Journal of Fundamental and Applied Sciences is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. Libraries Resource Directory. We are listed under Research Associations category.
GRADIENT REMEDIABILITY IN LINEAR DISTRIBUTED PARABOLIC SYSTEMS
ANALYSIS, APPROXIMATIONS AND SIMULATIONS
S. Rekkab* and S. Benhadid
Mathematics Department, Faculty of Exact Sciences, University Mentouri, Constantine, Algeria
Received: 28 April 2017 / Accepted: 22 August 2017 / Published online: 01 September 2017
ABSTRACT
The aim of this paper is the introduction of a new concept that concerned the analysis of a
large class of distributed parabolic systems. It is the general concept of gradient remediability.
More precisely, we study with respect to the gradient observation, the existence of an input
operator (gradient efficient actuators) ensuring the compensation of known or unknown
disturbances acting on the considered system. Then, we introduce and we characterize the
notions of exact and weak gradient remediability and their relationship with the notions of
exact and weak gradient controllability. Main properties concerning the notion of gradient
efficient actuators are considered. The minimum energy problem is studies, and we show how
to find the optimal control, which compensates the disturbance of the system. Approximations
and numerical simulations are also presented.
Keywords: actuators efficient; disturbance; gradient; parabolic systems; remediability;
sensors.
Author Correspondence, e-mail: rekkabsoraya@yahoo.fr doi: http://dx.doi.org/10.4314/jfas.v9i3.18
ISSN 1112-9867
1. INTRODUCTION
Various real problems can be formulated within certain concepts of distributed systems
analysis. These concepts consist of a set of notions as controllability, detectability,
observability, remediability…, which enable a better knowledge and understanding of the
system to be obtained. These concepts have been studies at different degrees (exact, weak,
extended). Systems analysis can be done from a purely theoretical viewpoint [1,2]. However,
the study may be also become concrete, in some sense, by using the structures of actuators
and sensors. Thus, one can study the different concepts of controllability via actuators
structures [3-5] or the different concepts of observability via sensors structures [6-8].
An extension of these concepts that is very important in practical applications is that of
gradient controllability [9], gradient detectability [10] and gradient observability [11-13].
These concepts are of great interest from a more practical and control point of view since
there exist systems that cannot be controllable but gradient controllable or that cannot be
observable but gradient observable or that cannot be detectable but gradient detectable and
they provide a means to deal with some problems from the real world, for example in the
thermic isolation problem it may be that the control is only required to vanish the
temperature-gradient before crossing the wall.
Hence, with the same preoccupation, we introduce in this paper, a new concept that is gradient
remediability of distributed parabolic systems. We recall that the notion of remediability
consists in studying the existence of a convenient input operator (efficient actuators), ensuring
the remediability of any disturbance acting in the considered systems. This problem is
particularly motivated by pollution problems and so called space compensation or
remediability problem. The notions of remediability and efficient actuators are introduced and
studied first for discrete systems and there for continuous systems of a finite time horizon and
for other situations (regional and asymptotic cases, internal or boundary actions of
disturbances) [14-16].
This paper is organized as follows: In the second paragraph, we start by presenting the
notations and some preliminary material. After, we recall the notions of exact and weak
of exact and weak gradient remadiability. We show how to find an input operator (actuators)
with respect to the output (gradient observation) that ensures the gradient remediability of a
disturbed parabolic system. By analogy with the relation between the remediability and the
controllability examined in the finite time case, it is then natural to study, in the paragraph 4,
the relationship between the gradient remediability and the gradient controllability. We show
that the gradient remediability is weaker and more supple than the gradient controllability of
the parabolic systems, that is to say, if any parabolic systems are gradient controllable, then it
is gradient remediable. The fifth paragraph recall the notion of gradient strategic actuators and
we give a characterization of gradient remediability which shows that the gradient
remediability of any system may depend on the structure of the actuators and sensors. Then
we introduce and we characterize the notion of gradient efficient actuators. In paragraph 6,
using an extension of Hilbert Uniqueness Method (H.U.M), we examine the problem of
gradient remediability with minimum energy, and we give the optimal control that
compensate an arbitrary disturbance. In the last paragraph, approximations and numerical
results are presented.
