NOT FOR QUOTATION WITHOUT THE PERMISSION OF THE AUTHOR
ON THE ANTI-HONOTONICITY OF DIFFERENTLAL MAPPINGS CONNEClXD WITH GENERAL
EQUJLIBRIUM PROBLM
S.P. Urias'ev
Working P a p e r s a r e interim r e p o r t s on work of t h e International Institute f o r Applied Systems Analysis and have received only limited review. Views o r opinions expressed h e r e i n do not necessarily r e p r e s e n t those of t h e Institute o r of its National Member Organizations.
INTERNATIONAL INSTITUTE FOR APPLIED SYSTEMS ANALYSIS A-2361 Laxenburg, Austria
FOREWORD
This p a p e r i s concerned with t h e anti-monotonicity of differential mappings connected with g e n e r a l equilibrium problems. These r e s u l t s c a n be used f o r the in- vestigation of different game t h e o r y problems, f o r example Nash equilibria f o r noncooperative n - p e r s o n games. Such an approach gives possibility t o c o n s t r u c t r e c u r r e n t algorithms f o r finding t h e equilibria point.
This r e s e a r c h w a s conducted within the framework of t h e Adaptation and Op- timization P r o j e c t in t h e System and Decision Sciences Program.
Alexander B. K u n h a n s k i Chairman System and Decision Sciences Program
AUTHOR
Senior research scientist Stanislav Urias'ev from the Institute of Cybernetics Academy of Sciences of the Ukr. SSR (252207 Kiev, USSR) worked on stochastic optimization and game theory and wrote this paper during his t w o week visit a t IIA- SA.
CONTENTS
1 Introduction
2 Weakly Convex Functions
3 On the Anti-Monotonicity of the Dierential Maps for the Weakly Convex Functions 4 The Monotonicity of Dierential Map f o r
Quasi Convex-Concave Functions References
ABSTRACT
L e t X be a s u b s e t o f a Hilbert s p a c e H a n d ck:X X X
- + R ,
(k(z, z ) = O f o r a l l z E X . Let G(z)= ay
ck(x,v)ly ==
denote generalized d i e r e n t i a l with respect to t h e second argument a t t h e point ( z , z ) . W e s h a l l b e concerned with t h e p r o p e r t i e s of t h e function ck sucient to e n s u r e t h e anti-monotonicity of t h e map G(x). I t will b e shown t h a t f o r t h e anti-monotonicity of t h e map G(x) i t i s sucient to assume convexity-concavity of t h e function ck. In t h e case of t h e weakly convex-concave function ck t h e map G(x) i s anti-monotone under some conditions o n t h e remainder terms. In t h e case of t h e quasi convex-concave function ck, t h e condition similar to t h e anti-monotonicity condition hold.Some p r o p e r t i e s of t h e weakly convex functions used in a r t i c l e will b e proved.
-
vii-
ON THE ANTI-MONOTONICITY OF DFFEFWMTIAL MF'PINGS CONNECTED WITH GENERAL
EQUILIBRIUM PROBLEM
SP. Urias'evThe mathematical problems discussed in this a r t i c l e were stimulated by t h e in- vestigations of simulation model f o r international oil t r a d e (SMIOT) developed at t h e International Institute f o r Applied Systems Analysis [ I ] .
Briey t h e main idea of t h i s model is t h e following. T h e r e i s a market of a single homogeneous p r o d u c t , which consists of some sellers ( e x p o r t e r s ) and a single b u y e r (importer). Let i
=
1,. . . ,
n be t h e e x p o r t e r s , f i (2) b e t h e marginal c o s t of which any e x p o r t e r i p r o d u c e s t h e amount z of t h e product f o r marketing andr ( z ) b e t h e p r i c e at which t h e i m p o r t e r would agree t o buy t h e amount z of t h e product. If z ; denotes t h e amount of t h e p r o d u c t sold by e x p o r t e r i , then t h e r e v e n u e qi ( z ) of t h e e x p o r t e r i , can be e x p r e s s e d as follows:
Let s b e a number of a time point, zf b e t h e amount of t h e product sold by ex- p o r t e r i at time s . The dynamics of t h e model i s given by t h e relation
xt
+l=
max 0,zt -
p,t
a q ia
zi ( z S )I , S =
0 , 1 , ...
where ps , s
=
0 , 1 , .. .
are t h e positive s c a l a r values.In more g e n e r a l form
where nX(-) denotes t h e o p e r a t i o n of projection on feasible set
is a d i e r e n t i a l map of p r e f e r e n c e .
The aim of t h e study is to formulate assumptions on t h e functions r ( 2 ) . f i ( z ) , i
=
1 ,. . .
, n s u c h t h a t t h e p r o c e s s (1) h a s a s t a b l e cycle, converges to a Nash equilibrium point o r converges to some point. The r e s u l t s of t h i s a r t i c l e allow us to formulate conditions o n t h e functions r ( z ) , f i ( z ) , i=
1,. . .
