See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/268250827

Proper Efficiency and Proper Karush–Kuhn–Tucker Conditions for Smooth

Multiobjective Optimization Problems

Article in Vietnam Journal of Mathematics · November 2014

DOI: 10.1007/s10013-014-0102-2

CITATIONS READS

4 603

2 authors:

Regina Burachik M. M. Rizvi

University of South Australia University of South Australia
142 PUBLICATIONS 3,046 CITATIONS 14 PUBLICATIONS 235 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by M. M. Rizvi on 19 November 2014.

The user has requested enhancement of the downloaded file.

Proper Efficiency and Proper Karush-Kuhn-Tucker
Conditions for Smooth Multiobjective
Optimization Problems ∗
† ‡
Regina S. Burachik M. M. Rizvi

September 18, 2014

Abstract Proper Karush-Kuhn-Tucker (PKKT) conditions are said to hold when

all the multipliers of the objective functions are positive. In 2012, Burachik and Rizvi
introduced a new regularity condition under which PKKT conditions hold at every
Geoffrion-properly efficient point. In general, the set of Borwein properly-efficient
points is larger than the set of Geoffrion-properly efficient points. Our aim is to
extend the PKKT conditions to the larger set of Borwein-properly efficient points.
Keywords Multiobjective optimization, Regularity conditions, Optimality condi-
tions for efficient solutions, Geoffrion-properly efficient solutions, Borwein-properly
efficient solutions, Proper Karush-Kuhn-Tucker conditions.
AMS subject classification. 90C29

1 Introduction
The process of optimizing a collection of objective functions simultaneously is called
multiobjective optimization (or vector optimization) problems (MOP). Because of
the conflicting nature of the objectives, it is hard to find a single solution that
optimizes all the objectives simultaneously. Therefore, a set of alternative solutions
is found and these solutions are called Pareto optimal points [1], vector minimum
points [2], non-inferior points [3] or efficient points [4, 5]. We use the term efficient
point for these solutions. More restricted definitions of efficiency, called proper
∗
M. M. Rizvi was supported by UniSA President’s Scholarships and the School of Information
Technology and Mathematical Sciences at University of South Australia.
†
School of Information Technology and Mathematical Sciences, University of South Australia
email: [email protected]
‡
School of Information Technology and Mathematical Sciences, University of South Australia,
email: [email protected]

1
efficiency, have been introduced by Kuhn-Tucker [6], Geoffrion [7] and Borwein [8].
We recall that every Geoffrion-properly efficient point is a Borwein-properly efficient
point. The converse is not true in general. More details on the relationship between
these notions is recalled in Section 2.1.
Optimality conditions are studied to find efficient points of multiobjective optimiza-
tion problems. Many authors derive the first-order and second-order necessary and
sufficient conditions for multiobjective optimization problems in the case of smooth
[5, 9, 10] and non-smooth functions [11, 12, 13]. Initially, the optimality conditions
of multiobjective problems used the same constraint qualifications as in the single-
objective case [14, 15]. For these constraint qualifications we could obtain at least
one positive multiplier corresponding to the objective function. This means that
some of the multipliers may be equal to zero. As a consequence, some Lagrange
multipliers have no role in the necessary conditions of optimality. To avoid this
situation and to obtain positive Lagrange multipliers associated with each of the
objective functions, we need a constraint qualification. In 1994, Maeda [5] intro-
duced a generalized constraint qualification and ensured the existence of positive
Lagrange multipliers. Later Bigi [2], and more recently, Burachik and Rizvi [4]
introduced a generalized regularity condition for multiobjective optimization prob-
lems. They were able to ensure positivity of all Lagrange multipliers under these
regularity conditions. The positivity of a multiplier corresponding to an objective
function indicates an active role of that function in the identification of the solutions
of the problem. When all the multipliers corresponding to the objective functions
are positive, we say that proper Karush-Kuhn-Tucker (PKKT) conditions hold.
In [4], PKKT conditions are established for Geoffrion-properly efficient points.
Our aim is to establish PKKT conditions for the larger set of Borwein-properly
efficient points (see Theorem 3.2).
Note that generalized constraint qualification and regularity conditions introduced
respectively in [5] and [4] do not guarantee KKT conditions for the non-smooth case.
This is why, in this short note, we focus on the differentiable case.
The outline of the present paper is as follows: Section 2 contains notations, defi-
nitions and preliminary materials. Still in this section, we describe our theoretical
setting, recall our regularity condition from [4], and prove a technical lemma. Fi-
nally, Section 3 contains the main results involving PKKT conditions under our
regularity condition.

