See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/241735203

Time optimal problems with Dirichlet boundary controls

Article in Discrete and Continuous Dynamical Systems · November 2003

DOI: 10.3934/dcds.2003.9.1549

CITATIONS READS

25 208

2 authors:

Nadir Arada Jean-Pierre Raymond

43 PUBLICATIONS 757 CITATIONS

Université Toulouse III - Paul Sabatier
134 PUBLICATIONS 3,329 CITATIONS
SEE PROFILE
SEE PROFILE

All content following this page was uploaded by Nadir Arada on 25 July 2015.

The user has requested enhancement of the downloaded file.

 Time optimal problems with Dirichlet boundary controls

Nadir Arada1 , Jean-Pierre Raymond2

Abstract
We consider time optimal control problems governed by semilinear parabolic equations with
Dirichlet boundary controls in the presence of a target state constraint. To establish opti-
mality conditions for the terminal time T , we define a new Hamiltonian functional. Due to
regularity results for the state and the adjoint state variables, this Hamiltonian belongs to
Lrloc (0, T ) for some r > 1. By proving that it satisfies a differential equation corresponding to
an optimality condition for T , we deduce that it belongs to W 1,1 (0, T ). This result answers
to the question: how to define Hamiltonian functionals for infinite dimensional problems
with variable endpoints (see [10], p. 282 and p. 595).
Key words. Time control problem, optimal control, Dirichlet controls, semilinear parabolic
equations, state constraints, parabolic equations with measures as data.

AMS Subject Classification. 49K20, 93C20, 93C50.

1 Introduction
Let Ω be a bounded connected subset of RN (N ≥ 2) with a regular boundary Γ. Consider a
system described by the equation

∂t y − ∆y + f (·, y) = 0 in Ω×]0, T [, y = u on Γ×]0, T [, y(0) = yo in Ω, (1.1)

where T > 0, u ∈ Uad ⊂ L∞ (Γ × R∗+ ), yo ∈ Cb (Ω). A target constraint is imposed on the state
variable at the terminal time T
g(·, T, y(·, T )) ∈ C, (1.2)
where C is a closed convex subset with a nonempty interior in Cb (Ω). We consider the control
problem

(P) inf{J(y, u, T ) | (y, u, T ) ∈ Cb (Ω×]0, T [) × Uad × R∗+ satisfies (1.1) and (1.2)}

where the cost functional J is defined on Cb (Ω × R+ ) × L∞ (Γ × R+ ) × R+ by

Z TZ Z TZ Z
J(y, u, T ) = F (·, y) dx dt + G(·, u) ds dt + L(·, T, y(·, T )) dx.
0 Ω 0 Γ Ω

Optimality conditions for u may be expressed in the form of a Pontryagin’s principle stated with
the Hamiltonian function
H(s, t, u, p, α) = α G(s, t, u) + p u (1.3)
1 Centro de Matemática e Aplicações, Instituto Superior Tecnico, Av. Rovisco Pais, 1049-001, Lisboa, Portugal.

E-mail: [email protected]
2 Laboratoire MIP, UMR 5640, Université Paul Sabatier, 31062 Toulouse Cedex 4, France. E-mail: ray-

[email protected]

1
where (s, t, u, p, α) ∈ Γ × R∗+ × R2 × R+ (see [2], [4]). When the terminal time T is a new
control variable, a new equation must be added to complete the first order optimality conditions.
For finite dimensional problems, optimality conditions for the optimal time T correspond to a
differential equation satisfied by the function

t −→ H(t, ȳ(t), ū(t), p̄(t), ᾱ),

where H is the Hamiltonian function of the corresponding control problem, (ȳ, ū) is an optimal
solution, ᾱ is the multiplier of the cost functional, and p̄ is the corresponding adjoint state.
For infinite dimensional problems we can still obtain a differential equation to characterize the
optimal time T , but the Hamiltonian function must be replaced by a Hamiltonian functional.
As noticed by Fattorini [10] (Remark 6.6.6, p. 282 and p. 595), even defining a Hamiltonian
functional is not obvious because of the lack of regularity of the state variable or of the adjoint
state. It is the reason why very few results have been obtained in this direction. Let us mention
that invariance principles for infinite dimensional control problems have been established by
Fattorini [7], [8]. Other results have been also obtained in the same direction by Barbu [5], [6], and
Yong [20]. Very recently, Raymond and Zidani have proved that some Hamiltonian functionals
for distributed controls [17], and Robin boundary controls [18], for parabolic equations, satisfy
differential equations characterizing optimal terminal time T . Our main purpose is to extend
these results for Dirichlet boundary controls. Such an extension is not at all obvious because
solutions to (1.1) are very irregular in the neighborhood of the boundary Γ×]0, T [. Since weak
solutions to equation (1.1) are defined by the transposition method, we introduce the Hamiltonian
functional:
Z Z
H(t, y, u, p, α) = α F (x, t, y(x)) dx + α G(s, t, u(s)) ds
Z  Ω Γ
 Z
− y(x) ∆p(x) + f (x, t, y(x)) p(x) dx + u(s) ∂ν p(s) ds (1.4)
Ω Γ
r0 1,r
for every (t, y, u, p, α) ∈ R∗+
× L (Ω) × L (Γ) × (W ∞ 2,r
(Ω) ∩ W0 (Ω)) × R+
for some r > 1 (∂ν p
denotes the normal derivative of p on the boundary Γ). We prove that, if (ȳ, ū, T̄ ) is a solution
of (P), then H satisfies the differential equation:
d
H(t, ȳ(t), ū(t), p̄(t), ᾱ) = Ht0 (t, ȳ(t), ū(t), p̄(t), ᾱ) in ]0, T̄ [, (1.5)
dt
Z D E
H(T̄ , ȳ(T̄ ), ū(T̄ ), p̄(T̄ ), ᾱ) = − ᾱ L0t (T̄ , ȳ(x, T̄ )) dx − µ̄, gt0 (T̄ , ȳ(T̄ )) ,
Ω ∗,Ω

where µ̄ is the multiplier associated with the terminal state constraint (1.2), ᾱ the multiplier
associated with the cost functional J, and p̄ is the adjoint state corresponding to (ᾱ, ȳ, ū),
Ht0 , G0t , and L0t denote the partial derivative of the corresponding functions with respect to t,
h·, ·i∗,Ω denotes the duality pairing between (Cb (Ω))0 and Cb (Ω). As a consequence, we obtain
an invariance principle for autonomous problems, which seems to be new for Dirichlet bound-
ary controls of semilinear parabolic equations. Equation (1.5) provides a regularity result for
the Hamiltonian functional H(·, ȳ(·), ū(·), p̄(·), ᾱ). Indeed, from the state equation it follows
that ȳ belongs to Cb (Ω×]0, T ]), and from the adjoint equation it follows that p̄ belongs to
Lrloc (0, T ; W 2,r (Ω)) for some r > 1. Therefore, from the state and the adjoint equations we
only deduce that H(·, ȳ(·), ū(·), p̄(·), ᾱ) belongs to Lrloc (0, T ). With equation (1.5) we prove that
H(·, ȳ(·), ū(·), p̄(·), ᾱ) belongs to W 1,1 (0, T ). To prove such a result we introduce a new control
variable, and we consider the time as a new state variable.
The plan of the paper is as follows. Assumptions and notation are stated in Section 2. Some

2
results concerning the state equation are recalled in Section 4. In Section 5, we transform the
problem (P) into a control problem (P) e with a fixed terminal time. Regularity results for the
linearized and adjoint equations are proved in Section 6. The Green formula needed to establish
optimality conditions for (P) e is proved in Section 7 (Proposition 7.3). Section 8 is devoted to
Taylor expansions corresponding to (P). e Since the adjoint state for a solution of (P)
e does not
r 2,r r 2,r
belong to L (0, 1; W (Ω)), but only to L (0, 1 − ε; W (Ω)) for all ε > 0, we first establish the
relationship between the Hamiltonian functional H and the adjoint state of the time variable on
the interval [0, 1 − ε] for some ε > 0 (see Section 9.1). But in this case we obtain multipliers αε ,
µε , and adjoint states φε and ψε , depending on ε. We pass to the limit when ε tends to zero in
Section 9.2. We come back to problem (P) in Section 10.
Comments on different notions of weak solutions. Since the control u in equation (1.1)
belongs to L∞ (Γ × (0, T )) and because we are interested in state constraints on y(T ) in Cb (Ω),
it is convenient to define solutions for equation (1.1) by the transposition method as in [2]. In
this way we prove the existence of a unique solution in Cb (QT \ ΣT ) (see Proposition 4.2). So-
lutions to equation (1.1) may be defined by using the semigroup approach as in [14] (see also
[15] for the case of linear equations). In that case, for L∞ -controls, the existence to equation
(1.1) is obtained in C([0, T ]; L2 (Ω)) (even in C([0, T ]; H 1/2−ε (Ω)) for all ε > 0). Therefore, for
problems with a fixed terminal time, optimality conditions may be obtained for state constraints
in L∞ (0, T ; L2 (Ω)), or with a suitable qualification condition in a smaller space, for example
in L∞ (Ω × (0, T )) [14]. To extend the results obtained in [14] to time optimal problems, it is
necessary to define the Hamiltonian functional H (here defined in (1.4)) by using the unbounded
operators A and B classically used to modelling of evolution equations in control theory. This
can surely be done. But this functional can only be defined for regular adjoint states p. Using
the adjoint equation stated in (3.3), and classical regularity results for parabolic equations we
are able to prove that the adjoint state belongs to L2 (0, T − ε; W 2,r (Ω)) for all ε > 0 and some
1 < r < ∞ (see Lemma 6.1). We also prove that the normal derivative of the adjoint state
belongs to L1 (Σ) (see Theorem 7.2). These results are crucial in the proof of necessary optimal-
ity conditions. Such results are not obtained in [14], where it is proved that the adjoint state
belongs to BV ([0, T ]; H −1 (Ω))∩L2 (0, T ; H01 (Ω))∩C([0, T ]; L2 (Ω)). However for state constraints
on y(T ) in L2 (Ω), the regularity results obtained in [14] for the adjoint state could be improved
enough to derive necessary optimality conditions for time optimal problems.
Finally mention that the issue of numerical approximations of the control of the heat equation
is discussed in [21].

