[1] Brown, B.M., Elliot, D. and D.F. Paget, Lipschitz constants for the Bernstein polynomials of a Lipschitz continuous funcitons, J. Approx. Theory 49 (1987), 196-199.
[2] Cobzas, S. and C. Mustata, Norm-Preserving Extension of Convex Lipschitz Functions, J. Approx. Theory 34 (1978), 236-244.
[3] Czipser, J. and L. Geher, Extension of Functions satisfying a Lipschitz Condition, Acta Math. Acad. Sci. Hungar 6(1955), 213-220.
[4] Deutsch, F., Li W. and S.H. Park, Tietze Extensions and Continuous Selections for Metric Projections, J. Approx. Theory 63 (1991), 55-88.
[5] Mc Shane, E.J., Extension of Range of Functions, Bull. Amer. Math. Soc. 40 (1934), 837-842.
[6] Mustat, C., Norm Preserving Extension of Starhaped Lipschitz Funcitons, Mathematica 19(42)2 (1977), 183-187.
[7] Mustata C., Best Approximaiton and Uniquje Extension of Lipschitz Functions, J. Approx. Theory 19(1977), 222-230.
[8] Mustata C., M-ideals in Metric Spaces, “Babes-Bolyai” University, Fac. of Math. and Physics, Research Seminars, Seminar on Math. Analysis, Preprint nr.7 (1988), 65-74.
[9] Mustata, C., Selections Associated to Mc Shane’s Extension Theorem for Lipschitz Functions, Revue d’Analyse Numerique et de Theorie de l’Approximation, 21, 2 (1992), 135-145.
Paper (preprint) in HTML form
1993-Mustata-Selections with values Bernstein polynomials-Jnaat
SELECTIONS WITH VALUES BERNSTEIN POLYNOMIALS ASSOCIATED TO THE EXTENSION OPERATOR FOR LIPSCHITZ FUNCTIONS
COSTICA MUSTATTA(Cluj-Napoca)
Let (X,d)(X, d) be a metric space and YY a subset of XX, containing at least two points. A function f:Y rarr Rf: Y \rightarrow R is called Lipschitz if there exists a constant K_(Y)(f) >= 0K_{Y}(f) \geqslant 0 such that
for all x,y in Xx, y \in X.
{:(2)" Lip "Y={f∣f:Y rarr R","f" is Lipschitz on "Y}.:}\begin{equation*}
\text { Lip } Y=\{f \mid f: Y \rightarrow R, f \text { is Lipschitz on } Y\} . \tag{2}
\end{equation*}
Equipped with the pointwise operations of addition and multiplication by real scalars, Lip YY is a real linear space. The smallest constant K_(Y)(f)K_{Y}(f) verifying (1) is denoted by ||f||_(Y)\|f\|_{Y} and is called the Lipschitz norm of ff.
For Y=XY=X one obtains the linear space Lip XX and for F inF \in Lip XX, ||F^(')||_(X)\left\|F^{\prime}\right\|_{X} is the smallest Lipsehitz constant for F^(')F^{\prime} (on XX ).
Obviously that for F in Lip XF \in \operatorname{Lip} X and Y sub X,F|_(Y)in Lip YY \subset X,\left.F\right|_{Y} \in \operatorname{Lip} Y and ||F|_(Y)||_(Y)⩽⩽||F^(')||_(X)\left\|\left.F\right|_{Y}\right\|_{Y} \leqslant \leqslant\left\|F^{\prime}\right\|_{X}.
Let Y sube XY \subseteq X and f inf \in Lip YY. A function F inF \in Lip XX is called norm preserving Lipschitz extension of ff if
Let:
{:(3)F|_(Y)=f" and "||F||_(X)=||f||_(Y):}\begin{equation*}
\left.F\right|_{Y}=f \text { and }\|F\|_{X}=\|f\|_{Y} \tag{3}
\end{equation*}
{:(4)E(f)={F in Lip X:F|_(Y)=f" and "||F||_(X)=||f||_(Y)}:}\begin{equation*}
E(f)=\left\{F \in \operatorname{Lip} X:\left.F\right|_{Y}=f \text { and }\|F\|_{X}=\|f\|_{Y}\right\} \tag{4}
\end{equation*}
be the set of all norm preserving Lipschitz extensions of f inf \in Lip YY.
By a result of Mc Shane [5], every f inf \in Lip YY has a least one norm preserving extension F in Lip XF \in \operatorname{Lip} X, i.e. E(f)!=O/E(f) \neq \emptyset for every f in Lip Yf \in \operatorname{Lip} Y.
Since for every constant function c in Lip Y,||c||_(Y)=0c \in \operatorname{Lip} Y,\|c\|_{Y}=0, it follows that the "norm" ||quad||_(Y)\|\quad\|_{Y} is only a seminorm on Lip YY. In order to obtain a genuine norm, fix a point x_(0)in Yx_{0} \in Y and let
Then ||||_(Y):}\left\|\|_{Y}\right. is a norm on Lip_(0)Y\operatorname{Lip}_{0} Y and Lip_(0)Y\operatorname{Lip}_{0} Y is a Banach space with respect to this norm.
Then, by the above quoted result of Me Shane, one obtains :
Theorem 1. Let ( X,dX, d ) be a metric space, x_(0)x_{0} a fixed point in XX and YY a subset of XX containing x_(0)x_{0}. Then for every f inLip_(0)Yf \in \operatorname{Lip}_{0} Y there exists F in Lip0_(0)XF \in \operatorname{Lip} 0_{0} X such that H^(')|_(Y)=f\left.H^{\prime}\right|_{Y}=f and ||F||_(X)=||f||_(Y)\|F\|_{X}=\|f\|_{Y}.
It is easily seen (see [3], [5]) that the following functions:
{:[F_(1)(x)=s u p{f(y)-||f||_(Y)*d(x,y):y in Y}","quad x in X],[(6)" and "],[F_(2)(x)=i n f{f(y)-||f||_(Y)*d(x,y):y in Y}","quad x in X]:}\begin{align*}
& F_{1}(x)=\sup \left\{f(y)-\|f\|_{Y} \cdot d(x, y): y \in Y\right\}, \quad x \in X \\
& \text { and } \tag{6}\\
& F_{2}(x)=\inf \left\{f(y)-\|f\|_{Y} \cdot d(x, y): y \in Y\right\}, \quad x \in X
\end{align*}
are two norm preserving Lipschitz extension of f inLip_(0)If \in \operatorname{Lip}_{0} I. The set W(f)subLip_(0)XW(f) \subset \operatorname{Lip}_{0} X is nonempty, convex and bounded, such that the extension operator
{:(7)D:Lip_(0)Y rarr2^(Lip_(0)X):}\begin{equation*}
D: \operatorname{Lip}_{0} Y \rightarrow 2^{\operatorname{Lip}_{0} X} \tag{7}
\end{equation*}
is well defined and multivalued.
The problem of the existence of a selection of the extension operator EE (i.e. a function e:Lip_(0)Y rarrLip_(0)Xe: \operatorname{Lip}_{0} Y \rightarrow \operatorname{Lip}_{0} X such that e(f)in E(f)e(f) \in E(f), for all f inLip_(0)Yf \in \operatorname{Lip}_{0} Y ) which is linear and continuous was considered in [9].
In the particular case X=RX=R and Y=[a,b],x_(0)in YY=[a, b], x_{0} \in Y, a linear and continuous selection ee of EE can be given explicitly (see [9]).
2. Suppose XX is a normed space and Y sub X,x_(0)in XY \subset X, x_{0} \in X. In this case there exist functions in E(f)E(f) which preserve some properties of ff such as starshapedness or convexity (in these case YY is supposed to be a starshaped, respectively a convex subset of XX and x_(0)=thetax_{0}=\theta ) (see [2], [6]). A natural question is to give explicit selections, with the values preserving some properties of the function ff and to study their linearity and continuity.
In the following we shall present an example of a homogeneous and continuous selection having values Bernstein polynomials.
Let X=[0,1],Y={0,1},x_(0)=0X=[0,1], Y=\{0,1\}, x_{0}=0 and d(x,y)=|x-y|d(x, y)=|x-y|. Tri this case
FF is Lipschitz on [0,1]}[0,1]\}
For f inLip_(0)Yf \in \operatorname{Lip}_{0} Y we have ||f||_(Y)=|f(1)|\|f\|_{Y}=|f(1)| and
{:(9)||T||_(X)=s u p{|F(x)-F(y)|//|x-y|:x","y in[0","1]","x!=y}.:}\begin{equation*}
\|T\|_{X}=\sup \{|F(x)-F(y)| /|x-y|: x, y \in[0,1], x \neq y\} . \tag{9}
\end{equation*}
for F inLip_(0)XF \in \operatorname{Lip}_{0} X.
In this case EE is single-valued, namely E(f)={F}E(f)=\{F\} where F(x)==f(1)x,x in[0,1]F(x)= =f(1) x, x \in[0,1] and the following result hold:
Theorem 2. The application E:Lip_(0)Y rarrLip_(0)XE: \operatorname{Lip}_{0} Y \rightarrow \operatorname{Lip}_{0} X, where E(f)={F}E(f)=\{F\} with F(x)=f(1)x,x in[0,1]F(x)=f(1) x, x \in[0,1] is linear and continuous.
Proof. The functions F_(1)F_{1} and F_(2)F_{2} given by (6) are equals and F_(1)(x)==F_(2)(x)=f(1)x,x in[0,1]F_{1}(x)= =F_{2}(x)=f(1) x, x \in[0,1]. In [9, Th. 4 and Corollary 5] it was proved that e(f)=(1//2)(F_(1)+F_(2))e(f)=(1 / 2)\left(F_{1}+F_{2}\right) is a linear and continuous selection for EE, so that e(f)=E(f)e(f)=E(f) is linear and continuous.
Remark 1. It is well known (see [1]) that for F inLip_(0)[0,1]F \in \operatorname{Lip}_{0}[0,1] the Bernstein polynomial of degree n(n >= 1)n(n \geqslant 1) given by
{:(10)B_(n)(F;x)=sum_(k=0)^(n)((n)/(k))F((k)/(n))x^(k)(1-x)^(n-k)","quad x in[0","1]:}\begin{equation*}
B_{n}(F ; x)=\sum_{k=0}^{n}\binom{n}{k} F\left(\frac{k}{n}\right) x^{k}(1-x)^{n-k}, \quad x \in[0,1] \tag{10}
\end{equation*}
is Lipschitz and ||B_(n)(F;)||_(X) <= ||F||_(X)\left\|B_{n}(F ;)\right\|_{X} \leqslant\|F\|_{X}. Because B_(n)(F;0)=F(0)==f(0)B_{n}(F ; 0)=F(0)= =f(0) and B_(n)(l^(');1)=F(1)=f(1)B_{n}\left(l^{\prime} ; 1\right)=F(1)=f(1) for every F in E(f)F \in E(f), it follows that B_(n)(F;)inM(f)B_{n}(F 😉 \in \mathcal{M}(f), for every f inLip_(0)Yf \in \operatorname{Lip}_{0} Y. In this case B_(n)(F;x)=f(1)xB_{n}(F ; x)=f(1) x, for all n in N,n >= 1n \in N, n \geqslant 1, so that the application
{:(11)f|->B_(n)(F;*)=E(f)=B_(1)(F;*):}\begin{equation*}
f \mapsto B_{n}(F ; \cdot)=E(f)=B_{1}(F ; \cdot) \tag{11}
\end{equation*}
is linear and continuous. Therefore, in this case, the extension operator E:Lip_(0)Y rarr2^("Lip "_(0)x)E: \operatorname{Lip}_{0} Y \rightarrow 2^{\text {Lip }_{0} x}, admits a linear and continuous selection with values Bernstein polynomials (of a fixed, but arbitrary, degree nn ).
