scieee Science in your language
[en] (orig)

Necessary and sufficient optimality conditions for vector equilibrium problems on Hadamard manifolds

Abstract

The aim of this paper is to show the existence and attainability of Karush-Kuhn-Tucker optimality conditions for weakly efficient Pareto points for vector equilibrium problems with the addition of constraints in the novel context of Hadamard manifolds, as opposed to the classical examples of Banach, normed or Hausdorff spaces. More specifically, classical necessary and sufficient conditions for weakly efficient Pareto points to the constrained vector optimization problem are presented. The results described in this article generalize results obtained by Gong (2008) and Wei and Gong (2010) and Feng and Qiu (2014) from Hausdorff topological vector spaces, real normed spaces, and real Banach spaces to Hadamard manifolds, respectively. This is done using a notion of Riemannian symmetric spaces of a noncompact type as special Hadarmard manifolds.

Read accessible full text

Necessary and sufficient optimality conditions for vector equilibrium problems on Hadamard manifolds

Author: Ruiz Garzón, Gabriel; Osuna Gómez, Rafaela; Ruiz Zapatero, Jaime
Publisher: MDPI
Year: 2019
DOI: 10.3390/sym11081037
Source: https://idus.us.es/bitstreams/a99b8e83-359b-40c3-8e78-04fdd1643992/download
symme y
S
S
A icle
Necessa y and Su icien Op imali y Condi ions o
Vec o Equilib ium P oblems on
Hadama d Mani olds
Gab iel Ruiz-Ga zón 1,*,†,‡ , Ra aela Osuna-Gómez 2,‡ and Jaime Ruiz-Zapa e o 3,‡
1Depa amen o de Es adís ica e I.O., Uni e sidad de Cádiz, 11405 Cádiz, Spain
2Depa amen o de Es adís ica e I.O., Uni e sidad de Se illa, 41012 Se illa, Spain
3Depa men o Physics and As onomy, Uni e si y College o London, London WC1E 6BT, UK
*Co espondence: gab iel. [email p o ec ed]
†
Cu en add ess: Depa amen o de Es adís ica e I.O., Uni e sidad de Cádiz, Campus de Je ez de la F on e a,
A da. de la Uni e sidad s/n, 11405, Je ez de la F on e a, Cádiz, Spain.
‡ These au ho s con ibu ed equally o his wo k.
Recei ed: 18 July 2019; Accep ed: 8 Augus 2019; Published: 12 Augus 2019


Abs ac :
The aim o his pape is o show he exis ence and a ainabili y o Ka ush–Kuhn–Tucke
op imali y condi ions o weakly e icien Pa e o poin s o ec o equilib ium p oblems wi h he
addi ion o cons ain s in he no el con ex o Hadama d mani olds, as opposed o he classical
examples o Banach, no med o Hausdo spaces. Mo e speci ically, classical necessa y and su icien
condi ions o weakly e icien Pa e o poin s o he cons ained ec o op imiza ion p oblem a e
p esen ed. The esul s desc ibed in his a icle gene alize esul s ob ained by Gong (2008) and Wei
and Gong (2010) and Feng and Qiu (2014) om Hausdo opological ec o spaces, eal no med
spaces, and eal Banach spaces o Hadama d mani olds, espec i ely. This is done using a no ion o
Riemannian symme ic spaces o a noncompac ype as special Hada ma d mani olds.
Keywo ds:
ec o equilib ium p oblem; gene alized con exi y; hadama d mani olds; weakly e icien
pa e o poin s
1. In oduc ion
The pu sui o equilib ium is a ubiqui ous ho izon in p ac ically all a eas o human ac i i y.
Fo example, in economics, he dynamics o o e and demand a e ypically desc ibed as equilib ium
p oblems. In he same way, physical o social phenomena such as he dis ibu ion o pa icles in
a con aine , a ic low o elecommunica ion ne wo ks can be accu a ely concep ualized in e ms
o equilib ium.
Howe e , i was no un il Fan [
1
] ha equilib ium heo y was applied in he con ex o Euclidean
spaces. Ma hema ically, he simples de ini ion o a equilib ium p oblem consis s in inding
x∈S
such ha
F(x,y)≥0, ∀y∈S
whe e
S⊆Rp
is a nonemp y closed se and
F:Rp×Rp→R
is an equilib ium bi unc ion, i.e.,
F(x,x) = 0 o all x∈S.
