INSTITUT NATIONAL DE RECHERCHE EN INFORMATIQUE ET EN AUTOMATIQUE
Control Problems with State Constraints for the Penrose-Fife Phase-field model Werner Horn, Jan Sokołowski, Ju¨rgen Sprekels
N˚ 2491 Fevrier 1995
PROGRAMME 6
ISSN 0249-6399
apport de recherche
Control Problems with State Constraints for the Penrose-Fife Phase-eld model
Werner Horn, Jan Sokoªowski , Jürgen Sprekels
Programme 6 Calcul scientique, modélisation et logiciel numérique Projet Numath Rapport de recherche n2491 Fevrier 1995 22 pages
Abstract: This article gives an optimality system for a control problem with state constraints for a Penrose-Fife model for phase transitions. Key-words: state constraints, optimality system, phase transition.
(Résumé : tsvp)
[email protected]
Unite´ de recherche INRIA Lorraine Technoˆpole de Nancy-Brabois, Campus scientifique, 615 rue de Jardin Botanique, BP 101, 54600 VILLERS LE`S NANCY (France) Te´le´phone : (33) 83 59 30 30 – Te´le´copie : (33) 83 27 83 19 Antenne de Metz, technopoˆle de Metz 2000, 4 rue Marconi, 55070 METZ Te´le´phone : (33) 87 20 35 00 – Te´le´copie : (33) 87 76 39 77
Problèmes de contrôle avec contraintes sur l'état pour le modèle de Penrose-Fife.
Résumé : On étudie des problèmes de contrôle avec des contraintes sur l'état pour des modéles non linéaires de type Penrose-Fife "Phase-eld". On dérive des conditions d'optimalité pour les problèmes de contrôle. Mots-clé : contraintes sur l'état, condition d'optimalité, transition de phase
Control Problems with State Constraints for the Penrose-Fife Phase-eld model 3
1 Introduction In this article we consider optimal control problems governed by the following system of quasi-linear parabolic equations, () = K ? s0 () ? (1) t
Tt =
1
0
1
?M1 T
T ? ()t + v;
(2)
on Q = (0; t), where IR3 is a bounded domain with a suciently smooth boundary @ . s0 denotes a double well potential. We let @Q = @ (0; t), and impose the following boundary conditions @T = ? (T ? w) @n @ =0 @n
on
@Q
(3)
on
@Q;
(4)
and the initial conditions (x; 0) = 0 (x); T (x; 0) = T0(x):
(5)
These equations arise in a model for phase transitions introduced by Penrose and Fife [10]. In this setting T denotes the absolute temperature, and a non-conserved order-parameter. Several papers have appeared in connection with the existence and uniqueness of solutions to this system as well as other analytical aspects of this system and related systems. We refer the reader to [7, 13, 4, 6, 8, 9] for some specic treatments. A more general discussion of systems of this type can be found in [2]. We will make similar assumptions in this article as in [7, 13], namely, for the potential s0 we will assume that either (A) s0 2 C 3(IR) and there exists a constant C > 0 such that s000 () > ?C for all 2 IR. or (B) s0 = log + (1 ? ) log(1 ? ). Furthermore, we will make the following simplifying assumptions:
RR n2491
4
Werner Horn, Jan Sokoªowski, Jürgen Sprekels
() = a + b, for a positive constant a. To simplify notations we will, without
loss of generality, use a = 1 and b = 0, i.e. use () = . In the boundary conditions we let = 1. To state an existence result we have to make some regularity assumptions and compatibility conditions. In particular we have
(H1) (H2)
@ 0 2 H 4( ); @n (x) = 0; 8x 2 @ ; @n@
T0 2 H 3( ); T~(x) =
?s00 (0) + T00 + 0 (x) = 0; 8x 2 @ :
(x) + T0 (x) > 0; 8x 2 @ ; T0(x) > 0; 8x 2 : Finally, we introduce some Banach spaces which will be widely used throughout this article. X1 X2 V W
= = = =
@T0 @n
C ([0; t]; H 4( )) \ C 1([0; t]; H 2( )) \ C 2 ([0; t]; L2( )); C ([0; t]; H 3( )) \ C 1([0; t]; H 1( )) \ H 4;2(Q); H 2(0; t; L2( )) \ H 1(0; t; H 2( )); 3 H 2(0; t; H 2 (@ )):
Using these conditions one can prove the following existence result (cf. [7, 13])
Proposition 1 Suppose (H1) and (H2) are satised. Then there exists a unique global smooth solution (; T ) 2 X1 X2 to the equations (1)(2) with the boundary conditions (3)(4). Furthermore, there exists a constant ct > 0 such that T (x; t) ct for all (x; t) 2 Q, and in the case (B) there exist constants 0 < at < bt < 1, such that at (x; t) bt for all (x; t) 2 Q. In Section 2 of this article we will state the optimal control problem with state constrains and discuss it. In Section 3 we will investigate the related observation operator and prove its dierentiability in the setting of Section 2. Finally, we will derive the necessary conditions for optimality in Section 4 of this paper.
