vol. 34 (2005) No. 4
Superlinear elliptic systems with distributed and boundary controls
by Dorota Bors
Faculty of Mathematics, University of L´ od´z ul. S. Banacha 22, 90-238 L´od´z, Poland
e-mail: bors@math.uni.lodz.pl
Abstract: The paper investigates the nonlinear partial differ- ential equations of the superlinear elliptic type with the Dirichlet boundary data. Some sufficient conditions, under which the solu- tions of considered equations depend continuously on distributed and boundary controls, are proved. The proofs of the main results are based on variational methods.
Keywords: boundary value problems, continuous dependence, stability, variational methods.
1. Introduction
In the paper we investigate the elliptic systems of nonlinear partial differential equations with variable distributed parameters (controls) and variable boundary conditions (controls). The systems considered are of the form
−∆z (x) = G z (x, z (x) , u (x)) (1)
with the boundary condition
z (x) = v (x) on ∂Ω (2)
where x ∈ Ω ⊂ R n , n ≥ 2, Ω is a bounded domain with Lipschitzian boundary
∂Ω, z ( ·) ∈ H 1 Ω, R N
. We shall assume that the distributed control u ( ·) varies in the space L ∞ (Ω, R m ) and the boundary control v ( ·) belongs to the space of traces H 1/2 ∂Ω, R N
, N, p, m ≥ 1 (for details see Section 2).
In the paper the terms: distributed parameters and distributed controls as
well as variable boundary data (conditions), boundary controls are used inter-
changeably.
The main result of this paper is stated in Theorem 4.1 (Section 4). Under some suitable assumption we show that for an arbitrary pair of controls (u, v) there exists a solution z u,v to the system (1)-(2), which is stable with respect to distributed and boundary controls. By stability we understand here con- tinuous dependence of solutions on variable controls. More precisely, we prove that z u,v → z u
0,v
0in H 1 Ω, R N
provided that u tends to u 0 in L ∞ (Ω, R m ) and v tends to v 0 in H 1/2 ∂Ω, R N
. Furthermore, by applying the above sta- bility results we show the existence of optimal solution to the control problem described by (1)-(2) with an integral performance index (see Theorems 5.1 and 5.2). Similar results to Theorems 5.1 and 5.2 were proved for optimal control systems of ordinary differential equations by Macki and Strauss (1982, Chapter IV).
It is easy to notice that system (1) represents the Euler-Lagrange equation for the following integral functional (the functional of action)
F (z) = Z
Ω
1
2 |∇z (x)| 2 − G (x, z (x) , u (x))
dx (3)
where z ( ·) ∈ H 1 Ω, R N
, z (x) = v (x) for x ∈ ∂Ω a.e., v (·) ∈ H 1/2 ∂Ω, R N , u ( ·) ∈ L ∞ (Ω, R m ) .
On the function G we shall impose, besides some technical assumptions, the following condition
a < pG (x, z, u) ≤ hG z (x, z, u) , z i (4)
for some a > 0, p > 2 and |z| sufficiently large, which guarantees that prob- lem (1)-(2) is referred to as a superlinear boundary value problem, where the functional of action is unbounded from above and below.
Generally, in the theory of boundary value problems and its applications we consider, first of all, the problem of the existence of a solution and then the questions of stability, uniqueness, smoothness, etc.
R. Courant and D. Hilbert write in their monograph : ”A mathematical
problem which is to correspond to physical reality should satisfy the follow-
ing basic requirements: (1) The solution must exist. (2) The solution should
be uniquely determined. (3) The solution should depend continuously on the
data (requirement of stability)” and, next, they write: ”The third requirement,
particularly incisive, is necessary if the mathematical formulation is to describe
observable natural phenomena. Data in nature cannot possibly be conceived as
rigidly fixed: the mere process of measuring them involves small errors...” (see
Courant, Hilbert, 1962, Vol II Ch.III § 6.2).
A wide presentation of the methods and results related to the existence theory of variational and boundary value problems can be found, in particular, in monographs: Partial Differential Equations by Evans (1998), Probl`emes de Dirichlet Variationnels Non-Lin´eaires by Mawhin (1987), Critical Point Theory and Hamiltonian Systems by Mawhin and Willem (1989), Minimax Methods in Critical Point Theory with Applications to Differential Equations by Rabinowitz (1986), Variational Methods by Struwe (1990), Minimax Theorems by Willem (1996). To obtain the existence result we apply the Mountain Pass Theorem presented in the above monographs.