, n insuring the con- v e r g e n c e of p r o c e s s (1) to a Nash equilibrium point. These questions were dis- cussed in p a p e r s [3]- 151
a n d i n more g e n e r a l situations in 161, P I , etc.We study some conditions on t h e payo functions sucient f o r t h e anti- monotonicity of mapping g ( z ) o r o t h e r s d i e r e n t i a l mappings. Hence we c a n formu- l a t e t h e conditions under which many i t e r a t i v e algorithms f o r s e a r c h equilibrium points are applicable (see, f o r example, [8]
-
1141).STATEMENT OF THE
PROBLEM
Let X be a s u b s e t of a Hilbert s p a c e H and<., .>
denote t h e i n n e r product. W e s a y t h e multivalued mapping G i s anti-monotone o n X i f < g ( z ) - g ( y ) , y - z > Z O f o r a l l z , y E X , g ( z ) E G ( z ) , g ( y ) € G ( y ) .Let Y:X XX
-
R b e a quasi o r weakly1151
convex-concave function, Y ( 2 , z )=
0 f o r a l l z E X. Let G ( 2 )=
8Y Y ( z . y ) &.,
denote generalized d i e r e n t i a l of t h e function Y with r e s p e c t t o t h e second argument at the point( 2 8 2 ) .
W e shall be concerned with t h e p r o p e r t i e s of t h e function Y sucient t o e n s u r e t h e anti-monotonicity of t h e map G ( z ) . I t will be shown t h a t f o r t h e anti- monotonicity of t h e map G ( z ) , i t i s sucient t o assume convexity-concavity of t h e function Y. In the c a s e of t h e weakly convex-concave function
*
the map G ( z ) i s anti-monotone under some conditions on the remainder terms. Some p r o p e r t i e s of t h e weakly convex functions used in a r t i c l e will b e proved. Investigation of such questions c a n b e motivated by t h e problem of nding t h e pointsz*
E X dened by variational inequalitywhere g ( z * ) € ~ ( z * ) .
If t h e multivalued map G ( z ) i s anti-monotone as s t r i c t l y anti-monotone w e c a n use r e s u l t s s t a t e d in [ 8 ]
-
[14], etc. f o r solving t h e problem (2).In t h e case of t h e quasi convex-concave function 9 ( 2 , y ) a n d 9 ( z , z )
=
0 f o r z E X t h e weaker condition t h e n t h e anti-monotonicity holds:<gr(z),z* - z > r O f o r a l l z ~ X , g ( z ) E G ( ~ ) ,
where G ( z ) i s a d i e r e n t i a l of quasi concave function 9 ( z , z ) with r e s p e c t to second argument.
Let us c o n s i d e r some problems which c a n b e r e d u c e d to t h e variational ine- quality (2).
EXAMPLE 1 N a s h e q u i l i b r i a f o r n o n c o o p e r a t i v e n - p e r s o n games. Let X b e a convex closed bounded s u b s e t of t h e production H I X XHn of a Hilbert s p a c e s H i , i
=
1,. . . ,
n . A point zi E Hi i s a s t r a t e g y of i -th p l a y e r i=
1,. . . ,
n and cpi(z)=
cpi ( z l ,. . .
,z,)
i s his payo function. The element (z1,. . .
, zi* * *
yi
,
zi +l , . . ,
zi ) i s denoted by (yi / z ). The point z=
( z l, . . .
, zn ) E X i s re- f e r r e d to as t h e Nash equilibrium of n - p e r s o n game if f o r i=
1,. . .
, nLet u s introduce t h e function 9 ( z
,
y ) :I t i s not d i c u l t to see t h a t '3) ( z , z )
=
0, z E X . W e suppose t h a t t h e functionsvi
( z ),i
=
1,. . . ,
n are continuous on X. The point z* E X i s dened as t h e normalized equilibrium point ifLEMMA 1 (See f o r example [16]). The normalized equilibrium point i s t h e equilibrium point, t h e r e v e r s e is t r u e if X
=
X l x- . X X , , &
C Hi.The condition ( 2 ) i s a n e c e s s a r y optimality condition f o r t h e problem (3), f o r t h i s r e a s o n t h e problem of nding Nash equilibrium is reduced to t h e problem ( Z ) ,
EXAMPLE 2 An equilibrium point is dened in [17] as follows. Let X be a con- vex closed subset of a n Euclidean space R n , the functions
be concave with r e s p e c t to y E X f o r each z E X and continuous with r e s p e c t t o
z ,
y on X X X . Let us denoteThe problem consists in nding the point z * such t h a t
* *
z * E x ( z * ) ; max ( Q , ( z * , y ) : y c ~ ( z * ) j
=
Q,(z . I.
If the condition X ( z ) = X is t r u e f o r all z f X I then t h e problem i s reduced t o problem (3), where Q ( z , y )
=
Q o ( z , y )-
Q , ( z , z ) . It is easy to s e e 9 ( z , z ) = O for all z f X. The necessary condition ( 2 ) can be used in this case too.EXAMPLE 3 Let us consider one more problem which can be reduced to (3).
We assume t h e X is a subset of a Hilbert space
H
(or m o r e general space), and 9 : X X X -4 R i s a function satisfying sup Q ( y , y ) S 0. The problem consists inY EX nding t h e point z
*
E X such t h a ts u p Q ( z * . y )
s o .
