Research Article Optimality Conditions for Nondifferentiable Multiobjective …downloads.hindawi.com/journals/aaa/2016/5367190.pdf · 2019-07-30 · Research Article Optimality Conditions
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Research ArticleOptimality Conditions for Nondifferentiable MultiobjectiveSemi-Infinite Programming Problems
D Barilla G Caristi and A Puglisi
Department of Economics University of Messina Via dei Verdi 75 Messina Italy
Correspondence should be addressed to G Caristi gcaristiunimeit
Received 2 May 2016 Accepted 18 August 2016
Academic Editor Jozef Banas
Copyright copy 2016 D Barilla et alThis is an open access article distributed under theCreative CommonsAttribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
We have considered a multiobjective semi-infinite programming problem with a feasible set defined by inequality constraintsFirst we studied a Fritz-John type necessary condition Then we introduced two constraint qualifications and derive the weak andstrong Karush-Kuhn-Tucker (KKT in brief) types necessary conditions for an efficient solution of the considered problem Finallyan extension of a Caristi-Ferrara-Stefanescu result for the (Φ 120588)-invexity is proved and some sufficient conditions are presentedunder this weak assumption All results are given in terms of Clark subdifferential
1 Preliminaries and Introduction
First we briefly overview some notions of convex analysisand nonsmooth analysis widely used in the formulations andproofs of the main results of the paper For more detailsdiscussion and applications see [1ndash3]
Given a nonempty set 119860 sube R119899 we denote with 119860 ri(119860)conv(119860) and cone(119860) the closure of 119860 the relative interiorof 119860 convex hull and convex cone (containing the origin)generated by 119860 respectively The polar cone and strict polarcone of 119860 are defined respectively by
119860minus fl 119909 isin R
119899
| ⟨119909 119886⟩ le 0 forall119886 isin 119860
119860119904 fl 119909 isin R
119899
| ⟨119909 119886⟩ lt 0 forall119886 isin 119860
(1)
It is easy to show that if119860119904 = 120601 then119860119904 = 119860minus The bipolar
theorem states that
119860minusminus
= cone (119860) fl cone (119860) (2)
The cone of feasible direction of119860 at isin 119860 is the cone definedby
It is worth observing that if is a minimizer of convexfunction 120601 on a convex set 119862 then
0 isin 120597120601 () + 119873 (119862 ) (4)
where 119873(119862 ) and 120597120601() denote respectively the normalcone of 119862 at and the convex subdifferential of 120601 at thatis
119873(119862 ) fl 119910 isin R119899
| ⟨119910 119909 minus ⟩ le 0 forall119909 isin 119862
120597120601 ()
fl 120585 isin R119899
| 120601 (119909) ge 120601 () + ⟨120585 119909 minus ⟩ forall119909 isin R119899
(5)
We observe that if 119870 sube R119899 is an arbitrary set and isin 119870then
119873(119870minus
) = 119870minusminus
(6)
If 119860120572
| 120572 isin Λ is a collection of convex sets in R119899 and119861 fl ⋃
120572isinΛ119860120572 then it is easy to see that
conv (119861) =
119896
sum
119895=1
120582120572119895119886120572119895
| 119886120572119895
isin 119860120572119895 119896 isin N 120582
120572119895
ge 0
119896
sum
119895=1
120582120572119895
= 1
(7)
cone (119861) =
119896
sum
119895=1
120582120572119895119886120572119895
| 119886120572119895
isin 119860120572119895 119896 isin N 120582
120572119895ge 0
(8)
Hindawi Publishing CorporationAbstract and Applied AnalysisVolume 2016 Article ID 5367190 6 pageshttpdxdoiorg10115520165367190
2 Abstract and Applied Analysis
Let isin R119899 and let120593 R119899 rarr R be a locally Lipschitz func-tion The Clarke directional derivative of 120593 at in thedirection V isin R119899 and the Clarke subdifferential of 120593 at arerespectively given by
1205930
( V) fl lim sup119910rarr119905darr0
120593 (119910 + 119905V) minus 120593 (119910)
119905
120597119888120593 () fl 120585 isin R
119899
| ⟨120585 V⟩ le 1205930
( V) forallV isin R119899
(9)
The Clarke subdifferential is a natural generalization of theclassical derivative since it is known that when function 120593 iscontinuously differentiable at 120597
119888120593() = nabla120593() Moreover
when a function 120593 is convex the Clarke subdifferential coin-cides with the subdifferential in the sense of convex analysis
In the following theorem we summarize some importantproperties of the Clarke directional derivative and the Clarkesubdifferential from [1] which are widely used in what fol-lows
Theorem 1 Let 120593 and 120601 be functions from R119899 to R which areLipschitz near Then
(i) the function V rarr 1205930
( V) is finite positively homoge-neous and subadditive on R119899 and
(ii) 120597119888120593() is a nonempty convex and compact subset of
R119899(iii) 120593
0
(119909 V) is upper semicontinuous as a function of (119909 V)
In this paper we have considered the following multiob-jective semi-infinite programming problem
(MOSIP) inf (1198911(119909) 119891
2(119909) 119891
119901(119909))
st 119892119905(119909) le 0 119905 isin 119879
119909 isin R119899
(12)
where 119891119894 119894 isin 119868 fl 1 2 119901 and 119892
119905 119905 isin 119879 are locally
Lipschitz functions fromR119899 toRcup +infin and the index set 119879is arbitrary not necessarily finite (but nonempty)
For differentiable MOSIP where 119879 is finite necessaryconditions of KKT type have been established under vari-ous constraint qualifications in [4] The Abadie constraintqualification and related constraint qualification for semi-infinite systems of convex inequalities and linear inequalitiesare also studied in [5] There the characterizations of variousconstraint qualifications in terms of upper semicontinuity ofcertain multifunctions are given
There are only a few works available that deal with opti-mality conditions for MOSIP For instance for differentiableMOSIPs some optimality conditions have been presented byCaristi et al in [6] Glover et al in [7] considered a nondiffer-entiable convex MOSIP and presented optimality theorems
for it For a nonsmooth MOSIP the ldquobasic constraint qualifi-cationrdquo has been studied by Chuong andKim in [8] who havegiven optimality and duality conditions of Karush-Kuhn-Tucker (KKT briefly) type for the problemwhich involves thenotion ofMordukhovich subdifferential Also Gao presentedsome sufficient and duality results for MOSIPs under thevarious generalized convexity assumptions in [9 10]
This paper is structured as follows In Section 2 we pro-pose a Fritz-John type necessary condition after we derive aKKT type necessary condition for optimality of the consid-ered problem under a suitable qualification condition andweestablish the strong KKT necessary conditions for an efficientsolution of the considered problem In Section 3 we obtain anextension of a Caristi-Ferrara-Stefanescu result for the (Φ 120588)-invexity
2 Necessary Conditions
As a starting point of this section we denote with 119872 thefeasible region of (MOSIP) that is
119872 fl 119909 isin R119899
| 119892119905(119909) le 0 forall119905 isin 119879 (13)
For a given isin 119872 let 119879() denote the index set of all activeconstraints at
A feasible point is said to be an efficient solution [respweakly efficient solution] to problem (MOSIP) iff there is no119909 isin 119872 satisfying 119891
119894(119909) le 119891
119894() 119894 isin 119868 and (119891
1(119909)
119891119901(119909)) = (119891
1() 119891
119901()) [resp 119891
119894(119909) lt 119891
119894() 119894 isin 119868]
For each isin 119872 set
119865fl ⋃
119894isin119868
120597119888119891119894()
119866fl ⋃
119905isin119879()
120597119888119892119905()
(15)
For each 119909 isin 119872 set
Ψ (119909) fl sup119905isin119879
119892119905(119909) (16)
Recall the following definition from [11]
Definition 2 We say that MOSIP has the Pshenichnyi-Levin-Valadire (PLV) property at 119909 isin 119872 if Ψ(sdot) is Lipschitz around119909 and
120597119888Ψ (119909) sube conv( ⋃
119905isin119879(119909)
120597119888119892119905(119909)) = conv (119866
119909) (17)
The following condition is well known even in differen-tiable cases (see eg [5 6])
Assumption A The index set119879 is a nonempty compact subsetofR119897 the function (119909 119905) rarr 119892
119905(119909) is upper semicontinuous on
R119899 times 119879 and 120597119888119892119905(119909) is an upper semicontinuous mapping in
119905 for each 119909
Abstract and Applied Analysis 3
The following lemma from [5Theorem 5] will be used insequel
Lemma 3 Suppose that Assumption A holds Then
(1) 119866is a compact set for each isin 119872
(2) the PLV property holds at each isin 119872
The following result is an extension of [6 Theorem 4]
Theorem 4 (FJ necessary condition) Let be a weaklyefficient solution ofMOSIP If condition A holds at then thereexist 120572
119894ge 0 (for 119894 = 1 2 119901) and 120573
119905ge 0 for 119905 isin 119879() with
120573119905
= 0 for finitely many indexes such that
0 isin
119901
sum
119894=1
120572119894120597119888119891119894() + sum
119905isin119879()
120573119905120597119888119892119905()
119901
sum
119894=1
120572119894+ sum
119905isin119879()
120573119905= 1
(18)
Proof Weknow fromLemma 3 that119866is a compact setThus
119865cup 119866is also a compact set and hence conv(119865
cup 119866) is
closedIf 0 notin conv(119865
cup119866) by strict separation theoremwe find
119902 isin R119899 such that ⟨119902 119906⟩ lt 0 for all 119906 isin conv(119865cup 119866) This
implies that
119902 isin (conv (119865cup 119866))119904
= (119865cup 119866)119904
= 119865119904
cap 119866119904
(19)
Since 119889 isin 119866119904
and the PLV property is satisfied at (by
Lemma 3) we have
119889 = (conv (119866))119904
sube (120597119888Ψ ())
119904
997904rArr
Ψ0
( 119889) lt 0
(20)
Thus there exists 120575 gt 0 such that Ψ( + 120576119889) minus Ψ() lt 0 forall 120576 isin (0 120575) The last inequality and the fact that Ψ() le 0
(since isin 119872) conclude that Ψ( + 120576119889) lt 0 and hence
For each 119894 = 1 119901 we find 120575119894gt 0 such that
119891119894( + 120576119889) minus 119891
119894() lt 0 forall120576 isin (0 120575
119894) (23)
Take fl min120575 1205751 120575
119901 By (21) and (23) for each 120576 isin
(0 ) we have
(1198911( + 120576119889) 119891
119901( + 120576119889))
lt (1198911() 119891
119901()) + 120576119889 isin 119872
(24)
which contradicts the weak efficiency of This contradictionimplies that
0 isin conv (119865cup 119866) (25)
Now (7) proves the result
Thenecessary conditions of Fritz-John type can be viewedas being degenerate when the multiplier corresponding tothe objective function vanishes because the function beingminimized is not involved In the next theorem we derive aKarush-Kuhn-Tuker type necessary condition for optimalityof MOSIP under a suitable qualification condition
Definition 5 Let isin 119872 We say that MOSIP satisfies theZangwill CQ (ZCQ briefly) at if
119866minus
sube 119863 (119872 ) (26)
Theorem 6 Let be a weakly efficient solution of MOSIPZCQ hold at and 119888119900119899119890(119866()) be a closed cone Then thereexist 120572
119894ge 0 (for 119894 = 1 2 119901) and 120573
119905ge 0 for 119905 isin 119879() with
120573119905
= 0 for finitely many indexes such that
0 isin
119901
sum
119894=1
120572119894120597119888119891119894() + sum
119905isin119879()
120573119905120597119888119892119905()
119901
sum
119894=1
120572119894= 1
(27)
Proof We first claim that
max1le119894le119901
1198910
119894( 119889) ge 0 forall119889 isin 119863 (119872 ) (28)
On the contrary suppose that there exists 119889 isin 119863(119872 ) suchthat 1198910
119894( 119889) lt 0 for all 119894 = 1 119901 Thus there exist positive
which contradicts the weak efficiency of Thus (28) is trueIf isin 119863(119872 ) there exists a sequence 119889
119896infin
119896=1in119863(119872 )
converging to Owing to (28) and continuity of function120593(119889) fl max
1le119894le1199011198910
119894( 119889) we deduce that
120593 () = lim119896rarrinfin
120593 (119889119896) ge 0 (31)
We thus proved that (by assumption of ZCQ at )
120593 (119889) = max1le119894le119901
1198910
119894( 119889) ge 0 forall119889 isin 119866
(32)
4 Abstract and Applied Analysis
Since 0 isin 119866minus
and 120593(0) = 0 the last relation implies that the
following convex problem has a minimum at 119889 fl 0
min 120593 (119889)
subject to 119889 isin 119866
(33)
Hence by (4) (6) and (11) we obtain that
