# Using ρ-cone arcwise connectedness on parametric set-valued optimization problems – Journal of Inequalities and Applications

#### ByKoushik Das and Mohammad Esmael Samei

May 8, 2022 ### The ρ–(mathbb{C})A(mathbb{C})

We introduce the notion of ρ(mathbb{C})A(mathbb{C}) of set-valued maps as a generalization of (mathbb{C})A(mathbb{CS})(mathbb{VM}) which was introduced by Das et al.  and Treanţă et al. . For (rho =0), we have the usual notion of cone convex set-valued maps introduced by Borwein .

### Definition 3.1

Let U and V be real normed spaces, (Wsubseteq U) be an arcwise connected, (mathrm{u}_{1}, mathrm{u}_{2}in W), (ein operatorname{int}(Omega )), and (mathfrak{F}: U to 2^{V}) be a set-valued map with (W subseteq operatorname{dom}(mathfrak{F})). Then (mathfrak{F}) is said to be ρ-Ω-arcwise connected (ρ-Ω-A(mathbb{C})) with respect to e on W for (mathrm{u}_{1}), (mathrm{u}_{2}) if there exists (rho in mathbb{R}) such that

$$(1 – upeta ) mathfrak{F}(mathrm{u}_{1}) + mathrm{u} mathcal{F}( mathrm{u}_{2}) subseteq mathrm{u}bigl( mathcal{H}_{mathrm{u}_{1}, mathrm{u}_{2}} (upeta )bigr) + rho upeta ( 1 – upeta ) Vert mathrm{u}_{1} – mathrm{u}_{2} Vert ^{2} e + Omega ,$$

(2)

(forall upeta in [0, 1]).

### Remark 3.1

If (rho > 0), then (mathfrak{F}) is said to be strongly ρ-Ω-arcwise connected ((mathbb{S}rho )-Ω-A(mathbb{C})); if (rho = 0), we have the usual notion of Ω-arcwise connectedness; and if (rho <0), then (mathfrak{F}) is said to be weakly ρ-Ω-arcwise connected ((mathbb{W}rho )-Ω-A(mathbb{C})). Obviously, (mathbb{S}rho )-Ω-A(mathbb{C}) Ω-arcwise connectedness (mathbb{W}rho )-Ω-A(mathbb{C}).

Further, we construct an example of ρ(mathbb{C})A(mathbb{CS})(mathbb{VM}), which is not cone arcwise connected.

### Example 3.1

Let (U =mathbb{R}^{2}), (V = mathbb{R}), (Omega = mathbb{R}_{+}), and

$$W = biggllbrace mathrm{u} = (mathrm{u}_{1}, mathrm{u}_{2}) : mathrm{u}_{1} + mathrm{u}_{2} geq frac{1}{4}, mathrm{u}_{1} geq 0, mathrm{u}_{2}geq 0 biggrrbrace subseteq U.$$

Define

$$mathcal{H}_{mathrm{u},acute{mathrm{u}}}(upeta ) = bigl(1 – upeta ^{2}bigr) upeta + upeta ^{2} acute{mathrm{u}},$$

where (mathrm{u}=(mathrm{u}_{1},mathrm{u}_{2})), (acute{mathrm{u}}=( acute{mathrm{u}}_{1}, acute{mathrm{u}}_{2})), and (upeta in [0,1]). Clearly, W is an arcwise connected set. Define a set-valued map (mathfrak{F} : mathbb{R}^{2} to 2^{mathbb{R}}) as follows:

$$mathfrak{F}(mathrm{u}) = textstylebegin{cases} [0,4], & mathrm{u}_{1} + mathrm{u}_{2} geq frac{1}{2}, mathrm{u}_{1}neq 3 mathrm{u}_{2}, mathrm{u}= ( mathrm{u}_{1}, mathrm{u}_{2}), \ [5,9],& text{otherwise}.end{cases}$$

We choose (mathrm{u}=(1,0)), (acute{mathrm{u}}=(0,1)), and (upeta = frac{1}{2}). Then

$$mathcal{H}_{mathrm{u},acute{mathrm{u}}} biggl( frac{1}{2} biggr) = biggl( frac{3}{4}, frac{1}{4} biggr)$$

and

begin{aligned} frac{1}{2} mathfrak{F}(1, 0)+frac{1}{2} mathfrak{F}(0, 1) & = frac{1}{2} [0,4] + frac{1}{2} [0,4] \ & = [0,4] nsubseteq [5,9] +mathbb{R}_{+} \ & = mathfrak{F} biggl( frac{3}{4}, frac{1}{4} biggr) + mathbb{R}_{+}. end{aligned}

Hence (mathfrak{F}) is not (mathbb{R}_{+})-arcwise connected. On the other hand, by considering (rho = -2) and (e = 5), we get that

$$(1- upeta ) mathfrak{F} (1, 0)+upeta mathfrak{F}(0, 1) = (1- upeta ) [0,4] + upeta [0,4] = [0,4]$$

and

$$mathfrak{F}bigl(mathcal{H}_{mathrm{u}, acute{mathrm{u}}} (upeta )bigr) + rho upeta ( 1 – upeta ) Vert mathrm{u} – acute{mathrm{u}} Vert ^{2} e = mathfrak{F} bigl( 1 – upeta ^{2}, upeta ^{2}bigr)- 20 upeta ( 1 – upeta ).$$

For (upeta neq 0.5), we have

$$mathfrak{F} bigl(1 – upeta ^{2}, upeta ^{2}bigr) = [0,4].$$

So,

begin{aligned} ( 1 – upeta ) mathfrak{F}(1, 0)+upeta mathfrak{F}(0, 1) + 20 upeta ( 1 – upeta ) & = [0,4] + 20 upeta ( 1 – upeta ) \ & subseteq [0,4] + mathbb{R}_{+} = mathbb{R}_{+}. end{aligned}

For (upeta = frac{1}{2}), we have

$$mathfrak{F} bigl(1-upeta ^{2}, upeta ^{2}bigr) = mathfrak{F} biggl( frac{3}{4}, frac{1}{4} biggr)= [3,5].$$

So,

begin{aligned} ( 1 – upeta ) mathfrak{F}(1, 0)+upeta mathfrak{F}(0, 1) + 20 upeta (1 – upeta ) & = [0,4] + 5 \ & = [5,9] subseteq [5,9]+ mathbb{R}_{+}. end{aligned}

Consequently, (mathfrak{F}) is ((-2))(mathbb{R}_{+})-A(mathbb{CS})(mathbb{VM}) with respect to 5 on W for ((1, 0)), ((0, 1)).

