SENSITIVITY ANALYSIS FOR EQUILIBRIUM PROBLEMS AND RELATED PROBLEMS
LAM QUOC ANH
Trang nhan đề
Lời cam đoan
Lời nói đầu
Mục lục
Phần 1
Chương_1: Semicontinuity of the solution sets of multivalued vector quasiequilibrium
Chương_2: Semicontinuity of the approximate solution sets of multivalued quasiequilibrium problems
Chương_3: Semicontinuity of the solution sets of symmetric quasiequilibrium problems
Chương 4: Semicontinuity of the solution sets to quasivariational inclusion problems
Phần 2
Chương_5: Uniqueness and holder continuity of the solution to equilibrium problems
Chương_6: Holder continuity of the unique solution to quasiequilibrium problems
Thông tin luận án
Tài liệu tham khảo
Table of Contents
Foreword i
Part I. Semicontinuity of solutions sets 1
Chapter 1. Semicontinuity of the solution set of multivalued vector
quasiequilibrium problems 2
Chapter 2. Semicontinuity of the approximate solution sets of multivalued
quasiequilibrium problems 41
Chapter 3. Semicontinuity of the solution sets of symmetric
quasiequilibrium problems 56
Chapter 4. Semicontinuity of the solution sets to quasivariational
inclusion problems 77
Part II. H¨older continuity of the unique solution 107
Chapter 5. Uniqueness and H¨older continuity of the solution to
equilibrium problems 108
Chapter 6. H¨older continuity of the unique solution to
quasiequilibrium problems 134
List of the papers related to the thesis 170
iv
15 trang |
Chia sẻ: maiphuongtl | Lượt xem: 1625 | Lượt tải: 0
Bạn đang xem nội dung tài liệu Đề tài Sensitivity analysis for equilibrium problems and related problems, để tải tài liệu về máy bạn click vào nút DOWNLOAD ở trên
Chapter 2
Semicontinuity of the approximate
solution sets of multivalued
quasiequilibrium problems
41
SEMICONTINUITY OF THE APPROXIMATE
SOLUTION SETS OF MULTIVALUED
QUASIEQUILIBRIUM PROBLEMS
Lam Quoc Anh 2 Department of Mathematics, Teacher College, Cantho
University, Cantho, Vietnam
Phan Quoc Khanh 2 Department of Mathematics, International University
of Hochiminh City, Linh Trung, Thu Duc, Hochiminh City, Vietnam
2 We consider two kinds of approximate solutions and approximate solution sets to multivalued
quasiequilibrium problems. Sufficient conditions for the lower semicontinuity, Hausdorff lower
semicontinuity, upper semicontinuity, Hausdorff upper semicontinuity and closedness of these
approximate solution sets are established. Applications in approximate quasivariational inequal-
ities, approximate fixed points and approximate quasioptimization problems are provided.
Keywords Lower and upper semicontinuity; Hausdorff lower and upper semicontinuity;
Closedness of multifunctions; ε-solutions; Quasiequilibrium problems; Quasivariational inequali-
ties; ε-fixed points; ε-quasioptimization problems.
1. INTRODUCTION AND PRELIMINARIES
Among various meanings of stability and sensitivity of the solutions of a problem the
semicontinuity has been increasingly interested recently in the literature. Upper semiconti-
nuity is investigated in [12, 15−17] for variational inequalities and in [3, 5, 7−10, 14, 18] for
equilibrium problems. Lower semicontinuity is studied in [19] for minimization problems, in
[12, 16, 17] for variational inequalities and in [3, 5, 7, 8, 14] for equilibrium problems. Beside
semicontinuity we observe [1, 2, 4, 6, 8−10] which deal with stability of equilibrium problems,
where Ho¨lder continuity is investigated.
On the other hand, in many practical problems, exact solutions do not exist since the
data of the problems are not sufficiently “regular”. Moreover, the data of the problems have
often been obtained approximately by measure devices or by statistical results. Mathematical
models of practical problems describe real situations also approximately and hence the existence
of exact solutions of mathematical models may become unimportant. That is why approximate
solutions are of real interest. Observing that multivalued quasiequilibrium problems are rather
general problems, which include quasivariational inequalities, complementarity problems, fixed-
point and coincidence-point problems, optimization, Nash equilibrium problems, etc as special
cases, in this note we consider the semicontinuity properties of the approximate solutions of
quasiequilibrium problems. In particular, many results of [16] are improved and extended to
more general settings.
The problems under our consideration are as follows. Throughout the paper, unless speci-
fied otherwise, let X, M , N , Λ be Hausdorff topological spaces and let Y be an invariant metric
Address corresponding to Phan Quoc Khanh, Department of Mathematics, International University of
Hochiminh City, Khu pho 6, Linh Trung, Thu Duc, Hochiminh City, Vietnam; E-mail: pqkhanh@hcmiu.edu.vn
1
linear space, i.e. d(y, z) = d(y + u, z + u),∀y, z, u ∈ Y . Let K : X ×Λ → 2X , G : X ×N → 2X
and F : X × X ×M → 2Y be multifunctions. Let C ⊆ Y be a closed subset with nonempty
interior. As usual, a problem involving single-valued mappings is split into many generalized
problems, while the mappings become multivalued. For the sake of simplicity we adopt the fol-
lowing notations. The Roman letters w, m, s are used for weak, middle, and strong problems.
For subsets A and B under consideration, we adopt the notations
(u, v)wA×B means ∀u ∈ A,∃v ∈ B,
(u, v)mA×B means ∃v ∈ B,∀u ∈ A,
(u, v) sA×B means ∀u ∈ A,∀v ∈ B,
α1(A,B) means A ∩B 6= ∅,
α2(A,B) means A ⊆ B,
(u, v) w¯A×B means ∃u ∈ A,∀v ∈ B and similarly for m¯, s¯, α¯1 and α¯2.
Let r ∈ {w,m, s}, α ∈ {α1, α2}. Our general parametric multivalued vector quasiequilibrium
problem is the following, for (λ, µ, η) ∈ Λ×M ×N ,
(Prα) find x¯ ∈ clK(x¯, λ), such that (y, x¯∗) rK(x¯, λ)×G(x¯, η),
α
(
F (x¯∗, y, µ), Y \ −intC).
Note that (Prα) represents six problems. This statement is not explicit but helps to shorten
the presentation of results. Let Srα(λ, µ, η) be the solution set of (Prα) corresponding to λ, µ
and η.
Recall first some notions. Let X and Y be as above and G : X → 2Y be a multifunction.
G is said to be lower semicontinuous (lsc) at x0 if G(x0) ∩ U 6= ∅ for some open set U ⊆ Y
implies the existence of a neighborhood N of x0 such that, for all x ∈ N,G(x) ∩ U 6= ∅. An
equivalent formulation is that: G is lsc at x0 if ∀xα → x0, ∀y ∈ G(x0),∃yα ∈ G(xα), yα → y. G
is called Hausdorff lower semicontinuous (Hlsc) at x0 if for each neighborhood B of the origin
in Y , there is a neighborhood N of x0 such that, for all x ∈ N , G(x0) ⊆ G(x) + B. G is
called upper semicontinuous (usc) at x0 if for each open set U ⊇ G(x0), there is a neighborhood
N of x0 such that U ⊇ G(N). G is termed Hausdorff upper semicontinuous (Husc) at x0
if for each neighborhood B of the origin in Y , there is a neighborhood N of x0 such that
G(N) ⊆ G(x0) + B. G is said to be continuous at x0 if it is both lsc and usc at x0 and to be
H-continuous at x0 if it is both Hlsc and H-usc at x0. G is called closed at x0 if for each net
(xα, yα) ∈ graphG := {(x, y) | y ∈ G(x)}, (xα, yα) → (x0, y0), y0 must belong to G(x0). The
closedness is closely related to the upper (and Hausdorff upper) semicontinuity (see e.g. [3],
Preposition 3.1). We say that G satisfies a certain property in a subset A ⊆ X if G satisfies it at
every points of A. If A = domG := {x ∈ X : G(x) 6= ∅} we omit “in domG” in the statement.
