Topology and its Applications 155 (2008) 830–850

www.elsevier.com/locate/topol

Topological convexities, selections and fixed points

Charles D. Horvath
Département de Mathématiques, Université de Perpignan, 66860 Perpignan cedex, France
Received 28 September 2006; received in revised form 4 January 2007; accepted 4 January 2007

Abstract
A convexity on a set X is a family of subsets of X which contains the whole space and the empty set as well as the singletons
and which is closed under arbitrary intersections and updirected unions. A uniform convex space is a uniform topological space
endowed with a convexity for which the convex hull operator is uniformly continuous. Uniform convex spaces with homotopically
trivial polytopes (convex hulls of finite sets) are absolute extensors for the class of metric spaces; if they are completely metrizable
then a continuous selection theorem à la Michael holds. Upper semicontinuous maps have approximate selections and fixed points,
under the usual assumptions.
© 2007 Elsevier B.V. All rights reserved.

Keywords: Uniform spaces; Generalized convexity; Continuous selections; Fixed points

1. Introduction

A convexity on a set X is an algebraic closure operator A → JAK from P(X) to P(X) such that J{x}K = {x} for
all x ∈ X or, equivalently, a family of subsets C of X, the convex sets, which contains the whole space and the empty
set as well as the singletons and which is closed under arbitrary intersections and updirected unions. Uniform convex
spaces, that is uniform topological spaces for which the convex hull operator is a uniformly continuous map, are
introduced in Section 2.1. The definition is slightly different from that given by Van de Vel in [26]; the language of
entourages is more natural and more suggestive when one is concerned with multivalued maps than the language of
coverings.
In [26] Van de Vel showed that a complete metric space endowed with a convexity which is compatible with
the metric, which has the Kakutani Property (i.e. disjoint convex sets C1 and C2 can be enlarged to disjoint convex
sets D1 and D2 such that D2 = X \ D1 ), and for which polytopes (i.e. sets of the form JAK where A is a finite set), are
compact and connected, is an absolute retract (Theorem 5.1 in [26]). That result is derived from a selection theorem
à la Michael, assuming of course that the metric is complete. We show in Section 4 that a uniform convex space for
which polytopes are homotopically trivial is an absolute extensor for the class of metric spaces; if it is metrizable it is
of course an absolute retract. Michael’s Selection Theorem is proved in Section 3.2 for completely metrizable uniform
spaces with homotopically trivial polytopes. From part (2) of Corollary 3.2 one can see that Van de Vel’s Theorem
follows from those results.

E-mail address: horvath@perp.fr.

0166-8641/$ – see front matter © 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.topol.2007.01.024
 C.D. Horvath / Topology and its Applications 155 (2008) 830–850 831

In Van de Vel’s paper compactness and connectedness of polytopes, coupled with the Kakutani Property, imply
an important property of nerves of coverings of convex sets by open convex sets; it is a minor modification of that
result that we call the Van de Vel Property. The Van de Vel Property is defined in Section 2.2. Section 3.3 deals with
approximate selections for upper semicontinuous maps. Section 5 deals with the fixed point property for single valued
and multivalued maps and related intersection properties.
Although selection theorems for lower semicontinuous maps, selection theorems for upper semicontinuous maps,
fixed point theorems and other related results do hold under different sets of hypotheses there is a framework (cor-
responding to that of locally convex topological vector spaces) under which all of those results hold (with complete
metrizability added for continuous selections of lower semicontinuous maps). That framework is in a sense a minimal
maximal framework; maximal from the point of view of the results that hold in such a context, all of the results alluded
to above, and minimal from the point of view of the hypotheses which are at the same time simple and trivially true
in the classical setting of topological vector spaces. The exact meaning of that discussion, which is the subject matter
of the following pages, is encapsulated in the following statement.

Metatheorem. Let (X, U, C) be a uniform convex space such that:

(1) polytopes are compact metrizable and connected;

(2) the Kakutani Property holds on polytopes equipped with the induced convexity.

Then,

(A) (Dugundji’s Theorem) X is an absolute extensor for the class of metric spaces; and, consequently, an absolute
retract if it is metrizable.
(B) (Michael’s Selection Theorem) if the uniformity U is completely metrizable then lower semicontinuous maps
Ω : Y → X defined on a paracompact space Y with nonempty closed convex values (i.e. Ωy ∈ C for all y ∈ Y )
have continuous selections.
(C) (Approximate selections for usc maps) If Ω : Y → X is an upper semicontinuous map with nonempty convex
values defined on a paracompact space Y then, for all elements R of the uniformity U , there exists a continuous
map f : Y → X such that, for all y ∈ Y , f (y) ∈ R(Ωy). Furthermore, if the values of Ω are convex and compact
then any neighborhood Θ ⊂ Y × X of the graph of Ω contains the graph of a continuous map f : Y → X.
(D) (Fan–Himmelberg’s Theorem) If the Kakutani Property holds then compact upper semicontinuous maps Ω : X →
X (i.e. the closure of Ω(X) is compact) with nonempty compact convex values have fixed points.

The paper ends with a few examples of uniform convex spaces. All spaces are assumed to be separated.

2. Convexities

2.1. Uniform convex spaces

Definition 2.1. A convexity on a set X is a family C of subsets of X such that:

(Conv1) ∅ and X belong to C and, for 

all x ∈ X, {x} ∈ C;
(Conv2) if A is a subfamily of C then A belongsto C;
(Conv3) if A is an updirected subfamily of C then A belongs to C.

Elements of C are called convex sets and (X, C) is a convex space.


Given A ⊂ X let C(A) = {C ∈ C: A ⊂ C}; by Conv1 C(A) = ∅ and by Conv2, C(A) ∈ C; it is the C-hull of A
and it is denoted by JAKC . Whenever a single convexity is under discussion we will drop the prefix, or the index, C.
Given a nonempty set S we will denote by S the family of nonempty finite subsets of S. An algebraic closure
operator on X is a map A → JAK from the power set of X to itself such that, for all A, B ⊂ X:
832 C.D. Horvath / Topology and its Applications 155 (2008) 830–850

(Alg1) A ⊂ JAK;
(Alg2) JJAKK = JAK;
(Alg3) if A ⊂ 
B then JAK ⊂ JBK;
(Alg4) JAK = {JSK: S ∈ A }.

If C is a convexity on X then A → JAKC is an algebraic closure operator on X and C = {C ⊂ X: JCKC = C};

reciprocally, if A → JAK is an algebraic closure operator on X such that ∅ = J∅K and {x} = J{x}K for all x ∈ X then
{C ⊂ X: JCK = C} is a convexity on X, call it C, for which J·K = J·KC .
If C is a convexity on X then, for all convex subset C of X, C|C = {C ∈ C: C ⊂ C} is a convexity on C, the
induced convexity, and, for all A ⊂ C, JAKC |C = JAKC .

Definition 2.2. A uniform convex space is a triple (X, U, C) where U is a separated uniformity on X and C is a
convexity such that:

(UnifConv) for all R ∈ U there exists S ∈ U such that, for all C ∈ C, JS(C)K ⊂ R(C).

If (X, U, C) is a uniform convex space we say that the uniformity U and the convexity C are compatible.

Lemma 2.1. If (X, U, C) is a uniform convex space then, for all convex sets C ∈ C the induced uniformity and the
induced convexity are compatible.

Proof. The induced uniformity on a convex set C ∈ C is U|C = {R ∩ C × C: R ∈ U}; let us see that (C, U|C , C|C ) is a
uniform convex space if (X, U, C) is one: let R and S be two elements of U such that JS(C )K ⊂ R(C ) for all C ∈ C;
from S(C ) ∩ C ⊂ JS(C )K ∩ C and the convexity of JS(C )K ∩ C we have JS(C ) ∩ CK ⊂ R(C ) ∩ C. 2

As one can easily see, locally convex topological vector spaces are uniform convex spaces, and therefore, so are
convex subsets of locally convex topological vector spaces. Uniform convex spaces are locally convex.

Proposition 2.1. In a uniform convex space (X, U, C) the closure of a convex set is convex. Furthermore, for all x ∈ X
and for all neighborhood W of x there exists a convex neighborhood V of x such that V ⊂ W.

Proof. Foreach R ∈ U chooseR ∈ U such that, for all C ∈  C, JR (C)K ⊂ R(C); from C ⊂ R (C) ⊂ JR (C)K and
from C = {R(C): R ∈ U} ⊂ {R (C): R ∈ U} we get C = {JR (C)K: R ∈ U}.
If W is a neighborhood of x there exists R ∈ U such that Rx ⊂ W ; let V = JR xK, it is a neighborhood of x since
it contains R x and, by construction, V ⊂ Rx ⊂ W . 2

Let Oconv be the set of elements of U such that, for all x ∈ X, Rx is open and belongs to C.
The second part of Proposition 2.1 is strengthened by the first part of the next proposition, which is essentially
Theorem 2.6 of Van de Vel [26].

Proposition 2.2. In a uniform convex space (X, U, C) the following properties hold:

(a) for all R ∈ U there exists S ∈ U such that, for all C ∈ C, there exists an open convex set V such that S(C) ⊂ V ⊂
R(C);
(b) Oconv is a base of the uniformity U .

Proof. Let R0 = R; if Rn has been defined, choose Rn+1 ∈ U such that Rn+1 ◦ Rn+1 ⊂ Rn and choose Sn ∈ U such
that, for all C ∈ C, JSn (C)K ⊂ Rn+1 (C). Put C0 = C and Cn+1 = JSn (Cn )K.
We have Cn+1 ⊂ Rn+1 (Cn ) and therefore R n+1 (Cn+1 ) ⊂ Rn (Cn ). From Cn ⊂ Sn (Cn ) ⊂ Cn+1 we have Cn ⊂
int Cn+1 ⊂ Cn+1 . It follows from this that V = n∈N Cn is open and convex.
We also have S0 (C) ⊂ JS0 (C)K = C1 ⊂ V ; let S = S0 .
 C.D. Horvath / Topology and its Applications 155 (2008) 830–850 833

Finally, we have seen that {Rn (Cn )}n∈N is a decreasing sequence of sets, and Cn ⊂ Rn (Cn ) by construction; from
this we have Cn ⊂ R0 (C0 ) for all n ∈ N, and consequently, V ⊂ R(C). This proves (a).
To prove (b), choose, for all R ∈ U an SR ∈ U , and for all x ∈ X an open convex set VR x such that SR x ⊂ VR x ⊂
Rx; from SR ⊂ VR we have VR ∈ Oconv and, from SR ⊂ VR ⊂ R, we see that {VR : R ∈ U} is a base of U . 2

Corollary 2.1. If (X, U, C) is a uniform convex space then, for all compact convex set K ⊂ X and for all open set
U ⊂ X containing K there exists an open convex set V such that K ⊂ V ⊂ U .