2 Preliminaries
In this section, we recall some basic tools of multiobjective optimization problems
(MOP) which are used in the rest of the analysis. Let Rn be the n-dimensional
Euclidean space. For xi , yi ∈ R, we say xi > yi iff (xi − yi ) is non negative. For

2
x, y ∈ Rn we use from now on the following conventions [16].

x = y, iff xi > yi , i = 1, ..., n.

In other words, = represents in Rn the usual order coordinate by coordinate.

x ≥ y, iff x = y and x 6= y,

so ≥ means that each coordinate of x is greater than or equal each coordinate of y,

and x 6= y.
x > y, iff xi > yi and i = 1, ..., n.
Hence > stands for the usual strict inequality, coordinate by coordinate. For `, m ∈
N, we consider in this paper the following multiobjective optimization Problem (P):

min f (x) := [f1 (x ), ..., f` (x )],
(P) (1)
subject to gj (x) 5 0, j = 1, 2, ..., m.

where fi : Rn → R, i = 1, ..., ` and gj : Rn → R, j = 1, ..., m are continuously

differentiable real-valued functions. We define X = {x ∈ Rn | gj (x) 5 0, j =
1, 2, ..., m} and I(x̄) := {j|gj (x̄) = 0} . In our analysis, we do not make any convexity
assumption on the functions involved in (P ).
We next recall the definitions of efficient and local efficient points.

Definition 2.1 A point x̄ is said to be efficient for Problem (P) if there is no x ∈ X

such that f (x) ≤ f (x̄). A point x̄ is said to be localTefficient to Problem (P) if there
exists δ > 0 such that x̄ is an efficient point in X B(x̄, δ).

The analysis of optimality conditions depends on the concept of the contingent and
linearizing cones, as well as their relationship. This relationship is used in many
constraint qualifications. More studies on the relationship between contingent and
linearizing cones can be found in [5, 13, 17, 18, 19, 20, 21]. We use the following
two definitions in our analysis.

Definition 2.2 Let C be a subset of Rn . The contingent cone to C at x̄ ∈ C is the

set defined by

T (C; x̄) := {d ∈ Rn | ∃{(xn , tn )} ⊆ (C×R++ ) such that xn → x̄ and tn (xn −x̄) → d}.

The set T (C; x̄) is a nonempty and closed cone. More properties of this cone can be
seen in [22, 23, 24, 25, 26].

Definition 2.3 The active linearizing cone to X at x̄ ∈ X is the set defined by

L̃(X; x̄) := d ∈ Rn | Ogj (x̄)T d 5 0, j ∈ I(x̄) .


(2)

3
2.1 Proper Efficiency
In 1952, Kuhn and Tucker [6] pointed out that certain efficient points show un-
favourable properties concerning the ratio of the rate of gain in one cost over the
corresponding rate of loss in another cost. To avoid such pathological behaviour,
they introduced the properness of the efficient solution of the multiobjective opti-
mization problems. More general definitions of proper efficiency are given by Ge-
offrion [7] and Borwein [8]. For discussions about the relationship among different
types of proper efficiency see [27]. In our analysis, we focus on the concepts of
Borwein-proper efficiency and Geoffrion-proper efficiency. We use these concepts
in Theorem 3.1 and Corollary 3.1 to ensure PKKT conditions under our proposed
regularity condition. We recall below the relevant definitions of proper efficiency.

