Next Article in Journal
Stability of λ-Harmonic Maps
Next Article in Special Issue
Some Simultaneous Generalizations of Well-Known Fixed Point Theorems and Their Applications to Fixed Point Theory
Previous Article in Journal
A Novel (R,S)-Norm Entropy Measure of Intuitionistic Fuzzy Sets and Its Applications in Multi-Attribute Decision-Making
Previous Article in Special Issue
Near Fixed Point Theorems in Hyperspaces
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Ekeland’s Variational Principle and Minimization Takahashi’s Theorem in Generalized Metric Spaces

1
Department of Mathematics, College of Basic Sciences, Karaj Branch, Islamic Azad University, Alborz 31485-313, Iran
2
Department of Mathematics, Iran University of Science and Technology, Tehran 16846-13114, Iran
*
Author to whom correspondence should be addressed.
Mathematics 2018, 6(6), 93; https://doi.org/10.3390/math6060093
Submission received: 28 April 2018 / Revised: 26 May 2018 / Accepted: 3 June 2018 / Published: 5 June 2018
(This article belongs to the Special Issue Fixed Point Theory)

Abstract

:
We consider a distance function on generalized metric spaces and we get a generalization of Ekeland Variational Principle (EVP). Next, we prove that EVP is equivalent to Caristi–Kirk fixed point theorem and minimization Takahashi’s theorem.

1. Introduction

After presentation of Ekeland Variational Principle (EVP) in 1972, it becomes clear that this principle is equivalent to Caristi fixed point theorem [1,2,3,4,5,6,7,8,9], Drop theorem [10,11], Flower Petal theorem [10,11] and Takahashi’s nonconvex minimization theorem. Many scholars have studied EVP on complete convex space and on locally convex space. EVP is proved for the investigation of best proximity points for Cyclic contractions, Reich type cyclic contractions, Kannan type cyclic contractions, Ciric type cyclic contractions, Hardy and Rogers type cyclic contraction, Chatterjee type cyclic, contractions, and Zamfirescu type cyclic contractions (for more details, see [12]).