Proof. For all x ∈ K there exists Sx ∈ U such that x ∈ Sx x ⊂ U ; for all x ∈ K choose Rx ∈ U such that Rx ◦ Rx ⊂ Sx .
m
By
m compactness there exists a finite subset {x 1 , . . . , x m } ⊂ K such that K ⊂ i=1 R xi x i ; choose R ∈ U such that R ⊂
i=1 Rxi . By (a) of Proposition 2.2 there exists an open convex set V such that K ⊂ V ⊂ R(K). For all x ∈ R(K)
there exists x ∈ K and i ∈ {1, . . . , m} such that x ∈ Rx and x ∈ Rxi xi , from which we obtain x ∈ (R ◦ Rxi )xi ⊂
(Rxi ◦ Rxi )xi ⊂ Sxi xi ⊂ U . We have shown that V ⊂ U . 2

The first part of Proposition 2.2 will be important for the approximation of upper semicontinuous maps while the
second part will be used in the proof of the main theorem of Section 3.
As a straightforward consequence of Proposition 2.2 we have:

Corollary 2.2. In a uniform convex space, a compact convex set is the intersection of the open convex sets containing
it.

2.2. The Van de Vel Property

Given a set X, a reflexive relation R ⊂ X × X and a nonempty subset F of X let, for all nonempty subsets C of X,
   
SR (C; F ) = A ∈ F : C ∩ Rx = ∅ .
x∈A

SR (C; F ) is a simplicial complex whose geometric realization is denoted by |SR (C; F )|; SR (C) stands for SR (C; X).
The following sections will show that a uniform convexity for which there are enough entourages R for which the
polyhedra SR (C), where C is convex, are homotopically trivial behaves, with respect to selections and fixed point
properties, like the linear convexity of a locally convex, or a normed, topological vector space. Under natural condi-
tions, in the sense that they trivially hold for locally convex vector spaces, we will see, following essentially Van de
Vel, that the polyhedra SR (C) are homotopically trivial. More surprising maybe is that from the fact that polyhedra
of the form SR (C) are homotopically trivial, when R is taken in a suitable basis of the uniformity, we obtain that the
convex sets themselves are homotopically trivial. We start with the following simple observation.

Lemma 2.2. Let K be a compact set in a uniform topological space (X, U) and B a basis of the uniformity U . Then
|SR (K)| is connected for all R ∈ B if and only if K is connected.

Proof. Assume that K is compact and not connected. There are two open sets U1 and U2 such that K ⊂ U1 ∪ U2 ,
K ∩ U1 ∩ U2 = ∅, K ∩ Ui = ∅ for i = 1, 2; for x ∈ K ∩ Ui we find Sxi ∈ B such that Sxi x ⊂ K ∩ Ui . Proceeding as
in the proof of Corollary 2.1 we find a finite set A = {x1 , . . . , xn } ⊂ K and R ∈ B such that for all x ∈ K there exists
y ∈ A such that Rx ⊂ Sy1 y or Rx ⊂ Sy2 y which implies that for all x ∈ K we have either Rx ⊂ U1 or Rx ⊂ U2 . Since
K ∩ U1 ∩ U2 = ∅, |SR (K)| is not connected.
Assume that K is connected and let {x} and {y} be two vertices of |SR (K)|. Choose x ∈ K ∩ Rx and y ∈ K ∩ Ry;
there is a finite sequence {x0 , . . . , xm } of points of X such that x ∈ Rx0 , y ∈ Rxm and K ∩ Rxi ∩ Rxi+1 = ∅ for
i = 0, . . . , m − 1. We also have K ∩ Rx ∩ Rx0 = ∅ and K ∩ Ry ∩ Rxm = ∅. By construction, the sets {x, x0 },
{xi−1 , xi } and {xm , y} belong to SR (K), this shows that the 1-skeleton of |SR (K)| is connected. 2

Definition 2.3. A uniform convex space (X, U, C) has the Van de Vel Property if there exists a base Uconv of the
uniformity U such that:
834 C.D. Horvath / Topology and its Applications 155 (2008) 830–850

(1) Uconv ⊂ Oconv and

(2) for all nonempty C ∈ C and all R ∈ Uconv , the polyhedron |SR (C)| is homotopically trivial.
(3) It is a Van de Vel space if for all convex sets C ∈ C, for all R ∈ Uconv and all E ⊂ X such that {Rx: x ∈ E} covers
C the polyhedron |SR (C; E)| is homotopically trivial.

Clearly, a uniform convex space which is a Van de Vel space has the Van de Vel Property. The following proposition
shows that in a Van de Vel space the Van de Vel Property is hereditary with respect to convex sets equipped with the
induced uniformity and the induced convexity.

Proposition 2.3. If (X, U, C) is a Van de Vel space then, for all C ∈ C, the uniform convex space (C, U|C , C|C ) has
the Van de Vel Property.

Proof. Let Uconv be a basis of U for which (1) and (3) of Definition 2.3 are verified and let C ⊂ X be a convex set. As
one can easily see the family Uconv|C = {R ∩ (C × C): R ∈ Uconv } is a basis of the induced convexity whose elements
have convex and open (in C), values. Let Q ⊂ C be convex and pick S ∈Uconv|C . Then S = R ∩ (C × C) for some
R ∈ Uconv . For all finite subsets A of C one has Q ∩ ( x∈A Sx) = Q ∩ ( x∈A Rx ∩ C) = (Q ∩ C) ∩ ( x∈A Rx) =
Q ∩ ( x∈A Rx). In other words, SS (Q) = SR (Q; C), where the first simplicial complex is computed in (C, U|C , C|C )
and the second in (X, U, C). By hypothesis, SR (Q; C) is homotopically trivial; this completes the proof. 2

Lemma 2.3. Let R ∈ Oconv where (X, U, C) is a uniform convex space.

(1) If polytopes are compact then for all C ∈ C, for all nonempty subsets E of X such that {Rx: x ∈ E} covers C
and for all compact subsets K of |SR (C; E)| there exists a polytope P ⊂ C and a finite set F ⊂ E such that
K ⊂ |SR (P ; F )| ⊂ |SR
(C; E)|.
(2) If C ∈ C and if C ⊂ x∈F Rx with F finite then there exists a polytope P ⊂ C such that |SR (C; F )| =
|SR (P ; F )|.

Proof. (1) A compact subset K of |SR (C; E)| is contained in a finite polyhedron; there is a finite  subset F of E
such that K ⊂ |{A ∈ F ∩ SR (C; E)}|. For each A ∈ F ∩ SR (C; E) choose a point pA in C ∩ ( x∈A Rx) and let
P = JpA : A ∈ F ∩ SR (C; E)K; P  is a polytope which is contained in C; P being compact by hypothesis, there is a
finite subset F0 of E such that P ⊂ x∈F0 Rx. Let F1 = F ∪ F0 ; since {A ∈ F ∩ SR (C; E)} ⊂ SR (P , F1 ) we have
K ⊂ |SR (P , F1 )| ⊂ |SR (C; E)|. 
(2) For all J ∈ SR (C; F ) choose a point xJ ∈ C ∩ ( x∈J Rx) and let P = J{xJ : J  ∈ SR (C; F )}K; it is a polytope
contained in C, therefore SR (P ; F ) ⊂ SR (C; F ). If J ∈ SR (C; F ) then xJ ∈ P ∩ ( x∈J Rx), and therefore J ∈
SR (P ; F ). We have shown that SR (P ; F ) = SR (C; F ), and therefore |SR (P ; F )| = |SR (C; F )|. 2

Lemma 2.4. Let R ∈ Oconv where (X, U, C) is a uniform convex space. Statements (1) and (2) below are equivalent.
If polytopes are compact then (3) is equivalent to (1) and (2).

(1) For all C ∈ C and all F ∈ X such that {Rx: x ∈ F } covers C |SR (C; F )| is homotopically trivial.
(2) For all polytope P ⊂ X and all F ∈ X such that {Rx: x ∈ F } covers P , |SR (P ; F )| is homotopically trivial.
(3) For all C ∈ C and all nonempty subset E of X such that {Rx: x ∈ E} covers C, |SR (C; E)| is homotopically
trivial.

Proof. The equivalence of (1) and (2) is a consequence of the second part of Lemma 2.3. To prove (3) from (1) and
(2) assuming that polytopes are compact recall that a polyhedron is homotopically trivial if each of its compact subset
is contained in a homotopically trivial subset. The conclusion follows from (1) of Lemma 2.3. 2

Definition 2.3 is rather abstract and one might wonder how one can check that a given uniform convex space is a
Van de Vel space. Surprisingly, under very simple natural conditions—compactness and connectedness of polytopes,
and the Kakutani Property—we have a Van de Vel space. This was essentially established by Van de Vel in [26] using
a different language; actually, we will prove this with a condition on polytopes formally weaker than the Kakutani
 C.D. Horvath / Topology and its Applications 155 (2008) 830–850 835

separation property. In the classical linear setting, the last condition is implied by the algebraic Hahn–Banach Theo-
rem; also, in linear spaces, polytopes are compact and connected. As a consequence, an arbitrary convex subset of a
locally convex topological vector space is a Van de Vel space.

Proposition 2.4. A uniform convex space (X, U, C) for which conditions (1) and (2) below hold is a Van de Vel space.

(1) Polytopes are compact and connected.

(2) (S4-weak ) For all polytopes P and for all disjoint nonempty open convex subsets W1 and W2 of P there exist
convex subsets C1 and C2 of P such that C1 ∩ W2 = C2 ∩ W1 = ∅ and P = C1 ∪ C2 (clearly, one can assume
that C1 and C2 are closed in P ).

Proof. Let P ⊂ X be a polytope, R an element of Oconv and F ⊂ X a nonempty finite subset such that {Rx: x ∈ F }
covers C. We show by induction on the number of simplexes of the finite polyhedron |SR (P ; F )| that it is homo-
topically trivial; if that number is one there is nothing to prove. Assume that number is m + 1 with m  1; if there
is only one maximal simplex,again there is nothing to prove, otherwise
 choose two maximal elements A1 and A2
of SR (P ; F). We have P ∩ ( x∈Ai Rx) = ∅, i = 1, 2, and P ∩ ( x∈A1 ∪A2 Rx) = ∅, since A1 ∪ A2 ∈ / SR (P ; F ); let
Wi = P ∩ ( x∈Ai Rx) = ∅. By (S4-weak ) we can find two convex subsets C1 and C2 such that C1 ∩ W2 = C2 ∩ W1 = ∅
and P = C1 ∪ C2 , and therefore SR (P ; F ) = SR (C1 ; F ) ∪ SR (C2 ; 
F ). 
Notice that Ai  ∈ SR (Ci ; F ), since Ci ⊂ x∈F Rx and Ci ∩ ( x∈Ai Rx) = Ci ∩ P ∩ ( x∈Ai Rx) = Ci ∩ Wi .
Also, from C2 ∩ ( x∈A1 Rx) = C2 ∩ P ∩ ( x∈A1 Rx) = C2 ∩ W1 = ∅ we see that A1 ∈ / SR (C2 ; F ); and, similarly,
A2 ∈ / SR (C1 ; F ). This shows that |SR (C1 ; F )| and |SR (C2 ; F )| have, each, at most m simplexes and are, by the
induction hypotheses, homotopically trivial.
Next, we show that SR (C1 ∩ C2 ; F ) = SR (C1 ; F ) ∩ SR (C2 ; F ); one inclusion being trivial, we only have to see
that SR (C1 ; F ) ∩ SR (C2 ; F ) ⊂ SR (C1 ∩ C2 ; F ). As we have seen, we can assume that C1 and C2 are closed in P ,
and consequently, P being connected,
 we must have C1 ∩ C2 = ∅. 
Let A ∈ F such that Ci ∩ ( x∈A Rx) = ∅, i = 1, 2. The set Q = P ∩ ( x∈A Rx) is connected since it is convex
and Ci ∩ Q is nonempty and closed in Q; from P = C1 ∪ C2 we have Q = (C1 ∩ Q) ∪ (C2 ∩ Q), and, from the
connectedness of Q, Q ∩ C1 ∩ C2 = ∅; this shows that A ∈ SR (C1 ∩ C2 ; F ).
Also, neither A1 nor A2 are in SR (C1 ∩ C2 ; F ); by the induction hypotheses, |SR (C1 ∩ C2 ; F )| is homotopically
trivial.
In conclusion, the finite polyhedron |SR (P ; F )| is the union of two homotopically trivial polyhedra |SR (C1 ; F )|
and |SR (C2 ; F )| whose intersection is also homotopically trivial. Since finite homotopically trivial polyhedra are
absolute retracts, and |SR (C1 ; F )| ∩ |SR (C2 ; F )| = |SR (C1 ∩ C2 ; F )| we have shown that |SR (P ; F )| is the union
of two absolute retracts whose intersection is an absolute retract, it is therefore an absolute retract and consequently
homotopically trivial. 2

Corollary 2.3. Convex subsets of locally convex vector spaces are Van de Vel spaces.

