Stabilizability of Two-Dimensional Navier-Stokes Equations With Help of A Boundary Feedback Control

J. math. uid mech.
3 (2001) 259301
1422-6928/01/030259-43 $ 1.50+0.20/0
c 2001 Birkhauser Verlag, Basel
Journal of Mathematical
Fluid Mechanics
Stabilizability of Two-Dimensional NavierStokes Equations
with Help of a Boundary Feedback Control
A. V. Fursikov
Abstract. For 2D NavierStokes equations dened in a bounded domain we study stabi-
lization of solution near a given steady-state ow v(x) by means of feedback control dened
on a part of boundary . New mathematical formalization of feedback notion is proposed.
With its help for a prescribed number > 0 and for an initial condition v
0
(x) placed in a
small neighbourhood of v(x) a control u(t, x
), x
, is constructed such that solution v(t, x)

of obtained boundary value problem for 2D NavierStokes equations satises the inequality:
v(t, ) v
H
1 ce
t
for t 0. To prove this result we rstly obtain analogous result on
stabilization for 2D Oseen equations.
Mathematics Subject Classication (2000). 93D15, 76D05.
Keywords. Oseen equations, NavierStokes equations, stabilization, extension operator, stable
invariant manifold.
1. Introduction
In this paper we study so-called stabilization problem for two-dimensional (2D)
NavierStokes equations dened in a bounded domain R
2
which is controlled
by Dirichlet boundary condition for velocity vector eld. Let ( v(x), p(x)), x
be a steady-state solution for NavierStokes equations
t
v(t, x) v(t, x) +(v, )v +p(t, x) = f(x), div v = 0, x , t > 0, (1.1)
supplied with an initial condition
v(t, x)[
t=0
= v
0
(x) (1.2)
and v ,= v
0
. We suppose that v[
= 0 and v is an unstable singular point for the

dynamic system generated by equation (1.1) supplied with zero condition v[
= 0
on the boundary of .
Hence, generally speaking the solution (v(t, x), p(t, x)) of (1.1)(1.2) goes
away ( v(x), p(x)) as t . But instead of zero boundary condition we introduce
This work was supported partially by RFBI grant 00-01-00415, and RFBI grant 00-15-96109.
260 A. V. Fursikov JMFM
the boundary control
v[
= u. (1.3)
Stabilization problem is as follows: Let
| v v
0
|
H
1 < where > 0 is suciently small. (1.4)
Given > 0 nd u(t, x), t > 0, x such that the solution (v, p) of (1.1)(1.3)
satises the inequality:
|v(t, ) v()|
H
1
()
ce
t
as t . (1.5)
Note that existence theorem for solution of problem (1.1)(1.5) can be derived
easily from the result of local exact controllability for NavierStokes equation (see
[5], [7] and references there). But this existence theorem can not be considered as
a complete theoretical foundation for numerical solution of problem (1.1)(1.5).
The point is that problem (1.1)(1.5) is ill-posed and therefore unpredictable uc-
tuations which usually appear during numerical realization, grow with time. As
a result this lead to completely wrong numerical calculations. The way how to
overcome this diculty is well-known: one has to nd and to use feedback control.
In other words control must react on appearing uctuations and suppress them.
Of course, feedback control was studied in applied sciences. A mathematical
formalization of feedback notion was proposed in the theory of controlled ordinary
dierential equations, and this notion was extended to the case of partial dier-
ential equations (PDE). In the case of problem (1.1)(1.5) this formalization is
reduced to the following denition:
A control u(t, x) from (1.3) is called feedback if there exists a map transfor-
ming vector elds dened on to vector elds dened on such that for each
t > 0
u(t, ) = (v(t, )) (1.6)
where v(t, x) is velocity vector eld of uid ow from (1.1)(1.5).
If we take
v(t, )) = v(t, )[
B(v(t, ), D
tx
v(t, ))[
(1.7)
where B is a nonlinear transformation and D
tx
denote derivatives with respect to
t, x, then (1.3), (1.6), (1.7) imply the equality
B(v(t, ), D
t,x
v(t, ))[
= 0. (1.8)
Thus, stabilization problem usually is reformulated as follows: nd a boundary
condition (1.8) such that the solution v(t, x) of boundary value problem (1.1),
(1.2), (1.8), (1.4) satises (1.5). Stabilization problem in this formulation was
studied for a number classes of evolution PDE in many papers. (See, for instance
[15], [12], [3], [1]
1
.) But for many linear and nonlinear parabolic equations as
well as for NavierStokes system satisfactory results for stabilization problem were
not obtained yet. From our point of view the reason of this is that formalization
1
Here we do not pretend on completeness of references, of course.
Vol. 3 (2001) NavierStokes Feedback Stabilizability 261
(1.6) of feedback motion is not adopted good enough for parabolic PDE. In [6]
some other mathematical formalization of feedback notion was proposed and this
formalization was applied to solve stabilization problem for linear and quasilinear
parabolic equations. In this paper we apply this new formalization of feedback
notion to investigate stabilization problem for 2D NavierStokes equations. More
precisely we solve stabilization problem (1.1)(1.5) with help of feedback control,
concentrated on a part of boundary . The idea of proposed methods is explained
in the next section for the case of Stokes equations.
2. Stabilizability of the Stokes system
2.1. Formulation of the problem
In this section we show the idea of method of stabilization with prescribed rate of
decay. We do this for the Stokes equations, dened on a bounded domain R
2
with C
-smooth connected boundary :
t
v(t, x) v +p(t, x) = 0, (t, x) Q, (2.1)
div v(t, x) = 0, (t, x) Q, (2.2)
v(t, x)[
t=0
= v
0
(x), x (2.3)
with Dirichlet boundary condition
v(t, x
) = u(t, x
) (t, x
) (2.4)
where u(t, x
) is a control, dened on the lateral surface = (0, ) of

space-time cylinder Q = (0, ) .
Here v(t, x) = (v
1
(t, x), v
2
(t, x)) is a velocity vector eld dened for (t, x)
(t, x
1
, x
2
) Q, p is a pressure gradient and v
0
(x) is a given solenoidal vector
eld: div v
0
(x) = 0.
Let a magnitude > 0 be given. The problem of stabilization from the bound-
ary with prescribed rate > 0 of a solution v(t, x) for evolution problem (2.1)
(2.4) is to construct a boundary control u(t, x
), (t, x
) such that the solution

(v, p) satises inequality:
|y(t, )|
L
2
()
ce
t
(2.5)
where constant c > 0 depends on v
0
and .
The main condition on control u, that this control satises the feedback condi-
tion, will be formulated after presentation the whole construction of stabilization.
2.2. Method of stabilization
We choose L > 0 so large that the following inclusion is true:
x = (x
1
, x
2
) R
2
: [x
i
[ < L, i = 1, 2. (2.6)
After that we identify the opposite sides of the square from (2.6) and obtain
the torus, which we denote by . Then (2.6) implies that
. (2.7)
We omit for a while condition (2.4) in problem (2.1)(2.4) and extend this
problem from to . As a result we obtain a boundary value problem for the
Stokes system with periodic boundary conditions
t
w(t, x) w +q(t, x) = 0, x , t > 0 (2.8)
div w = 0 (2.9)
w(t, x)[
t=0
= w
0
(x), div w
0
= 0. (2.10)
Let
w
0
(x) =
L
Z)
2
w
0
()e
ix
be the decomposition of w
0
(x) on Fourier series, where
= (
1
,
2
),
i
=

n
j
, n
j
N, j = 1, 2, x = x
1
1
+x
2
2
,
w
0
() = (2L)
2
_
w
0
(x)e
ix
dx
is a Fourier coecient for w
0
(x). Applying Fourier method we obtain that the
component w(t, x) of solution (w, q) for (2.8)(2.10) is dened by the formula:
w(t, x) =
L
Z)
2
w
0
()e
||
2
t+ix
. (2.11)
Solution (2.11) of problem (2.8)(2.10) satises the inequality
|w(t, )|
L
2
()
ce
t
(2.12)
if and only if the following equalities hold:
w
0
() = 0 : [[ <
. (2.13)
We introduce some functions spaces. Let be an open domain or a torus, then
V
k
() = v(x) = (v
1
(x), v
2
(x)) (H
k
())
2
: div v(x) = 0, k = 0, 1, 2, . . .
(2.14)
where H
k
() is the Sobolev space of functions which belong to L
2
() together
with all derivatives up to the order k; if k = 0 then div w = 0 is understood in the
meaning of distributions theory;
|v|
2
V
k
()
= |v|
2
(H
k
())
2 =
||k
_
[D
v(x)[
2
dx (2.15)
where = (
1
,
2
),
j
are nonnegative integer, [[ =
1
+
2
, D
v =
||
/x
1
1
x
2
2
.
We set
X
() =
_
w(x) V
0
() :
_
v(x)e
ix
dx = 0 if [[ <
_
. (2.16)
As always
| |
X
()
= | |
V
0
()
= | |
(L
2
())
2.
Below we prove the following assertion:
Theorem 2.1. For each > 0 there exists a continuous extension operator
Ext : V
0
() X
() (2.17)
(i.e. Extv(x) v(x) for x ).
Theorem 2.1 implies immediately the following method of solution for stabi-
lization problem (2.1)(2.5):
Theorem 2.2. Let v
0
V
0
(). For each > 0 the solution (v(t, x), u(t, x)) of
stabilization problem (2.1)(2.5) can be obtained by the formula
(v(t, ), u(t, )) = (
w(t, ),
w(t, )) (2.18)
where
and
are restriction operators on and on correspondingly,

w(t, x) is the solution of boundary value problem (2.8)(2.10) with initial con-
dition w
0
(x) = Extv
0
(x). Here Ext is extension operator (2.17) and v
0
(x) is the
initial condition from (2.3).
Proof. In virtue of (2.17) vector eld w
0
(x) = Extv
0
(x) satises (2.13). Therefore
(2.18), (2.17) imply:
|v(t, )|
L
2
()
|w(t, )|
L
2
()
ce
t
.
Denition 2.1. We say that control u from problem (2.1)(2.5) satises feedback
property if the solution (y, u) of this problem is dened by formula (2.18) where
w(t, x) is the solution of a certain articial boundary value problem.
From the physical point of view control satises feedback property if it can
react on unpredictable uctuations of a system. In Section 2.4 we will show that
control dened by (2.18) can react on such uctuations. But rstly we will prove
Theorem 2.1.
2.3. Proof of Theorem 2.1. Let v(x) V
0
(). Since is a bounded domain
with smooth boundary and div v = 0, we have v
(s)(s) H
1
2
() (see [17])
and
_
v(s) (s)ds = 0, (2.19)

where (s) is the vector eld of normals to .
Introduce the stream function F(x) by identities
x
2
F = v
1
,
x
1
F = v
2
. (2.20)
Recall that
H
1
0
() = (x) H
1
() : [
= 0,
H
1
() = completion in L
2
() by |f|
H
1 = sup
H
1
0
,=0
__
fdx/||
H
1
0
_
.
In virtue of (2.19), (2.20)
_
F[
(s) (s) ds = 0
where vector (s) is tangential to , and therefore we can set
F[
=
s
_
0
F[
(s)()(s)d H
1
2
() (2.21)
(recall that is connected set). Relations (2.20) imply:
F(x) = rotv
x
1
v
2

x
2
v
1
. (2.22)
Evidently, rotv H
1
(). That is why solution F of (2.22), (2.21) exists and
belongs to H
1
().
Extend F(x) from to torus by the formula
RF(x) = (1 (x))R
0
F(x) +(x)z(x) (2.23)
where R
0
: H
1
() H
1
() is an extension operator (i.e. R
0
F(x) F(x), x ),
(x) C
(), (x) = 0 for x
x : inf
y
[x y[ < and (x) = 1
for x
2
. The function z(x) from (2.23) is dened by the formula
z(x) =
L
Z)
2
, ||<
e
ix
(2.24)
where c
are coecients, which are determined from the system of equations
L
Z)
2
,||<
a
k
c
=
_
(1 (x))R
0
F(x)e
ikx
dx, k
_
L
Z
_
2
, [k[ <
.
(2.25)
Here
a
k
=
_
(x)e
ix(k)
dx.
Note that the matrix A = |a
k
| of system (2.25) is positively dened. Indeed, let
g(x) =
L
Z)
2
,||<
e
ix
.
Then
k
a
k
k
=
_
(x)[g(x)[
2
dx > 0,
if vector
,
_
L
Z
_
2
, [[ <

,= 0. Hence system (2.25) has a unique
solution c
and therefore extension (2.23) is well dened.

