Locally Confused

Does every finite topology originate from an order?

Warning! This is a draft. The information listed here may be incomplete, stupid, and on average will have been composed on 3:00AM.

Basics

\(\newcommand{\T}{\mathscr{T}}\newcommand{\N}{\mathbb{N}}\) Recall the following

Definition 1
Let \((P,\leq)\) be a poset. The order topology is the topology generated by the rays \((a,\infty), (-\infty,b)\) for all \(a,b\in P\).
Lemma 2

If \((P,\leq)\) originates from a lattice,

\begin{align*} (a,b) = \{x\in P: a\lt x \lt b\mid a\in P\cup\{-\infty\}, b\in P\cup\{\infty\} \} \end{align*}

forms a base of the associated order topology.

Proof
The key observation here is that in lattices, \((a,\infty)\cap (a^\prime,\infty) = (a\vee a^\prime,\infty)\). Because \((a,b) = (-\infty,b)\cap (a,\infty)\), it is easily seen that a) intervals are stable under intersection and b) the collection of rays and intervals are equivalent as subbases (i.e, both generate the same topology).

In an attempt to better understand finite topologies, my question is: Does every finite topology originate from this construction? If not, what are the necessary restrictions that must be fulfilled?

Focusing on finite sets, the following notions come in handy.

Definition 3
Let \(\T\) be a topology. An atom is a minimal nonempty set with respect to \((\T, \subseteq)\).

Atoms don't have to exist, which becomes apparent when considering an infinite descending chain like \(\{(-1/n,1/n)\mid n\in \N \}\) in the usual topology on \(\mathbb R\): The minimum of this chain would be the one-point set \(\{0\}\). It is not hard to see why this cannot happen it in the finite case:

Definition 4
A finite, nonempty topology \(\T\) has atoms. Furthermore, every open set contains an atom.
Proof
Note that every descending chain of nonempty open sets is finite and bounded by its intersection, which is again an open set. By Zorn's lemma (whose finite version can be proved without AC), we have a minimal element. Let now \(U\) be an open subset. Equipped with the subspace topology \(\T\vert_U\), which remains finite, it is guaranteed to contain an atom \(A\in \T\vert_U\), i.e. \(A = U\cap V\) for some \(V\) open in \(\T\). The minimality is inherited as follows: Every \(B\subseteq A\) that is open in the subspace topology already lies in \(\T\vert_U\), where \(A\) is minimal, ergo \(A=B\). Thus \(A\subseteq U\) is an atom.
Definition 5
Let \((P, \leq)\) be a poset, \(a, b\in P\). If \(a\) is maximal in \((-\infty, b)\), it is called “lower neighbor of \(b\)”. Similarly, if \(b\) is maximal in \((a, \infty)\), it is called an “upper neighbor of \(a\)”.
Lemma 6

Let \((P, \leq)\) poset, \(a < b \in P\). The following are equivalent:

  1. \(a\) is a lower neighbor of \(b\)
  2. \(b\) is an upper neighbor of \(a\)
  3. \((a,b)=\emptyset\).
Proof

Let \(a\) be a lower neighbor of \(b\), i.e. maximal in \((-\infty, b)\). By definition, every \(x\in (-\infty, b)\) satisfies \(a\leq x \Leftrightarrow a=x\), or equivalently,

which in turn holds precisely if \((a,b)=\emptyset\). The argument for \((2)\Leftrightarrow (3)\) follows analogously.

Some Examples

Example 7

Consider \(P:= \{1,2,3,6\}\) ordered by divisibility. It is a lattice, so we have a base consisting of the intervals \((1,6) = \{2,3\}\), \((-\infty,2)=(-\infty,3)= \{1\}\), \((-\infty,6)=\{1,2,3\}\), \((1,\infty)=\{2,3,6\}\), \((2,\infty)=(3,\infty) = \{6\}\), ergo

\begin{align*}\T_P = \left\{ \emptyset, \{1\}, \{6\}, \{2,3\}, \{1,2,3\}, \{2,3,6\}, P \right\}\end{align*}

The following observations seem not so bright in retrospect, but originated from an impulse to find open points, i.e. points \(x\) such that \(\{x\}\) is open already: Open points are pretty “isolated”, because the only sequences converging to \(x\) are the ones being eventually constant.

One might fall into the trap of thinking that maximal elements \(m\) with lower neighbors \(l\) (without the latter, they would only be comparable to themselves) have to be open points by representing them as \((l,\infty)\subseteq (m,\infty)\), but that does not have to be the case:

Example 8
Consider now \(\{1,2,3\}\), again ordered by divisibility. Then \(2,3\) are both maximal with lower neighbor \(1\), but the interval \((1,\infty)=\{2,3\}\) covers more than this point in both cases. Note however that both elements are topologically indistinguishable, which is not too, surprising, since they share the same upper and lower rays.

On a similar note, if \(a<b<c\), just because both \((a,b)\) and \((b,c)\) are empty does not mean \((a,c) = \{b\}\), as the first example shows: \((1,6)=\{2,3\}\). Note that again, both points in question are topologically indistinguishable.

Example 9
  1. Let \(X\) be a set. The poset topology of \((X, \Delta_X)\) has all rays empty except for \((-\infty,\infty)=X\), inducing the indiscrete topology.
  2. Let \(X\) be a finite set. Then any linear ordering induces the discrete topology.

Note that this means we have quite some amount of choice: The topology sees the ordering only on a “local” level, that is, the element in relation to some interval \((a,b)\ni x\).

Question 10
What is a poset-theoretic characterization of when two points are locally homeomorphic?
Question 11
How can we model the topology \(F_n:=(\{1,\ldots,n\}, \{\{1,\ldots,k\}\mid 1 \leq k \leq n\})\) with an order?

Small idea

Definition 12

Associate to a poset \((P, \leq)\) the generated equivalence relation of \(\leq\), i.e. the relation where \(a\sim b\) if and only if there is a sequence \(a=x_1, \ldots, x_n = b\) such that at each position, either \(x_i\leq x_{i+1}\) or \(x_{i+1} \leq x_i\) (check this!).

We call the resulting equivalence class “connected component”, and given a specific \(x\), the associated equivalence class is called the “connected component of \(x\)”.

Definition 13
Let \(x\in P\). Elements of \(\mathscr N_x := \bigcap_{a<x<b}(a,b)\) are called neighbors of \(x\).
Conjecture 14
  • Being a neighbor is an equivalence relation.
  • Neighbors lie in the same connected component.
  • They have the same upper and lower rays.
  • \(P\setminus \mathscr N_x\) decomposes into disjoint proper rays (does it tho?).

The linearly ordered case

(TODO)