Extension of Hölder functions and some related problems of best approximation

Abstract

Authors

Costica Mustata
“Tiberiu Popoviciu” Institute of Numerical Analysis, Romanian Academy, Romania

Keywords

Paper coordinates

C. Mustăţa, Extension of Hölder functions and some related problems of best approximation, ”Babeş-Bolyai” Univ., Research Seminars, Seminar on Mathematical Analysis, Preprint nr. 7 (1991), 71-86.

PDF

About this paper

Journal
Publisher Name
DOI
Print ISSN
Online ISSN

MR # 94f: 41038

google scholar link

[1] Aronsson, G., Extension of functions satisfying Lipschitz conditions, Arkiv for Mathematik, 6, 28(1967), 551-561
[2] Czipser, J. and Cheher, L., Extension of funcitons satisfying a Lipschitz condition, Acta Math. Acad. Sci. Hungar 6(1955), 213-220.
[3] Levy, R and Rice, M.D, The approximation of uniformly continuous mappings by Lipschitz and Holder mappings, (preprint) 1980, 29 p.
[4] McShane, E.J., Extension of range of functions, Bull. Amer. Soc. 40 (1934), 837-842.
[5] Mustăța, C., Best approximation and unique extension of Lipschitz funcitons, Journal Approx. Theory 19 (1977), 222-230.
[6] Singer, I., Best approximation in normed linear spaces by elements of linear subspaces, Springer-Verlag, 171, 1970.
[7] Shubert, B., A sequential method seeking the global maximum of a function, SIAM J.Num. Anal. 9(1972), 379-388.
[8] Shubert, B., The Otto Dunkel Memorial Problem Book, New York, 1956

