Obviously the question in the title alone doesn't make sense so I'll develop on the context and then I'll ask my question : Studying $AD$ (axiom of determinacy) I had to prove that $AD$ and $AC$ are incompatible (mod $ZF$). So to prove this I show that under $AC$, there are some undetermined games, and in order to prove this I use the fact that the Fréchet filter can be extended to an ultrafilter (necessarily non principal). As a remark, I thought "This shows that $AD$ is incompatible, not only with $AC$, but also with strictly weaker choice principles, such as the $BPI$ or the statement 'There exists a non-principal ultrafilter over $\omega$' ". Thinking this I wondered how strong that last statement (let $\Omega$ denote said last statement) was, and whether it could imply $BPI$. The answer seems to be "Obviously, no", as $\Omega$ is only about a specific set ($\omega$), whereas the $BPI$ is much more general. But then I wondered whether there was any "general choice principle", strong enough to prove $\Omega$, but not strong enough to prove the $BPI$. I then looked for a way to make "general choice principle" precise in order to look for an answer. So here are my questions :

Is there a satisfactory way to make the notion of "general choice principle" precise, i.e. characterize certain sentences such that $AC$, $BPI$, etc. fall under this scope, but $\Omega$ doesn't ?

If there is, let $\phi$ be such a general choice principle. Can we have $ZF + \phi \vdash \Omega$, but $ZF + \phi$ doesn't prove $BPI$ ?

I first thought of defining it as "$ZF^{-}+ \phi$ ($ZF^{-}$ being $ZF$ minus the axiom of infinity) does not prove the axiom of infinity", but 1.it was only for this particular example, 2. it didn't work, as "$Inf \implies \Omega$" would fall under this notion, but clearly wouldn't be satisfcatory, so a solution would have to be more clever than that.


2 Answers 2


It seems difficult to make the notion of "general choice principle" precise, but I would guess that the principle "every infinite set admits a nonprincipal ultrafilter" would qualify. If so, then it answers your second question. It obviously implies the existence of a nonprincipal ultrafilter on $\omega$, and, unless I'm making a stupid mistake, it doesn't imply BPI because it follows from the conjunction of "there is a nonprincipal ultrafilter on $\omega$" and countable choice. (The point is that countable choice implies that there are no infinite, Dedekind-finite sets, so every infinite set includes a copy of $\omega$.)

  • $\begingroup$ (You're not making a stupid mistake.) $\endgroup$
    – Asaf Karagila
    Dec 29, 2016 at 15:39
  • $\begingroup$ This is indeed what I would call a "general choice principle" (there is no mention of $\omega$, etc.). So this establishes that (with really big quotation marks) "the existence of a nonprincipal ultrafilter over $\omega$, coming from a general choice principle is stricly weaker than $BPI$". Thanks for your answer ! This was mostly the part I was interested in (the other part was some form of generalization, that could be helpful to solve this) $\endgroup$
    – Maxtimax
    Dec 29, 2016 at 15:48
  • $\begingroup$ @Maxtimax: So countable choice or dependent choice are not considered "general choices principles" since they do mention $\omega$ in their formulation? $\endgroup$
    – Asaf Karagila
    Dec 31, 2016 at 9:49
  • $\begingroup$ Of course what I'm about to answer is not precise at all (that was also one of the goals of the question; as well as the other question you mentionned in your answer), but I would say they are because they allow to "choose" from various sets, sets that themselves need not have any connection with $\omega$. So they do mention $\omega$, but not in the same way as what I called $\Omega$. Although I kind of feel that countable choice wouldn't be a "general choice principle" because it's restricted to countable sequences, whereas dependent choice is about any relation. But it's not precise at all $\endgroup$
    – Maxtimax
    Dec 31, 2016 at 10:54

First of all, not only the existence of a free ultrafilter on $\omega$ is far weaker than $\sf BPI$, even the statement that every filter on $\omega$ can be extended to an ultrafilter on $\omega$ is strictly stronger than "There is a free ultrafilter on $\omega$". This can be an example for a principle that you're looking for: Every filter on $\omega$ can be extended to an ultrafilter. It is weaker than $\sf BPI$, and it proves the existence of a free ultrafilter on $\omega$.

Secondly, there is no "formal definition" of what are choice principles. See What is a Choice Principle, really? for two possible answers, which do not agree at all on the meaning of a choice principle.

Finally, there is a notion of strength when it comes to statement provable from $\sf ZFC$ and not from $\sf ZF$ (or even just "consistent with $\sf ZFC$", like $V=L$ or $\sf GCH$ which imply choice but are not equivalent to it). This is the question whether or not one statement implies the other. So $\sf BPI$ implies $\sf BPI(\omega)$ which implies the existence of a free ultrafilter on $\omega$, and none of these can be reversed. So we have a good notion of being stronger: a stronger principle proves more.

Let me finish by saying that "Every countable family of non-empty sets admits a choice function" cannot prove, nor it is a consequence of $\sf BPI$, so if by a choice principle you mean something like "Every such and such family of sets admits a choice function", then in all likelihood you're expecting a failure; unless you allow something like "There is a choice function from every family which is used in the proof that a filter can be extended to an ultrafilter", or something like that.

  • $\begingroup$ Utlrafilter itself is a choice function of some sort, namely choice of an element within each pair of complementary sets. (And this choice function has to satisfy a couple of extra properties.) $\endgroup$ Mar 19, 2017 at 21:41
  • $\begingroup$ Sure, you could say that. But the important part is not the actual choice, but rather the fact that you have coherence between the choices. Saying that an ultrafilter is a choice function is like saying that the Mona Lisa is a drawing, which just happened to have some extra properties. Sure, it's technically correct, but it also sort of misses the point. $\endgroup$
    – Asaf Karagila
    Mar 19, 2017 at 21:43
  • $\begingroup$ You mean that ML has some extra properties and oh by the way it happens to be a drawing somewhere? $\endgroup$ Mar 21, 2017 at 21:30
  • 2
    $\begingroup$ You apparently don't like your own description of ML as "a drawing, which just happened to have some extra properties" (wording: "some associated artistic and social ideas" would be more adequate as ML is not a scientific term). So I took the liberty to permute the terms, expecting maybe you would like it better. But generally there is an old practice, well-rooted in human culture, to begin a description of something with a generic part and end with a special part (like: homo sapiens, painting known as ML, et cetera). $\endgroup$ Mar 22, 2017 at 2:40
  • 1
    $\begingroup$ (continuation) Sometimes it takes some effort though to properly figure out the generic affiliation, as I did with the qualiification of ultrafilters as choice functions of certain kind. $\endgroup$ Mar 22, 2017 at 2:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.