Relations (2.23)(2.25) imply that
_
RF(x)e
ikx
dx = 0 k : [k[ <
. (2.26)
We denote extension operator for v
0
(x), x as follows:
Extv
0
(x) =
_
v
0
(x), x
rotRF, x
(2.27)
where, recall, rotRF = (
x
2
RF,
x
1
RF). Since RF H
1
(), rotRF (L
2
())
2
and, evidently, div rotRF = 0. Hence rotRF V
0
() and in virtue of denition
(2.23)(2.25), (2.27) of extension operator we get:
|Extv
0
(x)|
V
0
()
c|v
0
|
V
0
()
. (2.28)
Using (2.26) we have after integration by parts
_
rotRF(x)e
ikx
dx =
_
_
ik
2
_
RF(x)e
ikx
dx, ik
1
_
RF(x)e
ikx
dx
_
_
= 0,
[k[ <
.
(2.29)
By (2.29) Extv
0
X
() and by (2.28) operator (2.17) is bounded.

2.3. Feedback property
We show that method of stabilization proposed below can react on unpredictable
uctuations of a system. The point is that if the solution v(t, x) of problem (2.1)
(2.4) satises inequality (2.5) and if at time moment

t
0
the system (2.1)(2.3) is
subjected by certain uctuation, then v(t, x) at t =

t
0
is pushed out X
and that
is why it will not tend to zero with prescribed rate. Therefore we check when
|v(t, )
w(t, )|
L
2
()
ce
t/2
(2.30)
(here w(t, x) is the solution of (2.8)(2.10) with w
0
(x) = Extv
0
(x)) and at this
moment, say t
1
, we regard v(t
1
, x) as initial condition and for t > t
0
we construct
the solution (v, u) of (2.1)(2.4) by formula (2.18) where w(t, x) is the solution of
(2.8), (2.9) with initial condition
w[
t=t
1
= Extv(t
1
, ). (2.31)
This construction can be written briey as impulse control for (2.8), concentrated
in the articial part of domain:
t
w(t, x) w +q(t, x) = (t t
1
)(Extv(t
1
, x) w(t
1
, x)) (2.32)
where w(t
1
, x) = v(t
1
, x), for x and w(t, x) = w(t, x) for x . After next
unpredictable pushing out X
() of the solution v(t, x) we do the same. Thus all

this process can be written by formula:
t
w(t, x) w +q(t, x) =
i=1
(t t
i
)(Extv(t
i
, x) w(t
i
, x)) (2.33)
where t
i
are moments when (2.30) became true.
Remark 2.1. Note that above we indicate only some possibility to organize reac-
tion of control on unpredictable uctuations. To realize this possibility the special
investigations should be made. They will be made in some other place. Note that
a certain previous result to this respect is obtained in [6].
3. Oseen equations
3.1. Formulation of the problem
Let R
2
be a bounded connected domain with C
-boundary , Q = R
+
.
In space-time cylinder Q we consider the Oseen equations
t
v(t, x) v + (a(x), )v + (v, )a +p(t, x) = 0 (3.1)
div v(t, x) = 0 (3.2)
with initial condition
v(t, x)[
t=0
= v
0
(x) (3.3).
Here (t, x) = (t, x
1
, x
2
) Q, v(t, x) = (v
1
, v
2
), is a velocity vector eld, p =
(
x
1
,
x
2
) is pressure gradient, initial velocity v
0
(x) = (v
01
(x), v
02
(x)) satises
condition div v
0
= 0, a(x) = (a
1
(x), a
2
(x)) is a solenoidal vector eld (div a = 0).
We suppose that the boundary of is decomposed on two parts:
=
0
, ,= (3.4)
where ,
0
are open sets (in topology of ). Here the line above means the
closure of a set. The case
0
= is also possible. We dene = R
+
,
0
=
R
+

0
, and we set the following boundary conditions:
v[
0
= 0, v[
= u (3.5)
where u is a control, concentrated on .
Let a magnitude > 0 be given. The problem of stabilization with the decay
rate of a solution to problem (3.1)(3.3), (3.5) is to construct a control u on
such that the solution v(t, x) of boundary value problem (3.1)(3.3), (3.5) satises
the inequality
|v(t, x)|
L
2
()
ce
t
(3.6)
where c > 0 depends on v
0
, and
0
. Moreover, we require that this control u
satises the feedback property in the meaning analogous to (2.18).
Let us give the exact formulation of this feedback property. Let R
2
be a
bounded domain such that
= , = . (3.7)
We set
G = Int( ) (3.8)
(the denotation Int A means, as always, the interior of the set A). We suppose
that G is a curve belonging to the smoothness class C
and in all points except

the set it possesses the C
smoothness. The exact conditions on

magnitude will be imposed below in an appropriate place.
We extend problem (3.1)(3.3) from to G. Let us assume that the given
vector eld a(x) from (3.1) actually is dened on G. Moreover, we suppose that
a(x) V
2
(G)
_
H
1
0
(G)
_
2
. (3.9)
That is why the extension of (3.1)(3.3) from to G can be written as follows:
t
w(t, x) w + (a(x), )w + (w, )a +p(t, x) = 0 (3.10)
div w(t, x) = 0 (3.11)
w(t, x)[
t=0
= w
0
(x). (3.12)
Besides, we impose on w the zero Dirichlet boundary condition
w[
S
= 0, (3.13)
where S = R
+
G. Note that, actually, w
0
from (3.12) will be some special
extension of v
0
(x) from (3.3).
For vector elds dened on G we introduce the operator
of restriction on
and the operator
of restriction on :
: V
k
(G) V
k
(), k 0;
: V
k
(G) V
k1/2
(), k > 1/2. (3.14)
Evidently, operators (3.14) are bounded.
Denition 3.1. A control u(t, x) in (3.1)(3.3), (3.5) is called the feedback if for
each t > 0 the following relations are true:
v(t, ) =
w(t, ), u(t, ) =
w(t, ) (3.15)
where (v(t, ), u(t, )) is the solution of stabilization problem (3.1)(3.3), (3.5) and
w(t, ) is the solution of boundary value problem (3.10)(3.13).
Below we prove that for given > 0, v
0
the problem (3.1)(3.3), (3.5) can be
stabilized with help of feedback control in the meaning of Denition 3.1.
3.2. Preliminaries
Let G be domain (3.8) and
V
0
0
(G) = v(x) V
0
(G) : v [
= 0 (3.16)
where V
0
(G) is space (2.14) with = G, (x) is the vector eld of outer normals
to G; for v V
0
(G) the function v [
is well-dened in the Sobolev space

H
1/2
(G) (see [17]). Denote by
: (L
2
(G))
2
V
0
0
(G) (3.17)
the operator of orthogonal projection. We consider the Oseen steady-state opera-
tor
Av v +[(a(x), )v + (v, )a] : V
0
0
(G) V
0
0
(G) (3.18)
where a(x) is vector eld (3.9). This operator is closed and its domain is dened
as follows:
T(A) = V
2
(G) (H
1
0
(G))
2
(3.19)
and it is dense in V
0
0
(G).
Assuming that spaces in (3.17), (3.19) are complex we denote by (A) the
resolvent set of operator A, i.e. the set of C such that the resolvent operator
R(, A) (I A)
1
: V
0
0
(G) V
0
0
(G) (3.20)
is dened and continuous. Here I is identity operator.
Recall that a closed operator B : X X (X is a Banach space) is called a
sectorial if there exist a magnitude (0, /2), M 1, a R such that
S
a,
= C : [arg( a)[ , ,= a (A) (3.21)
and
|(I A)
1
| M/[ a[, S
a,
. (3.22)
Denote by (A) C
1
(A) the spectrum of operator A.
Lemma 3.1. Oseen operator (3.18) with a(x) satisfying (3.9) is a sectorial opera-
tor. For (A) resolvent (3.20) is a compact operator. That is why the spectrum
(A) consists of a discrete set of points.
The proof of this lemma can be obtained by well-known methods: compactness
of resolvent and estimate (3.22) can be established by analogous estimates to that
were obtained below, in Lemmas 4.7 proof. All other assertions of the Lemma
follows from results of [10] and [20].
Let
0
(A). We decompose the resolvent R(, A) in a neighbourhood of
0
in a Laurent series:
R(, A) =
k=m
(
0
)
k
R
k
, R
k
= (2i)
1
_
|
0
|=
(
0
)
k1
R(, A)d.
(3.23)
Lemma 3.2. ([20, Ch YII Sect.8]).
a) All operators R
k
from (3.23) commutate among themselves and with A;
b) R
1
is a projector, i.e. R
2
1
= R
1
.
c) The following formulas hold:
R
(k+1)
+ (
0
I A)R
k
= 0 for k 1 (3.24)
R
1
+ (
0
I A)R
0
= E. (3.25)
Lemma 3.3.
a) The magnitude m from decomposition (3.23) is nite;
b) The images of each operator R
k
with k < 0 from (3.23) are nite dimen-
sional: dimImR
k
< for k < 0.
Moreover
ImR
1
= ker(
0
E A)
m
, R
k1
= (
0
E A)
k
R
1
, k = 1, 2, . . . , m1.
(3.26)
This lemma is derived in [6] from well-known results of [4], [20].
Let us consider the adjoint operator A
to Oseen operator (3.18):

A
w w [(a(x), )w (a)
w] : V
0
0
(G) V
0
0
(G) (3.27)
where
(a)
w = ((
1
a, w), (
2
a, w)), (
i
a, w) =
2
j=1
i
a
j
w
j
. (3.28)
Evidently, A
is a closed operator with domain coinciding to T(A)

T(A
) = T(A) = V
2
(G) (H
1
0
(G))
2
.
Moreover
(A
) = (A) and R(, A)
= R(
, A
) (A). (3.29)
Below we always assume that
vector eld a(x) from (3.9), (3.18), (3.27) is real valued. (3.30)
That is why we have
(A) = (A) = (A
) = (A
). (3.31)
Since B and B
are compact operators simultaneously, (3.29), (3.31) imply that

A
is a sectorial operator with a compact resolvent and (A) = (A) = (A
) =
(A
). Hence if
0
(A) then
0
(A
). Decomposition R(, A
) in Laurent
series around
0
yields:
R(, A
) =
k=m
(

0
)
k
R
k
, (3.32)
where, evidently poles orders in (3.32) and in (3.23) coincides and R
k
from (3.32)
are adjoint operators to the corresponding operators R
k
from (3.23). Therefore
by Lemma 3.2 (R
1
)
2
= R
1
,
R
(k+1)
+ (

0
I A
)R
k
= 0 for k 1 (3.33)
R
1
+ (

0
I A
)R
0
= I. (3.34)
Moreover, Lemma 3.3 implies that dimImR
k
= dimImR
k
, k < 0 and
ImR
1
= ker(

0
E A
)
m
, R
k1
= (

0
E A
)
k
R
1
, k = 1, 2, . . . , m1.
(3.35)
3.3. Structure of R
k
, R
k
with k < 0
Recall some denitions. If ker(
0
I A) ,= 0 then
0
C
1
is called eigenvalue of
A, and a vector e ,= 0 which belongs to ker(
0
E A) is called eigenvector. Vector
e
k
is called associated vector of order k to an eigenvector e if e
k
can be obtained
after solving the chain of equations:
(
0
I A)e = 0, e + (
0
I A)e
1
= 0, . . . , e
k1
+ (
0
I A)e
k
= 0. (3.36)
We say that e, e
1
, e
2
, . . . form a chain of associated vectors. If the maximal order
of vectors, associated to e equals m then the number m + 1 is called multiplicity
of the eigenvector e.
Denition 3.2. The set of eigenvectors and associated vectors
e
(k)
, e
(k)
1
, . . . , e
(k)
m
k
(k = 1, 2, . . . , N) (3.37)
corresponding to an eigenvalue
0
is called canonical system which corresponds to
the eigenvalue
0
if set (3.37) satises the properties:
i) Vectors e
(k)
, k = 1, 2, . . . , N form a basis in the space of eigenvectors cor-
responding to the eigenvalue
0
.
ii) e
(1)
is an eigenvector with maximal possible multiplicity.
iii) e
(k)
is an eigenvector which can not be expressed by a linear combination
of e
(1)
, . . . , e
(k1)
and multiplicity of e
(k)
achieves a possible maximum.
iv) Vectors (3.37) with xed k form a maximal chain of associated elements.
Evidently, numbers m
1
, m
2
, . . . , m
N
do not depend on a choice of canonical
system.
The number N(
0
) = m
1
+1 +m
2
+1 + +m
N
+1 is called multiplicity of
the eigenvalue
0
.
Besides canonical system (3.37) which corresponds to an eigenvalue
0
of op-
erator A we consider a canonical system
(k)
,
(k)
1
, . . . ,
(k)
m
k
(k = 1, 2, . . . , N) (3.38)
that corresponds to the eigenvalue
0
of the adjoint operator A
. Denition of
canonical system (3.38) is absolutely analogous to Denition 3.2 of canonical sys-
tem (3.37). Particularly the set (3.38) with xed k satises the relations which
are analogous to (3.36):
(

