Cardinality, The Well Ordering Principle

The Well Ordering Principle

Does every set have a well defined size? Can every set be mapped onto a cardinal? Can the elements of every set be well ordered? The "well ordering principle" says yes, but it really depends on the axiom of choice. In fact the well ordering principle and the axiom of choice are equivalent. Either implies the other.

First assume the well ordering principle holds, hence every set can be well ordered. We want to prove the axiom of choice.

Consider an indexed collection of sets Ax for every x in S. (Here S is the indexing set.) Assume each Ax is nonempty. Let U be the union over Ax and let P be the product set. Let w be any well ordering on U, which exists by the well ordering principle. Write a formula that pairs each x with the least element in Ax, according to w. Turn this formula into a set by comprehension, and realize that this set belongs to P. Thus P is nonempty, and the axiom of choice is true.

Conversely, assume the axiom of choice and let S be a set that cannot be well ordered. Let W be the set of subsets of S that can be well ordered. Verify that W is indeed a set, an application of the formula "can be well ordered" applied to the power set of S. Note that S ∉ W, since S cannot be well ordered.

Let f be any function from W into S. In other words, the well ordered subsets of S are mapped back into S. We wish to show there is some subset q ∈ W such that f(q) ∈ q.

If S is the integers, such a function f might start out looking like this. Each subset maps to a point not in the subset, for a while, but eventually this must fail.

(2,5,7) → 13
(5,6,9,23) → 4
(6) → 19
(even numbers) → 17
(prime numbers) → 77
…
q = (4,37,93,105,433,629,1383,…) → 93

Let f be any function as described above. Let S0 = f(∅). Let S1 = f(S0), S2 = f({S0,S1}), S3 = f({S0,S1,S2}), and so on. We are defining a new function e(ordinals) recursively. At each step the value of e() is determined by the value of f on a well ordered subset of S. At some point e is applied to an ordinal that will not fit into S, hence e is not a 1-1 function. Let b be the least ordinal such that e(b) = e(c) for some c < b. Let q be the set of points S0 S1 S2 .. up to but not including Sb. Now f(q) = Sc, which is contained in q. We have produced a subset q such that f(q) belongs to q.

Let P be the direct product of the power set of S, without the empty set and without S itself. Imagine selecting an element from each subset. In other words, a choice function maps each subset to an element in that subset. Let c() be such a function. We know c() exists, thanks to the axiom of choice.

Build a new function f(), on the subsets of S, such that f(x) is c applied to the complement of x. Now f maps every subset to an element outside of that subset. Bring in an ordered pair that maps the empty set to something in S. Now the domain of f is all the proper subsets of S, and f never maps a subset to any element of that subset. Our previous collection W, well ordered subsets of S, does not include S, and is part of the domain of f. Yet we showed that such a function cannot exist. There will always be some f(q) ∈ q. Therefore S can be well ordered.

The axiom of choice is equivalent to the well ordering principle, which asserts a well ordering on every set. Either implies the other.

Given the well ordering principle, every set is represented by a cardinal. Any two sets can be compared, and the "sizes" of sets are linearly ordered. In fact they are well ordered, according to the cardinals.

As a corollary, a function from S onto T implies |S| ≥ |T|. Take the inverse of f, selecting the least element in each preimage, and build a 1-1 map from T into S.

If separate functions map S onto T, and T onto S, then two injections embed T in S, and S in T. Use the Schroder Bernstein theorem to produce a bijection between S and T, hence they have the same cardinality. But again, this is only valid if the axiom of choice holds.