Numerical Approximation of the Exact Control

for the String Equation

M. A. Rincon
1
, M. Zegarra Garay
2
, M. Milla Miranda
3
1
Instituto de Matem atica, Universidade Federal do Rio de Janeiro, 21945-970, Brazil
E-mails: rincon@dcc.ufrj.br
2
Instituto de Matem atica, Universidade Federal do Rio de Janeiro, 21945-970, Brazil
E-mails: mgaray@posgrad.nce.ufrj.br
3
Instituto de Matem atica, Universidade Federal do Rio de Janeiro, 21945-970, Brazil
E-mails: milla@im.ufrj.br
Abstract
In this paper we implement the results obtained by Vasilyev et al [11] on the numerical
approximation of the exact control for the string equation. The computational part and the
respective graphs are made for a particular case. For that we have applied the Residues
Theorem of holomorphic functions, which, as far as we know, is the rst time that this
theorem is applied in the computational study of exact control problems.
AMS Subject Classication: 65M99, 93C20
Keywords: exact control, numerical series, Fourier series, residue theorem.
1 Introduction
Let us consider a exible elastic string of length L. Then its small transversal vibration can
be studied by the following problem:
_
_
_
y
tt
(x, t) y
xx
(x, t) = 0, in Q = (0, L) (0, T)
y(0, t) = v(t), y(L, t) = 0, 0 t T
y(x, 0) = y
0
(x), y
t
(x, 0) = y
1
(x), 0 x L
(1)
where y(x, t) denote the displacement of the point x of the string (0 x L), at the instant
t, (0 t T). The functions y
0
(x) and y
1
(x) represent, respectively, the initial position
and the initial velocity of the string . In (1), v(t) denotes the control variable, which acts
on the system through the extreme x = 0. The other extreme is held xed.
The exact control problem for (1) consists of nding, for each initial data y
0
(x) and y
1
(x)
that belong to some class, a control v(t) that lies in some class such that the solution y(x, t)
of Problem (1) with these data satises the nal condition
y(x, T) = 0, y
t
(x, T) = 0, 0 x L. (2)
Let us consider z(x, t) = y(x, T t). Then Problem (1) and nal condition (2) are,
respectively, equivalent to
_
_
_
z
tt
(x, t) z
xx
(x, t) = 0, in Q = (0, L) (0, T)
z(0, t) = v(T t), z(L, T t) = 0, 0 t T
z(x, 0) = 0, z
t
(x, 0) = 0, 0 x L
(3)
and
z(x, T) = y
0
(x), z
t
(x, T) = y
1
(x), 0 x L. (4)
The HUM (Hilbert Uniqueness Method) was introduced by Lions [5] in order to study
the exact control problem for distributed systems, in particular for (1) (or equivalently, (3)
with the nal condition (4)). A number of authors have studied this method, among them,
we can mention, Zuazua [13], Medeiros [6], Komornik [4] and Milla Miranda [8]. In these
works the exact control problem for (1) was studied when is an open bounded set of IR
n
,
which generalizes the case = (0, L). A study of the numerical analysis of the exact control
problem for the wave equation was done by Glowinski et al [2]. They used nite elements
and conjugate gradient methods. Similar results are obtained in Rincon [9], by using the
dierence nite method.
F.P.Vasilyev, M.A. Kurzhanskii and A. Razgulin made a numerical study which allows
us to obtain an approximation v
N
(t) for the exact control u(t) of Problem (3), where u(t)
is given by HUM. They used the Fourier series method. In this paper we implement these
results for the particular case
y
0
1, y
1
=
L/2
( = Dirac delta), T = 3L and L =
2
.
We present the graphics of the approximations u
N
(t) of u(t) and y
N
(x, t) of the solution
y(x, t) of (3). For that we use non standard results on numerics series, which are obtained by
applying the Residues Theorem of holomorphic functions (see, the Appendix). The novelty
in this paper is the use of these results in the computational study of an exact control
problem. As far as we know this is the rst time that it is done.
2 Preliminaries
In this part we describe the principal results obtained by Vasilyev et al [11], in order to
obtain an approximation u
N
(t) for the exact control u(t) of problem (3), where v(t) is given
by HUM.
Let us consider the problem
_
_
_
y
tt
(x, t) y
xx
(x, t) = 0, in Q = (0, L) (0, T)
y(0, t) = v(t), y(L, t) = 0, 0 t T
y(x, 0) = 0, y
t
(x, 0) = 0, 0 x L
(5)
and the nal condition
y(x, T) = y
0
(x), y
t
(x, T) = y
1
(x), 0 x L. (6)
Let us consider v(t) L
2
(0, T). Then the ultra weak solution y(x, t; v) of (5) has the
following form:
y(x, t; v) =
2
L

