What's so great about Kuratowski pairs?
I'm currently writing about countable order types — kinds of orderings of the naturals — and for various reasons want to do as much as I can without explicit talk about sets. But I obviously need to talk about ordered pairs of numbers so I can talk about "adding" and "multiplying" order types. So the trick I adopt is to code up the pair m, n by using the code 2m3n as 'pair numbers' (I also use the same powers-of-primes trick to code up finite sequences.)
It's been put to me that the fact I have resort to cheap tricks like this simply shows that there is something really a bit perverse about trying to do without sets. But not so, and perhaps it is worth explaining why (though I guess the following line of thought ought to be familiar — so don't expect novelties).
Why should handling ordered pairs of natural numbers be regarded as somehow inferior to other, albeit more familiar, devices? It might be said that (i) a single pair-number is really neither ordered nor a twosome; (ii) while a number m is a is one of the pair m, n, a number can't be a one of (or a member of) a pair-number; and in any case (iii) our construction was pretty arbitrary (we could equally well have used, e.g., 17m16n to code the pair).
Which is all true. But of course we can lay exactly analogous complaints against e.g. the Kuratowski definition that we all know and love, which treats the ordered pair (m, n) as the set {{m}, {m, n}}. For (i) that set is not intrinsically ordered (it is a set!), nor is it always two-membered (consider the case where m = n). (ii) Even when it is a twosome, its members are not the members of the pair: in standard set theories, m cannot be a member of {{m}, {m, n}}. And (iii) the construction again involves pretty arbitrary choices: thus {{n}, {m, n}} or {{{m}}, {{m, n}}} etc., etc., would have done just as well. Thus far, then, our coding of pairs of numbers using pair-numbers involves no worse a trick than coding them using Kuratowski's gadget.
To pursue the point further, suppose you are working in standard ZF (for example). Pure ZF knows only about pure sets. So to get natural numbers into the story — and hence to get Kuratowski pair-sets of natural numbers — you have to choose some omega-sequence of sets to implement the numbers, (or 'stand proxy' for them, 'simulate' them, 'play the role' of numbers, or even 'define' then — whatever your favourite way of describing the situation is). But someone who has opted to treat natural numbers within set theory by selecting some convenient sets to implement them is hardly in a position to complain about our opting to treat ordered pairs in arithmetic by choosing some convenient numbers to implement them. Suppose, alternatively, that you are working in ZFU, a set theory with urelements (so that at the base level of the iterative hierarchy you have not just the empty set but sets whose members are non-sets). You can then allow natural numbers as sui generis urelements, and then form Kuratowski pair-sets of these elements. But given the prior commitment to numbers as urelements and the availability of the numerical coding device, you don't need the extra commitment to sets to do the job of representing pairs of numbers. So why shoulder the extra ontological load?
You might respond that the Kuratowski trick at least has the virtue of being an all-purpose way of getting pairs of anything, while you can only use the powers-of-primes trick for coding pairs of numbers. But that's like saying that you can use sledgehammers to crack all sorts of things, while you can only use nutcrackers for nuts; true, but not to the point if it happens to be nuts you currently want to crack.
Putting the point more abstractly, suppose X are some objects and Y are some objects (maybe the same, maybe different). Then we'll say a pairing scheme for X with Y comprises some 'pair objects' O and a surjective two place function pr from X and Y to O, such that there are functions fst from O to X, and snd from O to Y which 'recover' the first and second items in a pair object in the obvious way. Set things up right, and pairing schemes are unique up to unique isomorphism. Then the 'pair numbers' trick and Kuratowski trick (with their obvious respective functions pr, fst, snd) yield equally good pairing schemes.
This is a common situation: we specify conditions on some mathematical gadgets, and then show that we have pinned them down uniquely up to unique isomorphism. And doing this, i.e. fixing gadgets up to unique structure-preserving bijection, arguably gives us all that we will normally need for mathematical purposes.
To take the case of natural numbers as the trite example, we might reasonably suppose that for mathematical purposes one omega-sequence is as good as another for implementing the numbers. True, to continue with the example, we will only be concerned as number-theorists with what holds true of numbers irrespective of the implementation. So, following Dedekind, perhaps we might insist that we should go on to abstract from the various 'concrete' implementations within richer frameworks (implementations which give numbers unwanted extraneous properties), and say that numbers themselves are distinguished as having no properties other than those they get in virtue of their place in the structure shared by different more concrete implementations. Well say that if you like: it doesn't matter one way or the other for most purposes.
And my present point is that we can be equally insouciant about what pairs of numbers 'really' are. My preferred scheme for pairing numbers and the more familiar set-theoretic scheme are just as good as each other, in that both 'define' pairs equally well for mathematical purposes. But if you do remain minded to go on in a Dedekinian spirit to say that, still, neither our pair-numbers nor the Kuratowski sets are really pairs (because both schemes treat pairs as objects with too much irrelevant structure), and we ought really to abstract away from these concrete implementations, then so be it. But that certainly doesn't spoil the symmetry and make my way of treating pairs of numbers any worse that the familiar standard way.