0 isin 120597120593 (0) + 119873 (119866minus
0)
= conv(
119901
⋃
119894=1
1205971198910
119894( sdot) (0)) + 119866
minusminus
= conv (119865) + cone (119866
)
(34)
Now the closeness of cone(119866()) (2) (7) and (8) prove theresults
In almost all examples we were not able to obtain positiveKKT multipliers associated with the vector-valued objectivefunction that is to say some of the multipliers may beequal to zero This means that the components of the vector-valued objective function do not have a role in the necessaryconditions for weak efficiency In order to avoid the casewhere some of the KKT multipliers associated with theobjective function vanish for the MOSIP an approach hasbeen developed in [5] and strong KKT necessary optimalityconditions have been obtained We say that strong KKTcondition holds for a multiobjective optimization problemwhen the KKTmultipliers are positive for all the componentsof the objective function Below we establish the necessarystrong KKT conditions for an efficient solution (not a weakefficient solution) of MOSIP under a suitable qualificationcondition
Let isin 119878 On the lines of [4] for each 119894 isin 119868 define the set
119876119894
() fl 119909 isin 119872 | 119891119897(119909) le 119891
119897() forall119897 isin 119868 119894
119876119894
() fl 119872 if 119901 = 1
(35)
For the sake of simplicity we denote119876119894() by119876119894 in this paper
Definition 7 Let isin 119872 We say that MOSIP satisfies thestrong Zangwill CQ (SZCQ briefly) at if
119866minus
sube
119901
⋂
119894=1
119863(119876119894 ) (36)
Theorem 8 (strong KKT necessary condition) Let be anefficient solution of MOSIP If in addition SZCQ and thecondition
(A) (
119901
⋃
119894=1
120597119888119891119894())
minus
0 sube
119901
⋃
119894=119894
(120597119888119891119894())119904 (37)
hold at then 120572119894gt 0 exist (for 119894 = 1 2 119901) and 120573
119905ge 0 for
119905 isin 119879() with 120573119905
= 0 for finitely many indexes such that
0 isin
119901
sum
119894=1
120572119894120597119888119891119894() + sum
119905isin119879()
120573119905120597119888119892119905() (38)
Proof We present the proof in four steps
Step 1 We claim that
(
119901
⋃
119894=1
(120597119888119891119894())119904
) cap (
119901
⋂
119894=1
119863(119876119894
)) = 0 (39)
It suffices only to prove that
(120597119888119891119897())119904
cap 119863(119876119897
) = 0 forall119897 isin 119868 (40)
On the contrary suppose that for some 119897 isin 119868 there is a vector119889 such that
119889 isin (120597119888119891119897())119904
cap 119863(119876119897
) (41)
By the definition of 119863(119876119897
) there exists 120575 gt 0 such that +
120576119889 isin 119876119897 for each 120576 isin (0 120575) Thus owing to the definition of
From these and 119894gt 0 (for all 119894 isin 119868) and
119905119904ge 0 (for all
119904 isin 1 119898) we obtain
119901
sum
119894=1
119894Φ(119909 120585
119894 120588 (119909 ))
+
119898
sum
119904=1
119905119904Φ(119909 120577
119905119904 120588119905119904(119909 )) lt 0
(66)
which contradicts (60)
Remark 12 Similar to [6] we can define some weaker (Φ 120588)-invexiy assumption for the function ℏ and then we can provesome weaker sufficient conditions for optimality of MOSIPSince the proof of these extensions is similar to previoustheorems we omit them
Competing Interests
The authors declare that they have no competing interests
References
[1] F H Clarke Optimization and Nonsmooth Analysis CanadianMathematical Society Series of Monographs and AdvancedTexts John Wiley amp Sons 1983
[2] J B Hiriart-Urruty and C Lemarechal Convex Analysis andMinimization Algorithms I amp II Springer Berlin Germany1991
[3] R T Rockafellar Convex Analysis Princeton University PressPrinceton NJ USA 1970
[4] T Maeda ldquoConstraint qualifications in multiobjective opti-mization problems differentiable caserdquo Journal of OptimizationTheory and Applications vol 80 no 3 pp 483ndash500 1994
[5] N Kanzi ldquoOn strong KKT optimality conditions for multiob-jective semi-infinite programming problems with lipschitziandatardquo Optimization Letters vol 9 no 6 pp 1121ndash1129 2015
[6] G Caristi M Ferrara and A Stefanescu ldquoSemi-infinite multi-objective programming with grneralized invexityrdquo Mathemati-cal Reports vol 62 pp 217ndash233 2010
[7] B M Glover V Jeyakumar and A M Rubinov ldquoDual con-ditions characterizing optimality for convex multi-objectiveprogramsrdquo Mathematical Programming vol 84 no 1 pp 201ndash217 1999
[8] T D Chuong and D S Kim ldquoNonsmooth semi-infinite multi-objective optimization problemsrdquo Journal of Optimization The-ory and Applications vol 160 no 3 pp 748ndash762 2014
[9] X Y Gao ldquoNecessary optimality and duality for multiobjectivesemi-infinite programmingrdquo Journal of Theoretical and AppliedInformation Technology vol 46 no 1 pp 347ndash354 2012
[10] X Y Gao ldquoOptimality and duality for non-smooth multipleobjective semi-infinite programmingrdquo Journal of Networks vol8 no 2 pp 413ndash420 2013
[11] N Kanzi ldquoConstraint qualifications in semi-infinite systemsand their applications in nonsmooth semi-infinite problemswith mixed constraintsrdquo SIAM Journal on Optimization vol 24no 2 pp 559ndash572 2014
Let isin R119899 and let120593 R119899 rarr R be a locally Lipschitz func-tion The Clarke directional derivative of 120593 at in thedirection V isin R119899 and the Clarke subdifferential of 120593 at arerespectively given by
1205930
( V) fl lim sup119910rarr119905darr0
120593 (119910 + 119905V) minus 120593 (119910)
119905
120597119888120593 () fl 120585 isin R
119899
| ⟨120585 V⟩ le 1205930
( V) forallV isin R119899
(9)
The Clarke subdifferential is a natural generalization of theclassical derivative since it is known that when function 120593 iscontinuously differentiable at 120597
119888120593() = nabla120593() Moreover
when a function 120593 is convex the Clarke subdifferential coin-cides with the subdifferential in the sense of convex analysis
In the following theorem we summarize some importantproperties of the Clarke directional derivative and the Clarkesubdifferential from [1] which are widely used in what fol-lows
Theorem 1 Let 120593 and 120601 be functions from R119899 to R which areLipschitz near Then