2 Optimal Control Problem The state equations (1)(2) give rise to several interesting optimal control problems. In this article we want to control the state (; T ) by using the source term v in (2) and the boundary term w in (4) as controls. However, we want to put local constraints on the state as well.
INRIA
Control Problems with State Constraints for the Penrose-Fife Phase-eld model 5
In order to formulate this problem in a precise manner we need to introduce some additional notation. We start by dening the cost functional I (; T ; v; w) =
2
1
^
2
(t ) ? (t ) L2 ( ) + 2
T ? T^
L2(Q) 2
2 Z 4 t
(6)
+ 3 kv k2L2 (Q) + kw(t)k2L2(@ ) dt; 2 2 0 for given target functions ^ 2 X1 and T^ 2 X2 . Next let n ~ = w 2 W : w(x; 0) = T~(x); 8x 2 @ ; W w(x; t) ; jwt(x; t)j < k; 8(x; t) 2 @Qg ; where T~ is the function introduced by (H2) and and k are suitably chosen positive constants. We use this set to introduce ~ K =V W The set Uad of admissible controls is a closed, convex and bounded subset of K . To state the local state constraints we use the constants 0 < K1 < K2 and K3 < K4 to dene
Yad = f(; T ) 2 X1 X2 : K1 T (x; t) K2 ^ K3 (x; t) K4; 8(x; t) 2 Qg ; the set of admissible states. Note, that this set has a nonempty interior. We can now state the optimal control problem under consideration.
(7)
Optimal Control Problem (CP)
Minimize I (; T ; v; w) under the following conditions: 1. (; T ) satises the state equations (1)(2) and the initial and boundary conditions (3)(5). 2. (v; w) 2 Uad . 3. (; T ) 2 Yad.
Remarks: RR n2491
6
Werner Horn, Jan Sokoªowski, Jürgen Sprekels
Clearly the initial values (0; T0) must also satisfy the constraints K1 T (x) K2 and K3 0 (x) K4 for all x 2 . The authors of [14] considered a similar but weaker control problem. In par-
ticular, they did not put local constraints on the state. Moreover, their treatment focuses on the function s00 () = ? 3 . However, this latter restriction can be easily removed and their argument extended to the cases (A) and (B) investigated here (see [5], for a sketch of this argument). We can therefore use their results whenever they are applicable. In the study of the control problem (CP) it is useful to introduce the observation operator S . To do this we dene the space of observations B by B = (C ([0; t]; H 2( ))) (C ([0; t]; H 2( ))): (8) Next dene S : K !B S : (v; w) 7! (; T );
(9) (10) i. e. S assigns to every pair (v; w) 2 K the pair (; T ) which solves (1)(5) for the given v and w. Since X1 X2 B and by virtue of Proposition 1 this operator S is well dened. Using this operator one sees that the cost functional I (; T ; v; w) depends only on the controls v and w i.e. we can rewrite it as J (v; w) = I (; T ; v; w)j(;T )=S (v;w) :
In the following section we will study the properties of this operator S . In Section 4 these properties will be used to derive the necessary conditions of optimality. Remark: Since the authors of [14] did not consider state constraints, they could use a larger space of observations with a coarser topology.
