3.2 Solutions to the Linear System
3.2.3 Approach by Galerkin-Approximation
d+1
X
j=1
|||fd+2−j|||md+2−j,p,Ω+
d+1
X
j=1
|||gd+2−j|||2−r
d+2−j−p1,p,∂Ω
(3.2.59)
holds where the constant c does not depend upon f, g, η and we use the norms depending on a parameter η ∈ C/{0}, namely let v ∈ Wps(Ω), g ∈ Ws−
1
p p(∂Ω) where s is an integer satisfying 1≤s≤m,
|||v|||s,p,Ω :=kvkWs
p(Ω)+|η|mskvkLp(Ω), (3.2.60)
|||g|||s−1
p,p,∂Ω :=kgk
Ws−
1p p (∂Ω)
+|η|
s−1 p
m kgkLp(∂Ω). (3.2.61) Proof. Since the boundary problem (3.2.32) is elliptic with parameter in L from Theorem 3.2.5, the existence and the a priori estimate follow directly from Theorem2.2.1.
3.2.3 Approach by Galerkin-Approximation
We use Galerkin’s method to solve the linear system (3.2.1). The idea of this method is that one first constructs a family of approximate problems whose so-lutions satisfy certain a priori estimates. This yields a sequence of approximate solutions that have uniform bounds. Uniform bounds imply the existence of a weakly convergent subsequence. One then shows that the weak limit is the solution we seek.
Preliminaries
We introduce some basic materials on functional analysis and Sobolev spaces which will be needed in the later procedure.
Theorem 3.2.7. (Rellich selection theorem)
Let Ω⊆Rn be a bounded open subset. Every bounded sequence in H01(Ω)has a subsequence, which converges in the norm ofL2(Ω).
Theorem 3.2.8. (Eigenvectors of a compact, symmetric operator)
Let H be a separable Hilbert space, and suppose S : H → H is a compact and symmetric operator. Then there exists a countable orthonormal basis ofH which consists of eigenvectors ofS.
We define a bilinear form inH01(Ω)∩H2(Ω) B : H01(Ω)∩H2(Ω)
× H01(Ω)∩H2(Ω)
→R with
B(u, v) = (u, v)L2+ (∇u,∇v)L2 + (4u,4v)L2. (3.2.62) It is easy to check that B(u, v) is an inner product in H01(Ω)∩H2(Ω). The norm defined by this inner product inH01(Ω)∩H2(Ω) is equivalent to the norm defined by
(u, v)H2(Ω)= X
|α|≤2
(Dαu, Dαv) from the following fact:
Theorem 3.2.9. Suppose that f ∈ H−1(Ω), Ω is C1, u ∈ H01(Ω) is a weak solution of the elliptic boundary-value problem
(4u=f, in Ω u= 0, on∂Ω
Assume finallyf ∈Hm(Ω), ∂Ω isCm+2 (m= 0,1,2,· · ·). Then u∈Hm+2(Ω)
and we have the estimate
kukHm+2(Ω)≤C4,m,ΩkfkHm(Ω), the constantC4,m,Ω depends only on m, Ω.
Proof. The proof can be found in [33] pp. 323.
Additionally we record the following embedding from [33] pp. 288
Theorem 3.2.10. Let Ω be open, bounded in Rd, T > 0, m be a nonnegative interger. Assume the boundary∂Ωis smooth.
Suppose u∈L2(0, T;Hm+2(Ω)), with ∂tu∈L2(0, T;Hm(Ω)), then u∈C([0, T];Hm+1(Ω)),
and the estimates
kukC([0,T];Hm+1(Ω))≤C kukL2(0,T;Hm+2(Ω))+k∂tukL2(0,T;Hm(Ω))
holds, where C >0 depends only on T, Ω, and m.
Then we obtain the following corollary:
Theorem 3.2.11. Let u ∈ L2(0, T;H01(Ω)∩H2(Ω)), ∂tu ∈ L2(0, T;L2(Ω)), then the function
t7→ k∇uk2 is absolutely continuous and
d
dtk∇uk2 =−2(∂tu,4u)L2(Ω) (3.2.63) for a.e 0≤t≤T.
Proof. FromTheorem 3.2.10 it follows direct thatu∈C([0, T];H1(Ω)).