2. EVP

Theorem 1. (Ekeland Theorem)
Assume that ( Z , d ) is a complete metric space (see [13]). Let f : Z R { + } be a proper, semicontinuous and bounded below function. Then, there exists y Z such that
f ( y ) f ( x ) ; d ( x , y ) 1   a n d   f ( z ) > f ( y ) ϵ d ( y , z )   f o r   a l l   y z .
Now, we study generalized metric spaces and their properties, for more details and application, we refer to [7,14,15,16,17,18,19,20,21,22,23].
Definition 1.
Assume that Z is a nonempty set and mapping
G : Z × Z × Z [ 0 , )
is satisfied in the following conditions (see [24]):
(i) 
G ( u , v , w ) = 0 if u = v = w ,
(ii) 
G ( u , u , v ) > 0 for all u , v X , where u v ,
(iii) 
G ( u , u , w ) G ( u , v , w ) for all u , v , w Z with v w ,
(iv) 
G ( u , v , w ) = G ( p { u , v , w } ) such that p is a permutation of u , v , w ,
(v) 
G ( u , v , w ) G ( u , α , α ) + G ( α , v , w ) for all u , v , w , α in Z.
Then, G is said to be G-metric and pair ( Z , G ) is said to be G-metric space.
Definition 2.
Let ( Z , G ) be a G-metric space (see [24]). A sequence { u n } in Z is said to be
(a) 
G-Cauchy sequence if, for all ϵ > 0 , there exists n 0 N such that for every m , n , l N and m , n , l n 0 then G ( u n , u m , u l ) < ϵ ,
(b) 
G-convergent to u Z if for all e p s i l o n > 0 , there exists natural number n 0 such that for all m , n n 0 , then G ( u n , u m , u ) < ϵ .
Proposition 1.
Assuming that ( Z , G ) is a G-metric space, then the following statements are equivalent (see [24]):
(a) 
{ u n } is a G-Cauchy sequence,
(b) 
for each ϵ > 0 , there exists natural number n 0 such that for all m , n n 0 , then G ( u n , u m , u m ) < ϵ .
Definition 3.
A function ξ : R + R + is sub-additive when ξ ( u + v ) ξ ( u ) + ξ ( v ) and ξ ( α u ) = α ξ ( u ) , for every α > 0 .
Definition 4.
Let Z be a nonempty set. A function
G : Z × Z × Z [ 0 , )
is said to be quasi-G-metric (shortly q-G-m) if the following conditions hold,
1. 
G ( u , v , w ) = 0 if u = v = w ,
2. 
G ( u , u , v ) > 0 for all u , v Z , u v ,
3. 
G ( u , u , w ) G ( u , v , w ) for all u , v , w Z , w v ,
4. 
G ( u , v , w ) G ( u , α , α ) + G ( α , v , w ) for all u , v , w , α X .
( Z , G ) is said to be q-G-m space when Z is a nonempty set and G is a q-G-m. The concept of Cauchy sequence, convergence and complete space are defined as G-metric space.
Throughout the paper, unless otherwise specified, Z is a nonempty set and ( Z , G ) is a q-G-m space. ψ : ( , ) ( 0 , ) is a nondecreasing function. A function h : Z ( , ) is said to be lower semicontinuous from above (shortly L s c a ) at u 0 , when, for each sequence { u n } in Z such that u n u 0 and h ( u 1 ) h ( u 2 ) h ( u n ) , we have h ( u 0 ) lim n h ( u n ) . The function h is said to be L s c a on Z, when h is L s c a at every point of Z, h is proper when h ¬ .
Definition 5.
Assume that Z is an order space with order ⪯ on Z.
(1) 
⪯ will be quasi-order on Z if it is reflexive and transitive;
(2) 
The sequence { u n } n N on Z will be decreasing if for all n N , u n + 1 u n ;
(3) 
Let P ( u ) = { v Z : v u } and if { u n } n N P ( u ) be decreasing and convergent to u ˜ Z , then u ˜ P ( u ) . Then, the quasi-order ⪯ is said to be lower closed if each u Z .
Definition 6.
Assume that ( Z , G ) is a q-G-m space with quasi-order ⪯. The set P ( u ) = { v Z : v u } u Z , would be said to be complete if every decreasing sequence in P ( u ) be convergent in it.
Definition 7.
Let ( Z , G ) be a q-G-m space. A function γ : Z × Z × Z [ 0 , ) is said to be γ-function when
(1) 
γ ( u , v , w ) γ ( u , α , α ) + γ ( α , v , w ) for all u , v , w , α Z ,
(2) 
if u Z , { v n } n N be a sequence in Z which is convergent to v in Z and γ ( u , v n , v n ) M , then γ ( u , v , v ) M ,
(3) 
for every ϵ > 0 , there exists δ > 0 such that γ ( u , α , α ) δ and γ ( α , v , w ) δ imply that G ( u , v , w ) ϵ .
We show the class of all the γ-function by Γ.
Remark 1.
Assume that ( Z , G ) is a q-G-m space and γ Γ . If ξ : R + R + is a nondecreasing and sub-additive function such that ξ ( 0 ) = 0 , then ξ γ Γ .
Example 1.
Assume that
G : Z 3 [ 0 , ) , G ( u , v , w ) = 1 2 ( w u + u v ) .
Then, G is a q-G-m but is not G-metric.
Proof. 
q-G-m is obvious. We show that G ( u , v , w ) G { p ( u , v , w ) } (p is a permutation of u , v , w ). Since
G ( 3 , 5 , 2 ) = 1 2 ( 2 3 + 3 5 ) = 3 2 , G ( 2 , 3 , 5 ) = 1 2 ( 3 2 + 5 2 ) = 2 ,
then G is not a G-metric. ☐
Example 2.
Let G ( u , v , w ) be the same as in the previous example. Then, γ = G is a γ-function.
