Dierential and Integral Equations Volume 18, Number 10 (2005), 1081-1110

RESOLVENT ESTIMATES FOR THE STOKES OPERATOR

ON AN INFINITE LAYER
Helmut Abels
Max Planck Institute for Mathematics in Science
Inselstr. 22, 04103 Leipzig, Germany
Michael Wiegner
Department of Mathematics I, RWTH Aachen, 52056 Aachen, Germany
(Submitted by: Y. Giga)
Abstract. In this paper we prove unique solvability of the general-
ized Stokes resolvent equations in an innite layer = R
n1
(1, 1),
n 2, in L
q
Sobolev spaces, 1 < q < , with nonslip boundary
condition u|

= 0. The unique solvability is proved for every

C\ (,
2
/4], where

2
4
is the least upper bound of the spectrum
of Dirichlet realization of the Laplacian and the Stokes operator in .
Moreover, we provide uniform estimates of the solutions for large spec-
tral parameter as well as close to

2
4
. Because of the special
geometry of the domain, a partial Fourier transformation is used to cal-
culate the solution explicitly. Then Fourier multiplier theorems are used
to estimate the solution operator.
1. Introduction
In the present contribution, we study the generalized Stokes resolvent
equation
( )u +q = f in , (1.1)
div u = g in , (1.2)
u[

= 0 on , (1.3)
where = R
n1
(1, 1), n 2, is an innite layer. The Stokes resolvent
equations are the starting point for the construction of strong solutions of
the Navier-Stokes equations using semigroup theory, cf. e.g. [10, 11, 18].
Our main result is
Accepted for publication: March 2005.
AMS Subject Classications: 35Q30, 76D07.
1081
1082 Helmut Abels and Michael Wiegner
Theorem 1.1. Let 1 < p < and C (,

2
4
]. Then for every
f L
p
()
n
and g W
1
p
()

W
1
p
() there is a unique solution (u, q)
W
2
p
()
n

W
1
p
() of (1.1)-(1.3). Moreover, for every > 0 and p r <
with
n1
2
(
1
p

1
r
) 1 there is a constant C

such that

+
2
/4

|u|
p
+

+
2
/4

(1 +[[)
|
2
u|
p
+|q|
p
C

_
|(f, g)|
p
+ (1 +[[)|g|

W
1
p
_

+
2
/4

1
n1
2
(
1
p

1
r
)
|u|
r
C

_
|(f, g)|
p
+ (1 +[[)|g|

W
1
p
_
uniformly in S

:= C V

, where V

:= C : [ Im[ (Re +

2
/4).
Here W
m
p
(), m N, denotes the usual Sobolev space of order m based
on the standard Lebesgue space L
p
(), 1 p , and

W
1
p
() :=
_
q L
p,loc
() : q L
p
()
_

W
1
p
() :=
_
g L
p,loc
() : sup
C

(0)
()
[g, )[||
1
p

<
_
,
where
1
p
+
1
p

= 1 and f L
p,loc
() if and only if f L
p
(B) for all balls
B R
n
with B ,= .
The result has been known since 1994 in an unpublished version [17]
and was used by several author; e.g. by Abels [6] the unique solvability
of the Stokes resolvent equations was used in order to prove the existence of
bounded imaginary powers of the Stokes operator in an innite layer.
An alternative proof of unique solvability of the Stokes resolvent equations,
i.e., g = 0, was later given by Abe and Shibata [1, 2] for f L
p
(), 1 < p <
, and C (, 0). An approach to the generalized Stokes resolvent
equations using a reduction to a pseudodierential boundary-value problem,
which works for C (, 0], may be found in [7, Section 5]; see also
[3, Remark 1.2]. Moreover, Nazarov and Pileckas [14, 15] considered the
solvability of the Stokes equations ( = 0) in weighted L
2
Sobolev spaces.
Since the unique solvability of the generalized Stokes resolvent equations,
i.e., g ,= 0, is important for perturbation argument, cf. [4, 5], the authors
decided to give a rigorous proof of Theorem 1.1 which has also been simplied
at some steps in comparison to its rst version [17]. Moreover, we note that
we give precise estimates of the solutions near
2
/4, which is the largest
Resolvent estimates for the Stokes operator 1083
value in the spectrum of the Laplace and the Stokes operator on . In
particular, Theorem 1.1 implies that the Stokes operator A
p
generates an
analytic semigroup e
tAp
, t 0, satisfying
|e
tAp
u
0
|
r
C
p,r
t

n1
2
(
1
p

1
r
)
e

2
4
t
|u
0
|
p
, t 0, u
0
L
p,
(), (1.4)
where 1 < p r < and L
p,
() = u C

0
()
n
: div u = 0
|.|p
. For
r = p, the latter estimate is a consequence of the well-known characterization
of analytic semigroups and their generators. For r > p, (1.4) is obtained
by estimating e
tAp
in a straightforward manner and using the semigroup
property.
The proof of Theorem 1.1 is based on (partial) Fourier transformations
in the tangential coordinates x
t
= (x
1
, . . . , x
n1
). By this (1.1)-(1.3) is
transformed to an ordinary two-point boundary-value problem depending on
(,
t
),
t
R
n1
, which can be solved explicitly. Then we use the Mikhlin
and Lizorkin multiplier theorems to estimate the solution operators. More
precisely, we rst estimate q and then use the corresponding estimates for
the Laplace resolvent equation to get the estimates for u.
Besides the study of the Stokes and Laplace resolvent equations, we give
an explicit formula for the Helmholtz decomposition of L
q
(), 1 < q < ,
which was proved by Miyakawa [13] and by Farwig [9] in a more general
context.
The structure of the article is as follows:
In Section 2 we study some Mikhlin multipliers, which will appear in
the solution operator of the Stokes and Laplace resolvent equations, and
estimate the corresponding operators. Then the explicit formula for the
Helmholtz decomposition in an innite layer is given in Section 3. In Section
4 we estimate the solutions of the Laplace resolvent equation, which can
easily be calculated by the same technique of partial Fourier transformations.
Finally, in Section 5 we derive an explicit formula for the pressure q in (1.1)-
(1.3). Then the main part of the proof consists in a careful analysis of the
solution operator for q. Once the pressure is estimated, the estimates of u
are obtained using the estimates for the Laplace resolvent equation.
2. Partial Fourier transformation and multiplier estimates
First let us introduce some notation. Let R
n
, n 2, be a domain.
Then C

0
() will denote the space of all smooth functions f : C with
compact support supp f . Moreover,
C

(0)
() := u = v[

: v C

0
(R
n
).
1084 Helmut Abels and Michael Wiegner
Finally, we note that by [4, Lemma 2.4] C

(0)
() is dense in

W
1
p
() for every
1 < p < .
The special type of domain = R
n1
(1, 1), we are dealing with,
suggests the use of a partial Fourier transformation. Denoting points of
by x = (x
t
, x
n
) with x
t
R
n1
, x
n
[1, 1], we dene for u: C
U(
t
, x
n
) := T
x

[u](
t
, x
n
) =
1
(2)
n1
2
_
R
n1
e
ix

u(x
t
, x
n
)dx
t
and the inverse transformation
u(x
t
, x
n
) := T
1

[U](x
t
, x
n
) =
1
(2)
n1
2
_
R
n1
e
i

U(
t
, x
n
)d
t
.
In the following we will use the abbreviation T T
x

. The Fourier and

inverse Fourier transformation of f : R
n
C with respect to x R
n
will be
denoted by

f and

f, respectively.
Throughout this section L will denote the smallest integer larger than
n
2
, where n N. Basic for deriving L
p
estimates is the Mikhlin multiplier
theorem, cf. e.g. [8]:
Theorem 2.1. Let f L
p
(R
n
) with 1 < p < and let m: R
n
0 C
denote an L-times dierentiable function with
[m] := sup
,=0,[[L
[[
[[
[D

m()[ < .
Then (m

f)

L
p
(R
n
) with
|(m

f)

|
p
C
p
[m]|f|
p
.
It is well known that m() :=

j
[[
, j = 1, . . . , n, satises the assumptions
of the latter theorem. This is a consequence of the fact that m() is homo-
geneous of degree 0.
Moreover, we will use the following variant of Theorem 2.1 due to Li-
zorkin [12].
Theorem 2.2. Let m: R
n
C be continuous with continuous derivatives

m() on R
n
:
i
,= 0, i = 1, . . . , n, 0, 1
n
, and let 1 < p q <
. Moreover, suppose that
sup

i
,=0,i=1,...,n

1
+
1

n+
n

m()