As far as we know the question of continuous dependence of solutions on distributed and boundary controls for nonlinear partial differential equation of elliptic type has not been investigated up to now. However, in the 1970s some papers were published in which authors dealt with Dirichlet problem for scalar ordinary differential equations with two-point boundary value conditions. For example in Ingram (1972), Klaasen (1970), Lepin, Ponomariev (1973), Sedziwy (1971) some stability results are proved by means of direct methods. The ques- tion of stability of vector systems of ordinary differential equations was investi- gated in Walczak (1995), where the proofs of the main results were obtained by variational methods.
The first result concerning the question of continuous dependence of solutions of the linear partial differential equation of elliptic type with the variable Dirich- let boundary data and parameters was published in Ole˘ınik (1952). In this work the linear partial differential systems are defined in the classical spaces. Similar results for scalar, but still linear partial differential equation with the Dirichlet boundary conditions defined in some Sobolev spaces are proved in Kok, Pen- ning (1980/81). Stability results for nonlinear partial differential equation are presented in Walczak (1998), where the author, applying variational methods, considers coercive functional of action.
2. Formulation of the problem and basic assumptions
By H 1 Ω, R N
we shall denote the Sobolev space of functions z = z (x) defined on a bounded domain Ω ⊂ R n , n ≥ 2, such that z(·) ∈ L 2 (Ω, R N ), whose (distributional) derivatives ∇z are elements of the space L 2 (Ω, R N n ) with the norm
kzk 2 H
1(Ω) = Z
Ω
|∇z (x)| 2 + |z (x)| 2
dx.
By H 1/2 Ω, R N
we denote the space of all functions v( ·) ∈ L 2 (Ω, R N ) for which I 0 (v) =
Z
Ω
Z
Ω
|v (x) − v (y)| 2
|x − y| n+1 dxdy < ∞, equipped with the norm
kvk 2 H
1/2(Ω) = kvk 2 L
2(Ω) + I 0 (v)
(see Theorem 7.48 in Adams, 1975 or Definition 6.8.2 in Kufner, John, Fucik, 1977).
Covering ∂Ω by coordinate patches, we define the space H 1/2 = H 1/2 ∂Ω, R N as before via such charts (see 7.51 in Adams, 1975 or § 6 in Kufner, John, Fucik, 1977) with an analogous norm.
H 1/2 ∂Ω, R N
is said to be the space of traces (boundary values or bound- ary controls) of functions from the space H 1 Ω, R N
. Throughout the paper, we shall assume that Ω satisfies any condition which guarantees a compact em- bedding of H 1 Ω, R N
into L s Ω, R N
with s ∈ [1, 2n/ (n − 2)) if n ≥ 3 and s ≥ 1 if n = 2; for example, ∂Ω may be Lipschitzian, i.e. Ω ∈ C 0,1 (see Kufner, John, Fucik, 1977).
Let us recall some facts from the trace theory. According to Theorem 6.8.13 of Kufner, John, Fucik (1977), we have the existence of a unique continuous linear mapping R acting from H 1 Ω, R N
into H 1/2 ∂Ω, R N
such that Rz = z | ∂Ω for all z ∈ C ∞ Ω
. The value Rz is often referred to as the trace of the function z on the boundary ∂Ω and we usually write z | ∂Ω instead of Rz.
Therefore, in system (1)-(2), the equality z = v on ∂Ω has to be understood in the trace sense.
In our considerations, an essential role is played by Theorem 6.9.2 from Kufner, John, Fucik (1977). We can prove that any function from the space H 1/2 ∂Ω, R N
is a trace of a function from H 1 Ω, R N
, namely, there exists a continuous linear operator T (lifting operator) acting from H 1/2 ∂Ω, R N
into H 1 Ω, R N
such that, for p = T v, we have p = v on ∂Ω (in the trace sense) and there exists a positive constant c such that the following inequality holds:
kT vk H
1≤ c kvk H
1/2where c depends on the mapping T and the description of ∂Ω. Besides, it is easy to check that there exists the mapping T such that
H 1 Ω, R N
= H 0 1 Ω, R N
⊕ Im T, i.e. for any y ∈ H 0 1 Ω, R N
and v ∈ H 1/2 ∂Ω, R N
(y, T v) H
1= 0,
where ( ·, ·) H
1denotes the scalar product in H 1 Ω, R N . Let v 0 be a fixed point from the space H 1/2 ∂Ω, R N
. By V denote the set of all boundary conditions of the form
V = n
v ∈ H 1/2 ∂Ω, R N
: kv − v 0 k H
1/2≤ k 1 o for k 1 > 0 and U the set of parameters
U = {u ∈ L ∞ (Ω, R m ) : u (x) ∈ U ⊂ R m and kuk L
∞≤ k 2 } for k 2 > 0 and some subset U of R m .