Proof. 
(a) and (b) are obvious. Let ϵ > 0 be given, put δ = ϵ 2 if γ ( u , α , α ) = 1 2 ( w α + α v ) < ϵ 2 then
G ( u , v , w ) = 1 2 ( w v + u v ) 1 2 ( w α + α u + u α + α v ) < ϵ .
 ☐
Example 3.
Assume that ( Z , . ) is a normed space. Then, the function γ : Z × Z × Z [ 0 , ) defined by γ ( u , v , w ) = v for each u,v,w Z , is a γ-function. However, it is not a q-G-m on Z.
Example 4.
Let Z = R . Define a function G : Z × Z × Z [ 0 , ) , by
G ( u , v , w ) = 0 , i f u = v = w , | u | + 1 , o t h e r w i s e .
Then, G is q-G-m but is not a γ-function.
Lemma 1.
Assume that ( Z , G ) is a G-metric space and γ Γ . Let { x n } and { y n } be two sequences in Z, { ρ n } and { φ n } be in [ 0 , ] , which are convergent to zero (see [20]). Let x , y , z , α Z , then
(1) 
If γ ( y , x n , x n ) ρ n and γ ( x n , y , z ) φ n for all n N , then G ( y , y , z ) < ϵ and hence z = y ,
(2) 
If γ ( y n , x n , x n ) ρ n and γ ( x n , x m , z ) φ n for every m > n , then G ( y n , y m , z ) convergent to zero and hence y n z ,
(3) 
If γ ( x n , x m , x l ) ρ n for all m , n , l N with n m l , then { x n } is a G cauchy sequence.
(4) 
If γ ( x n , α , α ) ρ n for all n N then { u n } is a G-cauchy sequence.
Lemma 2.
Let γ Γ . If sequence { u n } be in Z that
lim sup n { γ ( u n , u m , u l ) , n m l } = 0 .
Then, { u n } will be a G-Cauchy sequence in Z.
Proof. 
Assume ρ n = sup { γ ( u n , u m , u l ) } , then lim n ρ n = 0 . By Lemma 1 ( 3 ) , { u n } is a G-Cauchy sequence. ☐
Lemma 3.
Let h : Z [ , ] be a function and γ Γ . Let P ( u ) be defined by
P ( u ) = { v Z ; v u , γ ( u , v , v ) ψ ( h ( u ) ) ( h ( u ) h ( v ) ) } ·
If P ( u ) be nonempty, then, for every v P ( u ) ,
P ( v ) P ( u ) a n d h ( v ) h ( u ) .
Proof. 
Let v P ( u ) . Thus, v u and γ ( u , v , v ) ψ ( h ( u ) ) ( h ( u ) h ( v ) ) . Since γ ( u , v , v ) 0 and ψ is a nondecreasing and positive function, then h ( u ) h ( v ) . If P ( v ) = , then P ( v ) P ( u ) . Therefore, w v and γ ( v , w , w ) ψ ( h ( v ) ) ( h ( v ) h ( w ) ) as above h ( v ) h ( w ) . Since Γ Γ , then
γ ( u , w , w ) γ ( u , v , v ) + γ ( v , w , w ) ψ ( h ( u ) ) ( h ( u ) h ( w ) ) .
We claim that w u . Assume that w = u so γ ( u , w , w ) = 0 . On the other hand,
γ ( u , v , v ) ψ ( h ( u ) ) ( h ( u ) h ( v ) ) ψ ( h ( u ) ) ( h ( u ) h ( w ) ) = 0 γ ( u , v , v ) = 0 .
Then, γ ( u , v , v ) = 0 . For every ϵ > 0 , we have γ ( u , w , w ) = 0 < ϵ and γ ( w , v , v ) = 0 < ϵ . Then, by definition γ -function, we have G ( w , v , v ) < ϵ , so G ( w , v , v ) = 0 and w = v , which is a contradiction; therefore, w P ( u ) and P ( v ) P ( u ) . ☐
Proposition 2.
Assume that ( Z , G ) is a complete q-G- m space, h : Z [ , ] is a proper and bounded below function and γ Γ . Let
P ( u ) = { v Z ; v u , γ ( u , v , v ) ψ ( h ( u ) ) ( h ( u ) h ( v ) ) } ·
Let { u n } be a sequence in Z such that S ( u n ) is nonempty and for all n N , u n + 1 S ( u n ) . Then, there exists u 0 Z such that u n u 0 and u 0 n = 1 P ( u n ) . In addition, if, for every n N , h ( u n + 1 ) inf w S ( u n ) h ( w ) + 1 n , then n = 1 S ( u n ) is a singleton set.
Proof. 
At first, we prove that { u n } is a Cauchy sequence by Lemma 3, h ( u n ) h ( u n + 1 ) for all n N . Therefore, { h ( u n ) } is non-increasing. On the other hand, if h is bounded below, then lim n h ( u n ) = r , and h ( u n ) r , for all n N . We claim that
lim sup n { γ ( u n , u m , u m ) : m > n } = 0 ·
We have
γ ( u n , u m , u m ) γ ( u n , u n + 1 , u n + 1 ) + γ ( u n + 1 , u m , u m ) γ ( u n , u n + 1 , u n + 1 ) + γ ( u n + 1 , u n + 2 , u n + 2 ) + + γ ( u m 1 , u m , u m ) .
Thus,
γ ( u n , u m , u m ) j = n m 1 γ ( u n , u m , u m ) ψ ( h ( u ) ) ( h ( u n ) r )
for all m , n N with m > n .
Put ρ n = ψ ( h ( u ) ) ( h ( u n ) r ) ; then, sup { γ ( u n , u m , u m ) : m > n } ρ n , for all n N . Since lim n h ( u n ) = r , we have that
lim sup n { γ ( u n , u m , u m ) : m > n } = 0
and lim n ρ n = 0 . By Lemma 2, { u n } is a G-Cauchy sequence. Then, there exists u 0 Z such that u n u 0 . Now, we show that u 0 n = 1 P ( u n ) . Since h is L s c a , then h ( u 0 ) lim n h ( u n ) = r h ( u k ) .
Letting n N , we have
γ ( u n , u m , u m ) j = n m 1 γ ( u j , u j + 1 , u j + 1 ) ψ ( h ( u n ) ) ( h ( u n ) h ( u 0 ) )
for all m N and m > n . Since γ ( u , v , · ) and γ ( u , · , v ) : Z ( 0 , ) is semicontinuous from below, then
γ ( u n , u 0 , u 0 ) ψ ( h ( u n ) ( h ( u n ) h ( u 0 ) )