0
I A
)
(k)
= 0,
(k)
+ (

0
I A
)
(k)
1
= 0, . . . ,
(k)
m
k
1
+ (

0
I A
)
(k)
m
k
= 0.
(3.39)
The assertion which we formulate below is very close to well-known assertion from
[10], although their formulations are dier. That is why we give also a proof. Let
H be a Hilbert space with a scalar product , ). For u, v
H denote by uv
the
operator Bx = x, v
)u. Then v
u is adjoint operator B
g = u, g)v
.
Theorem 3.1. Let A : H H be a sectorial operator with a compact resolvent
and the operator A
: H H is adjoint to A. Suppose that

0
(A),
0
(A
)
and therefore
0
,
0
are the poles of R(, A), R(, A
) correspondingly which have

the equal multiplicity m. Then
i)
0
is an eigenvalue of A and
0
is an eigenvalue of A
and both of them

have eigenvectors of maximal multiplicity m.
ii) Let set (3.38) be an arbitrary canonical system of A
corresponding to the
eigenvalue
0
. Then (3.38) denes by a unique way a canonical system
(3.37) of A corresponding to
0
such that the main part of Laurent series
(3.23) has the form
k
_
e
(k)
(k)
(
0
)
m
k
+1
+
e
(k)
(k)
1
+e
(k)
1

(k)
(
0
)
m
k
+. . . .+
+
e
(k)
(k)
m
k
+e
(k)
1

(k)
m
k
1
+. . . .. +e
(k)
m
k
(k)
(
0
)
_
. (3.40)
Proof. As was mentioned, in virtue of (3.29) operators R
k
from (3.32) are ad-
joint to corresponding operators R
k
from (3.23), and by Lemma 3.3 dimImR
k
=
dimImR
k
< if k < 0. That is why
R
k
=
L
j=1
f
j
g
j
if and only if R
k
=
L
j=1
g
j
f
j
(3.41)
where f
j
H, g
j
H, j = 1, . . . , L are certain vectors. The formula (3.40) for the
main part of Laurent series (3.23) will be proved simultaneously with the following
formula for the main part of (3.32):
k
_

(k)
e
(k)
(
0
)
m
k
+1
+

(k)
1
e
(k)
+
(k)
e
(k)
1
((
0
)
m
k
+. . . .+
+
(k)
(m
k
)
e
(k)
+
(k)
(m
k
1)
e
(k)
1
+. . . .. +
(k)
e
(k)
(m
k
)
((
0
)
_
_
_
. (3.42)
Let e, e
1
, . . . , e
h
be a chain of associated vectors for A corresponding to eigen-
value
0
. Then for each from a neighbourhood of
0
the equality holds:
e
(
0
)
h+1
+
e
1
(
0
)
h
+ +
e
h
(
0
)
= R(, A)e
h
. (3.43)
To prove (3.43) one can apply to both parts of (3.43) operator (I A) = ((
0
)I +
0
I A) and use (3.36). Hence multiplicity m of pole (3.23) is not less
than maximal possible multiplicity of a eigenvector corresponding to
0
. If m is
the multiplicity of pole (3.23) then R
(m+1)
= 0 and by (3.33) with k = m we
have ImR
m
Ker(
0
I A). Therefore the multiplicity m of pole (3.23) equals to
maximal possible multiplicity of eigenvectors for A corresponding
0
. Analogously,
if ,
1
, . . . ,
h
be a chain of associated vectors for A
then
(
0
)
h+1
+

1
(
0
)
h
+ +

h
(
0
)
= R(, A
)
h
(3.44)
and the multiplicity m of pole (3.32) is equal to maximal possible multiplicity of
eigenvectors for A
corresponding
0
.
Let
(k)
, k = 1, . . . , j
1
be a given maximal set of linear independent eigenvectors
of maximal multiplicity m for A
. Then by (3.44) with

h
=
(k)
m1
for each x H
R
m
x =
j
1
k=1
< e
(k)
, x >
(k)
(3.45)
where, evidently, vectors e
(k)
are dened by
(k)
by unique way. Relations (3.41)
implies that
R
m
y =
j
1
k=1
< y,
(k)
> e
(k)
(3.46)
and in virtue of (3.43) e
(k)
, k = 1, . . . , j
1
are eigenvectors for A of maximal possible
multiplicity m. Since dimImR
m
= dimImR
m
, e
(k)
are linear independent and
there is no any other linear independent to e
(k)
, k = 1, . . . , j
1
eigenvectors for A of
multiplicity m.
By (3.43) with e
h
=
j
1
k=1
< y,
(k)
> e
(k)
m1
we see that the main part of (3.23)
contains
j
1
k=1
_
< y,
(k)
> e
(k)
(
0
)
m
+
< y,
(k)
> e
(k)
1
(
0
)
m1
+ +
< y,
(k)
> e
(k)
m1
(
0
)
_
where e
(k)
j
are vectors associated to eigenvectors e
(k)
. Analogously by (3.44) with
h
=
j
1
k=1
< e
(k)
, x >
(k)
m1
the main part of (3.32) contains:
j
1
k=1
_
< e
(k)
, x >
(k)
(
0
)
m
+
< e
(k)
, x >
(k)
1
(
0
)
m1
+ +
< e
(k)
, x >
(k)
m1
(
0
)
_
.
These two expressions and relation (3.41) imply that the main parts of poles (3.23)
and (3.32) contain correspondingly:
A
j
1
k=1
_
e
(k)
(k)
(
0
)
m
+
e
(k)
1

(k)
+e
(k)
(k)
1
(
0
)
m1
+ +
e
(k)
m1
(k)
+e
(k)
(k)
m1
(
0
)
_
(3.47)
A

j
1
k=1
_

(k)
e
(k)
(
0
)
m
+

(k)
1
e
(k)
+
(k)
e
(k)
1
(
0
)
m1
+ +

(k)
m1
e
(k)
+
(k)
e
(k)
m1
(
0
)
_
.
(3.48)
Now we continue this process: we supplement the sum in (3.47) using (3.43)
with e
h
=

j
1
k=1
< y,
(k)
l
> e
(k)
m1l
, h = m 1 l for l = 1, 2, . . . , m 1. After
that we add terms in (3.48) using new (added) sum in (3.47) and relation (3.41).
As a result we obtain that the main part of (3.23) contains terms
j
1
k=1
_
e
(k)
(k)
(
0
)
m
+
e
(k)
1

(k)
+e
(k)
(k)
1
(
0
)
m1
+
e
(k)
2

(k)
+e
(k)
1

(k)
1
+e
(k)
(k)
2
(
0
)
m2
+. . . .
+
e
(k)
m1
(k)
+e
(k)
m2
(k)
1
+ +e
(k)
(k)
m1
(
0
)
_
and the main part of (3.32) contains terms
A
1
=
j
1
k=1
_

(k)
e
(k)
(
0
)
m
+

(k)
1
e
(k)
+
(k)
e
(k)
1
(
0
)
m1
+ +. . .
+
(k)
m1
e
(k)
+
(k)
m2
e
(k)
1
+ +
(k)
1
e
(k)
m2
+
(k)
e
(k)
m1
(
0
)
_
. (3.49)
It is well-known that canonical system (3.38) for A
corresponding to eigenvalue
0
is a set of linear independent vectors.
2
That is why in virtue of (3.44) sum (3.49)
2
Amplication of this result is proved in [6] and in Lemma 3.7 below.
does not contain all main part of (3.32) but contains the operator R
m
/(
0
)
m
.
Therefore
1
k=m
R
k
(
0
)
k
A
1
(3.50)
is a pole which order is less than m. Applying to (3.50) arguments we used
above: we take the eigenvalues
(k)
, k = j
1
+ 1, . . . , j
r
for A
from (3.38) which

multiplicity is maximal possible after excluding
(k)
with k = 1, 2, . . . , j
1
. For
(k)
, k = j
1
+ 1, . . . , j
2
we repeat arguments written from (3.45) to (3.49). As a
result we obtain a sum of terms A
2
(analogous to (3.49)) such that the order of
pole
1
k=m
R
k
(
0
)
k
A
1
A
2
will be less than the order of pole (3.50). Repeating this process several times we
complete the proof of Theorem 2.1.
Corollary 3.1. Let A be a sectorial operator with compact resolvent R(, A) de-
composed at
0
(A) in Laurent series (3.23). Let (3.38) be a canonical system
of A
corresponding to the eigenvalue

0
, which we denote by E(
0
). Then
R
k
x = 0, k = 1, 2, . . . , m (3.51)
if and only if
x,
(k)
j
) = 0
(k)
j
E(
0
). (3.52)
This assertion follows immediately from representation (3.40) for the main part
of Laurent series (3.23) for R(, A).
3.4. Holomorphic semigroups
We consider boundary value problem (3.10)(3.13) for Oseen equations written in
the form
dw(t, )
dt
+Aw(t, ) = 0, w[
t=0
= w
0
(3.53)
where A is operator (3.18). In virtue of Lemma 3.1. Oseen operator is sectorial
with compact resolvent. It is known ([8, Sect. 1.3]) that the following assertion
holds.
Theorem 3.2. For each w
0
V
0
0
(G) the solution w(t, ) of problem (3.53) is
dened by w(t, ) = e
At
w
0
where e
At
is the holomorphic semigroup that is de-
termined by the formula
e
At
= (2i)
1
_
(I +A)
1
e
t
d, (3.54)
where is a contour belonging to (A) such that arg = for , [[ N
for certain (/2, ) and for suciently large N. Moreover, surrounds (A)
from the right.
Such contour exists, of course, because we can choose belonging to set
S
a,
from denition of sectorial operator (see (3.21)).
Let be a negative number satisfying
(A) C : Re = = . (3.55)
The case when there are certain points of (A) placed righter the line Re =
will be interesting for us.
Using contour described below (3.54) we dene the continuous contour
that is placed in C : Re and constructed from an interval of the line

Re = and from two branches of contour that transform to arg =
and arg = , (/2, ) for suciently large [[.
Evidently, integral in (3.54) can be transformed as follows:
e
At
= (2i)
1
_
(I +A)
1
e
t
d+
j
(2i)
1
_
|+
j
|=
(I +A)
1
e
t
d, (3.56)
where summation is made over points
j
(A) placed righter the line Re =
, and > 0 is small enough.
To calculate terms in the sum from (3.56) we decompose the resolvent R(, A)
at =
j
:
R(, A) =
k=m(
j
)
( +
j
)
k
R
k
(
j
). (3.57)
Residue of R(, A)e
t
at =
j
is calculated by the formula
Rese
t
R(, A) = e
j
t
m(
j
)
n=1
t
n1
(n 1)!
R
n
(
j
).
Hence (3.56) can be rewritten as follows:
e
At
= (2i)
1
_
(E +A)
1
e
t
d +
Re
j
<
e
j
t
m(
j
)
n=1
t
n1
(n 1)!
R
n
(
j
).
(3.58)
Lemma 3.4. The following estimate is true:
|(2i)
1
_
(E +A)
1
e
t
d|
V
0
0
(G)
ce
t
as t 0 (3.59)
where c > 0 does not depend on t.
Proof. Denition of
and (3.21) imply that for suciently large R
C : [[ > R S
a,
.
This inclusion, relation
(A) and (3.22) where A is changed on A and

