J. Math. Anal. Appl.

459 (2018) 203–216

Contents lists available at ScienceDirect

Journal of Mathematical Analysis and Applications

www.elsevier.com/locate/jmaa

Common fixed point properties for a family of set-valued

mappings
Anthony To-Ming Lau a,∗,1 , Liangjin Yao b
a
Department of Mathematical and Statistical Sciences, University of Alberta, Edmonton, Alberta,
T6G 2G1 Canada
b
International College of Manitoba, Winnipeg, Manitoba, R3T 2N2 Canada

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, we study common fixed point properties for a family of mappings.
Received 27 June 2017 We present the extensions of Caristi’s fixed point theorem and Markov–Kakutani
Available online 31 October 2017 fixed point theorem for a family of set-valued mappings.
Submitted by R. Timoney
© 2017 Elsevier Inc. All rights reserved.
Dedicated to Professor Simeon Reich
on the occasion of his 70th birthday
with warm friendship and great
respect

Keywords:
Common fixed point property
Set-valued mapping
Invariant mean
Fan–Glicksberg fixed point theorem
Caristi’s fixed point theorem
Markov–Kakutani fixed point
theorem

1. Introduction

Throughout this paper, we assume that X is a real Banach space with norm  · . We first introduce
some standard notation. Given f : X → ]−∞, +∞], we set dom f := f −1 (R). We say that f is proper if
dom f = ∅. Let f be proper. Set argmin f := {x ∈ X | f (x) = inf f }. Let f : ]−∞, +∞] → ]−∞, +∞]. We
say that f is increasing if f (x) ≤ f (y) for every x ≤ y with x, y ∈ dom f .
The distance function to a nonempty subset C of X, written as dist(·, C), is defined by

  
x −→ inf x − c | c ∈ C .

* Corresponding author.
E-mail addresses: [email protected] (A.T.-M. Lau), [email protected] (L. Yao).
1
Supported by NSERC Grant MS ZC912.

https://doi.org/10.1016/j.jmaa.2017.10.060
0022-247X/© 2017 Elsevier Inc. All rights reserved.
204 A.T.-M. Lau, L. Yao / J. Math. Anal. Appl. 459 (2018) 203–216

Let A : X ⇒ X be a set-valued operator (also known as multifunction) from X to X, i.e., for every x ∈ X,
   
Ax ⊆ X, and let gra A := (x, y) ∈ X × X | y ∈ Ax be the graph of A, and dom A := x ∈ X | Ax = ∅
be the domain of A.
Let C ⊆ X be a nonempty set and T : C ⇒ X be a set-valued mapping. The set of the fixed points of T
 
is Fix T := {x ∈ C | x ∈ T x}. Let S be a nonempty set, and S := Ts : C ⇒ X | s ∈ S . We say C has a
common fixed point for S if
  
Fix Ts = ∅, i.e., x ∈ C | x ∈ Ts x, ∀s ∈ S = ∅.
s∈S


In this case, Fix Ts is the set of all common fixed points of C for S.
s∈S
Set R+ := {x ∈ R | x ≥ 0} and N := {1, 2, 3, . . .}. We denote by −→ the norm convergence in X.
The rest of this paper are organized as follows. In Section 3, we present a new generalization of the Caristi’s
fixed point theorem for a family of set-valued mappings (Theorem 3.2). We also obtain an extension of the
Markov–Kakutani fixed point theorem: Theorem 4.3 in Section 4. Some open interesting problems are listed
and discussed in Section 5.

2. Preliminaries

All topologies in this paper are assumed to be Hausdorff. Let E be a separated locally convex space
and C ⊆ E. For x, y ∈ E, we set [x, y] := {tx + (1 − t)y | 0 ≤ t ≤ 1}. Note that [x, x] = {x}. Let

T : E ⇒ E be a set-valued mapping. Recall that T (C) := x∈C T x and the range of T is ran T := T (E).
Let T1 : E ⇒ E be another set-valued mapping. We say that T commutes with T1 on the right with respect
to C if T1 (T x) ⊆ T (T1 x), ∀x ∈ C.
The inverse of T is written as T −1 , which is defined by
 
gra T −1 := (x, y) ∈ E × E | (y, x) ∈ gra T ,

i.e.,

y ∈ T −1 x ⇐⇒ x ∈ T y.

Note that Fix T = Fix T −1 .

We recall the well known Fan–Glicksberg fixed point theorem: Lemma 2.1 (see [6,8]), which plays an
important role in the proofs of Lemma 4.1 and Theorem 4.3.