Y E X
We suppose 9 ( z
,
y )=
4 ( z , y )-
Q ( z , z ) . If the point z*
Ex
satises (3), then i t satises (4) too. For this reason the necessary condition (2) c a n be used in this case as w e l l .Theorems concerning t h e existence of problem (4) solutions were formulated in [18]. In t h e same book t h e r e a r e r e f e r e n c e s on the original p a p e r s related to this problem.
2.
WEAKLY
CONVM FUNCTIONSIn this section basic p r o p e r t i e s of weakly convex functions [15] a r e investi- gated. The family of weakly convex functions includes smooth and convex functions and i s closed with r e s p e c t to t h e summation and pointwise maximum. We give new denition of this family useful f o r applications. It will be shown t h a t this denition and t h e denition given by E. Nurminski [I51 a r e equivalent.
DEFINITION 1 Let X be a convez subset of
a
Hilbert space H. A c o n t i n u o u s f u n c t i o n f :X 4R
is called w e a k l y convex o n X i f for allz
E X, y E X,0 S a S 1 t h e following i n e q u a l i t y holds.
where t h e r e m a i n d e r
r
: X x X 4R
s a t i s e sfor all 2 EX.
The set Bf
( z )
i s c a l l e d a d i e r e n t i a l of a weakly c o n v e x function f( z )
at a p o i n t z ~ X o n X i f f o r a l l g ( z ) E B f ( z )f o r all
z ,
y E XW e s a y t h a t a function f
( z )
i s weakly c o n c a v e o n X, if-
f( z )
i s weakly con- v e x on X. A d i e r e n t i a l of t h e weakly c o n c a v e function f( z )
i s d e n e das
a d i e r e n t i a l of t h e weakly c o n v e x function-
f( z )
t a k e n with s i g n minus.THEOREM 1 Let X be
an
open convez subset ofa
filbert space H a n d the f i n c t i o n f ( z )i s
w e a k l y convez o n X. Then thR set B f ( z ) ,z
E Xi s
non-empty,convez, closed bounded a n d
where f l ( z , p ) is
a
d e r i v a t i v e of t h efimtwn
f ( z )at
a p o i n tz
alonga
direc- t i o n p .PROOF W e start with t h e following lemma.
LEMMA 2 For a n y
z
E X, p EH
a d e r i v a t i v ez +
A p ) - f ( z )/ ' ( z .
p )=
1imJ"A 4 0 A
e x i s t s a n d i s nite.
PROOF F i r s t of all we p r o v e a n e x i s t e n c e of t h e nite or innite d e r i v a t i v e f l ( z , p ) f o r a n y
z
E X, p E H . F o r 0 S A2<
A, inequality ( 5 ) impliesconsequently
The last inequality implies t h e existence of t h e derivative fl(z, p ) because (Z +.'P*
z,
-P 0 f o r A r 0.
The derivative fl(z, p ) c a n not t a k e t h e value
+ -.
Let us p r o v e t h a t t h e derfva- Live f'(z, p ) bounded below. For E>
0 , A>
0 t h e inequality (5) impliesAfter t h e equivalent transformation
If E and A are suciently small. then from (6) obtain - + P s A + & +
> 6 =
const.
consequently
LEMMA
3 The d e r i v a t i v e fl(z, p ) i s a c o n v e z positive-homogeneous * n otwn
c o n t i n u o u sat
t h e p o i n t 0 w i t h respect t o p.PROOF Let us p r o v e a positive-homogeneity of t h e function fl(z, p). Accord- ing to t h e denition of a d e r i v a t i v e f o r a
>
0fl(z, u p )
=
lim f(z +Asp) -f(z)A 4 0 h
For A,, A, r 0, A ,
+
X2=
1 t h e inequality (5) impliesPassing to limit A r 0
Prove t h a t t h e function f
' ( z
, p ) i s continuous with r e s p e c t t o p at t h e point 0 f o r anyz
E X . To check t h i s i t is sucient t o prove boundedness of the function f ' ( 2 , p ) ina
neighborhood of t h e point 0 (see f o r example [19]).Passing t o limit A2 r 0 in ( 9 ) obtain
The condition (6) implies t h a t t h e r e exists a neighborhood U ( 0 ) of t h e point 0
r ( z +
Alp,z )
such t h a t
1 I <
6=
const f o r a l l p E U(0). Since f( z )
is continuous, A1then w e can suppose that t h e function f
( z +
Alp ) is bounded on U ( 0 ) with r e s p e c tto
p.
Therefore t h e boundedness above of f ' ( 2 , p ) with r e s p e c t to p in the neigh- borhood U ( 0 ) follows from t h e last inequality.LEMMA
4 A dierential Bpf ' ( 2 , 0 ) of the convez &nction f ' ( 2 , p ) w t t h respect to argument p at t h e point 0 coincides w i t h 8 f ( 2 ) .PROOF Dierential Bpf ' ( 2 , 0 ) i s dened as followed
The inequality (5) implies
z
+ a ( y - 2 ) )- j ( z )
f ( 2 4 )
- f ( z )
21( +
(1- a ) r ( z ,
y ) ,a >
0.
a Passing to limit
a
& 0Let z belong t o $ f f ( z , 0 ) , then from t h e last inequality w e have
consequently Bpf ' ( 2 , 0 )
c
d f( z
).Let u s prove t h e r e v e r s e conclusion. If
z
E8 f ( z ) ,
thenHence
J ( z
+X P ) - f
( 2 )r ( z + X p , z ) A
Z< z , p > +
X
and f
' ( z , p )
2 <z,p >,
consequently8 1 ( z ) c 8 p f ' ( z , o
). According t o Minkovski duality, since f l ( z ,p )
i s a convex positive homogeneous function with r e s p e c t top
and continuous a t t h e point0
then t h e setB p f l ( z , 0 )
is non-empty bounded convex closed andConsequently t h e
set
a f( z )
i s non-empty bounded convex close and relation (8) is t r u e . The theorem h a s been proved.W e shall give equivalent denition of t h e weakly convex functions.
DEFINITION 2
Let
Xbe a convez subset of a Hilbat space
H .We s a y thut a f i n c t i o n c o n t i n u o u s o n
Xi s weakly convez o n
X Ii f f o r a n y z
E Xthe set
G( z
)consisting of the vectors
g ,s u c h thut
is empty, and remainder t a m <(z , y
)in each compact subset K c
X isuniform- l y small relatively to llz - y 11, i.e. for a n y
&> 0 there e z i s t s 6 > 0 t h a t
f i r l l z - y I I < & z , u fK.
E. Nurminski 1151 h a s introduced this denition in case X
=
H= R n
whereR n
i s an Euclidean n-dimenional space.THEOREM 2
Let
X bea convez open subset of a Hilbert space
H ,t h e n t h e d e n i t w n s
1a n d
2are equivalent in the fillowing sense:
a ) i f
a N n c t i o n i s weakly convez in the sense of t h e d e n i t w n
2,t h e n i t i s weakly convez in the sense of t h e d e n i t i o n
2and <(z
,y
)= r ( z
,y
);b) i f
a N n d i o n i s weakly convez in the sense of the d e n i t i o n
2,then i t i s
weakly convez in t h e sense of the d e n i t i o n
1and
PROOF
It i s not dicult to see t h a t (6) implies the uniform convergence on a compact s u b s e t K . Hence the statement a ) of t h e theorem follows from t h e denition of a weakly convex function and theorem 1.Let u s prove the statement b). In the view of
( l o ) ,
w e havef o r a l , a 2 2 0 , a 1
+
a 2=
1.From last t w o inequalities w e obtain
A s < ( z , y ) / l l z - y l ( - * O f o r z + z , y - + z , then
n'r
( ( ( 2 , Z+
a I + a E = l cri.%+"
a 2 ( v - z ) ) / a $ l y - z ( (
+
< ( y . y + a l ( z - y ) ) / a l l l y -211) - + O f o r z - 2 ,y --+ 2 .
The theorem h a s been proved.
W e shall note some c a s e s , where the remainder t e r m s in t h e denitions 1, 2 coincide.
COROLLARY 1 Let X be a convex open subset o f a Hilbert space H . u a weakly convez function
p
( z ) s a t i s e sf o r all z , y E X , then
Conversely the second inequality implies the rst o n e f o r a n y g E 8f ( z ).
This c o r o l l a r y i s w e l l known, if p
>
0, in this case the function f ( z ) i s strong- ly convex.Let u s consider t h e case when function f ( z ) is twice continuously dierentiable at e a c h z E
X,
whereX
i s an open s u b s e t of a Hilbert s p a c eH,
i.e. f o r all z , yE X
Vf ( z ) i s a g r a d i e n t at a point z ; A ( z )
:X
+H
i s a l i n e a r o p e r a t o r generating the symmetric bilinear functionThe function i s called twice continuously dierentiable on X
c
H, if i t i s dierentiable at e a c h point z EX
andCOROLLARY 2
V
a f u n c t i o n f ( z ) is twice c o n t i n u o u s l y d i e r e n t i a b l e o n an o p e n s u b s e tX
of a Hilbert s p a c e H , i.e. t h e c o n d i t i o n s @I), &?) a r e t r u e , t h e nw h e r e
C o n v e r s e l y , t h e i n e q u a l i t y (23) i m p l i e s @I).
PROOF
The c o n v e r s e statement follows from t h e a ) of t h e theorem 2. Let us p r o v e t h e d i r e c t statment using t h e b ) of t h e theorem 2. In t h i s case w e c a n denoteconsequently
Thus, t h e inequality (13) i s t r u e .
3. ON THE ANTI-HONOTONICITY OF THE DIFFERENTIAL
MAPS
FOETHE WEAKLY
CONVEX FUNCTIONSLet
*
:X
XX
-.,R
b e a function dened on a p r o d u c tX
XX,
whereX
i s a convex open s u b s e t of a Hilbert s p a c e H. The function 9 ( z , y ) i s weakly convex onX
with r e s p e c t t o t h e rst argument, i.e.f o r a l l z , y , z
E X ;
a l + a 2 = l ; a l , a z 2 0 a n d r , ( z , 24)- 0 if llz
-
y((-.,O f o r a l l zE X .
llz
- u I1
W e suppose t h a t t h e function *(z, y ) i s weakly concave with r e s p e c t t o t h e second argument on
X,
i.e.~ ( z S u )
- 0 if 112
-
y ( 1 - 0 f o r all zE X .
llz
- u II
We say t h e function *(z, y ) i s weakly convex-concave, if i t s a t i s e s (14), (15).
Let
Q(z, x ) i O f o r a l l z
E X .
(16)Denote G ( z )
= By
*(z , y )1 ==,
i.e. G ( z ) i s a d i e r e n t i a l of the function *(z, z ) with r e s p e c t to t h e second argument at a point ( z , z ) .We shall formulate t h e sucient conditions of the anti-monotonicity of the mul- tivalued map
G ( z ) ,
i.e. for allg ( z )
EG ( z ) ,
g ( y ) E G ( y )< g ( z ) - g ( y ) ,
y- z > r O
f o r a l lz ,
y E X.
THEOREM 3
Let
X bean open convez subset of a Hilbert space H, a mnction
9 : X x X --, R be
weakly convez-concave, the remainder r, ( z ,
y ) becontinuous with respect to
z ,the function
9satis& condition
(323). T h e njbr all z , y
E X ;g ( z )
EG(z),
g ( y ) EG ( y )
PROOF We c a n assume y
=
0. I t also can be assumed t h a t g ( 0 )=
0as
a n anti- monotonicity of the mapG ( z )
does not depend upon a linear t e r m of t h e function9(z
y ) with r e s p e c t to t h e second argument. It is necessary to prove thatIn t h e view of
(14)
w e g e ta Q ( z , a z ) + (1 - a)9(0, a z )
29 ( a z , a z ) + a(1 - a ) r a , ( z ,
0 ).
Taking into account the p r o p e r t i e s of t h e weakly concave functions f r o m the last inequality obtain
a [ 9 ( z , a z ) - 9 ( z , z ) ] r (1 - a)[9(0,
0 )-
9 ( 0 ,a z ) ] + a(1 - a ) r a , ( z ,
0 )Since
&(a z , 0 ) / a
-+ 0 ifa
4 0 , then t h e last inequality implies9 ( z ,
0 )-
9 ( z ,z ) =
lim[ 9 ( z , a z ) - 9 ( z , z ) ] r
a40
Consequently, taking into account the weakly concavity of the function
9 ( z ,
y ) with r e s p e c t to t h e second argument, we obtainTheorem has been proved.
REMARK W e c a n see f r o m t h e p r o o f o f t h e t h e o r e m t h a t c o n d i t i o n ( 1 4 ) c a n be r e p l a c e d b y
* ( z ,
y )- * ( z , z ) S
< g ( z ) , y- z >
+ @ , ( y ,z )
f o r a l l z , yE X
COROLLARY 1 Let a l l c o n d i t i o n s of t h e theorem 9 be fullled a n d
r g ( z ,
y )- & ( u s
Z ) 2 O f o r a l l z . yEX,
t h e n t h e m a p G ( z ) i s a n t i - m o n o t o n e o nX.
COROLLARY 2 Let
X
be a n open subset of a filbert space H , a f u n c t i o n@ ( z ,
-
y ) be c o n t i n u o u s a n d convex-concave o nX
XX,
t h e f?unction@ ( z , z )
beconcave o n
X, then
t h e m u l t i v a l u e d m a p G( z ) =
B y@(z
, y )l y =,
i s a n t i m o n o t i n e o nX.
PROOF I t i s easy to get t h i s s t a t e m e n t i f a s s u m e
COROLLARY 3 Let all c o n d i t i o n s o f t h e theorem 9 b e f u l l l e d a n d
jbr a l l z , y
EX,
w h e r e0 ' 1 ~
- Y ' " ) - 0 if2 - 2 , y - 2
llz -
Y112
un.igormly
with
respect t o zEX, then
jbr a l l z , y
E X
a n d j b r a l l g ( z ) E G ( z ) , g ( y ) E G ( y ) .PROOF L e t
z ,
yE X ; z #
y . In t h e v i e w o f ( 1 7 ) , ( 1 8 ) w e get< g ( z ) - g ( y ) , v - z > = lim
n + -
( 2
Corollary has been proved.
Let u s suppose t h a t function
O ( z , y )
i s twice dierentiable on e a c h argument.Denote by A
( z , y
)=
9=( z ,
y ) t h e second d e r i v a t i v e with r e s p e c t t o t h e rst argu- ment ( s e e ( l l ) ) , and i t h e same wayB ( z
,y
)=
9yy( z
,y
) with r e s p e c t t o t h e second argument andg ( z ) =
Vy9 ( z , y
)ly =, .
