G. S T R ¨ O H M E R (Iowa City, Ia.) W. M. Z A J A ¸ C Z K O W S K I (Warszawa)
LOCAL EXISTENCE OF SOLUTIONS OF THE FREE BOUNDARY PROBLEM
FOR THE EQUATIONS OF COMPRESSIBLE BAROTROPIC VISCOUS SELF-GRAVITATING FLUIDS
Abstract. Local existence of solutions is proved for equations describing the motion of a viscous compressible barotropic and self-gravitating fluid in a domain bounded by a free surface. First by the Galerkin method and reg- ularization techniques the existence of solutions of the linearized momentum equations is proved, next by the method of successive approximations local existence to the nonlinear problem is shown.
1. Introduction. In this paper we prove the existence of local solutions to equations describing the motion of a viscous compressible barotropic fluid under the self-gravitating force in a bounded domain Ω
t⊂ R
3with a free boundary S
t. Let v = v(x, t) be the velocity of the fluid, ̺ = ̺(x, t) the density, p = p(̺) the pressure, µ and ν the constant viscosity coefficients and p
0the external constant pressure. Then the problem is described by the following system of equations (see [1], Chs. 1,2):
(1.1)
̺(v
t+ v · ∇v) − div T(v, p) = ̺∇U in e Ω
T= S
0≤t≤T
Ω
t× {t},
̺
t+ div(̺v) = 0 in e Ω
T,
T · n = −p
0n on e S
T= S
0≤t≤T
S
t× {t},
̺|
t=0= ̺
0, v|
t=0= v
0in Ω = Ω
0, v · n = − φ
t|∇φ| on e S
T,
1991 Mathematics Subject Classification: 76N10, 35G30.
Key words and phrases: local existence, compressible viscous barotropic self-gravita- ting fluids.
The research of W. M. Zaj¸aczkowski was partially supported by Polish Government grant KBN 2 P301 008 06.
[1]
where φ(x, t) = 0 describes S
t, n is the unit outward vector normal to S
t, n = ∇φ/|∇φ|, Ω
tis the domain at time t, S
t= ∂Ω
t, Ω = Ω
t|
t=0= Ω
0, S = ∂Ω.
By T = T(v, p) we denote the stress tensor of the form (1.2) T (v, p) = {T
ij}
i,j=1,2,3= {−pδ
ij+ D
ij(v)}
i,j=1,2,3, where
(1.3) D(v) = {D
ij}
i,j=1,2,3= {µ(∂
xiv
j+ ∂
xjv
i) + (ν − µ)δ
ijdiv v}
i,j=1,2,3is the deformation tensor.
Moreover, U (x, t) is the self-gravitating potential such that
(1.4) U (x, t) = k
\
Ωt
̺(y, t)
|x − y| dy, where k is the gravitation constant.
By the continuity equation (1.1)
2and the kinematic condition (1.1)
5the total mass is conserved, so
(1.5)
\
Ωt
̺(x, t) dx =
\
Ω
̺
0(x) dx = M, where M is a given constant.
Let Ω be given. We introduce the Lagrangian coordinates ξ as the initial data for the Cauchy problem
(1.6) dx
dt = v(x, t), x|
t=0= ξ ∈ Ω, ξ = (ξ
1, ξ
2, ξ
3).
Integrating (1.6) we obtain a transformation between the Eulerian x and the Lagrangian ξ coordinates,
(1.7) x = x(ξ, t) ≡ ξ +
t
\
0
u(ξ, τ ) dτ ≡ x
u(ξ, t),
where u(ξ, t) = v(x
u(ξ, t), t) and the index u in x
u(ξ, t) will be omitted in evident cases.
Then, by (1.1)
5, Ω
t= {x ∈ R
3: x = x(ξ, t), ξ ∈ Ω} and S
t= {x ∈ R
3: x = x(ξ, t), ξ ∈ S}.
Let η(ξ, t) = ̺(x(ξ, t), t), q(ξ, t) = p(x(ξ, t), t), ∇
u= ∂
xξ
i∇
ξi, ∂
ξi= ∇
ξi,
T
u(u, q) = −qI + D
u(u), I = {δ
ij}
i,j=1,2,3is the unit matrix, D
u(u) =
{µ(∂
xiξ
k∇
ξku
j+ ∂
xjξ
k∇
ξku
i) + (ν − µ)δ
ij∇
u· u}, where ∇
u· u = ∂
xiξ
k∇
ξku
iand summation over repeated indices is assumed.
Since S
tis determined (at least locally) by the equation φ(x, t) = 0, S is described by φ(x(ξ, t), t)|
t=0= e φ(ξ) = 0. Moreover, we have
n
u= n(x
u(ξ, t), t) = ∇
xφ(x, t)
|∇
xφ(x, t)|
x=x
u(ξ,t)
,
n
0= n
u0(ξ, t) = ∇
ξφ(ξ) e
|∇
ξφ(ξ)| e .
In Lagrangian coordinates the problem (1.1) takes the form
(1.8)
ηu
t− div
uT
u(u, q) = U
u(η) in Ω
T= Ω × (0, T ), η
t+ η div
uu = 0 in Ω
T,
T
u(u, q) · n
u= −p
0n
uon S
T= S × (0, T ),
u|
t=0= v
0in Ω,
η|
t=0= ̺
0in Ω,
where
(1.9) U
u(η) = k
\
Ω
η(ϑ, t)
|x
u(ξ, t) − x
u(ϑ, t)| A(x
u(ϑ, t)) dϑ
and A is the Jacobian determinant of the transformation x = x(ξ, t).
The proof of the existence of solutions of problem (1.8) is divided into a few steps. First we consider the problem
(1.10)
u
t− div D(u) = f
1, D (u) · n = b
1, u|
t=0= v
0.
At the second step we examine the problem with a given positive function η(ξ, t):
(1.11)
ηu
t− div D(u) = f in Ω
T, D (u) · n = g on S
T, u|
t=0= v
0in Ω.
To examine the nonlinear problem (1.8) we need an existence result for the problem
(1.12)
ηu
t− div
wD
w(u) = f
3, D
w(u) · n
w= g
3, u|
t=0= u
0,
where η > 0 and w = w(ξ, t) are given functions.