S
a,
is changed on S
a,
lead to the inequality
|(E +A)
1
| M
1
/(1 +[[) for
. (3.60)
Besides, in virtue of denition of
[e
(+)t
[ ce
cos ||t/2
as
and [[ (3.61)
where cos < 0 since (
2
, ). By (3.60), (3.61)
|(2i)
1
_
(E +A)
1
e
t
d| e
t
_
M
1
ce
cos t/2
1 +[[
[d[ M
2
e
t
for t > 1.
(3.62)
As known, the norm of left-hand-side of (3.58) is bounded uniformly with respect
to t [0, 1]. Besides, the norm of the sum from right-hand-side of (3.58) is also
bounded uniformly to t [0, 1]. Hence
|(2i)
1
_
(E +A)
1
e
t
d| M
2
e
t
for t [0, 1]. (3.63)
Now (3.59) follows from (3.62), (3.63).
Denote by
(k)
(
j
),
(k)
1
(
j
), . . . ,
(k)
m
k
(
j
), k = 1, . . . , N(
j
) (3.64)
a certain canonical system of eigenvectors and associated vectors of operator A
corresponding eigenvalue
j
.
Theorem 3.3. Suppose that A is operator (3.18) and > 0 satises (3.55). Then
for each w
0
V
0
0
(G) that satises
w
0
,
(k)
l
(
j
) >= 0, l = 0, 1, . . . , m
k
, k = 1, 2, . . . , N(
j
), Re(
j
) <
(3.65)
(here by denition
(k)
0
(
j
) =
(k)
(
j
)) the following inequality holds:
|e
At
w
0
|
V
0
0
(G)
ce
t
|w
0
|
V
0
0
(G)
for t 0. (3.66)
Proof. By Corollary 3.1 if w
0
satises (3.65) then R
n
(
j
)w
0
= 0 for every oper-
ators R
n
(
j
) from (3.58). Therefore (3.58), (3.59) imply (3.66).
3.5. On linear independence of
(k)
(x)
We set some strengthening of well-known result on linear independence of eigenvec-
tors and associated vectors for operator (3.27), (3.28) conjugate to Oseen operator
(3.18).
Lemma 3.5. Let
(k)
(x) k = 1, . . . , N be complete linear independent system
of eigenvectors to an eigenvalue

0
. Then for an arbitrary subdomain < G
functions
(k)
(x), x , k = 1, . . . , N are linear independent.
Proof. Let
f(x) =
N
k=1
c
k
(k)
(x) = 0, x . (3.67)
Evidently, equality (
0
E A
)f(x) = 0 holds for x G and therefore by (3.27)

and denition of operator
f(x) (a(x), )f(x) + (a(x))
f(x) = p(x), x G (3.68)

with some function p(x) H
1
(G). Applying operator rot
g(x)
1
g
2

2
g
1
to
(3.68) we get
rot
f(x) (a(x), )rot
f(x) = 0, x G. (3.69)
(This formula can be veried by straightforward calculations). By (3.67) we have
rot
f(x) = 0, x . (3.70)
Relations (3.69), (3.70) imply (see [9]) that
rot
f(x) = 0, x G,
and this, as well-known (see, for example, [5, p. 204]) leads that f(x) const. By
the denition (3.67) and by relation
(k)
[
G
= 0 we get f(x) 0, x G. Hence
in (3.67) c
k
= 0, k = 1, . . . , N.
Theorem 3.4. Let (3.64) be a certain canonical system of eigenvectors and asso-
ciated vectors for A
corresponding eigenvalue
j
. Then for each satisfying
(3.55) and for arbitrary subdomain G the set of canonical systems (3.64) for
all eigenvalues
j
satisfying
Re
j
< (3.71)
are linear independent.
Proof. In proof of this assertion specic character of operator A
is used to prove
Lemma 3.5 only. The last part of prove is general and has been made in [6].
Impose on canonical systems (3.64) the following condition
(k)
(
j
) =
(k)
(
j
);
(k)
l
(
j
) =
(k)
l
(
j
). (3.72)
Indeed, since by (3.30) vector eld a(x) is real valued, if we act the operation
of complex conjugation to functions (3.64), we evidently get canonical system of
eigenvectors and associated vectors for A
j
. That
is why condition (3.72) can be realized easily.
In virtue of (3.72) canonical system corresponding to real
j
consists of real
valued vector elds. If Im
j
,= 0, instead of vector elds
(k)
(
j
),
(k)
l
(
j
),
l = 0, 1, . . . , we consider real valued vector elds
Re
(k)
l
(
j
), Im
(k)
l
(
j
), l = 1, . . . , k = 1, 2 . . . . (3.73)
(As above
(k)
0
(
j
) =
(k)
(
j
) by denition). We renumber all functions (3.73)
with Re
j
< (including elds with Im
j
= 0) as follows:
1
(x), . . . ,
K
(x). (3.74)
Lemma 3.6. For an arbitrary subdomain G vector elds (3.74) restricted on
are linear independent over the eld R of real numbers.
Lemma 3.6 follows easily from Theorem 3.4. (see [6])
Note that Theorem 3.3 implies immediately the following assertion.
Corollary 3.2. Assume that A is operator (3.18) and > 0 satises (3.55). Then
for each w
0
V
0
0
(G) satisfying
_
G
(w
0
(x),
j
(x))x = 0, j = 1, . . . , K (3.75)
with
j
from (3.74), inequality (3.66) is true.
4. Stabilization of the Oseen equation
4.1. Theorem on extension
The key step in stabilization method that we propose is construction of special
extension for vector elds from to G (see (3.8)). First of all we make more
precise the conditions imposed on and G. Recall that
G = Int(
) (4.1)
where and are open subsets of R
2
, = and is a closed curve of
C
-class and
=
0
, G =
0
(4.2)
where ,
0
are open subsets of , ,= , and is a nite number of points, or
= .
We suppose that
=
N
_
j=1
j
(4.3)
where
j
are connected components of . We assume that the following con-
ditions are true:
Condition 4.1. For each j = 1, . . . , N the set
j

0
is connected or it is
empty.
Condition 4.1. implies that for each j the set
j
also is connected or
empty.
We impose the following smoothness condition on and G:
Condition 4.2. Let C
, G C
and for each point P there

exist local coordinates (x, y) such that P is origin; P = (0, 0), (x, 0), x (0, )
, (x, 0), x (, 0)
0
and in a neighbourhood of P G can be represented
as follows:
G (x, y) = (x, x
), x (0, )(x, y) = (x, 0), x (, 0), > 1. (4.4)

Here > 0 is a small magnitude.
Remind that is a given domain where the Oseen system which should be sta-
bilized is determined. The domain we choose ourselves. That is why Condition
4.2 is not restrictive.
We introduce the following spaces
V
1
0
(G) = u(x) = (u
1
(x), u
2
(x)) V
1
(G) : u[
G
= 0
V
1
(,
0
) = u(x) = (u
1
(x), u
2
(x)) V
1
() :
u[
0
= 0, v V
1
0
(G) that u(x) =
v(x)
(4.5)
where V
1
(G) is dened in (2.14),
is the operator of restriction on for functions

from V
1
0
(G). The space V
1
(,
0
) is supplied with the following norm:
|u|
V
1
(,
0
)
= inf
L
|Lu|
V
1
0
(G)
where inmum is taken over all bounded extension operators L : V
1
(,
0
)
V
1
0
(G).
Denition (4.5) is convenient to prove the extension result but it is not con-
structive. Below, in Subsection 4.2 we give simple condition on vector eld which
guarantees its belonging to V
1
(,
0
).
We will use the following subspaces of Sobolov space H
2
(G) of scalar functions:
H
2
(G) = F H
2
(G) :
_
G
F(x) dx = 0, (4.6)
H
2
(G) = F(x)

H
2
(G) : F[
G
= 0 (4.7)
where, recall, F = (
x
1
F,
x
2
F).
Remind that for a scalar function F(x), x G R
2
rotF(x) = (
2
F(x),
1
F(x)). (4.8)
Lemma 4.1. The operator
rot :

H
2
(G) V
1
0
(G) (4.9)
is an isomorphism.
Proof. It is known (see [5, p. 204]) that the operator
rot :

H
2
(G) V
1
(G)
is isomorphism. If = (
1
,
2
) is the vector eld of normals to G, then =
(
1
,
2
) = (
2
,
1
) is the eld of vectors, tangent to G along G. Let v V
1
0
(G)
and F = rot
1
v

H
2
(G). Then
n
F[
G
= (rotF, )[
G
= (v, )[
G
= 0,
F[
G
= (rotF, )[
G
= (v, )[
G
= 0.
(4.10)
Therefore F[
G
= 0 and F

H
2
0
(G). If F

H
2
0
(G) and v = rotF then in virtue
of (4.10) and condition F[
G
= (
n
F,
F)[
G
= 0 we get that v V
1
0
(G).
For v V
1
(G) denote by rot
v the operator formally conjective to (4.8). It is

easy to see that
rot
v(x) =
1
v
2
(x)
2
v
1
(x) where v(x) = (v
1
(x), v
2
(x)). (4.11)
Lemma 4.2. Let
1
be a subdomain of G such that Int(
1
G) ,= . Suppose
that
f(x)
K
j=1
c
j
rot
j
(x) = 0, x
1
(4.12)
where c
j
are constants and
j
(x) are vector elds (3.74). Then c
j
= 0, j =
1, . . . , K.
Proof. Denote scalar functions
j
(x) = rot
1
j
(x). (4.13)
This denition is correct in virtue of Lemma 4.1 since
j
(x) V
1
0
(G). Let
g(x) =
N
j=1
c
j
j
(x). (4.14)
Since by (4.8), (4.11) = rot
rot we get by (4.12), (4.13), (4.14) that

g(x) f(x) = 0, x . (4.15)
By (4.13), Lemma 4.1 and (4.14) g[
G
= 0 and therefore
g[
G
1
= 0. (4.16)
Since Int(G
1
) ,= , by uniqueness of a solution for Cauchy problem for
Laplace operator (see [9]) relations (4.15), (4.16) imply that
g(x) const, x
1
,
and therefore applying to (4.14) operator rot we get in virtue of (4.13) that
N
j=1
c
j
j
(x) 0, x
1
.
This inequality and Lemma 3.6 imply that c
1
= = c
N
= 0.
We prove now the extension theorem. In the space of real valued vector elds
V
1
0
(G) we introduce the subspace
X
1
K
(G) = v(x) V
1
0
(G) :
_
G
v(x)
j
(x) dx = 0, j = 1, . . . , K (4.17)
where
j
(x) are functions (3.76).
Theorem 4.1. There exists a linear bounded extension operator
E
1
K
: V
1
(,
0
) X
1
K
(G) (4.18)
(i.e. E
K
(v)(x) v(x) for x ).
Proof. By denition (4.5) of V
1
(,
0
) there exists a linear continuous extension
operator
L : V
1
(,
0
) V
1
0
(G). (4.19)
Set
= x R
2
: dist(x, ) <
where dist(x, ) is the distance from x to . Let
0
(x) C
(

G), 0
0
(x) 1,
0
(x) =
_
1, x G
/2
,
0, x G
1
(x) = 1
0
(x) C
(

G). (4.20)
We look for extension operator E
1
K
in a form
E
1
K
v(x) = rot(
0
rot
1
Lv)(x) + rot
_
1
(x)
K
j=1
c
j
rot
j
(x)
_
, (4.21)
where c
j
are constants which we have to determine. Evidently, E
1
K
v(x) = v(x)
if x for any c
j
. To dene constants c
j
we note that by (4.17) inclusion
E
K
v X
K
(G) are fullled if
_
G
k
(x)rot
_
1
(x)
K
j=1
c
j
rot
j
(x)
_
dx =
_
G
k
(x)rot(
0
rot
1
Lv)(x) dx (4.22)
where k = 1, . . . , K.
Let
a
kj
=
_

k
(x) rot[
1
(x)rot
j
(x)] dx =
_
(rot
k
(x))
1
(x)rot
j
(x) dx. (4.23)
Then the matrix A = |a
kj
| is positively dened. Indeed, let = (
1
, ,
K
),
f(x) =
K
j=1
j
rot
j
(x). Then
(A, ) =
K
k,j=1
j
_

1
(rot
k
(x))rot
j
(x) dx =
_
G
1
(x)[f(x)[
2
dx 0.
Moreover, if for some (A, ) =
_