for all n N . In addition, u 0 u for all n N . Otherwise, there exists j N such that u 0 = u j . Since
γ ( u j , u j + 1 , u j + 1 ) ψ ( h ( u j ) ) ( h ( u j ) h ( u j + 1 ) ) ψ ( h ( u j ) ) ( h ( u j ) h ( u 0 ) ) = 0 ,
then we have γ ( u j , u j + 1 , u j + 1 ) = 0 and in the same way
γ ( u j + 1 , u j + 2 , u j + 2 ) = 0 .
Now assume that ϵ > 0 , γ ( u j , u j + 1 , u j + 1 ) = 0 < δ and γ ( u j + 1 , u j + 2 , u j + 2 ) = 0 < δ . Therefore, by Definition 7 ( 3 ) , we get G ( u j , u j + 2 , u j + 2 ) < ϵ . Then, u j = u j + 2 that is a contradiction since u j u j + 2 . Since u j + 1 P ( u j ) , then P ( u j + 1 ) P ( u j ) and u j + 2 P ( u j + 1 ) . Thus, u j + 2 P ( u j ) . Suppose that u j + 2 u j for all n N . We have u 0 n = 1 P ( u n ) , then n = 1 P ( u n ) . Let h ( u n + 1 ) inf w P ( u n ) h ( w ) + 1 n , for all u 0 u n . We show that n = 1 P ( u n ) = { u 0 } . Assume that z n = 1 S ( u n ) , then
γ ( u n , z , z ) ψ ( h ( u n ) ) ( h ( u n ) h ( z ) ) ψ ( h ( u 1 ) ) ( h ( u n ) inf w P ( u n ) h ( w ) ) ψ ( h ( u 1 ) ) ( h ( u n ) h ( u n + 1 ) + 1 n ) .
Let
ψ n = ψ ( h ( u 1 ) ) ( h ( u n ) h ( u n + 1 ) + 1 n )
for all n N , then lim n ψ n = 0 , so lim n γ ( u n , z , z ) = 0 . On the other hand, { u m } is a G-Cauchy sequence. Then, lim n γ ( u m , u m , u n ) = 0 and u n , by uniqueness z = u 0 . Then, n = 1 P ( u n ) = { u 0 } . ☐
Theorem 2. (Generalized Ekeland’s variational principle)
Assume that ( Z , G ) is a complete q-G-m space and h : Z ( , ] is a proper, bounded below and L s c a function. Suppose that γ is a γ-function on Z × Z × Z , then there exists u Z such that
γ ( y , u , u ) > ψ ( h ( u ) ) ( h ( u ) h ( y ) )
for all u Z with y u .
Proof. 
Suppose it is not true. Then, for every u Z , there exists v Z , v u such that γ ( u , v , v ) ψ ( h ( u ) ) ( h ( u ) h ( v ) ) . Then P ( u ) . We define the sequence { u n } as follows. Put u 1 = α , we choose u 2 S ( u 1 ) such that h ( u 2 ) inf u S ( u 1 ) h ( u ) + 1 . In the same way suppose that u n Z is given. We choose u n + 1 S ( u n ) such that h ( u n + 1 ) inf u P ( u n ) h ( u ) + 1 n · . By Proposition 2, there exists u 0 Z such that
n = 1 P ( u n ) = { u 0 } .
By Lemma 3, we have P ( u 0 ) n = 1 P ( u n ) = { u 0 } and then P ( u 0 ) = { u 0 } . This is a contradiction. Therefore, there exists y Z such that
γ ( y , u , u ) > ψ ( h ( y ) ) ( h ( y ) h ( u ) ) .
Now, we present two generalizations of Ekeland-type variational principles in the q-G-m spaces and complete q-G-m spaces. ☐
Theorem 3.
Assume that ( Z , G ) is a q-G-m space, γ : Z × Z × Z R + is a γ-function on Z. Suppose that ψ : ( , ) ( 0 , ) is a nondecreasing function and the function h : Z R { + } is proper and bounded below. We define the quasi-order ⪯ as follows:
v u i f f u = v o r γ ( u , v , v ) ψ ( h ( u ) ) ( h ( u ) h ( v ) ) .
Suppose that there exists u ^ Z such that inf u Z h ( u ) < h ( u ^ ) and P ( u ^ ) = { v Z : v u ^ } , is complete. Then, there exists u ¯ Z such that
(a) 
γ ( u ^ , u ¯ , u ¯ ) ψ ( h ( u ^ ) ) ( h ( u ^ ) h ( u ¯ ) ) ,
(b) 
γ ( u ¯ , u , u ) > ψ ( h ( u ¯ ) ) ( h ( u ¯ ) h ( u ) ) .
Proof. 
The reflexive property is obvious. We show that ⪯ is transitive. Let w v and v u , then
w = v o r γ ( v , w , w ) ψ ( h ( v ) ) ( h ( v ) h ( w ) )
and
v = u o r γ ( u , v , v ) ψ ( h ( u ) ) ( h ( u ) h ( v ) ) .
If w = v or u = v , the transitive is established. Let u v w , since γ ( u , v , w ) 0 and ψ ( u ) > 0 ; then, by Equations (2) and (3), we have h ( v ) h ( w ) and h ( u ) h ( v ) . Thus, h ( w ) h ( u ) . Since ψ is nondecreasing, then ψ ( h ( v ) ) ψ ( h ( u ) ) by Definition 7 ( 1 ) , (2) and (3), we get
γ ( u , w , w ) γ ( u , v , v ) + γ ( v , w , w ) ψ ( h ( u ) ) ( h ( u ) h ( v ) ) + ψ ( h ( v ) ) ( h ( v ) h ( w ) ) ψ ( h ( u ) ) ( h ( u ) h ( v ) ) + ψ ( h ( u ) ) ( h ( v ) h ( w ) ) = ψ ( h ( u ) ) ( h ( u ) h ( w ) ) .
Thus, w u , which means ⪯ is quasi-order on Z. We define the sequence { u n } in P ( u ^ ) as follows. Let
P ( u n ) = { v P ( u ^ ) : v = u n o r γ ( u n , v , v ) ψ ( h ( u n ) ) ( h ( u n ) h ( v ) ) } = { v P ( u ^ ) : v u n } .
Putting u ^ = u 0 , we choose u 2 p ( u 1 ) such that h ( u 2 ) inf u P ( u 1 ) h ( u ) + 1 2 . Let u n 1 be specified. Then, we choose u n P ( u n 1 ) such that
h ( u n ) inf u P ( u n 1 ) h ( u ) + 1 n .
Then, we have that u n u n 1 and { u n } is decreasing. Moreover,
γ ( u n 1 , u n , u n ) ψ ( h ( u n 1 ) ) ( h ( u n 1 ) h ( u n ) ) ,
