[1] Alfsen, E.M., Effros, E., Structure in real Banach spaces, Ann. of Math. 96 (1972), 98-173.
[2] Azipser, J., Géher, L., Extension of Functions satisfying a Lipschitz conditions Acta Math. Acad. Sci. Hungar 6 (1955), 213-220.
[3] Deutsch, F.., Wu Li, Sung-Ho Park, Tietze Extensions and Conditions Selections for Metric Projecitons J.A.T. 64 (1991), 55-68.
[4] Fakhoury, H., SéIections linéaires associées au théorème de Hahn-Banach, J. Funct. Analysis 11 (7972),436-452.
[5] Mc shane; E, J., Extensions of range of functions, Bull, Amer. Math. Soc. 40 (1934), 837-842.
[6] Mustata C., Best Approximation and Unique Extension of Lipschilz Functions’J.A.T. 19 (1977), 222-230.
[7] Mustata, C., M- ideals in metric spaccs, “Babes-Bolyai,, University, Fac. of Math. and Physics, Research Seminars, Seminar on Math. Anal., preprint Nr. 7 (1988), 65-74.
8. Rudin W., Functional Analysis, McGraw-Hill 1973.
Paper (preprint) in HTML form
1992-Mustata-Selections associated to McShane’s extension-Jnaat
SELECTIONS ASSOCIATED TO McSHANE'S EXTENSION THEOREM FOR LIPSCHITZ FUNCTIONS
COSTICA MUSTATA(Cluj-Napocal)
Let ( X,dX, d ) be a metric space and YY a nonvoid subset of XX. A fung tion f:Y rarr Rf: Y \rightarrow R is called Lipschitz on bar(Y)\bar{Y} if there exists K >= 0K \geqslant 0. such that (1,(1)/(1))_(1),quad:'|f(x)-f(y)| <= Id(x,y)\left(1, \frac{1}{1}\right)_{1}, \quad \because|f(x)-f(y)| \leqslant I d(x, y),
for all x,y in Yx, y \in Y. A number K >= 0K \geqslant 0 verifying (1.1) is called a Lipschiti constant for ff.
It is easy to show that the quantity ||f||_(X)\|f\|_{X} defined by
{:(1.2)||f||_(X)=s u p{|f(x)-f(y)|//d(x","y):x","y in Y","x!=y}:}\begin{equation*}
\|f\|_{X}=\sup \{|f(x)-f(y)| / \mathrm{d}(x, y): x, y \in Y, x \neq y\} \tag{1.2}
\end{equation*}
is the smallest Lipschitz constant for ff and we shall call it the Lipschitz norm of ff.
Denote by Lip YY the set of all real-valued Lipsehitz functions on Y_(". ")Y_{\text {. }} The set Lip XX and the quantity ||f||_(x)\|f\|_{x} are defined similarly.
Obviously, with respect to the usual operations of addition and multiplication by scalars for functions, Lip YY and Lip XX are linear spaces and the functionals ||||_(Y):}\left\|\|_{Y}\right. and ||||_(X)\| \|_{X} are seminorms on these spaces. The functionals ||||_(r):}\left\|\|_{r}\right. and ||||_(x)\| \|_{x} are not norms because they vanish on constant functions.
Mc Shane [5] proved the following extension theorem for Lipschitz functions:
Theorem 1. Let (X,d)(X, d) be a metric space and O/_(!=Y sub X)\varnothing_{\neq Y \subset X}. Then every Lipschitz function on YY has a norm preserving extensions to XX, i.e. for every f inf \in Lip YY there exists F inF \in Lip XX such that
F|_(Y)=f" and "||F||_(x)=||f||_(Y).\left.F\right|_{Y}=f \text { and }\|F\|_{x}=\|f\|_{Y} .
Now let x_(0)in Yx_{0} \in Y be fixed and denote by Lip _(0)Y{ }_{0} Y the subspace of Lip YY formed of all functions in Lip YY vanishing at x_(0)x_{0}, i.e.
s. As ||F||_(x)=||f||_(y)\|F\|_{x}=\|f\|_{y}, for all F in E(f)F \in E(f), it follows that E(f)E(f) is bounded.