Finally, we prove the existence of solutions to (1.8), hence also to (1.1), by the following method of successive approximations:
(1.13)
η
m∂
tu
m+1− div
umD
um
(u
m+1) = −∇
umq(η
m) + η
mU
um(η
m) in Ω
T, D
um
(u
m+1) · n
um= −(q(η
m) − p
0)n
umon S
T,
u
m+1|
t=0= v
0in Ω,
where η
mand u
mare treated as given, and
(1.14) ∂
tη
m+ η
mdiv
umu
m= 0 in Ω
T, η
m|
t=0= ̺
0in Ω, where u
mis treated as given, and m = 0, 1, . . .
We want to point out that the presented proof of existence uses the Galerkin method and some regularization techniques because the case con- sidered is singular in potential theory. This is related to H
3(Ω) regularity.
Ordinarily the Galerkin and regularization methods are connected with the energy method which is much more natural for (1.1) than the potential tech- nique. Moreover, this technique is applied in the stability proof for (1.1) in [4]. We have also to emphasize that H
3(Ω) regularity for v is the lowest pos- sible regularity in spaces with integer derivatives for solutions of nonlinear problems such as (1.1) to exist. As follows from [3] the existence of solutions to (1.1) can be shown in the spaces H
2+α,1+α/2(Ω
T), α ∈ (1/2, 1), but the norm of these spaces contains fractional derivatives and is not convenient for our considerations in [4].
In [2] local existence of solutions for the free boundary problem for the equations of a viscous compressible heat-conducting self-gravitating fluid is proved. However, the proof is done in a different way and the regularity obtained is not suitable for our considerations in [4].
2. Notation. To simplify considerations we introduce the following notation:
kuk
s,Q= kuk
Hs(Q), s ∈ N ∪ {0}, Q ∈ {Ω, Ω
t, S, S
t}, Ω
t= Ω × (0, t), S
t= S × (0, t),
|u|
p,Q= kuk
Lp(Q), p ∈ [1, ∞],
kuk
s,p,q,ΩT= kuk
Lq(0,T ;Wps(Ω)), p, q ∈ [1, ∞], 0 ≤ s ∈ Z.
We define the space Γ
lk(Ω) as part of T
k−li=0
C
i([0, T ]; H
k−i(Ω)) with the norm kuk
Γkl(Ω)
= P
k−li=0
k∂
tiuk
k−i,Ω.
Then we denote by L
p(0, T ; Γ
lk(Ω)) the closure of C
∞(Ω
T) with the norm
T\0
X
k−li=0
k∂
tiuk
k−i,Ω pdt
1/p, p ∈ [1, ∞].
Moreover, we introduce
|u|
k,l,p,ΩT= kuk
Lp(0,T ; Γkl(Ω))
.
3. Existence of solutions. We prove the existence of solutions to prob- lem (1.1) by the method of successive approximations described by problems (1.13) and (1.14). Therefore, we first consider the following auxiliary prob- lem:
(3.1)
ηu
t− div
wD
w(u) = F in Ω
T, D
w(u) · n
w= G on S
T,
u|
t=0= v
0in Ω,
where η and w are given functions. Moreover, η is such that (3.2) 0 < ̺
∗≤ η ≤ ̺
∗< ∞
and w = w(ξ, t) is such that
(3.3) x = ξ +
t
\
0
w(ξ, τ ) dτ ≡ x
w(ξ, t) ≡ x(ξ, t), and
(3.4) ∂x
∂ξ , ∂ξ
∂x
are matrices with determinants close to 1 for t ∈ [0, T ].
Definition 3.1. By a weak solution to problem (3.1) we mean a function u which satisfies the integral identity
(3.5)
\
Ω
[ηu
tϕ + D
′w(u) · D
′w(ϕ) − F · ϕ]J
wdξ −
\
S
GϕJ
wdξ
s= 0
for any sufficiently smooth function ϕ, where D
′w(u) · D
′w(ϕ) =
µ2(∇
wiu
j+
∇
wju
i)(∇
wiϕ
j+ ∇
wjϕ
i) + (ν − µ)∇
w· u∇
w· ϕ and J
wis the Jacobian determinant of the transformation x = x
w(ξ, t).
To obtain the integral formula for (3.1) we use the following integration
by parts:
\
Ω
div
wD
w(v(x
w(ξ, t), t)ϕ(x
w(ξ, t), t)J
wdξ
=
\
Ωt
div D(v(x, t))ϕ(x, t) dx
= −
\
Ωt
D
′(v) · D
′(ϕ)dx +
\
St
n · D(v)ϕ ds
= −
\
Ω
D
′w(v(x
w(ξ, t), t)) · D
′w(ϕ(x
w(ξ, t), t))J
wdξ +
\
S
n
w· D
w(v(x
w(ξ, t), t))ϕ(x
w(ξ, t), t)J
wdξ
S, where D
′(v) · D
′(ϕ) =
µ2(∂
xiv
j+ ∂
xjv
i)(∂
xiϕ
j+ ∂
xjϕ
i) + (ν − µ) div v div ϕ.
Take a basis {ϕ
k} in L
2(Ω). Then we are looking for an approximate solution of (3.5) in the form
(3.6) u
n=
X
n i=1c
in(t)ϕ
i(ξ),
where c
in, i = 1, . . . , n, are solutions of the following system of ordinary differential equations:
\
Ω
[ηu
ntϕ
i+ D
′w(u
n) · D
′w(ϕ
i) − F · ϕ
i]J
wdξ −
\
S
Gϕ
iJ
wdξ
S= 0, (3.7)
u
n|
t=0= X
n i=1c
in(0)ϕ
i(ξ), c
in(0) =
\
Ω
v
i0ϕ
ndξ,
where i = 1, . . . , n, and existence follows from the theory of ordinary differ- ential equations.
Next we have to obtain estimates for solutions of (3.7).