We propose the following two definitions of ε-solutions. Let us use the notations
comp(−intC)ε1 =
{
y ∈ Y : d(y, Y \ −intC) ≤ ε},
comp(−intC)ε2 = (Y \ −intC) +BεY ,
where BεY =
{
y ∈ Y | d(0, y) ≤ ε} and the notation “comp(.)” is related to the word “comple-
ment”.
2
Definition 1.1
(a) x¯ ∈ X is called an ε−solution of type 1 of problem (Prα) if x¯ ∈ clK(x¯, λ) and
(y, x¯∗)rK(x¯, λ)×G(x¯, η),
α
(
F (x¯∗, y, µ), comp(−intC)ε1
)
. (1.1)
(b) If (1.1) is replaced by
α
(
F (x¯∗, y, µ), comp(−intC)ε2
)
,
then x¯ is said to be an ε−solution of type 2 of problem (Prα).
Remark 1.1
(a) If x¯ is an ε−solution of type 2 then x¯ is an ε−solution of type 1. Indeed, it suffices
to show that
comp(−intC)ε2 ⊆ comp(−intC)ε1.
We check a more general fact that A + BεY ⊆ Aε := {y ∈ Y | d(y,A) ≤ ε}. Let
y ∈ A + BεY , i.e. y = a + z for some a ∈ A and z ∈ BεY . Then d(y, a) = d(y − a, 0) =
d(z, 0) ≤ ε, i.e. y ∈ Aε.
(b) If Y is finite dimensional, then the two types of ε−solutions coincide. Indeed, it
suffices to check that Aε ⊆ A+BεY , while A is closed. Let y ∈ Aε be arbitrary. Then,
for each n, there is an ∈ A with d(y, an) ≤ ε + 1n . We have some yn ∈ BεY with
d(y − an − yn, 0) = d(y − an, yn) ≤ 2n . By the compactness of BεY we can assume that
yn → y0 ∈ BεY . Let a = y− y0, un = y−an− y0 = a−an and vn = y−an− yn. Then,
0 ≤ d(an, a) = d(un, 0) ≤ d(un − vn, 0) + d(vn, 0) ≤ d(yn − y0, 0) + 2
n
.
Therefore, an → a and hence a ∈ A. Thus y = a+ y0 ∈ A+BεY .
The following two examples show that if Y is infinite dimensional, then in general an
ε−solution of type 1 is not guaranteed to be an ε−solution type 2.
Example 1.1. Let Y = l∞, A = {xk}, where xk1 = 1 + 1k , xkk+1 = 1 and xkj = 0, ∀j 6= 1
and j 6= k + 1. Then ‖xk − xl‖ = 1 if k 6= l and hence A is closed. Taking yk ∈ B1Y with
yk1 = −1, ykk+1 = −1 and ykj = 0 if j 6= 1 and j 6= k+1, we have xk + yk = ( 1k , 0, 0, ...) ∈ A+B1Y
and xk + yk → 0 /∈ A+B1Y . Thus, A+B1Y is not closed and hence A+B1Y is included properly
in A1.
The next example is more complicated but provides a case where A has the form Y \−intC
with C 6= Y as in Definition 1.1.
Example 1.2. Let Y = l∞. Let
U1 =
{
u ∈ l∞
∣∣∣ 1
2
≤ u1 ≤ 1, 2 ≤ u2 ≤ 3 and 0 ≤ uk ≤ 1 for k ≥ 3
}
,
U2 =
{
u ∈ l∞
∣∣∣ 1
4
≤ u1 ≤ 12 , 0 ≤ u2 ≤ 1, 2 ≤ u3 ≤ 3 and
0 ≤ uk ≤ 1 for k ≥ 4
}
,
· · ·
Un =
{
u ∈ l∞
∣∣∣ 1
2n
≤ u1 ≤ 12n−1 , 2 ≤ un+1 ≤ 3 and
0 ≤ uk ≤ 1 for k 6= 1, k 6= n+ 1
}
.
3
Then clearly Un are closed for all n and d(Un, Um) = 1 if n 6= m. Let A =
⋃∞
n=1 Un. We claim
that A is closed. Indeed, assume that ai ∈ A, ai → a0. Since d(Un, Um) = 1 for n 6= m, there
exist i0 and n0 such that ai ∈ Un0 for all i ≥ i0. Hence a0 ∈ Un0 ⊆ A. To see that A + B1Y is
not closed take sequences ak ∈ A and bk ∈ B1Y with components
ak1 =
1
2k
, akk+1 = 2 and a
k
j = 1,∀j 6= 1,∀j 6= k + 1,
bk1 = −1, bkk+1 = −1 and bkj = 0,∀j 6= 1,∀j 6= k + 1.
Then ak + bk = (−1 + 1
2k
, 1, 1, ...) → (−1, 1, 1, ...) /∈ A + B1Y . Thus A + B1Y is not closed and
contained properly in A1. Now taking −C = cl(l∞ \A) we get C 6= l∞ and A = l∞ \ −intC as
wanted.
The following example illustrates how the ε-solution may depend on ε and in particular
may be empty if ε is too small.
Example 1.3. For ε > 0, n ∈ N . Let X = Y = R,Λ ≡ M ≡ N = [0, 1], C = R+ and for
λ = 1n , K(x, λ) = [εn, εn+1], G(x, λ) = {x}, F (x, y, λ) = y−x− 1n . In this case the six cases of
(Prα) coincide since F and G are single-valued and the two types of the approximate solution
sets are equivalent. It is easy to see that for each λ = 1n , S
ε(λ) = [εn, εn+ ε− 1n ], if ε ≥ 1n and
Sε(λ) = ∅, if ε < 1n .
Definition 1.2
(a) The set of ε−solutions of type 1 (of type 2) of problem (Prα) at (λ, µ, η) is denoted
by Sε1rα(λ, µ, η)
(
Sε2rα(λ, µ, η), respectively
)
.
(b) Another kind of ε−solution sets is defined by
S˜εkrα(λ, µ, η) =
{
Srα(λ0, µ0, η0) if (λ, µ, η) = (λ0, µ0, η0),
Sεkrα(λ, µ, η) otherwise,
where (λ0, µ0, η0) is the point under consideration and k = 1, 2.
We propose the following weak semicontinuity.
Definition 1.3. Let X be a topological space and Y be a topological vector space, and C ⊆ Y
with intC 6= ∅ and C 6= Y .
(a) A multifunction H : X −→ 2Y is said to be C−upper semicontinuous (C−usc) at
x0 if, for any xα → x0, H(x0) ⊆ intC ⇒ ∃α¯,H(xα¯) ⊆ intC.
(b) H is called C−lower semicontinuous (C−lsc) at x0 if, for all xα → x0, H(x0)∩intC 6=
∅ ⇒ ∃α¯,H(xα¯) ∩ intC 6= ∅.