(i) the function V rarr 1205930
( V) is finite positively homoge-neous and subadditive on R119899 and
(ii) 120597119888120593() is a nonempty convex and compact subset of
R119899(iii) 120593
0
(119909 V) is upper semicontinuous as a function of (119909 V)
In this paper we have considered the following multiob-jective semi-infinite programming problem
(MOSIP) inf (1198911(119909) 119891
2(119909) 119891
119901(119909))
st 119892119905(119909) le 0 119905 isin 119879
119909 isin R119899
(12)
where 119891119894 119894 isin 119868 fl 1 2 119901 and 119892
119905 119905 isin 119879 are locally
Lipschitz functions fromR119899 toRcup +infin and the index set 119879is arbitrary not necessarily finite (but nonempty)
For differentiable MOSIP where 119879 is finite necessaryconditions of KKT type have been established under vari-ous constraint qualifications in [4] The Abadie constraintqualification and related constraint qualification for semi-infinite systems of convex inequalities and linear inequalitiesare also studied in [5] There the characterizations of variousconstraint qualifications in terms of upper semicontinuity ofcertain multifunctions are given
There are only a few works available that deal with opti-mality conditions for MOSIP For instance for differentiableMOSIPs some optimality conditions have been presented byCaristi et al in [6] Glover et al in [7] considered a nondiffer-entiable convex MOSIP and presented optimality theorems
for it For a nonsmooth MOSIP the ldquobasic constraint qualifi-cationrdquo has been studied by Chuong andKim in [8] who havegiven optimality and duality conditions of Karush-Kuhn-Tucker (KKT briefly) type for the problemwhich involves thenotion ofMordukhovich subdifferential Also Gao presentedsome sufficient and duality results for MOSIPs under thevarious generalized convexity assumptions in [9 10]
This paper is structured as follows In Section 2 we pro-pose a Fritz-John type necessary condition after we derive aKKT type necessary condition for optimality of the consid-ered problem under a suitable qualification condition andweestablish the strong KKT necessary conditions for an efficientsolution of the considered problem In Section 3 we obtain anextension of a Caristi-Ferrara-Stefanescu result for the (Φ 120588)-invexity
2 Necessary Conditions
As a starting point of this section we denote with 119872 thefeasible region of (MOSIP) that is
119872 fl 119909 isin R119899
| 119892119905(119909) le 0 forall119905 isin 119879 (13)
For a given isin 119872 let 119879() denote the index set of all activeconstraints at
A feasible point is said to be an efficient solution [respweakly efficient solution] to problem (MOSIP) iff there is no119909 isin 119872 satisfying 119891
119894(119909) le 119891
119894() 119894 isin 119868 and (119891
1(119909)
119891119901(119909)) = (119891
1() 119891
119901()) [resp 119891
119894(119909) lt 119891
119894() 119894 isin 119868]
For each isin 119872 set
119865fl ⋃
119894isin119868
120597119888119891119894()
119866fl ⋃
119905isin119879()
120597119888119892119905()
(15)
For each 119909 isin 119872 set
Ψ (119909) fl sup119905isin119879
119892119905(119909) (16)
Recall the following definition from [11]
Definition 2 We say that MOSIP has the Pshenichnyi-Levin-Valadire (PLV) property at 119909 isin 119872 if Ψ(sdot) is Lipschitz around119909 and
120597119888Ψ (119909) sube conv( ⋃
119905isin119879(119909)
120597119888119892119905(119909)) = conv (119866
119909) (17)
The following condition is well known even in differen-tiable cases (see eg [5 6])
Assumption A The index set119879 is a nonempty compact subsetofR119897 the function (119909 119905) rarr 119892
119905(119909) is upper semicontinuous on
R119899 times 119879 and 120597119888119892119905(119909) is an upper semicontinuous mapping in
119905 for each 119909
Abstract and Applied Analysis 3
The following lemma from [5Theorem 5] will be used insequel
Lemma 3 Suppose that Assumption A holds Then
(1) 119866is a compact set for each isin 119872
(2) the PLV property holds at each isin 119872
The following result is an extension of [6 Theorem 4]
Theorem 4 (FJ necessary condition) Let be a weaklyefficient solution ofMOSIP If condition A holds at then thereexist 120572
119894ge 0 (for 119894 = 1 2 119901) and 120573
119905ge 0 for 119905 isin 119879() with
120573119905
= 0 for finitely many indexes such that
0 isin
119901
sum
119894=1
120572119894120597119888119891119894() + sum
119905isin119879()
120573119905120597119888119892119905()
119901
sum
119894=1
120572119894+ sum
119905isin119879()
120573119905= 1
(18)
Proof Weknow fromLemma 3 that119866is a compact setThus
119865cup 119866is also a compact set and hence conv(119865
cup 119866) is
closedIf 0 notin conv(119865
cup119866) by strict separation theoremwe find
119902 isin R119899 such that ⟨119902 119906⟩ lt 0 for all 119906 isin conv(119865cup 119866) This
implies that
119902 isin (conv (119865cup 119866))119904
= (119865cup 119866)119904
= 119865119904
cap 119866119904
(19)
Since 119889 isin 119866119904
and the PLV property is satisfied at (by
Lemma 3) we have
119889 = (conv (119866))119904
sube (120597119888Ψ ())
119904
997904rArr
Ψ0
( 119889) lt 0
(20)
Thus there exists 120575 gt 0 such that Ψ( + 120576119889) minus Ψ() lt 0 forall 120576 isin (0 120575) The last inequality and the fact that Ψ() le 0
(since isin 119872) conclude that Ψ( + 120576119889) lt 0 and hence
For each 119894 = 1 119901 we find 120575119894gt 0 such that
119891119894( + 120576119889) minus 119891
119894() lt 0 forall120576 isin (0 120575
119894) (23)
Take fl min120575 1205751 120575
119901 By (21) and (23) for each 120576 isin
(0 ) we have
(1198911( + 120576119889) 119891
119901( + 120576119889))
lt (1198911() 119891
119901()) + 120576119889 isin 119872
(24)
which contradicts the weak efficiency of This contradictionimplies that
0 isin conv (119865cup 119866) (25)
Now (7) proves the result
Thenecessary conditions of Fritz-John type can be viewedas being degenerate when the multiplier corresponding tothe objective function vanishes because the function beingminimized is not involved In the next theorem we derive aKarush-Kuhn-Tuker type necessary condition for optimalityof MOSIP under a suitable qualification condition
