4
$\begingroup$

While trying to understand the motivation behind the definition of a filter I've stumbled upon the following notion, let's call it "almost first-countability":

Let $x$ be a point of a set $X$ and $\mathcal{F}$ be a filter on $X$ such that each $F\in\mathcal{F}$ contains $x$. We say $\mathcal{F}$ is almost first-countable if there is a countable collection $\mathcal{C}=\{C_1,C_2,\dots\}$ of nested subsets of $X$ (so $C_1\supset C_2\supset\cdots$) such that $\Sigma(\mathcal{C})=\Sigma(\mathcal{F})$, where for any $\mathcal{X}\in P(P(X))$ the symbol $\Sigma(\mathcal{X})$ denotes the set of sequences $\phi\colon\mathbb{N}\to X$ converging with respect to the collection $\mathcal{X}$, i.e. each element of $\mathcal{X}$ contains a tail of $\phi$.

Clearly if $\mathcal{F}$ is first-countable, i.e. it can be generated by a countable filter base, then it is almost first-countable (just take $\mathcal{C}$ to be $\{C_1,C_1\cap C_2,C_1\cap C_2\cap C_3,\dots\}$ where $\{C_1,C_2,C_3,\dots\}$ is a countable filter base generating $\mathcal{F}$).

Question: Is this notion strictly weaker than first-countability? In other words, are there a set $X$, a non-first-countable filter $\mathcal{F}$ at $x\in X$ and a countable collection $\mathcal{C}=\{C_1,C_2,\dots\}$ of nested subsets of $X$ such that $\Sigma(\mathcal{C})=\Sigma(\mathcal{F})$?

Observation: Call $\mathcal{C}'$ the filter generated by the base $\mathcal{C}$ by adding all supersets of elements of $\mathcal{C}$. Then $\Sigma(\mathcal{C}')=\Sigma(\mathcal{C})$. It follows that the above question boils down to finding two filters at $x$ sharing the same set of convergent sequences and such that one is first-countable while the other is not (or proving they do not exist, of course).


As an exercise for myself let me rewrite user126154's brilliant argument in a more detailed fashion. In order to fix the ideas, instead of letting $\alpha$ be any non-$\omega$-cofinal limit ordinal I decided to choose a concrete example by picking up the smallest limit ordinal which is not $\omega$-cofinal. If I'm not mistaken this should be the first uncountable ordinal $\omega_1$.

Consider the first uncountable ordinal $\omega_1$ and the filter $\mathcal{F}_\infty$ on $\omega_1$ generated by the filter base $\mathcal{B}_\infty:=\{(a,\infty)\mid a\in \omega_1\}$ where $(a,\infty):=\{x\in \omega_1\mid x>a\}$. Observe that $\Sigma(\mathcal{F}_\infty)$ is empty: indeed, assume there is a sequence $\phi\colon\mathbb{N}\to \omega_1$ converging with respect to $\mathcal{F}_\infty$; by induction $\phi$ has a strictly increasing subsequence $\phi'$ also converging w.r.t. $\mathcal{F}_\infty$; the limit of this sequence is just $s:=\sup_{n\in\mathbb{N}}\{\phi'(n)\mid n\in\mathbb{N}\}$, i.e. a countable union of countable ordinals which is still countable, hence $s\in \omega_1$ and the interval $(s+1,\infty)$ does not contain any tail of $\phi'$, which is absurd. Further observe that $\mathcal{F}_\infty$ cannot be generated by a countable filter base, otherwise we could construct a sequence converging w.r.t. the base and hence also w.r.t. $\mathcal{F}$.

Now consider the set $X:=(\sqcup_{n\in\mathbb{N}} (\omega_1)_n)\cup\{\infty\}$ consisting of the disjoint union of countably many copies of $\omega_1$ together with a new point $\infty$, and construct the filter $\mathcal{F}$ on $X$ consisting of all subsets $F\subset X$ satisfying all the three conditions below:

  1. $F\cap(\omega_1)_n\in\mathcal{F}_\infty$ for each $n\in\mathbb{N}$,
  2. there is $j\in\mathbb{N}$ such that $F\cap(\omega_1)_k=\omega_1$ for each $k\geq j$,
  3. $\infty\in F$.

Notice that $\mathcal{F}$ cannot be generated by a countable filter base, otherwise by restricting each element of the base to one copy of $\omega_1$ we would obtain a countable filter base for $\mathcal{F}_\infty$. Now consider the collection $\mathcal{C}=\{C_n\mid n\in\mathbb{N}\}$ of subsets of $X$ where $C_n$ is defined by the next three conditions:

  1. $C_n\cap(\omega_1)_j=\emptyset$ for $j<n$,
  2. $C_n\cap(\omega_1)_j=\omega_1$ for each $j\geq n$,
  3. $\infty\in C_n$ for each $n\in\mathbb{N}$.

