For topological spaces $X,Y$ let $\text{Cont}(X,Y)$ be the collection of continuous functions $f:X\to Y.$ We endow $\text{Cont}(X,Y)$ with the topology inherited from the product topology on $Y^X.$

Are there spaces $X,Y$ such that $X$ has more than one point and $Y\not\cong\mathbb{R}$ such that $\mathbb{R}\cong\text{Cont}(X,Y)$?


I must admit that I hesitated answering this question, but here it is.

The answer is "no". Assume there exist topological spaces $X$ and $Y$ such that $C(X,Y)\simeq \mathbb{R}$ and $Y\not\simeq\mathbb{R}$. Identifying $C(X,Y)$ with $\mathbb{R}$ and $Y$ with the constant functions in $Y^X$ we consider $Y$ as a closed subsapce of $\mathbb{R}$. Considering the surjection $\mathbb{R} \to C(X,Y)\to C(\{x\},Y) \to Y$, we see that $Y$ is connected. Thus $Y\subset\mathbb{R}$ is a closed convex subset. As $Y\not\simeq \mathbb{R}$ there must exist an extreme point $y\in Y$. Note that $C(X,Y)$ is a convex subset of $C(X,\mathbb{R})$ and the constant function $y$ is an extreme point of it. It follows that $C(X,Y)-\{y\}$ is also convex, hence contractible. But $C(X,Y)-\{y\}$ is homeomorphic to $\mathbb{R}$ minus a point. This is a contradiction.

  • 3
    $\begingroup$ Sorry to be dense, but I must admit that I'm not following the final sentence. First, I don't know how to parse postcomposing an object $C(X, Y)$ with a contraction map. Second, I'm not seeing where $\mathbb{R} - \{1\}$ is coming from. Third, we didn't use anything about $X$ having more than one point that I can see (maybe we don't actually need that condition?). $\endgroup$ – Todd Trimble Jan 8 '18 at 14:36
  • 1
    $\begingroup$ In fact, that X has more than one point follows from the assumption Y is not homeomorphic to $\mathbb{R}=C(X,Y)$ $\endgroup$ – Pietro Majer Jan 8 '18 at 15:06
  • 2
    $\begingroup$ @Todd, Third thing first, we didn't use anything about $X$ because we didn't need to. Second, $\mathbb{R}-\{1\}$ stands for $C(X,Y)$ minus the constant function 1, and for your first complain: that was my sloppy way to explain the obvious fact that $C(X,Y)-\{y\}$ is contactible where $Y$ is a convex set, $y\in Y$ is an extreme point and $y\in C(X,Y)$ represents the constant function. That seems obvious enough to me, so I allowed myself some informal presentation... $\endgroup$ – Uri Bader Jan 8 '18 at 16:44
  • 2
    $\begingroup$ Indeed, the proof of Uri Bader was correct (exploiting a very nice observation with the extreme point of $C(X,Y)$). Initially I did not catch the idea and because of that wrote my proof (actually a continuation of the proof of Uri Bader). $\endgroup$ – Taras Banakh Jan 8 '18 at 17:52
  • 1
    $\begingroup$ Uri, thanks for the edit. I wasn't trying to nitpick -- I honestly did find the original hard to follow. Nice proof. $\endgroup$ – Todd Trimble Jan 8 '18 at 17:55

Following the answer of Uri Bader, we can show that $Y$ is a retract of the real line, so can be identified with a closed convex subset of $\mathbb R$. Without loss of generality we can assume that $0,1\in Y$ and hence $[0,1]\subset Y$. It follows that the function $C(X,Y)$ is a convex subset of $Y^X\subset \mathbb R^X$ and $C(X,[0,1])\subset C(X,Y)$. The assumption $Y\not\cong\mathbb R\cong C(X,Y)$ implies that $C(X,Y)$ contains a non-constant function $f$. Consider the constant functions $\mathbf 0:X\to\{0\}\subset Y$ and $\mathbf 1:X\to\{1\}\subset Y$ and observe that the set $T=\{\mathbf 0,\mathbf 1,f\}\subset C(X,Y)\subset Y^X\subset\mathbb R^X$ is affinely independent and its convex hull $conv(T)\subset C(X,Y)$ is homeomorphic to the 2-dimensional symplex, which cannot be contained in the real line $\mathbb R\cong C(X,Y)$. This contradiction completes the proof.

The negative answer can also be deduced from the following theorem. We recall that a topological space $X$ is functionally Hausdorff if for any distinct points $x,y\in X$ there exists a continuous function $f:X\to\mathbb R$ such that $f(x)\ne f(y)$.

Theorem. If for non-empty topological spaces $X,Y$ the function space $C(X,Y)$ is functionally Hausdorff and path-connected, then either $C(X,Y)$ is homeomorphic to $Y^n$ for some $n\in\mathbb N$ or $C(X,Y)$ contains a topological copy of the Hilbert cube.

Proof. The space $Y\cong C(\{x\},Y)$ is functionally Hausdorff and path-connected, being a retract of the functionally Hausdorff path-connected space $C(X,Y)$. If $Y$ is a singleton, then $C(X,Y)\cong Y^1$ is a singleton, too. So, we assume that $Y$ contains more than one point. In this case $Y$ contains a subspace $I$, homeomorphic to the closed interval $[0,1]$.

Consider the canonical map $\delta:X\to Y^{C(X,Y)}$, $\delta:x\mapsto (f(x))_{f\in C(X,Y)}$. If the image $\delta(X)$ is finite of cardinality $n$, then $C(X,Y)$ is homeomorphic to $Y^n$ (since each function $f\in C(X,Y)$ is constant on each set $\delta^{-1}(y)$, $y\in \delta(X)$).

So, we assume that the set $\delta(X)$ is infinite. Taking into account that the space $Y^{C(X,Y)}$ is functionally Hausdorff, we can construct a continuous map $g:Y^{C(x,Y)}\to I$ such that the image $Z=g(\delta(X))$ is infinite. The surjective continuous map $p:=g\circ\delta:X\to Z$ induces a continuous injective map $p^*:C(Z,I)\to C(X,I)$, $p^*:f\mapsto f\circ p$. It is easy to see that the function space $C(Z,I)\subset I^Z$ contains a topological copy $Q$ of the Hilbert cube $I^\omega$. Then $p^*(Q)$ is a topological copy of the Hilbert cube in $C(X,I)\subset C(X,Y)$.

  • $\begingroup$ This proof I follow; +1. $\endgroup$ – Todd Trimble Jan 8 '18 at 15:04
  • 1
    $\begingroup$ Thanks Taras for this nice presentation! I accepted Uri's answer because it was the original one and I understood it after the edit $\endgroup$ – Dominic van der Zypen Jan 8 '18 at 19:56
  • 1
    $\begingroup$ @DominicvanderZypen No problems. You made a correct decision. Uri Bader was the first and had a brilliant idea. $\endgroup$ – Taras Banakh Jan 8 '18 at 20:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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