THEOREM 4
Let
Xbe a n open convez subset of a Hilbert space
H ,function
9 :X
XX
-+R be twice dierentiable with respect t o each argument, and
J J A ( z , ~ ) - A ( z , z ) J I - + o ,
ifz
- 2 ,y
4 2 uniformlyforz E X
;J I B ( z , y ) - B ( z , z ) l ( - - 0 ,
ifz 4 2 , y - + z
uniformly'forz E X
; t h e o p e r a t o rQ ( z , z ) = A ( z , z ) - B ( z
,z )
satisfyf o r a l l z
E X , h EH a n d
Ydo m t d e p e n c i o n z ,
h .Then
< g ( z ) - g ( y ) , y - z > r - v l l z
1- y
f o r allz , y E X .
2 (20
PROOF The statement of t h e theorem follows from t h e c o r o l l a r y 2 of t h e theorem
2,
and t h e c o r o l l a r y 3 of t h e theorem 3. According t o the c o r o l l a r y2
d t h e theorem2In t h e view of
(19)
Thus, t h e conditions of t h e c o r o l l a r y 3 of t h e theorem 3 are satised.
Let u s consider t h e example which i l l u s t r a t e s the last theorem.
EXAMPLE
4 Model f o r i n t e r n a t i o n a l o i l t r a d e[Z].
T h e r e i s a market of a single homogeneous p r o d u c t , which consists of some sellers ( e x p o r t e r s ) and a sin- g l e b u y e r (importer). Let i=
1,. . .
, n be t h e e x p o r t e r s , f i ( z ) be t h e marginal c o s t of which any e x p o r t e r i produces t h e amount z of t h e p r o d u c t f o r marketing and r ( z ) be t h e p r i c e at which t h e i m p o r t e r would a g r e e to buy t h e amount z of t h e product. If zi denotes t h e amount of t h e product sold by e x p o r t e r i , then t h e r e v e n u e p i ( z ) of t h e e x p o r t e r i , c a n be e x p r e s s e d as follows:N o t e t h a t t o t h e s e n s e o f t h e p r o b l e m z t r 0 , f i ( z ) 2 0 , r ( z ) r O , i = 1 ,
. . . ,
n . W e assume also t h a t e x p o r t e r i , i=
1,. . .
, n i s a b l e to s e l l no more t h e n &amount of t h e product. If w e suppase t h a t e a c h seller i s going to choose t h e amount zi in o r d e r to maximize his r e v e n u e in any market's situation c h a r a c t e r i z e d by a v e c t o r z
=
(zl,. . .
, z,), t h e n t h e problem w i l l b e as follows: t o nd a n* * *
equilibrium's situation z
=
( z l ,. . . ,
zn ) s u c h t h a tThe admissible set X
=
[z E R n :O s zt s p i , i=
1 ,. . . ,
n1
i s convex and compact, The function +(z,
y ) (see example 1 ) denes asW e assume t h a t t h e functions f z ) i
=
1. . .
, n are continuously dierentiable and t h e function r ( z ) i s twice continuously dierentiable on some open s u b s e t [ z E R : z=
C r = l z t , z EU c
R~1,
whereU
i s open s u b s e t such t h a t Xc U
c R n . DenoteIt is not dicult t o nd
Let us investigate under what conditions the matrix Q ( z , z ) satises (19). Let e
=
(1,. . .
, 1 ) be the ndimensional vector, andAssume that
r Z ( z ) 4 O r z z ( z ) 2 0 f o r z E X
.
We can write the following inequality
Hence, [ - 2 d n x : = l p f r , ( z )
+
2p(z)1 .+
v > O implies (19). The lastinequality is c o r r e c t iff o r a l l z E X I 1 6 i 6 n . Consequently (21), (22) imply (20) and t h e map g ( z )
=
(V, ,pl,
. . . ,
V,, p, ) i s anti-monotone.I t should be noted t h a t in t h e view of t h e (21), (22) the function *(z, y ) i s strongly concave with r e s p e c t t o y because t h e matrix B (z, y ) i s negatively dened and t h e Nash point equilibrium exists [20].
4 THE YONOTONICITY OF DIFFERENTIAL
MAP
FORQUASI
CONVEX-CONCAVE F'UNCTIONSLet X be a closed convex subset of a Hilbert s p a c e H, U be an open subset of H and X
c
U. A function*
:X X X --,R
i s quasiconvex with r e s p e c t t o t h e rst argu- ment, i.e.for all al, a 2 2 0; al
+
a 2=
1 and f o r all z , y,
z E X and quasiconcave with r e s p e c t t o t h e second argument, i.e.f o r a l l al, ,a2 2 0 ; al + a 2 = l a n d f o r a l l z , z , y E X .
For the f u r t h e r development we assume t h a t t h e function ck(z, y ) satises re- gularity condition with r e s p e c t to the second argument, i.e. f o r any z € X and any y such t h a t y
#
a r g max, ,X 9 ( z , z )int f u E U : +(z, u )
=
ck(z, y ) {=
0,
where int A denotes the i n t e r i o r of a set A and *(z, z )
=
0 for a l l z E X. The fam- ilies of weakly convex and quasiconvex functions intersect but are not embedded into e a c h other.A cone Df(z)
=
f g € H : < g , y - z > 6 0 for all y €CL(z)l is called a dierential of a quasiconvex function f (z ) at a point z on a s e t U CH ,
where M(z )=
f y : f (y ) 6 f (z ), y E U1.
