VARIABLE STRUCTURE EXTREMUM PROBLEM WITH
CONSTRAINTS FOR DELAY DISCRETE INCLUSIONS
GECİKMELİ AYRIK İÇERMELER İÇİN KISITLAMALI DEĞİŞKEN YAPILI EKSTREMUM PROBLEMİ
Hijran G. MIRZAYEVA
Baku State University Institute of Applied Mathematicshijan_2002@rambler.ru
ABSTRACT: The paper considers a variable structure extremum problem with
constraints for delay discrete inclusions. The necessary extremum conditions are obtained for the considered problem.
Keywords: Discrete Inclusion; Necessary Extremum Condition; Subdifferential;
Lipschitz Condition; Tangent Cone; Hypertangent cone
JEL Classifications: C61
ÖZET: Makalede gecikmeli diskret içermeler için kısıtlamalı değişken yapılı
ekstremum problemi araştırılmaktadır. Ele aldığımız problem için gerekli ekstremum koşulları bulunmuştur.
Anahtar Kelimeler: Diskret İçerme; Gerek Ekstremum Koşulu, Altdiferansiyel,
Lipschitz Koşulu, Teğet Koni, Hiperteğet Koni
JEL Sınıflaması: C61
1. Introduction
The extremum problems for discrete inclusions with delay are a generalization of the discrete problems of optimal control with delay. Application of nonsmooth analysis theory is of importance for investigation of such a problem. Since the definition of the subdifferential given by T. Rockafellar and F.Clarke is entirely a generalization of the smooth and convex problems, the application of the theory of nonsmooth analysis in this paper is advisable. As the necessary extremum conditions are formulated more naturally with the help of the subdifferential, the definition of subdifferential is of importance in the theory of extremum problems.
There exist a lot of papers devoted to the qualitative investigation of different problems of optimal control of discrete systems. The control problems of discrete systems, described by the different difference equations are of importance among various optimal control problems. In spite of this fact, there are many unstudied problems in the field of extremum problems for discrete inclusions. Recently, multivalued mappings became the subjectof intensivestudy.Different properties of multivalued mappings and their connection with the theory of optimization were considered in (Aubin, Ekeland, 1984: 510; Borisovich, Gelman, Mishkins, Obukhovskiy, 1986: 103).
Note that main models of mathematical economics are reduced to extremum problems for discrete inclusions.
If the state of the control system is characterized by a system of more than two equations and an operator connecting the equations of the system (connection operator) exists, then such a control system is called a variable structure system. This work generalizes some of the results, obtained in (Mirzayeva, Sadygov, 2005: 89-94; Mirzayeva, Sadygov, 2006: 106-112). An arbitrary order necessary extremum conditions are obtained for the nonconvex optimal control problem for discrete inclusions with delay in (Mirzayeva, Sadygov, 2006: 106-112). In this paper the extremum problem for delay discrete embedding is reduced to the mathematical programming problem using the method given in ( Boltyanskiy, 1975: 3-55). Further, using Clarke’s theory, necessary extremum conditions are obtained. Differently from the considered works (Mirzayeva, Sadygov, 2005: 89-94; Mirzayeva, Sadygov, 2006: 106-112; Mirzayeva, Sadygov, 2007: 67-72), in the paper the discrete extremum problem with constraints is considered. Note that the extremum problem for delay discrete inclusions, is a particular case of variable structure extremum problem for delay discrete inclusions. The known optimal control problems of a variable structure are obtained from variable structure extremum problems for delay discrete inclusions (Mirzayeva, 2007: 44-50). The optimal control problems of variable structure with discrete time delay are considered in (Mirzayeva, 2007: 44-50). Such problems are considered by many authors. Differently from the known works, in this paper a nonsmooth case is considered and the problem is reduced to the extremum problem for delay discrete inclusions with variable structure. Further, in the paper the necessary extremum condition is obtained for discrete systems with variable structure.
Variable structure extremum problem for delay discrete inclusions without constraints is considered in (Mirzayeva, Sadygov, 2007: 67-72). The necessary extremum conditions are obtained for one discrete systems class in (Mirzayeva, Sadygov, 2007: 67-72 ).
Note that optimal control problems of variable structure arise while investigating some chemical-technological processes, applied problems of economics and physics.
2. The formulation of the problem
Let X , be Banach spaces, Y : 2 2X,
t X a t0,1,...,k1, , 2 : 2 Y t Y
b tk,k1,...,m1 be the multivalued mappings, where 2V denotes the set of all subsets of
V
. We denote grF
(z,v)ZV:vF(z)
.Let us consider the delay discrete inclusions with variable structure
, ,..., 2 , 1 , ) ( 1 ,..., 1 , ), , ( 0 , 1 ,..., 1 , ) ( 1 ,..., 1 , 0 ), , ( 1 1 C y k h k h k h k t at x G y m k k t y y b y t at t c x k t x x a x m t t t h t t t t t t t t (2.1)
where c(t)X at t,1,...,1,0, CY,G:X Y is mapping, h
m
k, , are fixed natural numbers. As a trajectory (solution) , ({xt}, {yv}) of the discrete inclusion (2.1) we understand the process
, ,..., 1 , , , 1 ,..., 1 ,t k k y v k m
xt v for which (2.1) is satisfied.
Suppose that
t X R t k i n g k m k h k1, min{ 1, 1}, i , : , 1,..., , 1,...,
t Y R t k m i n fi , : , 1,..., , 1,..., . We denote x(x1,...,xk), y(yk1,...,ym).Consider the minimization of the function
k t m k t t t t f y t x g y x F 1 1 0 0 0( , ) ( , ) ( , ) (2.2)on the trajectories of discrete inclusion (2.1) and with the following constraints
k t m k t t t t f y t x g y x F 1 1 1 1 1 , , , 0,
k t m k t t j t j j x y g x t f y t F 1 1 , 0 , , ,
k t m k t t j t j j x y g x t f y t F 1 1 1 1 1 , , , 0, (2.3)
k t m k t t n t n n x y g x t f y t F 1 1 . 0 , , ,We note that as a trajectory of (2.1) we take the pairs ( yx, ), for which (2.1) is satisfied. Denote by M the set of solutions of problem (2.1). To reduce the formulated problem to the mathematical programming problem we use the following notation. We denote smk and define the sets in XkYsas
( 1,..., , 1,..., ) : 1 0( ( ), (0))
, 0 x x y y X Y x a c c M k s m k k
( 1,..., , 1,..., ) : 2 1( ( 1), 1)
, 1 x x y y X Y x a c x M k s m k k
( 1,..., , 1,..., ) : 1 ( (0), )
, x x y y X Y x a c x M k k m k s
( 1,..., , 1,..., ) : 2 1( 1, 1)
, 1 x x y y X Y x a x x M k k m k s
( 1,..., , 1,..., ) : 1( 1 , 1)
, 1 k k m k s k k k k k x x y y X Y x a x x M
( 1,..., k, k 1,..., m) k s: k 1 k( ( k h), ( k))
, k x x y y X Y y b G x G x M
( 1,..., , 1,..., ) : 2 1( ( 1 ), 1)
, 1 k k m k s k k k h k k x x y y X Y y b G x y M
( 1,..., k, k 1,..., m) k s: k h 1 k h( ( k), k h)
, h k x x y y X Y y b G x y M
( 1,..., , 1,..., ) : 2 1( 1, 1)
, 1 h k k m k s k h k h k k h k x x y y X Y y b y y M
( 1,..., , 1,..., ) : 1( 1 , 1)
, 1 k k m k s m m m h m m x x y y X Y y b y y M
(x1,...,x , y 1,...,y ) X Y :y C
. Mm k k m k s mIt is clear that the formulated problem will be reduced to the minimization of the function )F0(x,y on the set
m t t M M 0
with the constraints
, 0, ,
, 0,1 x y F x y
F j Fj1
x,y 0,,Fn
x,y 0.3. The solution of the problem
Let Z be a Banach space, E be a nonempty subset of Z . Clarke’s subdifferential of the function at the point z is denoted as 0 (z0) .
Suppose z0E. The set TE(z0)
zZ:dE0(z0;z)0
is called tangent cone to E in z . 0The set NE(z0)
zZ:z,z0at zTE(z0)
is called normal cone to E in z . 0Let D . The set of all hypertangents to D at the point Z z denote by D ID(z). By the definition (see (Clarke, 1988: 279) )
z vZ:0,thatytE atally
zB
E, vB,t
0,IE .