and then, h ( u n ) h ( u n 1 ) for all n N . This means that { h ( u n ) } is decreasing. On the other hand, h is bounded below, Thus, { h ( u n ) } is convergent. Letting lim n h ( u n ) = r , we prove that { u n } is a Cauchy sequence in P ( u ^ ) . Assuming that n < m , then
γ ( u n , u m , u m ) γ ( u n , u n + 1 , u n + 1 ) + γ ( u n + 1 , u m , u m ) γ ( u n , u n + 1 , u n + 1 ) + γ ( u n + 1 , u n + 2 , u n + 2 ) + + γ ( u m 1 , u m , u m ) ψ ( h ( u n ) ) ( h ( u n ) h ( u n + 1 ) ) + ψ ( h ( u n + 1 ) ) ( h ( u n + 1 ) h ( u n + 2 ) ) + + ψ ( h ( u m 1 ) ) ( h ( u m 1 ) h ( u m ) ) ψ ( h ( u n ) ) ( h ( u n ) h ( u n + 1 ) ) + ψ ( h ( u n ) ) ( h ( u n + 1 ) h ( u n + 2 ) ) + + ψ ( h ( u n ) ) ( h ( u m 1 ) h ( u m ) ) ψ ( h ( u n ) ) ( h ( u n ) h ( u m ) ) ψ ( h ( u n ) ) ( h ( u n ) r ) .
Put ρ n = ψ ( h ( u n ) ) ( h ( u n ) r ) , so lim n ρ n = 0 and by Lemma 1 ( 3 ) , { u n } is a Cauchy sequence in P ( u ^ ) . Since P ( u ^ ) , is complete, then { u n } is convergent to u ¯ P ( u ^ ) . Since ⪯ is transitive, P ( u n ) P ( u n 1 ) for all n N . Therefore, ( a ) holds. We show that { u ¯ } = P ( u ¯ ) . On the other hand, assuming that u p ( u ¯ ) and u u ¯ , we have:
γ ( u ¯ , u , u ) ψ ( h ( u ¯ ) ) ( h ( u ¯ ) h ( u ) ) .
Since γ and ψ are nonnegative, then h ( u ) h ( u ¯ ) . u ¯ P ( u n 1 ) , implies that u u ¯ and u ¯ u n 1 . Then, u u n 1 (transitive ⪯) for all n N . In addition, we have
h ( u ¯ ) h ( u n ) h ( u n ) + 1 n .
On the other hand lim n h ( u n ) = r , thus h ( u ¯ ) r h ( u ) h ( u ¯ ) . Therefore, h ( u ¯ ) = h ( u ) = r . Since u u n for all n N , we have
γ ( u n , u , u ) ψ ( h ( u n ) ) ( h ( u n ) h ( u ) ) = ψ ( h ( u n ) ) ( h ( u n ) r ) = ρ n .
In addition, we have
γ ( u n , u ¯ , u ¯ ) ψ ( h ( u n ) ) ( h ( u n ) h ( u ¯ ) ) = ψ ( h ( u n ) ) ( h ( u n ) r ) = ρ n .
Thus, lim n ρ n = 0 . By Equations (5) and (6) and Lemma (1), ( 1 ) we have u = u ¯ . Thus, { u ¯ } = P { u ¯ } . Hence,
γ ( u ¯ , u , u ) > ψ ( h ( u ¯ ) ) ( h ( u ¯ ) h ( u ) ·
for all u Z , u u ¯ . ☐
Theorem 4.
Let ( Z , G ) be a complete q-G-m space and
γ : Z × Z × Z R +
be a γ-function on Z, ψ : ( , ] ( 0 , ) be a nondecreasing function and function h : Z R { } be L s c a , proper and bounded below. Suppose that there is u ^ Z such that inf u Z h ( u ) < h ( u ^ ) . Then, there is u ¯ Z such that
(a) 
γ ( u ^ , u ¯ , u ¯ ) ψ ( h ( u ^ ) ) ( h ( u ^ ) h ( u ¯ ) ) ,
(b) 
γ ( u ¯ , u , u ) > ψ ( h ( u ¯ ) ) ( h ( u ¯ ) h ( u ) ) for all u Z , u u ¯ .
Proof. 
We define order ⪯ as follows:
v u i f f u = v o r γ ( u , v , v ) ψ ( h ( u ) ) ( h ( u ) h ( v ) ) .
We prove that ⪯ is quasi-order. We show that ⪯ is lower closed. Assume that { u n } is a sequence in Z, such that { u n } convergent to u Z and u n + 1 u n , thus
γ ( u n , u n + 1 , u n + 1 ) ψ ( h ( u n ) ) ( h ( u n ) h ( u n + 1 ) ) .
Since γ 0 and ψ 0 , then h ( u n + 1 ) h ( u n ) . Therefore, { h ( u n ) } is a nondecreasing sequence. Since h is bounded below, then there is lim n h ( u n ) . Let lim n h ( u n ) = r for all n N . In addition, if h be L s c a , then h ( u ) lim n h ( u n ) and we have h ( u ) r h ( u n ) . Let n N be fixed. Using Theorem 3, we have
γ ( u n , u m , u m ) ψ ( h ( u n ) ) ( h ( u n ) h ( u m ) ) ψ ( h ( u n ) ) ( h ( u n ) h ( u ) ) .
Therefore, h ( u ) h ( u n ) for all n N . Put M = ψ ( h ( u n ) ) ( h ( u n ) h ( u ) ) . By Definition 7, ( 2 ) , we have γ ( u n , u m , u m ) M for all n N . Then,
γ ( u n , u , u ) M = ψ ( h ( u n ) ) ( h ( u n ) h ( u ) ) · f o r a l l n N .
Thus, u u n , which means ⪯ is lower closed and P ( w ) = { v Z : v w } for all w Z , is lower closed. Let
P ( u n ) = { v Z : v = u n o r γ ( u n , v , v ) ψ ( h ( u n ) ) ( h ( u n ) h ( v ) ) } = { v Z : v u } .
Then, P ( u n ) is lower closed for every n N . Therefore, P ( u n ) is complete space. The rest of proof obtained of Theorem 3. ☐
Remark 2.
Theorem 3 and Theorem 4 are generalizations of Theorem 2.1 in [5], Theorem 1.1 in [25], Theorem 1 in [13], Theorem 3 in [26], Theorem 2.1 in [27] and Theorem 3 in [28].
Corollary 1.
Let ψ, Z, γ and h be the same as Theorem 4. Let ξ : R + R + be non-decreasing and sub-additive such that ξ ( 0 ) = 0 . Assume that there is u ^ Z such that inf u Z h ( u ) < h ( u ^ ) , then there exists u ¯ Z such that
(a) 
ξ ( γ ( u ^ , u ¯ , u ¯ ) ) ψ ( h ( u ^ ) ) ( h ( u ^ ) h ( u ¯ ) ) ,
(b) 
ξ ( γ ( u ¯ , u , u ) ) > ψ ( h ( u ¯ ) ) ( h ( u ¯ ) h ( u ) ) , u u ¯ .
Proof. 
By Remark 1, we conclude that η γ Γ . The rest of proof is obtained from Theorem 4. ☐