In this paper we shall consider a control system governed by an elliptic vector equation with variable parameters and boundary data of the form
−∆z (x) = G z (x, z (x) , u (x))
z (x) = v (x) on ∂Ω, (5)
where z ( ·) ∈ H 1 Ω, R N
, ∆z = ∆z 1 , ..., ∆z N
, ∆z i = ∂(x ∂
2z
1i)
2+ ... + ∂(x ∂
2nz
i)
2for i = 1, 2, ..., N, v ( ·) ∈ V, u (·) ∈ U and G : Ω × R N × R m → R, G z = (G z
1, ..., G z
N) .
Functional of action for system (5) has the form F (z) =
Z
Ω
1
2 |∇z (x)| 2 − G (x, z (x) , u (x))
dx, (6)
where z ∈ H 1 Ω, R N
, z (x) = v (x) a.e. on ∂Ω and ∇z = ∇z 1 , ..., ∇z N ,
∇z i =
∂z
i∂x
1, ..., ∂x ∂z
infor i = 1, 2, ..., N.
On the function G we shall impose the following conditions:
(2.1) G, G z are Carath´eodory functions, i.e. they are measurable with respect to x for any (z, u) ∈ R N × R m and continuous with respect to (z, u) ∈ R N × R m for x ∈ Ω a.e.;
(2.2) for any bounded subset U 0 ⊂ U, there exists c > 0 such that
|G (x, z, u)| ≤ c (1 + |z| s ) , |G z (x, z, u) | ≤ c
1 + |z| s−1 ,
for z ∈ R N , u ∈ U 0 and x ∈ Ω a.e., where s ∈ (1, 2 ∗ ) with 2 ∗ = 2n/ (n − 2) if n ≥ 3 and 2 ∗ = ∞ if n = 2;
(2.3) there exist p > 2, a > 0 and R > 0 such that
a < pG (x, z, u) ≤ hG z (x, z, u) , z i
for x ∈ Ω a.e., u ∈ U and |z| ≥ R;
(2.4) there exist ζ > 0 and 0 < b < 1 2 such that
G (x, z, u) + 1 2 |z| 2 ≤ 2 b |z − T v 0 (x) | 2
for |z| ≤ ζ, u ∈ U and x ∈ Ω a.e., where T is a fixed inverse operator to the trace operator such that H 1 Ω, R N
= H 0 1 Ω, R N
⊕ Im T.
3. Auxiliary lemmas
We begin with some definitions. Let I ( ·) : E → R be a functional of C 1 −class defined on real Banach space E (in our case on H 0 1 Ω, R N
). A point y ∈ E is called a critical point of the functional I ( ·) if I ′ (y) = 0 and moreover the number c = I (y) is referred to as a critical value.
We say that the functional I ( ·) satisfies the Palais-Smale (PS for short) condition if any sequence {y k } ⊂ E such that I ′ (y k ) → 0 and |I (y k ) | < C for some C > 0 is relatively compact in the strong topology of E.
In this part we shall use the following version of the Mountain Pass Theorem (see Mawhin, Willem, 1989; Struwe, 1990).
Theorem 3.1 If
1 0 there exist w 0 , w 1 ∈ E and a bounded neighborhood B of w 0 , such that w 1 ∈ E \ B,
2 0 inf y∈∂B I (y) > max {I (w 0 ) , I (w 1 ) } ,
3 0 c = inf g∈M max t∈[0,1] I (g (t)), where M =
g ∈ C ([0, 1] , E) : g (0) = w 0 , g (1) = w 1
,
4 0 I ( ·) satisfies the (PS) condition,
then c is a critical value and c > max {I (w 0 ) , I (w 1 ) } . In this section we shall use the following notations:
Let M r denote a set of continuous mappings g : [0, 1] → B r such that g (0) = w 0 , g (1) = w 1 and w 0 , w 1 ∈ B r , where B r = n
y ∈ H 0 1 Ω, R N
: kyk H
10< r o , r > 0.
Next, let I k : H 0 1 Ω, R N
→ R, k = 0, 1, 2, ... denote an arbitrary sequence of functionals of C 1 -class, and c k (r) be the value given by the formula
c k (r) = inf
g∈M
rt∈[0,1] max I k (g (t)) . (7)
In this case, the set of all critical points Y k (r) corresponding to the value c k (r) has the form
Y k (r) = {y ∈ B r : I k (y) = c k (r) and I k ′ (y) = 0 } (8) for k = 0, 1, 2, ....
For the sequence of the functionals {I k ( ·)} we shall prove:
Lemma 3.1 Assume that
1 0 the functionals I k ( ·) , k = 0, 1, 2, ... are of C 1 −class and I 0 ( ·) satisfies the (PS) condition,
2 0 the sequences {I k ( ·)} , {I k ′ ( ·)} tend uniformly on the ball B r to I 0 ( ·) , I 0 ′ ( ·) , respectively,
3 0 the sets Y k (r) defined by (8) are not empty for k = 0, 1, 2, ....
Then any sequence {y k } such that y k ∈ Y k (r) , k = 1, 2, ... is relatively compact in H 0 1 Ω, R N
, i.e. Lim sup Y k (r) is a nonempty set and Lim sup Y k (r) ⊂ Y 0 (r), where Lim sup Y k (r) is the upper limit of the sets Y k (r) , k = 1, 2, ..., i.e. the set of all cluster points with respect to the strong topology of H 0 1 Ω, R N of a sequence {y k } such that y k ∈ Y k (r) , k = 1, 2, ....
Proof. Let {y k } be an arbitrary sequence such that y k ∈ Y k (r) for k = 0, 1, 2, ....
By assumption 2 0
, we obtain 0 = lim k→∞ (I k ′ (y k ) − I 0 ′ (y k )) = − lim k→∞ I 0 ′ (y k ) because I k ′ (y k ) = 0 for k = 0, 1, 2.... Furthermore ky k k H
01< r hence the se- quence I 0 (y k ) is bounded. Since I 0 ( ·) satisfies the (P S) condition, we infer that {y k } is a relatively compact sequence with respect to the strong topology of H 0 1 Ω, R N
, i.e. Lim sup Y k (r) is not empty.
Let us notice that lim k→∞ c k (r) = c 0 (r). Indeed, by assumption 2 0 , we obtain
c k (r) = inf
g∈M
rt∈[0,1] max [(I k (g (t)) − I 0 (g (t))) + I 0 (g (t))]
≤ inf g∈M
r
t∈[0,1] max (ε + I 0 (g (t))) = ε + c 0 (r) for any ε > 0 and k sufficiently large.
Similar consideration is applied to c 0 (r) ≤ ε + c k (r) . Consequently,
k→∞ lim c k (r) = c 0 (r) . (9)
From 2 0
it follows that for any sequence {y k } such that y k ∈ Y k (r) for k = 1, 2, ..., A k = I 0 (y k ) − I k (y k ) → 0 and moreover by (9), we conclude that lim k→∞ I 0 (y k ) = lim k→∞ (A k + c k (r)) = c 0 (r).
We have proved that Lim sup Y k (r) is a nonempty set. Let ˜ y be an arbitrary point of this set. By definition of the upper limit of sequence of sets, ˜ y is a cluster point of some sequence {y k } such that y k ∈ Y k (r) . Passing, if necessary, to a subsequence, we may assume that y k → ˜y. Suppose that ˜y / ∈ Y 0 (r) , i.e.
I 0 (˜ y) 6= c 0 (r) or I 0 ′ (˜ y) 6= 0. Let us observe that the second inequality is false.
Indeed, assumption 2 0
and the first part of our proof imply I 0 ′ (˜ y) = lim
k→∞ I 0 ′ (y k ) = lim
k→∞ (I 0 ′ (y k ) − I k ′ (y k )) = 0.
Putting α = I 0 (˜ y) − I 0 (y 0 ) , where y 0 ∈ Y 0 (r) and α 6= 0, we see that
c k (r) − c 0 (r) = I k (y k ) − I 0 (y 0 ) = [I k (y k ) − I 0 (y k )] + [I 0 (y k ) − I 0 (˜ y)] + α.