Lemma 2.1 (Fan–Glicksberg). (See [6, Theorem 1] and [8].) Let E be a separated locally convex space and
let C ⊆ E be a nonempty compact convex set. Let T : C ⇒ C be such that gra T is closed and that T x is a
nonempty convex set for all x ∈ C. Then Fix T = ∅.

The following result is directly deduced by Lemma 2.1.

Corollary 2.2. Let E be a separated locally convex space, and C ⊆ E be a nonempty compact convex set.
Let T : E ⇒ E be such that gra T is closed and T x ∩ C is a nonempty convex set for every x ∈ C. Then
Fix T ∩ C is a nonempty compact set.

Proof. Let T : C ⇒ C be defined by

gra T := gra T ∩ C × C.
 A.T.-M. Lau, L. Yao / J. Math. Anal. Appl. 459 (2018) 203–216 205

Then we have gra T is closed since gra T is closed. By the assumption, Tx = T x ∩ C is a nonempty convex
set for every x ∈ C. Then applying Lemma 2.1 to the operator T and the compact set C, we have Fix T = ∅.
This implies that Fix T ∩ C = ∅. Since gra T is closed, Fix T is closed. Then by the compactness of C, we
have Fix T ∩ C is a nonempty compact set. 2

3. Generalization of the Caristi’s fixed point theorem

In this section, we shall present a new generalization of the Caristi’s fixed point theorem (i.e., our first
main result): Theorem 3.2. We begin with an extension of Ekeland’s variational inequality (Lemma 3.1)
that we shall use in Theorem 3.2.
The proof of Lemma 3.1 is inspired by the proof of [2, Theorem 1.45] as well as [5, Theorem 1] due to
M. Crandall.

Lemma 3.1. Let the function η : R+ → R+ be such that

(a) η is increasing on R+ .
(b) η(0) = 0 and η(t) > 0 for all t > 0.
(c) η(x + y) ≤ η(x) + η(y), ∀x, y ≥ 0.

Let f : X → ]−∞, +∞] be a proper lower semicontinuous and lower bounded function. Assume that there
exist x0 ∈ X and ε > 0 such that

f (x0 ) ≤ inf f + ε.

Then there exists y0 ∈ dom f such that

 
(i) η y0 − x0  ≤ ε.
 
(ii) η y0 − x0  ≤ f (x0 ) − f (y0 ).
 
(iii) f (x) + η x − y0  > f (y0 ), ∀x = y0 .

Proof. We obtain the sequence (xn )n∈N inductively in the following way. Set

 
Cn := x ∈ X f (x) + η x − xn  ≤ f (xn ) , ∀n ∈ N ∪ {0}.

Since xn ∈ Cn , we have Cn = ∅. Let xn+1 be such that

1 1
xn+1 ∈ Cn and f (xn+1 ) ≤ f (xn ) + inf{f (x) | x ∈ Cn }. (1)
2 2

Since xn+1 ∈ Cn ,

 
f (xn ) − f (xn+1 ) ≥ η xn+1 − xn  , ∀n ∈ N ∪ {0}. (2)

 
Then we have f (xn ) n∈N
is decreasing and lower bounded. Hence f (xn ) n∈N
is convergent.
206 A.T.-M. Lau, L. Yao / J. Math. Anal. Appl. 459 (2018) 203–216

Then every n, m ∈ N ∪ {0} with n < m, (2) shows that

m−1  
f (xn ) − f (xm ) = f (xi ) − f (xi+1 )
i=n
m−1 
m−1
   
≥ η xi+1 − xi  ≥ η xi+1 − xi  (by the assumption (c))
i=n i=n
 
≥ η xm − xn  (since η is increasing). (3)

Now we show that (xn )n∈N is convergent. It suffices to show that (xn )n∈N is a Cauchy sequence. Suppose to
the contrary that there exist ε0 > 0, and two subsequences of N: (nk )k∈N , (mk )k∈N with nk < mk such that
 
xm − xn  ≥ ε0 > 0, ∀k ∈ N.
k k

Then combining (3) and the assumptions (a) and (b), we have
 
f (xnk ) − f (xmk ) ≥ η xnk − xmk 

≥ η (ε0 ) > 0, (which contradicts that f (xn ) n∈N
is convergent).