M
for all 0, 1
n
and =
1
p

1
q
. Then |(m

f)

|
q
C
p,q
M|f|
p
.
Resolvent estimates for the Stokes operator 1085
In the following, we have to calculate estimates for various multipliers.
For convenience, let us dene for 0
1
<
2

[m]

1
:= sup

1
<[[
2
,[[L
[[
[[
[D

m()[
and [m] := [m]

0
.
Then we have the following lemma, the proof of which is elementary.
Lemma 2.3. There holds uniformly in 0
1
<
2
:
[m
1
m
2
]

1
c[m
1
]

1
[m
2
]

1
. (2.1)
If [m[ c
1
> 0, then
[1/m]

1
C(1 + [m]

1
)
L
. (2.2)
If m() = M([[), then
[m]

1
c sup
a[
1
,
2
]
_
L

k=0
[a
k
M
(k)
(a)[
_
. (2.3)
As a typical multiplier, we come across functions of the following type:
Lemma 2.4. Let f, g : [0, ) C be L-times dierentiable with
(i) Ref(a) c
1
Aa
(ii) sup
1lL
a
l1
[f
(l)
(a)[ c
2
A
(iii) sup
0lL
a
l
[g
(l)
(a)[ c
3
a
k
on 0
1
a
2
with some c
i
> 0, A > 0, k N
0
. Let h() =
g([[)e
f([[)
. Then
[h]

1
cA
k
. (2.4)
Proof. As h
(j)
(a), j = 1, . . . , L, is a sum of terms of the form
g
(l
0
)
(a)f
(l
1
)
(a) . . . f
(lm)
(a)e
f(a)
,
where l
0
+. . . +l
m
= j with l
1
, . . . , l
m
,= 0, and
sup

1
a
2
a
j

g
(l
0
)
(a)f
(l
1
)
(a) . . . f
(lm)
(a)e
f(a)

C
1
a
k
a
m
A
m
e
c
1
Aa
C
2
A
k
the statement is a consequence of (2.3).
The next lemma provides L
p
estimates for a typical operator.
1086 Helmut Abels and Michael Wiegner
Lemma 2.5. Let (t) =
[0,)
(t) or (t) = 1
[0,)
(t). Dene the operator
G by
(Gf)(x
t
, x
n
) = T
1
_
_
1
1
[
t
[e
[

[[xnz[
(x
n
z)T[f](
t
, z)dz
_
for f C

0
(). Then G extends to a bounded linear operator G: L
p
()
L
p
() for every 1 < p < .
Proof. Let f
0
denote the extension of f onto R
n
by zero and let g := Gf.
Then
T[g](
t
, x
n
) =
_
R
[
t
[e
[

[[xnz[
(x
n
z)T[f
0
](
t
, z)dz.
The right-hand side makes sense also for [x
n
[ > 1. Hence the inverse trans-
formation denes a function g on R
n
with g[

= g,
g(x
t
, x
n
) :=
1
(2)
n1
2
_
R
n
e
ix

[
t
[e
[

[[xnz[
(x
n
z)T[f
0
](
t
, z)d(z,
t
).
The claim follows if we can prove | g|
Lp(R
n
)
c|f
0
|
Lp(R
n
)
. Inserting the
denition of T[f
0
], we may write g as an R
n
convolution g = k f
0
or
equivalently g = (2)
n/2
(

k

f
0
)

with
k(x
t
, x
n
) =
1
(2)
n1
2
_
R
n1
e
ix

(x
n
)[
t
[e
[

[[xn[
d
t
.
Now k(x) = T

x
[h](x) with h(
t
, x
n
) = (x
n
)[
t
[e
[

[[xn[
hence

k() =
1
(2)
1
2
_
R
e
inxn
(x
n
)[
t
[e
[

[[xn[
dx
n
for = (
t
,
n
). As by simple integration
(2)
1/2

k() =
_
[
t
[([
t
[ +i
n
)
1
for (t) =
[0,)
(t)
[
t
[([
t
[ i
n
)
1
for (t) = 1
[0,)
(t)
we have [

k()] c, and the claim follows by Theorem 2.1.

Lemma 2.6. Let
h(t) =
_
1
1
s(x)
(2 t x)
dx, s L
p
(1, 1),
for t (1, 1). Then |h|
Lp(1,1)
C
p
|s|
Lp(1,1)
for every 1 < p < .
Resolvent estimates for the Stokes operator 1087
Proof. Extend s by zero to s and let

h(t) := h(2 t). Then

h(t) =
_

s(x)
tx
dx and
_
1
1
[h(t)[
p
dt
_
R
[

h(t)[
p
dt C
p
| s|
p
p
= C
p
|s|
p
p
by the Calderon-Zygmond estimate.
The following lemma is (partly) known; we include a proof, as it contains
a surprising borderline case.
Lemma 2.7. Suppose q is smooth, q L
p
(R
n
), n 1, 1 < p < . Then
there is a constant c
q
such that for R 1
|q c
q
|
Lp(B
R
)
c(R)|q|
p
with
c(R) =
_
c
p
R if p ,= n,
cR(lnR + 1)
11/n
if p = n.
Remark. The example q(x) = ln([x[
2
+ 1)

with 0 < < 1

1
n
shows
that the case p = n is indeed exceptional an estimate with some smaller
exponent for the logarithmic term does not hold for all R 1.
Proof. First we consider the case n 2. Let q

denote the mean value of q

over B

. Then, as
q(rRw) q(rw) =
_
rR
r
w q(sw)ds,
one gets
q
R
q

= c
n
_
1
0
r
n1
_
_
[w[=1
_
rR
r
w q(sw)dsdw
_
dr.
Hence
[q
R
q

[ c
n
_
1
0
r
n1
_
B
Rr
\Br
[q(x)[
[x[
n1
dxdr
c
n
_
1
0
r
n1
|q|
p

_
_
B
Rr
\Br
dx
[x[
(n1)p

_
1/p

dr.
As
_
_
Rr
r
s
(n1)(1p

)
ds
_
1/p

C
_
_
_
r
1n/p

R
pn
p1

pn
p1

11/p
if p ,= n,
ln(
R

)
1
1
p
if p = n
1088 Helmut Abels and Michael Wiegner
and
_
1
0
r
n1
r
1
n
p
dr =
_
1
0
r
n/p

dr c, one gets
[q
R
q

[ c|q|
p

_
_
_

R
pn
p1

pn
p1

1
1
p
if p ,= n,
ln(
R

)
1
1
p
if p = n.
Thus, for p > n, q

q(0) for 0, while for p < n, q

for ,
and we may take c
q
= q(0), c
q
= q

, respectively, to get [q
R
c
q
[ c|q|
p

R
1
n
p
. For p = n, take c
q
= q
1
. Together with Poincares inequality, the
claim follows.
Finally, if n = 1, we have [q(x) q(0)[ [x[
1
p

|
d
dx
q|
p
, which implies the
statement.
Other multipliers will involve a certain square root of + [
t
[
2
, where
S

, > 0. Besides the sector S

, it is useful to consider

:= z C 0 : [ arg z[ <
for (0, ). If

, (0, ), b 0, and
2
is dened by

2
= +b, we have
Re c

max[[
1
2
, b
1
2
for all

. (2.5)
The latter estimate can easily be proved by using the homogeneity of f(, b)
:= ( +b)
1
2
.
Finally, we recall that for every , N
n
0

+[[
2

C
,,
([[
1
2
+[[)
2+[[[[
(2.6)
for

with (0, ). The latter statement can be proved by using the

homogeneity of f

(, ) := D

e
i

2
+[[
2
, (0, ).
For S

, the following lemma provides some important estimates.

