10

There was recently a question that I cannot find about separation in compact spaces. The answer to that question was no for trivial reasons. Motivated by that, let me ask a less trivial version of this question.

Given two points $x,y$ in a (separable) compact Hausdorff space $X$, can we find two open sets $V,U$ in $X$ and a continuous function $f\colon X\to \mathbb{R}$ such that

  • $x\in V, y\in U$
  • $f(t)=1$ for all $t\in V$
  • $f(t)<0$ for all $t\in U$
  • $f^{-1}(\{0\})$ has empty interior

My feeling is that in general the answer should be no. Yet for familiar compact metric compacta such open sets and function $f$ can be constructed. (Of course this is not a proof.)

EDIT: I modified the question by removing one of the requirements; I believe now it will be more tractable. It seems that the answer is yes if $X$ is perfectly normal. Indeed, take $f$ that satisfies the first three clauses and by perfect normality take a function $g$ that is positive on the interior of $f^{-1}(\{0\})$ and $0$ otherwise. Then consider simply $f+g$.

TMK
  • 103
  • @egerg, thanks for spotting that. I am actually interested in (and thinking of) connected spaces; I added an extra hypothesis. – TMK Apr 01 '15 at 20:30
  • 1
    is this one question about separable compact spaces, or two questions, one about separable compact spaces and the other about compact spaces that are not necessarily separable. – Mirko Apr 04 '15 at 23:17
  • Is there a compact connected Hausdorff space in which every non-empty $G_\delta$ set has non-empty interior? Without connectedness, yes the remainder of the Stone-Cech compactification of the integers (easier yes for finite spaces). This might be relevant as $f^{-1}(0)=\cap_n\ f^{-1}(\frac{-1}n,\frac1n)$. If the answer to this question is yes, then the answer to the OP question is no in general (but if the answer to this question is no, then it doesn't say much about the answer to the OP question). http://arxiv.org/pdf/math/9805008.pdf – Mirko Apr 05 '15 at 03:07

1 Answers1

3

I believe there are two questions asked here, not just one. Namely:

Question 1. Given two points $x,y$ in a separable compact Hausdorff space $X$, can we find two open sets $V,U$ in $X$ and a continuous function $f:X\to\mathbb R$ satisfying the conditions given by the OP.

Question 2. Given two points $x,y$ in a compact Hausdorff space $X$, can we find two open sets $V,U$ in $X$ and a continuous function $f:X\to\mathbb R$ satisfying the conditions given by the OP. (That is, in this version the space $X$ is not required to be separable.)

Accordingly, I give two answers.

Question 1. Answer yes (and it is enough to assume $X$ is ccc).

Question 2. Answer no, not necessarily. (For example the answer is no for $X=H^*$, the remainder of the Stone-Cech compactification of the half-line $[0,\infty)$, although it is yes for $X=[0,\omega_1]$, as explained in more detail below.)

Separable case. If the question is about separable compact Hausdorff spaces, then the answer is yes. It is enough to assume the weaker condition that $X$ is a compact Hausdorff space that is ccc, which means that every disjoint collection of non-empty open sets is at most countable. (Every separable space is in general a ccc space, as is easily verified but not vice-versa.)

Pick $x,y\in X$ with $x\not=y$, where $X$ is a separable compact Hausdorff space that is ccc. Then $X$ is normal and hence there is a continuous $h:X\to[-2,2]$ with $h(x)=2$, $h(y)=-2$. Using $h$ define $g:X\to[-1,1]$ by $g(z)=-1$ if $h(z)\le-1$, $g(z)=1$ if $h(z)\ge1$, and $g(z)=h(z)$ if $h(z)\in(-1,1)$. It is easily seen that $g$ is continuous with $g(x)=1$, $g(y)=-1$. Let $U=h^{-1}([-2,-1))$ and $V=h^{-1}((1,2])$. Then $V$ and $U$ are open with $x\in V$, $y\in U$, $g(V)=1$, $g(U)=-1$.

There must be $s\in(-1,1)$ such that $g^{-1}(s)$ has empty interior (for otherwise the collection $\{{\mathrm{Int}}\, g^{-1}(t):t\in(-1,1)\}$ would be an uncountable collection of pairwise disjoint non-empty open sets, contradicting that $X$ is ccc.) (We do not rule out the possibility that $g^{-1}(s)$ itself is empty, but that does not change the rest of our argument.)