Definition 5 Let isin 119872 We say that MOSIP satisfies theZangwill CQ (ZCQ briefly) at if
119866minus
sube 119863 (119872 ) (26)
Theorem 6 Let be a weakly efficient solution of MOSIPZCQ hold at and 119888119900119899119890(119866()) be a closed cone Then thereexist 120572
119894ge 0 (for 119894 = 1 2 119901) and 120573
119905ge 0 for 119905 isin 119879() with
120573119905
= 0 for finitely many indexes such that
0 isin
119901
sum
119894=1
120572119894120597119888119891119894() + sum
119905isin119879()
120573119905120597119888119892119905()
119901
sum
119894=1
120572119894= 1
(27)
Proof We first claim that
max1le119894le119901
1198910
119894( 119889) ge 0 forall119889 isin 119863 (119872 ) (28)
On the contrary suppose that there exists 119889 isin 119863(119872 ) suchthat 1198910
119894( 119889) lt 0 for all 119894 = 1 119901 Thus there exist positive
which contradicts the weak efficiency of Thus (28) is trueIf isin 119863(119872 ) there exists a sequence 119889
119896infin
119896=1in119863(119872 )
converging to Owing to (28) and continuity of function120593(119889) fl max
1le119894le1199011198910
119894( 119889) we deduce that
120593 () = lim119896rarrinfin
120593 (119889119896) ge 0 (31)
We thus proved that (by assumption of ZCQ at )
120593 (119889) = max1le119894le119901
1198910
119894( 119889) ge 0 forall119889 isin 119866
(32)
4 Abstract and Applied Analysis
Since 0 isin 119866minus
and 120593(0) = 0 the last relation implies that the
following convex problem has a minimum at 119889 fl 0
min 120593 (119889)
subject to 119889 isin 119866
(33)
Hence by (4) (6) and (11) we obtain that
0 isin 120597120593 (0) + 119873 (119866minus
0)
= conv(
119901
⋃
119894=1
1205971198910
119894( sdot) (0)) + 119866
minusminus
= conv (119865) + cone (119866
)
(34)
Now the closeness of cone(119866()) (2) (7) and (8) prove theresults
In almost all examples we were not able to obtain positiveKKT multipliers associated with the vector-valued objectivefunction that is to say some of the multipliers may beequal to zero This means that the components of the vector-valued objective function do not have a role in the necessaryconditions for weak efficiency In order to avoid the casewhere some of the KKT multipliers associated with theobjective function vanish for the MOSIP an approach hasbeen developed in [5] and strong KKT necessary optimalityconditions have been obtained We say that strong KKTcondition holds for a multiobjective optimization problemwhen the KKTmultipliers are positive for all the componentsof the objective function Below we establish the necessarystrong KKT conditions for an efficient solution (not a weakefficient solution) of MOSIP under a suitable qualificationcondition
Let isin 119878 On the lines of [4] for each 119894 isin 119868 define the set
119876119894
() fl 119909 isin 119872 | 119891119897(119909) le 119891
119897() forall119897 isin 119868 119894
119876119894
() fl 119872 if 119901 = 1
(35)
For the sake of simplicity we denote119876119894() by119876119894 in this paper
Definition 7 Let isin 119872 We say that MOSIP satisfies thestrong Zangwill CQ (SZCQ briefly) at if
119866minus
sube
119901
⋂
119894=1
119863(119876119894 ) (36)
Theorem 8 (strong KKT necessary condition) Let be anefficient solution of MOSIP If in addition SZCQ and thecondition
(A) (
119901
⋃
119894=1
120597119888119891119894())
minus
0 sube
119901
⋃
119894=119894
(120597119888119891119894())119904 (37)
hold at then 120572119894gt 0 exist (for 119894 = 1 2 119901) and 120573
119905ge 0 for
119905 isin 119879() with 120573119905
= 0 for finitely many indexes such that
0 isin
119901
sum
119894=1
120572119894120597119888119891119894() + sum
119905isin119879()
120573119905120597119888119892119905() (38)
Proof We present the proof in four steps
Step 1 We claim that
(
119901
⋃
119894=1
(120597119888119891119894())119904
) cap (
119901
⋂
119894=1
119863(119876119894
)) = 0 (39)
It suffices only to prove that
(120597119888119891119897())119904
cap 119863(119876119897
) = 0 forall119897 isin 119868 (40)
On the contrary suppose that for some 119897 isin 119868 there is a vector119889 such that
119889 isin (120597119888119891119897())119904
cap 119863(119876119897
) (41)
By the definition of 119863(119876119897
) there exists 120575 gt 0 such that +
120576119889 isin 119876119897 for each 120576 isin (0 120575) Thus owing to the definition of
From these and 119894gt 0 (for all 119894 isin 119868) and
119905119904ge 0 (for all
119904 isin 1 119898) we obtain
119901
sum
119894=1
119894Φ(119909 120585
119894 120588 (119909 ))
+
119898
sum
119904=1
119905119904Φ(119909 120577
119905119904 120588119905119904(119909 )) lt 0
(66)
which contradicts (60)
Remark 12 Similar to [6] we can define some weaker (Φ 120588)-invexiy assumption for the function ℏ and then we can provesome weaker sufficient conditions for optimality of MOSIPSince the proof of these extensions is similar to previoustheorems we omit them
Competing Interests
The authors declare that they have no competing interests
References
[1] F H Clarke Optimization and Nonsmooth Analysis CanadianMathematical Society Series of Monographs and AdvancedTexts John Wiley amp Sons 1983
[2] J B Hiriart-Urruty and C Lemarechal Convex Analysis andMinimization Algorithms I amp II Springer Berlin Germany1991
[3] R T Rockafellar Convex Analysis Princeton University PressPrinceton NJ USA 1970
[4] T Maeda ldquoConstraint qualifications in multiobjective opti-mization problems differentiable caserdquo Journal of OptimizationTheory and Applications vol 80 no 3 pp 483ndash500 1994
[5] N Kanzi ldquoOn strong KKT optimality conditions for multiob-jective semi-infinite programming problems with lipschitziandatardquo Optimization Letters vol 9 no 6 pp 1121ndash1129 2015
[6] G Caristi M Ferrara and A Stefanescu ldquoSemi-infinite multi-objective programming with grneralized invexityrdquo Mathemati-cal Reports vol 62 pp 217ndash233 2010
[7] B M Glover V Jeyakumar and A M Rubinov ldquoDual con-ditions characterizing optimality for convex multi-objectiveprogramsrdquo Mathematical Programming vol 84 no 1 pp 201ndash217 1999