Hence (xn )n∈N is convergent. Thus, there exists y0 ∈ X such that

xn −→ y0 . (4)

Then by (3),
     
f (xn ) ≥ lim inf f (xm ) + η xm − xn  ≥ lim inf f (xm ) + lim inf η xm − xn 
m→∞ m→∞ m→∞
 
≥ f (y0 ) + lim inf η xm − xn  (by the lower semicontinuity of f ), ∀n ∈ N ∪ {0}. (5)
m→∞

Next we claim that

   
lim inf η xm − xn  ≥ η y0 − xn  , ∀n ∈ N ∪ {0}. (6)
m→∞

Set
 
δ := inf η(t) | t > 0 .

Then δ ≥ 0. We consider two cases.

Case 1: δ > 0.
 δ
Since f (xm ) m∈N is convergent, there exists K ∈ N such that f (xm ) − f (xm+1 ) < for all m ≥ K.
2
Then (2) and our assumption (that δ > 0) show that xm = xm+1 for all m ≥ K. Hence (4) shows that

xm = y0 , ∀m ≥ K.

Hence (6) holds.

Case 2: δ = 0.
Thus there exists a positive sequence (tm )m∈N in R such that

lim η(tm ) = 0. (7)

m→∞
 A.T.-M. Lau, L. Yao / J. Math. Anal. Appl. 459 (2018) 203–216 207

Let n ∈ N ∪ {0}. We have

   
η y0 − xn  = η y0 − xm + xm − xn 
   
≤ η y0 − xm  + xm − xn  (by the assumption (a))
   
≤ η y0 − xm  + η xm − xn  (by the assumption (c)).

Hence
     
η xm − xn  ≥ η y0 − xn  − η y0 − xm  . (8)
 
Since η(t) is increasing and y0 − xm  −→ 0, (7) implies that
 
lim η y0 − xm  = 0.
m→∞

Thus combining with (8), we have

   
lim inf η xm − xn  ≥ η y0 − xn  .
m→∞

Therefore, (6) holds.

Now by (5) and (6) together, we have
 
f (xn ) ≥ f (y0 ) + η y0 − xn  , ∀n ∈ N ∪ {0}. (9)

Taking n = 0 in the above inequality, we have (ii) holds. Then by (9) again, we have
 
η y0 − x0  ≤ f (x0 ) − f (y0 ) ≤ inf f + ε − f (y0 )
≤ f (y0 ) + ε − f (y0 ) = ε.

Thus (i) holds.

Now we show that (iii) holds. Suppose to the contrary that there exists z ∈ X with z = y0 such that
 
f (z) + η z − y0  ≤ f (y0 ) and thus f (z) < f (y0 ) (by the assumption (b)). (10)

Then by (9),
   
f (z) + η z − y0  ≤ f (y0 ) ≤ f (xn ) − η y0 − xn  , ∀n ∈ N ∪ {0}.

Thus
   
f (xn ) ≥ f (z) + η z − y0  + η y0 − xn 
 
≥ f (z) + η z − xn  , ∀n ∈ N (by the assumptions (c) and (a)).

Thus z ∈ Cn . Then by (1), we have

f (xn+1 ) + f (xn+1 ) − f (xn ) ≤ f (z), ∀n ∈ N.


Since f (xn ) n∈N
is convergent and xn −→ y0 , the above inequality shows that

f (y0 ) ≤ f (z), which contradicts (10).

Hence (iii) holds. 2

208 A.T.-M. Lau, L. Yao / J. Math. Anal. Appl. 459 (2018) 203–216

Applying the extension of Ekeland’s variational principle (Lemma 3.1), we may deduce the following
Theorem 3.2, which extends Caristi’s fixed point theorem (see [4,5]) as well as Feng and Liu’s result (see [7,
Theorem 4.2]) and Jachymski’s theorem (see [12, Theorem 6]) for a family of set-valued mappings.

Theorem 3.2. Let the function η : R+ → R+ be such that

(a) η is increasing on R+ .
(b) η(0) = 0 and η(t) > 0 for all t > 0.
(c) η(x + y) ≤ η(x) + η(y), ∀x, y ≥ 0.

Let f : X → ]−∞, +∞] be a proper lower semicontinuous and lower bounded function with C := dom f . Let
 
S be a nonempty set, and S := Ts : C ⇒ X with dom Ts = C | s ∈ S . Assume that for every x ∈ dom f
and every s ∈ S, there exists z ∈ Ts x such that f (z) + η (z − x) ≤ f (x). Then

(i) argmin f ⊆ Fix Ts = ∅. Therefore, C has a common fixed point for S. Moreover, if argmin f = ∅,
s∈S
then C has infinite common fixed points for S. 
(ii) C has the unique common fixed point for S ⇐⇒ argmin f = Fix Ts is a singleton.
s∈S

Proof. We follow an idea in our [16, Theorem 2.7].

