Return to Article Details Continuity of generalized convex and generalized concave set-valued functions

CONTINUITY OF GENERALIZED CONVEX AND GENERALIZED CONCAVE SET-VALUED FUNCTIONS

WOLFGANG W. BRECKNER(Cluj-Napoca)

1. Introduction

Criteria for the continuity of convex and concave functions play am important role in functional analysis and optimization theory, no matter whether these functions are single- or set-valued. But in the case of single-valued functions it suffices to state such criteria for one of these two classes of functions, because a real- or vector-valued function f f fff is convex (resp. concave) if and only if f f -f-ff is concave (resp. convex). Unfortunately, this advantage is lost in the case of set-valued functions. Although the continuity properties of convex set-valued functions are quite similar to those of concave set-valued functions, they have to be proved for each of these classes of functions separately. The present paper will show that the same phenomenon arises in the case of a pair of more general set-valued functions. We call these two new types of setvalued functions ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-convex and ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-concave, respectively. They are defined as follows.
Assume that A A AAA is a subset of the open interval ] 0 , 1 ] 0 , 1 ]0,1] 0,1]0,1 [.having zero as a cluster point, and that s s sss is a positive real number. Throughout the paper A A AAA and s s sss will always have this meaning. Let X X XXX and Y Y YYY be real topological linear spaces, let P 0 ( Y ) P 0 ( Y ) P_(0)(Y)\mathscr{P}_{0}(Y)P0(Y) denote the set consisting of all nonempty subsets of Y Y YYY, and let M M MMM be a nonempty convex subset of X X XXX. A function F 1 : M P 0 ( Y ) F 1 : M P 0 ( Y ) F^(1):M rarrP_(0)(Y)F^{1}: M \rightarrow \mathscr{P}_{0}(Y)F1:MP0(Y) is said to be :
(i) ( A , s ) -convex if ( 1 a ) s F ( x ) + a s F ( y ) F ( ( 1 a ) x + a y ) (i) ( A , s ) -convex if  ( 1 a ) s F ( x ) + a s F ( y ) F ( ( 1 a ) x + a y ) {:[(i)(A","s)"-convex if "],[quad(1-a)^(s)F(x)+a^(s)F(y)sub F((1-a)x+ay)]:}\begin{align*} & (A, s) \text {-convex if } \tag{i}\\ & \quad(1-a)^{s} F(x)+a^{s} F(y) \subset F((1-a) x+a y) \end{align*}(i)(A,s)-convex if (1a)sF(x)+asF(y)F((1a)x+ay)
whenever a A a A a in Aa \in AaA and x , y M x , y M x,y in Mx, y \in Mx,yM;
(ii) ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-concave if
F ( ( 1 a ) x + a y ) ( 1 a ) s F ( x ) + a s F ( y ) F ( ( 1 a ) x + a y ) ( 1 a ) s F ( x ) + a s F ( y ) F((1-a)x+ay)sub(1-a)^(s)F(x)+a^(s)F(y)F((1-a) x+a y) \subset(1-a)^{s} F(x)+a^{s} F(y)F((1a)x+ay)(1a)sF(x)+asF(y)
whenever a A a A a in Aa \in AaA and x , y M x , y M x,y in Mx, y \in Mx,yM.
Obviously, for special choices of A A AAA and s s sss in these definitions, we obtain certain kinds of convexity and concavity for set-valued functions which have already been investigated. So, when A = ] 0 , 1 [ A = ] 0 , 1 [ A=]0,1[A=] 0,1[A=]0,1[ and s = 1 s = 1 s=1s=1s=1, we get the convex and concave set-valued functions; when A = ] 0 , 1 [ Q A = ] 0 , 1 [ Q A=]0,1[nnQA=] 0,1[\cap \mathbf{Q}A=]0,1[Q and s = 1 s = 1 s=1s=1s=1, we get the rationally convex and rationally concave set-valued functions; finally, when A = { 2 n : n N } A = 2 n : n N A={2^(-n):n inN}A=\left\{2^{-n}: n \in \mathbf{N}\right\}A={2n:nN} and s = 1 s = 1 s=1s=1s=1, we get the sequentially midpoint conves and sequentially midpoint concave set-valued functions.
A relationship between our generalized convexity concept introduced for set-valued functions and some convexity concepts known hitherto merely for single-valued functions has also to be mentioned here. To this end, let K K KKK be a convex cone in Y Y YYY, i.e a nonempty convex set closed under multiplication by positive scalars. If f : M Y f : M Y f:M rarr Yf: M \rightarrow Yf:MY is a function satisfying
(1.1) ( 1 a ) s f ( x ) + a s f ( y ) f ( ( 1 a ) x + a y ) + K (1.1) ( 1 a ) s f ( x ) + a s f ( y ) f ( ( 1 a ) x + a y ) + K {:(1.1)(1-a)^(s)f(x)+a^(s)f(y)in f((1-a)x+ay)+K:}\begin{equation*} (1-a)^{s} f(x)+a^{s} f(y) \in f((1-a) x+a y)+K \tag{1.1} \end{equation*}(1.1)(1a)sf(x)+asf(y)f((1a)x+ay)+K
whenever a A a A a in Aa \in AaA and x , y M x , y M x,y in Mx, y \in Mx,yM, then F : M P 0 ( Y ) F : M P 0 ( Y ) F:M rarrP_(0)(Y)F: M \rightarrow \mathscr{P}_{0}(Y)F:MP0(Y), defined by F ( x ) = f ( x ) + + K F ( x ) = f ( x ) + + K F(x)=f(x)++KF(x)=f(x)+ +KF(x)=f(x)++K, is an ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-convex set-valued function. In particular, when Y = R Y = R Y=RY=\mathbf{R}Y=R and K = R + K = R + K=R_(+)K=\mathbf{R}_{+}K=R+, then (1.1) reduces to an inequality which obviously yields (for suitable A A AAA and s s sss ) diverse convexity concepts for real-valued functions used in optimization and functional analysis.
The goal of the present paper is to characterize the continuity of ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-convex and ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-concave set-valued functions. The results we shall prove reveal, for each of these classes of functions, the connection between continuity, on the one hand, and lower semi-continuity;.upper semicontinuity, local boundedness, uniform boundedness, on the other hand. They highlight the usefulness of the two types of set-valued functions introduced here. In a very general setting both necessary and sufficient conditions are obtained for continuity which have been missing even for those particular classes of functions which suggested the general concept of convexity and concavity, respectively. Another feature of these results is that they are a set-valued counterpart to well-known theorems concerning the continuity of convex and concave real-valued functions. By specializing our results, some sufficient conditions for continuity, recently given by K. Nikodem [3], [4], can be obtained.

2. Preliminaries

To make our paper self-contained we mention in this section all the notions concerning set-valued functions we shall need.
Let X X XXX and Y Y YYY be real topological linear spaces, let M M MMM be a nonempty subset of X X XXX, and let F F FFF be a function from M M MMM to P 0 ( Y ) P 0 ( Y ) P_(0)(Y)P_{0}(Y)P0(Y). For simplicity we shall write
F ( T ) = x T F ( x ) F ( T ) = x T F ( x ) F(T)=uuu_(x in T)F(x)F(T)=\bigcup_{x \in T} F(x)F(T)=xTF(x)
for any subset T T TTT of M M MMM.
If x 0 x 0 x_(0)x_{0}x0 is a point in M M MMM, we say that F F FFF is:
(i) lower semicontinuous at x 0 x 0 x_(0)x_{0}x0 if for every neighbourhood V V VVV of the origin of Y Y YYY there exists a neighbourhood U U UUU of x 0 x 0 x_(0)x_{0}x0 such that F ( x 0 ) F ( x ) + + V F x 0 F ( x ) + + V F(x_(0))sub F(x)++VF\left(x_{0}\right) \subset F(x)+ +VF(x0)F(x)++V for all x U M x U M x in U nn Mx \in U \cap MxUM;
(ii) upper semicontinuous at x 0 x 0 x_(0)x_{0}x0 if for every neighbourhood V V VVV of the origin of Y Y YYY there exists a neighbourhood U U UUU of x 0 x 0 x_(0)x_{0}x0 such that F ( x ) F ( x 0 ) + + V F ( x ) F x 0 + + V F(x)sub F(x_(0))++VF(x) \subset F\left(x_{0}\right)+ +VF(x)F(x0)++V for all x U M x U M x in U nn Mx \in U \cap MxUM;
(iii) continuous at x 0 x 0 x_(0)x_{0}x0 if F F FFF is both lower semicontinuous and upper semicontinuous at x 0 x 0 x_(0)x_{0}x0;
(iv) locally prebounded at x 0 x 0 x_(0)x_{0}x0 if for every neighbourhood V V VVV of the origin of Y Y YYY there exist a positive integer n n nnn and a neighbourhood U U UUU of x 0 x 0 x_(0)x_{0}x0 such that F ( x ) n V F ( x ) n V F(x)nn nV!=O/F(x) \cap n V \neq \emptysetF(x)nV for all x U M x U M x in U nn Mx \in U \cap MxUM;
(v) locally bounded at x 0 x 0 x_(0)x_{0}x0 if for every neighbourhood V V VVV of the origin of Y Y YYY there exist a positive integer n n nnn and a neigbourhood U U UUU of x 0 x 0 x_(0)x_{0}x0 such that F ( U M ) n V F ( U M ) n V F(U nn M)sub nVF(U \cap M) \subset n VF(UM)nV;
(vi) bounded at x 0 x 0 x_(0)x_{0}x0 if the set F ( x 0 ) F x 0 F(x_(0))F\left(x_{0}\right)F(x0) is bounded, i.e. if for every neighbourhood V V VVV of the origin of Y Y YYY there exists a positive integer n n nnn such that F ( x 0 ) n F x 0 n F(x_(0))sub n gradF\left(x_{0}\right) \subset n \nablaF(x0)n.