Lemma 3.2. Assume that ̺
∗≤ η, η
t∈ L
2(0, T ; H
1(Ω)), F ∈ L
2(Ω
T), G ∈ L
2(S
T), w ∈ L
2(0, T ; H
3(Ω)). Assume that
(3.8) sup
t∈[0,T ]
sup
ξ∈Ω
|I − ξ
x| ≤ δ,
where δ is sufficiently small and I is the unit matrix. Then for solutions of (3.7) the following inequality holds:
ku
nk
20,Ω+ c
0ku
nk
21,2,2,Ωt≤ ψ
1(1/̺
∗, t, kη
tk
1,2,2,Ωt, a(w, t)) (3.9)
× h
\Ω
̺
0v
20dx + kF k
20,Ωt+ kGk
20,Sti ,
where ψ
1is an increasing positive function, a(w, t) = t
1/2kwk
3,2,2,Ωt, and
t ≤ T .
P r o o f. Multiplying (3.7) by c
inand summing over i from 1 to n we get (3.10) 1
2
\
Ω
η d
dt u
2n+ |D
′w(u
n)|
2J
wdξ =
\
Ω
F · u
nJ
wdξ +
\
S
G · u
nJ
wdξ
S. Using the Korn inequality
(3.11) kuk
21,Ω≤ c(kD
′′(u)k
20,Ω+ kuk
20,Ω) and |D
′(v)|
2≥ c
0|D
′′(v)|
2, c
0= min
34
ν −
13µ ,
µ2, where D
′′(u) = {µ(∂
xiu
j+ ∂
xju
i)}, we have
kuk
21,Ω≤ c(kD
′′w(u)k
20,Ω+ kD
′′(u) − D
′′w(u)k
20,Ω+ kuk
20,Ω), so in view of (3.8) we get
(3.12) kuk
21,Ω≤ c(δ)(kD
′′w(u)k
20,Ω+ kuk
20,Ω).
Using (3.12) in (3.10) implies d
dt
\
Ω
ηu
2nJ
wdξ + c
0ku
nk
21,Ω≤ c
\
Ω
(|η
t| + η|div
ww|)|u
n|
2J
wdξ (3.13)
+ c(ku
nk
20,Ω+ kF k
20,Ω+ kGk
20,S).
Estimating the first term on the r.h.s. by εku
nξk
20,Ω+
c(ε)
̺
∗kη
tk
21,Ω+ |div
ww|
∞,Ω \Ω
ηu
2nJ
wdξ, ε ∈ (0, 1), from (3.13) we get
(3.14) d dt
\
Ω
ηu
2nJ
wdξ + c
0ku
nk
21,Ω≤
c
̺
∗(1 + kη
tk
21,Ω) + |div
ww|
∞,Ω \Ω
ηu
2nJ
wdξ + c(kF k
20,Ω+ kGk
20,S).
Integrating (3.14) with respect to time yields (3.15)
\
Ω
ηu
2nJ
wdξ + c
0ku
nk
21,2,2,Ωt≤
exp
c
̺
∗(t + kη
tk
21,2,2,Ωt+ ϕ(a(w, t))) h
\Ω
̺
0v
20dξ + kF k
20,Ωt+ kGk
20,Sti , where ϕ is an increasing positive function. From (3.15) we obtain (3.9).
This concludes the proof.
From (3.9) we can prove the existence of weak solutions such that u ∈
L
∞(0, T ; L
2(Ω)) ∩ L
2(0, T ; H
1(Ω)). However, we want to obtain more reg-
ular weak solutions simultaneously. Therefore we show
Lemma 3.3. Assume that ̺
∗≤ η, F
t∈ L
2(Ω
T), G
t∈ L
2(S
T), F ∈ L
∞(0, T ; L
2(Ω)), G ∈ L
∞(0, T ; L
2(S)), w ∈ L
2(0, T ; H
3(Ω)), η
t∈ L
2(0, T ; H
1(Ω)) and
T
Ω
̺
0u
2t(0) dξ < ∞. Assume (3.8). Then ku
ntk
20,Ω+ c
0ku
ntk
21,2,2,Ωt≤ ψ
2(1/̺
∗, a(w, t), t, kη
tk
1,2,2,Ωt) (3.16)
× h
\Ω
̺
0u
2t(0) dξ + kF
tk
20,Ωt+ kG
tk
20,St+ sup
t
(ku
nk
21,Ω+ kF k
20,Ω+ kGk
20,S)
×
t
\
0
ε
1kwk
23,Ω+ c(ε
1)kwk
20,Ωdt i
, where ψ
2is an increasing positive function and ε
1∈ (0, 1).
P r o o f. Differentiating (3.7) with respect to t, multiplying by ˙c
inand summing up over i from 1 to n we get
(3.17) d dt
\
Ω
ηu
2ntJ
wdξ + c
0ku
ntk
21,Ω≤ c
\
Ω
|η
t|u
2ntJ
wdξ + ϕ
1(a(w, t))
\
Ω
ηu
2ntJ
w|w
ξ| dξ + c(ku
ntk
20,Ω+ kF
tk
20,Ω+ kG
tk
20,S)
+ ϕ
1(a(w, t))|w
ξ|
2∞,Ω(ku
nk
21,Ω+ kF k
20,Ω+ kGk
20,S), where ϕ
1is an increasing positive function, a(w, t) was defined in Lemma 3.2, and the Korn inequality and condition (3.8) were used.
Estimating the first term on the r.h.s. of (3.17) by c
02 ku
ntk
21,Ω+ c(µ, ν)
̺
∗kη
tk
21,Ω\
Ω
ηu
2ntJ
wdξ we can write (3.17) in the form
(3.18) d dt
\
Ω
ηu
2ntJ
wdξ + c
0ku
ntk
21,Ω≤ ϕ
2(a(w, t))(1 + 1/̺
∗)(1 + kη
tk
21,Ω+ |w
ξ|
∞,Ω)
×
\
Ω
ηu
2ntJ
wdξ + c(kF
tk
20,Ω+ kG
tk
20,S) + ϕ
1(a(w, t))(εkwk
23,Ω+ c(ε)kwk
20,Ω)
× (ku
nk
21,Ω+ kF k
20,Ω+ kGk
20,S).