H is said to be C−continuous at x0, if H is both C−usc and C−lsc at x0.
(c) H is termed C− Hausdorff upper semicontinuous (C−Husc) at x0 if, for any xα → x0
and B (open neighborhood of 0 in Y ), H(x0) +B ⊆ intC ⇒ ∃α¯,H(xα¯) ⊆ intC.
Remark 1.2
(a) C−upper semicontinuity property is strictly stronger than C−Hausdorff upper semi-
continuity property;
(b) H is C−usc, C−lsc, C−Husc at x0 if and only if H is intC−usc, intC−lsc, intC−Husc,
respectively;
(c) H is usc at x0, iff H is C−usc at x0 for all subsets C of Y ;
(d) H is lsc at x0, iff H is C−lsc at x0 for all subsets C of Y .
4
Since the solution existence of quasiequilibrium problems has been intensively investigated, we
do not study this issue, and instead always assume the existence.
The organization of the paper is as follows. In Section 2 we give sufficient conditions for
the approximate solution sets to be lsc and Hlsc at the point in question. In Section 3 we
investigate sufficient conditions for the approximate solution sets to be usc, Husc and closed.
Section 4 is devoted to applications in approximate quasivariational inequalities, approximate
fixed points and approximate quasioptimization problems.
2. LOWER SEMICONTINUITY OF THE ε−SOLUTION SETS
Considering the lower semicontinuity of approximate solution sets for our quasiequilibrium
problem we will see below that the sufficient conditions for this lower semicontinuity are the
same for the two types of ε−solutions stated in Definition 1.1. The reason is that an ε−solution
of type 1, which is not an ε−solution of type 2, much lie on the boundary of comp(−int)Cε1 ,
since this set may have more points than comp(−int)Cε2 only in the boundary (see Remark 1.1
and Example 1.2). This difference does not affect the lower semicontinuity. But it affects the
upper semicontinuity as the next section makes it clear. Moreover, we will see that under usual
assumptions lower semicontinuity holds only for the ε−solution sets of the second kind S˜ε1rα and
S˜ε2rα (see Definition 1.2) while upper semicontinuity occurs only for the ε−solution sets of the
first kind Sε1rα and S
ε2
rα.
In the sequel let, for λ ∈ Λ,
E(λ) =
{
x ∈ X | x ∈ clK(x, λ)}.
We always assume that Srα(λ, µ, η) 6= ∅ for all (λ, µ, η) in a neighborhood of the considered
point (λ0, µ0, η0). In the sequel we will write Srα for the multifunction Srα(., ., .) and similarly
for other multifunctions.
Theorem 2.1. Assume that K is usc and has compact values in X × {λ0}, E is lsc at λ0 and
F is comp(−intC)εk− lsc in clK(X,Λ)× clK(X,Λ)× {µ0}, k = 1, 2.
(i) For r ∈ {w,m}, if G is lsc in clK(X,Λ)×{η0}, then S˜εkrα1 is lsc at (λ0, µ0, η0) for ε > 0.
(ii) If G is usc and compact-valued in clK(X,Λ) × {η0}, then S˜εksα1 is lsc at (λ0, µ0, η0) for
ε > 0.
Proof. Since the six assertions for the two types of ε - solutions can be proved by similar
techniques we present a proof only for S˜ε1wα1 of problem (Pwα1). Let ε > 0 be fixed. Suppose
that S˜ε1wα1 is not lsc at (λ0, µ0, η0), i.e., ∃x0 ∈ S˜ε1wα1(λ0, µ0, η0), ∃(λγ , µγ , ηγ) → (λ0, µ0, η0),
∀xγ ∈ S˜ε1wα1(λγ , µγ , ηγ), xγ 6→ x0. Since x0 ∈ S˜ε1wα1(λ0, µ0, η0) = Swα1(λ0, µ0, η0), ∀y0 ∈
K(x0, λ0), ∃x∗0 ∈ G(x0, η0),
F (x∗0, y0, µ0)∩ (Y \− intC) 6= ∅. (2.1)
By the lower semicontinuity of E at λ0, there is a net x¯γ ∈ E(λγ), x¯γ → x0. By the
contradiction assumption, there must be a subnet x¯β such that, ∀β, x¯β /∈ S˜ε1wα1(λβ , µβ , ηβ), i.e.,
∃yβ ∈ K(x¯β , λβ), ∀x¯∗β ∈ G(x¯β , ηβ),
F (x¯∗β , yβ , µβ)∩comp(−intC)ε1 = ∅. (2.2)
Since K is usc and K(x0, λ0) is compact, one can assume that yβ → y0, for some
y0 ∈ K(x0, λ0). By the lower semicontinuity of G at (x0, η0), there exists x¯∗β ∈ G(x¯β , ηβ),
x¯∗β → x∗0. In view of the comp(−intC)ε1− lower semicontinuity of F at (x∗0, y0, µ0) and of the
inclusion Y \ intC ⊆ int(comp(−intC)ε1), we see a contradiction between (2.1) and (2.2).
5
The following example shows that Theorem 2.1 is no longer true if we replace the
ε−solution sets of the second kind by that of the first kind.
Example 2.1. Let X = Y = R,Λ ≡ M ≡ N = [0, 1], C = R+, K(λ) = [0, 1], G(x, λ) = {x},
F (x, y, λ) = {y − x − ε − λ, x − y} and λ0 = 0. Since G is single-valued, (Pwα1), (Pmα1) and
(Psα1) coincide. From now on if Y is finite dimensional (then the two types of ε−solutions
coincide), instead of writing ε1 or ε2 in the index of ε−solution sets, we write simply ε, e.g.
Sεrα, S˜
ε
rα. It is easy to see that the assumptions of Theorem 2.1 are fulfilled and hence S˜εrα1
is lsc at 0 for r ∈ {w,m, s}. In fact Srα1(0) = {1} and Sεrα1(λ) = [1 − ε, 1], ∀λ ∈ (0, 1], but
Sεrα1(0) = {0, 1}. So Sεwα1 is not lsc at 0, for r ∈ {w,m, s}.
Theorem 2.2. Assume that K is usc and has compact values in clK(X,Λ)×{λ0}, E is lsc at
λ0 and F is comp(−intC)εk− usc in clK(X,Λ)× clK(X, Λ)× {µ0}, k = 1, 2.
(i) For r ∈ {w,m}, if G is lsc in clK(X,Λ)×{η0}, then S˜εkrα2 is lsc at (λ0, µ0, η0) for ε > 0.
(ii) If G is usc and has compact values in clK(X,Λ) × {η0}, then S˜εksα2 is lsc at (λ0, µ0, η0)
for ε > 0.