The function F F FFF is called :
(j) lower semicontinuous (resp. upper semicontinuous, continuous, locally prebounded, locally bounded) on M M MMM if it is lower semicontinuous (resp. upper semicontinuous, continuous, locally prebounded, locally bounded) at each point of M M MMM;
(jj) pointwise bounded on M M MMM if it is bounded at each point of M M MMM;
(jjj) uniformly bounded on M M MMM if for every neighbourhood V V VVV of the origin of Y Y YYY there exist a positive integer n n nnn and a nonempty open subset T T TTT of M M MMM such that F ( T ) n V F ( T ) n V F(T)sub nVF(T) \subset n VF(T)nV.
We note that the above-mentioned concepts of lower semicontinuity, upper semicontinuity and continuity of a set-valued function are well known, local preboundedness seems to be used here for the first time, while the other above-introduced concepts have been inspired by concepts used in functional analysis for families of functions.
Lemma 2.1. Let X X XXX and Y Y YYY be real topological linear spaces, let M M MMM be a nozempty subset of X X XXX, let x 0 x 0 x_(0)x_{0}x0 be a point in M M MMM, and let F : M P 0 ( Y ) F : M P 0 ( Y ) F:M rarrP_(0)(Y)F: M \rightarrow \mathscr{P}_{0}(Y)F:MP0(Y) be a function which is bounded at x 0 x 0 x_(0)x_{0}x0. Then the following assertions are true:
1 1 1^(@)1^{\circ}1 If F F FFF is lower semicontinuous at x 0 x 0 x_(0)x_{0}x0, then it is locally prebounded at x 0 x 0 x_(0)x_{0}x0.
2 2 2^(@)2^{\circ}2 If: F F FFF is upper semicontinuous at x 0 x 0 x_(0)x_{0}x0, then it is locally bounded at x 0 x 0 x_(0)x_{0}x0.
Proof. 1 1 1^(@)1^{\circ}1 Let V V VVV be any neighbourhood of the origin of Y Y YYY. Choose a neighbourhood V 0 V 0 V_(0)V_{0}V0 of the origin of Y Y YYY such that V 0 V 0 V V 0 V 0 V V_(0)-V_(0)sub VV_{0}-V_{0} \subset VV0V0V. Since F F FFF is bounded at x 0 x 0 x_(0)x_{0}x0, there exists a positive integer n n nnn such that F ( x 0 ) n V 0 F x 0 n V 0 F(x_(0))sub nV_(0)F\left(x_{0}\right) \subset n V_{0}F(x0)nV0. Taking now into consideration that F F FFF is lower semicontinuous at x 0 x 0 x_(0)x_{0}x0, we can conclude that there exists a neighbourhood U U UUU of x 0 x 0 x_(0)x_{0}x0 such that F ( x 0 ) ⊂⊂ F ( x ) + n V 0 F x 0 ⊂⊂ F ( x ) + n V 0 F(x_(0))⊂⊂F(x)+nV_(0)F\left(x_{0}\right) \subset \subset F(x)+n V_{0}F(x0)⊂⊂F(x)+nV0 for all x U M x U M x in U nn Mx \in U \cap MxUM. Thus we have
0 F ( x 0 ) F ( x 0 ) F ( x ) + n V 0 n V 0 F ( x ) n V 0 F x 0 F x 0 F ( x ) + n V 0 n V 0 F ( x ) n V 0in F(x_(0))-F(x_(0))sub F(x)+nV_(0)-nV_(0)sub F(x)-nV0 \in F\left(x_{0}\right)-F\left(x_{0}\right) \subset F(x)+n V_{0}-n V_{0} \subset F(x)-n V0F(x0)F(x0)F(x)+nV0nV0F(x)nV
for all x U M x U M x in U nn Mx \in U \cap MxUM, and hence F ( x ) n V F ( x ) n V F(x)nn nV!=O/F(x) \cap n V \neq \varnothingF(x)nV for all x U M x U M x in U nn Mx \in U \cap MxUM. Consequently, F F FFF is locally prebounded at x 0 x 0 x_(0)x_{0}x0.
2 2 2^(@)2^{\circ}2. Let V V VVV be any neighbourhood of the origin of Y Y YYY. Choose a neighbourhood V 0 V 0 V_(0)V_{0}V0 of the origin of Y Y YYY such that V 0 + V 0 V V 0 + V 0 V V_(0)+V_(0)sub VV_{0}+V_{0} \subset VV0+V0V. Since F ( x 0 ) F x 0 F(x_(0))F\left(x_{0}\right)F(x0) is bounded, there exists a positive integer n n nnn such that F ( x 0 ) n V 0 F x 0 n V 0 F(x_(0))sub nV_(0)F\left(x_{0}\right) \subset n V_{0}F(x0)nV0. In view of the upper semicontinuity of F F FFF at x 0 x 0 x_(0)x_{0}x0 there exists a neighbourhood U U UUU of x 0 x 0 x_(0)x_{0}x0 such that F ( U M ) F ( x 0 ) + n V 0 F ( U M ) F x 0 + n V 0 F(U nn M)sub F(x_(0))+nV_(0)F(U \cap M) \subset F\left(x_{0}\right)+n V_{0}F(UM)F(x0)+nV0. This inclusion implies F ( U M ) n V 0 + n V 0 n V F ( U M ) n V 0 + n V 0 n V F(U nn M)sub nV_(0)+nV_(0)sub nVF(U \cap M) \subset n V_{0}+n V_{0} \subset n VF(UM)nV0+nV0nV. Hence F F FFF is locally bounded at x 0 x 0 x_(0)x_{0}x0.

3. Continuity of ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-convex set-valued functions

First we investigate the continuity of an ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-convex set-valued function at an interior point of its domain.
Theorem 3.1. If X X XXX and Y Y YYY are real topological linear spaces, M M MMM a convex subset of X , x 0 X , x 0 X,x_(0)X, x_{0}X,x0 an interior point of M M MMM, and F : M P 0 ( Y ) F : M P 0 ( Y ) F:M rarrP_(0)(Y)F: M \rightarrow \mathscr{P}_{0}(Y)F:MP0(Y) an ( A , s ) ( A , s ) (A,s)(A, s)(A,s) -
-convex function, then the following statements are equivalent:
1 F 1 F 1^(@)F1^{\circ} F1F is both continuous and bounded at x 0 x 0 x_(0)x_{0}x0.
2 F 1 2 F 1 2^(@)F^(1)2^{\circ} F^{1}2F1 is both upper semicontinuous and bounded at c 0 c 0 c_(0)c_{0}c0.
3 F 3 F 3^(@)F3^{\circ} F3F is locally bounded at x 0 x 0 x_(0)x_{0}x0.
4 F 4 F 4^(@)F4^{\circ} F4F is both locally prebounded and bounded at x 0 x 0 x_(0)x_{0}x0.
5 F 5 F 5^(@)F5^{\circ} F5F is both lower semicontinuous and bounded at x 0 x 0 x_(0)x_{0}x0.
Proof. The implications 1 2 1 2 1^(@)=>2^(@)1^{\circ} \Rightarrow 2^{\circ}12 and 3 4 3 4 3^(@)=>4^(@)3^{\circ} \Rightarrow 4^{\circ}34 are obvious, while the implication 2 3 2 3 2^(@)=>3^(@)2^{\circ} \Rightarrow 3^{\circ}23 has been stated in Lemma 2.1.
Suppose now that 4 4 4^(@)4^{\circ}4 is true. Under this assumption F F F^(')F^{\prime}F is lower semicontinuous at x 0 x 0 x_(0)x_{0}x0. Indeed, let V V VVV be a neighbourhood of the origin of Y Y YYY. Choose a balanced neighbourhood V 0 V 0 V_(0)V_{0}V0 of the origin of Y Y YYY such that V 0 + + V 0 V V 0 + + V 0 V V_(0)++V_(0)sub VV_{0}+ +V_{0} \subset VV0++V0V. Since F F FFF is bounded at x 0 x 0 x_(0)x_{0}x0, there exists a positive integer m m mmm such that F ( x 0 ) m V 0 F x 0 m V 0 F(x_(0))sub mV_(0)F\left(x_{0}\right) \subset m V_{0}F(x0)mV0, and since F F FFF is locally prebounded at x 0 x 0 x_(0)x_{0}x0, there exist a positive integer n n nnn and a neighbourhood U 0 U 0 U_(0)U_{0}U0 of x 0 x 0 x_(0)x_{0}x0 such that F ( x ) n n V 0 F ( x ) n n V 0 F(x)nn nnV_(0)!=O/F(x) \cap n n V_{0} \neq \emptysetF(x)nnV0 for all x U 0 M x U 0 M x inU_(0)nn Mx \in U_{0} \cap MxU0M. Take now, on the one hand, a balanced neighbourhood W W WWW of the origin of X X XXX such that x 0 + W U 0 M x 0 + W U 0 M x_(0)+W subU_(0)nn Mx_{0}+W \subset U_{0} \cap Mx0+WU0M, and, on the other hand, a number a A a A a in Aa \in AaA which satisfies the following inequalities:
[ 1 ( 1 a ) s ] m 1 , a s n 1 . 1 ( 1 a ) s m 1 , a s n 1 . [1-(1-a)^(s)]m <= 1,quada^(s)n <= 1.\left[1-(1-a)^{s}\right] m \leqslant 1, \quad a^{s} n \leqslant 1 .[1(1a)s]m1,asn1.