We note if gi
,t ,i0,...,n, satisfy the Lipschitz condition in the neighbourhood of x where t, t1,...,k and fi
,t,i0,...,n, satisfy the Lipschitz condition in theneighbourhood of yt, where tk1,...,m, then Fi
x,y,i0,...,n satisfy the Lipschitz condition in the neighbourhood of
x,y . Denoted by the set of solutions of problem (2.1), satisfying condition (2.3). The pairs
x,y are called optimal , if F0
x,y F0
x,y at
x,y .We use Lagrange’s generalized function for the nonsmooth problem of mathematical programming. Let
n i M i iF x y r d x y r y x L 0 , , , , , , where
n
0, 1,, , i,rR ,i0,...,n. The following corollary follows from Theorem 6.1.1 ( Clarke, 1988: 279).
Corollary 1. Let
x,y minimize the functional F0 on the set , the functionsn
F F
F0, 1, satisfy the Lipschitz condition in the neighbourhood of
x,y, M be a closed set. Then for sufficiently large r the numbers 0 0,10,,j 0 andn
j
1,, may be found, not all equal to zero simultaneously, such that
j i i iF x y 1 0 , and 0L
x,y,,r
.In the paper, the necessary extremum conditions are obtained, considering the additional conditions A, B of R.T. Rockafellar (Rockafellar, 1998: 733), the conditions C, D (Sadygov, 2007: 59). Similarly to subdifferential calculation the certain additional conditions are required for calculating normal cones. Such problems (the problems of calculation of normal cones) were studied by some authors as F.H. Clarke, R.T.Rockafellar, Dubovitsky-Milutin and others. These conditions are not equivalent, though they are considered for the same purpose.
Condition A. m i IM z z M T i 1 ( )) . ( ) ( 0 Condition B.
X
R
n1,
Y
R
n2, from *N
(
z
)
i M i
and0
* * 1 * 0
m
it follows that
i*
0
at i0,...,m. Condition C. X Rn1,Y Rn2, ( ) ( ) 1 2 , 1 0 s n k n l i M M z T z R R T l i
l1,...,m,and Mi are closed sets at i0,...,m.
Condition D. X Rn1,Y Rn2, ( ) ( ) 1 2 , 1 s n k n m l i M M z T z R R T l i
l0,...,m1, and Mi are closed sets at i0,...,m.Proposition 1. If one of the conditions A, B, C or D holds, then .) ( ) ( 0
m i M M z N z N iFurther we suppose that bt:Y2 C
Y , where C(Y) denote the families of allnonempty closed subsets of .Y We denote
n i i iF x y y x F 0 . , ,
Theorem 1. Let z
x,y x1,,xk,yk1,ym
minimize the functional0
F on the set , grat at t0,...,k1, grbt at tk,...,m1 and C be closed sets, G:X Y continuous operator, the functions gi
,t :X R, i0,...,nsatisfy the Lipschitz condition in the neighbourhood of xt at t1,...,k, the functions fi
,t , i0,...,n satisfy the Lipschitz condition in the neighbourhood oft
y at tk1,...,m. If in addition one of the conditions A, B, C or D holds, then there exist vectors tNMt
z,t0,1,,m and the numbers0 , , 0 , 0 1 0 j
и j1,,n may be found, not all equal to zero simultaneously, where
j i i iF x y 1 0 , and F
x,y such that
m i i 0 .Proof. Using Theorem 1.3.12 (Borisovich, Gelman, Mishkins, Obukhovskiy, 1986: 103) we have that M is closed at s s ,...,k kh. So the intersection of finitely many closed sets is a closed set, therefore according to this condition we obtain that
M is closed. It is straightforward to check that the conditions of corollary 1 are satisfied. Then for enough large r the numbers 0 0,1 0,,j 0 and
n
j
1,, may be found, not all equal to zero simultaneously such that
j i i iF x y 1 0 , and
n i M i i n i M i iF r d x y F x y r d x y 0 0 , , , 0
n i M i iF x y N x y 0 , , . Since
m t M M x y N x y N t 0 , ,, then we have that 0
,
, .0
m t M x y N y x F tTherefore there exist tNMt
z,t 0,1,,m, and F
x,y such that
m t t 0 . The theorem is proved.
