1
$\begingroup$

$\DeclareMathOperator\Idl{Idl}$Let $P$ be a finite, connected poset with at least two elements, and let $\Idl_{\neq \emptyset, P}(P)$ be the set of downward closed sets $S \subset P$ such that $S \neq \emptyset$ and $S \neq P$, ordered under inclusion.

Question: Is $\Idl_{\neq \emptyset, P}(P)$ weakly contractible?

Notes:

  • If $P$ has less than two elements, then $\Idl_{\neq \emptyset, P}(P)$ is empty. If $P$ is a discrete poset with $\geq 2$ elements, then $|\Idl_{\neq \emptyset, P}(P)| \simeq S^{|P|-2}$. Hence the restriction to connected posets with $\geq 2$ elements.

  • If $P$ has four elements $a,b < c,d$, so that $|P| \simeq S^1$, then $\Idl_{\neq \emptyset, P}(P)$ has 5 elements arranged in an "$X$" shape, and is weakly contractible.

  • In general, if we add $\emptyset$ and $P$ back in, the poset $\Idl(P)$ of all ideals is always contractible (for at least 4 reasons -- it has an initial and a terminal object, and binary meets and binary joins). Similarly, $\Idl_{\neq \emptyset}(P)$ and $\Idl_{\neq P}(P)$ are contractible if $P \neq \emptyset$.

  • I checked a few random examples with Sage, which at least turned out to have vanishing reduced homology. (Although I'm not sure how much I trust my code!)

$\endgroup$
7
  • $\begingroup$ (Finite) distributive lattices are well known to be shellable. Does that answer your question? $\endgroup$ Commented Jan 11, 2022 at 3:00
  • $\begingroup$ For example, see Corollary 3.2 of Björner's "Shellable and Cohen-Macaulay partially ordered sets" (doi.org/10.1090/S0002-9947-1980-0570784-2) $\endgroup$ Commented Jan 11, 2022 at 3:04
  • $\begingroup$ @SamHopkins I'm not sure. $Idl_{\neq \emptyset, P}(P)$ is not necessarily a lattice -- the "meet" of two elements might be empty / their "join" might be $P$. On the other hand, I'm not terribly familiar with shellability or Cohen-Macaulayness of simplicial complexes, but it apparently doesn't imply being weakly contractible. So maybe you were suggesting a negative answer? $\endgroup$ Commented Jan 11, 2022 at 3:12
  • $\begingroup$ Don't worry about $\varnothing$ and $P$: when studying the topology of lattices it is common to implicitly remove the maximal and minimal elements (because these are cone points, as you note). The homotopy type of a distributive lattice is then known to be either a ball or a sphere. See the classical paper of Björner referenced above or arxiv.org/abs/1407.5311. $\endgroup$ Commented Jan 11, 2022 at 3:16
  • 1
    $\begingroup$ Regarding your restriction to connected $P$, see Theorem 3.7 of the paper of Hersh and Meszaros: if $P$ is connected (and has $\geq 2$ elements), then the whole distributive lattice $J(P)$ will not be a join of atoms, so in fact the homotopy type of $J(P)$ is that of a ball. $\endgroup$ Commented Jan 11, 2022 at 3:19

2 Answers 2

3
$\begingroup$

Let me be a little more historical. As mentioned, it is well known that (open intervals of) distributive lattices are shellable. This is mentioned in Corollary 3.2 of Björner's classic paper "Shellable and Cohen-Macaulay partially ordered sets" (https://doi.org/10.1090/S0002-9947-1980-0570784-2) but is an older result I believe.

The homotopy type of a shellable complex is a wedge of spheres (of the same dimension). How many spheres do we get? By a basic homology computation, we get $|\mu(\hat{0},\hat{1})|$ of them, where $\mu$ is the Möbius function and $\hat{0}$/$\hat{1}$ are the minimal/maximal elements of the lattice. In a distributive lattice $J(P)$ of order ideals, the Möbius function is given by $$\mu(I,I') = \begin{cases} (-1)^{\# I' \setminus I} &\textrm{if $I' \setminus I$ is an antichain}, \\ 0 &\textrm{otherwise}. \end{cases}$$ See e.g. Example 3.9.6 of Stanley's EC1, 2nd ed. So your restriction to connected $P$ with $\geq 2$ elements gives $\mu(\hat{0},\hat{1})=0$, hence the open interval $(\hat{0},\hat{1})$ in $J(P)$ is a (homotopy) ball.

