spaces on Ω , while ∥ • ∥Lq (Ω) , ∥ • ∥Hmq(Ω), ∥ • ∥Bsq,p (Ω) denote their norms, respectively. For a Banach space X with norm ∥•∥X,
Let {(f1, . . . , fd) | fi ∈X (i = 1, . . . , d)} , while the norm of Xd is written simply by ∥•∥X , which is defined by ∥f∥X =∑dj=1∥fj∥X for f = (f1, . . . , fd)∈Xd . Let ˆH1q,0(Ω)={θ∈Lq,loc(Ω)|∇θ∈Lq(Ω)N , θ|Γ= 0}
ˆH1q(Ω)=ˆH1q,0(Ω)+{p=p1+p2|p1∈H1q(Ω), p2∈ˆH1q,0(Ω)}
For 1 ≤ p ≤ ∞, Lp((a, b), X) and Hpm((a, b), X) denote the standard Lebesgue and Sobolev spaces of X-valued functions defined on an interval (a, b), while ∥•∥Lp((a,b),X) and ∥•∥Hpm((a,b),X) denote their norms, respectively. The letter C denotes generic constants and Ca,b,c,… means that the constant Ca,b,c,… depends on a, b, c... The values of C and Ca,b,c… may change from line to line.
After Introduction (Sect. 1), the paper is organized as follows: In Sect. 2, we reformulate problem (5) by using the formula:
∫t0∇(κ(ξ)v(ξ, s)ds = ∫T0∇(κ(ξ)v(ξ, s)ds−∫T0∇(κ(ξ)v(ξ, s)ds .
In Sect. 3, we give estimations of the nonlinear terms. In Sect. 4, we explain how to prolong local in time solutions to the infinite time interval (0,∞) . Finally, in Sect. 5, we prove Theorem 1 by using maximal Lp-Lq regularity and Lp-Lq decay estimate for solutions of the perturbed Stokes equations with free boundary condition in an exterior domain, which was proved in Shibata.30,47
Another formulation of Eq. (5)
Let T > 0 and let v∈H1p ((0,T),Lq (Ω )N )∩Lp((0,T),H2q (Ω )) , q∈Lp((0,T),H1q (Ω )+ˆH1q,0 (Ω )) (13)
be solutions of Eq. (5) satisfying the condition (3). In what follows, we rewrite Eq. (5) in order that the nonlinear terms have suitable decay properties.
Let A(t) = (aij(t)) be the N×N matrix given in (4), and let nt = (nt1, . . . , ntN)⊤ and n=(n1, . . . , nN)⊤, and then by the transformation (2), we have
aji(t)∂∂ξj, ∂∂xj= N∑j=1nti= d(t) N∑j=1aji(t)nj (14)
where d(t)=|T A(t)n|.
aji(t)=δij+˜aij (t), J(t)=1+˜J(t), lij(t)=δij+˜lij(t) , (15)
Where lij are given in (4) and J is the Jacobian of the transformation (2), that is J = J= det(lij), and then
˜aij (t)=bij (∫t0∇(κ(ξ)v(ξ, s))ds), ˜J(t) =K(∫t0∇(κ(ξ)v(ξ, s))ds),
˜lij (t)=mij (∫t0∇(κ(ξ)v(ξ, s))ds):= ∫t0∂∂ξj(κ(ξ)vi(ξ, t))ds) (16)
With some smooth functions bijand K defined on {w | |w| ≤ 1} such that bij(0) = K(0) = 0, where w is the corresponding variable to ∫t0∇(κ(ξ)v(ξ, s))ds Let u(x, t) =v(ξ, t) and p(x, t) =q(ξ, t) and then u and p satisfy Eq. (1). By (14),
∂ui∂xj+∂uj∂xi= Dij,t(v) :=Dij(v)+˜Dij(t)Δv
With
Dij(v)=∂vi∂ξj+∂vj∂ξj, ˜Dij(t)Δv=N∑K=1˜akj(t)∂vi∂ξk+˜aki(t)∂vj∂ξk (17)
We also have an important formula:
div u=N∑j=1∂uj∂xj=N∑j,k=1J(t)akj(t)∂vj∂ξk=N∑j,k=1∂∂ξk(J(t)akj(t)vj) (18)
which implies that
N∑j,k=1˜akj(t)=˜J(t)akj(t)∂vj∂ξk=N∑j,k=1∂∂ξk{(˜akj(t)+˜J(t)akj(t))vj} (19)
And then, Eq. (5) is written as follows:
{N∑i=1lis(t)(∂tvi+ (1−κ)N∑j,k=1vjakj(t)∂vi∂ξk) −μN∑i,j,k=1lis(t)akj(t)∂∂ξkDi,j,t(v)−∂q∂ξs=0 in Ω×(0,T),N∑j,k=1J(t)akj(t)∂vj∂ξk=N∑j,k=1∂∂ξk(J(t)akj(t)vj)=0 in Ω×(0,T) μN∑i,j,k=1lis(t)akj(t)Di,j,t(v)nk−qns=0 on Γ×(0,T), v|t=0= u0 in Ω (20)
Where s runs from 1 through N. Here, we have used the fact that (lij)=A−1 which follows from (4).
In order to get some decay properties of the nonlinear terms, we write
∫t0∇(κ(ξ)v(ξ, s)ds = ∫T0∇(κ(ξ)v(ξ, s)ds−∫T0∇(κ(ξ)v(ξ, s)ds .