In all the propositions and proofs that follow, Uconv will always stand for a basis of the uniformity U of a given
uniform convex space (X, U, C) for which (1) and (2) of Definition 2.3 holds.

3. Selections for lower semicontinuous maps

3.1. Selections for lower semicontinuous maps defined on finite dimensional spaces

Given a map Ω : Y → X and a binary relation R ⊂ X we say that Ω is R-small if for all y ∈ Y there exists x ∈ X
such that Ωy ⊂ Rx.

Theorem 3.1. A lower semicontinuous map with nonempty (closed) convex values Ω : Y → X from a finite dimen-
sional paracompact topological space to a uniform convex space (X, U, C) with the Van de Vel Property has, for all
R ∈ Uconv , a lower semicontinuous selection ΓR : Y → X with (closed) convex values which is R-small.
836 C.D. Horvath / Topology and its Applications 155 (2008) 830–850

Theorem 3.2. If (X, U, C) is a completely metrizable uniform convex space with the Van de Vel Property then any
lower semicontinuous map Ω : Y → X with closed and nonempty convex values defined on a paracompact finite
dimensional space Y has a continuous selection.

The proof of Theorem 3.1 will proceed as follows: given R ∈ Uconv we will show that there exists a map ω : Y →
X such that

(1) for all y ∈ Y Ωy ∩ ( x∈ω(y) Rx) = ∅;
(2) for all y ∈ Y there exists a neighborhood V of y such that, for all y ∈ V , ω(y ) ⊂ ω(y).

We will then put ΩR y = Ωy ∩ ( x∈ω(y) Rx); by (1), ΩR y is convex and  nonempty. If U ⊂ X is open and ΩR y ∩
U = ∅ then, taking into account that Ω is lower semicontinuous and that x∈ω(y) Rx is open we find a neighborhood

W of y such that, for all y ∈ W , Ωy ∩ ( x∈ω(y) Rx) ∩ U = ∅; next, taking V as in (2) above, we will have Ωy ∩

( x∈ω(y ) Rx) ∩ U = ∅ for all y ∈ V ∩ W , which shows that ΩR is lower semicontinuous; finally, we take ΓR y =
ΩR y.
That there is a map ω : Y → X for which (1) and (2) hold is a consequence of Lemma 3.1  below.
Given Ω : Y →  X, y ∈ Y and R ∈ Uconv , let, as previously, SR (Ωy) = {A ∈ X : Ωy ∩ ( x∈A Rx) = ∅} and put
SR = {A ∈ X : x∈A Rx = ∅}. |SR (Ωy)| is a subpolyhedron of |SR | which is itself a subpolyhedron of |X |; we
will denote by |SR (Ω)| the map y → |SR (Ωy)|.

Lemma 3.1. If Ω : Y → X is a lower semicontinuous map with nonempty convex values from a finite dimensional
paracompact topological space to a uniform convex space (X, U, C) with the Van de Vel Property then, for all R ∈
Uconv , the map |SR (Ω)| : Y → |SR | has a continuous selection.

The proof of Lemma 3.1 relies on a result of Valentin Gutev, Theorem 3.1 (and Remark 3) of [14]. For our purpose
the full strength of that result is not needed we therefore state it, with the necessary definitions, in a truncated form.
Following Gutev, we say that a map Φ : Y → Z is lower locally constant if, for all compact subset K of Z the set
{y ∈ Y : K ⊂ Φy} is open in Y . Equivalently, Φ is lower locally constant if z∈K Φ −1 z is open in Y whenever K is
a compact subset of Z. In particular, a lower locally constant map has open fibers (for all z ∈ Z, Φ −1 z is open in Y ),1
and is consequently lower semicontinuous.

Theorem 3.3 (A truncated formulation of Gutev’s theorem). A lower locally constant map Φ : Y → Z with nonempty
homotopically trivial values in a topological space Z, defined on a finite dimensional paracompact space Y , has a
continuous selection.

Proof of Lemma 3.1. We have to see that |SR (Ω)| : Y → |SR | is lower locally constant. First, we verify that SR (Ω) :
Y → SR , where SR is endowed with the discrete topology,  is lower locally constant. Let {A0 , . . . , Am } ⊂ SR and
y  ∈ Y such that {A0 , . . . , A
m } ⊂ SR (Ωy  ). We have Ωy  ∩ (
x∈Ai Rx) = ∅ for all i ∈ {0, . . . , m}. Since Ω is lower
semicontinuous and all the x∈Ai Rx are open we can find, for each i ∈ {0, . . . , m}, an open neighborhood Vi of y 
 
such that, for all y ∈ Vi , Ωy ∩ ( x∈Ai Rx) = ∅. For all y ∈ V = m i=0 Vi we have {A0 , . . . , Am } ⊂ SR (Ωy). This
shows that SR (Ω) is lower locally constant.
We show that |SR (Ω)| : Y → |SR | is lower locally constant.
Let K be a compact subset of |SR | contained m in |SR (Ωy )|,for some y ∈ Y . There
  is a finite number of simplices
{σ0 , . . . , σm } of |SR (Ωy )| such that K ⊂ i=0 σi ⊂ |SR (Ωy )|. Let Ai ∈ SR (Ωy  ) be the set of vertices of σi . By


the first part of the proof there is an open neighborhood V of y  such that, for all y ∈ V and all i ∈ {0, . . . , m},
σi ⊂ |SR (Ωy)|, and therefore, K ⊂ |SR (Ωy)|. 2

Proof of Theorem 3.1. Let r : Y → |SR | ⊂ |X | be a continuous selection of |SR (Ω)| obtained from Lemma 3.1.
For each vertex {x} of |SR |, that is for each x ∈ X such that Ωy ∩ Rx = ∅, let αx : |SR | → [0, 1] be the corresponding

1 The fibers of a Φ : Y → Z are the sets Φ −1 z.

 C.D. Horvath / Topology and its Applications 155 (2008) 830–850 837

barycentric map (if σ is a simplex of |SR | for which {x} is not a vertex then αx is identically 0 on σ , αx is therefore well
defined and continuous). Let SR0 be the set of vertices of |SR | and, for {x} ∈ SR0 , let rx = σx ◦ r; it is a continuous map
from Y to [0, 1] and {rx : {x} ∈ SR0 } is a partition of unity on Y . For all y ∈ Y there is a unique simplex σy ⊂ |SR | such
that r(y) belongs to the interior of that simplex; the set of vertices of that simplex is ρ(y) = {x: σx (r(y)) > 0}. The
map ζ : Y → [0, 1] defined by ζ (y) = sup{x}∈S 0 rx (y) is lower semicontinuous. Let ω(y) = {x ∈ X: rx (y)  12 ζ (y)},
R
it is nonempty by definition of ζ (y). Since for all y ∈ Y there exists x ∈ X such that rx (y) > 0 we have rx (y) > 0 for
all x ∈ ω(y) and therefore ω(y) ⊂ ρ(y); this shows that ω(y) is finite. Also, from {x}∈ρ(y) σx (r(y)){x}  = r(y) ∈
|SR (Ωy)| we have ρ(y) ∈ SR (Ωy); from ω(y) ⊂ ρ(y) we also have ω(y) ∈ SR (Ωy), that is Ωy ∩ ( x∈ω(y) Rx) = ∅.
For all y ∈ Y let Wy1 = {y ∈ Y : 1 − x∈ρ(y) rx (y ) < 12 ζ (y )}; from the lower semicontinuity of ζ it is an open
set and it contains y since 1 − x∈ρ(y) rx (y) = 0 < 12 ζ (y). Notice that if x̄ ∈ / ρ(y) and y ∈ Wy1 then rx̄ (y ) 
1 − x∈ρ(y) rx (y ) which yields rx̄ (y ) < 2 ζ (y ), and therefore x̄ ∈
1
/ ω(y ).
For all x ∈ ρ(y)\ω(y) choose an open neighborhood Wx,y of y such that, for all y ∈ Wx,y
2 2 , r (y ) < 1 ζ (y ) and let
x
 2
Wy = x∈ρ(y)\ω(y) Wx,y ; it is an open neighborhood of y. By construction, for all x ∈ ρ(y)\ω(y) and for all y ∈ Wy2
2 2

we have rx (y ) < 12 ζ (y ). Finally, let Vy = Wy1 ∩ Wy2 . We have seen that if y ∈ Vy and if x ∈ / ω(y) then x ∈ / ω(y ); in
other words, for all y ∈ Vy we have ω(y ) ⊂ ω(y). 2

Proof of Theorem 3.2. Let {Rn : n ∈ N} be a countable base of U such that, for all n„ and Rn+1 ⊂ Rn and Rn−1 = Rn ,
and, for all n and all x ∈ X, Rn x is open. For all n there exists R ∈ Uconv such that R ⊂ Rn therefore, by Theorem 3.1
there is a lower-semicontinuous map Γ0 : Y → X with nonempty closed convex values and a single valued map
γ0 : Y → X such that Γ0 y ⊂ Ωy ∩ R0 γ0 (y) for all y. Assuming that we have constructed Γn : Y → X, lower semi-
continuous with nonempty closed convex values, we obtain from Theorem 3.1 Γn+1 : Y → X with nonempty closed
convex values and a single valued map γn+1 : Y → X such that Γn+1 y⊂ Γn y ∩ Rn+1 γn+1 (y) for all y. Since the
uniformity is complete there is for all y ∈ Y a point f (y) ∈ X such that n∈N Γn y = {f (y)}.
By construction we have f (y) ∈ Ωy for all y ∈ Y . To see that f is continuous fix an arbitrary point y ∈ Y , let
W ⊂ X be a neighborhood of f (y), choose l such that Rl ◦ Rl (f (y)) ⊂ W and n such that Rn ◦ Rn ⊂ Rl . Then
Γn y ⊂ Rl (f (y)) for all y ∈ Y . From the lower semicontinuity of Γl there exists a neighborhood V of y in Y such that,
for all y ∈ V , Γn y ∩ Rl (f (y)) = ∅, consequently, Rl (f (y)) ∩ Rl (f (y)) = ∅ for all y ∈ V and f (y) ∈ Rl ◦ Rl (f (y))
for all y ∈ V ; we have shown that f (V ) ⊂ W. 2

Corollary 3.1.