3 Dierentiability of the Observation Operator We now turn our attention to the observation operator S dened in (9)(10). This operator is well-dened, and also due to Proposition 1 there exist positive constants and such that
kkX1 + kT kX2 ; 8(v; w) 2 Uad; T (x; t) > 0; 8(x; t) 2 Q:
(11) (12)
INRIA
Control Problems with State Constraints for the Penrose-Fife Phase-eld model 7
Moreover, if s0 () is of the form given in case B, there exist constants 0 < ^at < ^bt < 1 such that a^t (x; t) ^bt ; 8(x; t) 2 Q: (13) In order to prove dierentiability of the observation operator S one has to rst improve the stability result of [14]. To do this we let (i; Ti) = S (vi; wi); i = 1; 2 and (vi ; wi) 2 Uad . We dene = 1 ? 2 , T = T1 ? T2 , v = v1 ? v2 , and w = w1 ? w2. Using these notations we have the following result.
Proposition 2 There exists a constant C > 0 such that
2
2
2
2 Z t
2 max t (t) H 1 + (t) H 3 + T H 2 + T t (t) L2 +
tt(t) dt 0 0 such that (v h; w k) 2 Uadg ; (26) for (v; w) 2 Uad . Proposition 3 Suppose (H1) and (H2) hold and (v; w) 2 Uad. Then the observation operator S : K ! B; has a directional derivative ( ; ) = D(h;k)S (v; w) in the direction K + (h; k). Furthermore, at S (v; w) = (; T ), this directional derivative ( ; ) 2 X1 X2 is the unique solution of the linear parabolic initial-boundary value problem 1 00 ? = ? s () ? ; t
t ?
T2
T
0
T2
= ( )t + h;
@ @ = 0 ; + = k; on @ ; @n @n (0; x) = (0; x) = 0; on
The corresponding result holds for the directional derivative D(?h;?k) S (v; w) at (v; w) in direction (h; k) 2 K ? (v; w).
INRIA
Control Problems with State Constraints for the Penrose-Fife Phase-eld model 11
Proof: As in [14] we let
(; T ) = S (v + h; w + k): Furthermore, we use the notation of the previous Proposition and let 1 : = ? ; T = T ? T ; = TT
Finally, dene
p = ? ; q = T ? :
It is clear that the linear parabolic system in the statement admits a unique solution ( ; ) 2 X1 X2. To continue suppose that (h; k) 2 K + (v; w) and suppose that there is a > 0 such that (v + h; w + k) 2 Uad ; 8 2 (0; ). We have to show k(p; q)kB = o(); as ! 0+ : (27) Using our notation p and q observe the following system of linear parabolic boundary value problems. p T !
pt ? p = s00 () ? s00 () ? s000 () + qt ?
q T2
2 = t p + pt + t ? T
@q + q = 0; on @Q @n 0 = p(x; 0) = q (x; 0)
@p = 0; @n
We prove (27) in several steps. Step1: In [14] the authors show that
T
? T2 q + T T 2 ? T (28)
(29) (30) (31)
Z t 2 2 2 + kq (s)k2 + kp(s)k2 max k p ( t ) k + k q ( t ) k k p ( s ) k + ds C4; 1 1 2 t H H H 0tt 0 q(32)
for a suitable constant C > 0. We continue from there by multiplying (29) by T 2 t . After integrating the resulting equation over [0; t] we obtain
2 Z Z q @ q Z t
qt
2
ds + 1
r q2 (t)
? t dxds (33) 2 T 0 T 0 @ T 2 t @n T 2 Z t Z qtqTt Z t Z qt qTt dxds; f 2 ? 2 3 dxds ? 2 = T T 0 T3 0
(34)
RR n2491
12
Werner Horn, Jan Sokoªowski, Jürgen Sprekels
where f is given by
!
