Considering a metric space and its Lipschitz dual one defines the notion of M-ideal and HB-subspace of a metric space (with respect to its Lipschitz dual). One obtain some results analogous to these in the theory of M-ideal and HB-subspaces in a normed space. The results in the paper are based on an extension theorem of McShane [2], [3] and on a uniquenese theorem which is similar to one of R.R.Phelps [10], [11].
Authors
Costica Mustata
“Tiberiu Popoviciu” Institute of Numerical Analysis, Romanian Academy, Romania
Keywords
Paper coordinates
C. Mustăţa, M-ideals in metric spaces, ”Babeş-Bolyai” University, Faculty of Math. and Physics, Research Seminars, Seminar on Mathematica Analysis, Preprint Nr.7 (1988), 67-74 (MR # 90b: 54019)
[1] Alfsen, D.M., Effross, E., Structure in real Banach spaces, Ann. of Math. 96(1972), 98-173.
[2] Czisper, J.,Geher, L., Extension of Funcitons satisfying a Lipschitz conditions, Acta Math. Acad.Sci. Hungar 6(1955), 213-220.
[3] Fakhoury, E., Selections lineaires associees au Theoreme de Hahn-Banach, J. of Funcitonal analysis 11 (1972), 436-452.
[4] Hennefeld, J., M – ideas, HB – subspaces and Compact Operators, Indiana Univ. Math. J. 28 (6) (1979), 927-934.
[5] Hennefeld, J., A note on M – ideals in B(X), Mat. Soc. 98 (1) (1980), 89-92.
[6] Holmes, R.B., Scrantor, B., Ward, J.D., Approximation from the space of compact operators and other M – ideals Duke Math. J. 42 (1975), 259-269.
[7] Holmes, R.B., Geometric Functional Analysis and its Applications, Springer – Verlag – New York – Heidelberg – Berlin, 1975.
[8] Shane, E.J., Extension of range of funcitons, Bull. Amer. Math. Soc. 40 (1934), 837-842.
[9] Johnson, J.A., Banach Spaces of Lipschitz Functions and vector – valued Lipschitz Functions, Trans. Amer. Math. Soc. 148(1970), 147-169.
[10] Mustata, C., Best Approximation and Uniwuq Extension of Lipschitz Functions, J. Approx. Theory 19 (3) (1977), 222-230.
[11] Phelps, R.R., Uniqueness of Hahn-Banach Extension and Unique Best Approximation, Trans. Amer. Math. Sec. 25 (1960), 238-255.
[12] Oja, E., On the uniquess of the norm preserving extension of linear functional in this Hahn-Banach Theorem, Proc. Acad. Science Esteonian SSR 33 (4) (1984), 422-433 (Russian).
Paper (preprint) in HTML form
1988-Mustata-Seminar-UBB-Idelas-in-metric-spaces
M - IDEALS IN MRTRIC SPACES
Costică Mustăta
Abstrect. Considering a netric space and its Lipschitz dual one defines the notion of M-ideal and HB-subspace of a metric space (with respect to its Lipschitz dual). One obtain some results analogous to those in the theory of MMMMM-ideals and BB―BB¯bar(BB)\overline{B B}BB―-subspaces in a normed space. The rocults in the paper are based on an extension theorem of KCShane [2],[8][2],[8][2],[8][2],[8][2],[8] and on a uniqueness theorem which is similar to one of R.R. Paelps [10], [22].
Introduction. The notion of M-ideal, introduced by E.M. Alfsen and 2. Erross [1], has many applications in functional analysis as, for example, to the problem of best approximation of continuous linear operators by compact operators (see [6]). Following E.M. Alfsen and I. Effross [1], an U-ideal in a normed space XXXXX is a closed sibspace YYYYY of XXXXX whose annihilator Y⊥Y⊥Y^(_|_)Y^{\perp}Y⊥ admits a complement GGGGG in X∗X∗X^(**)X^{*}X∗ such that ‖g+h‖=‖g‖+‖h‖,g∈G,h∈I⊥‖g+h‖=‖g‖+‖h‖,g∈G,h∈I⊥||g+h||=||g||+||h||,g in G,h inI^(_|_)\|g+h\|=\|g\|+\|h\|, g \in G, h \in I^{\perp}‖g+h‖=‖g‖+‖h‖,g∈G,h∈I⊥. S. Henneleld [4], [5] defined a more general notion tAtAt quad At \quad AtA closed subspace YYYYY of a normed space XXXXX is called an HBHBHBH BHB-subspace if Y⊥Y⊥Y^(_|_)Y^{\perp}Y⊥ has a complement GGGGG 1n Z∗Z∗Z^(**)Z^{*}Z∗ such that for every f∈X∗,‖f‖⩾‖h‖,‖f‖>‖g‖f∈X∗,‖f‖⩾‖h‖,‖f‖>‖g‖f inX^(**),||f|| >= ||h||,||f|| > ||g||f \in X^{*},\|f\| \geqslant\|h\|,\|f\|>\|g\|f∈X∗,‖f‖⩾‖h‖,‖f‖>‖g‖, whenever f=g+hf=g+hf=g+hf= g+hf=g+h, with g∈G,h∈Y⊥,h≠0g∈G,h∈Y⊥,h≠0g in G,h inY^(_|_),h!=0g \in G, h \in Y^{\perp}, h \neq 0g∈G,h∈Y⊥,h≠0. It is immediate that overy μ−ide−μ−ide−mu-ide-\mu-i d e-μ−ide− al is an HBHBHB\mathbb{H B}HB-subspace and the converse is not true : The space K(i1,i)Ki1,iK(i_(1),i)\mathbb{K}\left(i_{1}, i\right)K(i1,i) of compact Iinear operators from l1l1l_(1)l_{1}l1 to l1l1l_(1)l_{1}l1 is an HB-subspace of A(l,n)A(l,n)A(l,n)A(l, n)A(l,n), the space of all continuous linear operators from ℓ1ℓ1ℓ_(1)\ell_{1}ℓ1 to ℓ1ℓ1ℓ_(1)\ell_{1}ℓ1, which 1 s not an H-ideal ( see [4], [5]).
The aim of this paper is to transpose these notions to a guneral
metric space XXXXX (without any linear structure) by appealing to the Iipschitz dual of XXXXX, 1.e. a Banach space of Lipschitz functions on XXXXX. The notions of M-ideal and HB-subspace cannot be eztonded automatically to a metric linear space by using its linear dual, for two reasons f first, this dual may be trivial, as is the case for the space Ip[0,1],0<p<1Ip[0,1],0<p<1I^(p)[0,1],0 < p < 1I^{p}[0,1], 0<p<1Ip[0,1],0<p<1, and second, there is no norm on the dual of a motric linear space.
Results. Let ( X,dX,dX,dX, dX,d ) be a metric space, x0x0x_(0)x_{0}x0 a fixed point in XXXXX and let XXXXX be a subset of XXXXX such that x0∈Yx0∈Yx_(0)in Yx_{0} \in Yx0∈Y. If XXXXX is a metric linear space we take always x0=0x0=0x_(0)=0x_{0}=0x0=0. A function f:Y→Rf:Y→Rf:Y rarr Rf: Y \rightarrow Rf:Y→R is called a lipschitz function on YYYYY if there exists K⩾0K⩾0K >= 0K \geqslant 0K⩾0 such that
(1) |f(x)−f(y)|⩽Kd(x,y)|f(x)−f(y)|⩽Kd(x,y)quad|f(x)-f(y)| <= Kd(x,y)\quad|f(x)-f(y)| \leqslant K d(x, y)|f(x)−f(y)|⩽Kd(x,y)
for all x,y∈Xx,y∈Xx,y in Xx, y \in Xx,y∈X :
Denote by Lip 00_(0){ }_{0}0 the following set
(2) Lip0I={f&I→R,fLip0I=f&I→R,fLip_(0)I={f&IrarrR,f:}\operatorname{Lip}_{0} \mathrm{I}=\left\{\mathrm{f} \& \mathrm{I} \rightarrow \mathbf{R}, \mathrm{f}\right.Lip0I={f&I→R,f is a Iipschitz function on I,f(x0)=0}I,fx0=0{:I,f(x_(0))=0}\left.\mathrm{I}, \mathrm{f}\left(\mathrm{x}_{0}\right)=0\right\}I,f(x0)=0}.
Equiped with the usual operations of addition and multiplication by scalars, Lifo YYYYY is a linear space and the application ‖‖I′‖I′||||_(I^(')):}\left\|\|_{I^{\prime}}\right.‖‖I′ ! Lip o X→o X→_("o ")X rarr_{\text {o }} X \rightarrowo X→ R , defined by coscri beanon & II
(3) ‖f‖Z=sup{|f(x)−f(y)|/d(x,y),x,y∈I,x≠y}‖f‖Z=sup{|f(x)−f(y)|/d(x,y),x,y∈I,x≠y}||f||_(Z)=s u p{|f(x)-f(y)|//d(x,y)quad,quad x,y in I,x!=y}\|f\|_{Z}=\sup \{|f(x)-f(y)| / d(x, y) \quad, \quad x, y \in I, x \neq y\}‖f‖Z=sup{|f(x)−f(y)|/d(x,y),x,y∈I,x≠y},
is a nord on Lip 0I0I_(0)I{ }_{0} I0I. It is easily seen that ‖f‖Y‖f‖Y||f||_(Y)\|f\|_{Y}‖f‖Y is the shallest of the numbers x⩾0x⩾0x >= 0x \geqslant 0x⩾0 for which the inequality (1) holds. The space ( Lip0γ,‖1‖YLip0γ,1YLip_(0)^(gamma),||_(1)||_(Y)L i p_{0}{ }^{\gamma},\left\|_{1}\right\|_{\mathbf{Y}}Lip0γ,‖1‖Y ) is a Banach space (even a dual Banach space , see [9]) and we call it the Lipschitz dual of YYYYY.
For X≡XX≡XX-=XX \equiv XX≡X the space Lip0XLip0XLip_(0)XL i p_{0} XLip0X and the norm ‖‖X‖X||||_(X):}\left\|\|_{X}\right.‖‖X are delined similarly.
In the following, one supposes always that the subset YYYYY of XXXXX contains x0x0x_(0)x_{0}x0 o the fixed element of xxxxx.
The following Hahn-Banach type extension theorem for Lipachitz functions was proved by Uc Shane [8]. (see also [2] ) :
TREOREM 1. Let (X,d)(X,d)(X,d)(X, d)(X,d) be a metric space, x0x0x_(0)x_{0}x0 a fixed point in XXXXX and let YYYYY be a subset of XXXXX such that x0∈Ix0∈Ix_(0)in Ix_{0} \in Ix0∈I. Then every function f∈Lp0If∈Lp0If in Lp_(0)If \in L p_{0} If∈Lp0I has a norm-preserving extension in Ldp0X,1.6Ldp0X,1.6Ldp_(0)X,1.6L d p_{0} X, 1.6Ldp0X,1.6. there exists Z∈Itg0XZ∈Itg0XZ in Itg_(0)XZ \in I \operatorname{tg}_{0} XZ∈Itg0X such that F|Y=fFY=fF|_(Y)=f\left.F\right|_{Y}=fF|Y=f and ‖‖X=‖P‖XX=PX||||_(X)=||P||_(X)\left\|\left\|_{X}=\right\| P\right\|_{X}‖‖X=‖P‖X.
In fact, He Shane [8], proved this theorem in the case of the space Lip YYYYY and Lip XXXXX of all Lipschitz functions on YYYYY and XXXXX, respectively, but the above formulation is more appropriate for our needs. We shall call sometimes briefly any function FFFFF, as given in Theorem 1, an extension of fffff.
In general, the extension of a function f∈Kp0Yf∈Kp0Yf in Kp_(0)Yf \in K p_{0} Yf∈Kp0Y to XXXXX is not unique. The functions
(4)II(x)=inf{f(y)+‖f‖Id(x,y):y∈I}I2(x)=sup{f(y)−‖f‖Id(x,y):y∈I}(4)II(x)=inff(y)+‖f‖Id(x,y):y∈II2(x)=supf(y)−‖f‖Id(x,y):y∈I{:[(4)I_(I)(x)=i n f{f(y)+||f||_(I)d(x,y):y in I}],[I_(2)(x)=s u p{f(y)-||f||_(I)d(x,y):y in I}]:}\begin{align*}
& I_{I}(x)=\inf \left\{f(y)+\|f\|_{I} d(x, y): y \in I\right\} \tag{4}\\
& I_{2}(x)=\sup \left\{f(y)-\|f\|_{I} d(x, y): y \in I\right\}
\end{align*}(4)II(x)=inf{f(y)+‖f‖Id(x,y):y∈I}I2(x)=sup{f(y)−‖f‖Id(x,y):y∈I}
are two extensions of fffff and they are extremal elements of the conver set R(f;Y)R(f;Y)R(f;Y)\mathbb{R}(f ; Y)R(f;Y) of all extensions of fffff. Frery extension IIIII of fffff verifies the inequalities :
(5)F2(x)⩽P(x)⩽F1(x),x∈X(5)F2(x)⩽P(x)⩽F1(x),x∈X{:(5)F_(2)(x) <= P(x) <= F_(1)(x)quad","quad x in X:}\begin{equation*}
F_{2}(x) \leqslant P(x) \leqslant F_{1}(x) \quad, \quad x \in X \tag{5}
\end{equation*}(5)F2(x)⩽P(x)⩽F1(x),x∈X
Therefore, the function f∈Lip0If∈Lip0If in Lip_(0)If \in L i p_{0} If∈Lip0I has a unique extension in Lip I0I0I_(0)I_{0}I0 if and only if F1=F2F1=F2F_(1)=F_(2)F_{1}=F_{2}F1=F2.
DEFIMITION 1. The subset YYYYY of XXXXX is said to have property ( UUUUU ) if every function f∈Lip0If∈Lip0If in Lip_(0)If \in L i p_{0} If∈Lip0I has a unique extension in Lip I0I0I_(0)I_{0}I0.
Hecessary and surficient conditions in order that a subset YYYYY of I have property (U) and relations of this property with the problem of best approximation in Lip 00_(0){ }_{0}0 by slements in x⊥⊥x⊥⊥x^(_|_)^(_|_){x^{\perp}}^{\perp}x⊥⊥ are given犃 [10].
For X⊆XX⊆XX sube XX \subseteq XX⊆X donote by X⊥X⊥X^(_|_)X^{\perp}X⊥ its annihilator in Lig0XLig0XLig_(0)X\operatorname{Lig}_{0} XLig0X, i.e.
(6) I⊥={F∈Lp0I,I|I=0}I⊥=F∈Lp0I,II=0quadI^(_|_)={F in Lp_(0)I, quad I|_(I)=0}\quad I^{\perp}=\left\{F \in L p_{0} I,\left.\quad I\right|_{I}=0\right\}I⊥={F∈Lp0I,I|I=0}.
Obviously, X⊥X⊥X^(_|_)X^{\perp}X⊥ is a closed subspace of It p0Xp0Xp_(0)X\mathrm{p}_{0} Xp0X.
DEFINITION 2. A subset VVVVV of a normed space ZZZZZ is called proyminal if for every z∈Zz∈Zz in Zz \in Zz∈Z there exists ∇0∈V∇0∈Vgrad_(0)in V\nabla_{0} \in V∇0∈V such that
(7) ‖z−v0‖=d(z,V^)z−v0=d(z,V^)quad||z-v_(0)||=d(z, hat(V))\quad\left\|z-v_{0}\right\|=d(z, \hat{V})‖z−v0‖=d(z,V^)
Where d(z,V)≡inf{|z−∇|,∇∈∇}d(z,V)≡inf{|z−∇|,∇∈∇}d(z,V)-=i n f{|z-grad|,grad in grad}d(z, V) \equiv \inf \{|z-\nabla|, \nabla \in \nabla\}d(z,V)≡inf{|z−∇|,∇∈∇} denotes the distance from zzzzz to VVVVV. An element V0V0V_(0)V_{0}V0 satisfying (7) is called a bost approximation element of zzzzz by elements in VVVVV. If every z∈Zz∈Zz in Zz \in Zz∈Z has a unique best approximation element in VVVVV then the set VVVVV is called Chebysherian .
R.R. Pholps [11] obtained some results concerning the relations between the property (U) and the unicity of best approximation in the dual of a normed space. As was shown in [10] similar results hold also in the Iipschitz case :
THSOREM 2. ([10]) A s joet YYYYY of a netric spacs XXXXX has proporty (V) if and oniy if its annihilator I⊥I⊥I^(_|_)I^{\perp}I⊥ is Chebyshorian subonce of Lipo 22^(2){ }^{2}2.
The proof of this theorem is based on the following lemma, which vill be used in the sequel :
LEVII 1. ([10]). Let XXXXX be a metric space, I a subset of XXXXX and Y⊥Y⊥Y^(_|_)Y^{\perp}Y⊥ the annihilator of YYYYY in Lip0XLip0XLip_(0)X\operatorname{Lip}_{0} XLip0X. If T∈Lip0XT∈Lip0XT inLip_(0)XT \in \operatorname{Lip}_{0} XT∈Lip0X then d(F,Y⊥)=′′I′′IdF,Y⊥=′′I′′Id(F,Y^(_|_))=^('')I^('')Id\left(F, Y^{\perp}\right)= { }^{\prime \prime} I^{\prime \prime} Id(F,Y⊥)=′′I′′I and an alement δ0∈I⊥δ0∈I⊥delta_(0)inI^(_|_)\delta_{0} \in I^{\perp}δ0∈I⊥. is a best aporoximation element for yyyyy by elements in I⊥I⊥I^(_|_)I^{\perp}I⊥ if and only if g0=F−F0g0=F−F0g_(0)=F-F_(0)g_{0}=F-F_{0}g0=F−F0, where I0I0I_(0)I_{0}I0 is a norm preserving extension of F|IFIF|_(I)\left.F\right|_{I}F|I to XXXXX.
The proporty ( UUUUU ) can be charactorized also in terms of some decompositions of the Lipschitz dual Lip 0I0I_(0)I{ }_{0} \mathrm{I}0I of X . To give this charactorization we noed first sone definitions and notations.
Lot It Id p0I→p0I→p_(0)Irarrp_{0} \mathrm{I} \rightarrowp0I→ Hip p0Ip0Ip_(0)Ip_{0} \mathrm{I}p0I denote the restriction operator, defined by :
(8) x(T)=F|I,F∈Lipp0Xx(T)=FI,F∈Lipp0Xquad x(T)=F|_(I),F in Lipp_(0)X\quad x(T)=\left.F\right|_{I}, F \in \operatorname{Lip} p_{0} Xx(T)=F|I,F∈Lipp0X,
and iev e: Liy 0I→∫(0I→∫_(0)I rarr int(:}{ }_{0} I \rightarrow \int\left(\right.0I→∫( IIp 0X)0X{:_(0)X)\left._{0} X\right)0X) denote the extension orerator, definod by :
(9)
e(f)=E(f;I),f∈LLp˙0X˙.e(f)=E(f;I),f∈LLp˙0X˙.e(f)=E(f;I)quad,quad f in LLp^(˙)_(0)X^(˙).e(f)=E(f ; I) \quad, \quad f \in L L \dot{p}_{0} \dot{X} .e(f)=E(f;I),f∈LLp˙0X˙.
Whore I(f;Y)I(f;Y)I(f;Y)\mathbb{I}(f ; Y)I(f;Y) denotes the set of all nom preserving extensions of I to XXXXX. Lot w:Iip0I→∫(Iip0I)w:Iip0I→∫Iip0Iw:I_(ip_(0))I rarr int(I_(ip_(0))I)w: I_{i p_{0}} I \rightarrow \int\left(I_{i p_{0}} I\right)w:Iip0I→∫(Iip0I) be the composition of the operators rrrrr and e, i.e.
(10)
w≡e∘x.w≡e∘x.w-=e@x.w \equiv e \circ x .w≡e∘x.
Then, for F∈Lip0IF∈Lip0IF inLip_(0)IF \in \operatorname{Lip}_{0} IF∈Lip0I, we have w(F)⇒e(r(F))=S(I(F);Y)w(F)⇒e(r(F))=S(I(F);Y)w(F)=>e(r(F))=S(I(F);Y)w(F) \Rightarrow e(r(F))=\mathbb{S}(I(F) ; Y)w(F)⇒e(r(F))=S(I(F);Y) and I(G)=I(F),‖G‖I=‖I(F)‖II(G)=I(F),‖G‖I=‖I(F)‖II(G)=I(F),quad||G||_(I)=||I(F)||_(I)I(G)=I(F), \quad\|G\|_{I}=\|I(F)\|_{I}I(G)=I(F),‖G‖I=‖I(F)‖I, for all G∈E(I(F);Y)G∈E(I(F);Y)G in E(I(F);Y)G \in E(I(F) ; Y)G∈E(I(F);Y) 。
In goneral, the operator wwwww is nulti-valued and w(F)w(F)w(F)w(F)w(F) is a conver subset of the ball of radius ‖I(F)‖Y‖I(F)‖Y||I(F)||_(Y)\|I(F)\|_{Y}‖I(F)‖Y and center 0 in Lip0XLip0XLip_(0)XL i p_{0} XLip0X.
Te can now state the theorem of characterization of property (U)(U)(U)(U)(U) :
THEOPI 3. If IIIII is a subset of a netric space XXXXX then the following enserions are equivalent : 2∘2∘2^(@)2^{\circ}2∘ I has sroperty (V) ; 2∘2∘2^(@)2^{\circ}2∘ Ivery function F∈Ljp0can be uniquely written in the F∈Ljp0can be uniquely written in the F inL_(j)p_(0)^("can be uniquely written in the ")F \in L_{j} p_{0}{ }^{\text {can be uniquely written in the }}F∈Ljp0can be uniquely written in the
form
(11) F=E+g,E∈w(P),g∈I⊥F=E+g,E∈w(P),g∈I⊥quad F=E+g,E in w(P),g inI^(_|_)\quad F=E+g, E \in w(P), g \in I^{\perp}F=E+g,E∈w(P),g∈I⊥,
and ‖I‖X>‖I‖X‖I‖X>‖I‖X||I||_(X) > ||I||_(X)\|I\|_{X}>\|I\|_{X}‖I‖X>‖I‖X, vinenever g≠0g≠0g!=0g \neq 0g≠0; 3∘G={H∈Lip0X,‖H‖Z=‖r(H)‖Y}3∘G=H∈Lip0X,‖H‖Z=‖r(H)‖Y3^(@)quadG={H inLip_(0)X,||H||_(Z)=||r(H)||_(Y)}3^{\circ} \quad \mathscr{G}=\left\{H \in \operatorname{Lip}_{0} X,\|H\|_{Z}=\|r(H)\|_{Y}\right\}3∘G={H∈Lip0X,‖H‖Z=‖r(H)‖Y} is the oniy subset of
Lipo XXXXX such that every I∈Lip0XI∈Lip0XI in Lip_(0)XI \in L i p_{0} XI∈Lip0X can be uniquely written in the 응펴 f=H+g,H∈G,g∈X⊥f=H+g,H∈G,g∈X⊥f=H+g,H in G,g inX^(_|_)f=H+g, H \in G, g \in X^{\perp}f=H+g,H∈G,g∈X⊥ and ‖F‖X>‖H‖X‖F‖X>‖H‖X||F||_(X) > ||H||_(X)\|F\|_{X}>\|H\|_{X}‖F‖X>‖H‖X if g≠0g≠0g!=0g \neq 0g≠0.
Proof. 10⇒2010⇒201^(0)=>2^(0)1^{0} \Rightarrow 2^{0}10⇒20. If the set YYYYY has property (U) then the oxtension operator 9 , delined by (9), is single-valued and so is the operator wwwww defined by (10). For F∈Ii0XF∈Ii0XF in Ii_(0)XF \in I i_{0} XF∈Ii0X the function D(F) ∈∈in\in∈ Lip XXXX_(X)X{ }_{X} XXX is the ouly norm preserving extension of I(F)I(F)I(F)I(F)I(F) to XXXXX, i.e. w(B)|Y=I(F)w(B)Y=I(F)w(B)|_(Y)=I(F)\left.w(B)\right|_{Y}=I(F)w(B)|Y=I(F) and ‖w(F)‖X=‖I(F)‖Y‖w(F)‖X=‖I(F)‖Y||w(F)||_(X)=||I(F)||_(Y)\|w(F)\|_{X}=\|I(F)\|_{Y}‖w(F)‖X=‖I(F)‖Y. It follows that g==F−w(F)∈Y⊥g==F−w(F)∈Y⊥g==F-w(F)inY^(_|_)g= =F-w(F) \in Y^{\perp}g==F−w(F)∈Y⊥ and F=w(F)+gF=w(F)+gF=w(F)+gF=w(F)+gF=w(F)+g is the unique decomposition of YYYYY With σ~∈I⊥σ~∈I⊥tilde(sigma)inI^(_|_)\tilde{\sigma} \in I^{\perp}σ~∈I⊥. By the definition (3) of Lipschitz norm we have ‖F→‖X⩾‖r(F)‖X‖F→‖X⩾‖r(F)‖X|| vec(F)||_(X) >= ||r(F)||_(X)\|\vec{F}\|_{X} \geqslant\|r(F)\|_{X}‖F→‖X⩾‖r(F)‖X. The equality ‖I‖X=‖I(F)‖Y‖I‖X=‖I(F)‖Y||I||_(X)=||I(F)||_(Y)\|I\|_{X}=\|I(F)\|_{Y}‖I‖X=‖I(F)‖Y implies that FFFFF is also a norm preserving extension of I(F)I(F)I(F)I(F)I(F) and, by the unicity of the axtension it follows F=w(F)F=w(F)F=w(F)F=w(F)F=w(F), so that B=F−w(F)=0B=F−w(F)=0B=F-w(F)=0B=F-w(F)=0B=F−w(F)=0. Hence ‖I‖X>‖I(F)‖Y‖I‖X>‖I(F)‖Y||I||_(X) > ||I(F)||_(Y)\|I\|_{X}>\|I(F)\|_{Y}‖I‖X>‖I(F)‖Y if g∈0g∈0g in0g \in 0g∈0. 2∘⇒3∘2∘⇒3∘2^(@)=>3^(@)2^{\circ} \Rightarrow 3^{\circ}2∘⇒3∘. Let F∈F∈F inF \inF∈ Lip 0X0X_(0)X{ }_{0} X0X and let F=H+gF=H+gF=H+gF=H+gF=H+g the decomposition of FFFFF given in 2∘2∘2^(@)2^{\circ}2∘. As H∈W(F)=O(I(R))H∈W(F)=O(I(R))H in W(F)=O(I(R))H \in W(F)=O(I(R))H∈W(F)=O(I(R)) it follows that I(F)≡r(H)I(F)≡r(H)I(F)-=r(H)I(F) \equiv r(H)I(F)≡r(H) and ‖H‖X=‖r(F)‖Y‖H‖X=‖r(F)‖Y||H||_(X)=||r(F)||_(Y)\|H\|_{X}=\|r(F)\|_{Y}‖H‖X=‖r(F)‖Y, i.e. H∈φjH∈φjH invarphi_(j)H \in \varphi_{j}H∈φj. The condition ‖F‖X>‖E‖X‖F‖X>‖E‖X||F||_(X) > ||E||_(X)\|F\|_{X} >\|E\|_{X}‖F‖X>‖E‖X, for g≠0g≠0g!=0g \neq 0g≠0, follows from the similar condition from 20202^(0)2^{0}20. 3∘⟹2∘3∘⟹2∘3^(@)Longrightarrow2^(@)3^{\circ} \Longrightarrow 2^{\circ}3∘⟹2∘. Let F∈Ip0XF∈Ip0XF in Ip_(0)XF \in I p_{0} XF∈Ip0X and let Z=H+g,H∈G,g∈I⊥Z=H+g,H∈G,g∈I⊥Z=H+g,H inG,g inI^(_|_)Z=H+g, H \in \mathscr{G}, g \in I^{\perp}Z=H+g,H∈G,g∈I⊥ be the unique decomposition of RRRRR given in 3∘3∘3^(@)3^{\circ}3∘. Then P−g=BP−g=BP-g=BP-g=BP−g=B, I(H)≡I(F)I(H)≡I(F)I(H)-=I(F)I(H) \equiv I(F)I(H)≡I(F) and, By Lomma 1, ‖I−g‖X=‖E‖X=‖I(H)‖Y=‖r(F)‖Y==a˙(B,Y⊥)‖I−g‖X=‖E‖X=‖I(H)‖Y=‖r(F)‖Y==a˙B,Y⊥||I-g||_(X)=||E||_(X)=||I(H)||_(Y)=||r(F)||_(Y)==a^(˙)(B,Y^(_|_))\|I-g\|_{X}=\|E\|_{X}=\|I(H)\|_{Y}=\|r(F)\|_{Y}= =\dot{a}\left(B, Y^{\perp}\right)‖I−g‖X=‖E‖X=‖I(H)‖Y=‖r(F)‖Y==a˙(B,Y⊥), which shows that ggggg is an element of best approximation for IIIII by elements in I⊥I⊥I^(_|_)I^{\perp}I⊥. If g1g1g_(1)g_{1}g1 is an other element of best approximation for TTTTT by elements in Y⊥Y⊥Y^(_|_)Y^{\perp}Y⊥ then, appealing again to Lema 1, there exists H1∈o(x(F))H1∈o(x(F))H_(1)in o(x(F))H_{1} \in o(x(F))H1∈o(x(F)) such that B1=F−B1B1=F−B1B_(1)=F-B_(1)B_{1}=F-B_{1}B1=F−B1 and ‖H1‖X=‖F−g1‖X=d(F,Y⊥)=‖r(F)‖Y=‖r(H2)‖YH1X=F−g1X=dF,Y⊥=‖r(F)‖Y=rH2Y||H_(1)||_(X)=||F-g_(1)||_(X)=d(F,Y^(_|_))=||r(F)||_(Y)=||r(H_(2))||_(Y)\left\|H_{1}\right\|_{X}=\left\|F-g_{1}\right\|_{X}=d\left(F, Y^{\perp}\right)=\|r(F)\|_{Y}=\left\|r\left(H_{2}\right)\right\|_{Y}‖H1‖X=‖F−g1‖X=d(F,Y⊥)=‖r(F)‖Y=‖r(H2)‖Y, which shows that H1∈GH1∈GH_(1)inGH_{1} \in \mathscr{G}H1∈G. Taking into acsount the unicity assumption in 3∘3∘3^(@)3^{\circ}3∘, it follows G≡G1G≡G1G-=G_(1)G \equiv G_{1}G≡G1. Therefore, Y⊥Y⊥Y^(_|_)Y^{\perp}Y⊥ is a Chebyshovian subspace of Iido XXXXX and, by Theorem 2, the set Y has property (U).
Theorem 3 is completely proved.
DEFINITION 3. A subset IIIII of a metric space IIIII is called an lifideel If its annihilator I1I1I^(1)I^{1}I1 has a complement GGGGG in Lip IIIII^(I)I^{I}II such that ‖F‖X=‖G‖X+‖E‖X‖F‖X=‖G‖X+‖E‖X||F||_(X)=||G||_(X)+||E||_(X)\|F\|_{X}=\|G\|_{X}+\|E\|_{X}‖F‖X=‖G‖X+‖E‖X, whonever I=G+EI=G+EI=G+EI=G+EI=G+E, with G∈GG∈GG inGG \in \mathcal{G}G∈G and E∈Y+,{0}E∈Y+,{0}E inY^(+),{0}E \in Y^{+},\{0\}E∈Y+,{0}. The subset YYYYY is said to have property (HB) if Y⊥Y⊥Y^(_|_)Y^{\perp}Y⊥ has a comple mentary subspace ξξxi\xiξ of Lip XX_(X){ }_{\mathrm{X}}X such that ‖F‖X⩾‖E‖X‖F‖X⩾‖E‖X||F||_(X) >= ||E||_(X)\|F\|_{\mathrm{X}} \geqslant\|E\|_{\mathrm{X}}‖F‖X⩾‖E‖X, ‖z‖X>‖G‖X‖z‖X>‖G‖X||z||_(X) > ||G||_(X)\|z\|_{\mathrm{X}}>\|G\|_{\mathrm{X}}‖z‖X>‖G‖X, Whenever F=G+EF=G+EF=G+EF=G+EF=G+E, with G∈YG∈YG inYG \in \mathscr{Y}G∈Y and H∈Σi,{0}H∈Σi,{0}H inSigma^(i),{0}H \in \Sigma^{i},\{0\}H∈Σi,{0}, for every function F∈Ip˙0IF∈Ip˙0IF in Ip^(˙)_(0)IF \in I \dot{p}_{0} IF∈Ip˙0I.
THEOREM 4. If a subset YYYYY of a metric space XXXXX has the property (BB) then YYYYY has the property (U) .
Proof. Suppose that YYYYY has the property (HB) and has not the properfy (J). Then there exists a function f∈f∈f inf \inf∈ Lip 0I0I_(0)I{ }_{0} I0I having two distinct extensions I1,I2I1,I2I_(1),I_(2)I_{1}, I_{2}I1,I2 Lip 0I0I_(0)I{ }_{0} \mathfrak{I}0I and the subspace I⊥I⊥I^(_|_)I^{\perp}I⊥ has a conplementary subspace GGGGG in Lip II^(I){ }^{I}I such that the condition in Definition 3 is fulfilled, implying Fi=Gi+EiFi=Gi+EiF_(i)=G_(i)+E_(i)F_{i}=G_{i}+E_{i}Fi=Gi+Ei, with Gi∈HGi∈HG_(i)inHG_{i} \in \mathcal{H}Gi∈H and Ei∈Y⊥Ei∈Y⊥E_(i)inY^(_|_)E_{i} \in Y^{\perp}Ei∈Y⊥ for 1=1,21=1,21=1,21=1,21=1,2. As F1−F2∈I⊥F1−F2∈I⊥F_(1)-F_(2)inI^(_|_)F_{1}-F_{2} \in I^{\perp}F1−F2∈I⊥ it follows that G1−G2=F1−I2G1−G2=F1−I2G_(1)-G_(2)=F_(1)-I_(2)G_{1}-G_{2}=F_{1}-I_{2}G1−G2=F1−I2 -- (E1−E2)∈I⊥E1−E2∈I⊥(E_(1)-E_(2))inI^(_|_)\left(E_{1}-E_{2}\right) \in I^{\perp}(E1−E2)∈I⊥, hence G1=G2=G(ζ∩I⊥={0}G1=G2=Gζ∩I⊥={0}G_(1)=G_(2)=G(zeta nnI^(_|_)={0}:}G_{1}=G_{2}=G\left(\zeta \cap I^{\perp}=\{0\}\right.G1=G2=G(ζ∩I⊥={0}, as ξξxi\xiξ and I⊥I⊥I^(_|_)I^{\perp}I⊥ are complementary subspaces of Ifipo XXXXX ). Therefore F1=G+H1F1=G+H1F_(1)=G+H_(1)F_{1}=G+H_{1}F1=G+H1 and F2=G+E2F2=G+E2F_(2)=G+E_(2)F_{2}=G+E_{2}F2=G+E2. Now, if E2≠0E2≠0E_(2)!=0E_{2} \neq 0E2≠0 then ‖F1‖Σ>‖G‖IF1Σ>‖G‖I||F_(1)||_(Sigma) > ||G||_(I)\left\|F_{1}\right\|_{\Sigma}>\|G\|_{I}‖F1‖Σ>‖G‖I so that ‖f‖I==‖F1‖Σ>‖G‖Σ⩾‖r(G)‖I=‖I‖Y‖f‖I==F1Σ>‖G‖Σ⩾‖r(G)‖I=‖I‖Y||f||_(I)==||F_(1)||_(Sigma) > ||G||_(Sigma) >= ||r(G)||_(I)=||I||_(Y)\|f\|_{I}= =\left\|F_{1}\right\|_{\Sigma}>\|G\|_{\Sigma} \geqslant\|r(G)\|_{I}=\|I\|_{Y}‖f‖I==‖F1‖Σ>‖G‖Σ⩾‖r(G)‖I=‖I‖Y. If H1≡0H1≡0H_(1)-=0H_{1} \equiv 0H1≡0 then H2≠H1=0H2≠H1=0H_(2)!=H_(1)=0H_{2} \neq H_{1}=0H2≠H1=0, hence G=F1G=F1G=F_(1)G=F_{1}G=F1 and, by Desinition 3 , the equality F1=F2+H2F1=F2+H2F_(1)=F_(2)+H_(2)F_{1}=F_{2}+H_{2}F1=F2+H2 implies ‖P‖I=‖F1‖X>‖F2‖X=‖S‖I‖P‖I=F1X>F2X=‖S‖I||P||_(I)=||F_(1)||_(X) > ||F_(2)||_(X)=||S||_(I)\|P\|_{I}=\left\|F_{1}\right\|_{X}>\left\|F_{2}\right\|_{X}=\|S\|_{I}‖P‖I=‖F1‖X>‖F2‖X=‖S‖I. The obtained contradictions shows that the set IIIII cannot have the propetty (HB). Theorem 4 is completejy proved.
THEOREIX 5. If the subset YYYYY of XXXXX has the property ( HBHBHBH BHB ) and F∈Lipp0,FF∈Lipp0,FF in Lipp_(0),FF \in \operatorname{Lip} p_{0}, FF∈Lipp0,F of 0 , then F∈GF∈GF inGF \in \mathscr{G}F∈G if and only if ‖r(F)‖I=‖F‖X‖r(F)‖I=‖F‖X||r(F)||_(I)=||F||_(X)\|r(F)\|_{I}=\|F\|_{X}‖r(F)‖I=‖F‖X, There ζζzeta\zetaζ is the complementary subspace of r⊥r⊥r^(_|_)r^{\perp}r⊥ given in Definition 3 .
Proof. Let F∈ξ,F∉0F∈ξ,F∉0F in xi,F!in0F \in \xi, F \notin 0F∈ξ,F∉0, and let GGGGG be a norm preserving extension of x(F)x(F)x(F)x(F)x(F) to I. I. I_(". ")I_{\text {. }}I. Let G=G1+H1,G1∈ξ,H2∈X⊥G=G1+H1,G1∈ξ,H2∈X⊥G=G_(1)+H_(1),G_(1)in xi,H_(2)inX^(_|_)G=G_{1}+H_{1}, G_{1} \in \xi, H_{2} \in X^{\perp}G=G1+H1,G1∈ξ,H2∈X⊥ be the
decomposition of GGGGG given by Definition 3. Supposing E1≠0E1≠0E_(1)!=0E_{1} \neq 0E1≠0, one obtains the contradiction ‖G‖I>‖G1‖I⩾‖I(G1)‖I=‖I(I)‖I==‖G1‖2‖G‖I>G1I⩾IG1I=‖I(I)‖I==G12||G||_(I) > ||G_(1)||_(I) >= ||I(G_(1))||_(I)=||I(I)||_(I)==||G_(1)||_(2)\|G\|_{I}>\left\|G_{1}\right\|_{I} \geqslant\left\|I\left(G_{1}\right)\right\|_{I}=\|I(I)\|_{I}= =\left\|G_{1}\right\|_{2}‖G‖I>‖G1‖I⩾‖I(G1)‖I=‖I(I)‖I==‖G1‖2. Therefore H1=0H1=0H_(1)=0H_{1}=0H1=0 and G=G1∈ζ1G=G1∈ζ1G=G_(1)inzeta_(1)G=G_{1} \in \zeta_{1}G=G1∈ζ1. As ζ1ζ1zeta_(1)\zeta_{1}ζ1 is a subspace of iip 00_(0){ }_{0}0 it follows F−G∈GF−G∈GF-G in GF-G \in GF−G∈G. But F−GF−GF-GF-GF−G is in I−I−I-I-I− too, bocauso r(F)=r(G)r(F)=r(G)r(F)=r(G)r(F)=r(G)r(F)=r(G), so that R−G∈ξ∩X⊥={0}R−G∈ξ∩X⊥={0}R-G in xi nnX^(_|_)={0}R-G \in \xi \cap X^{\perp}=\{0\}R−G∈ξ∩X⊥={0}, i.s. F=G∈GF=G∈GF=G inGF=G \in \mathcal{G}F=G∈G.
Conversely, suppose that F∈Kp0x~,F≠0F∈Kp0x~,F≠0F in Kp_(0) tilde(x),F!=0F \in K p_{0} \tilde{x}, F \neq 0F∈Kp0x~,F≠0, is such that ‖x(F)‖x=‖P‖X(>0)‖x(F)‖x=‖P‖X(>0)||x(F)||_(x)=||P||_(X)( > 0)\|x(F)\|_{x} =\|P\|_{X}(>0)‖x(F)‖x=‖P‖X(>0). Lot I=G+HI=G+HI=G+HI=G+HI=G+H with G∈ξ,H∈ILG∈ξ,H∈ILG in xi,H inI^(L)G \in \xi, H \in I^{L}G∈ξ,H∈IL. If E≠0E≠0E!=0E \neq 0E≠0 then ‖P‖I>‖G‖X‖P‖I>‖G‖X||P||_(I) > ||G||_(X)\|P\|_{I}>\|G\|_{X}‖P‖I>‖G‖X and the equality r(I)=r(G)r(I)=r(G)r(I)=r(G)r(I)=r(G)r(I)=r(G) gives the contradic tion ‖x(F)‖Z=‖F‖Z>‖G‖Z=‖x(G)‖Y=‖r(F)‖Y‖x(F)‖Z=‖F‖Z>‖G‖Z=‖x(G)‖Y=‖r(F)‖Y||x(F)||_(Z)=||F||_(Z) > ||G||_(Z)=||x(G)||_(Y)=||r(F)||_(Y)\|x(F)\|_{Z}=\|F\|_{Z}>\|G\|_{Z}=\|x(G)\|_{Y}=\|r(F)\|_{Y}‖x(F)‖Z=‖F‖Z>‖G‖Z=‖x(G)‖Y=‖r(F)‖Y, which shows that H=0H=0H=0H=0H=0 and F=G∈φSF=G∈φSF=G invarphi_(S)F=G \in \varphi_{S}F=G∈φS.
THFOPN 6. If the subset YYYYY of a metric snnce XXXXX has the property (HB) then the subsoace ζζzeta\zetaζ (given in Definition 3) is dropetrically isomorphic to the space Lip 0I0I_(0)^(I){ }_{0}{ }^{I}0I.
Proof. By Theorem 4, tae subset YYYYY has the property (U) , so that the restriction r1r1r_(1)r_{1}r1 of the restriction operator rrrrr to ξjξjxi_(j)\xi_{j}ξj is single-valued and linear. By Theorem 5, ‖r1(G)‖j=‖G‖Xr1(G)j=‖G‖X||r_(1)(G)||_(j)=||G||_(X)\left\|r_{1}(G)\right\|_{j}=\|G\|_{X}‖r1(G)‖j=‖G‖X, for all G∈ζξG∈ζξG inzeta_(xi)G \in \zeta_{\xi}G∈ζξ, showing that T1T1T_(1)T_{1}T1 is an loometry.
THEORSU 7. If the subset XXXXX of a motric space XXXXX pes the proper ty (HB) , tien the extension operator ⋅∣⋅∣*∣\cdot \mid⋅∣ ip 0Y→0Y→_(0)Yrarr_{0} \mathrm{Y} \rightarrow0Y→ Lip cXcX_(c)X_{c} \mathrm{X}cX de linear.
Proof. Let :~:~tilde(:)\tilde{:}:~ : Lip I→ζjI→ζjIrarrzeta_(j)\mathrm{I} \rightarrow \zeta_{j}I→ζj be the inverse of the restiliction operator r1=r|er1=rer_(1)=r|_(e)r_{1}=\left.r\right|_{e}r1=r|e & ⟶⟶longrightarrow\longrightarrow⟶ Lip 0I0I_(0)I{ }_{0} \mathrm{I}0I which, by Theorem 6, is an isoattrical isomorpicism between ζζzeta\zetaζ and Lipo YYYYY. Then is linear and e=j∘0~e=j∘0~e=j@ tilde(0)e=j \circ \tilde{0}e=j∘0~ where j:ζ→j:ζ→j:zeta rarrj: \zeta \rightarrowj:ζ→ Lip jxjxjxj xjx, denotes the imbedding operator of ζζzeta\zetaζ into Ling xxxxx.
RAITPRINCES
1 ALPSEN, E.M., EPFROSS, E. , Structure in real Banach speces , calt ac 1 ann. of Math. 96 (1972), 98-173.
2 CZIPSTR, J., GHENR, L., Extension of Functions Eatisfying a Liocchite conditions, Acta Hath, Acad. Sci. Hungar 6 (1955), 213-220
3 Tatrourr, E ., Sélections 11néaires essociées au Théorige de Fohn - Expach , J. of Functional Analysis 21 (2972), 436-452.
4 EETIMSILD, J., 14-jdec.ls, HB - subroaces and Conpact Operators, Indiana Univ. Math. J. 28 (6) (1979), 927 - 934.
5 EMTHIPID, J., Anote on M-ideals in B(X), Proc, Amer. Hat. Soc. 98 (1) (1980) , 89-92.
6 HOLIES, R.B., SCRANTON, B., WARD, J.D., Approzimation from the appe of compect operators and other x−x−x-x-x− ideals, Duke Math. J. 42 (1975), 259-26%.
7 BOLITS, R.B., Geonetric Functionol Anclssie and its Applicationg, Springer - Verlag, Now York - Heidelberg - Berlis , 1975 •
8 Mo SHANE, E.J., Thonsion of rango of functions, Bull. Amor. Math. Soc. 40 (1954) , 837-842.
9 JOEWSON, J.A., Benech Speces of Jipschitz Punctions and Vector - Velued Iipschitz Functions, Trans, knor. Meth. Soc. 248 (1970), 147-169 a pt at eccon abill Eipschite Functions, J. Appror. Theory 29 (3) (1977) , 222-230 。
11 RESTPS, R.R., Uniqueness of Bohn - Banach Extension and Uniqus Best Approximation, Trans. Amer. Math. Boc. 45 (1950),238−255(1950),238−255(1950),238-255(1950), 238-255(1950),238−255.
12 OJ1, S., On the uniquaness of the norr proservine extengion of linear luuctional in the Hohu - Bancon Feooren, Proc. Acad. Science Estonian SSR 33 (4) (2984), 422-433 (Russian)
Insvitutul do Calcul
Oficiul Postal 1
C. P. 68
3400 Cluj-Hiapoca
Romenia
This note is in final form and no version of 1t is or will be submitted for publication elsawhere.