2. CONSIDERED SYSTEM
Let be an open and bounded regular subset of IRn
n1,2,3
with a smooth boundary . ForT 0 , we denoted by Q
0,T
,
0,T
. Consider a parabolic system defined by
0 , 0 , , , , 0 t y x y x y Q t x f t Bu t x y A t x t y (1)Where A is a second order linear differential operator with compact resolvent and which generates a strongly continuous semi-group
S
t
t0on the Hilbert space L2
.
S*
t
t0 is considered for the adjoint semi-groupe of
S
t
t0 .
U X
u L
TU
B
L
, , 2 0, ; where U is a Hilbert space representing the control space and
H01The system (1) admits a unique solution yC
0,T;H10
C1
0,T;L2
[17] given by
t t ds s f s t S ds s Bu s t S y t S t y 0 0 0and it is augmented by the output equation
t C y
tz (2) where CL
L2
n,O
, O is a Hilbert space (observation space). In the case of an observation on
0,T
withq
sensors, we take generally qIR O .The operator
is defined by
n n x y x y x y y y L H , , , : 2 1 2 1 0 while
* its adjoint operator.3. GRADIENT REMEDIABILITY
The system (1) is disturbed by the force f assumed unknown and excited by a control u that will be chosen so as to compensate for the disturbance f . In the autonomous case,
without disturbance
f 0
and without control
u0
, this same system is written
0 , 0 , , , 0 t y x y x y Q t x y A t x t y (3)The system (3) admits a unique solution yC
0,T;H10
C1
0,T;L2
[17] given by
0y t S t
y then the observation on
0,T
is normal and it is given by z
t CS
t y0. But if the control u0and the disturbance f 0, the observation noted zu,f is disturbed such that
t zu,f
t t ds s f s t S C ds s Bu s t S C y t S C 0 0 0The problem consists to study the existence of an input operator B (actuators), with respect to the output operator C (sensors), ensuring at the timeT , the gradient remediability of any
disturbance on the system, that is:
For any f L2
0,T;X
, there exists uL2
0,T;U
such that
0 , T C ST yzu f this is
equivalent to CHuCFf 0where H and F are two operators defined by
T ds s Bu s T S Hu u X U T L H 0 2 ; , 0 :
T ds s f s T S Ff f X X T L F 0 2 ; , 0 :This leads to the following definitions
Definition 1:
1) We say that the system (1) augmented by the output equation (2) is exactly gradient f -
remediable on
0,T
, if there exists a control uL2
0,T;U
such that 0 Hu C Ff C2) We say that the system (1) augmented by the output equation (2) is weakly gradient f -
remediable on
0,T
, if for every
0,there exists a control uL2
0,T;U
such that O Ff C Hu C
3) We say that the system (1) augmented by the output equation (2) is exactly (resp. weakly)
gradient remediable on
0,T
, if for every f L2
0,T;X
the system (1)–(2) is exactly (resp. weakly) gradient f - remediable.Proposition 1:
Let us consider f L2
0,T;X
.1) The system (1)–(2) is exactly gradient f - remediable on
0,T
if and only if
C H
Ff
C Im Im
2) The system (1)–(2) is weakly gradient f - remediable on
0,T
if and only if
C H
Ff
C Im Im
Proof:
exists uL2
0,T;U
such that CHuCFf 0 , that is
u
C Hu1 H C Hu C Ff C with
u1 u
u1L2
0,T;U
then H C Ff C Im .Conversely, we assume that CFf ImCH, then there exists uL2
0,T;U
such thatHu C Ff
C that is CFf CHu0 this gives CFf CH
u
0. We put
T U
L u
u1 2 0, ; where the system (1)–(2) is exactly gradient f - remediable.