Definition 2.4 (See [6]) A point x̄ is said to be Kuhn-Tucker-properly efficient

(KTPE) of Problem (P) if it is efficient and the following system has no solution.
Ofi (x̄)T d 5 0,
∃i0 ∈ {1, 2, . . . , `}, such that Ofi0 (x̄)T d < 0,
Ogj (x̄)T d 5 0, j ∈ I(x̄).

Next we recall the definition of Geoffrion-properly efficient solution.

Definition 2.5 (See [7]) A point x̄ is said to be Geoffrion-properly efficient (GPE)

of Problem (P ) if it is efficient and there exists a scalar M > 0 such that, for each
i, we have
fi (x) − fi (x̄)
5 M,
fj (x̄) − fj (x)
for some j such that fj (x̄) < fj (x) whenever x ∈ X and fi (x̄) > fi (x).

An efficient point that is not properly efficient is said to be improperly efficient.

Thus a point x̄ is said to be improperly efficient of Problem (P ) when for every
scalar M > 0 (no matter how large) there is a point x ∈ X and an index i such that
fi (x̄) > fi (x) and
fi (x) − fi (x̄)
> M,
fj (x̄) − fj (x)
for all j such that fj (x̄) < fj (x).
The concept of Borwein proper efficiency relies on the definition of contingent cone
given in Definition 2.2.

Definition 2.6 (See [8]) A point x̄ is said to be Borwein-properly efficient (BPE)

of Problem (P ) if it is efficient and
\
T (f (X) + R`+ , f (x̄)) (−R`+ ) = {0}.

4
Remark 2.1 In general, we have the following inclusions.

BP E ⊇ GP E, (3)

(see [8, Proposition 1], where the inclusion “ ⊇ ” stands for “less restrictive”. How-
ever, if X is convex and fi are convex functions then we have

BP E = GP E,

(see [28, Theorems 2.45(2) and 2.48]).

The following example (taken from [28, Exercise 2.13]), illustrates the case of
a Borwein properly efficient point which is not properly efficient in the sense of
Definitions 2.5 and 2.4.

Example 2.1 Let f : R2 → R2 be given by f (x) := x, and consider the problem

min f (x),
[ [
s.t. x ∈ X = {x ∈ R2 | x1 + x2 = 0} {x ∈ R2 | x1 = 1} {x ∈ R2 | x2 = 1} ,
and X = f (X). Here, x̄ = (0, 0) is an efficient point. It is easy to check that
\ \
T (f (X) + R2+ , f (x̄)) (−R2+ ) = {x ∈ R2 | x1 + x2 = 0} (−R2+ ) = {0}.

Hence, x̄ is BPE. Note that

\
cl(cone(f (X) + R2+ − f (x̄)) (−R2+ ) = L1 ∪ L2 6= {0},

where

L1 := {x ∈ R2 | x2 = 0, x1 5 0} and L2 := {x ∈ R2 | x1 = 0, x2 5 0}.

We note that x̄ is not GPE. Indeed, let x = (−a, 1), a > 0, then we have that

f1 (x) − f1 (x̄)
= a. (4)
f2 (x̄) − f2 (x)

The value of (4) can be made arbitrarily large by taking a > 0. Hence x̄ is not
a GPE. Next we prove that x̄ is not KTPE. For this point, only the constraint
g1 (x) = x1 + x2 = 0 is active. If d = (−1, −1), we have that

Of1 (x̄)T d < 0, Of2 (x̄)T d < 0, Og1 (x̄)T d < 0, 1 ∈ I(x̄),

where Of1 (x̄) = (1, 0), Of2 (x̄) = (0, 1), and Og1 (x̄) = (1, 1). Hence, x̄ is not KTPE.