Fix any point t in (0, T) there exists σ > 0 such that t ∈ (ε, T −ε) for all ε≤σ. Setuε =ηε∗u,ηεdenoting the usual mollifier onR1, i.e.,ηε=ε−1η(t/ε) where
η(t) =
(ke−1/(1−|t|2) if|t|<1, 0 if|t| ≥1, k >0 is selected so that
Z
R
η(t)dt= 1.We find first that
ηε(t) = 0 if |t| ≥ε, Z ε
−ε
ηε(t)dt= 1.
Furthermore uε(t) =
Z T 0
ηε(t−y)u(y)dy= Z ε
−ε
ηε(y)u(t−y)dy, t∈(ε, T −ε) in the sense of Bochner integral with respect to H01(Ω)∩H2(Ω). It is easy to calculate that
dnuε(t) dtn =
Z T 0
dnηε
dtn (t−y)u(y)dy for any n∈N, which implies
uε∈C∞((ε, T −ε);H01(Ω)∩H2(Ω))).
Choose a s ∈ (ε, T −ε), without loss of generality, let s < t. Then uε ∈ C∞([s, t];H1(Ω)). Now forx∈[s, t]
kuε(x)−u(x)kH1(Ω) =
Z ε
−ε
ηε(y)u(x−y)dy− Z ε
−ε
ηε(y)u(x)dy H1(Ω)
=
Z ε
−ε
ηε(y)(u(x−y)−u(x))dy H1(Ω)
≤ sup
y∈(−ε,ε)
ku(x−y)−u(x)kH1(Ω).
Since u is uniformly continuous on [s, t] in H1(Ω), i.e., u ∈ C([s, t];H1(Ω)), then for anyδ > 0 there existsθ >0 independent of x∈[s, t] such that for all x1, x2 ∈[s, t] with|x1−x2|< θ ku(x1)−u(x2)kH1(Ω) < δ. Thus if ε < θ, we obtain
sup
y∈(−ε,ε)
ku(x−y)−u(x)kH1(Ω)< δ, which implies
kuε(x)−u(x)kH1(Ω) < δ
for all x ∈ [s, t]. Then we conclude that uε converges tou in C([s, t];H1(Ω)) asε→0.Furthermore from properties of mollifiers we obtain also
( uε−→u in L2loc(0, T;H01(Ω)∩H2(Ω))), (uε)0 −→u0 in L2loc(0, T;L2(Ω))
asε−→0. Then d
dtk∇uε(t)k2 = 2(∇uε(t),∇(uε(t))0)L2(Ω)=−2((uε(t))0,4uε(t))L2(Ω), consequently for anys∈(ε, T −ε)
k∇uε(s)k2 =k∇uε(t)k2+ Z s
t
−2((uε(τ))0,4uε(τ))L2(Ω)
dτ. (3.2.64) Next we find
k∇uε(s)k2− k∇u(s)k2 = Z
Ω
(∇uε(s)− ∇u(s))(∇uε(s) +∇u(s))dx
≤ k∇uε(s)− ∇u(s)kL2(Ω)k∇uε(s) +∇u(s)kL2(Ω)
≤ k∇uε(s)− ∇u(s)k2L2(Ω)
+ 2k∇u(s)kL2(Ω)k∇uε(s)− ∇u(s)kL2(Ω).
Since uε(s) converges to u(s) in H1(Ω) as ε→0, thenk∇uε(s)k2 → k∇u(s)k2 asε→0. Recalling (3.2.64), it follows
k∇u(s)k2 =k∇u(t)k2+ Z s
t
−2((u(τ))0,4u(τ))L2(Ω)
dτ.
Definition of Weak Solutions
Assume thatu is a classical solution to (3.2.1), then multiply (3.2.1) by (v0, v1,· · · , vd) =: (v0, vd) =:v∈(C0∞(ΩT))1+d
and integrate by parts, then we obtain the following equations
Z T
0
B1(u, v)dt= Z T
0
(F0, v0)dt, Z T
0
B2(u, v)dt= Z T
0
(Fd, vd)dt,
(3.2.65)
with
B1(u, v) := (∂tu0, v0)−ν0(v0,4u0)−(v0,divud), B2(u, v) := (∂tud, vd) +ν0(∇vd,∇ud) + 1
τ(ud, vd)−T0(∇u0, vd)−2
4(4u0,divvd), which give us the motivation for definition of weak solutions to (3.2.1).