By virtue of (9), assumption 2 0
and continuity of the functional I 0 ( ·) , we have the following convergences:
c k (r) − c 0 (r) → 0, I k (y k ) − I 0 (y k ) → 0 and I 0 (y k ) − I 0 (˜ y) → 0 as k → ∞.
This contradicts the fact that α 6= 0. Thus ˜y ∈ Y 0 (r) and consequently Lim sup Y k (r) is a nonempty set and Lim sup Y k (r) ⊂ Y 0 (r) . The proof is completed.
In our further consideration we need a specific form of the functional I k ( ·), which is strongly connected with the form of the functional of action given by (6).
Let T be a fixed lifting operator such that H 1 Ω, R N
= H 0 1 Ω, R N
⊕Im T.
For a fixed boundary value v, substituting z = y + T v into (6) we get F v,u (y) =
Z
Ω
1
2 |∇y (x) + ∇ (T v) (x)| 2 − G (x, y (x) + (T v) (x) , u (x)) dx
where y ( ·) ∈ H 0 1 Ω, R N
, v ( ·) ∈ V and u (·) ∈ U.
Let {v k } ∈ V, k = 0, 1, 2, ... be a sequence of boundary controls and {u k } ∈ U, k = 0, 1, 2... a sequence of distributed controls. Denote by {F k ( ·)} the following sequence of functionals
F k (y) = F v
k,u
k(y) + Z
Ω
h G (x, T v k (x) , u k (x)) − 1 2 |∇ (T v k ) (x) | 2 i dx
for which we define the value c k = inf
g∈M max
t∈[0,1] F k (g (t)) , (10)
where M is a set of continuous mappings g : [0, 1] → H 0 1 Ω, R N
such that g (0) = w 0 , g (1) = w 1 (w 0 , w 1 are some elements from H 0 1 Ω, R N
).
Let Y k denote the set of critical points corresponding to the value c k , Y k =
y ∈ H 0 1 Ω, R N
: F k (y) = c k and F k ′ (y) = 0
(11) for k = 0, 1, 2, .... In Section 4, we shall prove that for each k ∈ N the set Y k is not empty and the sequence of sets {Y k } possesses nonempty upper limit such that Lim sup Y k ⊂ Y 0 .
In the proof of the main theorem we shall use the following lemma:
Lemma 3.2 If the function G satisfies conditions (2.1)-(2.3), then there exists a ball B ρ in H 0 1 Ω, R N
such that Y k ⊂ B ρ = n
y ∈ H 0 1 Ω, R N
: kyk H
01< ρ o , ρ > 0 for any boundary controls v k ∈ V and distributed controls u k ∈ U.
Proof. Let us notice that the set {c k : v k ∈ V, u k ∈ U} is bounded from above.
Indeed, for any k ∈ N, from assumptions (2.2), (2.3) we get c k = inf
g∈M max
t∈[0,1] F k (g (t)) ≤ max
t∈[0,1] F k ((1 − t)w 0 + tw 1 )
= max
t∈[0,1]
Z
Ω 1
2 |(1 − t)∇w 0 + t ∇w 1 | 2 + ((1 − t)∇w 0 + t ∇w 1 , ∇ (T v k ))
−G (x, (1 − t)w 0 + tw 1 + T v k , u k ) + G (x, T v k , u k )
! dx
≤ max
t∈[0,1]
Z
Ω
2(1 − t) 2 |∇w 0 | 2 + 2t 2 |∇w 1 | 2 dx
− Z
Ω
+tG (x, (1 − t)w 0 + tw 1 + T v k , u k ) dx
− Z
Ω
−tG (x, (1 − t)w 0 + tw 1 + T v k , u k ) dx
!
+ Z
Ω
1
2 |∇ (T v k ) | 2 + G (x, T v k , u k ) dx
≤ max
t∈[0,1]
2(1 − t) 2 kw 0 k 2 + 2t 2 kw 1 k 2 − a p
Ω + t
+ c (1 + R s ) Ω − t + kT v k k 2 H
1+ c |Ω| + cc 1 kT v k k s H
1≤ 2 max n
kw 0 k 2 , kw 1 k 2 o
+ c (1 + R s ) |Ω| + kT v k k 2 H
1+ c |Ω| + cc 1 kT v k k s H
1≤ 2 max n
kw 0 k 2 , kw 1 k 2 o
+ D ≤ ˜c, i.e.
