Given topological spaces $X$ and $C$ we call $C$ a coordinate space for $X$ to mean that every open set $U \subset X$ is of the form $f^{-1}(V)$ for some open $V \subset C$ and continuous $f \colon X \to C$.

More generally if $\mathscr X$ is a class of spaces we call $C$ a coordinate space for $\mathscr X$ to mean it is a coordinate space for every member of $\mathscr X$.

Two familiar examples of coordinate spaces:

  1. The Seirpinski space $\{0,1\}$ with exactly one closed point and one open point is a coordinate space for the class of all topological spaces.

  2. The closed unit interval is a coordinate space for the class of all completely-regular spaces. Moreover the closed unit interval is itself completely-regular.

But what about the class of all Hausdorff spaces? Obviously the Seirpinski space is still a coordinate space here. But what if we demand the coordinate space itself be Hausdorff -- as with example 2?

I would reckon such a space does not exist but have no idea how to prove that?

Does this notion already have a name and has the question been addressed before?

  • 2
    $\begingroup$ For starters, I wonder if there is a Hausdorff coordinate space for the class of all ordinals. $\endgroup$ – Nate Eldredge Jan 23 '17 at 16:51
  • $\begingroup$ There was a related classical result in the past by Stanisław Mrówka, celebrated in the P.S.Aleksandrov's surveys on general topology. $\endgroup$ – Włodzimierz Holsztyński Jan 23 '17 at 20:25
  • $\begingroup$ A pedantic remark: it is Wacław Sierpiński, not Seirpinski ("ie", not "ei"). You may apply Polish diacritics by "cut & paste" from any non-sloppy article, say, from Wikipedia. BTW, the Polish combination "ie" is pronounced on its own (as a vowel) just like "e" alone but it affects the preceding "S" turning it into a soft " S' ". $\endgroup$ – Włodzimierz Holsztyński Jan 23 '17 at 20:50

First of all, let me point out that $[0,1]$ is not a coordinate space for the class of completely regular spaces.

The definition of completely regular spaces says that for any closed $C \subset X$ and any point $p \notin C$, there is a continuous $[0,1]$-valued function (say $f$) mapping $p$ to $1$ and mapping every point in $C$ to $0$. It does not say that $f^{-1}(0,1] = X-C$. In other words, $f$ does map $C$ to $0$, but it may also map some other points to $0$ as well.

For example, it can be shown that any continuous function $f: \omega_1 \rightarrow [0,1]$ is eventually constant. It follows that $[0,1]$ is not a coordinate space for $\omega_1+1$: the open set $\omega_1$ is never equal to $f^{-1}(U)$ for an open $U \subseteq [0,1]$ and continuous $f: \omega_1+1 \rightarrow [0,1]$, because $f$ maps a tail of $\omega_1$ to a single point $c$, and must (by continuity) map a tail of $\omega_1+1$ to $c$. So if $f^{-1}(U)$ contains $\omega_1$, it contains $\omega_1+1$.

A similar argument shows

There is no Hausdorff coordinate space for the class of Hausdorff spaces.

To prove this, suppose $C$ is such a space. Let $\kappa$ be a regular cardinal bigger than $|C|$. Any function $\kappa \rightarrow C$ must map an unbounded subset of $\kappa$ to a single point (there's no topology involved in this statement -- it's just because $\kappa$ is regular and bigger than $|C|$). Now consider the ordinal space $\kappa+1$. As above, the open set $\kappa$ is never equal to $f^{-1}(U)$ for an open $U \subseteq C$ and continuous $f: \kappa+1 \rightarrow C$.


A classical notion was considered and solved in an elegant paper by Stanisław Mrówka: (1) instead of arbitrary open subspaces of $\ X\ $ one simply considers just X itself; and (2) instead of a single space $\ C,\ $ one considered the class of all powers $\ C^S.\ $ Then Stanisław Mrówka answered that this kind of embeddability problem is characterized by the property of separating points from closed subsets of $\ X\ $ by functions into $\ C^F,\ $ where this time the exponents $\ f\ $ are just finite sets--a very elegant conceptual theorem.

Acknowledgment--thank you, Will Brian, for pointing to my earlier error.

  • 1
    $\begingroup$ It seems to me that you are overstating your case in your first paragraph. For example, the class of discrete spaces contains arbitrarily large spaces, but it has the discrete two-point space as a coordinate space (in the sense of the OP). A little less trivially, the class of perfectly normal spaces has $[0,1]$ for a coordinate space. (Recall: "perfectly normal" means normal and all closed sets are $G_\delta$, and this allows you to construct, for any closed $C$, a $[0,1]$-valued function $f$ with $f(x) = 0$ if and only if $x \in C$.) $\endgroup$ – Will Brian Jan 23 '17 at 21:13
  • $\begingroup$ W.B., I did a quick fix; I may write more perhaps. $\endgroup$ – Włodzimierz Holsztyński Jan 23 '17 at 21:25
  • 1
    $\begingroup$ Yes, this looks correct now -- and thanks for pointing out this result, which I was not aware of, but I agree that it is elegant. $\endgroup$ – Will Brian Jan 23 '17 at 21:26
  • $\begingroup$ Will Brian, I have misread OP statement. I somehow saw "homeomorphism" which was absent from the original author's formulation. I'll edit my answer. Thank you. $\endgroup$ – Włodzimierz Holsztyński Jan 23 '17 at 22:43

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.