Paper (preprint) in HTML form

1991-Mustata-UBB-Seminar-Extension-of-Holder-functions-and-some-related
Costică Mustata
  1. Let ( X , d ) ( X , d ) (X,d)(X, d)(X,d) be a metric space and α ( 0 , 1 ] α ( 0 , 1 ] alpha in(0,1]\alpha \in(0,1]α(0,1]. A function f : X R f : X R f:XrarrR\mathrm{f}: \mathrm{X} \rightarrow \mathrm{R}f:XR is called HÖlder of class α α alpha\alphaα on X if there exists K 0 K 0 K >= 0K \geq 0K0 such that
    (1.1) | f ( x ) f ( y ) | K d α ( x , y ) | f ( x ) f ( y ) | K d α ( x , y ) quad|f(x)-f(y)| <= Kd^(alpha)(x,y)\quad|f(x)-f(y)| \leq K d^{\alpha}(x, y)|f(x)f(y)|Kdα(x,y),
    for all x , y X x , y X x,y in Xx, y \in Xx,yX.
    Put
    (1.2) f α , x = sup { | f ( x ) f ( y ) | / d α ( x , y ) : x , y x , x v y } f α , x = sup | f ( x ) f ( y ) | / d α ( x , y ) : x , y x , x v y quad||f||_(alpha,x)=s u p{|f(x)-f(y)|//d^(alpha)(x,y):x,y in x,xvy}\quad\|f\|_{\alpha, x}=\sup \left\{|f(x)-f(y)| / d^{\alpha}(x, y): x, y \in x, x v y\right\}fα,x=sup{|f(x)f(y)|/dα(x,y):x,yx,xvy}. Then f α , X f α , X ||f||_(alpha,X)\|f\|_{\alpha, X}fα,X is the smallest constant K 0 K 0 K >= 0K \geq 0K0 for which the inequality (1.1) holds and is called Hölder norm of f f fff.
Denote by Λ α ( X , d ) Λ α ( X , d ) Lambda_(alpha)(X,d)\Lambda_{\alpha}(X, d)Λα(X,d) the set of all Hölder functions of class α α alpha\alphaα on X [ 3 ] X [ 3 ] X[3]X[3]X[3]. Ther Λ α ( X , d ) Λ α ( X , d ) Lambda_(alpha)(X,d)\Lambda_{\alpha}(X, d)Λα(X,d) is a vector lattice, that is, it is closed under the operations of addition, multiplication by scalars and formation of supremum and infimum of two of its elements.
For a nonvoid subset Y Y YYY of X X XXX, the Hölder norm f α , Y f α , Y ||f||_(alpha,Y)\|f\|_{\alpha, Y}fα,Y and the space Λ α ( Y , d ) Λ α ( Y , d ) Lambda_(alpha)(Y,d)\Lambda_{\alpha}(Y, d)Λα(Y,d) are defined similarly.
THEOREM 1. Let ( X , d X , d X,d\mathrm{X}, \mathrm{d}X,d ) be a metric space, Y X Y X YsubX\mathrm{Y} \subset \mathrm{X}YX and
z If f Λ α ( y , d ) then the functions ( 1.3 ) F 1 ( x ) = inf { f ( y ) + f α , y d α ( x , y ) : y y } , x X (1.3) F 2 ( x ) = sup { f ( y ) f α , y d ( x , y ) : y y } , x X z  If  f Λ α ( y , d )  then the functions  ( 1.3 ) F 1 ( x ) = inf f ( y ) + f α , y d α ( x , y ) : y y , x X (1.3) F 2 ( x ) = sup f ( y ) f α , y d ( x , y ) : y y , x X {:[zinquad" If "finLambda_(alpha)(y","d)" then the functions "],[(1.3)quadF_(1)(x)=i n f{f(y)+||f||_(alpha,y)d^(alpha)(x,y):yiny}","xinX],[(1.3)F_(2)(x)=s u p{f(y)-||f||_(alpha,y)(d)(x,y):yiny}","xinX]:}\begin{align*} & \mathrm{z} \in \quad \text { If } \mathrm{f} \in \Lambda_{\alpha}(\mathrm{y}, \mathrm{~d}) \text { then the functions } \\ & \mathrm{(1.3)} \quad \mathrm{~F}_{1}(\mathrm{x})=\inf \left\{\mathrm{f}(\mathrm{y})+\|\mathrm{f}\|_{\alpha, \mathrm{y}} \mathrm{~d}^{\alpha}(\mathrm{x}, \mathrm{y}): \mathrm{y} \in \mathrm{y}\right\}, \mathrm{x} \in \mathrm{X} \\ & \mathrm{~F}_{2}(\mathrm{x})=\sup \left\{\mathrm{f}(\mathrm{y})-\|\mathrm{f}\|_{\alpha, \mathrm{y}} \mathrm{~d}(\mathrm{x}, \mathrm{y}): \mathrm{y} \in \mathrm{y}\right\}, \mathrm{x} \in \mathrm{X} \tag{1.3} \end{align*}z If fΛα(y, d) then the functions (1.3) F1(x)=inf{f(y)+fα,y dα(x,y):yy},xX(1.3) F2(x)=sup{f(y)fα,y d(x,y):yy},xX
are extension of f f fff, i.e.
a) F 1 | Y = F 2 | Y = f F 1 Y = F 2 Y = f quadF_(1)|_(Y)=F_(2)|_(Y)=f\left.\quad F_{1}\right|_{Y}=\left.F_{2}\right|_{Y}=fF1|Y=F2|Y=f,
b) F 1 α , X = F 2 α , X = f α , Y F 1 α , X = F 2 α , X = f α , Y quad||F_(1)||_(alpha,X)=||F_(2)||_(alpha,X)=||f||_(alpha,Y)\quad\left\|F_{1}\right\|_{\alpha, X}=\left\|F_{2}\right\|_{\alpha, X}=\|f\|_{\alpha, Y}F1α,X=F2α,X=fα,Y.
Theorem 1 follows from Corollary 1.2 in [3].
For f Λ α ( Y , d ) f Λ α ( Y , d ) f inLambda_(alpha)(Y,d)f \in \Lambda_{\alpha}(Y, d)fΛα(Y,d) denote by E Y ( f ) E Y ( f ) E_(Y)(f)E_{Y}(f)EY(f) the set of all extensions of f f fff in Λ α ( X , d ) Λ α ( X , d ) Lambda_(alpha)(X,d)\Lambda_{\alpha}(X, d)Λα(X,d), i.e.
(1.4) E Y ( E ) = { F Λ α ( X , d ) : F | Y = f , F α , X = f a , Y } E Y ( E ) = F Λ α ( X , d ) : F Y = f , F α , X = f a , Y quadE_(Y)(E)={F inLambda_(alpha)(X,d):F|_(Y)=f,||F||_(alpha,X)=||f||_(a,Y)}\quad E_{Y}(E)=\left\{F \in \Lambda_{\alpha}(X, d):\left.F\right|_{Y}=f,\|F\|_{\alpha, X}=\|f\|_{a, Y}\right\}EY(E)={FΛα(X,d):F|Y=f,Fα,X=fa,Y}.
Let C C CCC be a convex subset of a vector space V V VVV. A A AAA subset: H H HHH of C C CCC is called a face of C C CCC if λ x + ( 1 λ ) y H λ x + ( 1 λ ) y H lambda x+(1-lambda)y in H\lambda x+(1-\lambda) y \in Hλx+(1λ)yH for some λ ( 0 , 1 ) λ ( 0 , 1 ) lambda in(0,1)\lambda \in(0,1)λ(0,1) and some x , y C x , y C x,y in Cx, y \in Cx,yC, implies x , y H x , y H x,y in Hx, y \in Hx,yH. A one-point face of C C CCC is called an extremal element of C C CCC.
In Theorem 2 below we present some properties of the set E y ( f ) E y ( f ) E_(y)(f)E_{y}(\mathrm{f})Ey(f).
THEOREM 2. Let ( X , d X , d X,d\mathrm{X}, \mathrm{d}X,d ) be a metric space, Y a nonvoid subset of X , α ( 0 , 1 ] X , α ( 0 , 1 ] X,alpha in(0,1]\mathrm{X}, \alpha \in(0,1]X,α(0,1] and f Λ α ( Y , d ) f Λ α ( Y , d ) finLambda_(alpha)(Y,d)\mathrm{f} \in \Lambda_{\alpha}(\mathrm{Y}, \mathrm{d})fΛα(Y,d). Then
a) E Y E Y E_(Y)\mathrm{E}_{Y}EY (f) is a convex subset of Λ α ( X , d ) Λ α ( X , d ) Lambda_(alpha)(X,d)\Lambda_{\alpha}(\mathrm{X}, \mathrm{d})Λα(X,d);
b) For every F E 1 y ( f ) , F 1 ( x ) F ( x ) F 2 ( x ) , x X F E 1 y ( f ) , F 1 ( x ) F ( x ) F 2 ( x ) , x X F inE_(1y)(f),F_(1)(x) >= F(x) >= F_(2)(x),quad x in XF \in E_{1 y}(f), F_{1}(x) \geq F(x) \geq F_{2}(x), \quad x \in XFE1y(f),F1(x)F(x)F2(x),xX, where the functions F 1 F 1 F_(1)\mathrm{F}_{1}F1 and F 2 F 2 F_(2)\mathrm{F}_{2}F2 are defined by (1.3);
c) The functions F 1 F 1 F_(1)\mathrm{F}_{1}F1 and F 2 F 2 F_(2)\mathrm{F}_{2}F2 defined by ( 1 , 3 ) ( 1 , 3 ) (1,3)(1,3)(1,3) are extremal elements of E Y ( f ) E Y ( f ) E_(Y)(f)\mathrm{E}_{Y}(f)EY(f).
Proof. a) For F , G E Y ( f ) F , G E Y ( f ) F,G inE_(Y)(f)F, G \in E_{Y}(f)F,GEY(f) and λ [ 0 , 1 ] λ [ 0 , 1 ] lambda in[0,1]\lambda \in[0,1]λ[0,1] we have
( λ F + ( 1 λ ) G ) | Y = λ F Y + ( 1 λ ) G Y = λ f + ( 1 λ ) f = f ( λ F + ( 1 λ ) G ) Y = λ F Y + ( 1 λ ) G Y = λ f + ( 1 λ ) f = f (lambdaF+(1-lambda)G)|_(Y)=lambdaF_(Y)+(1-lambda)G_(Y)=lambdaf+(1-lambda)f=f\left.(\lambda \mathrm{F}+(1-\lambda) \mathrm{G})\right|_{\mathrm{Y}}=\lambda \mathrm{F}_{\mathrm{Y}}+(1-\lambda) \mathrm{G}_{\mathrm{Y}}=\lambda \mathrm{f}+(1-\lambda) \mathrm{f}=\mathrm{f}(λF+(1λ)G)|Y=λFY+(1λ)GY=λf+(1λ)f=f.
Since
λ F + ( 1 λ ) G α , X λ F α , X + ( 1 λ ) G α , X = = λ f α , Y + ( 1 λ ) f α , Y = f α , Y λ F + ( 1 λ ) G α , X λ F α , X + ( 1 λ ) G α , X = = λ f α , Y + ( 1 λ ) f α , Y = f α , Y {:[||lambda F+(1-lambda)G||_(alpha,X) <= lambda||F||_(alpha,X)+(1-lambda)||G||_(alpha,X)=],[=lambda||f||_(alpha,Y)+(1-lambda)||f||_(alpha,Y)=||f||_(alpha,Y)]:}\begin{aligned} \|\lambda F+(1-\lambda) G\|_{\alpha, X} & \leq \lambda\|F\|_{\alpha, X}+(1-\lambda)\|G\|_{\alpha, X}= \\ & =\lambda\|f\|_{\alpha, Y}+(1-\lambda)\|f\|_{\alpha, Y}=\|f\|_{\alpha, Y} \end{aligned}λF+(1λ)Gα,XλFα,X+(1λ)Gα,X==λfα,Y+(1λ)fα,Y=fα,Y
and
λ F + ( 1 λ ) G α , X ( λ F + ( 1 λ ) G ) | Y | α , Y = f | α , Y λ F + ( 1 λ ) G α , X ( λ F + ( 1 λ ) G ) Y α , Y = f α , Y ||lambda F+(1-lambda)G||_(alpha,X) >= ||(lambda F+(1-lambda)G)|_(Y)|_(alpha,Y)=||f|_(alpha,Y)quad\|\lambda F+(1-\lambda) G\|_{\alpha, X} \geq\left.\left\|\left.\left.(\lambda F+(1-\lambda) G)\right|_{Y}\right|_{\alpha, Y}=\right\| f\right|_{\alpha, Y} \quadλF+(1λ)Gα,X(λF+(1λ)G)|Y|α,Y=f|α,Y.
it follows that
λ F + ( 1 λ ) G a , X = f α , Y λ F + ( 1 λ ) G E Y ( f ) λ F + ( 1 λ ) G a , X = f α , Y λ F + ( 1 λ ) G E Y ( f ) {:[||lambda F+(1-lambda)G||_(a,X)=||f||_(alpha,Y)],[lambda F+(1-lambda)G inE_(Y)(f)]:}\begin{aligned} & \|\lambda F+(1-\lambda) G\|_{a, X}=\|f\|_{\alpha, Y} \\ & \lambda F+(1-\lambda) G \in E_{Y}(f) \end{aligned}λF+(1λ)Ga,X=fα,YλF+(1λ)GEY(f)