5
2.2 Regularity Conditions
In scalar optimization problems, constraint qualifications are used to ensure the
positivity of the Lagrange multiplier associated with the objective function. How-
ever, constraint qualifications are not enough to guarantee the positivity of Lagrange
multipliers in multiobjective optimization problems. To obtain PKKT conditions,
i.e., positivity of Lagrange multipliers associated with each of the objective func-
tions, some special assumptions on the objective functions and the constraints are
required. These assumptions are called regularity conditions (RC). Fix x̄ ∈ X, and
for f , g as in (1) define
Q := {x ∈ Rn | g(x) 5 0, f (x) 5 f (x̄)} , and
Qi := {x ∈ Rn | g(x) 5 0, fk (x) 5 fk (x̄), k = 1, 2, . . . , ` and k 6= i} .
In our analysis we will need another concept of linearizing cone which is different
from the one given by (2.3). This concept of cone uses the set Q defined above, it
was introduced by Maeda in [5], and its definition is as follows.

Definition 2.7 The linearizing cone to Q at x̄ ∈ Q is the set defined by

L(Q; x̄) := d ∈ Rn | Ofi (x̄)T d 5 0, i = 1, 2, . . . , `, Ogj (x̄)T d 5 0, j ∈ I(x̄) .


Note that L(Q; x̄) ⊆ L̃(X; x̄), where L̃ as in Definition 2.3.

In 1994, Maeda [5] introduced the Guignard type regularity conditions.

Definition 2.8 (Generalized Guignard constraint qualification (GGCQ))

Let x̄ be a feasible point of (P). Then (1) is said to satisfy the GGCQ if
`
\
L(Q; x̄) ⊆ clconvT (Qi ; x̄). (5)
i=1

The regularity condition we focus on has been introduced by Burachik and Rizvi in
[4]. We recall now the definition of M i from Rizvi ([29, 30]). For a fixed x̄ ∈ X,
M i := {x ∈ Rn | g(x) 5 0, fi (x) 5 fi (x̄)} , i = 1, 2, . . . , `.
`
\
Note that, by defining M := M i , we obtain Q = M . The relationship between
i=1
these sets is obviously given by the equality
`
\
Qi = M j ; i = 1, ..., `.
j=1
j6=i

Based on these sets, we will define the following linearizing cones.

6
Definition 2.9 The linearizing cone to M i at x̄ ∈ M i is the set defined by
L(M i ; x̄) := d ∈ Rn | Ofi (x̄)T d 5 0, Ogj (x̄)T d 5 0, j ∈ I(x̄) , i = 1, 2, . . . , `,


and the linearizing cone to M at x̄ ∈ M is the set defined by

L(M ; x̄) := d ∈ Rn | Ofi (x̄)T d 5 0, i = 1, 2, . . . , `, Ogj (x̄)T d 5 0, j ∈ I(x̄) .


In Theorem 3.1, we will use the following three simple lemmas.

Lemma 2.1 Let h : Rp → R be a continuously differentiable function. Assume

that
i) zn → z̄,
ii) h(zn ) 5 h(z̄),
iii) v = limn→∞ sn (zn − z̄) with sn > 0 for all n then,
Oh(z̄)T v 5 0.

Proof. See, [4, Lemma 3.1]. 2

Lemma 2.2 Let f : Rn → R` be a continuously differentiable function. Assume

that {tn } ⊆ R++ , {zn } ⊆ Rn and z̄ ∈ Rn , such that zn → z̄ with h = limn→∞ tn (zn −
z̄). Then, we have that
lim tn (f (zn ) − f (z̄)) = (Of1 (z̄)T h, . . . , Of` (z̄)T h)T .
n→∞

Proof. It is enough to show that limn→∞ tn (fi (zn ) − fi (z̄)) = Ofi (z̄)T h, for all
i = 1, . . . , `. Fix i = 1, . . . , `. From Taylor Theorem we have
tn (fi (zn ) − fi (z̄)) = tn Ofi (z̄)T (zn − z̄) + oi (kzn − z̄k) ,

which can be rearranged as

tn (fi (zn ) − fi (z̄)) = Ofi (z̄)T (tn (zn − z̄)) + tn oi (kzn − z̄k).
Letting n → ∞ we have
lim tn (fi (zn ) − fi (z̄)) = Ofi (z̄)T h + lim tn oi (kzn − z̄k).
n→∞ n→∞