m=1
sin
mx
L
_
t
0
v() sin
m(t )
L
d. (7)
Note that
y C
0
([0, T]; L
2
(0, L)) and y
t
C
0
([0, T]; H
1
(0, L)).
Let us dene the application
A : L
2
(0, T) H
1
(0, L) L
2
(0, L)
v Av = y
t
(, T; v), y(, T; v),
where y(x, t; v) is the ultra weak solution of (5). The adjoint A

of A is the operador
A

: H
1
0
(0, L) L
2
(0, L) L
2
(0, T)

0
,
1
A

0
,
1
=

x
(0, ;
0
,
1
)
where is the ultra weak solution of the backward problem:
_
_
_

tt
(x, t)
xx
(x, t) = 0, in Q
(0, t) = 0, (L, t) = 0, 0 t T
(x, T) =
0
(x),
t
(x, T) =
1
(x), 0 x L
Note that C
1
([0, L]; L
2
(0, T)).
We dene = AA

. By the above expressions, we have

0
,
1
=
_
y
t
_
, T;

x
(0, ;
0
,
1
)
_
, y
_
, T;

x
(0, ;
0
,
1
)
__
; (8)
Thus
L
2
(0, T)
A
H
1
(0, L) L
2
(0, L) = F

H
1
0
(0, L) L
2
(0, L) = F
By Lions [5], we have that is an isomorphism from F to F

. We observe that the system

(5) is exactly controllable when T > 2L, see Lions [5]. Thus for each f = y
1
, y
0
F

,
there exists a unique c =
0
,
1
F such that

0
,
1
= y
1
, y
0
. (9)
From (8) and (9) it follows that
_

_
y
0
= y
_
, T;

x
(0, . ,
0
,
1
)
_
y
1
= y
t
_
., T;

x
(0, . ,
0
,
1
)
_
which is the nal condition (6) that we want to obtain.
Thus
u =

x
(0, . ;
0
,
1
). (10)
is one control which permits that the ultra weak solution y(x, t; u) of (5) veries the nal
condition (6).
The method described above to determine u(t) dened by (10) is the HUM, which de-
termine a unique control of Problem (3) for each y
1
, y
0
F

Note that
u = A

0
,
1

0
,
1
= y
1
, y
0
= f
In what follows one determines the approximations u
N
(t) of the exact control u(t), given
by (10). For that, we project the functions obtained before on nite-dimensional space and
use the properties of the isomorphism . In fact,

0
(x) =

k=1

0k
sin
kx
L
;
1
(x) =

k=1

1k
sin
kx
L
where the Fourier coecients are given by

0k
=
2
L
_
L
0

0
(x) sin
kx
L
dx,
1k
=
2
L
_
L
0

1
(x) sin
kx
L
dx
We are interested in a nite-dimensional subspace that contains the sum of the rst N
terms of
0
and
1
. Then we dene the subspace F
N
which is composed by elements
c
N
=
0N
,
1N
, where

