3

This is related to a previous question.

Consider the quasiordered set $Q = \{\bot, q,q', \top\}$ with $q \lesssim q'$ and $q \lesssim q',$ such that $\bot$ is the unique least element and $\top$ the unique greatest element. In this structure, there is clearly a strong sense in which $q$ and $q'$ are indistinguishable. For example, we have that:

$$(\forall x \in Q) \;\; x \lesssim q \iff x \lesssim q'$$

Now consider, on the other hand, the diamond poset $D = \{\bot,d,d',\top\}.$ Then okay, $d$ and $d'$ are "weakly" indistinguishable, for example no first-order formula in the language of posets can distinguish them, since after all, we can find an poset automorphism $\alpha : D \rightarrow D$ that permutes $d$ and $d'$. Which means, that for example, we can uniformly substitute $d$ with $d'$ without changing anything. For example, $d \leq d$, so therefore $d' \leq d'$. However, notice that we cannot just substitute $d$ with $d'$ willy-nilly, for example just because $d \leq d$ is true, does not mean $d \leq d'$ is true. So we might say that $d$ and $d'$ fail to be "strongly indistinguishable."

Along a similar vein, let $\alpha : D \rightarrow D$ denote the unique non-trivial such automorphism (the one that permutes $d$ and $d'$). Then although it holds universally that $$x \leq y \iff \alpha(x) \leq \alpha(y)$$

nonetheless it does not hold universally that

$$x \leq y \iff x \leq \alpha(y).$$

So in some sense, $\alpha$ is not a "strong automorphism" of $D$.

What I want to know is, what is the "correct" definition of "strongly indistinguishable" and/or "strong automorphism"?

This seems to be a very general phenomenon, and we can often get the "correct" definitions in an ad hoc manner. This allows us to take a quotient that identifies strongly indistinguishable elements. For example, we can define:

  1. In a quasiordered set... two points $x$ and $y$ are strongly indistinguishable iff $x \lesssim y$ and $y \lesssim x.$

  2. In a pseudometric space... iff $d(x,y)=0.$

  3. In a topological space... iff their neighbourhood filters are equal.

  4. In a measure space... two measurable sets $A$ and $B$ are strongly indistinguishable iff $\mu(A\, \Delta\, B) = 0.$

So, I am wondering, is there is a generic definition of "strongly indistinguishable" that yields the correct specific definition when applied to any given structure? We would like to be able to take a quotient that identifies indistinguishable elements while preserving basically everything that is interesting about the structure of interest.

Edit. I'm also interested to know whether there is an abstract category-theoretic definition of "strong automorphism."

goblin GONE
  • 67,744
  • The problems you point out come from being careless with the logical language. If you add to the signature of the language constant symbols for the elements of your model, then the available formulae and automorphisms change! – Zhen Lin Sep 28 '13 at 07:51
  • @ZhenLin, yes its well-known that we can end up with fewer automorphisms if we expand the signature. But, what is the significance here? – goblin GONE Sep 29 '13 at 06:09
  • All of your examples are really just "naturally occurring quasiorders". The question seems, to me at least, to ask whether or not "naturally occurring" can be given a generic definition. To that, I say, probably not. – Asaf Karagila Sep 29 '13 at 07:39
  • @AsafKaragila, I'm not sure I understand what you're getting at. I agree that we're talking about "naturally occurring quasiorders," because after all we're talking about "naturally occurring equivalence relations." However, you specifically said "quasiorders" rather than "equivalence relations," which makes me suspect I haven't grasped your meaning. – goblin GONE Sep 29 '13 at 09:24

1 Answers1

1

The only idea that springs to mind that is the same for all four is category-theoretic in nature.

If $Q$ is a preorder (I'm not familiar with the name quasi-order, but I think that's what you mean), then the poset $TQ$ you get by identifying equivalent elements, along with the quotient map $Q \to TQ$, has the following universal property: if $P$ is any poset, and $Q \to P$ is an order homomorphism, then there is a unique way to factor it as $Q \to TQ \to P$.

Put differently, there is a bijection between the set of homomorphisms $TQ \to P$ and the set of homomorphisms $Q \to P$, and the bijection is given by composing with the map $Q \to TQ$.

The above property uniquely determines (up to isomorphism) the poset $TQ$ and the map $Q \to TQ$. The only 'choice' we made was that we decided to single out the class of posets from the class of all preorders.

The general notion is that of a reflective subcategory: $\mathscr{C}$ is a full subcategory of $\mathscr{D}$, and the inclusion functor $i : \mathscr{C} \to \mathscr{D}$ has a left adjoint $T$. We have the property that

$$ \hom(TD, C) \cong \hom(D, C) \quad (= \hom(D, iC))$$

for $D \in \mathscr{D}$ and $C \in \mathscr{C}$. And as in the special case above, we have a natural transformation $\mathbf{1}_{\mathscr{D}} \to i \circ T$. e.g. we get maps $D \to TD$.

Your first example comes from $\mathbf{Poset} \subseteq \mathbf{Preorder}$.

The second example comes from the category of metric spaces as a full subcategory of the category of quasi-metric spaces.

The third example is, I think, the category of $T_0$ spaces as a subcategory of the category of all topological spaces.

The fourth one... well, I'm not versed enough in measure theory to know if there is a standard name for the two categories involved.


Of course, this is somewhat tangential to the question you actually asked: I've answered "what is the 'quotient' map that identifies 'indistinguishable' elements?" (I put 'quotient' in quotes, since it might not actually be a quotient). But from the category theoretic point of view, it's the maps that matter moreso than the elements anyways, so it's probably the question you should have asked.

  • See I thought of this. My concern, though, is that "reflective subcategory" is too lenient of a condition. For instance, in the transition from quasi-ordered set to partially-ordered set that occurs by identifying strongly indistinguishable elements, I feel that essentially no structure has been lost. On the other hand, the transition from a group to an Abelian group forgets a lot interesting structure. So it seems to me that we're looking for something like a notion of "strongly reflective subcategory," if that makes sense. – goblin GONE Sep 29 '13 at 06:00
  • @User: I think that's an artifact of what structure you're interested in. I've often been in the situation where the non-abelian structure is not interesting, thus passing from a group to its abelianization. I've also occasionally been in situations where it was useful not to identify isomorphic elements: for example, an affine space has a natural representation as a category where every homset has exactly one element: a quasiorder where all objects are in the same equivalence class. (arguably this is interesting because of additional strucutre) –  Sep 29 '13 at 17:06
  • Let me think about it a bit, and I'll get back to you. – goblin GONE Sep 30 '13 at 14:24