c k ≤ ˜c
where D, ˜ c are some constants, Ω + t = {x ∈ Ω : |((1 − t)w 0 + tw 1 + T v k ) (x) | ≥ R}
and Ω − t = {x ∈ Ω : |((1 − t)w 0 + tw 1 + T v k ) (x) | < R} . Furthermore, for any v k ∈ V, u k ∈ U and y ∈ Y k we obtain
p˜ c ≥ pc k = p F k (y) − hF k ′ (y) , y + T v k i
= p−2 2 kyk 2 H
01+ (p − 2) Z
Ω
( ∇y, ∇ (T v k )) dx − Z
Ω
|∇ (T v k ) | 2 dx
+ Z
Ω
− pG (x, y + T v k , u k ) + pG (x, T v k , u k ) + (G y (x, y + T v k , u k ) , y + T v k ) dx
≥ p−2 2 kyk 2 H
01− (p − 2) kyk H
01c 1 − c 2 1
+ Z
Ω
+( −pG (x, y + T v k , u k ) + pG (x, T v k , u k ) + (G y (x, y + T v k , u k ) , y + T v k )) dx
+ Z
Ω
−( −pG (x, y + T v k , u k ) + pG (x, T v k , u k ) + (G y (x, y + T v k , u k ) , y + T v k )) dx
≥ p−2 2 kyk 2 H
01+ D 1 kyk H
01+ D 2
where D 1 and D 2 are some constants, Ω + = {x ∈ Ω : |(y + T v k ) (x) | ≥ R} and Ω − = {x ∈ Ω : |(y + T v k ) (x) | < R} .
Thus
p˜ c ≥ p−2 2 kyk 2 H
10+ D 1 kyk H
01+ D 2 . (12) Since p − 2 > 0, there exists ρ > 0 such that y ∈ B ρ . Consequently, Y k ⊂ B ρ for any v k ∈ V and u k ∈ U.
Without loss of generality, we can assume that w 0 = 0. We shall prove that there exist a bounded neighborhood B of w 0 and some point w 1 ∈ B such that / the assumptions of the Mountain Pass Theorem are satisfied.
Lemma 3.3 Suppose that
1 0 conditions (2.1)-(2.4) are satisfied,
2 0 the sequence {v k } ⊂ V tends to v 0 in H 1/2 ∂Ω, R N
and the sequence {u k } ⊂ U tends to u 0 in L ∞ (Ω, R m ).
Then there exist a ball B η ⊂ H 0 1 Ω, R N
and element w 1 ∈ B / η such that inf ∂B
ηF k > 0 and F k (w 1 ) < 0 for any v k ∈ V and u k ∈ U, where B η =
y ∈ H 0 1 Ω, R N
: kyk H
1< η
, for η > 0.
Proof. In a similar way as in the proof of Theorem 3.3 in Mawhin (1987), we obtain that there exists a constant a 0 > 0 such that
G (x, z, u) ≥ a 0 |z| p ,
for |z| ≥ R, u ∈ U and x ∈ Ω a.e. and furthermore by (2.2) we have the existence of a positive constant a 1 such that G (x, z, u) ≥ a 0 |z| p − a 1 for z ∈ R N , u ∈ U and x ∈ Ω a.e., which lead to the fact that p, s ∈ (2, 2 ∗ ). By conditions (2.2), (2.4) we conclude that there exist b ∈ 0, 1 2
and A > 0 such
that G (x, z, u) + 1 2 |z| 2 ≤ b 2 |z − T v 0 (x) | 2 +A |z − T v 0 (x) | s for z ∈ R N , u ∈ U,
x ∈ Ω a.e. and s ∈ (1, 2 ∗ ) . For fixed k ∈ N, by the equality (y, T v k ) H
1= 0, we get
F k (y) = Z
Ω
1
2 |∇y| 2 + ( ∇y, ∇ (T v k )) − G (x, y + T v k , u k ) + G (x, T v k , u k ) dx
= Z
Ω
1
2 |∇y| 2 + ( ∇y, ∇ (T v k )) + 1 2 |y + T v k | 2 − 1 2 |y + T v k | 2
−G (x, y + T v k , u k ) + G (x, T v k , u k ) dx
= 1 2 kyk 2 H
1+ (y, T v k ) H
1− Z
Ω
G (x, y + T v k , u k ) + 1 2 |y + T v k | 2 dx +
Z
Ω
G (x, T v k , u k ) + 1 2 |T v k | 2 dx
≥ 1 2 − b
kyk 2 H
1− Ac 1 2 s−1 kyk s H
1− 3b 2 kT v k − T v 0 k 2 L
2− Ac 2 2 s kT v k − T v 0 k s H
1where c 1 , c 2 > 0. Since 1 2 − b > 0, v k → v 0 in the strong topology of H 1/2 ∂Ω, R N
and s > 2, there exists η > 0 such that inf ∂B
ηF k ≥ α > 0 for k sufficiently large.