Clearly $\mathcal{C}$ satisfies the axioms of a filter base, and the filter that it generates is obviously first-countable.

Now we show that $\Sigma(\mathcal{F})=\Sigma(\mathcal{C})$. Let $x_n$ be a sequence in $X$ converging w.r.t. $\mathcal{C}$. Each $F\in\mathcal{F}$ contains some $C_n$ so $x_n$ converges w.r.t. $\mathcal{F}$ as well. Conversely, assume $x_n$ converges w.r.t. $\mathcal{F}$ and define $\sigma(n)$ to be the index of the copy of $\omega_1$ where $x_n$ lies (put $\sigma(n)=\star$ if $x_n=\infty$). If for each $n$ there is $j_n$ such that for each $k\geq j_n$ $\sigma(k)$ is either greater than $n$ or equal to $\star$ then $x_n$ obviously converges w.r.t. $\mathcal{C}$ and we are done. Otherwise $\sigma$ has a constant subsequence at some natural number, that is there is a subsequence $x_{n_k}$ staying in a single copy of $\omega_1$. Since $x_n$ converges w.r.t. $\mathcal{F}$ by assumption, then $x_{n_k}$ must converge w.r.t. $\mathcal{F}_\infty$ and this is impossible since $\Sigma(\mathcal{F}_\infty)=\emptyset$.

Conclusion: we have found a set $X$, two filters at some point $x\in X$ one of which is first-countable while the other is not (the former being the filter generated by $\mathcal{C}$, the latter being $\mathcal{F}$) sharing the same set of converging sequences in $X$. In other words, $\mathcal{F}$ is "almost first-countable" but not first-countable, therefore the notion spelled out above of "almost first-countability" is really different than first-countability (in particular, it is strictly weaker).


In order to avoid ambiguities I add the definitions used. A filter on $X$ is a nonvoid collection of subsets of $X$ such that:

  1. the empty set does not belong to the collection,
  2. any finite intersection of its elements is a superset of some element of the collection,
  3. any superset of any of its elements belongs to the collection.

A filter base is like a filter but we drop the third assumption. By adding all supersets of all elements of a filter base $\mathcal{F}$ we obtain the filter $\mathcal{F}'$ generated by the base $\mathcal{F}$. It follows that a nonvoid collection $\mathcal{B}\in P(P(X))$ is a base of a given filter $\mathcal{F}$ if and only if $\mathcal{B}\subset\mathcal{F}$ and each $F\in\mathcal{F}$ contains some $B\in\mathcal{B}$.

$\endgroup$
  • $\begingroup$ what is your definition of first-countable? are we talking about open sets? if so, the $C_i$ are open? $\endgroup$ – user126154 Mar 10 '14 at 16:28
  • $\begingroup$ @user126154 in this context I say $\mathcal{F}$ is first-countable if it admits a countable filter base; no mention of openness, there is no topology on $X$ yet, just a filter at some point $x\in X$ $\endgroup$ – johndoe Mar 10 '14 at 17:17
  • $\begingroup$ in this case I think my answer applies, with the change that $\mathcal C$ is claimed to be a filter base at $x$. $\endgroup$ – user126154 Mar 10 '14 at 17:21
  • $\begingroup$ @user126154 could you please amend your proof accordingly? I can't follow your argument. $\endgroup$ – johndoe Mar 10 '14 at 17:24
  • $\begingroup$ could you give the precise definition that you are using? I'm a bit confused. $\endgroup$ – user126154 Mar 10 '14 at 17:41
2
$\begingroup$

Let $X$ be almost-first-countable. Let $\mathcal C=\{C_i\}$ be a nested countable sequence of sets containing $x$ so that $\Sigma(\mathcal C)=\Sigma(\mathcal F)$.

Suppose there is $F\in\mathcal F$ so that $x\in F$ and no $C_i$ is contained in $F$.

In particular there is $x_i\in C_i$ such that $x_i\notin F$. Therefore the sequence $\{x_i\}$ converges to $x$ for $\mathcal C$ but does not converge to $x$ for $\mathcal F$. This contradicts the hypothesis of almost first countability.

We have so proved that for any $F\in\mathcal F$ there is $C_i\in \mathcal C$ such that $C_i\subset F$.

If our filter comes from a topology and the $C_i$'s are open sets, then the converse is true.

Let's now see an example to show that the converse is not necessarily true in general.

Let $\omega$ be the typer of order of $\mathbb N$.