2) We assume that the system (1)–(2) is weakly gradient f - remediable on
0,T
, then
T U
L u 0, ; , 0 2 such that O Hu C Ff C that is 0,uL2
0,T;U
such that CFf CH
u
O . We put u u L2
0,T;U
1 ,
then 0,u1L2
0,T;U
such that O Hu C Ff C 1 , this gives
C H
Ff C Im .Conversely, we assume that CFf Im
CH
, then 0,u1L2
0,T;U
such that
Ff C Hu O
C 1 . We put u1 u L2
0,T;U
, then 0,uL2
0,T;U
such that CFf CHuO where the system (1)–(2) is weakly gradient f - remediable. Proposition 2:1) The system (1)–(2) is exactly gradient remediable on
0,T
if and only ifH C F
C Im Im
2) The system (1)–(2) is weakly gradient remediable on
0,T
if and only ifH C F C Im Im Proof:
1) We assume that the system (1)–(2) is exactly gradient remediable on
0,T
then
T X
L f 2 0, ;
the system (1)–(2) is exactly gradient f - remediable on
0,T
and from the Proposition 1 we have, f L2
0,T;X
: CFf ImCH , this givesH C F
C Im
Im .
Conversely, we assume that ImCF ImCH and we show that the system (1)–(2) is exactly gradient remediable on
0,T
. Let f L2
0,T;X
H C F
C Im
Im we have CFfImCH then there exists uL2
0,T;U
such that CFf CHu that is to say CFf CHu0 this gives CFf CH
u
0. We put u1 u L2
0,T;U
where the system (1)–(2) is exactly gradient remediable. 2) We assume that the system (1)–(2) is weakly gradient remediable on
0,T
thenf L2
0,T;X
the system (1)–(2) is weakly gradient f - remediable on
0,T
and from the Proposition.1 we have, f L2
0,T;X
: CFf ImCH , this givesH C F
C Im
Im .
Conversely, we assume that ImCF ImCHand we show that the system (1)–(2) is weakly gradient remediable on
0,T
. Let f L2
0,T;X
then CFfImCF. Since F
C
Im ImCH then CFf Im
CH
this leads to 0, u L2
0,T;U
such
that CFf CHu by putting u1 u L2
0,T;U
this gives 0,u1L2
0,T;U
such that O
Hu C Ff
C 1 where the system
(1)–(2) is weakly gradient remediable on
0,T
.4. GRADIENT REMEDIABILITY AND GRADIENT CONTROLLABILITY
By analogy with the relation between the remediability and the controllability examined in the
finite time case, it is then natural to study, in this paragraph, the relationship between the
gradient remediability and the gradient controllability. Firstly, we recall the definitions of
exact and weak controllability.
Definition 2:
1) The system (1) is said to be exactly gradient controllable on
0,T
if yd
L2
n there exist a control uL2
0,T;U
such thaty
T yd .2) The system (1) is said weakly gradient controllable on
0,T
if yd
L2
n ,
0 there exist a control uL2
0,T;U
such that
L n d y T y 2 .Proposition 3:
1) If the system (1)–(2) is exactly gradient controllable on
0,T
, then it is exactly gradient remediable on
0,T
.2) If the system (1)–(2) is weakly gradient controllable on
0,T
, then it is weakly gradient remediable on
0,T
.Proof:
1) We assume that the system (1)–(2) is exactly gradient controllable and let yd S
T y0then there exists uL2
0,T;U
such that y
T S
T y0 that is to say
0
0 y T S Ff Hu y T S this leads to HuFf 0 then let
0
Hu C Ff
C and then the system (1) – (2) is exactly gradient remediable.
2) We assume that the system (1)–(2) is weakly gradient controllable and let yd S
T y0then 0, u L2
0,T;U
such that
L n y T S T y 0 2 that is L n FfHu 2 . Since the operator C is continue, consequently we have
n L O k Hu Ff Ff C Hu C
2 with k 0 where the system (1)–(2) is
weakly gradient remediable.