3. EVP Results

Theorem 5.
Let ( Z , G ) be a complete q-G-m space and γ Γ . Let ψ : ( , ] ( 0 , ) be nondecreasing and the function h be a proper, bounded below and L s c a , then the following statements are equivalent to Theorem 4.
(a) (Caristi–Kirk type fixed point theorem) Assume that T : Z 2 Z is a multi-valued map with nonempty value. If
γ ( u , v , v ) ψ ( h ( u ) ) ( h ( u ) h ( v ) )
is satisfied for all v T ( u ) , and T has a fixed point, i.e., there is u ¯ Z such that { u ¯ } = T ( u ¯ ) . If
γ ( u , v , v ) ψ ( h ( u ) ) ( h ( u ) h ( v ) )
for some v T ( u ) , then there exists u ¯ Z , such that u ¯ T ( u ¯ ) .
(b) (Takahashi’s minimization theorem) Let inf w Z h ( w ) < h ( u ^ ) for every u ^ Z , and let there be u Z such that
γ ( u ^ , u , u ) ψ ( h ( u ^ ) ) ( h ( u ^ ) h ( u ) ) ; u u ^ .
Then, there exists u ¯ Z such that h ( u ¯ ) = inf v Z h ( v ) .
(c) (Equilibrium version of Ekeland-type variational principle) Assume that H : Z × Z R { } is a function with the following properties:
(E1)
H ( u , w ) H ( u , v ) + H ( v , w ) for every u , v , w Z ,
(E2)
H ( u , · ) : Z R { } , for every u Z is proper and L s c a ,
(E3)
There exists u Z that inf u Z H ( u ^ , u ) > .
Then, there is u Z such that
(i) 
ψ ( h ( u ^ ) ) H ( u ^ , u ¯ ) + γ ( u ^ , u ¯ , u ¯ ) 0 ,
(ii) 
ψ ( h ( u ¯ ) ) H ( u ¯ , u ) + γ ( u ¯ , u , u ) > 0 for each u Z and u u ¯ .
Proof. 
We show that Theorem 4 implies (a). ☐
By Theorem 4, there is u ¯ Z such that u u ¯
γ ( u ¯ , u , u ) > ψ ( h ( u ¯ ) ) ( h ( u ¯ ) h ( u ) ) f o r a l l u Z , u u ¯ .
We show that { u ¯ } = T ( u ¯ ) . Suppose that it is not true. By Equation (9), we have γ ( u ¯ , v , v ) ψ ( h ( u ¯ ) ) ( h ( u ¯ ) h ( u ) ) and γ ( u ¯ , v , v ) > ψ ( h ( u ¯ ) ) ( h ( u ¯ ) h ( u ) ) · , for every v ψ ( u ¯ ) with y u ¯ . This is a contradiction. Thus, { u ¯ } = T ( u ¯ ) .
( a ) ( b ) :
Let T : Z 2 Z and
T ( u ) = { v Z ; γ ( u , v , v ) ψ ( h ( u ) ) ( h ( u ) h ( v ) ) }
for all u Z and for every T in Equation (9).
By ( a ) , there is u ¯ Z such that { u ¯ } = T ( u ¯ ) . On the other hand, according to assumptions, for every u ^ Z with inf w Z h ( w ) < h ( u ^ ) , there is u Z such that γ ( u ^ , u , u ) ψ ( h ( u ^ ) ) ( h ( u ^ ) h ( u ) ) , u u ^ . Therefore, u T ( u ^ ) and T ( u ^ ) { u ^ } , thus f ( u ¯ ) = inf u Z h ( u ) .
( b ) ( c ) :
Suppose h : Z R { } is defined by h ( u ) = H ( u ^ , u ) , u ^ is the same as in ( E 3 ) . By ( E 3 ) , we have inf u Z h ( u ) > , and then h is bounded below. Let ( i i ) not be true. Then, for each u Z , v Z exists such that
ψ ( h ( u ) ) H ( u , v ) + γ ( u , v , v ) 0 .
By ( E 1 ) , we have H ( u ^ , v ) H ( u ^ , u ) + H ( u , v ) , then H ( u ^ , v ) H ( u ^ , u ) H ( u , v ) . Therefore, by Equation (13), we have
ψ ( h ( u ) ) ( H ( u ^ , v ) H ( u ^ , u ) ) + γ ( u , v , v ) ψ ( h ( u ) ) H ( u , v ) + γ ( u , v , v ) 0 .
Thus, for all u Z , there is v Z , such that
ψ ( h ( u ) ) ( h ( v ) h ( u ) ) + γ ( u , v , v ) 0 · , v u .
On the other hand, γ ( u , v , v ) ψ ( h ( u ) ) ( h ( v ) h ( u ) ) . Now, by ( b ) ,
h ( u ¯ ) = inf v Z h ( v ) h ( w ) ·
Replace u by u ¯ , then
ψ ( h ( u ¯ ) ) ( H ( u ^ , v ) H ( u ^ , v ) ) + γ ( u ¯ , v , v ) 0 , v u .
Thus,
ψ ( h ( u ¯ ) ) ( h ( v ) h ( u ¯ ) ) + γ ( u ¯ , v , v ) 0 o r γ ( u ¯ , v , v ) ψ ( h ( u ¯ ) ) ( h ( u ¯ ) h ( v ) ) .
Since v u , by Lemma 1, ( 1 ) , we get γ ( u ¯ , u ¯ , u ¯ ) 0 and γ ( u ¯ , u , u ) 0 . Thus, we have γ ( u ¯ , v , v ) > 0 and by Equation (15),
0 < ψ ( h ( u ¯ ) ) ( h ( u ¯ ) h ( v ) ) h ( v ) < h ( u ¯ ) ,
which is a contradiction.
( c ) Theorem 4:
Assume that H : Z × Z R { } is a function such that H ( u , v ) = h ( v ) h ( u ) , u , v Z . Then, by hypothesis, H is satisfied in condition ( c ) . By ( i ) , we have
ψ ( h ( u ^ ) ) H ( u ^ , u ¯ ) + γ ( u ^ , u ¯ , u ¯ ) 0 .
Then,
ψ ( h ( u ^ ) ) ( h ( u ¯ ) h ( u ^ ) ) + γ ( u ^ , u ¯ , u ¯ ) 0
and so
γ ( u ^ , u ¯ , u ¯ ) ψ ( h ( u ^ ) ) ( h ( u ^ ) h ( u ¯ ) ) .
In addition, by ( i i ) , we have
ψ ( h ( u ¯ ) ) H ( u ¯ , u ) + γ ( u ¯ , u , u ) > 0
for each u Z and u u ¯ . Then,
ψ ( h ( u ¯ ) ) ( h ( u ) h ( u ¯ ) ) + γ ( u ¯ , u , u ) > 0 .
Corollary 2.
(Equilibrium version of Ekeland-type variational principle.) Assume that ( Z , G ) is a complete q-G-m space and γ Γ . Let H : Z × Z R be a function such that:
(E1)
H ( u , w ) H ( u , v ) + H ( v , w ) for all u , v , w Z ,
(E2)
H ( u , · ) : Z R , for every u Z be a L s c a and bounded below, then for every ϵ > 0 and for each u ^ Z , there is u ¯ Z such that
(i) 