0N
(x) =
N

k=1
c
0Nk
sin
kx
L
,
1N
(x) =
N

k=1
c
1Nk
sin
kx
L
with c
0Nk
and c
1Nk
belonging to IR. One uses the notation F

N
= F
N
. Then dim F

N
=
dim F
N
= 2N.
We x f = y
1
, y
0
F

in order to determine the unique exact control u(t) of Problem

(3) given by (10). We denote by f
N
the orthogonal projection of f in F

N
, that is,
|f
N
f|
F
= min
g
N
F

N
|g
N
f|
F

Let us denote c
N
= f
N
(therefore c
N
=
1
f
N
) and u
N
= A

c
N
. Then
u
N
u in L
2
(0, T).
Thus u
N
is the required approximation of u(t). Note that, if f
N
is known then so is u
N
.
In what follows, we construct an algorithm to determine u
N
. Let us introduce
e
2k
= sin
kx
L
, 0, e
2k1
= 0, sin
kx
L
, k = 1, 2, ...
Then e
1
, e
2
, ..., e
2N
and e
1
, e
2
, ..., e
2N
are basis, respectively, of the subspace F
N
and F

N
.
The elements e
i
, see Vasilyev et al [11], are give explicitly by
e
2k
=
_

m=1
_
2km
2
L
3
_

mk
sin
_
mx
L
_
,

m=1
_
2k
L
2
_

mk
sin
_
mx
L
_
_
e
2k1
=
_

m=1
_
2m
L
2
_

mk
sin
_
mx
L
_
,

m=1
_
2
L
_

mk
sin
_
mx
L
_
_
where
_

mk
=
_
T
0
cos
_
m(T )
L
_
cos
_
k( T)
L
_
d

mk
=
_
T
0
sin
_
m(T )
L
_
cos
_
k( T)
L
_
d =
km

mk
=
_
T
0
sin
_
m(T )
L
_
sin
_
k( T)
L
_
d.
The function f
N
has the form
f
N
=
2N

k=1
f
Nk
e
k
because f
N
F

N
. Since we know e
k
, in order to determine f
N
, it is sucient to know the
coecients f
Nk
. This is obtained by solving the following system of linear equations:
2N

k=1
f
Nk
(e
k
, e
m
)
F
= (f, e
m
)
F
, m = 1, 2, , 2N (11)
This system is obtained from the relations
(f f
N
, e
m
)
F
= 0, m = 1, 2, , 2N
which hold because f f
N
is orthogonal to the subspace F

N
.
Using the coecients f
Nk
we obtain
c
N
=
1
f
N
=
2N

k=1
f
Nk

1
e
k
=
2N

k=1
f
Nk
e
k
u
N
= A

c
N
=
2N

k=1
f
Nk
A

e
k
(12)
where u
N
(t) is the approximate control that we seek.
From the above algorithm, u
N
(t) can be determined after solving the linear system (11).
Let M = (m
ij
),

f = (f
N1
, , f
N, 2N
)
T
and G = (g
1
, , g
2N
)
T
be the matrix of order
2N 2N, 2N 1 and 2N 1, respectively. Then the system (11) can be written in the
matrix form,
M

f = G
dened by
m
ij
= (e
i
, e
j
)
F
, g
j
= (f, e
j
)
F
, i, j = 1, 2, 2N. (13)
In the particular case T = 3L the entries m
ij
of the matrix M have the following form:
(e
2k
1
, e
2k
2
1
)
F
=
_

_
0 if (k
1
+ k
2
) is even
12k
1
k
2
k
2
1
k
2
2
if (k
1
+ k
2
) is odd.
(e
2k
1
1
, e
2k
2
1
)
F
=
_

_
0 if k
1
, k
2
are even and k
1
,= k
2
5L if k
1
is even and k
1
= k
2

4L

2
k
1
k
2
if k
1
, k
2
are odd and k
1
,= k
2
5L
4L

2
k
2
1
if k
1
is odd and k
1
= k
2
0 if k
1
is even and k
2
is odd
(e
2k
1
, e
2k
2
)
F
=
_

_
0 if k
1
, k
2
are even and k
1
,= k
2
5k
2
1

2
L
if k
1
is even and k
1
= k
2
0 if k
1
, k
2
are odd and k
1
,= k
2
5k
2
1

2
L
if k
1
is odd and k
1
= k
2
0 if k
1
is odd and k
2
is even
In the obtention of this result we used the equalities (26), , (33) of the Appendix.
In addition, for f =
L/2
, 1, by applying Proposition A.2 and result (25) of the
Appendix, the entries g
j
of the matrix G dened by (13) are given by
(f, e
2k1
)
F
=
_