Some o he main ma hema ical p oblems ha can be ph ased as equilib ium p oblems a e:
Symme y 2019,11, 1037; doi:10.3390/sym11081037 www.mdpi.com/jou nal/symme y
Symme y 2019,11, 1037 2 o 12
•
The weak minimum poin o a mul iobjec i e unc ion
= ( 1
,
. . .
,
p)
o e a closed se
S⊆Rp
is
any
¯
x∈S
such ha o any
y∈S
,
∃i
such ha
i(y)− i(¯
x)≥
0. Finding a weak minimum poin
can be educed o sol ing an equilib ium p oblem by i ue o se ing
F(x,y) = max
i=1,...,p[ i(y)− i(x)].
•The S ampacchia a ia ional inequali y p oblem demands inding ¯
x∈Ssuch ha
<G(¯
x),y−¯
x>≥0, ∀y∈S
whe e
G:Rp→Rp
and
S⊆Rp
is a closed se . This p oblem is also an equilib ium p oblem whe e
F(x,y) =<G(x),y−x>.
•
Nash equilib ium p oblems in a non-coope a i e game wi h
p
playe s whe e each playe
i
has a se
o possible s a egies
Ki⊆Rni
aim o minimize a loss unc ion
i:K→R
wi h
K=K1×. . . ×Kp
.
Thus, a Nash equilib ium poin is any
¯
x∈K
such ha no playe can educe i s loss by unila e ally
changing hei s a egy, i.e., any ¯
x∈Ksuch ha
i(¯
x)≤ i(¯
x(yi))
holds o any
yi∈Ki
o any
i=
1,
. . .
,
p
, wi h
¯
x(yi))
deno ing he ec o ob ained om
¯
x
by
eplacing ¯
xiwi h yi. The e o e, his p oblem amoun s o sol ing an equilib ium p oblem wi h
F(x,y) =
p
∑
i=1
[ i(x(yi)) − i(x)].
Despi e hei appa en di e si y, all he abo e-men ioned p oblems can be amed as pa icula
cases o he ec o equilib ium p oblem and hus can all be encompassed in a single ma hema ical
pic u e. Due o he powe o his o mula ion, i is o g ea in e es o ob ain and s udy he
Ka ush–Kuhn–Tucke (KKT) op imali y condi ions o he solu ion o such, mo e gene al p oblems.
Thanks o hei capaci y o p o ide such a undamen al insigh , ec o equilib ium p oblems
a e an ac i e b anch o non-linea analysis wi h plen y o publica ions being made up o his da e.
Fo example, in 2003, au ho s such as Iusem and Sosa [
2
] s udied he ela ion be ween equilib ium
p oblems and some auxilia y con ex p oblems. In addi ion, o e he pas cen u y, he ield o physics
depa ed om euclidean geome y as a space in which o alloca e i s heo ies, op ing ins ead o
mo e complex spaces also known as mani olds. A his o ical landma k ha illus a es his example is
Eins ein’s heo y o g a i y ha e ol es a ound he concep o space- ime cu a u e on a Riemannian
mani old. O he less known bu equally undamen al applica ions in he ields o physics in ol e he
appea ance o symplec ic mani olds in he ea men o Hamil onian ec o ields o Noe he ’s heo em.
Smoo h Riemannian mani olds a e spaces ha con ain cu a u e, as opposed o Euclidean spaces
which a e la e e ywhe e. This can be ma hema ically exp essed as
ax +by /∈M
,
∀x
,
y∈M
,
a
,
b∈R
,
whe e
M
is a Riemannian mani old. None heless, Riemannian geome y cons i u es a gene aliza ion o
he Euclidean case. This can be easily unde s ood by in oducing he no ion o angen planes. Fo any
poin o a smoo h cu ed space, say a 2-Sphe e, i is always possible o de ine a la angen plane
o ha poin ; i.e., a Euclidean space. We can hink o his in he same way we hink o he Ea h o
be la a local scales while o e all being sphe ical. Indeed, all cu ed mani olds locally esemble
Euclidean space, which is a i al p ope y o ou unde s anding o hem. Howe e , ca og aphy
can empi ically ell us ha la p ojec ions o cu ed su aces on o planes ails o ai h ully ep esen
he eal dimensions o he objec s ha li e on he o iginal cu ed su ace especially a la ge scales
Symme y 2019,11, 1037 3 o 12
whe e he locali y condi ion s a s weakening. Thus, me ici y is no longe i ial and measu emen s
o dis ances need o accoun o such cu a u e.