For any ordinal $\alpha$ let $\mathcal F_\alpha$ be the filter given by the order relation. (i.e. the filter of neighborhoods of $\infty$.)

Choose a limit ordinal $\alpha$ which is not $\omega$-cofinal. That is to say, there is no sequence $x_i\stackrel{\mathcal F_\alpha}{\to}\infty$ (i.e. $\mathcal F_\alpha$ is not first-countable.)

Let $X$ be the set obtained by countable disjoint union of copies of $\alpha$, plus the point $\infty$.

$$X=(\bigsqcup_{i\in\mathbb N}\alpha_i)\cup\{\infty\}$$

where $\alpha_i$ is a copy of $\alpha$.

Therefore, any subset $A\subset X$ is of the form $\bigsqcup_{i=0}^\infty A_i$ with $A_i\subset \alpha_i$ for $i\in\mathbb N$ and $A_\infty\subset\{\infty\}$.

Let $\mathcal F$ be the filter defined as follows: A set $F\subset X$ is in $\mathcal F$ if and only if

$\exists n_0=n_0(F)$ s.t.

1) $F_i=\alpha_i$ for any $n>n_0$

2) $\forall n\leq n_0$ we have $\emptyset\neq F_i\in\mathcal F_{\alpha_i}$

moreover,

3) $\infty\in F$

CLAIM: $\mathcal F$ is not first-countable.

PROOF: by conditions 1) and 2), any countable filter base for $F$ would give, by restriction to $\alpha_i$, a countable filter base on $\alpha$, which is assumed not $\omega$-cofinal. $\square$

Now, let's define $\mathcal C$. This is easy, let $$C_i=(\bigsqcup_{n\geq i} \alpha_i)\cup\{\infty\}$$

Clearly $\mathcal C$ is a countable filter.

CLAIM: for any sequence $(x_i)_{i\in\mathbb N}$ in $X$ we have $$x_i\stackrel{\mathcal C}{\to}\infty \Leftrightarrow x_i\stackrel{\mathcal F}{\to}\infty$$

PROOF. Let $x_i\stackrel{\mathcal C}{\to}\infty$. By condition $1)$ for all $F\in\mathcal F$ we have $x_i\in F$ for $i>n_0(F)$. Thus $x_i\stackrel{\mathcal F}{\to}\infty$.

On the other hand, suppose $x_i\stackrel{\mathcal F}{\to}\infty$

The point $x_i$ belongs to a set $\alpha_k$. Define $\sigma(i)$ so that $$x_i\in\alpha_{\sigma(i)}$$

If $\sigma(i)$ does not converge to infinity, then we can find a subsequence $i_k$ so that $\sigma(i_k)$ is a constant natural number, say $n$.

Therefore the sequence $x_{i_k}$ is a sequence in $\alpha_n$ which converges to infinity, which is impossible because $\alpha$ is not $\omega$-cofinal.

Therefore $\sigma(i)\to \infty$ and thus $x_i\stackrel{\mathcal C}{\to}\infty$. $\square$

$\endgroup$
  • $\begingroup$ Your argument shows that $\mathcal{C}$ is finer than $\mathcal{F}$; it still does not prove that $\mathcal{F}$ is generated by a countable filter base (clearly the condition $C_i\in\mathcal{F}$ would suffice but it is not true in general). Thanks for your input by the way, I appreciate it. $\endgroup$ – johndoe Mar 11 '14 at 1:18
  • $\begingroup$ ok, I'll try to complete with an example. $\endgroup$ – user126154 Mar 11 '14 at 11:35
  • $\begingroup$ There are several things I fail to understand, please bear with me. First, I don't understand the condition of non-$\omega$-cofinality on $\alpha$; e.g., does $\alpha=\omega+1$ satisfy it? Second, what is $\mathcal{F}_\alpha$? $\endgroup$ – johndoe Mar 11 '14 at 20:31
  • $\begingroup$ $\mathcal F_\alpha$ is the filter of sets of the type $\{x>a\}$ with $a\in\alpha$, is the filter fo "neigborhood of $\infty$" $\endgroup$ – user126154 Mar 11 '14 at 20:44
  • $\begingroup$ Given two ordinals $\omega$ and $\alpha$ we say that $\alpha$ is $\omega$-cofinal if there is an order preserving immersion of $\omega$ in $\alpha$ such that for all $a\in\alpha$ there is $b$ in the image of $\omega$ with $a\leq b$. So, you are right, $\omega+1$ is not $\omega$-cofinal. But it suffices to chose $\alpha$ a limit ordinal which is not cofinal in $\omega$. I edit the answer accordingly. $\endgroup$ – user126154 Mar 11 '14 at 20:50

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.