### Theorem 3.2

Let U, V be real normed spaces, (Wsubseteq U) be arcwise connected, (ein operatorname{int}(Omega )), and (mathfrak{F}: U to 2^{V}) be ρ-Ω-A(mathbb{C}) with respect to e on W. Let (acute{mathrm{u}}in W) and (acute{mathrm{v}}in mathfrak{F}(acute{mathrm{u}})). Then

$\mathfrak{F}\left(\mathrm{u}\right)-\stackrel{}{}$
v
´

D

F(
u
´
,
v
´
)
(

H
´

u
´

,
u

(
0
+
)
)
+ρ

u

u
´

2
e+Ω,uW,

where

$$acute{mathcal{H}}_{acute{mathrm{u}}, mathrm{u}}(0+) = lim_{ upeta to 0+} frac{ mathcal{H}_{acute{mathrm{u}}, mathrm{u}}( upeta ) – mathcal{H}_{acute{mathrm{u}}, mathrm{u}}(0)}{upeta},$$

assuming that (acute{mathcal{H}}_{acute{mathrm{u}}, mathrm{u}}(0+)) exists for all (mathrm{u}, acute{mathrm{u}}in W).

### Proof

Let (mathrm{u}in W). As (mathfrak{F}) is ρ-Ω-A(mathbb{C}) with respect to e on W, we have

$$(1-upeta ) mathfrak{F}(acute{mathrm{u}}) + upeta mathfrak{F}( mathrm{u}) subseteq mathfrak{F}bigl( mathcal{H}_{acute{mathrm{u}}, mathrm{u}}( upeta )bigr) + rho upeta (1 – upeta ) Vert mathrm{u} – acute{mathrm{u}} Vert ^{2} e + Omega ,$$

(forall upeta in [0, 1]). Let (mathrm{v}in mathfrak{F}(mathrm{u})). Consider a real sequence ({upeta _{n}}) with (upeta _{n} in (0, 1)), (n in mathbb{N}), such that (upeta _{n}to 0+) when (nto infty ). Suppose (mathrm{u}_{n} = mathcal{H}_{acute{mathrm{u}}, mathrm{u}} ( upeta _{n})) and

$$mathrm{v}_{n} = (1 – upeta _{n}) acute{mathrm{v}} + upeta _{n} mathrm{v} – rho upeta _{n} ( 1 – upeta _{n}) Vert mathrm{u} – acute{mathrm{u}} Vert ^{2} e.$$

Therefore, (mathrm{v}_{n}in mathfrak{F}( mathrm{u}_{n}) + Omega ). It is clear that

$${mathrm{u}_{n} = mathcal{H}_{acute{mathrm{u}}, mathrm{u}} ( mathrm{u}_{n}) to mathcal{H}_{acute{mathrm{u}}, mathrm{u}} (0) = acute{ mathrm{u}},}$$

(mathrm{v}_{n}to acute{mathrm{v}}), when n tends to ∞,

$$frac{mathrm{u}_{n} – acute{mathrm{u}}}{ upeta _{n}} = frac{ mathcal{H}_{acute{mathrm{u}}, mathrm{u}} (upeta _{n}) – mathcal{H}_{acute{mathrm{u}}, mathrm{u}} (0)}{ upeta _{n}} to acute{mathcal{H}}_{acute{mathrm{u}}, mathrm{u}}(0+),$$

when n tends to ∞, and

$$frac{mathrm{v}_{n} – acute{mathrm{v}}}{ upeta _{n}} = mathrm{v} – acute{mathrm{v}} – rho ( 1 -upeta _{n}) Vert mathrm{u} – acute{mathrm{u}} Vert ^{2} e to mathrm{v} – acute{mathrm{v}} – rho Vert mathrm{u} -acute{mathrm{u}} Vert ^{2} e,$$

when (nto infty ). Therefore,


(

H
´

u
´

,
u

(
0
+
)
,
v

v
´

ρ

u
´

u
´

2

e
)
T
(
epi
(
F
)
,
(

u
´

,

v
´

)
)
=epi
(

D

F
(

u
´

,

v
´

)
)
.

Consequently,

$\mathrm{v}-\stackrel{}{}$
v
´
ρ

u
´

u
´

2
e
D

F(
u
´
,
v
´
)
(

H
´

u
´

,
u

(
0
+
)
)
+Ω,

which is true for all (mathrm{v}in mathfrak{F}(mathrm{u})). Hence,

$\mathfrak{F}\left(\mathrm{u}\right)-\stackrel{}{}$
v
´

D

F(
u
´
,
v
´
)
(

H
´

u
´

,
u

(
0
+
)
)
+ρ

u

u
´

2
e+Ω,uW.

Hence the theorem follows. □

### Formulation of the main problem

Let U, (V_{1}), (V_{2}), and (V_{3}) be real normed spaces and (Omega _{1}), (Omega _{2}), and (Omega _{3}) be solid pointed convex cones in (V_{1}), (V_{2}), and (V_{3}), respectively. Let A be an arbitrary set and W be a nonempty subset of U. Suppose that

$$mathfrak{F} : Utimes A to 2^{V_{1}}, qquad mathfrak{G} : U times A to 2^{V_{2}}$$

are set-valued maps and (mathrm{p} : Utimes A to V_{3}) is a single-valued map with

$$W times A subseteq operatorname{dom}( mathfrak{F} ) cap operatorname{dom}( mathfrak{G}).$$

We consider a parametric (mathbb{S})(mathbb{VPOP}s) (1), where u is the state variable and a is the parameter. The feasible set Š of problem (1) is defined by

$$check{S} = bigl{ ( mathrm{u}, mathrm{a}) in Wtimes A : mathfrak{G}( mathrm{u}, mathrm{a})cap (-Omega _{2}) neq emptyset text{ and } mathrm{p}(mathrm{u}, mathrm{a})=0 bigr} .$$

The minimizer and weak minimizer of problem (1) are defined in the following ways. A point ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1})in U times A times V_{1}), with ((acute{mathrm{u}}, acute{mathrm{a}}) in check{S}) and (acute{mathrm{v}}_{1} in mathfrak{F}(acute{mathrm{u}}, acute{mathrm{a}})), is called a minimizer of problem (1) if there exists no point ((mathrm{u}, mathrm{a}, mathrm{v}_{1}) in Utimes A times V_{1}), with ((mathrm{u}, mathrm{a}) in check{S}) and (mathrm{v}_{1}in mathfrak{F}(mathrm{u}, mathrm{a})), such that

$$mathrm{v}_{1} – acute{mathrm{v}}_{1} in – Omega _{1} setminus {theta _{V_{1}}},$$

and is called a weak minimizer of problem (1) if there exists no point

$$(mathrm{u}, mathrm{a}, mathrm{v}) in Utimes A times V_{1},$$

with ((mathrm{u}, mathrm{a}) in check{S}) and (mathrm{v}_{1} in mathfrak{F}(mathrm{u}, mathrm{a})), such that (mathrm{v}_{1} – acute{mathrm{v}}_{1} in – operatorname{int}(V_{1})).