Note that,
 
oi kzn − z̄k
lim tn oi (kzn − z̄k) = lim tn (zn − z̄) = 0,
n→∞ n→∞ (zn − z̄)
oi kzn −z̄k
since (zn −z̄)
→ 0 as limn→∞ tn (zn − z̄) = h. This completes the proof. 2
The following lemma shows the relationship between the contingent cone T (M i ; x̄)
and the linearizing L(M ; x̄) cone.

7
Lemma 2.3 If x̄ is a feasible solution to problem P, then
`
\ \
cl conv T (M i , x̄) ⊆ L(M ; x̄) where M = M i. (6)
i=1

Proof. See, [4, Lemma 3.2]. 2

The reverse inclusion in (6) does not hold. Note that, when the sets M i are closed
and convex, then cl conv T (M i , x̄) = T (M i , x̄). Hence, we consider the assumption
`
\
L(M ; x̄) ⊆ T (M i , x̄), (7)
i=1

which is called generalized Abadie regularity condition (GARC), and was introduced
in [4].

3 Main Results

3.1 Proper Karush-Kuhn-Tucker Conditions

A standard way to prove PKKT conditions is to show that KTPE holds for (P ). In
other words, we need to show the inconsistency of the system
Ofi (x̄)T d 5 0,
∃i0 ∈ {1, 2, . . . , `}, such that Ofi0 (x̄)T d < 0,
Ogj (x̄)T d 5 0, j ∈ I(x̄).

Remark 3.1 Theorem 4.2 in [13] establishes that PKKT conditions hold for any
locally efficient solution (see Definition 2.1) of problem (P), under the following
assumption.
l
\
L(Q; x̄) ⊆ clconvT (Qi , x̄). (8)
i=1

Conditions (8) and (7) coincide when Qi and M i are all convex. However, condition
(7) does not imply PKKT conditions for arbitrary efficient points (see, for instance,
[4, Example 4.2]). On the other hand, it was recently established in [4, Theorem 5.3]
that condition (7) implies PKKT conditions for Geoffrion-properly efficient points.
Li (2000) presented an example in [25, Example 4.1] with a locally Lipschitz con-
straint, and shows that (8) does not guarantee KKT conditions for the non-smooth
case. Regarding GARC, it was shown in [25, Example 4.1], that regularity condition
(7) does not also guarantee KKT condition for the non-smooth case. This is why, as
we mentioned in the introduction, we restrict our analysis to the differentiable case.

8
Next we prove that the result established in [4, Theorem 5.3] can be extended to
the larger set of Borwein-properly efficient points.

Theorem 3.1 Let x̄ ∈ X. Assume that GARC (condition (7)) holds at x̄. Suppose
that x̄ ∈ X is a Borwein-properly efficient point of (P). Then the system

Ofi (x̄)T d 5 0, ∃ i0 ∈ {1, 2, . . . , `}, such that Ofi0 (x̄)T d < 0, (9)

Ogj (x̄)T d 5 0, j ∈ I(x̄), (10)

has no solution d ∈ Rp .

Proof. Assume that (9) and (10) have a solution d ∈ Rp . Without loss of generality,
one may assume
Of1 (x̄)T d < 0, (11)
Ofi (x̄)T d 5 0, i ∈ {2, . . . , `}. (12)
From (11) and (12), we can write d ∈ L(M ; x̄), since GARC holds. It implies that

d ∈ T (M i ; x̄), ∀i ∈ {1, 2, ..., `}.

Hence, for any i ∈ {2, 3, ..., `}, we have that d ∈ T (M i ; x̄). This implies that there
exist sequences {xn }n∈N ⊆ M i and

{sn } ⊂ R++ (13)

such that
lim xn = x̄, and lim sn (xn − x̄) = d. (14)
n→∞ n→∞

Fix i0 ∈ {2, 3, ..., `}. For this i0 , take the corresponding sequence {xn } ⊆ M i0 as
above. For this sequence we have

fi0 (xn ) − fi0 (x̄) 5 0.