b) Let F E Y ( f ) F E Y ( f ) F inE_(Y)(f)F \in E_{Y}(f)FEY(f) and x X x X x in Xx \in XxX. Then for y Y y Y y in Yy \in YyY we have F ( x ) f ( y ) = F ( x ) F ( y ) F α , x d α ( x , y ) = f α , x d α ( x , y ) F ( x ) f ( y ) = F ( x ) F ( y ) F α , x d α ( x , y ) = f α , x d α ( x , y ) F(x)-f(y)=F(x)-F(y) >= -||F||_(alpha,x)*d^(alpha)(x,y)=-||f||_(alpha,x)d^(alpha)(x,y)F(x)-f(y)=F(x)-F(y) \geq-\|F\|_{\alpha, x} \cdot d^{\alpha}(x, y)=-\|f\|_{\alpha, x} d^{\alpha}(x, y)F(x)f(y)=F(x)F(y)Fα,xdα(x,y)=fα,xdα(x,y) so that
F ( x ) f ( y ) f α , y d α ( x , y ) , for all y Y . F ( x ) f ( y ) f α , y d α ( x , y ) , for all  y Y F(x) >= f(y)-||f||_(alpha,y)d^(alpha)(x,y)", for all "y in Y". "F(x) \geq f(y)-\|f\|_{\alpha, y} d^{\alpha}(x, y) \text {, for all } y \in Y \text {. }F(x)f(y)fα,ydα(x,y), for all yY
Therefore
F ( x ) sup { f ( y ) f α , y d α ( x , y ) : y y } = F 2 ( x ) F ( x ) sup f ( y ) f α , y d α ( x , y ) : y y = F 2 ( x ) F(x) >= s u p{f(y)-||f||_(alpha,y)d^(alpha)(x,y):y in y}=F_(2)(x)F(x) \geq \sup \left\{f(y)-\|f\|_{\alpha, y} d^{\alpha}(x, y): y \in y\right\}=F_{2}(x)F(x)sup{f(y)fα,ydα(x,y):yy}=F2(x)
Similarly,
F ( x ) f ( y ) = F ( x ) F ( y ) F α , x d α ( x , y ) = f α , y d α ( x , y ) F ( x ) f ( y ) = F ( x ) F ( y ) F α , x d α ( x , y ) = f α , y d α ( x , y ) F(x)-f(y)=F(x)-F(y) <= ||F||_(alpha,x)d^(alpha)(x,y)=||f||_(alpha,y)d^(alpha)(x,y)F(x)-f(y)=F(x)-F(y) \leq\|F\|_{\alpha, x} d^{\alpha}(x, y)=\|f\|_{\alpha, y} d^{\alpha}(x, y)F(x)f(y)=F(x)F(y)Fα,xdα(x,y)=fα,ydα(x,y)
implies
F ( x ) f ( y ) + f α , y d α ( x , y ) , for all y Y F ( x ) f ( y ) + f α , y d α ( x , y ) ,  for all  y Y F(x) <= f(y)+||f||_(alpha,y)d^(alpha)(x,y)," for all "y in YF(x) \leq f(y)+\|f\|_{\alpha, y} d^{\alpha}(x, y), \text { for all } y \in YF(x)f(y)+fα,ydα(x,y), for all yY
so that
F ( x ) inf | f ( y ) + f α , y d α ( x , y ) : y Y | = F 1 ( x ) . F ( x ) inf f ( y ) + f α , y d α ( x , y ) : y Y = F 1 ( x ) . F(x) <= i n f|f(y)+||f||_(alpha,y)d^(alpha)(x,y):y in Y|=F_(1)(x).F(x) \leq \inf \left|f(y)+\|f\|_{\alpha, y} d^{\alpha}(x, y): y \in Y\right|=F_{1}(x) .F(x)inf|f(y)+fα,ydα(x,y):yY|=F1(x).
c) If F , G E Y ( f ) F , G E Y ( f ) F,G inE_(Y)(f)F, G \in E_{Y}(f)F,GEY(f) and λ ( 0 , 1 ) λ ( 0 , 1 ) lambda in(0,1)\lambda \in(0,1)λ(0,1) are such that λ F + ( 1 λ ) G = F 1 = λ F 1 + ( 1 λ ) F 1 λ F + ( 1 λ ) G = F 1 = λ F 1 + ( 1 λ ) F 1 lambda F+(1-lambda)G=F_(1)=lambdaF_(1)+(1-lambda)F_(1)\lambda F+ (1-\lambda) G=F_{1}=\lambda F_{1}+(1-\lambda) F_{1}λF+(1λ)G=F1=λF1+(1λ)F1, then λ ( F 1 F ) = ( 1 λ ) ( G F 1 ) λ F 1 F = ( 1 λ ) G F 1 lambda(F_(1)-F)=(1-lambda)(G-F_(1))\lambda\left(F_{1}-F\right)=(1-\lambda)\left(G-F_{1}\right)λ(F1F)=(1λ)(GF1) and since by b), G F 1 0 G F 1 0 G-F_(1) <= 0G-F_{1} \leq 0GF10 it follows that F 1 F F 1 F F_(1) <= FF_{1} \leq FF1F. But F 1 F F 1 F F_(1) >= FF_{1} \geq FF1F and hence F = F 1 F = F 1 F=F_(1)F=F_{1}F=F1. Then the relation λ F + ( 1 λ ) G = F 1 λ F + ( 1 λ ) G = F 1 lambda F+(1-lambda)G=F_(1)\lambda F+(1-\lambda) G=F_{1}λF+(1λ)G=F1 yields also G = F 1 G = F 1 G=F_(1)\mathrm{G}=\mathrm{F}_{1}G=F1.
The case of the function F 2 F 2 F_(2)F_{2}F2 can be treated similarly.
2. For a nónvoid subset Y Y YYY of metric space ( X , d ) ( X , d ) (X,d)(X, d)(X,d) denote (2.1) Y = { f Λ α ( X , d ) : 4 Y = 0 } Y = f Λ α ( X , d ) : 4 Y = 0 quadY^(_|_)={f inLambda_(alpha)(X,d):4_(Y)=0}\quad Y^{\perp}=\left\{f \in \Lambda_{\alpha}(X, d): 4_{Y}=0\right\}Y={fΛα(X,d):4Y=0}.
Obviously, Y Y Y^(_|_)Y^{\perp}Y is a closed subspace of A α ( X , d ) A α ( X , d ) A_(alpha)(X,d)A_{\alpha}(X, d)Aα(X,d).
A subset S S SSS of a normed space ( V , ) ( V , ) (V,||||)(V,\| \|)(V,) is called proximinal if for every x V x V x in Vx \in VxV there exists y 0 S y 0 S y_(0)in Sy_{0} \in Sy0S such that (2.2) x y 0 = d ( x , S ) = inf { x y : y S } x y 0 = d ( x , S ) = inf { x y : y S } quad||x-y_(0)||=d(x,S)=i n f{||x-y||:y in S}\quad\left\|x-y_{0}\right\|=d(x, S)=\inf \{\|x-y\|: y \in S\}xy0=d(x,S)=inf{xy:yS}.
An element Y 0 S Y 0 S Y_(0)in SY_{0} \in SY0S for which the infimum in (2.2) is attained is called an element of best approximation of x by elements in s. If for every x V x V x in Vx \in VxV there exists a unique element of best approximation of x x xxx in S S SSS, then the set S S SSS is called Chebyshevian [6]. Denote by P S ( x ) P S ( x ) P_(S)(x)P_{S}(x)PS(x) the set of all best approximation elements of x x xxx in S S SSS.
THEOREM 3. If Y is a nonvoid subset of a metric space ( x , d x , d x,d\mathrm{x}, \mathrm{d}x,d ) then
a) The subspace Y Y Y^(_|_)Y^{\perp}Y is proximinal and (2.3) d ( f , Y . ) = f Y α , Y d f , Y . = f Y α , Y quad d(f,Y_(.)^(_|_))=||f||_(Y)||_(alpha,Y)\quad d\left(f, Y_{.}^{\perp}\right)=\|f\|_{Y} \|_{\alpha, Y}d(f,Y.)=fYα,Y, for every f Λ α ( x , d ) f Λ α ( x , d ) f inLambda_(alpha)(x,d)f \in \Lambda_{\alpha}(\mathrm{x}, \mathrm{d})fΛα(x,d);
b) Every element g 0 Y g 0 Y g_(0)inY^(_|_)g_{0} \in Y^{\perp}g0Y of best approximation for f f f\mathbf{f}f has the form g 0 = f F g 0 = f F g_(0)=f-Fg_{0}=f-Fg0=fF, where F E Y ( f | Y ) F E Y f Y F inE_(Y)(f|_(Y))F \in E_{Y}\left(\left.f\right|_{Y}\right)FEY(f|Y), and, conversely, for every F E Y ( f | Y ) , f F F E Y f Y , f F F inE_(Y)(f|_(Y)),f-FF \in E_{Y}\left(\left.f\right|_{Y}\right), f-FFEY(f|Y),fF is an element of best approximation for f f fff in Y Y Y^(_|_)Y^{\perp}Y, i.e.
P Y ( f ) = f E Y ( f Y ) ; P Y ( f ) = f E Y f Y ; P_(Y _|_)(f)=f-E_(Y)(f_(Y));P_{Y \perp}(f)=f-E_{Y}\left(f_{Y}\right) ;PY(f)=fEY(fY);
c) The subspace Y Y Y^(_|_)Y^{\perp}Y is Chebyshevian if and only if for every f Λ α ( X , d ) f Λ α ( X , d ) f inLambda_(alpha)(X,d)f \in \Lambda_{\alpha}(X, d)fΛα(X,d) the function f Y f Y f_(Y)f_{Y}fY has a unique extension in Λ α ( x , d ) Λ α ( x , d ) Lambda_(alpha)(x,d)\Lambda_{\alpha}(x, d)Λα(x,d).
Proof. a) Let f Λ α ( X , d ) and F E Y ( f | Y ) f Λ α ( X , d ) and  F E Y f Y f inLambda_(alpha)(X,d)^("and ")F inE_(Y)(f|_(Y))f \in \Lambda_{\alpha}(X, d)^{\text {and }} F \in E_{Y}\left(\left.f\right|_{Y}\right)fΛα(X,d)and FEY(f|Y). Then (2.4)
f ( f F ) α , X = F α , X = f | Y α , Y . f ( f F ) α , X = F α , X = f Y α , Y . ||f-(f-F)||_(alpha,X)^(')=||F||_(alpha,X)=||f|_(Y)||_(alpha,Y).\|f-(f-F)\|_{\alpha, X}^{\prime}=\|F\|_{\alpha, X}=\left\|\left.f\right|_{Y}\right\|_{\alpha, Y} .f(fF)α,X=Fα,X=f|Yα,Y.
Since f F Y f F Y f-F inY^(_|_)f-F \in Y^{\perp}fFY, it follows
inf { f g α , x : g Y } f α , y . inf f g α , x : g Y f α , y i n f{||f-g||_(alpha,x):g inY^(_|_)} <= ||f||_(alpha,y)". "\inf \left\{\|f-g\|_{\alpha, x}: g \in Y^{\perp}\right\} \leq\|f\|_{\alpha, y} \text {. }inf{fgα,x:gY}fα,y
on the other hand
| f | x α , y = sup { | ( f g ) ( x ) ( f g ) ( y ) | / d α ( x , y ) : x , y Y ; x y } sup { | ( f g ) ( x ) ( f g ) ( y ) | / d α ( x , y ) : x , y X ; x y } = f g α , x | f | x α , y = sup | ( f g ) ( x ) ( f g ) ( y ) | / d α ( x , y ) : x , y Y ; x y } sup | ( f g ) ( x ) ( f g ) ( y ) | / d α ( x , y ) : x , y X ; x y = f g α , x {:[|f|x||_(alpha,y)= s u p{|(f-g)(x)-(f-g)(y)|//d^(alpha)(x,y):x,y in Y;:}],[x!=y} <= s u p{|(f-g)(x)-(f-g)(y)|//d^(alpha)(x,y):x,y in X;:}],[{:x!=y^(')}=||f-g||_(alpha,x)]:}\begin{aligned} |f| x \|_{\alpha, y}= & \sup \left\{|(f-g)(x)-(f-g)(y)| / d^{\alpha}(x, y): x, y \in Y ;\right. \\ x \neq y\} \leq & \sup \left\{|(f-g)(x)-(f-g)(y)| / d^{\alpha}(x, y): x, y \in X ;\right. \\ & \left.x \neq y^{\prime}\right\}=\|f-g\|_{\alpha, x} \end{aligned}|f|xα,y=sup{|(fg)(x)(fg)(y)|/dα(x,y):x,yY;xy}sup{|(fg)(x)(fg)(y)|/dα(x,y):x,yX;xy}=fgα,x
for every g Y g Y g inY^(_|_)g \in Y^{\perp}gY, so that
| f | y α , x inf { f g ˙ α , x : g X } . | f | y α , x inf f g ˙ α , x : g X . |f|_(y)||_(alpha,x) <= i n f{||f-(g^(˙))||_(alpha,x):g inX^(_|_)}.|f|_{y} \|_{\alpha, x} \leq \inf \left\{\|f-\dot{g}\|_{\alpha, x}: g \in X^{\perp}\right\} .|f|yα,xinf{fg˙α,x:gX}.