Lemma 2.8. Let S

, > 0. Then for a R dene = (a) uniquely

by
2
:= + a
2
with Re 0 and Im 0 if Re = 0. Then, if
max[[
1/2
, a

2
, > 1,
Re 2 max[[
1/2
, a [[ (2.7)
where = (, ) > 0. Furthermore, for any k N we have
sup
a0
a
k1
[
(k)
(a)[ C
,,k
Resolvent estimates for the Stokes operator 1089
uniformly in S

with [[
1
2

2
, > 1. Moreover, if a = [
t
[,
t
R
n1
,
then sup

R
n1
,[[k,,=0
[
t
[
[[1
[D

[ C
,,k
uniformly in S

with
[[
1
2

2
, > 1.
Proof. First of all, the estimate [[ 2 max[[
1/2
, a is trivial. Suppose
rst that max[[
1
2
, a = [[
1
2

2
with > 1. Then obviously

for some = (, ). Hence (2.7) is a consequence of (2.5) with b = a

2
.
Similarly, if max[[
1
2
, a = a

2
with > 1, (2.5) implies
[ Re [ c

max
_

+

2
4

1
2
,
_
a
2

2
4
_1
2
_
c

_
1
1

2
_1
2
a.
This proves (2.7).
The estimate for
(k)
(a) follows from the fact that
(k)
(a) = p
k
(
a

)a
1k
,
k 1, where p
k
is a polynomial. Then the last statement is a consequence of
the chain rule and the fact that

[
t
[

[
t
[
1[[
, which is a consequence
of the homogeneity of the mapping
t
[
t
[.
Lemma 2.9. For S

, > 0,
t
R
n1
, and a := [[ let be as
in Lemma 2.8. Moreover, let k N
0
, z, w 0 with z + w > 0, and let
M = max
[[
1
2
4
, . Then
_

a
_

M
C

,
_

+a
_

0
C

if [[
1
2
,
_

a
_1
2
[[
1
2
0
C

,
_
a
k
e
azw
_

(z +w)
k
if [[
1
2
,
_
a
k
e
azw
_

(z +w)
k
,
uniformly in S

.
Proof. If a max
[[
1
2
4
, , then [[ 5a and Lemma 2.8 yields
_

M
C

.
Moreover, because of Lemma 2.8 and
a

=
d
da
(a), [
a

0
C

uniformly in
S

with [[
1
2
. Hence

+a
= (1 +
a

)
1
, Re(1 +
a

) 1, and (2.2)
imply the statement for

+a
. Furthermore, if a
1
2
[[
1
2
, [[

3
2
[[
1
2
and
therefore [1
a

[ 1
1

3
, which implies [

a
]
1
2
[[
1
2
0
C

.
In order to prove
_
a
k
e
azw

0

C
(z+w)
k
if [[ and
_
a
k
e
azw

C
(z+w)
k
, we apply Lemma 2.4 with g(a) = a
k
, f(a) = az w, and A = z +
w. Here the assumptions of Lemma 2.4 are consequences of Lemma 2.8.
1090 Helmut Abels and Michael Wiegner
Lemma 2.10. Let

:= (e
a
e
(a)
)/( a) and let m

:= (
+

)
1
e
a+
, where and a are the same as in Lemma 2.8. Then there is a
constant a

such that
_
a
1
m

M
C

(2.8)
uniformly in S

, > 0, where M := maxa

,
1
4
[[
1
2
, and
_

1
m
+

2M
0
C

,
_

1
a
1 +a
m

_
2M
0
C

uniformly in S

, > 0, with
1
4
[[
1
2
a

.
Proof. First we consider (2.8). We will use (2.2), where we note that
m
1

=
1
+a
(1 e
2a2
)

e
(a+)
,

=
_
1
1
e
t(a)
dt.
By Lemma 2.4 with k = 1 and A = min
[t[1
(1t)2+(1+t) = min4, 2,
we have [a

e
a
]

M
C

. Because of Lemma 2.9, this implies [am

1

M

C

.
As [[ 5a if a M, we have [
a
a+
[
a
a+[[

1
6
and
[

[ =

_
1
1
e
t(a)
dt

2 maxe
Rea
, e
aRe
.
Hence using Lemma 2.8 again
[am
1

[
1
6
(1 e
2a2Re
) 2a maxe
2a
, e
2Re

1
8
for a a

and a

large enough (this is one of the conditions for a

).
Concerning the other estimates, we use (2.2) and prove for m

= (
+

)e
(a+)
that [ m
+
[ c on [0, 2M], [ m

[ c on [1, 2M], [a
1
m

[ c on
[0, 1] and that [ m
+
]
2M
0
, [ m

]
2M
1
, [a
1
m

]
1
0
are uniformly bounded in S

,
[[ a

. As m

=

+a
(1 e
2a2
)

a
(e
2a
e
2
), the estimates for
[ m
+
]
2M
0
and [ m

]
2M
1
follow by Lemma 2.9. Moreover, using
m

_
(1 e
2a
)(1 +e
2
) a(1 e
2
)(1 +e
2a
)
_
,
we conclude [a
1
m

]
1
0
C

uniformly in [[ a

, S

.
Last we need estimates from below. We remember that Re 2[[
1/2

8a

, c
1
[[ [ a[ c
2
[[, and Re [[. Hence
[ m
+
[ =

+a

1 e
2a2
+
+a
a
(e
2a
e
2
)

Resolvent estimates for the Stokes operator 1091

c
1
1
_
(1 +e
2a
) e
2Re

2a
[ a[
(e
2a
+e
2Re
) e
2Re
_
c
1
1
_
(1 +e
2a
)
c
[[
2e
2Re
_

1
2
c
1
1
for a

large enough (the second condition on a

).
Similarly,
[ m

+a

1 e
2a

2a
a
e
2a

e
2Re

e
2a
1
2a
a

_
and for a 1, we have
[ m

[ c
1
1
__
1 e
2

c
[[
_
ce
2Re
_
c
1
1
_
1
2
e
2
_
for a

suitably large (the third condition on a

), while for a 1
[a
1
m

[ c
1
1
__
1 e
2a
a
_

2e
2a
[ a[
e
2Re
_
1 e
2a
a
+
2
[ a[
__
c
1
1
_
2e
2

c
[[
ce
2Re
_
c
1
1
e
2
for a

large (the last condition on a

). This nishes the proof.

Lemma 2.11. Let f(a, ) := D/(a). Then f is a smooth function in
a R, C and f(a, ) ,= 0 if a 0 and Re 0 with
2
S

, > 0. In
particular, if a = a(
t
) and = (
t
, ) are dened as in Lemma 2.8, then
for every A, B > 0
[a/D]
A
0
c(A, B, )
uniformly in S

with [[ B.
Proof. First of all, since

=
_
1
1
e
t(a)
dt,

+
+

= 4
_
1
0
cosh(t) cosh(ta) dt = 4P
1
(a, )

a
= 4
_
1
0
sinh(t)

sinh(ta)
a
dt = 4P
2
(a, ).
Hence, f(a, ) = D/(a) = 16P
1
(a, )P
2
(a, ) is a smooth function for all
a R, C. Therefore, it remains to prove that f(a, ) ,= 0 if a 0,
Re 0, and
2
S

.
1092 Helmut Abels and Michael Wiegner
Firstly, let = x R. Then obviously P
1
(a, ), P
2
(a, ) > 0 and therefore
f(a, ) > 0. Moreover, if = iy, y R, then
P
1
(a, iy) =
_
1
0
cos ty cosh ta dt > 0, P
2
(a, iy) =
_
1
0
sin ty
ty
sinh ta
ta
t
2
dt > 0,
where we have used that
2
= y
2
S

implies [y[

2
.
Finally, let = x +iy, y ,= 0, x > 0. Using

+
+

4
= sinh cosh a a cosh sinh a
=
_

sinh
cosh
a
sinh a
cosh a
_
cosh cosh a,

4
= cosh sinh a a sinh cosh a
= a
_
sinh a
a cosh a

sinh
cosh
_
cosh cosh a,
where =
2
a
2
,= 0 if Im = y ,= 0 and cosh ,= 0 if Re > 0, it is
sucient to show that
Im
_

sinh
cosh
_
,= 0 and Im
_
sinh
cosh
_
,= 0.
By elementary calculations
Im
_

sinh
cosh
_
=
1
2
xsin 2y +y sinh 2x
cosh
2
xcos
2
y + sinh
2
xsin
2
y
Im
_
sinh
cosh
_
=
1
2[[
2
xsin 2y y sinh 2x
cosh
2
xcos
2
y + sinh
2
xsin
2
y
,
where [y sinh 2x xsin 2y[ [y[(sinh 2x 2x) > 0 if y ,= 0, x > 0.
The last statement is a trivial consequence of the rst part and (2.3).
Lemma 2.12. Let k(
t
, w, x
n
) be measurable and suciently smooth in
t

R
n1
0. Moreover, suppose that
[k(
t
, w, x
n
)] +
_

w
k(
t
, w, x
n
)[
t
[
1
_
c(2 w x
n
)
1
and dene
h(
t
, x
n
) = T
1
_
_
1
1
k(
t
, w, x
n
)[
t
[
1
T[g](
t
, w)dw
_
Resolvent estimates for the Stokes operator 1093
for g C