For every n ∈ N, define the set

Jn := {k = 2 | fk (xn ) > fk (x̄)} ⊂ {2, 3, ..., `}.

We claim that Jn 6= ∅ for all n ∈ N. Indeed, if Jn = ∅ for some n, then this means
that ∀k = 2 (in particular, this includes k = i0 ) we would have that

fk (xn ) 5 fk (x̄). (15)

Using also the fact that f1 (xn ) < f1 (x̄), we arrive at a contradiction with the
efficiency of x̄. Hence, Jn 6= ∅ for all n ∈ N.
Moreover, Jn ∈ P({2, 3, ..., `}) := P` , the family of subsets of the finite set
{2, 3, ..., `}. Consider the function ϕ : N → P` defined as ϕ(n) = Jn . Since the

9
set of natural numbers is infinite and P` has a finite number of elements, there must
exist a subset J in the range of ϕ such that ϕ−1 (J) = {n ∈ N | ϕ(n) = J} is
infinite. Note that J is nonempty because ϕ(n) = Jn 6= ∅, ∀ n. Hence, there exists
J ∈ P` such that ϕ−1 (J) =: I0 ⊆ N is infinite. For the subsequence {xn }n∈I0 we have
ϕ(n) = J. In other words, ϕ is constant when restricted to n ∈ I0 . By restricting
our analysis to the subsequence {xn }n∈I0 , we can assume without loss of generality
that ϕ(n) = Jn = J is constant for all n. Using Lemma 2.1 for h = fk , (14) and
(15), we have
Ofk (x̄)T d = 0, ∀k ∈ Jn = J, ∀n.
The above expression, together with (12) implies

Ofk (x̄)T d = 0, for all k ∈ J. (16)

Let
rn = (−1/2sn )(Of1 (x̄)T d, . . . , Of` (x̄)T d)T ∈ R`+ , (17)
where {sn } is as in (13) and define ĥ ∈ R` as

ĥ = lim sn (f (xn ) + rn − f (x̄)). (18)

n→∞

From (11) and (17),

lim sn rn = (−1/2)(Of1 (x̄)T d, . . . , Of` (x̄)T d)T ∈ R`+ \ {0}. (19)

n→∞

By Lemma 2.2 applied to zn := xn , z̄ := x̄ and h = d = limn→∞ sn (xn − x̄), we have

lim sn (f (xn ) − f (x̄)) = (Of1 (x̄)T d, . . . , Of` (x̄)T d)T . (20)

n→∞

Thus, combining (18), (19) and (20) we obtain,

ĥ = (1/2)(Of1 (x̄)T d, . . . , Of` (x̄)T d)T ∈ (−R`+ ) \ {0}.

From (18) we have that ĥ ∈ T (f (X) + R`+ , f (x̄)) and ĥ 6= 0, contradicting the
Borwein proper efficiency of x̄. Therefore, (9) and (10) have no solution. 2
The following corollary is an immediate consequence of Theorem 3.1. It was obtained
in [4, Theorem 5.3], using only the concept of Geoffrion-properly efficient points.

Corollary 3.1 Let x̄ ∈ X. Assume that GARC (condition (7)) holds at x̄. Suppose
that x̄ ∈ X is a Geoffrion-properly efficient point of (P). Then (9) and (10) have
no solution.

Proof. The conclusion follows from Theorem 3.1 and the fact that every Geoffrion-
properly efficient point is Borwein-properly efficient (see [8, Proposition 1]). 2
Using Theorem 3.1 we can establish PKKT conditions for multiobjective optimiza-
tion problems.

