The extension of starshaped bounded Lipschitz functions

Abstract

Authors

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

Keywords

Paper coordinates

C. Mustăţa, The extension of starshaped bounded Lipschitz functions, Anal. Numer. Théor. Approx. 9 (1980) 1, 93-99

PDF

About this paper

Journal

Revue d’Analyse Numer. Theor.Approximation

Publisher Name

Publishing Romanian Academy

Print ISSN

2457-6794

Online ISSN

2501-059X

google scholar link

[1] Cobzas, S., Mustata, C., Norm Preserving Extension of Convex Lipschitz Functions, J. Approx. Theory 24 (1978), 555-564.
[2] Holmes, R.B., A Course on Optimization and Best Approximation, Lectures Notes in Math. No. 257, Springer Verlag, Berlin-Heidelberg-New York, 1972.
[3] Johnson, J.A., Banach Spaces of Lipschitz Functions and Vector-Valued Lipschitz Functions, Trans. Amer. Math. Soc. 148 (1970), 147-169.
[4] McShane, R.J., Extension of Range of Functions, Bull. Amer. Math. Soc. 40 (1934), 837-842.
[5] Musata, C., Best Approximation and Unique Extension of Lipschitz Functions, H. Approx. Theory 19 (1977), 222-230.
[6] Mustata, C., A Characterization of Chebyshevian Subspace of Y^{}-Type, Mathematica – Revue Anal. Num. Teor. Approx., L’Analyse Num. Teor. approx. 6, 1 (1977), 51-56.
[7] Mustata, C., Norm Preserving Extension of Starshaped Lipschitz Functions, Mathematica 19 (42) 2 (1977), 183-187.
[8] Phelps, R.R., Uniqueness of Hahn-Banach Extension and Unique Best Approximation, Trans. Amer. Math. Soc. 95 (1960), 238-255.