2 Assumptions and notation

We suppose that the boundary Γ is of class C 2,ν̄ , for some 0 < ν̄ ≤ 1. Throughout the sequel
j denotes a positive number satisfying j > N2 + 1. For every 0 <  < t, we set Qt = Ω×]0, t[,
Σt = Γ×]0, t[, Qt = Ω ×], t[, where Ω = {x ∈ Ω | d(x, Γ) > }. For t ∈]0, +∞[ and r ∈]1, +∞[,
we set
Zt = {χ ∈ C 2 (Qt ) | χ|Σt = 0, χ(t) = 0},

Yr = {χ ∈ Lr (0, 1; W 2,r (Ω)) ∩ W 1,r (0, 1; Lr (Ω)) | χ|Σ1 = 0, χ(1) = 0}.

R µ ∈ Mb (E) (the space of bounded Radon measures on E),

If E is a locally compact subset of Ω,
and y ∈ Cb (E), we set hµ, yib,E = E y(x)dµ(x). We denote by int C, the interior of C for the
usual topology of Cb (Ω).
A1 - For every (t, y) ∈ R+ × R, f (·, t, y) is measurable on Ω. For almost every x ∈ Ω, f (x, ·) is

3
of class C 1 , and |f (x, t, 0)| ≤ f1 (x)η(t),
|ft0 (x, t, y)| ≤ f1 (x)η(t)η(|y|), 0 ≤ fy0 (x, t, y) ≤ f1 (x)η(t)η(|y|),
where f1 ∈ Lj (Ω), and η is a nondecreasing function from R+ into R+ .
A2 - For every (t, u) ∈ R+ × R, G(·, t, u) is measurable on Ω. For almost every s ∈ Γ, G(s, ·) is
of class C 1 and, with G1 ∈ L1 (Γ),
|G(s, t, u)| + |G0u (s, t, u)| + |G0t (s, t, u)| ≤ G1 (s) η(t) η(|u|),

A3 - For every (t, y) ∈ R+ × R, F (·, t, y) is measurable on Ω. For almost every x ∈ Ω, F (x, ·)

is of class C 1 on R+ × R, and
|F (x, t, y)| + |Ft0 (x, t, y)| ≤ F1 (x)η(t)η(|y|), |Fy0 (x, t, y)| ≤ F2 (x)η(t)η(|y|),
where F1 ∈ L1 (Ω), and F2 ∈ Lk (Ω) with k > 1.
A4 - For every (t, y) ∈ R+ × R, L(·, t, y) is measurable on Ω. For almost every x ∈ Ω, L(x, ·) is
of class C 1 and, with L1 ∈ L1 (Ω)
|L(x, t, y)| + |L0y (x, t, y)| + |L0t (x, t, y)| ≤ L1 (x)η(t)η(|y|),
A5 - g is a continuous function from Ω × R2 into R. For every x ∈ Ω, g(x, ·) is of class C 1 ,
gt0 and gy0 are continuous on Ω × R2 . For simplicity, we often write g(T, y) in place of g(x, T, y).
Recall that C is a closed convex subset with a nonempty interior in Cb (Ω).
A6 - The set of constraints on u is defined by
Uad = {u ∈ L∞ (Γ×]0, +∞[) | u(s, t) ∈ KU (s) for a.e. (s, t) ∈ Γ×]0, +∞[},
where KU is a measurable multimapping, with nonempty, closed and convex values in P(R). We
suppose that there exists ko ∈ L∞ (Γ) such that ko (s) ∈ KU (s) for almost every s ∈ Γ.

3 Statement of main results

As proved in [1], every ζ ∈ (Cb (Ω))0 may be identified with a measure ζb ∈ M(Ω × Ω# ), where
Ω# denotes the Stone-C̆ech compactification of Ω. (For notational simplicity, we identify ζb and
ζ.) We denote by Π the canonical projection from M(Ω × Ω# ) onto M(Ω) defined by:

Π : ζ ∈ M(Ω × Ω# ) −→ Πζ ∈ M(Ω),
D E D E
Πζ , ϕ = ζ, ϕ for all ϕ ∈ C(Ω).
M(Ω)×C(Ω) M(Ω×Ω# )×C(Ω×Ω# )

Throughout the sequel, for any ζ ∈ (Cb (Ω))0 , |ζ| stands for the total variation of ζ, and L∞
Π|ζ| (Ω)
denotes the space of Π|ζ| -essentially bounded, Π|ζ| -measurable functions on Ω.

Theorem 3.1 ([1], Corollary 4.8) Let ζ ∈ (Cb (Ω))0 . There exists a unique bounded linear trans-
formation Λζ : Cb (Ω) −→ L∞
Π|ζ| (Ω), such that:

D E D E Z
ζ, ϕ = Πζ , ϕ + Λζ (ϕ) dΠ|ζ| for all ϕ ∈ Cb (Ω),
∗,Ω b,Ω Γ
Z
Λζ (ϕ)dΠ|ζ| ≥ 0 for all 0 ≤ ϕ ∈ Cb (Ω),
Γ

4
 Z D E
Λζ (ϕ)dΠ|ζ| = Πζ , ϕ M(Γ)×C(Γ) e ∈ C(Ω).
for all ϕ
Γ

Now, we are in position to state the main result of the paper.

Theorem 3.2 Suppose that A1-A6 are fulfilled. If (ȳ, ū, T̄ ) is a solution of (P), then there
exist ᾱ ≥ 0, (p̄, q̄) ∈ L1 (0, T̄ ; W01,1 (Ω)) × W 1,1 (0, T̄ ), and µ̄ ∈ (Cb (Ω))0 such that the following
conditions hold:
• Nontriviality condition :
(ᾱ, µ̄) 6= 0, (3.1)
• Complementarity condition :
D E
µ̄, z − g(T̄ , ȳ(T̄ )) ≤0 for all z ∈ C, (3.2)
∗,Ω

• Adjoint equations :

 −∂t p̄ − ∆p̄ + fy0 (·, ȳ) p̄ = −ᾱ Fy0 (·, ȳ)


in QT̄ ,
(3.3)
p̄(T̄ ) = −ᾱ L0y (·, T̄ , ȳ(T̄ )) − [gy0 (T̄ , ȳ(T̄ ))∗ Πµ̄ ]|Ω in Ω,


 Z Z


 q̄ (t) = − (ᾱ Ft (·, t, ȳ) + ft (·, t, ȳ) p̄) dx − ᾱ G0t (·, t, v̄) ds
0 0 0
in ]0, T̄ [,
 Ω Γ
Z (3.4)
 D E
ᾱL0t (·, T̄ , ȳ(·, T̄ )) dx + µ̄, gt0 (T̄ , ȳ(T̄ ))

 q̄(T̄ ) =
 ,
Ω ∗,Ω

where [gy0 (T̄ , ȳ(T̄ ))∗ Πµ̄ ]|Ω is the bounded Radon measure on Ω defined by
D E
z −→ Πµ̄ , gy0 (T̄ , ȳ(T̄ )) z for all z ∈ Co (Ω),
b,Ω

• Optimality condition for ū :

Z   Z  
0
Hu ·, ū, ∂ν p̄, ᾱ (u − ū) ds dt+ Λµ̄ gy0 (T̄ , ȳ(T̄ ))(zu − zū )(T̄ ) d Π|µ̄| ≥ 0, (3.5)
ΣT̄ Γ

for all u ∈ Uad , where H is defined by (1.3), Λµ̄ is the operator in Theorem 3.1 corresponding to
µ̄, and zub (for u
b = u or u
b = ū) is the solution of

∂t z − ∆z = 0 in QT̄ , z=u
b on ΣT̄ , z(·, 0) = 0 in Ω, (3.6)

• Optimality condition for T̄ :

H(t, ȳ(t), ū(t), p̄(t), ᾱ) = −q̄(t) a.e. in ]0, T̄ [, (3.7)

where H is defined by (1.4).

4 State equation
We recall some existence and regularity results proved in [2].

5
Definition 4.1 A function y ∈ L1 (Q) is a weak solution of equation (1.1) if, and only if,
f (·, y(·)) ∈ L1 (QT ), and
Z   Z Z
y(−∂t χ − ∆χ) + f (·, y) χ dx dt = yo χ(0) dx − u ∂ν χ ds dt,
QT Ω ΣT

for every χ ∈ ZT = {z ∈ C 2 (QT ) | z|ΣT = 0 and z(T ) = 0}.