(1) A completely metrizable uniform convex space (X, U, C) with the Van de Vel Property is an absolute extensor for
the class of paracompact finite dimensional spaces. If X is finite dimensional then it is contractible.
(2) A completely metrizable uniform convex space (X, U, C) with the Van de Vel Property is homotopically trivial.

Corollary 3.2.

(1) A completely metrizable convex subset of a Van de Vel space is homotopically trivial.
(2) Completely metrizable convex subsets of a uniform convex space (X, U, C) with property (S4-weak ) and with
compact connected polytopes are homotopically trivial.

Proof. We have observed in the previous section that convex subsets of a Van de Vel space have, for the induced uni-
formity and the induced convexity, the Van de Vel Property. Part (1) is therefore a consequence of (2) of Corollary 3.1.
The second part is a consequence of the first and of Proposition 2.4. 2

3.2. Selections for lower semicontinuous maps

We show in this section that Michael’s Selection Theorem can be adapted to uniform convex spaces, more ex-
actly:
838 C.D. Horvath / Topology and its Applications 155 (2008) 830–850

Theorem 3.4 (Michael’s Selection Theorem). Let (X, U, C) be a uniform convex space for which polytopes are ho-
motopically trivial. If the uniformity is completely metrizable then all lower semicontinuous maps Ω : Y → X with
nonempty closed convex values defined a paracompact space Y have a continuous selection.
If A ⊂ Y is a closed set then any continuous selection of Ω restricted to A extends to a continuous selection of Ω.

Without assuming that the uniformity is metrizable, we show that a lower semicontinuous map with nonempty
convex values, defined on a paracompact topological space has, for all elements R of the uniformity, an R-selection,
in the sense that the each value of the selection is R-close to the corresponding value of the convex valued map.
The proof of the extension of Michael’s Theorem is done through the construction of an appropriate sequence of
such approximate selections and by taking the pointwise limit of these approximate selections. All the preliminary
results are presented first.

Lemma 3.2. Let (X, τ, C) be a topological space endowed with a convexity C for which polytopes are homotopically
trivial. Assume that we are given a paracompact topological space Y , an open covering R of Y and, for all U ∈ R,
a point η(U ) ∈ X; for all y ∈ Y let σ (y; R) = {U ∈ R: y ∈ U }. Then, there exists a continuous map g from the
geometric realization of the nerve of R to X such that, for all y ∈ Y , g(y) ∈ J{η(U ): U ∈ σ (y; R)}K.

Proof. Let |R| be the geometric realization of the nerve of R; identifying the set of vertices of |R| with R, we already
have a continuous map η : |R|0 → X. Assume that we have a continuous map ηm : |R|m → X such that, for all m-di-
mensional simplex σ , ηm (σ ) ⊂ J{η(U ): U ∈ σ 0 }K. If σ is an (m + 1)-dimensional simplex then ηm restricted to the
boundary of σ is a continuous map into J{η(U ): U ∈ σ 0 }K, and, since this set is homotopically trivial, we can extend
ηm restricted to the boundary of σ to a continuous map ηm,σ from σ to J{η(U ): U ∈ σ 0 }K; if σ and σ are two (m + 1)-
dimensional simplices which intersect then ηm,σ and ηm,σ agree on σ ∩ σ ; we therefore have a continuous map ηm+1
defined on |R|m+1 such that ηm+1 (σ ) ⊂ J{η(U ): U ∈ σ 0 }K for all (m + 1)-dimensional simplex. By induction, we
obtain a continuous map η̂ : |R| → X such that, for all simplices, η̂(σ ) ⊂ J{η(U ): U ∈ σ 0 }K.
Since Y is paracompact, there exists a locally finite partition of unity {κU : U ∈ R} on Y such that κU−1 (]0, 1]) ⊂ U
for all U ∈ R; let κ : Y → |R| be the associated canonical map. For all y ∈ Y denote by σ (y) the smallest simplex
of |R| to which κ(y) belongs; if U is a vertex of σ (y) the κU (y) > 0 and therefore y ∈ U , this shows that σ (y)0 ⊂
σ (y; R), and therefore η̂(κ(y)) ⊂ J{η(U ): U ∈ σ (y)0 }K ⊂ J{η(U ): U ∈ σ (y; R)}K. We can take g = η̂ ◦ κ. 2

Proposition 3.1. Let (X, U, C) be a uniform convex space for which polytopes are homotopically trivial and Ω : Y →
X a lower semicontinuous map with nonempty convex values from a paracompact topological space Y to X. Then,
for all R ∈ U , there exists a continuous map g : Y → X such that, for all y ∈ Y , g(y) ∈ R(Ωy).

Proof. Given R ∈ U there exists S ∈ U such that, for all C ∈ U , JS(C)K ⊂ R(C); without loss of generality, we can
assume that S −1 x is open for all x ∈ X. For all y ∈ Y , let Γ y = S(Ωy); Γ y is never empty, since Ω has nonempty
values, therefore {Γ −1 x: x ∈ X} is a covering of Y ; for all x ∈ X, Γ −1 x is open, by the lower semicontinuity of Ω
and by Γ −1 x = {y ∈ Y : Ωy ∩ S −1 x = ∅}. Let V be a locally finite open covering of Y finer than {Γ −1 x: x ∈ X}.
For all V ∈ V there exists xV ∈ X such that V ⊂ Γ −1 xV . From Lemma 3.2 there exists a continuous map g : Y → X
such that, for all y ∈ Y , g(y) ∈ J{xV : y ∈ V }K; notice that if y ∈ V then Ωy ∩ S −1 xV = ∅, and therefore xV ∈ S(Ωy);
taking the convex hull, we obtain J{xV : y ∈ V }K ⊂ JS(Ωy)K, and, from the choice of S, g(y) ∈ R(Ωy). 2

The conclusion of Proposition 3.1 can be formulated somewhat differently; U has a base of symmetric relations,
therefore, in the statement of Proposition 3.1 we can replace R by R −1 and the conclusion becomes R(g(y)) ∩ Ωy = ∅
for all y ∈ Y.

Lemma 3.3. Let X and Y be topological spaces, Ω : Y → X a lower semicontinuous map, g : Y → X a continuous
map and R ⊂ X × X an open graph relation. Then y → R(g(y)) ∩ Ωy defines a lower semicontinuous map from Y
to X.

Proof. Notice that y → {g(y)} × Ωy is lower semicontinuous from Y to X × X, call this map Ψ . If U ⊂ X is open
then R ∩ X × U is open in X × X and, for all y ∈ Y , R(g(y)) ∩ Ωy ∩ U = ∅ if and only if Ψ y ∩ (R ∩ X × U ) = ∅. 2
 C.D. Horvath / Topology and its Applications 155 (2008) 830–850 839

Lemma 3.4. If (X, U, C) is a uniform convex space and Ω : Y → X is a lower semicontinuous map then y → JΩyK
is also lower semicontinuous.

Proof. Let V ⊂ X be an open set and y ∈ Y such that JΩyK ∩ V = ∅. There is a finite set {x0 , . . . , xm } ⊂ Ωy such
that J{x0 , . . . , xm }K ∩ V = ∅. Choose u ∈ J{x0 , . . . , xm }K ∩ V and R ∈ U such that Ru ⊂ V . There exists S ∈ U such
that, for all convex set C ⊂ X, JS(C)K ⊂ R −1 (C); without loss of generality, we can assume that, for all x ∈ X, S −1 x
is open.
By reflexivity of S we have S −1 xi ∩ Ωy = ∅ for all i ∈ {0, . . . , n}; let Wi ⊂ Y be an open neighborhood of y such
that, for all y ∈ Wi , S −1 xi ∩ Ωy = ∅. Then for all y ∈ W = ∩ni=0 Wi we have {x0 , . . . , xn } ⊂ S(Ωy) and, a fortiori,
{x0 , . . . , xn } ⊂ S(JΩyK) and therefore J{x0 , . . . , xn }K ⊂ JS(JΩyK)K. From the choice of S we get, for all y ∈ W ,
J{x0 , . . . , xn }K ⊂ R −1 (JΩyK). Since u belongs to the left-hand side of the last inclusion we have Ru ∩ JΩyK = ∅ for
all y ∈ W , finally, from Ru ⊂ V we get V ∩ JΩyK = ∅ for all y ∈ W . 2

Proof of Theorem 3.4. Let d : X × X → R+ be a complete metric for which the relations Rn = {(x, y): d(x, y) <
2−n } form a base of the uniformity U . Since Uconv is also a base of U there exists T0 ∈ Uconv and n0 > 0 such that
Rn0 ⊂ T0 ⊂ R0 . By Proposition 3.1 there is a continuous map g0 : Y → X such that, for all y ∈ Y , Ωy ∩ Rn0 g0 (y) = ∅,
this implies Ωy ∩ R0 g0 (y) = ∅. Since Ωy is convex for all y ∈ Y and T0 x is convex for all x ∈ X we have, for all
y ∈ Y , JΩy ∩ Rn0 g0 (y)K ⊂ Ωy ∩ T0 g0 (y) ⊂ Ωy ∩ R0 g0 (y).
Let Ω1 y = JΩy ∩ Rn0 g0 (y)K; by Lemmas 3.3 and 3.4 Ω1 is lower semicontinuous, and by construction it has
convex values. Choose T1 ∈ Uconv and n1 > n0 such that Rn1 ⊂ T1 ⊂ R1 and repeat the previous construction to find a
continuous map g1 : Y → X such that, for all y ∈ Y , Ω1 y ∩ Rn1 g1 (y) = ∅ and JΩ1 y ∩ Rn1 g1 (y)K ⊂ Ω1 y ∩ T1 g1 (y) ⊂
Ω1 y ∩ R1 g1 (y).
By construction we also have Ω1 y ⊂ Ωy ∩R0 g0 (y) for all y ∈ Y , from which we can conclude that Ωy ∩R0 g0 (y)∩
R1 g1 (y) = ∅. An obvious induction yields a sequence gn : Y → X of continuous maps such that, for all n ∈ N and
for all y ∈ Y , Ωy ∩ R0 g0 (y) ∩ · · · ∩ Rn gn (y) = ∅; from Rn gn (y) ∩ Rn+1 gn+1 (y) = ∅ we have d(gn (y), gn+1 (y)) <
2−n + 2−n−1 , in other words, the sequence (gn )n∈N is uniformly Cauchy. The completeness of the metric implies that
there is a continuous map g : Y → X such that (gn (y))n∈N converges to g(y); from Ωy ∩ Rn gn (y) = ∅ we see that
g(y) belongs to the closure of Ωy.
The last part is proved as usual: if g : A → X is a continuous selection of Ω restricted to A then

g(y) if y ∈ A,
Ωy =
X if y ∈
/A
is lower semicontinuous with closed convex values. 2

3.3. Approximate selections for upper semicontinuous maps

We prove the following Cellina like result, [8]:

Theorem 3.5 (Approximate selections for usc maps). Let (X, U, C) be a uniform convex space for which polytopes
are homotopically trivial, Y a paracompact topological space and Ω : Y → X an upper semicontinuous map with
nonempty convex values. Then, for all R ∈ U there exists a continuous map f : Y → X such that, for all y ∈ Y ,
f (y) ∈ R(Ωy).
Furthermore, if the values of Ω are convex and compact then any neighborhood Θ ⊂ Y × X of the graph of Ω
contains the graph of a continuous map f : Y → X.

