It is proved by examples that there are (connected) non-closed sets with the fixed point property for contractions in complete metric spaces. In a Banach space, a convex set with nonvoid interior having the fixed point property for contractions is necessarily closed.
Authors
Mira-Cristiana Anisiu
Tiberiu Popoviciu Institute of Numerical Analysis Romanian Academy, Romania
Valeriu Anisiu
Babes-Bolyai University Faculty of Mathematics Cluj-Napoca, Romania
Keywords
?
Paper coordinates
M.-C. Anisiu, V. Anisiu, On the closedness of sets with the fixed point property for contractions, Rev. Anal. Numér. Théor. Approx. 26 (1-2) (1997), 13-17 (pdf filehere)
[1] E. H. Connell, Properties of fixed point spaces, Proc. Amer. Math. Soc. 10(1959), 974-979
[2] T. K. Hu, On a fixed point theorem for metric spaces, Amer. Math. Monthly 74(1967), 436-437
[3] P. V. Subrahmanyam, Completeness and fixed points, Monatsch. f¸r Math. 80(1975), 325-330
[4] I. A. Rus, Maximal fixed point structures, Zilele academice clujene, 18-23 nov. 19
1997-Anisiu-Anisiu-OnClosedness
On the closedness of sets with the fixed point property for contractions
Mira-Cristiana Anisiu"T. Popoviciu" Institute of Numerical Analysis37, Republicii st., 3400 Cluj-NapocaRomaniaValeriu Anisiu"Babeş-Bolyai" UniversityFaculty of Mathematics1, Kogălniceanu st., 3400 Cluj-NapocaRomaniaIn memoriamTiberiu Popoviciu
AMS Classification: 47H10, 54G20
Abstract
It is proved by examples that there are (connected) non-closed sets with the fixed point property for contractions in complete metric spaces. In a Banach space, a convex set with nonvoid interior having the fixed point property for contractions is necessarily closed.
1. Introduction
In his communication [4], I. A. Rus mentioned the following result of Hu from 1967, which gives a characterization of metric completeness:
Theorem 1.1. [2] A metric space ( X,dX, d ) is complete if and only if for each closed subset YY of XX any contraction f:Y longrightarrow Yf: Y \longrightarrow Y has a fixed point.
Hu has made the remark that "closed subset" can be replaced by "infinite denumerable closed set", and it is sufficient to consider contractions with a given constant rr.
As stated in [4], for the fixed point structure theory it would be desirable to have a result related to that of Hu , namely: given a complete metric space ( X,dX, d ) and a nonvoid subset YY such that any contraction f:Y longrightarrow Yf: Y \longrightarrow Y has a fixed point, the subset YY is necessarily closed. Unfortunately, some examples given by Connell [1] in 1959 for cross products show, as Subrahmanyam has emphasized in [3], that there are non-closed subsets of a complete metric space for which each contraction has a fixed point. The paper of Subrahmanyam includes an abstract generalization of such an example.
In the second section we describe an example of Connell and the way it provides a non-closed set on which each contraction (in fact any continuous function) has a fixed point. The set is connected but not path connected. A second example, also appearing in Connell's paper, provides a path connected, non-closed set in R^(2)\mathbb{R}^{2} on which each continuous function has a fixed point. In fact, Connell was not concerned about the connectedness properties of these sets.
In the third section we prove a theorem which, in the setting of Banach spaces, gives a class of sets which are necessarily closed if they have the fixed point property for contractions. So in this case Rus' problem has an affirmative answer.
The final section contains some remarks on another class of sets and on the way of providing contractions without fixed points in the case of normed spaces.
2. Examples in complete metric spaces of non-closed sets having the fixed point property for contractions
In the paper [1], Connell was interested in giving examples of bounded, but nonclosed sets YY, each continuous function f:Y longrightarrow Yf: Y \longrightarrow Y having a fixed point. Actually these examples are also good for our purposes. As these sets have in addition some connectedness properties, we mention the following definitions.