10
Theorem 3.2 Assume that the hypotheses of Theorem 3.1 hold. If x̄ ∈ X is a
Borwein properly efficient solution of Problem (P), then there exist vectors u ∈ R` ,
v ∈ Rm such that
X ` m
X
ui Ofi (x̄) + vj Ogj (x̄) = 0, (21)
i=1 j=1

vj gj (x̄) = 0, j = 1, 2, . . . , m. (22)
u > 0, v = 0.

Proof. Let x̄ be a Borwein properly efficient solution of Problem (P). Then, from
Theorem 3.1 and Tucker’s theorem [16], there exist u ∈ R` , ui > 0 and vj = 0,
j ∈ I(x̄) such that
X ` X
ui Ofi (x̄) + vj Ogj (x̄) = 0.
i=1 j∈I(x̄)

By setting vj = 0 for every j ∈

/ I(x̄), we can write (21). Since gj (x̄) = 0 for j ∈ I(x̄),
we have (22), which completes the proof. 2

Remark 3.2 Figure 1 demonstrates the relationship between proper efficiency and
proper Karush-Kuhn-Tucker conditions, where A:= Convexity assumptions (i.e., X
is convex and the functions fi , gj are convex) and B:= Generalized Abadie regularity
U
condition (7). Here, X − → Y means that X implies Y under assumption U , and
X → − Y means that X implies Y without any additional assumption. The impli-
B B
cations BP E − → KT P E and GP E − → KT P E are Theorem 3.1 and Corollary 3.1,
respectively.
Corollary 14.3 of [31] proves that, when X is polyhedral and fi are linear on Rn ,
then any efficient point is Geoffrion properly efficient. In this setting, efficient points,
Geoffrion properly efficient points and Borwein properly efficient points coincide (see
[28, Theorem 2.45 and 2.48]). We therefore have the following corollary.

Corollary 3.2 Let X be polyhedral and fi be linear on Rn . Let x̄ ∈ X and assume

that GARC (i.e., (7)) holds at x̄. If x̄ ∈ X is an efficient solution of Problem (P),
then there exist vectors u ∈ R` , v ∈ Rm such that
`
X m
X
ui Ofi (x̄) + vj Ogj (x̄) = 0,
i=1 j=1

vj gj (x̄) = 0, j = 1, 2, . . . , m.
u > 0, v = 0.

Proof. If X is polyhedral and fi are linear on Rn then every efficient point is Borwein
proper efficient point (see Remark 3.2). Hence by Theorem 3.2 we can conclude that
PKKT conditions hold. 2

11
 A
BPE GPE

B B
A A

KTPE

PKKT

Figure 1: Relationship between various forms of proper efficiency and PKKT

4 Acknowledgments
We are thankful to Jonathan M. Borwein, for pointing out to us that the results in
[4] only need local proper efficiency, such as the one introduced by him in [8].

References
1. Maciel, S. A., Santos, S. A., Sottosanto, G. N.: Regularity conditions in differ-
entiable vector optimization revisited. J. Optim. Theory Appl. 142, 385-398,
(2009)
2. Bigi, G.: Optimality and Lagrangian regularity in vector optimization. PhD
Thesis, Universita di Pisa (1999)
3. Chankong, V., Haimes, Y. Y.: Multiobjective Decision Making: Theory and
Methodology. North-Holland, Amsterdam (1983)
4. Burachik, R. S., Rizvi, M. M.: On weak and strong Kuhn-Tucker conditions
for smooth multiobjective optimization. J. Optim. Theory Appl., 155(2),
477-491 (2012)
5. Maeda, T.: Constraint qualification in multiobjective optimization problems:
differentiable case. J. Optim. Theory Appl. 80, 483-500 (1994)
6. Kuhn, H. W., Tucker, A. W.: Nonlinear programming, Proceedings of the
Second Berkeley Symposium on Mathematical Statistics and Probability. Uni-
versity of California Press, Berkeley, California, 481-492 (1952)

12
 7. Geoffrion, A. M.: Proper efficiency and the theory of vector maximization, J.
Math. Anal., 22, 618-630 (1968)

