Working Paper
IMPLICIT FUNCTION THEORE3IS FOR MULTI-VALUED MAPPINGS
B.N.
Pshenichn ySeptember 1986 WP-86-45
International Institute for Applied Systems Analysis
A-2361 Laxenburg, Austria
NOT FOR QUOTATION WITHOUT THE PERMISSION OF THE AUTHOR
IKPLICIT FUNCl'ION THEOREXS MIR MULTI-VALUED MAPPINGS
B.N. Pshenichn y
September 1986 WP-86-45
Working Papers 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 e x p r e s s e d h e r e i n d o not necessarily r e p r e s e n t those of t h e Institute o r of i t s National Member Organizations.
INTERNATIONAL INSTITUTE FOR APPLIED SYSTElMS ANALYSIS A-2361 Laxenburg, Austria
FOREWORD
This p a p e r d e a l s with t h e proof of implicit function t h e o r e m s f o r c e r t a i n c l a s s e s of set-valued functions. The techniques a p p l i e d h e r e are mainly b a s e d o n t h e duality t h e o r y of convex analysis. The p a p e r was written d u r i n g a visit of P r o - f e s s o r Pshenichny t o t h e System and Decision S c i e n c e s P r o g r a m of IIASA.
Alexander B. Kurzhanski Chairman System a n d Decision S c i e n c e s P r o g r a m
AUTHOR
P r o f e s s o r B.N. P s h e n i c h n y i s a n A s s o c i a t e Member of t h e Academy of S c i e n c e s of t h e U k r a i n e a n S S R a n d now i s t h e h e a d of t h e d e p a r t m e n t in V. Glushkov Insti- t u t e of C y b e r n e t i c s , Kiev, USSR.
CONTENTS
1 Implicit Function Theorems f o r Convex Mappings 2 Locally Smooth Maps
R e f e r e n c e s
-
vii-
W L I C I T
FLTNCTION THEOREXS FOR MULTI-VALUED MAPPINGS
B.N.
Pshenichn yV. Glushkov Institute of Cybernetics Kiev, USSR
Let us consider t h r e e Banach s p a c e s X , Y, Z and o p e r a t o r F : X x Y -+ Z. We are i n t e r e s t e d in t h e solution of t h e following equation:
Suppose t h a t t h e points (zo, yo) satisfy t h i s equation. Then implicit function theorems yield c e r t a i n sufficient conditions f o r solvability of t h e equation (1) with r e s p e c t t o y f o r all
z
f r o m t h e c e r t a i n neighborhood of zo.Let u s somewhat reformulate t h e problem now. Define
Then implicit function theorems give conditions f o r a (z) #
4
in t h e neighborhood of zo. W e shall b e concerned mostly with t h i s reformulation.Let us introduce some notations. Take Z
=
Xx
Y. Dual s p a c e s of continuous l i n e a r functionals will b e denoted by X* , Y* , Z*.
P a i r of points from X and Y will b e defined (z, y ) while C c , z*>
is r e s e r v e d f o r t h e value of t h e functionalz*
at* * *
t h e point
z .
Takingz =
(z , y ),z =
(z , y ) w e obtainMultivalued mapping t r a n s f o r m s e a c h point z E X into set a(z) G Y. The s e t a(z) may b e empty.
Some more notations:
aom a
= [z
:a(z) # @ {Mapping a is called convex if gf a is a convex set and closed if gf a is closed. Some more notations will b e introduced in due c o u r s e . Terminology is close t o [I].
I. mPLIcrr
F'JNCTION THEOREMS FOR CONVEX MAPPINGS Let us s t a r t with some definitions. For any convex set M e x i s twhich is convex cone associated with t h e set M . For convex mapping a l e t us de- fine
Suppose t h a t z E gf a . Then
a , ( Z )
= I G : ( Z , G )
M a ( z ) j* * *
a,
(v
)=
) I : ( - z * , v * ) E K ; ( Z ) { , where Ka ( z*
) is cone dual t o Ka ( z ). Thus,z
*
E a:(Y * ) if and only ifLet u s p r o v e two auxiliary lemmas:
* *
LEMMA 1 a, ( 0 )
= -
(dom a, ).
PROOF The mapping a, i s a positively homogenious convex mapping and t h e r e - f o r e dom a, is convex cone and dual cone t o t h i s cone exists. According t o defini-
* *
tion of a, ( 0 ) i t contains t h o s e and only t h o s e elements z which satisfy t h e follow- ing condition:
which is equivalent t o
ti,
- z*
> 2 O , Z E d o m a ,The l a s t inequality means t h a t
*
*
- z E (dom a, ) The proof i s completed.
LEMMA 2
S u p p o s e t h a t
K i scone
in X , int K # 0a n d
K*= to{. T h e n
K=
X . I f X=
Rnt h e n r e q u i r e m e n t
i n t k # dc a n be d r o p p e d .