In (16), by the Taylor formula we write
aij(t)=aij(T)+Aij(t), lij(t)=lij(T)+ Lij(t),Dij,t=Dij,T(v)+Dij(t)Δv J(t)=J(T)+J(t) (21)
With
Aij(t)=∫10−b′ij∫T0∇(κ(ξ)v(ξ, s)ds−θ∫Tt∇(κ(ξ)v(ξ, s)ds )dθ ∫Tt∇(κ(ξ)v(ξ, s)ds ,Lij(t)=−∫Tt∂∂ξjκ(ξ)vi(ξ, s)ds , Dij(t)Δv=N∑κ=1(Akj(t)∂vi∂ξk+Aki∂vj∂ξk) ,J(t)=−∫10K′(∫T0∇(κ(ξ)v(ξ, s)ds−θ∫Tt∇(κ(ξ)v(ξ, s)ds )dθ∫Tt∇(κ(ξ)v(ξ, s)ds
Where b′ij and K′ are derivatives of bij and K with respect to w. By the relation:
N∑s=1lis(T)asm(T) =δim , (22)
the first equation in (20) is rewritten as follows:
∂tvm−μN∑j,k=1ajk(T)∂∂ξk(μDmj,T(v)−δmjq=fm(v)
With
fm(v)=−N∑s=1asm(T){N∑i=1Lis(t)∂tvi+N∑i,j,k=1(1−κ)lis(t)akj(t)vi∂vi∂ξk} +μN∑s=1asm(T){N∑i,j,k=1lis(T)akj(T)∂∂ξk(Dij(t)∇v)+N∑i,j,k=1lis(T)Akj(t)∂∂ξkDij,t(v)} +N∑i,j,k=1Lis(t)akj(t)∂∂ξkDij,t(v) (23)
Next, by (18)
˜div v= g(v) = div g(v)
With
˜div v= N∑j,k=1J(T)akj(T)∂vi∂ξk=N∑j,k=1∂∂ξkJ(T)akj(T)vjg(v)=N∑j,k=1(J(T)Akj(t)+J(t)akj(t))∂vj∂ξkgk(v)=N∑j=1(J(T)Akj(t)+J(t)akj(t)vj, g(v)=(g1(v)....gN(v)) (24)
Finally, we consider the boundary condition. Let ˜n be an N -vector defined on ℝN such that ˜n=n on Γ and∥˜n∥H2∞ (ℝN ) ≤ C . In what follows, ˜n is simply written by n=(n1, . . . , nN )⊤ . By (14) and (22)
N∑j,k=1ajk(T)(μDmj,T(v)−δmjq)nk=hm(v)
with
hm(v)=−μN∑j,k=1(akj(T)Dmj(t)∇v+Akj(t)Dmj,t(v))nk −μN∑i,j,k,s=1asm(T)Lis(t)akj(t)Dij,t(v)nk (25)
By (18),
μN∑j,k=1akj(T)∂∂ξk(μDmj,T(v)−δmjq) = J(T)−1N∑j,k=1∂∂ξk{J(T)ajk(T)(Dmj,T(v)δmjq)}
And
N∑j,k=1akj(T)Dmj,T(v)nk−(N∑s=1asm(T)ns)q=N∑k=1[N∑j=1{J(T)akj(T)(Dmj,T(v)−δmjq}]nk
Thus, letting
Smk(v,q)=N∑j=1J(T)akj (T)(µDmj,T (v)−δmjq), ˜S(v,q)=(Sij (v,q)),
f(v)=(f1(v),...,fN(v))⊤, h(v)=(h1(v),...,hN(v))⊤,
and using (18), we see that v and q satisfy the following equations:
{∂tv−J(T−1)Div ˜S(v,q)=f(v) in Ω×(0,T), ˜ div v= g(v) = div g(v) in Ω×(0,T), ˜S(v,q)n=h(v) on Γ×(0,T), v|t=0=u0 in Ω . (26)
Let f (v), g(v), and h(v) are functions defined in Sect. 2. In this section, we estimate these functions. In what follows we write
∥< t > αw∥Lp((0,T),X) ={∫T0(< t>α∥w(·,t)∥X )pdt}1p1≤p<∞,∥< t > αw∥L∞((0,T),X)=esssup0<t<T< t>α∥w(·,t)∥X p=∞ .
First, we prove that
∥<t>bf ∥Lp((0,T), Lq1/2(Ω) + ∥ < t >f ∥Lp((0,T ), Lq2 (Ω)) ≤ C(I + [v]2T ). (27)
with I=‖u0‖B2(1−1/p)q1/2,p(Ω) +‖u0‖B2(1−1/p)q2,p(Ω) . Here and in what follows, C denotes generic constants independent of I, [v]T , δ, and T . The value of C may change from line to line. Since we choose I small enough eventually, we may assume that 0<I≤ 1 . Especially, we use the estimates:
I2 ≤I, I[v]T≤12(I2+[v]2T)≤I+[v]2T
Since
∫βα∥∇(κv(·,s))∥L∞(Ω)≤C(1 +α)−b+1p′(∫βα(<s>b∥v(·,s)∥H1∞ (Ω))pds)1p∫βα∥∇2(κv(·,s))∥Lq(Ω)≤C(1 +α)−b+N2q2+1p′(∫βα(<s>b−N2q2∥v(·,s)∥H2q2(Ω))pds)1p
for any 0≤α<β≤T , where q∈[1, q2] , we have
∫βα∥∇(κv(·,s))∥L∞(Ω)ds ≤C[v]T(1+α)−b+1p′∫βα∥∇2(κv(·,s))∥Lq(Ω)≤C[v]T (28)
for any 0≤ α<β≤ T , where q∈[1, q ] , because b > N2q2+1p′ as follows from (11). By real interpolation theorem, we have
sup t∈(0,T)<t>b− N2q2‖v(·,t)‖B2(1−1/p)q2,p(Ω)≤C‖u0‖B2(1−1/p)q2,p(Ω) + ‖< t >b− N2q2v‖Lp((0,T),H2q2 (Ω)) +‖<t >b− N2q2∂tv‖Lp((0,T)Lq2(Ω)) (29)
To prove (29), we introduce an operator T(t) acting of g∈B2(1−1/p)q2,pℝN defined by
T(t)g=F−1[e−(|ξ| +1)tF[g](ξ)], (30)
where F and F−1 denote the Fourier transform in ℝN and its inverse transform. We have
eγtT(t)g∥Lp((0,∞),H2q( ℝN)) +eγt∂tT(t)g∥Lp((0,∞),Lq( ℝN))≤C||g||B2(1−1/p)q2,p(ℝN) (31)
Given f(t) defned on (0, T) with f|t=0= 0 , let
[eTf](t)={ 0 (t <0),f(t) (0<t<T),f(2T−t) (T < t <2T), 0 (2T <t), (32)
and then [eTf](t)=f(t) for t∈(0, T) and
∂t[eTf](t)={ 0 (t <0),∂tf(t) (0<t<T),−(∂tf)(2T−t) (T < t <2T), 0 (2T <t), (33)
Let ˜u0 be an N-vector of functions in B2(1−1/p)q2,p(ℝN)N such that ˜u0=u0 in Ω and ‖˜u0‖B2(1−1/p)q1/2,p ℝN <C‖u0‖B2(1−1/p)q2,p(Ω) .