Proposition 4.2 ([2], Propositions 3.6 and 3.7) Let T ∈]0, +∞[, and let u ∈ L∞ (ΣT ). Equation
(1.1) admits a unique solution yuT ∈ Cb (QT \ ΣT ). This solution satisfies
 
kyuT k∞,QT ≤ C kuk∞,ΣT + kyo kC(Ω) + 1 ,

where C ≡ C(T, Ω). Moreover, for every  > 0, yuT is Hölder continuous on QT and satisfies,
with C() ≡ C(T, Ω, kuk∞,ΣT , kyo kC(Ω) , ),

kyuT kC ν,ν/2 (Q ) ≤ C() for some 0 < ν < 1,

T

5 Change of variable
To study optimality conditions for the time optimal problem (P), we first transform (P) into
a control problem with a fixed terminal time. To thisR end, we suppose that t is a function of
1
τ ∈ [0, 1]. Let w be in Wad := C([0, 1]; R∗+ ), satisfying 0 w(σ) dσ = T , and set
Z τ
t = λ(τ ) = w(σ) dσ.
0

The function w is introduced as a new control variable. Simple calculations show that if (y, u, T ) ∈
R1
R∗+ × Cb (QT \ ΣT ) × Uad is admissible for (P), then for every w ∈ Wad satisfying 0 w(τ ) dτ = T ,
the quadruplet (λ, z, v, w) defined by
Z τ
λ(τ ) = w(σ) dσ, z(x, τ ) = y(x, λ(τ )), v(s, τ ) = u(s, λ(τ )),
0

is admissible for the following control problem:

n
e z, v, w) | (λ, z, v, w) ∈ C([0, 1]) × Cb (Q1 \ Σ1 ) × Vad × Wad ,
inf J(λ,
(P)
e o
(λ, z, v, w) satisfies (5.2), (5.3), (5.4) ,

where
Z
J(λ,
e z, v, w) = w(τ )F (x, λ, z) dx dτ
Q1
Z Z (5.1)
+ w(τ ) G(s, λ, v) ds dτ + L(x, λ(1), z(1)) dx,
Σ1 Ω

 ∂τ z − w ∆z + w f (x, λ, z) = 0 in Q1 ,
(5.2)
z=wv on Σ1 , z(0) = yo in Ω,


6
 λ(0) = 0, λ0 (τ ) = w(τ ) in ]0, 1[, (5.3)

v ∈ Vad , g(λ(1), z(1)) ∈ C, (5.4)

Vad := {v ∈ L∞ (Σ1 ) | v(s, τ ) ∈ KU (s) for a.e. (s, τ ) ∈ Σ1 }.
To prove that y(·, λ(·)) is the solution of the previous partial differential equation, it is sufficient
to make a change of time variable in the weak formulation of (1.1). Notice that J(y, u, T ) =
e z, v, w). Moreover, (P)
J(λ, e is equivalent to (P) in the following sense.

Proposition 5.1 Let (λ, z, v, w) be in C 1 ([0, 1])×Cb (Q1 \Σ1 )×L∞ (Σ1 )×C([0, 1]). If (λ, z, v, w)
is admissible for (P),
e then the triplet defined by:

z(x, τ ) = y(x, λ(τ )), v(s, τ ) = u(s, λ(τ )), λ(1) = T,

is admissible for (P) and we have J(λ,

e z, v, w) = J(y, u, T ).

Proof. Let (λ, z, v, w) be an admissible quadruplet for (P).e Since w(τ ) > 0 on [0, 1], λ is an
increasing function. Therefore λ is a bijective mapping from [0, 1] onto [0, λ(1)]. We denote by
λ−1 the inverse mapping of λ. We set T = λ(1), y(·, t) = z(·, λ−1 (t)), u(·, t) = v(·, λ−1 (t)) if
t ≤ T and u(·, t) = ko (·) if t > T , where ko is given in Assumption A6. We can easily prove that
(y, u, T ) is an admissible triplet for (P) and that J(λ,
e z, v, w) = J(y, u, T ). 

Proposition 5.2 For every (v, w) ∈ Vad × Wad , equation (5.2) admits a unique solution zvw ∈
Cb (Q1 \ Σ1 ). This solution satisfies
 
kzvw kCb (Q1 \Σ1 ) ≤ C kwk∞,]0,1[ kvk∞,Σ1 + kyo kC(Ω) + 1 .

Moreover, for every  > 0, every M > 0, and every (v, w) ∈ Vad × Wad satisfying kvk∞,Σ1 +
kwkC([0,1]) ≤ M , we have

kzvw kC ν,ν/2 (Qε ) ≤ C() for some 0 < ν < 1,

1

where C() ≡ C(Ω, N, M, ).

Since problem (P)

e has the advantage of having a fixed terminal time, we firstly characterize
optimal solutions for (P),
e and next, with a change of variable, we deduce optimality conditions
for (P).

Theorem 5.3 If (λ̄, z̄, v̄, w̄) is a solution of (P), e then there exist ᾱ ≥ 0, (φ̄, ψ̄) ∈
L1 (0, 1; W01,1 (Ω)) × W 1,1 (0, 1), and µ̄ ∈ (Cb (Ω))0 such that the following conditions hold:
• Nontriviality and complementarity conditions :

ᾱ2 + kµ̄k2(Cb (Ω))0 = 1, (5.5)

D E
µ̄, z − g(λ̄(1), z̄(1)) ≤0 for all z ∈ C, (5.6)
∗,Ω

• Adjoint equations :

 −∂τ φ̄ − w̄ ∆φ̄ + w̄ fy0 (·, λ̄, z̄)φ̄ = −ᾱ Fy0 (·, λ̄, z̄)

in Q1 ,
(5.7)
φ̄(1) = −ᾱ L0y (·, λ̄(1), z̄(1)) − [gy0 (λ̄(1), z̄(1))∗ Πµ̄ ]|Ω in Ω,


7
  Z Z
0 0


 ψ̄ 0
(τ ) = − w̄ (ᾱ Ft (·, λ̄, z̄) + f t (·, λ̄, z̄) φ̄) dx − w̄ ᾱ G0t (·, λ̄, v̄) ds


 Ω Γ

in ]0, 1[, (5.8)


 Z
 D E
 ψ̄(1) = ᾱ L0t (·, λ̄(1), z̄(1)) dx + µ̄, gt0 (λ̄(1), z̄(1))

 ,
Ω ∗,Ω

• Optimality condition for v̄ :

Z  
w̄ Hu0 s, λ̄, v̄, ∂ν φ̄, ᾱ (v − v̄) ds dτ
Σ1
Z   (5.9)
+ Λµ̄ gy0 (λ̄(1), z̄(1)) (ξw̄,v − ξw̄,v̄ )(1) d Π|µ̄| ≥ 0 for all v ∈ Vad ,
Γ

where Λµ̄ is the operator in Theorem 3.1 corresponding to µ̄, and ξw̄,bv (for vb = v or vb = v̄) is
the solution of
∂τ ξ − w̄ ∆ξ = 0 in Q1 , ξ = w̄ vb on Σ1 , ξ(0) = 0 in Ω, (5.10)
• Optimality condition for w̄ :
H(λ̄(τ ), z̄(τ ), v̄(τ ), φ̄(τ ), ᾱ) = −ψ(τ ) a.e. in ]0, 1[. (5.11)

6 Some regularity results

Now, we establish regularity results needed in the other sections of the paper.
Lemma 6.1 ([13, Chapter 4, Theorem 9.1]) Let a be a nonnegative function in Ld (Q1 ) (d >
N r
2 + 1), and b be in L (Q1 ) (1 < r < ∞). Let β be in [0, 1], w, wo be in Wad , and set
wβ = w + β(wo − w). The equation
−∂τ χ − wβ ∆χ + wβ aχ = b in Q1 , χ = 0 on Σ1 , χ(1) = 0 in Ω, (6.1)
admits a unique weak solution in Lr (0, 1; W 2,r (Ω)) ∩ W 1,r (0, 1; Lr (Ω)). This solution satisfies
kχkLr (0,1;W 2,r (Ω)) + kχkW 1,r (0,1;Lr (Ω)) ≤ C(r) kbkr,Q1 ,
where C(r) ≡ C(Ω, r, d, M, w, wo ) is independent of β and a, and M is an upper bound for
kakd,Q1 + kwβ k∞,]0,1[ .
Proposition 6.2 Let a be a nonnegative function in Ld (Q1 ), ` in Ld (Q1 ) (d > N2 + 1), Φ in
Lr (Q1 ) (1 < r < ∞), Ψ in L∞ (Σ1 ). Let β be in [0, 1]. Let w, wo be in Wad , w
e in C([0, 1]), and
set wβ = w + β(wo − w). The equation
Z  
ζ − ∂τ χ − wβ ∆χ + wβ aχ dx dτ (6.2)
Q1
Z Z Z
= w
e Φ ∆χ dx dτ + ` χ dx dτ − w
e Ψ ∂ν χ ds dτ for all χ ∈ Yr0 .
Q1 Q1 Σ1

admits a unique solution ζ ∈ Lr (Q1 ). This solution satisfies

 
kζkr,Q1 ≤ C kwk e ∞,]0,1[ (kΦkr,Q1 + kΨkr,Σ1 ) + k`kd,Q1 , (6.3)

where C ≡ C(Ω, N, r, d, M, w, wo ) is independent of β and a, and M is an upper bound for

kakd,Q1 + kwβ k∞,]0,1[ .

8
 Proof. Observe that ζ = ζ1 + ζ2 , where ζ1 is the solution of

∂τ ζ1 − wβ ∆ζ1 + wβ aζ1 = ` in Q1 , ζ1 = 0 on Σ1 , ζ1 (0) = 0 in Ω,

and ζ2 is the solution of, for all χ ∈ Yr0 ,

Z   Z Z
ζ2 − ∂τ χ − wβ ∆χ + wβ aχ dx dτ = e Φ ∆χ dx dτ −
w w
e Ψ ∂ν χ ds dτ (6.4)
Q1 Q1 Σ1

where Yr0 is defined at the beginning of section 2.