It we are looking for a Lipschitz extension with a greater Lipschitz constant alpha||quad||_(y)\alpha\|\quad\|_{y}, where alpha > 1\alpha>1 is fixed, then the extension operator denoted by E_(alpha)E_{\alpha}, will be multivalued.
From Theorem 2 one obtains the following corollary :
Corollary 1. For every f inLip_(0)Yf \in \operatorname{Lip}_{0} Y there exists F inLip_(0)XF \in \operatorname{Lip}_{0} X such that
{:(12)F|_(Y)=f" and "||F||_(X)=alpha","|f(1)|=alpha||f||_(Y):}\begin{equation*}
\left.F\right|_{Y}=f \text { and }\|F\|_{X}=\alpha,|f(1)|=\alpha\|f\|_{Y} \tag{12}
\end{equation*}
Proof. It is easy to verify that the functions
{:[ bar(x)_(1)(x)=max{f(y)-alpha|f(1)||x-y|:y in{0","1}}],[(13)" and "],[ bar(I)_(2)(x)=min{f(y)+alpha|f(1)||x-y|:y in{0","1}}]:}\begin{align*}
& \bar{x}_{1}(x)=\max \{f(y)-\alpha|f(1)||x-y|: y \in\{0,1\}\} \\
& \text { and } \tag{13}\\
& \bar{I}_{2}(x)=\min \{f(y)+\alpha|f(1)||x-y|: y \in\{0,1\}\}
\end{align*}
denote the set of the Lipschitz extensions of the function ff which preserve the norm alpha||f||_(Y)\alpha\|f\|_{Y}. Then bar(F)_(1), bar(F)_(2)inE_(alpha)(f)\bar{F}_{1}, \bar{F}_{2} \in E_{\alpha}(f) and bar(F)_(1)(x)!= bar(F)_(2)(x)\bar{F}_{1}(x) \neq \bar{F}_{2}(x) for all x in(0,1)x \in(0,1), so that the extension operator
H_(alpha):Lip_(0)Y rarr2^(Lip_(0)X)H_{\alpha}: \operatorname{Lip}_{0} Y \rightarrow 2^{\operatorname{Lip}_{0} X}
is well defined and multivalued.
Concerning this operator E_(alpha)E_{\alpha} one can prove the following theorem:
Theorem 3. a) The operator E_(alpha)E_{\alpha} admits a homogeneous and continuous selection;
b) For every n in N,n >= 1n \in N, n \geqslant 1, the operator E_(alpha)E_{\alpha} admits a homogeneous and continuous selection with values Bernstein polynomials of degree nn.
Proof. a) Consider the following two selections e_(1),e_(2)e_{1}, e_{2} defined by
{:(15)e_(1)(f)= bar(F)_(1)" and "e_(2)(f)= bar(F)_(2)","f in Lipp_(0)I",":}\begin{equation*}
e_{1}(f)=\bar{F}_{1} \text { and } e_{2}(f)=\bar{F}_{2}, f \in \operatorname{Lip} p_{0} I, \tag{15}
\end{equation*}
where
vec(B_)_(1)(x)=max{-alpha|f(1)|x;f(1)-alpha|f(1)|(1-x},a in[0,1]\overrightarrow{\underline{B}}_{1}(x)=\max \{-\alpha|f(1)| x ; f(1)-\alpha|f(1)|(1-x\}, a \in[0,1]
and
{:(16) bar(E)_(2)(x)=min{a|f(1)|x;f(1)+alpha|f(1)|(1-x)}","x in[0","1].:}\begin{equation*}
\bar{E}_{2}(x)=\min \{a|f(1)| x ; f(1)+\alpha|f(1)|(1-x)\}, x \in[0,1] . \tag{16}
\end{equation*}
Then, for lambda >= 0,e_(1)(lambda f)=lambda*e_(1)(f)\lambda \geqslant 0, e_{1}(\lambda f)=\lambda \cdot e_{1}(f) and e_(2)(lambda f)=lambda*e_(2)(f)e_{2}(\lambda f)=\lambda \cdot e_{2}(f). By the definition of e_(1)e_{1} and e_(2),e_(1)(f)=-e_(2)(-f)e_{2}, e_{1}(f)=-e_{2}(-f), implying that the selection
(17)
e(lambda f)=lambda e(f),quad lambda in R,quad f inLip_(0)Ye(\lambda f)=\lambda e(f), \quad \lambda \in R, \quad f \in \operatorname{Lip}_{0} Y
Now, we show that e_(1),e_(2)e_{1}, e_{2} are continuous selections which will imply the continuity of ee, too.
Let epsi > 0\varepsilon>0 and 0 < delta < epsi0<\delta<\varepsilon. We shall show that for f,g in Lipp_(0)Yf, g \in \operatorname{Lip} p_{0} Y, a|f(1)-g(1)| < deltaa|f(1)-g(1)|<\delta implies || bar(F)_(1)- bar(G)_(1)||_(X) < epsi\left\|\bar{F}_{1}-\bar{G}_{1}\right\|_{X}<\varepsilon where bar(F)_(1)\bar{F}_{1} is defined by (16) and
bar(G)_(1)(x)=max{-alpha|g(1)|x;g(1)-alpha|g(1)|(1-x)},x in[0,1].\bar{G}_{1}(x)=\max \{-\alpha|g(1)| x ; g(1)-\alpha|g(1)|(1-x)\}, x \in[0,1] .