Let z=eT[<t>b− N2q2v−T(|t|)˜u0]+T(|t|)˜u0 for t∈ℝ . Since <t>b− N2q2v−T(|t|)˜u0|t=0=0 in Ω by (31), (32) and (33),
∥z∥Lp((0,∞),H2q2(Ω))+ ∥∂tz∥Lp((0,∞),Lq2 (Ω))<C(||<t>b− N2q2v‖Lp((0,T),H2q2 (Ω)) +< t >b− N2q2∂tv‖Lp((0,T)Lq2(Ω))+‖u0‖B2(1−1/p)q2,p(Ω) (34)
It is known (Tanabe51) that
Lp((0, ∞), E1) ∩ H1p((0, ∞), E0)
is continuously imbedded into
BUC((0, ∞), (E0,E1)1−1/p,p)
, where E0 and E1 are two Banach spaces such that E1 is a dense subset of E0, and BUC denotes the set of all uniformly bounded continuous functions . Noting that
z(t) =< t >b− N2q2v(t) for t ∈ (0, T )
, we have
sup 0<t<T)<t>b− N2q2‖v(t)‖B2(1−1/p)q2,p(Ω)≤ sup t∈(0,∞)‖z(t)‖B2(1−1/p)q2,p(Ω) ≤ C‖z‖Lp((0,∞),H2q2 (Ω)) +‖∂tz‖Lp((0,∞),Lq2(Ω))
which, combined with (34), furnishes (29).
Since 2p+Nq2<1, B2(1−1/p)q2,p(Ω) is continuously imbedded into H1∞(Ω) so that by (29)
<t>b− N2q2‖v(t)‖L∞((0,T),H1∞ (Ω)) ≤C(I+[v]T) (35)
Applying (3), (28) and (29) to the formulas in (15) and (16) and using the fact that −b+1p′N2q2≤−N2q2 and −b+N2q2≤−N2q2 ,which follows from (11), give
‖(aij(t),J(t), lij(t),Aij(t),J(t),Lij(t)‖L∞ (Ω)≤C ,
‖(Aij(t),J(t),Lij(t)‖L∞ (Ω)≤C∫Tt∇(κv(·,s))∥L∞(Ω)ds ≤C[v]T+<t>−b+1p′≤C[v]T<t>−N2q2,
‖∇(aij(t),J(t), lij(t),Aij(t),J(t),Lij(t)‖Lq (Ω)≤ C∫T0∇2(κv(·,s))∥Lq≤C[v]T, (36)
‖∂t(aij(t),J(t), lij(t),Aij(t),J(t),Lij(t)‖L∞ (Ω) ≤ C‖∇(κv(·,t))‖,L∞(Ω)≤C(I+[vT])<t>−b+ N2q2≤C(I+[vT])<t>N2q2
for any t ∈ (0, T] , where q ∈ [1, q2] . Moreover, we have
(˜aij,˜J, ˜lij,˜Aij,J,Lij)(x,t)=0 for x∉B2R and t∈[0,T] (37)
By (36) and (37),
‖(asm(T)Lis(t)∂tvi)‖Lq (Ω)C[v]T+<t>−b+1p′‖∂tvi(t)‖Lq2 (Ω)
for any q ∈ [1, q2] . Since 1p′<b−N2q2 as follows from (11), we have
‖<t>basm(T)Lis(t)∂tvi)‖Lp((0,T),Lq (Ω)) ≤C(I+[v]2T)
for any q ∈ [1, q2]
Next, by Holder’s inequality,
<t>b‖v·∇v(·,t)‖Lq1/2(Ω)≤<t>N2q1‖∇v(·,t)‖Lq1(Ω)
so that by (36), we have
‖<t>basm(T) lisakjvj∂vi∂ξk‖Lp((0,T),Lq1/2(Ω)) ≤C[v]2T.
Since
<t>b‖v·∇v(·,t)‖Lq2(Ω)≤<t>N2q2‖∇v(·,t)‖L∞(Ω)<t>b− N2q2‖∇v(·,t)‖Lq2(Ω),
by (36)
‖<t>basm(T) lisakjvj∂vi∂ξk‖Lp((0,T),Lq2 (Ω)) ≤C(I+[v]T)[v]T≤C(I+[v]2T)
Since
∂∂ξk(Dij(t)∇v)=N∑m=1(Amj(t)∂2vi∂ξk∂ξm+Ami(t)∂2vj∂ξk∂ξm)+N∑m=1((∂∂ξmAmj(t)∂vi∂ξm+∂∂ξkAmi(t))∂vj∂ξm)
by (36)
<t>b‖∂∂ξk(Dij(.)∇v)‖Lq(Ω)≤C[v]T{<t>b− N2q2‖∇2v(.t)‖Lq(Ω)+<t>b ‖∇v(·,t)‖L∞(Ω)} (38)
for any q ∈ [1, q2] and therefore
‖<t>basm(T) lis(T)akj(T)∂∂ξk(Dij(.)∇v)‖Lp((0,T),Lq (Ω))≤C[v]2T
for any q ∈ [1, q2] . Since
∂∂ξk(Dij,T(v))=N∑m=1(amj(T)∂2vi∂ξk∂ξm+ami(T)∂2vj∂ξk∂ξm)+N∑m=1((∂∂ξmamj(T)∂vi∂ξm+∂∂ξkami(T))∂vj∂ξm),
by (36)
<t>b‖asm(T) lis(T)Akj(t)∂∂ξk(Dij,T(v)‖Lq (Ω)≤C[v]T{<t>b− N2q2‖∇2v(.t)‖Lq(Ω)+<t>b ‖∇v(·,t)‖L∞(Ω)} ,
so that
<t>b‖asm(T) lis(T)Akj(T)∂∂ξk(Dij,T(v)‖Lp((0,T),Lq (Ω))≤C[v]2T
for any q ∈ [1, q2] . Analogously, we have
‖<t>basm(T) Lisakj∂∂ξk(Dij,T(v)‖Lp((0,T),Lq (Ω))≤C[v]2T
for any q ∈ [1, q2] . Summing up, we have obtained (27).
Next, we consider 𝔤 and g. To estimate the H12p norm, we use the following lemma.
Lemma 5.
Let f∈H1∞(ℝ, L∞(Ω)) and g∈H12p(ℝ, Lq(Ω)) . Assume that f(x, t) = 0for (x, t)∉BR×ℝ .