Integrating (3.18) with respect to t we get (3.19)
\
Ω
ηu
2ntJ
wdξ + c
0ku
ntk
21,2,2,Ωt≤ exp[ϕ
2(a(w, t))(1 + 1/̺
∗)
× (t + kη
tk
21,2,2,Ωt+ a(w, t))]
× h
\Ω
̺
0u
2t(0) dξ + kF
tk
20,Ωt+ kG
tk
20,St+ ϕ
1(a(w, t)) sup
t
(ku
nk
21,Ω+ kF k
20,Ω+ kGk
20,S)
×
t
\
0
(εkwk
23,Ω+ c(ε)kwk
20,Ω) dt i . From (3.19) we have (3.16). This concludes the proof.
To estimate the expression sup
tku
nk
21,Ωon the r.h.s. of (3.16) we need the following result.
Lemma 3.4. Let the assumptions of Lemma 3.3 be satisfied. Then (3.20) ku
ntk
20,Ωt+ c
0ku
nk
21,Ω≤ ψ
3t, 1/̺
∗, a(w, t),
t
\
0
(ε
1kwk
23,Ω+ c(ε
1)kwk
20,Ω) dt
× h
ku
0k
21,Ω+
\
Ω
̺
0v
20dx + kF k
20,Ωt+ c(ε
2)kGk
20,St+ ε
2ku
ntk
21,2,2,Ωti , where ψ
3is an increasing positive function and ε
1, ε
2∈ (0, 1).
P r o o f. Multiplying (3.7) by ˙c
inand summing over i from 1 to n we get (3.21)
\
Ω
ηu
2ntJ
wdξ +
\
Ω
D
′w(u
n) · D
′w(u
nt)J
wdξ
=
\
Ω
F · u
ntJ
wdξ +
\
S
G · u
ntJ
wdξ
S. From (3.21) in view of the H¨older and Young inequalities we obtain (3.22)
\
Ω
ηu
2ntJ
wdξ + d dt
\
Ω
|D
′w(u
n)|
2J
wdξ
≤ c(|w
ξ|
2∞,Ω+ 1)
\
Ω
|D
′w(u
n)|
2J
wdξ + c
\
Ω
|∇
ξu
n|
2J
wdξ
+ εku
ntk
21,Ω+ c(ε)kGk
20,S+ c(̺
∗)kF k
20,Ω.
Integrating (3.22) with respect to time, using the Korn inequality and (3.8) we get
(3.23)
\
Ωt
ηu
2ntJ
wdξ + c
0ku
nk
21,Ω≤ exp h c
\t0
(ε
1kwk
23,Ω+ c(ε
1)kwk
20,Ω) dt + t i
× [ku
0k
21,Ω+ ku
nk
21,2,2,Ωt+ ε
2ku
ntk
21,2,2,Ωt+ c(ε
2)kGk
20,St+ c(̺
∗)kF k
20,Ωt] + cku
nk
20,Ω. Using (3.9) in (3.23) yields
(3.24) ku
ntk
20,Ωt+ c
0ku
nk
21,Ω≤ exp h c
t\0
(ε
1kwk
23,Ω+ c(ε
1)kwk
20,Ω) dt + t i
× h
ku
0k
21,Ω+ ψ
1 \Ω
̺
0v
02dx + kF k
20,Ωt+ kGk
20,St+ ε
2ku
ntk
21,2,2,Ωt+ c(ε
2)kGk
20,St+ c(̺
∗)kF k
20,Ωti .
From (3.24) we obtain (3.20). This concludes the proof.
Inserting the estimate for ku
nk
21,Ωfrom (3.20) into the r.h.s. of (3.16) and assuming that
ε
2ψ
3c
0ψ
2b(t, ε
1, w) = c
02 , where
(3.25) b(t, ε
1, w) =
t
\
0
(ε
1kwk
23,Ω+ c(ε
1)kwk
20,Ω) dt, we obtain
(3.26) ku
ntk
20,Ω+ c
02 ku
ntk
21,2,2,Ωt≤ ψ
2h
\Ω
̺
0u
2t(0) dξ + kF
tk
20,Ωt+ kG
tk
20,St+ sup
t
(kF k
20,Ω+ kGk
20,S)b(t
1, ε
1, w) i + ψ
2ψ
3µ b
ku
0k
21,Ω+
\
Ω
̺
0v
02dξ + kF k
20,Ωt+ c
ψ
2ψ
3b µ
2kGk
20,St.
Simplifying the expression we get
Lemma 3.5.
(3.27) ku
ntk
20,Ω+ c
0ku
ntk
21,2,2,Ωt≤ ψ
4(t, 1/̺
∗, a(w, t), b(t, ε
1, w), kη
tk
1,2,2,Ωt)
× h
\Ω
̺
0v
02dξ +
\
Ω
̺
0u
2t(0) dξ + ku
0k
21,Ω+ kF
tk
20,Ωt+ kG
tk
20,St+ kF k
20,Ωt+ kGk
20,St+ sup
t
(kF k
20,Ω+ kGk
20,S) i . From (3.27), (3.20) and (3.9) we get
Lemma 3.6. Let the assumptions of Lemmas 3.2–3.4 be satisfied. Then (3.28) ku
nk
20,Ω+ ku
nk
21,Ω+ ku
ntk
20,Ω+ ku
nk
21,2,2,Ωt+ ku
ntk
21,2,2,Ωt≤ ψ
5(t, 1/̺
∗, a(w, t), b(t, ε
1, w), kη
tk
1,2,2,Ωt)
× h
\Ω
̺
0v
02dx +
\
Ω
̺
0u
2t(0) dx + ku
0k
21,Ω+ kF
tk
20,Ωt+ kG
tk
20,Ωt+ sup
t
(kF k
20,Ω+ kGk
20,S) i , where ψ
5is an increasing positive function of its arguments.
