This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Abstract. We consider a steady-state heat conduction problem Pα with mixedboundary conditions for the Poisson equation depending on a positive parameter α,which represents the heat transfer coefficient on a portion �1 of the boundary of agiven bounded domain in Rn . We formulate distributed optimal control problemsover the internal energy g for each α. We prove that the optimal control gopα andits corresponding system ugopα α
and adjoint pgopα αstates for each α are strongly
convergent to gop, ugop and pgop , respectively, in adequate functional spaces. We alsoprove that these limit functions are respectively the optimal control, and the systemand adjoint states corresponding to another distributed optimal control problem forthe same Poisson equation with a different boundary condition on the portion �1.We use the fixed point and elliptic variational inequality theories.
∗ This paper has been partially sponsored by the Project “Free Boundary Problems for the Heat-DiffusionEquation” from CONICET - UA, Rosario (Argentina).
214 C. M. Gariboldi and D. A. Tarzia
1. Introduction
We consider a bounded domain� inRn whose regular boundary � consists of the unionof two disjoint portions �1 and �2 with meas(�1) > 0 and meas(�2) > 0. We denotewith meas(�) the (n − 1)-dimensional Lebesgue measure of �.
We consider the following two steady-state heat conduction problems P and Pα (foreach parameter α > 0), respectively, with mixed boundary conditions:
−u = g in �, u|�1 = b, −∂u
∂n
∣∣∣∣�2
= q, (1)
and
−u = g in �, −∂u
∂n
∣∣∣∣�1
= α(u − b), −∂u
∂n
∣∣∣∣�2
= q, (2)
where g is the internal energy in�, b is the temperature on�1 for (1) and the temperatureof the external neighborhood of �1 for (2), q is the heat flux on �2 and α > 0 is the heattransfer coefficient of �1 (Newton’s law on �1), that satisfy the following assumptions:
g ∈ H = L2(�), q ∈ L2(�2), b ∈ H 1/2(�1). (3)
Problems (1) and (2) can be considered as the steady-state Stefan problem for suitabledata q , g and b [5], [8], [11], [17], [18], [20].
Let ug and ugα be the unique solutions of the mixed elliptic problems (1) and (2),respectively, whose variational equalities are given by [14]
a(ug, v) = Lg(v), ∀v ∈ V0, ug ∈ K , (4)
and
aα(ugα, v) = Lgα(v), ∀v ∈ V, ugα ∈ V, (5)
where
V = H 1(�), V0 = {v ∈ V/ v|�1 = 0},
K = v0 + V0, (g, h) = (g, h)H =∫�
gh dx, (6)
a(u, v) =∫�
∇u · ∇v dx, aα(u, v) = a(u, v)+ α∫�1
bv dγ,
Lg(v) = (g, v)H −∫�2
qv dγ, Lgα(v) = Lg(v)+ α∫�1
bv dγ
for a given v0 ∈ V, v0|�1 = b.We consider g as a control variable for the cost functionals J : H→R
+0 and
Jα: H→R+0 respectively given by
J (g) = 12‖ug − zd‖2
H +M
2‖g‖2
H (7)
Convergence of Distributed Optimal Controls Problems 215
and
Jα(g) = 12‖ugα − zd‖2
H +M
2‖g‖2
H , (8)
where zd ∈ H is given and M = const. > 0.Then we can formulate the following distributed optimal control problems [7], [9],
[10], [15]:
Find gop ∈ H such that J (gop) = ming∈H
J (g) (9)
and
Find gopα ∈ H such that Jα(gopα ) = ming∈H
Jα(g), (10)
respectively.The use of variational inequality theory in connection with optimal control problems
was done, for example, in [1]–[4], [6], [13] and [16]. In [12] an optimization problemcorresponding to (1) is studied in order to avoid a change phase process.
In Section 2 we prove that the functional J is coercive and Gateaux differentiableon H , and J ′ is a Lipschitzian and strictly monotone application on H . We also provethe existence and uniqueness of the distributed optimal control problem (9) and wecharacterize this optimal energy gop as a fixed point on H of a suitable operator W overits adjoint state pg for a large parameter M .
Similary, in Section 3 we prove that the functional Jα is coercive and Gateauxdifferentiable on H , and J ′α is a Lipschitzian and strictly monotone application on H forall α > 0. We also prove the existence and uniqueness of the distributed optimal controlproblem (10) and we characterize this optimal energy gopα as a fixed point on H of asuitable operator Wα over its adjoint state pgα for a large parameter M .