• Due to Proposition 3.6 in [2], ζ1 belongs to Co (Q1 ∪ Ω1 ), and satisfies

kζ1 k∞,Q1 ≤ C k`kd,Q1 . (6.5)

• From Lemma 6.1, it follows that the linear operator Λ : χ −→ −∂τ χ − wβ ∆χ + wβ aχ is

0
continuous, and bijective from Yr0 onto Lr (Q1 ). Equation (6.4) can be rewritten as
Z   Z
ζ2 − ∂τ χ − wβ ∆χ + wβ aχ dx dτ = ζ2 Λ(χ) dx dτ
Q1 Q1
D E D E
= Λ∗ (ζ2 ), χ = K, χ for all χ ∈ Yr0 ,
(Yr0 )0 ×Yr0 (Yr0 )0 ×Yr0

where K ∈ (Yr0 )0 is defined by:

D E Z Z
K, χ = e Φ ∆χ dx dτ −
w w
e Ψ ∂ν χ ds dτ.
(Yr0 )0 ×Yr0 Q1 Σ1

It follows that Λ∗ (ζ2 ) = K in (Yr0 )0 , and thus ζ2 = (Λ∗ )−1 (K) ∈ Lr (Q1 ). Since

k(Λ∗ )−1 kL((Yr0 )0 ;Lr (Q1 )) = kΛ−1 kL(Lr0 (Q1 );Yr0 ) ≤ C(r),

we deduce that
kζ2 kr,Q1 = k(Λ∗ )−1 (K)kr,Q1 ≤ C(r) kKk(Yr0 )0
  (6.6)
≤ Ckwk e ∞,]0,1[ kΦkr,Q1 + kΨkr,Σ1 .

The estimate (6.3) follows from (6.5) and (6.6). Observe that the constant C does not depend
neither on β nor on a. It depends on w, wo , and an upper bound for kakd,Q1 + kwβ k∞,]0,1[ . 

Rmeark 6.3 Since the space Z1 = {z ∈ C 2 (Q1 ) | z|Σ1 = 0, z(1) = 0} is dense in Yr0 , for all
1 < r < +∞, the variational equation (6.2) is equivalent to
Z  
ζ − ∂τ χ − wβ ∆χ + wβ aχ dx dτ (6.7)
Q1
Z Z Z
= w
e Φ ∆χ dx dτ + ` χ dx dτ − w
e Ψ ∂ν χ ds dτ for all χ ∈ Z1 .
Q1 Q1 Σ1

Proposition 6.4 Let ε be in ]0, 1[. Let a be a nonnegative function in Ld (Q1 ), ` in Ld (Q1 )
(d > N2 + 1), Φ in Lr (Q1 ) (1 < r < ∞), Ψ in L∞ (Σ1 ). Let β be in [0, 1], w, wo be in Wad ,

9
and set wβ = w + β(wo − w). Let ζ be the solution of (6.2). If w e ∈ C([0, 1]) satisfies we ≡ 0 on
[1 − ε, 1], then for every 0 < ε1 < ε, ζ|Ω×]1−ε1 ,1] belongs to Co (Ω×]1 − ε1 , 1]) and satisfies
 
kζkCo (Ω×]1−ε1 ,1]) ≤ C(ε, ε1 ) kwk
e ∞,]0,1[ (kΨk∞,Σ1 + kΦkr,Q1 ) + k`kd,Q1 , (6.8)

where C(ε, ε1 ) ≡ C(Ω, N, r, d, M, w, wo , ε1 , ε) is independent of β and a, and M is an upper

bound for kakd,Q1 + kwβ k∞,]0,1[ .

Proof. Observe that ζ = ζ1 + ζ2 , where ζ1 is the solution of

∂τ ζ1 − wβ ∆ζ1 + wβ aζ1 = ` in Q1 , ζ1 = w
eΨ on Σ1 , ζ1 (0) = 0 in Ω,

and ζ2 is the solution of

Z   Z
ζ2 − ∂τ χ − wβ ∆χ + wβ aχ dx dτ = w
e Φ ∆χ dx dτ for all χ ∈ Z1 .
Q1 Q1

• Due Proposition 3.6 in [2], ζ1 ∈ Cb (Q1 \ Σ1 ), satisfies ζ1 |Γ×[1−ε1 ,1] = 0, and

 
kζ1 kCb (Ω×]1−ε1 ,1]) ≤ kζ1 kCb (Q1 \Σ1 ) ≤ C kwk
e ∞,]0,1[ kΨk∞,Σ1 + kf kd,Q1 . (6.9)

• Let 0 < ε1 < ε2 < ε3 < ε4 < ε, and let θ : [0, 1] −→ [0, 1] be a regular function such that:
(
0 if τ ∈ [0, 1 − ε4 ],
θ(τ ) =
1 if τ ∈ [1 − ε3 , 1].

We can easily prove that ζe = θ ζ2 is the solution of

∂τ ζe − wβ ∆ζe + wβ aζe = −θ0 ζ2 in Q1 , ζe = 0 on Σ1 , ζ(0)

e = 0 in Ω.

Due to Proposition 6.2, ζ2 belongs to Lr (Q1 ). Thus ζe belongs to Lr̃ (Q1 ), for every r̃ <
N +2 r̃
max(0,N +2−2r) . Therefore ζ|Ω×]1−ε3 ,1[ = ζ2 |Ω×]1−ε3 ,1[ ∈ L (Ω×]1 − ε3 , 1[). By using an itera-
e
tive process, we can prove that
N
ζ2 |Ω×]1−ε2 ,1[ ∈ Ld (Ω×]1 − ε2 , 1[) with de > + 1, (6.10)
e
2
and
kζ2 kd,Ω×]1−ε
e 2 ,1[
≤ C kw
e Φkr,Q1 , (6.11)

where C ≡ C(Ω, N, r, d, d,
e ε2 , ε). Now, let θo : [0, 1] −→ [0, 1] be a regular function such that:
(
0 if τ ∈ [0, 1 − ε2 ],
θo (τ ) =
1 if τ ∈ [1 − ε1 , 1].

The function ζeo = θo ζ2 is the solution of

∂τ ζeo − wβ ∆ζeo + wβ aζeo = −θo0 ζ2 in Q1 , ζeo = 0 on Σ1 , ζeo (0) = 0 in Ω.

Due to (6.10) and (6.11), with regularity results for linear parabolic equations, we deduce that
ζeo belongs to L∞ (Q1 ), and

10
 kζeo k∞,Q1 ≤ C kθo0 ζ2 kd,Q
e 1 ≤ C kw
e Φkr,Q1 ≤ C kwk
e ∞,]0,1[ kΦkr,Q1 .

Since ζeo |Ω×]1−ε1 ,1] ≡ ζ2 |Ω×]1−ε1 ,1] , it follows that:

kζ2 kCo (Ω×]1−ε1 ,1]) ≤ C kwk

e ∞,]0,1[ kΦkr,Q1 . (6.12)

The estimate (6.8) immediately follows from (6.9) and (6.12). 

7 Adjoint equation
The adjoint equation for (Pe) is of the form

−∂τ φ − w ∆φ + w a φ = µ1 in Q1 , φ=0 on Σ1 , φ(1) = µ2 in Ω, (7.1)

where w ∈ Wad , a is a nonnegative function in Lj (Q1 ), µ1 ∈ Lk (Q1 ) (k > 1), and µ2 is a bounded
Radon measure in Ω.

Definition 7.1 A function φ ∈ L1 (0, T ; W01,1 (Ω)) is a weak solution of (7.1) if, and only if,
aφ ∈ L1 (Q1 ), and
Z   Z D E
φ ∂τ χ + w ∇χ ∇φ + w aχφ dx dτ = µ1 χ dx dτ + µ2 , χ(1)
Q1 Q1 b,Ω

for all χ ∈ C 1 (Q1 ) ∩ Co (Q1 ∪ Ω1 ).

Theorem 7.2 ([16, Theorem 4.1], [2, Theorem 4.2]) Let w be in Wad , a be a nonegative function
in Lj (Q1 ), µ1 in Lk (Q1 ) (k > 1), and µ2 in Mb (Ω). Equation (7.1) admits a unique weak solution
φ in L1 (0, T ; W01,1 (Ω)) satisfying
 
kφkLδ (0,1;W 1,d (Ω)) ≤ C(δ, d) kµ1 k1,Q1 + kµ2 kMb (Ω) , (7.2)
0

N
for every (δ, d) satisfying δ ≥ 1, d ≥ 1, 2d + 1δ > N2 + 12 . Moreover, there exist a function in
1 1
L (Σ), denoted by ∂ν φ, and a function in L (Ω), denoted by φ(0), such that:
Z  
∂τ χ − w ∆χ + w aχ φ dx dτ
Q1
Z D E Z Z
= µ1 χ dx dτ + µ2 , χ(1) + w χ ∂ν φ ds dτ − χ(0) φ(0) dx
Q1 b,Ω Σ1 Ω

for all χ ∈ X∞ = {y ∈ Cb (Q1 ∪ Ω1 ) | ∂τ y − w ∆y ∈ L∞ (Q1 ), y|Σ1 ∈ L∞ (Σ1 ), y(0) ∈ L∞ (Ω)}.

(X∞ is defined as the set of solutions of a Dirichlet boundary value problem when solutions are
defined by transposition [2].)

The Green formula stated in the proposition below, is crucial to prove optimality conditions
in Section 9.1.

Proposition 7.3 Let (w, a, µ1 , µ2 ) be as in the statement of Theorem 7.2. Let w e ∈ C([0, 1])
0
e ≡ 0 in [1 − ε, 1], Φ be in Lk (Q1 ) (k > 1), ` be in Lj (Q1 ), and Ψ in L∞ (Σ1 ). If φ is the
satisfy w

11
solution of (7.1) corresponding to (w, a, µ1 , µ2 ), and if ζ is the solution of (6.2) corresponding to
(wβ = w, w,
e a, Φ, `, Ψ), then we have:
Z D E
µ1 ζ dx dτ + µ2 , ζ(1)
Q1 b,Ω
Z Z Z (7.3)
= w
e Φ ∆φ dx dτ + ` φ dx dτ − w
e Ψ ∂ν φ ds dτ.
Q1 Q1 Σ1

0
Proof. First, notice that due to Proposition 6.2, ζ belongs to Lk (Q1 ). Let θ be in C 1 ([0, 1]) such
that 0 ≤ θ ≤ 1 and

 0 if t ∈ [0, 1 − ε],
θ(t) =
1 if t ∈ [1 − 21 ε, 1],


and set φ1 = (1 − θ) φ and φ2 = θ φ. The function φ1 is the solution of

 0
 −∂τ φ1 − w ∆φ1 + w a φ1 = θ φ + (1 − θ)µ1 in Q1 ,

φ1 = 0 on Σ1 , φ1 (1) = 0 in Ω.


From regularity results for parabolic equations with measures as data, it follows that φ ∈ Lσ (Q1 ),
for all 1 < σ < NN+2 . Hence, from Lemma 6.1 we deduce that φ1 belongs to Yσ∧k , for all
1 < σ < NN+2 . By setting χ = φ1 in (6.2), since wφ
e 1 ≡ wφ,
e it follows that,

Z  
ζ − ∂τ φ1 − w ∆φ1 + w a φ1 dx dτ
Q1
Z Z Z (7.4)
= w
e Φ ∆φ dx dτ + ` φ1 dx dτ − w
e Ψ ∂ν φ ds dτ.
Q1 Q1 Σ1

0
Now, let (Φn )n be a sequence of regular functions converging to Φ in Lk (Q1 ). Let ζn be the
solution of (6.2) corresponding to (w, w,
e a, Φn , `, Ψ). With Propositions 6.2 and 6.4, we can prove
that the function ζn − ζ satisfies

k(ζn − ζ)(1)kCo (Ω) + kζn − ζkk0 ,Q1 ≤ C kΦn − Φkk0 ,Q1 .

0
Therefore, the sequence (ζn (1), ζn )n converges to (ζ(1), ζ) in Co (Ω) × Lk (Q1 ). Moreover ζn is
the solution of

 ∂τ ζn − w ∆ζn + w a ζn = w e ∆Φn + ` in Q1 ,

ζn = w
eΨ on Σ1 , ζn (0) = 0 in Ω.


Using the Green formula of Theorem 7.2, we deduce that:

Z   D E
ζn − ∂τ φ2 − w ∆φ2 + w a φ2 dx dτ + φ2 (1), ζn (1)
Q1 b,Ω
Z   Z
= φ2 − ∂τ ζn − w∆ζn + w a ζn dx dτ − w ζn ∂ν φ2 ds dτ
Q1 Σ1

12
 Z Z Z
= w
e ∆Φn φ2 dx dτ + ` φ2 dx dτ − ww
e Ψ ∂ν φ2 ds dτ
Q1 Q1 Σ1
Z
= ` φ2 dx dτ,
Q1

because wθ
e = 0. By passing to the limit when n tends to infinity, we obtain
Z   D E Z
ζ − ∂τ φ2 − w ∆φ2 + w a φ2 dx dτ + φ(1), ζ(1) = ` φ2 dx dτ. (7.5)
Q1 b,Ω Q1

From (7.4) and (7.5), it follows that φ = φ1 + φ2 satisfies

Z   D E
ζ − ∂τ φ − w ∆φ + w a φ dx dτ + φ(1), ζ(1)
Q1 b,Ω
Z D E
= µ1 ζ dx dτ + µ2 , ζ(1)
Q1 b,Ω
Z Z Z
= w
e Φ ∆φ dx dτ + ` φ dx dτ − w
e Ψ ∂ν φ ds dτ.
Q1 Q1 Σ1

The proof is complete. 

8 Taylor expansions
To obtain optimality conditions we need some differential calculus rules, stated below. Let (v, w)
be in Vad ×Wad , and (b b be in L∞ (Σ1 )×C([0, 1]). As previously, we denote by zvw the solution
v , w)
of (5.2) corresponding to (v, w). For any w b ∈ C([0, 1]), the solution of (5.3) corresponding to w
b
will be denoted by λŵ . In all the sequel, hvw,ŵ stands for the solution of
Z  
h − ∂τ χ − w ∆χ + w fy0 (·, λw , zvw )χ dx dτ
Q1
Z Z
= b zvw ∆χ dx dτ −
w w
b v ∂ν χ ds dτ (8.1)
Q1 Σ1
Z  
− w λwb ft0 (·, λw , zvw ) + wf
b (·, λw , zvw ) χ dx dτ,
Q1

for all χ ∈ Z1 , and ξvw,v̂ stands for the solution of

∂τ ξ − w ∆ξ + w fy0 (·, λw , zvw ) ξ = 0 in Q1 ,

(8.2)
ξ = w vb on Σ1 , ξ(0) = 0 in Ω.

Theorem 8.1 Let ε and ρ be in ]0, 1[. Let v, vo be in Vad and set vρ = v + ρ(vo − v). Let w,
wo ∈ Wad satisfying wo ≡ w on [1 − ε, 1], and set wρ = w + ρ(wo − w). Then, the following
conditions hold.

• Taylor expansions for the state variable :

zvρ wρ (1) − zvw (1)

lim − (hvw,wo −w + ξvw,vo −v )(1) = 0,
ρ&0 ρ Co (Ω)

13
 zvρ wρ − zvw
lim − (hvw,wo −w + ξvw,vo −v ) =0 ∀r ∈]1, +∞[,
ρ&0 ρ Lr (Q1 )

where hvw,wo −w ∈ Lr (Q1 ) (for all 1 < r < ∞) is the solution of (8.1) corresponding to w
b =
wo − w, and ξvw,vo −v is the solution of (8.2) corresponding to vb = vo − v.

• Taylor expansion for the cost functional :

e w , zv w , vρ , wρ ) − J(λ
J(λ e w , zvw , v, w)
ρ ρ ρ

= ρ Je0 (λw , zvw , v, w) (λwo −w , hvw,wo −w + ξvw,vo −v , vo − v, wo − w) + o(ρ),

where
Je0 (λw , zvw , v, w) (λwo −w , hvw,wo −w + ξvw,vo −v , vo − v, wo − w)
Z
= L0y (·, λw (1), zvw (1))(hvw,wo −w + ξvw,vo −v )(1) dx
Ω
Z
+ w Fy0 (·, λw , zvw )(hvw,wo −w + ξvw,vo −v ) dx dτ
Q1
Z
+ w G0u (·, λw , v)(vo − v) ds dτ
Σ1
Z  
+ (wo − w) F (·, λw , zvw ) + w Ft0 (·, λw , zvw ) λwo −w dx dτ
Q1
Z  
+ (wo − w) G(·, λw , v) + w G0t (·, λw , v) λwo −w ds dτ
Σ1
Z
+ L0t (·, λw (1), zvw (1)) λwo −w (1) dx.
Ω

Proof. First, let us observe that hvw,wo −w is the solution of (6.2) corresponding to wβ = w (β =
e = wo − w, a = fy0 (·, λw , zvw ), Φ = zvw , ` = wλw−wo ft0 (·, λw , zvw ) + (w − wo )f (·, λw , zvw ),
0), w
Ψ = v. Since zvw belongs to L∞ (Q1 ), from Proposition 6.2, it follows that

hvw,wo −w ∈ Lr (Q1 ) for all r ∈]1, +∞[. (8.3)

zv w −zvw
The function ζρ = ρ ρρ − (hvw,wo −w + ξvw,vo −v ) is the solution of
Z  
ζρ − ∂τ χ − wρ ∆χ + wρ aρ χ dx dτ
Q1
Z Z
= fρ χ dx dτ + ρ (wo − w) (hvw,wo −w + ξvw,vo −v ) ∆χ dx dτ
Q1 Q1
Z
−ρ (wo − w)(vo − v) ∂ν χ ds dτ for all χ ∈ Z1 ,
Σ1

with Z 1
aρ = fy0 (λwρ , θzvρ wρ + (1 − θ)zvw ) dθ,
0

14
 fρ = (hvw,wo −w + ξvw,vo −v ) (w fy0 (·, λw , zvw ) − wρ aρ )
 Z 1 
+λw−wo − wρ ft0 (·, θλwρ + (1 − θ)λw , zvw ) dθ + w ft0 (·, λw , zvw ) .
0

Taking (8.3) and Assumption A1 into account, we can find d > N2 + 1 such that fρ belongs to
Ld (Q1 ). The function ζρ is the solution of (6.2) corresponding to wβ = wρ , w
e = ρ (w − wo ),
a = aρ , Φ = hvw,wo −w + ξvw,vo −v , ` = fρ , and Ψ = vo − v. Since w ≡ wo on [1 − ε, 1], from
Propositions 6.4 and 6.2, it follows that, for all r ∈]1, +∞[, we have

kζρ (1)kCo (Ω) + kζρ kr,Q1

 
≤ C ρ kw − wo k∞,]0,1[ (khvw,wo −w + ξvw,vo −v kr,Q1 + kv − vo k∞,Σ1 ) + kfρ kd,Q1

→0 when ρ → 0.

We have proved the Taylor expansion for the state variable. The Taylor expansion for the cost
functional may be proved with A2-A4, and with the Taylor expansion for the state variable. 