Proof. As an example we demonstrate only for problem (Psα2). Let ε > 0 be fixed. Suppose
to the contrary that S˜ε2sα2 is not lsc at (λ0, µ0, η0), i.e., ∃x0 ∈ S˜ε2sα2(λ0, µ0, η0), ∃λγ → λ0,∃µγ →
µ0,∃ηγ → η0, ∀xγ ∈ S˜ε2sα2(λγ , µγ , ηγ), xγ 6→ x0. Since x0 ∈ S˜ε2sα2(λ0, µ0, η0) = Ssα2(λ0, µ0, η0),
∀y0 ∈ K(x0, λ0), ∀x∗0 ∈ G(x0, η0),
F (x∗0, y0, µ0) ⊆ Y \− intC. (2.3)
By the lower semicontinuity of E at λ0, there is a net x¯γ ∈ E(λγ), x¯γ → x0. By virtue of
the contradiction assumption, there must be a subnet x¯β such that, ∀β, x¯β /∈ S˜ε2sα2(λβ , µβ , ηβ),
i.e., ∃yβ ∈ K(x¯β , λβ), ∃x¯∗β ∈ G(x¯β , ηβ),
F (x¯∗β , yβ , µβ) 6⊆ comp(−intC)ε2. (2.4)
Since K is usc and K(x0, λ0) is compact, one can assume that yβ → y0, for some
y0 ∈ K(x0, λ0). By the upper semicontinuity and the compactness of G at (x0, η0), one can
assume that x¯∗β → x∗0 for some x∗0 ∈ G(x0, η0). The comp(−intC)ε2− upper semicontinuity of F
at (x∗0, y0, µ0) and (2.3) together imply the existence of β¯ such that
F (x¯∗¯β , yβ¯ , µβ¯) ⊆ comp(−intC)ε2,
which contradicts (2.4).
The following two examples ensure that Theorem 2.2 is no longer true, for r ∈ {w,m, s}
and k = 1, 2, if we replace S˜εkrα2 by S
εk
rα2 .
Example 2.2. Let ε > 0 be fixed and small. Let X,Y,Λ,M,N and C be as in Example 2.1,
K(x, λ) = [λ, λ + 1], G(x, λ) = {−x + 1 − ε, x}, F (x, y, λ) = {x − 1 − λ} and λ0 = 0. Then,
it is clear that the assumption of Theorem 2.2 (i) holds. Direct calculations yield Swα2(0) =
Smα2(0) = {1}, Sεwα2(0) = Sεmα2(0) = {0}∪ [1−ε, 1] and Sεwα2(λ) = Sεmα2(λ) = [1+λ−ε, 1+λ],
∀λ ∈ (0, 1]. So, S˜εwα2 and S˜εmα2 are lsc at 0, while Sεwα2 and Sεmα2 are not.
Example 2.3. Let ε > 0 be fixed and small. Let X,Y,Λ,M,N and C be as in Example
2.1, K(x, λ) =
[
0, 1+ε+
√
(1+ε)2+4λ
2
]
, G(x, λ) = {x}, F (x, y, λ) = {x2 − (1 − ε)x − ε − λ} and
λ0 = 0. Then, it is not hard to see that the assumption (ii) of Theorem 2.2 is satisfied and
Ssα2(0) = {1 + ε}, Sεsα2(0) = {0, 1 + ε} and Sεsα2(λ) =
{
1+ε+
√
(1+ε)2+4λ
2
}
, ∀λ ∈ (0, 1]. Hence,
6
S˜εsα2 is lsc at 0, while S
ε
sα2 is not.
Passing to Hausdorff lower semicontinuity we have
Theorem 2.3. Let X be additionally a Hausdorff topological vector space. Assume that E is
lsc at λ0, E(λ0) is compact, K is usc and compact-valued in clK(X,Λ)× {λ0}, K(., λ0) is lsc
and closed in clK(X,Λ), F is comp(−intC)εk- lsc in clK(X,Λ) × clK(X,Λ) × {µ0}, k = 1, 2,
and F (., ., µ0) is −C -usc in clK(X,Λ)× clK(X,Λ).
(i) For r ∈ {w,m}, if G is lsc in clK(X,Λ)×{η0} and G(., η0) is usc and has compact values
in clK(X,Λ), then S˜εkrα1 is Hlsc at (λ0, µ0, η0) for ε > 0.
(ii) If G is usc and have compact values in clK(X,Λ)×{η0} and G(., η0) is lsc in clK(X,Λ),
then S˜εksα1 is Hlsc at (λ0, µ0, η0) for ε > 0.
Proof. We demonstrate only for S˜ε1wα1 . We first show that Swα1(λ0, µ0, η0) is compact. Let
xγ ∈ Swα1(λ0, µ0, η0), xγ → x0. If x0 /∈ Swα1(λ0, µ0, η0), then there exists y0 ∈ K(x0, λ0) such
that, ∀x∗0 ∈ G(x0, η0),
F (x∗0, y0, µ0) ⊆ −intC. (2.5)
Since K(., λ0) is lsc in clK(X,Λ), there is yγ ∈ K(xγ , λ0) such that yγ → y0. As xγ ∈
Swα1(λ0, µ0, η0), there exists x
∗
γ ∈ G(xγ , η0) with
F (x∗γ , yγ , µ0) 6⊆ −intC. (2.6)
Since G(., η0) is usc and G(x0, η0) is compact, we can assume that x∗γ → x∗0, for some x∗0 ∈
G(x0, η0). By the −C−upper semicontinuity of F (., ., µ0) in clK(X,Λ) × clK(X,Λ), we see a
contradiction between (2.5) and (2.6). Hence, Swα1(λ0, µ0, η0) is closed and then compact, as
E(λ0) is compact.
We show next that ∀(λγ , µγ , ηγ) → (λ0, µ0, η0), ∀x¯0 ∈ Swα1(λ0, µ0, η0), ∃x¯γ ∈ S˜ε1wα1(λγ ,
µγ , ηγ), x¯γ → x¯0. Suppose to the contrary that there exist (λγ , µγ , ηγ) → (λ0, µ0, η0) and
x¯0 ∈ S˜ε1wα1(λ0, µ0, η0) such that ∀xγ ∈ S˜ε1wα1(λγ , µγ , ηγ), xγ 6→ x¯0. Since E is lsc at λ0,
there is x¯γ ∈ E(λγ), x¯γ → x¯0. By the contradiction assumption, there exists a subnet
x¯β /∈ S˜ε1wα1(λβ , µβ , ηβ),∀β. The further argument to see a contradiction is similar to that
of Theorem 2.1.
Now suppose that S˜ε1wα1 is not Hlsc at (λ0, µ0, η0), i.e. ∃B (a neighborhood of the origin
in X), ∃(λγ , µγ , ηγ) → (λ0, µ0, η0) such that ∀γ,∃x0γ ∈ S˜ε1wα1(λ0, µ0, η0) = Swα1(λ0, µ0, η0),
x0γ /∈ S˜ε1wα1(λγ , µγ , ηγ) + B. Since Swα1(λ0, µ0, η0) is compact, we assume that x0γ → x0 for
some x0 ∈ Swα1(λ0, µ0, η0). So there are γ1, a neighborhood B1 of 0 in X with B1 + B1 ⊆ B
and bγ ∈ B1 such that, ∀γ ≥ γ1, x0γ = x0 + bγ . By the preceding part of the proof there is
x˜γ ∈ S˜ε1wα1(λγ , µγ , ηγ), x˜γ → x0 and hence, one can assume that there is γ2,∀γ ≥ γ2,
x˜γ ∈ x0 −B1,
i.e., there exists b′γ ∈ B1, x˜γ = x0 − b′γ . Hence ∀γ ≥ γ0 = max{γ1, γ2},
x0γ = x0 + bγ = x˜γ + b′γ + bγ ∈ x˜γ +B.
This is impossible due to the fact that x0γ /∈ S˜ε1wα1(λγ , µγ , ηγ) + B. Thus, S˜ε1wα1 is Hlsc at
(λ0, µ0, η0).
7
The following example shows that the compactness of E(λ0) is essential.