5. GRADIENT REMEDIABILITY, SENSORS AND ACTUATORS
We suppose that the system (1) is excited by p zone actuators
i,gi
1ip,giL2
i ,
i supp gi in this case the control space is U IRp and the operator:
p i i i i p p t u x g x Bu t u t u t u t u X IR B 1 2 1 , , , : Its adjoint is given by
p T p p z IR g z g z g z B , , , , , , 2 2 1 1 * (4)
Also suppose that the output of the system (2) is given by qsensors
Di,hi
1iq,hi L2
Di , being the spatial distribution,Di supphi , for i1 , ,qand D i Dj fori , then j the operator C is defined by:
T n i Dq i q n i D i n i D i q n t y h t y h t y h t Cy IR L C
1 2 1 2 1 1 1 2 , , , , , , : its adjoint is given by C with for*
1,2,,q
T IRq
q i i i i D q i i i i D q i i i i D x h x x h x x h x C 1 1 1 * , , , (5) Lemma 1 [18]:Let V,W and Z be reflexive Banach spaces, PL
V,Z
andQL
W,Z
. Then the following properties are equivalent:i. Im P ImQ
ii. 0 such that , * '
' * * ' * Z z z Q z P W V *
We have the following characterizations: Proposition 4:
1) The system (1)–(2) is exactly gradient remediable on
0,T
if and only if there exists0
such that for everyIRq, we have
p IR T L X T L B S T C C T S ; , 0 2 * * * * ' ; , 0 2 * * * . .
2) The system (1) – (2) is weakly gradient remediable on
0,T
if and only if
* * * *
* * *
ker
ker B F C F C
Proof:
1) It follows from the fact that F**C* S*
T .
*C* and that H**C*B*S*
T.
*C*and since the Proposition 2, we put PCF and QCH and using the Lemma 1. 2) We assume that the system (1)–(2) is weakly gradient remediable on
0,T
and we showthat ker
B*F**C*
ker
F**C*
. Let IRq such that B*F**C* 0, and we have
. . * * * * * T S B H T S Fthen, B*F**C* 0H**C* 0 this gives ker
H * *C*
and we have
ker * * *
ImC H H C . Since the hypothesis and the Proposition 2, we have
H C F C Im Im then ImCF
ker
H**C*
f L2
0,T;X
:CFf
* * *kerH C CFf, 0 because ker
H * *C*
, this gives
ImCF
ker
F**C*
, where the result.Conversely, assume that ker
B*F**C*
ker
F**C*
and we show thatH C F
C Im
Im . Let f L2
0,T;X
such that f ImCF , we have
ker * * *
ImC H H C . For every IRq such that H**C*0 , that is 0 * * * * C F
B we have F**C* 0 because ker
B*F**C*
ker
F**C*
then CFf, 0, where the result. Corollary 1:The system (1)–(2) is exactly gradient remediable on
0,T
if and only if 0 such thatq IR , we have
p i T L i T X ds g S T s C ds C s T S i 1 0 2 * * * 0 2 ' * * * 2 , Proof:Since the Proposition 4, the system (1)–(2) is exactly gradient remediable on
0,T
if and only if there exists 0 such that for every qIR , we have
2 ; , 0 * * * * 2 ' ; , 0 * * * 2 2 . . C L T X B S T C L T IRp T S by using (4) the formula of the operator B*, we have
p i T i T X ds g S T s C ds C s T S 1 0 2 * * * 0 2 ' * * * , where the result.
In the following, without loss of generality we consider, the system (1) with a dynamics Aof the form
1 0 2 1 2 1 , ,w w y D A H H y Ay m r j j m L j m m m where
1 1 m m r j j mw is an orthogonal basis in H10
of eigenvectors of A orthonormal in
2
L , associated to eigenvalues m0 with a multiplicity r . Then, the operator m A generates on the Hilbert space L2
a strongly continuous semi-group
S
t
t0 given by [1,19]:
m r j j m L j m m t m y w w e y t S 1 2 1 , (6) Corollary 2:The system (1)–(2) is exactly gradient remediable on
0,T
if and only if 0 such thatq IR , we have
p i T j m i m r j j m m s T m m m r j n L mj T m m ds w g w C e w C e 1 0 2 1 * 1 1 1 2 2 * 2 , , , 1 2 1 Proof:Since the Corollary 1, the system (1)-(2) is exactly gradient remediable on
0,T
if and only if there exists
0 such that for every IRq, we have
p i T i i T X ds g S T s C ds C s T S 1 0 2 * * * 0 2 ' * * * , By using (6) the formula of the operator S and since it is auto-adjoint, we obtain
T r j j m m s T T L ds e C w ds C s T S m m 0 1 2 * * 1 2 0 2 * * * , 2
m r j j m T m m m w C e 1 2 * 2 1 , 1 2 1 and
p i T j m i m r j j m m s T m p i T i L i S T s C ds e C w g w ds g 1 0 2 1 * * 1 1 0 2 2 * * * , , , where the result.