(i): We consider two cases.
Case 1 : argmin f = ∅.
Let x ∈ argmin f . Fix s ∈ S. By the assumption, there exists zs ∈ Ts x such that

f (zs ) + η (zs − x) ≤ f (x) ≤ f (zs ).

Thus η (zs − x) = 0 and then zs = x by the condition (b). Thus x ∈ Ts x and hence x ∈ Fix Ts . Then we
have

x ∈ Fix Ts , ∀s ∈ S.
 
Then we have argmin f ⊆ Fix Ts and then Fix Ts = ∅. Hence C has a common fixed point for S.
s∈S s∈S
Case 2 : argmin f = ∅. 
Clearly, ∅ = argmin f ⊆ Fix T . Now we show that Fix Ts has infinite elements.
s∈S
By the assumption, there exists xn ∈ X such that
1
f (xn ) ≤ inf f + , ∀n ∈ N. (11)
n
Fix n ∈ N. Applying Lemma 3.1, there exists yn ∈ dom f such that
   
η yn − xn  ≤ f (xn ) − f (yn ) and f (x) + η x − yn  > f (yn ), ∀x = yn . (12)

Let s ∈ S. By the assumption again, there exists zs,n ∈ Ts yn such that

 
f (zs,n ) + η zs,n − yn  ≤ f (yn ).

Then combining with the second part of (12), we have zs,n = yn and then yn ∈ Ts yn . Hence

yn ∈ Ts yn , ∀s ∈ S, ∀n ∈ N.
 A.T.-M. Lau, L. Yao / J. Math. Anal. Appl. 459 (2018) 203–216 209

This gives us that


yn ∈ Fix Ts , ∀n ∈ N. (13)
s∈S


Now suppose to the contrary that Fix Ts only has finite elements. By (13), there exists a subsequence
s∈S
of (yn )n∈N , for convenience, still written as (yn )n∈N such that

yn = y0 , ∀n ∈ N.

Then combining the first part of (12) with (11),

1
f (y0 ) = f (yn ) ≤ f (xn ) ≤ inf f + , ∀n ∈ N.
n

) ≤ inf f and hence y0 ∈ argmin f , which is a contradiction with that argmin f = ∅.

Thus f (y0
Hence Fix Ts has infinite elements. Therefore, C has infinite common fixed points for S. Combining
s∈S
all the above results, we have (i) holds.
(ii): ⇐=: Clear.

=⇒: Since Fix Ts is a singleton, (i) gives us that argmin f = ∅. Then by (i) again,
s∈S


∅ = argmin f ⊆ Fix Ts .
s∈S


Hence we have argmin f = Fix Ts is a singleton. 2
s∈S

By Theorem 3.2, we are able to obtain the following two corollaries.

Corollary 3.3. Let the function η : R+ → R+ be such that

(a) η is increasing on R+ .
(b) η(0) = 0 and η(t) > 0 for all t > 0.
(c) η(x + y) ≤ η(x) + η(y), ∀x, y ≥ 0.

Let C ⊆ X be a nonempty set, and T : C ⇒ C be a set-valued mapping such that T x is a nonempty closed
set for every x ∈ C. Suppose that the function

dist(x, T x), if x ∈ C;
f (x) :=
+∞, otherwise

is lower semicontinuous. Assume also that for every x ∈ C

   
dist(y, T y) + η x − y  ≤ x − y , ∀y ∈ T x. (14)

Then Fix T = ∅.

Proof. We have f is lower semicontinuous and lower bounded with dom f = C.

210 A.T.-M. Lau, L. Yao / J. Math. Anal. Appl. 459 (2018) 203–216

Let x ∈ C. Now we claim that there exists y ∈ T x such that

1  
f (y) + η x − y  ≤ f (x). (15)
2

We consider two cases.

Case 1 : f (x) = 0.
Since T x is a nonempty closed set, then x ∈ T x by the assumption that f (x) = 0. Then taking y = x,
we have (15) holds.
Case 2 : f (x) = 0.
Set d := f (x) = 0 and hence d > 0. By the assumption (b), η(d) > 0.
Then by the definition of f , there exists y ∈ T x such that

  η(d)   η(d)
d = f (x) > x − y  − and then d ≤ x − y  < d + . (16)
2 2

Since η(t) is increasing, we have

   
η x − y  ≥ η(d) and hence − η x − y  ≤ −η(d). (17)

Thus we have
1     1  
f (y) + η x − y  = dist(y, T y) + η x − y  − η x − y 
2 2
  1  
≤ x − y  − η x − y  (by (14))
2
  1
≤ x − y  − η(d) (by (17))
2
≤ f (x) (by (16)).

