BOUNDARY CONTROLLABILITY IN PROBLEMS
OF TRANSMISSION FOR A CLASS OF
SECOND ORDER HYPERBOLIC SYSTEMS
JOHN E. LAGNESE
Abstract. We consider transmission problems for general second or- der linear hyperbolic systems having piecewise constant coecients in a bounded, open connected set with smooth boundary and controlled through the Dirichlet boundary condition. It is proved that such a sys- tem is exactly controllable in an appropriate function space provided the interfaces where the coecients have a jump discontinuity are all star-shaped with respect to one and the same point and the coecients satisfy a certain monotonicity condition.
1. Introduction and main results
Let , 1be bounded, open, connected sets inRnwith smooth boundaries
; and ;1, respectively, such that 1 , and set 2 = n1, whose boundary is ;2 = ;;1. Let
m
andn
be positive integers. Forik
= 1::: m
, forj`
= 1::: n
and for = 12, leta
ijk` be real numbers having the symmetrya
ijk` =a
k`ij = 12ik
= 1::: m j`
= 1::: n:
(1.1) Whenm
=n
we further assume thata
ijk` =a
jik` =a
ij`k = 12ijk`
= 1::: n:
(1.2) Thea
ijk` are also assumed to satisfy the following ellipticity condition:if
m
6=n
thena
ijk` ijk`c
0ijijc
0>
0 8(ij)2Rm n (1.3) ifm
=n
thena
ijk` ijk`c
0ijijc
0>
0 8(ij)2S
n (1.4) whereS
n denotes the set of all real symmetricn
n
matrices. In (1.3), (1.4) and in what follows, a repeated Roman index indicates summation with respect to that index over an appropriate range (either 1 tom
or 1 ton
, as the context suggests), unless otherwise noted.Fix
T >
0 and letW
1 =
0
B
@
w
11w
...1m1
C
A
W2 =0
B
@
w
21w
...2m1
C
A
Department of Mathematics, Georgetown University, Washington, DC 20057 USA.
Received by the journal April 15, 1997. Accepted for publication October 13, 1997.
c Societe de Mathematiques Appliquees et Industrielles. Typeset by LATEX.
be vector functions dened on the cylinders
Q
1 = 1(0T
) andQ
2 = 2 (0T
), respectively. We set = ;(0T
) and 1 = ;1(0T
) and consider the following problem of transmission:(
w
1i;a
ijk`1w
1k`j= 0 inQ
1w
2i;a
ijk`2w
2k`j= 0 inQ
2i
= 1::: m
(1.5) where _=@=@t
and where a subscript following a comma indicates dieren- tiation with respect to the corresponding spatial variableW
2 =U on
(1.6)w
1i=w
2ia
ijk`1w
1k`j =a
ijk`2w
2k`j on 1i
= 1::: m
(1.7) where = (1:::
n) is the unit normal to ;1 pointing into 1 (and towards the exterior of 2),Wjt=0= _Wjt=0= 0 in ,
= 12:
(1.8) The function U in (1.6) is viewed as a control function which is chosen to eect the evolution of the dynamics. More precisely, we consider below the reachability problem, which is to determine the setRT =f(W1(