(0)
()

W
1
p
(). Then for every 1 < p <
|h|
Lp()
C
p
|g|

W
1
p
()
for all g C

(0)
()

W
1
p
().
Proof. Let C

0
() and dene by
T[](
t
, w) :=
_
1
1
k(
t
, w, z)[
t
[
1
T[](
t
, z)dz.
Then
g, ) =
_
1
1
_
1
1
_
R
n1
k(
t
, w, z)[
t
[
1
T[g](
t
, w)T[](
t
, z)d
t
dzdw
=
_
1
1
_
R
n1
T[h](
t
, z)T[](
t
, z)d
t
dz = h, )
and therefore [h, )[ |g|

W
1
p
()
||
p
.
Now T[

x
j
](
t
, w) =
_
1
1
k(
t
, w, z)i
j
[
t
[
1
T[](
t
, z)dz, for j = 1, . . . , n
1, which implies
_
_
_
_

x
j
(, w)
_
_
_
_
Lp(R
n1
)
C
p
_
1
1
|(, z)|
p
(2 w z)
1
dz
by Theorem 2.1 and the same estimate follows for j = n. By Lemma 2.6,
||
p
C
p
||
p
. Hence the claim is proved.
3. The Helmholtz Projection
Let L
p,
() = u C

0
()
n
: div u = 0
|.|p
. We want to give an explicit
construction for the Helmholtz projection P : L
p
()
n
L
p,
().
Theorem 3.1. For every f L
p
()
n
there is a unique decomposition f =
f
0
+ q with f
0
L
p,
(), q

W
1
p
= q L
p,loc
() : q L
p
()
n
.
Moreover,
|f
0
|
p
+|q|
p
C
p
|f|
p
(3.1)
and |q|
Lp(B
R
)
C
p
|f|
p
+C
p
R|(f
1
, . . . , f
n1
)|
p
for p ,= n1 if we choose
an appropriate constant for q.
Remark 3.2. The theorem (except for the L
p
(B
R
) estimate) was proved
by Miyakawa [13] and Farwig [9] in a more general context. Since we will
need the explicit formula for f
0
= Pf, we include a proof.
1094 Helmut Abels and Michael Wiegner
Proof. We may assume f C

0
()
n
, so that all integrals appearing are
well dened.
Dene F
j
(
t
, x
n
) := T[f
j
](
t
, x
n
), j = 1, . . . , n and F
n+1
(
t
, x
n
) :=
i

[

F
t
(
t
, x
n
), where F
t
= (F
1
, . . . , F
n1
). Note that
|T
1
[F
n+1
](, x
n
)|
Lp(R
n1
)
C
p
|f
t
(, x
n
)|
Lp(R
n1
)
.
Now dene a pressure q by
Q(
t
, x
n
) := T[q](
t
, x
n
)
=
1
2
_
1
1
e
a[xnz[
F
n+1
(
t
, z)dz
1
2
_
1
1
e
a[xnz[
sign(z x
n
)F
n
(
t
, z)dz
+e
axn
+e
axn
where a = [
t
[ and , are chosen such that
n
QF
n
[
xn=1
0.
Since (
n
a
2
)Q = i
t
F
t
+
n
F
n
, the Helmholtz projection f
0
= Pf is
then given by
F
0
j
= T[f
0
j
] := F
j
i
j
Q, j = 1, . . . , n 1,
F
0
n
= T[f
0
n
] := F
n

n
Q
(3.2)
But it remains to determine , . By dierentiation

n
QF
n
=
a
2
_
1
1
e
a[xnz[
sign(z w)F
n+1
(
t
, z)dz

a
2
_
1
1
e
a[xnz[
F
n
(
t
, z)dz +ae
axn
ae
axn
and
n
(
n
QF
n
) = a
2
Q+aF
n+1
. From the boundary conditions we conclude
ae
axn
=
a
2(1 e
4a
)
_
_
1
1
e
a(2xnz)
(F
n
(
t
, z) F
n+1
(
t
, z))dz

_
1
1
e
a(4+zxn)
(F
n
(
t
, z) +F
n+1
(
t
, z))dz
_
and
ae
axn
=
a
2(1 e
4a
)
_
_
1
1
e
a(4z+xn)
(F
n
(
t
, z) F
n+1
(
t
, z))dz

_
1
1
e
a(2+xn+z)
(F
n
(
t
, z) +F
n+1
(
t
, z))dz
_
.
Resolvent estimates for the Stokes operator 1095
These terms are of the type
R(
t
, x
n
) =
_
1
1
k(
t
, z, x
n
)T[g](
t
, z)dz.
Hence, P is a linear operation, and
|T
1
[R](, x
n
)|
Lp(R
n1
)
C
p
_
1
1
[k(, z, x
n
)]|g(, z)|
Lp(R
n1
)
dz.
As [z[, [x
n
[ 1 and the kernels are real analytic in a, we have [k(, z, x
n
)]
1
0

c. Furthermore, for b(a) = (1 e
4a
)
1
, we have [b]

1
c. Together with
Lemma 2.9, we obtain [k(, z, x
n
)] c((2 +z +x
n
)
1
+(2 z x
n
)
1
), and
by Lemma 2.6 |T
1
[R]|
p
C
p
|f|
p
. For the rst two terms of
n
Q F
n
,
we use Lemma 2.5 to conclude nally
_
_
_
q
x
n
f
n
_
_
_
p
C
p
|f|
p
.
For later purposes, let us note that
2a
1
F
0
n
(
t
, x
n
) =
_
1
1
k
n
(a, x
n
, z)F
n
(
t
, z)dz (3.3)
+
_
1
1
k
n+1
(a, x
n
, z)F
n+1
(
t
, z)dz
with
k
n
(a, w, z) = e
a[wz[
+ (1 e
4a
)
1

_
e
a(4z+w)
e
a(2+w+z)
e
a(2wz)
+e
a(4+zw)
_
and
k
n+1
(a, w, z) = e
a[wz[
sign(z w) (1 e
4a
)
1

_
e
a(4z+w)
+e
a(2+w+z)
e
a(2wz)
e
a(4+zw)
_
.
Note that these kernels are analytic in a (uniformly for [z[, [w[ 1).
Now for j = 1, . . . , n 1,
i
j
Q(
t
, x
n
) =
a
2
_
1
1
e
a[xnz[
i
j
[
t
[
F
n+1
(
t
, z)dz
+
a
2
_
1
1
e
a[xnz[
sign(z x
n
)
i
j
[
t
[
F
n
(
t
, z)dz ae
axn
i
j
[
t
[
ae
axn
i
j
[
t
[
.
1096 Helmut Abels and Michael Wiegner
This is obviously a sum of terms of the same type as in
n
Q F
n
with
F
n
, F
n+1
replaced by
i
j
[

[
F
n
,

j
[

[
F
n+1
, respectively. Hence also |
q
x
j
|
p

C
p
|f|
p
, j = 1, . . . , n 1. Thus P : L
p
()
n
L
p
()
n
is continuous. For the
uniqueness of the decomposition, we refer to [13].
Because of the construction, div f
0
= 0, f
0
n
[

= 0, and P
2
f = Pf.
Therefore P extends to a continuous projection P : L
p
()
n
L
p,
() such
that Pf = f q.
It remains to prove a local estimate for q. Take a cuto function with
(t) = 1 for t 1 and 0 for t 2. Let
Q = (a)Q+
_
1 (a)
a
_
Qa = Q
1
+Q
2
.
Now (1 (a))/a is a multiplier on R
n1
with [(1 (a))/a] c, hence
|T
1
[Q
2
]|
p
C
p
|T
1
[Qa]|
p
C
p
|f|
p
,
as shown above. Next Q
1
is of the type
Q
1
(
t
, x
n
) =
n

j=1
_
1
1
(a)k
j
(a, x
n
, z)F
j
(z)dz
(a)
2a
_
1
1
F
n+1
(z)dz
with [()k
j
(, x
n
, z)] c uniformly. Inverting,
q = q
0
+
_
1
1
n1

j=1
T
1
_
(a)
a
2

j
_
f
j
(, z)dz
with |q
0
|
p
c|f|
p
, |q
0
|
p
c|f|
p
. As T
1
[
(a)
a
2

j
] is smooth and bounded
if n 3, the convolution is well dened for f
j
C

0
in that case. If n = 2,
q(x) is well dened since it is determined up to a constant by
2
q(x) L
q
().
Hence q = q
0
+ q
1
(x
t
) with |
x
q
1
|
L
p
(R
n1
)
C|f
t
|
p
. The local estimates
follow now by Lemma 2.7.
It is well known that the existence of the Helmholtz decomposition of
L
p
(), 1 < p < , is equivalent to the unique solvability of the weak
Neumann problem for the Laplace equation; i.e., for every f (

