The Zermelo-Fraenkel axioms for set theory are more or less uncontroversial. However, there is another axiom, the axiom of choice, which is more controversial; it is therefore usually segregated from the others. When the Zermelo-Fraenkel axioms are accepted along with the axiom of choice, the whole axiom system is often called “ZFC” (for “Zermelo-Fraenkel with Choice”).

Axiom (Axiom of choice)   Let be a set of nonempty sets. Then there exists a function

such that for all .

For finite sets , a choice function can be constructed without appealing to the axiom of choice. It is only for infinite (and usually uncountable) sets that the existence of a choice function becomes an issue. Here one can see why it is not considered “obvious” and always taken for an axiom by everyone: one really cannot imagine any process which makes uncountably many selections without also imagining some rule for making the selections. Given such a rule, the axiom of choice is not needed. Thus, objects that are proved to exist using the axiom of choice cannot generally be described by any kind of systematic rule, for if they could it would not be necessary to their construction.

Let us consider a couple of examples. Imagine that there are infinitely many pairs of shoes (each consisting of one left shoe and one right shoe). Let denote the set of all pairs of shoes. In this scenario, it can be verified that the function defined by the left shoe of  is a choice function. Similarly, imagine that there are infinitely many pairs of socks. Let denote the set of all pairs of socks. In this scenario, one cannot assume that a function exists without appealing to the axiom of choice (or something equivalent to it). Note that this scenario cannot be resolved in the same manner as the previous scenario because most people do not differentiate between a "left sock" and a "right sock".

Strange objects that can be constructed using the axiom of choice include non-measurable sets (leading to the Hausdorff and Banach-Tarski paradoxa), and Hamel bases for any vector space. A Hamel basis may not seem strange, but try to imagine a set of continuous functions such that every continuous function can be expressed uniquely as a linear combination of finitely many elements of . Since in fact the existence of a basis for every vector space is equivalent to the axiom of choice, it is almost guaranteed that no such set can ever be described. It is for this reason that some mathematicians dislike the axiom of choice.

On the other hand, many very useful facts can be proven using the axiom of choice. For example, the fact that every vector space has a basis, every ring with identity element has a maximal ideal and many other algebraic theorems which are difficult or impossible to prove without using the axiom of choice.

In pure set theory, the axiom of choice is only relevant where most people's intuition more or less breaks down, when dealing with hierarchies of uncountable infinities.

The relevance of the axiom of choice to various branches of mathematics has led to a detailed study of its truth. It turns out that if the Zermelo-Fraenkel axioms are consistent, then they remain consistent upon adding the axiom of choice. But they also remain consistent upon adding the negation of the axiom of choice (see [G] and [C]).

Some mathematicians have suggested an axiom that would result in all subsets of the real numbers being measurable; this would of course imply the negation of the axiom of choice.

There are many alternative formulations of the axiom of choice, although it is not always trivial to prove equivalence. These include:

• The generalised Cartesian product of a non-empty family is the empty set if and only if there is a such that is the empty set. Due to this equivalence the axiom of choice is sometimes called the multiplicative axiom.
• A product of compact topological spaces is compact with respect to the product topology (Tychonoff's theorem)
• Every relation is a union of functions with the same domain as .
• Every ring with identity element has a maximal ideal.
• Every tree has a branch.
• Every vector space has a basis.
• Given any two sets and , there exists an injective mapping or .
• Hausdorff's maximum principle (Proof).
• König's theorem.
• Kuratowski's lemma.
• Tukey's lemma.
• Zermelo's postulate.
• Zermelo's well-ordering theorem (Proof).
• Zorn's lemma (Proof).

Figure  shows how these equivalences are proven on PlanetMath.

Figure: Structure of the equivalence proofs on PlanetMath. The abbreviations are explained in table . An arrow means that “ implies ” is proven on PlanetMath.

Table 1: Abbreviations

AC Axiom of Choice Hamel Every vector space has a basis Hausdorff Hausdorff's maximum principle König König's theorem Krull Every ring with identity element has a maximal ideal Tukey Tukey's lemma Well Zermelo's well-ordering theorem Zorn Zorn's lemma

Bibliography

C

P. J. COHEN, The independence of the continuum hypothesis. I, II, Proc. Natl. Acad. Sci. USA 50, 1143-1148 (1963); 51, 105-110 (1964).

G

K. GöDEL, The consistency of the axiom of choice and of the generalized continuum-hypothesis, Proc. Natl. Acad. Sci. USA 24, 556-557 (1938).