1
(x)[f(x)[
2
dx = 0 then f(x) = 0 for x
1
= G
and by Lemma 4.2

1
= =
K
= 0. That is why det A ,= .
Equations (4.22) can be rewritten as follows:
Ac = g
where c = (c
1
, . . . , c
K
) and g is the vector with components dened by right
size of (4.22). Thus c = A
1
g and denition of operator (4.21) is completed.
Boundedness of operator (4.18) follows immediately from its denition (4.21).
4.2. On the space V
1
(,
0
)
In this subsection we show that each vector eld v(x) V
1
() satisfying v[
0
= 0
belongs to V
1
(,
0
) if v(x) is smooth enough in a neighbourhood of
0
. This
result is not used directly for solution of the stabilization problem given below in
Subsection 4.4.
We introduce the spaces:
V
1
0
= u(x) = (u
1
, u
2
) V
1
() : v[
0
= 0 (4.24)
H
2
(,
0
) = F(x)

H
2
() : F[
0
= 0 (4.25)
where

H
2
() is dened analogously to (4.6).
Lemma 4.3. Operator
rot :

H
2
(,
0
) V
1
0
() (4.26)
is an isomorphism.
The proof of this lemma is completely analogous to the proof of Lemma 4.1.
We want to nd out what vector eld from V
1
0
() can be extended to a vector
eld belonging to V
1
0
(G). Firstly we consider the simplest case when
j
j
or
j
= . (4.27)
Proposition 4.1. If condition (4.27) is true then there exists an extension opera-
tor
L : V
1
0
() V
1
0
(G) (4.28)
and therefore V
1
0
() = V
1
(,
0
).
Proof. Let v V
1
0
(), F = rot
1
v

H
2
(,
0
) where rot
1
is constructed in
Lemma 4.3. Since by (4.27) is closed manifold we can extend F through by
means of usual Witney extension theorem and this operator we denote by L
0
.
Then operator (4.28) can be dened as follows:
L = rot
0
L
0
rot
1
(4.29)
where
0
is the operator of multiplication on function (4.20). Operator (4.29),
evidently, is bounded in spaces (4.28)
Condition (4.27) actually means that = . Now we consider the case when
,= . In this case we also look for an extension operator in the form (4.29)
where, however, L
0
is not usual Witney extension operator. Below we construct
new operator L
0
.
Lemma 4.4. Let domains and G satisfy (4.1), (4.2), (4.3) and Conditions 4.1,
4.2 with 2. Suppose that F(x)

H
2
(,
0
) satises the condition
F(x) H
k
( O()) with k >
3 + 1
2
(4.30)
where O() is a neighbourhood of and is magnitude from (4.4). Then F(x)
can be extended up to a function L
0
F(x) H
2
(G) where
H
2
(G) = f(x) H
2
(G) : f[
G
= 0. (4.31)
Proof. Step 1. If F

H
2
(,
0
) then by (4.25) and Condition 4.1 F[
j
= c
j
=
const. Dene f by the equality
F = f +c
j
.
We extend the constant c
j
through
j
by just the same constant c
j
. That
is why to extend F through
j
we have to extend f(x). The problem is to
extend f(x) to a neighbourhood of the point P
j
because outside this
neighbourhood we can use Witney extension method.
By Condition 4.2 we can suppose that f(x, y) H
k
(Q
) with Q
= (x, y)
R
2
: y 0, x
2
+y
2
< 1
3
f(x, 0) =
y
f(x, 0) = 0 for x (1, 0). (4.32)
We must extend f on Q = (x, y) R
2
: x
2
+y
2
< 1, y (x) where (x) = x
for x 0, (x) = 0 for x 0 such that extended function L

0
f H
2
(Q) and
L
0
f[
(x,(x))
= L
0
f[
(x,(x))
= 0.
Step 2. We dene
() =
_
1 2
1
, (0, 1/2),
2
1
(1 )
, (1/2, 1),
2. (4.33)
It is clear that () H
2
(0, 1)
(1) =
(1) =
(0) = 0, (0) = 1 (4.34)

and
[()[ c [
()[ c
1
, [
()[ c
2
. (4.35)
We determine the extension operator L
0
as follows:
L
0
f(x, y) =
_
f(x, y), y < 0,
g(x, y)(y/x
), y > 0, x > 0,
(4.36)
where g(x, y) = 4f(x,
y
2
) 3f(x, y).
In virtue of (4.34) L
0
f and L
0
f satisfy desired zero boundary condition on
(x, (x). Moreover the function
h(x, y) =
_
f(x, y), y < 0,
g(x, y), y > 0, x > 0
belongs to H
2
(Q). Therefore we have to prove only that
g(x, y)(y/x
) H
2
(Q Q
). (4.37)
3
In Condition 4.2 we have x
2
+ y
2
< instead of x
2
+ y
2
< 1. By multiplication f on a cut
function equal 1 for x
2
+y
2
< /2 and equal 0 outside x
2
+y
2
we can reduce the deal to the
case x
2
+ y
2
< 1.
Step 3. We prove that the following estimates for function g dened below
(4.36) are true:
[g(x, y)[ c(x
k1
+x
k2
y +y
2
), (x, y) Q Q
, > 0 (4.38)
[
x
g(x, y)[ c(x
k2
+x
k3
y +y
2
), [
y
g(x, y)[ c(x
k2
+y), (4.39)
(x, y) Q Q
, > 0.
Indeed, (4.32) implies that
g(x, 0) =
y
g(x, 0) = 0 for x (1, 0). (4.40)
By Sobolev embedding theorem and by (4.30)
g(x, y) H
k
(Q Q
) C
k1
(Q Q
). (4.41)
In virtue of Lagrange theorem, (4.40), (4.41) (4.30) we have for (x, y) Q Q
:
[g(x, y)[ [g(x, 0)[ +[g(x, y) g(x, 0)[ cx
k1
+c[
y
g(x, y)[y
cx
k1
+cy([
y
g(x, 0)[ +[
y
g(x, y)
y
g(x, 0)[)
c(x
k1
+x
k2
y +y
2
).
This proves (4.38). Inequalities (4.39) are proved analogously.
Step 4. We prove (4.37). We have
1
_
0
dx
x
_
0
[
yy
(g(x, y)(y/x
2
))[
2
dy
c
1
_
0
dx
x
_
0
[g
yy
[
2
+[(
y
g)
y
[
2
+[
yy
g[
2
.) dx
(4.42)
Since
yy
g L
2
and [[ c we have to estimate only the rst and second terms
from right side. Taking into account (4.35), (4.38), (4.39) we get
1
_
0
dx
x
_
0
[g
yy
[
2
+[(
y
g)
y
[
2
dy
c
1
_
0
dx
x
_
0
_
(x
2(k1)
+x
2(k2)
y
2
+y
4
)y
2(2)
x
2(2)4
+
+(x
2(k2)
+y
2
)y
2(1)
x
2(1)2
dy
c
1
_
0
_
x
2(k1)2+(23)
+
+x
2(k2)2+(21)
+x
2+(2+1)
_
dx =
(4.43)
= c
1
_
0
(x
2(k1)3
+x
2(k2)
+x
) dx.
The right side of (4.43) is nite because of (4.30). The terms
xy
(g),
xx
(g),
x
(g),
y
(g) can be estimated analogously.
Now we can prove the nal result:
Theorem 4.2. Let domains and G satisfy (4.1), (4.2), Conditions 4.1, 4.2 and
,= . Suppose that v(x) V
1
0
() satises the condition
v(x)
_
H
m
( O())
_
2
with m >
3 1
2
, 2, (4.44)
where O() is a neighbourhood of and is a magnitude from (4.4). Then
v(x) V
1
(,
0
), i.e. it can be extended up to a vector eld Lv V
1
0
(G).
Proof. Let v(x) satisfy conditions of the Theorem. By Lemma 4.3 the func-
tion F(x) = rot
1
v(x) belongs to

H
2
(,
0
) and by (4.44) F(x) satises (4.30)
with k = m + 1. Then by Lemma 4.4 F(x) can be extended up to a function
L
0
F(x) H
2
(G). Evidently, the vector eld Lv(x) rot(

0
(x)L
0
F(x))
rot(
0
(x)L
0
rot
1
v(x)) belongs to V
1
0
(G).
4.3. Extension Theorem for V
0
Extension operator constructed in Subsections 4.1, 4.2 is the main in our theory
because it can be used for NavierStokes equations. Nevertheless there are some
reasons to construct analogous extension operator for more wide function spaces
of V
0
-type. This construction is analogous but easier than that was written above.
That is why we expound it briey.
Set
V
0
(,
0
) = u(x) V
0
() : u [
0
= 0, v V
0
0
(G) that u(x) =
v(x)
(4.45)
where
is the operator of restriction on . Besides,

|u|
V
0
(,
0
)
= inf
v=u
|v|
V
0
0
(G)
.
Denote
H
1
(G) = F(x) H
1
(G) :
F[
G
= 0,
_
G
F(x)dx = 0, (4.46)
is the derivative along the vector eld tangent to G.

Lemma 4.5. The operator
rot :

H
1
(G) V
0
0
(G)
dened in (4.8) is an isomorphism. (V
0
0
(G) is dened in (3.16)).
This lemma is proved as Lemma 4.1.
Analogously to (4.17) we dene
X
0
K
(G) = v(x) V
0
0
(G) :
_
G
v(x)
j
(x) dx = 0, k = 1, . . . , K, (4.47)
where
j
(x) are functions (3.74).
Theorem 4.3. There exists a linear bounded extension operator
E
0
K
: V
0
(,
0
) X
0
K
(G). (4.48)
The proof is analogous to Theorems 4.1 proof.
At last let us give a condition on vector eld v V
0
0
= v V
0
(): v [
0
=
0, which guarantees that v belongs to space (4.45) when ,= .
We do it rstly for functions
F(x)

H
1
(,
0
) F(x) H
1
() :
F[
0
= 0,
_
G
F(x)dx = 0. (4.49)
Lemma 4.6. Let domains and G satisfy (4.1), (4.2), (4.3) and Conditions 4.1,
4.2. Suppose that F(x) satises (4.49) and satises the following condition:
F(x) H
1+
( O()) (4.50)
and parameter from (4.4) satises 1 < < 1 +, 0 < < 1.
Then F(x) can be extended to a function L
0
F(x) H
1
(G) where
H
1
(G) = f(x) H
1
(G) :
f[
G
= 0
and
is derivative along vector eld tangential to G.

Draft of proof. By (4.49) any F

H
1
(,
0
) equals constant c
j
on
j

0
for
each connected component
j
of , f(x, y) F c
j
H
1+
(Q
), f(x, 0) = 0
for x (1, 0) and we must extend f(x, y) on Q, where the sets Q
and Q are
dened near (4.32).
We set () = 1 for (0, 1) and dene the extension operator as follows:
L
0
f(x, y) =
_
f(x, y), y < 0,
f(x, y)(y/x
), y > 0, x (0, 1).