W
1
p

())
t
there is a unique u

W
1
p
() such that
u, v) = f, v) for all v

W
1
p
(), (3.4)
cf. e.g. [16]. Moreover, |u|
p
C
p
|f|
(

W
1
p

())

. This will be used in the

proof of the following lemma.
Resolvent estimates for the Stokes operator 1097
Lemma 3.3. Let 1 < p < . Then C

(0)
()

W
1
p
() is dense in W
1
p
()

W
1
p
().
Proof. Let g W
1
p
()

W
1
p
() be arbitrary. Since g (

W
1
p

())
t
and by
(3.4), there is a u L
p
() such that
g, v) = u, v), for all v

W
1
p
(). (3.5)
Moreover, since even g W
1
p
(), approximation of
j
u,
i

j
u, i, j = 1, . . . ,
n 1, by dierence quotients yields that
j
u,
i

j
u L
p
(). Using (3.5)
we obtain also
n
u,
n
u L
p
() and therefore u W
2
p
(). Furthermore,
(3.5) yields u = g almost everywhere and
n
u[

= 0.
Now let C

0
(R
n1
) with (0) = 1 and set g
R
(x) = div((Rx
t
)u).
Then g
R
W
1
p
()

W
1
p
() and lim
R
g
R
= g in W
1
p
()

W
1
p
(). Since
g
R
is compactly supported, [4, Lemma 2.8] implies that
_
g
R
dx = 0 if 1 < p
n 1
n 2
.
Hence, if 1 < p
n1
n2
, we can nd g
k,R
C

(0)
() with
_
g
k,R
dx = 0 such
that lim
k
g
k,R
= g
R
in W
1
p
(). Since
_
g
R
dx =
_
g
k,R
dx = 0, this implies
that lim
k
g
k,R
= g
R
in

W
1
p
() by Poincares inequality. In the case
p >
n1
n2
, by [4, Lemma 2.8] every h L
p
() with support in B
R
(0) is
in

W
1
p
() and |h|

W
1
p
()
C
R
|h|
p
. Hence, if g
k,R
C

(0)
() such that
lim
k
g
k,R
= g
R
in W
1
p
(), also lim
k
g
k,R
= g
R
in

W
1
p
().
In particular, we have proved that C

(0)
()g :
_
gdx = 0 if 1 < p
n1
n2
and C

(0)
() if p >
n1
n2
are dense in W
1
p
()

W
1
p
().
4. Laplace Resolvent Equation
We consider
( )u = f in , (4.1)
u[

= 0 on . (4.2)
In order to prove the estimates of Theorem 1.1, we will use the corresponding
statements for the latter system.
Theorem 4.1. Let 1 < q r < , n 2, and let > 0. Then for every
C (,

2
4
] and f L
q
() there is a unique solution u W
2
q
() of
1098 Helmut Abels and Michael Wiegner
(4.1)-(4.2). Moreover,

+

2
4

_
|u|
q
+ (1 +[[)

1
2
|u|
q
+ (1 +[[)
1
|
2
u|
q
_
C

|f|
q
,

+

2
4

1
n1
2
_
1
q

1
r
_
(1 +[[)

1
2
_
1
q

1
r
_
|u|
r
C

|f|
q
uniformly in S

provided that
n1
2
(
1
q

1
r
) 1.
First we calculate the solution u for f C

0
() and estimate the solution
operator. Then the general case is obtained by continuous extension to
L
p
(). Using partial Fourier transformation, (4.1)-(4.2) reduces to
( +[
t
[
2

2
n
)U = F in (1, 1), (4.3)
U[
xn=1
= 0 (4.4)
for
t
R
n1
and S

, where U(
t
, x
n
) = T
x

[u(., x
n
)] and F(
t
, x
n
) =
T
x

[f(., x
n
)]. The solution of this boundary-value problem is given by
U(
t
, x
n
) =
_
1
1
k(, x
n
, z
n
)F(
t
, z
n
)dz
n
,
where
k(, x
n
, z
n
)
=
1
2(1 e
4
)
_
e
(2xnzn)
+e
(2+xn+zn)
e
[xnzn[
e
(4[xnzn[)
_
with as in Lemma 2.8. Since
2
= +[
t
[
2
= ()
2
, we obtain k(, x
n
, z
n
)
= k(, x
n
, z
n
). Moreover, it is easy to observe that k(, x
n
, z
n
) is holomor-
phic in C
il
2
: l Z, l ,= 0 and depends smoothly on x
n
, z
n
R.
Since k(, x
n
, z
n
) has a pole of rst order for = i

2
and k(, x
n
, z
n
) =
k(, x
n
, z
n
), we have
k(, x
n
, z
n
) =
k
t
(
2
, x
n
, z
n
)

2
+

2
4
for all such that
2
B

2(0),
where k
t
(z, x
n
, z
n
) is holomorphic in z B

2(0).
Lemma 4.2. Let = (,
t
) be as in Lemma 2.8 with [[ < . Moreover,
let > 0, [0, 1], and let (0, 1). Then
[

k(, x
n
, z
n
)[ C
,,

+

2
4

1+
[
t
[
2[[
Resolvent estimates for the Stokes operator 1099
for all S

and
t
R
n1
such that [[ and uniformly in x
n
, z
n

[1, 1].
Proof. First of all,

_
1

2
+

2
4
_
=
[[

k=[[/2|
p
k
(
t
)
( +[
t
[
2
+

2
4
)
1+k
where p
k
(
t
) is a homogeneous polynomial of degree 2k [[. Hence

_
1

2
+

2
4
_

[[

k=[[/2|
C
k
[
t
[
2k[[
[ +
2
/4 +[
t
[
2
[
1+k

[[

k=[[/2|
C
k,
[[
2k[[
([ +
2
/4[ +[
t
[
2
)
1+k

C
,
[ +
2
/4[
(1)
[
t
[
2[[
,
where we have used [

+ [
t
[
2
[ c

([

[ + [
t
[
2
) for all

and some =
() (0, ), which is a consequence of (2.6). Moreover, since k
t
(z, x
n
, z
n
)
is holomorphic in B

2(0) and smooth in x

n
, z
n
R, [

k
t
(
2
, x
n
, z
n
)[ C

for all S

and
t
R
n1
such that [[ . Hence the statement of the
lemma is a consequence of the product and chain rule.
Proof of Theorem 4.1. Let C

0
(R) be such that (s) = 1 for [s[
1
2

2
and supp B3
4

2
(0). Then U = U
1
+U
2
, where
U
j
(
t
, x
n
) =
_
1
1
k
j
(, x
n
, z
n
)F(
t
, z
n
)dz
n
, j = 1, 2,
k
1
(, x
n
, y
n
)=([
2
[)k(, x
n
, y
n
) and k
2
(, x
n
, y
n
)=(1([
2
[))k(, x
n
, y
n
).
Since [

([
2
[)[ C

, we have for [0, 1] by Lemma 4.2

[

k
1
(, x
n
, y
n
)[
C
,
[ +
2
/4[
(1)
[
t
[
2[[
uniformly in
t
R
n1
, S

, and x
n
, z
n
[1, 1]. Hence
_
_
_T
1

[U
1
(
t
, x
n
)]
_
_
_
r
C sup
xn[1,1]
|T
1

[k
1
(, ., z
n
)F(
t
, .)]|
r,
C
,q,r

+

2
4

1+
n1
2
_
1
q

1
r
_
|f|
q,
.
1100 Helmut Abels and Michael Wiegner
by Theorem 2.2, where we have chosen =
n1
2
(
1
q

1
r
). Moreover, if
N
n1
0
, [[ 2, we get by choosing = [[/2
_
_
_

T
1

[U
1
(
t
, x
n
)]
_
_
_
q
C sup
xn[1,1]
|T
1

[(
t
)

k
1
(, x
n
, .)F(
t
, .)]|
q
C
,q

+

2
4

1+
|f|
q
C
,q

+

2
4

1
(1 +[[)
||
2
|f|
q
since k
1
(, x
n
, z
n
) = 0 for [[
2
. Because of (4.3),
_
_
_
2
xn
T
1