9 Proof of necessary optimality conditions for (P)

e
1
Let (λ̄, z̄, v̄, w̄) be a solution of problem (P).
e Let m̄ =
2 minτ ∈[0,1] w̄(τ ), M̄ = m̄ + kw̄kC([0,1]) ,
and 0 < ε < 1 be fixed. Set

Wad (ε) = {w ∈ Wad | m̄ ≤ w ≤ M̄ and w ≡ w̄ on [1 − ε, 1]},

and consider the following control problem

n
e z, v, w) | (λ, z, v, w) ∈ C([0, 1]) × Cb (Q1 \ Σ1 ) × Vad × Wad (ε),
inf J(λ,
(P
eε ) o
(λ, z, v, w) satisfies (5.2), (5.3), (5.4) .

It is clear that (λ̄, z̄, v̄, w̄) is an optimal solution for (P

eε ).

9.1 Optimality conditions for (P

eε )
Theorem 9.1 Let (λ̄, z̄, v̄, w̄) be a solution of (P).
e Then, for every ε ∈]0, 1[, there exist αε ≥ 0,
1,1
(φε , ψε ) ∈ L (0, 1; W0 (Ω)) × W (0, 1), and µε ∈ (Cb (Ω))0 such that the following conditions
1 1,1

hold:
(αε )2 + kµε k2(Cb (Ω))0 = 1, (9.1)

D E
µε , z − g(λ̄(1), z̄(1)) ≤0 for all z ∈ C, (9.2)
∗,Ω

 −∂τ φε − w̄ ∆φε + w̄ fy0 (·, λ̄, z̄)φε = −αε Fy0 (·, λ̄, z̄)

in Q1 ,
(9.3)
φε (1) = −αε L0y (·, λ̄(1), z̄(1)) − [gy0 (λ̄(1), z̄(1))∗ Πµε ]|Ω in Ω,


15
  Z Z
0 0


 ψ 0
ε (τ ) = − w̄ (α F
ε t (·, λ̄, z̄) + f t (·, λ̄, z̄) φ ε ) dx − w̄ αε G0t (·, λ̄, v̄) ds


 Ω Γ

in ]0, 1[, (9.4)


 Z
 D E
 ψε (1) = αε L0t (·, λ̄(1), z̄(1)) dx + µε , gt0 (λ̄(1), z̄(1))

 ,
Ω ∗,Ω
Z  
w̄ Hu0 s, λ̄, v̄, ∂ν φε , αε (v − v̄) ds dτ
Σ1
Z   (9.5)
+ Λµε gy0 (λ̄(1), z̄(1)) ξv̄w̄,v−v̄ (1) d Π|µε | ≥ 0 for all v ∈ Vad ,
Γ
where Λµε is the operator in Theorem 3.1 corresponding to µε , and ξv̄w̄,v−v̄ is the solution of
(8.2) corresponding to v − v̄,

H(λ̄(τ ), z̄(τ ), v̄(τ ), φε (τ ), αε ) = −ψε (τ ) a.e. in ]0, 1 − ε[. (9.6)

Proof. Let us consider the following sets:

n
Aε = (X, γ) ∈ Cb (Ω) × R | there exists (v, w) ∈ Vad × Wad (ε) s.t.
 
X = g(λ̄(1), z̄(1)) + g 0 (λ̄(1), z̄(1)) λw−w̄ (1), (hv̄w̄,w−w̄ + ξv̄w̄,v−v̄ )(1)
 o
γ ≥ Je0 (λ̄, z̄, v̄, w̄) λw−w̄ , hv̄w̄,w−w̄ + ξv̄w̄,v−v̄ , v − v̄, w − w̄ ,

B = int C×] − ∞, 0[,

where λ̄ ≡ λw̄ , hv̄w̄,ŵ is the solution of (8.1) corresponding to (v̄, w̄, w), b and ξv̄w̄,v̂ is the solution
of (8.2) corresponding to (v̄, w̄, vb). The sets Aε and B are convex, and B is open. Let us prove
that Aε ∩ B = ∅. Argue by contradiction and suppose that there exists (vo , wo ) ∈ Vad × Wad (ε),
γo ∈ R, such that:
 
g(λ̄(1), z̄(1)) + g 0 (λ̄(1), z̄(1)) λwo −w̄ (1), (hv̄w̄,wo −w̄ + ξv̄w̄,vo −v̄ )(1) ∈ int C, (9.7)
 
Je0 (λ̄, z̄, v̄, w̄) λwo −w̄ , hv̄w̄,wo −w̄ + ξv̄w̄,vo −v̄ , vo − v̄, wo − w̄ ≤ γo < 0. (9.8)

Let vρ = v̄ + ρ(vo − v̄), wρ = w̄ + ρ(wo − w̄), λρ = λ̄ + ρ(λwo − λ̄), and let zρ be the solution of
(5.2) corresponding to (vρ , wρ ). Set

g(λρ (1), zρ (1)) − g(λ̄(1), z̄(1))

Xρ = g(λ̄(1), z̄(1)) + .
ρ
From (9.7), (9.8), and Theorem 8.1, we deduce that:
e ρ , zρ , vρ , wρ ) − J(
J(λ e λ̄, z̄, v̄, w̄)
lim Xρ ∈ int C and lim < 0.
ρ&0 ρ&0 ρ
Therefore, there exists ρo > 0 such that, for every 0 < ρ ≤ ρo < 1, we have

g(λρ (1), zρ (1)) = ρ Xρ + (1 − ρ) g(λ̄(1), z̄(1)) ∈ int C,

and

16
 J(λ e λ̄, z̄, v̄, w̄) = inf(P
e ρ , zρ , vρ , wρ ) < J( eε ).

Since (λρ , zρ , vρ , wρ ) is admissible for (P

eε ), there is contradiction. Thus, Aε ∩ B = ∅. From a
geometric version of the Hahn-Banach theorem, there exists (αε , µε ) ∈ R × (Cb (Ω))0 , such that:

D E D E
αε γ
b + µε , X
b > αε γ + µε , X (9.9)
∗,Ω ∗,Ω

for all (X,

b γb) ∈ Aε , and all (X, γ) ∈ B.

• Nontriviality and complementarity conditions. From (9.9), we can easily check that αε is
nonnegative and that (αε , µε ) 6= 0. Then, possibly after a normalization procedure, we can
suppose that (9.1) is satisfied. Moreover, we have
D E D E
αε γ
b + µε , X b ≥ αε γ + µε , X (9.10)
∗,Ω ∗,Ω

for all (X,

b γb) ∈ Aε and all (X, γ) ∈ B = C×]−∞, 0]. We obtain (9.2) by setting X
b = g(λ̄(1), z̄(1)),
X = z ∈ C, and γ b = γ = 0 in (9.10).
• Optimality condition for v̄. Let v ∈ Vad . If we set
b = g(λ̄(1), z̄(1)) + gy0 (λ̄(1), z̄(1)) ξv̄w̄,v−v̄ (1),
X

b = Jez0 (λ̄, z̄, v̄, w̄) ξv̄w̄,v−v̄ + Jev0 (λ̄, z̄, v̄, w̄) (v − v̄),
γ

X = g(λ̄(1), z̄(1)), γ = 0,

in (9.10), we obtain

αε Jez0 (λ̄, z̄, v̄, w̄) ξv̄w̄,v−v̄ + αε Jev0 (λ̄, z̄, v̄, w̄) (v − v̄)
D E (9.11)
+ µε , gy0 (λ̄(1), z̄(1))ξv̄w̄,v−v̄ (1) ≥ 0.
∗,Ω

Let Λµε : Cb (Ω) −→ L∞Π|µ | (Ω) be the operator defined in Theorem 3.1 corresponding to µε .
ε
We have
D E D E Z
µε , ϕ = Πµε , ϕ + Λµε (ϕ) d Π|µε | for all ϕ ∈ Cb (Ω). (9.12)
∗,Ω b,Ω Γ

Therefore, if φε is the solution of (5.7) corresponding to (αε , µε ), with the Greeen formula of
Theorem 7.2, we obtain
Z Z
αε Fy0 (·, λ̄, z̄) ξv̄w̄,v−v̄ dx dτ + αε L0y (·, λ̄(1), z̄(1)) ξv̄w̄,v−v̄ (1) dx
Q1 Ω
D E
+ µε , gy0 (λ̄(1), z̄(1)) ξv̄w̄,v−v̄ (1) (9.13)
∗,Ω
 
0
R R
= Σ1
w̄ ∂ν φε (v − v̄) ds dτ + Γ
Λ µε gy ( λ̄(1), z̄(1)) ξv̄ w̄,v−v̄ (1) d Π|µε | .

From (9.11) and (9.13), we deduce that (9.5) is satisfied.

