Let $(E, \mathcal{T})$ be a compact Hausdorff space. It is well known that every topology $\mathcal{U}$ coarser than $\mathcal{T}$ such that $(E, \mathcal{U})$ is Hausdorff is equal to $\mathcal{T}$.

Is the converse true?

(that is: if $\mathcal{T}$ is a coarsest topology amongst Hausdorff topology on $E$, then $(E, \mathcal{T})$ is compact)

Thanks in advance.

  • 5
    $\begingroup$ No. See for example in this paper from A. Smythe and C. A. Wilkins. $\endgroup$ – martini Mar 27 '12 at 19:41
  • 2
    $\begingroup$ you can find the paper by searching "A. Smythe and C. A. Wilkins" on google. That link didn't work for some reason. $\endgroup$ – Dustan Levenstein Mar 27 '12 at 19:46
  • 2
    $\begingroup$ Perhaps this google-Link will do. It does for me, at least. $\endgroup$ – martini Mar 27 '12 at 19:48
  • 1
    $\begingroup$ Your question is probably sufficiently answered by the first comment. But a key word you might look for is "minimal Hausdorff space." Also, note that some spaces don't have "minimal Hausdorff" coarsenings -- the rational numbers for example. $\endgroup$ – ihaphleas Nov 15 '13 at 14:34

If $A$ is the linearly ordered set $\{1,2,3,...,\omega,...,-3,-2,-1\}$ with the order topology, and if $\Bbb Z^+$ is the set of positive integers with the discrete topology, we define $X=A\times\Bbb Z^+$ together with two ideal points $a$ and $-a$. The topology $\tau $ on $X$ is determined by the product topology on $A\times \Bbb Z^+$ together with the local base

$$M_n^+(a)=\{a\} \cup\{(i, j) \mid i<\omega, j>n\} \text{ and } M_{n}^-(-a)=\{-a\} \cup\{(i, j) \mid i>\omega,j<n\} $$

Visualizing $X$: enter image description here

  1. A straightforward consideration of cases shows that $X$ is Hausdorff.

  2. The collection of all basis neighborhoods form an open covering of $X$ with no finite subcovering (Hint: Consider the points $(\omega,j)$ for each $j\in \Bbb Z^+$). Thus $X$ is not compact.

  3. $X$ is almost compact (i.e. each open cover of the space has a finite subcollection the closures of whose member cover the space). This follows from the fact that the closures of any neighborhoods of $a$ and $-a$ contain all but finitely many of the points $(\omega,j)$. A straightforward consideration of cases shows similarly that the complement of any basis neighborhood is also almost compact.

  4. Now suppose $\tau'\subset\tau$, suppose $N$ is a basis neighborhood of $\tau$ and suppose $\{O_\alpha\}$ is a $\tau'$-open covering of $X \setminus N$. It is then a $\tau$ open covering, so there exist finitely many sets $O_1, O_2,..., O_n$ the union of whose $\tau$-closures covers $X \setminus N$. But the $\text{cl}_{\tau} (O_i)\subset\text{cl}_{\tau'} (O_i)$, so $X \setminus N$ is covered by the union of the $\tau'$-closures of $0_1, O_2, ..., O_n$. In otherwords, $X\setminus N$ is an almost compact subset of $(X,\tau')$

  5. Suppose $\tau'$ is a proper subtopology of $\tau$. Then there would be some basis neighborhood $N\in \tau$ for which $X\setminus N$ is not closed in $\tau'$, and so there would be a point $x\in N$ such that $x\in \text{cl}_{\tau'} (X\setminus N)$. Let $\{C_\alpha\}$ be the collection of all $\tau'$-neighborhoods of $x$, and suppose $\{X\setminus \overline{C_\alpha}\}$ covers $X\setminus N$, then there exists $\overline{C_1},\overline{C_2},...,\overline{C_n}$ such that $X\setminus N\subset \bigcup \left(\overline{X\setminus\overline{C_i}}\right)$ (by 4). But $\bigcup\ \left(\overline{X \setminus \overline{C}_{i}}\right)$ is closed, hence it must contain $x$ (as $x\in \text{cl}_{\tau'} (X\setminus N)$). But $\bigcap C_i$ is a neighborhood of $x$, implying $\left(\bigcap C_{i}\right) \cap\left(\bigcup\ \left(\overline{X \setminus \overline{C}_{i}}\right)\right)=\left(\bigcap C_{i}\right) \cap\left(\overline{\bigcup\ \left(X \setminus \overline{C}_{i}\right)}\right)$ is nonempty, which is clearly impossible. Hence $\{X\setminus \overline{C_\alpha}\}$ doesn't cover $X\setminus N$, which implies there exists $y\notin N$ such that $y\in \bigcap \overline{C_\alpha}$. Since $y\neq x$ are inseparable by open sets, $(X,\tau')$ is not Hausdorff.

In conclusion, $X$ is minimal Hausdorff but not compact.

The above is essentially my paraphrasing of Steen and Seebach's Counterexample in Topology example 100.

  • $\begingroup$ +1... My edit was for a typo in 5. ("contain" for "contains") and to change $X-N$ to $X\setminus N $ at one place in 4. to conform to the other two instances of it in 4.... In 5. I think it would be clearer in 5. to say that $\cap C_i$ is a nbhd of $x$ so if $x\in \cup(\overline {X\setminus \bar C_i})=$ $\overline {\cup (X\setminus \bar C_i)}$ then $ \phi\ne (\cap C_i)\cap (\cup (X\setminus \bar C_i) $ which is (more) clearly impossible. $\endgroup$ – DanielWainfleet May 6 at 21:24
  • $\begingroup$ The space A is $T_2$ as it is linear and $\Bbb Z^+$ is $T_2$ as it is discrete, so $A\times \Bbb Z^+$ is $T_2$. And $A\times \Bbb Z^+$ is open in $X$. So to check that $X$ is $T_2$ we need only to check the effect of adding $a$ and $-a$.....You refer to "the" basis and I know what you mean, but $X$ has many bases. Some readers may be puzzled.... I think the phrasing in 2. is hard to understand...... Altogether, though, a very nice argument. $\endgroup$ – DanielWainfleet May 6 at 21:36
  • $\begingroup$ @DanielWainfleet Thanks for your kind words! I will edit it if I have time. $\endgroup$ – YuiTo Cheng May 6 at 23:47

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.