A his poin , we can al eady see how Euclidean spaces a e simply Riemannian mani olds o
which he angen plane o any o i s poin s is iden ical o he plane i sel . Thus, in Euclidean spaces,
ec o s li ing o he su ace a e equi alen o ec o s li ing on i s angen space. I is his key ea u e o
Euclidean geome y ha allows o he simple de ini ion o dis ance as he do p oduc . Thus, gi en a
ec o
u
, i alloca ed in an Euclidean space, i s leng h is gi en by
|u|2=<u
,
u>
. On he o he
hand, in non- la spaces i is necessa y o accoun o he dis o ion o he dis ances when p ojec ed
o he angen space. Riemannian mani olds a e hose equipped wi h a so called “me ic enso ”;
commonly deno ed
kij
, ha allows us o adequa ely de ine dis ances; i.e.,
|u|2=kijuiuj
. (see Sec ion 2
o mo e de ails).
This new de ini ion o leng h has di ec sho comings in minimiza ion and equilib ium.
The Euclidean line elemen , he sho es connec ion be ween wo poin s on a la su ace, is eplaced on
mani olds by a geodesic equa ion which plays he ole o s aigh lines in non- la spaces. This can be
seen om he ac ha geodesic cu es a e solu ions o he Eule –Lag ange equa ions which minimize
he unc ional o he Lag angian gi en by he me ic o such space,
L=kijdxidxj
, and as such desc ibe
he ajec o ies ha minimize he ac ion necessa y o mo e om A o B. Fo example, he o bi s o
plane s obey geodesics despi e clea ly no being s aigh in a Euclidean sense.
A Hadama d mani old is a simply connec ed comple e Riemannian mani old o non-posi i e
sec ional cu a u e. The mo i a ion o he s udy o Hadama d spaces is ha hey sha e some p ope ies
wi h Euclidean spaces. One o hem is he sepa a ion heo em (see Fe ei a and Oli ei a [3]).
In addi ion, o any wo poin s in
M
, he e exis s a minimal geodesic joining hese wo poin s.
In a Hadama d mani old, he geodesic be ween any wo poin s is unique and he exponen ial map a
each poin o
M
is a global di eomo phism. Mo eo e , he
exp
map is de ined on he whole angen
space ([4]).
Howe e , he minimiza ion o unc ions on a Hadama d mani old is locally equi alen o he
smoo hly cons ained op imiza ion p oblem on a Euclidean space, due o he ac ha e e y
C∞
Hadama d mani old can be isome ically embedded in an Euclidean space by i ue o John Nash’s
embedding heo em. This is consis en wi h he in ui ion we p e iously laid ou .
The s udy o op imiza ion p oblems on Hadama d mani olds is a powe ul ool. This is due o
he ac ha , gene ally, sol ing noncon ex cons ained p oblems in
Rn
wi h he Euclidean me ic can
be also amed as sol ing he uncons ained con ex minimiza ion p oblem in he Hadama d mani old
easible se wi h he a ine me ic (see [
5
]). In Colao e al. [
5
] he exis ence o solu ions o equilib ium
p oblems unde some sui able condi ions on Hadama d mani olds and hei applica ions o Nash
equilib ium o non-coope a i e games was s udied. In he same way, in Néme h [
6
] he exis ence and
uniqueness esul s o a ia ional inequali y p oblems on Hadama d mani olds we e ob ained.
Mo eo e , many op imiza ion p oblems canno be sol ed in linea spaces, o example,
con olled he monuclea usion esea ch (see [
7
]), signal p ocessing, nume ical analysis and compu e
ision (see [
8
,
9
]) equi e Hadama d mani old s uc u es o hei modeling. Also, geome ical s uc u es
hidden in da a se s o machine lea ning p oblems a e s udied in e ms o mani olds. In he ield o
medicine, Hadama d mani olds ha e been used in he analysis o magne ic esonances o quan i y he
g ow h o umo s and consequen ly deduce hei s a e o p og ession, as shown by
Fle che e al. [10]
.
The geome y necessa y o unde s and and pe o m hese echniques is bes unde s ood h ough
he use o mani olds and symme ic s uc u es. Fo example, he se o symme ic posi i e de ini e
ma ices used in magne ic esonance imaging o s udy Alzheime ’s disease [
11
] is one case in which
his ansla ion o mani olds is necessa y. In addi ion, o he p oblems in compu e ision, signal
p ocessing o lea ning algo i hms employ geodesic cu es when add essing op imiza ion p oblems.