First, we show that in an arbitrary uniform convex space, a uniform neighborhood of an upper semicontinuous map
with convex values contains the graph of an open graph upper semicontinuous map with convex values; if the values
of the initial map are also compact then all neighborhoods of its graph must contain the graph of an open graph upper
semicontinuous map with convex values. From this we will deduce in a subsequent section a Kakutani like fixed point
theorem for compact upper semicontinuous maps (i.e. all the values are contained in a compact convex set), with
closed convex values in uniform convex spaces.
840 C.D. Horvath / Topology and its Applications 155 (2008) 830–850

Proposition 3.2. Let (X, U, C) be a uniform convex space and Ω : Y → X an upper semicontinuous map with non-
empty convex values defined on a paracompact topological space Y . Then, for all R ∈ U there exist a locally finite
open covering {Wy }y∈Y of Y , and an open graph map Γ : Y → X with convex values such that

(1) for all y ∈ Y , y ∈ Wy and

(2) for all y ∈ Y , Ωy ⊂ Γ y ⊂ Γ (Wy ) ⊂ R(Ωy).

Furthermore, if the values of Ω are compact and convex then, for all open neighborhood Θ ⊂ Y × X of the graph
of Ω, there exists an open graph and convex valued map Γ : Y → X such that Ω ⊂ Γ ⊂ Θ.

Proof. By Proposition 2.2 there exists, for all y ∈ Y , an open convex set Vy ⊂ X such that Ωy ⊂ Vy ⊂ R(Ωy). By
assumption, Ω is upper semicontinuous, let By ⊂ Y be an open neighborhood of y such that Ω(By ) ⊂ Vy and, by
regularity of Y , let Oy be an open neighborhood of y such that Oy ⊂ Oy ⊂ By , and by paracompactness, one can find
an open neighborhood Wy of y, contained in Oy such that {Wy }y∈Y is a locally finite covering of Y . For all y ∈ Y , the

set {z ∈ Y : y ∈ W z } is finite; let Γ y = y∈W z Vz ; it is an open convex set. By construction, if y ∈ W y then Γ y ⊂ Vy
therefore Γ (Wy ) ⊂ R(Ωy). Also, if y ∈ W z then y ∈ Bz , and consequently, Ωy ⊂ Vz ; this shows that Ωy ⊂ Γ y.

To complete this first part of the proof we show that the graph of Γ is open. For all y ∈ Y let Ny = Y \ y ∈W / z W z;
it is an open neighborhood of the point y, therefore Ny × Γ y is open in Y × X and, obviously, {y} × Γ y ⊂ Ny × Γ y.
If (y , x ) ∈ Ny × Γ y then, for all z ∈ Y , y ∈ W z implies y ∈ W z ; in other words, for all y ∈ Ny , Γ y ⊂ Γ y , and
therefore x ∈ Γ y . This shows that (y , x ) ∈ Γ , and consequently, that, for all y ∈ Y , Ny × Γ y ⊂ Γ . We have shown
that Γ is an open subset of Y × X.
To establish the last part of the proposition, notice first that for all compact set K ⊂ X and for all open set U ⊂ X
containing K there exists R ∈ C such that K ⊂ R(K) ⊂ U . Since Θy is an open set containing Ωy there exists, by
Corollary 2.1, an open convex set Vy such that Ωy ⊂ Vy ⊂ Θy; we can now proceed as in the first part of the proof,
with Θy replacing R(Ωy). 2

Proposition 3.2 and its proof are slight adaptations to our framework of Ancel’s Enlargement Lemma in [1].

Lemma 3.5. If (X, U, C) is a uniform convex space for which polytopes are homotopically trivial then any map
Γ : Y → X with nonempty convex values and open graph, from a paracompact topological space Y to X, has a
continuous selection.

Proof. Notice that {Γ −1 x: x ∈ X} is an open covering of Y and, taking into account that the values Γ y are convex,
we can proceed as in the proof of Proposition 3.1. 2

Proof of Theorem 3.5. From Proposition 3.2 there exists a convex valued map with open graph Γ : Y → X such
that, in the first case, Γ y ⊂ R(Ωy) for all y ∈ Y , in the second case, Γ ⊂ Θ. An appeal to Lemma 3.5 completes the
proof. 2

4. Dugundji’s theorem in uniform convex spaces

As the title of this section indicates we establish Theorem 4.1 below. All the necessary ingredients to closely follow
Dugundji’s proof in [12] are at our disposal.

Theorem 4.1. A uniform convex space for which polytopes are homotopically trivial is an absolute extensor for the
class of metric spaces.

Proof. Let (Y, d) be a metric space and f : A → X a continuous map defined on a closed subset A of Y . For all
y ∈ Y \ A choose an open ball By centered at y of radius strictly smaller than 2−1 d(y, A), the distance from y to A
and let R be a locally finite open covering of Y \ A finer than {By : y ∈ Y \ A}. To each U ∈ R associate a pair of
points (aU , xU ) ∈ A × U such that d(aU , xU ) < 2d(xU , A).
 C.D. Horvath / Topology and its Applications 155 (2008) 830–850 841

By Lemma 3.2, there exists a continuous map h : Y \ A → X such that, for all y ∈ A, h(y) ∈ J{f (aU ): U ∈
σ (y; R)}K. Let

f (y) if y ∈ A,
g(y) =
h(y) if y ∈ Y \ A.
Since g is continuous on the open set Y \ A we only have to check the continuity at all point of A to establish that g
is continuous on Y ; this is accomplished as in [12]. First, for all a ∈ A and for all neighborhood W (a) of a in X there
exists a neighborhood V (a) of a in X such that V (a) ⊂ W (a) and, for all U ∈ R, if U ∩ V (a) = ∅ then U ⊂ W (a)
and aU ∈ W (a).
Let a ∈ A and let W ⊂ X be a neighborhood of g(a); there is a convex neighborhood C ⊂ W of g(a) = f (a)
and a neighborhood W (a) of a in X such that f (A ∩ W (a)) ⊂ C. Let V (a) be as above, then f (A ∩ V (a)) ⊂ C; let
y ∈ Y \ A, if U ∈ R and y ∈ U ∩ V (a) then aU ∈ A ∩ W (a) and therefore f (aU ) ∈ C, this shows that {f (aU ): U ∈
σ (y; R)} ⊂ C, and, since C is convex, that h(y) ∈ C. We have shown that h(V (a) ∩ Y \ A) ⊂ C which, together with
f (A ∩ V (a)) ⊂ C, gives us g(V (a)) ⊂ C. 2

The main, and immediate, consequence of Theorem 4.1 is the following corollary:

Corollary 4.1. A metrizable uniform convex space with homotopically trivial polytopes is an absolute retract.

5. Fixed points

Brouwer’s Fixed Point Theorem can be derived from the well known result of Knaster–Kuratowski–Mazurkiewicz
known as the KKM Lemma, [13]; in this section we investigate convexities for which a KKM like property holds. In
Section 5.1 we first recall how to prove, in an almost formal way, fundamental intersection and fixed point theorems
for multivalued maps if we have a KKM like property at our disposal. Then, under an additional hypothesis (which in
the classical linear setting is verified if there are enough continuous linear maps on compact convex sets to separate
points), we derive a Schauder–Tychonov like Fixed Point Theorem. We establish the KKM Property for uniform
convex spaces with the Van de Vel Property and also for uniform convex spaces with compact connected polytopes
with the Kakutani Property.
In the second part, Section 5.2, we prove the Kakutani–Fan–Himmelberg Fixed Point Theorem for upper semicon-
tinuous maps, Theorem 5.2, and therefore, the Schauder–Tychonov Fixed Point Theorem, Corollary 5.2, in uniform
convex spaces with the Kakutani Property and with compact, connected metrizable polytopes.
Invoking Corollary 4.1 we can immediately state that a metrizable uniform convex space with homotopically
trivial polytopes has the fixed point property for continuous compact maps (recall that a continuous map f : X → X
is compact if the closure of its image is a compact subset of X). But neither the metrizability condition nor the
uniform continuity of the convex hull operator are a priori necessary conditions for fixed point properties, as shown,
for example, by Proposition 5.4.

5.1. Around KKM

The results of this section lay down a formal framework for fixed point properties. Accordingly, most of the proofs,
maybe with the exception of that of Theorem 5.1, cannot claim much originality.

Definition 5.1. Let X be a topological space endowed with a convexity C; we say that (X, C) has the weak Van de Vel
Property if polytopes are compact and are, with the induced convexity, uniform convex spaces with the Van de Vel
Property.

Convex subsets of arbitrary topological vector spaces have the weak Van de Vel Property.
The second condition is unambiguous since a compact space has a unique uniformity compatible with its topology.
Corollary 2.3 implies that arbitrary topological vector spaces have the weak Van de Vel Property.
If (X, C) has the weak Van de Vel Property then, by Lemma 2.2, polytopes are also connected. On the other hand,
if polytopes are compact and connected and if the Kakutani separation property holds for convex subsets of polytopes
842 C.D. Horvath / Topology and its Applications 155 (2008) 830–850

(or more generally the S4-weak property) with the induced convexity then (X, C) has the weak Van de Vel Property
holds.

Definition 5.2. An indexed family of subsets (A0 , . . . , Am ) of a set X is a KKM family with respect to a convexity
 a finite set of points {p0 , . . . , pm } ⊂ X such that for all set of indices J ⊂ {0, . . . , m} we have
C on X if there exists
J{pj : j ∈ J }K ⊂ j ∈J Aj .
The convexity C is KKM with respect to the family A of subsets of X if for  all indexed family of subsets (A0 ,
. . . , Am ) composed of elements of A which is KKM with respect to C we have m i=0 Ai = ∅.

One translation of the classical KKM Lemma in this language is that the usual convexity of a convex subset of an
arbitrary topological vector space is KKM with respect to the family of closed sets.
From Theorem 1 in [17] a convexity with homotopically trivial polytopes is KKM with respect to the family of
closed sets and with respect to the family of open sets; recalling Corollary 3.1 we obtain

Proposition 5.1. If (X, C) has the weak Van de Vel Property and if polytopes are metrizable then C is KKM with
respect to the family of closed subsets of X and also with respect to the family of open subsets of X.

Without the metrizability assumption polytopes might not be homotopically trivial, and therefore Theorem 1 in
[17] is not available anymore. But the KKM property still holds for closed sets.

Theorem 5.1. If (X, C) has the weak Van de Vel Property then C is KKM with respect to the family of closed sets.

Proof. First, we notice that we can assume that X itself is compact and that  C is compatible with the uniformity of
X; indeed, if P = J{p0 , . . . , pm }K is a polytope such that J{pj : j ∈ J }K ⊂ j ∈J Fj , where (F0 , . . . , Fm ) is a family
of closed subsets of X, we can replace X by P and Fj by P ∩ Fj .
(A) Next, we claim (and we will prove it in (C)), that if (X, U, C) is a uniform convex space with the Van de Vel
Property and if (A0 , . . . , Am ) is a family of open sets which is KKM with respect to C then, for all R ∈ U there exists
a lower semicontinuous map ΓR : Δm → X, where Δm is the standard m-dimensional simplex, with closed nonempty
convex values and a single valued map γR : Δm → X such that:

(1) for all set of indices J , ΓR (ΔJ ) ⊂ j ∈J Aj and
(2) for all y ∈ Δm , ΓR y ⊂ RγR (y).