8. Borwein, J. M.: Proper efficient points for maximizations with respect to

cones. SIAM J. Control Optim. 15, 57-63 (1977)

9. Chandra, S., Dutta, J., Lalitha, C. S.: Regularity conditions and optimality in
vector optimization, Numerical Functional Anal. Optim. 25, 479-501 (2004)

10. Wang, S. Y.: Second-order necessary and sufficient conditions in multiobjec-

tive programming. Numerical Functional Anal. Optim. 12, 237-252 (1991)

11. Maeda, T.: Second-order conditions for efficiency in non-smooth multiobjec-

tive optimization problems. J. Optim. Theory Appl. 122(3), 521-538 (2004)

12. Li, X. F., Zhang, J. Z.: Stronger Kuhn-Tucker type conditions in nonsmooth
multiobjective optimization: locally Lipschitz case. J. Optim. Theory Appl.
127, 367-388 (2005)

13. Giorgi, G., Jimenez, B., Novo, V.: Strong Kuhn-Tucker conditions and con-
straint qualifications in locally Lipschitz multiobjective optimization problems,
TOP 17(2), 288-304 (2009)

14. Marusciac, I.: On Fritz John type optimality criterion in multiobjective op-
timization. Analyse Numerique et Theorie de I’Approximation, 11, 109-114
(1982)

15. Singh, C.: Optimality conditions in multiobjective differentiable programming.

J. Optim. Theory Appl. 53, 115-123 (1987)

16. Mangasarian, O. L.: Nonlinear Programming. SIAM, Philadelphia (1994)

17. Bigi, G., Pappalardo, M.: Regularity conditions in vector optimization, J.

Optim. Theory Appl. 102, 83-96 (1999)

18. Preda, V., Chitescu, I.: On constraint qualification in multiobjective optimiza-

tion problems: semidifferentiable case. J. Optim. Theory Appl. 100, 417-433
(1999)

19. Bigi, G. and Pappalardo, M.: Regularity conditions for the linear separation
of sets, J. Optim. Theory Appl. 99, 533-540 (1998)

20. Abadie, J.: On the Kuhn-Tucker Theorem, Nonlinear Programming. John

Wiley, New York, 21-36 (1967)

21. Bazaraa, M., Shetty, C. M.: Nonlinear Programming: Theory and Algorithms.
John Wiley and Sons, New York (1979)

13
 22. Bazaraa, M. S., Goode, J. J., Nashed, M. Z.: On the cones of tangents with
applications to mathematical programming. J. Optim. Theory Appl. 13,
389-426 (1974)

23. Giorgi, G., Guerraggio, A.: On the notion of tangent cone in mathematical
programming. Optimization, 25, 11-23 (1992)

24. Giorgi, G., Guerraggio, A.: On a characterization of Clarke’s tangent cone. J.

Optim. Theory Appl., 74(2), 369-372 (1992)

25. Li, X. F.: Constraint qualifications in nonsmooth multiobjective optimization.

J. Optim. Theory Appl., 106(2), 373-398 (2000)

26. Borwein, J. M.: The geometry of Pareto efficiency over cones. Math. Opera-
tionsforschung 11(2), 235-248 (1980)

27. Guerraggio, A., Molho, E., Zaffaroni, A.: On the notion of proper efficiency
in vector optimization. J. Optim. Theory Appl. 82(1), 1-21 (1994)

28. Ehrgott, M.: Multicriteria Optimization. Springer Berlin Hiedelberg, New

York (2005)

29. Rizvi, M. M., Hanif, M., Waliullah, G. M.: First-order optimality conditions
in multiobjective optimization problems: - differentiable case. J. Bangladesh
Math. Soc. (Ganit), 29, 99-105 (2009)

30. Rizvi, M. M., Nasser, M.: New second-order optimality conditions in multi-
objective optimization problems: differentiable case, J. Indian Inst. Sci., 86,
279-286 (2006)

31. White D. J.: Concept of proper efficiency, European J. Oper. Res., 13, 180-188
(1983)

14

View publication stats