Now, we shall prove that for any v k ∈ V and u k ∈ U there exists w 1 ∈ B / η such that F k (w 1 ) < 0.
For fixed y ( ·) ∈ H 0 1 Ω, R N
, y 6= 0 and l > 0, we have F k (ly) =
Z
Ω
1
2 |l∇y| 2 + l ( ∇y, ∇ (T v k )) − G (x, ly + T v k , u k ) +G (x, T v k , u k )
dx
≤ 1 2 l 2 kyk 2 H
1+ lc 1 kyk H
1− Z
Ω
(a 0 |ly + T v k | p − a 1 ) dx + Z
Ω
G (x, T v k , u k ) dx
≤ 1 2 l 2 kyk 2 H
1+ lc 1 kyk H
1− a 0 l p Z
Ω
y + T v l
kp
dx + c 2 .
Since 2 < p < 2 ∗ and a 0 > 0 we infer that lim l→∞ F k (ly) = −∞. Accordingly, there exists l 0 > 0 such that for w 1 = l 0 y we have kw 1 k H
1≥ η and F k (w 1 ) < 0 for any v k ∈ V and u k ∈ U.
Now we formulate some sufficient conditions, which guarantee a uniform con- vergence of a sequence of functionals and a sequence of derivative of functionals on any ball from the space H 0 1 Ω, R N
. Lemma 3.4 If
1 0 conditions (2.1), (2.2) are satisfied,
2 0 the sequence {v k } ⊂ V tends to v 0 in H 1/2 ∂Ω, R N
and the sequence {u k } ⊂
U tends to u 0 in L ∞ (Ω, R m ) ,
3 0 for any u 0 ∈ U and ε > 0, there exists a constant c > 0 such that
|G (x, z, u 1 ) − G (x, z, u 2 ) | ≤ c
1 + |z| 2
|u 1 − u 2 |
|G z (x, z, u 1 ) − G z (x, z, u 2 ) | ≤ c (1 + |z|) |u 1 − u 2 |
for x ∈ Ω a.e., z ∈ R N and for any u 1 , u 2 ∈ U with |u 1 − u 0 | < ε and
|u 2 − u 0 | < ε,
then the sequences {F k ( ·)} , {F k ′ ( ·)} tend uniformly on any ball from H 0 1 Ω, R N to F 0 ( ·) and F 0 ′ ( ·) , respectively.
Proof. For any B ρ ⊂ H 0 1 Ω, R N
and y ∈ B ρ , we have
|F k (y) − F 0 (y) | ≤ Z
Ω
|∇y| |∇ (T v k ) − ∇ (T v 0 ) | dx
+ Z
Ω
(G (x, T v k , u k ) − G (x, T v 0 , u 0 )) dx
+ Z
Ω
(G (x, y + T v k , u k ) − G (x, y + T v 0 , u 0 )) dx
≤ ρ k∇ (T v k ) − ∇ (T v 0 ) k L
2+ Z
Ω
|G (x, T v k , u 0 ) − G (x, T v 0 , u 0 ) | dx
+ Z
Ω
|G (x, y + T v k , u 0 ) − G (x, y + T v 0 , u 0 ) | dx + ku k − u 0 k L
∞D 1 + D 2 k∇T v k k 2 L
2< ε
for k sufficiently large. In fact, from assumption 2 0
it follows that we have the strong convergence of T v k to T v 0 in H 1 Ω, R N
. By virtue of assumptions (2.1), (2.2), 3 0
and the Krasnosielskii theorem on continuity of Niemycki’s operator, we conclude that the right hand side of the above inequality tends to 0 for any y ∈ B ρ . It means that the sequence {F k ( ·)} tends uniformly to F 0 ( ·) on any ball B ρ .