We have to consider the following cases: 1^(@)f(1) > r,g(1) > 01^{\circ} f(1)>r, g(1)>0.
In this case
{:[ bar(F)_(1)(x)- bar(G)_(1)(x)=alpha[g(1)-f(1)]x","" for "x in[0,(alpha-1)/(2alpha)]],[=f(1)-g(1)-alpha[f(1)-g(1)](1-x)","" for "x in((alpha-1)/(2alpha),1]]:}\begin{aligned}
\bar{F}_{1}(x)-\bar{G}_{1}(x) & =\alpha[g(1)-f(1)] x, \text { for } x \in\left[0, \frac{\alpha-1}{2 \alpha}\right] \\
& =f(1)-g(1)-\alpha[f(1)-g(1)](1-x), \text { for } x \in\left(\frac{\alpha-1}{2 \alpha}, 1\right]
\end{aligned}
implying || bar(F)_(1)- bar(G)_(1)||_(x)=alpha|f(1)-g(1)| < delta < epsi\left\|\bar{F}_{1}-\bar{G}_{1}\right\|_{x}=\alpha|f(1)-g(1)|<\delta<\varepsilon. 2^(@)f(1) < 0,g(1) < 02^{\circ} f(1)<0, g(1)<0.
In this case
{:[ vec(F)_(1)(x)- bar(G)_(1)(x)=alpha[|f(1)|-|g(1)|]x","" for "x in[0,(alpha+1)/(2alpha)]=],[=f(1)-g(1)-alpha[|f(1)|-|g(1)|](1-x)","" for "x in((alpha+1)/(2alpha),1]]:}\begin{aligned}
\vec{F}_{1}(x)-\bar{G}_{1}(x) & =\alpha[|f(1)|-|g(1)|] x, \text { for } x \in\left[0, \frac{\alpha+1}{2 \alpha}\right]= \\
& =f(1)-g(1)-\alpha[|f(1)|-|g(1)|](1-x), \text { for } x \in\left(\frac{\alpha+1}{2 \alpha}, 1\right]
\end{aligned}
{:[ bar(F)_(1)(x)- bar(G)_(1)(x)=alpha[|g(1)|-|f(1)|]x","" for "x in[0,(alpha-1)/(2alpha)]],[=f(1)-alpha f(1)+alpha[|g(1)|-f(1)]*x","" for "x in((alpha-1)/(2alpha),(alpha+1)/(2alpha)]],[=f(1)-g(1)+alpha[|g(1)|-f(1)]+alpha[f(1)-|g(1)|]x],[" for "x in((alpha+1)/(2alpha),1]]:}\begin{aligned}
& \bar{F}_{1}(x)-\bar{G}_{1}(x)=\alpha[|g(1)|-|f(1)|] x, \text { for } x \in\left[0, \frac{\alpha-1}{2 \alpha}\right] \\
&=f(1)-\alpha f(1)+\alpha[|g(1)|-f(1)] \cdot x, \text { for } x \in\left(\frac{\alpha-1}{2 \alpha}, \frac{\alpha+1}{2 \alpha}\right] \\
&=f(1)-g(1)+\alpha[|g(1)|-f(1)]+\alpha[f(1)-|g(1)|] x \\
& \text { for } x \in\left(\frac{\alpha+1}{2 \alpha}, 1\right]
\end{aligned}
implying || bar(h)_(1)- bar(G)_(1)||_(x)=alpha|f^(')(1)-|g(1)|| <= alpha|f(1)-g(1)| < delta < epsi4^(@)f(1)=0\left\|\bar{h}_{1}-\bar{G}_{1}\right\|_{x}=\alpha\left|f^{\prime}(1)-|g(1)|\right| \leqslant \alpha|f(1)-g(1)|<\delta<\varepsilon 4^{\circ} f(1)=0 and g(1)!=0(g(1) \neq 0( or f(1)!=0f(1) \neq 0 and g(1)=0)g(1)=0)
In this case bar(F)_(1)(x)=0,x in[0,1]\bar{F}_{1}(x)=0, x \in[0,1] and || bar(F)_(1)- bar(G)_(1)||_(X)=|| bar(G)_(1)||_(X)==a|g(1)| < delta < epsi\left\|\bar{F}_{1}-\bar{G}_{1}\right\|_{X}=\left\|\bar{G}_{1}\right\|_{X}= =a|g(1)|<\delta<\varepsilon.
It follows that e_(1)e_{1} is a continuous selections. In a similar way one can show the continuty of the selection, e_(2)e_{2}, implying the continuity of the selection ee.
b) Let n in N,n >= 1n \in N, n \geqslant 1, be a fixed and for f inLip_(0)Yf \in \mathrm{Lip}_{0} Y let B_(n)(e(f)B_{n}(e(f); .) be the Bernstein operator asociated to the function e(f)e(f) :
{:(18)B_(n)(e(f);x)=sum_(k=0)^(n)((n)/(k))*e(f)((k)/(n))*x^(k)(1-x)^(n-k)","quad x in[0","1]:}\begin{equation*}
B_{n}(e(f) ; x)=\sum_{k=0}^{n}\binom{n}{k} \cdot e(f)\left(\frac{k}{n}\right) \cdot x^{k}(1-x)^{n-k}, \quad x \in[0,1] \tag{18}
\end{equation*}
Since B_(n)(e(f);0)=e(f)(0)=f(0)=0B_{n}(e(f) ; 0)=e(f)(0)=f(0)=0 and B_(n)(e(f);1)=e(f)(1)=f(1)B_{n}(e(f) ; 1)=e(f)(1)=f(1), it follows that B_(n)(e(f);.)inE_(alpha)(f)B_{n}(e(f) ;.) \in E_{\alpha}(f).
As the Bernstein operator is linear it follows that for lambda inR\lambda \in \mathbb{R}, b_(n)(lambda f)=B_(n)(e(lambda f);)=.B_(n)(lambda e(f);)=.lambdaB_(n)(e(f);)=.lambdab_(n)(f)b_{n}(\lambda f)=B_{n}(e(\lambda f) ;)=.B_{n}(\lambda e(f) ;)=.\lambda B_{n}(e(f) ;)=.\lambda b_{n}(f), showing that b_(n)b_{n} is a homogeneous selection.
If f,g inLip_(0){0,1}f, g \in \operatorname{Lip}_{0}\{0,1\} are such that alpha|f(1)-g(1)| < delta < epsi\alpha|f(1)-g(1)|<\delta<\varepsilon then || bar(H)_(1)- bar(G)_(1)||_(x) < epsi\left\|\bar{H}_{1}-\bar{G}_{1}\right\|_{x}<\varepsilon and || bar(F)_(2)- bar(G)_(2)||_(x) < epsi\left\|\bar{F}_{2}-\bar{G}_{2}\right\|_{x}<\varepsilon, so that ||b_(18)(f)-b_(n)(g)||_(X)=||B_(n)(e(f);.)-B_(n)(e(g);.)||_(X)=\left\|b_{18}(f)-b_{n}(g)\right\|_{X}=\left\|B_{n}(e(f) ; .)-B_{n}(e(g) ; .)\right\|_{X}=
showing that the selection b_(n)b_{n} is also continuous.
Remark 2. (a) Let C^(+)C^{+}be the cone of positive functions in Lip_(0){0,1}\operatorname{Lip}_{0}\{0,1\} and C^(-)C^{-}the cone of negative functions, i.e.
Then e_(1)(C^(-))subeK^(-)e_{1}\left(C^{-}\right) \subseteq K^{-}, where K^(-)={F in Lipp_(0)[0,1],P:}K^{-}=\left\{F \in \operatorname{Lip} p_{0}[0,1], P\right. is negative }\}, and e_(2)(C^(+))subeK^(+)e_{2}\left(C^{+}\right) \subseteq K^{+}, where K^(+)={F in Lip_(0)[0,1],F:}K^{+}=\left\{F \in \operatorname{Lip}{ }_{0}[0,1], F\right. is positive }\}.
be the restrictions of E_(alpha)E_{\alpha} to the cones C^(-)C^{-}and C^(+)C^{+}, respectively.
Obviously that E_(alpha)^(-)(f)!=O/E_{\alpha}^{-}(f) \neq \emptyset, for every f inC^(-)f \in C^{-}(the set E_(alpha)^(-)(f)E_{\alpha}^{-}(f) contains at least the function bar(F)_(1)inK^(-)\bar{F}_{1} \in K^{-}) and E_(alpha)^(+)(f)!=0E_{\alpha}^{+}(f) \neq 0, for every f inC^(+)f \in C^{+}(the set E_(alpha)^(+)(f)E_{\alpha}^{+}(f) contains at least the function bar(F)_(2)inK^(+)\bar{F}_{2} \in K^{+}).
We have the following corollary
Corollary 2. a) The selection e_(1)^(-)(f)= bar(F)_(1),f inC^(-)e_{1}^{-}(f)=\bar{F}_{1}, f \in C^{-}, associated to the operator E_(-)^(-)E_{-}^{-}is continuous, positively homogeneous and additive;
b) The selection e_(2)^(+)(f)= bar(F)_(2),f inC^(+)e_{2}^{+}(f)=\bar{F}_{2}, f \in C^{+}, associated to the operator D_(alpha)^(+)D_{\alpha}^{+}is continuous, positively homogeneous and additive;
e) The selections b_(n)^(-)(f)=B_(n)(e_(1)^(-)(f);.)b_{n}^{-}(f)=B_{n}\left(e_{1}^{-}(f) ;.\right) and b_(n)^(+)(f)=B_(n)(e_(2)^(+)(f);.)b_{n}^{+}(f)=B_{n}\left(e_{2}^{+}(f) ;.\right) are continuous, positively homogeneous and additive.
Proof. The continuity and the positive homogeneity of the selections e_(1)^(-)e_{1}^{-}and e_(2)^(+)e_{2}^{+}follow from the proofs of Cases 1^(@)1^{\circ} and 2^(@)2^{\circ} of Theorem 3.
If f(1) < 0f(1)<0 and g(1) < 0g(1)<0 then
{:[ bar(F)_(1)(x)=-alpha|f(1)|x","quad" for "quad a in[0,(alpha+1)/(2alpha)]],[=f(1)-alpha|f(1)|(1-x)","quad" for "quad a in((alpha+1)/(2alpha),1]]:}\begin{aligned}
\bar{F}_{1}(x) & =-\alpha|f(1)| x, \quad \text { for } \quad a \in\left[0, \frac{\alpha+1}{2 \alpha}\right] \\
& =f(1)-\alpha|f(1)|(1-x), \quad \text { for } \quad a \in\left(\frac{\alpha+1}{2 \alpha}, 1\right]
\end{aligned}
and
{:[ bar(G)_(1)(x)=-alpha|g(1)|x","" for "quad x in[0,(alpha+1)/(2alpha)]],[=g(1)-alpha|g(1)|(1-x)","" for "x in((alpha+1)/(2alpha),1]]:}\begin{aligned}
\bar{G}_{1}(x) & =-\alpha|g(1)| x, \text { for } \quad x \in\left[0, \frac{\alpha+1}{2 \alpha}\right] \\
& =g(1)-\alpha|g(1)|(1-x), \text { for } x \in\left(\frac{\alpha+1}{2 \alpha}, 1\right]
\end{aligned}
Similarly for f(1) > 0f(1)>0 and g(1) > 0g(1)>0 one obtains e_(2)^(+)(f+g)===e_(2)^(+)(f)+e_(2)^(+)(g)e_{2}^{+}(f+g)== =e_{2}^{+}(f)+e_{2}^{+}(g).