T
)W2(T
)W_ 1(T
)W_ 2(T
)) : U2Ug where U is an appropriate space of controls.The question of boundary controllability in problems of transmission has been considered by several authors. In particular, J.-L. Lions considered the system (1.5) - (1.8) in the special case of two wave equations, that is, he considered the problem
(
w
1;a
1w
1 = 0 inQ
1w
2;a
2w
2 = 0 inQ
2 (1.9) wherea
>
0 and is the ordinary Laplacian in Rn,w
2 =u
on (1.10)w
1 =w
2a
1@w
1@
=a
2@w
2@
on 1 (1.11)w
jt=0= _w
jt=0= 0 in ,= 12:
(1.12) For this problem the following result is proved in 5, Chapter VI].Theorem 1.1. Assume that ;1 is star-shaped with respect to some point
x
0 21 and set ;(x
0) =fx
2;j(x
;x
0)>
0g, (x
0) = ;(x
0)(0T
), where is the unit outer normal to;. LetRT =f(
w
1(T
)w
2(T
)w
_1(T
)w
_2(T
)) :u
2L
2()u
= 0 on n(x
0)g:
Ifa
1> a
2 andT > T
(x
0) := 2R
(x
0)=
pa
2, whereR
(x
0) = maxx22j
x
;x
0j, then RT =H
V
0 whereH
=L
2(1)L
2(2)V
=f('
1'
2)2H
1(1)H
1(2) :'
2j;= 0'
1j;1 ='
2j;1g:
ESAIM: Cocv, , ,
The main purpose of this paper is prove the following generalization of Theorem 1.1 to the system (1.5) - (1.8). Let us dene
j
A
2j=8
>
>
>
>
<
>
>
>
>
:
inf
06=(ij)2Rm n
a
ijk`2 klij ijijm
6=n
inf06=(ij)2Sn
a
ijk`2 klij ijijm
=n:
(1.13)Theorem 1.2. Assume that ;1 is star-shaped with respect to some point
x
021 and let;(x
0), (x
0) andR
(x
0) be as in Theorem 1.1. LetRT=f(W1(
T
)W2(T
)W_ 1(T
)W_ 2(T
)):U2L
2()]mU=0 on n(x
0)g:
If8
<
:
a
ijk`1 ijk`a
ijk`2 ijk` 8(ij)2Rm nm
6=n
a
ijk`1 ijk`a
ijk`2 ijk` 8(ij)2S
nm
=n
(1.14) and ifT > T
(x
0) =8
>
>
>
>
<
>
>
>
>
:
2
R
(x
0)p
j
A
2jm
6=n
2p2R
(x
0)p
j
A
2jm
=n:
(1.15) then RT =H
V
0 whereH
=L
2(1)]mL
2(2)]mV
=f(12)2H
1(1)]mH
1(2)]m : 2j;= 0 1j;1 = 2j;1g:
Remark 1.3. Whether or not the monotonicity condition (1.14)is necessary for exact controllability in dimension
n
2 is unknown, even in the case of ordinary wave equations considered in Theorem 1.1. It is known that no such condition is necessary for exact boundary controllability of transmission problems in dimension one or, more generally, for exact controllability of 1- dimensional networks in R3 3, Chapter IV].Remark 1.4. Set
H
;1(2) =f'
2H
1(2) :'
j;= 0g:
The space
V
is a closed subspace ofH
1(1)]mH
;1(2)]m andV
0 may be identied with the restriction toV
of elements in the product space (H
1(1))0]m(H
;1(2))0]m (cf. 5, Remark 5.1, p. 376].Remark 1.5. With only minor modications in the proof, Theorem 1.2 can be extended to the situation involving and
p
2 nested open sets!
1::: !
p with!
i!
i+1i
= 1::: p
;1!
p=:
Set 1 =!
1 i =!
in!
i;1i
= 1::: p:
Setting ;i:=
@!