Paper (preprint) in HTML form

1980-Mustata-The extension of starshaped bounded Lipschitz functions-Jnaat

THE EXTENSION OF STARSHAPED BOUNDED LIPSCHITZ FUNCTIONS

byc. MUSTATA

  1. Let X X XXX be a normed space and Y Y YYY a nonvoid subset of X X XXX. The set Y Y YYY is called starshaped (with respect to θ X θ X theta in X\theta \in XθX ) if α y Y α y Y alpha y in Y\alpha y \in YαyY for all α [ 0 , 1 ] α [ 0 , 1 ] alpha in[0,1]\alpha \in[0,1]α[0,1] and y Y y Y y in Yy \in YyY. A real function f f fff defined on Y Y YYY is called starshaped if
(1.1) f ( α y ) α f ( y ) , (1.1) f ( α y ) α f ( y ) , {:(1.1)f(alpha y) <= alpha f(y)",":}\begin{equation*} f(\alpha y) \leq \alpha f(y), \tag{1.1} \end{equation*}(1.1)f(αy)αf(y),
for all α [ 0 , 1 ] α [ 0 , 1 ] alpha in[0,1]\alpha \in[0,1]α[0,1] and y Y y Y y in Yy \in YyY. From (1.1) follows f ( θ ) 0 f ( θ ) 0 f(theta) <= 0f(\theta) \leq 0f(θ)0. In the following we consider only starshaped functions vanishing at θ . Y θ . Y theta.Y\theta . Yθ.Y will denote always a starshaped set.
A function f : Y R f : Y R f:Y rarr Rf: Y \rightarrow Rf:YR is called Lipschitz on Y Y YYY if there exists K 0 K 0 K >= 0K \geq 0K0 such that
(1.2) | f ( x ) f ( y ) | K x y (1.2) | f ( x ) f ( y ) | K x y {:(1.2)|f(x)-f(y)| <= K*||x-y||:}\begin{equation*} |f(x)-f(y)| \leq K \cdot\|x-y\| \tag{1.2} \end{equation*}(1.2)|f(x)f(y)|Kxy
for all x , y Y x , y Y x,y in Yx, y \in Yx,yY. Denote by Lip 0 Y Lip 0 Y Lip_(0)Y\operatorname{Lip}_{0} YLip0Y ( Lip 0 X Lip 0 X Lip_(0)X\operatorname{Lip}_{0} XLip0X ) the space of all Lipschitz functions on Y Y YYY (respectively on X X XXX ) vanishing at θ Y X θ Y X theta in Y sub X\theta \in Y \subset XθYX ([5]) and by BLip 0 Y BLip 0 Y BLip_(0)Y\mathrm{BLip}_{0} YBLip0Y ( BLip 0 X BLip 0 X BLip_(0)X\mathrm{BLip}_{0} XBLip0X ) their subspaces formed of bounded functions on Y Y YYY (respectively on X X XXX ). Denote also
(1.3) BS Y = { f : f BLip 0 Y , f , is starshaped } (1.4) BS X = { F : F BLip 0 X , F is starshaped } (1.3) BS Y = f : f BLip 0 Y , f ,  is starshaped  (1.4) BS X = F : F BLip 0 X , F  is starshaped  {:[(1.3)BS_(Y)={f:f inBLip_(0)Y,f," is starshaped "}],[(1.4)BS_(X)={F:F inBLip_(0)X,F" is starshaped "}]:}\begin{align*} & \mathrm{BS}_{Y}=\left\{f: f \in \mathrm{BLip}_{0} Y, f, \text { is starshaped }\right\} \tag{1.3}\\ & \mathrm{BS}_{X}=\left\{F: F \in \mathrm{BLip}_{0} X, F \text { is starshaped }\right\} \tag{1.4} \end{align*}(1.3)BSY={f:fBLip0Y,f, is starshaped }(1.4)BSX={F:FBLip0X,F is starshaped }
The sets BS Y BS Y BS_(Y)\mathrm{BS}_{Y}BSY and BS X BS X BS_(X)\mathrm{BS}_{X}BSX are convex cones in BLip 0 Y BLip 0 Y BLip_(0)Y\mathrm{BLip}_{0} YBLip0Y and in BLip 0 X BLip 0 X BLip_(0)X\mathrm{BLip}_{0} XBLip0X, respectively, i.e. f + g f + g f+gf+gf+g and λ f λ f lambda f\lambda fλf are in BS Y BS Y BS_(Y)\mathrm{BS}_{Y}BSY (in BS X BS X BS_(X)\mathrm{BS}_{X}BSX ) for all f , g f , g f,gf, gf,g in BS Y BS Y BS_(Y)\mathrm{BS}_{Y}BSY (in BS X BS X BS_(X)\mathrm{BS}_{X}BSX ) and λ 0 λ 0 lambda >= 0\lambda \geq 0λ0.
For f BS Y ( f BLip 0 Y ) f BS Y f BLip 0 Y f inBS_(Y)(f inBLip_(0)Y)f \in \mathrm{BS}_{Y}\left(f \in \mathrm{BLip}_{0} Y\right)fBSY(fBLip0Y) put
(1.5)) f d = sup { | f ( y 1 ) f ( y 2 ) | / y 1 y 2 : y 1 , y 2 Y , y 1 y 2 } (1.5)) f d = sup f y 1 f y 2 / y 1 y 2 : y 1 , y 2 Y , y 1 y 2 {:(1.5))||f||_(d)=s u p{|f(y_(1))-f(y_(2))|//||y_(1)-y_(2)||:y_(1),y_(2)in Y,y_(1)!=y_(2)}:}\begin{equation*} \|f\|_{d}=\sup \left\{\left|f\left(y_{1}\right)-f\left(y_{2}\right)\right| /\left\|y_{1}-y_{2}\right\|: y_{1}, y_{2} \in Y, y_{1} \neq y_{2}\right\} \tag{1.5)} \end{equation*}(1.5))fd=sup{|f(y1)f(y2)|/y1y2:y1,y2Y,y1y2}
the Lipschitz norm of f f fff, and
(1.6) f = sup { | f ( y ) | : y Y } , (1.6) f = sup { | f ( y ) | : y Y } , {:(1.6)||f||_(oo)=s u p{|f(y)|:y in Y}",":}\begin{equation*} \|f\|_{\infty}=\sup \{|f(y)|: y \in Y\}, \tag{1.6} \end{equation*}(1.6)f=sup{|f(y)|:yY},
the uniform norm of f f fff.
For F BLip 0 X F BLip 0 X F inBLip_(0)XF \in \mathrm{BLip}_{0} XFBLip0X, the Lipschitz and the uniform norms are defined similarly.
2. It is well known (see e.g. [4]) that a Lipschitz function f f fff defined on a nonvoid subset Y Y YYY of a metric space X X XXX has a norm preserving Lipschitz extension F F FFF on X X XXX, i.e. F | Y = f F Y = f F|_(Y)=f\left.F\right|_{Y}=fF|Y=f and F d = f d F d = f d ||F||_(d)=||f||_(d)\|F\|_{d}=\|f\|_{d}Fd=fd. In [1] it was shown that if Y is a nonvoid convex subset of a normed space X X XXX, then every convex Lipschitz function on Y Y YYY has a convex norm preserving Lipschitz extension F F FFF on X X XXX. In a similar result was established for starshaped Lipschitz function [7].
This Note is concerned with the problem of extension of bounded starshaped Lipschitz functions, i.e. for f BS Y f BS Y f inBS_(Y)f \in \mathrm{BS}_{Y}fBSY find F BS X F BS X F inBS_(X)F \in \mathrm{BS}_{X}FBSX such that F | Y = f , F d = f d F Y = f , F d = f d F|_(Y)=f,||F||_(d)=||f||_(d)\left.F\right|_{Y}=f,\|F\|_{d}=\|f\|_{d}F|Y=f,Fd=fd and F = f F = f ||F||_(oo)=||f||_(oo)\|F\|_{\infty}=\|f\|_{\infty}F=f. The function F F FFF is called briefly an extension of f f fff. We consider the problem of existence and unicity of such an extension.
Remark, that similar problem for bounded convex Lipschitz functions has a trivial answer: if Y Y YYY is a convex subset of X X XXX, such that θ Y θ Y theta in Y\theta \in YθY, then the nul function is the only bounded convex (Lipschitz) function on Y Y YYY which has a bounded convex Lipschitz extension on X X XXX. This follows from the fact that the constant functions are only bounded convex functions on X X XXX. Indeed, if F : X R F : X R F:X rarr RF: X \rightarrow RF:XR is a nonconstant there X X XXX. X X XXX unction, then there exist two points x 0 , x 1 X x 0 , x 1 X x_(0),x_(1)in Xx_{0}, x_{1} \in Xx0,x1X sich that F ( x 0 ) F ( x 1 ) F x 0 F x 1 F(x_(0))!=F(x_(1))F\left(x_{0}\right) \neq F\left(x_{1}\right)F(x0)F(x1), say F ( x 0 ) << F ( x 1 ) F x 0 << F x 1 F(x_(0))<<F(x_(1))F\left(x_{0}\right)< <F\left(x_{1}\right)F(x0)<<F(x1). But then, since the function φ : ( 0 , ) R φ : ( 0 , ) R varphi:(0,oo)rarr R\varphi:(0, \infty) \rightarrow Rφ:(0,)R, defined by
φ ( t ) = F ( x 0 + t ( x 1 x 0 ) ) F ( x 0 ) t , t > 0 φ ( t ) = F x 0 + t x 1 x 0 F x 0 t , t > 0 varphi(t)=(F(x_(0)+t(x_(1)-x_(0)))-F(x_(0)))/(t),t > 0\varphi(t)=\frac{F\left(x_{0}+t\left(x_{1}-x_{0}\right)\right)-F\left(x_{0}\right)}{t}, t>0φ(t)=F(x0+t(x1x0))F(x0)t,t>0
is nondecreasing (see [2] p. 17), it follows that
F ( x 0 + t ( x 1 x 0 ) ) F ( x 0 ) t F ( x 1 ) F ( x 0 ) > 0 F x 0 + t x 1 x 0 F x 0 t F x 1 F x 0 > 0 (F(x_(0)+t(x_(1)-x_(0)))-F(x_(0)))/(t) >= F(x_(1))-F(x_(0)) > 0\frac{F\left(x_{0}+t\left(x_{1}-x_{0}\right)\right)-F\left(x_{0}\right)}{t} \geq F\left(x_{1}\right)-F\left(x_{0}\right)>0F(x0+t(x1x0))F(x0)tF(x1)F(x0)>0
so that F ( x 0 + t ( x 1 x 0 ) ) F ( x 0 ) + t [ F ( x 1 ) F ( x 0 ) ] F x 0 + t x 1 x 0 F x 0 + t F x 1 F x 0 F(x_(0)+t(x_(1)-x_(0))) >= F(x_(0))+t[F(x_(1))-F(x_(0))]F\left(x_{0}+t\left(x_{1}-x_{0}\right)\right) \geq F\left(x_{0}\right)+t\left[F\left(x_{1}\right)-F\left(x_{0}\right)\right]F(x0+t(x1x0))F(x0)+t[F(x1)F(x0)], for all t 1 t 1 t >= 1t \geq 1t1, which shows that the function F F FFF is unbounded.
2a. The existence of extension. In Theorem 1 below will be shown that under some suplementary hypotheses on the function f BS Y f BS Y f inBS_(Y)f \in \mathrm{BS}_{Y}fBSY there exists an extension F BS X F BS X F inBS_(X)F \in \mathrm{BS}_{X}FBSX of f f fff.
Firstly, we prove two lemmas.
lemma 1. Let X X XXX be a real normed space and f f fff a starshaped function on X X XXX. Then, for every x X , x θ x X , x θ x in X,x!=thetax \in X, x \neq \thetaxX,xθ, the function Ψ : ( 0 , ) R Ψ : ( 0 , ) R Psi:(0,oo)rarr R\Psi:(0, \infty) \rightarrow RΨ:(0,)R, defined by
(2.1)
Ψ ( t ) = f ( t x ) / t , t > 0 Ψ ( t ) = f ( t x ) / t , t > 0 Psi(t)=f(tx)//t,quad t > 0\Psi(t)=f(t x) / t, \quad t>0Ψ(t)=f(tx)/t,t>0
is nondecreasing.
Proof. For 0 < t 1 < t 2 0 < t 1 < t 2 0 < t_(1) < t_(2)0<t_{1}<t_{2}0<t1<t2 and a fixed x X , x θ x X , x θ x in X,x!=thetax \in X, x \neq \thetaxX,xθ, we have
f ( t 1 x ) t 1 = f ( ( t 1 / t 2 ) t 2 x ) t 1 ( t 1 / t 2 ) f ( t 2 x ) t 1 = f ( t 2 x ) t 2 . f t 1 x t 1 = f t 1 / t 2 t 2 x t 1 t 1 / t 2 f t 2 x t 1 = f t 2 x t 2 . (f(t_(1)x))/(t_(1))=(f((t_(1)//t_(2))t_(2)x))/(t_(1)) <= ((t_(1)//t_(2))f(t_(2)x))/(t_(1))=(f(t_(2)x))/(t_(2)).\frac{f\left(t_{1} x\right)}{t_{1}}=\frac{f\left(\left(t_{1} / t_{2}\right) t_{2} x\right)}{t_{1}} \leq \frac{\left(t_{1} / t_{2}\right) f\left(t_{2} x\right)}{t_{1}}=\frac{f\left(t_{2} x\right)}{t_{2}} .f(t1x)t1=f((t1/t2)t2x)t1(t1/t2)f(t2x)t1=f(t2x)t2.
Now, for a function f : X R f : X R f:X rarr Rf: X \rightarrow Rf:XR define, as usually, the epigraph of f f fff, by
(2.2) epi f = { ( x , α ) X × R : f ( x ) α } (2.2)  epi  f = { ( x , α ) X × R : f ( x ) α } {:(2.2)" epi "f={(x","alpha)in X xx R:f(x) <= alpha}:}\begin{equation*} \text { epi } f=\{(x, \alpha) \in X \times R: f(x) \leq \alpha\} \tag{2.2} \end{equation*}(2.2) epi f={(x,α)X×R:f(x)α}
LEMMA 2. A function f : X R , f ( θ ) = 0 f : X R , f ( θ ) = 0 f:X rarr R,f(theta)=0f: X \rightarrow R, f(\theta)=0f:XR,f(θ)=0, is starshaped if and only if its epigraph is starshaped.
Proof. If f f fff is starshaped, f ( θ ) = 0 f ( θ ) = 0 f(theta)=0f(\theta)=0f(θ)=0, and ( x , α ) ( x , α ) (x,alpha)in(x, \alpha) \in(x,α) epi f f fff, then for every λ [ 0 , 1 ] , f ( λ x ) λ f ( x ) λ α λ [ 0 , 1 ] , f ( λ x ) λ f ( x ) λ α lambda in[0,1],f(lambda x) <= lambda f(x) <= lambda alpha\lambda \in[0,1], f(\lambda x) \leq \lambda f(x) \leq \lambda \alphaλ[0,1],f(λx)λf(x)λα, so that λ ( x , α ) = ( λ x , λ α ) epi f λ ( x , α ) = ( λ x , λ α ) epi f lambda(x,alpha)=(lambda x,lambda alpha)in epi f\lambda(x, \alpha)=(\lambda x, \lambda \alpha) \in \operatorname{epi} fλ(x,α)=(λx,λα)epif. Conversely, if epi f f fff is starshaped then ( x , f ( x ) ) epi f ( x , f ( x ) ) epi f (x,f(x))in epi f(x, f(x)) \in \operatorname{epi} f(x,f(x))epif implies ( λ x , λ f ( x ) ) ∈∈ ( λ x , λ f ( x ) ) ∈∈ (lambda x,lambda f(x))∈∈(\lambda x, \lambda f(x)) \in \in(λx,λf(x))∈∈ epi f f fff, i.e. f ( λ x ) λ f ( x ) f ( λ x ) λ f ( x ) f(lambda x) <= lambda f(x)f(\lambda x) \leq \lambda f(x)f(λx)λf(x), for all λ [ 0 , 1 ] λ [ 0 , 1 ] lambda in[0,1]\lambda \in[0,1]λ[0,1].
THEOREM 1. Let X X XXX be a normed space, Y Y YYY a starshaped subset of X X XXX, θ Y θ Y theta in Y\theta \in YθY, and f BS Y f BS Y f inBS_(Y)f \in \mathrm{BS}_{Y}fBSY. Then, there exists F BS X F BS X F inBS_(X)F \in \mathrm{BS}_{X}FBSX such that
(i) F | Y = f F Y = f F|_(Y)=f\left.F\right|_{Y}=fF|Y=f,
(ii) F d = f d F d = f d ||F||_(d)=||f||_(d)\|F\|_{d}=\|f\|_{d}Fd=fd,
(iii) F = f F = f ||F||_(oo)=||f||_(oo)\|F\|_{\infty}=\|f\|_{\infty}F=f,
if and only if f ( y ) 0 f ( y ) 0 f(y) <= 0f(y) \leq 0f(y)0, for all y Y y Y y in Yy \in YyY.
Proof. Let f BS Y f BS Y f inBS_(Y)f \in \mathrm{BS}_{Y}fBSY and suppose f 0 f 0 f <= 0f \leq 0f0 on Y Y YYY. Define G : X R G : X R G:X rarr RG: X \rightarrow RG:XR by
(2.3) G ( x ) = inf y Y [ f ( y ) + f d x y ] (2.3) G ( x ) = inf y Y f ( y ) + f d x y {:(2.3)G(x)=i n f_(y in Y)[f(y)+||f||_(d)||x-y||]:}\begin{equation*} G(x)=\inf _{y \in Y}\left[f(y)+\|f\|_{d}\|x-y\|\right] \tag{2.3} \end{equation*}(2.3)G(x)=infyY[f(y)+fdxy]
The function G G GGG defined by (2.3) is starshaped and satisfies G | Y = f G Y = f G|_(Y)=f\left.G\right|_{Y}=fG|Y=f, G d = f d G d = f d ||G||_(d)=||f||_(d)\|G\|_{d}=\|f\|_{d}Gd=fd (see [7]).
Let
(2.4) F ( x ) = { 0 if G ( x ) > 0 G ( x ) if G ( x ) 0 (2.4) F ( x ) = 0  if  G ( x ) > 0 G ( x )  if  G ( x ) 0 {:(2.4)F(x)={[0," if "G(x) > 0],[G(x)," if "G(x) <= 0]:}:}F(x)= \begin{cases}0 & \text { if } G(x)>0 \tag{2.4}\\ G(x) & \text { if } G(x) \leq 0\end{cases}(2.4)F(x)={0 if G(x)>0G(x) if G(x)0
Since G | Y = f 0 G Y = f 0 G|_(Y)=f <= 0\left.G\right|_{Y}=f \leq 0G|Y=f0, it follows F | Y = f F Y = f F|_(Y)=f\left.F\right|_{Y}=fF|Y=f and F d = f d F d = f d ||F||_(d)=||f||_(d)\|F\|_{d}=\|f\|_{d}Fd=fd. Obviously F f F f ||F||_(oo) >= ||f||_(oo)\|F\|_{\infty} \geq\|f\|_{\infty}Ff. If x X x X x in Xx \in XxX is such that G ( x ) 0 G ( x ) 0 G(x) <= 0G(x) \leq 0G(x)0, then
0 F ( x ) = G ( x ) f ( y ) + f d x y f ( y ) , 0 F ( x ) = G ( x ) f ( y ) + f d x y f ( y ) , 0 >= F(x)=G(x) >= f(y)+||f||_(d)*||x-y|| >= f(y),0 \geq F(x)=G(x) \geq f(y)+\|f\|_{d} \cdot\|x-y\| \geq f(y),0F(x)=G(x)f(y)+fdxyf(y),
so that 0 F ( x ) f ( y ) 0 F ( x ) f ( y ) 0 <= -F(x) <= -f(y)0 \leq-F(x) \leq-f(y)0F(x)f(y), for all y Y y Y y in Yy \in YyY. Therefore F = inf x X ( F ( x ) ) ≤≤ f F = inf x X ( F ( x ) ) ≤≤ f ||F||_(oo)=i n f_(x in X)(-F(x))≤≤||f||_(oo)\|F\|_{\infty}=\inf _{x \in X}(-F(x)) \leq \leq\|f\|_{\infty}F=infxX(F(x))≤≤f and F = f F = f ||F||_(oo)=||f||_(oo)\|F\|_{\infty}=\|f\|_{\infty}F=f. Since the epigraph of F F FFF is starshaped, by Lemma 2 , F 2 , F 2,F2, F2,F is starshaped. Consequently, F F FFF is the required extension of f f fff.
Suppose now, that there exists y 0 Y y 0 Y y_(0)in Yy_{0} \in Yy0Y such that f ( y 0 ) > 0 f y 0 > 0 f(y_(0)) > 0f\left(y_{0}\right)>0f(y0)>0, and let F F FFF be a starshaped extension of f f fff. By Lemma 1
0 < f ( y 0 ) = F ( y 0 ) ( F ( t y 0 ) / t ) , 0 < f y 0 = F y 0 F t y 0 / t , 0 < f(y_(0))=F(y_(0)) <= (F(ty_(0))//t),0<f\left(y_{0}\right)=F\left(y_{0}\right) \leq\left(F\left(t y_{0}\right) / t\right),0<f(y0)=F(y0)(F(ty0)/t),
so that F ( t y 0 ) t f ( y 0 ) F t y 0 t f y 0 F(ty_(0)) >= tf(y_(0))F\left(t y_{0}\right) \geq t f\left(y_{0}\right)F(ty0)tf(y0), for all t 1 t 1 t >= 1t \geq 1t1, which shows that the function F F FFF is unbounded. Therefore f f fff has no bounded strashaped extension, which ends the proof of Theorem 1.
Let
(2.5)
BS S Y = { f BS Y : f 0 } BS X = { F BS X : F 0 } BS S Y = f BS Y : f 0 BS X = F BS X : F 0 {:[BSS_(Y)={f inBS_(Y):f <= 0}],[BS_(X)^(-)={F inBS_(X):F <= 0}]:}\begin{gathered} \mathrm{BS} S_{Y}=\left\{f \in \mathrm{BS}_{Y}: f \leq 0\right\} \\ \mathrm{BS}_{X}^{-}=\left\{F \in \mathrm{BS}_{X}: F \leq 0\right\} \end{gathered}BSSY={fBSY:f0}BSX={FBSX:F0}
By Theorem 1 follows:
corolyary 1. Every function f BS Y ¯ f BS Y ¯ f inBS_( bar(Y))f \in \mathrm{BS}_{\bar{Y}}fBSY¯ has an extension F BS X ¯ F BS X ¯ F inBS_( bar(X))F \in \mathrm{BS}_{\bar{X}}FBSX¯.
2b. The unicity of extension. By Theorem 1 and Corollary 1, every nonpositive bounded starshaped function, defined on a starshaped subset Y Y YYY of a normed space X X XXX, has a nonpositive bounded starshaped extension to whole X X XXX. Furthemore, these are the only bounded starshaped function on Y Y YYY admitting bounded starshaped Lipschitz extension on X X XXX.
Equiped with the norms
f Y = max ( f d , f ) , f BLip 0 Y (2.6) F X = max ( F d , F ) , F BLip 0 X f Y = max f d , f , f BLip 0 Y (2.6) F X = max F d , F , F BLip 0 X {:[||f||_(Y)=max(||f||_(d),:}{:||f||_(oo))","quad f inBLip_(0)Y],[(2.6)||F||_(X)=max(||F||_(d),:}{:||F||_(oo))","quad F inBLip_(0)X]:}\begin{align*} \|f\|_{Y}=\max \left(\|f\|_{d},\right. & \left.\|f\|_{\infty}\right), \quad f \in \mathrm{BLip}_{0} Y \\ \|F\|_{X}=\max \left(\|F\|_{d},\right. & \left.\|F\|_{\infty}\right), \quad F \in \mathrm{BLip}_{0} X \tag{2.6} \end{align*}fY=max(fd,f),fBLip0Y(2.6)FX=max(Fd,F),FBLip0X
BLip 0 Y BLip 0 Y BLip_(0)Y\mathrm{BLip}_{0} YBLip0Y and BLip 0 X BLip 0 X BLip_(0)X\mathrm{BLip}_{0} XBLip0X become Banach spaces (see [3]). Let H = BS X ¯ H = BS X ¯ H=BS_( bar(X))H=\mathrm{BS}_{\bar{X}}H=BSX¯ - BS X BS X -BS_(X)-\mathrm{BS}_{X}BSX the subspace of BLip 0 X BLip 0 X BLip_(0)X\mathrm{BLip}_{0} XBLip0X generated by the convex cone BS X BS X BS_(X)\mathrm{BS}_{X}BSX and
Y ⊥= { g H : g | Y = 0 } Y ⊥= g H : g Y = 0 Y⊥={g in H:g|_(Y)=0}Y \perp=\left\{g \in H:\left.g\right|_{Y}=0\right\}Y⊥={gH:g|Y=0}
the anihilator of the set Y Y YYY in H H HHH. Obviously, Y Y Y _|_Y \perpY is a subspace of H H HHH. A subset Z Z ZZZ of a normed space X X XXX is called proximinal for W X W X W sub XW \subset XWX if for every f W f W f in Wf \in WfW there exists g 0 Z g 0 Z g_(0)in Zg_{0} \in Zg0Z, such that
(2.7) f g 0 = d ( f , Z ) = inf { f g : g Z } . (2.7) f g 0 = d ( f , Z ) = inf { f g : g Z } . {:(2.7)||f-g_(0)||=d(f","Z)=i n f{||f-g||:g in Z}.:}\begin{equation*} \left\|f-g_{0}\right\|=d(f, Z)=\inf \{\|f-g\|: g \in Z\} . \tag{2.7} \end{equation*}(2.7)fg0=d(f,Z)=inf{fg:gZ}.
If for every f W f W f in Wf \in WfW the element g 0 Z g 0 Z g_(0)in Zg_{0} \in Zg0Z satisfying (2.7) is unique then the set Z Z ZZZ is called Chebyshevian for W W WWW. An element g 0 Z g 0 Z g_(0)in Zg_{0} \in Zg0Z satisfying (2.7) is called an element of best approximation of f f fff by elements of Z Z ZZZ.
THEOREM 2. Y Y Y _|_Y \perpY is a Chebyshevian subsapce for BS X ¯ BS X ¯ BS_( bar(X))\mathrm{BS}_{\bar{X}}BSX¯ if and only if, every f BS f BS f inBSf \in \mathrm{BS}fBS has a unique (preserving the uniform and Lipschitz norms) extension F F FFF in BS x ¯ BS x ¯ BS_( bar(x))\mathrm{BS}_{\bar{x}}BSx¯.
Proof. Follows from Theorem 1 in [6].
Remark. Observe that Theorem 2 remains true if the spaces BLip 0 Y BLip 0 Y BLip_(0)Y\mathrm{BLip}_{0} YBLip0Y and BLip 0 X BLip 0 X BLip_(0)X\mathrm{BLip}_{0} XBLip0X are equiped with the norms
f 1 = f a + f , for f BLip 0 Y (respectively BLip 0 X ) f 1 = f a + f ,  for  f BLip 0 Y  (respectively  BLip 0 X {:||f||_(1)=||f||_(a)+||f||_(oo)," for "f inBLip_(0)Y" (respectively "BLip_(0)X)\left.\|f\|_{1}=\|f\|_{a}+\|f\|_{\infty}, \text { for } f \in \mathrm{BLip}_{0} Y \text { (respectively } \mathrm{BLip}_{0} X\right)f1=fa+f, for fBLip0Y (respectively BLip0X)
Theorem 2 is analogous with a theorem of PHELPS [8], in the linear case.
3. Now, we try to find conditions on the function f f fff ensuring the unicity of the extension.
Consider, firstly, the case X = R X = R X=RX=RX=R with the usual norm | | | | |*||\cdot||| (the absolute value).
THEOREM 3. Let Y = [ a , b ] R , a < 0 < b Y = [ a , b ] R , a < 0 < b Y=[a,b]sub R,a < 0 < bY=[a, b] \subset R, a<0<bY=[a,b]R,a<0<b. A function f BS Y f BS Y f inBS_(Y)f \in \mathrm{BS}_{Y}fBSY has a unique extension F BS R F BS R F inBS_(R)^(-)F \in \mathrm{BS}_{R}^{-}FBSRif and only if f ( a ) = f ( b ) = 0 f ( a ) = f ( b ) = 0 f(a)=f(b)=0f(a)=f(b)=0f(a)=f(b)=0.
Proof. Suppose that f BS Y f BS Y f inBS_(Y)^(-)f \in \mathrm{BS}_{Y}^{-}fBSYhas two distinct extensions F 1 , F 2 F 1 , F 2 F_(1),F_(2)F_{1}, F_{2}F1,F2 in BS R ¯ BS R ¯ BS_( bar(R))^(-)\mathrm{BS}_{\bar{R}}^{-}BSR¯. Let x R [ a , b ] x R [ a , b ] x in R\\[a,b]x \in R \backslash[a, b]xR[a,b] be such that F 1 ( x ) F 2 ( x ) F 1 ( x ) F 2 ( x ) F_(1)(x)!=F_(2)(x)F_{1}(x) \neq F_{2}(x)F1(x)F2(x), say F 1 ( x ) < F 2 ( x ) 0 F 1 ( x ) < F 2 ( x ) 0 F_(1)(x) < F_(2)(x) <= 0F_{1}(x)<F_{2}(x) \leq 0F1(x)<F2(x)0. Suppose x > b x > b x > bx>bx>b. The function F 1 F 1 F_(1)F_{1}F1 being starshaped it follows
F 1 ( λ x ) λ F 1 ( x ) < 0 , F 1 ( λ x ) λ F 1 ( x ) < 0 , F_(1)(lambda x) <= lambdaF_(1)(x) < 0,F_{1}(\lambda x) \leq \lambda F_{1}(x)<0,F1(λx)λF1(x)<0,
for all λ ( 0 , 1 ] λ ( 0 , 1 ] lambda in(0,1]\lambda \in(0,1]λ(0,1]. In particular, since b = λ b x b = λ b x b=lambda_(b)xb=\lambda_{b} xb=λbx for λ b ( 0 , 1 ) λ b ( 0 , 1 ) lambda_(b)in(0,1)\lambda_{b} \in(0,1)λb(0,1), it follows f ( b ) = F 1 ( b ) < 0 f ( b ) = F 1 ( b ) < 0 f(b)=F_(1)(b) < 0f(b)=F_{1}(b)<0f(b)=F1(b)<0. If x < a x < a x < ax<ax<a, then a = λ a x a = λ a x a=lambda_(a)xa=\lambda_{a} xa=λax, for a λ a ( 0 , 1 ) λ a ( 0 , 1 ) lambda_(a)in(0,1)\lambda_{a} \in(0,1)λa(0,1), and similarly, f ( a ) = F 1 ( a ) < 0 f ( a ) = F 1 ( a ) < 0 f(a)=F_(1)(a) < 0f(a)=F_{1}(a)<0f(a)=F1(a)<0.
Conversely, we shall show that if f BS Y f BS Y f inBS_(Y)^(-)f \in \mathrm{BS}_{Y}^{-}fBSYis such that f ( a ) < 0 f ( a ) < 0 f(a) < 0f(a)<0f(a)<0 of f ( b ) < 0 f ( b ) < 0 f(b) < 0f(b)<0f(b)<0, then f f fff has at least two distinct extensions F 1 F 1 F_(1)F_{1}F1 and F 2 F 2 F_(2)F_{2}F2 in BS R ¯ BS R ¯ BS_( bar(R))\mathrm{BS}_{\bar{R}}BSR¯. If f ( b ) < 0 f ( b ) < 0 f(b) < 0f(b)<0f(b)<0, then
F 1 ( x ) = { f ( x ) , x [ 0 , b ] f ( b ) + f d ( x b ) , x ( b , b f ( b ) ( f d ) 1 ] 0 , x ( , a ) ( b f ( b ) ( f d ) 1 , ) F 1 ( x ) = f ( x ) ,      x [ 0 , b ] f ( b ) + f d ( x b ) ,      x b , b f ( b ) f d 1 0 ,      x ( , a ) b f ( b ) f d 1 , F_(1)(x)={[f(x)",",x in[0","b]],[f(b)+||f||_(d)(x-b)",",x in(b,b-f(b)(||f||_(d))^(-1)]],[0",",x in(-oo","a)uu(b-f(b)*(||f||_(d))^(-1),-oo)]:}F_{1}(x)= \begin{cases}f(x), & x \in[0, b] \\ f(b)+\|f\|_{d}(x-b), & x \in\left(b, b-f(b)\left(\|f\|_{d}\right)^{-1}\right] \\ 0, & x \in(-\infty, a) \cup\left(b-f(b) \cdot\left(\|f\|_{d}\right)^{-1},-\infty\right)\end{cases}F1(x)={f(x),x[0,b]f(b)+fd(xb),x(b,bf(b)(fd)1]0,x(,a)(bf(b)(fd)1,)
and
F 2 ( x ) = { f ( x ) , x [ 0 , b ] ( f ( b ) / b ) x , x ( b , ( f / f ( b ) ) b ) f , x [ ( f / f ( b ) ) b , + ) 0 , x ( , a ) F 2 ( x ) = f ( x )      , x [ 0 , b ] ( f ( b ) / b ) x      , x b , f / f ( b ) b f      , x f / f ( b ) b , + 0      , x ( , a ) F_(2)(x)={[f(x),","x in[0","b]],[(f(b)//b)x,","x in(b,(-||f||_(oo)//f(b))b)],[-||f||_(oo),","x in[-(||f||_(oo)//f(b))b,+oo)],[0,","x in(-oo","a)]:}F_{2}(x)= \begin{cases}f(x) & , x \in[0, b] \\ (f(b) / b) x & , x \in\left(b,\left(-\|f\|_{\infty} / f(b)\right) b\right) \\ -\|f\|_{\infty} & , x \in\left[-\left(\|f\|_{\infty} / f(b)\right) b,+\infty\right) \\ 0 & , x \in(-\infty, a)\end{cases}F2(x)={f(x),x[0,b](f(b)/b)x,x(b,(f/f(b))b)f,x[(f/f(b))b,+)0,x(,a)
are two distinct extensions of f f fff, i.e. F 1 F 2 , F 1 | Y = F 2 | Y = f , F 1 a == F 2 d = f d , F 1 = F 2 = f F 1 F 2 , F 1 Y = F 2 Y = f , F 1 a == F 2 d = f d , F 1 = F 2 = f F_(1)!=F_(2),F_(1)|_(Y)=F_(2)|_(Y)=f,||F^(')_(1)||_(a)==||F_(2)||_(d)=||f||_(d),||F_(1)||_(oo)=||F_(2)||_(oo)=||f||_(oo)F_{1} \neq F_{2},\left.F_{1}\right|_{Y}=\left.F_{2}\right|_{Y}=f,\left\|F^{\prime}{ }_{1}\right\|_{a}= =\left\|F_{2}\right\|_{d}=\|f\|_{d},\left\|F_{1}\right\|_{\infty}=\left\|F_{2}\right\|_{\infty}=\|f\|_{\infty}F1F2,F1|Y=F2|Y=f,F1a==F2d=fd,F1=F2=f.
7 - Mathematica - Revue d'analyse numérique et de théorie de l'approximation, Tome 9. nr. 1/1980
F 1 ( x ) F 2 ( x ) F 1 ( x ) F 2 ( x ) F_(1)(x)!=F_(2)(x)F_{1}(x) \neq F_{2}(x)F1(x)F2(x) for all x > max { b f ( b ) ( f a ) 1 ; f ( f ( b ) ) 1 } x > max b f ( b ) f a 1 ; f ( f ( b ) ) 1 x > max{b-f(b)*(||f||_(a))^(-1);-||f||_(oo)*(f(b))^(-1)}x>\max \left\{b-f(b) \cdot\left(\|f\|_{a}\right)^{-1} ;-\|f\|_{\infty} \cdot(f(b))^{-1}\right\}x>max{bf(b)(fa)1;f(f(b))1}.
If f ( a ) < 0 f ( a ) < 0 f(a) < 0f(a)<0f(a)<0, then two distinct extensions F 1 , F 2 F 1 , F 2 F_(1),F_(2)F_{1}, F_{2}F1,F2, may be given, in a similar way.
Remark. The hypothesis a < 0 < b a < 0 < b a < 0 < ba<0<ba<0<b in Theorem 3 is essential as it is shown by the following example. Take Y = [ 0 , b ] , b > 0 Y = [ 0 , b ] , b > 0 Y=[0,b],b > 0Y=[0, b], b>0Y=[0,b],b>0 or Y == [ a , 0 ] , a < 0 Y == [ a , 0 ] , a < 0 Y==[a,0],a < 0Y= =[a, 0], a<0Y==[a,0],a<0. Then every f BSS Y f BSS Y f inBSS_(Y)^(-)f \in \mathrm{BSS}_{Y}^{-}fBSSYhas an infinite set of extensions in BS R ¯ BS R ¯ BS_( bar(R))^(-)\mathrm{BS}_{\bar{R}}^{-}BSR¯. For exemple, if Y = [ 0 , b ] Y = [ 0 , b ] Y=[0,b]Y=[0, b]Y=[0,b] and f BS Y ¯ f BS Y ¯ f inBS_( bar(Y))^(-)f \in \mathrm{BS}_{\bar{Y}}^{-}fBSY¯is such that f ( 0 ) = 0 f ( 0 ) = 0 f(0)=0f(0)=0f(0)=0, f ( b ) = 0 f ( b ) = 0 f(b)=0f(b)=0f(b)=0, then
F λ ( x ) = { f ( x ) , x [ 0 , b ] 0 , x ( 0 , + ) f d x , x ( λ f ( f d ) 1 , 0 ) λ f , x ( , λ f ( f d ) 1 ) F λ ( x ) = f ( x ) ,      x [ 0 , b ] 0 ,      x ( 0 , + ) f d x ,      x λ f f d 1 , 0 λ f ,      x , λ f f d 1 F_(lambda)(x)={[f(x)",",x in[0","b]],[0",",x in(0","+oo)],[||f||_(d)x",",x in(-lambda||f||_(oo)*(||f||_(d))^(-1),0)],[-lambda||f||_(oo)",",x in(-oo,-lambda||_(f)∣||_(oo)(||f||_(d))^(-1))]:}F_{\lambda}(x)= \begin{cases}f(x), & x \in[0, b] \\ 0, & x \in(0,+\infty) \\ \|f\|_{d} x, & x \in\left(-\lambda\|f\|_{\infty} \cdot\left(\|f\|_{d}\right)^{-1}, 0\right) \\ -\lambda\|f\|_{\infty}, & x \in\left(-\infty,-\lambda\left\|_{f} \mid\right\|_{\infty}\left(\|f\|_{d}\right)^{-1}\right)\end{cases}Fλ(x)={f(x),x[0,b]0,x(0,+)fdx,x(λf(fd)1,0)λf,x(,λf(fd)1)
is an extension of f f fff for every λ [ 0 , 1 ] λ [ 0 , 1 ] lambda in[0,1]\lambda \in[0,1]λ[0,1].
Consider now the general case. For x X , x θ x X , x θ x in X,x!=thetax \in X, x \neq \thetaxX,xθ, the ray θ x θ x vec(theta x)\overrightarrow{\theta x}θx is defined by
θ x = { α x : α 0 } θ x = { α x : α 0 } vec(theta x)={alpha x:alpha >= 0}\overrightarrow{\theta x}=\{\alpha x: \alpha \geq 0\}θx={αx:α0}
If the set Y Y YYY is starshaped and y Y , y θ y Y , y θ y in Y,y!=thetay \in Y, y \neq \thetayY,yθ, then θ y Y θ y Y vec(theta y)sub Y\overrightarrow{\theta y} \subset YθyY or θ y Y θ y Y vec(theta y)nn Y\overrightarrow{\theta y} \cap YθyY is a segment. In the second case put
α y = sup { α : α y Y } , α y = sup { α : α y Y } , alpha_(y)=s u p{alpha:alpha y in Y},\alpha_{y}=\sup \{\alpha: \alpha y \in Y\},αy=sup{α:αyY},
and z y = α y y z y = α y y z_(y)=alpha_(y)*yz_{y}=\alpha_{y} \cdot yzy=αyy. The set { z y : y Y } z y : y Y {z_(y):y in Y}\left\{z_{y}: y \in Y\right\}{zy:yY} is called the algebric starshapes boundary of Y Y YYY and is denoted by F r s Y F r s Y F_(r)^(s)Y\mathrm{F}_{\mathrm{r}}^{\mathrm{s}} YFrsY.
Evidently, every z F r s Y z F r s Y z inF_(r)^(s)Yz \in \mathrm{~F}_{\mathrm{r}}^{\mathrm{s}} Yz FrsY is a limit point of Y Y YYY, i.e. Y F r s Y ( Y ¯ Y F r s Y ( Y ¯ Y uuF_(r)^(s)Y( bar(Y)Y \cup \mathrm{~F}_{\mathrm{r}}^{\mathrm{s}} Y(\bar{Y}Y FrsY(Y¯. Since every f BS Y f BS Y f inBS_(Y)f \in \mathrm{BS}_{Y}fBSY is uniformly continuous (as Lipschitz) it can be uniquely extended to Y Fr r s Y Y Fr r s Y Y uuuFr_(r)^(s)YY \bigcup \mathrm{Fr}_{r}^{s} YYFrrsY. Therefore with no restrition of generality, we can suppose F r s Y Y F r s Y Y F_(r)^(s)Y sub Y\mathrm{F}_{\mathrm{r}}^{\mathrm{s}} Y \subset YFrsYY.
THEOREM 4. Let Y Y YYY be an absorbing starshaped subset of the normed space X X XXX, such that. F r s Y Y F r s Y Y F_(r)^(s)Y sub Y\mathrm{F}_{\mathrm{r}}^{\mathrm{s}} Y \subset YFrsYY. If f BS Y f BS Y f inBS_(Y)^(-)f \in \mathrm{BS}_{Y}^{-}fBSYis such that f ( z ) = 0 f ( z ) = 0 f(z)=0f(z)=0f(z)=0, for all z ∈∈ F r s Y z ∈∈ F r s Y z∈∈F_(r)^(s)Yz \in \in \mathrm{F}_{\mathrm{r}}^{\mathrm{s}} Yz∈∈FrsY, then f f fff has a unique extension F BSS X F BSS X F inBSS_(X)^(-)F \in \mathrm{BSS}_{X}^{-}FBSSX.
Proof. Suppose f BS Y , f ( z ) = 0 f BS Y , f ( z ) = 0 f inBS_(Y)^(-),f(z)=0f \in \mathrm{BS}_{Y}^{-}, f(z)=0fBSY,f(z)=0, for all z F I s Y z F I s Y z inF_(I)^(s)Yz \in \mathrm{~F}_{\mathrm{I}}^{s} Yz FIsY, and suppose that f f fff has two distinct extensions F 1 , F 2 F 1 , F 2 F_(1),F_(2)F_{1}, F_{2}F1,F2 in BS X BS X BS_(X)^(-)\mathrm{BS}_{X}^{-}BSX. Let x X Y x X Y x in X\\Yx \in X \backslash YxXY be such that F 1 ( x ) F 2 ( x ) F 1 ( x ) F 2 ( x ) F_(1)(x)!=F_(2)(x)F_{1}(x) \neq F_{2}(x)F1(x)F2(x), say F 1 ( x ) < F 2 ( x ) 0 F 1 ( x ) < F 2 ( x ) 0 F_(1)(x) < F_(2)(x) <= 0F_{1}(x)<F_{2}(x) \leq 0F1(x)<F2(x)0. The set Y Y YYY being absorbing and starshaped, there exists λ > 0 λ > 0 lambda > 0\lambda>0λ>0 such that λ x F r s Y λ x F r s Y lambda x inF_(r)^(s)Y\lambda x \in \mathrm{~F}_{\mathrm{r}}^{\mathrm{s}} Yλx FrsY. But then, one obtains the contradiction
0 = f ( λ x ) = F 1 ( λ x ) λ F 1 ( x ) < 0 0 = f ( λ x ) = F 1 ( λ x ) λ F 1 ( x ) < 0 0=f(lambda x)=F_(1)(lambda x) <= lambdaF_(1)(x) < 00=f(\lambda x)=F_{1}(\lambda x) \leq \lambda F_{1}(x)<00=f(λx)=F1(λx)λF1(x)<0
Theorem 4 is proved.

BIBLIOGRAPHY

[1] Cobzas, S., Mustăta, C., Norm Preserving Extension of Convex Lipschitz Functionş. J. Approx. Theory 2 2 4 2 2 4 22^(4)2 \boldsymbol{2}^{4}224 (1978), 555-564.
[2] Holmes, R. B., A Course on Optimisation and Best Approximation, Lectures Notes in s, R. B., A Course on Poringer Verlag, Berlin-Heidelberg-New York, 1972.
[3] J. A. Banach Spaces of Lipschitz Functions and Vector-Valued Lipschitz Fi Mons, Trans. Amer. Math. Soc. 148 (1970), 147-169.
Funch. Extension of Range of Functions, Bull. Amer. Math. Soc. 40 (1934),
[4] McSh 837-842.
[5] Lipschitz Functions, J. [5] Mustăta, C.,
[5] Must Approx. The Approximation and 230.
Approx. Theory 19 (1977), 222-230.
[6] Mustăta, C., A Characterisainon of Cheox., L'Analyse Num. Teor. Approx. 6, 1 (1977), 51-56.
[7] Musiăta, C., Norm Preserving Extension of Starshoped Lipschitz Functions, Mathe-
matica 19 (12) 2 Haness of Hahn-Banach Extension and Unique Best Approximation, s, R. R., Uniqueness of Han- Amer. Math. Soc. 95 (1960), 238-255.
Received. 21.XII.1989.
1980

Related Posts