Definition 3.2.2. Assume that
F0 ∈L2(0, T;L2(Ω)), Fd ∈L2(0, T; (H−1(Ω))d), g0 ∈H01(Ω), gd ∈L2(Ω)
We say a function
u∈L2(0, T; (H01(Ω)∩H2(Ω))×(H01(Ω))d)∩H1(0, T;L2(Ω)×(H−1(Ω))d) is a weak solution of the system (3.2.1) provided (3.2.65) holds for each v ∈ L2(0, T; (H01(Ω))1+d), andu(0) = (g0, gd)T.
Existence and Uniqueness
Theorem 3.2.12. Assume that the given functions
F0 ∈L2(0, T;L2(Ω)), Fd ∈L2(0, T; (H−1(Ω))d), g0 ∈H01(Ω), gd ∈L2(Ω)
then (3.2.1) has a unique weak solution which satisfies the following a priori estimates
ku0k2L2(0,T;H01(Ω)∩H2(Ω))+kudk2L2(0,T;H01(Ω))+k∂tuk2L2(0,T;L2(Ω)×(H−1(Ω))d)
≤C
kF0k2L2(0,T;L2(Ω))+kFdk2L2(0,T;(H−1(Ω))d)+kg0k2H1
0(Ω))+kgdk2L2(Ω)
, where the constantC does not depend upon F.
Remark 3.2.2. From Definition (3.2.2) and Theorem3.2.10 it followsu∈ C([0, T], H1(Ω)×(L2(Ω))d), which allows us to interpret the initial condition.
Proof. Our idea is the Faedo-Galerkin method. More precisely, we expect a set of functions φj = φj(x) (j = 1,2, ...) in H01(Ω)∩H2(Ω) such that {φj}∞j=1 is an orthonormal basis ofL2(Ω) and an orthogonal basis in H01(Ω) andH01(Ω)∩ H2(Ω) respectively. We take {φj}∞j=1 to be the complete set of normalized eigenfunctions for the operator−4in H01(Ω). SetS the solution operator
S :L2(Ω)→L2(Ω) f 7→u=Sf, uis the weak solution of
−4u=f, u|∂Ω = 0.
It is obvious thatS is symmetric. Furthermore from Rellich selection theorem we check that S is compact, which implies all the eigenvalues of S are real, nonzero and there are corresponding eigenfunctions suitably normalized, which make up an orthonormal basis in L2(Ω). But observe as well that for η 6= 0, we have Sf =ηf if and only if −4f = λf for λ = 1
η. Let {ηj}∞j=1 comprise the sequence of distinct eigenvalues ofS,Hj the eigenspace with respect toηj, we obtain thereby an orthonormal basisφj =φj(x) (j = 1,· · ·) of normalized eigenvectors inHj of L2(Ω). Actuallyφj ∈H01(Ω)∩C∞(Ω),4φj ∈H01(Ω) and we have the following calculations
(∇φi,∇φj)L2 =ηi(φi, φj)L2 = 0,
−4φi =ηiφi =⇒ 42φi =ηi(−4φi) =ηi2φi,
=⇒(4φi,4φj)L2 =η2i(φi, φj)L2 = 0.
Therefore{φj}∞j=1is orthogonal inH01(Ω)∩H2(Ω) with respect to (3.2.62). We claim further that{φj}∞j=1 is complete inH01(Ω)∩H2(Ω). To see this, it suffices to verify that for u ∈ H01(Ω)∩H2(Ω), B(φj, u) = 0 implies u ≡0. From the
properties of φj, B(φj, u) = (1 +ηj +ηj2)(φj, u)L2. The completeness follows immediately since {φj}∞j=1 is complete in L2(Ω). That {φj}∞j=1 is complete in H01(Ω) follows from the same reason. Via normalization of {φj}∞j=1 in H01(Ω)
ωj := φj
kφjkH1 0(Ω)
!
we obtain an orthonormal basis {ωj}∞j=1 inH01(Ω).