Assertion c) follows from the fact that the Bernstein operator is linear and positive.
(b) Remark that the selections e_(1)^(-)e_{1}^{-}and e_(1)^(+)e_{1}^{+}are monotonically increasing with respect to the pointwise order, i.e. 0 < f(1) < g(1)0<f(1)<g(1) implies bar(F)_(2)(a)⩽⩽ bar(G)_(2)(x),x in[0,1]\bar{F}_{2}(a) \leqslant \leqslant \bar{G}_{2}(x), x \in[0,1] and 0 > f(1) > g(1)0>f(1)>g(1) implies F_(1)(x) >= G_(1)(x),x in[0,1]F_{1}(x) \geqslant G_{1}(x), x \in[0,1]
Furthermore, e_(1)^(-)(f)e_{1}^{-}(f) is a convex lunction for f inC^(-)f \in C^{-}and e_(2)^(+)(f)e_{2}^{+}(f) is a concave function for f inC^(+)f \in C^{+}.
3. Selections associated to the operator of metric projection
Let Y^(_|_)Y^{\perp} be the anihilator of the set Y={0,1}Y=\{0,1\} in Lip_(0)[0,1]\operatorname{Lip}_{0}[0,1], i.e.
{:(22)Y⊥={G inLip_(0)[0,1]:G(0)=G(1)=0}:}\begin{equation*}
Y \perp=\left\{G \in \operatorname{Lip}_{0}[0,1]: G(0)=G(1)=0\right\} \tag{22}
\end{equation*}
Then X^(_|_)X^{\perp} is a closed ideal in Lipp_(0)[0,1]\operatorname{Lip} p_{0}[0,1]. For X^(1)in Lipp_(0)[0,1]X^{1} \in \operatorname{Lip} p_{0}[0,1] let
{:(23)d(F,Y^(_|_))=i n f{||I^(Y)-G||_(X):quad G inY^(_|_)}:}\begin{equation*}
d\left(F, Y^{\perp}\right)=\inf \left\{\left\|I^{Y}-G\right\|_{X}: \quad G \in Y^{\perp}\right\} \tag{23}
\end{equation*}
An element G_(0)inY^(_|_)G_{0} \in Y^{\perp} for which the infimum in (23) is attained is called the nearest point to F^(1)F^{1} in Y^(_|_)Y^{\perp}.
for all F inLip_(0)[0,1]F \in \operatorname{Lip}_{0}[0,1]. X^(_|_)X^{\perp} is called proximinal (resp. Chebyshev) if for each F inJip_(0)[0,1]F \in \mathrm{Jip}_{0}[0,1] the set P_(Y^(_|_))(F^(r))P_{Y^{\perp}}\left(F^{r}\right) is nonempty (resp. a singleton).
The following proposition holds :
Proposition 1. a) The formula
(25)
is valid for every H^(')inLip_(0)H^{\prime} \in \operatorname{Lip}_{0} [0,1]. In particular Y^(_|_)Y^{\perp} is a proximinal subspace of Lip Lip_(0)[0,1]\operatorname{Lip}_{0}[0,1];
b) If G inP_(x^(_|_))(F)G \in P_{x^{\perp}}(F) then G=H^(TT)-HG=H^{\top}-H, where H inE_(alpha)(F^(TT)|_(Y))∣H \in E_{\alpha}\left(\left.F^{\top}\right|_{Y}\right) \mid is such that ||H||_(X)=|F(1)|:\|H\|_{X}=|F(1)|:
c) There holds the equality:
{:(26)d(F","X _|_)=d(F,F-E_(x)(F|_(Y))):}\begin{equation*}
d(F, X \perp)=d\left(F, F-E_{x}\left(\left.F\right|_{Y}\right)\right) \tag{26}
\end{equation*}
where F-D_(alpha)(F|_(X))={F-Pi:H inE_(alpha)(F|_(Y))},quad F in Lip_(0)[0,1];F-D_{\alpha}\left(\left.F\right|_{X}\right)=\left\{F-\Pi: H \in E_{\alpha}\left(\left.F\right|_{Y}\right)\right\}, \quad F \in \operatorname{Lip}{ }_{0}[0,1] ;
d) The equality
{:(27)s u p{||F-G||_(X):G in F-W_(alpha)(F|_(Y))}=alpha|F(1)|:}\begin{equation*}
\sup \left\{\|F-G\|_{X}: G \in F-W_{\alpha}\left(\left.F\right|_{Y}\right)\right\}=\alpha|F(1)| \tag{27}
\end{equation*}
holds for every W inLip_(0)[0,1]W \in \operatorname{Lip}_{0}[0,1].
Proof. a) Let F in Lipp_(0)[0,1]F \in \operatorname{Lip} p_{0}[0,1]. Then for every G in Y _|_G \in Y \perp one has |G(1)|==|F(1)-G(1)| <= ||F-G||_(X)|G(1)|= =|F(1)-G(1)| \leqslant\|F-G\|_{X}. Taking the infimum with respect to G inY^(_|_)G \in Y^{\perp} one obtains
Let G_(0)(x)=H(x)-H(1)x,x in[0,1]G_{0}(x)=H(x)-H(1) x, x \in[0,1]. It follows that G_(0)in Y _|_(G_(0)(0)=0=:}=G_(0)(1)G_{0} \in Y \perp\left(G_{0}(0)=0=\right. =G_{0}(1) and. ||F-G_(0)||_(X)=|F^(')(1)|\left\|F-G_{0}\right\|_{X}=\left|F^{\prime}(1)\right| so that ||F^(')-G_(0)||_(X)=d(F^(I),Y _|_)\left\|F^{\prime}-G_{0}\right\|_{X}=d\left(F^{I}, Y \perp\right). This shows that Y^(_|_)Y^{\perp} is a proximinal subspace of Lip_(0)[0,1]\operatorname{Lip}_{0}[0,1].
b) If G inP_(Y^(_|_))(F^(l))G \in P_{Y^{\perp}}\left(F^{l}\right), then
showing that F-G inE_(x)(P^(')|_(Y))F-G \in E_{x}\left(\left.P^{\prime}\right|_{Y}\right). If follows that there exists UU in E_(alpha)(F|_(Y^(')))E_{\alpha}\left(\left.F\right|_{Y^{\prime}}\right) such that F^(')-G=HF^{\prime}-G=H and ||W||_(x)=||I-G_(x^(˙))=|H^(')(1)|\|W\|_{x}=\| I-G_{\dot{x}}=\left|H^{\prime}(1)\right|.
c) Follows from a) and b).
d) For every G in F-H_(3)(F|_(Y))G \in F-H_{3}\left(\left.F\right|_{Y}\right) we have
where H inD_(alpha)(H_(∣J))H \in D_{\alpha}\left(H_{\mid J}\right).
Taking the supremum with respect to G in I-E_(alpha)(F|_(x))G \in I-E_{\alpha}\left(\left.F\right|_{x}\right) we find
s u p{||F-G||_(X):G in F-D_(alpha)(F|_(Y))} <= alpha|F(1)|.\sup \left\{\|F-G\|_{X}: G \in F-D_{\alpha}\left(\left.F\right|_{Y}\right)\right\} \leqslant \alpha|F(1)| .
Let
G_(1)(x)=H^(')(x)-max{-alpha|H^(')(1)|x;H^(')(1)-alpha|H^(')(1)|(1-x)},G_{1}(x)=H^{\prime}(x)-\max \left\{-\alpha\left|H^{\prime}(1)\right| x ; H^{\prime}(1)-\alpha\left|H^{\prime}(1)\right|(1-x)\right\},
and
G_(2)(x)=l^(')(x)-min{alpha|F^(')(1)|x;l^(')(1)+alpha|F^(')(1)|(1-x)},G_{2}(x)=l^{\prime}(x)-\min \left\{\alpha\left|F^{\prime}(1)\right| x ; l^{\prime}(1)+\alpha\left|F^{\prime}(1)\right|(1-x)\right\},
x in[0,1]x \in[0,1].
Obviously that G_(1),G_(2)in F-H_(2)(I^(')|_(Y))G_{1}, G_{2} \in F-H_{2}\left(\left.I^{\prime}\right|_{Y}\right) and,
proving the assertion dd ).
Remark 3. By Proposition 1 it follows that the nearest points to F in Lip[0,1]F \in \operatorname{Lip}[0,1] in Y _|_Y \perp are the functions G in F-E_(alpha)(F|_(Y))subY^(_|_),G=F-HG \in F-E_{\alpha}\left(\left.F\right|_{Y}\right) \subset Y^{\perp}, G=F-H with H inE_(alpha)(F|_(Y))H \in \mathbb{E}_{\alpha}\left(\left.F\right|_{Y}\right) of minimal Lipsehitz norm and the farthest points for F^(')F^{\prime} in F^(')-E_(a)(F^(')|_(Y))subX^(_|_)F^{\prime}-E_{a}\left(\left.F^{\prime}\right|_{Y}\right) \subset X^{\perp} are the functions G=F-I^(')G=F-I^{\prime}, with H inB_(a)(H^(Y)|_(Y))H \in B_{a}\left(\left.H^{Y}\right|_{Y}\right) of the maximal norm ( ||H||_(X)=alpha∣H(1)||\|H\|_{X}=\alpha \mid H(1) \| ).
Let r:Lip_(0)[0,1]rarrLip_(0){0,1}r: \operatorname{Lip}_{0}[0,1] \rightarrow \operatorname{Lip}_{0}\{0,1\} be the restriction operator
where I:Lip_(0)[0,1]rarrLip_(0)[0,1]I: \operatorname{Lip}_{0}[0,1] \rightarrow \operatorname{Lip}_{0}[0,1] is the identity operator, i.e.
is a multivalued operator for alpha > 1\alpha>1.
it follows that P_(Y^(_|_))(Y^(r))subeQ_(chi)(Y^(r))P_{Y^{\perp}}\left(Y^{r}\right) \subseteq Q_{\chi}\left(Y^{r}\right), for all F in LipP_(0)[0,1]F \in \operatorname{Lip} P_{0}[0,1].
Let T_(alpha):Lip_(0)[0,1]rarr2r^(_|_)T_{\alpha}: \operatorname{Lip}_{0}[0,1] \rightarrow 2 r^{\perp} be defined by
{:(30)T_(alpha)(H^('))={II inQ_(alpha)(I):||II-H^(')||_(X)=s u p{||U-I^(')||_(X):U inQ_(alpha)(I^('))}}.:}\begin{equation*}
T_{\alpha}\left(H^{\prime}\right)=\left\{I I \in Q_{\alpha}(I):\left\|I I-H^{\prime}\right\|_{X}=\sup \left\{\left\|U-I^{\prime}\right\|_{X}: U \in Q_{\alpha}\left(I^{\prime}\right)\right\}\right\} . \tag{30}
\end{equation*}
The following theorem holds:
Teorem 1. a) The operator P_(Sigma^(_|_))P_{\Sigma^{\perp}} is a linear and continuous selection of the operator Q_(alpha)Q_{\alpha};
b) The subspace I^(_|_)I^{\perp} is complemented in L_(1)P_(0)[0,1]L_{1} P_{0}[0,1] by the subspace
{:(31)W={H in Lip0[0,1]:H(x)=a alpha,alpha in[0,1],a inR_(j);:}:}\begin{equation*}
W=\left\{H \in \operatorname{Lip} 0[0,1]: H(x)=a \alpha, \alpha \in[0,1], a \in R_{j} ;\right. \tag{31}
\end{equation*}
tions.
c) The operators T_(alpha)T_{\alpha} and Q_(alpha)Q_{\alpha} admit continuous and homogeneous selec-
Proof. a) The operator P_(y^(_|_))P_{y^{\perp}} is single-valued since for every F inLip_(0)[0,1]F \in \operatorname{Lip}_{0}[0,1] there exists a unique element H inU_(x)(F|_(Y^(')))H \in U_{x}\left(\left.F\right|_{Y^{\prime}}\right) such that ||U||_(X)=∣F^(')(1)\|U\|_{X}= \mid F^{\prime}(1) and by Proposition 1. b), it tollows that I^(')I^{\prime} has a unique nearest point in Y _|_Y \perp. Then
so that ||P_(Y^(_|_))(F^('))-P_(Y^(_|_))(G)||_(X) < 2\left\|P_{Y^{\perp}}\left(F^{\prime}\right)-P_{Y^{\perp}}(G)\right\|_{X}<2 e for ||I^(')-G||_(X) < epsi\left\|I^{\prime}-G\right\|_{X}<\varepsilon, proving the continuity of the operator P_(gamma _|_)P_{\gamma \perp}.
b) Let F inLip_(0)[0,1]F \in \operatorname{Lip}_{0}[0,1]. Then G(x)=F(x)-W(1)x,x in[0,1]G(x)=F(x)-W(1) x, x \in[0,1] is an element of Y^(_|_)Y^{\perp} and, since F(1)xF(1) x is an element of WW it follows that P(x)===G(x)+E(1)x,quad x in[0,1]P(x)== =G(x)+E(1) x, \quad x \in[0,1].
If F_(n)rarrF^(')F_{n} \rightarrow F^{\prime} in Lip _(0)[0,1]{ }_{0}[0,1], i.e. ||F_(n)-F^(')||_(x)rarr0\left\|F_{n}-F^{\prime}\right\|_{x} \rightarrow 0, then the inequality
implies |F_(n)(1)|rarr|F(1)|\left|F_{n}(1)\right| \rightarrow|F(1)|, showing that the projection operator on WW is continuous. Consequently Lip_(0)[0,1]=Y _|_ o+W\operatorname{Lip}_{0}[0,1]=Y \perp \oplus W.
c) Consider the selections of the metric projections
{:[t_(alpha,1)(F)=F-e_(1)(F|_(Y))","quad F inLip_(0)[0","1]","],[t_(alpha,2)(F)=F-e_(2)(F_(Y)∣)","quad F inLip_(0)[0","1]","]:}\begin{aligned}
& t_{\alpha, 1}(F)=F-e_{1}\left(\left.F\right|_{Y}\right), \quad F \in \operatorname{Lip}_{0}[0,1], \\
& t_{\alpha, 2}(F)=F-e_{2}\left(F_{Y} \mid\right), \quad F \in \operatorname{Lip}_{0}[0,1],
\end{aligned}
where e_(1),e_(2)e_{1}, e_{2} are the selections defined by formulae (15) and (16) (with f=F^(')|_(Y)f=\left.F^{\prime}\right|_{Y} ). Then the selection
is homogeneous and continuous (according to assertion a of Theorem 3).
Since T_(alpha)(F)subeQ_(alpha)(F)T_{\alpha}(F) \subseteq Q_{\alpha}(F), for all F inLip_(0)[0,1]F \in \operatorname{Lip}_{0}[0,1], it follows that the selection t_(alpha)t_{\alpha} defined by (32) is a selection for Q_(alpha)Q_{\alpha}, too.
Remark 4. For alpha=1\alpha=1 one obtains P_(x^(_|_))=T_(1)=Q_(1)P_{x^{\perp}}=T_{1}=Q_{1} implying that T_(1)T_{1} and Q_(1)Q_{1} are single valued and therefore are linear and continuous applications from Lip_(0)[0,1]\operatorname{Lip}_{0}[0,1] to Y^(_|_)Y^{\perp}.
REFERENCES
Brown, B. M., Elliot, D. and D. F. Paget, Lipschitz constants for the Bernstein polynomials of a Lipschitz continuous functions, J. Approx. Theory 49 (1987), 196-199.
Cobzaş, S. and C. Mustăţa, Norm-Preserving Extension of Convex Lipschitz Functions, J. Approx. Theory 34 (1978), 236-244.
Czipser, J. and L. Géher, Extension of Functions satisfying a Lipschitz Condition, Acta Math. Acad. Sci. Hungar 6(1955), 213-220.
Deutsch, F., Li, W. and S. H. Park, Tielze Extensions and Conlinuous Selections for Metric Projections, J. Approx. Theory 63 (1991), 55-88.
Mc Shane, E. J., Extension of Range of Functions, Bull. Amer. Math. Soc. 40 (1934), 837-842.
Mustăţa, C., Best Approximation and Unique Extension of Lipschilz Funclions, J. Approx. Theory 19(1977), 222-230.
Mustăta, C., M-ideals in Metric Spaces, „Babeş-Bolyai" Universily, Fac. of Math. and Physics, Research Seminars, Seminar on Math. Analysis, Preprint № 7 (1988), 65-74.
Mustăţa, C., Selections Associated to Mc Shane's Extension Theorem for Lipschitz Functions, Revue d'Analyse Numérique et de Théorie de l'Approx. 21, 2 (1992), 135-145.
Received 2.IV,1993
Institutul de Calcul
Str. Republicii Nr. 37
Oficiul Postal 1
3400 Cluj-Napoca
România