### Sufficient optimality conditions

Let

$$(mathrm{u}, mathrm{a}),(acute{mathrm{u}}, mathrm{a}),( mathrm{u}, acute{ mathrm{a}}),(acute{mathrm{u}}, acute{mathrm{a}}) in Utimes A,$$

(acute{mathrm{v}}_{1} in mathfrak{F}(acute{mathrm{u}}, acute{mathrm{a}})), and (acute{mathrm{v}}_{2} in mathfrak{G}(acute{mathrm{u}}, acute{mathrm{a}})). Throughout the paper, we use the following assumptions:

begin{aligned}& mathfrak{F}(mathrm{u}, mathrm{a}) – mathfrak{F}( acute{ mathrm{u}}, mathrm{a}) subseteq Omega _{1}, \& mathfrak{G}(mathrm{u}, mathrm{a}) – mathfrak{G}( acute{mathrm{u}}, mathrm{a}) subseteq Omega _{2}, \& mathfrak{F}(mathrm{u}, acute{mathrm{a}}) – acute{mathrm{v}}_{1} subseteq -Omega _{1}, \& mathfrak{G}(mathrm{u}, acute{mathrm{a}}) – acute{mathrm{v}}_{2} subseteq -Omega _{1}, \& mathrm{p}(mathrm{u}, acute{mathrm{a}}) + mathrm{p}( acute{mathrm{u}}, mathrm{a}) in -Omega _{3}. end{aligned}

(3)

We now prove the following lemma which assists in establishing the sufficient KKT optimality conditions of the parametric (mathbb{S})(mathbb{VPOP}s) (1).

### Lemma 3.3

Let W be an arcwise connected subset of U and ((acute{mathrm{u}}, acute{mathrm{a}})in Utimes A) with (acute{mathrm{v}}_{1} in mathfrak{F}(acute{mathrm{u}}, acute{mathrm{a}})), (acute{mathrm{v}}_{2} in mathfrak{G}(acute{mathrm{u}}, acute{mathrm{a}})), and (mathrm{p}(acute{mathrm{u}}, acute{mathrm{a}})geq 0). Let (ein operatorname{int}(Omega _{1})), (e^{prime}in operatorname{int}(Omega _{2})), and (e^{prime prime}in operatorname{int}(Omega _{3})). Suppose that (mathfrak{F}(cdot, acute{mathrm{a}}) : U to 2^{V_{1}}) is (rho _{1})-Ω-A(mathbb{C}) with respect to e, (mathfrak{G}(cdot, acute{mathrm{a}}) : Uto 2^{V_{2}}) is (rho _{2})-Ω-A(mathbb{C}) with respect to (e^{prime}), and (mathrm{p}(cdot, acute{mathrm{a}}) : U to V_{3}) is (rho _{3})-Ω-A(mathbb{C}) with respect to (e^{prime prime}) on W. Assume that the contingent epiderivatives

${}_{}$
D

F(,
a
´
)(
u
´
,

v
´

1
)

and

${}_{}$
D

G(,
a
´
)(
u
´
,

v
´

2
)

exist and the Gâteaux derivative (mathrm{p}^{prime}(cdot, acute{mathrm{a}})(acute{mathrm{u}})) exists. If equations in (3) are satisfied, then we have

$\begin{array}{}\end{array}$

v
1

,
F
(
u
,
a
)

v
´

1

+

v
2

,
G
(
u
,
a
)

v
´

2

v
1

,

D

F
(

,

a
´

)
(

u
´

,

v
´

1

)

(

H
´

u
´

,
a

(
0
+
)
)

+
F
(

u
´

,
a
)

v
´

1

+

v
2

,

D

G
(

,

a
´

)
(

u
´

,

v
´

2

)

(

H
´

u
´

,
u

(
0
+
)
)

+
G
(

u
´

,
a
)

v
´

2

+

v
3

,

p

(

,

a
´

)
(

u
´

)

(

H
´

u
´

,
u

(
0
+
)
)

+
p
(

u
´

,
a
)

+

u

u
´

2

(

ρ
1

v
1

,
e

+

ρ
2

v
2

,

e

+

ρ
3

v
3

,

e

)

,

(4)

(forall (mathrm{u}, mathrm{A})in Wtimes A).

### Proof

Let ((mathrm{u}, mathrm{a})in Wtimes A). As (mathfrak{F}(cdot, acute{mathrm{a}}) : U to 2^{V_{1}}) is (rho _{1})-Ω-A(mathbb{C}) with respect to e on W and (acute{mathrm{v}}_{1}in mathfrak{F}(acute{mathrm{u}}, acute{mathrm{a}})), we have

$\mathfrak{F}\left(\mathrm{u},\stackrel{}{}$
a
´
)

v
´

1

D

F(,
u
´
)(
u
´
,

v
´

1
)
(

H
´

u
´

,
u

(
0
+
)
)
+
ρ
1

u

u
´

2
e+
Ω
1
.

(5)

As (mathfrak{G}(cdot, acute{mathrm{a}}) : U to 2^{V_{2}}) is (rho _{2})-Ω-A(mathbb{C}) with respect to (e^{prime}) on W and (acute{mathrm{v}}_{2}in mathfrak{G}(acute{mathrm{u}}, acute{mathrm{a}})), we have

$\mathfrak{G}\left(\mathrm{u},\stackrel{}{}$
a
´
)

v
´

2

D

G(,
a
´
)(
u
´
,

v
´

2
)
(

H
´

u
´

,
u

(
0
+
)
)
+
ρ
2

u

u
´

2

e

+
Ω
2
.

(6)

Again, as (mathrm{p}(cdot, acute{mathrm{a}}) : U to V_{3}) is (rho _{3})-Ω-A(mathbb{C}) with respect to (e^{prime prime}) on W, we have

$$mathrm{p}(mathrm{u}, acute{mathrm{a}}) – mathrm{p}( acute{mathrm{u}}, acute{mathrm{a}}) in mathrm{p}^{prime}(cdot, acute{mathrm{a}}) (acute{mathrm{u}}) bigl( acute{mathcal{H}}_{ acute{mathrm{u}}, mathrm{u}}(0+) bigr) + rho _{3} Vert mathrm{u} – acute{mathrm{u}} Vert ^{2} e^{prime prime}+ Omega _{3}.$$

(7)

Hence, from Eq. (5), we have

$\begin{array}{}\end{array}$

v
1

,
F
(
u
,

a
´

)

v
´

1

+
F
(

u
´

,
a
)

v
´

1

+

v
2

,
G
(
u
,

a
´

)

v
´

2

+
G
(

u
´

,
a
)

v
´

2

+

v
3

,
p
(
u
,

a
´

)

p
(

u
´

,

a
´

)
+
p
(

u
´

,
a
)

v
1

,

D

F
(

,

a
´

)
(

u
´

,

v
´

1

)

(

H
´

u
´

,
u

(
0
+
)
)

+
F
(

u
´

,
a
)

v
´

1

+

v
2

,

D

G
(

,

a
´

)
(

u
´

,

v
´

2

)

(

H
´

u
´

,
u

(
0
+
)
)

+
F
(

u
´

,
a
)

v
´

2

+

v
3

,

p

(

,

a
´

)
(

u
´

)

(

H
´

u
´

,
u

(
0
+
)
)

+
p
(

u
´

,
a
)

+

u

u
´

2

(

ρ
1

v
1

,
e

+

ρ
2

v
2

,

e

+

ρ
3

v
3

,

e

)

.

(8)

By Eq. (3), we have

begin{aligned}& bigllangle mathrm{v}_{1}^{ast}, mathfrak{F}( mathrm{u}, mathrm{a}) – acute{mathrm{v}}_{1} bigrrangle geq bigllangle mathrm{v}_{1}^{ast}, mathfrak{F}(acute{ mathrm{u}}, mathrm{a}) – acute{mathrm{v}}_{1} bigrrangle , \& bigllangle mathrm{v}_{2}^{ast},mathfrak{G}( mathrm{u}, mathrm{a}) – acute{mathrm{v}}_{2} bigrrangle geq bigllangle mathrm{v}_{2}^{ast}, mathfrak{G}(acute{ mathrm{u}}, mathrm{a}) – acute{mathrm{v}}_{2} bigrrangle , \& bigllangle mathrm{v}_{1}^{ast}, mathfrak{F}( mathrm{u}, acute{mathrm{a}})-acute{mathrm{v}}_{1} bigrrangle leq 0, bigllangle mathrm{v}_{2}^{ast}, mathfrak{G}( mathrm{u}, acute{mathrm{a}}) – acute{mathrm{v}}_{2} bigrrangle leq 0, end{aligned}

and (langle mathrm{v}_{3}^{ast}, mathrm{p}(mathrm{u}, acute{mathrm{a}}) + mathrm{p}(acute{mathrm{u}}, mathrm{a}) rangle leq 0). By assumption, we have (mathrm{p}(acute{mathrm{u}}, acute{mathrm{a}})geq 0). Therefore,

begin{aligned} &bigllangle mathrm{v}_{1}^{ast}, mathfrak{F}( mathrm{u}, mathrm{a}) – acute{mathrm{v}}_{1} bigrrangle + bigllangle mathrm{v}_{2}^{ast}, mathfrak{G}( mathrm{u}, mathrm{a}) – acute{mathrm{v}}_{2} bigrrangle \ &quad geq bigllangle mathrm{v}_{1}^{ast}, mathfrak{F} (mathrm{u}, acute{mathrm{a}}) – acute{mathrm{v}}_{1} + mathfrak{F}( acute{mathrm{u}}, mathrm{a}) – acute{mathrm{v}}_{1} bigrrangle \ &qquad {} + bigllangle mathrm{v}_{2}^{ast}, mathfrak{G}( mathrm{u}, acute{mathrm{a}}) – acute{mathrm{v}}_{2} + mathfrak{G}( acute{mathrm{u}}, acute{mathrm{a}}) – acute{ mathrm{v}}_{2} bigrrangle \ &qquad {} + bigllangle mathrm{v}_{3}^{ast}, mathrm{p}( mathrm{u}, acute{mathrm{a}}) – mathrm{p}(acute{mathrm{u}}, acute{ mathrm{a}}) + mathrm{p}( acute{mathrm{u}}, mathrm{a}) bigrrangle . end{aligned}

(9)

Consequently,

$\begin{array}{}\end{array}$

v
1

,
F
(
u
,
a
)

v
´

1

+

v
2

,
G
(
u
,
a
)

v
´

2

v
2

,

D

F
(

,

a
´

)
(

u
´

,

v
´

1

)

(

H
´

u
´

,
u

(
0
+
)
)

+
F
(

u
´

,
a
)

v
´

1

+

v
2

,

D

G
(

,

a
´

)
(

u
´

,

v
´

2

)

(

H
´

u
´

,
u

(
0
+
)
)

+
G
(

u
´

,
a
)

v
´

2

+

v
3

,

p

(

,

a
´

)
(

u
´

)

(

H
´

u
´

,
u

(
0
+
)
)

+
p
(

u
´

,
a
)

+

u

u
´

2

(

ρ
1

v
1

,
e

+

ρ
2

v
2

,

e

+

ρ
3

v
2

,

e

)

.

(10)

It completes the proof of Lemma 3.3. □

We establish the sufficient KKT optimality conditions of the parametric (mathbb{S})(mathbb{VPOP}s) (1) under contingent epiderivative and ρ(mathbb{C})A(mathbb{C}) assumptions.

### Theorem 3.4

(Sufficient optimality conditions)

Let W be an arcwise connected subset of U and ((acute{mathrm{u}}, acute{mathrm{a}})in Utimes A), with ((acute{mathrm{u}}, acute{mathrm{a}})in check{S}),

$$acute{mathrm{v}}_{1} in mathfrak{F}(acute{mathrm{u}}, acute{ mathrm{u}}), qquad acute{mathrm{v}}_{2} in mathfrak{G}( acute{ mathrm{u}}, acute{mathrm{a}}) cap (-Omega _{2}),$$

and (mathrm{p}(acute{mathrm{u}}, acute{mathrm{a}})geq 0). Let (ein operatorname{int}(Omega _{1})), (e^{prime}in operatorname{int}(Omega _{2})), and (e^{prime prime}in operatorname{int}(Omega _{3})). Suppose that (mathfrak{F}(cdot, acute{mathrm{a}}): U to 2^{V_{1}}) is (rho _{1})(Omega _{1})A(mathbb{C}) with respect to e, (mathfrak{G}(cdot, acute{mathrm{a}}) : U to 2^{V_{2}}) is (rho _{2})(Omega _{2})A(mathbb{C}) with respect to (e^{prime}), and (mathrm{p}(cdot, acute{mathrm{a}}) : U to V_{3}) is (rho _{3})(Omega _{3})A(mathbb{C}) with respect to (e^{prime prime}) on W. Assume that the contingent epiderivatives

