2

I'd like to prove the following (Engelking, exercise 6.3.2(c)).

Theorem: Every second countable LOTS is embeddable in $\mathbb R$.

Here, LOTS = linearly ordered topological space. "Embeddable in $\mathbb R$" means there is a homeomorphic embedding as topological spaces.

Since LOTS are regular Hausdorff, by Urysohn's metrization theorem this is equivalent to:

Theorem: Every separable metrizable LOTS is embeddable in $\mathbb R$.

In the answer below I have put a sketch of a proof and would appreciate if people could check the argument. It's a little long, but I don't see how it can be shortened.

PatrickR
  • 7,165
  • Suppose we believe the fact that every countable LOTS embeds into $\mathbb Q$ (or prove it with the standard back-and-forth argument). Given a countable base, it can be refined to segments $(a,b)$. Then the endpoints of the base embed in $\mathbb Q$. I feel like we can then extend this embedding to get the entire LOTS embedded into $\mathbb R$. – Steven Clontz May 10 '24 at 00:39
  • @StevenClontz Interesting approach. It would be instructive to see a careful justification following these lines. Part of what you are proposing have maybe already been discussed in other posts? I'll take a look. – PatrickR May 10 '24 at 19:26
  • Well I took a little time to try and work out the details, but ran into a roadblock: think about ${0}\cup{\pm\frac{1}{n}:n\in\mathbb Z^+}$: the segments might have endpoints ${\pm\frac{1}{n}:n\in\mathbb Z^+}$ which form a discrete space and could be sent to $\mathbb Z$ without some extra care - in particular, we cannot extend that embedding to send $0$ to the right place. – Steven Clontz May 11 '24 at 00:23
  • @StevenClontz https://math.stackexchange.com/questions/3722543 (specifically the answer by Eric Wofsey) may be the missing ingredient. Any countable LOTS can be embedded into $\mathbb Q$ simultaneously as an order embedding and a topological embedding. So maybe taking a countable topologically dense subset $D$ of $X$, embedding into $\mathbb Q$ as per that question, and somehow showing the closure of $D$ can be mapped to $\mathbb R$ (as Dedekind-MacNeille completion of $\mathbb Q$) and show continuity? Not completely clear. – PatrickR May 12 '24 at 03:55
  • @StevenClontz regarding your roadblock with the $1/n$ example, I think we should show the space could be mapped into the closed interval $[0,1]$ instead of into $\mathbb R$. That would preclude this type of problem. – PatrickR May 12 '24 at 04:33

1 Answers1

2

Let $(X,<)$ be a totally ordered set with corresponding order topology $\tau$ and suppose the topology is second countable. Also suppose $X$ contains at least two points, otherwise the result is trivial.

(side note: Henno Brandsma made repeated use of the "gap" and "jump" terminology below. I thought it was his own, but found out it's used with the same meaning on p. 5 of Engelking.)

Step 1: Construct the Dedekind-MacNeille completion $(X^*,<)$ of $(X,<)$ (using the same symbol $<$ for the order relation) and let $\tau^*$ be the corresponding order topology. So $X^*$ is obtained by adding an element $\alpha$ for each "gap" (= Dedekind cut) in $X$, that is, for each pair of nonempty sets $(A,B)$ such that $X=A\cup B$, $a<b$ for each $a\in A$ and $b\in B$, $A$ has no maximum element and $B$ has no minimum element; and also adding a smallest element and a largest element if $X$ did not already have one. The resulting $X^*$ is order-complete (i.e., every subset has a least upper bound) and $\tau^*$ is compact.

Claim: The order topology $\tau$ on $X$ coincides with the subspace topology induced by the order topology $\tau^*$ on $X^*$ and $X$ is topologically dense in $X^*$.

This is easy to check.

Claim: There are only countably many "jumps" in $(X,<)$, i.e., pairs of elements $x,y\in X$ with $x$ an immediate predecessor of $y$.

Reason: Let $\mathscr B\subseteq\tau$ be a countable base for the topology on $X$. For each jump $(x,y)$, the open set $(\leftarrow,x]$ contains an element of $\mathscr B$ containing $x$. So there is a element $B_x$ of the countable base having $x$ as a maximum. Therefore these $B_x$ must all be distinct, and there are only countably many possible $x$.

Step 2: Fill in the jumps. For each jump $\langle x,y\rangle$ in $X$ (which are the same as the jumps in $X^*$), insert a copy of the open real interval $(0,1)$. This constructs a larger ordered set $(X',<)$ into which the ordered set $(X^*,<)$ embeds. Let $\tau'$ be the corresponding order topology on $X'$. $X'$ does not have any jump and does not have any gap (not hard to check).

Claim: The order topology $\tau^*$ on $X^*$ coincides with the subspace topology induced by the order topology $\tau'$ on $X'$.

This is easy to check by comparing neighborhoods of points of $X^*$ for the two topologies, separately for cases when the point is an endpoint of one of the jumps and not.

So by transitivity the ordered set $(X,<)$ embeds into $(X',<)$ as an ordered set and the inclusion is a homeomorphic embedding for the corresponding order topologies.

The space $X$ is separable and dense in $X^*$. And the real interval $(0,1)$ is also separable. So by taking the union of a countable dense subset of $X$ with a countable dense subset for each copy of $(0,1)$ used to fill the countably many gaps, one obtains a countable dense subset of $X'$, showing that $X'$ is separable.

Claim: $X'$ is homeomorphic to the real interval $[0,1]$.

Reason: Let $D$ be a countable (topologically) dense subset of $X'$. Since $X'$ has no jumps, $D$ is order-dense (for each $x,y\in D$ with $x<y$, there is some $z\in D$ with $x<z<y$) and also we can assume $D$ does not contain the smallest and largest element of $X'$, i.e., $(D,<)$ is unbounded. And $X'$ contains no gaps, so is (order-isomorphic to) the Dedekind-MacNeille completion of $(D,<)$. Now, by Cantor's isomorphism theorem any two countable order-dense unbounded linear orders are order-isomorphic. So $D$ is order-isomorphic to $(0,1)\cap\mathbb Q$. And their respective Dedekind-MacNeille completions are also order-isomorphic. So $X'$ is order-isomorphic to the real interval $[0,1]$, and their corresponding order topologies are homeomorphic.

Finally, this shows that $X$ is homeomorphic to a subspace of $[0,1]$ and embeds into $\mathbb R$.


NOTE: The homeomorphic embedding of $X$ into $\mathbb R$ in the proof above is also order-preserving.

PatrickR
  • 7,165