PROOF I t follows from t h e well-known t h e o r e m s of convex analysis t h a t K**
=
c l K , w h e r e symbol c l defines c l o s u r e of t h e s e t . The f a c t t h a t K*=
101 im- p l i e s c l K=
K* * =
X. T h e r e f o r e cone X i s d e n s e e v e r y w h e r e in X a n d i t i s l e f t t o p r o v e t h a t i t coincides with X. L e t u s assume t h e o p p o s i t e , namely s u p p o s e t h a t ex- i s t sz o
s u c h t h a tz o
K . T a k ez l
E i n t K a n d s e l e c t2 = 2 z 0 - z l .
The c o n e K i s d e n s e e v e r y w h e r e in X, t h u s e x i s t s e q u e n c ezk
E K such t h a tzk
42.
Let u s t a k ez l k = 2 z 0 - z k . -
I t i s c l e a r t h a tz l k
4 X a n d f o r l a r g e k w e h a v ez l k
E K . A c -1 1
-
c o r d i n g t o definition
z 0 = zzk + 2 z k ,
a n d d u e t o convexity of K w e o b t a i nz o
E K.The proof i s completed b e c a u s e in t h e finite-dimensional case assumption i n t K # q5 i s fulfilled automatically, which c a n b e v e r i f i e d in t h e s t a n d a r d way. These two lemmas l e a d t o t h e following r e s u l t :
THEOREM 1 S u p p o s e t h a t a i s a c o n v e z m a p p i n g , z
= ( z O ,
y o ) E gfa a n d t h e following c o n d i t i o n s a r e s a t i s f i e d :
1 int dom
a
# I0T h e n f o r a n y element z e z i s t s u c h n u m b e r
S>
0 ,t h a t a ( z o +
h z ) # q5f o r
all h E [O, dl. X i sof f i n i t e d i m e n s i o n t h e n t h e f i r s t r e q u i r e m e n t c a n be d r o p p e d a n d in a d d i t i o n , e z i s t s u c h a member
S>
0t h a t a ( z o + z)
# q5for
allZ , 11; 11 r
S.B e f o r e s t a r t i n g t h e proof l e t u s make o n e comment. W e h a v e in t h e s t a t e m e n t of t h e t h e o r e m c e r t a i n point y o E
a ( z O )
a n d i t is n o t defined how t o s e l e c t it. I t might seem t h e r e f o r e t h a t t h e r e s u l t of t h e t h e o r e m d e p e n d s on a p p r o p r i a t e s e l e c - tion of t h i s point. L e t u s show t h a t t h i s is n o t t h e case. F i r s t l y w e s h a l l i n t r o d u c e t h e following notations [I]:Notice t h a t Wa
( z
, y * )= + -
ifa ( z ) = 6
a n d t h e function Wa( z
, y * ) i s convex with r e s p e c t t oz .
Let us t a k ea ( z , y ) ( ~ * ) * = 1 if y F a ( z ,
y * )
a,
W a ( z , Y * ) if Y Ea ( z ,
Y * )If y
* =
0 t h e nI
T h e r e f o r e
ZZ
Wa ( x , 0 )= a
6 ( x (dom a )= -
[con(dom a-
z ) ]* .
Comparison of t h e s e s t a t e m e n t s l e a d s t o t h e conclusion t h a tf o r any point y E a ( z ) . T h e r e f o r e w e c a n t a k e a n a r b i t r a r y point in t h e s t a t e m e n t of t h e t h e o r e m 1 . Let u s start t h e proof now. If assumption 1 i s s a t i s f i e d t h e n i t i s e a s y t o g e t t h a t doma,
=
X. L e t u s s e l e c t a r b i t r a r y5 .
We h a v ez
E doma, a n d t h e r e f o r e e x i s t v e c t o ry
- =
7 ( y - y o ) , 7>
0 , ( z , y ) E gf a Taking now 6=
7 - 1 w e o b t a i n( z o
+ X Z ,
y o+
A ; )=
( ( 1 - X 7 ) z o+
X 7 z , ( 1-
X 7 ) y o+
X 7 y ) E g f af o r X E [0, dl i.e. a ( z o
+ Xz)
# 0 . In t h e c a s e of a f i n i t e dimension t h e f i r s t as- sumption i s n o t n e c e s s a r y . F u r t h e r m o r e , if X= R n
t h e n i t i s possible t o find s u c h- -
v e c t o r s
z i ,
i=
1 , .. .
, n+
1 t h a t simplex S=
I X l z l+ .
+ An Z , + I : X i 2 0,z r 2
X i=
lj c o n t a i n s 0 as i n n e r point. If w e r e d u c e t h e length of t h e v e c t o r sZi
a p p r o p r i a t e l y w e c a n o b t a i n t h a t all sets a ( z o
+ z i )
are not empty. T h e r e f o r e a n y point from t h e c e r t a i n neighborhood of z e r o c a n b e r e p r e s e n t e d as follows:This t o g e t h e r with p r o p e r t i e s of t h e convex maps implies
The proof i s completed.