Now choosing a subsequence and passing with n to infinity we get Lemma 3.7. Assume that ̺
∗≤ η ≤ ̺
∗, w ∈ L
2(0, T ; H
3(Ω)), η
t∈ L
2(0, T ; H
1(Ω)), v
0∈ H
1(Ω), u
t(0) ∈ L
2(Ω), F
t∈ L
2(Ω
T), G
t∈ L
2(S
T), F ∈ L
∞(0, T ; L
2(Ω)) and G ∈ L
∞(0, T ; L
2(S)). Then there exists a weak solution of problem (3.1) such that u ∈ L
∞(0, T ; H
1(Ω)) ∩ L
2(0, T ; H
1(Ω)), u
t∈ L
2(0, T ; H
1(Ω)) ∩ L
∞(0, T ; L
2(Ω)), and
(3.29) kuk
21,Ω+ ku
tk
20,Ω+ kuk
21,2,2,Ωt+ ku
tk
21,2,2,Ωt≤ ψ
5(t, 1/̺
∗, a(w, t), b(t, ε
1, w), kη
tk
1,2,2,Ωt)
× h
\Ω
̺
0v
20dx +
\
Ω
̺
0u
2t(0) dx + kv
0k
21,Ω+ kF
tk
20,Ωt+ kG
tk
20,St+ sup
t
(kF k
20,Ω+ kGk
20,S) i
.
Having proved the existence of weak solutions to problem (3.1) expressed
by Lemma 3.7 we obtain by regularization techniques (see Appendix, The-
orem 4.1 and Remark 4.2) the following result:
Lemma 3.8. Let the assumptions of Lemma 3.7 be satisfied. Let v
0∈ H
2(Ω), w ∈ L
2(0, T ; H
3(Ω)), F ∈ L
2(0, T ; H
1(Ω)), G ∈ L
2(0, T ; H
3/2(S)), η ∈ L
∞(0, T ; H
2(Ω)), and S ∈ H
5/2. Then there exists a unique solution to problem (3.1) such that u ∈ L
∞(0, T ; H
1(Ω)) ∩ L
2(0, T ; H
3(Ω)), u
t∈ L
∞(0, T ; L
2(Ω)) ∩ L
2(0, T ; H
1(Ω)), and
(3.30) ku
tk
20,Ω+ kuk
21,Ω+ kuk
23,2,2,Ωt+ ku
tk
21,2,2,Ωt≤ ψ
6(t, 1/̺
∗, a(w, t), b(t, ε
1, w), kη
tk
1,2,2,Ωt, kηk
2,2,∞,Ωt)
× h
\Ω
̺
0v
20dx +
\
Ω
̺
0u
2t(0) dx + kv
0k
22,Ω+ kF k
21,Ωt+ kG
tk
20,St+ kGk
23/2,2,2,St+ kF k
20,2,∞,Ωt+ kGk
20,2,∞,Sti , where ψ
6is an increasing positive function.
Now we prove the existence of solutions of (1.1) by the method of suc- cessive approximations determined by problems (1.13) and (1.14). First we show the boundedness of the sequence described by (1.13) and (1.14) in the norm defined in Lemma 3.8.
To simplify considerations let us introduce
(3.31) α
m(t) = ku
mtk
20,Ω+ ku
mk
21,Ω+ ku
mk
23,2,2,Ωt+ ku
mtk
21,2,2,Ωt. Lemma 3.9. Assume that v
0∈ H
2(Ω), ̺
0∈ H
2(Ω), and there exist two positive constants ̺
∗and ̺
∗, ̺
∗< ̺
∗and ̺
∗≤ ̺
0≤ ̺
∗.
T (v
0, p(̺
0))n = −p
0n on S.
Then for A such that G(0, 0, F
0) < A, α
m(0) ≤ A, where F
0= kv
0k
22,Ω+ ku
t(0)k
20,Ω+ k̺
0k
22,Ωand G is defined by the r.h.s. of (3.41), there exists T
∗such that for t ≤ T
∗,
(3.32) α
m(t) ≤ A, m = 1, 2, . . . Moreover , in view of (1.1)
1we have
ku
t(0)k
0,Ω≤ ckv
0k
22,Ω+ p
′(̺
∗)
̺
∗k̺
0k
1,Ω+ µ + ν
̺
∗kv
0k
2,Ω+ ck̺
0k
0,Ω. P r o o f. First we obtain estimates for solutions of (1.14). Integrating (1.14) we get
(3.33) η
m(ξ, t) = ̺
0(ξ) exp h
−
t
\
0
div
umu
mdτ i
.
From (3.33) we have
(3.34)
sup
Ωt
η
m+ sup
Ωt
1 η
m≤ k̺
0k
2,Ωϕ
1(a(u
m, t)), sup
t
kη
mk
2,Ω≤ k̺
0k
2,Ωϕ
1(a(u
m, t))ϕ
2(a(u
m, t)), where a(u
m, t) = t
1/2(
Tt
0
ku
mk
23,Ωdt)
1/2. Moreover,
η
mt= ̺
0(ξ) exp h
−
t
\
0
div
umu
mdτ i
(− div
umu
m).
Therefore
(3.35) kη
mtk
1,2,2,Ωt≤ k̺
0k
2,Ωϕ
3(a(u
m, t))b(t, ε, u
m), where b(t, ε, u
m) is defined by (3.25).
Comparing (3.1) with (1.13) we have
(3.36) F = −∇
umq(η
m) + η
mU
um(η
m), G = −(p(η
m) − p
0)n
um. From (3.36) we have
(3.37) kF k
21,2,2,Ωt+ kF k
20,2,∞,Ωt≤ ϕ
4(t, a(u
m, t), k̺
0k
2,Ω) and
(3.38) kF
tk
20,Ωt≤ ϕ
5(a(u
m, t), sup
t
kη
mk
2,Ω)b(t, ε, u
m).
Moreover,
(3.39) kGk
23/2,2,2,St+ kGk
20,2,∞,Ωt≤ ϕ
6(a(u
m, t), t, sup
t
kη
mk
2,Ω) and
(3.40) kG
tk
20,St≤ ϕ
7(a(u
m, t), sup
t
kη
mk
2,Ω)b(t, ε, u
m).
Using the fact that
a
2(u
m, t) ≤ tα
m, b(t, ε, u
m) ≤ t
aα
m+ cF
0, a > 0, and inserting all the above estimates into (3.30) we get
(3.41) α
m+1(t) ≤ G(t, t
aα
m(t), F
0),
where a > 0, F
0= kv
0k
22,Ω+ ku
t(0)k
20,Ω+ k̺
0k
22,Ω, and G is an increasing positive function.
Let A be such that G(0, 0, F
0) < A. Since G is a continuous increasing
function of its arguments there exists T
∗> 0 such that for t ≤ T
∗we have
(3.42) G(t, t
aA, F
0) ≤ A.