i, one should assume that ;iis star-shaped with respect to a pointx
0 21,i
= 1::: p
;1, and that the parametersa
ijk` ,= 1::: p
,ESAIM: Cocv, , ,
associated with satisfy appropriate monotonicity conditions with respect to
.Example 1.6. (Wave equations) Suppose that
m
= 1. In this case the system (1.5) may be written(
w
1;a
j`1w
1j` inQ
1w
2;a
j`2w
2j` inQ
2where
a
j`=a
`j. If we setA
= (a
j`), the condition (1.14) withm
= 1< n
becomes the requirement that the matrixA
1;A
2 be positive semidenite.The control time
T
(x
0) is given by 2R
(x
0)=
pjA
2j, wherej
A
2j= inf06=2Rn
A
2j
j2:
In particular, if
A
=a
I
, whereI
is the identity matrix in Rn, we recover Theorem 1.1.Example 1.7. (Linear isotropic elastodynamics) Here
n
=m
= 3 and the system of equations is (we do not use summation convention in this example)8
>
>
>
>
>
<
>
>
>
>
>
:
1w
1i;1w
1i;(1+1)X3j=1
w
1jji= 0 inQ
12
w
2i;2w
2i;(2+2)X3j=1
w
2jji= 0 inQ
2i
= 123where
is the (constant) mass density per unit of reference volume and, are the Lam e parameters for the elastic material occupying the region . We assume that>
0 and 0:
(1.16)By a change of notation we may assume that
1 =2= 1. Then the nonzero coe!cientsa
ijk` are given bya
iiii =+ 2i
= 123a
ijij =a
ijji =a
iijj =i
6=j
= 123 and therefore for (ij)2S
3 we haveX
a
ijk` ijk`= (+ 2)X3i=1
2ii+3
X
iji=1
6=j
iijj+2
3
X
iji=1
6=j
(
ij +ji)2=
3
X
i=1
ii!
2
+ 2
3
X
ij=1
2ij 23
X
ij=1
ij2:
In the same way, the monotonicity condition (1.14) (with strict inequality) will be satised if
1 2 and 1 2:
The control time
T
(x
0) is given by 2p2R
(x
0)=
pjA
2j= 2R
(x
0)=
p2, which is what is expected since the two wave speeds associated with the region 2 are p2 and p2+ 22.ESAIM: Cocv, , ,
Let us brie"y recount other works on boundary controllability in problems of transmission or, more generally, in interface problems. These works are for the most part disjoint from the present paper. For one-dimensional interface problems (for example, string or beam equations on one-dimensional graphs in R3), one may refer to 3] for a comprehensive analysis of the reachabil- ity problem utilizing controls supported at the nodes of the graph. These systems, however, do not generally fall within the framework of the prob- lem (1.5) - (1.8). For higher dimensional systems, one may cite 2], where boundary controllability in transmissions problems for the Reissner-Mindlin system of thin plate theory is considered. This system consists of two cou- pled second order equations with constant coe!cients but with lower order terms as well, so they too fall outside of the scope of the present work. There are also several works dealing with boundary controllability in problems of transmission or in interface problems for the wave equation in the singular case. If, for example, ;1\; 6= in the problem (1.9) - (1.12) above, the solution will generally suer a loss of regularity in a neighborhood of points belonging to this set. In this case the proof technique for Theorem 1.1 given in 5] fails that is the reason for the assumption 1 . In the singular case, we can refer to two recent papers of S. Nicaise 6], 7], where the question of exact boundary controllability for the wave equation on two-dimensional networks in Rnis treated. In these papers it is shown, among other things, that a result analogous to Theorem 1.1 may be obtained provided that, in addition to boundary controls, controls are also imposed near the singular vertices. The reader is referred to the cited papers for the precise statements of Nicaise's results. References to other works related to the topic of the present paper may be found in the bibliographies of the books and papers cited above.