In Section 4 we study the convergence when α→∞ of the optimal control problem(10) corresponding to the state system (2). We prove that the optimal state system ugopα α
and the optimal adjoint system pgopα αof problem (10) are strongly convergent in V to
the corresponding ugop and pgop for problem (9), respectively, when α→∞. Finally,the strong convergence in H of the optimal control gopα of problem (10) to the optimalcontrol gop of problem (9) is also proved when α→∞.
2. Problem P and Its Corresponding Optimal Control Problem
Let C : H→ V0 be the application such that
C(g) = ug − u0, (11)
where u0 is the solution of problem (4) for g = 0 whose variational equality is given by
a(u0, v) = L0(v), ∀v ∈ V0, u0 ∈ K , (12)
with
L0(v) = −∫�2
qv dγ.
216 C. M. Gariboldi and D. A. Tarzia
Let �: H × H→R and L: H→R be defined by the following expressions:
�(g, h) = (C(g),C(h))+ M(g, h), ∀g, h ∈ H, (13)
L(g) = (C(g), zd − u0), ∀g ∈ H.
We have that a is a bilinear, continuous and symmetric form on V and coercive onV0, that is [14],
∃λ > 0 such that a(v, v) ≥ λ‖v‖2V , ∀ v ∈ V0. (14)
Lemma 2.1.
(i) C is a linear and continuous application.(ii) � is linear, continuous, symmetric and coercive form on H, that is,
�(g, g) ≥ M‖g‖2H , ∀g ∈ H. (15)
(iii) L is linear and continuous on H .(iv) J can be also written as
J (g) = 12�(g, h)− L(g)+ 1
2‖u0 − zd‖2H , ∀g ∈ H. (16)
(v) There exists a unique optimal control gop ∈ H such that
J (gop) = ming∈H
J (g). (17)
(vi) The application g ∈ H→ ug ∈ V is Lipschitzian, that is,
‖ug2 − ug1‖V ≤ 1
λ‖g2 − g1‖H , ∀g1, g2 ∈ H. (18)
Proof. (i)–(iii) This follows as in [12] and [15].(iv) From the definitions of J , � and L , we have
J (g) = 12‖ug + u0 − u0 − zd‖2
H +M
2‖g‖2
H
= 12‖ug − u0‖2
H − (ug − u0, zd − u0)+ 12‖u0 − zd‖2
H +M
2‖g‖2
H
= 12�(g, h)− L(g) + 1
2‖u0 − zd‖2H .
(v) This is a result of (ii)–(iv) [14], [15].(vi) If we take v = ug1 − ug2 ∈ V0 in the variational equality (4) for ug2 , that is,
a(ug2 , ug1 − ug2) = (g2, ug1 − ug2)−∫�2
q(ug1 − ug2) dγ,
and if we take v = ug2 − ug1 ∈ V0 in the variational equality (4) for ug1 , that is,
a(ug1 , ug2 − ug1) = (g1, ug2 − ug1)−∫�2
q(ug2 − ug1) dγ,
Convergence of Distributed Optimal Controls Problems 217
then we obtain
a(ug2 − ug1 , ug2 − ug1) = (g2 − g1, ug2 − ug1)
and taking into account that a is a coercive form we get
Remark 1. We note the double dependence on the parameter α for the optimal statesystem ugopα α
and the adjoint state pgopα α.
Let the operator Wα: H → V ⊂ H be defined by
Wα(g) = − 1
Mpgα, ∀g ∈ H. (53)
We have the following property:
Lemma 3.3. Wα is a Lipschitz operator over H, that is,
‖Wα(g2)−Wα(g1)‖H ≤ 1
λ2αM‖g1 − g2‖H , ∀g1, g2 ∈ H, (54)
and it is a contraction for all M > 1/λ2α .
Proof. If we take v = pg2α− pg1α in variational equality (48) for g2 and g1, respectively,by substracting them and by using the coerciveness of aα we have
λα‖pg2α − pg1α‖2H ≤ aα(pg2α − pg1α, pg2α − pg1α)
= (ug2α − ug1α, pg2α − pg1α)H
≤ ‖ug2α − ug1α‖H‖pg2α − pg1α‖H ,
therefore
‖pg2α − pg1α‖V ≤ 1
λα‖ug2α − ug1α‖H . (55)
Next, taking into account inequalities (46) and (55) we obtain
‖Wα(g2)−Wα(g1)‖H = 1
M‖pg2α − pg1α‖V
≤ 1
λαM‖ug2α − ug1α‖H
≤ 1
λ2αM‖g1 − g2‖H ,
that is, (54).