[U
1
(
t
, x
n
)]
_
_
_
q
C

(1 +[[)

+

2
4

1
|f|
q
.
Using |
n
v|
q
C|v|
1
2
q
_
|v|
q
+|
2
n
v|
q
_1
2
for v W
2
q
(), we obtain the
corresponding estimate for
xn
T
1

[U
1
(
t
, x
n
)].
In order to estimate U
2
(
t
, x
n
), we use
T
1

_
(1 ([
2
[))
_
1
1
e
(2xnzn)
2(1 e
4
)
F(
t
, z
n
)dz
n
_
=
1
T
1
x
_
(1 ([
2
[))
(1 e
4
2
)( +[[
2
)
T
x
[
1
ef]
_
,
where (
y
f)(x) = f(x + y) and ef denotes the extension by 0 of f to R
n
.
Because of (2.6) and 1 ([
2
[) = 0 if [[
2

1
2

2
,
_

(1 ([
2
[))
(1 e
4
)( +[[
2
)
_
C

(1 +[[)
1+
||
2
for every [[ 2 uniformly in S

, > 0. Hence we obtain by Theorem 2.1

_
_
_D

x
T
1

_
(1 ([
2
[))
_
1
1
e
(2xnzn)
2(1 e
4
)
F(
t
, z
n
)dz
n
__
_
_
q
C
,q
(1 +[[)
1+
||
2
|f|
q
uniformly in S

for all [[ 2 and 1 < q < . In the same way, we

obtain by Theorem 2.2
_
_
_T
1

_
(1 ([
2
[))
_
1
1
e
(2xnzn)
2(1 e
4
)
F(
t
, z
n
)dz
n
__
_
_
r
C

(1 +[[)
1+
n
2
(
1
q

1
r
)
|f|
q
Resolvent estimates for the Stokes operator 1101
uniformly in S

for all 1 < q r < . Moreover,

e
[xnzn[
+e
(4[xnzn[)
2(1 e
4
)
=
e
[xnzn[
2
+
e
(4xn+zn)
+e
(4+xnzn)
2(1 e
4
)
,
where
T
1

_
(1 ([
2
[))
_
1
1
e
[xnzn[
2
F(
t
, z
n
)dz
n
_
= T
1
x
_
1 ([
2
[)
+[[
2
T
x
[ef]
_
,
and
T
1

_
(1 ([
2
[))
_
1
1
e
(4xnzn)
2(1 e
4
)
F(
t
, z
n
)dz
n
_
=
2
T
1
x
_
(1 ([
2
[))
(1 e
4
)( +[[
2
)
T
x
[
2
ef]
_
.
Hence the last terms can be estimated as before.
It remains to prove the uniqueness of the solution. Let
q
: T(
q
)
L
q
(), 1 < q < , with T(
q
) = u W
2
q
(): u[

= 0. Then
q
is surjective for every C (,
2
/4] by the rst part of the proof.
Moreover, since (
q
)
t

q
, the adjoint of
q
is surjective.
Hence
q
is injective for every C (,
2
/4], which nishes the
proof.
5. Proof of Theorem 1.1
We may start with two simplications. First we may restrict ourselves to
f C

0
() and g C

(0)
()

W
1
p
() by Lemma 3.3. Next we may use
the Helmholtz decomposition f = f
0
+q, where f
0
is dened explicitly via
its partial Fourier transformation F
0
, being well dened and smooth. The
pressure part may be absorbed on the left side. Hence we may assume for
simplicity that f = f
0
. Thus applying the partial Fourier transformation we
have to solve (with a = [
t
[)
( +a
2
)U
t

2
n
U
t
+i
t
Q = F
t
, (5.1)
( +a
2
)U
n

2
n
U
n
+
n
Q = F
n
, (5.2)

n
U
n
+i
t
U
t
= G, (5.3)
U[
xn=1
= 0, (5.4)
1102 Helmut Abels and Michael Wiegner
where U
t
= (U
1
, . . . , U
n1
) and F
t
= (F
1
, . . . , F
n1
). Dierentiating (5.2)
with respect to x
n
, multiplying (5.1) by i
t
and adding, we obtain
( +a
2
)G
2
n
Ga
2
Q+
2
n
Q = 0.
Hence,

2
n
(QG) a
2
(QG) = G. (5.5)
Thus, we dene
Q(
t
, x
n
) = G(
t
, x
n
)+

2a
_
1
1
e
a[xnw[
G(
t
, w)dw+
A
a
e
axn

B
a
e
axn
(5.6)
with parameters A, B to be chosen below. Setting
L
j
:= i
j
QF
j
, j = 1, . . . , n 1, L
n
:=
n
QF
n
(5.7)
we may solve

2
n
U
j
( +a
2
)U
j
= L
j
, U
j
= 0 on z = 1 for j = 1, . . . , n. (5.8)
It remains to satisfy (5.3). Let R := i
t
U
t
+
n
U
n
G. We have to show
that R 0. Now

n
R = i
t

n
U
t
+ (
2
n
U
n

n
Q) +
n
(QG)
= i
t

n
U
t
+ ( +a
2
)U
n
F
n
+
n
(QG) and

2
n
R = i
t

2
n
U
t
+ ( +a
2
)
n
U
n

n
F
n
+
2
n
(QG)
= (i
t
U
t
+
n
U
n
)( +a
2
) a
2
Q(i
t
F
t
+
n
F
n
) +
2
n
(QG)
= ( +a
2
)R
by (5.5) and div f = 0. Therefore, with as dened in Lemma 2.8,
R = R(
t
, x
n
) = a(
t
)e
xn
+b(
t
)e
xn
for ,= 0
resp. = a(
t
)x
n
+b(
t
) for = 0
Suppose we can choose A, B such that R(
t
, 1) = 0. Then R 0, as
e
4
1 ,= 0 due to + [
t
[
2
,=
k
2
4

2
(the case = 0 is trivial), hence the
claim follows.
A and B will of course depend on
t
and , and we have to calculate them
in order to estimate the solution. Skipping the
t
dependence, we get from
(5.2) and (5.4)

n
U
n
(z)e
z
[
1
1
=
_
1
1
(
2
n
U
n
(z)
2
U
n
(z))e
z
dz
=
_
1
1
(
n
Q(z) F
n
(z))e
z
dz
Resolvent estimates for the Stokes operator 1103
= Q(z)e
z

1
1

_
1
1
(Q(z) +F
n
(z))e
z
dz,
hence,
R(z)e
z
[
1
1
= (Q(z) G(z))e
z
[
1
1

_
1
1
(Q(z) +F
n
(z))e
z
dz.
Again due to e
4
1 ,= 0, the right-hand sides should vanish for + as well
as for . For the +-sign, this means
0 =

2a
_
_
1
1
e
a(1x)+
G(x)dx
_
1
1
e
a(1+x)
G(x)dx
_
+
A
a
(e
a+
e
(a+)
)

B
a
_
e
a
e
(a)
_

_
1
1
F
n
(z)e
z
dz
_
1
1
Q(z)e
z
dz.
Calculating the last term:
_
1
1
Q(z)e
z
dz =
_
1
1
G(z)e
z
dz +

2a
_
1
1
_
1
1
e
a[wz[
G(w)e
z
dwdz
+
A
a
_
1
1
e
(a+)z
dz
B
a
_
1
1
e
(a+)z
dz
and noting that
_
1
1
e
a[wz[
e
z
dz =
2a

e
w
+e
aw
e
a
a
e
aw
e
(+a)
+a
we get
_
1
1
Q(z)e
z
dz =

2a
_
1
1
_
e
aw
e
a
a
e
aw
e
(+a)
+a
_
G(w) dw
+
A
a
_
e
a+
a +

e
(a+)
a +
_

B
a
_
e
a
a

e
(a)
a
_
.
Inserting, we end with
A
+
+B

=
_
1
1
F
n
(
t
, x)e
x
dx +
1
2
_
1
1
G(
t
, x)k(
t
, x)dx,
where k(
t
, x) = e
ax e
a
a
+e
ax e
(+a)
+a
and