H ( u ^ , u ¯ ) + γ ( u ^ , u ¯ , u ¯ ) 0 ,
(ii) 
H ( u ¯ , u ) + ϵ Γ ( u ¯ , u , u ) > 0 for every u Z , u u ¯ .
Proof. 
Put h : Z R { + } , that h ( u ^ ) = H ( u , u ^ ) and u ^ Z . Then, function h is proper, L s c a and bounded by the below. For every ϵ > 0 and u ^ Z , we define ψ ( h ( u ^ ) ) = 1 ϵ . Then, by Theorem 5, we get ( c ) , ( i ) and ( i i ) . ☐
Corollary 3. (Nonconvex equilibrium theorem).
Assume that the function H : Z × Z ( , ) is a proper and lsca and bounded below in the first argument, and ψ : ( , ) ( 0 , ) is a nondecreasing function. Let, for each u Z with { x Z : H ( u , x ) < 0 } , there exist v = v ( u ) Z with v u such that
γ ( u , v , v ) ψ ( H ( u , t ) ) ( H ( u , t ) H ( v , t ) )
holds for all z Z . Then, there exists y Z such that H ( y , v ) 0 for all v Z .
Proof. 
By Theorem 2 for each w Z , there exists y ( w ) Z such that γ ( y ( w ) , u , u ) > ψ ( H ( y ( w ) , w ) ) ( H ( y ( w ) , w ) H ( u , w ) ) for each u Z with u y ( w ) . We show that there exists y Z such that H ( y , v ) 0 for all v Z . On the contrary, for each u Z , there exists v Z such that H ( u , v ) < 0 . Then, for each u Z , { x Z : H ( u , x ) < 0 } . According to the assumption, there exists v = v ( y ( w ) ) , v y ( w ) such that γ ( y ( w ) , v , v ) ψ ( H ( y ( w ) , w ) ) ( H ( y ( w ) , w ) H ( v , w ) ) , which is a contradiction. ☐
Example 5.
Assume that Z = [ 0 , 1 ] and G ( u , v , w ) = max { | u v | , | u w | , | v w | } . Then, ( Z , G ) is a complete q-G-m space. Let a,b be positive real numbers with a b . Assume that H : Z × Z R with H ( u , v ) = a u b v . Therefore, function x H ( u , v ) is proper, lower semicontinuous and bounded below, and H ( 1 , v ) 0 for every v Z . In addition, H ( u , v ) 0 for every u [ b a , 1 ] and for each v Z . In fact, for each u [ 0 , b a ] , H ( u , v ) = a u b v < 0 when v [ a b u , 1 ] . Then, set { x Z : H ( u , x ) < 0 } for each u [ 0 , b a ] . Supposing u , v Z , u v , we have u v = 1 a { ( a u b x ) ( a v b x ) } , for each x z . Let ψ : [ 0 , ) [ 0 , ) with ψ ( t ) = 1 a be define. Therefore, G ( u , v , v ) ψ ( H ( u , x ) ) ( H ( u , x ) H ( v , x ) ) , for each u v , and u , v , x Z . By Corollary 3, there exists y Z such that H ( y , v ) 0 for every v Z .
Definition 8.
Let M be a nonempty subset of metric space (see [29]). Assume that H : M × M R is a real valued function and γ Γ . Let ϵ > 0 , if there exists u ¯ Z , such that
H ( u ¯ , v ) + ϵ γ ( u ¯ , v , v ) 0
for every v M , then u ¯ is a σ-solution of E P if the inequality in Equation (17) is strict for all u v . Then, u ¯ is said to be strictly a σ-solution of E P .
Theorem 6.
Assume that M is a nonempty compact of complete metric space Z and γ Γ Assume that H : M × M R is a real valued function such that:
(E1)
H ( u , w ) H ( u , v ) + H ( v , w ) for each u , v , w M ,
(E2)
For all u M , H ( u , · ) : M R is an L s c a and bounded below function,
(E3)
For all v M , H ( · , v ) : M R is a lower semicontinuous function.
Then, u ¯ M is a solution of E P .
Proof. 
By using Corollary 2, for all n N , there is u n M such that
H ( u n , v ) + 1 n γ ( u n , v , v ) 0
for every v M .
On the other hand, for n N , u n M is a σ -solution of E P for σ = 1 n . M is compact. Therefore, there is subsequence { u n k } of { u n } such that u n k u ¯ . Since H ( · , v ) is upper semicontinuous, then
F ( u ¯ , v ) lim sup n ( H ( u n , v ) + 1 n k γ ( u n k , v , v ) ) 0
u ¯ is a solution of E P . ☐
Definition 9.
Assume that ( Z , G ) is a complete q-G-m space and γ is a γ-function on Z. We said that u 0 Z is satisfied in condition Ξ if and only if every sequence { u n } in Z, H ( u 0 , u n ) 1 n for each n N and H ( u n , u ) + 1 n γ ( u n , u , u ) 0 for all u Z and n N has a convergent subsequence.
Theorem 7.
Assume that ( Z , G ) be a complete q-G-m space. Suppose γ Γ and H : Z × Z R satisfied in conditions ( E 1 ) and ( E 2 ) of Corollary 2. In addition, H will be upper semicontinuous in first argument, if u 0 Z be satisfied in condition Ξ. Then, there is a solution u ¯ Z of E P .
Proof. 
Put σ = 1 n in Corollary 2, then for each n N and for every u 0 X , there is u n Z such that
H ( u 0 , u n ) + 1 n γ ( u 0 , u n , u n ) 0
and
H ( u n , u ) + 1 n γ ( u n , u , u ) > 0 ,
for all u Z . Since γ ( u 0 , u n , u n ) 0 , by Equation (18), we have, H ( u 0 , u n ) 0 for all n N . ☐
By using condition Ξ , there is subsequence { u n } that is convergent to point u ¯ Z . On the other hand, by using semicontinuous h ( · , u ) and Equation (19), we find that u ¯ is a solution of E P .