Corollary 3:
The system (1)–(2) is exactly gradient remediable on
0,T
if and only if 0 such thatq IR , we have
x ds w h w g e x w h e p i m r j n k q l k mj l l i L j m i m s T m T m m r j n k q l k L Dl mj l l T m m 2 1 1 1 1 2 1 0 1 1 1 1 2 2 2 , , , 1 2 1
Proof:Since the Corollary 2, the system (1)-(2) is exactly gradient remediable on
0,T
if and only if there exists
0 such that for every IRq, we have
p i T j m i m r j j m m s T m m m r j n L mj T m m ds w g w C e w C e 1 0 2 1 * 1 1 1 2 2 * 2 , , , 1 2 1 and by using (5) the formula of the operator C , we have *
n L n mj mj mj q l l l l D q l l l l D q l l l l D n L mj x w x w x w h h h w C
2 2 1 1 1 1 2 * , , Dl L n k q l k mj l l x w h 2 1 1 ,
where the result.
This characterization shows that the remediability of a system may depend on the structure of
the actuators and sensors.
By analogy with the concept of gradient strategic actuator, we introduce the notion of gradient
efficient actuator, as follows:
Definition 3:
The actuators
i,gi
1ip,gi L2
i are said to be gradient efficient if the system (1)–(2) so excited is weakly gradient remediable.The actuators gradient efficient define actions with the structure (spatial distribution, location
the following characterization of the gradient efficient actuators.
Form1, let Mm be the matrix of order
p rm
defined by M
g w
i pj i j m i m , ,1
and 1 j rm and let Gm be the matrix of order
q rm
defined byq i x w h G j i n k k j m i m
1 , , 1 and 1 j rm. Proposition 5:The actuators
i,gi
1ip,giL2
i are gradient efficient if and only if
m m
m f M C ker ker 1 * * . Where, for IRqand m1,
m f
T rm m r m m m C w C w IR w C * *, 1 , * *, 2 ,, * *, Proof:We assume that the actuators
i,gi
1ip,giL2
i are gradient efficient and we showthat
m m
m f M C ker ker 1 * * . Since the Proposition 4, the system (1)-(2) is weakly gradient remediable on
0,T
if and only if ker
B*F**C*
ker
F**C*
. Let qIR , we have
* * * * * * * * . C T S B C F B
m r j L p mj P L mj m T m m r j L mj L mj m T m m r j L mj L mj m T m w g w C e w g w C e w g w C e 1 2 2 * * 1 . 1 2 2 2 2 * * 1 . 1 1 2 1 2 * * 1 . , , , , , , and we have m1,
m r j L p mj P L mj m r j L mj L mj m r j L mj L mj m m w g w C w g w C w g w C f M 1 2 2 * * 1 2 2 2 2 * * 1 1 2 1 2 * * , ; , ; , ; If we assume that
m m
m f M ker 1 , then
, 0,
1,2, ,
, 1 ; 1 , ker 1 * * 2
g w i p m w C m f M m r j mj i L mj m m
1 , , , 2 , 1 , 0 , , 1 1 * * .
m p i w g w C e m m r j mj i j m T m
* * * *
* * * * ker 0 B F C C F B where
* * * *
1 ker ker Mmfm B F C m that is
* * * *
1 ker ker Mmfm B F C m .On the other hand, we have for every q
IR ,
j m m m r j j m T m C w w e C T S C F
1 1 * * . * * * * * * , . We assume that
* * *
ker F C , then ** * 0 C F
0 , . 1 1 * * . * * * * * *
j m m m r j j m T m C w w e C T S C F 0 , 1 1 * *
j m m m r j j m w w C *C* 0ker
*C*
. Then,
* * *
* *
kerker F C C . If we assume thatker
*C*
, then *C* 0 that is 0 , 1 1 * * . * * *
j m m m r j j m T m C w w e C F
* * *
ker F C , then
* *
* * *
kerker C F C that is ker
*C*
ker
F**C*
.Where the result.