:= (e
a
e
(a)
)/( a).
Changing from to ,
+
goes over to

and vice versa and the second

condition is
A

+B
+
=
_
1
1
F
n
(
t
, x)e
x
dx
1
2
_
1
1
G(
t
, x)k(
t
, x)dx.
1104 Helmut Abels and Michael Wiegner
Finally,
A =
_
1
1
F
n
(
t
, x)H(, x)dx +
1
2
_
1
1
G(
t
, x)K(
t
, x)dx (5.9)
B =
_
1
1
F
n
(
t
, x)H(
t
, x)dx
1
2
_
1
1
G(
t
, x)K(
t
, x)dx (5.10)
with
H(
t
, x) = (
+
e
x

e
x
)/D
K(
t
, x) = (
+
k(
t
, x) +

k(
t
, x))/D.
Thus (5.6) - (5.10) gives the explicit solution of our problem.
Now we may start to estimate the quantities of (5.7) after inverting the
partial Fourier transformation. Obviously,
|T
1
[
n
G]|
p
+|T
1
[i
t
G]|
p
c|g|
p
.
Next, multiplying by i
j
and dierentiating with respect to x
n
, respectively,
we have to treat the following three types of integrals
h
1
(x
t
, x
n
) = T
1
_
_
1
1
e
[

[[wxn[
(w x
n
)G(
t
, w)dw
_
(5.11)
h
2
(x
t
, x
n
) = T
1
_
_
1
1
e
[

[xn
F
n
(
t
, w)H(
t
, w)dw
_
(5.12)
h
3
(x
t
, x
n
) = T
1
_
_
1
1
e
[

[xn
G(
t
, w)K(
t
, w)dw
_
(5.13)
with (t) = signt or (t) = 1. We start with
Lemma 5.1. Let h
1
be dened as in (5.11) and let 1 < p < . Then
|h
1
|
Lp()
C
p
|g|

W
1
p
for all g C

(0)
()

W
1
p
().
Proof. Let C

0
(). Dene by
T[](
t
, x
n
) =
_
1
1
e
[

[[xnz[
(x
n
z)T[](
t
, z)dz.
Then
_

g(x
t
, x
n
)(x
t
, x
n
)d(x
t
, x
n
)
=
_
R
n1
_
1
1
_
1
1
G(
t
, x
n
)e
[

[[xnz[
(x
n
z)T[](
t
, z)dx
n
dzd
t
Resolvent estimates for the Stokes operator 1105
=
_
R
n1
_
1
1
T[h
1
](
t
, z)T[](
t
, z)dzd
t
=
_

h
1
(x
t
, x
n
)(x
t
, x
n
)dx.
As for j = 1, . . . , n 1
T
_

x
j
_
(
t
, x
n
) =
_
1
1
e
[

[[xnz[
[
t
[(x
n
z)
i
j
[
t
[
T[](
t
, z)dz,
T
_

x
n
_
(
t
, x
n
) =
_
1
1
e
[

[[xnz[
[
t
[sign(z x
n
)(w x
n
)T[](
t
, z)dz
_
+2T[](
t
, x
n
) in the case of (t) = signt
_
,
Lemma 2.5 implies that ||
p
C
p
||
p
. Therefore,

h
1
dx

C
p
|g|

W
1
p
||
p
,
which implies the claim.
The most important estimates are contained in the following lemma.
Lemma 5.2. Let h
2
be dened as in (5.12) and let 1 < p < . Then
|h
2
|
p
C
p
|f|
p
for all f C

0
().
Proof. Here we need a little trick in splitting the phase space into two parts,
depending on the size of [[. Therefore let M := max
1
4
[[
1/2
, a

with
a

as in Lemma 2.10. Moreover, we choose a cuto function

M
(a),
which vanishes outside [0, 2M] and equals 1 on [0, M]. We may assume that
0
M
(a) 1 and [
M
] C, independent of M. Now we split h
2
= h
1
2
+h
2
2
with
h
j
2
(x
t
, x
n
) = T
1
_
_
1
1
e
axn
F
n
(
t
, w)H
j
(, w)dw
_
, j = 1, 2,
where H
1
(
t
, w) = (1
M
(a))H(
t
, w) and H
2
(
t
, w) =
M
(a)H(
t
, w). If
we now set
H
3
(
t
, w) =
M
(a)a(D)
1
(
+
e
w
+

e
w
),
then
H
3
w
(
t
, w) = H
2
(
t
, w)a.
Since F
n
(
t
, 1) = 0 and
n
F
n
= i
t
F
t
, one gets
h
2
2
(x
t
, x
n
) = T
1
_
_
1
1
e
axn
i
t
[
t
[
F
t
(
t
, w)H
3
(
t
, w)dw
_
.
1106 Helmut Abels and Michael Wiegner
Thus by Theorem 2.1
|h
2
(, x
n
)|
Lp(R
n1
)
C
p
_
1
1
[e
axn
H
1
(
t
, w)]|f
n
(, w)|
Lp(R
n1
)
dw
+C
p
_
1
1
[e
axn
H
3
(
t
, w)]|f
t
(, w)|
Lp(R
n1
)
dw.
By (2.1) and the denition of
[e
az
H
1
(
t
, w)] C

+,
[D
1
e
az

e
w
]

M
and (5.14)
[e
az
H
3
(
t
, w)] C

+,
[D
1
(a
1
)e
az

e
w
]
2M
0
. (5.15)
We will show that the right-hand sides are bounded by C(2 z w)
1
+
C(2 z + w)
1
. Then the statements of the lemma are a consequence of
Lemma 2.6.
In order to prove (5.14), we write the multipliers as a product m
1
m
2
with
m
1
= ae
a(z1)+(w1)
and m
2
= (e
a+

)(aD)
1
.
Since a M = max
1
4
[[
1/2
, a

, we have [m
1
]

M
C

(2 z w)
1
by
Lemma 2.9. Instead of estimating m
2
, we may as well estimate
m

2
= e
a+
(
+

)(aD)
1
= a
1
(
+

)
1
e
a+
as D =
2
+

, which was done in Lemma 2.10.

For the proof of (5.15), suppose rst that
1
4
[[
1/2
a

, which implies
a
1
2
[[
1/2
. Similarly as before, the necessary estimates follow from the
estimates of
_
a

e
az+w
(
+

)
_
2M
0
(and by symmetry also for w), as D =
2
+

.
For the case of the +-sign, we factor into
ae
a(z1)+(w1)
and
1
(
+
+

)
1
e
a+
.
For the --sign, we take the factors
(1 +a)e
a(z1)+(w1)
and
1
a
1 +a
(
+

)
1
e
a+
on [0, 2M]. All these terms were estimated in Lemma 2.9 and Lemma 2.10.
For the second part we have to consider the other possibility [[
1/2
/4
a

= M, where now a

is xed. Then we do not change the original form of

Resolvent estimates for the Stokes operator 1107
h
2
2
, but still consider
h
2
2
(x
t
, x
n
) = T
1
_
_
1
1
e
axn
F
n
(
t
, w)
M
(a)H(
t
, w)dw
_
.
The crucial observation now is that, because of (3.3) and F
n
= F
0
n
, F
n
(
t
, w)
contains factor a = [
t
[! Hence we estimate
|h
2
2
(, x
n
)|
p
C
p
_
1
1
[aH(
t
, w)e
axn
]
2M
0
_
_
T
1
_
F
n
(
t
, w)a
1

M
(a)
_
_
p
dw.
For the second factor, (3.3) and M = a

imply for all [w[ 1 that

_
_
T
1
_
F
n
(
t
, w)a
1

M
(a)
_
_
p
C
p
(M)|f|
p
.
As M = a

is xed, we only need to show that [aH(

t
, w)e
axn
]
2M
0
C
(uniformly for [x
n
[, [w[ 1), which follows from [aH(
t
, w)]
2M
0
c.
Now aH(
t
, w) = a(
+
e
w

e
w
)/D. As
+
e
w

e
w
is odd in
and analytic in a and , one gets

+
e
w

e
w
= Z
w
(a,
2
)
with Z
w
(a, b) analytic in a, b. Thus
[aH(
t
, w)]
2M
0
c[Z
w
(a, +a
2
)]
2M
0
[a/D]
2M
0
and the conclusion follows with the help of Lemma 2.11.
Lemma 5.3. Let h
3
be dened as in (5.13) and let 1 < p < . Then
|h
3
|
p
C
_
|g|
p
+ (1 +[[)|g|

W
1
p
_
for all g C

(0)
()