To prove the closedness of E(f)E(f), let (F_(n))\left(F_{n}\right) be a sequence in E(f)E(f) converging to a function F^(')in Lipp_(0)XF^{\prime} \in \operatorname{Lip} p_{0} X. As l^(')(x_(0))=0=F_(n)(x_(0))l^{\prime}\left(x_{0}\right)=0=F_{n}\left(x_{0}\right), for all n in Nn \in N, it follows that
for every x in Xx \in X, implying the pointwise convergence of the sequence (F_(n)(x))\left(F_{n}(x)\right) to F^(')(x)F^{\prime}(x), for all x in Xx \in X. Since F_(n)(x)=f(x)F_{n}(x)=f(x) for all x inT^(TT)x \in T^{\top} and all n in Nn \in N it. follows F(x)=f(x)F(x)=f(x), for all x in Yx \in Y. Also F_(n)rarrF^(2)F_{n} \rightarrow F^{2} in Lip _(0)X{ }_{0} X and ||F_(n)||_(x)=||f||_(x),n in N\left\|F_{n}\right\|_{x}=\|f\|_{x}, n \in N, imply ||F^(')||_(x)=lim_(n rarr oo)||F_(n)||_(x)=||f||_(y)\left\|F^{\prime}\right\|_{x}=\lim _{n \rightarrow \infty}\left\|F_{n}\right\|_{x}=\|f\|_{y} showing that F^(')inF(f)F^{\prime} \in \mathbb{F}(f); 2^(@)2^{\circ}. To prove the second inequality in (1.8) suppose, on the contrary, that there exists x_(1)in Xx_{1} \in X such that F^(')(x_(1)) > F_(2)(x_(1))F^{\prime}\left(x_{1}\right)>F_{2}\left(x_{1}\right). Since the functions FF, F_(2)F_{2} are continuous on XX and F^(')|_(X)=f=F_(2)|_(Y)\left.F^{\prime}\right|_{X}=f=\left.F_{2}\right|_{Y}, it follows that they agree on the closure bar(Y)\bar{Y} of YY. Therefore x_(1)in X\\ bar(Y)x_{1} \in X \backslash \bar{Y}, implying d(x_(1),y) > 0\mathrm{d}\left(x_{1}, y\right)>0 for all y in Yy \in Y. Taking into account definition (1.6) of F_(2)F_{2} and the inequality F_(2)(x_(1)) < F(x_(1))F_{2}\left(x_{1}\right)< F\left(x_{1}\right), there exists an element y_(1)in Yy_{1} \in Y such that
As f(y_(1))=F(y_(1))f\left(y_{1}\right)=F\left(y_{1}\right), the inequality (1.9), gives the contradiction ||f||_(r)=\|f\|_{r}= : =||F||_(X) >= (F^(')(x_(1))-F^(')(y_(1)))//d(x_(1),y_(1)) > ||f||_(Y)=\|F\|_{X} \geqslant\left(F^{\prime}\left(x_{1}\right)-F^{\prime}\left(y_{1}\right)\right) / \mathrm{d}\left(x_{1}, y_{1}\right)>\|f\|_{Y}.
The first inequality in (1,8) can be proved in a similar way. 3^(@)3^{\circ}. Let B_(Lip_(0)Y),B_(Lip_(0)x)B_{\mathrm{Lip}_{0} \mathrm{Y}}, B_{\mathrm{Lip}_{0} \mathrm{x}} be the closed unit balls of the spaces Lip _(0)Y{ }_{0} Y respectively Lip_(0)X\operatorname{Lip}_{0} X. Let ff be an extremal point of B_("Lipy ")B_{\text {Lipy }} and stippose that H_(1),H_(2)inB_("Lip ",x)H_{1}, H_{2} \in B_{\text {Lip }, \mathrm{x}} and lambda in(0,1)\lambda \in(0,1) are such that lambdaH_(1)+(1-lambda)H_(2)in E(f)\lambda H_{1}+(1-\lambda) H_{2} \in E(f). Since ff is extremal the equality lambdaH_(1)|_(Y)+(1-lambda)H_(2)|_(Y)=f\left.\lambda H_{1}\right|_{Y}+\left.(1-\lambda) H_{2}\right|_{Y}=f implies H_(1)|_(Y)==f=H_(2)|_(x)\left.H_{1}\right|_{Y}= =f=\left.H_{2}\right|_{x}. Also 1=||f||_(Y) <= ||lambdaH_(1)+(1-lambda)H_(2)||_(x) <= lambda||H_(x)||_(x)+(1quad lambda)||H_(2)||_(x) <= 11=\|f\|_{Y} \leqslant\left\|\lambda H_{1}+(1-\lambda) H_{2}\right\|_{x} \leqslant \lambda\left\|H_{x}\right\|_{x}+(1 \quad \lambda) \left\|H_{2}\right\|_{x} \leqslant 1 implies ||H_(1)||_(x)=||H_(2)||_(x)=1=||f||_(x)\left\|H_{1}\right\|_{x}=\left\|H_{2}\right\|_{x}=1=\|f\|_{x}, showing that H_(1),H_(2)in E(f)H_{1}, H_{2} \in E(f).
Now, suppose that f_(2)||f||_(Y)=1f_{2}\|f\|_{Y}=1, is not an extremal point of B_(Lip_(0)Y)B_{\mathrm{Lip}_{0} Y}. Then there exist two distinct elements f_(1),f_(2)f_{1}, f_{2} in B_("Lip "_(0))YB_{\text {Lip }_{0}} Y and lambda in(0,1)\lambda \in(0,1) such that f=lambdaf_(1)+(1-lambda)f_(2)f=\lambda f_{1}+(1-\lambda) f_{2}. If H_(i)in E(f_(i)),i=1,2H_{i} \in E\left(f_{i}\right), i=1,2, then H_(1)|_(x)+(1-lambda)H_(2)|_(x)==lambdaf_(1)+(1-lambda)f_(2)=f\left.H_{1}\right|_{x}+\left.(1-\lambda) H_{2}\right|_{x}= =\lambda f_{1}+(1-\lambda) f_{2}=f and 1=||f||_(x)=||lambdaH_(1)||_(x)+(1-lambda)H_(2)|_(x)||_(x)⩽⩽||lambdaH_(1)+(1-lambda)H_(2)||_(x) <= 11=\|f\|_{x}=\left\|\lambda H_{1}\right\|_{x}+\left.(1-\lambda) H_{2}\right|_{x} \|_{x} \leqslant \leqslant\left\|\lambda H_{1}+(1-\lambda) H_{2}\right\|_{x} \leqslant 1, showing that lambdaH_(1)+(1-lambda)H_(2)in E(f)\lambda H_{1}+(1-\lambda) H_{2} \in E(f). Since H_(i)|_(r)=f_(i)!=f,i=1,2\left.H_{i}\right|_{r}=f_{i} \neq f, i=1,2, it follows that H_(i)in E(f),i=1,2H_{i} \in E(f), i=1,2 and E(f)E(f) is not a face of B_(/_\p_(0)x)xB_{\triangle \mathrm{p}_{0} x} x.
Remark 1. The extensions F_(1),F_(2)F_{1}, F_{2} of xx function f in Lipp_(0)Yf \in \operatorname{Lip} p_{0} Y, given by (1.5)(1.5) and (1.6)(1.6), are extremal points of the face E(f)E(f) and consequently they are extremal elements of the unit ball B_(1+D_(0)x)B_{1+D_{0} x}, too. Indeed, if F,G∈∈B(f)F, G \in \in B(f) and lambda in(0,1)\lambda \in(0,1) are such that F_(1)=lambda F+(1-lambda)GF_{1}=\lambda F+(1-\lambda) G then, by (1.8), F_(1) <= FF_{1} \leqslant F and F_(1) <= GF_{1} \leqslant G implying F=F_(1)=GF=F_{1}=G. The extremality of F_(2)F_{2} is proved similarly.
{:(1.8)F_(1)(x) <= F(x) <= F_(2)(x)","x in X",":}\begin{equation*}
F_{1}(x) \leqslant F(x) \leqslant F_{2}(x), x \in X, \tag{1.8}
\end{equation*}
where F_(1),F_(2)F_{1}, F_{2} are the extensions of ff given by (1.5) and (1.6) ; 3^(@)3^{\circ} For ||f||_(Y)=1\|f\|_{Y}=1, the set E(f)E(f) is a face of the unit ball B_("Lip "_(X))B_{\text {Lip }{ }_{X}} if and only if ff is an extremal point of the unit ball of Lip _(0)Y{ }_{0} \mathrm{Y}.
Proof. 1^(@)1^{\circ}. Let F,G in E(f)F, G \in E(f) and lambda in[0,1]\lambda \in[0,1]. Obviously that (lambda F+(1--lambda)G)|_(X)=f(\lambda F+(1- -\lambda) G)\left.\right|_{X}=f and ||lambda F+(1-lambda)G||_(X) <= lambda||F||_(X)+(1-lambda)||G||_(X)==lambda||f^(')||_(X)+(1-lambda)||f||_(Y)=||f||_(Y)\|\lambda F+(1-\lambda) G\|_{X} \leqslant \lambda\|F\|_{X}+(1-\lambda)\|G\|_{X}= =\lambda\left\|f^{\prime}\right\|_{X}+(1-\lambda)\|f\|_{Y}=\|f\|_{Y}. Since ||lambda F+(1-lambda)G||_(X) >= ||(lambda F++(1-lambda)G|_(Y)||_(Y)=||f||_(Y)\|\lambda F+(1-\lambda) G\|_{X} \geqslant \|(\lambda F+ +\left.(1-\lambda) G\right|_{Y}\left\|_{Y}=\right\| f \|_{Y}, it follows that lambdaF^(')+(1-lambda)G in E(f)\lambda F^{\prime}+(1-\lambda) G \in E(f), proving the convexity of E(f)E(f).
A subset Lambda\Lambda of Lip _(0)X{ }_{0} X is called proximinal in Lip _(0)X{ }_{0} X if every F inLip_(0)XF \in \operatorname{Lip}_{0} X has a nearest point in Lambda\Lambda, i.e. there exists G in AG \in A such that ||F-G||_( bar(x))= ⇋d(F,Lambda)\|F-G\|_{\bar{x}}= \leftrightharpoons \mathrm{d}(F, \Lambda), where
d(F,Lambda)=i n f'{||F-H||_(X):H in Lambda}.\mathrm{d}(F, \Lambda)=\inf ^{\prime}\left\{\|F-H\|_{X}: H \in \Lambda\right\} .
The metric projection P_(Lambda):P_{\Lambda}: Lip _(0)X rarr2^(Lambda){ }_{0} X \rightarrow 2^{\Lambda} is defined by
P_(Lambda)(F)={G in Lambda:||F-G||_(x)=d(F,(A))}.P_{\Lambda}(F)=\left\{G \in \Lambda:\|F-G\|_{x}=\mathrm{d}(F, \mathrm{~A})\right\} .
If P_(Lambda)(F)P_{\Lambda}(F) is a singleton for every F inLip_(0)XF \in \operatorname{Lip}_{0} X then Lambda\Lambda is called a Ohebyshevian subset of: Lip _(0)X_{0} X.
There is a closed relation between the extension operator EE and the projection pperator P_(Y^(_|_))P_{Y^{\perp}}, ilustrated in the following theorem:
Theorem 3. The following assertions hold: 1^(@)1^{\circ} The subspace Y^(_|_)Y^{\perp} is proximinal in Lip _(0)X_{0} X; 2^(@)2^{\circ} The equality
is true for all F inLip_(0)XF \in \operatorname{Lip}_{0} X; 3^(@)3^{\circ} A function G in Y _|_G \in Y \perp is a best approximation element for FF if and only if there exists H in E(F|_(Y))H \in E\left(\left.F\right|_{Y}\right) such that G=F-HG=F-H, or equivalently
Proof. First we prove formula (2.1). If F inLip_(0)X_("then for any ")G inY^(_|_)F \in \operatorname{Lip}_{0} X_{\text {then for any }} G \in \mathbf{Y}^{\perp},
{:[||F|_(x)||_(Y)=s u p{|F^(')(y)-F(y^('))|//d(y,y^(')):y,y^(')in Y,y!=y^(')}=],[=s u p{|(F-G)(y)-(F-G)(y^('))|//d(y,y^(')):y,y^(')in Y,y!=y^(')}],[ <= s u p{|(F-G)(x)-(F-G)(x^('))|//d(x,x^(')):x,x^(')in X,x!=x^(')}],[=]:}\begin{aligned}
\left\|\left.F\right|_{x}\right\|_{Y} & =\sup \left\{\left|F^{\prime}(y)-F\left(y^{\prime}\right)\right| / \mathrm{d}\left(y, y^{\prime}\right): y, y^{\prime} \in Y, y \neq y^{\prime}\right\}= \\
& =\sup \left\{\left|(F-G)(y)-(F-G)\left(y^{\prime}\right)\right| / \mathrm{d}\left(y, y^{\prime}\right): y, y^{\prime} \in Y, y \neq y^{\prime}\right\} \\
& \leqslant \sup \left\{\left|(F-G)(x)-(F-G)\left(x^{\prime}\right)\right| / \mathrm{d}\left(x, x^{\prime}\right): x, x^{\prime} \in X, x \neq x^{\prime}\right\} \\
& =
\end{aligned}
implying that ||F|_(Y)||_(Y) <= d(F,Y^(_|_))\left\|\left.F\right|_{Y}\right\|_{Y} \leqslant \mathrm{~d}\left(F, Y^{\perp}\right).
On the other hand, by Theorem 1, there exists H inH \in Lip _(0)X{ }_{0} X such that H|_(Y)=F|_(Y)\left.H\right|_{Y}=\left.F\right|_{Y} and ||H||_(X)=||F^(Y)|_(Y)||_(Y)\|H\|_{X}=\left\|\left.F^{Y}\right|_{Y}\right\|_{Y}. It follows that FF. H inY^(_|_)H \in Y^{\perp} and : ||F^(Y)|_(Y)||_(Y)=||F-(F-H)_(x)|| >= i n f{||F-G||_(x):G inY^(_|_)}=d(F^('),Y _|_)\left\|\left.F^{Y}\right|_{Y}\right\|_{Y} =\left\|F-(F-H)_{x}\right\| \geqslant \inf \left\{\|F-G\|_{x}: G \in Y^{\perp}\right\}=\mathrm{d}\left(F^{\prime}, Y \perp\right), showing that ||F|_(Y)||_(Y)=d(F,Y^(_|_))\left\|\left.F\right|_{Y}\right\|_{Y}=\mathrm{d}\left(F, Y^{\perp}\right).
Assertion 3^(@)3^{\circ} and formula (2.2) follow from [6], Lemma 1, p. 223 and assertion 1^(@)1^{\circ} follows from 3^(@)3^{\circ}.
Now, by Theorem 2, 1^(@)1^{\circ}, the set P_(Y)(T)=F-E(F|_(Y))P_{Y}(T)=F-E\left(\left.F\right|_{Y}\right) is bounded, convex and closed, for any F inLip_(0)XF \in \operatorname{Lip}_{0} X.
We shall say that the set Y sub XY \subset X has the property (U)(U) if every f∈∈Lip_(0)Yf \in \in \operatorname{Lip}_{0} Y has a unique Lipschitz extension F^(')inLip_(0)XF^{\prime} \in \operatorname{Lip}_{0} X, i.e. E(f)E(f) is a singleton for every f in Lipp_(0)Yf \in \operatorname{Lip} p_{0} Y. By Theorem 3, the set YY has property (U)(U) if and only if Y^(_|_)Y^{\perp} is a Chebyshevian subspace of Lip_(0)X\operatorname{Lip}_{0} X.
2. A natural question is when have the set valued operators EE and P_(Y _|_)P_{Y \perp} continuous selections. If S:A rarr2^(B)S: A \rightarrow 2^{\mathrm{B}} is a set-valued application, a function s:A rarr Bs: A \rightarrow B is called a selection for SS if s(x)in S(x)s(x) \in S(x), for all x in Ax \in A.
In the following theorems, we shall prove the existence of continuous selections for the operators EE and P_(Y.L)P_{Y . L} in the particular case X=RX=R, with the usual distance d(x,y)=|x-y|\mathrm{d}(x, y)=|x-y|.
Theorem 4. Let X=R,Y=[a,b]in RX=R, Y=[a, b] \in R and x_(0)in[a,b]x_{0} \in[a, b] fixed. Then the extension operator E:Lip_(0)Y rarr2^(Lip_(0)R)E: \operatorname{Lip}_{0} Y \rightarrow 2^{\operatorname{Lip}_{0} R} has a continuous and positively homogeneous selection ee.
Proof. Define epsi_(2):Lip_(0)Y rarrLip_(0)R^(˙)\varepsilon_{2}: \operatorname{Lip}_{0} Y \rightarrow \operatorname{Lip}_{0} \dot{R} by
e_(2)(f)=F_(2),quad f inLip_(0)Ye_{2}(f)=F_{2}, \quad f \in \operatorname{Lip}_{0} Y
where F_(2)F_{2} is the maximal extension of ff given by (1.6).
If alpha > 0\alpha>0 then
{:[e_(2)(af)(x)=in{af(y)+||af||_(Y)|x-y|:y in[a,b]}],[=a:i n f{f(y)+||f||_(Y)|x-y|_(1):y in[a*b]}=],[=a*F_(2)(x)]:}\begin{aligned}
e_{2}(a f)(x) & =\operatorname{in}\left\{a f(y)+\|a f\|_{Y}|x-y|: y \in[a, b]\right\} \\
& =a: \inf \left\{f(y)+\|f\|_{Y}|x-y|_{1}: y \in[a \cdot b]\right\}= \\
& =a \cdot F_{2}(x)
\end{aligned}
showing that e_(2)e_{2} is positively homogeneous.
To prove the continuity of e_(2)e_{2} for epsi > 0\varepsilon>0, take delta=epsi//3\delta=\varepsilon / 3 and show that
for all f,g inLip_(0)Yf, g \in \operatorname{Lip}_{0} Y such that ||f-g||_(Y) < delta\|f-g\|_{Y}<\delta.
It is easy to check that the maximal extension F_(2)F_{2} of ff has the form
{:[F_(2)(x)=f(a)-||f||_(Y)(x-a)","" for "x < a],[=f(x)","quad" for "x in[a","b]],[=f(b)+||f||_(Y)(x-b)","" for "x > b.]:}\begin{aligned}
F_{2}(x) & =f(a)-\|f\|_{Y}(x-a), \text { for } x<a \\
& =f(x), \quad \text { for } x \in[a, b] \\
& =f(b)+\|f\|_{Y}(x-b), \text { for } x>b .
\end{aligned}
If g inLip_(0)Yg \in \operatorname{Lip}_{0} Y is such that ||f-g||_(Y) < delta=epsi//3\|f-g\|_{Y}<\delta=\varepsilon / 3, then similarly the maximal extension G_(2)G_{2} of gg has the expression:
{:[theta_(2)(x)=g(a)-||g||_(Y)(x-a)","" for "x < a],[=g(x)","quad" for "x in[a","b]],[=g(b)+||g||_(Y)(x-b)","" for "x > b.]:}\begin{aligned}
\theta_{2}(x) & =g(a)-\|g\|_{Y}(x-a), \text { for } x<a \\
& =g(x), \quad \text { for } x \in[a, b] \\
& =g(b)+\|g\|_{Y}(x-b), \text { for } x>b .
\end{aligned}
It follows that the difference H_(2)=F_(2)-G_(2)H_{2}=F_{2}-G_{2} has the expression :
{:[H_(2)(x)=f(a)-g(a)-(||f||_(Y)-||g||_(Y))(x-a)","" for "x < a],[=f(x)-g(x)","],[=f(b)-g(b)+(||f||_(Y)-||g||_(Y))(x-b)","" for "x > b]:}\begin{aligned}
H_{2}(x) & =f(a)-g(a)-\left(\|f\|_{Y}-\|g\|_{Y}\right)(x-a), \text { for } x<a \\
& =f(x)-g(x), \\
& =f(b)-g(b)+\left(\|f\|_{Y}-\|g\|_{Y}\right)(x-b), \text { for } x>b
\end{aligned}
We have to consider several cases:
(... Case 1^(@):.x,y > b,x!=y1^{\circ} \therefore x, y>b, x \neq y. In this case
ast Taking into account the inequalities (2.4)-(2.9), it follows that ||F_(2)-G_(2)||_(X)=||H_(2)||_(X)=s u p{|H_(2)(x)-H_(2)(y)|||x-y∣:x,y in R,x!=y} < epsi\left\|F_{2}-G_{2}\right\|_{X}=\left\|H_{2}\right\|_{X}=\sup \left\{\left|H_{2}(x)-H_{2}(y)\right|| | x-y \mid: x, y \in R, x \neq y\right\}<\varepsilon i.e.
Remark 2. The selection e_(1)(f)=F_(1)e_{1}(f)=F_{1}, where F_(1)F_{1} is the minimal extension of ff defined by (1.5) is also continuous and positively homogeneous. This can be proved directly or taking into account the equality
which holds for all f in Lip_(0)Yf \in \operatorname{Lip}{ }_{0} Y.
Combining these two results one obtains the following consequence:
Corollary 5. The extension operator E_("- Lip "0)Y rarr2^("Lip "0)E_{\text {- Lip } 0} Y \rightarrow 2^{\text {Lip } 0} R has a contimious and homogeneous selection given by
e(f)=(1//2)*(e_(1)(f)+e_(2)(f)),f inLip_(0)Y.e(f)=(1 / 2) \cdot\left(e_{1}(f)+e_{2}(f)\right), f \in \operatorname{Lip}_{0} Y .
Proof. Obviously that ee is continuous and positively homogeneous. On the other hand by (2.10), it follows
e(-f)≐-e(f),f inLip_(0)Y,e(-f) \doteq-e(f), f \in \operatorname{Lip}_{0} Y,
implying the homogeneity of e:e(alpha f)⇌alpha:e(f),alpha in R,f in Lip]_(0)Ie: e(\alpha f) \rightleftharpoons \alpha: e(f), \alpha \in R, f \in \operatorname{Lip}]_{0} \mathcal{I}.
3. This section is concerned with the existence of selections for the metric projection P_(Y _|_)P_{Y \perp} in the case X=>R,Y=[a,b],x_(0)in[a,b]X \Rightarrow R, Y=[a, b], x_{0} \in[a, b]. : usal selection pp : Lip X rarrY^(_|_)X \rightarrow Y^{\perp} of P_(Y_(_|_))P_{Y_{\perp}} is called additive modulo Y^(_|_)Y^{\perp}, provided:
formall e^(-1)p_(0)Xe^{-1} p_{0} X and G inY^(_|_)G \in Y^{\perp}.
Theorem 6. The metric projection P_(Y _|_)P_{Y \perp} has a homogeneous, additive modulo Y _|_Y \perp and continuous setection pp.
Proof. Let ee be the homogeneous and continuous selection of the extension operator EE, given in Corollary 5. Taking into account Theorem 3 and equality (2.10), define p:Lip_(0)X rarr Y _|_p: \operatorname{Lip}_{0} X \rightarrow Y \perp by the formula
p=I-e@r,p=I-e \circ r,
where r:Lip_(0)X rarrLip_(0)Yr: \operatorname{Lip}_{0} X \rightarrow \operatorname{Lip}_{0} Y is the restriction operator given by r(F)==F^(')|_(Y)r(F)= =\left.F^{\prime}\right|_{Y} and I:Lip_(0)X:Lip_(0)XI: \operatorname{Lip}_{0} X: \operatorname{Lip}_{0} X this the identity map. Then
Indeed F-e(F|_(Y))=(1//2)(F-F_(1))+(1//2)(F-F_(2))inP_(Y _|_)(F)F-e\left(\left.F\right|_{Y}\right)=(1 / 2)\left(F-F_{1}\right)+(1 / 2)\left(F-F_{2}\right) \in P_{Y \perp}(F); since the set P_(Y _|_)(F^('))P_{Y \perp}\left(F^{\prime}\right) is convex.
for all T in Lipp_(0)XT \in \operatorname{Lip} p_{0} X and G inY^(_|_)G \in Y^{\perp}, since for G inY^(_|_),P_(Y^(_|_))(G)={G}G \in Y^{\perp}, P_{Y^{\perp}}(G)=\{G\} and p(G)=Gp(G)=G
By (3.2), the selection pp is additive modulo Y _|_Y \perp and Theorem 6 is. proved.
It is easily seen that the kernel of P_(YL)P_{Y L},
Cozollary 7. For X=R,Y=[a;b]X=R, \mathrm{Y}=[a ; b] and x_(0)in[a;b]x_{0} \in[a ; b] the following. assertions are true:
(a) The extension operator Thes a linear and continuous setection;
(b) The metric projection P_(Y _|_)P_{Y \perp} has linear and continuous selection;
(c) There exists a subspace WW of the subspace Ker P_(Y _|_)P_{Y \perp} such that every T^(')inLip_(0)XT^{\prime} \in \mathrm{Lip}_{0} X can be uniquely represented in the form F=H+GF=H+G, with H inW^(TT),G inY^(_|_)H \in W^{\top}, G \in Y^{\perp}, i.e. the subspace Y^(_|_)Y^{\perp} is complemented in Lip _(0)X{ }_{0} X.
Proof. (a) We show that the application e:Lipp_(0)Psi rarre: \operatorname{Lip} p_{0} \Psi \rightarrow Lip p_(0)X_("defi- ")p_{0} X_{\text {defi- }} ned by
where H_(1),F_(2)H_{1}, F_{2} are the extremal extensions of ff given by (1,5)(1,5) and (1,6)(1,6), is linear and continuous selection of EE.
Writing explicitly ee we find that
{:[e(f)(x)=f(a)","" for "x < a],[=f(x)","" for "x in[a","b]],[=f(b)","" for "x > b]:}\begin{aligned}
e(f)(x) & =f(a), \text { for } x<a \\
& =f(x), \text { for } x \in[a, b] \\
& =f(b), \text { for } x>b
\end{aligned}
for any f inLip_(0)Yf \in \operatorname{Lip}_{0} \mathbf{Y}. Obviously e(alpha f+beta g)(x)=alpha*e(f)(x)+beta*e(g)(x)e(\alpha f+\beta g)(x)=\alpha \cdot e(f)(x)+\beta \cdot e(g)(x) for all x in Rx \in R and all f,g in Liptheta_(theta)Y,alpha,beta in Rf, g \in \operatorname{Lip} \theta_{\theta} Y, \alpha, \beta \in R, showing that ee is linear. The continuity of eiwas proved in Corollary 5.
(b) The application p:Lip_(0)Y rarrY^(_|_)p: \operatorname{Lip}_{0} Y \rightarrow Y^{\perp} defined, for F in Lipp_(0)XF \in \operatorname{Lip} p_{0} X, by
it follows that W sub KerP_(Y _|_)W \subset \operatorname{Ker} P_{Y \perp}.
For F in Lipp_(0)XF \in \operatorname{Lip} p_{0} X define
{:[(3.7)G(x)=0","" for "x in[a","b]],[=F^(')(x)-F(a)","" for "x < a],[=F^(')(x)-F(b)","" for "x > b]:}\begin{align*}
G(x) & =0, \text { for } x \in[a, b] \tag{3.7}\\
& =F^{\prime}(x)-F(a), \text { for } x<a \\
& =F^{\prime}(x)-F(b), \text { for } x>b
\end{align*}
and
{:[(3.8)H(x)= bar(F)(x)","" for "x in[a","b]],[=F(a)","" for "x < a],[=F(b)","" for "x > b]:}\begin{align*}
H(x) & =\bar{F}(x), \text { for } x \in[a, b] \tag{3.8}\\
& =F(a), \text { for } x<a \\
& =F(b), \text { for } x>b
\end{align*}
Then F=H+G,G inY^(_|_)F=H+G, G \in Y^{\perp} and H=e(F|_(Y))in WH=e\left(\left.F\right|_{Y}\right) \in W.
To prove that Lip_(0)X\operatorname{Lip}_{0} X is the topological direct sum of the subspaces Y^(_|_)Y^{\perp} and WW, it is sufficient to prove that the projection operator on bar(X)^(_|_)\bar{X}^{\perp} is continuous, i.e. that the application F rarr GF \rightarrow G, where G in Y _|_G \in Y \perp is the function defined in (3.7) is a linear and continuous operator from Lip_(0)X\operatorname{Lip}_{0} X to Y^(_|_)Y^{\perp}.
寻 1 The linearity is obvious. To prove the continuity suppose that F_(n),T_(0),FF_{n}, T_{0}, F in Lip_(0)X\operatorname{Lip}_{0} X, i.e. ||F_(n)-F||_(x)rarr0\left\|F_{n}-F\right\|_{x} \rightarrow 0. Then
for all x,y in Rx, y \in R, implying ||D_(n)||_(X) <= 2||F_(n)-F||_(X)rarr0\left\|D_{n}\right\|_{X} \leqslant 2\left\|F_{n}-F\right\|_{X} \rightarrow 0.
It follows that F_(n)rarr FF_{n} \rightarrow F implies G_(n)rarr GG_{n} \rightarrow G, showing that the projection operator on Y^(_|_)Y^{\perp} is continuous and consequently L^(1)p_(0)XL^{1} p_{0} X is the direct sum of Y^(i)Y^{i} and WW. Corollary 7 is completely proved.
: : Remarks 3. (a) In the considered case (X=R,(Y^(˙))=[a,b],x_(0)in:}in[a,b])\left(X=R, \dot{Y}=[a, b], x_{0} \in\right. \in[a, b]), we have e_(1)(f)!=e_(2)(f)e_{1}(f) \neq e_{2}(f), for all f inLip_(0)Y,f!=0f \in \operatorname{Lip}_{0} Y, f \neq 0. Tn fact e_(1)(f) < e(f)<<e_(2)(f),fLip_(0)Ye_{1}(f)<e(f)< <e_{2}(f), f \operatorname{Lip}_{0} Y.
(b) Let X=[0,1]X=[0,1] with d(x,y)=|x-y|,Y={0,1}\mathrm{d}(x, y)=|x-y|, Y=\{0,1\} and x_(0)=0x_{0}=0 Then e_(1)(f)=e_(2)(f)=e(f)e_{1}(f)=e_{2}(f)=e(f), for all f inLip_(0)Yf \in \operatorname{Lip}_{0} Y. It follows that Y^(_|_)={H^(')inLip_(0)X:}:F(0)=F(1)=0}Y^{\perp}=\left\{H^{\prime} \in \operatorname{Lip}_{0} X\right. : F(0)=F(1)=0\} is a Chebyshevian subspace of Lip _(0)X{ }_{0} X. In this case Lip_(0)X=KerP_(Y _|_)o+Y^(_|_)\operatorname{Lip}_{0} X=\operatorname{Ker} P_{Y \perp} \oplus Y^{\perp} and the extension operator EE and the metric projection P_(Y^(_|_))P_{Y^{\perp}} are linear and single valued.
REFERENCES
Alfsen, E. M., Effross, E., Stutucture in real Banach spaces, Ann. of Math. 96 (1972),98-173. 2. Czipser, J., Gêher, L., Extension of Functions sotisfying a Lipschitz.conditions. Acta Math. Acad. Sci. Hungạr 6 (1955), 213-220.
Deutsch, F., Wu Li, Sung-Ho Park, Tiedze Explensions and Continuous Selections for Metric Projections. J.A.T. 64 (1991), 55-68.
Fakhoury, H., Sélections linéaires associées au théorème de Hahn-Banach, J. Funct. Analysis 11 (1972), 436-452.
Mc Shane, E. J., Extension of range of functions, Bull. Amer. Math. Soc. 40 (1934), 837-842.
Mustăta, C., Best Approximation and Unique Extension of Lipschitz Functions, J.A.T. 19 (1977), 222-230.
Mustăja, C., M-ideals in metric spaces, "Babeş-Bolyai" University, Fac. of Math. and Physics, Research Seminars, Seminar on Math. Anal., Preprint Nr. 7, 1988, 65-74. 8. Rudin W., Functional Analysis, McGraw-Hill 1973.
Received 15.V. 1992
Institutul de Calcul
Oficiul Poștal 1 C.P. 68