From (3.42) we see that if α
m(t) ≤ A then α
m+1(t) ≤ A for t ≤ T
∗. Here A must be so large that α
m(0) ≤ A.
To end the proof we have to construct the zero approximation function u
0. We use the solution of the problem
u
0t− div D(u
0) = 0 in Ω
T, n · D(u
0) = (p(̺
0) − p
0)n on S
T, u
0|
t=0= v
0in Ω.
The existence of solutions to the above problem follows from the Galerkin method and can be proved in the classes determined by α
0(t) < ∞. More- over, the compatibility condition is satisfied. Finally, A must be so large that α
0(t) ≤ A, t ≤ T
∗. This concludes the proof.
Now we prove the convergence of the sequence {u
m, η
m}.
To show this we obtain from (1.13) and (1.14) the following system of problems for the differences U
m= u
m− u
m−1and H
m= η
m− η
m−1:
(3.43)
η
m∂
tU
m+1− div
umD
um
U
m+1= − H
m∂
tu
m− (div
umD
um
(u
m) − div
um−1D
um−1
(u
m))
− (∇
um− ∇
um−1)q(η
m) − ∇
um−1(q(η
m) − q(η
m−1)) + H
mU
um(η
m) + η
m−1U
um(H
m)
+ η
m−1(U
um(η
m−1) − U
um−1(η
m−1))
≡ X
7 i=1F
i≡ e F , D
um
(U
m+1) · n
um= − (D
um(u
m) · n
um− D
um−1(u
m) · n
um−1)
− q(η
m)(n
um− n
um−1)
− (q(η
m) − q(η
m−1))n
um−1+ p
0(n
um− n
um−1)
≡ X
4 i=1G
i≡ e G, U
m+1|
t=0= 0,
and
(3.44) ∂
tH
m+ H
mdiv
umu
m= −η
m−1(div
umu
m− div
um−1u
m−1),
H
m|
t=0= 0.
Now we write the expressions on the r.h.s. of (3.43)
1in qualitative forms:
(3.45)
F
2= f
1 t\
0
U
mξdτ u
mξξ+ f
2 t\
0
U
mξdτ
t
\
0
dτ u
mξξu
mξ+ f
3 t\
0
U
mξdτ
t
\
0
u
m−1,ξξdτ u
mξ+ f
4 t\
0
U
mξξdτ u
mξ,
F
3= f
5f
1′t
\
0
U
mξdτ η
mξ,
F
4= f
6f
2′(η
mξ+ η
m−1ξ)H
m+ f
7f
3′H
mξ, G
1= f
8t
\
0
U
mξdτ u
mξ,
G
2= f
9f
4′t
\
0
U
mξdτ,
G
3= p
0f
10 t\
0
U
mξdτ, G
4= f
11f
5′H
m, where f
i= f
i(I+
Tt
0
u
mξdτ, I+
Tt
0
u
m−1,ξdτ ), i = 1, . . . , 11, f
j′= f
j′(η
m, η
m−1), j = 1, . . . , 5, are C
∞functions of their arguments and I is the unit matrix.
Moreover, we have the estimates
(3.46) |f
i| ≤ ϕ
1(A), |f
j′| ≤ ϕ
2(A)
where ϕ
1, ϕ
2are increasing positive functions, for i, j as above.
Therefore we have
Lemma 3.10. Let the assumptions of Lemma 3.9 be satisfied. Then there exists 0 < T
∗∗sufficiently small such that
(3.47) kU
m+1k
21,Ω+ kU
m+1,tk
20,Ωt+ kU
m+1k
22,2,2,Ωt≤ δkU
mk
22,2,2,Ωt, where δ = δ(t) < 1 for t ≤ T
∗∗.
P r o o f. To show (3.47) we multiply (3.43) by U
m+1J
umand integrate over Ω. Therefore after integration by parts we get
(3.48) 1 2
\
Ω
η
md
dt U
m+12J
umdξ +
\
Ω
|D
′um(U
m+1)|
2J
umdξ
=
\
Ω
F U e
m+1J
umdξ +
\
S
GU e
m+1J
umdξ
S.
First we estimate all terms on the r.h.s.:
\
Ωt
H
mu
mtU
m+1J
umdξ dt ≤ εkU
m+1k
21,2,2,Ωt+ c(ε)ϕ(A)t sup
t
ku
mtk
20,Ωsup
t
kH
mk
21,Ω,
\
Ωt
F
2U
m+1J
umdξ dt ≤ εkU
m+1k
21,2,2,Ωt+ c(ε)ϕ(A)t
t
\
0
kU
mk
22,Ωdt,
\
Ωt
F
3U
m+1J
umdξ dt ≤ εkU
m+1k
21,2,2,Ωt+ c(ε)ϕ(A)t
t
\
0
kU
mk
22,Ωdt,
\
Ωt
F
4U
m+1J
umdξ dt ≤ εkU
m+1k
21,2,2,Ωt+ c(ε)ϕ(A)t sup
t
kH
mk
21,Ω,
\
Ωt
(F
5+ F
6)U
m+1J
umdξ dt ≤ εkU
m+1k
20,Ωt+ c(ε)ϕ(A)t sup
t
kH
mk
20,Ω,
\
Ωt
F
7U
m+1J
umdξ dt ≤ εkU
m+1k
20,Ωt+ c(ε)ϕ(A)tkU
mk
21,2,2,Ωt. Next we estimate the boundary term (3.48):
\
St
(G
1+ G
2+ G
3)U
m+1J
umdξ
sdt ≤ εkU
m+1k
21,2,2,Ωt+ c(ε)ϕ(A)tkU
mk
22,2,2,Ωt,
\
St
G
4U
m+1J
umdξ
sdt ≤ εkU
m+1k
21,2,2,Ωt+ c(ε)ϕ(A)t sup
t
kH
mk
21,Ω. Using the Korn inequality in (3.48), integrating with respect to time, using the above estimates and taking ε sufficiently small we obtain
(3.49) kU
m+1k
20,Ω+kU
m+1k
21,2,2,Ωt≤ ϕ(A)t(kU
mk
22,2,2,Ωt+sup
t
kH
mk
21,Ω).