2. A priori estimates Let
H
andV
be as in Theorem 1.2. Fork
0 setH
k() =
H
k(1)]mH
k(2)]m:
Let := (1
2), $ := ($1$2) be elements ofV
and consider the bilinear form ($) =X2=1
Z
a
ijk`'
k`ijdx
$2V:
This form is continuous on
V
V
in the topology induced byH1() and is symmetric by virtue of (1.1). Ifm
6=n
, is also coercive onV
because of the ellipticity assumption (1.3) and 2j; = 0, that is, for somek >
0, ()k
kk2H1() 82V:
On the other hand, if
m
=n
, because of the symmetry assumptions (1.2) we may write ($) =X2=1
Z
a
ijk`"
k`()"
ij($)dx
ESAIM: Cocv, , ,
where
"
k`() = 12('
k`+'
`k):
If 2V
then the functions'
i(x
) =(
'
1i(x
)x
21'
2i(x
)x
22i
= 1::: m
are in
H
01() and therefore the following Korn's inequality holds 8, Theorem 2.1]:2
X
=1
Z
"
ij()"
ij()dx
12X2=1
Z
'
ij'
ijdx:
(2.1) It follows from (2.1) and (1.4) that ()c
o2
X
=1
Z
"
ij()"
ij()dx
k
kk2H1() 82V
Therefore in all cases ()]1=2denes a norm onV
equivalent to the one induced byH1(). We retopologizeV
with this norm and letV
0denote the dual space ofV
. WithH
identied with its dual space we have the dense and continuous embeddingsV H V
0.Let us consider the homogeneous boundary value problem
(
'
1i;a
ijk`1'
1k`j= 0 inQ
1'
2i;a
ijk`2'
2k`j= 0 inQ
2i
= 1::: m
(2.2)2= 0 on
(2.3)'
1i ='
2ia
ijk`1'
1k`j =a
ijk`2'
2k`j on 1i
= 1::: m
(2.4) jt=0= 0 _jt=0= 1 in , = 12:
(2.5) Assume that 0 := (0102) 2V
, 1 := (1112) 2H
. According to standard variational theory (2.2) - (2.5) has a unique solution in the fol- lowing sense: there is a unique function := (12) 2C
(0T
]V
) \C
1(0T
]H
)\C
2(0T
]V
0) such thath
$iV +($) = 0 8$2V
0< t
T
and which satises (2.5), where h
$iV denotes the pairing between ele- ments 2V
0 and $2V
in theV
0;V
duality. Furthermore, the mappingS
(t
) : (01)7!((t
) _(t
)) is a denes a unitary group onV
H
, that isk(
t
)k2V +k_(t
)k2H =k0k2V +k1k2Ht
0:
(2.6) LetA
denote the canonical isomorphism ofV
ontoV
0, so that ($) =h
A
$iV,8$2V
, and setD
(A
) =f$2V
:A
$2H
g k$kD(A)=kA
$kH:
If (0
1)2D
(A
)V
, then 2C
(0T
]D
(A
))\C
1(0T
]V
)\C
2(0T
]H
) andS
(t
) is a unitary group onD
(A
)V
. Because of our assumptions on the regions and 1, an element 2D
(A
) has the regularity 2ESAIM: Cocv, , ,
H ()\
V
andA
=;(a
ijk`1'
1kj`a
ijk`2'
2kj`). Moreover, satises both of the transmission conditions in (2.4) in the trace sense and the normk kD(A) is equivalent to that induced on
D
(A
) byH2(). These results are standard and may be proved by the usual methods.Lemma 2.1. Suppose that (0
1) 2D