T 2 t p + pt + t ? : T
From Proposition 1 and the earlier estimates we see that
Z t 0
kf (s)k2 ds C14;
for a suitable constant C1.Furthermore, we have
Z t
qTt
2
3 (s)
ds C24; T 0
due to earlier estimates. For the boundary term we observe @ q q w = T2 T ? 1 : @n T 2
Therefore we have
Z t Z q @ q Z t Z q q w dxds = 1 ? T dxds 0 @ T 2 t @n T 2 0 @ T 2 t T 2
2
c1
Tq2 (t)
Z t Z w dxds +c2 q 2 T t 0 @
2 c1
r Tq2 (t)
+ c3 kq(t)k2
Zt +c2 kq (s)k2L4 (@ )
Tw
ds t
0 q
2 Zt c1
r T 2 (t)
+ c44 + c5 0 kq(s)k2H 1 ds
In the last line of this estimate we used (25). Combining these estimates, using Young's inequality and choosing suciently small we obtain
q
2 Z t
q
2
r 2 (t)
+
t
ds C34: max 0tt T T 0
INRIA
Control Problems with State Constraints for the Penrose-Fife Phase-eld model 13
It immediately follows max kq (t)kH 1 + 2
0tt
Z t 0
kqt k2 ds C44:
(35)
Step 2: In the next step, we take the derivative of (28) with respect to t to get
0
ptt ? pt = s0 () ? s00 () ? s000 ()
p 2 + T ? T 2 q + T T ? T : (36) t t
We observe that
jF1;tj = s00() ? s00() ? s000 () t t s000 () ? s000 () ? s0000 () + s0000 ()tp + s000 ()pt + s000 () ? s000 () :
Using the mean-value theorem one easily sees that
Z t 0
kF1;t(s)k2 ds c64;
(37)
for a suitable constant c6. Next we observe that pt pTt t T ? ? q + 2 3t q ? 2 qt + t T 2 ? 2 TtT 2 2 2 T T T T T T T 2 +2 T T T t + T T t ? t T ? T t ? T t Since both t and Tt are elements of C ([0; t]; H 1( )) we see that F2;t =
Z t 0
kF2;t(s)k2 ds c74;
(38)
for a suitable constant c7. So if one multiplies (36) by pt and integrates the result over [0; t] one gets immediately max kpt(t)k2 + 0tt
Z t 0
kpt(s)k2H 1 ds C54
We can now apply the standard elliptic regularity estimates to get max kp(t)k2H 2 C64 : 0tt RR n2491
(39) (40)
14
Werner Horn, Jan Sokoªowski, Jürgen Sprekels
Furthermor, we can multiply (36) by ptt, integrate the result over [0; t] and use (37) and (38) again to get max kpt kH 1 + 2
0tt
Z t 0
kptt(s)k2 ds C74;
(41)
for a suitable constant C7. Step 3: To continue we take the time derivative of (29) to obtain qtt ?
where
q T2
t
= F3;t (x; t);
(42)
T 2 F3;t = t p + pt + t ? T
!!
: t
To simplify notations we introduce ^ = T , which has the same properties as . We observe that T 2 ^ = 2T t^T + 4^rT rT t + 4T t rT r^ + 2T ^T t + 4T rT t r^ t
2 +2T T t ^ + 2 rT ^t + 2^tT T + 4T rT r^t + T 2 ^t :
Using the results of Proposition 1, we can bound
T (t)
H 2 by c9 for a suciently large constant c9 . Furthermore, we know that T has the same regularity as ^ which enables us to bound terms of the form
Z t
2
T H 2 ds; and 0
max
T t (t)
H 1 0tt
by constants. Combining these properties we see that
Z t
2
2
T ^ t (s) ds c102 0
for a suitable constant c10. It follows that
Z t 0
kF3;t(s)k2 ds c112:
(43)
INRIA
Control Problems with State Constraints for the Penrose-Fife Phase-eld model 15
We multiply (42) by qt and integrate the result over [0; t] to get
1 kq (t)k2 + Z t Z rq r q dxds ? Z t Z q @ q dxds t t 2 t T2 t 0
0 @ @n T 2 t
Z t 0
! 21 Z t
kF3;t(s)k ds 2
0
for a suitable constant c12. We next observe that
q
Z t Z 0
Z t Z
! 21
kqt(s)k ds 2
q
c123;
? 2 qT3t dxds 2 T T
rqt r Z t
rqt
2
(s)
ds = 0 q Z tTZ Tt rqt T 3 rT + T 3 rq dxds ?2 qTt Z0t Z
q +2 rqt 3 T 4 rT ? T 3 rTt dxds
rqt r T 2 dxds = t
0
0
t
One sees that the mixed terms on the right can be treated via Young's inequality, and that we can use the fact that
Z t 0
kqk2H 2 ds c144;
and the other earlier estimates on q . Finally we observe
Z t Z 0
@
qt
@ q @n T 2
t
1 @q qt @T t ? 2 dxds T 2 @n T 3 @n 0 @
Z t Z qTt @T Tt @q q @Tt ? ? qt 3 4 +2 dxds T @n T 3 @n T 3 @n 0 @