${}_{}$
D

F(,
a
´
)(
u
´
,

v
´

1
)

and

${}_{}$
D

G(,
a
´
)(
u
´
,

v
´

2
)

exist and the Gâteaux derivative (mathrm{p}^{prime}(cdot, acute{mathrm{a}})(acute{mathrm{u}})) exists. Suppose that the conditions of Lemma 3.3hold at ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) for some

$$bigl(mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast}bigr) in Omega _{1}^{+} times Omega _{2}^{+} times Omega _{3}^{+},$$

with (mathrm{v}_{1}^{ast}neq theta _{V_{1}}) and

$$rho _{1} bigllangle mathrm{v}_{1}^{ast}, ebigrrangle + rho _{2} bigllangle mathrm{v}_{2}^{ast}, e^{prime}bigrrangle + rho _{3} bigllangle mathrm{v}_{3}^{ast}, e^{prime prime} bigrrangle geq 0,$$

(11)

such that

$\begin{array}{}\end{array}$

v
1

,

D

F
(

,

a
´

)
(

u
´

,

v
´

1

)

(

H
´

u
´

,
u

(
0
+
)
)

+
F
(

u
´

,
a
)

v
´

1

+

v
2

,

D

G
(

,

a
´

)
(

u
´

,

v
´

2

)

(

H
´

u
´

,
u

(
0
+
)
)

+
G
(

u
´

,
a
)

v
´

2

+

v
3

,

p

(

,

a
´

)
(

u
´

)

(

H
´

u
´

,

u
´

(
0
+
)
)

+
p
(

u
´

,
a
)

0
,

(12)

(forall (cdot, mathrm{a}) in Wtimes A), and

$$bigllangle mathrm{v}_{3}^{ast}, acute{mathrm{v}}_{2}bigrrangle =0,$$

(13)

then ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1})) is a weak minimizer of problem (1).

### Proof

Suppose that ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1})) is not a weak minimizer of problem (1). Then there exist ((mathrm{u}, mathrm{a})in check{S}) and (mathrm{v}_{1}in mathfrak{F}(mathrm{u}, mathrm{a})) such that (mathrm{v}_{1} <acute{mathrm{v}}_{1}). As

$$mathrm{v}_{1}^{ast}in Omega _{1}^{+} setminus {theta _{V_{1}}},$$

(langle mathrm{v}_{1}^{ast}, mathrm{v}_{1} – acute{mathrm{v}}_{1} rangle < 0). As ((mathrm{u}, mathrm{a})in mathrm{a}), there exists

$$mathrm{v}_{2} in mathfrak{G}( mathrm{u}, mathrm{a})cap (- Omega _{2}).$$

So, (langle mathrm{v}_{2}^{ast},mathrm{v}_{2} rangle leq 0) as (mathrm{v}_{2}^{ast}in Omega _{2}^{+}). Since (langle mathrm{v}_{2}^{ast}, acute{mathrm{v}}_{2} rangle = 0), we have

$$bigllangle mathrm{v}_{2}^{ast}, mathrm{v}_{2} – acute{mathrm{v}}_{2} bigrrangle = bigllangle mathrm{v}_{2}^{ast}, mathrm{v}_{2} bigrrangle leq 0.$$

Therefore,

$$bigllangle mathrm{v}_{1}^{ast}, acute{mathrm{v}}_{1} – acute{mathrm{v}}_{2} bigrrangle + bigllangle mathrm{v}_{2}^{ast}, mathrm{v}_{2} – acute{mathrm{v}}_{2} bigrrangle < 0.$$

(14)

As the conditions of Lemma 3.3 hold at ((acute{mathrm{u}}, acute{mathrm{u}}, acute{mathrm{u}}, acute{mathrm{u}}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})), from Eqs. (4), (11), and (12), we have

$$bigllangle mathrm{v}_{1}^{ast}, mathfrak{F}( mathrm{u}, mathrm{u})- acute{mathrm{v}}_{1} bigrrangle + bigllangle mathrm{v}_{2}^{ast}, mathrm{v}_{2}^{ast}(mathrm{u}, mathrm{a}) – acute{mathrm{v}}_{2} bigrrangle geq 0.$$

Hence,

$$bigllangle mathrm{v}_{1}^{ast}, mathrm{v}_{1}-acute{mathrm{v}}_{1} bigrrangle + bigllangle mathrm{v}_{2}^{ast}, mathrm{v}_{2} – acute{mathrm{v}}_{2} bigrrangle geq 0,$$

which contradicts (14). Consequently, ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1})) is a weak minimizer of problem (1). □

### Wolfe type dual

We consider a Wolfe type dual (15), where (mathfrak{F}(cdot, acute{mathrm{a}})) and (mathfrak{G}(cdot, acute{mathrm{a}})) are contingent epiderivable set-valued maps and (mathrm{p}(cdot, acute{mathrm{a}})) is a Gâteaux derivable single-valued map, where (acute{mathrm{a}}in A).

$\begin{array}{}\end{array}$

maximize

v
´

1

+

v
1

,

v
´

2

e
,

subject to

v
1

,

D

F
(

,

a
´

)
(

u
´

,

v
´

1

)

(

H
´

u
´

,
u

(
0
+
)
)

+
F
(

u
´

,
a
)

v
´

1

+

v
2

,

D

G
(

,

a
´

)
(

u
´

,

v
´

2

)

(

H
´

u
´

,
u

(
0
+
)
)

+
G
(

u
´

,
a
)

v
´

2

+

v
3

,

p

(

,

a
´

)
(

u
´

)

(

H
´

u
´

,
u

(
0
+
)
)

+
p
(

u
´

,
a
)

0
,

(15)

(forall (mathrm{u}, mathrm{a}) in Wtimes A), (acute{mathrm{u}}in W), (acute{mathrm{a}}in A), (acute{mathrm{v}}_{1}in mathfrak{F}(acute{mathrm{u}}, acute{mathrm{a}})), (acute{mathrm{v}}_{2}in mathfrak{G}(acute{mathrm{u}}, acute{mathrm{a}})), (mathrm{p}(acute{mathrm{u}}, acute{mathrm{a}}) geq 0),

$$bigl(mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast}bigr) in Omega _{1}^{+} times Omega _{2}^{+} times Omega _{3}^{+},$$

and (langle mathrm{v}_{1}^{ast}, erangle =1).

