Math Studies Algebra:

Axiom of Choice∗.
(Version 1d: 30 September 2021)

About the Axiom of Choice

The usual foundation of mathematics is set theory. Every mathematical object is a set.
For example, earlier in the course we defined a group as a set equipped with a certain
kind of a binary operation. The most popular axioms for the set theory are those of
Zermelo and Fraenkel. We will not discuss these axioms here, and retain the informal
approach.
One of the axioms of set theory, the Axiom of Choice, is less self-evident than the
rest. It had been speculated that it might be possible to derive it from the remaining
axioms. However, it has been proved that the axiom can neither be proved nor refuted
from the remaining axioms.
These notes discuss the Axiom of Choice and two statements that are equivalent to
it, Zorn’s lemma and the well-ordering principle.

Axiom of Choice Informally, the axiom of choice says that it is possible to choose an el-
ement from every set. Formally, a choice function on a set X is a function f : 2X \{∅} → X
such that f (S) ∈ S for every non-empty S ⊂ X. The Axiom of Choice asserts that on
every set there is a choice function.

Zorn’s Lemma A partially ordered set (poset) is a set whose elements can be compared,
but not every pair of elements is comparable. Formally, a partially ordered set is a set P
together with a binary relation 4 satisfying

1. x 4 x,

2. x 4 y and y 4 z imply x 4 z,

3. x 4 y and y 4 x imply x = y.
∗
These notes are available from the course webpage, and directly from http://www.borisbukh.org/
MathStudiesAlgebra1718/notes_ac.pdf

1
21-237: Math Studies Algebra Axiom of Choice notes

If x 4 y and x 6= y, then we write x ≺ y and say that x is smaller than y. It might

happen, for two elements x and y, that neither of them is smaller than the other. In that
case we say that x and y are incomparable.
An element x is called maximal if in P there is no larger element. It is called maximum
if it is larger than all the other elements. Minimal and minimum elements are defined
similarly.
Example 1: In Z endowed with the usual ordering, every two elements are compa-
rable. However, there are no minimal and no maximal elements.
Example 2: The set Z+ = {0, 1, 2, . . . } has the minimum element 0, but no maximal
elements.
Example 3: Consider Z+ ∪ {ω} = {0, 1, 2, . . . , ω} where ω is a new element that is
greater than each element of Z+ . In this poset ω is the maximum.
Example 4: On any set one can define trivial partial order in which no two distinct
elements are comparable. In such a poset, every element is both minimal and maximal.
There are no maximum and minimal elements (if the poset contains at least two elements).
Example 5: If X is any set, then the relation ⊆ makes 2X = {S : S ⊆ X} into a
poset. Here, the empty set is the minimum, and X is the maximum.
A chain is a set C all of whose elements are comparable to each other. Note that a
subset of chain is also a chain. The Examples 1–3 are chains.
We say that an element u is an upper bound for a chain C if x 4 u for each x ∈ C.
Zorn’s lemma asserts that if P is a non-empty poset in which each chain has an upper
bound, then P has a maximal element.

Well-ordering principle A poset P is called well-ordered if it is a chain, and every

non-empty subset S ⊂ P has a minimum. The well-ordering principle asserts that every
set can be well-ordered by a suitable relation.

Equivalence of Axiom of Choice, Zorn’s Lemma and the well-

ordering principle
Zorn’s lemma implies Axiom of Choice Let X be any non-empty set. Aided by
Zorn’s lemma, we will construct a choice function on X. Consider pairs (Y, f ) consisting
of a subset Y ⊆ X and a choice function f on Y . We introduce a partial order on the
set of all such pairs by defining (Y, f ) 4 (Y 0 , f 0 ) whenever Y ⊆ Y 0 and f = f 0 |Y .
The poset is non-empty because for every x ∈ X, there is an (obvious) partial choice
S
function on {x}. If C is a chain in this poset, then we can define Y = (Y,f )∈C Y and
f (S) = f (S) for any S such that f is defined on S. Then (Y , f ) is an upper bound for C.
Hence, by Zorn’s lemma there is some maximal element, which we call (f, Y ). If
x ∈ X \ Y , then we can extend f from Y to Y ∪ {x} by defining f (S) to be equal to x for
any S containing x. This contradicts maximality, and so X \ Y = ∅, and so f is a choice
function for X.