As well as in Lemma 4.4 all proof is reduced to establishing of the inclusions
x
_
f(x, y)(y/x
)
_
L
2
(QQ
),
y
_
f(x, y)(y/x
)
_
L
2
(QQ
). (4.51)
To do it we take into account that by Sobolev embedding theorem H
1+
(Q
)
C
1
(Q
) and therefore the inequality is true:

[f(x, y)[ c(x
1
+y
1
) (x, y) Q Q
, where 1 <
1
< .
Now using this inequality and do estimates analogous to (4.42), (4.43) we prove
(4.51).
The following assertion succeeds from Lemma 4.6 as Theorem 4.2 was derived
from Lemma 4.4.
Theorem 4.4. Let and G satisfy (4.1), (4.2), Conditions 4.1, 4.2, and ,= .
Suppose that v(x) V
0
0
(), v(x) H
1+
( O()) where 0 < < 1, and
parameter from (4.4) satises inequalities 1 < < 1+. Then v(x) V
0
(, ).
4.4. Results on stabilization
We prove now the main theorem of this section on stabilizability by feedback
boundary control of 2D Oseen equations.
Theorem 4.5. Let domains and G satisfy (4.1), (4.2), and Conditions 4.1,
4.2. Then for each initial condition v
0
(x) V
0
(,
0
) and each > 0 there exists
a feedback control u dened on the part of boundary (0, ) such that the
solution v(t, x) of boundary value problem (3.1)(3.3), (3.5) satises the inequality:
|v(t, )|
(L
2
())
2 ce
t
as t . (4.52)
Proof. Let A be operator (3.18) and the magnitude > 0 satises condition (3.55).
In the case if it is not so we get satisfying (3.55) by small increasing of . We act on
initial condition v
0
V
0
(,
0
) by the operator E
0
K
from (4.48). Then by Theo-
rem 4.3 w
0
= E
0
K
v
0
satises including w
0
X
0
K
(G). By denition (4.47) of X
0
K
(G)
and denition (3.74) of vector elds
1
(x), . . . ,
K
(x) one yields that w
0
and sat-
isfy all conditions of Theorem 3.3. In virtue of this theorem inequality (3.66) is
true. We dene desired solution (v, u) of stabilization problem (3.1)(3.3), (3.5)
by formula (3.15), where w(t, ) = e
At
w
0
is the solution of problem (3.10)(3.13)
which can be rewritten in the equivalent form (3.53). Then (3.66) and evident
inequality
|v(t, )|
L
2
()
|w(t, )|
V
0
0
(G)
imply (4.52).
Theorem 4.6. Let domains and G satisfy (4.1), (4.2), and Conditions 4.1,
4.2 with 2. Then for each initial condition v
0
(x) V
1
(,
0
) and for each
> 0 there exists a feedback control u dened on such that the solution v(t, x)
of (3.1)(3.3), (3.5) satises the inequality
|v(t, )|
(H
1
())
2 ce
t
as t . (4.53)
Proof. As in Theorem 4.5 we can assume that satises conditions (3.55). We
act on initial condition v
0
V
1
(,
0
) by the operator E
1
K
from (4.18) and by
Theorem 4.1 we obtain that w
0
= E
1
K
v
0
X
1
K
(G). Since X
1
K
(G) V
1
0
(G)
V
0
0
(G), the solution w(t, x) of problem (3.1)(3.3), (3.5) can be written in the
form w(t, ) = e
At
w
0
where A is operator (3.18). We want to prove that
|w(t, )|
V
1
0
(G)
ce
t
for t 0. (4.54)
Since by well-known results on smoothness of solutions for Oseen equations in-
equality (4.54) is true for t [0, 1], it sucient to prove (4.54) for t > 1. In virtue
of representation (3.58) for e
At
, inclusion w
0
X
1
K
(G) and Corollary 3.1 we have
w(t, ) = e
At
w
0
=
_
(I +A)
1
w
0
e
t
d (4.55)
where the contour
is dene lower formula (3.55).

To estimate (4.55) we use the following assertion which will be proved after
nishing of Theorems 4.6 proof.
Lemma 4.7. Let A be operator (3.18) and > 0 satises (3.55). Then
|(I +A)
1
w|
V
1
0
(G)
c|w|
V
0
0
(G)
, (4.56)
where constant c > 0 does not depend on w V
0
0
(G) and
and the contour
is dened lower (3.55).

End of proof of Theorem 4.5. We estimate integral in (4.55) using arguments
of proof of the Lemma 3.4, but applying estimate (4.56) instead of (3.60). As a
result we get:
|w(t, )|
V
1
0
(G)
ce
t
|w
0
|
V
0
0
(G)
, t > 1.
Hence we established (4.54). Now we dene (v(t,), u(t, )) = (
w(t,),
w(t,))
and derive (4.53) from (4.54) and evident inequality |v(t,)|
H
1
()
c|w(t,)|
V
1
0
()
.
Proof of Lemma 4.7. In virtue if Lemma 3.1, denition of contour
, and condi-
tion (3.55), operator (I +A)
1
is well-dened on V
0
0
(G) if
.
4
We denote v = (I + A)
1
w. When to prove (4.56) we have to establish the
bound
|(A+I)v|
V
0
0
(G)
c
1
|v|
V
1
0
(G)
. (4.57)
4
Well-deniteness of (I + A)
1
on V
0
0
(G) for
, || 1 can be obtained easily also

by estimates analogous to that we get below in the proof of this Lemma.
If
and [[ 1 then by denition of
(see lower (3.55))

= (1 +i) where 1 and [[ < 1 ( R is xed). (4.58)
We consider rstly the Stokes operator
A
0
= : V
0
0
(G) V
0
0
(G) (4.59)
As well-known (see [17], [19]), the closure of map (4.59) is self-adjoint positive
operator which possesses a compact inverse operator. By HilbertShmidt theorem
the spectrum (A
0
) consists of a countable set of positive numbers 0 <
1

2

k
, as k , and corresponding eigenfunctions e
j
form an
orthogonal basis in V
0
0
(G) and in V
1
0
(G) (see [5]), and for v =
j=1
v
j
e
j
we have:
|v|
2
V
0
0
(G)
=
j=1
[v
j
[
2
, |v|
2
V
1
0
(G)
=
j=1
j
[v
j
[
2
. (4.60)
In virtue of (4.58), (4.60) we get that for
, [[ 1
|(I+A
0
)v|
2
V
0
0
(G)
=
j=1
[(1+i)+
j
[
2
[v
j
[
2
=
j=1
[(
j
)
2
+
2
2
][v
j
[
2
. (4.61)
Since for function f(x) = ((x )
2
+
2
2
)/x
inf
xR
+
f(x) = f(
_
1 +
2
) = 2(
_
1 +
2
1)
we get from (4.61) that
|(I +A
0
)v|
2
V
0
0
(G)
2(
_
1 +
2
1)
j=1
j
[v
j
[
2
= c
0
|v|
2
V
1
0
(G)
(4.62)
Consider now operator A from (3.18) which can be written in the form
Av = A
0
v +[(a(x), )v + (v, )a)] (4.63)
Since by (3.9) and in virtue of Sobolev embedding theorem
a(x) C(G), a(x) L
4
(G),
we get the following estimate:
|[(a, )v + (v, )a)]|
V
0
0
(G)

|a|
C(G)
|v|
V
1
0
(G)
+|a|
L
4
(G)
|v|
L
4
(G)
c
1
|v|
V
1
0
(G)
.
(4.64)
Taking into account (4.62), (4.63), (4.64) we obtain
|(I +A)v|
V
0
0
(G)

|(I +A
0
)v|
V
0
0
(G)
|[(a, )v + (v, )a)]|
V
0
0
(G)
(c
0
c
1
)|v|
V
1
0
(G)
.
This estimate and (4.58) imply (4.57) and therefore imply (4.56) for
, [[
1. For other
estimate (4.56) is true because
(A).
5. Stabilization of 2D NavierStokes equations
5.1. Formulation of the stabilization problem
As in Subsections 3.1, 4.1 we suppose that R
2
is a bounded connected domain
with C
-boundary , and =

0
, ,= where ,
0
are open sets
satisfying Condition 4.1. Recall denotations of the space-times sets: Q = R
+
,
0
= R
+

0
, = R
+
. In Q we consider the NavierStokes equations
t
v(t, x) v(t, x) + (v, )v +p(t, x) = f(x), (t, x) Q (5.1)
div v = 0 (5.2)
(v = (v
1
, v
2
)) with initial condition
v(t, x)[
t=0
= v
0
(x), x (5.3)
and boundary conditions
v[
0
= 0, v[
= u, (5.4)
where u = (u
1
, u
2
) is a control dened on .
We suppose also that a steady-state solution ( v(x), p(x)) of NavierStokes
system with the same right-hand side f(x) as in (5.1) is given:
v(x) + ( v, ) v + p = f(x), div v(x) = 0, x (5.5)
v[
0
= 0. (5.6)
Let > 0 be given. The problem of stabilization with the rate reduced to
look for a solution of problem (5.1)(5.4) which satises the inequality
|v(t, ) v|
(H
1
())
2 ce
t
as t . (5.7)
The important additional condition is that u is a feedback control. Denition
of feedback notion is analogous to Denition 3.1. Nevertheless we give now exact
formulation of feedback property.
We extend to a domain G through (see (4.1), (4.2)) such that Condition 4.2
holds. After that we extend problem (5.1)(5.4) to analogous problem dened on
= R
+
G:
t
w(t, x) w + (w, )w +q(t, x) = g(x), div w(t, x) = 0 (5.8)
w(t, x)[
t=0
= w
0
(x) (5.9)
with additional condition
w[
S
= 0 (5.10)
where S = R
+
G. Moreover we assume that solution ( v, p) of (5.5), (5.6) is
extended on G to a pair (a(x), q(x)), x G such that
a(x) + (a, )a + q(x) = g(x), div a(x) = 0, x G (5.11)
a[
G
= 0 (5.12)
where right side g(x) is the same as in (5.8). (We show below how to construct
such extension.)
Denition 5.1. A control u(t, x) in stabilization problem (5.1)(5.4) is called
feedback if the solution (v(t, x), u(t, x)) of (5.1)(5.4) is dened by the equality:
(v(t, x), u(t, x)) = (
w(t, ),
w(t, )) (5.13)
where w(t, x) satises to (5.8)(5.10), and
are operators of restriction of a

function dened on G to and to respectively.
5.2. Invariant manifolds
Let g(x) from (5.8) satisfy the condition:
g(x) (L
2
(G))
2
. (5.14)
Then as well-known (see, for instance [18]) equations (5.8) are equivalent to the
following equation with respect to one unknown function w(t, x):
t
w(t, x) w +(w, )w = g(x) (5.15)
where is orthoprojector (3.17) on V
0
0
(G) (see (3.16)). We assume in addition
that solution w of (5.15) (as well as solution w of (5.8)) belongs to the space
V
1,2
(
T
) w(t, x) L
2
(0, T; V
2
(G) (H
1
0
(G))
2
) :
t
w L
2
(0, T; V
0
0
(G)
(5.16)
for each T > 0, where
T
= (0, T) G. It is proved (see [13], [17]) that for each
T > 0, g(x) (L
0
2
(G))
2
, w
0
(x) V
1
0
(G) there exists unique solution w(t, x)
V
1,2
(Q
T
) of problem (5.15), (5.9). Solution w(t, x) of (5.15), (5.9) taken at time
moment t we denote as S(t, w
0
)(x):
w(t, x) = S(t, w
0
)(x). (5.17)
Since embedding V
1,2
(Q
T
) C(0, T; V
1
0
(G)) is continuous, the family of op-
erators S(t, w
0
) is continuous semigroup on the space V
1
0
(G) : S(t + , w
0
) =
S(t, S(, w
0
)).
Note that we can rewrite (5.11) in the form analogous to (5.15):
a(x) +(a, )a = g, a(x) V
1
0
(G) V
2
(G). (5.18)
Since a(x) is steady-state solution of (5.15), S(t, a) = a for each t 0. We can
decompose semigroup S(t, w
0
) in a neighbourhood of a in the form
S(t, w
0
+a) = a +L
t
w
0
+B(t, w
0
) (5.19)
where L
t
w
0
= S
w
(t, a)w
0
is derivative of S(t, w
0
) with respect to w
0
at point a,
and B(t, w
0
) is nonlinear operator with respect to w
0
. Dierentiability of S(t, w
0
)
is proved, for instance in [2, Ch. 7. Sect. 5]. Therefore
B(t, 0) = 0, B
w
(t, 0) = 0. (5.20)
Moreover in [2, Ch. 7. Sect. 5] is proved that B(t, w) belongs to class C
1+
with = 1/2 with respect to w. This means that for each w
0
V
1
0
(G)
|B
w
(t, w
0
)|
C
sup
0uw
0
V
1
0
(G)
1
|B
w
(t, u) B
w
(t, w
0
)|
V
1
0
(G)
|u w
0
|
V
1
0
(G)
<
and left side is a continuous function with respect to w.
We study now semigroup L
t
w
0
= S
w
(t, a)w
0
of linear operators. First of all
note that w(t, x) = L
t
w
0
is the solution of problem (3.10)(3.12) in which the
coecient a is the solution of (5.18). Therefore
L
t
w
0
= e
At
w
0
(5.21)
where A is operator (3.18). We remind (see [2])
Denition 5.2. Semigroup of linear bounded operators S
t
: X X (X is a
Hilbert space) is called almost stable if
i) S
t
are bounded in X for 0 t T by a constant depending on T.
ii) S
t
u, ) is continuous with respect to t u X,
where
is a
dense set in X
.
iii) For each t
0
> 0 and for each r
0
(0, 1) there exists not more than nite set
+
= (
1
, . . . ,
N
) of points belonging to spectrum of operator S
t
0
which
are placed outside the disk [[ r
0
.
iv) Invariant subspace of operator S
t
0
corresponding to
+
is nite dimen-
sional.
Below we suppose that r
0
(0, 1) satises the property:
C : [[ = r
0
(e
At
0
) = (5.22)
where, recall, (e
At
) is the spectrum of operator (5.21).
It is clear that
j
(e
At
0
) if and only if
j
= e
j
t
0
and
j
(A).
That is why condition (5.22) is equivalent to condition (3.55) where = lnr
0
/t
0
.
Besides, if [
j
[ > r
0
then Re
j
> .
The following assertion holds:
Theorem 5.1. Family of operators e
At
: V
1
0
(G) V
1
0
(G) where A is operator
(3.18) is well dened for each t 0 and it is almost stable semigroup (see Denition
5.2 where S
t
= e
At
, X = V
1
0
(G)). Let
+
=
1
, . . . ,
K
:
j
(e
At
0
), [
j
[ > r
0
, j = 1, . . . , K (5.23)
where r
0
(0, 1) and satises (5.22). Let X
+
V
1
0
(G) be the invariant subspace
for e
At
0
corresponding to
+
,
+
: V
1
0
(G) X
+
be the projector on X
+
(i.e.,
+
V
1
0
(G) = X
+
,
2
+
= ) and X
= (I
+
)V
1
0
(G) be complementary invariant
subspace. Let L
+
t
0
= e
At
0
[
X
+
: X
+
X
+
, L
t
0
= e
At
0
[
X
: X
. Then
operator L
+
t
0
has inverse operator (L
+
t
0
)
1
. For some t
0
there exist constants r,
+
,
(0, 1) such that