Fix now a positive integer m. We will look for 1 +d functions u0m, udm :=
(u1m,· · · , udm)T with
u0m : [0, T]→H01(Ω)∩H2(Ω), (3.2.66) udm : [0, T]→(H01(Ω))d, (3.2.67) of the form
u0m(t) :=
m
X
j=1
dju0
m(t)ωj, (3.2.68)
udm(t) :=
m
X
j=1
dju1 m(t)φj,
m
X
j=1
dju2
m(t)φj,· · ·,
m
X
j=1
djud m(t)φj
T
, (3.2.69)
where we hope to select the coefficients dju0
m(t), djul
m(t) (0 ≤ t ≤ T, l = 1, ..., d, j = 1, ..., m) so that
dju0
m(0) = (g0, ωj)H1
0(Ω), djul
m(0) = (gl, φj) (3.2.70) and
(∂tu0m, ωj) +ν0(∇ωj,∇u0m) + (∇ωj, udm) = (F0, ωj), (∂tulm, φj) +ν0(∇φj,∇ulm) +1
τ(ulm, φj)
−T0(∂xlu0m, φj) +2
4(∂xl4u0m, φj) = (Fl, φj).
(3.2.71)
Thus we seek functionsu0m,udm of the form (3.2.68) and (3.2.69) respectively, that satisfy the ”projection” (3.2.71) of problem (3.2.1) onto the finite dimen-sional subspaces spanned by {φj}mj=1. Assuming u0m, udm have the structures (3.2.68), (3.2.69) respectively, we first note that
(∂tu0m(t), ωj) = 1 kφjk2
H01(Ω)
(dju0
m(t))0, (3.2.72) (∂tulm(t), φj) = (djul
m(t))0, (3.2.73)
since {φj}∞j=1 forms an orthonormal basis in L2(Ω). Then (3.2.71) becomes a linear system of ODE subject to the initial conditions (3.2.70), which may be
written in the form
d0(t) =Ad(t) +b(t), d(0) =a,
wherea is via (3.2.70) solvable. Forb(t)∈L2(0, T) this equation has a unique solution in{v∈H1(0, T)|v(0) = 0}, more precisely,d(t) has the form
d(t) = Z t
0
eA(t−s)b(s)ds+eAta.
Multiplying (3.2.71) by dju0
m(t), djul
m(t) respectively, sum forj = 1,· · ·, m, l= 1,· · · , d, and then recall (3.2.68) and (3.2.69), we obtain
(∂tu0m, u0m) +ν0(∇u0m,∇u0m) + (∇u0m, udm) = (F0, u0m), (∂tudm, udm) +ν0(∇udm,∇udm) +1
τ(udm, udm)
−T0(∇u0m, udm) +2
4(∇4u0m, udm) = (Fd, udm).
(3.2.74)
Next we will get the energy estimates of (u0m, udm)T. Since−4ωj =ηjωj, then
−4u0m =
m
X
j=1
dju0
m(−4ωj) =
m
X
j=1
dju0 mηjωj. Multiplying the first equation of (3.2.71) bydju0
mηj, summing up forj= 1,· · · , m yields
(∂t∇u0m,∇u0m) +ν0(4u0m,4u0m)−(∇4u0m, udm) =−(F0,4u0m). (3.2.75) Combining (3.2.74) and (3.2.75) we obtain the following equation
1
2∂t T0ku0mk2+2
4k∇u0mk2+kudmk2
+ν0T0k∇u0mk2 +2
4ν0k4u0mk2+ν0k∇udmk2+ 1 τkudmk2
=T0(F0, u0m) + (Fd, udm)−2
4(F0,4u0m).