A dierential of t h e quasiconcave function p(z ) i sdened as a d i e r e n t i a l of t h e quasiconvex function
-
q ( z ) t a k e n with sign minus.We s a y a function Q ( z , y ) i s quasi convex-concave if i t i s quasiconvex with r e s p e c t t o t h e rst argument a n d i s quasiconcave with r e s p e c t t o t h e second one.
Denote G ( z )
= Dy
Q( z
, y )ly =,
whereDy
i s a dierential of a function Q ( z , y ) with r e s p e c t to y in t h e s e n s e d e s c r i b e d above.Let a point
z *
E X be a solution of t h e equation max* ( z S ,
y )= o .
Y E X
W e will p r o v e t h a t ( 2 5 ) implies
f o r all
z
E X and f o r all g( z )
E G ( z ) . Consequently f o r nding a pointz *
r e s u l t s of t h e p a p e r [14] c a n be used, f o r example.THEOREM 5 Let U be
an
o p e n s u b s e t o f a Hilbert space H , X bea
closed con- v e z s u b s e t of H a n d Xc
U ,a
f i n c t i o n*
: X X X 4 R be c o n t i n u o u s o n X X X a n d q u a s i conuez-concave o n X X X,at
l e a s t one of t h e i n e q u a l i t i e s @), @4) be s t r i c t @oral # O ,al
# I ) . Y ( z , z ) = O f b r a l l z ~ X . f l a ~ o i n t z * ~ ~ s a t i r a s (23) t h e n t h e v a r i a t i o n a l i n e q u a l i t y (26) h o l d s for allz
E X a n d for a l l~ ( 2 ) f G ( z ) .
PROOF W e c a n assume
z* =
0 . Consequently i t i s n e c e s s a r y t o p r o v e t h a tAssume at rst t h a t t h e inequality ( 2 3 ) i s s t r i c t , t h e n max [ Q ( z , az), Q ( 0 ,
a z ) ] >
Q ( a z , a m )=
0 , 0< a <
1The equality ( 2 5 ) implies
Q(0, a m ) r Q(0,O)
=
0 f o r 0 ra s
1 ,z
f X.
W e g e t Q ( z , a m )
>
0 f o r 0< a <
1 taking i n t o account (27). Passing to limit whena
.( 0 obtain Q ( z , 0 ) 2 0 and Q ( z , 0 )-
Q ( z ,z )
2 0 . In a c c o r d a n c e to t h e denition of t h e d i e r e n t i a l of a quasiconcave function w e obtain statement of t h e theorem from t h e last inequality.Let u s consider t h e case with t h e s t r i c t inequality (24). In t h i s case Q ( O , ~ U C ) < Q ( O , O ) = O f o r O < a < l , z EX,^ # O
.
The inequality (23) implies
m a x [ Q ( z , a z ) , Q ( O , c l u : ) ] 2 Q ( a ~ , a z ) = O f o r 0 4 a 4 l .
Taking into account t h e last two inequalities w e g e t Q ( z , az) 2 0 f o r 0
<
a<
1.F u r t h e r consideration coincides with t h e previous case. The theorem i s proved.
Let us c o n s i d e r t h e case when a function Q ( z , y ) i s dierentiable with r e s p e c t to t h e argument y
.
DenoteCOROLLARY Let all conditions of the theorem 5 be satised a n d t h e fiLnction
* ( z ,
y ) i s dierentiable w i t h respect to the second argument. Then the e q u a l i t y (25) implies.PROOF The theorem statement follows from t h e inclusion q
( z )
€ G ( z ) , where G ( z ) i s a d i e r e n t i a l of quasiconcave functionQ(z
,z
) with r e s p e c t to t h e second argument.Let us consider an example illustrating t h i s corollary.
EXAMPLE Wald's production model [ Z ] . Let n p r o d u c t s are produced and
r
r e s o u r c e s used in a n economy,
a . . .
,a ,
be t h e amounts of t h e s e r e s o u r c e s . The values a i j ( i=
1,. . .
, n ; j=
1,. . .
, r ) denote t h e input of j - t h r e s o u r c e neces- sary to produce t h e unit of i - t h product. The p r i c e s of t h e p r o d u c t s depend on t h e amounts of t h e produced products. Let f i( z ) =
f i( z l , . . .
,z n )
(i=
1,. . .
,n )
be t h e p r i c e of a unit of i - t h e p r o d u c t if t h e p r o d u c t s are produced in t h e amountsz . . .
,zi ,... z,;
X be a feasible setn
X
= I z (
i = laijzi
4a,,
j = I , .. .
, r ; z i 2 0 , i = I , .. . , n t
Under some conditions on t h e functions f i
( z ) ,
(i=
1,. . .
,n )
t h e existence of a* *
non-negative production v e c t o r
z * = ( z
l ,. . .
