Let (X,d)(X, d) and (Y,rho)(Y, \rho) be two metric spaces. For alpha in(0,1]\alpha \in(0,1] a function f:X rarr Yf: X \rightarrow Y is called Hölder of class alpha\alpha if there exists M >= 0M \geqslant 0 such that
(1) quad(E(x),P(y)) <= U(d(x,y))^(alpha)quad\quad(E(x), P(y)) \leq U(d(x, y))^{\alpha} \quad,
for all x_(v),y in Xx_{v}, y \in X.
Denote by Lambda_(alpha)(X,Y)\Lambda_{\alpha}(X, Y) the set of all Hölder functions of class alpha\alpha from XX to YY.
If II is a metric linear space then, equiped with the pointvise operations of addition and multiplication by scalars, ^^_(a)(X,Y)\wedge_{a}(X, Y) become a linear space. If Y=RY=R then ^^^_(alpha)(X,R)\bigwedge_{\alpha}(X, R) is also a lattice (the order is defined pointwisely too).
For f in^^_(alpha)(X,R)f \in \wedge_{\alpha}(X, R) put
(2) ||f||_(alpha)=s u p{|f(x)-f(y)|//(d(x,y))^(alpha):quad x,y in X,quad x in y}\|f\|_{\alpha}=\sup \left\{|f(x)-f(y)| /(d(x, y))^{\alpha}: \quad x, y \in X, \quad x \in y\right\}, the smallest number M >= 0M \geqslant 0 for which the inequality
(3) quad|f(x)-f(y)| <= M(d(x,y))^(alpha)\quad|f(x)-f(y)| \leq M(d(x, y))^{\alpha},
holds for all x,y in Xx, y \in X.
Obviously ||f||_(alpha) >= 0\|f\|_{\alpha} \geqslant 0 and ||f||_(alpha)=0\|f\|_{\alpha}=0 if and only if f=f= const. , for all f inLambda_(alpha)(X,R)f \in \Lambda_{\alpha}(X, R).
Let x_(0)in Xx_{0} \in X be fixed and let
.. (4) quadLambda_(alpha)(x_(0),X,R)={f inLambda_(alpha)(X,R),f(x_(0))=0}\quad \Lambda_{\alpha}\left(x_{0}, X, R\right)=\left\{f \in \Lambda_{\alpha}(X, R), f\left(x_{0}\right)=0\right\}.
Then ^^^_(alpha)(x_(0),x,R)\bigwedge_{\alpha}\left(x_{0}, x, R\right) is a subspace of Lambda_(alpha)(X,R){\Lambda_{\alpha}}(X, R) and the functional defined by (2) is a norm on this subspace and is called the Hölder norm of ff.
We say that two functions f,g in^^^_(alpha)(X,R)f, g \in \bigwedge_{\alpha}(X, R) are equivalent if f-g=f-g= const. and we shall denote this by frug.
It is immediate that the quatient space of nnn_(alpha)(X,R)\bigcap_{\alpha}(X, R) by this equivalence relation is isomorphic to ^^^_(alpha)(x_(0),X,R)\bigwedge_{\alpha}\left(x_{0}, X, R\right).
A very important problem in the theory of Hölder functions is the extension problem. Hore exactly, let (X,d),(Y,rho)(X, d),(Y, \rho) be two metric spaces and let ZCX . The extension problem is the following: for f inLambda_(alpha)(Z,Y)f \in \Lambda_{\alpha}(Z, Y) find F inLambda_(alpha)(X,Y)F \in \Lambda_{\alpha}(X, Y) such that
{:(5)f= eta|_(Z)quad" and "quad||f||_(alpha)=||F||_(alpha)quad.:}\begin{equation*}
f=\left.\eta\right|_{Z} \quad \text { and } \quad\|f\|_{\alpha}=\|F\|_{\alpha} \quad . \tag{5}
\end{equation*}
The function FF is called a norm preserving extension of ff.