Of course, U = x 0 + a W U = x 0 + a W U=x_(0)+aWU=x_{0}+a WU=x0+aW is a neighbourhood of x 0 x 0 x_(0)x_{0}x0 contained in U 0 M U 0 M U_(0)nn MU_{0} \cap MU0M. We claim that
(3.1) F ( x 0 ) F ( x ) + V for all x U (3.1) F x 0 F ( x ) + V  for all  x U {:(3.1)F(x_(0))sub F(x)+V" for all "x in U:}\begin{equation*} F\left(x_{0}\right) \subset F(x)+V \text { for all } x \in U \tag{3.1} \end{equation*}(3.1)F(x0)F(x)+V for all xU
Let x x xxx be a point in U U UUU. Then there exists y W y W y in Wy \in WyW such that x = x 0 + x = x 0 + x=x_(0)+x=x_{0}+x=x0+ + ay. From T ( x 0 + y ) n V 0 T x 0 + y n V 0 T(x_(0)+y)nn nV_(0)!=O/T\left(x_{0}+y\right) \cap n V_{0} \neq \varnothingT(x0+y)nV0 it follows that 0 F ( x 0 + y ) n V 0 0 F x 0 + y n V 0 0in F(x_(0)+y)-nV_(0)0 \in F\left(x_{0}+y\right)-n V_{0}0F(x0+y)nV0. Consequently, we have
0 a s F ( x 0 + y ) a s n V 0 a s F ( x 0 + y ) + V 0 0 a s F x 0 + y a s n V 0 a s F x 0 + y + V 0 0ina^(s)F(x_(0)+y)-a^(s)nV_(0)suba^(s)F(x_(0)+y)+V_(0)0 \in a^{s} F\left(x_{0}+y\right)-a^{s} n V_{0} \subset a^{s} F\left(x_{0}+y\right)+V_{0}0asF(x0+y)asnV0asF(x0+y)+V0
Taking into account this result as well as the ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-convexity of F F FFF, we obtain
F ( x 0 ) ( 1 a ) s F ( x 0 ) + a s F ( x 0 + y ) + [ 1 ( 1 a ) s ] I ( x 0 ) + V 0 F ( x ) + [ 1 ( 1 a ) s ] m V 0 + V 0 F ( x ) + V 0 + V 0 F ( x ) + V . F x 0 ( 1 a ) s F x 0 + a s F x 0 + y + 1 ( 1 a ) s I x 0 + V 0 F ( x ) + 1 ( 1 a ) s m V 0 + V 0 F ( x ) + V 0 + V 0 F ( x ) + V . {:[F(x_(0))sub(1-a)^(s)F(x_(0))+a^(s)F(x_(0)+y)+[1-(1-a)^(s)]I^(')(x_(0))+V_(0)sub],[ sub F(x)+[1-(1-a)^(s)]mV_(0)+V_(0)subF^(')(x)+V_(0)+V_(0)sub F(x)+V.]:}\begin{aligned} & F\left(x_{0}\right) \subset(1-a)^{s} F\left(x_{0}\right)+a^{s} F\left(x_{0}+y\right)+\left[1-(1-a)^{s}\right] I^{\prime}\left(x_{0}\right)+V_{0} \subset \\ & \subset F(x)+\left[1-(1-a)^{s}\right] m V_{0}+V_{0} \subset F^{\prime}(x)+V_{0}+V_{0} \subset F(x)+V . \end{aligned}F(x0)(1a)sF(x0)+asF(x0+y)+[1(1a)s]I(x0)+V0F(x)+[1(1a)s]mV0+V0F(x)+V0+V0F(x)+V.
Hence (3.1) is proved. Therefore F F FFF is lower semicontinuous at x 0 x 0 x_(0)x_{0}x0. So the implication 4 5 4 5 4^(@)=>5^(@)4^{\circ} \Rightarrow 5^{\circ}45 is stated.
To complete the proof we show now that 5 5 5^(@)5^{\circ}5 implies 1 1 1^(@)1^{\circ}1. Let V V VVV be any neighbourhood of the origin of Y Y YYY. Choose a balanced neighbourhood V 0 V 0 V_(0)V_{0}V0 of the origin of Y Y YYY such that V 0 + V 0 + V 0 V V 0 + V 0 + V 0 V V_(0)+V_(0)+V_(0)sub VV_{0}+V_{0}+V_{0} \subset VV0+V0+V0V. Since F F F^(')F^{\prime}F is bounded at x 0 x 0 x_(0)x_{0}x0, there exists a positive integer n n nnn such that F ( x 0 ) n V 0 F x 0 n V 0 F^(')(x_(0))sub nV_(0)F^{\prime}\left(x_{0}\right) \subset n V_{0}F(x0)nV0. On the other hand, since F F FFF is lower semicontinuous at x 0 x 0 x_(0)x_{0}x0, there exists for V 0 V 0 V_(0)V_{0}V0 a neighbourhood U 0 U 0 U_(0)U_{0}U0 of x 0 x 0 x_(0)x_{0}x0 such that F ( x 0 ) F ( x ) + V 0 F x 0 F ( x ) + V 0 F(x_(0))sub F(x)+V_(0)F\left(x_{0}\right) \subset F(x)+V_{0}F(x0)F(x)+V0 for all x U 0 M x U 0 M x inU_(0)nn Mx \in U_{0} \cap MxU0M.
Take now a balanced neighbourhood W W WWW of the origin of X X XXX such that x 0 + W U 0 M x 0 + W U 0 M x_(0)+W subU_(0)nn Mx_{0}+W \subset U_{0} \cap Mx0+WU0M. Also select a number a A a A a in Aa \in AaA which satisfies the following inequalities:
a < 1 2 , [ ( 1 1 a ) s 1 ] n 1 , ( a 1 a ) s n 1 . a < 1 2 , 1 1 a s 1 n 1 , a 1 a s n 1 . a < (1)/(2),quad[((1)/(1-a))^(s)-1]n <= 1,quad((a)/(1-a))^(s)n <= 1.a<\frac{1}{2}, \quad\left[\left(\frac{1}{1-a}\right)^{s}-1\right] n \leqslant 1, \quad\left(\frac{a}{1-a}\right)^{s} n \leqslant 1 .a<12,[(11a)s1]n1,(a1a)sn1.
The set
U = x 0 + a a 1 W U = x 0 + a a 1 W U=x_(0)+(a)/(a-1)WU=x_{0}+\frac{a}{a-1} WU=x0+aa1W
is a nieghbourhood of x 0 x 0 x_(0)x_{0}x0 contained in U 0 M U 0 M U_(0)nn MU_{0} \cap MU0M. Furthermore, we have
(3.2) F ( x ) F ( x 0 ) + V for all x U . (3.2) F ( x ) F x 0 + V  for all  x U . {:(3.2)F(x)sub F(x_(0))+V" for all "x in U.:}\begin{equation*} F(x) \subset F\left(x_{0}\right)+V \text { for all } x \in U . \tag{3.2} \end{equation*}(3.2)F(x)F(x0)+V for all xU.
Indeed, if x x xxx is in U U UUU, it can expressed in the form
x = x 0 + a a 1 y x = x 0 + a a 1 y x=x_(0)+(a)/(a-1)yx=x_{0}+\frac{a}{a-1} yx=x0+aa1y
where y W y W y in Wy \in WyW. Since F ( x 0 ) F ( x 0 + y ) + V 0 F x 0 F x 0 + y + V 0 F(x_(0))sub F(x_(0)+y)+V_(0)F\left(x_{0}\right) \subset F\left(x_{0}+y\right)+V_{0}F(x0)F(x0+y)+V0 and x 0 = ( 1 a ) x + a ( x 0 + y ) x 0 = ( 1 a ) x + a x 0 + y x_(0)=(1-a)x+a(x_(0)+y)x_{0}=(1-a) x+a\left(x_{0}+y\right)x0=(1a)x+a(x0+y), it follows by the ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-convexity of F F F^(')F^{\prime}F that
( 1 a ) s F ( x ) + a s F ( x 0 ) ( 1 a ) s F ( x ) + a s F ( x 0 + y ) + a s V 0 F ( x 0 ) + a s V 0 ( 1 a ) s F ( x ) + a s F x 0 ( 1 a ) s F ( x ) + a s F x 0 + y + a s V 0 F x 0 + a s V 0 (1-a)^(s)F(x)+a^(s)F(x_(0))sub(1-a)^(s)F(x)+a^(s)F(x_(0)+y)+a^(s)V_(0)sub F(x_(0))+a^(s)V_(0)(1-a)^{s} F(x)+a^{s} F\left(x_{0}\right) \subset(1-a)^{s} F(x)+a^{s} F\left(x_{0}+y\right)+a^{s} V_{0} \subset F\left(x_{0}\right)+a^{s} V_{0}(1a)sF(x)+asF(x0)(1a)sF(x)+asF(x0+y)+asV0F(x0)+asV0, and henoe
F ( x ) + ( a 1 a ) s F ( x 0 ) ( 1 1 a ) s F ( x 0 ) + ( a 1 a ) s V 0 ( 1 1 a ) s F ( x 0 ) + V 0 F ( x ) + a 1 a s F x 0 1 1 a s F x 0 + a 1 a s V 0 1 1 a s F x 0 + V 0 F(x)+((a)/(1-a))^(s)F(x_(0))sub((1)/(1-a))^(s)F(x_(0))+((a)/(1-a))^(s)V_(0)sub((1)/(1-a))^(s)F(x_(0))+V_(0)F(x)+\left(\frac{a}{1-a}\right)^{s} F\left(x_{0}\right) \subset\left(\frac{1}{1-a}\right)^{s} F\left(x_{0}\right)+\left(\frac{a}{1-a}\right)^{s} V_{0} \subset\left(\frac{1}{1-a}\right)^{s} F\left(x_{0}\right)+V_{0}F(x)+(a1a)sF(x0)(11a)sF(x0)+(a1a)sV0(11a)sF(x0)+V0.