, 2,) and non-negative r e s o u r c e p r i c e v e c t o r such t h a tr
* * *
f:
a i j z ; ~ a j ( j = I . .. .
, r ) ;x
a y U j Z Y i ( z l n . . ..z,,)i = l j = l
w a s p r o v e d in [Z], [I?]. For a l l non-negative and continuous o n t h e s e t z Z 0 func- tions t h e existence of a non-negative production v e c t o r z
*
E X being the solution of t h e following l i n e a r programming problemn
max
C li
( z*
)zi= C
f t (2 *)z;E € X i = l i = l
w a s proved in [Zl]. The e x i s t e n c e of v e c t o r s z * , U* satisfying (29) follows from t h e last equality and t h e duality theorem of t h e l i n e a r programming.
Denote
Since the function 9 ( Z , y ) i s l i n e a r with r e s p e c t t o y t h e n i t i s concave in y . If t h e point z
*
i s a solution of t h e problem (30), t h e n z*
i s a l s o the solution of t h e problem (25). Hence in o r d e r to satisfy inequality (28) it i s s u c i e n t t h a t t h e func- tion 9 ( z , y ) be s t r i c t l y quasiconvex with r e s p e c t t o z on X, i.e.max [*(z, y ),
W z ,
y ) l>
9 ( a l z + 4 2 2 I Yf o r all a l , a 2 0 ,
#
1 , Q l+
a2=
1 and f o r a l l z , z , y E X , z#
2 .1 Ermoliev, Ju. and A. Papin: An a p p r o a c h to simulating international oil t r a d e . Working P a p e r May 1981: Laxenburg, Austria, International Institute f o r Ap- plied Systems Analysis.
2 Wald A.: Uber d i e Productionsgleichungen d e r Okonomischen Wertlehre, Ergeb- nisse eines Mathematischen Kolloquiums, No. 7 (1934-1935).
3 Rosen I.B.: Existence and Uniqueness of Equilibrium Points f o r Concave
N-
P e r s o n Games. Econometrics, 1965, 33, No. 3.
4 Primak, M.E.: On t h e algorithm f o r s e a r c h of t h e equilibrium points. Kiberneti- ka, 1973, No. 1 , (in Russian).
5 Ermoliev, Ju. M. and S.P. Urias'ev: On s e a r c h f o r equilibrium by Nash in many- person G a m e s . Kibernetika, 1982, 3, (in Russian).
6 Pau, Z.F.: Dierential games and a Nash equilibrium s e a r c h i n g algorithm. SIAM J . Control, 1975, V. 13, No. 4.
7 Bensoussan A., J.L. Lions a n d R. Teman: Methodes numeriques d'analyse de sys- temes, Tome 2: methodes de decomposition. Cahier N o . 11, IRIA, P a r i s , 1972.
8 Gajewski, H., K . G r o g e r a n d K. Zacharias: Nichtlineare operatorgleichungen und Operatordierentialgleichungen. Akademie
-
Verlag Berlin, 1974.9 Bakushinskij, A.B. and B.T. Poljak: On t h e solving of t h e variational inequality.
Dokl. Akad. Nauk SSSR, 1974, V. 219, N o . 5, (in Russian).
10 Goldshtein, E.G.: The method of modication of t h e monotone mappings. Econ. and Math. Meth., 1975, V. XI, 6, (in Russian).
11 Rockafellar, R.T.: Monotone o p e r a t o r s and t h e proximal point algorithm. SIAM J. on Control and Optimization, 1976, 14.
12 Lions P.L.: Une methode i t e r a t i v e d e resolution d'une inequation variationnelle.
I.J. Math., 1978, V. 31, No. 2.
13 Bruck, R.: On weak convergence of a n e r g o d i c i t e r a t i o n f o r t h e solution of variational inequalities f o r monotone cperators i n Hilbert s p a c e . J. Math. and Appl. V. 61, 1.
14 Nemirovskij, A.S.: The e c i e n t methods of solving equations with monotone o p e r a t o r s . Econ. and Math. 1981, V. 17, 2, (in Russian).
15 Nurminski, E.A.: Numerical methods f o r solving Deterministic and Stochastic Minimax Problem, Naukova Dumka, Kiev, 1979, (in Russian).
16 Aubin, J.P.: Mathematical methods of game and economic theory. North Holland Publishing Company, 1979.
17 Primak, M.E.: On t h e generalized equilibrium-optimal problems and some economic models. Dokl. Acad. Nauk SSSR, 1971, 200, No. 3, (in Russian).
18 Nirenberg, L.: Topics in nonlinear functional analysis. Mir, Moscow, 1977, (in Russian).
19 Pshenichnyi, B.N.: Necessary conditions f o r an extremum. Dekker, N e w York, 1971.
20 Nikaido, H., and K. Isoda: A note on noncooperative convex games. Collection.
Innite antagonistic games, Fizmatgiz. Moscow, 1963, (in Russian).
21 Kuhn, H.W.: Collection. Linear inequalities and o t h e r questions. Inostrannaja L i t e r a t u r a , Moscow, 1959 (in Russian).