Example 2.4. Let ε > 0 be fixed and small. Let X = R2, Y = R, Λ ≡ M ≡ N = [0, 1],
C = R+, K(x, λ) = {(x1, λx1)} (for x = (x1, x2)), G(x, λ) = {x}, λ0 = 0 and
F (x, y, λ) =
{
{1} if λ = 0,
{λ− ε2} otherwise.
Then our six problems coincide and E(λ) = {x ∈ R2 | x2 = λx1}. So the assumptions of
Theorem 2.3, but the compactness of E(λ0), are satisfied. However, S˜εrα(λ) = {x ∈ R2 | x2 =
λx1} is not Hlsc at 0 (although S˜εrα is lsc at 0). The reason is that E(0) = {x ∈ R2 | x2 = 0}
is not compact (but E and K are continuous and closed).
Theorem 2.4. Assume that E is lsc at λ0, E(λ0) is compact, K is usc and compact-valued
in clK(X,Λ) × {λ0}, K(., λ0) is lsc and closed in clK(X,Λ), F is comp(−intC)εk - usc in
clK(X,Λ)× clK(X,Λ)× {µ0}, k = 1, 2, and F (., ., µ0) is −C -lsc in clK(X,Λ)× clK(X,Λ).
(i) For r ∈ {w,m}, if G is lsc in clK(X,Λ)×{η0} and G(., η0) is usc and has compact values
in clK(X,Λ), then S˜εkrα2 is Hlsc at (λ0, µ0, η0) for ε > 0.
(ii) If G is usc and have compact values in clK(X,Λ)×{η0} and G(., η0) is lsc in clK(X,Λ),
then S˜εksα2 is Hlsc at (λ0, µ0, η0) for ε > 0.
We omit the proof since it is similar to that of Theorem 2.3.
Example 2.4 shows also that the compactness of E(λ0) in Theorem 2.4 cannot be dropped.
Moreover, Examples 2.1 - 2.3 also present that Theorems 2.3 and 2.4 are no longer true if we
replace S˜εkrα by S
εk
rα (for r ∈ {w,m, s} and k = 1, 2.)
3. UPPER SEMICONTINUITY OF THE ε−SOLUTION SETS
In this section we consider upper semicontinuity, Hausdorff upper semicontinuity and
closedness. We will establish sufficient conditions for the ε−solutions of the two types to have
these upper semicontinuity properties. We will also see that these results hold only for the
ε−solution sets Sεkrα not for S˜εkrα , r ∈ {w,m, s}, α ∈ {α1, α2} and k = 1, 2.
Theorem 3.1. Assume that K is lsc in clK(X,Λ)× {λ0}, E is usc, E(λ0) is compact and F
is Y \comp(−intC)εk−usc in clK(X,Λ)× clK(X,Λ)× {µ0}, k = 1, 2.
(i) For r ∈ {w,m}, if G is usc and has compact values in clK(X,Λ) × {η0}, then Sεkrα1 is
both usc and closed at (λ0, µ0, η0) for ε ≥ 0.
(ii) If G is lsc in clK(X,Λ)×{η0}, then Sε1sα1 is both usc and closed at (λ0, µ0, η0) for ε ≥ 0.
Proof. By the similarity we present only a proof for problem (Psα1) and k = 1. Let
ε ≥ 0 be fixed. Reasoning ab absurdo suppose the existence of an open neighborhood U
of Sε1sα1(λ0, µ0, η0), of nets λγ → λ0, µγ → µ0, ηγ → η0 and xγ ∈ Sε1sα1(λγ , µγ , ηγ) such that
xγ /∈ U,∀γ. By the upper semicontinuity of E and the compactness of E(λ0), we assume
that xγ → x0 for some x0 ∈ E(λ0). If x0 /∈ Sε1sα1(λ0, µ0, η0), there are y0 ∈ K(x0, λ0) and
x∗0 ∈ G(x0, η0) such that
F (x∗0, y0, µ0) ∩ comp(−intC)ε1 = ∅. (3.1)
Since K and G are lsc at (x0, λ0) and (x0, η0), respectively, there are nets yγ ∈ K(xγ , λγ)
and x∗γ ∈ G(xγ , ηγ) such that yγ → y0 and x∗γ → x∗0. Since xγ ∈ Sε1sα1(λγ , µγ , ηγ), one has
F (x∗γ , yγ , µγ) ∩ comp(−intC)ε1 6= ∅. (3.2)
8
By the closedness of comp(−intC)ε1 and the Y \ comp(−intC)ε1− upper semicontinuity of
F at (x∗0, y0, µ0) one has a contradiction between (3.1) and (3.2). Thus, x0 ∈ Sε1sα1(λ0, µ0, η0),
which is again a contradiction, since xγ /∈ U,∀γ.
The closedness of Sε1sα1 can be proved similarly.
Theorem 3.1 is no longer true if we replace Sεkrα1 by S˜
εk
rα1 , r ∈ {w,m, s}, respectively, as
shown by the example below.
Example 3.1. Let ε > 0 be fixed and small. Let X = Y = R, Λ ≡ M ≡ N = [0, 1], C = R+,
K(x, λ) = [0, 1], G(x, λ) = {x}, F (x, y, λ) = {y − x+ λ} and λ0 = 0. Since G is single-valued,
(Pwα1), (Pmα1) and (Psα1) coincide. It is easy to see that the conditions of Theorem 3.1 hold
and accordingly, Sεkrα1 is usc at 0 (for all r ∈ {w,m, s}). In fact Srα1(0) = {0}, Sεkrα1(0) = [0, ε]
and Sεkrα1(λ) = [0, ε+ λ]. Thus, S˜
εk
rα1 is not usc at 0.
Theorem 3.2. Assume that K is lsc in clK(X,Λ)×{λ0}, E is Husc and compact-valued at λ0,
and that F is Y \ comp(−intC)εk - Husc in clK(X,Λ)× clK(X,Λ)×{µ0}. Assume further that
∀BX (open neighborhood of 0 in X), ∀x /∈ Sε1rα1(λ0, µ0, η0) + BX , ∃ρ > 0, (y, x∗) r¯K(x, λ0) ×
G(x, η0), α¯1
(
F (x∗, y, µ0)+intB
ρ
Y , comp(−intC)ε1
)
(remember that BρY = {y ∈ Y | d(0, y) ≤ ρ}).
(i) For r ∈ {w,m}, if G is Husc and compact-valued in clK(X,Λ)×{η0}, then Sεkrα1 is Husc
at (λ0, µ0, η0) for ε ≥ 0.
(ii) If G is lsc in clK(X,Λ)× {η0}, then Sεksα1 is Husc at (λ0, µ0, η0) for ε ≥ 0.