Finally, in economics, he sea ch o Nash–S ampacchia equilib ia poin s using Hadama d mani olds
has been used by K is ály [12].
Symme y 2019,11, 1037 4 o 12
I is known ha a con ex en i onmen has good p ope ies o he sea ch o op imal poin s.
In Fe ei a [
13
], he au ho gi es necessa y and su icien condi ions o con ex unc ions on Hadama d
mani olds. A signi ican gene aliza ion o he con ex unc ions a e he in ex unc ions, in oduced
by Hanson [
14
], whe e he x-y ec o is eplaced by any unc ion
η(x
,
y)
. The main esul o
in ex unc ions s a es ha a scala unc ion is in ex i and only i e e y c i ical poin is a global
minimum solu ion. This p ope y is essen ial o ob ain op imal poin s h ough algo i hms, due o
he coincidence o c i ical poin s and solu ions being always assu ed. In Ba ani and Pou yayeli [
15
]
and Hosseini and Pou yaye ali [
16
], he ela ion be ween in exi y and mono onici y using he mean
alue heo em is s udied. Ruiz-Ga zón e al. [
17
] showed ha in exi y can be cha ac e ized in he
con ex o Riemannian mani olds o bo h scala and ec o cases, in a simila way o Euclidean spaces.
Recen ly, in Ahmad e al.
[18]
he au ho s in oduced he log-p ein ex and log-in ex unc ions on
Riemannian mani olds and he mean alue heo em on Ca an-Hadama d mani olds.
In he same way, se e al au ho s ha e s udied ec o equilib ium p oblems. Ansa i and
Flo es-Bazán [
19
] we e capable o p o iding a heo em o exis ence o solu ions o ec o
quasi-equilib ium p oblems. Fu he mo e, a cha ac e iza ion o a weakly e icien Pa e o poin o he
ec o equilib ium p oblems wi h cons ain s unde con exi y condi ions on eal Hausdo opological
ec o spaces we e p esen ed by Gong [
20
]. In he ollowing yea s, scala iza ion esul s o he solu ions
o he ec o equilib ium p oblems we e also gi en by Gong [
21
]. La e , op imali y condi ions o
weakly e icien Pa e o poin s o ec o equilib ium p oblems wi h cons ain s in eal no med spaces
we e in es iga ed by Wei and Gong [
22
]. Also, su icien condi ions o weakly e icien Pa e o poin s
on eal Banach spaces o ec o equilib ium and ec o op imiza ion p oblems wi h cons ain s unde
gene alized in exi y we e ob ained by Feng and Qiu [23].
Mo i a ed by Gong’s wo ks men ioned abo e, ou objec i e will ocus on ex ending he KKT
necessa y and su icien condi ions o cons ained ec o equilib ium p oblems ob ained in opological
o no med spaces o o he en i onmen s like he Hadama d mani olds, no p esen in he li e a u e up
o da e o publica ion. Hence, we p opose a gene aliza ion ha ex ends he linea space de ini ion o
Hadama d mani olds, by i ue o subs i u ing line segmen s by geodesic a cs. We will see ha he
KKT classic condi ions o cons ained ec o op imiza ion a e a pa icula case o he ones ob ained
o cons ained ec o equilib ium p oblem.
The o ganiza ion o he pape is as ollows: In Sec ion 2, we discuss no a ion, di e en ials and
in ex unc ion concep s on Hadama d mani olds. Sec ion 3is de o ed o p o ing he main esul s
ob ained in his pape , and s udying he necessa y and su icien op imali y condi ions o weakly
e icien poin s o he cons ained ec o equilib ium p oblem. Sec ion 4dwells on how he p e ious
esul s can be educed o classical KKT condi ions o cons ained ec o op imiza ion p oblems,
i s ob ained by William Ka ush [
24
] and edisco e ed by Ha old Kuhn and Albe Tucke [
25
].
Finally, an example is p esen ed as well as he inal conclusions.
2. P elimina ies
Le
M
be a
C∞
-mani old modeled on a Hilbe space
H
endowed wi h a Riemannian me ic
gx
on
a angen space
TxM
. We deno e by
TxM
he angen space o
M
a
x
, by
TM =Sx∈MTxM
he angen
bundle o
M
, by
¯
TM
an open neighbo hood o he submani old
M
o
TM
. The co esponding no m is
deno ed by k.kxand he leng h o a piecewise C1cu e α:[a,b]→Mis de ined by
L(α) = Zb
akα0( )kα( )d .