Assuming that (A) holds, let ωR,j = {y ∈ Δm : ΓR y ∩ Aj = ∅}; from the lower semicontinuity of ΓR we have that
ωR,j is an open subset of Δm and from (2) we have ΔJ ⊂ j ∈J ωR,j for all sets of indices J ⊂ {0, . . . , m}. From
 m
the standard KKM Theorem, for open sets, we have m j =0 ωR,j = ∅; if y ∈ Δm is a point of j =0 ωR,j then, by (2),
RγR (y) ∩ Aj = ∅ for all j . We have shown that
m −1 A
(3) for all R ∈ U , j =0 R j = ∅.

(B) Now we assume that (X, U, C) is a compact (uniform) convex space with the Van de Vel Property and that
(F0 , . . . , Fm ) is a family of closed sets which is KKM with respect to C. For all R ∈ U choose −1
m S ∈ U such that S = S
and S ◦ S ⊂ R; then SA ⊂ RA for all subsets A ⊂ X. From (3) applied to S we obtain j =0 RAj = ∅. If R ⊂ R1 ∩ R2
 m m  m
then m j =0 RAj ⊂ ( j =0 R1 Aj ) ∩ ( j =0 R2 Aj ); since X is compact we must have R∈U ( j =0 RAj ) = ∅ and this
m
implies that j =0 Aj = ∅.
(C) To complete the proof we establish the claim made  in A. For all y = (t0 , . . . , tn ) ∈ Δn let σ (y) = {i: ti >
0}; given {p0 , . . . , pn } ⊂ X such that J{pj : j ∈ J }K ⊂ j ∈J Aj for all nonempty subset of indices J ⊂ {0, . . . , n},
let Ωy = J{pj : j ∈ σ (y)}K. By definition, Ω : Δn → J{p0 , . . . , pn }K has convex values; we show that it is lower
semicontinuous. Let V (y) = {y ∈ Δn : σ (y) ⊂ σ (y )}; by continuity of the barycentric coordinate maps V (y) is
an open neighborhood of y, furthermore, for all y ∈ V (y) we have Ω(y) ⊂ Ω(y ) and consequently, for all U ⊂
J{p0 , . . . , pn }K, if Ωy ∩ U = ∅ then Ωy ∩ U = ∅ for all y ∈ V (y). The claim now follows from Theorem 3.1. 2
 C.D. Horvath / Topology and its Applications 155 (2008) 830–850 843

Corollary 5.1. A convexity C on a topological space X for which

(1) polytopes are compact and connected;

(2) for all polytopes, the induced convexity is compatible with the uniformity;
(3) all polytopes satisfy S4-weak

is KKM with respect to the family of closed sets. Furthermore, if polytopes are also metrizable then C is also KKM
with respect to the family of open sets.

Given a topological space X endowed with a convexity C we say that a map Ω : Y → X is KKM with respect to
the convexity C, or simply KKM, if, for all nonempty finite subset {y0 , . . . , ym } ⊂ Y the family {Ωy0 , . . . , Ωym } is
KKM. From the definitions we have the well known KKM principle.

Proposition 5.2. If C is a convexity on the topological space which is KKM with respect tothe family of closed sets
and if Ω : Y → X is a KKM map with closed values, at least one of which is compact, then y∈Y Ωy = ∅.

If Y is a subset of X we say that Ω : Y → X is a Ky Fan map, with respect to the convexity C on X, if:

(KF1) for all y ∈ Y , y ∈ Ωy;

(KF2) for all x ∈ X, Y \ Ω −1 x is convex.

Lemma 5.1. Ky Fan maps are KKM maps.


Proof. We show that for all finite subsets {y0 , . . . , ym } of Y , J{y0 , . . . , ym }K ⊂ ni=0 Ωyi . For a contradiction, n assume
that there exists a finite nonempty subset {y 0 , . . . , y m } of Y such that J{y 0 , . . . , y m }K is not contained in i=0 Ωyi ;
/ ni=0 Ωyi , this is equivalent to {y0 , . . . , ym } ⊂ Y \ Ω −1 z, and therefore,
there exists z ∈ J{y0 , . . . , ym }K such that z ∈
by the convexity assumption, we obtain J{y0 , . . . , ym }K ⊂ Y \ Ω −1 z, which implies z ∈ / Ωz, in contradiction with
(KF1). 2

Proposition 5.2 and Lemma 5.1 yield the following proposition:

Proposition 5.3 (KKM principle). If (X, C) is a topological space which is endowed with a convexity which is KKM
with respect to closed sets and if Yis a subset of X then, for all Ky Fan maps with closed values Ω : Y → X, at least
one of which is compact, we have y∈Y Ωy = ∅.

The most immediate fixed point result is:

Proposition 5.4 (Fan–Browder’s Fixed Point Theorem). Let (X, C) be a topological space which is endowed with a
convexity which is KKM with respect to closed sets and let Y be a compact convex subset of X. If Γ : X → X is a map
with nonempty convex values such that, for all y ∈ Y , Γ −1 y is open and Γ y ∩ Y = ∅ then there exists y  ∈ Y such
that y  ∈ Γ y  .

Proof. For a contradiction, assume that Γ : X → X has no fixed point; then for all x ∈ X, x ∈ X \ Γ −1 x. Let
Ωx = X \ Γ −1 x; since X \ Ω −1 x = Γ x for all x ∈ X, we have that Ω : X → X is a Ky Fan map with closed values.
For all y ∈ Y , y ∈ Ωy ∩ Y , and since
  Y is convex, y → Ωy ∩ Y is a Ky Fan map with compact values and therefore
−1 y) = ∅, in other words, there exists y  ∈ Y such that Γ y  ∩ Y = ∅, which
y∈Y Ωy ∩ Y = ∅, or Y ∩ (X \ y∈Y Γ
is a contradiction. 2

Definition 5.3. A convexity C on a topological space X is a Φ-convexity if for all compact convex subset Y of X and
for all open set U ⊂ Y × Y containing ΔY = {(y, y) ∈ Y × Y } there exists R : X → X reflexive with nonempty convex
values and open fibers such that R ∩ (Y × Y ) ⊂ U . If C is a Φ-convexity on X we will say that (X, C) is Φ-convex.
844 C.D. Horvath / Topology and its Applications 155 (2008) 830–850

Let X and Y be to sets endowed with convexities CX and CY ; following Van de Vel, we will say that a map
f : X → Y is convexity preserving if, for all C ∈ CY the set f −1 (C) belongs to CX . If for all pair of distinct points
(x, x ) of X there is a continuous convexity preserving map f : X → R such that f (x) = f (x ) we say that (X, CX )
has enough convexity preserving maps.

Lemma 5.2. A convexity C on a topological space X for which there are enough convexity preserving maps is a
Φ-convexity.

Proof. Let Y be a compact convex subset of X and U ⊂ Y × Y an open set such that ΔY ⊂ U . For all pairs (x, x ) ∈
Y ×Y \U there exists a continuous convexity preserving map ϕ(x,x ) : X → R such that ϕ(x,x ) (x)−ϕ(x,x ) (x ) > 0; de-
fine f(x,x ) : X × X → R by f(x,x ) (y, y ) = ϕ(x,x ) (y) − ϕ(x,x ) (y ) and choose λ(x,x ) in ]0, f(x,x ) (x, x )[. From com-

pactness of Y × Y \ U there is a finite set {(x1 , x1 ), . . . , (xm , xm )} such that Y × Y \ U ⊂ m
i=1 {(x, x ): f(x ,x ) (x, x ) >
i i
λ(x ,x ) }; choose λ in the interval ]0, min1im {λ(x ,x ) }[ and let f (x, x ) = max1im {f(x ,x ) (x, x )}. Notice that f
i i i i i i
is continuous, that for all x ∈ X, f (x, x) = 0 and therefore Y × Y \ U ⊂ {(x, x ): λ  f (x, x )}.
The relation R = {(x, x ) ∈ X × X : f (x, x ) < λ} is reflexive, since 0 < λ; also Rx = {x ∈ X: f (x, x ) < λ}
= m −1
i=1 {x ∈ X: f(xi ,xi ) (x, x ) < λ} is convex and R x = {x ∈ X: f (x, x ) < λ} is open; and we have ΔY ⊂
(Y × Y ∩ R) ⊂ U by construction. 2

Lemma 5.3. Let (X, C) be Φ-convex and assume that C is KKM with respect to closed sets; then, any continuous map
f : X → Y from X to a compact convex subset Y of X has a fixed point.

Proof. Let R : X → X be a reflexive map with nonempty convex values and open inverse images and let Γ x = Rf (x).
Since f : X → X is continuous and Γ −1 x = f −1 (R −1 (x)) we have that Γ has nonempty convex values and Γ −1 x is
open for all x ∈ X; furthermore, for y ∈ Y we have f (y) ∈ Y ∩ Rf (y), by the reflexivity of R. Both ΔY and GY (f ) =
{(y, f (y)): y ∈ Y } are closed subsets of the compact set Y × Y ; if f (y) = y for all y ∈ Y then ΔY ∩ GY (f ) = ∅. By
compactness, there is an open set U ⊂ Y × Y containing ΔY such that U ∩ GY (f ) = ∅; the convexity being Ky Fan
there exists R such that ΔY ⊂ (R ∩ Y × Y ) ⊂ U , and therefore (R ∩ Y × Y ) ∩ GY (f ) = ∅, contrary to what has been
established at the beginning of the proof. 2

The next statement is an obvious consequence of Lemmas 5.2 and 5.3.

Lemma 5.4. If there is on a compact topological space X a convexity C which is KKM with respect to closed sets and
for which there are enough convexity preserving maps, then any continuous map f : X → X has a fixed point.

5.2. Fixed points

We begin with an intermediary result which surprisingly links the fixed point property to separation properties. In its
final version, without the completeness assumption on the uniformity, that result will simply be Schauder–Tychonov’s
Fixed Point Theorem in uniform convex spaces.

Proposition 5.5. Let (X, U, C) be a complete uniform convex space with compact and connected polytopes. If one of
the following sets of hypotheses holds then all continuous compact maps f : X → X have a fixed point.

(A) all polytopes satisfy S4-weak and there are enough convexity preserving maps or
(B) the convexity has the Kakutani Property.

Proof. According to Theorem 2.8 in [26] the closed convex hull of a compact set is compact. If (A) holds then the
conclusion follows from Lemma 5.4 and Corollary 5.1.
Assume that (B) holds. Then, according to Theorem 2.7 in [26] there are enough continuous convexity preserving
maps from X to R, and therefore (A) is also the case. 2
 C.D. Horvath / Topology and its Applications 155 (2008) 830–850 845

Lemma 5.5. Let (X, U) be a Hausdorff uniform space and Ω : X → X an upper semicontinuous compact map with
closed values, then Ω has a fixed point if and only if, for all R ∈ U , there exists xR ∈ X such that xR ∈ R(ΩxR ).