Corollary 4:
If there exists m0 1 such that
q G
rank mT0 (7)
then the actuators
i,gi
1ip,giL2
i are efficient if and only if ker
01 T m m m G M . Proof: Let q IR , then
m mT
m G M ker 1
MmGmT
0, m1 p i m x w h w g q l l m r j n k k j m l j m i, , 0, 1, 1, , 1 1 1
m i p w C w g m r j L mj j m i, , 0, 1, 1, , 1 2 * *
0, 1 Mmfm m
m m
m f M ker 1 this gives
T m m m G M ker 1
m m
m f M ker 1 .On the other hand,ker
*C*
*C*0, then for m that appears in the hypothesis 0and by using (5) the formula of the operator C , we obtain *
0 1 1 2 0 0 * * , , 1 , 0 , , m q l l D L n k k j m l l j m j r x w h w C
T m T m G G 0 0 0 ker andsince rankGmT0 q , then kerGmT0
0 this gives 0. That is ker
C* *
0Finally, the proof follows directly from the Proposition 5. Corollary 5:
If there exists m0 1 such that rank
GmT
q0 and if
M G
q rank m mT 0 0 (8) Or
Mm0
rm0 rank (9)then the actuators
i,gi
1ip,giL2
i are gradient efficient. Proof:Assume that there exists m0 1 such that rank
Mn0GnT0
q. The matrix
Mm0GmT0
is of order p q . From the theorem of rank to matrices [20], we haverank
Mm0GmT0
dim
ker
Mm0GmT0
q, and then dim
ker
Mm0GmT0
0 witch is equivalent toker
0 0 0 T m m G M ker
0 1 T m m m G M . Since the Corollary 4, that is equivalent to the gradient efficient of the actuators
i,gi
1ip,giL2
i .Now, we suppose that rank
GmT0
q and
0
0 m
m r
M
rank . The matrix
GmT0
is of order rm0 q . By using the theorem of rank to matrices [20], we have
G
G
qrank mT mT
0
0 dim ker then, dim
ker
0
0T m
G . That is equivalent to
0ker GmT0 (10) The same, the matrix
Mm0
is of order p rm0. By using the theorem of rank for matrices[19], we have
0 0 0 dim ker m m m M r Mrang . And from (9), we obtain
ker
0dim Mm0 which is equivalent to
0ker Mm0 (11)
On the other hand, let
Mm GmT
0 0 ker , then
0 0 0 T m m G M which gives
0 0 0 T m m G M . From (11), we obtain 0 0 T mG and from (10), we obtain 0 then ker
00 0 T m m G M
and then, ker
01 T m m m G M
which is equivalent, from the Corollary 4, to the gradient efficient of the actuators
i,gi
1ip,giL2
i . Remark 1:1) The condition (8) q p.
2) The condition q p is not necessary for actuators to be gradient efficient. Indeed, in the case of a single actuator
1,g
and of q sensors
Di,hi
1iq , with q1 ,
m r j j m m g w M 1 , is of order
1rm
and q i m r j n k k j m i T m x w h G
1 1 1 , is of order
rm q
consequently q i m r j n k k j m i j m T m m x w h w g G M
1 1 1 , , is of order
1q
.From the Corollary 4, if there exists m0 1 such that rankGTm q
0 , then
1,g
isgradient efficient if and only if ker
01 T m m m G M
. Then there exists n1,n2,,nm such that n i nj for i and j
0 ker , 1 T i n i n m i G M (12) then
1,g
is gradient efficient. In particular ifm , the condition (12) is equivalent to q0 , , , , , , , , 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
q n r j n k k j q n q j q n q n r j n k k j q n j q n n r j n k k j n q j n n r j n k k j n j n x w h w g x w h w g x w h w g x w h w g 6. GRADIENT REMEDIABILITY WITH MINIMUM ENERGY
For f L2
0,T;X
, we study the existence and the unicity of an optimal control
P
IR T L
u 2 0, ; ensuring, at the timeT, the gradient remediability of the disturbance f
such that CHuCFf 0. That is the set defined by
2 0, ; / 0
u L T IR C Hu C Ff
D P (13)
is non empty.