$\endgroup$
4
  • $\begingroup$ So in fact you don't even need that $P$ is connected, just that it is not a disjoint union of points. $\endgroup$ Commented Jan 11, 2022 at 3:34
  • $\begingroup$ Yeah, that's really surprising, somehow! $\endgroup$ Commented Jan 11, 2022 at 3:35
  • 1
    $\begingroup$ @TimCampion: If you're interested in homotopy types of other lattices, I recommend the "Poset Topology" monograph by Wachs (math.miami.edu/~wachs/papers/toolnotes.pdf), though it doesn't mention distributive lattices as far as I can tell (perhaps because they are basically topologically trivial as explained here). $\endgroup$ Commented Jan 11, 2022 at 3:48
  • $\begingroup$ Thanks! This came up in a rather particular setting -- if I haven't messed up too badly, computing the "space of composites" of an $n$-categorical pasting diagram (which, for various definitions of "pasting diagram", is known to be connected, and hoped to be contractible), can be reduced to computing the homotopy type of a poset of ideals like this one. In general, I think it should be not uncommon in higher category theory to be able to reduce some computation to finding the homotopy type of a particular poset, so actually knowing something about how to do that does seem pretty valuable! $\endgroup$ Commented Jan 11, 2022 at 3:53
2
$\begingroup$

Here is another proof. Let $M$ be the set of minimal elements of $P$, and let $2_{\ne \emptyset}^M$ be the poset of non-empty subsets of $M$. There is an obvious inclusion of posets $2_{\ne \emptyset}^M \hookrightarrow Idl_{\ne \emptyset}(P)$. Assuming that $P$ is not a disjoint union of points, this restricts to an inclusion $$i\colon 2_{\ne \emptyset}^M \hookrightarrow Idl_{\ne \emptyset, P}(P).$$

There also is a map of posets $$r\colon Idl_{\ne \emptyset, P}(P) \to 2_{\ne \emptyset}^M,$$ which sends an ideal $I$ to $I\cap M$. The composition $ri$ is the identity, while $ir$ is related to the identity by the inequality $ir(I)\le I$ for every ideal $I$. It follows that $i$ and $r$ induce homotopy equivalences of geometric realizations $$|Idl_{\ne \emptyset, P}(P)| \simeq |2_{\ne \emptyset}^M|.$$ But $2_{\ne \emptyset}^M$ has a maximum element $M$, so its geometric realization is contractible.

$\endgroup$
9
  • $\begingroup$ Oh nice! I had been hoping to find some kind of adjointness argument like this. I was trying to work with maximal elements, though, and the condition dual to not being a discrete poset never occurred to me! $\endgroup$ Commented Jan 11, 2022 at 4:07
  • 2
    $\begingroup$ This is essentially an application of Rota's cross-cut theorem. Here M is the cross-cut. $\endgroup$ Commented Jan 11, 2022 at 13:21
  • 1
    $\begingroup$ @BenjaminSteinberg: yes, and the cross-cut theorem is behind the Mobius function computation mentioned in my answer. $\endgroup$ Commented Jan 11, 2022 at 15:44
  • $\begingroup$ @BenjaminSteinberg Thanks for the pointer -- it's a shame that there seems to be a bit of a "cultural (and terminological) gap" between the study of simplicial complexes and categorical homotopy theory. $\endgroup$ Commented Jan 11, 2022 at 18:25
  • $\begingroup$ @TimCampion, Quillen seemed to play both sides successfully. Anyway many proof use that a Galois connection (aka adjunction) of posets gives an equivalence of categories. $\endgroup$ Commented Jan 11, 2022 at 19:20

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .