MOND-WEIR TYPE DUALITY OF NONLINEAR PROGRAMMING PROBLEM UNDER

THE GENERALIZED CONVEXITY CONDITIONS

PARAS BHATNAGAR
1
, PRASHANT CHAUHAN
2
& AMRITA PAL
3

1
G.L. Bajaj Institute of Management and Technology, Greater Noida, Uttar Pradesh, India
2,3
IFTM University, Moradabad, Uttar Pradesh, India

ABSTRACT
In this paper we will establish duality between the nonlinear programming problem and Mond-weir type dual,
using Hanson-Mond generalized convexity conditions.
KEYWORDS: Pseudo-Convex Functions, Quasiconvex Functions, Nonlinear Programming Problems, Weak and Strong
Duality
INTRODUCTION
Hanson and Mond [5] defined generalized convexity by use of sub-linear functional which satisfy certain
convexity type conditions. Wolfe duality was shown to hold under the assumption that a sub-linear functional exists such
that the Lagrangian satisfies generalized convexity conditions.
Definition.1. A functional F is sublinear over a space S if
(A) ( ) ( ) ( ) S y x y F x F y x F + + +
(B) ( ) ( ) S x R x F x F = 0, , for .
From (B) it follows that F(0)=0.
Definition.2. A differentiable function ) (x defined on some set
n
R X is said to be F-convex if
[ ] ) ( ) ( ) (
2 , 2 1
2 1
x F x x
x x
(1)
X x x
2 1
, and for some arbitrary given sublinear functional F.
Definition 3. A differentiable function ) (x defined on some set
n
R X is said to be F-quasiconvex if
[ ] 0 ) ( ) ( ) (
2 , 2 1
2 1
x F x x
x x
(2)
X x x
2 1
, and for some arbitrary given sublinear functional F.
Definition.4. A differentiable function ) (x defined on a subset
n
R X of is said to be F-pseudo convex if
[ ] ) ( ) ( 0 ) (
2 1 2 ,
2 1
x x x F
x x
(3)
X x x
2 1
, and for some arbitrary given sublinear functional F.
Definition.5. A differentiable function ) (x defined on some set
n
R X is said to be strictly F-pseudo convex if
for X x x
2 1
, ,
2 1
x x ,
[ ] ) ( ) ( ) (
2 1 2 ,
2 1
x x x F
x x
(4)
for some arbitrary given sub linear functional F.
International Journal of Mathematics and Computer
Applications Research (IJMCAR)
ISSN 2249-6955
Vol. 3, Issue 1, Mar 2013, 43-50
TJPRC Pvt. Ltd.
44 Paras Bhatnagar, Prashant Chauhan & Amrita Pal
As pointed out in Mond [8], F-convex functions can be regarded as a generalization of convex functions.
Similarly F-pseudo convex functions and F-quasiconvex functions can be regarded as generalizations of pseudo convex
and quasiconvex functions respectively.
Now consider the following pair of nonlinear programming problems:
(P) Minimize f(x)
Subject to 0; g(x) (5)
(DW) Maximize g(u) y f(u)
t
+ (6)
Subject to 0, g(u) y f(u)
t
= +
0; y (7)
where R R f
n
: ,
m n
R R g : are differentiable functions.
Let S denote the feasible region of (P) i.e., { } 0 ) ( : = x g x S .
Hanson and Mond [5] have established the following results.
Theorem.1. (Weak duality) If for all feasible x in (P) and feasible (u,y) in (DW) there exists a sublinear functional F such
that the Lagrangian g y f
t
+ is F-pseudoconvex, then ) ( ) ( DW Max P Min .
Theorem.2. (Strong duality) If
0
x is a local or global optimal for (P) and a constraint qualification is satisfied at
0
x , then
there exists a
m
R y
0
such that ) , (
0 0
y x is feasible in (DW) and the corresponding values of (P) and (DW) are equal.
If, also, for all feasible x in (P) and feasible (u,y) in (DW), there exists a sublinear functional F such that g y f
t
+ is F-
pseudoconvex, then
0
x and ) , (
0 0
y x are global optima for (P) and (DW) respectively.
Mond-Weir Type Duality
We will establish duality between (P) and the following Mond-Weir dual (DMW).
(DMW) Maximize f(u)
Subject to 0, g(u) y f(u)
t
= + (8)

t
y g(u) 0 , (9)
0 y . (10)
Theorem .3. (Weak duality) If for all feasible x in (P) and feasible (u,y) in (DMW) there exists a sublinear functional F
such that, for feasible (x,u,y) f is F-pseudoconvex and g
t
y is F-quasiconvex , then
Minimum (DMW) Maximum ) ( P .
Proof. Since x is feasible for (P) and (u,y) is feasible for (DMW), we have from (5),(9) and (10) that
0 g(u) y - g(x) y
t t
(11)
And, since g
t
y is F-quasiconvex (3.51) implies

t
,
y ( ) 0
x u
F g u (

(12)
Mond-Weir Type Duality of Nonlinear Programming Problem under the Generalized Convexity Conditions 45
From sublinearity of F we have

[ ]
t t
, , ,
f(u) y ( ) f(u) y ( )
x u x u x u
F g u F F g u ( ( + +

(13)
Also, from (8) and sublinearity of F we obtain
0 )] ( ) f(u [
,
= + u g y F
t
u x
.
Now this and (3.53) yield

[ ]
t
, ,
f(u) y ( )
x u x u
F F g u (


0 . (14)
where (14) follows from (12). Now (14) and F-pseudoconvexity of f yields ) ( ) ( u f x f .
Theorem.4. (Strong duality) Let
0
x solve (P) and assume a constraint qualification is satisfied at
0
x . Suppose also that a
sublinear functional F exists such that f is F-pseudoconvex and g
t
y is F-quasiconvex for all feasible x in (P) and (u,y) in
(DMW). Then there exists a
0
y such that ) , (
0 0
y x solves (DMW) and Minimum (P) = Maximum (DMW) .
Proof. Since
0
x solves (P) and a constraint qualification is satisfied at
0
x , from Kuhn-Tucker conditions there exists a
m
R y
+

0
, such that ) , (
0 0
y x is feasible for (DMW). Clearly the objective functions of (P) and (DMW) are equal, so the
value of (P) equals the value of (DMW) at
0
x . Optimality now follows from weak duality.
We now give an example where Theorem.1 and 2 fail to apply while Theorems 3 and 4 hold.
Example .1
Minimize
3
3
3
3
2
3
1
) ( x x x x x f + +
Subject to 0 1 2 2 ) (
3
2
2
2
3
1
2
1 1
+ + + x x x x x g ,
, 0 1 ) (
2
2 2
+ x x g
. 0 1 ) (
3 3
+ x x g
An optimal solution is attained at ) 1 , 1 , 0 (
0
= x . Now ) 1 , 1 , 0 (
0
= x is feasible for (P) and
) 4 , 1 , 1 , 1 , 0 , 0 ( ) , (
0 0
= y u is feasible for the Wolfe dual, that is
Maximize ) 1 2 2 (
3
2
2
2
3
1
2
1 1 3
3
3
3
2
3
1
+ + + + + + u u u u y u u u u
) 1 ( ) 1 (
3 3
2
2 2
u y u y + +
Subject to
(
(
(

(
(
(

(
(
(

=
(
(
(

1
0
0
0
2
0
0
6 2
6 2
1 3
3
3
3 2 2
2
2 2
2
1 1
1
2
3
2
2
2
1
y u y u u
u u
y
u
u
u
.
Now f(0,-1,1)=3, but 4 ) ( ) (
0 0 0
= + u g y u f
t
, where ) 4 , 1 , 1 ( ), 1 , 0 , 0 (
0 0
= = y u . Hence weak duality does
not hold. Also by considering the third constraint, that is 1 3
2
3 3
+ = u y , we find that the objective function tends to
infinity as
3
u tends to minus infinity for any feasible
2 1 2 1
and , , y y u u . Hence strong duality does not hold.
46 Paras Bhatnagar, Prashant Chauhan & Amrita Pal
Examining
3
3
3
3
2
3
1
) ( x x x x x f + + , we find that f(x) is not pseudo-convex over the feasible region. Hence
Mond-Weir results [10] do not apply between Example.1. and its Mond-Weir dual,
Maximize
3
3
3
3
2
3
1
u u u u + +
Subject to
(
(
(

(
(
(

(
(
(

=
(
(
(

1
0
0
0
2
0
0
6 2
6 2
1 3
3
3
3 2 2
2
2 2
1 1
1
2
3
2
2
2
1
y u y u u
u u
y
u
u
u
,
0 ) 1 2 2 (
3
2
2
2
3
1
2
1 1
+ + + u u u u y ,
0 ) 1 (
2
2 2
u y ,
, 0 ) 1 (
3 3
u y
. 0 , 0 , 0
3 2 1
y y y
However, if we define
[ ] )), ( ) ( ) ( ) ( (
2 1
3
1
,
x g x g u f x f z a z z F
i
t
i u x
+ + |

\
|
=

=

where ) 1 , 1 , 1 ( =
t
a and x is feasible in (P) and u is feasible in (DMW), then
[ ] )) ( ) ( ) ( ) ( )( 2 ) ( 6 ( ) (
2 1
2
3
2
2
2
1 ,
x g x g u f x f u u u u f F
u x
+ + + + + =
)), ( ) ( )( 2 ) ( 6 (
2
3
2
2
2
1
u f x f u u u + + +
Since ) ( g and ) (
2 1
x x g are non-positive for feasible x in (P). Hence
[ ]. ) (
) 2 ) ( 6 (
1
) ( ) (
,
2
3
2
2
2
1
u f F
u u u
u f x f
u x

+ + +

So [ ] ) ( ) ( 0 ) (
,
u f x f u f F
u x
. Therefore, f is F-pseudoconvex for all feasible x and u. Also, since x is feasible
for (P) and (u,y) is feasible for (DMW), we have
0 ) ( ) ( u g y x g y
t t

and

[ ] [ ]
(
(
(
(

|
|
|
|

\
|

= =
1 3
3
3
) ( ) (
2
3
2
2
2
1
, , ,
u
u
u
F u f F u g y F
u x u x
t
u x

)). ( ) ( ) ( ) ( )( 1 3 3 3 1 3 3 3 (
2 1
2
3
2
2
2
1
2
3
2
2
2
1
x g x g u f x f u u u u u u + + + + + =
=0.
Hence, g y
t
is F-quasiconvex for all feasible (x,u,y). So Theorem 3 is applicable to this pair of nonlinear
programs. Now ) 1 , 1 , 0 (
0
= x solves the primal program and for ), 4 ,
2
1
,
2
1
, 1 , 1 , 0 ( ) , (
0 0
= y x ) , (
0 0
y x is feasible
and hence optimal in (DMW) with equality of objective functions.
We now give general results which subsume the different Mond-Weir type duals for the primal problem (PE).
Mond-Weir Type Duality of Nonlinear Programming Problem under the Generalized Convexity Conditions 47
(P) Minimize f(x),
Subject to . 0 ) ( x g (15)
h(x)=0. (16)
The general Mond-Weir dual to (PE) is (DEG) [10].
(DEG) Maximize ) ( ) ( ) (
0 0 0 0
u h z u g y u f
J
t
J I
t
I
+ +
Subject to 0 ) ( ) ( ) ( = + + u f u h z u g y
t t
(17)
, 0 ) ( ) ( + u h z u g y
J
t
J I
t
I


r ,..., 2 , 1 = , (18)
, 0 y (19)
Where
k n m n n
R R h R R g R R f : , : , : are differentiable functions.
'
, I J

,
0,1, 2,..., r = are partitions of the sets M={1,2,3,m}, K={1,2,3,k} respectively. Also { }
2 1
, max r r r = where
1
r and
2
r is the number of partitions of M and K respectively and { }
2 1
, Min for or r r I J > = =

.Here

I
y denotes the vector consisting of the components
i
y of y such that

I i . Similar meanings apply to

J J I
y z g and , .
Theorem.5. (Weak Duality) If all feasible x in (PE) and (u,y,z) in (DEG) there is a sublinear functional
u x
F
,
such
that
0 0 0 0
J
t
J I
t
I
h z g y f + + is F-pseudoconvex and

J
t
J I
t
I
h z g y + , r ,..., 2 , 1 = is F-quasiconvex, then Minimum
(PE) Maximum(DEG) .
Proof. Since x and (u,y,z) are feasible
r u h z u g y x h z x g y
J
t
J I
t
I J
t
J I
t
I
,..., 2 , 1 , 0 ) ( ) ( ) ( ) ( = +

(20)
And since

J
t
J I
t
I
h z g y + is F-quasiconvex, r ,..., 2 , 1 = , we have
[ ] , ,..., 2 , 1 , 0 )) ( ) ( (
,
r u h z u g y F
J
t
J I
t
I u x
= +

(21)
And since
u x
F
,
is sublinear we have
[ ]

= =
+
(

+
r
J
t
J I
t
I u x
r
J
t
J I
t
I u x
u h z u g y F u h z u g y F
1
,
1
,
)) ( ) ( ( )) ( ) ( (



0 (22)
where (22) follows from (21). Also from the equality constraint of (DEG) i.e. (17) and sublinearity of
u x
F
,
we have
[ ] [ ] )) ( ) ( ( )) ( ) ( ) ( (
0 0 0 0 0 0 0 0
\ \ \ \ , ,
u h z u g y F u h z u g y u f F
J K
t
J K I M
t
I M u x J
t
J I
t
I u x
+ + + (23)

But

=
+ = +
r
J
t
J I
t
I
t
J K I M
t
I M
h z u g y z g y
1
\ \ \
). ) ( (
0 0 0 0 0 0 0
Therefore (22) and (23) yield
48 Paras Bhatnagar, Prashant Chauhan & Amrita Pal
[ ] 0 )) ( ) ( ) ( (
0 0 0 0
,
+ + u h z u g y u f F
J
t
J I
t
I u x
(24)
and since
0 0 0 0
J
t
J I
t
I
h z g y f + + is F-pseudoconvex, then from (24) we have
) ( ) ( ) ( ) ( ) ( ) (
0 0 0 0 0 0 0 0
u h z u g y u f x h z x g y x f
J
t
J I
t
I J
t
J I
t
I
+ + + +
and since 0 ) ( ) (
0 0 0 0
+ x h z x g y
J
t
J I
t
I
we have
) ( ) ( ) ( ) (
0 0 0 0
u h z u g y u f x f
J
t
J I
t
I
+ + .
Theorem.6. (Strong duality) If
0
x is a local or global minimum of (PE) at which a constraint qualification is satisfied, then
there exists a ) , (
0 0
z y such that ) , , (
0 0 0
z y x is feasible for (DEG) and the corresponding values of (PE) and (DEG) are
equal. If also there exists a sublinear functional
u x
F
,
such that
0 0 0 0
J J I
t
I
h z g y f + + is F-pseudoconvex and
r h z g y
J
t
J I
t
I
,..., 2 , 1 , = +

is F-quasiconvex for all feasible (x, u, y, z) then
0
x and ) , , (
0 0 0
z y x are global optimal
solutions of (PE) and (DEG) respectively.
Proof. Since
0
x solves (PE) either locally or globally and a constraint qualification is satisfied; then by the Kuhn-Tucker
conditions [6], there exists a
m k
R z R z
+

0 0
and such that

. 0 and 0 ) ( , 0 ) (
, 0 ) ( ) ( ) (
0 0 0 0
0 0 0 0 0
=
= + +
y x g x g y
x h z x g y x f
t
t t

From 0 ) (
0
x g , 0
0
y and 0 ) (
0 0
= x g y
t
we conclude that
0 0
( ) 0
i i
y g x = ,
1, 2,..., i m = and since 0 ) (
0
= x h we must have
r. 0,1,2,..., , 0 ) ( ) (
0 0 0 0
= = +

x h z x g y
J
t
J I
t
I

Hence ) , , (
0 0 0
z y x is feasible in (DEG) and the values of (PE) and (DEG) are equal. Now since there exists a sublinear
functional
u x
F
,
such that
0 0 0 0
J
t
J I
t
I
h z g y f + + is F-quasiconvex and

J
t
J I
t
I
h z g y + is F-quasiconvex 1,2,...,r =
for all feasible (x,u,y,z), optimality follows from weak duality.
CONCLUSIONS
Mond and weir gave a dual for (P), different from the Wolfe dual where the convexity requirements for duality
are considerably weakened. Here we presented Mond-weir type duality under the Hanson and Mond generalized convexity
conditions.
REFERENCES
1. Bector, C.R. ( 1968). Duality in fractional and indefinite programming, ZAMM, Vol.48, 418-420.
2. Bector C.R.( 1973). Duality in nonlinear fractional programming, Z Operations Research Series A, Vol.17, 183-
193.
3. Bector C.R. and Bector M.K.,(1987). On various theorems in nonlinear programming,Journal of Optimization
Theory and Application, Vol. 53, No.3, 509-515.
Mond-Weir Type Duality of Nonlinear Programming Problem under the Generalized Convexity Conditions 49
4. Gulati T.R., and Craven, (1983). B.D., A strict converse duality theorem in nonlinear
programming,J.Inform.Optim.Sci.4, 301-306.
5. Hanson M.A. and Mond, B., (1982). Further generalizations of convexity in mathematical programming,
J.Inform. Optim. Sci.3, 25-32.
6. Kuhn H.W. and Tucker, A.W., (1951). Nonlinear Programming, Prov. 2
nd
Berkeley Symposiu7m on
Mathematical Statistics and Probability, (University of California Press), 481-492.
7. Mangasarian, O.L.,( 1969). Nonlinear programming (McGraw-Hill, New York, London, Sydney).
8. Mond, B., (1983). Generalized convexity in mathematical programming, Bull. Austral. Math. Soc.27, 185-202.
9. Mond B. and Egudo, R.R., (1985). On strict converse duality in nonlinear programming, J. Inform. Optim.
Sci.6, 113-16.
10. Mond B. and Weir, T.,( 1981). Generalized concavity and duality, in Generalized concavity in optimization and
economics (eds. S. Schaible and W.T. Ziemba), (Academic Press), 263-279.
11. Weir, T.,( 1982). Generalized convexity and duality in mathematical programming, Ph.D. thesis, La Trobe
University, Bundoora, Melbourne, Australia.
12. Wolfe, P., (1961). A duality theorem for nonlinear programming, Quart. Appl. Math.19, 239-244.