(3.2.76)
Integrating (3.2.76) from 0 toT yields T0
2 ku0m(T)k2+1
2kudm(T)k2+2
8k∇u0m(T)k2 +ν0T0k∇u0mk2L2(0,T;L2(Ω)) +2
4ν0k4u0mk2L2(0,T;L2(Ω))
+ν0k∇udmk2L2(0,T;L2(Ω))+ 1
τkudmk2L2(0,T;L2(Ω))
= Z T
0
T0(F0, u0m) + (Fd, udm)−2
4(F0,4u0m)
dt +T0
2 ku0m(0)k2+1
2kudm(0)k2+2
8k∇u0m(0)k2, which implies
T0
2 ku0m(T)k2+1
2kudm(T)k2+2
8k∇u0m(T)k2 +ν0T0k∇u0mk2L2(0,T;L2(Ω)) +2
4ν0k4u0mk2L2(0,T;L2(Ω))
+ν0k∇udmk2L2(0,T;L2(Ω))+ 1
τkudmk2L2(0,T;L2(Ω))
≤c1 Z T
0
kF0kH−1(Ω)ku0mkH1(Ω)+c2
Z T 0
kFdkH−1(Ω)kudmkH1(Ω)
+c3
Z T 0
kF0kL2(Ω)k4u0mkL2(Ω)
+ max T0
2 ,1 2,2
8
(kg0k2H1
0(Ω)+kgdk2L2(Ω))
(3.2.77)
sinceku0m(0)k2
H01(Ω) ≤ kg0k2
H01(Ω),kudm(0)k2L2(Ω)≤ kgdk2L2(Ω) by (3.2.70).
Applying Poincare’s inequality
ku0mk2L2(Ω) ≤c4k∇u0mk2L2(Ω), (c4 is a constant independent ofm) and Cauchy’s inequality we obtain
ku0mk2L2(0,T;H2(Ω))+kudmk2L2(0,T;H1(Ω))
≤c5
kF0k2L2(0,T;L2(Ω))+kFdk2L2(0,T;(H−1(Ω))d)+kg0k2H1
0(Ω)+kgdk2L2(Ω)
. (3.2.78) Fix anyh ∈L2(Ω), with khkL2(Ω) ≤1, and write h =h1+h2, where h1 ∈ span{ωj}mj=1 and (h2, ωj) = 0 (j = 1,· · ·, m). Since the functions {ωj}∞j=1 are
orthogonal in L2(Ω), kh1kL2(Ω) ≤ khkL2(Ω) ≤1. Utilizing (3.2.71), we deduce for a.e. 0≤t≤T that
(∂tu0m, h1)−ν0(h1,4u0m)−(h1,divudm) = (F0, h1) (3.2.79) then we obtain
|(∂tu0m, h)|=|(∂tu0m, h1)|=|(F0, h1) +ν0(h1,4u0m) + (h1,divudm)|
≤c7(kF0kL2(Ω)+k4u0mkL2(Ω)+kdivudmkL2(Ω)), which yields
k∂tu0mkL2(Ω)≤c7(kF0kL2(Ω)+ku0mkH2(Ω)+kudmkH1(Ω)), and
Z T 0
k∂tu0mk2L2(Ω)dt≤c8 Z T
0
(kF0k2L2(Ω)+ku0mk2H2(Ω)+kudmk2H1(Ω))dt, since (3.2.78) holds, we obtain the uniform bounds of∂tu0m inL2 0, T;L2(Ω)
k∂tu0mk2L2(0,T;L2(Ω))
≤c9 kF0k2L2(0,T;L2(Ω))+kFdk2L2(0,T;(H−1(Ω))d)+kg0k2H1
0(Ω)+kgdk2L2(Ω)
. (3.2.80) Next fix anyh∗∈H01(Ω), with kh∗kH1
0(Ω) ≤1, and write h=h1∗+h2∗, where h1∗ ∈span{ωj}mj=1 and (h2∗, ωj) = 0 (j= 1,· · ·, m). Since the functions {ωj}∞j=1 are orthogonal inH01(Ω), kh1∗kH1
0(Ω)≤ kh∗kH1
0(Ω) ≤1. By an analogous reason we find
k∂tudmk2L2(0,T;(H−1(Ω))d)
≤c10 kF0k2L2(0,T;L2(Ω))+kFdk2L2(0,T;(H−1(Ω))d)+kg0k2H1
0(Ω)+kgdk2L2(Ω)
. (3.2.81) Then (3.2.78), (3.2.80), and (3.2.81) are sufficient to ensure the existence of a subsequence {u0m
s}∞s=1 ⊂ {u0m}∞m=1 and a subsequence {udm
s}∞s=1 ⊂ {udm}∞m=1 and functions
u0 ∈L2(0, T;H01(Ω)∩H2(Ω)), (3.2.82) v0 ∈L2 0, T;L2(Ω)
, (3.2.83)
ud ∈L2(0, T; (H01(Ω))d), (3.2.84) vd ∈L2(0, T; (H−1(Ω))d), (3.2.85)
such that
By (3.2.86) we obtain Z T Fix an integerN and choose functions
f ∈C1([0, T];H01(Ω))
where {djf(t)}Nj=1, {djhs(t)}Nj=1 (s = 1,2,· · ·, d) are given smooth functions.