Proof. Let K be the closure of Ω(X); for all R ∈ U there exists xR ∈ X and uR ∈ ΩxR such that xR ∈ RuR ; (uR )R∈U
is a generalized sequence in K. By compactness of K, there exists a base B of U such that the generalized sequence
(uR )R∈B converges in K; let x be its limit. Let T ∈ U be an arbitrary element of the uniformity and choose S ∈ U
such that S ◦ S ⊂ T . There exists R0 ∈ B such that, for all R ∈ B with R ⊂ R0 , uR ∈ Sx; without loss of generality
we can assume that R0 ⊂ S, from which we obtain RuR ⊂ S(Sx) if R ⊂ R0 ; since xR ∈ RuR for all R, we have
shown that for all R ∈ B with R ⊂ R0 , xR ∈ T x. Since T was an arbitrary element of U the generalized sequence
R )R∈B converges in X to x. If x ∈
(x / Ωx then there exists R ∈ U such that x ∈ / R(Ωx) (since, for all A ⊂ X, A =
R∈U R(A)); choose R 1 ∈ U , symmetrical, such that R1 ◦ R1 ⊂ R, then R 1 x ∩ R1 (Ωx) = ∅. By upper semicontinuity
of Ω there is a neighborhood V of x such that ΩV ⊂ R1 (Ωx); we can also find R0 ∈ B such that, for all R ∈ B
with R ⊂ R0 , xR and uR are in V , but then uR ∈ R1 (Ωx). If V ⊂ R1 x, which we can always assume, we obtain
a contradiction. 2

Lemma 5.6. Let (X, U, C) be a uniform convex space such that convex hulls of compact sets are compact. If Γ : Y →
X is an upper semicontinuous map with compact values then y → JΓ yK is upper semicontinuous.

Proof. Let U ⊂ X be an open set and y a point in Y such that JΓ yK ⊂ U . By Corollary 2.1 there exists an open
convex set V ⊂ X such that JΓ yK ⊂ V ⊂ U . From Γ y ⊂ JΓ yK and the upper semicontinuity of Γ there exists an
open neighborhood Wy of y such that Γ (Wy ) ⊂ V , finally, by convexity of V , we have JΓ yK ⊂ V for all y ∈ Wy . 2

Theorem 5.2 (Fan–Himmelberg’s theorem). A uniform convex space (X, U, C) with compact, metrizable and con-
nected polytopes with the Kakutani Property has the fixed point property for upper semicontinuous compact maps
with closed convex nonempty values.

Proof. First, we assume that there exists a compact convex set K such that Ω(X) ⊂ K. Consider K with the induced
uniformity U|K and the induced convexity C|K ; (K, U|K , C|K ) is a uniform convex space with compact and connected
polytopes, and with the Kakutani Property. The map x → Ωx ∩ K = Ωx is upper semicontinuous with nonempty
convex and compact values. Let ΔK = {(x, x): x ∈ K}; if Ω has no fixed point then (K × K) \ ΔK is an open
neighborhood of the graph of Ω : K → K. Being compact, K is paracompact, there is therefore, by Theorem 3.5, a
continuous map f : K → K whose graph is contained in (K × K) \ ΔK .
By compactness, the induced uniformity C|K is complete, and by part (B) of Proposition 5.5, there is a point x ∈ K
such that f (x) = x. We have reached a contradiction.
Next, we remove the convexity assumption on K. That is K ⊂ X is compact and Ω(X) ⊂ K.
Since all uniformities have a base of symmetric closed graph relations, given R ∈ U we can find S ∈ U , symmetric
and closed, such that, for all convex sets C, JS(C)K ⊂ R(C). Since K is compact we can find a finite subset F ⊂ K
such that K ⊂ S(F ); let P = JF K, we have K ⊂ S(P ). For x ∈ P we have Ωx ⊂ K and therefore Ωx ⊂ S(P ) and,
since S is symmetric S(Ωx) ∩ P = ∅. From the compactness of the values of Ω and the closedness of S we have that
S(Ωx) is closed for all x and that x → S(Ωx) ∩ P is an upper semicontinuous map, with compact values, from P
to P . Since P is compact, the induced uniformity is complete, and therefore the convex hull of a compact set in P is
compact (Theorem 2.8 in [26]); from Lemma 5.6 we infer that x → JS(Ωx)∩P K is an upper semicontinuous map with
compact convex values from P to P . From the first part of the proof there exists x ∈ P such that x ∈ JS(Ωx) ∩ P K;
from JS(Ωx) ∩ P K ⊂ JS(Ωx)K ⊂ R(Ωx) we have x ∈ R(Ωx). Lemma 5.5 completes the proof. 2

Corollary 5.2 (Schauder–Tychonov’s Theorem). A uniform convex space (X, U, C) with compact, metrizable and
connected polytopes with the Kakutani Property has the fixed point property for continuous compact maps.
846 C.D. Horvath / Topology and its Applications 155 (2008) 830–850

Appendix A. On the Kakutani Property

Definition A.1. A convex space (X, C) is join-hull commutative if, for all subsets S of X and for all point p ∈ X we
have
q y
S ∪ {p} = Jx, pK.
x∈JSK

Definition A.2. A convex space (X, C) has the Pash–Peano Property if for all quintuple (a, b1 , b2 , c1 , c2 ) of points of
X such that ci ∈ Ja, bi K one has
Jb1 , c2 K ∩ Jb2 , c1 K = ∅.

Proposition A.1. A join-hull commutative convex space (X, C) has the Pash–Peano Property if and only if it has the
Kakutani Property.

Proof. Assume that the Pash–Peano Property holds. Let Z be the family of pairs of disjoint convex sets (D1 , D2 )
such that Ci ⊂ Di partially ordered by (D1 , D2 ) ⊂ (D1 , D2 ) if Di ⊂ Di . The pair (C  to Z and if
1 , C2 ) belongs
C = {(D1,λ , D2,λ ): λ ∈ Λ} is a chain in Z, that is, a totally ordered subset of Z, then ( λ∈Λ D1,λ  λ∈Λ D2,λ ) ∈ Z
,
since an up-directed union of convex sets is convex and, as can easily be seen, ( λ∈Λ D1,λ ) ∩ ( λ∈Λ D2,λ ) = ∅; by
Zorn’s Lemma there is a maximal element (H1 , H2 ) in Z. Assume that there is a point a in X \ (H1 ∪ H2 ); from
the maximality of the pair (H1 , H2 ) we have JH1 ∪ {a}K ∩ H2 = ∅ and JH2 ∪ {a}K ∩ H1 = ∅, take a point c1 in the
first set and a point c2 in the second set. By join-hull commutativity there exists bi ∈ Hi such that ci ∈ Ja, bi K. By
the Pash–Peano Property there exists a point u in Jb1 , c2 K ∩ Jb2 , c1 K. From b1 , c2 ∈ H1 and b2 , c1 ∈ H2 we obtain
u ∈ H1 ∩ H2 , which is impossible since the pair (H1 , H2 ) is in Z.
Assume now that the Kakutani Property holds but not the Pash–Peano Property. Let (a, b1 , b2 , c1 , c2 ) be a quintuple
of points of X such that ci ∈ Ja, bi K and Jb1 , c2 K ∩ Jb2 , c1 K = ∅. There exists a convex set C such that X \C is convex,
Jb2 , c1 K ⊂ C and Jb1 , c2 K ⊂ X \C. We can assume that a ∈ C; then c2 ∈ Ja, b2 K ⊂ C. But c2 ∈ Jb1 , c2 K ⊂ X \C; we
have reached a contradiction. 2

The first part of the proof is exactly the proof of the standard algebraic Hahn–Banach Theorem in vector spaces as
it can be found, for example, in the book of Holmes [15]. Proposition A.1 has no claim to originality, the proof can be
found, for example, in [9], Proposition 22.
If the convexity is not joint-hull commutative then the Kakutani Property is equivalent to a generalized Pash–Peano
Property.

Definition A.3. A convex space (X, C) has the Generalized Pash–Peano Property if for all quintuple (a, B1 , B2 , c1 , c2 ),
where a, c1 , c2 are points of X and B1 , B2 are nonempty finite subsets of X such that ci ∈ Ja, Bi K one has
q y q y
{c2 } ∪ B1 ∩ {c1 } ∪ B2 = ∅.

Proposition A.2. A convex space (X, C) has the Kakutani Property if and only if it has the Generalized Pash–Peano
Property.

Proof. Assume, for a contradiction, that (X, C) has the Kakutani Property and does not have the Generalized Pash–
Peano Property. We can then find (a, B1 , B2 , c1 , c2 ) as above and a convex set C such that J{c2 } ∪ B1 K ⊂ C and
J{c1 } ∪ B2 K ⊂ X \C. We can assume that a ∈ C; from B1 ⊂ J{c2 } ∪ B1 K ⊂ C and from the convexity of C we obtain
J{a} ∪ B1 K ⊂ C. From c1 ∈ J{a} ∪ B1 K we finally have c1 ∈ C ∩ J{c1 } ∪ B2 K ⊂ C ∩ X \C.
To see that the Generalized Pash–Peano Property implies the Kakutani Property one proceeds exactly as in the
proof of Proposition A.1 with the final argument modified as follows. We have two points c1 and c2 such that c1 ∈
J{a} ∪ H1 K ∩ H2 and c2 ∈ J{a} ∪ H2 K ∩ H1 . From J{a} ∪ Hi K = B∈{a}∪Hi JBK we find two finite sets B1 ⊂ H1 and
B2 ⊂ H2 such that c1 ∈ J{a} ∪ B1 K ∩ H2 and c2 ∈ J{a} ∪ B2 K ∩ H1 . From the Generalized Pash–Peano Property we
have J{c1 } ∪ B2 K ∩ J{c2 } ∪ B1 K = ∅. Finally, from c1 ∈ H2 , B2 ⊂ H2 , c2 ∈ H1 , B1 ⊂ H1 and the convexity of H1 and
H2 we have J{c1 } ∪ B2 K ∩ J{c2 } ∪ B1 K ⊂ H1 ∩ H2 , which is in contradiction with H1 ∩ H2 = ∅. 2
 C.D. Horvath / Topology and its Applications 155 (2008) 830–850 847

Notice that the Generalized Pash–Peano Property holds if and only if it holds for all polytopes with the induced
convexity. As a consequence of Proposition A.2 we obtain the following result of Keimel and Wieczorek [18].

Corollary A.1 (Keimel–Wieczorek). A convex space (X, C) has the Kakutani Property if and only if all polytopes have,
with respect to the induced convexity, the Kakutani Property.

Property S4-weak is related to the screening property introduced by De Groot and Aarts in [11] where two subsets
A and B of a set X are said to be screened by a pair (C, D) of subsets of X if X = C ∪ D, A ∩ C = ∅ and B ∩ D = ∅.
Condition S4-weak says that every pair of disjoint open convex subsets of a polytope is screened by a pair of closed
convex subsets of that polytope.

Appendix B. Some examples

B.1. Hyperconvex metric spaces

A metric space (X, d) is an hyperconvex  metric space if for all collections {(xi , ri )}i∈I ⊂ X × R+ such that for all
(i, j ) ∈ I × I d(xi , xj )  ri + rj one has i∈I B[xi , ri ] = ∅, where B[x, r] is the closed ball of radius r centered at x.
An hyperconvex metric space is complete and is a nonexpansive retract of any metric space in which it is embed-
ded [2]. Fixed point and selection properties for hyperconvex metric spaces have been studied in [3,16,23] and KKM
like properties in [19].
Let C be the collection of subsets C of X such that for all nonempty finite subset A of C one has

B[x, r]: A ⊂ B[x, r] ⊂ C.