### Definition 3.5

A point ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) satisfying all the constraints of (15) is called a feasible point of problem (15). The feasible point

$$bigl(acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast}bigr)$$

of problem (15) is called a weak maximizer of (15) if there exists no feasible point ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, tilde{mathrm{v}}_{1}^{ast}, tilde{mathrm{v}}_{2}^{ast}, tilde{mathrm{v}}_{3}^{ast})) of (15) such that

$$bigl(mathrm{v}_{1} + bigllangle tilde{mathrm{v}}_{2}^{ast}, mathrm{v}_{2} bigrrangle ebigr) – bigl(acute{ mathrm{v}}_{1} + bigllangle mathrm{v}_{2}^{ast}, acute{mathrm{v}}_{2} bigrrangle ebigr)in operatorname{int}(Omega _{1}).$$

We prove the duality results of Wolfe type of problem (1). The proofs are very similar to Theorems 3.103.12, and hence omitted.

### Theorem 3.6

(Weak duality)

Let W be an arcwise connected subset of U, ((acute{mathrm{u}}_{0}, acute{mathrm{a}}_{0}) in check{S}), ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) be a feasible point of problem (15), and (mathrm{p}(acute{mathrm{u}}, acute{mathrm{a}})geq 0). Let

$$ein operatorname{int}(Omega _{1}),qquad e^{prime}in operatorname{int}( Omega _{2}), qquad e^{prime prime}in operatorname{int}(Omega _{3}).$$

Suppose that (mathfrak{F} (cdot, acute{mathrm{a}}) : U to 2^{V_{1}}) is (rho _{1})(Omega _{1})A(mathbb{C}) with respect to e, (mathfrak{G}(cdot, acute{mathrm{a}}) : U to 2^{V_{2}}) is (rho _{2})(Omega _{2})A(mathbb{C}) with respect to (e^{prime}), and (acute{mathrm{a}}(cdot, acute{mathrm{a}}) : U to V_{3}) is (rho _{3})(Omega _{3})A(mathbb{C}) with respect to (e^{prime prime}) on W. Assume that the contingent epiderivatives

${}_{}$
D

F(,
a
´
)(
u
´
,

v
´

1
)

and

${}_{}$
D

G(,
a
´
)(
u
´
,

v
´

2
)

exist and the Gâteaux derivative (mathrm{p}^{prime}(cdot, acute{mathrm{a}})(acute{mathrm{u}})) exists. Suppose that the conditions of Lemma 3.3hold at ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) and (17) is satisfied. Then

$$mathfrak{F}(mathrm{u}_{0}, mathrm{a}_{0}) – bigl( acute{mathrm{v}}_{1} + bigllangle mathrm{v}_{2}^{ast}, acute{mathrm{v}}_{2}bigrrangle ebigr) subseteq V_{1} setminus -operatorname{int}(Omega _{1}).$$

### Theorem 3.7

(Strong duality)

Let ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1})) be a weak minimizer of problem (1) and (acute{mathrm{v}}_{2}in mathfrak{G}(acute{mathrm{u}}, acute{mathrm{a}})cap (-Omega _{2})). Assume that for some

$$bigl(mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast}bigr) in Omega _{1}^{+} times Omega _{2}^{+} times Omega _{3}^{+},$$

with (langle mathrm{v}_{1}^{ast}, erangle =1), Eqs. (12) and (13) are satisfied at the point ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})). Then ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) is a feasible solution for problem (15). If the weak duality Theorem 3.6between (1) and (15) holds, then the point ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) is a weak maximizer of problem (15).

### Theorem 3.8

(Converse duality)

Let W be an arcwise connected subset of the space U and ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) be a feasible point of problem (15) with (langle mathrm{v}_{2}^{ast}, acute{mathrm{v}}_{2}rangle geq 0) and (mathrm{p}(acute{mathrm{u}}, acute{mathrm{a}})geq 0). Let (ein operatorname{int}(Omega _{1})), (e^{prime}in operatorname{int}(Omega _{2})), and (e^{prime prime} in operatorname{int}(Omega _{3})). Suppose that (mathfrak{F} (cdot, acute{mathrm{a}}) : Uto 2^{V_{1}}) is (rho _{1})(Omega _{1})A(mathbb{C}) with respect to e, (mathfrak{G}(cdot, acute{mathrm{a}}) : U to 2^{V_{2}}) is (rho _{2})(Omega _{2})A(mathbb{C}) with respect to (e^{prime}), and (mathrm{p}(cdot, acute{mathrm{a}} ) : Uto V_{3}) is (rho _{3})(Omega _{3})A(mathbb{C}) with respect to (e^{prime prime}) on W. Assume that the contingent epiderivatives

${}_{}$
D

F(,
a
´
)(
u
´
,

v
´

1
),
D

G(,
a
´
)(
u
´
,

v
´

2
),

exist and the Gâteaux derivative (mathrm{p}^{prime}(cdot, acute{mathrm{a}})( acute{mathrm{u}})) exists. Suppose that the conditions of Lemma 3.3hold at ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) and (17) is satisfied. If ((acute{mathrm{u}}, acute{mathrm{a}})in check{S}), then ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{2})) is a weak minimizer of (1).

### Mond–Weir type dual

We consider a Mond–Weir type dual (16), where (mathfrak{F}(cdot, mathfrak{F})) and (mathfrak{G}(cdot, acute{mathrm{a}})) are contingent epiderivable and (mathrm{p}(cdot, acute{mathrm{a}})) is a Gâteaux derivable single-valued map, where (acute{mathrm{a}}in A).

$\begin{array}{}\end{array}$

maximize

v
´

1

,

subject to

v
1

,

D

F
(

,

a
´

)
(

a
´

,

v
´

1

)

(

H
´

u
´

,
u

(
0
+
)
)

+
F
(

u
´

,
a
)

v
´

1

+

v
2

,

D

G
(

,

a
´

)
(

u
´

,

v
´

2

)

(

H
´

u
´

,
u

(
0
+
)
)

+
G
(

u
´

,
a
)

v
´

2

+

v
3

,

p

(

,

a
´

)
(

u
´

)

(

H
´

u
´

,

u
´

(
0
+
)
)

+
p
(

u
´

,
a
)

0
,

(16)

(forall (mathrm{u}, mathrm{u}) in Wtimes A), (langle mathrm{v}_{2}^{ast}, acute{mathrm{v}}_{2} rangle geq 0), (acute{mathrm{u}}in W), (acute{mathrm{a}}in A), (acute{mathrm{v}}_{1} in mathfrak{F}(acute{mathrm{u}}, acute{mathrm{a}})), (acute{mathrm{a}}in mathfrak{G}(acute{mathrm{u}}, acute{mathrm{a}})), (mathrm{p}(acute{mathrm{u}}, acute{mathrm{a}}) geq 0),

$$bigl(mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast}bigr) in Omega _{1}^{+} times Omega _{2}^{+} times Omega _{3}^{+},$$

with (langle mathrm{v}_{1}^{ast}, erangle =1).