We de ine das he dis ance which induces he o iginal opology on Msuch ha
d(x,y) = in {L(α)|αis a piecewise C1cu e joining xand y∀x,y∈M}.
Symme y 2019,11, 1037 5 o 12
I
d
is he dis ance induced by he Riemannian me ic
kij
hen any Riemannian mani old
(M
,
kij)
can
be con e ed in o a me ic space
(M
,
d)
. The de i a i es o he cu es a a poin
x
on he mani old lies
in a ec o space TxM. Wha e e pa h αjoining xand yin Msuch ha L(α) = d(x,y)is a geodesic.
Le
exp : ¯
TM →M
be he Riemannian exponen ial map de ined as
expx(V) = αV(
1
)
o e e y
V∈¯
TM, whe e αVis he geodesic s a ing a xwi h eloci y V(i.e., α(0) = x,α0(0) = V).
Assume now ha ηis a map η:M×M→TM de ined on he p oduc mani old such ha
η(x,y)∈TyM,∀x,y∈M.
De ini ion 1.
[
26
] A subse
S1
o
M
is conside ed o ally con ex i
S1
con ains e e y geodesic
αx,y
o
M
whose
endpoin s x and y belong o S1.
On a Hadama d mani old
M
, we can de ine he unc ion
η
as
η(x
,
y) = α0
x,y(
0
)
o all
x
,
y∈M
.
This unc ion plays he same ole o
x−y∈Rn
. He e
αx,y
is he unique minimal geodesic joining
y
o
xas ollows
αx,y=expy(λexp−1
yx)∀λ∈[0, 1].
Example 1.
Le
M=R++ ={y∈R:y>
0
}
endowed wi h he Riemannian me ic de ined by
g(y) = y−2
be a Hadama d mani old. Hype bolic spaces and geodesic spaces, mo e p ecisely, a Busemann non-posi i e
cu a u e (NPC) space a e examples o Hada ma d mani olds.
We will need an adequa e concep o he di e en ial:
De ini ion 2.
[
27
] A mapping
i:M→R
is said o be a di e en ial map along he geodesic
αx,y
a
y∈M
i
and only i he limi
0
i(y) = lim
λ→0
i(expy(λη(x,y))) − i(y)
λkη(x,y)k
exis s.
The g adien o a eal- alued
C∞
unc ion
= ( 1
,
. . . p):S1⊆M→Rn
on
M
in
x
, deno ed by
g ad x= ( 0
1(x)
,
0
2(x)
,
. . .
,
0
n(x))
, is he unique ec o in
TxM
such ha
d x(X) = hg ad x
,
Xi
o all
X
in
TxM is he di e en ial o a ¯
x o X.
Rema k 1.
The di e en ial o
a
¯
x
o
X
is simila o he de ini ion o di ec ional de i a i e in he
Euclidean space.
Le
S1⊂M
be a nonemp y open o ally con ex subse and le
F:S1×S1→Rp
,
g:S1→Rp
be mappings.
De ini ion 3.
We de ine he cons ain se
S={x∈S1:g(x)∈ −Rp
+}
and conside he ec o equilib ium
p oblem wi h cons ain s (VEPC): ind x ∈S such ha
F(x,y)/∈ −Rp
+ {0},∀y∈S
whe e Rp
+is he non-nega i e o han o Rp.
We ecall he classical concep :
De ini ion 4.
A ec o
x∈S
sa is ying
F(x
,
y)/∈ −in Rp
+
,
∀y∈S
is called a weakly e icien Pa e o poin
o he VEPC.
No a ion 1. We deno e as Hx(y) = F(x,y),∀y∈S1,gi en x ∈S, whe e H :S1→Rpis a mapping.

Symme y 2019,11, 1037 6 o 12
Inspi ed by he concep o con exi y on a linea space, he no ion o in exi y unc ion concep on
Hadama d mani olds has become a success ul ool in ec o op imiza ion. This gene alized de ini ion
was no ably p o ided by Hanson in [14].
De ini ion 5.
Le
S1
be a nonemp y open o ally con ex subse o a Hadama d mani old
M
. A di e en iable
h:S1→Rp
unc ion is said o be a
Rp
+
-in ex a
¯
x∈S1
espec o
η:M×M→TM
i he e exis
η(x,¯
x)∈T¯
xM such ha
h(x)−h(¯
x)−dh ¯
x(η(x,¯
x)) ∈Rp
+.
Using he p e iously s a ed de ini ions, we can ob ain he su icien condi ions o op imali y by
i ue o he assump ion o in exi y o he unc ions o he p oblem.
3. Main Resul s
Nex , we will ob ain a cha ac e iza ion o he weakly e icien poin s o VEPC h ough he
applica ion o necessa y and su icien op imali y condi ions. We s a wi h he necessa y condi ions:
Theo em 1.
[Necessa y KKT-condi ions] Le
S1
be a nonemp y open o ally con ex subse o a Hadama d
mani old
M
and le
F:S1×S1→Rp
,
g:S1→Rp
,
η:M×M→TM
be mappings. Le
F(¯
x
,
¯
x) =
H¯
x(¯
x) =
0. Assume ha
H
and
g
a e di e en iable a
¯
x∈S
. Fu he mo e, assume ha he e exis s
x1∈S1
such ha
g(¯
x) + dg¯
x(η(x1
,
¯
x)) ∈ −in Rp
+
. I
¯
x
is a weakly e icien Pa e o poin o he VEPC, hen he e
exis s ∈Rp
+ {0}, u ∈Rp
+such ha
dH¯
x(η(x,¯
x)) + udg ¯
x(η(x,¯
x)) ≥0, ∀x∈S1(1)
ug(¯
x) = 0. (2)
P oo . Le he e be ¯
x∈Sas a weakly e icien Pa e o poin o he VEPC. We deno e by
W={(y,z)∈Rp×Rp: he e exis s x∈S1, such ha y−dH¯
x(η(x,¯
x)) ∈in Rp
+,
z−[g(¯
x) + dg¯
x(η(x,¯
x))] ∈in Rp
+}.
I may be no ed ha W is a nonemp y open o ally con ex se . This p oo can be di ided in o
i e s eps:
S ep 1.
We ha e o p o e ha
(
0, 0
)/∈W
. By educ ion ad absu dum, i
(
0, 0
)∈W⇒ ∃x0∈S1
,
such ha
dH¯
x(η(x0,¯
x)) ∈ −in Rp
+,g(¯
x) + dg¯
x(η(x0,¯
x)) ∈ −in Rp
+. (3)
F om he di e en iabili y we ob ain ha
dH¯
x(η(x0,¯
x)) = lim
λ→0
1
λ[H¯
x(exp ¯
x(λη(x0,¯
x)) −H¯
x(¯
x)] ∈ −in Rp
+(4)
g(¯
x) + dg¯
x(η(x0,¯
x)) = g(¯
x) + lim
λ→0
1
λ[g(exp ¯
x(λη(x0,¯
x))) −g(¯
x)] ∈ −in Rp
+. (5)
As −in Rp
+is an open se , hen ∃λ0, 0 <λ0<1 such ha
1
λ0
[H¯
x(exp ¯
x(λ0η(x0,¯
x))) −H¯
x(¯
x)] ∈ −in Rp
+(6)
g(¯
x) + 1
λ0
[g(exp ¯
x(λ0η(x0,¯
x))) −g(¯
x)] ∈ −in Rp
+. (7)
Symme y 2019,11, 1037 7 o 12
By hypo hesis, om g(¯
x)∈ −Rp
+,F(¯
x,¯
x) = H¯
x(¯
x) = 0, and 1
λ0
>1, hen
H¯
x[exp ¯
x(λ0η(x0,¯
x))] ∈ −in Rp
+and g(exp ¯
x(λ0η(x0,¯
x)))∈ −in Rp
+. (8)
As S1is a o ally con ex se we ha e ha
exp ¯
x(λ0η(x0,¯
x)) ∈S1,F(¯
x,exp ¯
x(λ0η(x0,¯
x)))∈ −in Rp
+(9)
and
g(exp ¯
x(λ0η(x0,¯
x)))∈ −in Rp
+(10)
s ands in con adic ion wi h
¯
x∈S
as a weakly e icien Pa e o poin o he VEPC, consequen ly
(0, 0)/∈W.
S ep 2.
We will p o e ha he e exis s a mul iplie
∈Rp
+
. As
W
is an open se and he sepa a ion
heo em holds (see Theo em 2.13 and Rema k 2.14 in [
28
]) o [
3
]), he e exis s
(
,
u)6= (
0, 0
)∈Rp×Rp
such ha
y +uz >0, ∀(y,z)∈W. (11)
Le (y,z)∈Wbe a poin hen ∃x∈S1such ha
y−dH¯
x(η(x,¯
x)) ∈in Rp
+,z−[g(¯
x) + dg¯
x(η(x,¯
x))] ∈in Rp
+. (12)
Fo any ∈in Rp
+,s∈in Rp
+, 0, 00 >0, we ha e (y+ 0 ,z)∈Wand (y,z+ 00s)∈W.
F om Equa ion (11) we ha e ha
(y+ 0 ) + u(z)>0, ∀ ∈in Rp
+, 0>0. (13)
Then
>−uz − y
0. (14)
Le ing
0→∞
we ge
≥
0,
∀ ∈in Rp
+
and he e o e
≥
0 o all
∈Rp
+
, ha is
∈Rp
+
.
In he same way, we can show ha u∈Rp
+.
S ep 3.
We will p o e ha
6=
0, hus is,
∈Rp
+ {
0
}
. By educ ion ad absu dum, i
=
0,
om Equa ion (11) we ge
uz >0, ∀(y,z)∈W. (15)
Acco ding o he hypo hesis,
∃x1∈S1
such ha
g(¯
x) + dg¯
x(η(x1
,
¯
x)) ∈ −in Rp
+
; hen, we ob ain
(dH¯
x(η(x1,¯
x)) + ,g(¯
x) + dg¯
x(η(x1,¯
x)) + s)∈W,∀ ∈in Rp
+∀s∈in Rp
+. (16)
The e o e, om Equa ion (11) we ha e ha
u[g(¯
x) + dg¯
x(η(x1,¯
x)) + s]>0, ∀s∈in Rp
+(17)
us >−u[g(¯
x) + dg¯
x(η(x1,¯
x))]. (18)
As
[g(¯
x) + dg¯
x(η(x1
,
¯
x))] ∈ −in Rp
+
, and i
s=
0, we ge
u·
0
=
0
>
0, which implies a
con adic ion, hus 6=0.
S ep 4. We will p o e he i s KKT condi ion.
Since
(dH¯
x(η(x,¯
x)) + ,g(¯
x) + dg¯
x(η(x,¯
x)) + s)∈W,x∈S1, ∈in Rp
+,s∈in Rp
+. (19)
Symme y 2019,11, 1037 8 o 12
F om Equa ion (11) we ge
[dH¯
x(η(x,¯
x)) + ] + u[g(¯
x) + dg¯
x(η(x,¯
x)) + s]>0, ∀x∈S1, ∈in Rp
+,s∈in Rp
+. (20)
Le ing →0, s→0, we ob ain
dH¯
x(η(x,¯
x)) + u[g(¯
x) + dg¯
x(η(x,¯
x))] ≥0, ∀x∈S1. (21)
S ep 5. We will p o e he second KKT condi ion. As
dH¯
x(η(¯
x,¯
x)) + 0 ,g(¯
x) + dg¯
x(η(¯
x,¯
x)) + 0s∈W,∀ ∈in Rp
+,s∈in Rp
+, 0>0. (22)
F om Equa ion (11) we ha e ha
[dH¯
x(η(¯
x,¯
x)) + 0 ] + u[g(¯
x) + dg¯
x(η(¯
x,¯
x)) + 0s] = 0 +ug(¯
x) + 0us >0. (23)
Le ing
0→
0, we ob ain
ug(¯
x)≥
0. No ing ha
g(¯
x)∈ −Rp
+
and
u∈Rp
+
, we ha e ha
ug(¯
x)≤0, in consequence
ug(¯
x) = 0 (24)
and he e o e ∃ ∈Rp
+ {0},u∈Rp
+such ha KKT condi ions
dH¯
x(η(x,¯
x)) + udg ¯
x(η(x,¯
x)) ≥0, ∀x∈S1(25)
ug(¯
x) = 0 (26)
hold.
Le us see now he ecip ocal o he p e ious heo em. To ob ain i we i s need condi ions
o in exi y.
Theo em 2.
[Su icien KKT-condi ions] Le
S1
be a nonemp y open o ally con ex subse o Hadama d mani old
M
and le
F:S1×S1→Rp
,
g:S1→Rp
be mappings. Le
F(¯
x
,
¯
x) = H(¯
x) =
0. Assume ha
H
and
g
a e
di e en iable a
¯
x∈S
.
H
and
g
a e
Rp
+
-in ex a
¯
x
espec o
η
on
S1
. I he e exis
∈Rp
+ {
0
}
and
u∈Rp
+
such ha
dH¯
x(η(x,¯
x)) + udg ¯
x(η(x,¯
x)) ≥0, ∀x∈S1(27)
ug(¯
x) = 0 (28)
hen ¯
x is a weakly e icien Pa e o poin o he VEPC.