This result implies
F ( x ) F ( x ) + ( a 1 a ) s F ( x 0 ) ( a 1 a ) s F ( x 0 ) ( 1 1 a ) s F ( x 0 ) ( a 1 a ) s F ( x 0 ) + V 0 F ( x 0 ) + [ ( 1 1 a ) s 1 ] F ( x 0 ) ( a 1 a ) s F ( x 0 ) + V 0 F ( x 0 ) + [ ( 1 1 a ) s 1 ] n V 0 ( a 1 a ) s n V 0 + V 0 F ( x 0 ) + V 0 + V 0 + V 0 F ( x 0 ) + V . F ( x ) F ( x ) + a 1 a s F x 0 a 1 a s F x 0 1 1 a s F x 0 a 1 a s F x 0 + V 0 F x 0 + 1 1 a s 1 F x 0 a 1 a s F x 0 + V 0 F x 0 + 1 1 a s 1 n V 0 a 1 a s n V 0 + V 0 F x 0 + V 0 + V 0 + V 0 F x 0 + V . {:[F(x)sub F(x)+((a)/(1-a))^(s)F(x_(0))-((a)/(1-a))^(s)F(x_(0))sub],[ sub((1)/(1-a))^(s)F(x_(0))-((a)/(1-a))^(s)F(x_(0))+V_(0)sub],[ sub F(x_(0))+[((1)/(1-a))^(s)-1]F(x_(0))-((a)/(1-a))^(s)F(x_(0))+V_(0)sub],[ sub F(x_(0))+[((1)/(1-a))^(s)-1]nV_(0)-((a)/(1-a))^(s)nV_(0)+V_(0)sub],[ sub F(x_(0))+V_(0)+V_(0)+V_(0)sub F(x_(0))+V.]:}\begin{aligned} & F(x) \subset F(x)+\left(\frac{a}{1-a}\right)^{s} F\left(x_{0}\right)-\left(\frac{a}{1-a}\right)^{s} F\left(x_{0}\right) \subset \\ & \subset\left(\frac{1}{1-a}\right)^{s} F\left(x_{0}\right)-\left(\frac{a}{1-a}\right)^{s} F\left(x_{0}\right)+V_{0} \subset \\ & \subset F\left(x_{0}\right)+\left[\left(\frac{1}{1-a}\right)^{s}-1\right] F\left(x_{0}\right)-\left(\frac{a}{1-a}\right)^{s} F\left(x_{0}\right)+V_{0} \subset \\ & \subset F\left(x_{0}\right)+\left[\left(\frac{1}{1-a}\right)^{s}-1\right] n V_{0}-\left(\frac{a}{1-a}\right)^{s} n V_{0}+V_{0} \subset \\ & \subset F\left(x_{0}\right)+V_{0}+V_{0}+V_{0} \subset F\left(x_{0}\right)+V . \end{aligned}F(x)F(x)+(a1a)sF(x0)(a1a)sF(x0)(11a)sF(x0)(a1a)sF(x0)+V0F(x0)+[(11a)s1]F(x0)(a1a)sF(x0)+V0F(x0)+[(11a)s1]nV0(a1a)snV0+V0F(x0)+V0+V0+V0F(x0)+V.
Thus (3.2) holds. In other words, we have proved that F F FFF is upper semicontinuous at x 0 x 0 x_(0)x_{0}x0. Being both lower semicontinuous and upper semicontinuous at x 0 x 0 x_(0)x_{0}x0, the function F F FFF is continuous at x 0 x 0 x_(0)x_{0}x0. Se we have shown that 5 5 5^(@)5^{\circ}5 implies 1 1 1^(@)1^{\circ}1.
Corollary 3.2. Let X X XXX and Y Y YYY be real topological linear spaces, let M M MMM be a convex subset of X X XXX, let x 0 x 0 x_(0)x_{0}x0 be an interior point of M M MMM, and let F : M P 0 ( Y ) F : M P 0 ( Y ) F:M rarrP_(0)(Y)F: M \rightarrow \mathscr{P}_{0}(Y)F:MP0(Y) be an ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-convex function which is bounded at x 0 x 0 x_(0)x_{0}x0. Then F F FFF is continuous at x 0 x 0 x_(0)x_{0}x0 if and only if there exists a function G : M P 0 ( Y ) G : M P 0 ( Y ) G:M rarrP_(0)(Y)G: M \rightarrow \mathscr{P}_{0}(Y)G:MP0(Y) satisfying the following conditions :
(i) G G GGG is locally prebounded at x 0 x 0 x_(0)x_{0}x0;
(ii) G ( x ) F ( x ) G ( x ) F ( x ) G(x)sub F(x)G(x) \subset F(x)G(x)F(x) for all x M x M x in Mx \in MxM.
Proof. Necessity. Obvious in view of Theorem 3.1.
Sufficiency. It is easily seen that F F FFF is locally prebounded at x 0 x 0 x_(0)x_{0}x0. By applying Theorem 3.1 it results that F F FFF is continuous at x 0 x 0 x_(0)x_{0}x0.
By means of Theorem 3.1 we can now investigate the continuity of an ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-convex set-valued function on its whole domain.
Theorem 3.3. If X X XXX and Y Y YYY are real topological linear spaces, M M MMM a nonempty open convex subset of X X XXX, and F : M P 0 ( Y ) F : M P 0 ( Y ) F:M rarrP_(0)(Y)F: M \rightarrow \mathscr{P}_{0}(Y)F:MP0(Y) an ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-concex function, then the following statements are equivalent :
1 F 1 F 1^(@)F1^{\circ} F1F is both continuous and pointwise bounded on M M MMM.
2 F 2 F 2^(@)F2^{\circ} F2F is both lower semicontinuous and pointwise bounded on M M MMM.
3 F 3 F 3^(@)F^(')3^{\circ} F^{\prime}3F is both upper semicontinuous and pointwise bounded on M M MMM.
4 F 4 F 4^(@)F4^{\circ} F4F is both locally prebounded and pointwise bounded on M M MMM.
5 F 5 F 5^(@)F5^{\circ} F5F is locally bounded on M M MMM.
6 F 6 F 6^(@)F6^{\circ} F6F is both uniformly and pointwise bounded on M M MMM.
Proof. According to Theorem 3.1 we have only to show that the statements 5 5 5^(@)5^{\circ}5 and 6 6 6^(@)6^{\circ}6 are equivalent. Clearly, 5 5 5^(@)5^{\circ}5 implies 6 6 6^(@)6^{\circ}6. So it remains to prove that 6 6 6^(@)6^{\circ}6 implies 5 5 5^(@)5^{\circ}5.
Fix any point x 0 M x 0 M x_(0)in Mx_{0} \in Mx0M. Let V V VVV be a neighbourhood of the origin of Y Y YYY. Take a balanced neighbourhood V 0 V 0 V_(0)V_{0}V0 of the origin of Y Y YYY such that V 0 + V 0 V V 0 + V 0 V V_(0)+V_(0)sub VV_{0}+V_{0} \subset VV0+V0V. Since F F FFF is uniformly bounded on M M MMM, there exist for V 0 V 0 V_(0)V_{0}V0 a positive integer m m mmm and a nonempty open subset T T TTT of M M MMM such that F ( T ) m V 0 F ( T ) m V 0 F(T)sub mV_(0)F(T) \subset m V_{0}F(T)mV0.
Select a point t 0 T t 0 T t_(0)in Tt_{0} \in Tt0T. Since M M MMM is a neighbourhood of t 0 t 0 t_(0)t_{0}t0 and zero a cluster point of A A AAA, we can find a number a A a A a in Aa \in AaA such that the point
y = t 0 + a 1 a ( t 0 x 0 ) y = t 0 + a 1 a t 0 x 0 y=t_(0)+(a)/(1-a)(t_(0)-x_(0))y=t_{0}+\frac{a}{1-a}\left(t_{0}-x_{0}\right)y=t0+a1a(t0x0)
lies in M M MMM. Because F ( y ) F ( y ) F(y)F(y)F(y) is bounded, there exists a positive integer n n nnn such that
( 1 a a ) s F ( y ) n V 0 1 a a s F ( y ) n V 0 ((1-a)/(a))^(s)F(y)sub nV_(0)\left(\frac{1-a}{a}\right)^{s} F(y) \subset n V_{0}(1aa)sF(y)nV0
Finally, choose a positive integer p p ppp such that
m a s p m a s p (m)/(a^(s)) <= p\frac{m}{a^{s}} \leqslant pmasp
Then the set
U = a 1 a y + 1 a T U = a 1 a y + 1 a T U=(a-1)/(a)y+(1)/(a)TU=\frac{a-1}{a} y+\frac{1}{a} TU=a1ay+1aT
satisfies the following inequality :
(3.3) F ( U M ) ( n + p ) V (3.3) F ( U M ) ( n + p ) V {:(3.3)F(U nn M)sub(n+p)V:}\begin{equation*} F(U \cap M) \subset(n+p) V \tag{3.3} \end{equation*}(3.3)F(UM)(n+p)V
Indeed, if x x xxx is in U M U M U nn MU \cap MUM, then there exists t T t T t in Tt \in TtT such that
a = a 1 a y + 1 a b a = a 1 a y + 1 a b a=(a-1)/(a)y+(1)/(a)ba=\frac{a-1}{a} y+\frac{1}{a} ba=a1ay+1ab
and consequently we have t ( 1 a ) y + a x t ( 1 a ) y + a x t≐(1-a)y+axt \doteq(1-a) y+a xt(1a)y+ax. By the ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-convexity of F F FFF it follows that
( 1 a ) s F ( y ) + a s F ( x ) F ( t ) m V 0 . ( 1 a ) s F ( y ) + a s F ( x ) F ( t ) m V 0 . (1-a)^(s)F^(')(y)+a^(s)F(x)sub F(t)sub mV_(0).(1-a)^{s} F^{\prime}(y)+a^{s} F(x) \subset F(t) \subset m V_{0} .(1a)sF(y)+asF(x)F(t)mV0.
which implies that
( 1 a a ) s F ( y ) + F ( x ) m a s V 0 p V 0 1 a a s F ( y ) + F ( x ) m a s V 0 p V 0 ((1-a)/(a))^(s)F(y)+F(x)sub(m)/(a^(s))V_(0)sub pV_(0)\left(\frac{1-a}{a}\right)^{s} F(y)+F(x) \subset \frac{m}{a^{s}} V_{0} \subset p V_{0}(1aa)sF(y)+F(x)masV0pV0
So we obtain
F ( x ) ( 1 a a ) s F ( y ) + F ( x ) ( 1 a a ) s F ( y ) p V 0 n V 0 ( n + p ) ( V 0 + V 0 ) ( n + p ) V F ( x ) 1 a a s F ( y ) + F ( x ) 1 a a s F ( y ) p V 0 n V 0 ( n + p ) V 0 + V 0 ( n + p ) V {:[F(x)sub((1-a)/(a))^(s)F(y)+F(x)-((1-a)/(a))^(s)F(y)sub],[sub pV_(0)-nV_(0)sub(n+p)(V_(0)+V_(0))sub(n+p)V]:}\begin{gathered} F(x) \subset\left(\frac{1-a}{a}\right)^{s} F(y)+F(x)-\left(\frac{1-a}{a}\right)^{s} F(y) \subset \\ \subset p V_{0}-n V_{0} \subset(n+p)\left(V_{0}+V_{0}\right) \subset(n+p) V \end{gathered}F(x)(1aa)sF(y)+F(x)(1aa)sF(y)pV0nV0(n+p)(V0+V0)(n+p)V
Thus (3.3) is true, as elaimed.