[8] T D Chuong and D S Kim ldquoNonsmooth semi-infinite multi-objective optimization problemsrdquo Journal of Optimization The-ory and Applications vol 160 no 3 pp 748ndash762 2014
[9] X Y Gao ldquoNecessary optimality and duality for multiobjectivesemi-infinite programmingrdquo Journal of Theoretical and AppliedInformation Technology vol 46 no 1 pp 347ndash354 2012
[10] X Y Gao ldquoOptimality and duality for non-smooth multipleobjective semi-infinite programmingrdquo Journal of Networks vol8 no 2 pp 413ndash420 2013
[11] N Kanzi ldquoConstraint qualifications in semi-infinite systemsand their applications in nonsmooth semi-infinite problemswith mixed constraintsrdquo SIAM Journal on Optimization vol 24no 2 pp 559ndash572 2014
For each 119894 = 1 119901 we find 120575119894gt 0 such that
119891119894( + 120576119889) minus 119891
119894() lt 0 forall120576 isin (0 120575
119894) (23)
Take fl min120575 1205751 120575
119901 By (21) and (23) for each 120576 isin
(0 ) we have
(1198911( + 120576119889) 119891
119901( + 120576119889))
lt (1198911() 119891
119901()) + 120576119889 isin 119872
(24)
which contradicts the weak efficiency of This contradictionimplies that
0 isin conv (119865cup 119866) (25)
Now (7) proves the result
Thenecessary conditions of Fritz-John type can be viewedas being degenerate when the multiplier corresponding tothe objective function vanishes because the function beingminimized is not involved In the next theorem we derive aKarush-Kuhn-Tuker type necessary condition for optimalityof MOSIP under a suitable qualification condition
Definition 5 Let isin 119872 We say that MOSIP satisfies theZangwill CQ (ZCQ briefly) at if
119866minus
sube 119863 (119872 ) (26)
Theorem 6 Let be a weakly efficient solution of MOSIPZCQ hold at and 119888119900119899119890(119866()) be a closed cone Then thereexist 120572
119894ge 0 (for 119894 = 1 2 119901) and 120573
119905ge 0 for 119905 isin 119879() with
120573119905
= 0 for finitely many indexes such that
0 isin
119901
sum
119894=1
120572119894120597119888119891119894() + sum
119905isin119879()
120573119905120597119888119892119905()
119901
sum
119894=1
120572119894= 1
(27)
Proof We first claim that
max1le119894le119901
1198910
119894( 119889) ge 0 forall119889 isin 119863 (119872 ) (28)
On the contrary suppose that there exists 119889 isin 119863(119872 ) suchthat 1198910
119894( 119889) lt 0 for all 119894 = 1 119901 Thus there exist positive
which contradicts the weak efficiency of Thus (28) is trueIf isin 119863(119872 ) there exists a sequence 119889
119896infin
119896=1in119863(119872 )
converging to Owing to (28) and continuity of function120593(119889) fl max
1le119894le1199011198910
119894( 119889) we deduce that
120593 () = lim119896rarrinfin
120593 (119889119896) ge 0 (31)
We thus proved that (by assumption of ZCQ at )
120593 (119889) = max1le119894le119901
1198910
119894( 119889) ge 0 forall119889 isin 119866
(32)
4 Abstract and Applied Analysis
Since 0 isin 119866minus
and 120593(0) = 0 the last relation implies that the
following convex problem has a minimum at 119889 fl 0
min 120593 (119889)
subject to 119889 isin 119866
(33)
Hence by (4) (6) and (11) we obtain that
0 isin 120597120593 (0) + 119873 (119866minus
0)
= conv(
119901
⋃
119894=1
1205971198910
119894( sdot) (0)) + 119866
minusminus
= conv (119865) + cone (119866
)
(34)
Now the closeness of cone(119866()) (2) (7) and (8) prove theresults
In almost all examples we were not able to obtain positiveKKT multipliers associated with the vector-valued objectivefunction that is to say some of the multipliers may beequal to zero This means that the components of the vector-valued objective function do not have a role in the necessaryconditions for weak efficiency In order to avoid the casewhere some of the KKT multipliers associated with theobjective function vanish for the MOSIP an approach hasbeen developed in [5] and strong KKT necessary optimalityconditions have been obtained We say that strong KKTcondition holds for a multiobjective optimization problemwhen the KKTmultipliers are positive for all the componentsof the objective function Below we establish the necessarystrong KKT conditions for an efficient solution (not a weakefficient solution) of MOSIP under a suitable qualificationcondition
Let isin 119878 On the lines of [4] for each 119894 isin 119868 define the set
119876119894
() fl 119909 isin 119872 | 119891119897(119909) le 119891
119897() forall119897 isin 119868 119894
119876119894
() fl 119872 if 119901 = 1
(35)
For the sake of simplicity we denote119876119894() by119876119894 in this paper
Definition 7 Let isin 119872 We say that MOSIP satisfies thestrong Zangwill CQ (SZCQ briefly) at if
119866minus
sube
119901
⋂
119894=1
119863(119876119894 ) (36)
Theorem 8 (strong KKT necessary condition) Let be anefficient solution of MOSIP If in addition SZCQ and thecondition
(A) (
119901
⋃
119894=1
120597119888119891119894())
minus
0 sube
119901
⋃
119894=119894
(120597119888119891119894())119904 (37)
hold at then 120572119894gt 0 exist (for 119894 = 1 2 119901) and 120573
119905ge 0 for
119905 isin 119879() with 120573119905
= 0 for finitely many indexes such that
0 isin
119901
sum
119894=1
120572119894120597119888119891119894() + sum
119905isin119879()
120573119905120597119888119892119905() (38)
Proof We present the proof in four steps
Step 1 We claim that
(
119901
⋃
119894=1
(120597119888119891119894())119904
) cap (
119901
⋂
119894=1
119863(119876119894
)) = 0 (39)
It suffices only to prove that
(120597119888119891119897())119904
cap 119863(119876119897
) = 0 forall119897 isin 119868 (40)
On the contrary suppose that for some 119897 isin 119868 there is a vector119889 such that
119889 isin (120597119888119891119897())119904
cap 119863(119876119897
) (41)
By the definition of 119863(119876119897
) there exists 120575 gt 0 such that +
120576119889 isin 119876119897 for each 120576 isin (0 120575) Thus owing to the definition of
From these and 119894gt 0 (for all 119894 isin 119868) and
119905119904ge 0 (for all
119904 isin 1 119898) we obtain
119901
sum
119894=1
119894Φ(119909 120585
119894 120588 (119909 ))
+
119898
sum
119904=1
119905119904Φ(119909 120577
119905119904 120588119905119904(119909 )) lt 0
(66)
which contradicts (60)