Z t Z q 2 1 @T dxds t 1 + 2 = ? 0 @ T 2 Z t Z qqt T@T@nt Tt @T ?2 ? 3 T 2 @n dxds T + T 3 t @n
dxds =
Z t Z
0
In the rst term we observe that
qt
@
2 L1(@Q): 1 + 2 T1 T@T @n
RR n2491
16
Werner Horn, Jan Sokoªowski, Jürgen Sprekels
In the second term one has 1 T + @Tt ? 3 Tt @T 2 L2(0; t; L1 (@ )): T
t
@n
T 2 @n
Using this we get
Z t Z ! 12 Z t
qt
2 qqt Tt @T @Tt 2 2
0 @ T 3 Tt + @n ? 3 T 2 @n dxds C9 0 T L2 (@ ) kqkL2 (@ ) ds :
Observe that
kq(t)k2L2(@ ) C104:
This implies that we are left to treat estimate of the form
Z t
qt
2
ds: T L2 (@ ) 0
We do this by using
Z t 0
kg(s)k2L2 (@ ) ds
Z t 0
krg(s)k2 ds + C^
Z t 0
kg(s)k2 ds;
for a suitable constant C^ . We can now combine all these estimates and use the properties of T to conclude max kqt (t)k +
0tt
Z t 0
krqtk2 ds C113;
(44)
for a suitable constant C11. From elliptic regularity estimates it follows, that the same estimate holds for max kq (t)k2H 2 : 0tt This nishes the proof of the proposition.
4 Optimality conditions We return to the optimal control problem (CP) of Section 2. We introduced the non-linear observation operator S (9)(10). We can write S in components (S1; S2) as follows. ! ! S ( v; w ) 1 S (v; w) = S (v; w) = T (45) 2
INRIA
Control Problems with State Constraints for the Penrose-Fife Phase-eld model 17
Proposition 3 states that this operator is Gateaux dierentiable with Gateaux derivative ! ! DS ( v; w )( h; k ) 1 (46) DS (v; w)(h; k) = DS (v; w)(h; k) = ; 2
given by the following system of linearized equations 1 00 ? s0 () ? T 2 ; = t? T t ?
T2
= ( )t + h;
(47) (48)
@ @ = 0 ; + = k; on @ ; @n @n (0; x) = (0; x) = 0; on
(49) (50) An application of the Lagrange multiplier rule implies that there exists 0 and Borel measures 1 ; 2 ; 3; 4, with the properties: i (f(x; t) 2 QjT (x; t) 6= Ki g) = 0; i = 1; 2; (51) (52) i (f(x; t) 2 Qj(x; t) 6= Ki g) = 0; i = 3; 4; such that + j1 j + j2 j + j3 j + j4 j > 0; where ji j; i = 1; :::; 4; denotes the norm of the measure i . The constants Ki are the ones given in the state constraints (7). To continue, we denote = 1 ? 2 ; = 3 ? 4 . The abstract optimality system for the control problem under consideration is given below by () and (). The rst condition takes the form ()
Z Z 8(; ) 2 Yad : ( ? T )d + ( ? )d 0;
where (; T ) = S (v; w) is a solution to the state equations for optimal controls (v; w) 2 Uad . For the second condition,we recall the notation. We denote by I (; T ; v; w) the cost functional, i.e. J (v; w) = I (S1(v; w); S2(v; w); v; w), then the gradient of the cost functional, with respect to the controls takes the form hDJ (v; w); (h; k)i = hD1I (; T ; v; w); D1S (; T )(h; k)i + hD2I (; T ; v; w); D2S (; T )(h; k)i + hD3I (; T ; v; w); hi + hD4I (; T ; v; w); ki: RR n2491
18
Werner Horn, Jan Sokoªowski, Jürgen Sprekels
The second optimality condition is of the form () hDJ (v; w); (h ? v; k ? w)i + h[DS2(v; w)](h ? v; k ? w)]; i +h[DS1(v; w)](h ? v; k ? w)]; i 0; for all (h; k) 2 Uad , where [DSi (v; w)] denotes the adjoint to [DSi(v; w)]; i = 1; 2. Assuming that the Slater condition is satised, we can take = 1. In the present case the Slater condition (S ) means, that there exists an optimal control (h0 ; k0) 2 Uad such that for all (x; t) 2 Q, K1 < T (x; t) + [DS2(v; w)(h0 ? v; k0 ? w)](x; t) < K2 K3 < (x; t) + [DS1(v; w)(h0 ? v; k0 ? w)](x; t) < K4 Furthermore, an adjoint state is introduced in order to simplify the latter optimality condition. To this end, we rewrite the linearized equations in the form. L11( ) + L12() = 0; (53) L21( ) + L22() = h; (54) with nonhomogeneous boundary condition where we denote
@ + = k; on @ ; @n
L11( ) = t ? ?