224 C. M. Gariboldi and D. A. Tarzia
Now, we prove other properties of the functional Jα .
Lemma 3.4.
(i) The operator g ∈ H → pgα ∈ V is strictly monotone, that is,
4. Convergence of Problem Pα and Its Corresponding Optimal Control asα→∞
In this section we prove that the optimal control gopα of problem (10) and its correspondingadjoint state pgopα α
(48) are convergent to the optimal control gop of problem (9) and itscorresponding adjoint state pgop (20), respectively, when the parameter α (heat transfercoefficient on �1) goes to infinity.
Theorem 4.1. Let M > 1/λ21. Then we have:
(i) If pgop and pgopα αare the corresponding adjoint states of problems (9) and (10),
respectively, then
limα→∞‖pgopα α
− pgop‖V = 0. (68)
Convergence of Distributed Optimal Controls Problems 227
(ii) If gop and gopα are the solutions of problems (9) and (10), respectively, then
limα→∞‖gopα − gop‖H = 0. (69)
(iii) If ugop and ugopα αare the corresponding solutions of problems P and Pα ,
respectively, then
limα→∞‖ugopα α
− ugop‖V = 0. (70)
Proof. We prove some preliminary results for the three cases.Since gopα is the solution of problem (10), we have the following inequality:
12‖ugopα α
− zd‖2H +
M
2‖gopα‖2
H ≤ 12‖ugα − zd‖2
H +M
2‖g‖2
H , ∀g ∈ H,
then, taking g = 0, we have
12‖ugopα α
− zd‖2H +
M
2‖gopα‖2
H ≤ 12‖u0α − zd‖2
H ≤ C3, ∀α > 0,
where C3 is a constant independent of parameter α because u0α is convergent whenα→∞. Therefore
‖gopα‖H ≤ C4 and ‖ugopα α‖H ≤ C5, (71)
where C4 and C5 are constants independent of α.Now, if we take v = ugopα α
− ugop in the variational equality (5), following [18] weobtain, for α > 1,
λ1‖ugopα α− ugop‖2
V + (α − 1)∫�1
(ugopα α− ugop)
2 dγ
≤ aα(ugopα α− ugop , ugopα α
− ugop)
≤ C6‖ugopα α− ugop‖V ,
where C6 = C6(gop, q, ugop) is independent of α. Next, we have
(a) ‖ugopα α− ugop‖2
V ≤C6
λ1,
(b) (α − 1)∫�1
(ugopα α− ugop)
2 dγ ≤ (C6)2
λ1,
(72)
and therefore we deduce that
∃ η ∈ V such that ugopα α⇀ η weakly in V, (73)
228 C. M. Gariboldi and D. A. Tarzia
and because of the following inequalities,
0 ≤∫�1
(η − ugop)2 dγ ≤ lim inf
α→∞
∫�1
(ugopα α− ugop)
2 dγ = 0,
we obtain that η ∈ K .Next, if we take v = pgopα α
− pgop in the variational equality (48) we get
λ1‖pgopα α− pgop‖2
V + (α − 1)∫�1
(pgopα α− pgop)
2 dγ
≤ aα(pgopα α− pgop , pgopα α
− pgop)
≤ C7‖pgopα α− pgop‖V ,
with C7 = C7(C5, pgop). Next, we obtain
(a) ‖pgopα α− pgop‖2
V ≤C7
λ1,
(b) (α − 1)∫�1
(pgopα α− pgop)
2 dγ ≤ (C7)2
λ1,
(74)
and therefore we deduce that
∃ξ ∈ V such that pgopα α⇀ ξ weakly in V (75)
and by the following inequality,
0 ≤∫�1
(ξ − pgop)2 dγ ≤ lim inf
α→∞
∫�1
(pgopα α− pgop)
2 dγ = 0,
we obtain ξ ∈ V0.Now we consider v ∈ V0 and, taking into acount (73) and (75), from the variational
equality (48) we have
a(ξ, v) = (η − zd , v), ∀v ∈ V0, ξ ∈ V0. (76)
Next, from (71) we deduce that there exists f ∈ H such that gopα ⇀ f weakly inH . Therefore if we put v ∈ V0 in the variational equality (5) and we pass to the limitα→∞, we obtain
a(η, v) = ( f, v)−∫�2
qv dγ, ∀v ∈ V0, η ∈ K . (77)
Now, taking into account Lemma 3.3 and the facts that gopα ⇀ f weakly in H andpgopα α
⇀ ξ weakly in V , we have
f = − 1
Mξ in H. (78)
Convergence of Distributed Optimal Controls Problems 229
Therefore from the uniqueness of fixed point we have
gop = − 1
Mpgop in H, (79)
and then we obtain that f = gop, η = ugop and ξ = pgop .Moreover, from (75) and the following computation,