Remark 12 Similar to [6] we can define some weaker (Φ 120588)-invexiy assumption for the function ℏ and then we can provesome weaker sufficient conditions for optimality of MOSIPSince the proof of these extensions is similar to previoustheorems we omit them
Competing Interests
The authors declare that they have no competing interests
References
[1] F H Clarke Optimization and Nonsmooth Analysis CanadianMathematical Society Series of Monographs and AdvancedTexts John Wiley amp Sons 1983
[2] J B Hiriart-Urruty and C Lemarechal Convex Analysis andMinimization Algorithms I amp II Springer Berlin Germany1991
[3] R T Rockafellar Convex Analysis Princeton University PressPrinceton NJ USA 1970
[4] T Maeda ldquoConstraint qualifications in multiobjective opti-mization problems differentiable caserdquo Journal of OptimizationTheory and Applications vol 80 no 3 pp 483ndash500 1994
[5] N Kanzi ldquoOn strong KKT optimality conditions for multiob-jective semi-infinite programming problems with lipschitziandatardquo Optimization Letters vol 9 no 6 pp 1121ndash1129 2015
[6] G Caristi M Ferrara and A Stefanescu ldquoSemi-infinite multi-objective programming with grneralized invexityrdquo Mathemati-cal Reports vol 62 pp 217ndash233 2010
[7] B M Glover V Jeyakumar and A M Rubinov ldquoDual con-ditions characterizing optimality for convex multi-objectiveprogramsrdquo Mathematical Programming vol 84 no 1 pp 201ndash217 1999
[8] T D Chuong and D S Kim ldquoNonsmooth semi-infinite multi-objective optimization problemsrdquo Journal of Optimization The-ory and Applications vol 160 no 3 pp 748ndash762 2014
[9] X Y Gao ldquoNecessary optimality and duality for multiobjectivesemi-infinite programmingrdquo Journal of Theoretical and AppliedInformation Technology vol 46 no 1 pp 347ndash354 2012
[10] X Y Gao ldquoOptimality and duality for non-smooth multipleobjective semi-infinite programmingrdquo Journal of Networks vol8 no 2 pp 413ndash420 2013
[11] N Kanzi ldquoConstraint qualifications in semi-infinite systemsand their applications in nonsmooth semi-infinite problemswith mixed constraintsrdquo SIAM Journal on Optimization vol 24no 2 pp 559ndash572 2014
and 120593(0) = 0 the last relation implies that the
following convex problem has a minimum at 119889 fl 0
min 120593 (119889)
subject to 119889 isin 119866
(33)
Hence by (4) (6) and (11) we obtain that
0 isin 120597120593 (0) + 119873 (119866minus
0)
= conv(
119901
⋃
119894=1
1205971198910
119894( sdot) (0)) + 119866
minusminus
= conv (119865) + cone (119866
)
(34)
Now the closeness of cone(119866()) (2) (7) and (8) prove theresults
In almost all examples we were not able to obtain positiveKKT multipliers associated with the vector-valued objectivefunction that is to say some of the multipliers may beequal to zero This means that the components of the vector-valued objective function do not have a role in the necessaryconditions for weak efficiency In order to avoid the casewhere some of the KKT multipliers associated with theobjective function vanish for the MOSIP an approach hasbeen developed in [5] and strong KKT necessary optimalityconditions have been obtained We say that strong KKTcondition holds for a multiobjective optimization problemwhen the KKTmultipliers are positive for all the componentsof the objective function Below we establish the necessarystrong KKT conditions for an efficient solution (not a weakefficient solution) of MOSIP under a suitable qualificationcondition
Let isin 119878 On the lines of [4] for each 119894 isin 119868 define the set
119876119894
() fl 119909 isin 119872 | 119891119897(119909) le 119891
119897() forall119897 isin 119868 119894
119876119894
() fl 119872 if 119901 = 1
(35)
For the sake of simplicity we denote119876119894() by119876119894 in this paper
Definition 7 Let isin 119872 We say that MOSIP satisfies thestrong Zangwill CQ (SZCQ briefly) at if
119866minus
sube
119901
⋂
119894=1
119863(119876119894 ) (36)
Theorem 8 (strong KKT necessary condition) Let be anefficient solution of MOSIP If in addition SZCQ and thecondition
(A) (
119901
⋃
119894=1
120597119888119891119894())
minus
0 sube
119901
⋃
119894=119894
(120597119888119891119894())119904 (37)
hold at then 120572119894gt 0 exist (for 119894 = 1 2 119901) and 120573
119905ge 0 for
119905 isin 119879() with 120573119905
= 0 for finitely many indexes such that
0 isin
119901
sum
119894=1
120572119894120597119888119891119894() + sum
119905isin119879()
120573119905120597119888119892119905() (38)
Proof We present the proof in four steps
Step 1 We claim that
(
119901
⋃
119894=1
(120597119888119891119894())119904
) cap (
119901
⋂
119894=1
119863(119876119894
)) = 0 (39)
It suffices only to prove that
(120597119888119891119897())119904
cap 119863(119876119897
) = 0 forall119897 isin 119868 (40)
On the contrary suppose that for some 119897 isin 119868 there is a vector119889 such that
119889 isin (120597119888119891119897())119904
cap 119863(119876119897
) (41)
By the definition of 119863(119876119897
) there exists 120575 gt 0 such that +
120576119889 isin 119876119897 for each 120576 isin (0 120575) Thus owing to the definition of
From these and 119894gt 0 (for all 119894 isin 119868) and
119905119904ge 0 (for all
119904 isin 1 119898) we obtain
119901
sum
119894=1
119894Φ(119909 120585
119894 120588 (119909 ))
+
119898
sum
119904=1
119905119904Φ(119909 120577
119905119904 120588119905119904(119909 )) lt 0
(66)
which contradicts (60)
Remark 12 Similar to [6] we can define some weaker (Φ 120588)-invexiy assumption for the function ℏ and then we can provesome weaker sufficient conditions for optimality of MOSIPSince the proof of these extensions is similar to previoustheorems we omit them
Competing Interests
The authors declare that they have no competing interests
References
[1] F H Clarke Optimization and Nonsmooth Analysis CanadianMathematical Society Series of Monographs and AdvancedTexts John Wiley amp Sons 1983
[2] J B Hiriart-Urruty and C Lemarechal Convex Analysis andMinimization Algorithms I amp II Springer Berlin Germany1991
[3] R T Rockafellar Convex Analysis Princeton University PressPrinceton NJ USA 1970
[4] T Maeda ldquoConstraint qualifications in multiobjective opti-mization problems differentiable caserdquo Journal of OptimizationTheory and Applications vol 80 no 3 pp 483ndash500 1994
[5] N Kanzi ldquoOn strong KKT optimality conditions for multiob-jective semi-infinite programming problems with lipschitziandatardquo Optimization Letters vol 9 no 6 pp 1121ndash1129 2015
[6] G Caristi M Ferrara and A Stefanescu ldquoSemi-infinite multi-objective programming with grneralized invexityrdquo Mathemati-cal Reports vol 62 pp 217ndash233 2010
[7] B M Glover V Jeyakumar and A M Rubinov ldquoDual con-ditions characterizing optimality for convex multi-objectiveprogramsrdquo Mathematical Programming vol 84 no 1 pp 201ndash217 1999
[8] T D Chuong and D S Kim ldquoNonsmooth semi-infinite multi-objective optimization problemsrdquo Journal of Optimization The-ory and Applications vol 160 no 3 pp 748ndash762 2014
[9] X Y Gao ldquoNecessary optimality and duality for multiobjectivesemi-infinite programmingrdquo Journal of Theoretical and AppliedInformation Technology vol 46 no 1 pp 347ndash354 2012
[10] X Y Gao ldquoOptimality and duality for non-smooth multipleobjective semi-infinite programmingrdquo Journal of Networks vol8 no 2 pp 413ndash420 2013
[11] N Kanzi ldquoConstraint qualifications in semi-infinite systemsand their applications in nonsmooth semi-infinite problemswith mixed constraintsrdquo SIAM Journal on Optimization vol 24no 2 pp 559ndash572 2014
From these and 119894gt 0 (for all 119894 isin 119868) and
119905119904ge 0 (for all
119904 isin 1 119898) we obtain
119901
sum
119894=1
119894Φ(119909 120585
119894 120588 (119909 ))
+
119898
sum
119904=1
119905119904Φ(119909 120577
119905119904 120588119905119904(119909 )) lt 0
(66)
which contradicts (60)
Remark 12 Similar to [6] we can define some weaker (Φ 120588)-invexiy assumption for the function ℏ and then we can provesome weaker sufficient conditions for optimality of MOSIPSince the proof of these extensions is similar to previoustheorems we omit them
Competing Interests
The authors declare that they have no competing interests
References
[1] F H Clarke Optimization and Nonsmooth Analysis CanadianMathematical Society Series of Monographs and AdvancedTexts John Wiley amp Sons 1983
[2] J B Hiriart-Urruty and C Lemarechal Convex Analysis andMinimization Algorithms I amp II Springer Berlin Germany1991
[3] R T Rockafellar Convex Analysis Princeton University PressPrinceton NJ USA 1970
[4] T Maeda ldquoConstraint qualifications in multiobjective opti-mization problems differentiable caserdquo Journal of OptimizationTheory and Applications vol 80 no 3 pp 483ndash500 1994
[5] N Kanzi ldquoOn strong KKT optimality conditions for multiob-jective semi-infinite programming problems with lipschitziandatardquo Optimization Letters vol 9 no 6 pp 1121ndash1129 2015
[6] G Caristi M Ferrara and A Stefanescu ldquoSemi-infinite multi-objective programming with grneralized invexityrdquo Mathemati-cal Reports vol 62 pp 217ndash233 2010
[7] B M Glover V Jeyakumar and A M Rubinov ldquoDual con-ditions characterizing optimality for convex multi-objectiveprogramsrdquo Mathematical Programming vol 84 no 1 pp 201ndash217 1999
[8] T D Chuong and D S Kim ldquoNonsmooth semi-infinite multi-objective optimization problemsrdquo Journal of Optimization The-ory and Applications vol 160 no 3 pp 748ndash762 2014
[9] X Y Gao ldquoNecessary optimality and duality for multiobjectivesemi-infinite programmingrdquo Journal of Theoretical and AppliedInformation Technology vol 46 no 1 pp 347ndash354 2012
[10] X Y Gao ldquoOptimality and duality for non-smooth multipleobjective semi-infinite programmingrdquo Journal of Networks vol8 no 2 pp 413ndash420 2013
[11] N Kanzi ldquoConstraint qualifications in semi-infinite systemsand their applications in nonsmooth semi-infinite problemswith mixed constraintsrdquo SIAM Journal on Optimization vol 24no 2 pp 559ndash572 2014
From these and 119894gt 0 (for all 119894 isin 119868) and
119905119904ge 0 (for all
119904 isin 1 119898) we obtain
119901
sum
119894=1
119894Φ(119909 120585
119894 120588 (119909 ))
+
119898
sum
119904=1
119905119904Φ(119909 120577
119905119904 120588119905119904(119909 )) lt 0
(66)
which contradicts (60)
Remark 12 Similar to [6] we can define some weaker (Φ 120588)-invexiy assumption for the function ℏ and then we can provesome weaker sufficient conditions for optimality of MOSIPSince the proof of these extensions is similar to previoustheorems we omit them
Competing Interests
The authors declare that they have no competing interests
References
[1] F H Clarke Optimization and Nonsmooth Analysis CanadianMathematical Society Series of Monographs and AdvancedTexts John Wiley amp Sons 1983
[2] J B Hiriart-Urruty and C Lemarechal Convex Analysis andMinimization Algorithms I amp II Springer Berlin Germany1991
[3] R T Rockafellar Convex Analysis Princeton University PressPrinceton NJ USA 1970
[4] T Maeda ldquoConstraint qualifications in multiobjective opti-mization problems differentiable caserdquo Journal of OptimizationTheory and Applications vol 80 no 3 pp 483ndash500 1994
[5] N Kanzi ldquoOn strong KKT optimality conditions for multiob-jective semi-infinite programming problems with lipschitziandatardquo Optimization Letters vol 9 no 6 pp 1121ndash1129 2015
[6] G Caristi M Ferrara and A Stefanescu ldquoSemi-infinite multi-objective programming with grneralized invexityrdquo Mathemati-cal Reports vol 62 pp 217ndash233 2010
[7] B M Glover V Jeyakumar and A M Rubinov ldquoDual con-ditions characterizing optimality for convex multi-objectiveprogramsrdquo Mathematical Programming vol 84 no 1 pp 201ndash217 1999
[8] T D Chuong and D S Kim ldquoNonsmooth semi-infinite multi-objective optimization problemsrdquo Journal of Optimization The-ory and Applications vol 160 no 3 pp 748ndash762 2014
[9] X Y Gao ldquoNecessary optimality and duality for multiobjectivesemi-infinite programmingrdquo Journal of Theoretical and AppliedInformation Technology vol 46 no 1 pp 347ndash354 2012
[10] X Y Gao ldquoOptimality and duality for non-smooth multipleobjective semi-infinite programmingrdquo Journal of Networks vol8 no 2 pp 413ndash420 2013
[11] N Kanzi ldquoConstraint qualifications in semi-infinite systemsand their applications in nonsmooth semi-infinite problemswith mixed constraintsrdquo SIAM Journal on Optimization vol 24no 2 pp 559ndash572 2014