• Optimality condition for w̄. Let w ∈ Wad (ε). If we set

17
 b = g(λ̄(1), z̄(1)) + g 0 (λ̄(1), z̄(1)) hv̄w̄,w−w̄ (1) + g 0 (λ̄(1), z̄(1)) λw−w̄ (1),
X y t

b = Jeλ0 (λ̄, z̄, v̄, w̄) λw−w̄ + Jez0 (λ̄, z̄, v̄, w̄) hv̄w̄,w−w̄ + Jew0 (λ̄, z̄, v̄, w̄) (w − w̄),
γ

X = g(λ̄(1), z̄(1)), γ = 0,

in (9.10), we obtain
 
αε Jeλ0 (λ̄, z̄, v̄, w̄) λw−w̄ + Jez0 (λ̄, z̄, v̄, w̄) hv̄w̄,w−w̄ + Jew0 (λ̄, z̄, v̄, w̄) (w − w̄)
D E (9.14)
+ µε , gy0 (λ̄(1), z̄(1)) hv̄w̄,w−w̄ (1) + gt0 (λ̄(1), z̄(1)) λw−w̄ (1) ≥ 0.
∗,Ω

Since w ≡ w̄ on [1 − ε, 1], it follows that hv̄w̄,w−w̄ (1) belongs to Co (Ω), and from (9.12) and the
Green formula (7.3), we obtain
Z Z
αε Fy0 (·, λ̄, z̄) hv̄w̄,w−w̄ dx dτ + αε L0y (·, λ̄(1), z̄(1)) hv̄w̄,w−w̄ (1) dx
Q1 Ω
D E
+ µε , gy0 (λ̄(1), z̄(1)) hv̄w̄,w−w̄ (1)
∗,Ω
Z Z (9.15)
=− (w − w̄) (z̄ ∆φε + f (·, λ̄, z̄) φε ) dx dτ + (w − w̄) v̄ ∂ν φε ds dτ
Q1 Σ1
Z
+ w̄ λw−w̄ ft0 (·, λ̄, z̄) φε dx dτ.
Q1

Taking (9.14) and (9.15) into account, it follows that:

Z 1 Z
(w − w̄) H(λ̄, z̄, v̄, φε , αε ) dτ + w̄ λw−w̄ ft0 (·, λ̄, z̄) φε dx dτ
0 Q1
Z Z
+αε w̄ G0t (·, λ̄, v̄) λw−w̄ ds dτ + αε w̄ Ft0 (·, λ̄, z̄) λw−w̄ dx dτ
Σ1 Q1 (9.16)
Z
+αε L0t (·, λ̄(1), z̄(1)) λw−w̄ (1) dx
Ω
D E
+ µε , gt0 (λ̄(1), z̄(1)) λw−w̄ (1) ∗,Ω ≥ 0,

where Z Z
H(λ̄, z̄, v̄, φ̄ε , αε ) = αε F (·, λ̄, z̄) dx + αε G(·, λ̄, v̄) ds
Ω Γ
Z Z
− (z̄ ∆φε + f (·, λ̄, z̄) φε ) dx + v̄ ∂ν φε ds.
Ω Γ
Let ψε be the solution of the differential equation (5.8) corresponding to (αε , µε ). With an
integration by parts, we obtain
Z 1 Z
ψε (w − w̄) dτ = λw−w̄ (1) αε L0t (·, λ̄(1), z̄(1)) dx
0 Ω
D E Z
0
+λw−w̄ (1) µε , gt (λ̄(1), z̄(1)) ∗,Ω + αε w̄ G0t (·, λ̄, v̄) λw−w̄ ds dτ
Σ1

18
 Z Z
+ αε Ft0 (·, λ̄, z̄) λw−w̄ dx dτ + w̄ ft0 (·, λ̄, z̄) φε λw−w̄ dx dτ.
Q1 Q1

Therefore, from (9.16), it follows that

Z 1  
(w − w̄) H(λ̄, z̄, v̄, φε , αε ) + ψε dτ ≥ 0 for all w ∈ Wad (ε). (9.17)
0

m̄
Let wo be in Co (]0, 1 − ε[) satisfying kwo k∞,]0,1−ε[ < 2, and define w
e by

 wo (τ ) + w̄(τ ) if τ ∈ [0, 1 − ε],
w(τ
e ) = wo (τ ) + w̄(τ ) =
w̄ if τ ∈ [1 − ε, 1].


It is clear that w
e belongs to Wad (ε). Therefore, by setting w = w
e in (9.17), it follows that:
Z 1−ε  
wo H(λ̄, z̄, v̄, φε , αε ) + ψε dτ ≥ 0 (9.18)
0

for all wo ∈ Co (]0, 1 − ε[) satisfying kwo k∞,]0,1−ε[ < m̄ 2 . Let w be in Co (]0, 1 − ε[), not identically
± w(τ ) −
zero. If we set wo (τ ) = ± 2kwk∞,]0,1[ m̄, then kwo k∞,]0,1−ε[ ≤ m̄
+ m̄
2 , kwo k∞,]0,1−ε[ ≤ 2 , and (9.18)
implies

m̄  Z 1−ε 
± w (H(λ̄, v̄, z̄, φ̄, ᾱ) + ψ̄) dτ ≥ 0.
2kwk∞,]0,1[ 0
R 1−ε  
Thus for every w ∈ Co (]0, 1 − ε[), we have 0 w H(λ̄, z̄, v̄, φε , αε ) + ψε dτ = 0. Since τ −→
H(λ̄(τ ), z̄(τ ), v̄(τ ), φε (τ ), αε ) + ψε (τ ) belongs to L1 (]0, 1 − ε[), we obtain (9.6). The proof is
complete. 
Rmeark 9.2 Due to Theorem 8.1, we have been able to prove that Aε ∩ B = ∅. If we define A
by replacing Wad (ε) by Wad in the definition of Aε , it is not necessarily true that A ∩ B = ∅.
This is the reason why we first write optimality conditions for (P eε ) and not directly for (P).
e In
Section 9.2 we are able to pass to the limit when ε tends to zero by using the nontriviality and the
complementarity conditions for (P eε ), and the fact that the interior of C in Cb (Ω) is nonempty.

9.2 Proof of Theorem 5.3

Due to (9.1), the sequence (αε , µε )ε is bounded in [0, 1] × (Cb (Ω))0 . Therefore, (αε )ε (or at least
a subsequence) converges to ᾱ ≥ 0, and there exists a generalized sequence, still indexed by ε,
such that (µε )ε converges to a limit µ̄ for the weak-star topology of (Cb (Ω))0 . From Theorem
7.2, it follows that

kφε kLδ (0,T ;W 1,d (Ω)) ≤ C αε kL0y (·, λ̄(1), z̄(1))k1,Ω + αε kF (·, λ̄, z̄)k1,Q1

+kgy0 (λ̄(1), z̄(1))k∞,Ω kµε k(Cb (Ω))0 ,
N
for every δ ≥ 1, d ≥ 1 satisfying 2d + 1δ > N2+1 . Then there exist a subsequence, still indexed
by ε, and φ̄ such that (φε )ε converges to φ̄ weakly in Lδ (0, T ; W 1,d (Ω)) for every (δ, d) satisfying
N 1 N +1 j0
2d + δ > 2 . Due to classical imbeddings, (φε )ε converges to φ̄ weakly in L (Q). For every

19
χ ∈ C 1 (Q1 ) ∩ Co (Q1 ∪ Ω1 ), we have
Z  
φε ∂τ χ + w̄ ∇χ ∇φε + w̄ fy0 (·, λ̄, z̄) χ φε dx dτ
Q1
Z Z
+ αε Fy0 (·, λ̄, z̄) χ dx dτ + αε L0y (·, λ̄(1), z̄(1)) χ(1) dx
Q1 Ω
D E D E
= − Πµε , gy0 (λ̄(1), z̄(1)) χ(1) = − µε , gy0 (λ̄(1), z̄(1)) χ(1) .
b,Ω ∗,Ω

By passing to the limit in this formulation, we obtain

Z  
φ̄ ∂τ χ + w̄ ∇χ ∇φ̄ + w̄ fy0 (·, λ̄, z̄) χ φ̄ dx dτ
Q1
Z Z
+ ᾱ Fy0 (·, λ̄, z̄) χ dx dτ + ᾱ L0y (·, λ̄(1), z̄(1)) χ(1) dx
Q1 Ω
D E D E
= − µ̄, gy0 (λ̄(1), z̄(1)) χ(1) = − Πµ̄ , gy0 (λ̄(1), z̄(1)) χ(1) ,
∗,Ω b,Ω

for every χ ∈ C 1 (Q1 ) ∩ Co (Q1 ∪ Ω1 ) . Therefore, φ̄ is the solution of (5.7). The sequence
(φε , dφ δ
dτ )ε is bounded in L (0, 1; W
ε 1,d
(Ω))×L1 (0, 1; (W 1,σ (Ω))0 ), for every δ ≥ 1, d ≥ 1, satisfying
N 1 N +1
2d + δ > 2 , and some σ big enough. By a compactness argument [19], we deduce that (φε )ε
is relatively compact in L1 (Q1 ). Therefore, we can assume that (φε )ε converges to φ̄ almost
everywhere on Q1 . We already know that (φε )ε is bounded in Lr (Q1 ) for every r < NN+2 , thus
by classical arguments, it follows that (φε )ε converges to φ̄ in Lr (Q1 ), for every 1 < r < NN+2 .
Since (φε )ε converges to φ̄ in L1 (Q1 ), we can easily prove that (ψε )ε converges to the solution ψ̄
of (5.8), in W 1,1 (0, 1).
• By passing to the limit in (9.2), we deduce (5.6). Moreover, by passing to the limit in (9.1),
we obtain ᾱ2 + (limε kµε k(Cb (Ω))0 )2 = 1. If ᾱ > 0, the proof is complete. If ᾱ = 0, we must prove
that kµ̄k(Cb (Ω))0 > 0. Since intCb (Ω) C =
6 ∅, there exists a ball B(z, 2ρ) ⊂ C in Cb (Ω), with center
z and radius 2ρ > 0. We can choose z̃ε ∈ B(0, 2ρ) such that hµ , z̃ε i∗,Ω = ρkµ k(Cb (Ω))0 . Since
z + z̃ε ∈ C, we have
D E
µε , z + z̃ε − g(λ̄(1), z̄(1)) ≤ 0.
∗,Ω
D E
By passing to the limit, we obtain: ρ + µ̄, z − g(λ̄(1), z̄(1)) ≤ 0, and thus µ̄ 6= 0.
∗,Ω