We claim that functions of the form (3.2.87), (3.2.88) are dense in L2 0, T;H01(Ω)
and L2(0, T; (H01(Ω))d)
respectively. To prove this let g ∈ L2([0, T], H01(Ω)), ϑ > 0. The assertion follows if we can show that there is a functionp of the form (3.2.87) with
kp−gkL2(0,T;H01(Ω))< ϑ.
Let gn:=
The sequence of functions
converges pointwise to kg(t)k2H1 0(Ω)
sincegm is an interior point of{f ∈L2(0, T;H01(Ω))|kf−gkL2(0,T;H01(Ω)) < ϑ}.
Using the dense embedding C0∞((0, T)) ,→ L2((0, T)) then for each aj (j = 1,· · · , m) we can select a sequence (aj,n)n⊂C0∞((0, T)) such that aj,n→aj in L2((0, T)) forn→ ∞.
fn:=
m
X
j=1
aj,n(t)wj −→gm
inL2(0, T;H01(Ω)) since
kfn−gmkL2(0,T;H01(Ω)) =
m
X
j=1
(aj,n−aj)wj
L2(0,T;H01(Ω))
≤
m
X
j=1
k(aj,n−aj)wjkL2(0,T;H1 0(Ω))=
m
X
j=1
kaj,n−ajkL2((0,T)). Thus there existsM ∈Nsuch that for alln≥M fn belongs to
{f ∈L2(0, T;H01(Ω))|kf −gmkL2(0,T;H01(Ω))< r}.
Consequently from (3.2.91)kfn−gkL2(0,T;H01(Ω))< ϑ.
The system (3.2.89) holds for all f ∈L2 0, T;H01(Ω)
and hd∈L2(0, T; (H01(Ω))d), since functions of the form (3.2.87), (3.2.88) are dense in
L2 0, T;H01(Ω)
and L2(0, T; (H01(Ω))d) respectively.
In order to proveu0(0) =ud(0) =g, we first note from (3.2.86) that
u0ms *∗ u0 weakly star in L2(0, T;H−1(Ω)),
∂tu0ms *∗ ∂tu0 weakly star in L2 0, T;H−1(Ω) , udms *∗ ud weakly star in L2(0, T; (H−1(Ω))d),
∂tudms *∗ ∂tud weakly star in L2(0, T; (H−1(Ω))d),
(3.2.92)
then from TheoremB.0.7 we obtain
(u0ms, udms)(0)*∗(u0, ud)(0) weakly star in H−1(Ω)×(H−1(Ω))d. By the uniqueness of the limit, we get
u(0) =g.
Let (u01, ud1) and (u02, ud2) be two weak solutions of (3.2.1), define u04 =u01−u02, ud4=ud1 −ud2
then we get the system
∂tu04−ν04u04−divud4 = 0, (3.2.93)
∂tud4−ν04ud4+2
4∇4u04−T0∇u04+ 1
τud4 = 0 (3.2.94) with
(u04, ud4)(0) =0, (u04, ud4)|∂Ω=0.
To get the uniqueness we first note− Z
Ω
∂tu044u04dx= 1
2∂tk∇u04k2by theorem 3.2.11.
Taking the L2(Ω) scalar product of (3.2.93) with −2
44u04 then for a.e. 0 ≤ t≤T
1 2
2
4∂tk∇u04k2+2
4ν0k4u04k2−2
4(∇4u04, ud4) = 0. (3.2.95) Taking the L2(Ω) scalar product of (3.2.93) with T0u04; (3.2.94) with ud4 we obtain for a.e. 0≤t≤T
1
2T0∂tku04k2+ν0T0k∇u04k2−T0(divud4, u04) = 0, (3.2.96) and
1
2∂tkud4k2+1
τkud4k+ν0k∇ud4k2+T0(divud4, u04) +2
4(∇4u04, ud4) = 0.
