2
$\begingroup$

Let $(X,\tau)$ be a topological space. Let $\text{Cont}(X,X)$ denote the set of continuous functions $f:X\to X$.

What can be said about spaces $(X,\tau)$ where $|\text{Cont}(X,X)| = |X|$? For instance, is it impossible that they are zero-dimensional? (Note that trivially spaces with $|\text{Cont}(X,X)| = |X|$ must be infinite, because all constant functions as well as the identity are members of $\text{Cont}(X,X)$.)

$\endgroup$
2
  • 5
    $\begingroup$ X must be infinite or a point. $\endgroup$ Feb 12, 2015 at 19:34
  • 2
    $\begingroup$ Dominic, you could include as a second part of the question what you already have implicitly, about the spaces which have no other self-maps but identity and constants (you could ask about different classes of spaces in this, theorems and counter-examples, in this context). $\endgroup$ Feb 13, 2015 at 0:06

4 Answers 4

11
$\begingroup$

If a Hausdorff space $\ X\ $ admits a dense subset $ A\subseteq X\ $ such that

$$|X|^{|A|}\ =\ |X|$$

then indeed $\ \left|Cont(X\ X)\right|\ =\ |X|$.

This holds in particular for the separable metric spaces of cardinality continuum, as was already noted by Tomek Kania in the first Answer.

$\endgroup$
2
  • 1
    $\begingroup$ What's nice about this is that it lets us produce examples of such spaces of arbitrarily large cardinality. $\endgroup$ Feb 12, 2015 at 23:54
  • $\begingroup$ For Hausdorff spaces, is the converse true? Suppose $|\text{End}_{\text{Top}}(X)| = |X|$, is it true that there is a dense subspace such that...? $\endgroup$ Nov 23, 2017 at 10:12
5
$\begingroup$

If $X$ denotes the Cantor set, then $\mathfrak{c} = |X|=|{\rm Cont}(X,X)|$, so yes, such spaces can be zero-dimensional. You may replace $X$ with any separable metric space of cardinality continuum. What is your motivation?

$\endgroup$
5
$\begingroup$

Suppose that $\lambda$ is an infinite cardinal with $\lambda^{\aleph_{0}}=\lambda$. Then Saharon Shelah has shown in his paper Existence of endo-rigid Boolean algebras that there is some Boolean algebra $B$ with $|B|=\lambda$ but where the only endomorphisms $f:B\rightarrow B$ are the trivial endomorphisms. When we transfer by Stone duality these facts over to topological spaces, for each infinite cardinal $\lambda$ with $\lambda^{\aleph_{0}}=\lambda$, there is a compact totally disconnected space $X$ with $w(X)=\lambda$ (here $w$ denotes the weight of the space) but where the only continuous maps $f:X\rightarrow X$ are the mappings where there is some partition $A_{1},...,A_{n}$ of $X$ into clopen sets where $f|_{A_{i}}$ is a constant function or where $f|_{A_{i}}$ is the identity function for each $i\in\{1,...,n\}$ (i.e. the only mappings are the trivial mappings). Take note that if $X$ is a compact space, then $w(X)\leq|X|$ and if $X$ is compact and totally disconnected, then $|\mathfrak{B}(X)|\leq|X|$ where $\mathfrak{B}(X)$ is the Boolean algebra of clopen subsets of $X$. Therefore if $X$ is a compact zero-dimensional space where the only mappings $f:X\rightarrow X$ are the trivial mappings, then $|C(X,X)|=|X|$ since there are at most $|X|$ ways to partition $X$ into finitely many clopen sets.

$\endgroup$
3
$\begingroup$

E-D-I-T 1   @RasmusBentmann has pointed out to my error (in the $\,T_0\,$ case (I considered $\ f(X)\ $ instead of $X).\ $ Thus virtually nothing of my attempt is left. Perhaps Rasmus Bentmann can post his proof if it's ready. I'll have to think a bit more about the situation. Perhaps I'll remove my post unless I can provide a correct result, either one pretty soon.

*

E-D-I-T 2   My "proof" was false. Eric Vofsey has provided a correct proof though in a comment below--I hope that Eric will his argument into an Answer (then I'll remove my answer immediately upon seen it, and with a relief).

*

THEOREM   $\ \left|Cont(X\ X)\right|\ \ge\ |X|+2\,\ $ for every finite topological space $\ X\ $ such that $\ |X|\ge 3$.

PROOF   We need to show that there is a continuous function $\ f:X\rightarrow X\ $ such that it is not the identity, nor constant.

First, let $\ X\ $ be not a $T_0$-space, i.e. there are two different points $\ a\ b\in X\ $ such $\ \left|G\cap\{a\ b\}\right|\ne 1\ $ for arbitrary open $\ G\subseteq X.\ $ Define $\ f:X\rightarrow X\ $ by conditions:

  • $\ f\,|\,X\!\setminus\!\{a\}\ $ is the identity on $\ X\!\setminus\!\{a\}$
  • $\ f(a)\, :=\, b$

Then obviously $\ f^{-1}(G)\,=\,G\ $ for every open $\ G\subseteq X.\ $ Thus $\ f\ $ is continuous, it's not an identity on $\ X,\ $ and it's not a constant map (since $\ |X|\ge 3$. Thus the theorem holds in this case.

Now let $\ X\ $ be a $T_0$-space. Then see the Eric's argument in the comment below (until and if Eric expands his comment into an Answer).

$\endgroup$
7
  • 1
    $\begingroup$ In the second case, unless $f(a)=a$, we have $f^{-1}(A)=\emptyset$, contradicting what you wrote about $f^{-1}(G)$. In fact, the function $f$ will not always be continuous: if $X=3\to 2\to 1$, $A=\{3\}$ and $f(3)=1$ then $f$ will not be monotonous, hence not continuous ($f^{-1}(\{3,2\}=\{2\})$ is not open). I don't know how to repair the proof but haven't found a counterexample either. $\endgroup$
    – Rasmus
    Feb 13, 2015 at 9:29
  • $\begingroup$ If there was a counterexample, it had to have the fixed-point property, so by a result of Ivan Rival from 1976 it could not contain a "crown." $\endgroup$
    – Rasmus
    Feb 13, 2015 at 9:57
  • $\begingroup$ That is to say, a counterexample $X$ had to be connected and every "convex" subset of $X$ had to be simply connected. $\endgroup$
    – Rasmus
    Feb 13, 2015 at 10:38
  • $\begingroup$ Is think one can do this: choose a chain $C$ of maximal length in $X$. Define $f\colon X\to X$ such that $f(x)$ is the open point in $C\cap\overline{\{x\}}$. If this intersection is empty, set $f(x)$ to be the closed point in $C$. Then $f$ is continuous and is the identity on $C$. Unless, $X$ is totally ordered or discrete, $f$ is not the identity or constant. $\endgroup$
    – Rasmus
    Feb 13, 2015 at 11:08
  • 2
    $\begingroup$ If $X$ is finite and not discrete, it contains a 2-point subspace $A$ that is not discrete. Unless $X=A$ is the Sierpinski 2-point space, there will be a nonconstant map $X\to A$ that is not the identity. $\endgroup$ Feb 13, 2015 at 18:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.