Therefore f X α , Y = f ( f F ) α , X = d ( f , Y ) f X α , Y = f ( f F ) α , X = d f , Y quad||f||_(X)||_(alpha,Y)=||f-(f-F)||_(alpha,X)=d(f,Y^(_|_))\quad\|f\|_{X}\left\|_{\alpha, Y}=\right\| f-(f-F) \|_{\alpha, X}=d\left(f, Y^{\perp}\right)fXα,Y=f(fF)α,X=d(f,Y), which shows that Y Y Y^(_|_)Y^{\perp}Y is proximinal and that the formula (2.3) holds.
b) By (2.4), it follows that f = F Y f = F Y f=F inY^(_|_)f=F \in Y^{\perp}f=FY is an element of best approximation for f f fff in Y Y Y^(_|_)Y^{\perp}Y, where F E Y ( f | Y ) F E Y f Y F inE_(Y)(f|_(Y))F \in E_{Y}\left(\left.f\right|_{Y}\right)FEY(f|Y). If g 0 Y i g 0 Y i g_(0)inY^(i)g_{0} \in Y^{i}g0Yi is an element of best approximation for f f fff by elements of Y Y Y^(_|_)Y^{\perp}Y, then
f g 0 α , X = f | Y α , Y and ( f g 0 ) | Y = f | Y f g 0 α , X = f Y α , Y  and  f g 0 Y = f Y ||f-g_(0)||_(alpha,X)=||f|_(Y)||_(alpha,Y)quad" and " quad(f-g_(0))|_(Y)=f|_(Y)\left\|f-g_{0}\right\|_{\alpha, X}=\left\|\left.f\right|_{Y}\right\|_{\alpha, Y} \quad \text { and }\left.\quad\left(f-g_{0}\right)\right|_{Y}=\left.f\right|_{Y}fg0α,X=f|Yα,Y and (fg0)|Y=f|Y
so that f g 0 E Y ( f Y ) f g 0 E Y f Y f-g_(0)inE_(Y)(f_(Y))f-g_{0} \in E_{Y}\left(f_{Y}\right)fg0EY(fY) and F 0 = f g 0 F 0 = f g 0 F_(0)=f-g_(0)F_{0}=f-g_{0}F0=fg0 is an extension of f Y f Y f∣Yf \mid YfY.
c) If χ χ chi^(_|_)\chi^{\perp}χ is Chebyshevian, then every f Λ α ( X , d ) f Λ α ( X , d ) f inLambda_(alpha)(X,d)f \in \Lambda_{\alpha}(X, d)fΛα(X,d) has a unique element of best approximation in Y Y Y^(_|_)Y^{\perp}Y and by b b bbb ), the set E Y ( f Y ) E Y f Y E_(Y)(f_(Y))E_{Y}\left(f_{Y}\right)EY(fY) contains only one element.
3. Let now ( x , d x , d x,d\mathrm{x}, \mathrm{d}x,d ) be a metric space of finite diameter, i.e. d ( X ) = sup { d ( x , y ) : x p y X } < d ( X ) = sup d ( x , y ) : x p y X < d(X)=s u p{d(x,y):x_(p)y in X} < ood(X)=\sup \left\{d(x, y): x_{p} y \in X\right\}<\inftyd(X)=sup{d(x,y):xpyX}<. Then every function f Λ α ( X , d ) f Λ α ( X , d ) f inLambda_(alpha)(X,d)f \in \Lambda_{\alpha}(X, d)fΛα(X,d) is bounded, for if x 0 X x 0 X x_(0)in Xx_{0} \in Xx0X is fixed, then | f ( x ) | | f ( x ) f ( x 0 ) | + | f ( x 0 ) | | f ( x 0 ) | + f α , x d α ( x , x 0 ) ≤≤ | f ( x 0 ) | + f α , x [ d ( X ) ] α , | f ( x ) | f ( x ) f x 0 + f x 0 f x 0 + f α , x d α x , x 0 ≤≤ f x 0 + f α , x [ d ( X ) ] α , |f(x)| <= |f(x)-f(x_(0))|+|f(x_(0))| <= |f(x_(0))|+||f||_(alpha,x)d^(alpha)(x,x_(0))≤≤|f(x_(0))|+||f||_(alpha,x)[d(X)]^(alpha),quad|f(x)| \leq\left|f(x)-f\left(x_{0}\right)\right|+\left|f\left(x_{0}\right)\right| \leq\left|f\left(x_{0}\right)\right|+\|f\|_{\alpha, x} d^{\alpha}\left(x, x_{0}\right) \leq \leq\left|f\left(x_{0}\right)\right|+\|f\|_{\alpha, x}[d(X)]^{\alpha}, \quad|f(x)||f(x)f(x0)|+|f(x0)||f(x0)|+fα,xdα(x,x0)≤≤|f(x0)|+fα,x[d(X)]α, for every x X x X x in Xx \in XxX.
In this case, we can define the uniform norm on Λ α ( x , d ) Λ α ( x , d ) Lambda_(alpha)(x,d)\Lambda_{\alpha}(x, d)Λα(x,d), (3.1)
f u , x = sup { | f ( x ) | : x X } . f u , x = sup { | f ( x ) | : x X } . ||f||_(u,x)=s u p{|f(x)|:x in X}.\|f\|_{u, x}=\sup \{|f(x)|: x \in X\} .fu,x=sup{|f(x)|:xX}.
Let X X XXX be a subset of metric space of finite diameter ( X , d X , d X,dX, dX,d ) and let X X X^(_|_)X^{\perp}X be defined by (2.1). For f Λ α ( X , d ) f Λ α ( X , d ) f inLambda_(alpha)(X,d)f \in \Lambda_{\alpha}(X, d)fΛα(X,d) let G ( f ) G ( f ) G(f)G(f)G(f) denote
the set of all best approximation elements, with respect to the Hölder norm, of f f fff by elements in Y Y Y^(-)Y^{-}Y. Consider the following problem:
Find g g ggg, and g g g^(**)g^{*}g in G ( f ) G ( f ) G(f)G(f)G(f) such that
f g u , x = inf { f g u , x : g G ( f ) } , (3.2) f g u , x = sup { f g u , x : g G ( f ) } . f g u , x = inf f g u , x : g G ( f ) , (3.2) f g u , x = sup f g u , x : g G ( f ) . {:[||f-g||_(u,x)=i n f{||f-g||_(u,x):g in G(f)}","],[(3.2)||f-g^(**)||_(u,x)=s u p{||f-g||_(u,x):g in G(f)}.]:}\begin{align*} & \|f-g\|_{u, x}=\inf \left\{\|f-g\|_{u, x}: g \in G(f)\right\}, \\ & \left\|f-g^{*}\right\|_{u, x}=\sup \left\{\|f-g\|_{u, x}: g \in G(f)\right\} . \tag{3.2} \end{align*}fgu,x=inf{fgu,x:gG(f)},(3.2)fgu,x=sup{fgu,x:gG(f)}.
By Theorem 3 b), the problem (3.2) is equivalent to the following problem:
Find two extensions F s F s F_(s)F_{s}Fs and F F F^(')F^{\prime}F in E Y ( f Y ) E Y f Y E_(Y)(f_(Y))E_{Y}\left(f_{Y}\right)EY(fY) such that
{ F a u , x = inf { F u , x : F E Y ( f Y ) } , (3.3) F u , x = sup { F u , x : F E Y ( f X ) } . F a u , x = inf F u , x : F E Y f Y , (3.3) F u , x = sup F u , x : F E Y f X . {:[{F_(a)||_(u,x)=i n f{||F||_(u,x):F inE_(Y)(f||_(Y))},:}],[(3.3)||F^(**)||_(u,x)=s u p{||F||_(u,x):F inE_(Y)(f||_(X))}.]:}\begin{align*} & \left\{F_{\mathrm{a}} \|_{\mathrm{u}, \mathrm{x}}=\inf \left\{\|F\|_{\mathrm{u}, \mathrm{x}}: F \in E_{\mathrm{Y}}\left(f \|_{\mathrm{Y}}\right)\right\},\right. \\ & \left\|F^{*}\right\|_{\mathrm{u}, \mathrm{x}}=\sup \left\{\|F\|_{\mathrm{u}, \mathrm{x}}: F \in E_{\mathrm{Y}}\left(f \|_{\mathrm{X}}\right)\right\} . \tag{3.3} \end{align*}{Fau,x=inf{Fu,x:FEY(fY)},(3.3)Fu,x=sup{Fu,x:FEY(fX)}.
The next theorem shows that the problem considered above has always a solution.
THEOREM 4. a) The infimum in (3.2) is attained, namely for every function g G ( f ) g G ( f ) g_(**)in G(f)g_{*} \in G(f)gG(f) of the form g = f r ¯ g = f r ¯ g_(**)=f- bar(r)g_{*}=f-\bar{r}g=fr¯, where F q E Y ( f | Y ) F q E Y f Y F_(q)inE_(Y)(f|_(Y))F_{q} \in E_{Y}\left(\left.f\right|_{Y}\right)FqEY(f|Y) is such that F U , X = | f | Y U , X F U , X = | f | Y U , X ||F_(**)||_(U,X)=|f|_(Y)||_(U,X)\left\|F_{*}\right\|_{U, X}=|f|_{Y} \|_{U, X}FU,X=|f|YU,X;
b) The supremum in (3.2) is attained, namely for g = f F 1 g = f F 1 g^(**)=f-F_(1)g^{*}=f-F_{1}g=fF1 or g = f F 2 g = f F 2 g^(**)=f-F_(2)g^{*}=f-F_{2}g=fF2 or for both of these functions, where F 1 F 1 F_(1)F_{1}F1 and F 2 F 2 F_(2)F_{2}F2 are de ned by (1.3).
proof. a) First, observe that there exists F v E Y ( f x ) F v E Y ( f x ) F_(v)inE_(Y)(f∣x)F_{v} \in E_{Y}(f \mid x)FvEY(fx) such that F U , X = f Y U , X F U , X = f Y U , X ||F_(**)||_(U,X)=||f||_(Y)||_(U,X)\left\|F_{*}\right\|_{U, X}=\|f\|_{Y} \|_{U, X}FU,X=fYU,X. Indeed, if for F E Y ( f Y ) F E Y ( f Y ) F inE_(Y)(f∣Y)F \in E_{Y}(f \mid Y)FEY(fY) put F a ( x ) = | f | Y u , Y F a ( x ) = | f | Y u , Y F_(a)(x)=|f|_(Y)||_(u,Y)quadF_{a}(x)=|f|_{Y} \|_{u, Y} \quadFa(x)=|f|Yu,Y if F ( x ) > | f | Y u , Y F ( x ) > | f | Y u , Y F(x) > |f|_(Y)||_(u,Y)F(x)>|f|_{Y} \|_{u, Y}F(x)>|f|Yu,Y, (3.4) = F ( x ) = F ( x ) =F(x)quad=F(x) \quad=F(x) if f | Y , Y F ( x ) f Y U , Y , = | f | Y U , Y f Y , Y F ( x ) f Y U , Y = | f | Y U , Y -||f|_(Y)||_(,Y) <= F(x) <= ||f||_(Y)||_(U,Y", ")=-|f|_(Y)||_(U,Y)quad-\left\|\left.f\right|_{Y}\right\|_{, Y} \leq F(x) \leq\|f\|_{Y} \|_{U, Y \text {, }} =-|f|_{Y} \|_{U, Y} \quadf|Y,YF(x)fYU,Y=|f|YU,Y if F ( x ) < | f | Y U , Y F ( x ) < | f | Y U , Y F(x) < -|f|_(Y)||_(U,Y)F(x)<-|f|_{Y} \|_{U, Y}F(x)<|f|YU,Y,
then F q F q F_(q)F_{q}Fq is in E Y ( f | Y ) E Y f Y E_(Y)(f|_(Y))E_{Y}\left(\left.f\right|_{Y}\right)EY(f|Y) and F q U , X = f Y U , Y F q U , X = f Y U , Y ||F_(q)||_(U,X)=||f||_(Y)||_(U,Y)\left\|F_{q}\right\|_{U, X}=\|f\|_{Y} \|_{U, Y}FqU,X=fYU,Y.
Since
| F u , x sup { | F ( y ) | : y y } = | f | y u , y , F u , x sup { | F ( y ) | : y y } = | f | y u , y , |F||_(u,x) >= s u p{|F(y)|:y in y}=|f|y||_(u,y),:}\left|F\left\|_{u, x} \geq \sup \{|F(y)|: y \in y\}=|f| y\right\|_{u, y},\right.|Fu,xsup{|F(y)|:yy}=|f|yu,y,
for every F E Y ( f | Y ) F E Y f Y F inE_(Y)(f|_(Y))F \in E_{Y}\left(\left.f\right|_{Y}\right)FEY(f|Y), it follows that
inf { F u , x : F E X ( f | Y ) } = F , u , x inf F u , x : F E X f Y = F , u , x i n f{||F||_(u,x):F inE_(X)(f|_(Y))}=||F,||_(u,x)\inf \left\{\|F\|_{u, x}: F \in E_{X}\left(\left.f\right|_{Y}\right)\right\}=\|F,\|_{u, x}inf{Fu,x:FEX(f|Y)}=F,u,x