W
1
p
().
Proof. Remember that
h
3
(x
t
, x
n
) = T
1
_
_
1
1
e
axn
K(
t
, w)G(
t
, w)dw
_
with K(
t
, w) = (
+
k(
t
, w) +

k(
t
, w))/D and
k(
t
, w) = e
aw
e
a
a
+e
aw
e
(+a)
+a
.
Due to = ( a)( +a), there is no singularity for = a.
We use the same cuto function
M
(a) as in the proof of the previous
lemma to split h
3
into h
1
3
+ h
2
3
, where the support of the multiplier in h
2
3
is
contained in [0, 2M], and M = max[[
1/2
/4, a

.
Let us estimate h
1
3
rst. Using (2.1)
[a
1
e
az
K(, w)]

M
c[e
a+

a
1
D
1
]

M
[ae
a(zw2)
]

M
1108 Helmut Abels and Michael Wiegner

__
a
1
e
a
e
a
a
_

M
+
_
a
1
e
a
e
a
+a
_

M
_
.
The rst factor is bounded by a constant because of Lemma 2.10, the second
by C(2 z w)
1
because of Lemma 2.9. The last factor may be simplied
to
_
+a
a
_

M
+
_
a
a
e
2
_

M
,
which is bounded due to Lemma 2.9. Thus,
|h
1
3
(, x
n
)|
Lp(R
n1
)
C
p
_
1
1
(2 x
n
w)
1
|T
1
[aG](, w)|
Lp(R
n1
)
dw.
As |T
1
[aG](, w)|
Lp(R
n1
)
C
p
|g(, w)|
Lp(R
n1
)
, the estimate for h
1
3
in
L
p
() follows by Lemma 2.6.
Turning to h
2
3
, we have to estimate multipliers on [0, 2M]. Suppose rst
that
1
4
[[
1/2
a

and therefore [[

3a.
We calculate
K(
t
, w)e
az
= e
a(w+z)
_
e
2
e
2
_
D
1
+e
a(wz)
_
a
+a
(1 e
22a
)
+a
a
(1 e
22a
)
_
D
1
.
In order to apply Lemma 2.12, we estimate the multiplier of G[
t
[
1
, which
is
h(
t
, w, z) =
_
e
2+2a
e
2a2
_
ae
a(w+z2)
D
+
_
a
+a
(e
2a
e
2
)
+a
a
(e
2a
e
2
)
_
ae
a(zw2)
D
.
With = ( a)( +a), we estimate on [0, 2M] using Lemma 2.9
[h(
t
, w, z)]
2M
0

C
2 w z
_
e
2+2a
D
1
( +a)
2
a
1 +a
_
2M
0
_
+a
a
(1 e
4
)
_
2M
0
+
C
2 z +w
_
e
2+2a
D
1
( +a)
2
a
1 +a
_
2M
0

__
e
2
e
42a
_

_
+a
a
_
2
_
e
2
e
2a
__
2M
0
.
Because of Lemma 2.9, Lemma 2.10, and
e
2+2a
D
1
( +a)
2
a
1 +a
=
_

+a
_
2

1
m
+

1
a
1 +a
m

,
Resolvent estimates for the Stokes operator 1109
we conclude [e
2+2a
D
1
( + a)
2 a
1+a
]
2M
0
C

. The remaining terms can

also be estimated by Lemma 2.9. Hence we get
[h(
t
, w, z)]
2M
0
c(2 z w)
1
.
Now [
t
[
1 h
w
(
t
, w, z) is of precisely the same structure, apart from a sign,
which does not matter. Hence by Lemma 2.12, we get |h
2
3
|
p
c|g|

W
1
p
.
It remains to consider the case [[
1/2
4a

, a 2a

. We calculate
K(
t
, w)e
az
= a
1
e
az
a
D
_
e
aw
e
2
e
2

+e
aw
_
2
0
e
ta
(e
t
+e
t
)dt +ae
aw
_
2
0
e
ta
e
t
e
t

dt
_
,
where by Lemma 2.11 [
a
D
]
2a
0
c, and the terms in brackets are analytic in a
and
2
. Thus K(
t
, w)e
az
= a
1
R(
t
, w, z), with
_

w
R(
t
, w, z)[
t
[
1

2a
0
+
[R(
t
, w, z)]
2a
0
c. By Lemma 2.12, we get in this case |h
2
3
|
p
c|g|

W
1
p
.
This nishes the proof.
Combining Lemma 5.1-Lemma 5.3, we have proved that
|q|
p
C
,p
_
|f|
p
+|g|
p
+ (1 +[[)|g|

W
1
p
_
uniformly in S

, > 0. Since ( )u = f q and u[

= 0,
Theorem 4.1 implies the estimates for u stated in Theorem 1.1. Hence ex-
tending the solution operator by continuity we have proved the solvability
of (1.1)-(1.3) for every C (,

2
4
].
Finally, it remains to prove uniqueness of the solution. Let A
p
= P
p

p
with T(A
p
) = T(
p
) L
p,
(), 1 < p < , be the Stokes operator. Then,
by the solvability of (1.1)-(1.3), +A
p
is surjective for every C(,

2
4
]
and 1 < p < . Because of ( + A
p
)
t
+ A
p
, ( + A
p
)
t
is surjective.
Hence +A
p
is injective and the solution of (1.1)-(1.3) is unique.
References
[1] T. Abe and Y. Shibata. On a resolvent estimate of the Stokes equation on an innite
layer, J. Math. Soc. Japan, 55 (2003), 469497.
[2] T. Abe and Y. Shibata, On a resolvent estimate of the Stokes equation on an innite
layer, part 2, = 0 case, J. Math. Fluid Mech., 5 (2003), 245274.
[3] H. Abels, Generalized Stokes resolvent equations in an innite layer with mixed bound-
ary conditions, Preprint No. 2292, TU Darmstadt, 2003, to appear in Math. Nachr.
1110 Helmut Abels and Michael Wiegner
[4] H. Abels, Reduced and generalized Stokes resolvent equations in asymptotically at
layers, part I: unique solvability, Preprint No. 2293, TU Darmstadt, 2003, to appear
in J. Math. Fluid. Mech.
[5] H. Abels, Reduced and generalized Stokes resolvent equations in asymptotically at
layers, part II: H-calculus, Preprint No. 2294, TU Darmstadt, 2003, to appear in J.
Math. Fluid. Mech.
[6] H. Abels, Bounded imaginary powers of the Stokes operator in an innite layer, J.
Evol. Eq., 2 (2002), 439457.
[7] H. Abels, Stokes Equations in Asymptotically Flat Domains and the Motion of a Free
Surface, PhD thesis, TU Darmstadt, Shaker Verlag, Aachen, 2003.
[8] J. Bergh and J. Lofstrom, Interpolation Spaces, Springer, Berlin - Heidelberg - New
York, 1976.
[9] R. Farwig, Weighted L
q
-Helmholtz decompositions in innite cylinders and in innite
layers, Adv. Di. Eq., 8 (2003), 357384.
[10] Y. Giga and T. Miyakawa, Solutions in Lr of the Navier-Stokes Initial Value Problem,
Arch. Rat. Mech. Anal., 89 (1985), 267281.
[11] T. Kato, Strong L
p
-solutions of the Navier-Stokes equation in R
n
, with applications
to weak solutions, Math. Z., 187 (1984), 471480.
[12] P. I. Lizorkin, (Lp, Lq)-multipliers of Fourier integrals, Sov. Math., Dokl., 4 (1963),
14201424.
[13] T. Miyakawa, The Helmholtz decomposition of vector elds in some unbounded do-
mains, Math. J. Toyama Univ., 17 (1994), 115-149.
[14] S. A. Nazarov and K. Pileckas, On the solvability of the Stokes and Navier-Stokes
problems in the domains that are layer-like at innity, J. Math. Fluid Mech., 1 (1999),
78-116.
[15] S. A. Nazarov and K. Pileckas, On th Fredholm property of the Stokes operator in a
layer-like domain, Z. Anal. Anwend., 20 (2001), 155-182.
[16] C.G. Simader and H. Sohr, A new approach to the Helmholtz decomposition and the
Neumann problem in Lq-spaces for bounded and exterior domains, Series on Advances
in Mathematics for Applied Sciences, Vol. 11, Singapore, World Scientic, 1-35, 1992.
[17] M. Wiegner, Resolvent estimates for the Stokes operator on an innite layer, Personal
manuscript, 1994.
[18] M. Wiegner, The Navier-Stokes equations a neverending challenge? Jber. d. Dt.
Math.-Verein., 101 (1999), 1-25.