_
L
k
((1)
k/2
1) if k is even

6L
k
if k is odd
(f, e
2k
)
F
=
_

_
0 if k is even
3(1)
(k1)/2
if k is odd.
3 Computational Results
In this section we determine the graphs of the approximate control u
N
(t) and of the approx-
imate ultra weak solution y
N
(x, t) at the instant t = T. Here y(x, t) denotes the solution of
Problem (5) with T = 3L, exact control u(t) and initial data y
0
1 , y
1
=
L/2
.
By the characteristic of the matrix M dened by (13), the Crout form or (L
T
DL) (see
Golub et al [3]) is an appropriate method for solving the system, that is, for obtaining the
coecients f
Nk
.
Note that if we know the f
Nk

s, then we will know u

N
(t) dened by (12), because
A

e
k
=
k
L
cos
k(t T)
L
k = 1, 2 .
A

e
k1
= sin
k(t T)
L
k = 1, 2 .
In order to obtain y
N
(x, T; u), we substitute the exact control u(t) in (7) by u
N
(t), and
compute the respective integral,
y(x, t; u
N
) =
2
L

m=1
sin
mx
L
_
T
0
u
N
() sin
m(T )
L
d.
We have the following graphs with L =
2
.
Figure 1 describes the variation of the approximate control u
N
(t), when N = 20, 100
and 1000 and we note that u
N
(t) approximates a continuous function almost everywhere
when N increases. Since we use the Fourier series, the Gibbs phenomenon appears near the
points of discontinuity.
Figure 2 shows the variation of the approximate solution y
N
(x, T) at the instant t =
T = 3L. We note that y
N
(x, T), for all x (0, L), approximates the function y
0
1, except
near the extremes x = 0 and x = L. This is due to the Gibbs phenomenon.
4 Error Estimates
In this part we analyze the error of the results obtained when N increase. In order to reduce
the inuence of the other numerical errors, we have considered suciently large values of
N
x
= L/x = 1000 and N
t
= T/t = 1000.
N=20
N=100
N=1000
t
u
N
(t)
30 25 20 15 10 5 0
0.6
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8
Figure 1: Approximation Control u
N
(t)
The error estimate in the L
2
(0, L) norm for the solution y(x, t) at the instant t = T is
given by
|E
y
| =
_
_
(1 y
N
(0, T))
2
+ (1 y
N
(L, T))
2
_
x
2
+
N
x
1

j=1
((1 y
N
(jx, T))
2
x
_
1/2
Since we do not know the exact control u(t), for the computation of the error, the approx-
imation u
N
(t) with N = 1600 is used instead of u(t). The formula that approximates the
error of the control u
N
(t), in the L
2
(0, T) norm, is given by
|E
u
| =
_
_
(u
N
(0) u
1600
(0))
2
+ (u
N
(T) u
1600
(T))
2
_
t
2
+
N
t
1

j=1
(u
N
(jt) u
1600
(jt))
2
t
_
1/2
We conrm that each of the above errors depend algebraically on N, that is, |E| a(N)