For alpha=1\alpha=1 (the case of Lipschitz functions) the problem was extensively studied. The existence of a norm preserving extension for every f inLambda_(d)(Z,Y)f \in \Lambda_{d}(Z, Y) depends on the properties of the sets ZZ and YY. A positive solution for the extension problem in the case Y=BY=B and for XX arbitrary was given by Mc SHANE [14] and by GG. MLNTY [11] in the case when XX and YY are Hilber spaces.
If XX and YY are arbitrary metric spaces (ever Banach spaces) the extensions is not always possible as was shown by B. Gi JNBAUIA [5] and s.o. SCHÖENBECK [12], [13] .
T.U. PLETT [4] proved that if XX and YY are normed spaces and ZsubX\mathrm{Z} \subset \mathrm{X} is convex, closed, bounded of diameter delta\delta and contains a ball of radius r > 0r>0 then forf o r every f inLambda_(1)(Z,X)f \in \Lambda_{1}(Z, X) there exists
86 - F inLambda_(1)(X,Y)F \in \Lambda_{1}(X, Y) such that F|_(2)=f\left.F\right|_{2}=f and ||F||_(1)=(delta )/(r)*||f||_(1)\|F\|_{1}=\frac{\delta}{r} \cdot\|f\|_{1}.
If every function f inLambda_(alpha)(Z,Y)f \in \Lambda_{\alpha}(Z, Y) has an extension F inLambda_(alpha)(X,Y)F \in \Lambda_{\alpha}(X, Y)
it is natural to ask if this extension is unique or not. It was shown that the question of the unicity of the norm preserving extension is closely related to some approximation problems in the space Lambda_(alpha)(X,Y)\Lambda_{\alpha}(X, Y) (see [7], [8], [10] ) .
In the following we shall denote Lip(X,Y)=Lambda_(1)(X,Y)\operatorname{Lip}(X, Y)=\Lambda_{1}(X, Y). If XX is a Banach space and SS is a closed ball of radius r > 0r>0 in XX then as was shown by T.M. FLETT [4] there exists a function F in Lip(X,X)F \in \operatorname{Lip}(X, X) such that
(6) quad||F||_(1)=2∣I||_(1)\quad\|F\|_{1}=2 \mid I \|_{1}
where f=F|_(S)f=\left.F\right|_{S}.
THEOREM 1. Let XX be a Banach space and let f in Ilp(X,X)f \in \operatorname{Ilp}(X, X). Suppose that the following conditions hold true :
a) There exists a convex, closed, bounded set C of diameter rho\rho and containing a ball of radius delta > 0\delta>0 guch that
b) Every extension FF of. ff yexifies
(7) quad|f-F|_(2) < 1-|I|_(G)|_(2)(rho )/(delta)\quad|f-F|_{2}<1-\left.|I|_{G}\right|_{2} \frac{\rho}{\delta}.
Then there exists a unique x^(')in Xx^{\prime} \in X auch that f(x^(-))=x^(**)f\left(x^{-}\right)=x^{*}. (The function ff has a unique fix point x^(**)in xx^{*} \in x ).
Proof. Let f in Lip(X,I)f \in \operatorname{Lip}(X, I) and C sub XC \subset X such that condition a) is verified. By the above quated result of Mett there exists F in Lip(x,x)F \in \operatorname{Lip}(x, x) such that F|_(c)=4|_(c)\left.F\right|_{c}=\left.4\right|_{c} and |F|_(1)=|f|_(0)|_(1)f|F|_{1}=\left.|f|_{0}\right|_{1} f. Then ||f||_(1)=||f-F+FI_(2) <= ||f-F||_(2)+|I_(1) < 1-|F|_(C)|_(I)(rho )/(delta)+||P|_(C)||_(I)(rho )/(sigma)=1\|f\|_{1}=\left\|f-F+F I_{2} \leq\right\| f-\left.F\left\|_{2}+\left|I_{1}<1-|F|_{C}\right|_{I} \frac{\rho}{\delta}+\right\| P\right|_{C} \|_{I} \frac{\rho}{\sigma}=1.
87 -
Since ||f(x)-f(y)|| <= ||P||_(4)||x-y||\|f(x)-f(y)\| \leq\|P\|_{4}\|x-y\| for all x,y in Xx, y \in X it follows that II is a contraction on XX and by Banach contraction principle there exists a unique x^(**)in Xx^{*} \in X such that f(x^(**))=x^(**)f\left(x^{*}\right)=x^{*}. Theorem is proved .
COROLLARY 1. Let XX be a Banach space and f in Lip(X,X)f \in \operatorname{Lip}(X, X). Suppose that there exists a closed ball s in Xs \in X of radius delta > 0\delta>0 such that every extension FF of f//sf / s verifies the condition :
Then ff has a unique fix point in XX.
Proof. The diameter of SS is rho=2delta\rho=2 \delta and by (8) ||f|_(S)||_(1) < (1)/(2)\left\|\left.f\right|_{S}\right\|_{1}<\frac{1}{2} so that the condition a) and b) from Theorem 1 are verified.
Bemark 1. Let cc be as in Theorem 1 and f in Lip(X,X)f \in \operatorname{Lip}(X, X). If ||P|_(C)||_(1)=0\left\|\left.P\right|_{C}\right\|_{1}=0 then ||P(x)-f(y)||=0\|P(x)-f(y)\|=0 for all x,y in Cx, y \in C and f(x)=f(y)=z in Xf(x)=f(y)=z \in X for all x,y in Cx, y \in C. Since F|_(C)=f\left.F\right|_{C}=f and ||F||_(1)==0(rho )/(f)=0\|F\|_{1}= =0 \frac{\rho}{f}=0 it follows that F(x)=zF(x)=z for all x in Xx \in X. Therefore the condition (7) from Theorem 1 becomes
( {:7^(@))\left.7^{\circ}\right)
||x||_(1) < 1.\|x\|_{1}<1 .
1.e. ff is a contraction on XX.
If ||P|_(G)||_(1)=0\left\|\left.P\right|_{G}\right\|_{1}=0 then P=P= const. on CC and the extension I in I4p( bar(s),X)I \in I 4 p({ } \bar{s}, X) is uniqus.
3. We consider the following problem tt for a metric space XX, a subset MM of XX and a function f inLambda_(alpha)(M,R)f \in \Lambda_{\alpha}(M, R) find
{:(9)min{f(y):y in M}:}\begin{equation*}
\min \{f(y): y \in M\} \tag{9}
\end{equation*}
In conorete problems the set 1 I is usually determined by some restriotions and the function ff is replaced by the function bar(f)\bar{f}
defined by
P(x)={[f(x),",",x in M],[+oo,",",x in XM]:}P(x)=\left\{\begin{array}{lll}
f(x) & , & x \in M \\
+\infty & , & x \in X M
\end{array}\right.
Obviously min{f(y):y inM}=min{ bar(I)(x):x in X}\min \{f(y): y \in \mathbb{M}\}=\min \{\bar{I}(x): x \in X\}.
HIRIART-URRUTY [6] proved that if XX is a Banach space M sub XM \subset X is closed and f in Lip(M,R)f \in \operatorname{Lip}(M, R), then the problem
min{f(y):y in M}\min \{f(y): y \in M\}
can be replaced by the problem :
min{F_(1)(x):x in X},\min \left\{F_{1}(x): x \in X\right\},
where F_(1)(x)=i n f_(y in M)[f(y)+||f||_(1)*||x-y||],x in XF_{1}(x)=\inf _{y \in M}\left[f(y)+\|f\|_{1} \cdot\|x-y\|\right], x \in X.
In this note we shall give some similar results in the case of a metric space XX and for a function f inLambda_(a)(M,R),0 < pi <= 1f \in \Lambda_{a}(M, R), 0<\pi \leq 1.
Let XX be a metric space, let MM be a closed subset of XX and let f inLambda_(alpha)(M,R)f \in \Lambda_{\alpha}(M, R). By a result in [9] the function F_(1)F_{1} defined by
(10) quadF_(1)(x)=i n f_(y in M)[P(y)+||Ih_(alpha)(a(x,y))^(alpha)],x in X\quad F_{1}(x)=\inf _{y \in M}\left[P(y)+\| I h_{\alpha}(a(x, y))^{\alpha}\right], x \in X
is in Lambda_(alpha)(X,R)\Lambda_{\alpha}(X, R) and
for all y inMy \in \mathbb{M}.
THEORIM 2. Let XX be a metric space, MM a closed subset of XX and f in^^_(alpha)(H,R)f \in \wedge_{\alpha}(H, R). Then y_(0)in My_{0} \in M is a minimum point for ff on MM if and only if J_(0)J_{0} is a minimum point for F_(1)F_{1} on XX.
Proof. Let y_(0)y_{0} be a minimum point for ff on UU and let E_(1)E_{1} be defined by (10). For every x in Mx \in M we have
If x!in Ix \notin I, the set II being closed, there exists delta > 0\delta>0 such that a(x,y) >= 6 > 0a(x, y) \geq 6>0 for all y in My \in M. Therefore
{:[I_(I)(x)=i n f_(j in f)[f(y)+||f||_(alpha)(d(x,y))^(alpha d)] <= ],[ >= i n f_(j in M)[f(y)+||f||_(alpha)delta]=||f||_(alpha)delta^(alpha)+f(y_(0)) > f(y_(0))]:}\begin{aligned}
I_{I}(x) & =\inf _{j \in f}\left[f(y)+\|f\|_{\alpha}(d(x, y))^{\alpha d}\right] \leq \\
& \geqslant \inf _{j \in M}\left[f(y)+\|f\|_{\alpha} \delta\right]=\|f\|_{\alpha} \delta^{\alpha}+f\left(y_{0}\right)>f\left(y_{0}\right)
\end{aligned}
so that y_(0)y_{0} is a minimum point for F_(1)F_{1} on XX.
Conversely, suppose that Y_(0)Y_{0} is a minimum point for P_(1)P_{1} on XX. If we would aher that y_(0)in My_{0} \in M then, as I_(1)|_(M)=f\left.I_{1}\right|_{M}=f, it would follow that y_(0)y_{0} is a minimum point for ff on MM.
Suppose, on the cuntrary, that y_(0)!in My_{0} \notin M. Then, since MM is clossd,
d(J_(0),M)=i n f{d(J_(0),J),y inM}=q > 0.d\left(J_{0}, M\right)=\inf \left\{d\left(J_{0}, J\right), y \in \mathbb{M}\right\}=q>0 .
By the definition of F_(1)F_{1} we have
F_(1)(y_(0))=i n f_(g in M)[f(y)+||f||_(alpha)(alpha(y_(0),y))^(alpha)],F_{1}\left(y_{0}\right)=\inf _{g \in M}\left[f(y)+\|f\|_{\alpha}\left(\alpha\left(y_{0}, y\right)\right)^{\alpha}\right],
so that, for every epsi > 0\varepsilon>0, there exists y_(epsi)inIy_{\varepsilon} \in \mathbb{I} such that
If ||f||_(alpha)=0\|f\|_{\alpha}=0 then f=f= const on MM and J_(0)quad(J_{0} \quad( as every other point in MM ). will be a minimum point for ff on MM.
Letting n rarr oon \rightarrow \infty in the inequality 0 < q <= ((q)/(n))^((1)/(4))0<q \leqslant\left(\frac{q}{n}\right)^{\frac{1}{4}} one obtains a contradiction. Theorem 2 is proved.
Let f in^^_(a)(M,R)f \in \wedge_{a}(M, R) and let
F_(2)(x)=s u p_(y in M)[f(y)^(')-||f||_(d)(d(x,y))^(d)],x in XF_{2}(x)=\sup _{y \in M}\left[f(y)^{\prime}-\|f\|_{d}(d(x, y))^{d}\right], x \in X
The function B_(2)B_{2} has the properties :
F_(2)|_(M)=P quad" and "quad||F_(2)||_(alpha)=||PR_(alpha),\left.F_{2}\right|_{M}=P \quad \text { and } \quad\left\|F_{2}\right\|_{\alpha}=\| P R_{\alpha},
(see [9] ) .
THEOREM 3. Let XX be a metric space, M a closed subset of XX and f inLambda_(alpha)(M,R)f \in \Lambda_{\alpha}(M, R). Then y_(0)in My_{0} \in M is a maximum point for ff on MM if and anly if y_(0)y_{0} is a maximum point for F_(2)F_{2} on XX.
The proof of this Theorem is simillar to the proof of Theorem 2.
Remark 2. If XX is a metric linear space, MM a closed convex subset of XX and f in^^_(alpha)(M,R)f \in \wedge_{\alpha}(M, R) is convex, then ff has minimum point on MM. The function F_(1)F_{1}, defined by (10), has the same minimum on XX as ff on MM. Furthermore the function FF is conver too (see [8]).
If ff is a concave function on MM, then the function F_(2)F_{2} is concave too on XX and the maximum of F_(2)F_{2} on XX equals the maximum of ff on M\mathbb{M}.
REFFERENCES
ARONSSON, G. Brtension of runctions satisfying Lipschitz conditions, Arictiv for Mathematik 6 (1967), 551 - 561.
CZIPBKR, J., GEHER, Io, Extension of functions satisfying a Lipgchite condition , Acta Math. Sci. Hungar 6 (1955), 213-220213-220.
DANZER, L., GRUNBAMM, B., KLKE, Psi_(0)\Psi_{0}, Helly's theorem and its relatives , Proc. Sympos. Pure Mathem. 7, Amer. Math. Soc., Providence, R.I. (1963), 101-180.
FIEIT, T.M., Extension of Iipschits functions, J. London Math. Soc. 7 (1974), 604-608.
GRUNBAUM, B., A generalization of Theorems of Kirszbraum and Mintri . Proc. Amer. Math. Soc, 13 (1962) , 812-814812-814.
HIRIART-URRUTY, J.B., Extension of Lipschitz functions, Preprint 1980 ( 23 pp. ).
MUSTATA, C., Best Approximation and Unique Extension of Ilpschitz Functions, Journal of Approx. Theory 19, 3 (1977), 222-230.
MUSTATTA, C., COBZAS, B., Norm Preserving Extension of Convex Lipschitz Functions, Journal of Approx. Theory 24, 3 (1978), 236-244.
MUSTATA, C., About the determination of extrema of a
Hölder functions, Seminar of Funct.Anal. Nunerical Methods, Babes-Bolyai University, Fac. of Math. Preprint Nr. 1 (1983), 107-116.
10. MUSTÁTA, C., Asupra unicitătil prelungiril functilior Hölder reale , Seminarul itinerant de Bcuatii Functionale, Aproximars si Convexitate, ClujNapoca, 1930 .
11. UINIY, G.J., on the extension of Lipschitz, LipschitzHölder contruous, and monotone functions, Bull. Amer. Math. Soc. 76 (1970), 334-339 .
12. SCHONBACK, S.O., Extension of nonlinear contractions, Bull. Amer. Math . Soc. 72 (1966), 99-101 .
13. SCHONBECK, S.O., On the extension of Lipschitz maps. Arkiv för Mathematik 7 (1967-1969), 201-209.
14. MCSHANE, E.J., Extension of range of functions, Bull. Amer. Math. Soc., 40 (1934), 837 - 842 .
15. VALENTINL, F.A., A Lipschitz condition preserving extension for a vector function, Amer. J. Math. 67 (1945) , 83-93.