|L
t
0
| r(1
), |(L
+
t
0
)
1
| r
1
(1
+
). (5.24)
Proof. Well posedness of operator e
At
and the point i) of Denition 5.2 is proved,
for instance, in [13], [17]. Denote in the point ii) of Denition 5.2 by , ) the duality
between V
1
0
(G) and V
1
(G) = (V
1
0
(G))
. Then ii) also follows from [13], [17].

Semigroup e
At
: V
1
0
(G) V
1
0
(G) is restriction of semigroup e
At
: V
0
0
(G)
V
0
0
(G). Hence formula (3.54) is true for e
At
: V
1
0
(G) V
1
0
(G):
e
At
w
0
= (2i)
1
_
(I +A)
1
w
0
e
t
d, w
0
V
1
0
(G), (5.25)
where C
1
is the contour described lower (3.54). In Section 3 was established
that for each < 0 there exists not more than nite number of
j
(A)
for which Re(
j
t
0
) > t
0
. Hence, for r
0
= e
t
0
the number of
j
= e
j
t
0
(e
At
0
) such that [
j
[ > r
0
is nite. This established iii) of Denition 5.2.
In correspondence with general denition ([4], [20]) operator
+
dened in
formulation of Theorem 5.1 can be calculated as follows:
+
= (2i)
1
_
(I +e
At
0
)
1
d (5.26)
where = z C
1
: [[ = r
0
C
1
: [[ = R and R > 0 is so large
that outside of the disk [[ R there is no eigenvalues of operator e
At
0
; going
around external circle is counterclockwise and circuit internal circle is clockwise.
In [6, Lemma 5.3] is shown that
+
=
Re
j
<
R
1
(
j
) (5.27)
where R
1
(
j
) is the operator-coecient R
k
(
j
) with k = 1 in Laurent
decomposition of the resolvent R(, A) around eigenvalue
j
. Since
dimImR
1
(
j
) < (see Lemma 3.3 above), and number of
j
with Re(
j
) >
is nite, we proved point iv) of Denition 5.2.
Operators
+
and e
At
0
commutate. (see Lemma 3.2) Therefore operator
L
+
t
0
= e
At
0
[
X
+
with X
+
=
+
V
1
0
(G) is dened by the formula
L
+
t
0
=
+
e
At
0
= (2i)
1
_
(I e
At
0
)
1
d. (5.28)
As we showed in [6, Lemma 5.3] this formula can be transformed as follows:
L
+
t
0
=
Re
j
<
e
j
t
0
m(
j
)
n=1
t
n1
0
R
n
(
j
)
(n 1)!
. (5.29)
Comparing (5.26), (5.29) and (3.58) and taking into account (3.26) we see that
operator L
t
0
= e
At
0
[
X
where X
= (I
+
)V
1
0
(G) is dened by the formula
L
= (2i)
1
_
(I +A)
1
e
t
0
d (5.30)
where
is the counter described lower formula (3.55). Integral (5.30) was esti-
mated in Theorem 4.6 (see (4.53)). As a result we have:
|L
| ce
(+)t
0
(5.31)
where | | is the norm of operator acting in V
1
0
(G).
In [6, Lemma 5.3] we showed that the inverse operator (L
+
t
0
)
1
exists and is
dened by the formula
(L
+
t
0
)
1
=
Re
j
<(+
0
)
e
+
j
t
0
m(
j
)
n=1
t
n1
0
(n 1)!
R
n
(
j
).
It follows from this formula that
|(L
+
t
0
)
1
| c
1
e
+()t
0
(5.32)
where c
1
does not depend on t
0
and > 0 is a suciently small magnitude.
In virtue of (5.31), (5.32) inequalities (5.24) are true for enough large t
0
if
r = e
(
2
)t
0
and
+
,
(0, 1).
Generally speaking, eigenvalues of operators A and e
At
are complex-valued.
That is why all spaces in Theorem 5.1 are complex. But to apply obtained results
to (nonlinear) NavierStokes equations we need to have analogous results for the
real spaces of the same type. Actually for this we have to dene the projector
of (5.27) in real spaces.
The form (3.40) of the main part for Laurent series (3.23) implies that
R
1
(
j
)v =
(k)
_
e
(k)
0
(
j
)v,
(k)
m
k
(
j
)) +e
(k)
1
(
j
)v,
(k)
m
k
1
(
j
))+
+ +e
(k)
m
k
(
j
)v,
(k)
0
(
j
))
_
(5.33)
where , ) is scalar product in complex space (L
2
(G)),
(k)
l
(
j
) is canonical
system (3.64) of operator A
j
and e
(k)
l
(
j
)
is canonical system of A corresponding to eigenvalue
j
, which is constructed
by
(k)
l
(
j
) in Theorem 3.1. We suppose also that
(k)
l
(
j
) satisfy (3.72).
Then e
(k)
l
(
j
) satisfy analogous property that follows from the proof of Theo-
rem 3.1.
We dene restriction of operator (5.33) on the real space V
1
0
(G) with help of
vector elds (3.73) and
Ree
(k)
l
(
j
, x), Ime
(k)
l
(
j
, x). (5.34)
Lemma 5.1. Restriction of operator (5.33) on the real space V
1
0
(G) can be written
in the form
(
+
)(x) =
K
j=1
e
j
(x)
_
G
v(x)
j
(x) dx (5.35)
where e
j
is the set of functions (5.34) which is renumbered and is renormalized by
suitable away and
j
is renumbered and is renormalized set of functions (3.74).
The proof of this simple lemma one can nd in [6, Lemma 6.2].
The following assertion holds.
Lemma 5.2. For an arbitrary subdomain G vector elds e
j
(x), j = 1, . . . , K
from (5.35) restricted on are linear independent over R.
Since there is evident symmetry between operators A and A
and therefore
between e
j
(x) and elds (3.74), Lemma 5.2 is proved similarly to Lemma 3.6.
Using (5.35) we can easily restrict spaces X
+
and X
as well as operators L
+
t
0
,
L
t
0
dened in formulation of Theorem 5.1 on the real subspaces of V
1
0
(G). We
denote this new real spaces and operators also by X
+
, X
, L
+
t
0
, L
t
0
. This will not
lead to misunderstanding because below we do not use their complex analogs.
In a neighbourhood of steady-state solution a of (5.18) we establish existence of
a manifold M
which is invariant with respect to semigroup S(t, w) (i.e., S(t, w)

M
w M
). This manifold can be represented as the graph:

M
= u V
1
0
(G) : u = a +u
+g(u
), u X
O (5.36)
where O is a neighbourhood of origin in V
1
0
(G), g : X
O X
+
is an operator-
function of class C
3/2
and
g(0) = 0, g
(0) = 0. (5.37)
Note that condition (5.37) means that manifold (5.36) is tangent to X
at
point a.
The following theorem is true.
Theorem 5.2. Let a satisfy (5.18), > 0 satisfy (3.55), O = O
= v V
1
0
(G) :
|v|
V
1
0
(G)
< and is suciently small. Then there exists unique operator-
function g : X
O X
+
of class C
3/2
satisfying (5.37) such that the mani-
fold M
dened in (5.36) is invariant with respect to semigroup S(t, w

0
) connected
with (5.15).
5
There exists a constant c > 0 such that
|S(t, w
0
) a|
V
1
0
(G)
c|w
0
a|
V
1
0
(G)
e
t
as t 0 (5.38)
for each w
0
M
.
5
I.e. S(t, w
0
) is the resolving semigroup of equation (5.15) (see (5.17)).
This theorem follows form results of [2, Ch. 5, Sect. 2; Ch. 7, Sect. 5] and from
Theorem 5.1.
5.3. Extension operator
Here we construct extension operator for NavierStokes equations. This operator
is analog of extension operators (4.18), (4.48) constructed for Oseen equations.
Recall that the domain and its extension G satisfy (4.1), (4.2) and Conditions
4.1, 4.2. Besides
V
1
(,
0
) = u(x) V
1
() : u[
0
= 0, v V
1
0
(G) that u(x) =
v(x). (5.39)
This space is equipped with the norm
|u|
V
1
(,
0
)
= inf
Eu
|Eu|
V
1
0
(G)
(5.40)
where inmum is taken over all bounded extension operators
E : V
1
(,
0
) V
1
0
(G).
Let us prove the theorem on extension.
Theorem 5.3. Let a(x) be a steady-state solution of (5.18), v(x) =
a, and M
is the invariant manifold constructed in neighbourhood a + O of a in V

1
0
(G) in
Theorem 5.2. Let B
1
= v
0
V
1
(,
0
) : |v
0
v|
V
1
(,
0
)
<
1
. Then for
suciently small
1
one can construct a continuous operator
Ext : v +B
1
M
, (5.41)
which is operator of extension for vector elds from to G:
(Extv)(x) v(x), x . (5.42)
Proof. By denition of V
1
(,
0
) there exists a bounded linear extension operator
R : V
1
(,
0
) V
1
0
(G).
Let (x) C
(

G), (x) 1 for x and (x) 0 outside a neighbourhood
of . Similarly to (4.21) we introduce the following operator of extension:
Qv(x) =
_
v(x), x ,
rot(rot
1
Rv)(x) +w(x), x ,
(5.43)
where w(x) is a vector eld concentrated in which is constructed by v(x). We
describe its construction below. At last we dene the desired operator Ext by the
formula
Extv =
Qz +g(
Qz) +a, with z = v a, (5.44)

where
= I
+
,
+
is operator (5.35) of projection on X
+
=
+
V
1
0
(G),
X
V
1
0
(G), and g : X
X
+
is the operator constructed in Theorem 5.2.
By denition (5.36) of M
we have Extv M
. Hence we have to provide that

the equality
(Extv)(x) v(x), x (5.45)
is true, which shows that Ext is an extension operator. By (5.35) e
j
(x) generates
X
+
and therefore the map g(u) can be written in the form
g(u) =
N
j=1
e
j
g
j
(u).
That is why taking into account (5.35) we can rewrite (5.44) in the form
Extv = a(x) +Qz(x)
K
j=1
e
j
(x)
_
Q
Qz(y)
j
(y) dy +
K
j=1
e
j
(x)g
j
(
Qz), (5.46)
(z = v a).
In virtue of Lemma 5.2 e
j
(x), x are linear independent and therefore
(5.45), (5.46) imply
_
G
Qz(x)
j
(x) dx = g
j
(
Qz), j = 1, . . . , K. (5.47)
Let (x) C
(

