There are concise and elegant characterisations of the real line as a topological space and as an ordered space in the literature. I am interested in the harder case of characterising subsets of the reals in this manner. There are satisfactory answers to the topological version (e.g., de Groot, Mary Ellen Rudin) which are, as to be expected, more complicated and inticrate in proof than for the whole space. I recall reading a solution for the corresponding result in the category of ordered spaces but the standard search methods have failed to locate it. Can anybody on this site assist me with a reference?
-
4$\begingroup$ Doesn't "linear order with a countable dense set" do the trick? $\endgroup$– bofCommented Mar 13, 2015 at 6:04
-
1$\begingroup$ Well, if I had known the answer to YOUR question, I wouldn't have posed MINE. Could you indicate a proof or provide a reference? $\endgroup$– weatherCommented Mar 13, 2015 at 8:23
-
$\begingroup$ I think that what bof meant was: Are you not happy with the (well known) characterization of the subsets of the reals that he/she has given? One direction follows from the old result (Cantor's?) that every countable linear order embeds into the rationals. $\endgroup$– GoldsternCommented Mar 13, 2015 at 11:00
-
3$\begingroup$ @bof I think the proposed characterization is not quite right, since you have to worry about discrete intervals. If you duplicate every real number twice, then you've got separable linear order that does not embed into $\mathbb{R}$. $\endgroup$– Joel David HamkinsCommented Mar 13, 2015 at 14:31
-
$\begingroup$ Not directly relevant, but you may find it interesting: one variation on your question is, "How much information do I need to provide to specify a suborder of $\mathbb{R}$ of size $\kappa$, up to order isomorphism?" This is of course insanely complicated, so it's reasonable to restrict to "nicely behaved" suborders - say, those $L$ which are $\kappa$-dense. Then in general the situation is quite complicated, but some set-theoretic hypotheses can make things extremely nice - e.g. PFA implies that all $\aleph_1$-dense suborders of $\mathbb{R}$ are isomorphic. $\endgroup$– Noah SchweberCommented Mar 13, 2015 at 14:50
2 Answers
The suggestion in the comments that a linear order embeds into $\mathbb{R}$ just in case it has a countable dense set is not quite true. For example, let $2\times\mathbb{R}$ be the doubled real line ($\mathbb{R}$ copies of $2$), the order arising from the reals by replacing each real number with two copies, a lower one and an upper one, adjacent to each other. This order has a countable dense set (dense in the order topology) — that is, it is separable in the order topology — since every nonempty open interval $(a,b)$ continues to contain (two duplicates of) a rational number. But $2\times\mathbb{R}$ cannot embed into $\mathbb{R}$ because any dense order extending $2\times\mathbb{R}$ will not be separable.
Meanwhile, there is a nearby characterization that does succeed:
Theorem. The following are equivalent for a linear order $\langle L,<\rangle$.
$\langle L,<\rangle$ embeds into the real line $\langle\mathbb{R},<\rangle$.
$\langle L,<\rangle$ is separable and has at most countably many closed intervals $[a,b]$ with $a<b$ and $(a,b)=\emptyset$.
Proof. ($1\to 2$) This is clear, since once you have embedded $L$ into $\mathbb{R}$, then you can pick an element from $L$ from each rational interval having any elements from the image of $L$, and this will be dense in $L$, and also $L$ can have at most countably many discrete intervals $[a,b]$, because each will have a distinct rational in $\mathbb{R}$.
($2\to 1$) Here, you use the suggestion from the comments. Let $Q\subset L$ be a countable dense set that also includes all the endpoints $a$ and $b$ from any discrete closed interval $[a,b]$ in $L$. This is a countable linear order, which we may map into the rationals $\mathbb{Q}$ using the usual back-and-forth (but just forth) construction of Cantor. Now, we may extend this to all of $L$, since any element of $L$ is the least upper bound in $L$ of the elements of $Q$ below it. QED
Meanwhile, the separable linear orders are characterized as those that embed into $2\times\mathbb{R}$. We just map the countable dense set into the rationals, using always the lower rational number, say. And then we can extend this to the whole order since every element of $L$ is either the LUB or the GLB of a subset of that countable dense set, and so we can map into $2\times\mathbb{R}$ accordingly.
Theorem. The following are equivalent for any linear order$\langle L,<\rangle$.
$\langle L,<\rangle$ embeds into the doubled real line $2\times\mathbb{R}$.
$\langle L,<\rangle$ has a countable dense set.
-
1$\begingroup$ I think 2 can be put more simply: there is a countable $Q\subseteq L$ such that $a,b\in L,a\lt b\implies[a,b)\cap Q\ne\emptyset.$ Let $Q=\{q_n:n\lt\omega\}$ and define $f:L\to\mathbb R$ by $f(x)=\sum_{q_n\lt x}2^{-n}$. Or something like that. $\endgroup$– bofCommented Mar 13, 2015 at 15:47
-
$\begingroup$ Yes, that is equivalent. Also, one can assert the dual: there is a countable $Q\subset L$ with $a<b\to (a,b]\cap Q\neq\emptyset$. $\endgroup$ Commented Mar 13, 2015 at 15:50
-
$\begingroup$ There should be a condition like $a<b$ in 2. otherwise there are uncountably many intervals $[a,a]$ with $(a,a)=\emptyset$ in $\mathbb{R}$ itself. $\endgroup$ Commented Mar 13, 2015 at 15:55
-
$\begingroup$ Thanks for this nice answer. The result is so definitive and elegant that I feel that it must exist in the literature. I would like to include a reference and will be happy to cite this response but, of course, I am still interested in the question of whether there are earlier sources. $\endgroup$– weatherCommented Mar 16, 2015 at 9:27
-
2$\begingroup$ The paper "Separable linear orders and universality" is related. $\endgroup$ Commented Mar 16, 2015 at 12:02
Just for the record, the result seems to be due to Isodore Fleischer ("Numerical representation of utility", Jour. Soc. Ind. Appl. Math, 9 (1961) 48-50). As the title indicates, it is useful in determining when a suitable ordering on a state space is induced by a numerical function (price, utility function, temperature, entropy) and unifies many such results in fields such as economics, thermodynamics, philosophy ... This (surprisingly late) reference is the earliest one that I can trace.
-
1$\begingroup$ Mehta, G. On a theorem of Fleischer. J. Austral. Math. Soc. Ser. A 40 (1986), no. 2, 261–266. From its MR (by M. Erné) "The theorem alluded to in the title states that a linearly ordered set is isomorphic to a subset of the real line if and only if it is separable in its order topology and has at most countably many jumps (i.e. two-element intervals) [I. Fleischer, SIAM J. 9 (1961), 48–50; MR0122551]". The MR of the latter says nothing on the contents, and it has no review on ZBmath. $\endgroup$– YCorCommented Jan 13, 2022 at 16:11