### Definition 3.9

A point ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) satisfying all the constraints of problem (16) is called a feasible point of (16). The feasible point is called a weak maximizer of problem (16) if there exists no feasible point ((mathrm{u}, mathrm{a}, mathrm{v}_{1}, mathrm{v}_{2}, tilde{mathrm{v}}_{1}^{ast}, tilde{mathrm{v}}_{2}^{ast}, tilde{mathrm{v}}_{3}^{ast})) of (16) such that (mathrm{v}_{1}- acute{mathrm{v}}_{1} in operatorname{int}(Omega _{1})).

### Theorem 3.10

(Weak duality)

Let W be an arcwise connected subset of U, ((mathrm{u}_{0}, mathrm{a}_{0}) in check{S}), ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) be a feasible point of problem (16), and (mathrm{p}(acute{mathrm{u}}, acute{mathrm{a}})geq 0). Let

$${ein operatorname{int}(Omega _{1}), qquad e^{prime}in operatorname{int}(Omega _{2}),qquad e^{prime prime}in operatorname{int}( Omega _{3}).}$$

Suppose that (mathfrak{F}(cdot, acute{mathrm{a}}) : U to 2^{V_{2}}) is (rho _{1})(Omega _{1})A(mathbb{C}) with respect to e, (mathfrak{G}(cdot, acute{mathrm{a}}) : U to 2^{V_{2}}) is (rho _{2})(Omega _{2})A(mathbb{C}) with respect to (e^{prime}), and (mathrm{p}(cdot, acute{mathrm{a}}) : U to V_{3}) is (rho _{3})(Omega _{3})A(mathbb{C}) with respect to (e^{prime prime}) on W. Assume that the contingent epiderivatives

${}_{}$
D

F(,
u
´
)(
u
´
,

v
´

1
)

and

${}_{}$
D

G(,
a
´
)(
u
´
,

v
´

2
)

exist and the Gâteaux derivative (mathrm{p}^{prime}(cdot, acute{mathrm{a}})(acute{mathrm{u}})) exists. Suppose that the conditions of Lemma 3.3hold at ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})). Assume that

$$rho _{1} +rho _{2} bigllangle mathrm{v}_{2}^{ast}, e^{prime}bigrrangle + rho _{3} bigllangle mathrm{v}_{3}^{ast}, e^{prime prime} bigrrangle geq 0.$$

(17)

Then (mathfrak{F}(mathrm{u}_{0}, mathrm{a}_{0}) – acute{mathrm{v}}_{1} subseteq V_{1} setminus – operatorname{int}(Omega _{1})).

### Proof

We prove the theorem by the method of contradiction. Suppose that for some

$$mathrm{v}_{1}^{circ}in mathfrak{F}( mathrm{u}_{0}, mathrm{a}_{0}), qquad mathrm{v}_{1}^{circ}- acute{mathrm{v}}_{1} in – operatorname{int}(Omega _{1}).$$

Therefore, (langle mathrm{v}_{1}^{ast}, mathrm{v}_{1}^{circ}- acute{mathrm{v}}_{1} rangle < 0) as (theta _{V_{1}} neq mathrm{v}_{1}^{ast}in Omega _{1}^{+}). Again, since ((mathrm{u}_{0}, mathrm{a}_{0})in check{S}), we have

$$mathfrak{G} (mathrm{u}_{0}, mathrm{a}_{0})cap (- Omega _{2} ) neq emptyset ,$$

and (mathrm{p}(mathrm{u}_{0}, mathrm{a}_{0}) =0). We choose

$$mathrm{v}_{2}^{circ}in mathfrak{G}( mathrm{u}_{0}, mathrm{a}_{0}) cap (-Omega _{2}).$$

So, (langle mathrm{v}_{2}^{ast}, mathrm{v}_{2}^{circ}rangle leq 0) as (mathrm{v}_{2}^{ast}in Omega _{2}^{+}). Again, from the constraints of (16), we have (langle mathrm{v}_{2}^{ast}, acute{mathrm{v}}_{2} rangle geq 0). Therefore,

$$bigllangle mathrm{v}_{2}^{ast},mathrm{v}_{2}^{circ}- acute{mathrm{v}}_{2} bigrrangle = bigllangle mathrm{v}_{2}^{ast}, mathrm{v}_{2}^{circ}bigrrangle – bigllangle mathrm{v}_{2}^{ast}, acute{ mathrm{v}}_{2} bigrrangle leq 0.$$

Hence,

$$bigllangle mathrm{v}_{1}^{ast}, mathrm{v}_{1}^{circ}- acute{mathrm{v}}_{1} bigrrangle +bigllangle mathrm{v}_{2}^{ast}, mathrm{v}_{2}^{circ}- acute{mathrm{v}}_{2} bigrrangle < 0.$$

(18)

As the conditions of Lemma 3.3 hold at ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})), from Eqs. (4), (17) and the constraints of (16), we have

$$bigllangle mathrm{v}_{1}^{ast}, mathfrak{F} bigl( mathrm{v}_{1}^{circ}, mathrm{a}bigr) – acute{ mathrm{v}}_{1} bigrrangle + bigllangle mathrm{v}_{2}^{ast}, mathfrak{G}(mathrm{u}_{0}, mathrm{a}) – acute{ mathrm{v}}_{2} bigrrangle geq 0.$$

Hence,

$$bigllangle mathrm{v}_{1}^{ast}, mathrm{v}_{1}^{circ}- acute{mathrm{v}}_{1} bigrrangle + bigllangle mathrm{v}_{2}^{ast}, mathrm{v}_{2}^{circ}- acute{mathrm{v}}_{2} bigrrangle geq 0,$$

$$mathfrak{F}(mathrm{u}_{0}, mathrm{a}_{0}) – acute{ mathrm{v}}_{1} subseteq V_{1}setminus – operatorname{int}(Omega _{1}).$$

It completes the proof of the theorem. □

### Theorem 3.11

(Strong duality)

Let ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1})) be a weak minimizer of problem (1) and (acute{mathrm{v}}_{2}in mathfrak{G}(acute{mathrm{u}}, acute{mathrm{a}}) cap (-Omega _{2})). Assume that for some

$$bigl(mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast}bigr) in Omega _{1}^{+} times Omega _{2}^{+} times Omega _{3}^{+},$$

with (langle mathrm{v}_{1}^{ast}, erangle = 1), Eqs. (12) and (13) are satisfied at the point

$$bigl(acute{mathrm{u}}, acute{mathrm{a}},acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast}bigr).$$

Then ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) is a feasible solution for problem (16). If the weak duality Theorem 3.10between (1) and (16) holds, then the point ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) is a weak maximizer of (16).