(3.2.97)
Summing (3.2.95)-(3.2.97) yields 1
2∂t
T0ku04k2+2
4k∇u04k2+kud4k2
+ν0T0k∇u04k2+2
4ν0k4u04k2 +ν0k∇ud4k2+ 1
τkud4k2= 0,
(3.2.98) thenu04=ud4 = 0.
Regularity of the Weak Solution
Theorem 3.2.13. Concerning the linear problem
∂tu(t, x) +A(∂x)u(t, x) =F(t, x), (t, x)∈[0, T)×Ω u(0, x) = 0,
u|∂Ω= 0 for a.e. t∈[0, T].
(3.2.99)
Let T >0 and
F0 ∈L∞(0, T;H1(Ω)), Fd∈L∞(0, T; (L2(Ω))d);
F˙0 ∈L∞(0, T;L2(Ω)),F˙d ∈L∞(0, T; (H−1(Ω))d);
F0(0, x)∈H01(Ω), Fd(0, x)∈L2(Ω),
(3.2.100)
then the weak solutionu= (u0, ud)of (3.2.99)has the following global regularity
u0 ∈L∞(0, T;H3(Ω))∩C([0, T], C1(Ω))∩C([0, T], H2(Ω)), ud∈L∞(0, T;H2(Ω))∩C([0, T], C(Ω))∩C([0, T], H01(Ω)),
∂tu0 ∈L∞(0, T;H01(Ω))∩C([0, T], L2(Ω))∩L2(0, T;H2(Ω)),
∂tud∈L∞(0, T;L2(Ω))∩C([0, T], L2(Ω))∩L2(0, T;H01(Ω)),
(3.2.101)
and the a priori estimates
ku0k2L∞(0,T;H3(Ω))+kudk2L∞(0,T;H2(Ω))
≤C kF0k2L∞(0,T;H1(Ω))+kFdk2L∞(0,T;L2(Ω))
+kF0(0,·)k2H1(Ω)+kFd(0,·)k2L2(Ω)
!
+C
kF˙0k2L∞(0,T;L2(Ω))+kF˙dk2L∞(0,T;H−1(Ω))
Z T 0
eCsds,
(3.2.102)
and
ku˙0k2L∞(0,T;H1(Ω))+ku˙dk2L∞(0,T;L2(Ω))
≤C
kF0(0,·)k2H1(Ω)+kFd(0,·)k2L2(Ω)
+C
kF˙0k2L∞(0,T;L2(Ω))+kF˙dk2L∞(0,T;H−1(Ω))
Z T 0
eCsds
(3.2.103)
hold, where the constantC depends only onΩand the coefficients ofA(∂x), but doesn’t depend upon F.
Proof. The usual way of proving regularity for solutions of problem (3.2.99) is to first establish temporal regularity and then use elliptic estimates for the operator A(∂x) to show spatial regularity. More precisely, we can formally differentiate equation (3.2.99) with respect to time. This yields
∂tu˙ +A(∂x) ˙u= ˙F ,
˙
u(0, x) =F(0, x),
u|˙ ∂Ω = 0 for a.e. t∈[0, T].
(3.2.104)
We now consider ˙u as a new variableQ= (Q0, Qd), then from theorem 3.2.12 Q∈L2(0, T;V)∩H1(0, T; (H−1(Ω))1+d)∩C([0, T], L2(Ω)) solves
∂tQ+A(∂x)Q= ˙F , Q(0, x) =F(0, x),
Q|∂Ω= 0 for a.e. t∈[0, T].
(3.2.105)
Below we shall prove that actually Q= ˙u. For this set Z(t) :=
Z t 0
Qdτ (3.2.106)
in the sense of the Bochner integral. Then we have Z T
0
kZ(t)k2Vdt= Z T
0
Z t 0
Qdτ
2
V
dt≤ Z T
0
Z T 0
kQk2Vdτ dt <∞
=⇒Z(t)∈L2(0, T;V).
We conclude that ˙Z =Q in the sense of distribution and A(∂x)Z(t) =
Z t 0
A(∂x)Qdτ
=F(t)−Q(t).
With W denotingZ−u, then W ∈L2(0, T;V). We have ˙Z(t) = ˙W(t) + ˙u(t), combining ˙Z(t) =Q(t) we thus obtain
(W˙ +A(∂x)W = 0, W(0) = 0.