Proof. We demonstrate only for Sε1wα1 . Let ε ≥ 0 be fixed. Suppose that ∃BX (open
neighborhood of 0 in X), ∃(λγ , µγ , ηγ) → (λ0, µ0, η0), ∃xγ ∈ Sε1wα1(λγ , µγ , ηγ) such that xγ /∈
Sε1wα1(λ0, µ0, η0)+BX ,∀γ. By the Hausdorff upper semicontinuity of E and the compactness of
E(λ0), we can assume that xγ → x0 for some x0 ∈ E(λ0). If x0 /∈ Sε1wα1(λ0, µ0, η0) +BX , there
is some ρ > 0 and some y0 ∈ K(x0, λ0) such that, ∀x∗0 ∈ G(x0, η0),
[F (x∗0, y0, µ0) + intB
ρ
Y ]∩ comp(−intC)ε1 = ∅. (3.3)
Since K is lsc at (x0, λ0), there is yγ ∈ K(xγ , λγ), yγ → y0. As xγ ∈ Sε1wα1(λγ , µγ , ηγ),
∃x∗γ ∈ G(xγ , ηγ),
F (x∗γ , yγ , µγ) ∩ comp(−intC)ε1 6= ∅. (3.4)
As G is H-usc and G(x0, η0) is compact, one has a subnet x∗β such that x
∗
β → x∗0 for some
x∗0 ∈ G(x0, η0). By the Y \comp(−intC)ε1− Hausdorff upper semicontinuity of F at (x∗0, y0, µ0),
we see a contradiction between (3.3) and (3.4). Hence, x0 ∈ Sε1wα1(λ0, µ0, η0) + BX , which is
another contradiction, since xβ /∈ Sε1wα1(λ0, µ0, η0) +BX ,∀β.
We see that to ensure the Hausdorff upper semicontinuity of Sεrα1 , r ∈ {w,m, s}, the
upper semicontinuity assumed in Theorem 3.2 is reduced to Hausdorff upper semicontinuity.
However, we have to add some assumption in Theorem 3.2. The following example shows that
this additional assumption is essential.
Example 3.2. Let ε = 0, X = Y = R, Λ ≡ M ≡ N = [0, 1], C = R+, K(x, λ) = [0, 1],
G(x, λ) = {x}, F (x, y, λ) = (−∞, λx) and λ0 = 0. As G is single-valued, (Pwα1), (Pmα1) and
(Psα1) coincide. It is clear that S
01
rα1(0) = {0}, S01rα1(λ) = [0, 1], ∀r ∈ {w,m, s}, ∀λ ∈ (0, 1]. So,
S01rα1 is not Husc at 0. The reason is that the additional assumption is violated. Indeed, take
BX = (−1, 1) and x = 1. Then, for each ρ > 0 and each y ∈ [0, 1], one has F (1, y, 0)+ intBρX =
(−∞, ρ). So, [F (1, y, 0) + intBρX ] ∩ [−ε,+∞) 6= ∅.
Theorem 3.3. Assume that K is lsc in clK(X,Λ)× {λ0}, E is usc, E(λ0) is compact and F
is Y \comp(−intC)εk−lsc in clK(X,Λ)× clK(X,Λ)× {µ0}, k = 1, 2.
9
(i) For r ∈ {w,m}, if G is usc and compact-valued in clK(X,Λ) × {η0}, then Sεkrα2 is both
usc and closed at (λ0, µ0, η0) for ε ≥ 0.
(ii) If G is lsc in clK(X,Λ)×{η0}, then Sεksα2 is both usc and closed at (λ0, µ0, η0) for ε ≥ 0.
We omit the proof since it is similar to the previous ones.
Example 3.1 ensures also that Theorem 3.3 is no longer true if we replace Sε1rα2 by S˜
ε1
rα2 ,
since F and G are single-valued.
4. APPLICATIONS
Quasiequilibrium problems include as special cases many important problems such as qua-
sivariational inequalities, complementarity problems, fixed-point and coincidence-point prob-
lems, optimization problems, etc. Therefore, applying the results presented in the preceding
sections we obviously obtain sufficient conditions for semicontinuity of approximate solution
sets of these particular cases. In this section we derive some consequences of the theorems of
Sections 2 and 3 as examples.
4.1. Quasivariational inequalities
If Y = R, F (x, y, µ) = 〈T (x, µ), y − g(x, µ)〉 and G(x, η) = {x}, where T : X ×M → 2X∗
and g : X × M → X is a continuous mapping, then (Prα1) coincides with (QVI) in [15, 16],
and (Prα2) becomes (SQVI) in [15, 16], r ∈ {w,m, s}. We state three direct consequences of
Theorems 2.1, 3.1 and 3.3 as examples.
Corollary 4.1. The ε−solution set S˜ε1 of (QVI) is lsc at (λ0, µ0) for ε > 0, if K is usc and
compact-valued in clK(X,Λ) × {λ0}, E is lsc at λ0 and (x, y, µ) 7→ 〈T (x, µ), y − g(x, µ)〉 is
[−ε,+∞)−lsc in clK(X,Λ)× clK(X,Λ)× {µ0}.
Corollary 4.2. Let K be lsc in clK(X,Λ)× {λ0}, E be usc and E(λ0) be compact. Let (x, y,
µ) 7→ 〈T (x, µ), y − g(x, µ)〉 be (−∞,−ε)−usc in clK(X,Λ) × clK(X,Λ) × {µ0}. Then Sε1 of
(QVI) is both usc and closed at (λ0, µ0) for ε ≥ 0.
Corollary 4.3. The ε−solution set Sε2 of (SQVI) is both usc and closed at (λ0, µ0) for ε ≥ 0,
if K and E are as in Corollary 4.2, (x, y, µ) 7→ 〈T (x, µ), y − g(x, µ)〉 is (−∞,−ε)−lsc in
clK(X,Λ)× clK(X,Λ)× {µ0}.
These three corollaries improve Theorems 5.1, 6.1 and 6.3, respectively, in [16]. Now we
give an example, where the assumptions of our corollaries are strictly weaker than the corre-
sponding ones imposed in the mentioned paper.
Example 4.1. Let ε > 0 be fixed. Let X = Y = R, Λ ≡ M = [0, 1], C = R+, K(x, λ) = [0, 1],
λ0 = 0 and
T (x, λ) =
{
{ε} if λ = 0,
[ ε4 ,
ε
2 ] otherwise.
Then 〈T (x, 0), y − g(x, 0)〉 = [−ε, ε] and 〈T (x, λ), y − g(x, λ)〉 = [− ε2 , ε2 ],∀λ ∈ (0, 1]. Hence
all the assumptions of Corollaries 4.1 and 4.2 are satisfied. Applying these corollaries we
know that S˜ε1 is lsc at 0 and S
ε
1 is both usc and closed at 0. In fact, direct calculations give
S˜ε1(λ) = S
ε
1(λ) = [0, 1],∀λ ∈ Λ. However, 〈T (., .), .〉 is not lsc and T (., .) is not usc in X×X×{0}
as required in assumption (ii) of Theorems 5.1 and 6.1 of [16].
Now we pass to quasivariational inequalities with operator solutions introduced in [13].
10
Let X,M,N,Λ and Y be as defined in Section 1. Let C ⊆ Y be closed convex cone with
intC 6= ∅ and C 6= Y . Let K : L(X,Y ) × Λ → 2L(X,Y ), T : L(X,Y ) × L(X,Y ) ×M → 2X be
multifunctions. Our quasivariational inequalities with operator solutions are
(QVI’) Find f¯ ∈ clK(f¯ , λ) such that, for each f ∈ K(f¯ , λ),(
f − f¯ , T (f¯ , µ)) ∩ (Y \ −intC) 6= ∅;
(SQVI’) Find f¯ ∈ clK(f¯ , λ) such that, for each f ∈ K(f¯ , λ),(
f − f¯ , T (f¯ , µ)) ⊆ Y \ −intC.
We have not observed stability results for these problems in the literature. As examples
we state now three consequences, which are easily derived from Theorems 2.3, 2.4 and 3.2.
