CONSTRAINT QUALIFICATIONS FOR NONLINEAR PROGRAMMING
RODRIGO G. EUST
´
AQUIO
§
, ELIZABETH W. KARAS
¶
,
AND
ADEMIR A. RIBEIRO
¶
Abstract.
This paper deals with optimality conditions to solve nonlinear programming problems. The classical
KarushKuhnTucker (KKT) optimality conditions are demonstrated through a cone approach, using the well known
Farkas’ Lemma.
These conditions are valid at a minimizer of a nonlinear programming problem if a constraint
qualification is satisfied. First we prove the KKT theorem supposing the equality between the polar of the tangent
cone and the polar of the first order feasible variations cone. Although this condition is the weakest assumption, it
is extremely difficult to be verified. Therefore, other constraints qualifications, which are easier to be verified, are
studied, as: Slater’s, linear independence of gradients, MangasarianFromovitz’s and quasiregularity. The relations
among them are discussed.
Key words.
Optimality conditions, KarushKuhnTucker, constraint qualifications.
1. Introduction.
We shall study the nonlinear programming problem
(
P
)
minimize
f
(
x
)
subject to
h
(
x
) = 0
g
(
x
)
≤
0
,
where the functions
f
: IR
n
→
IR,
g
: IR
n
→
IR
p
and
h
: IR
n
→
IR
m
are continuously differentiable.
The feasible set is Ω =
{
x
∈
IR
n

h
(
x
) = 0
,
g
(
x
)
≤
0
}
.
Given
x
*
∈
Ω, the classical
KarushKuhnTucker
(KKT) conditions say that there exist
La
grangian
multipliers
λ
*
∈
IR
m
and
μ
*
∈
IR
p
such that:
∇
f
(
x
*
) =
m
X
i
=1
λ
*
i
∇
h
i
(
x
*
) +
p
X
j
=1
μ
*
j
∇
g
j
(
x
*
)
,
μ
*
j
≥
0
,
j
= 1
, . . . , p,
μ
*
j
g
j
(
x
*
) = 0
,
j
= 1
, . . . , p.
In nonlinear programming we would like that KKT are necessary conditions for a given point to
be a solution to the problem. When the problem is unconstrained (Ω = IR
n
), the KKT conditions
reduce to
∇
f
(
x
*
) = 0 which is a necessary optimality condition. However, this not always happens,
as shown in the following example.
Example 1.
Consider the problem
(
P
)
with
f
: IR
2
→
IR
and
g
: IR
2
→
IR
2
defined by
f
(
x
) =
x
1
and
g
(
x
) = (
x
2

(1

x
1
)
3
,

x
2
)
T
. Note that
x
*
= (1
,
0)
T
is a minimizer of the problem
but the KKT conditions do not hold.
In this paper we shall discuss assumptions on the constraints in order to ensure that the
KKT conditions hold at a minimizer. Such an assumption is called
constraint qualification (CQ)
.
Formally, we say that the constraints
h
(
x
) = 0 and
g
(
x
)
≤
0 satisfy a constraint qualification at
x
*
∈
Ω when, given any differentiable function
f
minimized at
x
*
with respect to Ω, the KKT
conditions are valid.
Several authors have obtained different constraint qualifications. In this work, we will discuss
many of them as well as some relations between them. A special interest is devoted to show the
weakest such qualification. In this context, the concept of cones and their polars will be useful.
§
Master Program in Numerical Methods in Engineering, Federal University of Paran´a, Cx. Postal 19081, 81531
980, Curitiba, PR, Brazil; email :
[email protected]
.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
This is the end of the preview.
Sign up
to
access the rest of the document.
 Fall '10
 riley
 Linear Algebra, Vector Space, Mathematical optimization, lagrange multipliers, tk, Karush–Kuhn–Tucker conditions

Click to edit the document details