• By using the Green formula of Theorem 7.2, satisfied by φ̄ and φε , we have

Z Z
0
αε Fy (·, λ̄, z̄) ξv̄w̄,v−v̄ dx dτ + αε L0y (·, λ̄(1), z̄(1)) ξv̄w̄,v−v̄ (1) dx
Q1 Ω
D E
+ µε , gy0 (λ̄(1), z̄(1)) ξv̄w̄,v−v̄ (1) (9.19)
∗,Ω
Z   Z
= Λµε gy0 (λ̄(1), z̄(1)) ξv̄w̄,v−v̄ (1) d Π|µε | + w̄ ∂ν φε (v − v̄) ds dτ,
Γ Σ1

20
and Z Z
ᾱ Fy0 (·, λ̄, z̄) ξv̄w̄,v−v̄ dx dτ + ᾱ L0y (·, λ̄(1), z̄(1)) ξv̄w̄,v−v̄ (1) dx
Q1 Ω
Z   Z (9.20)
= Λµ̄ gy0 (λ̄(1), z̄(1)) ξv̄w̄,v−v̄ (1) d Π|µ̄| + w̄ ∂ν φ̄ (v − v̄) ds dτ,
Γ Σ1

where Λµ̄ : Cb (Ω) −→ L∞Π|µ̄| (Ω) is the operator defined in Theorem 3.1, and corresponding to µ̄.
Let us set
   
I = Σ1 w̄ ∂ν φε − ∂ν φ̄ (v − v̄) ds dτ + Γ Λµε gy0 (λ̄(1), z̄(1)) ξv̄w̄,v−v̄ (1) d Π|µε |
R R
Z  
− Λµ̄ gy0 (λ̄(1), z̄(1)) ξv̄w̄,v−v̄ (1) d Π|µ̄| .
Γ

From (9.19) and (9.20), it follows that

 
Iε ≤ (α − αε ) kFy0 (·, λ̄, z̄)k1,Q1 + kL0y (·, λ̄(1), z̄(1))k1,Ω kξv̄w̄,v−v̄ k∞,Q1
D E
+ µ̄ − µε , gy0 (λ̄(1), z̄(1)) ξv̄w̄,v−v̄ (1) .
∗,Ω

With the convergence results stated above, we obtain

lim I = 0. (9.21)
&0

Observe that, for v ∈ Vad , the function ξv̄w̄,v − ξw̄,v (ξw̄,v is the solution of (5.10) corresponding
to v, and ξv̄w̄,v is the solution of (8.2) corresponding to (v̄, w̄, v)) belongs to Cb (Q1 \ Σ1 ) and
satisfies ξv̄w̄,v − ξw̄,v ≡ 0 on Σ1 ∪ Ωo . It follows that
Z   Z  
0
Λµ̄ gy (λ̄(1), z̄(1)) ξv̄w̄,v−v̄ (1) d Π|µ̄| = Λµ̄ gy0 (λ̄(1), z̄(1)) ξw̄,v−v̄ (1) d Π|µ̄| . (9.22)
Γ Γ

Therefore, by taking (9.21) and (9.22) into account, and by passing to the limit in (9.5), we
obtain (5.9).
• Now, we study the convergence of (∆φε )ε and (∂ν φε )ε . Let εo > 0. For 0 < ε < 12 εo , we define
θε ∈ C 1 ([0, 1]), such that 0 ≤ θε ≤ 1 and
 3
 1 if t ∈ [0, 1 − εo ], |θε0 (t)| ≤ ,
θε (t) = ε o

if t ∈ [1 − ε, 1].

0

The function θε (φε − φ̄) is the solution of

−∂τ φ − w̄∆φ + w̄ fy0 (·, λ̄, z̄) φ = θε (αε − ᾱ)Fy0 (·, λ̄, z̄) + θε0 (φε − φ̄) in Q1 ,

φ = 0 on Σ1 , φ(1) = 0 in Ω.

Using regularity results for parabolic equations, we obtain

kφε − φ̄kLk∧r (0,1−εo ;W 2,k∧r (Ω)) ≤ kθε (φε − φ̄)kLk∧r (0,1;W 2,k∧r (Ω))
 
≤ C (αε − ᾱ) kFy0 (·, λ̄, z̄)kk,Q1 + kθε0 k∞,]0,1[ kφε − φ̄kr,Q1
 3 
≤C (αε − ᾱ) kFy0 (·, λ̄, z̄)kk,Q1 + kφε − φ̄kr,Q1
εo

21
for all 1 < r < NN+2 . With convergence results stated above, by passing to the limit when ε tends
to zero, we deduce that (∆φε )ε converges to ∆φ̄ in Lk∧r (Ω×]0, 1 − εo [), for all εo > 0. Since εo is
arbitrary chosen, we can suppose that the sequence (or at least a subsequence) ((∆φε − ∆φ̄)(τ ))ε
converges to zero in Lk∧r (Ω), for all τ ∈ I, where I ⊂]0, 1[ and L(I) = 1. In the same way we
prove that ((∂ν φε − ∂ν φ̄)(τ ))ε converges to zero in L1 (Γ), for all τ ∈ I. Thus, passing to the
limit when ε tends to zero in (9.6), we obtain (5.11). 

10 Proof of Theorem 3.2

Let (ȳ, ū, T̄ ) be a solution of (P). Set z̄(x, τ ) = ȳ(x, λ̄(τ )), v̄(s, τ ) = ū(s, λ̄(τ )), w̄(τ ) = T̄ ,
λ̄(τ ) = τ T̄ on [0, 1]. Due to Proposition 5.1, (λ̄, z̄, v̄, w̄) is a solution of (P).
e Let ᾱ, µ̄, φ̄, and
ψ̄ be the multipliers and the adjoint states of Theorem 5.3. Set p̄(x, t) = φ̄(x, λ̄−1 (t)), and
q̄(x, t) = ψ̄(x, λ̄−1 (t)). We can easily see that p̄, and q̄ are the solutions of (3.3) and (3.4). The
optimality conditions (3.1)-(3.7) easily follow from (5.5)-(5.11). 

References
[1] N. Arada, J. P. Raymond, Necessary optimality conditions for control problems and the
Stone-C̆ech compactification, SIAM J. Cont. and Optim., Vol. 37 (1999), 1011-1032.
[2] N. Arada, J. P. Raymond, Dirichlet boundary control of semilinear parabolic equations. Part
1: Problems with no state constraints, Applied Math. and Optim., Vol. 45 (2002), 125-143.
[3] N. Arada, J. P. Raymond, Dirichlet boundary control of semilinear parabolic equations. Part
2: Problems with pointwise state constraints, Applied Math. and Optim., Vol. 45 (2002),
145-167.
[4] N. Arada, J. P. Raymond, Optimality conditions for state constrained Dirichlet boundary
control problems, Journal of Optimization, Theory and Applications, Vol. 102 (1999), 51-68.
[5] V. Barbu, The time optimal control of the Navier-Stokes equations, Systems and Control
Letters, Vol. 30, (1997), 93-100.
[6] V. Barbu, ”Analysis and control of nonlinear infinite dimensional systems”, Academic Press,
New-York, 1993.
[7] H. O. Fattorini, Constancy of the Hamiltonian in infinite dimensional problems, Inter. Series
Num. Math., Vol. 91 (1989), 123-133.
[8] H. O. Fattorini, Invariance of the Hamiltonian in control problems for semilinear parabolic
distributed parameter systems, Inter. Series Num. Math., Vol. 118 (1994), 115-129.
[9] H. O. Fattorini, The time optimal control problems in Banach spaces, Appl. Math. Optim,
Vol. 1 (1974), 168-188.
[10] H. O. Fattorini, ”Infinite Dimensional Optimization and Control Theory”, Encyclopedia of
Mathematics and its Applications, Cambridge University Press, 1999.
[11] H. O. Fattorini, T. Murphy, Optimal problems for nonlinear parabolic boundary control
systems, SIAM J. Control Optim., Vol. 32 (1994), 1577-1596.
[12] H. O. Fattorini, T. Murphy, Optimal control problems for nonlinear parabolic boundary
control systems: The Dirichlet boundary condition, Diff. and Integ. Equat., Vol. 7 (1994),
1367-1388.

22
 [13] O. A. Ladyzenskaja, V. A. Solonnikov, and N.N. Ural’ceva, ”Linear and quasilinear equa-
tions of parabolic type”, Translations of Mathematical Monographs, Vol. 23, Amer. Math.
Soc., Providence, R. I., 1968.
[14] B. S. Mordukovich, K. Zhang, Minimax control of parabolic systems with Dirichlet boundary
conditions and state constraints, Appl. Math. Optim. Vol. 36 (1997), 323-360.
[15] I. Lasiecka, R. Triggiani, Th regulator problem for parabolic equations with Dirichlet bound-
ary control. Part I: Riccati’s feedback synthesis and regularity of optimal solutions, Appl.
Math. Optim., Vol. 16 (1987), 147-168.
[16] J. P. Raymond, Nonlinear boundary control of semilinear parabolic equations with pointwise
state constraints, Disc. Cont. Dyn. Syst., Vol. 3 (1997), 341-370.
[17] J. P. Raymond, H. Zidani, Pontryagin’s principle for time optimal problems, Journal of
Optimization, Theory and Applications, Vol. 101 (1999), 375-402.
[18] J. P. Raymond, H. Zidani, Time optimal problems with boundary controls, Diff. Int. Eq.,
Vol. 13 (2000), 1039-1072.
[19] J. Simon, Compact sets in Lp (0, T ; B), Ann. Mat. Pura. Appl., Vol. 196 (1987), 65-96.
[20] J. Yong, Time optimal control for semilinear distributed parameter systems - existence
theory and necessary conditions, Kodai. Math. J., Vol. 14 (1991), 239-253.
[21] E. Zuazua, Controllability of partial differential equations and its semi-discrete approxima-
tions, Discrete and Continuous Dynamical Systems, Vol. 8 (2002), 469-513.

23

View publication stats