(A
)V
, and leth
2W
11()]n. Then the solution of (2.2) - (2.5) satises the following identity:2
X
=1
Z
'
_i(h
r'
i)dx
T
0
+ 12
Z
Q(div
h
) _'
i'
_i;a
ijk`'
k`'
ij]dxdt
+Z
Q
a
ijk`'
k`'
iph
pjdxdt
= 12
Z
(
h
)a
ijk`2@'
2k@
@'
2i@
j`
d
+ 12Z
1
(
h
)h(a
ijk`1 ;a
ijk`2 )'
1ij'
1k`+
a
ijk`2@'
1i@
;@'
2i@
@'
1k@
;@'
2k@
j`
d :
(2.7) Proof. The proof utilizes classical multipliers. Form0 =
Z
Q(
'
i;a
ijk`'
k`j)h
r'
idxdt
(2.8) and integrate by parts. One hasZ
Q
'
i(h
r'
i)dxdt
=Z
'
_i(h
r'
i)dx
T
0
+ 12
Z
Q(div
h
) _'
i'
_idxdt
;12Z(
h
) _'
i'
_id
(2.9) where =@
(0T
) and is the unit normal vector pointing into the exterior of . In addition,Z
Q(
a
ijk`'
k`j)h
r'
idxdt
=Z
a
ijk`'
k`'
iph
pjd
; Z
Q
a
ijk`'
k`'
iph
pjdxdt
;ZQ
a
ijk`'
k`'
ipjh
pdxdt:
From the symmetry property (1.1) we obtain
a
ijk`'
k`'
ipj = 12(a
ijk`'
k`'
ij)pESAIM: Cocv, , ,
and therefore
Z
Q(
a
ijk`'
k`j)h
r'
idxdt
=
Z
a
ijk`'
k`'
iph
pjd
;1 2Z
(
h
)a
ijk`'
k`'
ijd
; Z
Q
a
ijk`'
k`'
iph
pjdxdt
+ 12ZQ(div
h
)a
ijk`'
k`'
ijdxdt:
(2.10) We then obtain from (2.8) - (2.10) the identityZ
'
_i(h
r'
i)dx
T
0
+
Z
Q
a
ijk`'
k`'
iph
pjdxdt
+ 12Z
Q(div
h
) _'
i'
_i;a
ijk`'
k`'
ij]dxdt
= 12
Z
(
h
)( _'
i'
_i;a
ijk`'
k`'
ij)d
+Z
a
ijk`'
k`'
iph
pjd :
(2.11) We sum (2.11) over = 12, noting the relations'
1ij1 ='
2ij1'
2ijj =@'
2i@
ji
= 1::: m j
= 1::: n
and observing that along the boundary region ;1 that is common to 1 and 2 the unit normal vector pointing out of 2 points into 1. We obtain2
X
=1
(
Z
'
_i(h
r'
i)dx
T
0
+Z
Q
a
ijk`'
k`'
iph
pjdxdt
+ 12Z
Q(div
h
)( _'
i'
_i;a
ijk`'
k`'
ij)dxdt
= 12
Z
(
h
)a
ijk`2@'
2k@
@'
2i@
j`
d
+ 12Z
1
(
h
)(a
ijk`1'
1k`'
1ij;a
ijk`2'
2k`'
2ij)d
+Z
1
(
a
ijk`2'
2k`'
2ip;a
ijk`1'
1k`'
1ip)h
pjd :
(2.12) To complete the proof we need to show that the last two integrals in (2.12) agree with the last integral in (2.7).Consider the last integral in (2.12). By choosing suitable extensions of 1 and 2 we may assume that
'
i2H
2() for= 12 andi
= 1::: n
. We have(
a
ijk`2'
2k`'
2ip;a
ijk`1'
1k`'
1ip)h
pj=
a
ijk`2'
2k`jh
p('
2ip;'
1ip) = (h
)a
ijk`2'
2k`j@'
2i@
;@'
1i@
(2.13)
ESAIM: Cocv, , ,
on 1, since (
'
2ip;'
1ip)j;1 =@'
2i@
;@'
1i@
p
i
= 1::: m p
= 1::: n:
(2.14) It also follows from (2.14) that on 1a
ijk`1'
1k`'
1ij =a
ijk`1'
1k`
'
2ij+@'
1i@
;@'
2i@
j
and
a
ijk`2'
2k`'
2ij=a
ijk`2'
2k`
'
1ij;@'
1i@
;@'
2i@
j
:
If the last expression is subtracted from the preceding one we obtaina
ijk`1'
1k`'
1ij;a
ijk`2'
2k`'
2ij= (
a
ijk`1 ;a
ijk`2 )'
1k`'
2ij+ 2a
ijk`2'
