Best approximation and unique extension of Lipschitz functions

Abstract

Authors

Costica Mustata
Institutul de Calcul, Romania (ICTP)

Keywords

Paper coordinates

C. Mustata, Best approximation and unique extension of Lipschitz functions,  J. Approx. Theory 19 (1977), 222-230, doi: 10.1016/0021-9045(77)90053-3

PDF

About this paper

Journal

Journal of Approximation Theory

Publisher Name
Print ISSN

?

Online ISSN

?

Google Scholar Profile

?

[1] D.E BOOR, “On “Best” Interpolation,” Universky of Wisconsin, MRC 7% No. 1426, 1971.
[2] J. CZIPSER AND L. GEHER, Extension of function  satisfying a Eipschitz condition, hia Mut,k. ACM/. SC. Hungm. 6 (1955), 213-220.
[3] C. B. DUNHAM, Chebyshev approximation with a null space, Proc. Amer. Math. Sot. 41 (1973), 557-558.
[4] J. A. JOHNSON, Banach space of Lipschitz functions and vector-valued Lipschitz functions, Tuarzs. Amer. Math. Sot. 148 (1970), 147-169.
[5] C. MUSTATA, Asupra unor subspalii cebiseviene din spatiul normat al functiilor lipschitziene, Reo. Anal Mumer. Teoria Aproximatiei 2 (1973), 81-87.
[6] R. R. PHELPS, Uniqueness of Hahn-Banach extension and unique best approximation, Trans. Amer. Math. Sot. 95 (1960), 238-255.
[7] A. K. ROY, Extreme points and linear isometries of Banach space of Lipschitz functions, Canad. J. IWZ~A. 20 (1968), 1150-1164.
[8] I. SINGER, Cea mai buna aproximare in spalii vectoriale normate prin elemente din subspalii vectoriale, Edit. Acad. R. S. Romania, Bucuresti, (1967).
[9] D. R. SHERBERT, The structure of ideals and point derivations in Banach algebras of Lipschitz functions, Trans. Amer. Math. Sot. 111 (1964), 240-272.

Paper (preprint) in HTML form

1977-Mustata-J.-Approx.-Th.-Best-approximation-and-unique-extension-of-Lipschitz-functions

Best Approximation and Unique Extension of Lipschitz Functions

Costică MustățaInstitutul de Calcul, Cluj-Napoca, R. S. RontaniaCommunicated by Lothar Collatz

Received July 21, 1975

1. Introduction

In the sequel, X X XXX will always denote a metric space with the metric d , x 0 d , x 0 d,x_(0)d, x_{0}d,x0 a fixed point from X X XXX, and Y Y YYY a subset of X X XXX such that x 0 Y x 0 Y x_(0)in Yx_{0} \in Yx0Y. If f f fff is a real-valued function defined on X X XXX, denote
(1.1) f Y = sup { | f ( x ) f ( y ) | / d ( x , y ) : x , y Y , x y } . (1.1) f Y = sup { | f ( x ) f ( y ) | / d ( x , y ) : x , y Y , x y } . {:(1.1)||f||_(Y)=s u p{|f(x)-f(y)|//d(x","y):x","y in Y","x!=y}.:}\begin{equation*} \|f\|_{Y}=\sup \{|f(x)-f(y)| / d(x, y): x, y \in Y, x \neq y\} . \tag{1.1} \end{equation*}(1.1)fY=sup{|f(x)f(y)|/d(x,y):x,yY,xy}.
A Lipschitz function on X X XXX is a function f : X R f : X R f:X rarr Rf: X \rightarrow Rf:XR such that f X < f X < ||f||_(X) < oo\|f\|_{X}<\inftyfX<. Denote by Lip 0 X 0 X _(0)X{ }_{0} X0X the Banach space of all Lipschitz functions on X X XXX which vanish at x 0 x 0 x_(0)x_{0}x0, with the norm f = f X f = f X ||f||=||f||_(X)\|f\|=\|f\|_{X}f=fX. Put also
(1.2) Y = { f : f Lip 0 X , f | Y = 0 } . (1.2) Y = f : f Lip 0 X , f Y = 0 . {:(1.2)Y^(_|_)={f:f inLip_(0)X,f|_(Y)=0}.:}\begin{equation*} Y^{\perp}=\left\{f: f \in \operatorname{Lip}_{0} X,\left.f\right|_{Y}=0\right\} . \tag{1.2} \end{equation*}(1.2)Y={f:fLip0X,f|Y=0}.
A Lipschitz extension of a function f Lip 0 Y f Lip 0 Y f inLip_(0)Yf \in \operatorname{Lip}_{0} YfLip0Y is a function F Lip 0 X F Lip 0 X F inLip_(0)XF \in \operatorname{Lip}_{0} XFLip0X such that F | Y = f F Y = f F|_(Y)=f\left.F\right|_{Y}=fF|Y=f and F X = f Y F X = f Y ||F||_(X)=||f||_(Y)\|F\|_{X}=\|f\|_{Y}FX=fY. It is known (see, e.g., [2]) that every f Lip 0 Y f Lip 0 Y f inLip_(0)Yf \in \operatorname{Lip}_{0} YfLip0Y has a Lipschitz extension in Lip Lip 0 X Lip 0 X Lip_(0)X\operatorname{Lip}_{0} XLip0X.
For a subset Y Y YYY of X X XXX and x X x X x in Xx \in XxX we put
(1.3) d ( x , Y ) = inf { d ( x , y ) : y Y } . (1.3) d ( x , Y ) = inf { d ( x , y ) : y Y } . {:(1.3)d(x","Y)=i n f{d(x","y):y in Y}.:}\begin{equation*} d(x, Y)=\inf \{d(x, y): y \in Y\} . \tag{1.3} \end{equation*}(1.3)d(x,Y)=inf{d(x,y):yY}.
Now, let E E EEE be a normed linear space, G G GGG a nonempty subset of E , x E , x E,xE, xE,x an element from E E EEE, and
(1.4) P G ( x ) = { y G : x y = d ( x , G ) } . (1.4) P G ( x ) = { y G : x y = d ( x , G ) } . {:(1.4)P_(G)(x)={y in G:||x-y||=d(x","G)}.:}\begin{equation*} P_{G}(x)=\{y \in G:\|x-y\|=d(x, G)\} . \tag{1.4} \end{equation*}(1.4)PG(x)={yG:xy=d(x,G)}.
An element from P G ( x ) P G ( x ) P_(G)(x)P_{G}(x)PG(x) is called a best approximation to x x xxx from G G GGG. If M M MMM is a subset of E E EEE we say that G G GGG is M M MMM-proximinal if P G ( x ) P G ( x ) P_(G)(x)!=O/P_{G}(x) \neq \varnothingPG(x), for all x M x M x in Mx \in MxM. If P G ( x ) P G ( x ) P_(G)(x)P_{G}(x)PG(x) contains exactly one element for every x M x M x in Mx \in MxM, then G G GGG is called M M MMM-chebyshevian. If the set G G GGG is E E EEE-proximinal (respectively E E EEE-chebyshevian) then we say, simply, that G G GGG is proximinal (respectively chebyshevian).
We say that a linear subspace Z Z ZZZ of E E EEE has the property ( U U UUU ) if every continuous linear functional on Z Z ZZZ has a unique Hahn-Banach extension to E E EEE (i.e., linear and norm preserving) [6]. Let us denote by E E E^(**)E^{*}E the conjugate spacs of E E EEE and by Z l Z Z^("l ")Z^{\text {l }}Z the annihilator of the subspace Z Z ZZZ in E E E^(**)E^{*}E, i.e.,
(1.5) Z Ω = { φ E ; φ | Z = 0 } (1.5) Z Ω = φ E ; φ Z = 0 {:(1.5)Z^(Omega)={varphi inE^(**); varphi|_(Z)=0}:}\begin{equation*} Z^{\Omega}=\left\{\varphi \in E^{*} ;\left.\varphi\right|_{Z}=0\right\} \tag{1.5} \end{equation*}(1.5)ZΩ={φE;φ|Z=0}
Phelps [6] showed that the subspace Z Z ZZZ of E E EEE has property ( U U UUU ) if and only if its annihilator Z 1 Z 1 Z^(1)Z^{\mathbb{1}}Z1 is chebyshevian. This result can be extended to Lipschitz functions:
Theorem 1 ([5, Lemma 2]). Let X X XXX be a metric space, x 0 x 0 x_(0)x_{0}x0 in X X XXX, and Y X Y X Y sube XY \subseteq XYX such that x 0 Y x 0 Y x_(0)in Yx_{0} \in Yx0Y. The space Y Y Y^(_|_)Y^{\perp}Y is chebyshevian for f Lip 0 X f Lip 0 X f inLip_(0)Xf \in \operatorname{Lip}_{0} XfLip0X if and only if f | Y Lip 0 Y f Y Lip 0 Y f|_(Y)inLip_(0)Y\left.f\right|_{Y} \in \operatorname{Lip}_{0} Yf|YLip0Y has a unique Lipschitz extension in Lip 0 X Lip 0 X Lip_(0)X\operatorname{Lip}_{0} XLip0X.
We also need the following lemma.
Lemma 1. Every best approximation to f Lip 0 X f Lip 0 X f inLip_(0)Xf \in \operatorname{Lip}_{0} XfLip0X from Y 1 Y 1 Y^(-1)Y^{-1}Y1 is of the form f F f F f-Ff-FfF, where F F FFF is a Lipschitz extension of f | Y f Y f_(|_(Y))f_{\left.\right|_{Y}}f|Y to X X XXX.
Proof. Suppose F F FFF is a Lipschit extension of f | y f y f|_(y)\left.f\right|_{y}f|y to X X XXX. Then, by [ 5 [ 5 [5[5[5, Theorem 2 and Lemma 1], we get
| f ( f I ) | X F | X f X d ( f , Y ) | f ( f I ) | X F X f X d f , Y |f-(f-I)|_(X)-||F|_(X)-||f||_(X)-d(f,Y^(_|_))|f-(f-I)|_{X}-\left\|\left.F\right|_{X}-\right\| f \|_{X}-d\left(f, Y^{\perp}\right)|f(fI)|XF|XfXd(f,Y)
Conversely, if g Y g Y g inY^(_|_)g \in Y^{\perp}gY is a best approximation to f f fff, then f g X = d ( f , Y ) = | i f | Y f g X = d f , Y = | i f | Y ||f-g||_(X)=d(f,Y^(_|_))=|if|_(Y)\|f-g\|_{X}= d\left(f, Y^{\perp}\right)=|i f|_{Y}fgX=d(f,Y)=|if|Y and ( f g ) | Y = f | Y ( f g ) Y = f Y (f-g)|_(Y)=f|_(Y)\left.(f-g)\right|_{Y}=\left.f\right|_{Y}(fg)|Y=f|Y. Therefore F = f g F = f g F=f-gF=f-gF=fg is a Lipschitz extension of f | Y f Y f|_(Y)\left.f\right|_{Y}f|Y,

2. Main Theorem

A metric space X X XXX is called uniformly discrete if there exists a number δ > 0 δ > 0 delta > 0\delta>0δ>0, such that d ( x , y ) δ d ( x , y ) δ d(x,y) >= deltad(x, y) \geqslant \deltad(x,y)δ for all x , y X x , y X x,y in Xx, y \in Xx,yX with x y x y x!=yx \neq yxy. The following theorem appears in [5], in the hypothesis that Y Y YYY has an accumulation point in X X XXX. The main result is:
Theorem 2. Let X , x 0 X , x 0 X,x_(0)X, x_{0}X,x0, and Y Y YYY be as in Theorem 1. Suppose, further, that Y Y YYY is nonuniformly discrete. If every f Lip 0 Y f Lip 0 Y f inLip_(0)Yf \in \operatorname{Lip}_{0} YfLip0Y has a unique Lipschitz extension. then Y = X Y = X vec(Y)=X\vec{Y}=XY=X (or equivalently Y L = { 0 } Y L = { 0 } Y^(L)={0}Y^{\mathrm{L}}=\{0\}YL={0} ).
Proof. Since Y Y YYY is nonuniformly discrete, for every n N n N n in Nn \in NnN, there exist x n , y n Y , x n y n x n , y n Y , x n y n x_(n),y_(n)in Y,x_(n)!=y_(n)x_{n}, y_{n} \in Y, x_{n} \neq y_{n}xn,ynY,xnyn such that d ( x n , y n ) < 1 / n d x n , y n < 1 / n d(x_(n),y_(n)) < 1//nd\left(x_{n}, y_{n}\right)<1 / nd(xn,yn)<1/n. Defining f n : X R f n : X R f_(n):X rarr Rf_{n}: X \rightarrow Rfn:XR by
f n ( x ) = d ( x , x n ) d ( x , y n ) d ( x 0 , x n ) + d ( x 0 , y n ) , n = 1.2 , 3 , . f n ( x ) = d x , x n d x , y n d x 0 , x n + d x 0 , y n , n = 1.2 , 3 , . f_(n)(x)=d(x,x_(n))-d(x,y_(n))-d(x_(0),x_(n))+d(x_(0),y_(n)),quad n=1.2,3,dots.f_{n}(x)=d\left(x, x_{n}\right)-d\left(x, y_{n}\right)-d\left(x_{0}, x_{n}\right)+d\left(x_{0}, y_{n}\right), \quad n=1.2,3, \ldots .fn(x)=d(x,xn)d(x,yn)d(x0,xn)+d(x0,yn),n=1.2,3,.
we have
f n ( x 0 ) = 0 , n = 1 , 2 , 3 , , 2 d ( x n , y n ) f n ( x n ) = d ( x n , y n ) d ( x 0 , x n ) + d ( x 0 , y n ) 0 , n = 1 , 2 , 3 , , 0 f n ( y n ) = d ( x n , y n ) d ( x 0 , x n ) + d ( x 0 , y n ) 2 d ( x n , y n ) , n = 1 , 2 , 3 , , f n X = sup { | d ( x , x n ) d ( x , y n ) d ( y , x n ) + d ( y , y n ) | d ( x , y ) : x , y Y , x y } 2 , n = 1 , 2 , 3 , , f n x 0 = 0 , n = 1 , 2 , 3 , , 2 d x n , y n f n x n = d x n , y n d x 0 , x n + d x 0 , y n 0 , n = 1 , 2 , 3 , , 0 f n y n = d x n , y n d x 0 , x n + d x 0 , y n 2 d x n , y n , n = 1 , 2 , 3 , , f n X = sup d x , x n d x , y n d y , x n + d y , y n d ( x , y ) : x , y Y , x y } 2 , n = 1 , 2 , 3 , , {:[f_(n)(x_(0))=0","quad n=1","2","3","dots","],[-2d(x_(n),y_(n)) <= f_(n)(x_(n))=-d(x_(n),y_(n))-d(x_(0),x_(n))+d(x_(0),y_(n))],[ <= 0","quad n=1","2","3","dots","],[0 <= f_(n)(y_(n))=d(x_(n),y_(n))-d(x_(0),x_(n))+d(x_(0),y_(n))],[ <= 2d(x_(n),y_(n))","quad n=1","2","3","dots","],[||f_(n)||_(X)=s u p{|d(x,x_(n))-d(x,y_(n))-d(y,x_(n))+d(y,y_(n))|d(x,y)::}],[x","y in Y","x!=y} <= 2","quad n=1","2","3","dots","]:}\begin{aligned} & f_{n}\left(x_{0}\right)= 0, \quad n=1,2,3, \ldots, \\ &-2 d\left(x_{n}, y_{n}\right) \leqslant f_{n}\left(x_{n}\right)=-d\left(x_{n}, y_{n}\right)-d\left(x_{0}, x_{n}\right)+d\left(x_{0}, y_{n}\right) \\ & \leqslant 0, \quad n=1,2,3, \ldots, \\ & 0 \leqslant f_{n}\left(y_{n}\right)=d\left(x_{n}, y_{n}\right)-d\left(x_{0}, x_{n}\right)+d\left(x_{0}, y_{n}\right) \\ & \leqslant 2 d\left(x_{n}, y_{n}\right), \quad n=1,2,3, \ldots, \\ &\left\|f_{n}\right\|_{X}= \sup \left\{\left|d\left(x, x_{n}\right)-d\left(x, y_{n}\right)-d\left(y, x_{n}\right)+d\left(y, y_{n}\right)\right| d(x, y):\right. \\ &x, y \in Y, x \neq y\} \leqslant 2, \quad n=1,2,3, \ldots, \end{aligned}fn(x0)=0,n=1,2,3,,2d(xn,yn)fn(xn)=d(xn,yn)d(x0,xn)+d(x0,yn)0,n=1,2,3,,0fn(yn)=d(xn,yn)d(x0,xn)+d(x0,yn)2d(xn,yn),n=1,2,3,,fnX=sup{|d(x,xn)d(x,yn)d(y,xn)+d(y,yn)|d(x,y):x,yY,xy}2,n=1,2,3,,
so that f n Lip 0 X f n Lip 0 X f_(n)inLip_(0)Xf_{n} \in \operatorname{Lip}_{0} XfnLip0X for n = 1 , 2 , 3 , n = 1 , 2 , 3 , n=1,2,3,dotsn=1,2,3, \ldotsn=1,2,3,.
Let a n = d ( x 0 , y n ) d ( x 0 , x n ) a n = d x 0 , y n d x 0 , x n a_(n)=d(x_(0),y_(n))-d(x_(0),x_(n))a_{n}=d\left(x_{0}, y_{n}\right)-d\left(x_{0}, x_{n}\right)an=d(x0,yn)d(x0,xn), and suppose that the set I = { n N I = { n N I={n in NI=\{n \in NI={nN : a n 0 } a n 0 {:a_(n) <= 0}\left.a_{n} \leqslant 0\right\}an0} is infinite, say I = { n j : j N } I = n j : j N I={n_(j):j in N}I=\left\{n_{j}: j \in N\right\}I={nj:jN}. Then, we have f n j ( x 0 ) = 0 , f n j ( x n j ) < 0 f n j x 0 = 0 , f n j x n j < 0 f_(n_(j))(x_(0))=0,f_(n_(j))(x_(n_(j))) < 0f_{n_{j}}\left(x_{0}\right)=0, f_{n_{j}}\left(x_{n_{j}}\right)<0fnj(x0)=0,fnj(xnj)<0, f n j ( y n j ) 0 , j = 1 , 2 , 3 , f n j y n j 0 , j = 1 , 2 , 3 , f_(n_(j))(y_(n_(j))) >= 0,j=1,2,3,dotsf_{n_{j}}\left(y_{n_{j}}\right) \geqslant 0, j=1,2,3, \ldotsfnj(ynj)0,j=1,2,3,. Now, we consider the sequence { ψ j } ψ j {psi_(j)}\left\{\psi_{j}\right\}{ψj} of functions ψ i : f n j ( X ) [ 0 , 1 ] ψ i : f n j ( X ) [ 0 , 1 ] psi_(i):f_(n_(j))(X)rarr[0,1]\psi_{i}: f_{n_{j}}(X) \rightarrow[0,1]ψi:fnj(X)[0,1] defined by
ψ j ( t ) = 1 , t < f n j ( x n j ) , = t / f n j ( x n j ) , f n j ( x n j ) t < 0 = f n j ( x 0 ) , = 0 , t 0 , ψ j ( t ) = 1 , t < f n j x n j , = t / f n j x n j , f n j x n j t < 0 = f n j x 0 , = 0 , t 0 , {:[psi_(j)(t)=1","t < f_(n_(j))(x_(n_(j)))","],[=t//f_(n_(j))(x_(n_(j)))","f_(n_(j))(x_(n_(j))) <= t < 0=f_(n_(j))(x_(0))","],[=0","t >= 0","]:}\begin{aligned} \psi_{j}(t) & =1, & & t<f_{n_{j}}\left(x_{n_{j}}\right), \\ & =t / f_{n_{j}}\left(x_{n_{j}}\right), & & f_{n_{j}}\left(x_{n_{j}}\right) \leqslant t<0=f_{n_{j}}\left(x_{0}\right), \\ & =0, & & t \geqslant 0, \end{aligned}ψj(t)=1,t<fnj(xnj),=t/fnj(xnj),fnj(xnj)t<0=fnj(x0),=0,t0,
for j = 1 , 2 , 3 , j = 1 , 2 , 3 , j=1,2,3,dotsj=1,2,3, \ldotsj=1,2,3,. Putting q j = ψ j f n j q j = ψ j f n j q_(j)=psi_(j)@f_(n_(j))q_{j}=\psi_{j} \circ f_{n_{j}}qj=ψjfnj, we have
q j Y | ψ j ( f n j ( x n j ) ) ψ j ( f n j ( y n j ) ) | / d ( x n j , y n j ) n j . q j Y ψ j f n j x n j ψ j f n j y n j / d x n j , y n j n j . ||q_(j)||_(Y) >= |psi_(j)(f_(n_(j))(x_(n_(j))))-psi_(j)(f_(n_(j))(y_(n_(j))))|//d(x_(n_(j)),y_(n_(j))) >= n_(j).\left\|q_{j}\right\|_{Y} \geqslant\left|\psi_{j}\left(f_{n_{j}}\left(x_{n_{j}}\right)\right)-\psi_{j}\left(f_{n_{j}}\left(y_{n_{j}}\right)\right)\right| / d\left(x_{n_{j}}, y_{n_{j}}\right) \geqslant n_{j} .qjY|ψj(fnj(xnj))ψj(fnj(ynj))|/d(xnj,ynj)nj.
By [5, Corollary 2] it follows that
d ( x , Y ) ( sup { ψ j ( f n j ( y ) ) : y Y } inf { ψ j ( f n j ( y ) ) : y Y } ) / ( 2 q j Y ) = 1 / ( 2 q j Y ) 1 / n j 0 d ( x , Y ) sup ψ j f n j ( y ) : y Y inf ψ j f n j ( y ) : y Y / 2 q j Y = 1 / 2 q j Y 1 / n j 0 {:[d(x","Y) <= (s u p{psi_(j)(f_(n_(j))(y)):y in Y}-i n f{psi_(j)(f_(n_(j))(y)):y in Y})//(2||q_(j)||_(Y))],[=1//(2||q_(j)||_(Y)) <= 1//n_(j)rarr0]:}\begin{aligned} d(x, Y) & \leqslant\left(\sup \left\{\psi_{j}\left(f_{n_{j}}(y)\right): y \in Y\right\}-\inf \left\{\psi_{j}\left(f_{n_{j}}(y)\right): y \in Y\right\}\right) /\left(2\left\|q_{j}\right\|_{Y}\right) \\ & =1 /\left(2\left\|q_{j}\right\|_{Y}\right) \leqslant 1 / n_{j} \rightarrow 0 \end{aligned}d(x,Y)(sup{ψj(fnj(y)):yY}inf{ψj(fnj(y)):yY})/(2qjY)=1/(2qjY)1/nj0
so that x Y ¯ x Y ¯ x in bar(Y)x \in \bar{Y}xY¯, for all x X x X x in Xx \in XxX, that is Y ¯ = X Y ¯ = X bar(Y)=X\bar{Y}=XY¯=X.
By Theorems 1 and 2, we have
Corollary 1. Suppose that Y Y YYY is nonuniformly discrete. Then Y Y Y^(_|_)Y^{\perp}Y is chebyshevian in Lip 0 X Lip 0 X Lip_(0)X\operatorname{Lip}_{0} XLip0X if and only if Y = { 0 } Y = { 0 } Y^(_|_)={0}Y^{\perp}=\{0\}Y={0}.
We can also prove the following result.
Theorem 3. Let X , x 0 X , x 0 X,x_(0)X, x_{0}X,x0, and Y Y YYY be as in Theorem 1. If ( Y ) 1 Y 1 (Y^(_|_))^(1)\left(Y^{\perp}\right)^{1}(Y)1 has the property ( U U UUU ) then every f Lip 0 Y f Lip 0 Y f inLip_(0)Yf \in \operatorname{Lip}_{0} YfLip0Y has a unique Lipschitz extension F Lip 0 X F Lip 0 X F inLip_(0)XF \in \operatorname{Lip}_{0} XFLip0X.
Proof. Follows from [8, Corollary 3.1.b)] and the above Theorem 1.
Corollary 2. Let X , x 0 X , x 0 X,x_(0)X, x_{0}X,x0, and Y Y YYY be as in Theorem 1. Suppose that Y Y YYY is nonuniformly discrete. If ( Y ) 1 Y 1 (Y^(_|_))^(1)\left(Y^{\perp}\right)^{\mathbb{1}}(Y)1 has the property ( U U UUU ), then Y ¯ = X Y ¯ = X bar(Y)=X\bar{Y}=XY¯=X (or equivalently Y = { 0 } Y = { 0 } Y^(_|_)={0}Y^{\perp}=\{0\}Y={0} ).

3. Examples

(a) Let X = [ 0 , 1 ] , d ( x , y ) = | x y | , x 0 = 0 X = [ 0 , 1 ] , d ( x , y ) = | x y | , x 0 = 0 X=[0,1],d(x,y)=|x-y|,x_(0)=0X=[0,1], d(x, y)=|x-y|, x_{0}=0X=[0,1],d(x,y)=|xy|,x0=0, and Y = { 0 , 1 } Y = { 0 , 1 } Y={0,1}Y=\{0,1\}Y={0,1}. Then every f Lip 0 { 0 , 1 } f Lip 0 { 0 , 1 } f inLip_(0){0,1}f \in \operatorname{Lip}_{0}\{0,1\}fLip0{0,1} has a unique Lipschitz extension F Lip 0 [ 0 , 1 ] F Lip 0 [ 0 , 1 ] F inLip_(0)[0,1]F \in \operatorname{Lip}_{0}[0,1]FLip0[0,1], namely, F ( x ) = f ( 1 ) x F ( x ) = f ( 1 ) x F(x)=f(1)xF(x)=f(1) xF(x)=f(1)x. This example shows that the supposition that Y Y YYY is nonuniformly discrete is essential in Theorem 2.
(b) Let f Lip 0 [ 0 , 1 ] f Lip 0 [ 0 , 1 ] f inLip_(0)[0,1]f \in \operatorname{Lip}_{0}[0,1]fLip0[0,1] and let Y Y YYY be the set of points 0 = x 0 < x 1 < < x n + 1 = 1 0 = x 0 < x 1 < < x n + 1 = 1 0=x_(0) < x_(1) < cdots < x_(n+1)=10=x_{0}<x_{1}<\cdots <x_{n+1}=10=x0<x1<<xn+1=1. Then, we have:
Theorem 4. The following conditions are equivalent:
( 1 ) Y 1 Y (1^(@))quadY^(_|_)\left(1^{\circ}\right) \quad Y^{\perp}(1)Y is f-chebyshevian,
(2) f Y = [ x k , x i + 1 ; f ] , k = 0 , 1 , 2 , , n f Y = x k , x i + 1 ; f , k = 0 , 1 , 2 , , n quad||f||_(Y)=||[x_(k),x_(i+1);f]||,k=0,1,2,dots,n\quad\|f\|_{Y}=\left\|\left[x_{k}, x_{i+1} ; f\right]\right\|, k=0,1,2, \ldots, nfY=[xk,xi+1;f],k=0,1,2,,n,
where [ x k , x k + 1 ; f ] = ( f ( x k + 1 ) f ( x k ) ) / ( x k + 1 x k ) x k , x k + 1 ; f = f x k + 1 f x k / x k + 1 x k [x_(k),x_(k+1);f]=(f(x_(k+1))-f(x_(k)))//(x_(k+1)-x_(k))\left[x_{k}, x_{k+1} ; f\right]=\left(f\left(x_{k+1}\right)-f\left(x_{k}\right)\right) /\left(x_{k+1}-x_{k}\right)[xk,xk+1;f]=(f(xk+1)f(xk))/(xk+1xk).
Proof. ( 1 ) ( 2 ) 1 2 quad(1^(@))=>(2^(@))\quad\left(1^{\circ}\right) \Rightarrow\left(2^{\circ}\right)(1)(2) Obviously,
L ( x ) = [ x k , x k + 1 ; f ] ( x x k ) + f ( x k ) , x ( x k , x k + 1 ) , k = 0 , 1 , , n L ( x ) = x k , x k + 1 ; f x x k + f x k , x x k , x k + 1 , k = 0 , 1 , , n L(x)=[x_(k),x_(k+1);f](x-x_(k))+f(x_(k)),quad x in(x_(k),x_(k+1)),k=0,1,dots,nL(x)=\left[x_{k}, x_{k+1} ; f\right]\left(x-x_{k}\right)+f\left(x_{k}\right), \quad x \in\left(x_{k}, x_{k+1}\right), k=0,1, \ldots, nL(x)=[xk,xk+1;f](xxk)+f(xk),x(xk,xk+1),k=0,1,,n,
is a Lipschitz extension of f | Y f Y f|_(Y)\left.f\right|_{Y}f|Y and f | Y [ x k , x k + 1 ; f ] | i k = 0 , 1 5 , , n f Y x k , x k + 1 ; f i k = 0 , 1 5 , , n ||f|_(Y) >= [x_(k),x_(k+1);f]|_(i)k=0,1_(5),dots,n\|\left. f\right|_{Y} \geqslant\left.\left[x_{k}, x_{k+1} ; f\right]\right|_{i} k=0,1_{5}, \ldots, nf|Y[xk,xk+1;f]|ik=0,15,,n. Suppose that k 0 , 0 k 0 < n k 0 , 0 k 0 < n k_(0),0 <= k_(0) < nk_{0}, 0 \leqslant k_{0}<nk0,0k0<n is such that f Y >∣ [ x h 0 , x I 0 + 1 ; f ] f Y >∣ x h 0 , x I 0 + 1 ; f ||f||_(Y)>∣[x_(h_(0)),x_(I_(0)+1);f]\|f\|_{Y}>\mid\left[x_{h_{0}}, x_{I_{0}+1} ; f\right]fY>∣[xh0,xI0+1;f]. We have to consider the following cases:
(i) f ( x k 0 ) < f ( x k 0 + 1 ) f x k 0 < f x k 0 + 1 f(x_(k_(0))) < f(x_(k_(0)+1))f\left(x_{k_{0}}\right)<f\left(x_{k_{0}+1}\right)f(xk0)<f(xk0+1),
(ii) f ( x k 0 ) > f ( x k 0 + 1 ) f x k 0 > f x k 0 + 1 f(x_(k_(0))) > f(x_(k_(0)+1))f\left(x_{k_{0}}\right)>f\left(x_{k_{0}+1}\right)f(xk0)>f(xk0+1),
(iii) f ( x k 0 ) = f ( x k 0 + 1 ) f x k 0 = f x k 0 + 1 f(x_(k_(0)))=f(x_(k_(0)+1))f\left(x_{k_{0}}\right)=f\left(x_{k_{0}+1}\right)f(xk0)=f(xk0+1).
If condition (i) holds, put z 1 = x k 0 + ( f ( x k 0 + 1 ) f ( x k 0 ) ) / f r z 1 = x k 0 + f x k 0 + 1 f x k 0 / f r z_(1)=x_(k_(0))+(f(x_(k_(0)+1))-f(x_(k_(0))))//||f||_(r)z_{1}=x_{k_{0}}+\left(f\left(x_{k_{0}+1}\right)-f\left(x_{k_{0}}\right)\right) /\|f\|_{r}z1=xk0+(f(xk0+1)f(xk0))/fr and define the function F 1 : [ 0 , 1 ] R F 1 : [ 0 , 1 ] R F_(1):[0,1]rarr RF_{1}:[0,1] \rightarrow RF1:[0,1]R by
F 1 ( x ) = L ( x ) , x [ 0 , 1 ] [ x k 0 , x k 0 1 ] , (3.2) = f ( x k 0 ) f Y ( x x k 0 ) , x ( x k 1 , z 1 ) , = f ( x k 0 ) , x [ z 1 , x k 0 + 1 ) . F 1 ( x ) = L ( x ) , x [ 0 , 1 ] x k 0 , x k 0 1 , (3.2) = f x k 0 f Y x x k 0 , x x k 1 , z 1 , = f x k 0 , x z 1 , x k 0 + 1 . {:[F_(1)(x)=L(x)","x in[0","1]-[x_(k_(0)),x_(k_(0)-1)]","],[(3.2)=f(x_(k_(0)))-∣f||_(Y)(x-x_(k_(0)))","x in(x_(k_(1)),z_(1))","],[=f(x_(k_(0)))","x in[z_(1),x_(k_(0)+1)).]:}\begin{align*} F_{1}(x) & =L(x), & & x \in[0,1]-\left[x_{k_{0}}, x_{k_{0}-1}\right], \\ & =f\left(x_{k_{0}}\right)-\mid f \|_{Y}\left(x-x_{k_{0}}\right), & & x \in\left(x_{k_{1}}, z_{1}\right), \tag{3.2}\\ & =f\left(x_{k_{0}}\right), & & x \in\left[z_{1}, x_{k_{0}+1}\right) . \end{align*}F1(x)=L(x),x[0,1][xk0,xk01],(3.2)=f(xk0)fY(xxk0),x(xk1,z1),=f(xk0),x[z1,xk0+1).
It is easy to see that F 1 F 1 F_(1)F_{1}F1 is a Lipschitz extension of f | Y f Y f|_(Y)\left.f\right|_{Y}f|Y, distinct from L , L , L_(,)L_{,}L, and then, by Theorem 1 , Y 1 , Y 1,Y^(_|_)1, Y^{\perp}1,Y is not f f fff-chebyshevian.
In case (ii) the proof proceeds similarly. If condition (iii) holds, put z 2.2 = ( 2 x k 11 + x k 0 + 1 ) / 3 z 2.2 = 2 x k 11 + x k 0 + 1 / 3 z_(2.2)=(2x_(k_(11))+x_(k_(0)+1))//3z_{2.2}=\left(2 x_{k_{11}}+x_{k_{0}+1}\right) / 3z2.2=(2xk11+xk0+1)/3 and define
F 2 ( x ) = L ( x ) , x [ 0 , 1 ] [ x k 0 , x k 0 + 1 ] , (3.3) = f ( x k 0 ) + :∣ f Y ( x x k 0 ) , x ( x k 0 , z 2 ] , = f ( x k 0 + 1 ) f Y Y ( x x k 0 + 1 ) , x ( z 2 , x k 0 1 ) , F 2 ( x ) = L ( x ) , x [ 0 , 1 ] x k 0 , x k 0 + 1 , (3.3) = f x k 0 + :∣ f Y x x k 0 , x x k 0 , z 2 , = f x k 0 + 1 f Y Y x x k 0 + 1 , x z 2 , x k 0 1 , {:[F_(2)(x)=L(x)","x in[0","1]-[x_(k_(0)),x_(k_(0)+1)]","],[(3.3)=f(x_(k_(0)))+:∣f||_(Y)(x-x_(k_(0)))","x in(x_(k_(0)),z_(2)]","],[=f(x_(k_(0)+1))-||f_(Y)||_(Y)(x-x_(k_(0)+1))","x in(z_(2),x_(k_(0)-1))","]:}\begin{align*} F_{2}(x) & =L(x), & & x \in[0,1]-\left[x_{k_{0}}, x_{k_{0}+1}\right], \\ & =f\left(x_{k_{0}}\right)+: \mid f \|_{Y}\left(x-x_{k_{0}}\right), & & x \in\left(x_{k_{0}}, z_{2}\right], \tag{3.3}\\ & =f\left(x_{k_{0}+1}\right)-\left\|f_{Y}\right\|_{Y}\left(x-x_{k_{0}+1}\right), & & x \in\left(z_{2}, x_{k_{0}-1}\right), \end{align*}F2(x)=L(x),x[0,1][xk0,xk0+1],(3.3)=f(xk0)+:∣fY(xxk0),x(xk0,z2],=f(xk0+1)fYY(xxk0+1),x(z2,xk01),
Then F 2 F 2 F_(2)F_{2}F2 is a Lipschitz extension of f | Y f Y f|_(Y)\left.f\right|_{Y}f|Y, different from L L LLL. By Theorem 1, Y Y Y^(_|_)Y^{\perp}Y is not f f fff-chebyshevian.
( 2 ) ( 1 ) 2 1 (2^(@))=>(1^(@))\left(2^{\circ}\right) \Rightarrow\left(1^{\circ}\right)(2)(1) If | [ x k , x k + 1 ; f ] | = f Y x k , x k + 1 ; f = f Y |[x_(k),x_(k+1);f]|=||f||_(Y)\left|\left[x_{k}, x_{k+1} ; f\right]\right|=\|f\|_{Y}|[xk,xk+1;f]|=fY for k = 0 , 1 , 2 , , n k = 0 , 1 , 2 , , n k=0,1,2,dots,nk=0,1,2, \ldots, nk=0,1,2,,n, then the function L L LLL defined by (3.1) is the only Lipschitz extension of f | Y f Y f|_(Y)\left.f\right|_{Y}f|Y.
A consequence of Theorem 4 is:
Corollary 3. Let Y Y YYY be the set of points 0 = x 0 < x 1 < < x n + 1 = 1 0 = x 0 < x 1 < < x n + 1 = 1 0=x_(0) < x_(1) < cdots < x_(n+1)=10=x_{0}<x_{1}<\cdots<x_{n+1}=10=x0<x1<<xn+1=1, f Lip 0 [ 0 , 1 ] f Lip 0 [ 0 , 1 ] f inLip_(0)[0,1]f \in \operatorname{Lip}_{0}[0,1]fLip0[0,1] and
(3.4) K = { h : h Lip 0 [ 0 , 1 ] , h ( x k ) = f ( x k ) , k = 0 , 1 , 2 , , n + 1 } . (3.4) K = h : h Lip 0 [ 0 , 1 ] , h x k = f x k , k = 0 , 1 , 2 , , n + 1 . {:(3.4)K={h:h inLip_(0)[0,1],h(x_(k))=f(x_(k)),k=0,1,2,dots,n+1}.:}\begin{equation*} K=\left\{h: h \in \operatorname{Lip}_{0}[0,1], h\left(x_{k}\right)=f\left(x_{k}\right), k=0,1,2, \ldots, n+1\right\} . \tag{3.4} \end{equation*}(3.4)K={h:hLip0[0,1],h(xk)=f(xk),k=0,1,2,,n+1}.
Then Y Y Y^(_|_)Y^{\perp}Y is K K KKK-chebyshevian if and only if Y Y Y^(-)Y^{-}Yis f f fff-chebyshevian.
(c) Let C 1 [ 0 , 1 ] C 1 [ 0 , 1 ] C^(1)[0,1]C^{1}[0,1]C1[0,1] be the space of all continuously differentiable functions on [ 0,1 ] and let Y Y YYY be the set of points 0 = x 0 < x 1 < < x n + 1 = 1 0 = x 0 < x 1 < < x n + 1 = 1 0=x_(0) < x_(1) < cdots < x_(n+1)=10=x_{0}<x_{1}<\cdots<x_{n+1}=10=x0<x1<<xn+1=1. Put
(3.5) Z = C 1 [ 0 , 1 ] Lip 0 [ 0 , 1 ] , W = C 1 [ 0 , 1 ] Y . (3.5) Z = C 1 [ 0 , 1 ] Lip 0 [ 0 , 1 ] , W = C 1 [ 0 , 1 ] Y . {:(3.5)Z=C^(1)[0","1]nnLip_(0)[0","1]","quad W=C^(1)[0","1]nnY^(_|_).:}\begin{equation*} Z=C^{1}[0,1] \cap \operatorname{Lip}_{0}[0,1], \quad W=C^{1}[0,1] \cap Y^{\perp} . \tag{3.5} \end{equation*}(3.5)Z=C1[0,1]Lip0[0,1],W=C1[0,1]Y.
For f Z f Z f in Zf \in ZfZ, we have
(3.6) f [ 0 , 1 ] = max { | f ( x ) | : x [ 0 , 1 ] } . (3.6) f [ 0 , 1 ] = max f ( x ) : x [ 0 , 1 ] . {:(3.6)||f||_([0,1])=max{|f^(')(x)|:x in[0,1]}.:}\begin{equation*} \|f\|_{[0,1]}=\max \left\{\left|f^{\prime}(x)\right|: x \in[0,1]\right\} . \tag{3.6} \end{equation*}(3.6)f[0,1]=max{|f(x)|:x[0,1]}.
Let us define the function set S S SSS by
S = { h : h Z , [ x k , x k + 1 ; h ] [ x k + 1 , x k + 2 ; h ] (3.7) h Y 2 , k = 0 , 1 , 2 , , n 1 } . S = h : h Z , x k , x k + 1 ; h x k + 1 , x k + 2 ; h (3.7) h Y 2 , k = 0 , 1 , 2 , , n 1 . {:[S={h:h in Z,[x_(k),x_(k+1);h][x_(k+1),x_(k+2);h]:}],[(3.7){:!=-||h||_(Y)^(2),k=0,1,2,dots,n-1}.]:}\begin{align*} S & =\left\{h: h \in Z,\left[x_{k}, x_{k+1} ; h\right]\left[x_{k+1}, x_{k+2} ; h\right]\right. \\ & \left.\neq-\|h\|_{Y}^{2}, k=0,1,2, \ldots, n-1\right\} . \tag{3.7} \end{align*}S={h:hZ,[xk,xk+1;h][xk+1,xk+2;h](3.7)hY2,k=0,1,2,,n1}.
We need the following two lemmas:
Lemma 2. Let [ p , q ] R , f ( x ) = a x + b , a , b R , a > 0 [ p , q ] R , f ( x ) = a x + b , a , b R , a > 0 [p,q]sub R,f(x)=ax+b,a,b in R,a > 0[p, q] \subset R, f(x)=a x+b, a, b \in R, a>0[p,q]R,f(x)=ax+b,a,bR,a>0, and M > a M > a M > aM>aM>a. Then there exists a function g C 1 [ p , q ] g C 1 [ p , q ] g inC^(1)[p,q]g \in C^{1}[p, q]gC1[p,q] such that f ( p ) = g ( p ) , f ( q ) = g ( q ) f ( p ) = g ( p ) , f ( q ) = g ( q ) f(p)=g(p),f(q)=g(q)f(p)=g(p), f(q)=g(q)f(p)=g(p),f(q)=g(q), f ( p ) = M ( f ( q ) = M ) , f ( q ) = g ( q ) ( f ( p ) = g ( p ) ) f ( p ) = M f ( q ) = M , f ( q ) = g ( q ) f ( p ) = g ( p ) f^(')(p)=M(f^(')(q)=M),f^(')(q)=g^(')(q)(f^(')(p)=g^(')(p))f^{\prime}(p)=M\left(f^{\prime}(q)=M\right), f^{\prime}(q)=g^{\prime}(q)\left(f^{\prime}(p)=g^{\prime}(p)\right)f(p)=M(f(q)=M),f(q)=g(q)(f(p)=g(p)) and max { | g ( x ) | : x [ p , q ] } = M max g ( x ) : x [ p , q ] } = M max{|g^(')(x)|::}x in[p,q]}=M\max \left\{\left|g^{\prime}(x)\right|:\right. x \in[p, q]\}=Mmax{|g(x)|:x[p,q]}=M.
Proof. The proof of the lemma is obvious from Fig. 1:
Figure 1
( AC ) s 1 ( x ) = f ( p ) + M ( x p ) , ( BC ) s 2 ( x ) = f ( q ) M ( x q ) , ( DE ) s 3 ( x ) = f ( r ) M ( x r ) , r ( p , q ) . ( AC )      s 1 ( x ) = f ( p ) + M ( x p ) , ( BC )      s 2 ( x ) = f ( q ) M ( x q ) , ( DE )      s 3 ( x ) = f ( r ) M ( x r ) , r ( p , q ) . {:[(AC),s_(1)(x)=f(p)+M(x-p)","],[(BC),s_(2)(x)=f(q)-M(x-q)","],[(DE),s_(3)(x)=f(r)-M(x-r)","quad r in(p","q).]:}\begin{array}{ll} (\mathrm{AC}) & s_{1}(x)=f(p)+M(x-p), \\ (\mathrm{BC}) & s_{2}(x)=f(q)-M(x-q), \\ (\mathrm{DE}) & s_{3}(x)=f(r)-M(x-r), \quad r \in(p, q) . \end{array}(AC)s1(x)=f(p)+M(xp),(BC)s2(x)=f(q)M(xq),(DE)s3(x)=f(r)M(xr),r(p,q).
Lemma 3. If h S h S h in Sh \in ShS, then h | Y h Y h|_(Y)\left.h\right|_{Y}h|Y has at least one Lipschitz extension H Z H Z H in ZH \in ZHZ.
Proof. Let h S h S h in Sh \in ShS and k 0 N , 0 k 0 < n + 1 k 0 N , 0 k 0 < n + 1 k_(0)in N,0 <= k_(0) < n+1k_{0} \in N, 0 \leqslant k_{0}<n+1k0N,0k0<n+1, such that
(3.8) [ x k 0 , x k 0 + 1 ; h ] = 1 h Y (3.8) x k 0 , x k 0 + 1 ; h = 1 h Y {:(3.8)[x_(k_(0)),x_(k_(0)+1);h]=||_(1)h||_(Y):}\begin{equation*} \left[x_{k_{0}}, x_{k_{0}+1} ; h\right]=\left\|_{1} h\right\|_{Y} \tag{3.8} \end{equation*}(3.8)[xk0,xk0+1;h]=1hY
By the definition of S S SSS, we have
; h Y < [ x k 0 1 , x k 0 ; h ] | h | Y . h Y < [ x h 0 + 1 , x k 0 + 2 ; h ] h | Y Y . ; h Y < x k 0 1 , x k 0 ; h | h | Y . h Y < x h 0 + 1 , x k 0 + 2 ; h h Y Y . {:[-;h||_(Y) < [x_(k_(0)-1),x_(k_(0));h] <= |h|_(Y)],[-.h||_(Y) < [x_(h_(0)+1),x_(k_(0)+2);h] <= ||h|_(Y_(Y)).]:}\begin{aligned} & -; h \|_{\mathrm{Y}}<\left[x_{k_{0}-1}, x_{k_{0}} ; h\right] \leqslant|h|{ }_{\mathrm{Y}} \\ & -.\left.h\left\|_{\mathrm{Y}}<\left[x_{h_{0}+1}, x_{k_{0}+2} ; h\right] \leqslant\right\| h\right|_{\mathrm{Y}_{Y}} . \end{aligned};hY<[xk01,xk0;h]|h|Y.hY<[xh0+1,xk0+2;h]h|YY.
Applying Lemma 2 to the intervals [ x k 0 1 , x k 6 ] x k 0 1 , x k 6 [x_(k_(0)-1),x_(k_(6))]\left[x_{k_{0}-1}, x_{k_{6}}\right][xk01,xk6] and [ x k 0 + 1 , x k 0 + 2 ] x k 0 + 1 , x k 0 + 2 [x_(k_(0)+1),x_(k_(0)+2)]\left[x_{k_{0}+1}, x_{k_{0}+2}\right][xk0+1,xk0+2], twice it follows that there exists a function H 1 H 1 H_(1)H_{1}H1 in C 1 [ x k 0 1 , x k 0 + 2 ] C 1 x k 0 1 , x k 0 + 2 C^(1)[x_(k_(0)-1),x_(k_(0)+2)]C^{1}\left[x_{k_{0}-1}, x_{k_{0}+2}\right]C1[xk01,xk0+2] such that max { | H 1 ( x ) | : x [ x k 0 1 , x k 0 + 2 ] } = h Y max H 1 ( x ) : x x k 0 1 , x k 0 + 2 = h Y max{|H_(1)^(')(x)|:x in[x_(k_(0)-1),x_(k_(0)+2)]}=||h||_(Y)\max \left\{\left|H_{1}{ }^{\prime}(x)\right|: x \in\left[x_{k_{0}-1}, x_{k_{0}+2}\right]\right\}=\|h\|_{Y}max{|H1(x)|:x[xk01,xk0+2]}=hY and which interpolates the function h h hhh at the points x k 0 1 , x k 0 , x k 0 + 1 , x k 0 + 2 x k 0 1 , x k 0 , x k 0 + 1 , x k 0 + 2 x_(k_(0)-1),x_(k_(0)),x_(k_(0)+1),x_(k_(0)+2)x_{k_{0}-1}, x_{k_{0}}, x_{k_{0}+1}, x_{k_{0}+2}xk01,xk0,xk0+1,xk0+2.
Appiying Lemma 2 to the intervals [ x i , x i + 1 ] , i = 0 , 1 , , k 0 2 x i , x i + 1 , i = 0 , 1 , , k 0 2 [x_(i),x_(i+1)],i=0,1,dots,k_(0)-2\left[x_{i}, x_{i+1}\right], i=0,1, \ldots, k_{0}-2[xi,xi+1],i=0,1,,k02, k 0 + 2 , , n k 0 + 2 , , n k_(0)+2,dots,nk_{0}+2, \ldots, nk0+2,,n, we get a function H Z H Z H in ZH \in ZHZ, which is a Lipschitz extension of h v h v h^(')vh^{\prime} vhv to [ 0 , 1 ] [ 0 , 1 ] [0,1][0,1][0,1].
If [ x k 0 , x k 0 + 1 ; h ] = h Y x k 0 , x k 0 + 1 ; h = h Y [x_(k_(0)),x_(k_(0)+1);h]=-||h||_(Y)\left[x_{k_{0}}, x_{k_{0}+1} ; h\right]=-\|h\|_{Y}[xk0,xk0+1;h]=hY we can proceed analogously.
Theorem 5. The subspace W W WWW is S S SSS proximinal and for each h S h S h in Sh \in ShS the following equality holds:
(3.9) d ( h , W ) = d ( h , Y ) (3.9) d ( h , W ) = d h , Y {:(3.9)d(h","W)=d(h,Y^(_|_)):}\begin{equation*} d(h, W)=d\left(h, Y^{\perp}\right) \tag{3.9} \end{equation*}(3.9)d(h,W)=d(h,Y)
Proof. Let h S h S h in Sh \in ShS. By Lemma 3, h | Y h Y h|_(Y)\left.h\right|_{Y}h|Y has a Lipschitz extension H Z H Z H in ZH \in ZHZ. Then, h H W h H W h-H in Wh-H \in WhHW, and this is a best approximation to h h hhh, from Y Y Y^(_|_)Y^{\perp}Y.
But then
d ( h , Y ) d ( h , W ) h ( h H ) | X = d ( h , Y ) d h , Y d ( h , W ) h ( h H ) X = d h , Y d(h,Y^(_|_)) <= d(h,W) <= ||h-(h-H)|_(X)=d(h,Y^(-))d\left(h, Y^{\perp}\right) \leqslant d(h, W) \leqslant \| h-\left.(h-H)\right|_{X}=d\left(h, Y^{-}\right)d(h,Y)d(h,W)h(hH)|X=d(h,Y)
so that
h ( h H ) X = d ( h , W ) = d ( h , Y ) h ( h H ) X = d ( h , W ) = d h , Y ||h-(h-H)||_(X)=d(h,W)=d(h,Y^(_|_))\|h-(h-H)\|_{X}=d(h, W)=d\left(h, Y^{\perp}\right)h(hH)X=d(h,W)=d(h,Y)
Remark 1. Let f Z S f Z S f in Z-Sf \in Z-SfZS; that is, there exists 0 k 1 < n + 1 0 k 1 < n + 1 0 <= k_(1) < n+10 \leqslant k_{1}<n+10k1<n+1 such that
[ x k 1 1 , x k 1 ; f ] [ x k 1 , x k 1 + 1 ; f ] = | f | Y 2 x k 1 1 , x k 1 ; f x k 1 , x k 1 + 1 ; f = | f | Y 2 [x_(k_(1)-1),x_(k_(1));f][x_(k_(1)),x_(k_(1)+1);f]=-|f|_(Y)^(2)\left[x_{k_{1}-1}, x_{k_{1}} ; f\right]\left[x_{k_{1}}, x_{k_{1}+1} ; f\right]=-|f|_{Y}^{2}[xk11,xk1;f][xk1,xk1+1;f]=|f|Y2
In this case, it is possible that no Lipschitz extension to f f fff exists in Z Z ZZZ : e.g.for f ( x ) = 4 x 2 + 4 x , Y = [ 0 , 1 2 , 1 } f ( x ) = 4 x 2 + 4 x , Y = 0 , 1 2 , 1 f(x)=-4x^(2)+4x,Y=[0,(1)/(2),1}f(x)=-4 x^{2}+4 x, Y=\left[0, \frac{1}{2}, 1\right\}f(x)=4x2+4x,Y=[0,12,1} we have
[ 0 , 1 2 ; f ] [ 1 2 , 1 ; f ] = 4 0 , 1 2 ; f 1 2 , 1 ; f = 4 [0,(1)/(2);f][(1)/(2),1;f]=-4\left[0, \frac{1}{2} ; f\right]\left[\frac{1}{2}, 1 ; f\right]=-4[0,12;f][12,1;f]=4
and the only Lipschitz extension of f | Y f Y f|_(Y)\left.f\right|_{Y}f|Y is
F ( x ) = 2 x , x [ 0 , 1 2 ) , = 2 ( x 1 ) , x [ 1 2 , 1 ] , F ( x ) = 2 x , x 0 , 1 2 , = 2 ( x 1 ) , x 1 2 , 1 , {:[F(x)=2x","x in[0,(1)/(2))","],[=-2(x-1)","x in[(1)/(2),1]","]:}\begin{aligned} F(x) & =2 x, & & x \in\left[0, \frac{1}{2}\right), \\ & =-2(x-1), & & x \in\left[\frac{1}{2}, 1\right], \end{aligned}F(x)=2x,x[0,12),=2(x1),x[12,1],
which, obviously, does not belong to Z Z ZZZ.
By Lemmas 2 and 3, every h S h S h in Sh \in ShS has a best approximation in W W WWW, namely, h H h H h-Hh-HhH, where H H HHH is a Lipschitz extension of h h hhh, such that H Z H Z H in ZH \in ZHZ. We can show that every best approximation is of this form (Lemma 1). It follows that W W WWW is chebyshevian for h S h S h in Sh \in ShS if and only if h | Y h Y h|_(Y)\left.h\right|_{Y}h|Y has a unique Lipschitz extension in Z Z ZZZ. A class of such functions is given by
(3.11) S 1 = { h : h S , h ( x k ) = h ( 1 ) x k , k = 0 , 1 , 2 , , n + 1 } . (3.11) S 1 = h : h S , h x k = h ( 1 ) x k , k = 0 , 1 , 2 , , n + 1 . {:(3.11)S_(1)={h:h in S,h(x_(k))=h(1)x_(k),k=0,1,2,dots,n+1}.:}\begin{equation*} S_{1}=\left\{h: h \in S, h\left(x_{k}\right)=h(1) x_{k}, k=0,1,2, \ldots, n+1\right\} . \tag{3.11} \end{equation*}(3.11)S1={h:hS,h(xk)=h(1)xk,k=0,1,2,,n+1}.
Theorem 6. W W WWW is S 1 S 1 S_(1)S_{1}S1-chebyshevian.
Proof. If h S 1 h S 1 h inS_(1)h \in S_{1}hS1, then the unique Lipschitz extension of h h hhh in Z Z ZZZ is H ( x ) = h ( 1 ) x H ( x ) = h ( 1 ) x H(x)=h(1)xH(x)=h(1) xH(x)=h(1)x. Therefore h ( x ) h ( 1 ) x h ( x ) h ( 1 ) x h(x)-h(1)xh(x)-h(1) xh(x)h(1)x is the only element of best approximation for h h hhh in W W WWW.
Remark 2. J. Favard and recently de Boor [1] considered a problem analogous to that in Example (c).
(d) Finally, let X X XXX be a metric space of finite diameter (i.e., sup { d ( x , y ) sup { d ( x , y ) s u p{d(x,y)\sup \{d(x, y)sup{d(x,y) : x , y X } < x , y X } < x,y in X} < oox, y \in X\}<\inftyx,yX}< ), x 0 x 0 x_(0)x_{0}x0 a fixed element in X X XXX, and Y Y YYY a subset of X X XXX such that x 0 Y x 0 Y x_(0)in Yx_{0} \in Yx0Y. Let f Lip 0 X f Lip 0 X f inLip_(0)Xf \in \operatorname{Lip}_{0} XfLip0X and let G ( f ) G ( f ) G(f)G(f)G(f) be the set of best approximation to f f fff from Y Y Y^(_|_)Y^{\perp}Y. We can define on Lip 0 X Lip 0 X Lip_(0)X\operatorname{Lip}_{0} XLip0X the uniform norm u : Lip 0 X R u : Lip 0 X R ||*||_(u):Lip_(0)X rarr R\|\cdot\|_{u}: \operatorname{Lip}_{0} X \rightarrow Ru:Lip0XR by
(3.12) f u = sup { | f ( x ) | : x X } , f Lip 0 X . (3.12) f u = sup { | f ( x ) | : x X } , f Lip 0 X . {:(3.12)||f||_(u)=s u p{|f(x)|:x in X}","quad f inLip_(0)X.:}\begin{equation*} \|f\|_{u}=\sup \{|f(x)|: x \in X\}, \quad f \in \operatorname{Lip}_{0} X . \tag{3.12} \end{equation*}(3.12)fu=sup{|f(x)|:xX},fLip0X.
Obviously, the set G ( f ) Y G ( f ) Y G(f)subY^(_|_)G(f) \subset Y^{\perp}G(f)Y is closed, convex, and bounded, for every f Lip 0 X f Lip 0 X f inLip_(0)Xf \in \operatorname{Lip}_{0} XfLip0X. We consider the following problems: Find g , g G ( f ) g , g G ( f ) g_(**),g^(**)in G(f)g_{*}, g^{*} \in G(f)g,gG(f) such that
(3.13) f g u = inf { f g u : g G ( f ) } (3.13) f g u = inf f g u : g G ( f ) {:(3.13)||f-g_(**)||_(u)=i n f{||f-g||_(u):g in G(f)}:}\begin{equation*} \left\|f-g_{*}\right\|_{u}=\inf \left\{\|f-g\|_{u}: g \in G(f)\right\} \tag{3.13} \end{equation*}(3.13)fgu=inf{fgu:gG(f)}
and
(3.14) f g u = sup { f g u : g G ( f ) } ; (3.14) f g u = sup f g u : g G ( f ) ; {:(3.14)||f-g^(**)||_(u)=s u p{||f-g||_(u):g in G(f)};:}\begin{equation*} \left\|f-g^{*}\right\|_{u}=\sup \left\{\|f-g\|_{u}: g \in G(f)\right\} ; \tag{3.14} \end{equation*}(3.14)fgu=sup{fgu:gG(f)};
i.e., find the nearest and the farthest point to f f fff in G ( f ) G ( f ) G(f)G(f)G(f), in the uniform norm.
Since every element in G ( f ) G ( f ) G(f)G(f)G(f) is of the form f F f F f-Ff-FfF, where F F FFF is a Lipschitz extension of f | X f X f|_(X)\left.f\right|_{X}f|X it follows that the problems (3.13) and (3.14) are equivalent to the following problems: Find two Lipschitz extensions F F F_(**)F_{*}F and F F F^(**)F^{*}F of f | Y f Y f|_(Y)\left.f\right|_{Y}f|Y such that
( ) F u = inf { F u : F is a Lipschitz extension of f | Y } ( ) F u = inf F u : F  is a Lipschitz extension of  f Y {:('")"||F_(**)||_(u)=i n f{||F||_(u):F" is a Lipschitz extension of "f|_(Y)}:}\begin{equation*} \left\|F_{*}\right\|_{u}=\inf \left\{\|F\|_{u}: F \text { is a Lipschitz extension of }\left.f\right|_{Y}\right\} \tag{$\prime$} \end{equation*}()Fu=inf{Fu:F is a Lipschitz extension of f|Y}
and
(3.14') F u = sup { F | u : F is a Lipschitz extension of f | Y } . (3.14') F u = sup F u : F  is a Lipschitz extension of  f Y . {:(3.14')||F^(**)||_(u)=s u p{||F|_(u):F" is a Lipschitz extension of "f|_(Y)}.:}\begin{equation*} \left\|F^{*}\right\|_{u}=\sup \left\{\|\left. F\right|_{u}: F \text { is a Lipschitz extension of }\left.f\right|_{Y}\right\} . \tag{3.14'} \end{equation*}(3.14')Fu=sup{F|u:F is a Lipschitz extension of f|Y}.
THEOREM 6. The infimum (3.13) is attained for every g = f F g = f F g_(**)=f-F_(**)g_{*}=f-F_{*}g=fF such that F F F_(**)F_{*}F is a Lipschitz extension of f | Y f Y f|_(Y)\left.f\right|_{Y}f|Y and F = f | Y F = f Y ||F_(**)||_(||)=_(||)f|_(Y)||_(||)\left\|F_{*}\right\|_{\|}=\left.{ }_{\|} f\right|_{Y} \|_{\|}F=f|Y. The set of these extensions is nonempty.
Proof. If F F FFF is a Lipschitz extension of f | Y f Y f|_(Y)\left.f\right|_{Y}f|Y then
| F | u sup { | F ( y ) | : y Y } = sup { | f ( y ) | : y Y } = | f | Y | a F u sup { | F ( y ) | : y Y } = sup | f ( y ) | : y Y = | f | Y a |F^(')|_(u) >= s u p{|F(y)|:y in Y}=s u p{|f(y)|_(:)y in Y}=|f|_(Y)|_(a)\left|F^{\prime}\right|_{u} \geqslant \sup \{|F(y)|: y \in Y\}=\sup \left\{|f(y)|_{:} y \in Y\right\}=\left.|f|_{Y}\right|_{a}|F|usup{|F(y)|:yY}=sup{|f(y)|:yY}=|f|Y|a
Therefore, if F u = f Y u F u = f Y u ||F_(**)||_(u)=||f^(')_(Y)||_(u)\left\|F_{*}\right\|_{u}=\left\|f^{\prime}{ }_{Y}\right\|_{u}Fu=fYu then inf { F u : F inf F u : F i n f{||F||_(u):F:}\inf \left\{\|F\|_{u}: F\right.inf{Fu:F is a Lipschitz extension of f | Y } = { F | u , = f | Y | u f Y = F u , = f Y u {:f|_(Y)}={F_(**)|_(u,)=||f|_(Y)|_(u):}\left.\left.f\right|_{Y}\right\}=\left\{\left.F_{*}\right|_{u,}=\|\left.\left. f\right|_{Y}\right|_{u}\right.f|Y}={F|u,=f|Y|u. Now, if F F FFF is a Lipschitz extension of f | Y f Y f|_(Y)\left.f\right|_{Y}f|Y. we define a new Lipschitz function F F F_(**)F_{*}F by
( { 3.15 } ) F Y ( x ) = f | X u b if F ( x ) > { f | Y u , = F ( x ) if f | Y | u F ( x ) f | Y ! u , = f | Y u b if F ( x ) < | f | Y | u . ( { 3.15 } ) F Y ( x ) = f X u b  if  F ( x ) > f Y u , = F ( x )  if  f Y u F ( x ) f Y ! u , = f Y u b  if  F ( x ) < f | Y | u . {:(({3.15})")"{:[F_(Y)(x),=||f|_(X)||_(ub),," if ",F(x), > {f|_(Y)||_(u),:}],[,=F(x),," if ",-||f|_(Y)|_(u) <= F(x) <= ||f|_(Y)!_(u)","],[,=-||f|_(Y)||_(ub),," if ",F(x) < -|_(f)|Y|_(u).]:}:}\begin{array}{rlrlrl} F_{Y}(x) & =\left\|\left.f\right|_{X}\right\|_{u b} & & \text { if } & F(x) & >\left\{\left.f\right|_{Y} \|_{u},\right. \\ & =F(x) & & \text { if } & -\left.\left\|\left.\left.f\right|_{Y}\right|_{u} \leqslant F(x) \leqslant\right\| f\right|_{Y}!_{u}, \tag{$\{3.15\}$}\\ & =-\left\|\left.f\right|_{Y}\right\|_{u b} & & \text { if } & F(x)<-\left.\right|_{f}|Y|_{u} . \end{array}({3.15})FY(x)=f|Xub if F(x)>{f|Yu,=F(x) if f|Y|uF(x)f|Y!u,=f|Yub if F(x)<|f|Y|u.
It is easy to see that F F F_(**)F_{*}F is a Lipschitz extension of f Y f Y f_(∣Y)f_{\mid Y}fY such that | F x | u = | f u | u F x u = f u u |F_(x)|_(∣u)=|^(')f_(∣u)|_(u)\left|F_{x}\right|_{\mid u}= \left.\left.\right|^{\prime} f_{\mid u}\right|_{u}|Fx|u=|fu|u.
Theorem 7. The supremum (3.14) is attained for f F 1 f F 1 f-F_(1)^(**)f-F_{1}^{*}fF1 or f F 2 f F 2 f-F_(2)^(**)f-F_{2}^{*}fF2 or for both of these functions, where
( { 3.16 } ) F 1 ( x ) = inf { [ f ( y ) + | f | 1 d ( x , y ) ] : y Y } . ( { 3.16 } ) F 1 ( x ) = inf f ( y ) + | f | 1 d ( x , y ) : y Y . {:(({3.16})")"F_(1)^(**)(x)=i n f{[f(y)+|f|_(1)^(')d(x,y)]:y in Y}.:}\begin{equation*} F_{1}^{*}(x)=\inf \left\{\left[f(y)+|f|_{1}^{\prime} d(x, y)\right]: y \in Y\right\} . \tag{$\{3.16\}$} \end{equation*}({3.16})F1(x)=inf{[f(y)+|f|1d(x,y)]:yY}.
and
(3.17) F 2 ( x ) = sup { [ f ( y ) f Y d ( x , y ) ] : y Y } . (3.17) F 2 ( x ) = sup f ( y ) f Y d ( x , y ) : y Y . {:(3.17)F_(2)^(**)(x)=s u p{[f(y)-∣f_(||)||_(Y)d(x,y)]:y in Y}.:}\begin{equation*} F_{2}^{*}(x)=\sup \left\{\left[f(y)-\mid f_{\|} \|_{Y} d(x, y)\right]: y \in Y\right\} . \tag{3.17} \end{equation*}(3.17)F2(x)=sup{[f(y)fYd(x,y)]:yY}.
Proof. By [2], F 1 F 1 F_(1)^(**)F_{1}{ }^{*}F1 and F 2 F 2 F_(2)^(**)F_{2}{ }^{*}F2 are Lipschitz extensions of f | r f r f|_(r)\left.f\right|_{r}f|r and obviously. for every Lipschitz extension F F FFF of f | Y f Y f|_(Y)\left.f\right|_{Y}f|Y we have
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),quad x inX^(')F_{2}^{*}(x) \leqslant F(x) \leqslant F_{1}^{*}(x), \quad x \in X^{\prime}F2(x)F(x)F1(x),xX
From these inequalities, it follows that
; F u max ( F 1 u , | F e | u ) ; F u max F 1 u , F e u ||;F||_(u) <= max(||F_(1)**||_(u),|F_(e)**|_(u))\|; F\|_{u} \leqslant \max \left(\left\|F_{1} *\right\|_{u},\left|F_{\mathrm{e}} *\right|_{u}\right);Fumax(F1u,|Fe|u)
Remark 3. Dunham [3] has considered a problem similar to the problem in (d) in the case when G ( f ) G ( f ) G(f)G(f)G(f) has the betweenness property (see [3] for definition). In (d) the set G ( f ) G ( f ) G(f)G(f)G(f), being convex, has the betweenness property. We found explicitly the nearest and the farthest points of f f fff in G ( f ) G ( f ) G(f)G(f)G(f).

References

I. C. DE BOOR, "On "Best" Interpolation," University of Wisconsin, MRC TSR No. 1426, 1974.
2. I. Czipsfr and L. Gehér, Extension of function satisfying a Lipschitz condition, Acta Math. Acad. Sci. Hungar. 6 (1955), 213-220.
3. C. B. Dunham, Chebyshev approximation with a null space, Proc. Amer. Math. Soc. 41 (1973), 557-558.
4. J. A. Johnson, Banach space of Lipschitz functions and vector-valued Lipschitz functions, Trans. Amer. Math. Soc. 148 (1970), 147-169.
5. C. Mustăța, Asupra unor subspații cebîșeviene din spațiul normat al funcțiilor lipschitziene, Rev. Anal Numer. Teoria Aproximației 2 (1973), 81-87.
6. R. R. Phelps, Uniqueness of Hahn-Banach extension and unique best approximation, Trans. Amer. Math. Soc. 95 (1960), 238-255.
7. A. K. Rox, Extreme points and linear isometries of Banach space of Lipschitz functions, Canad. J. Math. 20 (1968), 1150-1164.
8. I. Singer, Cea mai bună aproximare în spații vectoriale normate prin elemente din subspații vectoriale, Edit. Acad. R. S. Romania, București, (1967).
9. D. R. Sherbert, The structure of ideals and point derivations in Banach algebras of Lipschitz functions, Trans. Amer. Math. Soc. 111 (1964), 240-272.
1977

Related Posts