Corollary 4.4. Assume for (QVI’) that K is usc and compact-valued in clK(L(X, Y ), Λ) ×
{λ0}, K(., λ0) is lsc and closed in clK(L(X,Y ), Λ), E is lsc at λ0 and E(λ0) is compact,
(f, g, µ) 7→ (f − g, T (g, µ)) is comp(−intC)ε1-lsc or comp(−intC)ε2-lsc in clK(L(X,Y ), Λ) ×
clK(L(X,Y ), Λ)×{µ0}, and (f, g) 7→
(
f−g, T (x, µ0)
)
is −C-usc in clK(L(X,Y ), Λ)×clK(L(X,
Y ), Λ). Then the ε−solution set S˜ε1 is Hlsc at (λ0, µ0) for ε > 0.
Corollary 4.5. Assume that K and E are as in Corollary 4.4 and (f, g, µ) 7→ (f − g, T (g, µ))
is comp(−intC)ε1−usc or comp(−intC)ε2−usc in clK(L(X,Y ), Λ) × clK(L(X,Y ), Λ) × {µ0}.
Assume further that E(λ0) is compact, K(., λ0) is lsc and closed in clK(L(X,Y ), Λ) and
(f, g) 7→ (f−g, T (x, µ0)) is −C-lsc in clK(L(X,Y ), Λ)×clK(L(X,Y ), Λ). Then the ε−solution
set S˜ε2 of (SQVI’) is Hlsc at (λ0, µ0) for ε > 0.
Corollary 4.6. Let K be lsc in clK(L(X,Y ), Λ)×{λ0}, E be Husc and E(λ0) be compact and
(f, g, µ) 7→ (f − g, T (g, µ)) be Y \comp(−intC)ε1− Husc in clK(L(X,Y ), Λ)×clK(L(X,Y ), Λ)×
{µ0}. If ∀B (open neighborhood of 0 in L(X,Y )), ∀f0 /∈ Sε1(λ0, µ0) +B, ∃BY (open neighbor-
hood of 0 in Y ), ∃f ∈ K(f0, λ0),
(
(f − f0, T (f0, µ0)) + BY
) ∩ comp(−intC)ε1 = ∅, then Sε1 of
(QVI’) is Husc at (λ0, µ0) for ε ≥ 0.
4.2. Approximate fixed points
Let X be a Hilbert space, Λ and M be as in Section 1, ε > 0, K : Λ → 2X and
T : X ×M → 2X . The problem of finding ε - fixed points of T at (λ, µ) is defined as (see, e.g.
a resent paper [16])
(Pε) Find x¯ ∈ K(λ) such that
x¯ ∈ T ε(x¯, µ) := {x ∈ X | d(x, T (x¯, µ)) ≤ ε}.
Assume that for each λ ∈ Λ, K(λ) is closed and contains 0. The problem of finding
ε−fixed points is related to the following problem of finding ε−solutions of a quasiequilibrium
problem:
(QEPε) Find x¯ ∈ K(λ) such that ∃t¯ ∈ clT (x¯, µ), ∀y ∈ K(λ),
〈x¯− t¯, y + t¯ − x¯〉 ≥ −ε. (4.1)
Proposition 4.1. If x¯ is a solution of (QEPε) then x¯ is a solution of (P
√
ε).
Proof. By the assumption we have (4.1). Taking y = 0 we get ‖x¯ − t¯‖2 ≤ ε and hence
d
(
x¯, T (x¯, µ)
) ≤ √ε.
Corollary 4.7. If K is lsc at λ0 and T is lsc in K(Λ) × {µ0}, then the ε−solution set S˜ε of
(QEPε) is lsc at (λ0, µ0) for ε > 0.
11
Proof. By the lower semicontinuity of T , (x, y, µ) 7→ 〈x − clT (x, µ), y + clT (x, µ) − x〉 is lsc
too. Applying Theorem 2.1 yields the corollary.
The following two results are easily derived from Theorems 2.3 and 3.1. Corollary 4.8.
Assume the assumptions of Corollary 4.7. Assume further that K(λ0) is compact and T (., µ0) is
usc and has compact values in K(Λ). Then the ε−solution set S˜ε of (QEPε) is Hlsc at (λ0, µ0)
for ε > 0.
Corollary 4.9. If K is usc, K(λ0) is compact and T is usc and has compact values in K(Λ)×
{µ0}, then the ε−solution set Sε of (QEPε) is usc and closed at (λ0, µ0) for ε ≥ 0.
By Proposition 4.1, Corollary 4.5 yields the semicontinuity with respect to (λ, µ) of the
part of the
√
ε−fixed points of T (x, µ). To deal with the whole set of ε - fixed points we modify
problem (QEPε) as follows.
(QEPε1) Find x¯ ∈ K(λ) such that ∃t¯ ∈ T ε(x¯, µ), ∀y ∈ K(λ),
〈x¯ − t¯, y + t¯ − x¯〉 ≥ 0. (4.2)
Proposition 4.2. x¯ is a solution of (Pε) if and only if x¯ is a solution of (QEPε1).
Proof. Being a solution of (QEPε1), x¯ yields t¯ ∈ T ε(x¯, µ) satisfying (4.2). Taking y = 0 we
see that ‖x¯− t¯‖ = 0, and hence x¯ ∈ T ε(x¯, µ). Conversely, if x¯ is an ε−fixed point of T (., µ), i.e.
x¯ ∈ T ε(x¯, µ). Taking t¯ = x¯ we see that x¯ satisfies (4.2).
Let us denote the fixed-point set and the ε - fixed-point set of T at (λ, µ) by P (λ, µ) and
P ε(λ, µ), respectively. Similarly as for quasiequilibrium problems we consider also the following
second kind of ε - fixed-point set
P˜ ε(λ, µ) =
{
P (λ0, µ0) if (λ, µ) = (λ0, µ0),
P ε(λ, µ) otherwise.
Proposition 4.3. If K is lsc at λ0 and T is lsc in K(Λ)× {µ0}, then P˜ ε is lsc at (λ0, µ0) for
ε > 0.
Proof. Suppose to the contrary that there are λγ → λ0, µγ → µ0 and x0 ∈ P˜ ε(λ0, µ0)
such that, ∀xγ ∈ P˜ ε(λγ , µγ), xγ 6→ x0. As K is lsc at λ0, there exists x¯γ ∈ K(λγ) such that
x¯γ → x0. Then there must be a subnet x¯β with x¯β /∈ P˜ ε(λβ , µβ) for all β. Since x0 ∈ P˜ ε(λ0, µ0),
x0 ∈ T (x0, µ0). By the lower semicontinuity of T at (x0, µ0), there is tβ ∈ T (x¯β , µβ) such that
tβ → x0. Since, for all µβ 6= µ0, x¯β /∈ P˜ ε(λβ , µβ), x¯β /∈ T ε(x¯β , µ), i.e. ‖x¯β − tβ‖ ≥ ε. This is
impossible since x¯β → x0 and tβ → x0.
Proposition 4.4. If K is usc, K(λ0) is compact and T is usc and has compact values in
K(Λ)× {µ0}, then P ε is usc and closed at (λ0, µ0) for ε ≥ 0.
Proof. Arguing by contradiction suppose the existence of a neighborhood U of P ε(λ0, µ0), of
nets (λγ , µγ) → (λ0, µ0) and xγ ∈ P ε(λγ , µγ), xγ /∈ U,∀γ. By the assumption on K we have
a point x0 ∈ K(λ0) and a subnet xβ → x0. Since xβ ∈ P ε(λβ , µβ), there is tβ ∈ clT (xβ , µβ),
‖tβ − xβ‖ ≤ ε. By the assumption on T we can assume that tβ → t0, for some t0 ∈ T (x0, µ0).