This defines a convexity on X, the ball convexity, for which polytopes are exactly sets of the form {B[x, r]:
A ⊂ B[x, r]} with A finite. With the induced metric this set is itself hyperconvex. Furthermore, for all C ∈ C and for
all ε > 0
B(x, ε) ∈ C.
x∈C

This shows that the ball convexity is compatible with the metric uniformity. In conclusion:

An hyperconvex metric space equipped with the ball convexity is a uniform convex space with homotopically trivial
polytopes.

B.2. Normally supercompact spaces

A topological space (X, τ ) is normally supercompact if there exists a subbase of closed sets B which is a normal
family and such that any subfamily A ⊂ B whose members have nonempty pairwise intersections has a nonempty
intersection. Normally supercompact spaces have been extensively studied. The information needed here can be found
in one of the first paper on the subject [25]; one could also look at [16]. Given a normally supercompact topological
space and a closed subbase B as above let, for all finite subsets A of X,

JAK = {B ∈ B: A ⊂ B}

and let C be the family of subsets C of X such that, for all finite subsets A ⊂ C, JAK ⊂ C. This defines a convexity on
X whose polytopes are exactly the sets JAK; since they are closed by construction they are compact. If the topology is
metrizable then there exists a metric d on X for which ε-neighborhood of polytopes are in C; also, if X is connected
then polytopes are contractible. In conclusion:

A normally supercompact connected and metrizable topological space can be endowed with a convexity for which
it becomes a uniform convex space with homotopically trivial polytopes.
848 C.D. Horvath / Topology and its Applications 155 (2008) 830–850

B.3. Topological semilattices

A Topological semilattice is a pair (X, ) where X is a topological space,  is a partial order on X for which each
pair of elements (x1 , x2 ) ∈ X 2 has a least upper bound x1 ∨ x2 and the map (x1 , x2 ) → x1 ∨ x2 is continuous. For
finite nonempty subsets A of X let
  
JAK = a, A
a∈A

where, for a  b, [a, b] = {x ∈ X: a  x  b} and A is the least upper bound of the nonempty finite set A. Declare
C ⊂ X convex if for all nonempty finite subsets A of C one has JAK ⊂ C. This is a convexity on X whose nonempty
polytopes are the sets JAK. One can check that C is convex if and only if, for all x1 , x2 ∈ C, [x1 , x1 ∨ x2 ] ⊂ C holds
and that the Pash–Peano Property holds. If X is compact then this convexity is compatible with the uniformity of X
if each point has a neighborhood base consisting of subsemilattices and (x1 , x2 ) → [x1 , x1 ∨ x2 ] is continuous map
from X 2 to the space of nonempty compact subspaces of X, [26], in which case, by Theorem 2.7 of [26] there are
enough convexity preserving maps to separate points of X.
A lemma of Brown [7] implies that a pathconnected bounded topological semilattice is homotopically trivial;
consequently if, for all x1 , x2 ∈ X such that x1  x2 the interval [x1 , x1 ∨ x2 ] is path connected, polytopes will be path
connected bounded semilattices and therefore homotopically trivial. Brown’s Lemma does not assume compactness.

B.4. B-convexity

In [5] and [6] a subset C of Rn+ is called B-convex if for mall m ∈ N \{0}, for all (x1 , . . . , xm ) ∈ C and for all
m

(t1 , . . . , tm ) ∈ [0, 1] such that max{t1 , . . . , tm } = 1 one has i=1 ti xi ∈ C, where

m


m  
yi = max {yi,1 }, . . . , max {yi,n }
1im 1im
i=1

is the least upper bound of the vectors (y1 , . . . , ym ) with respect to the partial order of Rn defined by the standard
positive cone Rn+ . One can check that C ⊂ Rn+ is B-convex if and only if, for all x, y ∈ C and for all t ∈ [0, 1], tx ∨ y ∈
C. This is a convexity on Rn+ for which the Pash–Peano Property holds. Polytopes are compact and contractible. With
respect to the metric d(x, y) = x − y∞ = max1j n |xj − yj | ε-neighborhoods of B-convex sets are B-convex.

B-convexity is a uniform join-hull commutative convexity on Rn+ with the Pash–Peano Property and with compact
and connected polytopes.

For B-convexity one can directly and easily show that convex sets are contractible.

B.5. Max-Plus convexity

A subset C of (R ∪ {−∞})n is Max-Plus convex if, for all m ∈ N \{0}, for all (x1 , . . . , xm ) ∈ C m and for all
(t1 , . . . , tm ) ∈ [−∞, 0]m such that max{t1 , . . . , tm } = 0 one has
 
max {xi,1 + ti }, . . . , max {xi,n + ti } ∈ C.
1im 1im

One defines a metric on (R ∪ {−∞})n by DMax (x, y) = max1in |exi − eyi |.

The map x → (ex1 , . . . , exn ) is a distance preserving homeomorphism from (R ∪ {−∞})n with the DMax metric
to Rn+ with the metric d(x, y) = x − y∞ . Max-Plus convex sets are mapped to B-convex sets and the inverse map,
that is x → (ln x1 , . . . , ln xn ), maps B-convex sets to Max-Plus convex sets.
The literature on Max-Plus convexity and its applications is very large. Michael’s Selection Theorem for Max-Plus
convex sets is proved by M. Zarichnyi in [28], where additional references on Max-Plus convexity can also be found.
 C.D. Horvath / Topology and its Applications 155 (2008) 830–850 849

Appendix C. Conclusion and acknowledgement

Convexity is multifaceted; its study, and therefore its diverse generalizations, could favor the algebraic, the combi-
natorial, the topological, the functional or the geometric aspect depending on one’s motivations.
A convexity like structure on a topological space might yield fixed point properties, selection theorems or ANR or
AR properties. In this context the important objects are the “convex sets” or, more precisely, the “polytopes”; there
are quasiconvex maps (real valued maps whose sublevel sets are convex), quasiconcave maps (maps whose negative
is quasiconvex) and quasiaffine maps (simultaneously quasiconvex and quasiconcave) but, in this context, one could
ask what a convex map should be; and, even in a framework like geodesic convexity, where a natural definition of
convex maps is available, the existence of nonconstant convex maps might be problematic.
For an easily formulated, but apparently very difficult problem, on the convex hull of three points with respect to
the geodesic convexity on a Riemannian manifold see page 231 of [4].
Convexity—of maps—is an important and useful property in optimization and mathematical programming and one
has therefore been led to various generalizations of the concept of convex maps, as, for example, in [21,22] or [24].
Convex sets play a very modest role, if they are defined at all, and when they are defined, as in [10], for example, their
properties are not very well understood.
One could say that convexity has grown two branches, generalized convex sets and generalized convex maps, with
practically no interaction. B-convexity is one example of a generalized convexity which has both nice topological and
functional properties.
Murota’s book [20], which is mainly on the functional side of generalized convexity, finds its motivation, and its
applications, in discrete optimization. Coppel’s book [9] offers a beautiful axiomatic study of convexity.
The ideas and results presented in this paper owe a lot to the work of Maurice Van de Vel [26,27].
Professor Valentin Gutev read a first draft of this paper. In that version Lemma 3.1 had a much more restricted
form and a rather long and difficult proof. He showed the author how to use Theorem 3.1 of [14]. Also, the proof of
Theorem 3.1 of this paper incorporates some of his suggestions.

References

[1] F.D. Ancel, The role of countable dimensionality in the theory of cell-like relations, Trans. Amer. Math. Soc. 287 (1) (1985) 1–40.
[2] N. Aronszajn, P. Panitchpakdi, Extensions of uniformly continuous transformations and hyperconvex metric spaces, Pacific J. Math. 6 (1956)
405–439.
[3] J.B. Baillon, Nonexpensive mappings and hyperconvex metric spaces, Contemp. Math. 72 (1988) 11–19.
[4] M. Berger, Riemannian Geometry, a Panoramic View, Springer-Verlag, 2004.
[5] W. Briec, C. Horvath, B-convexity, Optimization 53 (2) (2004) 103–127.
[6] W. Briec, C. Horvath, A. Rubinov, Separation in B-convexity, Optimization 53 (2) (2004) 103–127.
[7] D.R. Brown, Topological semilattices on the two cell, Pacific J. Math. 15 (1) (1965) 35–46.
[8] A. Cellina, Approximation of set valued functions and fixed point theorems, Ann. Mat. Pura Appl. 82 (1969) 17–24.
[9] W.A. Coppel, Foundations of Convex Geometry, Cambridge University Press, 1998.
[10] B. Dacorogna, Some geometrical and algebraic properties of various types of convex hulls, in: P. Alart, O. Maisonneuve, R.T. Rockafellar
(Eds.), Nonsmooth Mechanics and Analysis, Theoretical and Numerical Advances, Springer, 2006, pp. 25–34.
[11] J. De Groot, J.M. Aarts, Complete regularity as a separation axiom, Canad. J. Math. 21 (1969) 96–105.
[12] J. Dugundji, Topology, Allyn and Bacon, 1966.
[13] J. Dugundji, A. Granas, Fixed Point Theory, Springer-Verlag, 2003.
[14] V. Gutev, Selections and approximations in finite dimensional spaces, Topology Appl. 146–147 (2005) 353–383.
[15] R.B. Holmes, Geometric Functional Analysis and its Applications, Springer-Verlag, 1975.
[16] C. Horvath, Extensions and selections in topological spaces with a generalized convexity structure, Annales de la Faculté des Sciences de
Toulouse 2 (2) (1993) 253–269.
[17] C. Horvath, M. Lassonde, Intersection of sets with n-connected unions, Proc. Amer. Math. Soc. 125 (4) (1997) 1209–1212.
[18] K. Keimel, A. Wieczorek, Kakutani property of the polytopes implies Kakutani property of the whole space, Math. Ann. Appl. 130 (1988)
97–109.
[19] M.A. Khamsi, KKM and Ky Fan theorems in hyperconvex metric spaces, JMAA 204 (1996) 298–306.
[20] K. Murota, Discrete Convex Analysis, SIAM Monographs on Discrete Mathematics and Applications, 2003.
[21] D. Pallaschke, S. Rolewicz, Foundations of Mathematical Optimization, Convex Analysis without Convexity, Kluwer Academic Publishers,
1997.
[22] A. Rubinov, Abstract Convexity and Global Optimization, Kluwer, 2000.
[23] R. Sine, Hyperconvexity and approximate fixed points, Nonlinear Anal. 13 (7) (1989) 863–869.
[24] I. Singer, Abstract Convex Analysis, Canadian Mathematical Society Series of Monographs and Advanced Texts, John Wiley, 1997.
850 C.D. Horvath / Topology and its Applications 155 (2008) 830–850

[25] J. Van Mill, M. Van de Vel, Convexity preserving mappings in subbase convexity theory, Proc. Kon. Ned. Akad. Wet. A 81 (1977) 76–90.
[26] M. Van de Vel, A selection theorem for topological convex structures, Trans. Amer. Math. Soc. 336 (2) (1993) 463–495.
[27] M. Van de Vel, Theory of Convex Structures, North-Holland Mathematical Library, vol. 50, 1993.
[28] M. Zarichnyi, Michael Selection theorem for Max-Plus compact convex sets, Topology Proc., submitted for publication.