2k`j@'
1i@
;@'
2i@
on 1
(2.15) where we have used the symmetry property (1.1) and the second transmis- sion condition in (2.4). It follows from (2.13) and (2.15) that12
Z
1
(
h
)(a
ijk`1'
1k`'
1ij;a
ijk`2'
2k`'
2ij)d
+Z
1
(
a
ijk`2'
2k`'
2iph
pj;a
ijk`1'
1k`'
1iph
pj)d
= 12
Z
1
(
h
)(a
ijk`1 ;a
ijk`2 )'
1k`'
2ijd :
(2.16) We use (2.14) once again to obtain on 1(
a
ijk`1 ;a
ijk`2 )'
1k`'
2ij=(
a
ijk`1 ;a
ijk`2 )'
1k`'
1ij+ (a
ijk`1 ;a
ijk`2 )'
1k`j@'
2i@
;@'
1i@
=(
a
ijk`1 ;a
ijk`2 )'
1k`'
1ij+a
ijk`2'
2k`j@'
2i@
;@'
1i@
;
a
ijk`2'
1k`j@'
2i@
;@'
1i@
=(
a
ijk`1 ;a
ijk`2 )'
1k`'
1ij+a
ijk`2 ('
2k`;'
1k`)j@'
2i@
;@'
1i@
=(
a
ijk`1 ;a
ijk`2 )'
1k`'
1ij+a
ijk`2@'
2k@
;@'
1k@
@'
2i@
;@'
1i@
j`
:
(2.17) Now substitute (2.17) into (2.16).Theorem 2.2. Let (0
1) 2V
H
. Then the solution of (2.2) - (2.5) satisesZ
a
ijk`2@'
2k@
@'
2i@
j`
d
C
(T
+ 1)(k0k2V +k1k2H) (2.18) for some constantC
.ESAIM: Cocv, , ,
Proof. By density it su!ces to prove (2.18) for (
) 2D
(A
)V
. Since; and ;1 are smooth, there is a vector eld
h
2C
1()]n such thath
= on ; andh
= 0 on ;1 (see Komornik 1, Lemma 2.1]). We utilize this vector eld in (2.7) and observe that the left side of (2.7) is bounded above in absolute value byC
o(k_(T
)k2H+k(T
)k2V+k1k2H+k0k2V)+C
1T
sup0T](k(
t
)k2V+k_(t
)k2H)C
(T
+ 1)(k1k2H+k0k2V) in view of (2.6).Remark 2.3. If
m
6=n
is follows from (2.18) and (1.3) that@'
2i@
2L
2()i
= 1::: m:
(2.19) Ifm
=n
, then (2.18) and (1.4) imply that@'
2i@
j +@'
2j@
i 2L
2()ij
= 1::: n
so, in particular, i@'
2i=@
2L
2() for eachi
= 1::: n
. Sincen
X
ij=1
@'
2i@
j+@'
2j@
i
2 = 2Xn
i=1
@'
2i@
2+ 2 Xn
ij=1
ij@'
2i@ @'
2j@
and the second sum is inL
1(), it follows that (2.19) holds true also in this case.Theorem 2.4. Assume that ;1 is star-shaped with respect to some point
x
021 and suppose that the coecientsa
ijk` satisfy (1.14). Let (01)2V
H
. Then the solution of (2.2) - (2.5) satises(
T
;T
(x
0))(k1k2H+k0k2V)Z
(x0)(
h
)a
ijk`2@'
2k@
@'
2i@
j`
d
(2.20) whereT
(x
0) is given by (1.15) andh
(x
) =x
;x
0.Proof. Because of (2.18) it su!ces to prove (2.20) for (0
1)2D
(A
)V
. We apply (2.7) withh
(x
) =x
;x
0. Thenh
0 on ;1. On the other hand, assumption (1.14) and the standing assumption (1.1) (or (1.2)) imply that(
a
ijk`1 ;a
ijk`2 )'
1ij'
1k`+a
ijk`2@'
1i@
;@'
2i@
@'
1k@
;@'
2k@
j` 0
:
Therefore2
X
=1
Z
'
_i(h
r'
i)dx
T
0
+
n
2Z
Q
'
_i'
_idxdt
+ (1;n
2)
Z
Q
a
ijk`'
k`'
jdxdt
+Z
Q
a
ijk`'
k`'
iph
pjdxdt
12
Z
(x0)(
h
)a
ijk`2@'
2k@
@'
2i@
j`
d :
(2.21)ESAIM: Cocv, , ,