Hence ‖x0 − t0‖ ≤ ε and then x0 ∈ P ε(λ0, µ0), a contradiction, since xβ /∈ U,∀β.
The closedness of P ε at (λ0, µ0) is similarly verified.
12
4.3. Approximate quasioptimization problems
Let X,Y,M,Λ,K and C be as in Section 1. Let T : X ×M → 2Y be a multifunction.
We consider the following quasioptimization problem, for (λ, µ) ∈ Λ×M ,
(QOP) Find x¯ ∈ clK(x¯, λ) such that
T (x¯, µ) ∩ wMin{T (K(x¯, λ), µ) ∣∣C} 6= ∅,
where wMin
{
H
∣∣C} denotes the set of all weakly efficient points y∗ of the set H ⊆ Y , with
respect to the ordering set C, i.e. points y∗ ∈ H such that, ∀y ∈ H, y − y∗ ∈ Y \ −intC. Note
that unlike for the usual weak efficiency here C needs not be a cone.
We will show now that (QOP) can be expressed as a case of problem (Pmα1). Set
X1 = X × Y and define K1 : X1 ×Λ → 2X1 , G1 : X1 ×M → 2X1 and F1 : X1 ×X1 ×M → 2Y
by, for x1 = (x1, y1), x2 = (x2, y2) in X1,
K1(x1, λ) = K1
(
(x1, y1), λ
)
= K(x, λ)× {0Y },
G1(x2, µ) = G1
(
(x2, y2), µ
)
= {0X} × T (x2, µ),
F1(x1, x2, µ) = F
(
(x1, y1), (x2, y2), µ
)
= T (x2, µ)− y2.
We consider the following problem (Pmα1):
(MQEP) Find x¯ ∈ clK(x¯, λ) such that ∃y¯ ∈ T (x¯, µ), ∀x ∈ K(x¯, λ),
F1
(
(0, y¯), (x, 0), µ
) ≡ T (x, µ)− y¯ ⊆ Y \ −intC.
Proposition 4.5. x¯ is a solution of (QOP) if and only if x¯ is a solution of (MQEP).
The proof is direct and so is omitted. The following two approximate problems of (QOP)
and (MQEP) are also equivalent
(QOPε) Find x¯ ∈ clK(x¯, λ) such that
T (x¯, µ) ∩ wMin{T (K(x¯, λ), µ)∣∣comp(−intC)ε1} 6= ∅;
(MQEPε) Find x¯ ∈ clK(x¯, λ) such that ∃y¯ ∈ T (x¯, µ), ∀x ∈ K(x¯, λ),
T (x, µ)− y¯ ⊆ comp(−intC)ε1.
The following corollaries about the approximate solution sets S˜ε and Sε of (QOP) are
direct consequences of Theorems 2.2, 2.4 and 3.3, respectively.
Corollary 4.10. Assume that K is usc and has compact values in clK(X,Λ) ×{λ0}, E is
lsc at λ0, T is lsc in clK(X,Λ) × {µ0} and (x, y, µ) 7→ T (x, µ) − y is comp(−intC)ε1−usc in
clK(X,Λ)× clK(X,Λ)× {µ0}. Then S˜ε is lsc at (λ0, µ0) for ε > 0.
Corollary 4.11. Assume the assumptions of Corollary 4.10. Assume further that E(λ0) is
compact, K(., λ0) is lsc and closed in clK(X,Λ), T (., µ0) is usc and has compact values in
clK(X,Λ) and (x, y) 7→ T (x, µ0)− y is −C-lsc in clK(X,Λ)× clK(X,Λ). Then S˜ε is Hlsc at
(λ0, µ0) for ε > 0.
Corollary 4.12. Sε is usc and closed at (λ0, µ0) for ε ≥ 0, if K is lsc in clK(X,Λ) × {λ0},
E is usc at λ0 and E(λ0) is compact, T is usc, Y \ comp(−intC)ε1- lsc and compact-valued in
clK(X,Λ)× {µ0}.
13
REFERENCES
1. M. Ait Mansour, and L. Scrimali (2007). Ho¨lder continuity of solutions to elastic traffic
network models. J. Glob. Optim., online.
2. M. Ait Mansour and H. Riahi (2005). Sensitivity analysis for abstract equilibrium prob-
lems. J. Math. Anal. Appl. 306: 684-691.
3. L.Q. Anh and P.Q. Khanh (2004). Semicontinuity of the solution set of parametric
multivalued vector quasiequilibrium problems. J. Math. Anal. Appl. 294: 699-711.
4. L.Q. Anh and P.Q. Khanh (2006). On the Ho¨lder continuity of solutions to parametric
multivalued vector equilibrium problems. J. Math. Anal. Appl. 321: 308-315.
5. L.Q. Anh and P.Q. Khanh (2007). On the stability of the solution sets of general multi-
valued vector quasiequilibrium problems. J. Optim. Theory Appl. 135: 271-284.
6. L.Q. Anh and P.Q. Khanh (2007). Uniqueness and Ho¨lder continuity of the solution to
multivalued equilibrium problems in metric spaces. J. Glob. Optim. 37: 449-465.
7. L.Q. Anh and P.Q. Khanh (2007). Various kinds of semicontinuity and the solution sets
of parametric multivalued symmetric vector quasiequilibrium problems. J. Glob. Optim.,
to appear.
8. L.Q. Anh and P.Q. Khanh. Sensitivity analysis for multivalued quasiequilibrium prob-
lems in metric spaces: Ho¨lder continuity of solutions. J. Glob. Optim.,to appear.
9. M. Bianchi and R. Pini (2003). A note on stability for parametric equilibrium problems.
Oper. Res. Lett. 31: 445-450.
10. M. Bianchi and R. Pini (2006). Sensitivity for parametric vector equilibria. Optimization
55: 221-230.
11. R. Braˆnzei, J. Morgan, V. Scalzo, and S. Tijs. (2003). Approximate fixed point theorems
in Banach spaces with applications in game theory. J. Math. Anal. Appl. 285: 619-628.
12. Y.H. Cheng and D.L. Zhu (2005). Global stability results for the weak vector variational
inequality. J. Glob. Optim. 32: 543-550.
13. A. Domokos and J. Kolumba´n (2002). Variational inequalities with operator solutions.
J. Glob. Optim. 23: 99-110.
14. N.J. Huang, J. Li, and H.B. Thompson (2006). Stability for parametric implicit vector
equilibrium problems. Math. Comput. Model. 43: 1267-1274.
15. P.Q. Khanh and L.M. Luu (2005). Upper semicontinuity of the solution set of parametric
multivalued vector quasivariational inequalities and applications. J. Glob. Optim. 32:
569-580.
16. P.Q. Khanh and L.M. Luu (2007). Lower and upper semicontinuity of the solution sets
and approximate solution sets to parametric multivalued quasivariational inequalities. J.
Optim. Theory Appl. 133: no. 1, in press.
17. S.J. Li, G.Y. Chen, and K.L. Teo (2002). On the stability of generalized vector quasi-
variational inequality problems. J. Optim. Theory Appl. 113: 283-295.
18. L.D. Muu (1984). Stability property of a class of variational inequalities. Math. Opera-
tionsforsch. Statist., ser: Optim. 15: 347-351.
19. J. Zhao (1977). The lower semicontinuity of optimal solution sets. J. Math. Anal. Appl.
207: 240-254.
14