1
L12() = T2 ; L21( ) = ? ( )t ; L22() = t ? T2 :
? s000 () ; T
(55) (56) (57) (58) (59)
Therefore, for any functions (q; p) 2 V V it follows that (L11 ( ) + L12 (); q )V = 0; (L21 ( ) + L22 (); p)V = (h; p)V ; and the latter term, by an application of the associated Green formula, can be written in the form
@ + ; p (L22 (); p)V = A(; p) ? ` @n
(60)
INRIA
Control Problems with State Constraints for the Penrose-Fife Phase-eld model 19
with an appropriate bilinear form A(; ), and a boundary form `(; ), which will be @ + = 0 it follows that specied below. In particular, for @n (L22(); p)V = A(; p) : Hence, the system becomes (L11( ) + L12(); q )V = 0; (L21( ); p)V + A(; p) = (h; p)V + `(k; p): In order to identify the boundary form `(k; p), we need Green formulae for the subsequent terms in the scalar product of the space L2 (Q) which are given below.
? T2 ;
L2 (Q)
= r T 2 ; r
L2 (Q)
Z t @ ? ; @n T 2 0
L2 (@ )
and, in view of the boundary conditions, it follows that
@ ? @n T 2 ; 2 L (@ ) @ 1 1 = ( ? k) T 2 ; 2 ? @n T 2 ; 2 1 @ L 1(@ ) L1 (@ ) = 2? ? kT2; ; T @n T 2 2 L (@ )
Similarly,
? r T2 ; r ()
L2 ( )
=
L2 (@ )
:
2 ; 2 T L ( ) @ ? T 2 ; @n 2
L (@ )
and
? T2 ;
L2 ( )
=
r T 2 ; r () 2 @ L ( ) ? @n T 2 ; 2 : L (@ )
RR n2491
;
dt;
20
Werner Horn, Jan Sokoªowski, Jürgen Sprekels
We have also the following relation on the boundary @ , (cf. [12]),
2 2 = ? 2 + @ 2 + @ 2 2 ; T
T
@n T
@n T where ? is the LaplaceBeltrami operator on ? = @ and where denotes the tangential divergence of the normal vector eld on ?, i.e. = div ? n, in the notation
of [12]. The adjoint state equations are introduced in the following way. Assume that the functions (q; p) 2 V V satisfy the following variational equation (L11 ( ); q )V + (L12( ); q )V + (L21( )Z; p)V + (L22 ( ); p)V Z = hD1I (; T ; v; w); i + d + hD2I (; T ; v; w); i + d;(61) for all suciently smooth functions ; satisfying homogeneous initial conditions and the homogeneous boundary conditions @ = 0; @n
@ and @n + = 0:
(62)
Using the Lions projection theorem, see e.g. [15] for a variant of this theorem, one can show that these functions are uniquely determined. The system (61) can be rewritten in the form (L11 ( ); q )V + (L12( ); q )V + (L21( )Z; p)V + A(; p) Z = hD1I (; T ; v; w); i + d + hD2I (; T ; v; w); i + d; @ where the boundary condition @n + = 0 is imposed directly in the equation. If we replace ; by ; , it follows that
Z
Z
hD1I (; T ; v; w); i + hD2I (; T ; v; w); i + d + d = (L11( ); q )V + (L12(); q )V + (L21( ); p)V + A(; p) = (L11( ); q )V + (L12(); q )V +(L21( ) + L22(); p)V + `(k; p) = (h; p)V + `(k; p) :
Using the above construction, it follows that for = 1 the necessary optimality conditions can be rewritten in the following form.