2
21-237: Math Studies Algebra Axiom of Choice notes

Zorn’s lemma implies well-ordering principle We may assume that the set X is
non-empty, for the empty set is trivially well-ordered.
An initial segment of a chain C is subchain C 0 such that x ∈ C, y ∈ C 0 and x ≺ y
imply that x ∈ C 0 .
Consider pairs (Y, ≤Y ), consisting of a subset Y ⊆ X and a well-ordering ≤Y on Y .
We define a partial order on the set of all such pairs in the similar manner to the preceding
proof. Namely, (Y, ≤Y ) 4 (Y 0 , ≤Y 0 ) whenever Y ⊆ Y 0 , the set Y is an initial segment of
Y 0 in ≤Y 0 , and the two orderings ≤Y and ≤Y 0 agree on the set Y .
Since X is non-empty, the poset is non-empty. Furthermore, if C is a chain in this
S
poset, we can define Y = (Y,f )∈C Y and x ≤Y y whenever x ≤Y for some (Y, ≤Y ) ∈ C.
Then ≤Y is a well-ordering on Y . Indeed, suppose that a set S ⊆ Y is non-empty and
def
(Y, ≤Y ) ∈ C is any pair in the chain such that S ∩ Y 6= ∅. Let u = min≤Y (S ∩ Y ), where
the minimum is with respect to ≤Y . Then u is a minimum for S with respect to ≤Y ,
for if s ∈ S is arbitrary, then either s ∈ Y in which case u ≤Y s follows from u ≤Y s,
or s ∈/ Y , in which case u ≤Y s follows from the fact that Y is an initial segment of Y .
Hence, the pair (Y , ≤Y ) is an upper bound for C.
So, by Zorn’s lemma the poset contains a maximal element (Y, ≤Y ). If Y 6= X and
x ∈ X \ Y , then we can extend (Y, ≤Y ) to a set Y ∪ {x} by defining x to be greater
than every element of Y . This contradicts maximality, and so Y = X, i.e., X can be
well-ordered.

Well-ordering principle implies Axiom of Choice Suppose X is a set, and ≤ is a

well-ordering of X. Then f (S) = min S defines a choice function on X.

Axiom of Choice implies Zorn’s Lemma (intuition for the proof ) Let P be any
non-empty poset such that every chain has an upper bound. Assume for contradiction’s
sake that P has no maximal element. Pick an element x0 from P (using the choice
function). Since x0 is not maximal, there is some x1 such that x0 ≺ x1 . Again, x1 is
not maximal, and so there is some x2 that is greater. We thus obtain a chain x0 ≺
x1 ≺ x2 ≺ . . . . By the assumption, this chain admits an upper bound xω . Since xω is
not maximal, there is xω+1 that is greater than xω . Keep on going to obtain the chain
x0 ≺ x1 ≺ x2 ≺ · · · xω ≺ xω+1 ≺ · · · . There is an upper bound, xω+ω , etc. Then, if we
believe in magic, then we can argue that this way we can build chains “longer” than P .
The contradiction shows that maximal elements do exist after all. This magic does exist,
and is called transfinite induction. Our formal proof below avoids formal discussion
of transfinite discussion. Instead it uses a shortcut (suggested to me by Prof. James
Cummings).

Axiom of Choice implies Zorn’s Lemma (formal proof ) Let P be any non-empty
poset such that every chain has an upper bound. Assume for contradiction’s sake that
def
P has no maximal element. Let f be a choice function on P , and let x0 = f (P ). If C is
def
chain, let Upp(C) = {u 6∈ C : ∀x ∈ C, x ≺ u} be set of all strict upper bounds for C.
Lemma 1. For any chain C, the set Upp(C) is non-empty.

3
21-237: Math Studies Algebra Axiom of Choice notes

Proof. Let u be an upper bound for C (which exists by the assumption on P ). If C has
no maximum element, then u 6∈ C, and so u ∈ Upp(C). Suppose next that C contain a
maximum element, which we call m. Since P has no maximal element, there is u that is
greater than m. Then x 4 m ≺ u for each x ∈ C, and so u ∈ Upp(C).

For any chain C, let

def

g(C) = f Upp(C) .
We finally come to the point where we can formalize our informal attempts to build a
chain x0 , x1 , x2 , . . . , xω , xω+1 , . . . . For purpose of this proof, an attempt is a well-ordered
set A ⊂ P satisfying the following:

1. min A = x0 ,

2. For every proper initial segment C ⊂ A, we have min A \ C = g(C).

Lemma 2. If A and A0 are two attempts, then either A ⊆ A0 or A0 ⊆ A.

Proof. Suppose the opposite, and let z = min A \ A0 and z 0 = min A0 \ A. These are
well-defined since A and A0 are well-ordered, respectively.
Since z 6= z 0 , we cannot have both z 4 z 0 and z 0 4 z. Without loss of generality,
suppose z 0 64 z. Let C = {x ∈ A : x ≺ z}. From the definition of z it follows that C ⊆ A0 .
It is clear that z = min A \ C, and so z = g(C).
If C = A0 , then A0 ⊂ A, and we are done. So, suppose that C 6= A0 . If z 0 4 x for some
x ∈ C, then transitivity would have implied that z 0 ≺ z, contrary to our assumption. So,
since A0 is chain, x 4 z 0 for every x ∈ C. Therefore C is a proper initial segment of A0 ,
and so g(C) ∈ A0 . However, g(C) = z 6∈ A0 . The contradiction completes the proof.

A consequence of the preceding lemma is that union of any set of attempts is an

def S
attempt. So, let A be the set of all attempts, and put A = A∈A A. Then A is an
attempt. However, A ∪ {g(A)} is an attempt that contains A. The contradiction shows
P does have a maximal element after all.