Multiplying (3.43)
1by U
m+1,tJ
um, integrating over Ω and by parts we have (3.50)
\
Ω
η
m|U
m+1,t|
2J
umdξ +
\
Ω
D
′um
(U
m+1) · D
′um(U
m+1,t)J
umdξ
=
\
S
GU e
m+1,tJ
umdξ +
\
Ω
F · U e
m+1,tJ
umdξ.
Continuing, we have (3.51)
\
Ω
η
m|U
m+1,t|
2J
umdξ +
\
Ω
D
′um
(U
m+1) · d dt D
′um
(U
m+1)J
umdξ
−
\
Ω
D
′um
(U
m+1) · ∂
t(D
′um)(U
m+1)J
umdξ
= d dt
\
S
G · U e
m+1J
umdξ
S−
\
S
G e
t· U
m+1J
umdξ
S−
\
S
G · U e
m+1J
umdiv
umu
mdξ
S+
\
Ω
F · U e
m+1,tJ
umdξ.
In view of the H¨older and Young inequalities we get (3.52)
\
Ω
η
m|U
m+1,t|
2J
umdξ + d dt
\
Ω
|D
′um(U
m+1)|
2J
umdξ
≤ d dt
\
S
G · U e
m+1J
umdξ
S+ cku
mk
23,Ω·
\
Ω
|D
′um(U
m+1)|
2J
umdξ
+ c
\
Ω
|∇
ξU
m+1|
2J
umdξ + ck e F k
20,Ω+ ε
1(k e G
tk
20,S+ ϕ(A)ku
mk
23,Ωk e Gk
20,S) + c(ε
1)kU
m+1k
21,Ω,
where ε
1∈ (0, 1).
Integrating with respect to time and using the Korn inequality we obtain from (3.52)
(3.53) kU
m+1,tk
20,Ω+ kU
m+1k
21,Ω≤ [ε
2k e Gk
20,S+ c(ε
2)(ε
3kU
m+1,ξk
20,Ω+ c(ε
3)kU
m+1k
20,Ω) + ckU
m+1k
21,2,2,Ωt+ ck e F k
20,Ωt+ ε
1ϕ(A)(k e G
tk
20,St+ sup
t
k e Gk
20,S)]e
A+ ckU
m+1k
20,Ω,
where we used the facts that
\
S
G · U e
m+1J
umdξ ≤ ε
2k e Gk
20,S+ c(ε
2)kU
m+1k
20,S, kU
m+1k
20,S≤ ε
3kU
m+1ξk
20,Ω+ c(ε
3)kU
m+1k
20,Ω,
t
\
0
ku
mk
23,Ωk e Gk
20,Sdt ≤ sup
t
k e Gk
20,St
\
0
ku
mk
23,Ωdt ≤ A sup
t
k e Gk
20,S. Using (3.49) in (3.53) implies
(3.54) kU
m+1k
21,Ω+ kU
m+1,tk
20,Ωt+ kU
m+1k
21,2,2,Ωt≤ ϕ(A)[ε(k e G
tk
20,St+ sup
t
k e Gk
20,S) + k e F k
20,Ωt] + ϕ(A)t(kU
mk
22,2,2,Ωt+ sup
t
kH
mk
21,Ω).
Now from the regularity result for the parabolic problem η
mU
m+1,t− div
umD
um
(U
m+1) = e F in Ω
T, D
um(U
m+1) · n
um= e G on S
T,
U
m+1|
t=0= 0 in Ω,
we obtain (see Theorem 4.1 and Remark 4.2)
(3.56) kU
m+1k
22,2,2,Ωt≤ c(k e F k
20,Ωt+ k e Gk
21/2,2,2,St) + ckU
m+1k
20,Ωt. Now collecting (3.49), (3.54) and (3.56) together, we get
(3.57) kU
m+1k
21,Ω+ kU
m+1,tk
20,Ωt+ kU
m+1k
22,2,2,Ωt≤ ϕ(A)[ε sup
t
k e Gk
20,S+ εk e G
tk
20,St+ k e F k
20,Ωt+ k e Gk
21/2,2,2,St] + ϕ(A)t(kU
mk
22,2,2,Ωt+ sup
t
kH
mk
21,Ω).
Using the form of e F and e G we estimate the terms in the first bracket on the r.h.s. of (3.57):
(3.58)
k e Gk
20,S≤ ϕ(A) h t
t
\
0
kU
mk
22,Ωdt + sup
t
kH
mk
21,Ωi ,
k e G
tk
20,St≤ ϕ(A) h
t\0
kU
mk
22,Ωdt +
t
\
0
kH
mtk
21,Ωdt i ,
k e Gk
21/2,2,2,St+ k e F k
20,Ωt≤ ϕ(A)t h
t\0
kU
mk
22,Ωdt + sup
t
kH
mk
21,Ωi
.
Using the equation (3.47) we have the estimate (3.59)
t
\
0
kH
mtk
21,Ωdt ≤ ϕ(A)(t sup
t
kH
mk
21,Ω+ kU
mk
22,2,2,Ωt).
From (3.57)–(3.59) it follows that
(3.60) kU
m+1k
21,Ω+ kU
m+1,tk
20,Ωt+ kU
m+1k
22,2,2,Ωt≤ ϕ(A)(c(ε)t + ε)(kU
mk
22,2,2,Ωt+ sup
t
kH
mk
21,Ω).
Integrating (3.44) we respect to time yields (3.61) H
m(ξ, t)
= − exp h
−
t
\
0
div
umu
mdτ i
t\0
h η
m−1(div
umu
m− div
um−1u
m−1)
× exp
t\′0
div
umu
mdt
′′i dt
′, hence one has
(3.62) sup
t
kH
mk
21,Ω≤ ϕ(A)tkU
mk
22,2,2,Ωt. Using (3.62) in (3.60) yields
(3.63) kU
m+1k
21,Ω+ kU
m+1,tk
20,Ωt+ kU
m+1k
22,2,2,Ωt≤ ϕ(A)(c(ε)t + ε)kU
mk
22,2,2,Ωt. Therefore for t so small that
(3.64) (c(ε)t + ε)ϕ(A) < 1
we have convergence of the sequence {u
m, η
m} to a solution. Assume that (3.64) holds for t ≤ T
∗∗. This concludes the proof.