Then,
‖ fg‖H12p(ℝ, Lq(Ω))≤C‖f‖H1∞(ℝ, L∞(Ω)) ‖g‖H12p(ℝ, Lq2(Ω)) (39)
Proof: To prove the lemma, we use the fact that
H12p(ℝ, Lq(Ω))=(Lp(ℝ, Lq(Ω)),H1p(ℝ, Lq(Ω)))[12] , (40)
where (·,·)[σ] denotes a complex interpolation functor. Let q ∈ [1, q2] . Noting that f(x, t) = 0 for (x, t)∉BR×ℝ , we have
‖∂t (fg)‖Lq(Ω)≤‖∂tf‖L∞(Ω) ‖g‖Lq2(Ω)+‖f‖L∞(Ω)+‖∂tg‖Lq2(Ω),
and therefore
‖∂t (fg)‖Lp(ℝ, Lq(Ω))≤C‖f‖H1∞(ℝ, L∞(Ω)) ‖g‖H1p(ℝ, Lq2(Ω)) .
for any q ∈ [1, q2] . Moreover, we easily see that
‖ (fg)‖Lp(ℝ, Lq(Ω))≤C‖f‖L∞(ℝ, L∞(Ω)) ‖g‖Lp(ℝ, Lq2(Ω)).
Thus, by (40), we have (39), which completes the proof of Lemma 5.
To use the maximal Lp-Lq estimate, we have to extend 𝔤, g and h to R. For this purpose, we introduce an extension operator ˜eT . Let f be a function defined on (0, T ) such that f|t=T = 0 , and then ˜eT is an operator acting on f defined by
[˜eTf](t)={0 (t>T),f(t) (0<t<T),f(−t) (−T<t<0),0 (t<−T) . (41)
Lemma 6
Let 1 < p < ∞, 1 ≤ q ≤ q2 and 0 ≤ a ≤ b . Let f∈H1∞(0,T, L∞(Ω)) and g∈H1p((0,T),Lq2(Ω))∩Lp((0,T)H2q(Ω)) . Assume f|t=T = 0 and f=0 for (x, t)∉BR×ℝ . Let <t>=(1+t2)1/2 . Then we have
‖ ˜eT (<t>a f∇g) ‖H12p(ℝ, Lq(Ω)) ≤C<t>N2q2 f‖H1∞((0,∞),L∞(Ω)) ×(‖<g‖Lp((0,T)H2q(Ω))+‖<t>b− N2q2∂tg‖Lp((0,T),Lq2(Ω))+‖g|t=0‖B2(1−1/p)q2,p(Ω)) (42)
Proof: Let f0(t)=<t>a−b+N2q2f(t) and g0 (t) =< t >b+N2q2g (t) , and then < t >a f∇g =f0∇g0 . Let T (t) be the operator given in (30) and let h be a function in B2(1−1/p)q2,p(ℝN) such that h=g|t=0 in Ω and h=g|t=0 ‖h‖B2(1−1/p)q2,p(Ω)≤‖g|t=0‖B2(1−1/p)q2,p(Ω) . Recall the operator eT defined in (32) and note that g0 |t=0 =g |t=0=T(t)h|t=0 in Ω . Let ˜g(t)=eT [g0 −T(·)h](t)+T(t)h
for t > 0 and let
[ιg](t)={˜g(t) (t>0),˜g(−t) (t<0), [ιf](t)={0 (t>T),f0(t) (0<t<T),f0(−t) (−T<t<0),0 (t<−T) .
Since ˜g(t) = g0 (t) for 0< t<T , we have
˜eT [<t>a f∇g](t)={0 (t>T),f0(t) ∇g0(t) (0<t<T),f0(−t)∇g0(−t) (−T<t<0),0 (t<−T) .={0 (t>T),f0(t) ∇˜g(t) (0<t<T),f0(−t)∇˜g(−t) (−T<t<0),0 (t<−T) . =[ιf](t)∇[ιg](t).
By Lemma 5,
‖˜eT [<t>a f∇g] ‖H12p(ℝ, Lq(Ω)) = ‖ [ιf]∇[ιg] ‖H12p(ℝ, Lq(Ω)) ≤C‖ [ιf] ‖H1∞(ℝ, Lq(Ω))‖∇(ιg)‖H12p(ℝ, Lq(Ω))
Since, f0|t=T = 0 we have
‖[ιf]‖H1∞(ℝ, Lq(Ω))=2‖[f0]‖H1∞((0,T),L∞(Ω))≤ ‖<t>N2q2 f‖H1∞((0,T),L∞(Ω)) ,
Because a− b≤0 .
To estimate ‖∇(ιg)‖H12p(ℝ, Lq(Ω)) , we use the fact that H1p(ℝ, Lq2(Ω))∩ Lp(ℝ, H2q2(Ω)) is continuously imbedded into H12p(ℝ, H1q2(Ω)) which was proved by Meyries and Schnaubelt52 in case of p = q2 and by Shibata30 for any 1<p, q2 <∞ . Using this fact and (31), we have
‖∇(ιg)‖H12p(ℝ, Lq2(Ω)) ≤C ‖(ιg)‖H1p(ℝ, Lq2(Ω)) +‖(ιg)‖Lp(ℝ, H2q2(Ω))
≤C(‖˜g‖H1p((0,∞),Lq2(Ω))= ‖˜g‖Lp((0,∞),H2q2(Ω)))
≤C( ‖g0-T(·)h‖H1p((0,T),Lq2(Ω))+ ‖g0-T(·)h‖Lp((0,T),H2q2(Ω))+‖T(·)h‖H1p((0,T),Lq2(Ω))+‖T(·)h‖Lp((0,∞),H2q2(Ω)))
≤C(‖<t>b− N2q2∂tg‖Lp((0,T),Lq2(Ω)))+‖<t>b− N2q2∂tg‖Lp((0,T)H2q(Ω))+‖g|t=0‖B2(1−1/p)q2,p(Ω)
This completes the proof of Lemma 6.