G), (x) = 1 outside small neighbourhood of and (x) = 0,
x . Similarly to (4.21) we look for the vector eld w(x) from (5.43) in the form
w(x) = rot
_
(x)
K
k=1
c
k
rot
k
(x)
_
. (5.48)
To nd coecients (c
1
, . . . , c
K
) c we substitute (5.48) into (5.47) taking into
account (5.43), (5.35). As a result we get
z +Ac = g
_
z + (c, rot[rot
]) (e, z +Ac)
_
, (5.49)
where z = rot(rot
1
Rz), z = (z
1
, . . . , z
K
), A = |a
jk
| and
z
j
=
_
G
rot(rot
1
Rz(x))
j
(x) dx, a
jk
=
_
G
rot[(x)rot
k
(x)]
j
(x) dx,
g(u) = (g
1
(u), . . . , g
K
(u)),
= (
1
(x), . . . ,
K
(x)), e = (e
1
(x), . . . , e
K
(x)), (c, e) =
K
j=1
c
j
e
j
.
We prove as in Theorem 4.1 that matrix A = |a
jk
| is positive dened and therefore
it is invertible.
Applying to both parts of (5.49) the matrix A
1
we get the equality
c = G
z
(c) (5.50)
where the map G
z
: R
K
R
K
is dened by the relation
G
z
(c) = A
1
g(z (e, z) + (c, rot[rot
]) (e, Ac)) A
1
z. (5.51)
In virtue of Theorem 5.2 the map A
1
g : R
K
R
K
belongs to the class C
1+1/2
and A
1
g(0) = 0, A
1
g
(0) = 0. Therefore for suciently small |c

1
|
R
K, |c
2
|
R
K,
|z|
V
1
0
(G)
we derive from (5.51) that
|G
z
(c
1
) G
z
(c
2
)| sup
[0,1]
|A
1
g
(z (e, z)+
+ (c
1
+ (1 )c
2
, rot[rot
]) (e, A[c
1
+ (1 c
2
)]))| |c
1
c
2
|
(z, c
1
, c
2
)|c
1
c
2
|, where (z, c
1
, c
2
)
1
(|z|
1/2
V
1
0
(G)
+|c
1
|
1/2
R
K
+|c
2
|
1/2
R
K
),
and
1
> 0 is a constant. Therefore the map G
z
is a contraction one. Hence
by contraction mapping principle ([11]) equation (5.50) has a unique solution c =
(c
1
, . . . , c
K
) if |z|
V
1
0
(G)
is suciently small. For these |z|
V
1
0
(G)
the operator Ext
dened in (5.44), (5.43), (5.48) is the desired extension operator.
5.4. Theorem on stabilization
First of all we make more precise conditions connected with solution ( v, p) of
steady-state problem (5.5), (5.6). We set
V
2
(,
0
) = v(x) V
2
() : v[
0
= 0, w V
2
(G) V
1
0
(G), v(x) =
w
(5.52)
where
is the operator of restriction on .

Proposition 5.1. Let f (L
2
())
2
and a pair ( v(x), p(x)) belong to V
2
(, )
(L
2
())
2
and satises equations (5.5), (5.6). Then there are extension g(x)
(L
2
(G))
2
of f(x) from to G and an extension (a(x), q(x)) (V
2
(G)V
1
0
(G))
L
2
(G) of ( v(x), p(x)) from to G such that the pair (a(x), q(x)) is a solution
of (5.11), (5.12).
Proof. By denition of the space V
2
(,
0
) there exists an extension a(x)
V
2
(G) V
1
0
(G) of v(x) from to G. Since p L
2
(), one can choose p(x)
H
1
() and after that extend p(x) from to G up to a function q(x) H
1
(G).
We substitute (a(x), q) to the left side of the rst equation in (5.11) and denote
the obtained right side as g(x). Evidently g(x) L
2
(G). Moreover
g(x) = f(x)
because
(a(x), q(x)) = ( v(x), p(x)), and (5.5) is true.

Remark 5.1. Condition v V
2
(,
0
) which we impose on a given steady-state
solution is not constructive. But using arguments expounded in Subsection 4.2 one
can prove that vector eld v(x) belongs to V
2
(,
0
) if it satises the following
condition: v(x) V
2
(), v[
0
= 0 and v(x) is suciently smooth in a neighbour-
hood of . Moreover we suppose that assumptions (4.1), (4.2), Conditions 4.1,
4.2 are fullled and in Condition 4.2 restriction 3 holds.
We now are in position to prove the main result of this paper.
Theorem 5.4. Let R
2
be a bounded domain with C
-boundary and =
0

1
, where ,
0
are open curves, ,= , is a nite number of points
or = . Suppose that a domain G R
2
is chosen such that assumptions (4.1),
(4.2) and Conditions 4.1, 4.2 with 3 are fullled. Let f(x) (L
2
())
2
and
( v(x), p(x)) V
2
(,
0
) (L
2
())
2
satisfy (5.5), (5.6). Then for an arbitrary
> 0 there exists suciently small
1
> 0 such that for each v
0
V
1
(,
0
)
satisfying
| v v
0
|
V
1
0
()
<
1
(5.53)
there exists a feedback boundary control u(t, x) R
+
which stabilized
NavierStokes boundary value problem (5.1)(5.4) with the rate (5.7), i.e. the so-
lution v of (5.1)(5.4) satises (5.7).
Proof. Using Proposition 5.1 we extend v(x) to a(x) V
1
(G), and f(x) to
g(x) (L
2
(G))
2
. As a result we get boundary value problem (5.8)(5.10) (with
certain w
0
) and steady-state solution (a(x), q(x)) of this problem. We can sup-
pose that > 0 satises (3.55): otherwise we increase a little bit and get (3.55).
In virtue of Theorem 5.2 in a neighbourhood of a there exists a manifold M
which
is invariant with respect of semigroup S(t, w
0
) connected with equation (5.15) and
such that for each w
0
M
inequality (5.38) holds. Let

1
be so small that it
satises condition of Theorem 5.3. Then we apply extension operator Ext con-
structed in Theorem 5.3 to initial condition v
0
of problem (5.1)(5.3) and take
w
0
= Extv
0
as initial condition for problem (5.8)(5.10) or for equation (5.15)(that
is equivalent). Then since w
0
M
, S(t, w
0
) M
for each t 0, and esti-

mate (5.38) holds. We dene solution (v, u) of stabilization problem (5.1)(5.4)
by formula (5.13) where w(t, x) = S(t, w
0
) is the solution of (5.15), (5.9). Then
(5.7) follows from (5.13), (5.38)
References
[1] A. Balogh and M. Krstic, Burgers equation with Nonlinear boundary Feedback: H
1
Stability, Well-posedness and Simulation, Math. Probl. in Engineering 6 (2000), 189200.
[2] A. V. Babin and M. I. Vishik, Attractors of evolution equations, North-Holland, Amster-
dam, London, New York, Tokio, 1992.
[3] Coron J.-M., On null asymptotic stabilization of the two-dimensional incompressible Euler
equations in a simply connected domains, SIAM J. Control Optim. 37 no. 6 (1999), 1874
1896 (electronic).
[4] N. Dunford and J. T. Schwartz, Linear Operatos. Part 1. General Theory, Interscience
Publishers, New York, London, 1958.
[5] A. V. Fursikov, Optimal control of distributed systems. Theory and applications, Trans-
lations of Math. Monographs 187 Amer. Math. Soc., Providence, Rhode Island, 2000.
[6] A. V. Fursikov, Stabilizability of quazilinear parabolic equation by feedback boundary
control, Matem. Sbornik 192 no. 4 (2001), 115160 (in Russian).
[7] A. V. Fursikov and O. Yu. Emanuilov, Exact controllability of NavierStokes and
Boussinesq equations, Russian Math. surveys 54 (3) (1999), 565618.
[8] D. Henry, Geometric theory of semilinear parabolic equations, Lecture Notes in Mathem.
840, Springer-Verlag, 1981.
[9] L. Hormander, Linear Partial Dierential Operators, Die Grundlehren der Mathematis-
chen Wissenschaften 116, Springer-Verlag, 1963.
[10] M. V. Keldysh, On completeness of eigenfunctions for certain classes of not self-adjoint
linear operators, Russian Math Surveys 26 (4) (1971), 1541 (in Russian).
[11] A. N. Kolmogorov and S. V. Fomin, Introductory real analysis, N.Y. Dover Publications,
1975.
[12] V. Komornik, Rapid boundary stabilization of linear distributed systems, SIAM J. Control
Optim. 35 (1997), 15911613.
[13] O. A. Ladyzhenskaya, The Mathematical Theory of Viscous Incompressible Flow, Gordon
and Breach, New York, 1963.
[14] J. Lagnese, Boundary stabilization of thin plates, SIAM, Philadelphia, 1989.
[15] J.-L. Lions, Exact controllability, stabilizability, and perturbations for distributed systems,
SIAM Rev. 30 (1988), 168.
[16] V. A. Solonnikov, A priori Estimates for Second-Order Parabolic Equations, Transl.
Amer. Math. Soc. Ser. 2 65 (1967), 51131.
[17] R. Temam, NavierStokes equations, North-Holland, Amsterdam, 1979.
[18] M. I. Vishik and A. V. Fursikov, Mathematical Problems of Statistical Hydromechnics,
Kluwer Academic Publishers, Dordrecht, Boston, London, 1988.
[19] I. I. Vorovich and V. I. Yudovich, Steady Flow in a viscous Incompressible Fluid, Mat.
Sbornik 53 no. 4 (1961), 393428 (in Russian).
[20] K. Yosida, Functional Analysis, Springer-Verlag, 1965.
A. V. Fursikov
Department of Mechanics and Mathematics
Moscow State University
119899 Moscow
Russia
e-mail: fursikov@dial01.msu.ru
(accepted: November 24, 2000)
To access this journal online:
http://www.birkhauser.ch

Stabilizability of Two-Dimensional Navier-Stokes Equations With Help of A Boundary Feedback Control

Uploaded by

Copyright:

Available Formats

Stabilizability of Two-Dimensional Navier-Stokes Equations With Help of A Boundary Feedback Control

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Stabilizability of Two-Dimensional Navier-Stokes Equations With Help of A Boundary Feedback Control

Uploaded by

Copyright:

Available Formats

J. math. uid mech.

, is constructed such that solution v(t, x)

= 0 and v is an unstable singular point for the

-smooth connected boundary :

) is a control, dened on the lateral surface = (0, ) of

) such that the solution

are restriction operators on and on correspondingly,

v(s) (s)ds = 0, (2.19)

(), (x) = 0 for x

are coecients, which are determined from the system of equations

and therefore extension (2.23) is well dened.

() and by (2.28) operator (2.17) is bounded.

() of the solution v(t, x) we do the same. Thus all

and in all points except

smoothness. The exact conditions on

is well-dened in the Sobolev space

to Oseen operator (3.18):

is a closed operator with domain coinciding to T(A)

) = (A) and R(, A)

are compact operators simultaneously, (3.29), (3.31) imply that

is a sectorial operator with a compact resolvent and (A) = (A) = (A

: H H is adjoint to A. Suppose that

) correspondingly which have

and both of them

. Then by (3.44) with

from (3.38) which

corresponding to the eigenvalue

that is placed in C : Re and constructed from an interval of the line

and (3.21) imply that for suciently large R

(A) and (3.22) where A is changed on A and

)f(x) = 0 holds for x G and therefore by (3.27)

f(x) = p(x), x G (3.68)

f(x) (a(x), )rot

and for each point P there

), x (0, )(x, y) = (x, 0), x (, 0), > 1. (4.4)

is the operator of restriction on for functions

v the operator formally conjective to (4.8). It is

rot we get by (4.12), (4.13), (4.14) that

and by Lemma 4.2

for x 0, (x) = 0 for x 0 such that extended function L

(0) = 0, (0) = 1 (4.34)

(G). Evidently, the vector eld Lv(x) rot(

is the operator of restriction on . Besides,

is the derivative along the vector eld tangent to G.

is derivative along vector eld tangential to G.

), y > 0, x (0, 1).

) and therefore the inequality is true:

is dene lower formula (3.55).

and the contour

is dened lower (3.55).

Proof of Lemma 4.7. In virtue if Lemma 3.1, denition of contour

, || 1 can be obtained easily also

and [[ 1 then by denition of

(see lower (3.55))

estimate (4.56) is true because

are operators of restriction of a

(0, 1) such that

. Then ii) also follows from [13], [17].