On the other hand, notice that
U = x 0 + 1 a ( T t 0 ) . U = x 0 + 1 a T t 0 . U=x_(0)+(1)/(a)(T-t_(0)).U=x_{0}+\frac{1}{a}\left(T-t_{0}\right) .U=x0+1a(Tt0).
Therefore U U UUU is a neighbourhood of x 0 x 0 x_(0)x_{0}x0. Together with (3.3) this remark expresses that F F F^(')F^{\prime}F is locally bounded at x 0 x 0 x_(0)x_{0}x0.
Corollary 3.4. Let X X XXX and Y Y YYY be real topological linear spaces, let M M MMM be a nonempty open convex subset of X X XXX, and let I A : M P 0 ( Y ) I A : M P 0 ( Y ) I^(A):M rarrP_(0)(Y)I^{A}: M \rightarrow \mathscr{P}_{0}(Y)IA:MP0(Y) be an ( A , s ) ( A , s ) (A,s)(A, s)(A,s) convex function which is pointwise bounded on M M MMM and for which there exists a point x 0 M x 0 M x_(0)in Mx_{0} \in Mx0M such that one (and hence all) of the following statements is true:
(i) F F FFF is continuous at x 0 x 0 x_(0)x_{0}x0 :
(ii) F F FFF is lower semicontinuous at x 0 x 0 x_(0)x_{0}x0;
(iii) F F FFF is upper semicontinuous at x 0 x 0 x_(0)x_{0}x0;
(iv) F F FFF is locally prebounded at x 0 x 0 x_(0)x_{0}x0;
(v) F F FFF is locally bounded at x 0 x 0 x_(0)x_{0}x0.
Then F F FFF is continuous on M M MMM.
Proof. The function F F FFF is uniformly bounded on M M MMM, because it is locally bounded at x 0 x 0 x_(0)x_{0}x0. By applying Theorem 3.3 it follows that F F FFF is continuous on M M MMM.
Corollary 3.5. Let X X XXX and Y Y YYY be real topological linear spaces, let M M MMM be a nonempty open convex subset of X X XXX, and let F : M P 0 ( Y ) F : M P 0 ( Y ) F:M rarrP_(0)(Y)F: M \rightarrow \mathscr{P}_{0}(Y)F:MP0(Y) be an A , s ) A , s {:A,s)\left.A, s\right)A,s)-convex function which is pointwise bounded on M M MMM and for which there exists a nonempty open subset T T TTT of M M MMM such that F ( T ) F ( T ) F(T)F(T)F(T) is bounded. Then F F FFF is continous on M M MMM.
Proof. Since F ( T ) F ( T ) F(T)F(T)F(T) is bounded, the function F F FFF is uniformly bounded on M M MMM. By applying Theorem 3.3 it follows that F F FFF is continuous on M M MMM.

4. Continuity of ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-eoneave set-valued functions

We start by proving an auxiliary lemma.
Lemma 4.1. If X X XXX and Y Y YYY are real topological linear spaces, M M MMM a convex subset of X , x 0 X , x 0 X,x_(0)X, x_{0}X,x0 an interior point of M M MMM, and F : M P 0 ( Y ) F : M P 0 ( Y ) F:M rarrP_(0)(Y)F: M \rightarrow \mathscr{P}_{0}(Y)F:MP0(Y) an ( A , s ) ( A , s ) (A,s)(A, s)(A,s) -
concave function which is bounded at x 0 x 0 x_(0)x_{0}x0, then the following statements are equivalent:
1 F 1 F 1^(@)F1^{\circ} F1F is upper semicontinuous at x 0 x 0 x_(0)x_{0}x0.
2 2 2^(@)2^{\circ}2 For every neighbourhood V V VVV of the origin of Y Y YYY there exist a positive integer m m mmm and a neighbourhood U U UUU of x 0 x 0 x_(0)x_{0}x0 such that F ( U M ) F ( x 0 ) + m V F ( U M ) F x 0 + m V F(U nn M)sub F(x_(0))+mVF(U \cap M) \subset F\left(x_{0}\right)+m VF(UM)F(x0)+mV.
3 3 3^(@)3^{\circ}3 For every neighbourhood V V VVV of the origin of Y Y YYY there exist a positive integer n n nnn and a neighbourhood U U UUU of x 0 x 0 x_(0)x_{0}x0 such that
(4.1) F ( U M ) F ( x 0 ) n V (4.1) F ( U M ) F x 0 n V {:(4.1)F(U nn M)-F(x_(0))sub nV:}\begin{equation*} F(U \cap M)-F\left(x_{0}\right) \subset n V \tag{4.1} \end{equation*}(4.1)F(UM)F(x0)nV
Proof. The implication 1 2 1 2 1^(@)=>2^(@)1^{\circ} \Rightarrow 2^{\circ}12 is obvious. In order to prove that 2 3 2 3 2^(@)=>3^(@)2^{\circ} \Rightarrow 3^{\circ}23 holds, consider any neighbourhood V V VVV of the origin of Y Y YYY. Then choose a balanced neighbourhood V 0 V 0 V_(0)V_{0}V0 of the origin of Y Y YYY such that V 0 + + V 0 V V 0 + + V 0 V V_(0)++V_(0)sub VV_{0}+ +V_{0} \subset VV0++V0V. By 2 2 2^(@)2^{\circ}2 there exist a positive integer m m mmm and a neighbourhood U U UUU of x 0 x 0 x_(0)x_{0}x0 such that
(4.2) F ( U M ) F ( x 0 ) + m V 0 . (4.2) F ( U M ) F x 0 + m V 0 . {:(4.2)F(U nn M)sub F(x_(0))+mV_(0).:}\begin{equation*} F(U \cap M) \subset F\left(x_{0}\right)+m V_{0} . \tag{4.2} \end{equation*}(4.2)F(UM)F(x0)+mV0.
On the other hand, since F ( x 0 ) F ( x 0 ) F x 0 F x 0 F(x_(0))-F(x_(0))F\left(x_{0}\right)-F\left(x_{0}\right)F(x0)F(x0) is a bounded set, there exists a positive integer p p ppp such that
(4.3) F ( x 0 ) F ( x 0 ) p V 0 . (4.3) F x 0 F x 0 p V 0 . {:(4.3)F(x_(0))-F(x_(0))sub pV_(0).:}\begin{equation*} F\left(x_{0}\right)-F\left(x_{0}\right) \subset p V_{0} . \tag{4.3} \end{equation*}(4.3)F(x0)F(x0)pV0.
From (4.2) and (4.3) we obtain
F ( U M ) F ( x 0 ) F ( x 0 ) F ( x 0 ) + m V 0 p V 0 + m V 0 ( p + m ) ( V 0 + V 0 ) ( p + m ) V . F ( U M ) F x 0 F x 0 F x 0 + m V 0 p V 0 + m V 0 ( p + m ) V 0 + V 0 ( p + m ) V . {:[F(U nn M)-F(x_(0)) sub F(x_(0))-F(x_(0))+mV_(0)sub pV_(0)+mV_(0)sub],[ sub(p+m)(V_(0)+V_(0))sub(p+m)V.]:}\begin{aligned} F(U \cap M)-F\left(x_{0}\right) & \subset F\left(x_{0}\right)-F\left(x_{0}\right)+m V_{0} \subset p V_{0}+m V_{0} \subset \\ & \subset(p+m)\left(V_{0}+V_{0}\right) \subset(p+m) V . \end{aligned}F(UM)F(x0)F(x0)F(x0)+mV0pV0+mV0(p+m)(V0+V0)(p+m)V.
If we set n = p + m n = p + m n=p+mn=p+mn=p+m, we can conclude that (4.1) holds. Thus the implication 2 3 2 3 2^(@)=>3^(@)2^{\circ} \Rightarrow 3^{\circ}23 is true.
Suppose now that 3 3 3^(@)3^{\circ}3 holds. Let V V VVV be a neighbourhood of the origin of Y Y YYY. Choose a balanced neighbourhood V 0 V 0 V_(0)V_{0}V0 of the origin of Y Y YYY such that V 0 + V 0 V V 0 + V 0 V V_(0)+V_(0)sub VV_{0}+V_{0} \subset VV0+V0V. By 3 3 3^(@)3^{\circ}3 there exist a positive integer n n nnn and a neighbourhood U 0 U 0 U_(0)U_{0}U0 of x 0 x 0 x_(0)x_{0}x0 such that
F ( U 0 M ) F ( x 0 ) n V 0 . F U 0 M F x 0 n V 0 . F(U_(0)nn M)-F(x_(0))sub nV_(0).F\left(U_{0} \cap M\right)-F\left(x_{0}\right) \subset n V_{0} .F(U0M)F(x0)nV0.