### Proof

As Eqs. (12) and (13) are satisfied at ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})),

$\begin{array}{}\end{array}$

v
1

,

D

F
(

,

a
´

)
(

u
´

,

v
´

1

)

(

H
´

u
´

,
u

(
0
+
)
)

+
F
(

u
´

,
a
)

v
´

1

+

v
2

,

D

G
(

,

a
´

)
(

u
´

,

v
´

2

)

(

H
´

u
´

,
u

(
0
+
)
)

+
G
(

u
´

,
u
)

v
´

2

+

v
3

,

p

(

,

a
´

)
(

u
´

)

(

H
´

u
´

,
u

(
0
+
)
)

+
p
(

u
´

,
a
)

0
,

(19)

(forall (mathrm{u}, mathrm{a}) in Wtimes A), and (langle mathrm{v}_{2}^{ast}, acute{mathrm{v}}_{2} rangle =0). As ((acute{mathrm{u}}, acute{mathrm{a}})in check{S}), (mathrm{p}(acute{mathrm{u}}, acute{mathrm{a}})=0). Hence, ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) is a feasible solution for (16). Suppose that the weak duality Theorem 3.10 between problems (1) and (16) holds and the point ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) is not a weak maximizer of problem (16). Let

$$bigl(mathrm{u}, mathrm{a}, mathrm{v}_{1}, mathrm{v}_{2}, tilde{mathrm{v}}_{1}^{ast}, tilde{ mathrm{v}}_{2}^{ast}, tilde{mathrm{v}}_{3}^{ast}bigr)$$

be a feasible point for (16) such that (acute{mathrm{v}}_{1} – mathrm{v}_{1} in -operatorname{int}( Omega _{1})). It contradicts the weak duality Theorem 3.10 between (1) and (16). Consequently, ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) is a weak maximizer for (16). □

### Theorem 3.12

(Converse duality)

Let W be an arcwise connected subset of U, (mathrm{p}(acute{mathrm{u}}, acute{mathrm{a}}) geq 0), and ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) be a feasible point of problem (16). Let

$$ein operatorname{int}(Omega _{1}),qquad e^{prime}in operatorname{int}(Omega _{2}),qquad e^{prime prime}in operatorname{int}(Omega _{3}).$$

Suppose that (mathfrak{F}(cdot, acute{mathrm{a}}) : U to 2^{V_{1}}) is (rho _{1})(Omega _{1})A(mathbb{C}) with respect to e, (acute{mathrm{a}}(cdot,acute{mathrm{a}}) : U to 2^{V_{2}}) is (rho _{2})(Omega _{2})A(mathbb{C}) with respect to (e^{prime}), and (acute{mathrm{a}}(cdot, acute{mathrm{a}}) : U to V_{3}) is (rho _{3})(Omega _{3})A(mathbb{C}) with respect to (e^{prime prime}) on W. Assume that the contingent epiderivatives

${}_{}$
D

F(,F)(
u
´
,

v
´

1
)

and

${}_{}$
D

G(,
a
´
)(
u
´
,

v
´

2
)

exist and the Gâteaux derivative (mathrm{p}^{prime}(cdot, acute{mathrm{a}})(acute{mathrm{u}})) exists. Suppose that the conditions of Lemma 3.3hold at ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})) and (17) is satisfied. If ((acute{mathrm{u}}, acute{mathrm{a}})in check{S}), then ((acute{mathrm{u}}, acute{mathrm{a}},acute{mathrm{v}}_{1})) is a weak minimizer of (1).

### Proof

Suppose that ((acute{mathrm{u}}, acute{mathrm{a}},acute{mathrm{v}}_{1})) is not a weak minimizer of problem (1). Then there exist ((mathrm{u}, mathrm{a})in check{S}) and (mathrm{v}_{1} in mathfrak{F}(mathrm{u}, mathrm{a})) such that (mathrm{v}_{1} < acute{mathrm{v}}_{1}). As (mathrm{v}_{1}^{ast}in Omega _{1}^{+} setminus {theta _{V_{1}} }), (langle mathrm{v}_{1}^{ast}, mathrm{v}_{1} – acute{mathrm{v}}_{1} rangle < 0). As ((mathrm{u}, mathrm{a})in check{S}), there exists

$$mathrm{v}_{2}in mathfrak{G}(mathrm{u},mathrm{a}) cap (-Omega _{2}).$$

So, (langle mathrm{v}_{2}^{ast}, mathrm{v}_{2} rangle leq 0) as (mathrm{v}_{2}^{ast}in Omega _{2}^{+}). By the constraints of (16), we have (langle mathrm{v}_{2}^{ast}, acute{mathrm{v}}_{2}rangle geq 0). Therefore,

$$bigllangle mathrm{v}_{2}^{ast}, mathrm{v}_{2} – acute{mathrm{v}}_{2} bigrrangle = bigllangle mathrm{v}_{2}^{ast}, mathrm{v}_{2} bigrrangle – bigllangle mathrm{v}_{2}^{ast}, acute{ mathrm{v}}_{2}bigrrangle leq 0.$$

Indeed,

$$bigllangle mathrm{v}_{1}^{ast}, mathrm{v}_{1} – acute{mathrm{v}}_{1} bigrrangle + bigllangle mathrm{v}_{2}^{ast}, mathrm{v}_{2} – acute{mathrm{v}}_{2} bigrrangle < 0.$$

(20)

As the conditions of Lemma 3.3 hold at ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1}, acute{mathrm{v}}_{2}, mathrm{v}_{1}^{ast}, mathrm{v}_{2}^{ast}, mathrm{v}_{3}^{ast})), from Eqs. (4), (17) and the constraints of (16), we have

$$bigllangle mathrm{v}_{1}^{ast}, mathfrak{F}( mathrm{u}, mathrm{a}) – acute{mathrm{v}}_{1} bigrrangle + bigllangle mathrm{v}_{2}^{ast}, mathfrak{G}(mathrm{u}, mathrm{a}) – acute{mathrm{v}}_{2}bigrrangle geq 0.$$

Hence,

$$bigllangle mathrm{v}_{1}^{ast}, mathrm{v}_{1} -acute{mathrm{v}}_{1} bigrrangle + bigllangle mathrm{v}_{2}^{ast}, mathrm{v}_{2} – acute{mathrm{v}}_{2}bigrrangle geq 0,$$

which contradicts (20). Consequently, ((acute{mathrm{u}}, acute{mathrm{a}}, acute{mathrm{v}}_{1})) is a weak minimizer of problem (1). □