(for some > 0) for N large enough. This conclusion is based on the analysis of the following
data:
N=20
N=100
N=1000
x
y
N
10 8 6 4 2 0
1.2
1.1
1
0.9
0.8
0.7
0.6
0.5
Figure 2: Approximation Solution y
N
(x, T, u)
N |E
y
| |E
v
|
1600 0,06342 0
800 0,088386 0,030861
400 0,124107 0,077274
200 0,17467 0,1217316
100 0,24622 0,179067
50 0,350747 0,256549
25 0,506850 0,358164
In fact, we note that if at N, we have the numerical error E
y
and E
u
, then at 4N, we will
have, approximately, the numerical errors E
y
/2 and E
u
/2. This suggests us to formulate
the following hypotheses:
|E
y
| a N
1/2
; a = tan (14)
|E
u
| b N
1/2
; b = tan (15)
In the rst case, making the graph of the error for the exact solution, |E
y
| vs N
1/2
(see Figure 3) we observe that the error can be approximated by a straight line passing
through the origin of coordinates. This conrms our hypothesis (14). In the second case,
for the graph of the control error, |E
u
| vs N
1/2
(see Figure 4), we consider only the values
N = 25, 50, 100, 200, 400, 800, since we have taken u
N
(t) with N = 1600 instead of the
exact control u(t) in the computation of the errors, and it is natural that we can not obtain
good approximations for u(t) when N is near of the value 1600. Based on the hypothesis
(14) and (15) we arm that the order of convergence of the applied method is 1/2. In what
follows we present the graphs of the estimate errors.
N
1/2
|E
y
|
0.2 0.16 0.12 0.08 0.04 0
0.6
0.5
0.4
0.3
0.2
0.1
0
Figure 3: Error for Approximation Solution y
N
(x, T, u)
Appendix
In this part by using the Residue Theorem of holomorphic functions we obtain some
results on numerical series. For other similar results, see Weinberger [12].
In what follows we x some notations and write some results. Let O be an open set of
I C and f : O I C a holomorphic function. Let z
0
be a pole of order k of f(z). The residue
of f(z) at z
0
, which is denoted by '
z
0
, is by denition
'
z
0
[f] =
1
(k 1)!
lim
zz
0
_
d
k1
dz
k1
(z z
0
)
k
f(z)
_
.
Note that if f(z) = g(z)/h(z), where g(z) and h(z) are holomorphic functions in O and z
0
is a pole of order one of f(z), then
'
z
0
[f] =
g(z
0
)
h

(z
0
)
(16)
In this case g(z
0
) ,= 0, h(z
0
) = 0 and h

(z
0
) ,= 0.
Proposition A.1 Let f(z) be a holomorphic function in I C, except in the poles z
1
, z
2
,
such that [z
j
[ , j . Suppose that there exists a sequence (R
q
) of real positive
numbers with R
q
q , such that
lim
q
_
R
q
max
|z|=R
q
[f(z)[
_
= 0 (17)
Then

j=1
'
z
j
[f] = 0
N
1/2
|E
u
|
0.2 0.16 0.12 0.08 0.04
0.4
0.3
0.2
0.1
0
Figure 4: Error for the exact control u
N
(T)
Proof We denote by
q
the curve [z[ = R
q
. By hypothesis (17) we have that there exists
q
0
such that the curves
q
, q > q
0
, do not intersect the poles z
j
since each time that
q
intersect some pole, one has max
|z|=R
q
[f(z)[ = .
Let us consider
q
with q q
0
. By Residue Theorem we have
_

q
f(z) dz = 2i

j
'
z
j
[f] (18)
where the summation is made on all poles that are inside q
q
. We have
[
_

q
f(z)dz[
_
max
|z|=R
q
[f(z)
_
2R
q
0, q
which implies
_

q
f(z)dz 0, q (19)
Expressions (18) and (19) give the proposition.
Let z = x + iy be a complex number. Then
[ cos z[
2
= cos
2
x + sinh
2
y and [ sin z[
2
= sin
2
x + sinh
2
y (20)
In the sequel we obtain results on numerical series which have been used in the paper.
Proposition A.2 Let k
e
> 0 be an even number. Then

p=1
1
(2p 1)
2
k
2
e
= 0
Proof Let us consider the function
f(z) =
1
z
2
k
2
e
cos

2
(z + 1)
sin

2
(z + 1)
.
As cos

2
(k
e
+1) = 0 it follows that z = k
e
are removable singularities. The poles of f(z)
are z
n
= n, n = 1, 3, 5, , and their order is one.
By (16) and noting that
d
dz
_
sin

2
(z + 1)
_
=

2
cos

2
(z + 1), we have that
'
n
[f] =
2
(n
2
k
2
e
)
; n = 1 3, .
Thus, the sum of all residues is

n=
n odd
2
(n
2
k
2
e
)
(21)
On the other hand, we have
[f(z)[
2
=
1
[z
2
k
2
e
[
2
[ cos