A topological space XX is connected if it cannot be written as a disjoint union of two open nonvoid sets; it is path connected if for every pair x_(1),x_(2)x_{1}, x_{2} of points in XX there exists a continuous function phi:[0,1]longrightarrow X\phi:[0,1] \longrightarrow X such that phi(0)=x_(1)\phi(0)=x_{1} and phi(1)=x_(2)\phi(1)=x_{2}.
Example 1. A set Y subeR^(2)Y \subseteq \mathbb{R}^{2} which is non-closed but has the fixed point property for each contraction f:Y longrightarrow Yf: Y \longrightarrow Y. The set YY is connected but not path connected.
Let there be given the function varphi:[0,1]longrightarrow[0,1]\varphi:[0,1] \longrightarrow[0,1],
In R^(2)\mathbb{R}^{2} one considers the connected, but not path connected set
Y={(t,varphi(t))inR^(2):0 <= t <= 1},Y=\left\{(t, \varphi(t)) \in \mathbb{R}^{2}: 0 \leq t \leq 1\right\},
for which bar(Y)=Y uu({1}xx[-1,1])\bar{Y}=Y \cup(\{1\} \times[-1,1]), so YY is non-closed. The set YY is connected because Y_(0)sube Y sube bar(Y_(0))Y_{0} \subseteq Y \subseteq \overline{Y_{0}}, where Y_(0)={(t,varphi(t)inR^(2):0 <= t < 1}:}Y_{0}=\left\{\left(t, \varphi(t) \in \mathbb{R}^{2}: 0 \leq t<1\right\}\right. is connected as a continuous image of the interval [0,1)[0,1). The assertion that YY is not path connected follows from the fact that no path can join the point (0,0)(0,0) to (1,1)(1,1), because otherwise YY would be locally connected, which obviously is not the case.
We show that each continuous function f:Y longrightarrow Yf: Y \longrightarrow Y (hence each contraction) has a fixed point.Let us suppose that there is such a function ff without fixed points. Denoting with indices 1 and 2 the first, respectively the second component of a point in R^(2)\mathbb{R}^{2}, we define
Both AA and BB are open in YY, since ff is continuous. They are nonvoid, because (1,1)in A(1,1) \in A and (0,0)in B(0,0) \in B. More than that, X=A uu BX=A \cup B. Indeed, we have f(x)_(1)!=x_(1)f(x)_{1} \neq x_{1} (if we suppose f(x)_(1)=x_(1)f(x)_{1}=x_{1}, it will follow f(x)_(2)=x_(2)f(x)_{2}=x_{2}, since varphi\varphi is a function, and x=(x_(1),x_(2))x=\left(x_{1}, x_{2}\right) would be a fixed point for ff, contradiction with our assumption).
To summarize, AA and BB are open disjoint nonvoid sets such that Y=A uu BY=A \cup B, which contradicts the fact that YY is connected.
Example 2. A set Z subeR^(2)Z \subseteq \mathbb{R}^{2} which is non-closed but has the fixed point property for each contraction g:Z longrightarrow Zg: Z \longrightarrow Z. Obviously, the set ZZ is in this case path connected and a fortiori connected.
Let I_(n)subeR^(2),I_(0)=[0,1]xx{0}I_{n} \subseteq \mathbb{R}^{2}, I_{0}=[0,1] \times\{0\} and I_(k)={(1)/(k)}xx[0,1],k inN^(**)I_{k}=\left\{\frac{1}{k}\right\} \times[0,1], k \in \mathbb{N}^{*}. The set Z=uu_(n=0)^(oo)I_(n)subeR^(2)Z= \cup_{n=0}^{\infty} I_{n} \subseteq \mathbb{R}^{2} is path connected, hence connected. We have bar(Z)=Z uu({0}xx[0,1])\bar{Z}=Z \cup(\{0\} \times[0,1]),
so ZZ is non-closed. Each continuous function g:Z longrightarrow Zg: Z \longrightarrow Z (hence each contraction) has a fixed point.
Let us suppose that there is a continuous function g:Z longrightarrow Zg: Z \longrightarrow Z without fixed points. The continuous function widetilde(g):I_(0)longrightarrowI_(0)\widetilde{g}: I_{0} \longrightarrow I_{0} given by widetilde(g)(x)=(g(x)_(1),0)\widetilde{g}(x)=\left(g(x)_{1}, 0\right) has obviously a fixed point (p,0)inI_(0)(p, 0) \in I_{0}. Since gg has no fixed point, there exists k_(0)inN^(**)k_{0} \in \mathbb{N}^{*} such that p=(1)/(k_(o))p=\frac{1}{k_{o}}, hence g((1)/(k_(0)),0)=((1)/(k_(0)),y),0 < y <= 1g\left(\frac{1}{k_{0}}, 0\right)=\left(\frac{1}{k_{0}}, y\right), 0<y \leq 1.
Let us denote y_(1)=s u p{y in[0,1]:EE z in[0,1],y < z,g((1)/(k_(0)),y)=((1)/(k_(0)),z)}y_{1}=\sup \left\{y \in[0,1]: \exists z \in[0,1], y<z, g\left(\frac{1}{k_{0}}, y\right)=\left(\frac{1}{k_{0}}, z\right)\right\}. By the continuity of gg and the definition of y_(1)y_{1}, there exists z_(1)in[0,1],y_(1) <= z_(1)z_{1} \in[0,1], y_{1} \leq z_{1} such that g((1)/(k_(0)),y_(1))=((1)/(k_(0)),z_(1))g\left(\frac{1}{k_{0}}, y_{1}\right)=\left(\frac{1}{k_{0}}, z_{1}\right). If we suppose y_(1) < z_(1)y_{1}<z_{1}, using again the continuity of gg, one can find y_(2),z_(2)in[0,1],y_(1) < y_(2) < z_(2)y_{2}, z_{2} \in[0,1], y_{1}<y_{2}<z_{2} such that g((1)/(k_(0)),y_(2))=((1)/(k_(0)),z_(2))g\left(\frac{1}{k_{0}}, y_{2}\right)=\left(\frac{1}{k_{0}}, z_{2}\right), contradiction with the definition of y_(1)y_{1}. It follows that y_(1)=z_(1)y_{1}=z_{1} and ((1)/(k_(0)),y_(1))\left(\frac{1}{k_{0}}, y_{1}\right) is a fixed point for gg, contradiction with our assumption that gg has no fixed points.
3. Convex subsets with nonvoid interior in Banach spaces
In this section, in the setting of Banach spaces, we give a class of sets for which one can prove that if any of their contractions has a fixed point, they are necessarily closed.
We mention the following definitions. A set AA in a linear space is convex if from x,y in Ax, y \in A it follows that (1-lambda)x+lambda y in A(1-\lambda) x+\lambda y \in A for each lambda in(0,1)\lambda \in(0,1); the relative interior of the convex set AA is ri A={a in A:AA x in A\\{a},EE y in AA=\{a \in A: \forall x \in A \backslash\{a\}, \exists y \in A such that a=(1-lambda)x+lambda ya=(1-\lambda) x+\lambda y, for some lambda in(0,1)}\lambda \in(0,1)\}. In a normed space, from int A!=O/A \neq \varnothing, it follows obviously that ri A!=O/A \neq \emptyset, but the converse is not true.
We can prove now
Theorem 3.1. Let EE be a Banach space, A sube EA \subseteq E a convex set with int A!=O/\operatorname{int} A \neq \varnothing. If each contraction h:A rarr Ah: A \rightarrow A has a fixed point, then AA is closed.
Proof. We have to prove that b in bar(A)b \in \bar{A} implies b in Ab \in A. Let us suppose by contradiction that there is an element b in bar(A)\\Ab \in \bar{A} \backslash A. Making (if necessary) a translation, we can take b=0b=0. Since int A!=O/\operatorname{int} A \neq \varnothing, there is a in int Aa \in \operatorname{int} A. Let alpha,beta > 0\alpha, \beta>0 be given such alpha+beta < 1,r=alpha+beta||a|| < 1\alpha+\beta<1, r=\alpha+\beta\|a\|<1.
We define a function H:E longrightarrow EH: E \longrightarrow E,
H(x)=alpha x+beta(||x||)/(||x||+1)a.H(x)=\alpha x+\beta \frac{\|x\|}{\|x\|+1} a .
For x in Ax \in A, we have u=(||x||)/(||x||+1)a in int Au=\frac{\|x\|}{\|x\|+1} a \in \operatorname{int} A (since 0in bar(A),a in int A0 \in \bar{A}, a \in \operatorname{int} A ). Then the convex combination
v=(alpha)/(alpha+beta)x+(beta)/(alpha+beta)uv=\frac{\alpha}{\alpha+\beta} x+\frac{\beta}{\alpha+\beta} u
is contained in int A\operatorname{int} A, so alpha x+beta u=(alpha+beta)v+(1-alpha-beta)0in int A(alpha+beta < 1)\alpha x+\beta u=(\alpha+\beta) v+(1-\alpha-\beta) 0 \in \operatorname{int} A(\alpha+\beta<1). It follows that H(A)sube int A sube AH(A) \subseteq \operatorname{int} A \subseteq A, hence H( bar(A))sube bar(H(A))sube bar(A)H(\bar{A}) \subseteq \overline{H(A)} \subseteq \bar{A}.
We prove now that HH is a contraction on EE with the constant rr :
Applying Banach's theorem for H|_( bar(A))\left.H\right|_{\bar{A}}, it follows that it has a unique fixed point, which is equal to 0 (because H(0)=0H(0)=0 ).
By the hypothesis, the fixed point set of h=H|_(A)h=\left.H\right|_{A} is nonvoid; it is included in that of H|_( bar(A))\left.H\right|_{\bar{A}}, which contains exactly the point 0 , so it follows that 0in A0 \in A, contradiction. It remains that the set AA has to be closed.
The class of convex sets with nonvoid relative interior is larger than that considered in Theorem 3.1 The problem if the condition int A!=O/A \neq \emptyset could be replaced by ri A!=O/A \neq \emptyset remains open.
4. Remarks
In the case of normed spaces, there are some further comments to be done.
Remark 1. Suppose that the problem at the end of the previous section can be answered in the affirmative, i.e. Theorem 3.1 is true with ri A!=O/A \neq \emptyset. Then an immediate consequence would be that any normed space with the property that each contraction has a fixed point is in fact a Banach space. Indeed, a normed space AA can be considered as a convex set with ri A!=O/A \neq \emptyset in its completion widetilde(A)=E\widetilde{A}=E; applying the theorem, it follows that the normed space AA is closed, hence a Banach space.
Remark 2. In a normed space which is not Banach, contractions without fixed points may exist. But to provide such contractions is not an easy task. For example, let us consider the space l^(0)l^{0} of all real sequences with a finite number of nonzero terms, endowed with the sup norm, and the shift operator s:l^(0)longrightarrowl^(0),s(x_(1),x_(2),dots)=(0,x_(1),x_(2),dots)s: l^{0} \longrightarrow l^{0}, s\left(x_{1}, x_{2}, \ldots\right)=\left(0, x_{1}, x_{2}, \ldots\right). For lambda in(0,1)\lambda \in(0,1) and e_(1)=(1,0,0,dots)e_{1}=(1,0,0, \ldots), we define f:l^(0)longrightarrowl^(0),f(x)=lambda s(x)+e_(1)f: l^{0} \longrightarrow l^{0}, f(x)=\lambda s(x)+e_{1}, which is a lambda\lambda-contraction. But it has no fixed points, because f(u)=uf(u)=u implies u=(1,lambda,lambda^(2),lambda^(3),dots)!inl^(0)u=\left(1, \lambda, \lambda^{2}, \lambda^{3}, \ldots\right) \notin l^{0}.
References
[1] E. H. Connell, Properties of fixed point spaces, Proc. Amer. Math. Soc. 10(1959), 974-979
[2] T. K. Hu, On a fixed point theorem for metric spaces, Amer. Math. Monthly 74(1967), 436-437
[3] P. V. Subrahmanyam, Completeness and fixed points, Monatsch. für Math. 80(1975), 325-330
[4] I. A. Rus, Maximal fixed point structures, "Zilele academice clujene", 18-23 nov. 1996.