When a lattice $(L,\leqslant)$ is a lattice of open (or closed) sets of some topological space $(X,\tau)$? Which conditions have to be satisfied? We may assume that $X$ is $T_1$.
2 Answers
A lattice $L$ is isomorphic to the lattice of open sets in a topological space iff the following two conditions hold:
$L$ is complete: every subset $S\subseteq L$ has a least upper bound.
$L$ has enough prime elements (elements $p\in L\setminus\{1\}$ such that $a\wedge b\leq p$ implies $a\leq p$ or $b\leq p$): that is, if $a,b\in L$ and $a\not\leq b$, then there exist a prime element $p\in L$ such that $b\leq p$ and $a\not\leq p$.
Such a lattice is called a spatial frame. More generally, a frame is a complete lattice which also satisfies the infinite distributive law $a\wedge\bigvee S=\bigvee\{a\wedge s:s\in S\}$ (this infinite distributive law is a consequence of condition (2)). Frames are studied as a generalization of (sober) topological spaces; roughly speaking, a frame is like the lattice of "open sets" in a "topological space" in which an open set may not be determined by the points it contains.
As a sketch of a proof, if $L$ is the lattice of open sets in a topological space $X$, then it satisfies (1) since an arbitrary union of open sets is open. For (2), note that for any $x\in X$, the open set $p_x=X\setminus\overline{\{x\}}$ is prime (the open sets it contains are exactly the open sets that do not contain $x$). If $a,b\subseteq X$ are open with $a\not\subseteq b$, let $x$ be such that $x\in a$ and $x\not\in b$, and then $b\subseteq p_x$ and $a\not\subseteq p_x$.
Conversely, suppose $L$ satisfies (1) and (2). Let $X$ be the set of all prime elements of $L$. We can then put a topology on $X$ by saying that the open sets are the sets of the form $U_a=\{p\in X:a\not\leq p\}$. The map $a\mapsto U_a$ turns arbitrary joins into unions and finite meets into intersections, so (1) implies that this is indeed a topology. Moreover, (2) implies that $a\leq b$ iff $U_a\subseteq U_b$, so that $a\mapsto U_a$ is an order-isomorphism.
If you want to require the topological space to be $T_1$, you need to modify condition (2) to say that $L$ has enough maximal prime elements (that is, prime elements $p$ such that if $q\geq p$ and $q$ is prime then $q=p$). The proof is roughly the same: if $X$ is a $T_1$ space, then $\{x\}=\overline{\{x\}}$ for all $x\in X$ and so the open sets $p_x$ used above are not just prime but maximal prime (since the only set larger than them is $X$ itself). Conversely, if $L$ is complete and has enough maximal primes, you can take $X$ to be the set of maximal primes and topologize it as before, and maximality of the primes will guarantee that $X$ is $T_1$.

- 330,363
-
This is a neat fact - do you know a reference for it? – Noah Schweber Sep 18 '18 at 23:51
-
I don't know a citation for this exact statement. The statement that a frame is spatial iff it satsifies (2) can be found implicitly in section II.1.5 of Peter Johnstone's Stone spaces. I don't recall seeing the slightly stronger statement that omits the infinite distributive law and just uses (1) and (2) before (in fact, my memory is that I first noticed the infinite distributive law is not necessary in the course of writing up this answer). – Eric Wofsey Sep 19 '18 at 00:16
-
To be more precise, Johnstone proves that the frame of open sets in any topological space satisfies (2). He also, given a frame $A$, constructs a space $pt(A)$ (the space of prime elements I used in my answer) and a homomorphism from $A$ to the frame of open subsets of $pt(A)$, and remarks (without proof) that this map is an isomorphism iff $A$ satisfies (2). (The proof is very straightforward from the definition, though.) – Eric Wofsey Sep 19 '18 at 00:20
-
According to Johnstone's historical notes, this characterization of lattices of open sets was first published by J. R. Büchi in 1952 in the paper "Representation of complete lattices by sets". – Eric Wofsey Sep 19 '18 at 00:38
-
To be clear, I don't mean to imply that there are a lot of people going around stating this fact with the infinite distributive law as a redundant extra hypothesis. Rather, it is often stated in the context of discussing the duality between frames and spaces, so all the lattices under discussion are assumed to be complete and infinitely distributive from the beginning. – Eric Wofsey Sep 19 '18 at 00:51
-
Thanks a lot, that's interesting background! – Noah Schweber Sep 19 '18 at 01:32
A necessary condition (I don't know if it is sufficient) is that it must be a distributive lattice.

- 427,504
-
1It must also be complete, since the lattice of open sets of a topological space is complete. – amrsa Jan 15 '18 at 13:02
-
-
Additionally, the lattice of open sets must be a complete Heyting Algebra. However, the one of closed sets doesn't, because it may not satisfy the infinite distributive law; it satisfies its dual. I think this is as far as I can go... – amrsa Jan 15 '18 at 14:28