Recall the definitions of g(v) and hm(v) given in (24) and (25). By Lemma 6 and (36)
‖ ˜eT (<t>a g(v) ‖H12p(ℝ, Lq(Ω))
≤N∑j,k=1‖<t>N2q2(J(T)Akj (·)+T(·)akj (·))∥H1∞ ((0,T ),L∞(Ω))×(‖<t>b− N2q2v‖Lp((0,T)H2q2(Ω))+‖<t>b− N2q2∂tv‖Lp((0,T),Lq2(Ω))+‖u0‖B2(1−1/p)q2,p(Ω)≤C(I+[v]2T) (43)
for any a∈[0, b] and q ∈ [1, q2] . Analogously, we have
‖ ˜eT (<t>a h(v) ‖H12p(ℝ, Lq(Ω)) ≤C(I+[v]2T) (44)
for any a∈[0, b] and q ∈ [1, q2] . Analogously, we have
Next, by (36), (37) and (41),
‖ ˜eT [<t>a g(v)] ‖Lp(ℝ, H1q(Ω)) ≤N∑j,k=1‖<t>N2q2(J(T)Akj(·)+J(·)akj (·))‖L∞((0,T),L∞(Ω))‖<t>b− N2q2v‖Lp((0,T)H2q2(Ω))≤C[v]2T (45)
for any a∈[0, b] and q ∈ [1, q2] . Analogously, we have
‖ ˜eT (<t>b h(v) ‖Lp((0,T)H1q(Ω))≤C[v]2T (46)
for any a∈[0, b] and q ∈ [1, q2] . Since
∂tgk(v)=N∑j=1‖(J(T)∂tAkj(t)+(∂tJ(t)akj (t)+J(t)∂takj (t))vj+N∑j=1‖(J(T)Akj(t)+J(t)akj (t))vj
and since ‖J(T)akj(t)J(t)‖L∞(Ω)≤C as follows from (36), by (37) we have
‖ ˜eT [<t>a ∂tgk(v)] ‖Lp((0,T),Lq (Ω))≤N∑j=1(<t>N2q2∂t(Akj,J,akj)‖L∞((0,T),L∞(Ω))‖<t>b− N2q2v‖Lp((0,T)Lq2(Ω) +(<t>N2q2(Akj,J)‖L∞((0,T),L∞(Ω))‖<t>b− N2q2∂tv‖Lp((0,T)Lq2(Ω) ,
which, combined with (36), furnishes that
‖ ˜eT [<t>a g(v)] ‖Lp(ℝ,Lq(Ω)) ≤C(I+[v]2T) (47)
for any a∈[0, b] and q ∈ [1, q2] .
Before proving Theorem 1, we state a unique existence theorem of locally in time solutions to Eq. (5), which can be proved by a standard argumentation based on maximal Lp-Lq regularity theorem for the Stokes equations with free boundary condition.29,40
Theorem 7
Let N<q2<∞ and 2<p<∞ . Assume that 2/p + N/q2 < 1. Then, given T > 0, there exists an ∈0> 0 depending on T such that if initial data u0∈B2(1−1/p)q2,p(Ω) satisfies the condition:
‖u0‖B2(1−1/p)q2,p(Ω)≤∈0 (48)
and the compatibility condition (9), then Eq. (5) admits unique solutions v and q with
v∈H1p((0,T),Lq2(Ω)N)∩Lp((0,T),H2q2(Ω)N), q∈Lp((0,T),H1q2 (Ω)+ˆH1q2,0)
Let T be a positive number > 2 and let v and q be solutions of Eq. (5) satisfying (13) and (3). In
view of Theorem 7, such solutions v and q exist uniquely provided that
‖u0‖B2(1−1/p)q2,p(Ω)≤∈0 (49)
Thus, we assume that 0< ∈ ≤ ∈0 in Theorem 1. Let [v]T and I be the quantities defined in Theorem 1 in Sect. 1 Introduction. And then, if we prove that there exists a constant M > 0 independent of ϵ and T such that
[v]T≤M(I+ [v]2T) (50)
then we can prolong v and q beyond T. Namely, there exist v1 and q1 with
v1∈H1p ((T,T+1),Lq2(Ω)N)∩Lp((T,T+1)H2q2 (Ω)N), q1∈Lp((T,T+1),H1q2 (Ω)ˆH1q2,0(Ω))
such that v1 and q1 are solutions to the equations:
{∂tv1−Div(µD(v1)−qI)=f(v1), in Ω×(T,T+1), div v1=g(v1) =div g(v1) in Ω×(T,T+1), (µD(v1)−qI)n=h(v1) on Γ×(0,T), v1|t=T =v(·,T) in Ω. (51)
Here, f (v1) is consisting of some linear combinations of nonlinear terms of the forms
V0(∫T0(∇(κv)ds+∫tT(∇(κv1)ds))∇2v1, V1(∫T0(∇(κv)ds+∫tT(∇(κv1)ds)∇2v1))∂tv1
V1(∫T0(∇(κv)ds+∫tT(∇(κv1)ds)) (∫T0(∇(κv)ds+∫tT(∇(κv1)ds)∇2v1))∇v1,(1−κ)(A⊤)−1{v1 (A⊤∇v1)};
and g(v1) , g(v1) and h(v1) have the following forms:
g(v1)=V3(∫T0(∇(κv)ds+∫tT(∇(κv1)ds))∇v1 ; g(v1)=V4(∫T0(∇(κv)ds+∫tT(∇(κv1)ds))v1 ;h(v1)=V5(∫T0(∇(κv)ds+∫tT(∇(κv1)ds))∇v1 ,
where Vi are the same nonlinear functions as in (6) and (7).
In fact, the inequality (50) yields that there exists a small constant ε∈(0,ε0] such that if I≤ε , then
[v]T ≤(2M)−1 −√(2M)−2 −I=2MI+O(I2).
Thus, we may assume that
[v]T ≤3MI (52)
By (29) and (52) we have
‖v(·, T )∥‖B2(1−1/p)q2,p(Ω)≤M1ε
with some positive constant M1 independent of T . Thus, noting that
∫T0∥∇v(·,t)∥L∞(Ω) dt≤C (∫T0< t>−bp′ds)1/p′[v]T ≤M2ϵ
with some positive constant M2 independent of T, choosing ϵ >0 smaller if necessary, we can show the existence of v1 and q1. Thus, setting
v2(,t)={v(,t) 0< t <T,v1(·,t) T < t < T+1, q2={q(,t) 0< t<T,q2(·,t) T< t< T+1,
we see that v2 and q2 satisfy the regularity condition:
v2 ∈H1p ((0,T+1), Lq2(Ω)N) ∩ Lp((0,T+1),H2q2 (Ω)), q2 ∈Lp((0,T+1),H1q2 (Ω)+ˆH1q2,0 (Ω))
and Eq. (5) replacing T by T + 1. Repeating this argument, we can prolong v to time interval (0,∞) .
This completes the proof of Theorem 1. Therefore, we prove (50).
A Proof of Theorem 1
Let v and q be the same N-vector of functions and the function as in Sect. 4. We prove that v satisfies (50). And, we recall that T > 2. As was seen in Sect. 2, v and q satisfy Eq. (26). To estimate v, we write v by, v = w + u where w is a solution to the equations:
{∂tw+λ0w−J(T−1)Div˜ s(w,r)=f(v) in Ω × (0,T), div w= ˜eT [g(v)]=div ˜eT[g(v)] in Ω × (0,T), ˜ s(w,r)= ˜eT[h] on Γ × (0,T), w|t=0 =u0 in Ω (53)
with some pressure term r, and u is a solution to the equations:
{∂tu−J(T−1)Div ˜S(u,p)=−λ0w in Ω × (0,T), div u = 0 in Ω × (0,T), ˜S(u,p) = 0 on Γ × (0,T), u|t=0 =0 in Ω (54)
To estimate w, we quote the maximal Lp-Lq regularity theorem due to Shibata.33 Let us consider the equations:
{∂tw+λ0w−J(T−1) Div ˜S(w,r)=f in Ω × (0,T), ˜div w=g=div g in Ω × (0,T), ˜S(w,r)=h on Γ × (0,T), w|t=0 =w0 in Ω (55)
And then, we have
Theorem 8
Let Ω be an exterior domain in ℝN whose boundary Γ is a C2 hyper surface. Let 1< p, q<∞ and T > 0. Assume that
w0∈B2(1−1/p)q,p(Ω), f∈Lp((0,T),Lq(ΩN)), g∈H12p(ℝ, Lq(Ω)) ∩Lp(ℝ, H1q (Ω)),g∈H1p(ℝ, Lq(ΩN)), h∈H12p(ℝ, Lq(ΩN))∩Lp(ℝ, H1q (ΩN))
and that w0 satisfies the compatibility condition:
w0−g|t=0∈B2(1−1/p)q,p(Ω)
and in addition
(µD(w0)−h|t=0)τ =0 on Γ
If 2/p+ 1/q< 1 , where dτ = d− <d, n>n . Then, there exists a positive number λ0 such that Eq. (55) admits unique solutions w and r with
w∈Lp((0,T),H2q(Ω)N)∩H1p ((0,T),Lq (Ω)N), r∈Lp((0,T),H1q(Ω)+ˆH2q,0(Ω))
Possessing the estimate:
‖w‖Lp((0,T ),H2q(Ω)) +‖∂tw‖ Lp((0,T ),Lq(Ω)) ≤ Cq{‖w0‖B2(1−1/p)q,p(Ω)+∥f ∥Lp((0,T ),Lq(Ω)) +‖(g,h)‖H12p(ℝ, Lq(Ω))‖(g, h)‖Lp(ℝ,Lq (Ω)) + ‖∂tg‖Lp(ℝ,Lq (Ω))},
where Cq is a constant that depends on q but is independent of T. Applying Theorem 8 yields that there exists a large λ0>0 such that Eq. (53) admits unique solutions w and r with
w ∈ H1p((0,T),Lq(Ω)N) ∩ Lp ((0,T),H2q(Ω)N) (q ∈ q12,q2)
Possessing the estimate:
(‖<t>b∂tw‖Lp((0,T)Lq1/2(Ω) ∩ Lq2(Ω))+‖<t>b∂tw‖Lp((0,T),H2q1/2(Ω)∩H2q2(Ω) ≤ C(I+[v]2T) . (56)
In fact, f(v), ˜eT [g(v)], ˜eT [g(v)] and ˜eT [h(v)] satisfy (27), (43), (44), (45), (46), and (47), so that we know the existence of w possessing the estimate:
∂tw‖Lp((0,T)Lq1/2(Ω) ∩ Lq2(Ω))+‖<t>b∂tw‖Lp((0,T),H2q1/2(Ω) ∩ H2q2(Ω)≤ C(I+[v]2T)
with some constant C depending on q>sub>1
/2 and q2. Let a = min(1, b), and then ˜w:=<t>aw satisfies the equations:{∂t˜w+λ0˜w−J(T−1) Div ˜S(˜w, < t >a r)= < t >a f+λ0 at< t >a−2w in Ω × (0,T), ˜div ˜w= ˜eT[< t >a g(v)] =div˜eT[< t >a g(v)] in Ω × (0,T), ˜S(˜w,< t >a r)=˜eT[< t >ah] on Γ × (0,T), ˜w|t=0 =u0 in Ω (57)
Since
‖t< t >a−2w ‖Lp((0,T),Lq(Ω))≤ ‖w ‖Lp((0,T),Lq(Ω)) ≤ C(I+[v]2T)
as follows from the fact that a − 1 ≤ 0 , we have
(‖<t>a∂tw‖Lp((0,T)Lq1/2(Ω) ∩ Lq2(Ω))+‖<t>aw‖Lp((0,T),H2q1/2(Ω) ∩H2q2(Ω) ≤ C(I+[v]2T) .
Repeating this argument finite times yields (56). In particular, by (56) we have
[w]T≤ C(I+[v]2T) (58)
Next, we consider g. Let {T(t)}t≥0 be a C0 analytic semigroup associated with problem (54). Shibata33 proved the existence of {T(t)}t≥0 satisfying the estimates:
‖T(t)f‖Lp(Ω)≤Ct−N2(1q−1p)‖f‖Lq(Ω), ‖ΔT(t)f‖Lp(Ω)≤Ct−12−N2(1q−1p)‖f‖Lq(Ω) (59)
for any t > 0 and f∈ Lq(Ω)N provided that 1 < q ≤ p ≤ ∞ and q ≤ q2 . To represent u by using {T(t)}t≥0 , we introduce the solenoidal space Jq(Ω) defined by
Jq(Ω)={f∈ Lq(Ω)N|(f,JAT∇φ)Ω= 0 for any φ ∈ˆH1q′,0(Ω)} (60)
Here, A is the matrix defined in (4) and J the function given in (15), and
ˆH1q′,0(Ω)={φ ∈ Lq′,loc (Ω) |∇φLq′ (Ω)N, φ|Γ=0} .
As was proved by Shibata,30 we know that for any f∈ Lq(Ω)N there exists a unique solution ψ∈ˆH1q′,0(Ω) of the variational equation
(A∇ψ, JATφ)Ω = (f,JAT∇φ)Ω= 0 for any φ ∈ˆH1q′,0(Ω) . (61)
which possesses the estimate ∥∇ψ∥ Lq (Ω) ≤Cq∥f∥Lq (Ω) . Here Cq is a constant that is independent of v and T in view of (36). Given f∈ Lq(Ω)N let ψ∈ˆH1q′,0(Ω) be a unique solution of Eq.(61), and let Pqbe an operator acting on f defined Pqf= f−∇ψ . And then, Pqf ∈ Jq(Ω) and
∥Pq f∥Lq(Ω) ≤ Cq ∥f∥Lq(Ω) (62)
with some constant Cq that is independent of v and T. By Proposition 21 in Shibaata,33 we have
u(·, t)=−λ0∫t0T(t−s)(Pw)(·,s)ds . (63)
Combining (59) and (62) yields that
‖∇ju(·,t)‖Lr (Ω) ≤ Cr,˜q1 ∫t−10(t−s− j2 − N2(1˜q1−1r))∥w(·,s)∥ L˜q1 (Ω)ds +Cr,˜q2∫t−10(t−s− j2 − N2(1˜q2−1r))∥w(·,s)∥ L˜q2 (Ω)ds (64)
for j=0, 1, for any t > 1 and for any indices r, ˜q1 and ˜q2 such that 1 < ˜q1, ˜q2 ≤ r ≤ ∞ and ˜q1 , ˜q2 ≤ q2, where ∇0u = u and ∇1u =∇u .