Theorem 2. If z(x,y)
x1,,xk,yk1,,ym
minimizes functional F on 0 the set , gra at t t0,...,k1, grb at t tk,...,m1 and C are closed sets,Y X
G: is continuous operator, the functions gi
,t :X R,i0,...,n, satisfy the Lipschitz condition in the neighbourhood of x at t t1,...,k, the functions
t i nfi , , 0,..., , satisfy the Lipschitz condition in the neighbourhood of y at t m
k
t 1,..., and in addition one of the conditions A, B, C or D holds, then the numbers 00 0,1 0,,j and j1,,n may be found, not all equal to zero simultaneously, where
j i i iF x y 1 0 ,
k t t x g x i t n i i t ( , ) at 1,..., 0 * 0
, ( , )at 0 * 0 f y t x i t n i i t
tk1,...,m,
, 0
1 1 0 N 0 x x a c c ,
(), * ()
( ( ),)( , 1), 1,..., , 1 * t N x x t x t xt t grat c t t t
xt t,xt t,xt1 t
Ngrat
xt ,xt,xt1
att1,...,k1, , ) , , ( )) ( ), ( ), ( ( ( (), ()) 1 * 1 * * h k k grb G G k h k k k k x k y k N x x y x k
xk t h(k t),yk t(k t),yk* t (k t) Ngrbk t(G(),)(xk t h,yk t,yk t 1),t 1,...,h 1 * * ), , , ( )) ( ), ( ), ( ( * * * 1 1 k h t k h t grb k t k h t k h t t k k h t y k h t y k h t N y y y y t h k , 1 , 1m k h t *( ) ( ) m C m m N y y such that in the case h the relations are fulfilled:
0 ) ( ) ( ) 1 ( * * * * 0 , x t x t x t xt t t t at t1,...,k; 0 ) ( ) ( ) 1 ( * * * * 0 , y t y t y t xt t t t at tk1,...,mh1; (3.1)
0 ) 1 ( * * 0 , y t y t xt t t at tmh,...,m;in the case h the relations are fulfilled:
0 ) ( ) ( ) 1 ( * * * * 0 , x t x t x t xt t t t at t1,...,k1; 0 ) ( ) 1 ( * * * 0 , x t x t xt t t at tk,...,kh1; 0 ) ( ) ( ) 1 ( * * * * 0 , x t x t x th xt t t t attkh;...,k; (3.2)
0 ) 1 ( * * 0 , y t y t y th xt t t t at tk1,...,mh1; 0 ) ( ) 1 ( * * * 0 , y t y t xt t t at tmh,..., m;in the case h the relations are fulfilled:
; 1 ,..., 1 at 0 ) ( ) ( ) 1 ( * * * * 0 , x t x t x t t kh xt t t t ; 1 ..., , at 0 ) ( ) ( ) ( ) 1 ( * * * * * 0 , x t x t x t x th tkh k xt t t t t (3.3) ; ,..., at 0 ) ( ) ( ) 1 ( * * * * 0 , x t x t x t h t k k xt t t t ; 1 ,..., 1 at 0 ) ( ) ( ) 1 ( * * * * 0 , y t y t y th tk mh xt t t t 0 ) ( ) 1 ( * * * 0 , y t y t xt t t at tmh,..., m;
Proof. Using the corollary of Theorem 2.4.5. (Clarke, 1988: 279), from the definition of
M
0,
M
1,
,
M
m we obtain that
( ,..., , ,..., ) : ( )
, ) ( 1 1 1 0( ( ),(0)) 1 0 z x x y y X Y x T x Т k s a c c m k k М
( ,..., , ,..., ) :( , ) ( , )
) ( 1 k k1 m k s t t1 gra (c(t),) t t1 М z x x y y X Y x x T x x Т t t ,...,1 t
,
( 1,..., , 1,..., ) k s:( t, t, t1) gra ( t, t, t1)
m k k M z x x y y X Y x x x T x x x T t t at t 1,...,k1,
( ,..., , ,..., ) :( , , ) ) ( 1 k k 1 m k s k h k k 1 М z x x y y X Y x x y Т k ,
, , 1
TgrbkG G xk h xk yk
( ,..., , ,..., ) :( , , ) ( , , )
) ( 1 1 k s th t t1 grb(G(),) th t t1 m k k М z x x y y X Y x y y T x y y Т t t h k k t 1,..., ,
( ,..., , ,..., ) :( , , ) ( , , )
) ( 1 k k1 m k s th t t1 grb th t t1 М z x x y y X Y y y y T y y y Т t t 1 ,..., 1 k h m t ,
( 1,..., k, k 1,..., m) k s: m C ( m)
. М z x x y y X Y y T y Т m Therefore we have
( ,0,...,0) : ( )
, ) ( 1 1 0(( ),(0)) 1 0 z x X Y x N x N a c c k М s
( , , , ,0,...,0) :( , ) ( , ) ) ( ( ( ),) 1 * 1 * 2 1 k k t t gra ct t t М z x x x X Y x x N x x N t s t ,
1 , at 0 i t t xi at t ,...,1 ,
1 1 * 1,..., ,0, ,0) : , , , , ( ) ( t t t gra t t t k k М z x x X Y x x x N x x x N t s t ,
1 , , at 0 i t t t xi at t1,...,k1,
1 * 1 * 1,..., , ,0,...,0) : , , ( ) ( k k k k h k k М z x x y X Y x x y N s k NgrbkG ,G
xk h,xk,yk 1
, xi 0at ikh,k
(3.4)
, ,..., 1 at 1 , at 0 , at 0 ), , , ( ) , , (: ) ,..., , ,..., ( ) ( 1 ) ), ( ( * 1 * 1 1 h k k t t t i y h t i x y y x N y y x Y X y y x x z N i i t t h t G b gr t t h t k m k k М t s t
1 , , at 0 ), , , ( ) , , (: ) ,..., , ,..., ( ) ( 1 * 1 * 1 1 t t h t i y y y y N y y y Y X y y x x z N i t t h t b gr t t h t k m k k М t s t , 1 ,..., 1 attkh m (0,...,0,0,...,0, ) : ( ) ) ( *m k m C m М z y X Y y N y N s m .According to Theorem 1 the numbers 0 0,10,,j 0andj1,,n may be found, not all equal to zero simultaneously, where
j
m t k t m k t k t t t t M n i i iF x y N t x y g x t f y t g x t 0 1 1 1 1 1 0 0 0 0 0 , , , , , 0
m t n i i i M m k t k t m k t t n n t n n t t g x t f y t N x y g x y f t 0 0 1 1 1 1 1 1 , , , , ,1
m t M n i n i n i m i i k i i k i ig x k f y k f y m N t x y 0 0 0 0 1, 1 , , , (3.5)
k t m k t m t M n i t i i n i t i ig x t f y t N t x y 1 0 1 0 0 . , , , Using the relations (3.4) and (3.5) it is straightforward to check the correctness of Theorem 2.
Theorem 3. Let z
x,y minimize the functional F on the set 0 ,grat at, 1 ,..., 0
t k grbtat tk,...,m1and С be closed sets, G:X Y be continuous operator, the functions gi
,t :X R,i0,...,n satisfy the Lipschitz condition in the neighbourhood of xt att1,...,k, the functions fi
,t ,i0,...n, satisfy the Lipschitz condition in the neighbourhood of yt attk1,...,m, ,
x,x1
att1,...,, Igratc t t t Igrat
xt,xt,xt1
at t1,...,k1, G(),G()( kh, k, k1) grb x x y I k , Igrbi(G(),)(xth,yt,yt1) at tk1,...,kh, ) , , ( th t t1 grb y y yI t at tkh1,...,m1 and IC
ym be nonempty sets. Then the numbers 00 0,1 0,,j and j1,,n may be found, not all equal tozero simultaneously, where
j
i 1 iFi x,y 0,
and there vectors exist , 0 0 t t z x where 0,1, ( , )at 1,... , 0 * 0 g x t t k z i t n i i t
( , ) 0 * 0 f y t z i t n i i t
attk1,...,m, x1
0 Na0c ,c0
x1 ,
() ), ( *1 * t x t xt t ) , ( 1 ) ), ( ( Ngrat c t xt xt at t1,...,,
xt
t
,xt
t
,xt1
t
Ngrat(xt,xt,xt1)att1,...,h1,
( ), ( ), * ( ) ( (),)( , , 1), 1 * * t h k t k t grb G k t h k t k t k k t y k t y k t N x y y x t k 1,...,h,(y* (k h t),y* (k h t),y* 1(k h t)) t k t k h t k h t ) ( ) ( , 1 , 1 ), , , ( k t k h t k ht 1 m* C m grb y y y t m k h y m N y N t h k such that in the case h the relations (3.1) are fulfilled, in the case h the relations (3.2) are fulfilled, in the case h the relations (3.3) are fulfilled.