Author Contributions

Supervision, R.S.; Writing–Review and Editing, E.H.

Acknowledgments

We would like to thank the three anonymous referees for their comments and suggestions on the manuscript.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Aubin, J.-P.; Siegel, J. Fixed points and stationary points of dissipative multivalued maps. Proc. Am. Math. Soc. 1980, 78, 391–398. [Google Scholar] [CrossRef]
  2. Bae, J.S. Fixed point theorems for weakly contractive multivalued maps. J. Math. Anal. Appl. 2003, 284, 690–697. [Google Scholar] [CrossRef]
  3. Bae, J.S.; Cho, E.W.; Yeom, S.H. A generalization of the Caristi–Kirk fixed point theorem and its application to mapping theorems. J. Korean Math. Soc. 1994, 31, 29–48. [Google Scholar]
  4. Caristi, J. Fixed point theorems for mappings satisfying inwardness conditions. Trans. Am. Math. Soc. 1976, 215, 241–251. [Google Scholar] [CrossRef]
  5. Chen, Y.; Cho, Y.J.; Yang, L. Note on the results with lower semi-continuity. Bull. Korean Math. Soc. 2002, 39, 535–541. [Google Scholar] [CrossRef]
  6. Daffer, P.Z.; Kaneko, H.; Li, W. Variational principle and fixed points. In Set Valued Mappings with Applications in Nonlinear Analysis; Agarwal, R.P., O’Regan, D., Eds.; Taylor and Francis: London, UK, 2002. [Google Scholar]
  7. Demma, M.; Saadati, R.; Vetro, P. Fixed point results on b-metric space via Picard sequences and b-simulation functions. Iran. J. Math. Sci. Inform. 2016, 11, 123–136. [Google Scholar]
  8. Gajek, L.; Zagrodny, D. Geometric variational principle. Dissertationes Math. 1995, 340, 55–71. [Google Scholar]
  9. Pluptieng, S.; Seangwattana, T. The Borwein-preiss variational principle for nonconvex countable systems of equilibrium problems. J. Nonlinear Sci. Appl. 2016, 9, 2224–2232. [Google Scholar] [CrossRef]
  10. Hyers, D.H.; Isac, G.; Rassias, T.M. Topics in Nonlinear Analysis and Applications; World Scientific: Singapore, 1997. [Google Scholar]
  11. Penot, J.-P. The drop theorem, the petal theorem and Ekeland’s variational principle. Nonlinear Anal. 1986, 10, 813–822. [Google Scholar] [CrossRef]
  12. Ivanov, M.; Zlatanov, B.; Zlateva, N. A variational principle and best proximity points. Acta Math. Sin. Engl. Ser. 2015, 31, 1315–1326. [Google Scholar] [CrossRef]
  13. Ekeland, I. On convex minimization problems. Bull. Am. Math. Soc. 1979, 1, 445–474. [Google Scholar] [CrossRef]
  14. Azhini, M.; Kenari, H.M.; Saadati, R. Strong ergodic theorem for commutative semigroup of non-Lipschitzian mappings in multi-Banach space. Proc. Indian Acad. Sci. Math. Sci. 2017, 127, 657–672. [Google Scholar] [CrossRef]
  15. Banaei, S.; Ghaemi, M.B.; Saadati, R. An extension of Darbo’s theorem and its application to system of neutral differential equations with deviating argument. Miskolc Math. Notes 2017, 18, 83–94. [Google Scholar] [CrossRef]
  16. Demma, M.; Saadati, R.; Vetro, P. Multi-valued operators with respect wt-distance on metric type spaces. Bull. Iran. Math. Soc. 2016, 42, 1571–1582. [Google Scholar]
  17. Eslamian, M.; Saadati, R.; Vahidi, J. Viscosity iterative process for demicontractive mappings and multivalued mappings and equilibrium problems. Comput. Appl. Math. 2017, 36, 1239–1253. [Google Scholar] [CrossRef]
  18. Naeem, R.; Anwar, M. Jessen type functionals and exponential convexity. J. Math. Comput. Sci. 2017, 17, 429–436. [Google Scholar] [CrossRef] [Green Version]
  19. Piri, H.; Rahrovi, S.; Kumam, P. Generalization of Khan fixed point theorem. J. Math. Comput. Sci. 2017, 17, 76–83. [Google Scholar] [CrossRef] [Green Version]
  20. Saadati, R.; Vaezpour, S.M.; Vetro, P.; Rhoades, B.E. Fixed point theorems in generalized partially ordered G-metric spaces. Math. Comput. Model. 2010, 52, 797–801. [Google Scholar] [CrossRef]
  21. Shoaib, A.; Azam, A.; Arshad, M.; Ameer, E. Fixed point results for multivalued mappings on a sequence in a closed ball with applications. J. Math. Comput. Sci. 2017, 17, 308–316. [Google Scholar] [CrossRef] [Green Version]
  22. Singh, D.; Chauhan, V.; Kumam, P.; Joshi, V.; Thounthong, P. Applications of fixed point results for cyclic Boyd-Wong type generalized F-Psi-contractions to dynamic programming. J. Math. Comput. Sci. 2017, 17, 200–215. [Google Scholar] [CrossRef]
  23. Suzuki, T. On Downing-Kirk’s theorem. J. Math. Anal. Appl. 2003, 286, 453–458. [Google Scholar] [CrossRef]
  24. Mustafa, Z.; Sims, B. A new approach to generalized metric spaces. J. Nonlinear Convex Anal. 2006, 7, 289–297. [Google Scholar]
  25. Ekeland, I. On the variational principle. J. Math. Anal. Appl. 1974, 47, 324–353. [Google Scholar] [CrossRef]
  26. Kada, O.; Suzuki, T.; Takahashi, W. Nonconvex minimization theorems and fixed point theorems in complete metric spaces. Math. Jpn. 1996, 44, 381–391. [Google Scholar]
  27. Lin, L.-J.; Du, W.-S. Ekeland’s variational principle, minimax theorems and existence of nonconvex equilibria in complete metric spaces. J. Math. Anal. Appl. 2006, 323, 360–370. [Google Scholar] [CrossRef]
  28. Park, S. On generalizations of the Ekeland-type variational principles. Nonlinear Anal. 2000, 39, 881–889. [Google Scholar] [CrossRef]
  29. Al-Homidan, S.; Ansari, Q.H.; Yao, J.-C. Some generalizations of Ekeland-type variational principle with applications to equilibrium problems and fixed point theory. Nonlinear Anal. 2008, 69, 126–139. [Google Scholar] [CrossRef]

Share and Cite

MDPI and ACS Style

Hashemi, E.; Saadati, R. Ekeland’s Variational Principle and Minimization Takahashi’s Theorem in Generalized Metric Spaces. Mathematics 2018, 6, 93. https://doi.org/10.3390/math6060093

AMA Style

Hashemi E, Saadati R. Ekeland’s Variational Principle and Minimization Takahashi’s Theorem in Generalized Metric Spaces. Mathematics. 2018; 6(6):93. https://doi.org/10.3390/math6060093

Chicago/Turabian Style

Hashemi, Eshagh, and Reza Saadati. 2018. "Ekeland’s Variational Principle and Minimization Takahashi’s Theorem in Generalized Metric Spaces" Mathematics 6, no. 6: 93. https://doi.org/10.3390/math6060093

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop