International Journal of Engineering Research & Science (IJOER)

ISSN: [2395-6992]

[Vol-2, Issue-5 May- 2016]

Gap Functions and Error Bound to Set-Valued Variational

Inequalities
Yuexian Li
Department of Mathematics and Statistics, Inner Mongolia Agricultural University, P.R.China

Abstract In this paper, the gap function for set-valued variational inequalities is introduced and the finiteness of the gap
function is discussed. Furthermore, under -strongly monotone condition, we obtain error bounds for set-valued variational
inequalities, i.e. upper estimates forthe distance to the solution set of the variational inequalities.
Keywords variational inequalities, gap functions, set-valued maps, error bounds.

I.

INTRODUCTION

A vector variational inequality (for short,VVI) in a finite-dimensional Euclidean space was introduced first by Giannessi [1]
in 1980.In the recent years, variational inequalities have become a very popular field of research in optimization theory.
From the computational point of view, one important research direction in variational inequalities is the study of gap
functions. One advantage of the introduction of gap functions in variational inequality is that variational inequalities can be
transformed into optimization problem. Thus, powerful optimization solution methods and algorithms can be applied to find
solutions of variational inequalities. Meanwhile, the gap functions can be used to devise error bounds for variational
inequalities. There have been many results regarding gap functions and error bounds for classical variational inequalities
see[2-5].
In this paper, we are interested in studying variational inequalities with set-valued maps. The solution of the variational
inequalities with set-valued maps is a natural extension of the classic generalized variational inequalities studied in[6,7].
These kind of variational inequalities arise as generalization of optimality conditions for a constrained optimization problem
with non-smooth objective function. See for example[8], where some equivalence of some particular set-valued VVI and a
nondifferentiable and nonconvex vector optimization problem is established. Our aim in this article is to construct gap
functions which can be used to devise error bounds for variational inequalities with set-valued maps.
Given a set-valued map T : R R and a nonempty and convex subset
n

x C and t T ( x )

C of Rn ,

then VI (T , C ) consists in finding

such that

t , y x 0,

y C .

This problem is also referred to as Stampacchiavariational inequality defined by T and

(1.1)

C .on the vector y C , this leads us

to define what is called the weak Stampacchiavariational inequality,

denoted by VI w (T , C ) which consists in finding

x C

such that, y C , t ( y ) T ( x ) satisfying

t ( y ), y x 0,

y C

(1.2)

We shall denote by S (T , C ) the solution set of the VI (T , C ) , that is

x S (T , C ) x * T ( x ) such that x * , y x 0, y C ,
while S w (T , C ) stands for the solution set of the VI w (T , C ) :

x Sw (T , C ) y C , x y* T ( x ) such that x y* , y x 0.
Clearly one alwayshas S (T , C ) S w (T , C ) .

Page | 60

International Journal of Engineering Research & Science (IJOER)

II.

ISSN: [2395-6992]

[Vol-2, Issue-5 May- 2016]

GAP FUNCTIONS

In this section,we introduce the concept of gap functions for VI (T , C ) and VI w (T , C ) .

Definition 2.1. h : C R R is said to be a gap functions of the VI (T , C ) or the
n

VI w (T , C ) if
() h( x) 0,

x C ;

() 0 h( x ) if and only if x is a solution of the VI (T , C ) or the VI w (T , C ) .

Given a set-valued map T : R R , for any
n

the normalized operator T defined by:

if 0 T ( x)
0

T
1

*
*
x
:
x

T
(
x
)
, otherwise

In [9], Yang and Yao introduced the gap function and established necessary and sufficient conditions for the existence of a
solution for variational inequalities with set-valued maps. In [10], D.Aussel. J.Dutta first introduced a simple gap function.

g ( x) : sup *inf x* , x y
yC x T ( x )

In order to prove error bounds, it is important to know whether the gap function is finite or not. If
functions is finite. In order to overcome this boundedness

C is bounded, the gap

Assumption about C , D.Aussel. J.Dutta again introduced a class of gap functions.

g ( , ) ( x) : sup[ *inf x* , x y
yC x T ( x )

for any

1
2
x y ]
2

0 and 0 .

Inspired and motivated by the above research work, in this paper ,the new gap functions for VI (T , C ) and VI w (T , C ) are
introduced.
Definition 2.2. An operator T : R R is
n

strongly monotone, if there exists 0 such that, for any x, y R n and

any x T ( x), y T ( y ) ,one has

*

y* x* , y x y x .
2

Proposition 2.1. If T : R R is
n

(2.1)

strongly monotone with 0 and C is a nonempty convex set, then the

VI (T , C ) and the VI w (T , C ) have at most one solution.

Proof. If x and

all is solution of VI (T , C ) , then t T ( x ), t T ( x ) such that

t , y x 0, y C and t , y x 0, y C.

There is a y x such that t , x x 0 and a y

x such that t , x x 0 .Combining above two inequalities, we

have
Page | 61

International Journal of Engineering Research & Science (IJOER)

ISSN: [2395-6992]

[Vol-2, Issue-5 May- 2016]

t t , x x 0 .
strongly monotone with 0 , it follows that,

Since T : R R is
n

2
t t , x x x x 0 ,
and consequently,
Let

x x .With the same proof as above one can obtain same consequence about VI w (T , C ) .

: R n R n R be a real-valued function. We make the following hypothesis.

Assumption 2.1.
() ( x, y ) is continuously differentiable in the second argument;
()

( x, y ) 0, ( x, y ) R n R n ;

() for all x R ,
n

( x, y ) is

0 ,for all x R

uniformly strong convex in the second argument with respect to constant

0;

i.e.,

( x, y1 ) ( x, y ) 2 ( x, y2 ), y1 y2 y1 y2 , y1 , y2 R n ;
2

()

( x, y ) 0 if and only if x y ;

() 2 ( x, y ) is Lipschitz in the second argument with Lipschitz constant

2 ( x, y1 ) 2 ( x, y2 ) L y1 y2 ,

L ; i.e.,

y1 , y2 C.

The following examples show that Assumption 2.1 is satisfied by some functions.
Example 2.1. ( x, y) k x y

(k 0).

Example 2.2. ( x, y ) x y, B ( x)( x y ) ,

Where B ( x ) is positive definite symmetric matrices and B ( x) is uniformly continuous differentiable.
Lemma 2.1. ([11]) Let Assumption 2.1 () - () hold, then 2 ( x, y ) 0 if and only if x y .
Proposition 2.2. Let Assumption 2.1 hold, then,

x y ( x, y) ( L ) x y , x, y C.
2

Proof. In Assumption 2.1 (), let y1 y, y2 x and hence,

( x, y) ( x, x) 2 ( x, x), y x y x .
2

By Assumption 2.1(), ( x, x ) 0 and by Lemma 2.1 , 2 ( x, x) 0 .Hence,

x y ( x, y), x, y C.
2

In Assumption 2.1 (), let y1 x, y2 y and hence,

( x, x) ( x, y) 2 ( x, y), x y x y .
2

Page | 62

International Journal of Engineering Research & Science (IJOER)

ISSN: [2395-6992]

[Vol-2, Issue-5 May- 2016]

By Assumption 2.1() and Lemma 2.1 , we have that

( x, y ) 2 ( x, y ), x y x y

2 ( x, y ) 2 ( x, x), x y x y

2 ( x, y ) 2 ( x, x) x y x y .
2

L x y x y ( L ) x y
2

Definition 2.3. A set-valued map T : R R is said to be directionally closed if and only if there
n

any

such that for

x R , T ( x) is closed.
n

Consider VI (T , C ) and VI w (T , C ) . We define the following function:

G ( , ) ( x) : sup[ *inf x* , x y ( x, y)]

yC x T ( x )

for any

0 and 0 .

Proposition 2.3. Let the set-valued map T : R R be directionally closed with nonempty values and let
n

convex subset. Then for any

0 and 0 , the function G( , ) ( x)

C be a closed

is a finite-valued function on R .

Proof. Let ( x, y ) *inf x , x y and we immediately obtain

*

x T ( x )

G ( , ) ( x) sup[ ( x, y ) ( x, y )].
yC

While x being fixed, the function y ( x, y ) is concave and further T has non-empty closed values when T is
directionally closed with nonempty values. By T definition, we have that T is bounded, thus T is compact-valued. Hence
the function y ( x, y ) is finite.
Moreover the function y ( x, y ) ( x, y ) is uniformly strong convex and thus it attains a unique minimum over

when

is closed.

Remark 2.1. For any x R , any

n

and any

0 let us denote by y( , ) ( x) the unique global minimum of the

function y ( x, y ) ( x, y ) on the closed constraint set

Lemma 2.2. ([10]) Let T be any set-valued map from
assertions are equivalent:

C.

Rn to Rn and C is a nonempty subset of Rn . Then the following

() x S (T , C ) ;
() there exists
() for any

0 , such that x S (T , C ) ;

0 , x S (T , C ) .

Theorem 2.1. Let the set-valued map T : R R be directionally closed with nonempty values and let
n

convex subset. Then the function

Proof. It is simple to observe that

C be

a closed

G( , ) ( x) is a gap function for the VI w (T , C ).

G( , ) ( x) 0 for all x C just by setting x y .

Page | 63

International Journal of Engineering Research & Science (IJOER)

Let us now assume that G( , ) ( x ) 0 , there

ISSN: [2395-6992]

[Vol-2, Issue-5 May- 2016]

x C . i.e.,

G ( , ) ( x ) sup[ *inf x* , x y ( x , y )] 0.
yC x T ( x )

Thus,

inf x* , x y ( x , y) 0, y C.

x T ( x )
*

By Proposition 2.2, we have

( x , y) ( L ) x y , y C.
2

Hence

inf x* , x y ( x , y) ( L ) x y , y C.
2

x*T ( x )

Let us consider a fixed but arbitrary point x0 C . Consider the sequence of vector

1
yk x ( x0 x ),
k
Since

C is convex it is clear that yk C

for each

yk given as

k N.

k N . Therefore , for any k ,

inf x* , x yk ( x , yk ) ( L ) x yk .
2

x T ( x )
*

Combining above two inequalities, we have

inf x* , x x0

x*T ( x )

As

1
2
( L ) x x0 .
k

k we have
inf x* , x x0 0.

x*T ( x )

Since T is compact-valued, there exists

x0* T ( x ) such that

x0* , x x0 *inf x* , x x0 0.
x T ( x )

Hence

x0* , x0 x 0.
Since

x0 C is arbitrary, it is clear that x S w (T , C ) .

Conversely, let us now assume x S w (T , C ) . By Lemma 2.2, we obtain x S w (T , C ) , that is, for

y C , x y* T ( x ) such that
x y* , y x 0.
This immediately implies that
Page | 64

International Journal of Engineering Research & Science (IJOER)

ISSN: [2395-6992]

[Vol-2, Issue-5 May- 2016]

G( , ) ( x) 0.
This completes the proof.
Theorem 2.2.Let the set-valued map T : R R
n

convex subset. Then x S w (T , C ) if and only if

Proof. From the definition of

be directionally closed with nonempty values and let

C be a closed

x y( , ) ( x ) for any 0 and 0 .

y( , ) ( x) , if x y( , ) ( x ) , we immediately obtain

G ( , ) ( x ) sup[ ( x , y ) ( x , y )] ( x , x ) ( x , x ) 0.
yC

By Theorem 2.1 , we have x S w (T , C ) .Now let us suppose, for a contradiction, that

x S w (T , C ) and that

x y( , ) ( x ) .By x S w (T , C ) , for each y C ,there exists x y* T ( x ) such that x y* , y x 0. Hence, for each
y C , ( x , y ) 0 . By Assumption 2.1(), while x y( , ) ( x ) we have ( x , y( , ) ( x ) 0 . Since

G ( , ) ( x ) ( x , y( , ) ( x )) ( x , y( , ) ( x )),
thus we have

G( , ) ( x ) 0 , which is a contradiction with Theorem 2.1.

Remark 2.2.Theorem 2.2 provide an interesting fixed point characterization of the solutions of VI w (T , C ) .
3.Error Bounds
Theorem 3.1. Suppose that T : R R is
n

strongly monotone with 0 and let x C be the unique minimum of

VI w (T , C ) .Then, for any 0 and 0 , such that ( L ) 0 , one has for any x C ,

xx

1
G
( x) .
( L ) ( , )

Proof. By x S w (T , C ) , for each y C ,there exists

x y* T ( x ) such that

x y* , y x 0.
Thus, for x T ( x) , we have
*

x* , x x x* xy* , x x x x .
2

By Proposition 2.2, we have

( x , y) ( L ) x y , y C.
2

Hence,

G ( , ) ( x) sup[ *inf x* , x y ( x, y )]
yC x T ( x )

*inf x* , x x ( x, x )
x T ( x )

x x ( L ) x x
2

[ ( L )] x x

.
2

Therefore, we have
Page | 65

International Journal of Engineering Research & Science (IJOER)

xx

ISSN: [2395-6992]

[Vol-2, Issue-5 May- 2016]

1
G
( x) .
( L ) ( , )

ACKNOWLEDGEMENT
The work Supported by the Colleges and Universities Scientific Research Project of Inner Mongolia (NJZY12079 and
NJZC13088).

REFERENCES
[1] F.GIANNESSI, Theorems of Alternative, Quadratic Programmes and Complementarity Problems, in: R. W. Cottle, F.Giannessi,
J.L.Lions(Eds.), Variational Inequalities and Complementarity Problems, Wiley, Chichester, 1980,pp.151-186.
[2] Auslender A. Optimization Methods Numeriques. Masson: Paris,1976.
[3] Fukushima M. Equivalent Differentiate Optimization Problems and Descent Methods for Asymmetric Variational Inequality
Problems. Mathematical Programming, 1992,53:99-110.
[4] Solodov M V. Merit Functions and Error Bounds for Generalized Variational Inequalities. Journal of Mathematical Analysis and
Applications, 2003,287:405-414.
[5] Noor M A. Merit Functions for General Variational Inequalities. Journal of Mathematical Analysis and Applications, 2006,316:736752.
[6] BROWDER, F.E.,HESS,P.,Nonlinear Mappings of Monotone Type in Banach Spaces,Journal of Functional Analysis. Vol.11,pp.251294,1972.
[7] KRAVVARITTIS, D., Nonlinear Equations and Inequalities in Banach Spaces, Journal of Mathematical Analysis and
Applications,Vol.67,pp.205-214,1979.
[8] ANSARI,Q.H., and YAO,J.C., Nondifferentiable and Nonconvex Optimization Problems, Journal of Optimization Theory and
Applications, Vol.106, pp.475-488,2000.
[9] X. Q. YANG, J. C. YAO, Gap Functions and Existence of Solutions to Set-Valued Vector Variational Inequalities, Journal of
Optimization Theory and Applications, Vol.115, pp.407-417,2002.
[10] D.Aussel.J.Dutta, On Gap Functions Multivalued StampacchiaVariational Inequalities,Journal of Optimization Theory and
Applications, Vol.149, pp.513-527,2011.
[11] Yamashita N., Taji K., Fukushima M. Unconstrained Optimization Reformulations of Variational Inequality Problems, Journal of
Optimization Theory and Applications, Vol.92, pp.439-456,1997.

Page | 66