λ1‖pgopα α− pgop‖2
V ≤ aα(pgopα α− pgop , pgopα α
− pgop)
= aα(pgopα α, pgopα α
− pgop)− a(pgop , pgopα α− pgop)
= (ugopα α− zd , pgopα α
− pgop)− a(pgop , pgopα α− pgop)
we have (68).From Lemmas 2.3 and 3.3 it results that
‖gopα − gop‖H = 1
M‖pgop − pgopα α
‖H ≤ 1
M‖pgop − pgopα α
‖V
and therefore (69) holds.Now we have
λ1‖ugopα α− ugop‖2
V ≤ aα(ugopα α− ugop , ugopα α
− ugop)
= aα(ugopα α, ugopα α
− ugop)− aα(ugop , ugopα α− ugop)
= Lgopα α(ugopα α
− ugop)− a(ugop , ugopα α− ugop)
− α∫�1
b(ugopα α− b) dγ
= a(ugopα, ugopα α
− ugop)− a(ugop , ugopα α− ugop)
and taking into account (69) and the fact that ugopα→ ugop strongly in V when α→∞
because of (18), we get (70).
References
1. Abergel F (1998) A non-well-posed problem in convex optimal control. Appl Math Optim 17:133–1752. Adams DR, Lenhart SM, Yong J (1998) Optimal control of the obstacle for an elliptic variational
inequality. Appl Math Optim 38:121–1403. Bensoussan A (1974) Teorıa moderna de control optimo. Cuadern Inst Mat Beppo Levi # 7. Univ Nac
Rosario, Rosario4. Bergouniux M (1997) Optimal control of an obstacle problem. Appl Math Optim 36:147–1725. Berrone LR, Garguichevich GG (1992) On a steady Stefan problem for the Poisson equation with flux
and Fourier’s type boundary conditions. Math Notae 36:49–616. Casas E (1986) Control of an elliptic problem with pointwise state constraints. SIAM J Control Optim
24:1309–13187. Cea J (1971) Optimisation: theorie et algoritmes. Dunod, Paris8. Duvaut G (1976) Problemes a frontiere libre en theorie des milieux continus. Rapport de Recherche
No 185, LABORIA-IRIA, Rocquencourt9. Ekeland I, Teman R. (1973) Analyse convexe et problemes variationnelles. Dunod-Gauthier Villars, Paris
230 C. M. Gariboldi and D. A. Tarzia
10. Faurre P (1988) Analyse numerique. Notes d’optimization. Ellipses, Paris11. Garguichevich GG, Tarzia DA (1991) The steady-state two-phase Stefan problem with an internal energy
and some related problems. Atti Sem Mat Fis Univ Modena 39:615–63412. Gonzalez RLV, Tarzia DA (1990) Optimization of heat flux in a domain with temperature constraints.
J Optim Theory Appl 65:245–25613. Haslinger J, Roubıcek T (1986) Optimal control of variational inequalities. Approximation theory and
numerical realization. Appl Math Optim 14:187–20114. Kinderlehrer D, Stampacchia G (1980) An Introduction to Variational Inequalities and Their Applications.
Academic Press, New York15. Lions JL (1968) Controle optimal des systemes gouvernes par des equations aux derivees partielles.
Dunod-Gauthier Villars, Paris16. Mignot F, Puel JP (1984) Optimal control in some variational inequalities. SIAM J Control Optim
22:466–47617. Tabacman ED, Tarzia DA (1989) Sufficient and/or necessary condition for the heat transfer coefficient
on �1 and the heat flux on �2 to obtain a steady-state two-phase Stefan problem. J Differential Equations77:16–37
18. Tarzia DA (1979) Sur le probleme de Stefan a deux phases. C. R. Acad Sci Paris Ser A 288:941–94419. Tarzia DA (1979) Una familia de problemas que converge hacia el caso estacionario del problema de
Stefan a dos fases. Math Notae 27:157–16520. Tarzia DA (1988) An inequality for the constant heat flux to obtain a steady-state two-phase Stefan
problem. Eng Anal 5:177–181
Accepted 16 September 2002. Online publication 22 April 2003.