3
$\begingroup$

Working in the context of set theory $\sf ZF$, selection may be defined as a function from nonempty sets to their elements. Formally:

$\operatorname {selective}(c) \iff \operatorname {function}(c) \land \forall x \in dom(c): x \neq \emptyset \to c(x) \in x$

Restricted selections to be defined in a conditional manner by letting $c$ itself be the result of a function from prior parameters, i.e. we can have $c=F(x_1,..,x_n)$, this way we may say that $c$ is conditioned after $x_1,..,x_n$ and the function $F$, also we can impose conditions on the kind of sets from which selections are made. To capture that formally we write:

$ \forall x_1,..,\forall x_n \exists F \forall y \neq \emptyset : \Omega \to \exists z \in y: F(x_1,..,x_n)(y)=z $

Where $F$ may be used in $\Omega$.

This conditional selection principle is to be denoted $\mathcal S^\Omega$.

For example we write the axiom of dependent choice $\sf DC$, along those lines:

If $R$ is a set implementing a total relation on set $S$, denoted $R|^T S$, then we take the relation $$R^*=\{ \langle x, s \rangle \mid x \in S, s=\{y \in S \mid x \ R \ y \} \}$$. For short, we'll denote the above formula as: $\Psi$

$ \forall S \forall R \forall h \exists F \forall y \neq \emptyset: \\ \Big{(}R|^T S \land \exists R^* \exists x : \Psi \land h \in S \land y=R^*(x) \land \\ \big{[}x=h \lor \exists d \in S : x=F(S,R,h)(R^*(d)) \big{]} \Big{)} \\ \to \\ \exists z \in y: z=F(S,R, h)(y)$

So, dependent choice is one form of conditional selection!

To get the full axiom of choice, we set $n=1$, and $\Omega$ to be $y \in x$, then we get: $$\forall x \exists F \forall y \neq \emptyset : y \in x \to \exists z \in y: z=F(x) (y)$$

To get countable choice, we set $n=1$, and $\Omega$ to be $|x|=\omega \land y \in x $, then we get: $$\forall x \exists F \forall y \neq \emptyset: |x|=\omega \land y \in x \to \exists z \in y: z=F(x)(y)$$

Now, if we define choice principle over $\sf ZF$, in the following manner:

$\sf H$ is a choice principle if and only if we have a formula $\Omega$ such that both of the following are fulfilled:

\begin{align} \bullet \ \sf (ZF+H) \vdash (ZF+\mathcal S^\Omega)\\ \bullet \ \sf (ZF + \mathcal S^\Omega) \vdash (ZF + H) \end{align}

. And provided that: $\sf ZF \not \vdash \mathcal S^\Omega$

Then can we prove (in $\sf ZF$ or some suitable extension of it) that the Ordering Principle [every set can be linearly ordered] is a choice principle?

[Addendum] I should relate my personal output on that issue: principles like the Ordering Principle and the Small Violations of Choice principle [see Asaf's posting], which I think are spoken about in the context of what may be called as weak choice principles, meaning that they are fragments of $\sf AC$ that are not provable in $\sf ZF$. As, a terminology I'd prefer to call those pre-choice Principles. So, if they prove to be inequivalent with any selection principle, then they are proper pre-choice principles. This is to discriminate them from weak choice principles like $\sf DC, CC, etc..$ which are equivalent to selection principles, but weaker than $\sf AC$.

$\endgroup$
4
  • 1
    $\begingroup$ This seems more like a blog post than a question, since not everyone will buy into your characterization of what a choice principle is. $\endgroup$ Commented Dec 23, 2023 at 15:25
  • 1
    $\begingroup$ @JoelDavidHamkins, but there are two related questions specifically asked. Specially whether the Ordering principle is equivalent to a selection principle? $\endgroup$ Commented Dec 23, 2023 at 15:39
  • $\begingroup$ The question begins with a contentious claim about what a choice principle "is", and I believe much of your audience simply doesn't read past that. I would recommend a severe edit. $\endgroup$ Commented Dec 23, 2023 at 15:50
  • $\begingroup$ @JoelDavidHamkins, OK! I've edited it. $\endgroup$ Commented Dec 23, 2023 at 18:17

2 Answers 2

9
$\begingroup$

Here is one way to view the so-called ordering principle as a selection principle.

Theorem. The following are equivalent over ZF set theory:

  1. Every set admits a linear order.
  2. For every set $X$, there is a choice function choosing for each linear order of a proper subset of $X$ a proper extension of that order to a linear order of a subset of $X$.

In 2, the proper extension can add just one or a lot of new points, as it likes.