Hence (15) holds.

Then combining all the above results, we have (15) holds.
By Theorem 3.2 and (15), we have Fix T = ∅. 2

Let A, B ⊆ X be nonempty sets. The generalized Hausdorff metric is defined by

H(A, B) := max sup dist(y, A), sup dist(x, B) .

y∈B x∈A

Corollary 3.4. Let the function η : R+ → R+ be such that

(a) η is increasing on R+ .
(b) η(0) = 0 and η(t) > 0 for all t > 0.
(c) η(x + y) ≤ η(x) + η(y), ∀x, y ≥ 0.

Let T : X ⇒ X be a set-valued mapping such that T x is a nonempty closed set for every x ∈ X. Assume
that
   
H(T x, T y) + η x − y  ≤ x − y , ∀x, y ∈ X. (18)

Then Fix T = ∅.
 A.T.-M. Lau, L. Yao / J. Math. Anal. Appl. 459 (2018) 203–216 211

Proof. Let f (x) := dist(x, T x), ∀x ∈ X. Then f is lower bounded. Now we show that f is lower semicon-
tinuous. Let zn −→ z. By the definition of f , there exists z
n ∈ T zn such that

  1
f (zn ) ≥ zn − z 
n − , ∀n ∈ N. (19)
n

By (18), we have
     
zn , T z) + η zn − z  ≤ H(T zn , T z) + η zn − z  ≤ zn − z .
zn , T z) ≤ dist(
dist(

Thus there exists yn ∈ T z such that

  1   1
z  zn , T z) + ≤ zn − z  + ,
n − yn ≤ dist( ∀n ∈ N. (20)
n n

Then combining with (19),

  1   1
f (zn ) ≥ zn − z 
n − = zn − yn + yn − z 
n −
n n
    1     1 1
≥ zn − yn  − yn − z 
n − ≥ zn − yn  − zn − z  − − (by (20))
n n n
    2
= zn − z + z − yn  − zn − z  −
n
    2
≥ z − yn  − 2zn − z  −
n
  2   2
≥ dist(z, T z) − 2zn − z  − = f (z) − 2zn − z  − .
n n

Taking the limit inf in the above inequality as n −→ ∞, we have

lim inf f (zn ) ≥ f (z).

n→∞

Hence f is lower semicontinuous.

Let x ∈ X. For every y ∈ T x, we have
     
dist(y, T y) + η x − y  ≤ H(T x, T y) + η x − y  ≤ x − y .

Thus, applying Corollary 3.3, we have Fix T = ∅. 2

Remark 3.5. Corollary 3.4 generalizes [1, Theorem 3.3], in which it was assumed that T x is a compact set
(for every x ∈ X) and η is continuous.

4. An extension of Markov–Kakutani fixed point theorem

In this section, we obtain an extension of the Markov–Kakutani fixed point theorem (see Theorem 4.3
below), which is our second main result.
The result below is the key technical result for Theorem 4.3. Lemma 4.1 is inspired by [10, Theorem 1],
which slightly generalizes the Fan–Glicksberg fixed point theorem.

Lemma 4.1. Let E be a separated locally convex space and let C ⊆ E be a nonempty compact convex set.
Let T0 : C ⇒ C be such that gra T0 is closed and that T0 x is a nonempty convex set for every x ∈ C. Let
T1 : C ⇒ C be such that gra T1 is closed and that (for all x ∈ Fix T0 ) T1 x is a nonempty convex set. Assume
212 A.T.-M. Lau, L. Yao / J. Math. Anal. Appl. 459 (2018) 203–216

that T0−1 y is convex for every y ∈ T1 (Fix T0 ), and that T0 commutes with T1 on the right with respect to
Fix T0 . Then

T1 x ∩ Fix T0 = ∅, ∀x ∈ Fix T0 .

Moreover, if Fix T0 is convex, then Fix T0 ∩ Fix T1 = ∅.