Next choose a positive integer p p ppp for which F ( x 0 ) p V 0 F x 0 p V 0 F(x_(0))sub pV_(0)F\left(x_{0}\right) \subset p V_{0}F(x0)pV0 holds. Taking into account that zero is a cluster point of A A AAA, we can find an a A a A a in Aa \in AaA satisfying both
a s n 1 and | ( 1 a ) s + a s 1 | p 1 . a s n 1  and  ( 1 a ) s + a s 1 p 1 . a^(s)n <= 1" and "|(1-a)^(s)+a^(s)-1|p <= 1.a^{s} n \leqslant 1 \text { and }\left|(1-a)^{s}+a^{s}-1\right| p \leqslant 1 .asn1 and |(1a)s+as1|p1.
Note that U = ( 1 a ) x 0 + a ( U 0 M ) U = ( 1 a ) x 0 + a U 0 M U=(1-a)x_(0)+a(U_(0)nn M)U=(1-a) x_{0}+a\left(U_{0} \cap M\right)U=(1a)x0+a(U0M) is a neighbourhood of x 0 x 0 x_(0)x_{0}x0 contained in M M MMM. Let x x xxx be a point in U U UUU. Then there exists y U 0 M y U 0 M y inU_(0)nn My \in U_{0} \cap MyU0M such that x = ( 1 a ) x 0 + a y x = ( 1 a ) x 0 + a y x=(1-a)x_(0)+ayx=(1-a) x_{0}+a yx=(1a)x0+ay, and hence we have
F ( x ) ( 1 a ) s F ( x 0 ) + a s F ( y ) F ( x 0 ) + a s [ F ( y ) F ( x 0 ) ] + [ ( 1 a ) s + a s 1 ] F ( x 0 ) F ( x 0 ) + a s n 0 + [ ( 1 a ) s + a s 1 ] p V 0 F ( x 0 ) + V 0 + V 0 F ( x 0 ) + V . F ( x ) ( 1 a ) s F x 0 + a s F ( y ) F x 0 + a s F ( y ) F x 0 + ( 1 a ) s + a s 1 F x 0 F x 0 + a s n 0 + ( 1 a ) s + a s 1 p V 0 F x 0 + V 0 + V 0 F x 0 + V . {:[F(x)sub(1-a)^(s)F(x_(0))+a^(s)F(y)sub],[sub F(x_(0))+a^(s)[F(y)-F(x_(0))]+[(1-a)^(s)+a^(s)-1]F(x_(0))sub],[sub F(x_(0))+a^(s)ngrad_(0)+[(1-a)^(s)+a^(s)-1]pV_(0)sub],[sub F(x_(0))+V_(0)+V_(0)sub F(x_(0))+V.]:}\begin{gathered} F(x) \subset(1-a)^{s} F\left(x_{0}\right)+a^{s} F(y) \subset \\ \subset F\left(x_{0}\right)+a^{s}\left[F(y)-F\left(x_{0}\right)\right]+\left[(1-a)^{s}+a^{s}-1\right] F\left(x_{0}\right) \subset \\ \subset F\left(x_{0}\right)+a^{s} n \nabla_{0}+\left[(1-a)^{s}+a^{s}-1\right] p V_{0} \subset \\ \subset F\left(x_{0}\right)+V_{0}+V_{0} \subset F\left(x_{0}\right)+V . \end{gathered}F(x)(1a)sF(x0)+asF(y)F(x0)+as[F(y)F(x0)]+[(1a)s+as1]F(x0)F(x0)+asn0+[(1a)s+as1]pV0F(x0)+V0+V0F(x0)+V.
Since x x xxx was arbitrarily chosen in U U UUU, we have F ( x ) F ( x 0 ) + V F ( x ) F x 0 + V F(x)sub F(x_(0))+VF(x) \subset F\left(x_{0}\right)+VF(x)F(x0)+V for all x U x U x in Ux \in UxU. Thus F F FFF is upper semicontinuous at x 0 x 0 x_(0)x_{0}x0. Consequently, the impllication 3 1 3 1 3^(@)=>1^(@)3^{\circ} \Rightarrow 1^{\circ}31 is also true.
Like in Section 3 we begin with the investigation of the continuity of an ( A , s A , s A,sA, sA,s )-concave set-valued function at an interior point of its domain.
Theorem 4.2. If X X XXX and Y Y YYY are real topological linear spaces, M M MMM a convex subset of X , x 0 X , x 0 X,x_(0)X, x_{0}X,x0 an interior point of M M MMM, and F : M P 0 ( Y ) F : M P 0 ( Y ) F:M rarrP_(0)(Y)F: M \rightarrow \mathscr{P}_{0}(Y)F:MP0(Y) an ( A , s ) ( A , s ) (A,s)(A, s)(A,s) concave function, then the following statements are equivalent:
1 F 1 F 1^(@)F1^{\circ} F1F is both continuous and bounded at x 0 x 0 x_(0)x_{0}x0.
2 F 2 F 2^(@)F2^{\circ} F2F is both upper semicontinuous and bounded at x 0 x 0 x_(0)x_{0}x0.
3 F 3 F 3^(@)F3^{\circ} F3F is locally bounded at x 0 x 0 x_(0)x_{0}x0.
Proof. The implication 1 2 1 2 1^(@)=>2^(@)1^{\circ} \Rightarrow 2^{\circ}12 is obvious, while the implication 2 3 2 3 2^(@)=>3^(@)2^{\circ} \Rightarrow 3^{\circ}23 has been stated in Lemma 2.1. Therefore it is sufficient to prove that 3 3 3^(@)3^{\circ}3 implies 1 1 1^(@)1^{\circ}1.
If F F FFF is locally bounded at x 0 x 0 x_(0)x_{0}x0, it is evident that F F FFF is bounded at x 0 x 0 x_(0)x_{0}x0. Moreover, F F FFF is lower semicontinuous at x 0 x 0 x_(0)x_{0}x0. Indeed, let V V VVV be any neighbourhood of the origin of Y Y YYY. Choose a balanced neighbourhood V 0 V 0 V_(0)V_{0}V0 of the origin of Y Y YYY such that V 0 + V 0 V V 0 + V 0 V V_(0)+V_(0)sub VV_{0}+V_{0} \subset VV0+V0V. Since F F FFF is locally bounded at x 0 x 0 x_(0)x_{0}x0, there exist a positive integer n n nnn and a neighbourhood U 0 U 0 U_(0)U_{0}U0 of x 0 x 0 x_(0)x_{0}x0 such that
(4.4) F ( U 0 M ) n V 0 (4.4) F U 0 M n V 0 {:(4.4)F(U_(0)nn M)sub nV_(0):}\begin{equation*} F\left(U_{0} \cap M\right) \subset n V_{0} \tag{4.4} \end{equation*}(4.4)F(U0M)nV0
Taking into account that zero is a cluster point of A A AAA, we can find an a A a A a in Aa \in AaA satisfying the following inequalities :
a < 1 2 , a s n 1 , [ 1 ( 1 a ) s ] n 1 a < 1 2 , a s n 1 , 1 ( 1 a ) s n 1 a < (1)/(2),quada^(s)n <= 1,quad[1-(1-a)^(s)]n <= 1a<\frac{1}{2}, \quad a^{s} n \leqslant 1, \quad\left[1-(1-a)^{s}\right] n \leqslant 1a<12,asn1,[1(1a)s]n1
Select now a balanced neighbourhood W W WWW of the origin of X X XXX such that x 0 + W U 0 M x 0 + W U 0 M x_(0)+W subU_(0)nn Mx_{0}+W \subset U_{0} \cap Mx0+WU0M, and then set
U = x 0 + a a 1 W . U = x 0 + a a 1 W . U=x_(0)+(a)/(a-1)W.U=x_{0}+\frac{a}{a-1} W .U=x0+aa1W.
Obviously, U U UUU is a neighbourhood of x 0 x 0 x_(0)x_{0}x0 contained in U 0 M U 0 M U_(0)nn MU_{0} \cap MU0M. Let x x xxx be a point in U U UUU. Then there exists y W y W y in Wy \in WyW such that
x = x 0 + a a 1 y . x = x 0 + a a 1 y . x=x_(0)+(a)/(a-1)y.x=x_{0}+\frac{a}{a-1} y .x=x0+aa1y.
Accordingly we have x 0 = ( 1 a ) x + a ( x 0 + y ) x 0 = ( 1 a ) x + a x 0 + y x_(0)=(1-a)x+a(x_(0)+y)x_{0}=(1-a) x+a\left(x_{0}+y\right)x0=(1a)x+a(x0+y). In view of the ( A , s ) ( A , s ) (A,s)(A, s)(A,s) concavity of F F FFF it follows that
(4.5) F ( x 0 ) ( 1 a ) s F ( x ) + a s F ( x 0 + y ) F ( x ) + + [ ( 1 a ) s 1 ] F ( x ) + a s F ( x 0 + y ) . (4.5) F x 0 ( 1 a ) s F ( x ) + a s F x 0 + y F ( x ) + + ( 1 a ) s 1 F ( x ) + a s F x 0 + y . {:[(4.5)F(x_(0)) sub(1-a)^(s)F(x)+a^(s)F(x_(0)+y)sub F(x)+],[+[(1-a)^(s)-1]F^(')(x)+a^(s)F(x_(0)+y).]:}\begin{align*} F\left(x_{0}\right) & \subset(1-a)^{s} F(x)+a^{s} F\left(x_{0}+y\right) \subset F(x)+ \tag{4.5}\\ & +\left[(1-a)^{s}-1\right] F^{\prime}(x)+a^{s} F\left(x_{0}+y\right) . \end{align*}(4.5)F(x0)(1a)sF(x)+asF(x0+y)F(x)++[(1a)s1]F(x)+asF(x0+y).