Recall that T > 2. In what follows, we prove that
(∫T2(<t>b‖ u (·,t)‖H1∞(Ω))pdt)1p≤ C(I+[v]2T) (65)
Sup2≤t≤T(<t>N2q1‖u(·,t)‖Lq1≤ C(I+[v]2T) (66)
(∫T2(<t>b−N2q1‖u(·,t)‖H1q1(Ω))pdt)1p≤ C(I+[v]2T) (67)
(∫T2(<t>b−N2q2‖u(·,t)‖H1q2(Ω))pdt)1p≤ C(I+[v]2T) (68)
By (64) with r = ∞ , ˜q1 = q1/2 and ˜q2 = q2 ,
‖ u (·,t)‖H1∞(Ω) ≤ C ∫t0‖T(t−s)w(·,s)‖H1∞(Ω)ds = C(I∞(t)+II∞(t)+III∞(t))
With
I∞(t) = ∫t/20(t−s− Nq1)‖w(·,s)‖Lq1/2 (Ω)ds,II∞(t)=∫t−1t/2(t−s− Nq1)‖w(·,s)‖Lq1/2 (Ω)ds,III∞(t)=∫tt−1(t−s− N2q2− 12)‖w(·,s)‖Lq2/2 (Ω)ds.
Since
I∞(t) ≤ (t/2− Nq1) (∫t/20< s >−bp′ds)1/p′(∫t/20< s >b‖w(·,s)‖Lq1/2 (Ω)pds)1/p ≤ C(bp′−1−1/p′)(I+[v]2T)t−Nq1
as follows from the condition: bp′> 1 in (11), by the condition: (Nq1− b)p > 1 in (11), we have
∫T2(< t >bI∞(t))pdt ≤ C ∫T2< t >−(Nq1− b)pdt(I+[v]2T)p≤ C((Nq1− b)p−1)−1(I+[v]2T)p .
By Holder’s inequality
< t >bII∞(t) ≤ C ∫t−1t/2(t−s)−Nq1 < s>b‖w(·,s)‖Lq1/2 (Ω)ds ≤ C (∫t−1t/2(t−s)−Nq1ds1/p′) (∫t−1t/2(t−s−Nq1) < s>b‖w(·,s)‖Lq1/2 (Ω)ds1/p) ≤ C(Nq1− 1)−1/p′(∫t−1t/2(t−s)−Nq1 < s>b‖w(·,s)‖Lq1/2 (Ω)ds1/p)
Because N/q1 = N/q2 + 1 > 1 . By the change of integration order and (56),
∫T2(< t >bII∞(t))p ≤ C(Nq1− 1)− pp′ ∫T2dt ∫t−1t/2(t−s−Nq1) (< s >b‖w(·,s)‖Lq1/2 (Ω)p)ds ≤ C(Nq1− 1)− pp′∫T−11(<s>b‖w(·,s)‖Lq1/2 (Ω))pds∫2ss+1(t−s)−Nq1dt ≤ C(Nq1− 1)−p(I+[v]2T)p.
Since N2q2+12 < 1 as follows from q2 > N, by Holder’s inequality,
< t >bIII∞(t) ≤ C ∫tt−1(t−s)− N2q2 − 12 < s>b‖w(·,s)‖Lq2 (Ω)ds ≤ C ( ∫tt−1(t−s)− N2q2 − 12 ds1/p′) (∫tt−1(t−s)− N2q2 − 12 (< s>b‖w(·,s)‖Lq2 (Ω)p)ds1/p) ≤ C(−N2q2 − 12 )−1/p′(∫tt−1(t−s)− N2q2 − 12 (< s>b‖w(·,s)‖Lq2 (Ω)p)ds1/p).
By the change of integration order, we have
∫T2(< t >bIII∞(t)p )dt≤ C(1−N2q2)− pp′∫T2dt ∫tt−1(t−s−N2q2−12) (< s >b‖w(·,s)‖Lq2 (Ω)p)ds ≤ C(1−N2q2)− pp′∫T1 (< s >b‖w(·,s)‖Lq2 (Ω)p)ds ∫s+1s (t−s−N2q2−12) dt ≤ C(1−N2q2)− p(I+[v]2T)p.
Summing up, we have obtained (65). Next, we prove (66).
By (64) with r=q1, ˜q1=q1/2 and ˜q2=q1 ,
‖ u (·,t)‖Lq1 (Ω)≤ C(Iq1,∞(t)+IIq1,1(t)+IIIq1,1(t))
With
Iq1,1(t) = ∫t/20(t−s)−N2q1‖w(·,s)‖Lq1/2 (Ω)ds,IIq1,1(t) = ∫t−1t/2(t−s)−N2q1‖w(·,s)‖Lq1/2 (Ω)ds,IIIq1,1(t) = ∫tt−1 ‖w(·,s)‖Lq1 (Ω)ds.
By (56)
Iq1,1(t) ≤ (t/2−N2q1) (∫t/20< s >−bp′ds) 1/p′ (∫t/20< s >b ‖w(·,s) ‖Lq1/2 (Ω))pds) 1/p ≤Ct−N2q1(I+[v]2T)p.
Analogously, by Holder’s inequality and (56),
IIq1,1(t) = ∫t−1t/2(t−s−N2q1)<s>−b <s>b ‖w(·,s)‖Lq1/2 (Ω)ds,≤ C < t >−b (∫t−1t/2(t−s−Np′2q1)ds1/p′) (∫T0 (< s >b‖w(·,s)‖Lq1/2 (Ω)p)ds1/p) =C(1−−Np′2q11/p′)< t >−b−N2q1+1p′ (I+[v]2T)≤ C(1−−Np′2q11/p′)< t >N2q1 (I+[v]2T)
because b>1p′ . Finally, by (56),
IIIq1,1(t) ≤C t−b∫tt−1 <s>b ‖w(·,s)‖Lq1/2 (Ω)ds ≤C t−b(∫tt−1ds 1/p′)) (∫T0 < s >b‖w(·,s)‖Lq1/2 (Ω)p)ds)1/p ≤C t−b(I+[v]2T).