INRIA
Control Problems with State Constraints for the Penrose-Fife Phase-eld model 21
Theorem 1 Assume that condition (S ) is satised. Then there exist ; and the adjoint state (q; p) such that the optimality system for the control problem includes the state equation, the adjoint state equation, and the condition (), as well as the following condition hD3I (; T ; v; w); h ? vi + (h ? v; p)V + hD4I (; T ; v; w); k ? wi + `(k ? w; p) 0 for all (h; k) 2 Uad .
References [1] R. A. Adams, Sobolev Spaces, Academic Press, New York (1984). [2] H. Amann, Nonhomogeneous linear and quasilinear elliptic and parabolic boundary value problems, preprint, 1993. [3] E. Casas, Boundary control of semilinear elliptic equations with pointwise state constraints, SIAM J. Control Optim., 31(1993), 9931006. [4] P. Colli, J. Sprekels, On a Penrose-Fife model with zero interfacial energy leading to a Phase-eld system of relaxed Stefan type, Ann. Math. Pura Appl. (4), to appear. [5] W. Horn, Mathematical Aspects of the Penrose-Fife Phase-eld model, Control and Cybernetics, Vol. 23, No. 4, 677690, 1994. [6] W. Horn, Ph. Laurençot, J. Sprekels, Global solutions to a Penrose-Fife phaseeld model under ux boundary conditions for the inverse temperature, submitted. [7] W. Horn, J. Sprekels, S. Zheng, Global Smooth solutions to the Penrose-Fife Model for Ising ferromagnets, Adv. Math. Sci. Appl., to appear. [8] N. Kenmochi, M. Niezgódka, Systems of nonlinear parabolic equations for phase change problems, Adv. Math. Sci. Appl. 3 (1993/94), 89117. [9] Ph. Laurençot, Solutions to a Penrose-Fife model of phase-eld type, J. Math. Anal. Appl. 185 (1994), 262274.
RR n2491
22
Werner Horn, Jan Sokoªowski, Jürgen Sprekels
[10] O. Penrose, P. C. Fife, Thermodynamically consistent models of phase-eld type for the kinetics of phase transitions, Physica D 43 (1990) 4462. [11] J. Sokoªowski, J. Sprekels, Control Problems for Shape Memory Alloys with Constraints on the Shear Strain, Lecture Notes in Pure and Applied Mathematics, Vol. 165 (1994) Marcel Dekker, G. Da Prato, L. Tubaro (Eds), 189-196. [12] J. Sokoªowski, J.P. Zolesio, Introduction to Shape Optimization Springer Verlag (1992). [13] J. Sprekels, S. Zheng, Global smooth solutions to a thermodynamically consistent model of phase-eld type in higher space vvdimensions, J. Math. Anal Appl. 176, 200223, 1993. [14] J. Sprekels, S. Zheng, Optimal Control problems for a thermodynamically consistent model of phase-eld type for phase transitions, Adv. in Math. Sci. and Appl., Vol. 1, No. 1 (1992), 113125. [15] Tapas Mazumdar, Generalized projection theorem and weak noncoercive evolution problems in Hilbert spaces, J. Math. Anal. Appl. 46 (1974), 143168.
INRIA
Unite´ de recherche INRIA Lorraine, Technopoˆle de Nancy-Brabois, Campus scientifique, 615 rue du Jardin Botanique, BP 101, 54600 VILLERS LE`S NANCY Unite´ de recherche INRIA Rennes, Irisa, Campus universitaire de Beaulieu, 35042 RENNES Cedex Unite´ de recherche INRIA Rhoˆne-Alpes, 46 avenue Fe´lix Viallet, 38031 GRENOBLE Cedex 1 Unite´ de recherche INRIA Rocquencourt, Domaine de Voluceau, Rocquencourt, BP 105, 78153 LE CHESNAY Cedex Unite´ de recherche INRIA Sophia-Antipolis, 2004 route des Lucioles, BP 93, 06902 SOPHIA-ANTIPOLIS Cedex
E´diteur INRIA, Domaine de Voluceau, Rocquencourt, BP 105, 78153 LE CHESNAY Cedex (France) ISSN 0249-6399