But the points x x xxx and x 0 + y x 0 + y x_(0)+yx_{0}+yx0+y belong to U 0 M U 0 M U_(0)nn MU_{0} \cap MU0M. Therefore (4.4) and (4.5) yield
F ( x 0 ) F ( x ) + [ ( 1 a ) s 1 ] n V 0 + a s n V 0 F ( x ) + V 0 + + V 0 F ( x ) + V . F x 0 F ( x ) + ( 1 a ) s 1 n V 0 + a s n V 0 F ( x ) + V 0 + + V 0 F ( x ) + V . {:[F(x_(0))sub F(x)+[(1-a)^(s)-1]nV_(0)+a^(s)nV_(0)sub F(x)+V_(0)+],[+V_(0)sub F(x)+V.]:}\begin{aligned} F\left(x_{0}\right) \subset F(x)+\left[(1-a)^{s}-1\right] n V_{0}+a^{s} n V_{0} \subset F(x)+V_{0}+ \\ +V_{0} \subset F(x)+V . \end{aligned}F(x0)F(x)+[(1a)s1]nV0+asnV0F(x)+V0++V0F(x)+V.
Since x x xxx was arbitrarily chosen in U U UUU, we have F ( x 0 ) F ( x ) + V F x 0 F ( x ) + V F(x_(0))sub F(x)+VF\left(x_{0}\right) \subset F(x)+VF(x0)F(x)+V for all x U x U x in Ux \in UxU. Hence F F FFF is lower semicontinuous at x 0 x 0 x_(0)x_{0}x0.
Finally, we prove that F F FFF is also upper semicontinuous at x 0 x 0 x_(0)x_{0}x0. Let V V VVV be any neighbourhood of the origin of Y Y YYY. Choose a neighbourhood V 0 V 0 V_(0)V_{0}V0 of the origin of Y Y YYY such that V 0 V 0 V V 0 V 0 V V_(0)-V_(0)sub VV_{0}-V_{0} \subset VV0V0V. Since F F FFF is locally bounded at x 0 x 0 x_(0)x_{0}x0, there exist a positive integer n n nnn and a neighbourhood U U UUU of x 0 x 0 x_(0)x_{0}x0 such that
(4.6) F ( U M ) n V 0 . (4.6) F ( U M ) n V 0 . {:(4.6)F(U nn M)sub nV_(0).:}\begin{equation*} F(U \cap M) \subset n V_{0} . \tag{4.6} \end{equation*}(4.6)F(UM)nV0.
In particular, we have
(4.7) F ( x 0 ) n V 0 . (4.7) F x 0 n V 0 . {:(4.7)F(x_(0))sub nV_(0).:}\begin{equation*} F\left(x_{0}\right) \subset n V_{0} . \tag{4.7} \end{equation*}(4.7)F(x0)nV0.
From (4.6) and (4.7) it results that
F ( U M ) F ( x 0 ) n V 0 n V 0 n V . F ( U M ) F x 0 n V 0 n V 0 n V . F(U nn M)-F(x_(0))sub nV_(0)-nV_(0)sub nV.F(U \cap M)-F\left(x_{0}\right) \subset n V_{0}-n V_{0} \subset n V .F(UM)F(x0)nV0nV0nV.
By applying Lemma 4.1 it follows that F F FFF is upper semicontinuous at x 0 x 0 x_(0)x_{0}x0.
Being both lower semicontinuous and upper semicontinuous at x 0 x 0 x_(0)x_{0}x0, the function F F FFF is continuous at x 0 x 0 x_(0)x_{0}x0. Hence the implication 3 1 3 1 3^(@)=>1^(@)3^{\circ} \Rightarrow 1^{\circ}31 is proved.
Corollary 4.3. Let X X XXX and Y Y YYY be real topological linear spaces, let M M MMM be a convex subset of X X XXX, let x 0 x 0 x_(0)x_{0}x0 be an interior point of M M MMM, and let F : M P 0 ( Y ) F : M P 0 ( Y ) F:M rarrP_(0)(Y)F: M \rightarrow \mathscr{P}_{0}(Y)F:MP0(Y) be an ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-concave function. Then F F FFF is both continuous and bounded at x 0 x 0 x_(0)x_{0}x0 if and only if there exists a function G : M P 0 ( Y ) G : M P 0 ( Y ) G:M rarrP_(0)(Y)G: M \rightarrow \mathscr{P}_{0}(Y)G:MP0(Y) which satisfies the following conditions:
(i) G G GGG is locally bounded at x 0 x 0 x_(0)x_{0}x0;
(ii) F ( x ) G ( x ) F ( x ) G ( x ) F(x)sub G(x)F(x) \subset G(x)F(x)G(x) for all x M x M x in Mx \in MxM.
Proof. Necessity. Obvious in view of Theorem 4.2.
Sufficiency. It is evident that F F FFF is locally bounded at x 0 x 0 x_(0)x_{0}x0. By applying Theorem 4.2 it result that F F FFF is both continuous and bounded at x 0 x 0 x_(0)x_{0}x0.
Concernig the continuity of an ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-concave set-valued function on its whole domain we have the following results.
Theorem 4.4. If X X XXX and Y Y YYY are real topological linear spaces, M M MMM a nonempty open convex subset of X X XXX, and F : M P 0 ( Y ) F : M P 0 ( Y ) F^('):M rarrP_(0)(Y)F^{\prime}: M \rightarrow \mathscr{P}_{0}(Y)F:MP0(Y) and ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-concave function, then the following statements are equivalent:
1 F 1 F 1^(@)F1^{\circ} F1F is both continuous and pointwise bounded on M M MMM.
2 F 2 F 2^(@)F2^{\circ} F2F is both upper semicontinuous and pointwise bounded on M M MMM.
3 F 3 F 3^(@)F^(')3^{\circ} F^{\prime}3F is locally bounded on M M MMM.
4 F 4 F 4^(@)F4^{\circ} F4F is both uniformly and pointwise bounded on M M MMM.
Proof. According to Theorem 4.2 we have only to show that the statements 3 3 3^(@)3^{\circ}3 and 4 4 4^(@)4^{\circ}4 are equivalent. Clearly, 3 3 3^(@)3^{\circ}3 implies 4 4 4^(@)4^{\circ}4. So it remains to prove that 4 4 4^(@)4^{\circ}4 implies 3 3 3^(@)3^{\circ}3.
Fix any point x 0 M x 0 M x_(0)in Mx_{0} \in Mx0M. Let V V VVV be a neighbourhood of the origin of Y Y YYY. Take a balanced neighbourhood V 0 V 0 V_(0)V_{0}V0 of the origin of Y Y YYY such that V 0 + V 0 V V 0 + V 0 V V_(0)+V_(0)sub VV_{0}+V_{0} \subset VV0+V0V. Then there exist a positive integer m m mmm and a nonempty open subset T T TTT of M M MMM such that F ( T ) m V 0 F ( T ) m V 0 F(T)sub mV_(0)F(T) \subset m V_{0}F(T)mV0.
Choose a point t 0 T t 0 T t_(0)in Tt_{0} \in Tt0T. Since M M MMM is a neighbourhood of x 0 x 0 x_(0)x_{0}x0 and zero a cluster point of A A AAA, we can find a number a A a A a in Aa \in AaA such that the point
y = x 0 + a 1 a ( x 0 t 0 ) y = x 0 + a 1 a x 0 t 0 y=x_(0)+(a)/(1-a)(x_(0)-t_(0))y=x_{0}+\frac{a}{1-a}\left(x_{0}-t_{0}\right)y=x0+a1a(x0t0)
lies in M M MMM. Because F ( y ) F ( y ) F(y)F(y)F(y) is bounded, there exists a positive integer n n nnn such that F ( y ) n V 0 F ( y ) n V 0 F(y)sub nV_(0)F(y) \subset n V_{0}F(y)nV0. Set now U = ( 1 a ) y + a T U = ( 1 a ) y + a T U=(1-a)y+aTU=(1-a) y+a TU=(1a)y+aT. By the convexity of M M MMM we conclude that U U UUU is contained in M M MMM. Furthermore, the inclusion
holds.
(4.8) F ( U ) ( m + n ) V (4.8) F ( U ) ( m + n ) V {:(4.8)F(U)sub(m+n)V:}\begin{equation*} F(U) \subset(m+n) V \tag{4.8} \end{equation*}(4.8)F(U)(m+n)V
Indeed, if x x xxx is in U U UUU, then there exists t T t T t in Tt \in TtT such that x = ( 1 a ) y + a t x = ( 1 a ) y + a t x=(1-a)y+atx=(1-a) y+a tx=(1a)y+at. Therefore it follows that
F ( x ) ( 1 a ) s F ( y ) + a s F ( t ) ( 1 a ) s n V 0 + a s m V 0 m V 0 + n V 0 ( m + n ) ( V 0 + V 0 ) ( m + n ) V F ( x ) ( 1 a ) s F ( y ) + a s F ( t ) ( 1 a ) s n V 0 + a s m V 0 m V 0 + n V 0 ( m + n ) V 0 + V 0 ( m + n ) V {:[F(x) sub(1-a)^(s)F(y)+a^(s)F(t)sub(1-a)^(s)nV_(0)+a^(s)mV_(0)sub],[ sub mV_(0)+nV_(0)sub(m+n)(V_(0)+V_(0))sub(m+n)V]:}\begin{aligned} F(x) & \subset(1-a)^{s} F(y)+a^{s} F(t) \subset(1-a)^{s} n V_{0}+a^{s} m V_{0} \subset \\ & \subset m V_{0}+n V_{0} \subset(m+n)\left(V_{0}+V_{0}\right) \subset(m+n) V \end{aligned}F(x)(1a)sF(y)+asF(t)(1a)snV0+asmV0mV0+nV0(m+n)(V0+V0)(m+n)V
Thus (4.8) is true.
On the other hand, notice that the equality U = x 0 + a ( T t 0 ) U = x 0 + a T t 0 U=x_(0)+a(T-t_(0))U=x_{0}+a\left(T-t_{0}\right)U=x0+a(Tt0) holds. Therefore U U UUU is a neighbourhood of x 0 x 0 x_(0)x_{0}x0. Together with (4.8) this remark expresses that F F FFF is locally bounded at x 0 x 0 x_(0)x_{0}x0.