Proof. It is straightforward to check that
( 1,..., , 1,..., ) :( , 1) ( ( ),)( , 1
) ( 1 t t gra ct t t s k m k k M z x x y y X Y x x I x x I t ,...,1 at t ;
( ,..., , ,..., ) :( , , ) ) ( 1 1 k s t t t 1 m k k M z x x y y X Y x x x I t
, , 1
Igrat xt xt xt att1,...,k1;
( ,..., , ,..., ) :( , , ) ) ( 1 k k 1 m k s k h k k 1 M z x x y y X Y x x y I k
1
)) ( ), ( ( , , Igrbk G G xk h xk yk ;
( 1,..., , 1,..., ) :( , , 1) ( (),)
, , 1
) ( k s th t t grb G th t t m k k M z x x y y X Y x y y I x y y I t t ; ,..., 1 attk kh
( 1,..., , 1,..., ) :( , , 1)
, , 1
) ( k s th t t grb1 th t t m k k M z x x y y X Y y y y I y y y I t k h ; 1 ,..., 1 attkh m
( ,..., , ,..., ) : ( )
. ) ( 1 k k 1 m k s m C m M z x x y y X Y y I y I m If m t M M z I z T t 1 ( )) , ( ) ( 0 then according to lemma 5.11 ( Girsanov, 1970: 118) we can find the linear functionals * N (z),t 0,1,...,m,
t
M
t
not all equal to zero,
such that *01*...m* 0. Then we obtain that at 0 the statement of Theorem 3 is satisfied.
According to Theorem 3 we have that IMt
z intTMt
z att1,...,m. Therefore, if
mt Mt
M
z
I
z
T
1
0
(
)
(
(
))
, then the conditions of Theorem 2 withthe condition A are satisfied. Then it follows from Theorem 2, that at
1
the statement of Theorem 3 is satisfied. The theorem is proved.Theorem 4. If z(x,y)
x1,,xk,yk1,,ym
, gra at t t0,...,k1, tgrb at t k,...,m1 and C are convex sets, G:X Y is a linear operator, the functions gi
,t :X R,i0,...,n are convex at t1,...,k, the functions
t Y R i nfi , : , 0,..., are convex at t k1,...,m, besides j and n 0 , , 0 , 1 1 0 j
may be found, where
j i 1iFi x,y 0 , , ,... 1 at ) , ( 0 * 0 g x t t k x n i t i i t 0 ( , )at * 0 f y t x n i t i i t tk1,...,m,
and there the vectors exist
, 0
1 1 0 N 0 x x a c c ,
xt*(t),x*t1(t)
Ngrat(c(t),)(xt ,xt1), t1,...,,
xt t,xt t,xt1 t
Ngrat
xt ,xt,xt1
att1,...,k1, , ) , , ( )) ( ), ( ), ( (xk*h k xk* k y*k1 k Ngrbk(G(),G()) xkh xk yk1
xk*th(kt),yk*t(kt),yk*t1(kt)Ngrbkt(G(),)(xkth,ykt,ykt1),, ... 1 h t ( ( ), ( ), * ( )) 1 * * k h t y k h t y k h t yk t k h t k h t h k m t y y y Ngrb k t k h t k h t t h k ( , , 1), 1,..., 1 ,
(
)
(
)
* m C mm
N
y
y
such that in the case h the relations are fulfilled:
0 ) ( ) ( ) 1 ( * * * * 0 , x t x t x t xt t t t at t1,...,k; 0 ) ( ) ( ) 1 ( * * * * 0 , y t y t y t xt t t t at tk1,...,mh1; (3.6)
0 ) 1 ( * * 0 , y t y t xt t t at tmh,..., m;in the case h the relations are fulfilled:
0 ) ( ) ( ) 1 ( * * * * 0 , x t x t x t xt t t t at t1,...,k1; 0 ) ( ) 1 ( * * * 0 , x t x t xt t t at tk,...,kh1; 0 ) ( ) ( ) 1 ( * * * * 0 , x t x t x th xt t t t at tkh,...,k; (3.7)
0 ) 1 ( * * 0 , y t y t y th xt t t t at tk1,...,mh1; 0 ) ( ) 1 ( * * * 0 , y t y t xt t t at tmh, m... ;in the case h the relations are fulfilled:
; 1 ,..., 1 at 0 ) ( ) ( ) 1 ( * * * * 0 , x t x t x t t kh xt t t t ; 1 ,..., at 0 ) ( ) ( ) ( ) 1 ( * * * * * 0 , x t x t x t x th tkh k xt t t t t k k t h t x t x t x xt*,0 t*( 1) t*() t*( )0at ,..., ; ; 1 ,..., 1 at 0 ) ( ) ( ) 1 ( * * * * 0 , y t y t y th tk mh xt t t t (3.8) 0 ) ( ) 1 ( * * * 0 , y t y t xt t t at tmh,...,m
Then z
x,y x1,,xk,yk1,ym
minimizes the functional F on the 0 set .Proof. Consider the case h. Since ( , )
0 * 0 g x t x n i t i i t at t1,...,k and ) , ( 0 * 0 f y t x n i t i i t
at tk1,...,m, then from the relation (3.6) we have
) , ( ) ( ) ( ) 1 ( 0 * * * t x t x h t g x t x n i t i i t t t at t1,...,k; ) , ( ) ( ) ( ) 1 ( 0 * * * t y t y t h f y t y n i t i i t t t at tk1,...,mh1;
( , ) ) 1 ( 0 * t y t f y t y n i t i i t t at tmh,..., m; where), ( ) 0 ( ( ( ), (0)) 1 * 1 N 0 x x a c c
xt*(t),x*t1(t)
Ngrat(c(t),)(xt ,xt1), t1,...,h,
xt ht,x*t(ht),xt1
ht
Ngrat (xt,xt,xt1), ,t1,...k1h , ) , , ( )) ( ), ( ), ( (xk*h k x*k k yk*1 k Ngrbk(G(),G()) xkh xk yk1
( ), ( ), * ( )
( (),)( , , 1), 1 * * t h k t k t grb G k t h k t k t k k t y k t y k t N x y y x k t h t1,..., ,(yk*t(kht),yk*ht(kht),yk*ht1(kht)) , 1 ,..., 1 ), , , (y y y 1 t m k h Ngrb k t k h t k h t t h k ( ) ( ) * m C m m N y y . Therefore t t t t t t i n i i t i n i i x x t h x t x t x t x g t x g ( , ) ( , ) ( 1) () ( ), * * * 0 0 at t 1,...,k; t t t t t t i n i i t i n i if y t f y t y t y t y t h y y ( , ) ( , ) ( 1) () ( ), * * * 0 0 at ;tk1,...,mh1 t t t t t i n i i t i n i if y t f y t y t y t y y ( , ) ( , )) ( 1) (), * * 0 0 at tmh,..., m. Hence follows
( 1) () ( ),
( 1) (), . ), ( ) ( ) 1 ( ) , ( ) , ( ) ) , ( ) , ( ( )) , ( ) , ( ( * * * 1 1 * * 1 * * * 0 0 1 1 0 0 1 0 0
m h m t t t t t t t t h m k t t t k t t t t t t m h m t t i n i i t i n i i h m k t t i n i i t i n i i k t t i n i i t i n i i y y t y t y y y h t y t y t y x x t h x t x t x t y f t y f t y f t y f t x g t x g Since according to the condition that grat,grbt,C are convex sets and G:X Y is a linear operator, then we have
0 ), 0 ( 1 1 * 1 x x x at x1a0(c(h),c(0)), 0 ) , ( ) , ( )), ( ), ( (xt* t xt*1 t xt xt1 xt xt1 at (xt,xt1)grat(c(ht),) and ,t1,...,h
, ( ), ,( , , ) 1 1 * t t t t t t h t x h t x h t x x x x 0 ) , , ( 1 xt xt xt at (xt,xt,xt1)grat and t1,...,k1h, 0 ) , , ( ) , , ( )), ( ), ( ), ( (xk*h k x*k k y*k1 k xkh xk yk1 xkh xk yk1 at , (.)) (.), ( ) , , (xkh xk yk1 grbk G G
xk*th(kt),y*kt(kt),yk*t1(kt)
, 0 ) , , ( ) , , (xkth ykt ykt1 xkth ykt ykt1 at t1,...,h,
( ), ( ), ( ),( , , ) (yk* t k h t yk* h t k h t y*k h t 1 k h t yk t yk h t yk h t 10 ) , , ( 1 ykt ykht ykht at (ykt,ykht,ykht1)grbkht and , 1 ,..., 1 m k h t *( ),( ) 0 m m m m y y y at ym . C If ({xt},{yv})M , then hence we have
h t xt t xt t xt xt xt xt x x x 1 1 1 * 1 * 1 1 * 1(0), ( (), ()),( , ) ( , )