This proof i s f a i r l y simple, b u t t h e r e s u l t i s q u i t e i n t e r e s t i n g . Let u s i l l u s t r a t e t h i s with some examples:
EXAMPLE 1 Take X
= R ~
Y, = R ~
A,
a n d B are m a t r i c e s r x n a n d r x m .-
0. Then Ka (0)=
Define g f a
=
) ( z , y ) : A z - B y = 0 ] a n d s e l e c t z o =O, y o -I(Z, y)
: f i - B ;=
01,* *
T h e r e f o r e a. ( y )
=
)A* u=
B* u , u E Rr j . Condition a : ( 0 )=
)O j means now t h a t B * U=
0 which implies A * u=
0. Thus, f o r solvability of t h e system of equa- tion Ax - B y=
0 with r e s p e c t t o y f o r a l l x , i t is sufficient t h a t Kern B*c
Kern A
* .
H e r e , a s usual, Kern C=
! v : C v = O jEXAMPLE 2 Take t h e same assumption as in t h e previous example and consid- e r g f a
=
) ( x , y ) : h - B y SO, y 2 0 1 . Nowwe havewhere Rf, and R y a r e positive o r t h a n t s . This gives
Hence condition a: ( y * )
=
101 implies in this c a s e t h a t from inequalities B * u 5 0 , u 2 0 t h e equality A * u=
0 follows. T h e r e f o r e f o r solvability of t h e system Ax - B y 5 0 , y r u f o r all z i t i s sufficient t h a t t h e following inclusion is satis- fied:ern
A*>=
Iu 2 0 : B * us
01.EXAMPLE 3 Suppose now t h a t X, Y, W a r e Banach s p a c e s , Z
=
X x Y , F - convex multivalued map from Z into W , i.e. F ( x , y ) C W , M-convex s u b s e t of W.Define a ( x )
=
) y : F ( x , y )n
M+ $1.
I t is c l e a r t h a t points ( x , y ) E g f a if and only if exist such point w t h a t ( x , y , w ) E gf F , w E M . Let us s e l e c t t h e point s o such t h a t a ( x o )+ 4,
y o E a ( x O ) i.e. ( s o , y o ) E gf a . Suppose t h a t w o i s such a point from M t h a t ( x o , y o , w o ) E g f F . Let us denote z o=
( s o , y o ) and define~ : ~ ( y * ) a s follows. According t o definition x * E az* ( y * ) if and only if
-
<z-
x O .0
z *
> +
<y-
y o , y* >
2 0 , ( x , y ) E gf a . Taking into account t h e description of gf a given above, w e a r r i v e a t t h e conclusion t h a t t h e l a s t inequality i s equivalent t o t h e following:f o r a l l ( x , y , w ) E g f F , w E M , where 0: i s z e r o of t h e s p a c e
w * .
But t h e last inequality is equivalent t o t h e following:Suppose now t h a t e x i s t t h e point ( x , y , y ) E gf F such t h a t (2) u E int M . This im- plies g f F
5
int (X X Y X M)+
0. I t is consequence of t h e well-known r e s u l t s of convex analysis t h a t u n d e r t h i s assumption t h e cone in t h e r i g h t hand side of (2) is equal t o t h e sum of convex cones dual t o t h e i n t e r s e c t e d ones. Taking into account easily established r e l a t i o n s* * * *
(con (X x Y x M
-
( z O , Y 0 , ~ 0 ) ) )=
(Ox I O y , (con (M - w o ) ) ) w e obtain t h a t (2) is equivalent t o t h e following statement:* *
o r , in o t h e r words, e x i s t functional w E (con (M
-
wo)) such t h a t t h e following inclusion holds:And, finally using a definition of t h e conjugate mapping introduced b e f o r e w e ob-
* * *
tain t h a t x E G* ( y
*
) only if f o r some w E (con (M-
w 0)) t h e following inclu-2 0
sion holds:
Hence a,
*
,(O)=
10{ only if t h e following inclusion* *
implies equality x
=
0, , i.e.THEOREM 2 Suppose t h a t X, Y, W a r e Banach s p a c e s , M i s a convex set w h i c h belongs to W , F i s c o n v e z m u l t i v a l u e d m a p p i n g from X X Y to W a n d
s u p p o s e t h a t
y o Ea ( x
O )a n d w
E Ma r e p o i n t s s u c h t h a t ( x o , y w o )
E gf F .T h e n t h e following c o n d i t i o n s a r e s u m c i e n t for e x i s t e n c e of n u m b e r
d>
0for a n y s u c h t h a t a ( x o + h z ) + 4 , h
E [O,63:
1 int dom
a +
@;2 g f F
9
( X XY X i n t M )+
@;* * * *
F ( z ~ w ~ ) (- (con (M
- w o > > > = lo,,
0,I .
I f X
i so f f i n i t e d i m e n s i o n t h e n i t
i sn o t n e c e s s a r y to check t h e f i r s t c o n d i t i o n . The proof follows d i r e c t l y f r o m theorem
1a n d t h e above a r g u m e n t .
EXAMPLE
4 Suppose t h a tb
andc
a r e convex multivalued mappings from X t o Y anda ( x ) = b ( x ) r.~ c ( x ) .
THEOREM
3S u p p o s e t h a t
y Ea ( x o ) a n d t h e following c o n d i t i o n s a r e s a - t i s f i e d :
1 int d o m a
+ 4;
2 gf
b n
int gfc + 4;
3
for a n y y * e i t h e r one of t h e s e t s b:o(y
*),cze0(-
* ) i se m p t y , o r e x i s t s
* * *
s u c h f u n c t i o n a l x * t h a t bz*o(y *
)= Ix 1, c
( -y
)=
)- x * 1. T h e n f o r a n y
Z E Xe x i s t
b> 0 s u c h t h a t a ( x o + h z ) + 4 f o r h
E[O,
61.I f X
i so f f i n - i t e d i m e n s i o n t h e n i t
i sn o t n e c e s s a r y to check c o n d i t i o n
1.The proof of t h e theorem follows from theorem 1, t h e f a c t t h a t gf
a =
gfb n
gfc
and d i r e c t calculation ofa z O ( O ) *
by using convex analysis tech- niques.Let us show how t o use this theorem by t h e following example. Take
b ( x ) = Ix : A x
- B y 501, c ( x )
= M where M i s a fixed convex s e t inR m , A
andB
a r e matrices of dimensionr
xn andr
Xm
,X = R n
, Y= R m .
Suppose t h a t 0 E M and e x i s t pointsx l ,
y such t h a t Axl- B y
50 ,
y E int M . A straightforward a r - gument shows t h a tKc
(0)= X
x con MW e can obtain more from t h e following r e s u l t by applying theorem 3. In o r d e r t h a t t h e system of inequalities Ax - B y 4 0, y E M h a s a solution with r e s p e c t t o y f o r a l l z from some neighborhood of z e r o i t i s sufficient t h a t conditions
* *
imply equality A* u
=
0. In f a c t b o ( y )=
@ if y*
c a n not b e r e p r e s e n t e d in t h e* *
form y = B * u , u 2 0 . In case if y = B * u , u 2 0 , and - B * U Z ( c o n ~ ) * . If, however,
* * *
t h e n f o r f *
=
B u w e h a v e c o ( - y )=
101. T h e r e f o r e according t o t h e theorem* *
3 t h e set b o ( y ) a l s o should contain only one point i.e. z e r o . I t follows from
* * *
r e p r e s e n t a t i o n of t h e set b o ( y ) t h a t A u
=
0.If M
= Ry
i.e. M consists of nonnegative v e c t o r s t h e n t h i s r e s u l t coincides with one obtained from example 2.2. LOCALLY SMOOTH MAF'S
Let K b e a convex cone in a Banach s p a c e X. I t is obvious t h a t LinK = K
- K
is t h e minimal l i n e a r manifold containing K. If M i s a convex set, t h e n LinM =con(M - 2 ) -con(M - 2 ) ,
where z i s a n a r b i t r a r y point of M . I t is not difficult t o show t h a t LinM does not depend upon z E M
.
W e s h a l l s a y t h a t t h e point x belongs t o t h e r e l a t i v e i n t e r i o r of M (denote i t r i M ) , if f o r some E
>
0x
+
(EB)n
Lin Mc
M ,wnere B i s t h e unit ball of t h e s p a c e X with t h e c e n t e r in t h e origin.
If X
= R",
t h e n i t i s well known [Z], [3], t h a t r i M+ 4.
In g e n e r a l t h i s r e s u l t i s n o t t r u e .L e t M b e a n a r b i t r a r y set of X.
DEFINITION Call t h e set K m a r q u e e for M
at
t h e p o i n t x € M, if'K i s t h e con- v e x cone a n d for e a c h € K e x i s t v a l u e s E>
0, d>
0 a n d c o n t i n u o u s l y d i f - ferentiablein
t h e neighborhood of t h e o r i g i n f u n c t i o n+:
X -4 X e x i s t s u c h t h a t1 +(O)
=
0, +'(O)=
I ( u n i t operator);2 z
+ +(y)
E M for a l lThis definition i s b a s e d on i d e a s of V.G. Boltyanski, h e developed them f o r
n-
dimensional s p a c e . But t h e s e i d e a s c a n n o t b e generalized o n infinite dimensional s p a c e without c h a n g e s . F o r t h i s r e a s o n t h e i n t r o d u c e d definition d i f f e r s f r o m V.G.Boltyanski's definition [3].
F u r t h e r i t i s convenient t o s u p p o s e without r e d u c i n g g e n e r a l i t y t h a t point z coincide with t h e origin.
THEOREM 4 Let M o i s
a
s e t , 0 € M o . KO t h e m a r q u e e for M oat
p o i n t x=
0 a n d f u n c t i o n s f i ( z ) , i=
1,. . .
, K satisf'g t h e c o n d i t i o n s :1 f i ( 0 ) = 0 , i = ,
. . . .
K ;2
yi
a r e c o n t i n u o u s l y Freshet-dinerentiablein
t h e neighborhood of zero a n d d e r i v a t i v e s f i ( 0 ) a r e l i n e a r l y i n d e p e n d e n t o n L o=
Lin KO.T h e n
i s t h e Local m a r q u e e a t t h e p o i n t x
=
0 for t h e setPROOF Since f i ( 0 ) are linearly independent on l i n e a r manifold L o , t h e n vec- t o r s ej E L o e x i s t [ I ] , such t h a t
Let E K. Then
and smooth in t h e neighborhood of z e r o . T h e r e f o r e e x i s t s function q0 such t h a t q O ( o )
=
0 ,q&o) =
I ,f o r a l l
i7
E [ c o n ( z+
( E B )n
L O ) ]n
( 6 B ) ,f o r E
>
0 , 6>
0 . Letk
where z i s t h e v e c t o r with components z j , j
=
1 ,. . .