We consider the function
2 ; , 0 2 2 P q L T IR IR u Ff C Hu C u J
The considered problem becomes J
uD u
min . For its resolution, we will use an extension of the Hilbert Uniqueness Methods (H.U.M). ForIRq, let us note
2 1 0 2 * * * * *
T IR ds C s T S B P *is a semi-norm on qIR . If the condition (7) is verified then it is a norm if and only if the system (1)-(2) is weakly gradient remediable on
0,T
. The corresponding inner product isgiven by B S
T s
C B S
T s
C ds T
0 * * * * * * * * * , , and the operator :IR q IRq defined by
T ds C s T S BB s T S C C HH C 0 * * * * * * * Then, we have the following proposition:
Proposition 6:
If the condition (7) is verified, then *is a norm on IRqif and only if the system (1)-(2) is weakly gradient remediable on
0,T
and the operator is invertible.Proof :
0 2 1 0 2 * * * * *
T IR ds C s T S B P
0 ; , 0 * * * * 2 P IR T L C s T S B
* *
* *
* *
* * * *
* * ker . ker 0 . C B S T C B F C T S B But ker
B*F**C*
m m
m f M ker 1 (see the Proof of Proposition 5) and we have also
T m m m G M ker 1
m m
m f M ker 1 (see the Proof of Corollary 4) then,ker
B*F**C*
m mT
m G M ker 1 this gives
m mT
m G M ker 1 and since the
Corollary 4 we obtain the result.
On the other hand the operator is symmetric, indeed
q IR q IR C HH C , ** * , q IRq IR C HH C , * * * ,
and positive definite, indeed
0 , 0 , , , , 2 * 0 * * * * * * * * ; , 0 * * * * * * * * * 2
for ds C s T S B C s T S B C H C H C HH C T IR IR T L IR IR P P q qand then is invertible.
We give hereafter the expression of the optimal control ensuring the gradient remediability of
a disturbance f at the timeT. Proposition 7:
For f L2
0,T;X
, there exists a unique q f IR such that f CFf and the
control
f f B S C u * * * * .. verifiesCHuf CFf 0. Moreover, it is optimal and
0, ; * 2 T IR f L P f u . Proof :
From the Proposition 6, the operator is invertible then, for f L2
0,T;X
, there exists aunique q
f IR
Ff C f and if we put
f f B S C u * * * * . . , we obtain
f T f f C S T s BB S T s C ds C Hu
0 * * * * C Ff C Huf CHuf CFf 0.The set D defined by (13) is closed, convex and not empty. For u D , we
have
2 ; , 0 2 T IRP L u uJ . J is strictly convex on D, and then has a unique minimum at
D u * , characterized by 0; , ; , 0 * * 2 P IR T L u v u v D. For v D, we have P
P f f v u L T IR B S C f v B S C f L T IR u ; , 0 * * * * * * * * ; , 0 2 2 . , . , 0 , q IR f f C Hv Since u*is unique, then u* uf andf u is optimal with
2 * 2 ; , 0 * * * * 2 ; , 0 2 2 T IR . f L T IR f L P P f B S C u .6. APPROXIMATIONS AND NUMERICAL SIMULATIONS
This section concerns approximations and numerical simulations of the problem of gradient
remediability. First we give an approximation of f as a solution of a finite dimension linear system Ax b and then the optimal control
f
u , with a comparison between the
corresponding observation noted zu,f and the normal case. 7.1 Approximations
Coefficients of the system: For i, j1,let
q IR j i j i e e
a , where
ei 1iqis the canonical basis of IRq, we have
T i i C S T s BB S T s C e ds e 0 * * * *
n k L Dj j k l m n k L Di i k h m L h m L l m M m m r l N m m r h p m m T m m j i h x w h x w w g w g e a 1 2 1 ' ' 2 ' 2 ' 2 1 1 '1 ' 1 1 ' ' , , , , 1 and f CFf q IR j j C Ff e b ,For N sufficiently large, we have
T L h m s T m N m m r h n k L Dj j k h m j h e f s w ds x w b 0 2 ' ' 1 ' ' 1 1 2 ' , , The optimal control: In this part, we give an
approximation of the optimal control
f u which is defined by
f f B S T C u * * * * . . .Its function coordinates uj, f
. are given by
j
f f j g S T C u * * * , . , . i k h m N m m r h L j h m j T m n k q i f i h x w w g e , ' , 1 ' ' 1 2 ' . ' 1 1 ,
for a large integer N .