P oo . On he assump ion ha Hand ga e Rp
+-in ex a ¯
x espec o ηon S1 hen
dH¯
x(η(x,¯
x)) ∈H¯
x(x)−H¯
x(¯
x)−Rp
+=H¯
x(x)−Rp
+,∀x∈S1(29)
dg¯
x(η(x,¯
x)) ∈g(x)−g(¯
x)−Rp
+,∀x∈S1. (30)
F om ∈Rp
+ {0},u∈Rp
+and (27) we ob ain ha
H¯
x(x) + u(g(x)−g(¯
x)) = dH¯
x(η(x,¯
x)) + udg ¯
x(η(x,¯
x)) ≥0, ∀x∈S1. (31)
F om hypo hesis (28), we ge on he one hand ha :
H¯
x(x) + ug(x)≥0, ∀x∈S1. (32)
Symme y 2019,11, 1037 9 o 12
On he o he hand, we will show ha
¯
x
is a weakly e icien Pa e o poin o he VEPC. I no ,
consequen ly by de ini ion ∃y0∈Ssuch ha
F(¯
x,y0)∈ −in Rp
+. (33)
F om ∈Rp
+ {0} ⇒ F(¯
x,y0)<0.
Since y0∈S, we ha e g(y0)∈ −Rp
+, so ug(y0)≤0 because o u∈Rp
+and hen
F(¯
x,y0) + ug(y0)<0 (34)
s ands in con adic ion wi h (32) and he e o e ¯
xis a weakly e icien Pa e o poin o he VEPC.
Rema k 2.
Theo em 3.1 in [
20
] on eal Hausdo opological ec o spaces and Theo em 3.2 and Theo em 3.4
in [
22
] on eal no med spaces a e pa icula cases o Theo ems 1and 2ob ained in his pape on Hadama d
mani olds. The same is ue o Theo ems 3.1 and 3.3 in [23] on eal Banach spaces.
To sum up, we ob ain he KKT op imali y condi ions o weakly e icien Pa e o poin s o he
ec o equilib ium p oblems wi h cons ain s. These esul s a e no only necessa y bu also su icien .
4. Applica ion
As a pa icula case o he esul s ob ained in he p e ious sec ion, we will ob ain he op imali y
condi ions o KKT o cons ained ec o op imiza ion p oblems.
Le us conside he cons ained mul iobjec i e p og amming (CVOP) de ined as:
(CVOP) min (x)
subjec o:
g(x)≤0
x∈X⊆M
whe e
= ( 1
,
. . . p):X⊆M→Rp
,
g= (g1
,
. . .
,
gm):X⊆M→Rm
a e di e en iable
mul iobjec i e unc ions on he open se X⊆Mand le Mbe a Hadama d mani old.
As a consequence o he p e ious heo ems and conside ing CVOP as a pa icula case o VEPC
we ha e he KKT classical condi ions.
Co olla y 1.
Le
S1
be a nonemp y open o ally con ex subse o Hadama d mani old
M
and le
,
g:S1→Rp
be mappings. Assume ha
and
g
a e di e en iable a
¯
x∈S
. Fu he mo e, assume ha he e exis s
x1∈S1
such ha
g(¯
x) + dg¯
x(η(x1
,
¯
x)) ∈ −in Rp
+
. I
¯
x
is a weakly e icien Pa e o poin o he CVOP, hen he e exis
∈Rp
+ {0}, u ∈Rp
+such ha
d ¯
x(η(x,¯
x)) + udg ¯
x(η(x,¯
x)) ≥0, ∀x∈S1(35)
ug(¯
x) = 0. (36)
Co olla y 2.
Le
S1
be a nonemp y open o ally con ex subse o Hadama d mani old
M
and le
,
g:S1→Rp
be mappings. Assume ha
and
g
a e di e en iable a
¯
x∈S
. Assume ha
and
g
a e di e en iable a
¯
x∈S
and and g a e Rp
+-in ex espec a ¯
x o ηon S1. I he e exis ∈Rp
+ {0}, u ∈Rp
+such ha
d ¯
x(η(x,¯
x)) + udg ¯
x(η(x,¯
x)) ≥0, ∀x∈S1(37)
ug(¯
x) = 0 (38)
hen ¯
x is a weakly e icien Pa e o poin o he CVOP.