The Universe of Disco

The nonassociativity of the cartesian product

!!\def\pr#1#2{\langle{#1},{#2}\rangle}!!

Set theory doesn't include the ordered pair as a primitive type; ordered pairs have to be represented as sets somehow. This caused technical problems in Principia Mathematica, as I explained a couple of years ago:

In 1910 … a crucial piece of machinery was missing: the ordered pair.

This technical problem was solved the following year, by Norbert Wiener. We usually use the simpler definition invented later by Kazimierz Kuratowski:

!!\def\kp#1#2{\{\{{#1}\}, \{{#1},{#2}\}\}}!!

$$\pr ab = \kp ab$$

This works well enough (and it has worked well enough that we have used it for a hundred years) but it has some warts. We define the cartesian product of two sets like this: $$A\times B = \{\pr ab \mid a\in A\text{ and }b\in B\}$$

With Kuratwoski pairs, and this definition of product, it is not the case that $$A\times(B\times C) = (A\times B)\times C\qquad\color{darkred}{Not!}$$ because on the left side the elements look like $$\pr a{\pr bc} = \kp a{\kp bc}$$ and on the right side the corresponding element looks like $$\pr{\pr ab}c = \kp{\kp ab}c.$$

This doesn't present any serious technical problems; see Henning Makholm's discussion here. He says:

The nitty-gritty of getting such details to work completely formally is not generally worth the trouble.

Even with highly fussy technical details like this, you can almost always find some mathematician who said “Ah, but you have to be a little bit careful, because…”. As far as I know, this matter is an exception. If anyone thinks there is something interesting going on here that deserves further examination, I have not heard of it. Everyone ignores the matter.

Category theory takes the high road away from the issue, breezily dismissing it with a remark about how !!A\times(B\times C)!! and !!(A\times B)\times C!! are isomorphic, and equal up to unique isomorphism. This is the right answer, because as Wiener said in 1911, the structure of the pairs as actual sets is “largely a matter of choice”, and the whole point of category theory is to ignore this sort of irrelevant distraction, focusing on the API rather than on the internal implementation.

But last week I wondered: what if you do actually want !!A\times(B\times C)!! be the exact same set as !!(A\times B)\times C!!, and not just naturally isomorphic? Is there some clever way to define ordered pairs so that this happens?

I got the question all written up for Math Stack Exchange, when I realized the answer: No.

Consider the special case where !!A = \{a\}, B=\{b\},!! and !!C = \{c\}!!. Then !!A\times(B\times C) = (A\times B)\times C!! implies $$\pr a{\pr bc} = \pr{\pr ab}c.$$

But this violates the one property that we absolutely require of ordered pairs, which is that !!\pr pq = \pr rs!! if and only if !!p=r!! and !!q=s!!. To have !!\pr a{\pr bc} = \pr{\pr ab}c!! would imply !!a=\pr ab!!, which would imply !!\pr a{b'} = a = \pr ab!! even when !!b'\ne b!!.

Oh well, it would have been fun.

[Other articles in category /math] permanent link