Summing up, we have obtained (66). Next, we prove (67). By (64),
‖ u (·,t)‖H1q1 (Ω)≤ C(Iq1,2(t)+IIq1,2(t)+IIIq1,2(t))
with
Iq1,2(t) = ∫t/20(t−s)−N2q1‖w(·,s)‖Lq1/2 (Ω) ds,IIq1,2(t) = ∫t−1t/2(t−s)−N2q1‖w(·,s)‖Lq1/2 (Ω) ds,IIIq1,2(t) = ∫tt−1 (t−s)−12‖w(·,s)‖Lq1 (Ω) ds.
By (56),
Iq1,2(t) ≤ (t/2−N2q1) (∫t/20< s >−bp′ds) 1/p′ (∫t/20< s >b ‖w(·,s) ‖Lq1/2 (Ω))pds) 1/p ≤Ct−N2q1(I+[v]2T).
so that by the condition: (Nq1−b)p>1 in (11)
( ∫T2(< t >b−N2q1Iq1,2(t))pdt)1/p≤C((Nq1−b)p−1)−1/p(I+[v]2T).
By Holder’s inequality,
< t >b−N2q1IIq1,2(t) ≤ C < t >−N2q1 ∫t−1t/2(t−s)−N2q1 < s>b‖w(·,s)‖Lq1/2 (Ω)ds ≤ C < t >−N2q1(∫t−1t/2(t−s)−Np′2q1ds1/p′) (∫To(< s >b‖w(·,s)‖Lq1/2 (Ω)p)ds1/p) ≤ C(1+t)−(Nq1−1p′)(I+[v]2T) .
Since (Nq1−1p′)p > 1 as follows from Nq1=1+Nq2> 1=1p+1p′ , we have
(∫T2(< t >b−Nq1 IIq1,2(t))pdt)1/p≤ C((Nq1−b)p−1)−1/p(I+[v]2T)
Since q1/2 < q1 < q2 , we have
‖w(·,t)‖Lq1 (Ω) ≤ ‖w(·,t)‖q2N+2q2Lq1/2 (Ω)‖w(·,t)‖N+q2N+2q2Lq2 (Ω) .
Let α=q2N+2q2 and β=N+q2N+2q2 then α+β =1 , so that by (56) and Holder’s inequality
‖< t>bw ‖Lp((0,T),Lq1(Ω)) ≤ (∫T0(< t>b‖w(·,t) ‖Lq1/2 (Ω))pα(< t>b‖w(·,t) ‖Lq2 (Ω))pβdt)1/p≤ (∫T0(< t>b‖w(·,t) ‖Lq1/2 (Ω))pdt)α/p(∫T0(< t>b‖w(·,t) ‖Lq2 (Ω))pdt)β/p≤ C(I+[v]2T) (69)
Since
< t >b− N2q1IIIq1,2(t) ≤ ∫tt−1(t−s) − 12 < s>b− N2q1‖w(·,s)‖Lq1 (Ω)ds ≤ ( ∫tt−1(t−s) − 12 ds1/p′) (∫tt−1(t−s− 12) ds (< s>b‖w(·,s)‖Lq1 (Ω)p)ds1/p),
by the change of integration order, we have
∫T2(< t >b− N2q1IIIq1,2(t)p)dt ≤ 2pp′∫T2dt ∫tt−1(t−s)−12(< s>b‖w(·,s)‖Lq1 (Ω)p)ds≤ 2pp′∫T0(< s>b‖w(·,s)‖Lq1 (Ω)p)ds ∫s+1s(t−s)−12 dt=2p ‖< t>bw ‖Lp((0,T),Lq1(Ω)) ,
which, combined with (69), furnishes that
(∫T2(< t >b− N2q1IIIq1,2(t))pdt)1/p≤ C(I+[v]2T)
Summing up, we have obtained (67).
Finally, we prove (68). By (64) with r = q2, ˜q1 = q1/2 and ˜q2= q2 ,
‖ u (·,t)‖Lq2 (Ω)≤ C(Iq2(t)+IIq2(t)+IIIq2(t))
with
Iq2(t) = ∫t/20(t−s)−N2(2q1+ 1q2) ‖w(·,s)‖Lq1/2 (Ω) ds,IIq2(t) = ∫t−1t/2(t−s)−N2(2q1+ 1q2) ‖w(·,s)‖Lq1/2 (Ω) ds,IIIq2(t) = ∫tt−1 ‖w(·,s)‖Lq2 (Ω) ds.
By Holder’s inequality,
Iq2(t)≤ (t/2−N2(2q1+ 1q2)) (∫t/20< s>bp′ds)1/p′(∫t/20< s>b‖w(·,s)‖Lq1/2 (Ω))pds)1/p ≤ C < t >−N2(2q1+ 1q2) (I+[v]2T) .
for t ≥ 2 . Since N2(2q1+ 1q2)−(b− N2q2)= Nq1−b
by the condition: (Nq1−b)p>1 in (11),
(∫T2(< t >b− N2q2Iq2(t))pdt)1/p≤ C (∫T2t−(Nq1−b)pdt)1/p(I+[v]2T) ≤ C((Nq1−b)p−1)−1/p(I+[v]2T) .
Since
N2(2q1−1q2)=N2(1q2+2N)=N2q2+1 > 1,
by Holder’s inequality
so that by the change of integration order and (56)
Analogously, by Holder’s inequality
so that by the change of integration order and (56)
Summing up, we have obtained (68).
Recalling that , applying the maximal Lp-Lq regularity theorem due to Shibata33 to Eq. (54) and using (56) give that
(70)
For any . Employing the same argumentation as that in proving (29), by real interpolation,we have
(71)
for any . Combining (65), (66), (67), (68), (70), (71) and the Sobolev imbedding theorem, we have
(72)
From (54), u satisfies the equations:
so that by Theorem 8,
which, combined with (72), furnishes that
(73)
Since v = w+u, by (58) and (73), we see that v satisfies the inequality (50), which completes the proof of Theorem 1.53-56
None.
Author declares that there is no conflict of interest.
©2017 Shibata. This is an open access article distributed under the terms of the, which permits unrestricted use, distribution, and build upon your work non-commercially.