where F z F z F_(z)F_{z}Fz is defined by (3.4).
b) Since
F 2 ( x ) F ( x ) F 1 ( x ) , x X , F 2 ( x ) F ( x ) F 1 ( x ) , x X , F_(2)(x) <= F(x) <= F_(1)(x),x in X,F_{2}(x) \leq F(x) \leq F_{1}(x), x \in X,F2(x)F(x)F1(x),xX,
for every F E Y ( f | Y ) F E Y f Y F inE_(Y)(f|_(Y))F \in E_{Y}\left(\left.f\right|_{Y}\right)FEY(f|Y) (Theorem 2. b)), it follows that
F u , X max { F 1 u , x , F 2 u , x } . F u , X max F 1 u , x , F 2 u , x . ||F||_(u,X) <= max{||F_(1)||_(u,x),||F_(2)||_(u,x)}.\|\mathrm{F}\|_{\mathrm{u}, \mathrm{X}} \leq \max \left\{\left\|\mathrm{F}_{1}\right\|_{\mathrm{u}, \mathrm{x}},\left\|\mathrm{~F}_{2}\right\|_{\mathrm{u}, \mathrm{x}}\right\} .Fu,Xmax{F1u,x, F2u,x}.
Remarks. a) The set of the functions of the form (3.4) is a convex subset of E Y ( f Y ) E Y ( f Y ) E_(Y)(f∣Y)E_{Y}(f \mid Y)EY(fY);
b) The functions F 1 , F 2 F 1 , F 2 F_(1),F_(2)F_{1}, F_{2}F1,F2 for which the supremum in (3.3) is attained are extremal elements of the convex set E Y ( f | Y ) E Y f Y E_(Y)(f|_(Y))E_{Y}\left(\left.f\right|_{Y}\right)EY(f|Y).
4. Let ( X , d ) ( X , d ) (X,d)(X, d)(X,d) be a metric space of finite diameter. Consider on A α ( X , d ) A α ( X , d ) A_(alpha)(X,d)A_{\alpha}(X, d)Aα(X,d) the norms
f s = f α , X + f u , X (4.1) f m = max | f α , X ; f u , X | f s = f α , X + f u , X (4.1) f m = max f α , X ; f u , X {:[||f||_(s)=||f||_(alpha,X)+||f||_(u,X)],[(4.1)||f||_(m)=max|||f||_(alpha,X);||f||_(u,X)|]:}\begin{align*} & \|f\|_{\mathrm{s}}=\|f\|_{\alpha, \mathrm{X}}+\|f\|_{\mathrm{u}, \mathrm{X}} \\ & \|f\|_{\mathrm{m}}=\max \left|\|f\|_{\alpha, \mathrm{X}} ;\|f\|_{\mathrm{u}, \mathrm{X}}\right| \tag{4.1} \end{align*}fs=fα,X+fu,X(4.1)fm=max|fα,X;fu,X|
called the "sum-norm" and the "max-norm", respectively.
If Y Y YYY is a subset of X X XXX and f f fff is in Λ α ( Y , d ) Λ α ( Y , d ) Lambda_(alpha)(Y,d)\Lambda_{\alpha}(Y, d)Λα(Y,d), then it is natural to ask i f f i f f iffi f fiff has an extension F A α ( X , d ) F A α ( X , d ) F inA_(alpha)(X,d)F \in A_{\alpha}(X, d)FAα(X,d) preserving the norms (4.1). An afirmative answer to this question and some consequences will be objects of the following theorems.
THEOREM 5. If ( X , d ) ( X , d ) (X,d)(\mathrm{X}, \mathrm{d})(X,d) is a metric space of finite diameter and Y Y YYY is a nonvoid subset of X X XXX, then for every f Λ α ( Y , d ) f Λ α ( Y , d ) f inLambda_(alpha)(Y,d)f \in \Lambda_{\alpha}(Y, d)fΛα(Y,d) there exists F A α ( X , d ) F A α ( X , d ) FinA_(alpha)(X,d)\mathrm{F} \in \mathrm{A}_{\alpha}(\mathrm{X}, \mathrm{d})FAα(X,d) such that
F F y = f and | F | s = | f | s . F F y = f  and  | F | s = | f | s FF_(y)=f quad" and "quad|F|_(s)=|f|_(s)". "F F_{y}=f \quad \text { and } \quad|F|_{s}=|f|_{s} \text {. }FFy=f and |F|s=|f|s
proof. Let F 1 F 1 F_(1)F_{1}F1 be defined by (1.3) and let
F ¯ 1 ( x ) = F 1 ( x ) if F 1 ( x ) f u , Y (4.3) = f u , Y if F 1 ( x ) > F u , Y F ¯ 1 ( x ) = F 1 ( x )  if  F 1 ( x ) f u , Y (4.3) = f u , Y  if  F 1 ( x ) > F u , Y {:[ bar(F)_(1)(x)=F_(1)(x)quad" if "F_(1)(x) <= ||f||_(u,Y)],[(4.3)=||f||_(u,Y)quad" if "quadF_(1)(x) > ||F||_(u,Y)]:}\begin{align*} \bar{F}_{1}(x) & =F_{1}(x) \quad \text { if } F_{1}(x) \leq\|f\|_{u, Y} \\ & =\|f\|_{u, Y} \quad \text { if } \quad F_{1}(x)>\|F\|_{u, Y} \tag{4.3} \end{align*}F¯1(x)=F1(x) if F1(x)fu,Y(4.3)=fu,Y if F1(x)>Fu,Y
Then F ¯ 1 ] Y = f F ¯ 1 Y = f bar(F)_(1)]_(Y)=f\left.\bar{F}_{1}\right]_{Y}=fF¯1]Y=f and F ¯ 1 F ¯ 1 bar(F)_(1)\bar{F}_{1}F¯1 preserves both of the Holder and uniform norms, so that it preserves the "sum-norm", also.
similarly, the function
F ¯ 2 ( x ) = F 2 ( x ) if F 2 ( x ) f u , y , (4.4) = f u , y if F 2 ( x ) < f u , x , F ¯ 2 ( x ) = F 2 ( x )  if  F 2 ( x ) f u , y , (4.4) = f u , y  if  F 2 ( x ) < f u , x , {:[ bar(F)_(2)(x)=F_(2)(x)" if "F_(2)(x) >= -||f||_(u,y)","],[(4.4)=||f||_(u,y)" if "F_(2)(x) < -||f||_(u,x)","]:}\begin{align*} \bar{F}_{2}(x) & =F_{2}(x) \text { if } F_{2}(x) \geq-\|f\|_{u, y}, \\ & =\|f\|_{u, y} \text { if } F_{2}(x)<-\|f\|_{u, x}, \tag{4.4} \end{align*}F¯2(x)=F2(x) if F2(x)fu,y,(4.4)=fu,y if F2(x)<fu,x,
where F 2 F 2 F_(2)F_{2}F2 is defined by (1.3), is also an extension of f f fff, preserving both of the Hölder and uniform norms.
Denote by E g ( f ) E g ( f ) E_(g)(f)E_{g}(f)Eg(f) the set of all extensions of f Λ α ( Y , d ) f Λ α ( Y , d ) f inLambda_(alpha)(Y,d)f \in \Lambda_{\alpha}(Y, d)fΛα(Y,d) which preserve the "sum-norm", i.e.
(4.5)
E g ( f ) = { F A α ( X , d ) : F X Y = f , F g = f g } . E g ( f ) = F A α ( X , d ) : F X Y = f , F g = f g . E_(g)(f)={F inA_(alpha)(X,d):FX_(Y)=f,quad||F||_(g)=||f||_(g)}.E_{g}(f)=\left\{F \in A_{\alpha}(X, d): F X_{Y}=f, \quad\|F\|_{g}=\|f\|_{g}\right\} .Eg(f)={FAα(X,d):FXY=f,Fg=fg}.
THEOREA 6. If ( x , d x , d x,d\mathrm{x}, \mathrm{d}x,d ) is a metric space of finite diameter, Y a subset of X and f Λ a ( Y , d ) f Λ a ( Y , d ) finLambda_(a)(Y,d)\mathrm{f} \in \Lambda_{\mathrm{a}}(\mathrm{Y}, \mathrm{d})fΛa(Y,d), then
(4.6) a) F a , X = f a , Y F a , X = f a , Y quad||F||_(a,X)=||f||_(a,Y)\quad\|F\|_{a, X}=\|f\|_{a, Y}Fa,X=fa,Y and F u , X = f u , X F u , X = f u , X ||F||_(u,X)=||f||_(u,X)\|F\|_{u, X}=\|f\|_{u, X}Fu,X=fu,X
for every F E g ( f ) F E g ( f ) F inE_(g)(f)F \in \mathrm{E}_{g}(f)FEg(f);
b) r x F ¯ 1 r x F ¯ 1 rx quad bar(F)_(1)r x \quad \bar{F}_{1}rxF¯1 and F ¯ 2 F ¯ 2 bar(F)_(2)\bar{F}_{2}F¯2 are defined by (4.3) and (4.4) then (4.7) F ¯ 1 ( x ) E ( x ) F ¯ 2 ( x ) , x X F ¯ 1 ( x ) E ( x ) F ¯ 2 ( x ) , x X quad bar(F)_(1)(x) >= E(x) >= bar(F)_(2)(x)quad,x in X\quad \bar{F}_{1}(x) \geq E(x) \geq \bar{F}_{2}(x) \quad, x \in XF¯1(x)E(x)F¯2(x),xX,
for every F E s ( f ) F E s ( f ) F inE_(s)(f)F \in E_{s}(f)FEs(f) :
c) The set E s E s E_(s)\mathrm{E}_{s}Es (t) is a convex subset of the ball (with respect to "sum-norm") of center 0 and radius f s f s ||f||_(s)\|f\|_{s}fs in A α ( X , d ) A α ( X , d ) A_(alpha)(X,d)\mathrm{A}_{\alpha}(\mathrm{X}, \mathrm{d})Aα(X,d);
d) The functions F ¯ 1 F ¯ 1 bar(F)_(1)\bar{F}_{1}F¯1 and F ¯ 2 F ¯ 2 bar(F)_(2)\bar{F}_{2}F¯2 are extremal elements of E g ( f ) E g ( f ) E_(g)(f)E_{g}(f)Eg(f);
e) If f g = 1 f g = 1 ||f||_(g)=1\|f\|_{g}=1fg=1, then f f fff is an extremal element of the unit ball of A α ( Y , d ) A α ( Y , d ) A_(alpha)(Y,d)\mathrm{A}_{\alpha}(\mathrm{Y}, \mathrm{d})Aα(Y,d) (with respect to the "sum-norm") if and only if
E s ( x ) E s ( x ) E_(s)(x)\mathrm{E}_{\mathrm{s}}(\mathrm{x})Es(x) is a face of the unit ball of Λ α ( x , d ) Λ α ( x , d ) Lambda_(alpha)(x,d)\Lambda_{\alpha}(\mathrm{x}, \mathrm{d})Λα(x,d).
Pxoof. a) Let F E g ( f ) F E g ( f ) F inE_(g)(f)F \in E_{g}(f)FEg(f). Since F Y = f F Y = f F_(Y)=fF_{Y}=fFY=f, it follows that = F α , X f α , Y = F α , X f α , Y =||F||_(alpha,X) >= ||f||_(alpha,Y)=\|F\|_{\alpha, X} \geq\|f\|_{\alpha, Y}=Fα,Xfα,Y and F U , X f U , Y F U , X f U , Y ||F||_(U,X) >= ||f||_(U,Y)\|F\|_{U, X} \geq\|f\|_{U, Y}FU,XfU,Y. If F α , X > f α , Y F α , X > f α , Y ||F||_(alpha,X) > ||f||_(alpha,Y)\|F\|_{\alpha, X}>\|f\|_{\alpha, Y}Fα,X>fα,Y then f α , Y + f u , Y < F α , X + F u , X = F s = f s = f α , Y + f u , Y f α , Y + f u , Y < F α , X + F u , X = F s = f s = f α , Y + f u , Y ||f||_(alpha,Y)+||f||_(u,Y) < ||F||_(alpha,X)+||F||_(u,X)=||F||_(s)=||f||_(s)=||f||_(alpha,Y)+||f||_(u,Y)\|f\|_{\alpha, Y}+\|f\|_{u, Y}<\|F\|_{\alpha, X}+\|F\|_{u, X}=\|F\|_{s}=\|f\|_{s}=\|f\|_{\alpha, Y}+\|f\|_{u, Y}fα,Y+fu,Y<Fα,X+Fu,X=Fs=fs=fα,Y+fu,Y, which is impossible. Therefore F α , v = f α , Y F α , v = f α , Y ||F||_(alpha,v)=||f||_(alpha,Y)\|F\|_{\alpha, v}=\|f\|_{\alpha, Y}Fα,v=fα,Y and F u , X == F s F α , X = f s f α , Y = f u , ν F u , X == F s F α , X = f s f α , Y = f u , ν ||F||_(u,X)==||F||_(s)-||F||_(alpha,X)=||f||_(s)-||f||_(alpha,Y)=||f||_(u,nu)\|F\|_{u, X}= =\|F\|_{s}-\|F\|_{\alpha, X}=\|f\|_{s}-\|f\|_{\alpha, Y}=\|f\|_{u, \nu}Fu,X==FsFα,X=fsfα,Y=fu,ν.