Fix $s\in(-1,1)$ such that $g^{-1}(s)$ has empty interior. Let $\gamma:[-1,1]\to[-1,1]$ be a monotonically increasing (piecewise linear and) continuous function that fixes the endpoints (i.e. $\gamma(-1)=-1$ and $\gamma(1)=1$), and $\gamma(s)=0$. Let $f=(\gamma\circ g) : X\to [-1,1]$. Then $f(x)=1$, $f(y)=-1$, $f(V)=1$, $f(U)=-1$, and $f^{-1}(0)=g^{-1}(s)$ has empty interior.

Non-separable case. If the question is about compact spaces that are not necessarily separable, then the answer is no (as felt by the OP, and as suggested in one of my comments above). (As a clarification prompted by the comments to my answer, in this case the required $V,U$ and $f$ do not, in general, exist, although they may sometimes exist, even for non-separable spaces.)

Let $H^*$ be the Stone-Cech remainder of the half-line $H=[0,\infty)$. Then $H^*$ is compact and connected, and has the property that every non-empty $G_\delta$ set has non-empty interior. See this answer by Alessandro Vignati and Adam Przeździecki who cleared my confusion about it (see also the answer for the LOTS case by Eric Wofsey and answer by Joseph Van Name about almost $P$-spaces, and perhaps more comments or answers coming there).

We now show that the answer is no whenever $X$ is any compact, connected, Hausdorff space (with at least two points) in which every non-empty $G_\delta$ set has non-empty interior. Take any two different points $x,y\in X$ and any continuous function $f:X\to(-\infty,\infty)$, $f(x)=1$, $f(y)<0$. Since $f(X)$ is connected, it follows that $0\in f(X)$, that is $f^{-1}(0)$ is not empty. Also, $f^{-1}(0)=\bigcap_{n\ge1}\, f^{-1}((\frac{-1}n,\frac1n))$ is a $G_\delta$ set. Hence $f^{-1}(0)$ has non-empty interior. Since the points $x,y$ and the continuous function $f$ were arbitrary, this answers the question negatively. (To be more specific, when $X=H^*$ for example and $x,y\in X$ are any two different points, then $U,V$ and $f$ as asked by the OP, do not exist. We could not find $f$ such that $f^{-1}(0)$ has empty interior, even if we disregard the conditions about $U,V$.)

Edit. Prompted by the comments, is seems my answer needs clarification (although the above is correct). Namely, the answer is in general no, that is, if $x,y$ are two points in a compact Hausdorff space $X$ we cannot conclude the existence of $V,U$ and $f$ as asked by the OP. If $X$ is in addition separable (or more generally, if $X$ is ccc), then yes, we could find such $V,U$ and $f$.

It so happens that for some $X$ that are not separable, we could find such $V,U$ and $f$. For example let $X=[0,\omega_1]$ be the space of all countable ordinals together with the first uncountable ordinal, with the order topology. Then $X$ is compact and not separable. If $x,y\in X$ with $x<y$ then $V=[0,x]$ and $U=(x,\omega_1]=[x+1,\omega_1]$ are disjoint closed-and-open neighborhoods of $x$ and $y$ respectively, with $X=V\cup U$. If we define $f(v)=1$ for every $v\in V$, and $f(u)=-1$ for every $u\in U$, then $f$ is continuous with $f^{-1}(\{0\})$ being empty, in particular having empty interior. So all conditions asked by the OP are satisfied for this particular choice of $X=[0,\omega_1]$. But, unless the answer is yes for every choice of $X$ and $x,y$, then the answer is, in general, no. The answer is no for $H^*$ (as discussed above), hence the answer is no.

Mirko
  • 13,445
  • Did you mean to say that the answer is no for non-separable spaces rather than not necessarily separable? – Myridium Apr 05 '15 at 21:27
  • 2
    @Myridium There are non-separable spaces for which the answer is yes. For example let $N^=\beta N\setminus N$ be the Stone-Cech remainder of the integers. Then $N^$ is compact, non-separable, and extremely disconnected. The latter implies that if $x$ and $y$ are two different points, then $X$ could be partitioned into disjoint open $V,U$ with $x\in V$, $y\in U$, hence we could define continuous $f$ by $f(V)=1$, $f(U)=-1$ and clearly $f^{-1}(0)$ is empty (and has empty interior). An easier example is the space $[0,\omega_1]$. – Mirko Apr 05 '15 at 21:58
  • So what you mean is that for non-separable spaces, the answer is not always rather than no? This is not what your answer says; the wording there is quite confusing. – Myridium Apr 05 '15 at 21:59
  • 1
    @Myridium Not always, hence no. The answer could be yes or no, in this case the answer is no. If you prefer to rephrase it in a different way, you could say "not always", but the answer remains "no". Or if you prefer, "no in general". By default, "no" means "no in general". – Mirko Apr 05 '15 at 22:20