Proof. By Corollary 2.2,

Fix T0 is a nonempty compact set. (21)

Let x ∈ Fix T0 . Since T0 commutes with T1 on the right with respect to Fix T0 ,

T1 x ⊆ T1 (T0 x) ⊆ T0 (T1 x) . (22)

We have T1 x is a nonempty compact convex set since gra T1 ⊆ C × C is a closed set and C is compact. For
every y ∈ T1 x, (22) shows that there exists z ∈ T1 x such that y ∈ T0 z, i.e., z ∈ T0−1 y. Thus z ∈ T0−1 y ∩ T1 x.
Hence

T0−1 y ∩ T1 x = ∅, ∀y ∈ T1 x. (23)

Since gra T0 is closed, gra T0−1 is closed. We also have T0−1 y is a nonempty convex (for every y ∈ T1 x) by
(23) and the assumption. Then applying Corollary 2.2 to T0−1 and the compact convex set T1 x, (23) shows
that Fix T0−1 ∩ T1 x = ∅. Since Fix T0−1 = Fix T0 ,

T1 x ∩ Fix T0 = ∅, ∀x ∈ Fix T0 . (24)

Now we assume that Fix T0 is convex. Then (21) shows that Fix T0 is a nonempty compact and con-
vex set. Thus applying Corollary 2.2 to T1 and the compact convex set Fix T0 , (24) shows that Fix T1 ∩
Fix T0 = ∅. 2

The following example shows that we cannot remove the constraint that T0 commutes with T1 on the
right in Lemma 4.1.

Example 4.2. Let E be a separated locally convex space with e ∈ E\{0}. Let C := [0, e], and let T0 , T1 :
C ⇒ C be defined by
      
1 1 t2 t2
T0 (te) := (1 − t)e, e , T1 (te) := + e, 1 − e , ∀0 ≤ t ≤ 1.
2 2 1 + t2 1 + t2

Then we have
⎧
⎪ 1
⎪
⎪ [(1 − t)e, e] , if 0 ≤ t < ;
⎪
⎨ 2
1
T0−1 (te) = [0, e] , if t = ;
⎪
⎪ 2
⎪
⎪[0, (1 − t)e] , 1
⎩ if < t ≤ 1.
2
Clearly, C is a nonempty compact convex set. We have gra T0 and gra T1 both are closed. For every 0 ≤
t ≤ 1, T0 (te), T0−1 (te), and T1 (te) are all nonempty convex sets. Then Fix T0 = ∅ and Fix T1 = ∅ (see
7 
Lemma 2.1). But Fix T0 ∩ Fix T1 = ∅. Indeed, we have Fix T0 = { 12 e} and T1 ( 12 e) = 10 e, 45 e . Therefore,
Fix T0 ∩ Fix T1 = ∅. Note that
 A.T.-M. Lau, L. Yao / J. Math. Anal. Appl. 459 (2018) 203–216 213

   
1 1 7 4
T1 T0 ( e) = T1 ( e) = e, e
2 2 10 5
     
1 1 7 4 1
 e, e = T0 e, e = T0 T1 ( e) .
5 2 10 5 2

This shows that T0 does not commute with T1 on the right with respect to Fix T0 . Hence we cannot
significantly weaken the constraints in Lemma 4.1.

Let C be a nonempty convex subset of E. Recall that a set-valued mapping A : C ⇒ E is convex if, for

each x, y ∈ C and λ ∈ [0, 1], λAx + (1 − λ)Ay ⊆ A λx + (1 − λ)y . Then gra A is convex if and only if the
mapping A is convex.
Now we come to our second main result below, which also provides an answer to a question in [18, Page 3].

Theorem 4.3. Let E be a separated locally convex space and let C ⊆ E be a nonempty compact convex set.
Let S be a nonempty set, and S := {Ts : C ⇒ C with dom Ts = C | s ∈ S}. Assume that gra Ts is closed
convex and that (for all s, t ∈ S) Ts (Tt x) ⊆ Tt (Ts x), ∀x ∈ C. Then

Fix Ts ∩ Fix Tt = ∅, ∀s, t ∈ S. (25)

Moreover, if there are only at most two mappings in S that are not singled-valued mappings, then C has a
common fixed point for S.