Corollary 4.5. Let X X XXX and Y Y YYY be real topological linear spaces, let M M MMM be a nonempty open convex subset of X X XXX, and let F : M P 0 ( Y ) F : M P 0 ( Y ) F:M rarrP_(0)(Y)F: M \rightarrow \mathscr{P}_{0}(Y)F:MP0(Y) be an ( A , s ) ( A , s ) (A,s)(A, s)(A,s) --concave function which is pointwise bounded on M M MMM and for which there exists a point x 0 M x 0 M x_(0)in Mx_{0} \in Mx0M such that one (and hence all) of the following properties. holds:
(i) F F FFF is continuous at x 0 x 0 x_(0)x_{0}x0;
(ii) F F FFF is upper semicontinuous at x 0 x 0 x_(0)x_{0}x0;
(iii) F F FFF is locally bounded at x 0 x 0 x_(0)x_{0}x0.
Then F F FFF is continuous on M M MMM.
Proof. The function F F FFF is uniformly bounded on M M MMM, because it is locally bounded at x 0 x 0 x_(0)x_{0}x0. By applying Theorem 4.4 it follows that F F FFF is continous on M M MMM.
Corollary 4.6. Let X X XXX and Y Y YYY be real topological linear spaces, let M M MMM be a nonempty open convex subset of X X XXX, and let F : M P 0 ( Y ) F : M P 0 ( Y ) F:M rarrP_(0)(Y)F: M \rightarrow \mathscr{P}_{0}(Y)F:MP0(Y) be an ( A , s ) ( A , s ) (A,s)(A, s)(A,s) --concave function which is pointwise bounded on M M MMM and for which there exists a nonempty open subset T T TTT of M M MMM such that F ( T ) F ( T ) F(T)F(T)F(T) is bounded. Then F F FFF is continuous on M M MMM.
Proof. Since F ( T ) F ( T ) F(T)F(T)F(T) is bounded, the function F F FFF is uniformly bounded on M M MMM. By applying Theorem 4.4 it follows that F F FFF is continuous on M M MMM. 圆
Theorem 4.7. Let X X XXX be a real topological linear space of the second category, let Y Y YYY be a real topological linear space, let M M MMM be a nonempty open convex subset of X X XXX, and let F : M P 0 ( Y ) F : M P 0 ( Y ) F^('):M rarrP_(0)(Y)F^{\prime}: M \rightarrow \mathscr{P}_{0}(Y)F:MP0(Y) be an ( A , s ) ( A , s ) (A,s)(A, s)(A,s)-concave function which is pointwise bounded on M M MMM. Then F F FFF is continuous on M M MMM if and only if it is lower semicontinuous on M M MMM.
Proof. Necessity. Obvious.
Sufficiency. Let V V VVV be any neighbourhood of the origin of Y Y YYY. Take a closed neighbourhood V 0 V 0 V_(0)V_{0}V0 of the origin of Y Y YYY such that V 0 V V 0 V V_(0)sub VV_{0} \subset VV0V, and then set
M n = { x M : F ( x ) n V 0 } M n = x M : F ( x ) n V 0 M_(n)={x in M:F(x)sub nV_(0)}M_{n}=\left\{x \in M: F(x) \subset n V_{0}\right\}Mn={xM:F(x)nV0}
for each positive integer n n nnn. Notice that all the sets M n ( n N ) M n ( n N ) M_(n)(n in N)M_{n}(n \in N)Mn(nN) are closed in the induced topology on M M MMM. Indeed, fix any positive integer n n nnn. Let x 0 x 0 x_(0)x_{0}x0 be a point in M M n M M n M\\M_(n)M \backslash M_{n}MMn. By the definition of M n M n M_(n)M_{n}Mn it follows that
I ( x 0 ) ( Y n V 0 ) . I x 0 Y n V 0 . I(x_(0))nn(Y\\nV_(0))!=O/.I\left(x_{0}\right) \cap\left(Y \backslash n V_{0}\right) \neq \emptyset .I(x0)(YnV0).
Choose a point y F ( x 0 ) ( Y n V 0 ) y F x 0 Y n V 0 y in F(x_(0))nn(Y\\nV_(0))y \in F\left(x_{0}\right) \cap\left(Y \backslash n V_{0}\right)yF(x0)(YnV0). Since Y n V 0 Y n V 0 Y\\nV_(0)Y \backslash n V_{0}YnV0 is a nieghbourhood of y y yyy, there exists a neighbourhood W W WWW of the origin of Y Y YYY such that y W ⊂⊂ Y n V 0 y W ⊂⊂ Y n V 0 y-W⊂⊂Y\\nV_(0)y-W \subset \subset Y \backslash n V_{0}yW⊂⊂YnV0. Taking now into consideration that F F FFF is lower semicontinuous at x 0 x 0 x_(0)x_{0}x0, it follows that there is a neighbourhood U U UUU of x 0 x 0 x_(0)x_{0}x0 such that
F ( x 0 ) F ( x ) + W for all x U M . F x 0 F ( x ) + W  for all  x U M F(x_(0))sub F(x)+W" for all "x in U nn M". "F\left(x_{0}\right) \subset F(x)+W \text { for all } x \in U \cap M \text {. }F(x0)F(x)+W for all xUM
This inclusion implies that y F ( x ) + W y F ( x ) + W y in F(x)+Wy \in F(x)+WyF(x)+W for all x U M x U M x in U nn Mx \in U \cap MxUM. Hence there exists for each x U M x U M x in U nn Mx \in U \cap MxUM a pair ( z x , w x ) z x , w x (z_(x),w_(x))\left(z_{x}, w_{x}\right)(zx,wx) in F ( x ) × W F ( x ) × W F(x)xx WF(x) \times WF(x)×W such that y = z x + + w x y = z x + + w x y=z_(x)++w_(x)y=z_{x}+ +w_{x}y=zx++wx. So we have z x F ( x ) z x F ( x ) z_(x)in F(x)z_{x} \in F(x)zxF(x) as well as z x = y w x y W Y n V 0 z x = y w x y W Y n V 0 z_(x)=y-w_(x)in y-W sub Y\\nV_(0)z_{x}=y-w_{x} \in y-W \subset Y \backslash n V_{0}zx=ywxyWYnV0 for all x U M x U M x in U nn Mx \in U \cap MxUM. This shows that
F ( x ) ( Y n V 0 ) for all x U M . F ( x ) Y n V 0  for all  x U M . F(x)nn(Y\\nV_(0))!=O/" for all "x in U nn M.F(x) \cap\left(Y \backslash n V_{0}\right) \neq \emptyset \text { for all } x \in U \cap M .F(x)(YnV0) for all xUM.
Consequently, we have U M M M n U M M M n U nn M sub M\\M_(n)U \cap M \subset M \backslash M_{n}UMMMn. Therefore x 0 x 0 x_(0)x_{0}x0 is an interior point of M M n M M n MM_(n)M M_{n}MMn. Since x 0 x 0 x_(0)x_{0}x0 was arbitrarily chosen in M M n M M n M\\M_(n)M \backslash M_{n}MMn, the set M M n M M n M\\M_(n)M \backslash M_{n}MMn is open. Hence M n M n M_(n)M_{n}Mn is closed in the induced topology on M M MMM.
By a well-known result from topology (see A. Császár [2, p. 386, (9.1.11)]) the set M M MMM is of the second category in the induced topology. Taking into account that the pointwise boundedness of F F F^(')F^{\prime}F implies the equality.
M = n N M n M = n N M n M=uuu_(n in N)M_(n)M=\bigcup_{n \in N} M_{n}M=nNMn
we conclude that there is a positive integer n n nnn such that M n M n M_(n)M_{n}Mn has interior points in the induced topology on M M MMM. Thus there exist a point x 0 M x 0 M x_(0)in Mx_{0} \in Mx0M and an open neighbourhod U U UUU of x 0 x 0 x_(0)x_{0}x0 such that U M M n U M M n U nn M subM_(n)U \cap M \subset M_{n}UMMn. So it follows that F ( U M ) n V F ( U M ) n V F(U nn M)sub nVF(U \cap M) \subset n VF(UM)nV.
In conclusion, we have shown that for any neighbourhood V V VVV of the origin of Y Y YYY there exist a positive integer n n nnn and a nonempty open subset T T TTT of M M MMM such that F ( T ) n V F ( T ) n V F(T)sub nVF(T) \subset n VF(T)nV. In other words, we have shown that F F FFF is uniformly bounded on M M MMM. By Theorem 4.4 it follows that F F FFF is continuous on M M MMM. 圆
Acknowledgements. This paper was written while the author was visiting the Department of Mathematics of the University of Duisburg in 1991. He wishes to express his gratitude to both Deutscher Akademische Austauschdienst for the financial support of this visit and Professor Werner Haußmann for the kind assistance in Duisburg.
Note. After this paper has been finished the author was aware of the following paper related to the topic of the present paper : T. Cardinali and F. Papalini, Una estensione del concetto di midpoint convessità per multifunzioni. Riv. Mat. Univ. Parma (4) 15 (1989), 119-131.

REFERENCES

  1. J. M. Borwein, Convex relations in analysis and optimization. Generalized concavity in optimization and economics, pp. 335-377, edited by S. Schaible and W.T. Ziemba, Academic Press, New York, 1981.
  2. A. Császár , General topology. Akadémiai Kiadó, Budapest, 1978.
  3. K. Nikodem, On midpoint convex set-valued functions. Aequationes Math. 33 (1987), 46-56.
  4. K. Nikodem, On concave and midpoint concave set-valued functions. Glasnik Mat., Ser. III 22 (42) (1987), 69-76.
  5. W. Smajdor, Subadditive and subquadratic set-valued functions. Uniwersytet Slaski, Katowice, 1987.
  6. W. S majdor , Superadditive set-valued functions and Banach-Steinhaus theorem. Rad. Mat. 3 (1987), 203-214.
Received 1.X. 1993