The axiom of choice, by which I mean that every collection of non-empty sets has a choice function, is usually an axiom most working mathematicians accept without further thought. But in set theory we usually get ourselves into situations where we simply cannot have (full) choice — most notably in determinacy scenarios, giving rise to several weakened forms of choice. might seem like an isolated axiom without much direct connection to other axioms, as we usually simply assume choice and get on with our day. But choice is in fact *implied* by the generalised continuum hypothesis , which can then also be seen as a choice principle, and choice even *forces* us to work in classical logic.

First of all, let’s recall a few well-known weakened versions of . Firstly there’s the Axiom of Dependent Choices , stating that whenever R is a binary relation satisfying that , there’s an -sequence such that for every . Secondly we got the Axiom of Countable Choice , saying that every non-empty countable set has a choice function. It’s not too hard to show that , but these have further been shown to *not* be reversible. Solovay built his famous Solovay Model in 1970, which in particular satisfies and *not* , separating the two. As for the separation of and , this is accomplished by building a certain symmetric extension — see Theorem 8.12 in Jech’s “The Axiom of Choice” for more details.

is already a powerful axiom, as is perhaps manifested in the theorem stating that it actually implies *classical* logic, i.e. the law of excluded middle , a result which was proven by Diaconescu. Curiously, this does indeed require full choice, as it’s still consistent to work intuitionistically with weakenings such as .

Theorem (Diaconescu ’75)..

If we now choose to strengthen the principles even further, we for instance have Global Choice, which is a class principle saying that there’s a class function such that for every non-empty set — a sort of a “uniform choice”. As is shown in Jech’s chapter on HOD, this turns out to be equivalent to . A natural way to strengthen this even further then is to consider the axiom , which can then be seen as a (very strong) choice principle. Along the way to it seems we’ve skipped an axiom however, as in particular implies , which a surprising theorem of Sierpiński’s implies !

Theorem (Sierpiński ’45)..

A proof of this theorem can also be found in Paul Cohen’s “Set Theory and the Continuum Hypothesis”. It would then seem reasonable that should also imply , but this is *false*, as shown in Theorems 3.2 and 3.6 of Truss (’73). We *can* get some of the way however: if we assume both and then we get that the reals can be well-ordered, a result which Caicedo attributes to Specker. Whether is enough to get this conclusion is open (at least it was back in 2013, again according to Caicedo).

Okay, so that was quite a handful of axioms. Note that *all* of these choice principles are equiconsistent (modulo ), simply because is equiconsistent with . The interesting connections between the choice principles are therefore the direct implications — here’s an overview:

As mentioned above, we cannot reverse the implications . This still remains true for all the other implications:

- McAloon (’70) shows that is consistent;
- Hamkins-Reitz-Woodin (’08) contains a proof that is consistent. This is
*not*the original source however, as it’s attributed to folk lore.

This shows that and are independent of each other (modulo ), which then also implies that the implications , and are irreversible.

We could modify the above diagram by replacing with any fine-structural extender model, as these satisfy both and . Of course we weaken the axiom by doing this, so another way to phrase the inner model programme is then that we (in particular) want to find the weakest such axiom implying both and .

Another curious recent development is concerned with the so-called *choiceless large cardinals*, which are large cardinals such as Reinhardt cardinals and beyond. All of these are inconsistent with by the Kunen inconsistency, but Koellner has even shown that a so-called Berkeley cardinal of countable cofinality is inconsistent with ! Whether these cardinals can actually exist is a different story.

*I want to thank Mohammad Golshani for bringing my attention to several of the results I’ve used here, and Asaf Karagila for correcting my statement that whether implies is open — it’s false, by the above reference to Truss (’73).*