, k . Let u s consider t h e system of equationsg i ( y , z ) = 0 , i = I , .
. .
, k with r e s p e c t t o z.
I t i s evidentSince gi are compositions of smooth functions, t h e n they a r e smooth functions. U s - ing r u l e s f o r differentiation of t h e complex function i t i s not difficult t o g e t
gi& ( 0 , 0 )
= P i
( O ) q i , ( 0 )- P i
( 0 )=
0 , Because 4'6 ( 0 )=
I . F u r t h e rThus, t h e matrix with elements g i , , (0, 0) i s non-degenerate a n d a c c o r d i n g t o t h e
1
implicit f u n c t i o n s t h e o r e m [4] t h e solution z (y') of t h e system (7) e x i s t s , s u c h t h a t z (0)
=
0. F u r t h e r a c c o r d i n g t o t h e same t h e o r e m z(c)
i s continuously d i f f e r e n t i - a b l e in t h e neighborhood of t h e origin of c o o r d i n a t e s a n d t a k i n g i n t o a c c o u n t (8)Let now
Then Q(0)
=
0 , a n da c c o r d i n g t o (10). Denote
Let u s c h o o s e el
>
0, dl>
0 sufficiently small s o t h a t t h e inclusiony' E [con (5 + (clB)
fl
LO)]n
( ~ I B ) (12)e n s u r e t h a t
c
s a t i s f i e s t h e inclusion (5). I t is possible, b e c a u s e ( l o ) , b u t ifc
sa-t i s f i e s (5) t h e n
Denote
L1
= ) Z :
f i ( 0 ) Z = O f i=
1 , .. .
, K {.
I t i s e a s y t o see t h a t LinK G L o r-?L1
.
Let now
y
E [con(5 +
(elB) r\ LinK)]n
(dlB).
(13)The set from t h e r i g h t p a r t of (13) i s t h e s u b s e t of set from t h e r i g h t p a r t of (12).
F o r t h i s r e a s o n (13) implies t h e inclusion Q(G) E Mo. F u r t h e r , s i n c e E L1, t h e n
E L 1 . T h e r e f o r e due to (7) and (6)
Thus (13) assumes
'k(y)
E M and consequently K is t h e local marquee f o r M .Since t h e p r o o f s of t h e following theorems r e p e a t t h e similar p a r t s of t h e theorem 4 proof t h e details will b e omited.
THEOREM 5 Let M 1 a n d M 2 be t w o s e t s , M
=
M 1n
M 2 , 0 E M a n d K1, K2 be r e s p e c t i v e l y m a r q u e e s f o r M 1 a n d M 2 .V
t h e Linear m a n i f o l d s e x i s t s u c h t h a t : 1 L 1 c L i n K 1 , L 2 ~ L i n K 2 ;2 L 1
+
L 2 = X ;3 for a l l x l E L l , x2 E L 2
t h e n K
=
K1n
K2 i s t h e m a r q u e e for M a t t h e p o i n t 0.PROOF In a c c o r d a n c e t o t h e condition 2 a r b i t r a r y x E X can b e r e p r e s e n t e d as follows
That r e p r e s e n t a t i o n i s unique, because if d i f f e r e n t r e p r e s e n t a t i o n e x i s t s
then
i.e. x l = x i , x 2 = x i .
Consequently o p e r a t o r s P 1 and P 2 are defined such t h a t
Due t o t h e condition 3
i t i s e a s y t o see t h a t o p e r a t o r s Pi are l i n e a r a n d consequently t h e y a r e l i n e a r a n d continuous o p e r a t o r s .
Let now 5 EK1
n
K 2 a n d E , d i , \ki c o r r e s p o n d t o5
in t h e m a r q u e e Ki. Consid- e r t h e equationwith r e s p e c t t o z . I t i s e a s y t o see
In a c c o r d a n c e t o t h e implicit functions t h e o r e m t h e smooth function z ( c ) i s de- fined in t h e neighborhood of z e r o a n d z ( 0 )
=
0 , z ' ( 0 )=
0 .Suppose
Taking into a c c o u n t z ' ( 0 )
=
0 , i.e.z
(-1
+ II
YI1
0 , f o rI I ~ O --,o
i t i s n o t difficult t o p r o v e t h a t \k i s t h e d e s i r e d function f o r
E .
This p r o o f i s b a s e d o n t h e f a c t t h a t f o r
c
c l o s e t o t h e d i r e c t i o n 'kl(ii + p l z( 5 ) )
E M 1.
\k2(Y + P ~ Z
(GI)
E M 2 Ia n d consequently d u e t o ( 1 5 )
\k(y)
E M ln
M 2 . In t h i s connection ify
E LinKi, i=
1 , 2 , t h e n5 +
P l z(c)
E L l ,-
P 2 z( 5 )
E L 2 a n d taking i n t o a c c o u n t t h e smallness of z( y )
all conditions r e l a t e d with t h e c h o i c e of E , 6 c a n b e satisfied. Be- s i d e sb e c a u s e \k; ( 3 )
=
I , z ' ( 0 )=
0 . The proof i s completed.REMARK Conditions 1-3 f o r X
=
Rn c a n b e r e p l a c e d by t h e conditionIn a g e n e r a l c a s e t h a t condition is not sufficient b e c a u s e condition 2 means t h a t t h e s e t s M1 a n d M 2 h a v e sufficiently l a r g e dimensions. The condition 3 means t h a t l i n e a r manifolds i n t e r s e c t at some angle. Indeed, if X i s a H i l b e r t s p a c e with i n n e r p r o d u c t [ x , y ] t h e n condition 3 i s equivalent t o condition
Let u s c o n s i d e r now t h e t h e o r e m a b o u t implicit functions f o r nonconvex multivalued maps.
THEOREM 6 Let a be a m u l t i v a l u e d m a p , K be a m a r q u e e f o r g~' a L X X Y at t h e p o i n t z
=
( x o , y o ) ,If t h e following conditions are s a t i s f i e d 1 i n t dom a,, # @ ;
2 e x i s t a l i n e a r r e s t r i c t e d o p e r a t o r P : X -+ Y s u c h t h a t ( z , P z ) E Lin K ; a,O
*
(0)=
I O j , t h e n f o r a n y e x i s t s d>
0 s u c h t h a ta ( x o + X z ' ) #
4
f o r X E [ O , d l.
If X and I' h a v e f i n i t e dimensions t h e n t h e f i r s t two conditions are satisfied and be- s i d e s value d
>
0 e x i s t s s u c h t h a t a ( x o+
z ) # @ f o r a l l5
E (dB).PROOF Without l o s s of t h e g e n e r a l i t y w e c o n s i d e r x
=
0 , y o=
0.Lemma 2 implies
i.e. f o r any ji EX t h e v e c t o r e x i s t s s u c h t h a t
(z, c)
E K. Consequently f o r anyz
E X t h e v e c t o rc
EY e x i s t s s u c h t h a t ( z , c ) E Lin K.Thus, l e t Z o b e a v e c t o r f r o m X and z o
=
( Z o , g o ) E K . S i n c e K i s t h e m a r q u e e t h e n E>
0 , 6>
0 and e x i s t s smooth function 9 ( 0 )=
0 , 9'(O)=
I,, w h e r e I, i s t h e unit o p e r a t o r in Z a n d 9 ( z ) E gf a f o r a l lw h e r e B, i s t h e unit ball in Z. S i n c e
Z =
X x Y, t h e na n d condition 9 ' ( 0 )
=
I, c a n b e r e w r i t t e n as followsi.e. 91;-(0, 0 ) =I,, 9 ' ( 0 , 0 ) = 0 , ' k k ( 0 , 0 ) = 0 , 'k&(O, 0 ) = I y . lfi
Consider t h e system of e q u a t i o n s
w h e r e A E
R',
z E X. Taking i n t o a c c o u n t p r e v i o u s r e l a t i o n s w e g e tThus d u e to t h e t h e o r e m a b o u t implicit f u n c t i o n s t h e system (17) h a s t h e solution r ( A ) a n d a l s o
L e t u s c o n s i d e r now t h e point
; ( A )
=
(Ago+
r ( A ) ,Ago +
Pr ( A ) ).
Taking i n t o a c c o u n t r ' ( 0 )
=
0 , r (A)=
o ( A ) f o r sufficiently s m a l l A>
0 inclusion (16) i s s a t i s f i e d , b e c a u s e by definition of t h e o p e r a t o r P w e h a v e ;(A) E LinK.Consequently
. k ( z ( A ) ) E gf a f o r small A. But
h e r e t h e condition (17) w a s used. Thus
i.e. a ( A s o ) # t$ f o r small A. Thus t h e f i r s t p a r t of t h e t h e o r e m i s p r o v e d . L e t u s c o n s i d e r t h e case with t h e f i n i t e dimension. Condition 1 c a n b e omitted in a c c o r - d a n c e t o lemma 1 a n d 2. W e will show t h a t condition 2 c a n b e omitted too.
Indeed b e c a u s e LinK G R n X R m t h e n e x i s t t h e m a t r i c e s A a n d B with dimen- sions r x n a n d r X m r e s p e c t i v e l y s u c h t h a t t h e points ( x , y ) E LinK a n d only t h e y s a t i s f y t h e equations
w h e r e rows of t h e matrix (A, - B ) with t h e dimension r x ( n
+
m ) are l i n e a r l y in- dependent. This follows f r o m t h e f a c t t h a t in t h e f i n i t e dimensional s p a c e l i n e a r manifold c a n b e d e s c r i b e d as set of solution of some l i n e a r equations system.S i n c e t h e rows of t h e m a t r i x (A, - B ) are l i n e a r l y independent t h e n e x i s t non-degenerate submatrix of t h i s matrix with dimension r x r . Consequently system
(18) h a s solution y f o r a r b i t r a r y x
.
F o r t h i s r e a s o n ( s e e example 15
1)K e r n B * c_ K e r n A * (19)
* *
However K e r n B *
=
f O1.
Indeed l e t y E R ~ , y # 0 a n d B* y* =
0, t h e n d u e t o (19) A * y* =
0. This means t h a t e x i s t t h e non-zero v e c t o r y*
o r t h o g o n a l t o a l l columns of t h e matrix (A, - B ) . The l a s t s t a t e m e n t c o n t r a d i c t s t h e e x i s t e n c e of non-degenerated r X r s u b m a t r i x of t h e m a t r i x ( A ,-
B ) . Thus K e r n B* =
fOj, i.e.columns of t h e matrix B* are l i n e a r l y independent. F o r t h i s r e a s o n B* (and conse- quently B ) c o n t a i n s non-degenerated submatrix B1 with r a n k r X r . L e t B
=
(B1, B2). Consider t h e v e c t o r s yw h e r e y E
R~
a n d 0,-,
i s t h e non-zero v e c t o r with dimension m-
r . I t i s c l e a r t h a t By=
Bly l. If y=
B1-lAx, t h e n v e c t o rs a t i s f i e s (18). I t i s obvious t h a t t h e l i n e a r o p e r a t o r
s a t i s f i e s condition 2 of t h e t h e o r e m . Q.E.D.
REMARK Let u s c o n s i d e r t h e condition 2 in g e n e r a l . Let L C X XY i s a l i n e a r manifold. Denote
I t i s n o t difficult t o see t h a t 1 1 ( x ) i s a n a f f i n e manifold;
2 1 ( x l
+
x z )=
1 ( x l )+
1 ( x z ) ; 3 L ( X x ) = X l ( z ) f o r X + O .Let in a c c o r d a n c e t o t h e t h e o r e m 6 dom 1
=
X. Thus t h e map 1 from t h e s p a c e X t o t h e set of a f f i n e manifolds of y is l i n e a r . Condition 2 of t h e o r e m 6 implies t h a t ex- i s t a l i n e a r continuous map P such t h a t Pz € 1 ( x ) . S i n c e 1 is a l i n e a r map t h e n ex- i s t e n c e of P i s a n a t u r a l condition. A s was shown e a r l i e r in a f i n i t e dimensional s p a c e , t h i s o p e r a t o r e x i s t s . I t would b e i n t e r e s t i n g in g e n e r a l t o formulate a n addi- tional condition f o r L g u a r a n t e e i n g e x i s t e n c e of a continuous l i n e a r selector of P.Let u s c o n s i d e r now t h e questions connected with t h e c o n s t r u c t i o n of a mar- q u e e f o r a convex set.
THEOREM 7 Let M be a convex s e t i n B a n a c h s p a c e X, 0 E M and r i M
+ 4.
Then t h e c o n e K
=
c o n ( r i M )is t h e m a r q u e e f o r s e t M a t t h e p o i n t z
=
0.PROOF I t i s e v i d e n t t h a t r i M
c
M , f o r t h i s r e a s o n Lin ( r i M)c
LinM a n d LinL L LinM.Let
5
E K i.e. z- =
y x , y > 0 , z E r i M . If x f O t h e n f o r some E > O x+
(EB)n
LinMc
M in a c c o r d a n c e t o t h e definition of t h e r e l a t i v e i n t e r i o r of a set M.Let
where y i s a n a r b i t r a r y point of Lin
K , 11
y1)
g 1. Ift h e n y' E M d u e t o t h e definition ri M . Since 5
+
0 , t h e nF o r t h i s r e a s o n if
< (Ig (1
t h e nThus if
>
0 , 6>
0 are chosen s o t h a tt h e n f o r
y
satisfying conditiony
E [con( g + ( c l B ) n
LinK)]n
( d B )the, following inciusion is t r u e :
c
E M . Thus, in t h i s c a s e9(5) = c
c a n be taken.If 5
=
0 t h e n in a c c o r d a n c e with t h e definition of t h e r e l a t i v e i n t e r i o r e x i s t s such E>
0 t h a t( E B )
n
Lin K L Mi.e. any point of LinK L K with norm less t h a n E belongs t o M. lt is c l e a r t h a t in t h i s c a s e
9 ( y ) = 5
too. Q.E.D.Let u s now consider some applications of t h e s e r e s u l t s . In p a r t i c u l a r it is in- t e r e s t i n g f o r u s t o generalize t h e implicit functions theorem in c a s e s when solu- tions belong t o some set M. I t is formulated below.
THEOREM 8
Let t h e f u n c t i o n s f c ( z ) ,
i=
1,. . .
, Kbe d e f i n e d o n t h e s p a c e
Z=
Xx Y, t h e s e m n c t i o n s be smooth in t h e neighborhood of t h e o r i g i n of coor- d i n a t e s ,
Mbe
ac o n v e x s e t c o n t a i n i n g
0.Let in a d d i t i o n :
1
g r a d i e n t s f i ( z o ) a r e L i n e a r l y i n d e p e n d e n t o n s u b s p a c e
LinM;2
e x i s t p o i n t z s u c h that
3
for a n y vector
u E R kt h e set
i s
e m p t y or c o n s i s t s from t h e u n i q u e vector
f ;* u .
Then for a n y vector z, [I 2 1) <
d e x i s t s vectors u c h t h a t
PROOF Define
a ( x ) = l y : f i ( x , y ) = O , i =I. . . . ,
k ,( x , y )
E M J.
In a c c o r d a n c e with t h e theorems 4 and 7 t h e cone
is t h e marquee f o r gf
a
at t h e pointz o = ( x o , y o ) .
Taking into account assumptions and well known theorems of convex analysis w e g e twhere
f ; ( z O )
is t h e F r e s h e t derivative of t h e map f : R n + m -+ R k , i.e. matrix with rowspi, ( z o )
E R n + m. Condition 3 of t h e theorem 6 means t h a t r e l a t i o n s( X I * , y * )
E (con(M- z o ) ) *
* *
Y + f $ ( z 0 b
= D assume t h e equality* *
x +p3; ( z 0 ) u = o .
The l a s t condition i s equivalent t o condition 3 of t h e theorem.
THEOREM 9
Let
Z=
R n X R m ,Pi ( z ) ,
i=
1,. . .
, kbe a smooth f u n c t i o n a n d
Ube a convex set
i n R n.
Ip( x o , y
O ) i sa point s u c h t h a t
t h e n for t h e existence of t h e v a l u e d >
0s u c h t h a t for a n y
E R ne x i s t vector
5
E R ms a t i s j 5 i n g
i t i s su.tj%cient
1 the vectors f f ; ( z o ) a r e LinearLy independent;
2 e x i s t vector
( z l , y l )
s u c h t h a t3 The set
c o n t a i n s onLy zero.
The proof follows d i r e c t l y from t h e p r e v i o u s t h e o r e m , t a k i n g i n t o a c c o u n t t h e e q u i v a l e n c e of e q u a l i t i e s f g ( z o ) u
=
0 a n d u=
0 which, in t u r n follows f r o m l i n e a r independence of v e c t o r s ft;
( z O ) , i=
1,. . .
, kL e t u s c o n s i d e r now t h e solvability of t h e system of i n e q u a l i t i e s
for a n y x from vicinity of some point x o . S u p p o s e t h a t t h e point ( x o , y o ) i s o n e of t h e solutions of t h i s system.
This p r o b l e m c a n b e r e d u c e d to t h e p r e v i o u s o n e by introducing a u x i l i a r y v a r i a b l e s w i , i
=
1,. . .
, k a n d c o n s i d e r i n g t h e following system:The t h e o r e m 8 c a n b e a p p l i e d now. To d o t h i s let u s t a k e X
=
Rn a n d t h e s p a c e Y from t h i s t h e o r e m will b e t h e s p a c e of p a i r s ( y , w ) E Rm XR ~ .
T h e set M i s now t h e set ( R n , R m , R : ) . T h e r e f o r e LinM=
( R n . R m , R k ) . Let us n o t e t h a t in t h e conditions ( 2 1 ) e a c h new v a r i a b l e c o r r e s p o n d s to s e p a r a t e equality, t h e r e f o r e condition 1 of t h e o r e m 8 i s t r u e . F u r t h e r m o r e , w e c a n assume without l o s s of gen- e r a l i t y t h a tThis assumption will considerably simplify t h e argument. What is needed now f o r fulfillment of t h e second condition of t h e theorem is existence of t h e v e c t o r z l
- =
(5 l, GI) such t h a tDue t o t h e f a c t t h a t M
=
( R n , R m , Rk+) w e have [con (M-
Zo)l* =
(On, O m , Rk,).
The third condition of theorem 8 easily follows now from t h e assumption t h a t condi-
* *
tions u r 0 ,
f $
( z 0 ) u=
0 i m p l yf $
( z o ) u=
0. O r in o t h e r words Kern f; ( z O ) 2 (Kern* fi*
(20))f7
R*Thus, we have obtained t h e following r e s u l t :
THEOREM 1 0 S u p p o s e t h a t x E R n , y E k m , j'unctions
Pi
( z ) , i=
1,. . .
, k a r e smooth f o r z=
( x , y ) a n d t h e p o i n t z o=
(xo, y o ) is s u c h t h a tLet US t a k e i n a d d i t i o n t h e following a s s u m p t i o n s : 1 E x i s t s vector
zl =
( z l , GI) s u c h t h a t2 Kern
I; *
( z O ) 2 (Kern f $ ( z O ) )n
Rk+Then e z i s t s d
>
0 s u c h t h a t f o r a n yz,
llz'11 <
d e z i s t s s u c h t h a tREFERENCES
1 Pshenichny, B.N.: Convex analysis and extremal problems. Moscow, Nauka, 1980 (in Russian).
2 Rockafellar, T. : Convex analysis. Princeton Univ. P r e s s , Princeton NJ, 1970.
3 Boltiansky, B.G.: The method of marquees in t h e t h e o r y of extremal problems.
Uspekhi motematicheskih nauk, 1975, v. 30, No. 3 , pp. 1-55 (in Russian).