Cost: The minimum energy (cost) is defined by
2 1 0 2 * * * * ; , 0 2
T p IR f p IR T L f B S T s C ds u 2 1 1 0 2 1 ' ' 1 ' ' 1 , 1 , ,
p j T N m m r h h m j k h m i n k f i q i s T m g w ds x w h e for N sufficiently large.
The corresponding observation: The observation corresponding to the control is given by
t t f f f u t C S t y C St s Bu s ds C St s f sds z 0 0 0 , Its coordinates
q j f f u j z 1 ,, . are obtained for a large integer N , as follows:
N m r h n k t h m s t j k h m N m r h n k p i t i s t j k h m h m i N m r h k L D h m j L h m n k t f u j m m m f m m j m f ds w s f e h x w ds s u e h x w w g x w h w y e t z 1 ' 1 1 0 ' ' 1 ' 1 1 1 0 , ' ' 1 ' 1 ' ' 0 1 , , ' ' ' ' ' 2 2 ' , , , , , , 7.2 Numerical simulations
We consider without loss of generally the following diffusion system
T t y x y x y T t x f t u x g t x y t x t y p i i i i , 0 0 , 0 , , 0 , , , 0 1 with
0,1 and a Dirichlet boundary condition. In this case, the functions wm
. are defined bywm
x 2sin
mx
;m1. The associated eigenvalues are simple and given bymm22; m1. Then in the case of: an initial state: y0
. 0, a sensor:
D,h
with D
0,1 and h
x 2x2
q1
an efficient actuator:
,g
with
0,1 and g
x 2x3
p1
a disturbance function: defined by
, 240 10 ; 0 t e t x f x t
For
M
N
1
and T 70, we obtain numerical results illustrating the theoretical results established in previous sections. Hence, in figure1, we give the representations of the discreteobservation zu,f corresponding to the control uuf and the disturbance f and z0,0
which represent the normal observation, that is u0 and f 0.
This figure show that for t sufficiently large
t50
, the disturbance f is compensate by the control optimalf
u at the time T
T 70
that is, we have z f
t z
tf
u , 0,0 .
The optimal control uf ensuring the gradient remediability of the disturbance f , is
represented in figure 2.
Fig.2.Representation of the optimal control
f
u (blue line)
8. CONCLUSION
In this paper, which is an extension of previous works to the analysis of the gradient of a
large class of parabolic systems, new notions of weak and exact gradient remediability are
introduced and characterized. The relation between the notion of gradient remediability and
the notion of gradient controllability is also studied. We have shown that a parabolic system
is gradient remediable if it is gradient controllable. Furthermore, we have shown that the
exact and weak gradient remediability of a system may depend on the structure and the
number of the actuators and sensors. Using an extension of the Hilbert Uniqueness Method,
we have shown how to find the optimal control ensuring the gradient remediability of the
known or unknown disturbance. The results of illustrative examples and numerical
approximations are acceptable.
These results are developed for a class of discrete linear distributed parabolic systems, but
the considered approach can be extended to other class of systems with a convenient choice