The proof of b), c), d) proceedes similarly to the proofs of assertions a), b), c) of Theorem 2.
e) Let f f fff be an extremal element of the unit ball of Λ α ( Y , d ) Λ α ( Y , d ) Lambda_(alpha)(Y,d)\Lambda_{\alpha}(Y, d)Λα(Y,d) and let λ ϵ ( 0 , 1 ) λ ϵ ( 0 , 1 ) lambda epsilon(0,1)\lambda \epsilon(0,1)λϵ(0,1). If F 1 , F 2 F 1 , F 2 F_(1),F_(2)F_{1}, F_{2}F1,F2 are elements of the unit ball of A α ( x , d ) A α ( x , d ) A_(alpha)(x,d)A_{\alpha}(x, d)Aα(x,d) such that λ F 1 + ( 1 λ ) F 2 E s ( f ) λ F 1 + ( 1 λ ) F 2 E s ( f ) lambdaF_(1)+(1-lambda)F_(2)inE_(s)(f)\lambda F_{1}+(1-\lambda) F_{2} \in E_{s}(f)λF1+(1λ)F2Es(f), then λ F 1 | Y + ( 1 λ ) F 2 | Y = f λ F 1 Y + ( 1 λ ) F 2 Y = f lambdaF_(1)|_(Y)+(1-lambda)F_(2)|_(Y)=f\left.\lambda F_{1}\right|_{Y}+\left.(1-\lambda) F_{2}\right|_{Y}=fλF1|Y+(1λ)F2|Y=f. Since f f fff is extremal it follows that F 1 | Y = F 2 | Y = £ F 1 Y = F 2 Y = £ F_(1)|_(Y)=F_(2)|_(Y)=£\left.F_{1}\right|_{Y}=\left.F_{2}\right|_{Y}=£F1|Y=F2|Y=£ and F 1 s = F 2 s = 1 F 1 s = F 2 s = 1 ||F_(1)||_(s)=||F_(2)||_(s)=1\left\|F_{1}\right\|_{s}=\left\|F_{2}\right\|_{s}=1F1s=F2s=1, i.e. F 1 , F 2 E s ( £ ) F 1 , F 2 E s ( £ ) F_(1),F_(2)inE_(s)(£)F_{1}, F_{2} \in E_{s}(£)F1,F2Es(£). We have shown that E s ( f ) E s ( f ) E_(s)(f)E_{s}(f)Es(f) is a face of the unit ball of Λ α ( x , d ) Λ α ( x , d ) Lambda_(alpha)(x,d)\Lambda_{\alpha}(x, d)Λα(x,d).
Conversely, suppose that f s = 1 f s = 1 ||f||_(s)=1\|f\|_{s}=1fs=1 and f f fff is not an extremal element of the unit ball of Λ α ( Y , d ) Λ α ( Y , d ) Lambda_(alpha)(Y,d)\Lambda_{\alpha}(\mathrm{Y}, \mathrm{d})Λα(Y,d). Then there exists f 1 , f 2 f 1 , f 2 f_(1),f_(2)f_{1}, f_{2}f1,f2 in Λ α ( Y , d ) , f 1 f 2 , f i s = 1 , i = 1 , 2 Λ α ( Y , d ) , f 1 f 2 , f i s = 1 , i = 1 , 2 Lambda_(alpha)(Y,d),f_(1)!=f_(2),||f_(i)||_(s)=1,i=1,2\Lambda_{\alpha}(Y, d), f_{1} \neq f_{2},\left\|f_{i}\right\|_{s}=1, i=1,2Λα(Y,d),f1f2,fis=1,i=1,2, and λ ( 0 , 1 ) λ ( 0 , 1 ) lambda in(0,1)\lambda \in(0,1)λ(0,1) such that λ f 1 + ( 1 λ ) f 2 = f λ f 1 + ( 1 λ ) f 2 = f lambdaf_(1)+(1-lambda)f_(2)=f\lambda f_{1}+(1-\lambda) f_{2}=fλf1+(1λ)f2=f. If F i E s ( f i ) F i E s f i F_(i)inE_(s)(f_(i))F_{i} \in E_{s}\left(f_{i}\right)FiEs(fi), i = 1 , 2 i = 1 , 2 i=1,2i=1,2i=1,2, then λ F 1 Y + ( 1 λ ) F 2 Y = f λ F 1 Y + ( 1 λ ) F 2 Y = f lambdaF_(1)^(')||_(Y)+(1-lambda)F_(2)^(')||_(Y)=f\lambda F_{1}^{\prime}\left\|_{Y}+(1-\lambda) F_{2}^{\prime}\right\|_{Y}=fλF1Y+(1λ)F2Y=f and 1 = λ F 1 Y + + ( 1 λ ) F 2 Y s λ F 1 + ( 1 ˙ λ ) F 2 s 1 1 = λ F 1 Y + + ( 1 λ ) F 2 Y s λ F 1 + ( 1 ˙ λ ) F 2 s 1 1=||lambdaF_(1)^(')||_(Y)++(1-lambda)F_(2)^(')||_(Y)||_(s) <= ||lambdaF_(1)+((1^(˙))-lambda)F_(2)^(')||_(s) <= 11=\left\|\lambda F_{1}^{\prime}\right\|_{Y}+ +(1-\lambda) F_{2}^{\prime}\left\|_{Y}\right\|_{s} \leq\left\|\lambda F_{1}+(\dot{1}-\lambda) F_{2}^{\prime}\right\|_{s} \leq 11=λF1Y++(1λ)F2YsλF1+(1˙λ)F2s1, which show that λ F 1 + ( 1 λ ) F 2 E 8 ( f ) λ F 1 + ( 1 λ ) F 2 E 8 ( f ) lambdaF_(1)^(')+(1-lambda)F_(2)^(')inE_(8)(f)\lambda F_{1}^{\prime}+(1-\lambda) F_{2}^{\prime} \in E_{8}(f)λF1+(1λ)F2E8(f). Consequently E g ( f ) E g ( f ) E_(g)(f)E_{g}(f)Eg(f) is not a face of the unit ball of Λ α ( x , d ) Λ α ( x , d ) Lambda_(alpha)(x,d)\Lambda_{\alpha}(x, d)Λα(x,d).
THEOREM 7. Let ( X , d X , d X,d\mathrm{X}, \mathrm{d}X,d ) be a metric space of finite diameter, Y a subset of X and suppose Λ α ( Y , d ) Λ α ( Y , d ) Lambda_(alpha)(Y,d)\Lambda_{\alpha}(\mathrm{Y}, \mathrm{d})Λα(Y,d) and Λ α ( X , d ) Λ α ( X , d ) Lambda_(alpha)(X,d)\Lambda_{\alpha}(\mathrm{X}, \mathrm{d})Λα(X,d) endoved with the "max-norm". Then -
a) For every F E s ( f ) , F m = f m F E s ( f ) , F m = f m F inE_(s)(f),||F||_(m)=||f||_(m)F \in \mathrm{E}_{\mathrm{s}}(\mathrm{f}),\|\mathrm{F}\|_{\mathrm{m}}=\|\mathrm{f}\|_{\mathrm{m}}FEs(f),Fm=fm;
E m ( f , = { F Λ α ( X , d ) : F Y = f , F m = f m } , E m f , = F Λ α ( X , d ) : F Y = f , F m = f m , E_(m)(f,={F inLambda_(alpha)(X,d):F||_(Y)=f,||F||_(m)=||f||_(m)},:}E_{m}\left(f,=\left\{F \in \Lambda_{\alpha}(X, d): F\left\|_{Y}=f,\right\| F\left\|_{m}=\right\| f \|_{m}\right\},\right.Em(f,={FΛα(X,d):FY=f,Fm=fm},
then there exists F ~ 1 F ~ 1 widetilde(F)_(1)\widetilde{F}_{1}F~1 and F ~ 2 F ~ 2 widetilde(F)_(2)\widetilde{F}_{2}F~2 in E m ( f ) E m ( f ) E_(m)(f)\mathrm{E}_{m}(f)Em(f) such that
( 4.9 ) F ~ 1 ( x ) F ( x ) F ~ 2 ( x ) , x X , ( 4.9 ) F ~ 1 ( x ) F ( x ) F ~ 2 ( x ) , x X , (4.9)quad tilde(F)_(1)(x) >= F(x) >= tilde(F)_(2)(x),x in X,(4.9) \quad \tilde{F}_{1}(x) \geq F(x) \geq \tilde{F}_{2}(x), x \in X,(4.9)F~1(x)F(x)F~2(x),xX,
for every F E m ( f ) F E m ( f ) F inE_(m)(f)F \in \mathrm{E}_{m}(f)FEm(f);
c) The set E m ( f ) E m ( f ) E_(m)(f)E_{m}(f)Em(f) is a convex subset of the ball (in the norm m m ||||_(m):}\left\|\|_{m}\right.m ) of center 0 and radius f m f m ||f||_(m)\| f \|_{m}fm in Λ α ( X , d ) Λ α ( X , d ) Lambda_(alpha)(X,d)\Lambda_{\alpha}(\mathrm{X}, \mathrm{d})Λα(X,d);
d) The functions F ~ 1 F ~ 1 widetilde(F)_(1)\widetilde{\mathrm{F}}_{1}F~1 and F ~ 2 F ~ 2 widetilde(F)_(2)\widetilde{\mathrm{F}}_{2}F~2 satisfying (4.9) are extremal elements of the set E m ( f ) E m ( f ) E_(m)(f)\mathrm{E}_{\mathrm{m}}(\mathrm{f})Em(f);
e) The function f f fff is an extremal element of the unit ball of Λ α ( Y , d ) Λ α ( Y , d ) Lambda_(alpha)(Y,d)\Lambda_{\alpha}(\mathrm{Y}, \mathrm{d})Λα(Y,d) if and only if E m ( f ) E m ( f ) E_(m)(f)\mathrm{E}_{\mathrm{m}}(\mathrm{f})Em(f) is a face of the unit ball of A a ( x , d ) A a ( x , d ) A_(a)(x,d)A_{a}(x, d)Aa(x,d).
Proof. a) If F E s ( f ) F E s ( f ) F inE_(s)(f)F \in E_{s}(f)FEs(f) then, by Theorem 6. a),
F α , x = f α , y and F u , x = f u , y , F α , x = f α , y  and  F u , x = f u , y , ||F||_(alpha,x)=||f||_(alpha,y)quad" and "||F||_(u,x)=||f||_(u,y),\|F\|_{\alpha, x}=\|f\|_{\alpha, y} \quad \text { and }\|F\|_{u, x}=\|f\|_{u, y},Fα,x=fα,y and Fu,x=fu,y,
so that F m = f m F m = f m quad||F||_(m)=||f||_(m)\quad\|F\|_{m}=\|f\|_{m}Fm=fm.
b) Let
H ˙ 1 ( x ) = inf { f ( y ) + f m d a ( x , y ) : y y } , x X . H ˙ 1 ( x ) = inf f ( y ) + f m d a ( x , y ) : y y , x X . H^(˙)_(1)(x)=i n f{f(y)+||f||_(m)d^(a)(x,y):y in y},x in X.\dot{H}_{1}(x)=\inf \left\{f(y)+\|f\|_{m} d^{a}(x, y): y \in y\right\}, x \in X .H˙1(x)=inf{f(y)+fmda(x,y):yy},xX.
Then (see [2]) the function H 1 H 1 H_(1)H_{1}H1 has the properties
H 1 | Y = f , H 1 α , X = f m . H 1 Y = f , H 1 α , X = f m . H_(1)|_(Y)=f,||H_(1)||_(alpha,X)=||f||_(m).\left.H_{1}\right|_{Y}=f,\left\|H_{1}\right\|_{\alpha, X}=\|f\|_{m} .H1|Y=f,H1α,X=fm.
The function
(4.10) F ~ 1 ( x ) = H 1 ( x ) if H 1 ( x ) f m , = f m if H 1 ( x ) > f m , (4.10) F ~ 1 ( x ) = H 1 ( x )  if  H 1 ( x ) f m , = f m  if  H 1 ( x ) > f m , {:[(4.10) widetilde(F)_(1)(x)=H_(1)(x)quad" if "quadH_(1)(x) <= ||f||_(m)","],[=||f||_(m)quad" if "quadH_(1)(x) > ||f||_(m)","]:}\begin{align*} \widetilde{F}_{1}(x) & =H_{1}(x) \quad \text { if } \quad H_{1}(x) \leq\|f\|_{m}, \tag{4.10}\\ & =\|f\|_{m} \quad \text { if } \quad H_{1}(x)>\|f\|_{m}, \end{align*}(4.10)F~1(x)=H1(x) if H1(x)fm,=fm if H1(x)>fm,
has the properties
F ~ 1 | Y = f , F ~ 1 m = f m F ~ 1 Y = f , F ~ 1 m = f m tilde(F)_(1)|_(Y)=f,quad|| tilde(F)_(1)||_(m)=||f||_(m)\left.\tilde{F}_{1}\right|_{\mathrm{Y}}=\mathrm{f}, \quad\left\|\tilde{F}_{1}\right\|_{\mathrm{m}}=\|\mathrm{f}\|_{\mathrm{m}}F~1|Y=f,F~1m=fm
that is F ~ 1 E m ( f ) F ~ 1 E m ( f ) widetilde(F)_(1)inE_(m)(f)\widetilde{F}_{1} \in E_{m}(f)F~1Em(f).
Similarly, by truncating the function
H 2 ( x ) = sup { f ( y ) γ f m d α ( x , y ) : y y } , x X H 2 ( x ) = sup f ( y ) γ f m d α ( x , y ) : y y , x X H_(2)(x)=s u p{f(y)-gamma f||_(m)d^(alpha)(x,y):y in y},x in XH_{2}(x)=\sup \left\{f(y)-\gamma f \|_{m} d^{\alpha}(x, y): y \in y\right\}, x \in XH2(x)=sup{f(y)γfmdα(x,y):yy},xX
one obtains the function
(4.11) F ~ 2 ( x ) = H 2 ( x ) if H 2 ( x ) f m , = f m if H 2 ( x ) < f m , (4.11) F ~ 2 ( x ) = H 2 ( x )  if  H 2 ( x ) f m , = f m  if  H 2 ( x ) < f m , {:[(4.11) tilde(F)_(2)(x)=H_(2)(x)" if "H_(2)(x) >= -||f||_(m)","],[=-||f||_(m)" if "H_(2)(x) < -||f||_(m)","]:}\begin{align*} \tilde{F}_{2}(x) & =H_{2}(x) & \text { if } & H_{2}(x) \geq-\|f\|_{m}, \tag{4.11}\\ & =-\|f\|_{m} & \text { if } & H_{2}(x)<-\|f\|_{m}, \end{align*}(4.11)F~2(x)=H2(x) if H2(x)fm,=fm if H2(x)<fm,
which is an extension of f f fff with respect to the "max-norm", i.e. F ~ 2 E m ( f ) F ~ 2 E m ( f ) quad widetilde(F)_(2)inE_(m)(f)\quad \widetilde{F}_{2} \in E_{m}(f)F~2Em(f).
The inequalities (4.9) can be obtained reasoning like in the proof of assertion b) in Theorem 2.
The proofs of c) and d) are similar to the proofs of assertion a) and c) of Theorem 2 and the proof of e) is similar to the proof of e) in Theorem 6.
From Theorem 6 and 7 obtains the following corolary:
COROLLARY 1. If ( X , d ) ( X , d ) (X,d)(\mathrm{X}, \mathrm{d})(X,d) is a metric space of finite diameter, Y Y YYY is a subset of X X XXX and f Λ α ( Y , d ) f Λ α ( Y , d ) f inLambda_(alpha)(Y,d)f \in \Lambda_{\alpha}(Y, d)fΛα(Y,d), then
a) E g ( f ) E m ( f ) E g ( f ) E m ( f ) quadE_(g)(f)subE_(m)(f)\quad E_{g}(f) \subset E_{m}(f)Eg(f)Em(f);
b) F ~ 1 ( x ) F ¯ 1 ( x ) F ¯ 2 ( x ) F ~ 2 ( x ) , x X F ~ 1 ( x ) F ¯ 1 ( x ) F ¯ 2 ( x ) F ~ 2 ( x ) , x X quad tilde(F)_(1)(x) >= bar(F)_(1)(x) >= bar(F)_(2)(x) >= tilde(F)_(2)(x),x in X\quad \tilde{F}_{1}(x) \geq \bar{F}_{1}(x) \geq \bar{F}_{2}(x) \geq \tilde{F}_{2}(x), x \in XF~1(x)F¯1(x)F¯2(x)F~2(x),xX,
where the functions F ~ 1 , F ~ 2 , F ¯ 1 , F ¯ 2 F ~ 1 , F ~ 2 , F ¯ 1 , F ¯ 2 tilde(F)_(1), tilde(F)_(2), bar(F)_(1), bar(F)_(2)\tilde{F}_{1}, \tilde{F}_{2}, \bar{F}_{1}, \bar{F}_{2}F~1,F~2,F¯1,F¯2 are defined by (4.3), (4.4), (4.10) and (4.11) , respectively.
5. In the following we shall give a procedure to find the global extrema of a function f Λ α ( X , d ) f Λ α ( X , d ) f inLambda_(alpha)(X,d)f \in \Lambda_{\alpha}(X, d)fΛα(X,d) by using the extensions of the restriction of f f fff to some finite subset of X X XXX.
Let ( X , d X , d X,dX, dX,d ) be a compact metric space and let f f fff be in Λ α ( X , d ) Λ α ( X , d ) Lambda_(alpha)(X,d)\Lambda_{\alpha}(X, d)Λα(X,d). If Y Y YYY is a subset of X X XXX and q Y α , Y q Y α , Y q >= ||||_(Y)||_(alpha,Y)q \geq\| \|_{Y} \|_{\alpha, Y}qYα,Y (here Y Y ||_(Y)\|_{Y}Y denotes the restriction of f f fff to Y Y YYY ), then the functions
(5.1) U q ( x ) = inf { f ( y ) + q d α ( x , y ) : y Y } , x X , u q ( x ) = sup { f ( y ) q d α ( x , y ) : y Y } , x X , (5.1) U q ( x ) = inf f ( y ) + q d α ( x , y ) : y Y , x X , u q ( x ) = sup f ( y ) q d α ( x , y ) : y Y , x X , {:[(5.1)U_(q)(x)=i n f{f(y)+qd^(alpha)(x,y):y in Y}","x in X","],[u_(q)(x)=s u p{f(y)-qd^(alpha)(x,y):y in Y}","x in X","]:}\begin{align*} & U_{q}(x)=\inf \left\{f(y)+q d^{\alpha}(x, y): y \in Y\right\}, x \in X, \tag{5.1}\\ & u_{q}(x)=\sup \left\{f(y)-q d^{\alpha}(x, y): y \in Y\right\}, x \in X, \end{align*}(5.1)Uq(x)=inf{f(y)+qdα(x,y):yY},xX,uq(x)=sup{f(y)qdα(x,y):yY},xX,
are extensions of f | Y f Y f|_(Y)\left.f\right|_{Y}f|Y which belongs to Λ α ( X , d ) Λ α ( X , d ) Lambda_(alpha)(X,d)\Lambda_{\alpha}(X, d)Λα(X,d) and have the Hölder norms at most q q qqq (see [2]). If u u uuu and u u uuu denotes the
functions defined by (5.1) for q = f α , x q = f α , x q=||f||_(alpha,x)q=\|f\|_{\alpha, x}q=fα,x then
(5.2) u ( x ) f ( x ) U ( x ) , x X u ( x ) f ( x ) U ( x ) , x X quad u(x) <= f(x) <= U(x),x in X\quad u(x) \leq f(x) \leq U(x), x \in Xu(x)f(x)U(x),xX,
and for q > f α , x q > f α , x q > ||f||_(alpha,x)q>\|f\|_{\alpha, x}q>fα,x we have
(5.3) u q ( x ) u ( x ) f ( x ) U ( x ) U q ( x ) , x X u q ( x ) u ( x ) f ( x ) U ( x ) U q ( x ) , x X quadu_(q)(x) <= u(x) <= f(x) <= U(x) <= U_(q)(x),x in X\quad u_{q}(x) \leq u(x) \leq f(x) \leq U(x) \leq U_{q}(x), x \in Xuq(x)u(x)f(x)U(x)Uq(x),xX.
A maximum (respectively a minimum) point of a function
f : X R f : X R f:X rarr Rf: X \rightarrow Rf:XR is a point x X x X x^(**)in Xx^{*} \in XxX such that
(5.4) f ( x ) f ( x ) f x f ( x ) quad f(x^(**)) >= f(x)quad\quad f\left(x^{*}\right) \geq f(x) \quadf(x)f(x) (respectively f ( x ) f ( x ) f x f ( x ) f(x^(**)) <= f(x)f\left(x^{*}\right) \leq f(x)f(x)f(x) )
for all x X x X x in Xx \in XxX.
For a bounded real function f f fff on X X XXX put
(5.5)
M f = sup { f ( x ) : x X } , m f = { f ( x ) : x X } E f = { x X : f ( x ) = M f } , e f = { x X : f ( x ) = m f } M f = sup { f ( x ) : x X } , m f = { f ( x ) : x X } E f = x X : f ( x ) = M f , e f = x X : f ( x ) = m f {:[M_(f)=s u p{f(x):x in X}","m_(f)={f(x):x in X}],[E_(f)={x in X:f(x)=M_(f)}","e_(f)={x in X:f(x)=m_(f)}]:}\begin{aligned} & M_{f}=\sup \{f(x): x \in X\}, m_{f}=\{f(x): x \in X\} \\ & E_{f}=\left\{x \in X: f(x)=M_{f}\right\}, e_{f}=\left\{x \in X: f(x)=m_{f}\right\} \end{aligned}Mf=sup{f(x):xX},mf={f(x):xX}Ef={xX:f(x)=Mf},ef={xX:f(x)=mf}
Let now ( X , d ) ( X , d ) (X,d)(X, d)(X,d) be a compact metric and let f f fff be a function in Λ α ( x , d ) Λ α ( x , d ) Lambda_(alpha)(x,d)\Lambda_{\alpha}(x, d)Λα(x,d).
We define now inductively two sequences ( x n ) n 0 x n n 0 (x_(n))_(n >= 0)\left(x_{n}\right)_{n \geq 0}(xn)n0 and ( M n ) n 0 M n n 0 (M_(n))_(n >= 0)\left(M_{n}\right)_{n \geq 0}(Mn)n0 of points in X X XXX and of real numbers, respectively, as follows:
Let q f α , x q f α , x q >= ||f||_(alpha,x)q \geq\|f\|_{\alpha, x}qfα,x be fixed and let x 0 x 0 x_(0)x_{0}x0 be a fixed point in X X XXX. Let u 0 ( x ) = f ( x 0 ) + q d α ( x , x 0 ) , x X u 0 ( x ) = f x 0 + q d α x , x 0 , x X u^(0)(x)=f(x_(0))+qd^(alpha)(x,x_(0)),x in Xu^{0}(x)=f\left(x_{0}\right)+q d^{\alpha}\left(x, x_{0}\right), x \in Xu0(x)=f(x0)+qdα(x,x0),xX, the greatest extension of f f fff obtained from (5.1) for Y = { x 0 } Y = x 0 Y={x_(0)}Y=\left\{x_{0}\right\}Y={x0} and let
M 0 = sup { U 0 ( x ) : x X } : M 0 = sup U 0 ( x ) : x X : M_(0)=s u p{U^(0)(x):x in X}:M_{0}=\sup \left\{U^{0}(x): x \in X\right\}:M0=sup{U0(x):xX}:
Let x 1 X x 1 X x_(1)in Xx_{1} \in Xx1X be a point with u 0 ( x 1 ) = M 0 u 0 x 1 = M 0 u^(0)(x_(1))=M_(0)u^{0}\left(x_{1}\right)=M_{0}u0(x1)=M0.
Suppose now that for a natural number n 1 n 1 n >= 1n \geq 1n1, the points x 0 , x 1 , , x n 1 x 0 , x 1 , , x n 1 x_(0),x_(1),dots,x_(n-1)x_{0}, x_{1}, \ldots, x_{n-1}x0,x1,,xn1 and the numbers M 0 , M 1 , , M n 1 M 0 , M 1 , , M n 1 M_(0),M_(1),dots,M_(n-1)M_{0}, M_{1}, \ldots, M_{n-1}M0,M1,,Mn1 were defined. Let U n 1 U n 1 U^(n-1)U^{n-1}Un1 be the greatest extension of I Y I Y I_(Y)I_{Y}IY obtained from (5.1) for y = { x 0 , x 1 , , x n 1 } y = x 0 , x 1 , , x n 1 y={x_(0),x_(1),dots,x_(n-1)}y=\left\{x_{0}, x_{1}, \ldots, x_{n-1}\right\}y={x0,x1,,xn1}. Put
M n = sup { U n 1 ( x ) : x X } M n = sup U n 1 ( x ) : x X M_(n)=s u p{U^(n-1)(x):x in X}M_{n}=\sup \left\{U^{n-1}(x): x \in X\right\}Mn=sup{Un1(x):xX}
and let x n x n x_(n)x_{n}xn be a point in x x xxx such that u n 1 ( x n ) = M n u n 1 x n = M n u^(n-1)(x_(n))=M_(n)u^{n-1}\left(x_{n}\right)=M_{n}un1(xn)=Mn.
The properties of the so defined sequences ( x n ) n 0 x n n 0 (x_(n))_(n >= 0)\left(x_{n}\right)_{n \geq 0}(xn)n0 and
Theorem a. Let ( X , d ) ( X , d ) (X,d)(\mathrm{X}, \mathrm{d})(X,d) be a compact metric space and let f Λ α ( x , d ) f Λ α ( x , d ) f inLambda_(alpha)(x,d)f \in \Lambda_{\alpha}(x, d)fΛα(x,d). For a fixed, q f α , x q f α , x q >= ||f||_(alpha,x)q \geq\|f\|_{\alpha, x}qfα,x let the sequences ( x n ) n 0 x n n 0 (x_(n))_(n >= 0)\left(x_{n}\right)_{n \geq 0}(xn)n0 and ( M n ) n 0 M n n 0 (M_(n))_(n >= 0)\left(M_{n}\right)_{n \geq 0}(Mn)n0 be defined as above. Then
a) lim n M n = M f lim n M n = M f quadlim_(n rarr oo)M_(n)=M_(f)^(')\quad \lim _{n \rightarrow \infty} M_{n}=M_{f}^{\prime}limnMn=Mf;
b) lim n [ inf { d ( x , x n ) : x E f } ] = 0 lim n inf d x , x n : x E f = 0 quadlim_(n rarr oo)[i n f{d(x,x_(n)):x inE_(f)}]=0\quad \lim _{n \rightarrow \infty}\left[\inf \left\{d\left(x, x_{n}\right): x \in E_{f}\right\}\right]=0limn[inf{d(x,xn):xEf}]=0;
c) The sequence ( f ( x n ) ) n 0 f x n n 0 (f(x_(n)))_(n >= 0)\left(f\left(x_{n}\right)\right)_{n \geq 0}(f(xn))n0 has the number M f M f M_(f)M_{f}Mf as a limit point.
Proor. Since U n U n 1 U n U n 1 U^(n) <= U^(n-1)U^{n} \leq U^{n-1}UnUn1 for n = 1 , 2 , n = 1 , 2 , n=1,2,dotsn=1,2, \ldotsn=1,2, it follows that the sequence ( M n ) n 0 M n n 0 (M_(n))_(n >= 0)\left(M_{n}\right)_{n \geq 0}(Mn)n0 is nonincreasing. By (5.3), M n = U n 1 ( x n ) ≥≥ f ( x n ) min { f ( x ) : x X } M n = U n 1 x n ≥≥ f x n min { f ( x ) : x X } M_(n)=U^(n-1)(x_(n))≥≥f(x_(n)) >= min{f(x):x in X}M_{n}=U^{n-1}\left(x_{n}\right) \geq \geq f\left(x_{n}\right) \geq \min \{f(x): x \in X\}Mn=Un1(xn)≥≥f(xn)min{f(x):xX} so that the sequence. ( M n ) n 0 M n n 0 (M_(n))_(n >= 0)\left(M_{n}\right)_{n \geq 0}(Mn)n0 is also bounded. Therefore there exists M = lim n M n B y ( 5.3 ) M = lim n M n B y ( 5.3 ) M=lim_(n rarr oo)M_(n)*By(5.3)M=\lim _{n \rightarrow \infty} M_{n} \cdot B y(5.3)M=limnMnBy(5.3), f ( x ) U n ( x ) M n f ( x ) U n ( x ) M n f(x) <= U^(n)(x) <= M_(n)f(x) \leq U^{n}(x) \leq M_{n}f(x)Un(x)Mn, for all x N x N x in Nx \in NxN so that (5.6) f ( x ) M f ( x ) M quad f(x) <= M\quad f(x) \leq Mf(x)M, for all x X x X x in Xx \in XxX.
The metric space X X XXX being compact, the sequence ( x n ) n 0 x n n 0 (x_(n))_(n >= 0)\left(x_{n}\right)_{n \geq 0}(xn)n0 contains a subsequence ( x n k ) k 0 x n k k 0 (x_(n_(k)))_(k >= 0)\left(x_{n_{k}}\right)_{k \geq 0}(xnk)k0 converging to a point z X z X z in Xz \in XzX. Since the function f f fff is continuous it follows that
(5.7) f ( x n k ) f ( z ) , k f x n k f ( z ) , k quad f(x_(n_(k)))longrightarrow f(z),k longrightarrow oo\quad f\left(x_{n_{k}}\right) \longrightarrow f(z), k \longrightarrow \inftyf(xnk)f(z),k.
But, for k 1 k 1 k >= 1k \geq 1k1,
| U n k 1 ( z ) M n k 1 | = | 0 n k 1 ( z ) 0 n k 1 ( x n k ) | q d α ( z , x n k ) 0 U n k 1 ( z ) M n k 1 = 0 n k 1 ( z ) 0 n k 1 x n k q d α z , x n k 0 |U^(n_(k)-1)(z)-M_(n_(k)-1)|=|0^(n_(k)-1)(z)-0^(n_(k)-1)(x_(n_(k)))| <= qd^(alpha)(z,x_(n_(k)))rarr0\left|U^{n_{k}-1}(z)-M_{n_{k}-1}\right|=\left|0^{n_{k}-1}(z)-0^{n_{k}-1}\left(x_{n_{k}}\right)\right| \leq q d^{\alpha}\left(z, x_{n_{k}}\right) \rightarrow 0|Unk1(z)Mnk1|=|0nk1(z)0nk1(xnk)|qdα(z,xnk)0
for k k k longrightarrow ook \longrightarrow \inftyk, and M n k 1 M M n k 1 M M_(n_(k)-1)longrightarrow MM_{n_{k}-1} \longrightarrow MMnk1M for k k k longrightarrow ook \longrightarrow \inftyk, so that (5.8) u n k 1 ( z ) M u n k 1 ( z ) M quadu^(n_(k)-1)(z)rarrM\quad \mathrm{u}^{\mathrm{n}_{\mathrm{k}}-1}(\mathrm{z}) \rightarrow \mathrm{M}unk1(z)M, for k k krarr oo\mathrm{k} \rightarrow \inftyk.
By the relation
| U n k ( z ) f ( x n k ) | = | U n k ( z ) U n k ( x n k ) | q d α ( z , x n k ) 0 U n k ( z ) f x n k = U n k ( z ) U n k x n k q d α z , x n k 0 |U^(n_(k))(z)-f(x_(n_(k)))|=|U^(n_(k))(z)-U^(n_(k))(x_(n_(k)))| <= qd^(alpha)(z,x_(n_(k)))longrightarrow0\left|U^{n_{k}}(z)-f\left(x_{n_{k}}\right)\right|=\left|U^{n_{k}}(z)-U^{n_{k}}\left(x_{n_{k}}\right)\right| \leq q d^{\alpha}\left(z, x_{n_{k}}\right) \longrightarrow 0|Unk(z)f(xnk)|=|Unk(z)Unk(xnk)|qdα(z,xnk)0,
k k k rarr ook \rightarrow \inftyk, and by (5.7) it follows that (5.9) u n k ( z ) f ˙ ( z ) , k u n k ( z ) f ˙ ( z ) , k quadu^(n_(k))(z)longrightarrowf^(˙)(z),quad k longrightarrow oo\quad u^{n_{k}}(z) \longrightarrow \dot{f}(z), \quad k \longrightarrow \inftyunk(z)f˙(z),k. Therefore, if in the inequalities
f ( z ) U n k 1 ( z ) U n k 1 ( z ) , k 1 , f ( z ) U n k 1 ( z ) U n k 1 ( z ) , k 1 , f(z) <= U^(n_(k)-1)(z) <= U^(n_(k-1))(z),quad k >= 1,f(z) \leq U^{n_{k}-1}(z) \leq U^{n_{k-1}}(z), \quad k \geq 1,f(z)Unk1(z)Unk1(z),k1,
we let k k k longrightarrow ook \longrightarrow \inftyk one obtains f ( z ) M f ( z ) f ( z ) M f ( z ) f(z) <= M <= f(z)f(z) \leq M \leq f(z)f(z)Mf(z), so that f ( z ) = M f ( z ) = M f(z)=Mf(z)=Mf(z)=M. Taking into account (5.6) it follows that
M = f ( z ) = max { f ( x ) : x X } . M = f ( z ) = max { f ( x ) : x X } . M=f(z)=max{f(x):x in X}.M=f(z)=\max \{f(x): x \in X\} .M=f(z)=max{f(x):xX}.
To prove b) , observe that if contrary, then there exist > 0 > 0 > 0>0>0 and an infinite subset J J JJJ of N N NNN such that
(5.10) inf { d ( x , x j ) : x E f } ε , (5.10) inf d x , x j : x E f ε , {:(5.10)i n f{d(x,x_(j)):x inE_(f)} >= epsi",":}\begin{equation*} \inf \left\{d\left(x, x_{j}\right): x \in E_{f}\right\} \geq \varepsilon, \tag{5.10} \end{equation*}(5.10)inf{d(x,xj):xEf}ε,
for all j J j J j in Jj \in JjJ. The space x x xxx being compact there exists a subsequence ( x j k ) k 0 x j k k 0 (x_(j_(k)))_(k >= 0)\left(x_{j_{k}}\right)_{k \geq 0}(xjk)k0 of ( x j ) j J x j j J x_(j))_(j in J)\left.x_{j}\right)_{j \in J}xj)jJ converging to a point y X y X y in Xy \in XyX. But then, repeating the above arguments will follows that y E f y E f y inE_(f)y \in \mathrm{E}_{f}yEf, which contradicts (5.10).
The affirmation c) follows from (5.7).
Remarks. 1) In the clase X = [ a , b ] X = [ a , b ] X=[a,b]\mathrm{X}=[\mathrm{a}, \mathrm{b}]X=[a,b] and α = 1 α = 1 alpha=1\alpha=1α=1 a similar result is proved in [7].
2) If the extensions u n u n u^(n)u^{n}un are replaced by the extensions u n u n u^(n)u^{n}un and m n = inf { u n ( x ) : x x } , u n ( x n + 1 ) = m n m n = inf u n ( x ) : x x , u n x n + 1 = m n m_(n)=i n f{u^(n)(x):x in x},u^(n)(x_(n+1))=m_(n)m_{n}=\inf \left\{u^{n}(x): x \in x\right\}, u^{n}\left(x_{n+1}\right)=m_{n}mn=inf{un(x):xx},un(xn+1)=mn, then one obtains a procedure to find the minimum m f m f m_(f)m_{f}mf of a function f Λ α ( x , d ) f Λ α ( x , d ) f inLambda_(alpha)(x,d)f \in \Lambda_{\alpha}(x, d)fΛα(x,d).
Example. Let x = [ 0 , 1 ] , d ( x , y ) = | x y | x = [ 0 , 1 ] , d ( x , y ) = | x y | x=[0,1],d(x,y)=|x-y|\mathrm{x}=[0,1], \mathrm{d}(\mathrm{x}, \mathrm{y})=|\mathrm{x}-\mathrm{y}|x=[0,1],d(x,y)=|xy| and
f ( x ) = x sin ( 1 / x ) , if x ( 0 , 1 ] , = 0 , if x = 0 . f ( x ) = x sin ( 1 / x ) ,  if  x ( 0 , 1 ] , = 0 ,  if  x = 0 . {:[f(x)=x*sin(1//x)","" if "x in(0","1]","],[=0","" if "x=0.]:}\begin{aligned} f(x) & =x \cdot \sin (1 / x), \text { if } x \in(0,1], \\ & =0, \text { if } x=0 . \end{aligned}f(x)=xsin(1/x), if x(0,1],=0, if x=0.
It is known that f Λ α ( x , d ) f Λ α ( x , d ) f inLambda_(alpha)(x,d)f \in \Lambda_{\alpha}(x, d)fΛα(x,d) if and only if α ( 0 , 1 / 2 ] α ( 0 , 1 / 2 ] alpha in(0,1//2]\alpha \in(0,1 / 2]α(0,1/2] ( see [8], Problem 153 ) and in this case
f α , x [ 1 + 2 ln ( 1 + 2 π ) + 2 π ] 1 / 2 < 4 . f α , x [ 1 + 2 ln ( 1 + 2 π ) + 2 π ] 1 / 2 < 4 . ||f||_(alpha,x) <= [1+2ln(1+2pi)+2pi]^(1//2) < 4.\|f\|_{\alpha, x} \leq[1+2 \ln (1+2 \pi)+2 \pi]^{1 / 2}<4 .fα,x[1+2ln(1+2π)+2π]1/2<4.
  1. ARONSSON, G. : Extension of function satisfying Lipschitz conditions, Arkiv für Mathematik 6, 28 (1967) , 551-561.
  2. CZIPSER, J. and GHEHER, L. : Extension of functions satisfying a Lipschitz condition , Acta Math. Acad. Sci. Hungar 6(1955), 213-220.
  3. LEVY, R. and RICE, M. D. : The approximation of uniformly continuous mappings by Lipschitz and Hölder mappings, (preprint) 1980, 29 p .
  4. McSHANE, E. J. : Extension of range of functions, Bull. Amer. Soc. 40 (1934) , 837-842 .
  5. MUSTĂTA, C. : Best approximation and unique extension of Lipschitz functions, Journal Approx. Theory 19(1977), 222--230.
  6. SINGER, I.: Best approximation in normed linear spaces by elements of linear subspaces, Springer-Verlag, 171, 1970.
  7. SHUBERT, B. : A sequential method seeking the global maximum of a function, SIAM J.Num.Anal. 9(1972), 379-388.
  8. . The Otto Dunkel Memorial Problem Book, New York, 1957.
1991

Related Posts