Proof. For the implication $(1\to 2)$, suppose that we have a set $X$. We may fix a linear order $\leq$ of $X$. Now, given a linear order $\preceq$ of a proper subset $Y\subseteq X$, we can simply extend $\preceq$ by placing all the other elements of $X-Y$ on top and in the order induced by $\leq\upharpoonright(X-Y)$.

For the implication $(2\to 1)$, suppose that we have a function $F$ that chooses larger linear orders for any given linear order of a proper subset of $X$. Define $\langle X_\alpha,\leq_\alpha\rangle$ by recursively applying the function, starting with the empty order $X_0=\emptyset$, applying $F$ at successor stages, and taking unions at limits. Since the set of all linearly ordered subsets of $X$ is a set in ZF, it has a Hartog number, and so this recursion cannot go forever through the ordinals. But the only way it stops is if $X_\alpha=X$ for some $\alpha$. And thus $X$ is linearly orderable as desired. $\Box$

(We can formulate statement 2 as a selection principle, by considering families of sets of pairs $\langle \preceq,\leq\rangle$, where these are linear orders of subsets of $X$ and the first extends properly to the second, and where in each set in family, the first order $\preceq$ is the same. So the desired choice function is selecting such a pair, which in effect is selecting a particular extension $\leq$ of $\preceq$.)

If the choice function in statement 2 always adding just one more point, then this would provide a well-ordering of $X$. So in the cases where AC fails and $X$ is not well-orderable, then the choice function will generally be placing a lot of new points into the chosen larger order.

$\endgroup$
8
  • $\begingroup$ Feels a bit convoluted, to be honest. "A set can be linearly ordered if and only if for every two points in the set we can choose a linear ordering which puts one above the other", although admittedly not as contrived. $\endgroup$
    – Asaf Karagila
    Commented Dec 23, 2023 at 20:32
  • $\begingroup$ @AsafKaragila I don't quite understand your comment. Are you saying that if we can choose from pairs, then every set is linearly orderable? (But in any case, I'm just trying to answer the question.) $\endgroup$ Commented Dec 23, 2023 at 20:36
  • $\begingroup$ But why is it convoluted? Obviously every linearly ordered proper subset of $X$ can be extended to slightly larger linearly ordered subsets, and the ordering principle is exactly equivalent to having a choice function for this. Well-orderability is equivalent to being able to do this while adding just one new point. $\endgroup$ Commented Dec 23, 2023 at 20:39
  • $\begingroup$ In particular, in statement 2 we're just choosing a strictly larger linear order on a subset, not a linear order of the whole of $X$. $\endgroup$ Commented Dec 23, 2023 at 20:58
  • $\begingroup$ Regarding my remark, choice for pairs does not prove every set is linearly orderable, since then it would imply choice for finite sets, which it doesn't. $\endgroup$ Commented Dec 23, 2023 at 21:48
6
$\begingroup$

Here is a variation on the same theme as Joel David Hamkins' answer.

Theorem. The following are equivalent over ZF set theory:

  • Every set admits a linear order.
  • For every set $X$, there is a function choosing for each subset $Y$ of $X$ a total preordering of $Y$ which is nontrivial unless $|Y| \leq 1$.

By preordering, I mean a reflexive transitive relation on a set; a total preordering is one where any two elements are comparable (perhaps both ways). A total preordering is nontrivial if there are two elements that are not comparable.

$\endgroup$
5
  • $\begingroup$ Nice! You iteratively apply it to the equivalence classes to gradually resolve the whole order. $\endgroup$ Commented Dec 25, 2023 at 21:04
  • $\begingroup$ For example, it suffices to pick nontrivial subsets of a given set, since such a choice function gives a nontrivial preorder. $\endgroup$ Commented Dec 25, 2023 at 21:22
  • $\begingroup$ I made a tweet explaining this latter point. (Perhaps this is already known?) twitter.com/JDHamkins/status/1739421469426266335 $\endgroup$ Commented Dec 25, 2023 at 23:20
  • 1
    $\begingroup$ Yes, this is known as the Kinna-Wagner Selection Principle. It is actually equivalent to: every set injects into $2^\alpha$ for some ordinal $\alpha$. (IIRC this is in Jech's The Axiom of Choice.) Since $2^\alpha$ is linearly ordered, the Ordering Principle follows. $\endgroup$ Commented Dec 26, 2023 at 17:15
  • $\begingroup$ Ah, yes, that's what the argument gives, since we are splitting each equivalence class in two at each step. And the converse is immediate. Great! $\endgroup$ Commented Dec 26, 2023 at 17:24

You must log in to answer this question.

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