Proof. By Corollary 2.2 and the convexity of gra Ts (for all s ∈ S),

∅ = Fix Ts is compact and convex. (26)

Since gra Ts is convex with dom Ts = C (for every s ∈ S), Ts x is nonempty convex for every x ∈ C and
Ts−1 y is nonempty convex for every y ∈ ran Ts . Then Lemma 4.1 directly shows that

Fix Ts ∩ Fix Tt = ∅, ∀s, t ∈ S. (27)

Let s, t ∈ S. By Lemma 4.1 again,

Ts x ∩ Fix Tt = ∅, ∀x ∈ Fix Tt . (28)

Now we assume that there are only at most two mappings in S that are not singled-valued. We claim
that

Fix Ts = ∅, for all nonempty finite set Λ ⊆ S. (29)
s∈Λ

We use the induction method to show the above statement. Let Λ ⊆ S contain finite elements only, and |Λ|
be denoted by the amount of the total elements in the set Λ. When |Λ| = 2, the statement for (29) holds
by (25).
Let k ≥ 2 and k ∈ N. Assume that (29) holds when |Λ| = k. We consider the case when |Λ| = k + 1. By
the assumption, there exists s0 ∈ Λ such that Ts0 is a single-valued mapping on C. Set

D := Fix Ts . (30)
s∈Λ\ {s0 }

We have Λ\ {s0 } = k, thus our assumption and (26) show that D is a nonempty compact and convex set.
214 A.T.-M. Lau, L. Yao / J. Math. Anal. Appl. 459 (2018) 203–216

By (28), we have (for every x ∈ D)

Ts0 x ∈ Fix Ts , ∀s ∈ Λ\ {s0 } (since Ts0 x is a singleton).

Hence

Ts0 x ∈ D, ∀x ∈ D. (31)

Then applying Corollary 2.2 to Ts0 and the compact convex set D, we have

Fix Ts0 ∩ D = ∅.

Therefore, according to the definition of the set D (see (30)),


Fix Ts = ∅.
s∈Λ

Thus the statement for |Λ| = k + 1 holds. Hence by the induction methods, (29) holds.
Then applying the finite intersection property, (26) shows that

Fix Ts = ∅.
s∈S

Hence C has a common fixed point for S. 2

5. Remarks and open problems

Remark 5.1. Some other extensions of the Fan–Glicksberg fixed point theorem (Lemma 2.1) can be found
in [10,11].

Remark 5.2. If let η(t) := λt for some λ > 0 in Lemma 3.1, then we obtain the classical Ekeland’s variational
principle (see also [19, Theorem 3.13], [21, Corollary 1.4.2] or [3]). Lemma 3.1 also holds in a complete metric
space.

Remark 5.3. All results in Section 3 hold in the setting of a complete metric space.

We recall the following interesting open problem posed by S. Reich in 1974 [20], which is related to
Corollary 3.4.

Open Problem 5.4. Let T : X ⇒ X be a set-valued mapping such that T x is a nonempty closed bounded
set for every x ∈ X. Let the function η : ]0, +∞[ → [0, 1[ be such that

lim sup η(t) < 1, ∀r > 0

t→r +

Assume that
   
H(T x, T y) ≤ η x − y  · x − y , ∀x, y ∈ X with x = y. (32)

Is Fix T necessarily nonempty?

 A.T.-M. Lau, L. Yao / J. Math. Anal. Appl. 459 (2018) 203–216 215

Remark 5.5. In [17], Mizoguchi and Takahashi showed that Problem 5.4 has an affirmative answer under
the constraint that lim supt→r+ η(t) < 1 for all r ≥ 0.
In Problem 5.4, let η(0) := 0 and η(t) := t − t · η(t), ∀t > 0. Then (32) is equivalent to
   
H(T x, T y) + η x − y  ≤ x − y , ∀x, y ∈ X.

Thus η(t) > 0 for all t > 0. Suppose that η satisfies the following conditions:

(i) η is increasing on R+ and η(x + y) ≤ η(x) + η(y), ∀x, y ≥ 0.

Then Corollary 3.4 shows that Problem 5.4 has an affirmative answer with the above assumptions: (i).

Open Problem 5.6. Can Theorem 4.3 be generalized to left amenable semigroups S (i.e., there is a left
translation variant positive linear functional of norm one on the space of bounded real valued functions) or
to the class of left reversible semigroups S (i.e., any two right ideals in S has non-void intersection) as in
[13,14,9].

Let C be a nonempty bounded closed convex subset of X. We say that C has the fixed point property if
Fix T = ∅ for every nonexpansive mapping T : C → C. We say that X has the weak fixed point property if
every weakly compact convex subset of X has the fixed point property. We say X has the weak fixed point
property for left reversible semigroups if whenever S is a left reversible semitopological semigroup and C is
a nonempty weakly compact convex subset of X for which the action of S on C (with the norm topology)
is separately continuous and nonexpansive, then C has a common fixed point for S [15].
In [15], Lau and Mah showed that

Theorem 5.7. (See [15, Theorem 4.2].) Let G be a separable compact group. Then the Banach algebra of G,
B(G), has the weak∗ fixed point property for left reversible semigroups.

There raises the following interesting open problem [15].

Open Problem 5.8. Let X be with the weak fixed point property. Does X have the weak fixed point property
for left reversible semigroups?

The following two open problems and the comments are kindly suggested by the referee.

Open Problem 5.9. Is there a common fixed point for a commuting family of non-convex set-valued mappings
which are convex-valued (even for two mappings), similar to Theorem 4.3?

Open Problem 5.10. Let E be a locally convex space, which is not assumed to be separated, and let C ⊆ E
be a nonempty compact convex set. Let T : C ⇒ C be such that gra T is closed and convex, and that T x
is nonempty for all x ∈ C. Is Fix T necessarily nonempty?

Note that an answer to Problem 5.10 would directly lead to the extension of Theorem 4.3 to the non-
separated case.

Acknowledgment

The authors would like to thank the referee for his or her careful reading of the paper, and valuable
suggestions. Furthermore, Problem 5.9 and Problem 5.10 are suggested by the referee.
216 A.T.-M. Lau, L. Yao / J. Math. Anal. Appl. 459 (2018) 203–216

References

[1] A. Amini-Harandi, Some generalizations of Caristi’s fixed point theorem with applications to the fixed point theory of
weakly contractive set-valued maps and the minimization problem, Nonlinear Anal. 72 (2010) 4661–4665.
[2] H.H. Bauschke, P.L. Combettes, Convex Analysis and Monotone Operator Theory in Hilbert Spaces, Springer, 2011.
[3] J.M. Borwein, J.D. Vanderwerff, Convex Functions, Cambridge University Press, 2010.
[4] J. Caristi, Fixed point theorems for mappings satisfying inwardness conditions, Trans. Amer. Math. Soc. 215 (1976)
241–251.
[5] I. Ekeland, Nonconvex minimization problems, Bull. Amer. Math. Soc. 1 (1979) 443–474.
[6] K. Fan, Fixed-point and minimax theorems in locally convex topological linear spaces, Proc. Natl. Acad. Sci. USA 38
(1952) 121–126.
[7] Y. Feng, S. Liu, Fixed point theorems for multi-valued contractive mappings and multi-valued Caristi type mappings,
J. Math. Anal. Appl. 317 (2006) 103–112.
[8] I.L. Glicksberg, A further generalization of the Kakutani fixed theorem, with application to Nash equilibrium points, Proc.
Amer. Math. Soc. 3 (1952) 170–174.
[9] R.D. Holmes, A.T.-M. Lau, Non-expansive actions of topological semigroups and fixed points, J. Lond. Math. Soc. 5 (1972)
330–336.
[10] S. Itoh, W. Takahashi, Single-valued mappings, multivalued mappings and fixed-point theorems, J. Math. Anal. Appl. 59
(1977) 514–521.
[11] S. Itoh, W. Takahashi, The common fixed point theory of singlevalued mappings and multivalued mappings, Pacific J.
Math. 79 (1978) 493–508.
[12] J.R. Jachymski, Caristi’s fixed point theorem and selections of set-valued contractions, J. Math. Anal. Appl. 227 (1998)
55–67.
[13] A.T.-M. Lau, Invariant means on almost periodic functions and fixed point properties, Rocky Mountain J. Math. 3 (1973)
69–76.
[14] A.T.-M. Lau, Invariant means on almost periodic functions and equicontinuous actions, Proc. Amer. Math. Soc. 49 (1975)
379–382.
[15] A.T.-M. Lau, P.F. Mah, Fixed point property for Banach algebras associated to locally compact groups, J. Funct. Anal.
258 (2010) 357–372.
[16] A.T.-M. Lau, L. Yao, Hahn–Banach extension and optimization related to fixed point properties and amenability, J. Non-
linear Var. Anal. 1 (2017) 127–143.
[17] N. Mizoguchi, W. Takahashi, Fixed point theorems for multivalued mappings on complete metric spaces, J. Math. Anal.
Appl. 141 (1989) 177–188.
[18] I. Mohamadi, The first common fixed point theorem for commutative set-valued mappings, arXiv:1707.06098, July 2007.
[19] R.R. Phelps, Convex Functions, Monotone Operators and Differentiability, 2nd edition, Springer-Verlag, 1993.
[20] S. Reich, Some fixed point problems, Atti Accad. Naz. Lincei 57 (1974) 194–198.
[21] C. Zălinescu, Convex Analysis in General Vector Spaces, World Scientific Publishing, 2002.