2
(z + 1)[
2
[ sin

2
(z + 1)[
2
Noting that, for z = x + iy,
[z
2
k
2
e
[
_
[z[ k
e
_
2
and

2
(z + 1) =

2
(x + 1) + i

2
y
and using (20), we obtain from the last equality
[f(z)[
2

1
([z[ k
e
)
4
cos
2
2
(x + 1) + sinh
2
2
y
sin
2
2
(x + 1) + sinh
2
2
y
(22)
Let us consider, R
q
= 2q, q = 1, 2, , and [z[ = 2q ; z = x + iy. We have:
1. If [x[ 2q
1
2
, then
_

_
q +

4


2
(x + 1) q +

2
, x > 0
q +

2


2
(x + 1) q +
3
4
, x < 0
For these values of x, one has
cos
2

2
(x + 1) sin
2

2
(x + 1).
This inequality and (22) imply that
[f(z)[
2

1
([z[ k
e
)
4
=
1
(2q k
e
)
4
. (23)
2. If [x[ < 2q
1
2
, then
y
2
> 4q
2

_
2q
1
2
_
2
= 2q
1
4

7
4
that is, [y[ >

7
2
. From (22) it follows that
[f(z)[
2

1
([z[ k
e
)
4
(y) (24)
where (y) is the function dened by
(y) =
1 + sinh
2
2
y
sinh
2
2
y
, whenever y ,= 0
Noting that
sinh
2

2
y =
1
4
(e
y
2 + e
y
) and
d
dy
_
sinh
2

2
y
_
=

4
(e
y
e
y
),
we obtain
d
dy
(y) =
1
(sinh
2
2
y)
2

4
(e
y
e
y
), whenever y ,= 0.
From that we conclude that (y) is decreasing in ]0, [ and increasing in ], 0[. Therefore
(y) <
_
(

7
2
) for y >

7
2
(

7
2
) for y <

7
2
Nothing that (y) = (y) , for y IR, y ,= 0, we derive then
(y) <
_
7
2
_
, [y[ >

7
2
.
Thus by (24) we obtain
[f(z)[
2

1
(2q k
e
)
4

_

7
2
_
The above inequality and (23) give that
[f(z)[
1
(2q k
e
)
2
M, [z[ = 2q, 2q > k
e
where M
2
=
_

7
2
_
. The last inequality implies
lim
q
_
2q max
|z|=2q
[f(z)[
_
= 0.
Proposition A.1, (21) and this limit give us that

n=
n odd
2
(n
2
k
2
e
)
= 0,
which implies the proposition.
By applying similar arguments as in the proof of Proposition A.2, we derive the following
results on numerical series. We will present results and the respective function f(z) that
allow us to obtain each one of them.
Let k
e
> 0 be an even number. Then

p=1
(1)
p+1
(2p 1)((2p 1)
2
k
2
e
)
=

4
(
(1)
k/2
k
2
e

1
k
2
e
)
f(z) =
1
z(z
2
k
2
e
)
1
sin

2
(z + 1)
.
(25)
Let k
1e
> 0 and k
2e
> 0 be even numbers with k
1e
,= k
2e
. Then

p=1
1
((2p 1)
2
k
2
1e
)
1
((2p 1)
2
k
2
2e
)
= 0
f(z) =
1
(z
2
k
2
1e
)(z
2
k
2
2e
)
cos

2
(z + 1)
sin

2
(z + 1)
.
(26)
Let k
e
be an even number. Then

p=1
1
((2p 1)
2
k
2
e
)
=

2
16k
2
e
f(z) =
1
(z
2
k
e
)
2
cos

2
(z + 1)
sin

2
(z + 1)
.
(27)
Let k
1o
> 0 and k
2o
> 0 be odd numbers with k
1o
,= k
2o
. Then

p=1
1
((2p)
2
k
2
1o
)
1
((2p)
2
k
2
2o
)
=
1
2 k
2
1o
k
2
2o
f(z) =
1
(z
2
k
2
1o
)(z
2
k
2
2o
)
cos