From Lemmas 3.9 and 3.10 we have
Theorem 3.11. Let the assumptions of Lemmas 3.9 and 3.10 be satisfied.
Then there exists T
∗∗sufficiently small such that for T ≤ T
∗∗there exists a solution to problem (1.1) such that
u ∈ L
∞(0, T ; H
1(Ω)) ∩ L
2(0, T ; H
3(Ω)), u
t∈ L
∞(0, T ; L
2(Ω)) ∩ L
2(0, T ; H
1(Ω)) and
(3.65) kuk
1,2,∞,ΩT+ kuk
3,2,2,ΩT+ ku
tk
0,2,∞,ΩT+ ku
tk
1,2,2,ΩT≤ A,
where A is defined in Lemma 3.9. Moreover, η, 1/η ∈ L
∞(Ω
T) ∩
L
∞(0, T ; H
2(Ω)), η
t, (1/η)
t∈ L
∞(0, T ; L
2(Ω))∩L
2(0, T ; H
2(Ω)), η
tt, (1/η)
tt∈ L
2(Ω
T), and
(3.66) kχk
2,2,∞,ΩT+ kχ
tk
0,2,∞,ΩT+ kχ
tk
2,2,2,ΩT+ kχ
ttk
0,ΩT≤ ϕ(A), where χ replaces either η or 1/η and ϕ is some positive function.
P r o o f. We only have to show the last statement and the estimate (3.66).
They follow from the expression for η, (3.67) η(ξ, t) = ̺
0(ξ) exp
−
t
\
0
div
uu(ξ, τ ) dτ .
The most difficult part is to estimate η
tt. Taking the second derivative of η with respect to time we obtain
η
tt= ̺
0exp
−
t
\
0
div
uu(ξ, τ ) dτ
(−(div
uu)
t+ (div
uu)
2).
Since the first two factors are bounded we only consider the last bracket.
Qualitatively,
div
uu = f
1 t\0
u
ξdτ u
2ξ, where f
1is a smooth function and f
1(
Tt
0
u
ξdτ ) is bounded. Next in view of (3.65),
|u
2ξ|
2,Ωt≤
\t0
|u
ξ|
2∞,Ωdt
1/2sup
t
|u
ξ|
2,Ω≤ A
2. Similarly,
(div
uu)
t= f
2 t\0
u
ξdτ
u
ξt+ f
3 t\0
u
ξdτ u
2ξ,
where f
2, f
3are smooth functions and the same considerations as above can be applied. This concludes the proof.
4. Appendix. In this section we show the regularity of solutions to problem (1.12). First we consider the problem
(4.1)
ηu
t− div D(u) = F in Ω
T, n · D(u) = G on S
T, u|
t=0= v
0in Ω.
We examine (4.1) using the following weak formulation:
(4.2)
\
Ω
ηu
tϕ dx +
\
Ω
D
′(u) · D
′(ϕ) dx =
\
Ω
F · ϕ dx +
\
S
Gϕ ds.
To examine regularity we only have to consider the integral
(4.3) K(ϕ) :=
\
Ω
D
′(u) · D
′(ϕ) dx.
Set ϕ = ζϕ
1, where ζ is a smooth function with a support in e Ω ⊂ Ω and ϕ
1is a test function. Then we get
(4.4) K(ζϕ
1) =
\
Ωe
[D
′(u) · D
′(ζ)ϕ
1+ D
′(u
′) · D
′(ϕ
1) − uD
′(ζ) · D
′(ϕ
1)] dx,
where u
′= uζ.
In further considerations we choose e Ω such that e Ω ∩ S 6= ∅. Therefore we apply the transformation Φ : e Ω → b Ω which straightens locally the boundary of Ω. Hence (4.4) takes the form
(4.5) K(b ζ e ϕ
1) =
\
Ωb
[D
′Φ(b u)·D
′Φ(b ζ) b ϕ
1+D
′Φ(e u)·D
′Φ( b ϕ
1)−b uD
′Φ(b ζ)·D
′Φ( b ϕ
1)]J
Φdz,
where e Ω ∋ x → Φ(x) = z ∈ b Ω, b u = u ◦ Φ
−1, e u = b ub ζ, D
Φis such that ∇
xin D are replaced by ∇
xΦ(x)|
x=Φ−1(z)· ∇
zand J
Φis the Jacobian determinant of the transformation z = Φ(x). We also need the fact that b Ω = {z ∈ R
3:
|z
i| < d, i = 1, 2, 0 < z
3< d}, b S = Φ( e S) = {z ∈ R
3: |z
i| < d, i = 1, 2, z
3= 0}, and e S = e Ω ∩ S. Since the integrand in (4.5) vanishes on ∂ b Ω \ b S it can be extended by zero on R
3+= {z ∈ R
3: z
3> 0}. Therefore, we assume b ϕ
1= δ
h−1δ
he u, where δ
hu(z) =
1h(u(z
′+ h, z
3) − u(z)), z
′= (z
1, z
2), corresponds only to the tangent directions, which will also be denoted by τ . Then from (4.5) under the assumption that S and hence Φ are smooth, we have the estimate
(4.6) K(b ζδ
h−1δ
hu) ≥ e c
02 kδ
huk e
21, bΩ− εkD
′Φ(δ
hb u)k
20, bΩ− ckb uk
21, bΩ, where ε ∈ (0, 1).
Now we consider the first term in (4.2):
\
Ω
ηu
tϕ dx =
\
Ωe
ηu
′tϕ
1dx =
\
Ωb
b
ηe u
tϕ b
1J
Φdz (4.7)
=
\
Ωb
δ
hηJ b
Φu e
tδ
hu dz + b
\
Ωb
b
ηδ
hJ
Φe u
tδ
hb u dz +
\
Ωb
b
ηδ
hu e
tδ
huJ b
Φdz,
where the last term is equal to 1
2
\
Ωb
b η d
dt |δ
he u|
2J
Φdz = 1 2
d dt
\
Ωb
b
η|δ
he u|
2J
Φdz − 1 2
\
Ωb