(3.2.107) Since (3.2.107) has only one solution and 0 is a solution, we conclude W = 0, which implies that the times derivatives of u satisfies the same boundary
conditions as u, namely ˙u|∂Ω = u|∂Ω = 0 for a.e. 0 ≤ t ≤ T. Using a same calculation as (3.2.76) on (3.2.105):
1
2∂t T0kQ0k2+2
4k∇Q0k2+kQdk2
+ν0T0k∇Q0k2 +2
4ν0k4Q0k2+ν0k∇Qdk2+1 τkQdk2
=T0( ˙F0, Q0) + ( ˙Fd, Qd)−2
4( ˙F0,4Q0).
(3.2.108)
Define the energy
E(t) := 1 2
T0kQ0k2+2
4k∇Q0k2+kQdk2
,
then by H¨older’s inequality and Cauchy’s inequality and the assumption that F˙0 ∈L∞(0, T;L2(Ω)), F˙d ∈L∞(0, T; (H−1(Ω))d)
we obtain the following type of differential inequality:
d
dtE(t) +C1E(t)≤C2, ∀t≥0, whereC2>0 depends only on
kF˙0kL∞(0,T;L2(Ω))+kF˙dkL∞(0,T;(H−1(Ω))d). Solving this differential inequality leads to
E(t)≤E(0)e−C1t+C2
C1, ∀t≥0.
Since
E(0) = 1 2
T0kQ0(0)k2+2
4k∇Q0(0)k2+kQd(0)k2
= 1 2
T0kF0(0,·)k2+ 2
4k∇F0(0,·)k2+kFd(0,·)k2
, we conclude
k∂tu0kL∞(0,T;H01(Ω))+k∂tudkL∞(0,T;L2(Ω))≤C, C >0 depends only on F(0, x) and ˙F.
From theorem 3.2.5 and 2.2.1 we have the estimate
ku0kL∞(0,T;H3(Ω))+|η|3/2ku0kL∞(0,T;L2(Ω))+kudkL∞(0,T;H2(Ω))
+|η|kudkL∞(0,T;L2(Ω))
≤c kF0kL∞(0,T;H1(Ω))+|η|1/2kF0kL∞(0,T;L2(Ω))+kFdkL∞(0,T;L2(Ω))
+k∂tu0kL∞(0,T;H01(Ω))+|η|1/2k∂tu0kL∞(0,T;L2(Ω))+k∂tudkL∞(0,T;L2(Ω))
≤C,
C >0 depends only onF,F(0, x) and ˙F. For 0≤t0 ≤t00≤T, the H¨older estimates
ku0(t0,·)−u0(t00,·)kH2(Ω)≤ Z t00
t0
k∂tu0kH2(Ω)dt
≤ |t0−t00|1/2k∂tu0kL2(0,T;H2(Ω))
holds. Fix a number β with 0< β < 1
2 then by Sobolev’s embedding theorem kud(t0)−ud(t00)kC(Ω)≤Ckud(t0)−ud(t00)kH2−β(Ω), (3.2.109) here letC denote the embedding constant. Interpolation yields
kud(t0)−ud(t00)kH2−β(Ω)
≤Ckud(t0)−ud(t00)kβ/2L2(Ω)kud(t0)−ud(t00)k(2−β)/2H2(Ω) .
(3.2.110) Finally
kud(t0)−ud(t00)kL2(Ω)≤ Z t00
t0
k∂tudkL2(Ω)dt
≤ |t0−t00|k∂tudkL∞(0,T;L2(Ω))
(3.2.111) together with (3.2.109) and (3.2.110) yields
kud(t0)−ud(t00)kC(Ω)
≤C|t0−t00|β/2k∂tudkβ/2
L∞(0,T;L2(Ω))kudk(2−β)/2
L∞(0,T;H2(Ω)).
(3.2.112) In like manner
k∇u0(t0)− ∇u0(t00)kC(Ω)
≤C|t0−t00|β/2k∂tu0kβ/2
L∞(0,T;H01(Ω))ku0k(2−β)/2L∞(0,T;H3(Ω)).
(3.2.113) Then
(u0 ∈C([0, T], C1(Ω)), ud ∈C([0, T], C(Ω)).
Thus we have completed the proof of the theorem.