2
z
sin

2
z
.
(28)
Let k
o
> 0 be an odd number. Then

p=1
1
((2p)
2
k
2
o
)
2
=

2
16 k
2
o

1
2 k
4
o
f(z) =
1
(z
2
k
2
o
)
cos

2
z
sin

2
z
.
(29)
Let k
1e
> 0 and k
2e
> 0 be even numbers with k
1e
,= k
2e
. Then

p=1
(2p 1)
2
((2p 1)
2
k
2
1e
)
2
((2p 1)
2
k
2
2e
)
= 0
f(z) =
z
2
(z
2
k
2
1e
)(z
2
k
2
2e
)
cos

2
(z + 1)
sin

2
(z + 1)
.
(30)
Let k
e
> 0 be an even number. Then

p=1
(2p 1)
2
((2p 1)
2
k
2
e
)
2
=

2
16
f(z) =
z
2
(z
2
k
2
e
)
cos

2
(z + 1)
sin

2
(z + 1)
.
(31)
Let k
1o
> 0 and k
2o
> 0 be odd numbers with k
1o
,= k
2o
. Then

p=1
(2p)
2
((2p)
2
k
2
1o
) ((2p)
2
k
2
2o
)
= 0
f(z) =
z
2
(z
2
k
2
1o
)(z
2
k
2
2o
)
cos

2
z
sin

2
z
.
(32)
Let k
o
> 0 be an odd number. Then

p=1
(2p)
2
((2p)
2
k
2
o
)
2
=

2
16
f(z) =
z
2
(z
2
k
2
o
)
cos

2
z
sin

2
z
.
(33)
References
[1] H. Brezis, Analyse Fonctionelle, Theorie et Aplications, Masson, Paris, 1983.
[2] R. Glowinski, Chin-Hsien Li, J.L. Lions, A Numerical Approach to the Exact Bound-
ary Controllability of the Wave Equation (I) Dirichlet Controls: Description of the
Numerical Methods, Japan Journal of Applied Mathematics, Vol.7, N 1,(1990), pp.
1-76.
[3] G.H. Golub e Ch. F. Van Loan , Matrix Computations, The Johns Hopkins University
Press, Baltimore, 1989.
[4] V. Komornik, Exact Controllability and Stabilizations, The Multiplier Method, John
Wiley and Sons e Masson, New York and Paris, 1994.
[5] J.L. Lions, Contr olabilite Exacte et Stabilization de Syst`emes Distribues, Vol. 1, Mason,
Paris, 1988.
[6] L.A. Medeiros, Exact Controllability for Wave Equations - HUM, Proceedings of the
37

Brazilian Seminar of Analysis, 1993.

[7] L.A. Medeiros and M. Milla Miranda, Introdu c ao aos Espa cos de Sobolev e as Equa c oes
Diferenciais Parciais, IM-UFRJ, Rio, 2000.
[8] M. Milla Miranda, HUM and the wave equation with variable coecients, Asymptotic
Analysis 11, (1995), pp. 317-341.
[9] M.A. Rincon, Metodos Numericos para Problemas de Controle, Thesis, IM-UFRJ, Rio,
1994.
[10] M.A. Rincon e I. Shih Liu, Introdu c ao ao Metodo de Elementos Finitos, An alise e
Aplica c ao, IM-UFRJ, Rio, 2000.
[11] F. P. Vasilyev, M. A. Kurzhanskii and A. V. Razgulin. Moscow University, Computa-
tional Mathematics and Cibernetics; No. 2,(1993), pp. 1-5.
[12] H.F. Weinberger, A First Course in Partial Dierential Equations with Complex Vari-
ables and Transform Methods, Blaisdell Publishing Company, New York, 1965.
[13] E. Zuazua, Controlabilidad Exacta y Estabilizaci on de la Ecuaci on de Ondas, Textos
de Metodos Matem aticos 23, IM-UFRJ, Rio, 1990.