Similar arguments are applied to the case of a uniform convergence of the sequence {F k ′ ( ·)} to F 0 ′ ( ·) on any ball from H 0 1 Ω, R N
. Let us take any ball B ρ ⊂ H 0 1 Ω, R N
. For any y ∈ B ρ and h ∈ H 0 1 Ω, R N
such that h ∈ B 1 by simple calculations we have
|hF k ′ (y) − F 0 ′ (y) , h i| ≤ k∇ (T v k ) − ∇ (T v 0 ) k L
2+ Z
Ω
|(G y (x, y + T v k , u k ) − G y (x, y + T v 0 , u 0 ) , h) | dx < ε
for k sufficiently large, and the lemma follows.
4. The main result
Theorem 4.1 Suppose that
1 0 the function G satisfies conditions (2.1)-(2.4) and assumption 3 0 from Lemma 3.4,
2 0 the sequence {v k } ⊂ V tends to v 0 in H 1/2 ∂Ω, R N
and the sequence {u k } ⊂ U tends to u 0 in L ∞ (Ω, R m ).
Then
(a) for any k the set of critical points Y k of the functional F k ( ·) is nonempty and does not contain the trivial solution,
(b) any sequence {y k } such that y k ∈ Y k , k = 1, 2, ..., is relatively compact in H 0 1 Ω, R N
and Lim sup Y k ⊂ Y 0 .
Proof. Applying the Mountain Pass Theorem, we shall prove assertion (a) of our theorem. The functional F k ( ·) , k = 0, 1, 2, ... is of C 1 −class on H 0 1 Ω, R N
. From Lemma 3.3 it follows that there exist the ball B η and the point w 1 ∈ H 0 1 Ω, R N
(independent of the choice of k) such that w 1 ∈ B / η and inf ∂B
ηF k >
0 = max {F k (0) , F k (w 1 ) } , k = 0, 1, 2, .... Using assumptions (2.2) and (2.3), we shall demonstrate that the functional F k ( ·) , k = 0, 1, 2, ... satisfies the (PS) condition. For fixed k, let {y i } denote a sequence such that {F k (y i ) } is bounded and F k ′ (y i ) → 0 as i → ∞. Thus, there exist constants C 1 , C 2 > 0 such that
|F k (y i ) | ≤ C 1 and kF k ′ (y i ) k ≤ C 2 for i ∈ N. In the same manner as in the proof of Lemma 3.2, we obtain the following inequality
C 1 p + C 2
p 1 + d 2 ky i k H
10+ C 2 kT v k k H
1≥ C 1 p + C 2 ky i + T v k k H
1≥ pF k (y i ) − hF k ′ (y i ) , y i + T v k i ≥ p−2 2 ky i k 2 H
10+ D 1 ky i k H
10+ D 2 , where D 1 and D 2 are some constants. Hence
ky i k 2 H
01≤ p−2 2
pC 1 + D 3 ky i k H
01+ C 2 kT v k k H
1− D 2
for i ∈ N,
where D 3 = C 2
√ 1 + d 2 − D 1 and D 1 , D 2 , C 1 , C 2 are described above. It means that the sequence {y i } is bounded in H 0 1 Ω, R N
and therefore contains a subsequence, denoted by {y i } , such that y i tends to y 0 weakly in H 0 1 Ω, R N
. It is a well-known fact that the space H 0 1 Ω, R N
is compactly embedding into the space L s Ω, R N
with s ∈ [1, 2 ∗ ) . Consequently,
hF k ′ (y i ) − F k ′ (y 0 ) , y i − y 0 i → i→∞ 0
and because of
hF k ′ (y i ) − F k ′ (y 0 ) , y i − y 0 i
= ky i − y 0 k 2 H
10+ Z
Ω
(G y (x, y 0 + T v k , u k ) − G y (x, y i + T v k , u k ) , y i − y 0 ) dx
and the growth conditions (2.2), we get
Z
Ω
(G y (x, y 0 + T v k , u k ) − G y (x, y i + T v k , u k ) , y i − y 0 ) dx
≤ ky i − y 0 k L
s
Z
Ω
|G y (x, y i + T v k , u k ) − G y (x, y 0 + T v k , u k ) |
s−1sdx
s−1 s
≤ ky i − y 0 k L
sc2
s+1s
Z
Ω
(1 + |y i + T v k | s + |y 0 + T v k | s ) dx
s−1 s