14.

Naive set theory

14.1
14.2
14.3
14.4
14.5
14.6
14.7

Sets
Posets, ordinals
Transfinite induction
Finiteness, infiniteness
Comparison of infinities
Example: transfinite induction in Lagrange replacement
Equivalents of the Axiom of Choice

1. Sets
Naive definition: A set is an unordered collection of things (not counting multiplicities), its elements.
Write x S or S 3 x for an element x of S. Sets are described either as comma-separated lists (whose order
is not supposed to be significant)
S = {x1 , x2 , . . .}
or by a rule
S = {x : some condition on x is met}
The empty set is
= {}

[1.0.1] Theorem: There is no set S such that x S if and only if x 6 x.

Proof: Suppose there were such S. Then S S if and only if S 6 S, contradiction.

///

Extension Principle (Leibniz) Two sets are equal if and only if they have the same elements.

[1.0.2] Corollary: There is only one empty set .

///

Idea: Everything is a set.

A subset T of S is a set such that for all elements x of T also x is an element of S. Write T S or S T .
199

200

Naive set theory

A subset of S is proper if it is neither S itself nor . The union of a set F of sets is

[

S = {x : x S for some S F }

SF

The intersection is
\

S = {x : x S for all S F }

SF

We make an exception in the case of intersections over F for F = , since the defining condition would be
vacuous, and (supposedly) every set would be an element of that intersection, which is not viable. The union
and intersection of a finite number of sets can also be written, respectively, as
S1 . . . Sn
S1 . . . Sn
Proto-definition: The ordered pair construct (x, y) with first component x and second component y should
have the property that
(x, y) = (z, w) x = z and y = w

[1.0.3] Remark: As sets, taking (x, y) = {x, y} fails, since the elements of a set are not ordered. Taking
(x, y) = {x, {y}} fails, since it may be that x = {y}.
[1.0.4] Proposition: We can construct ordered pairs as sets by defining
(x, y) = {{x}, {x, y}}

Proof: We must prove that (x, y) = (z, w) if and only if the respective components are equal. One direction
of the implication is clear. For the other implication, from
{{x}, {x, y}} = {{z}, {z, w}}
{x} is either {z} or {z, w}, and {x, y} is either {z} or {z, w}. Treat cases, using the Extension Principle.
///
For finite n, define recursively ordered n-tuples by
(x1 , . . . , xn1 , xn ) = ((x1 , . . . , xn1 ), xn )

[1.0.5] Remark: Subsequently we ignore the internal details of the construction of ordered pair, and
only use its properties. This is a typical ruse.
The Cartesian product X Y of two sets X and Y is the set of ordered pairs
X Y = {(x, y) : x X, y Y }
A function or map f : X Y from X to Y is a subset of X Y such that for all x X there is a unique
y in Y such that (x, y) f . As usual, this is written f (x) = y or f x = y. The image f (X) of f is
f (X) = image of f = {f (x) : x X}

[1.0.6] Remark: This definition identifies a function with its graph, rather than by a formula or algorithm
by which to compute the function.

Garrett: Abstract Algebra

201

[1.0.7] Definition: A function f : X Y is surjective if for every y Y there is x X such that

f (x) = y. It is injective if f (x) = f (x0 ) implies x = x0 . If f is both surjective and injective is it bijective.
The composition f g of two functions f : Y Z and g : X Y is defined by
(f g)(x) = f (g(x))
A left inverse g (if it exists) to a function f : X Y is a function g : Y X such that g f = 1X ,
where 1X is the identity function on X, defined by 1X (x) = x for all x X. A right inverse g (if it
exists) to a function f : X Y is a function g : Y X such that f g = 1Y
Let F be a set of sets. A choice function f on F (if it exists) is any function
f : F

SF

such that
f (S) S
for all S in F . To postulate that at least one choice function exists for any set F of sets is a non-trivial thing,
and, roughly, is the Axiom of Choice. The collection of all choice functions on F is the direct product
of the sets, denoted
Y
S
SF

Again, to know that this is non-empty (for F infinite) requires something!

K. Godel and P. Cohen proved that the Axiom of Choice is not only not provable from other more mundane
axioms for sets, but is independent of them, in the sense that it is equally consistent to assume the negation
of the Axiom of Choice.
A relation R between sets X and Y is a subset of X Y . A (binary) relation on a set X is a subset of
X X. A relation R on X is
reflexive if (x, x) R for all x X
symmetric if (x, y) R implies (y, x) R for all x, y X
transitive if (x, y) R and (y, z) R implies (x, z) R An equivalence relation is a relation that
enjoys all three of these properties. For an equivalence relation R, the equivalence class of x is
equivalence class of x = {y X : (x, y) R}

202

Naive set theory

2. Posets, ordinals
A partial order on a set X is a relation R on X, written x y if (x, y) R, such that
(Reflexivity) x x for all x X
If x y and y x then x = y
(Transitivity) If x y and y z then x z Then X is a partially ordered set or poset. We may write
x < y if x y and x 6= y.
A partial ordering on X is a total ordering if for all x, y X either x y or y x.
A well ordering [sic] on a set X is a total ordering on X such that any non-empty subset Y of X has a
minimal element (also called least element). That is, there is an element y Y such that for all z Y
we have y z.

[2.0.1] Proposition: Let X be a well-ordered set. Let f : X X be an order-preserving injective

map (so x x0 implies f (x) f (x0 )). Then
f (x) x
for all x X.

Proof: Let Z be the subset of X consisting of elements x such that f (x) < x. If Z is non-empty, then it
has a least element x. Thus, on one hand, f (x) < x. On the other hand, f (x) 6 Z, so f (f (x)) > f (x). But,
since f preserves order and is injective, f (x) < x implies f (f (x)) < f (x), contradiction.
///

[2.0.2] Corollary: The only order-preserving bijection of a well-ordered set X to itself is the identity
map.

///

[2.0.3] Corollary: There is no order-preserving bijection of a well-ordered set X to a proper initial

segment
X <x = {y X : y < x}
of it for any x X.

///

[2.0.4] Example: The set

Z = {X <x = {y X : y < x} : x X}
of initial segments X <x of a well-ordered set X, with ordering
z w z w
has an order-preserving bijection to X by
X <x x
An ordinal is a well-ordered set X such for every element x X
x = X <x
That is, x is the set X <x = {y X : y < x} of its predecessors in X.

[2.0.5] Example: The empty set is an ordinal, since the defining condition is met vacuously. Let X be
an ordinal that is not the empty set. Then X (being non-empty) has a least element x. Since x is the union
of its predecessors, of which there are none, x = . So is the least element of every ordinal.

Garrett: Abstract Algebra

203

[2.0.6] Example: If X is an ordinal, and x X, then the initial segment below x

X <x = {y X : y < x}
is also an ordinal. Indeed, the well-ordering is preserved, and by transitivity the predecessors of y in X <x
are exactly the predecessors of y in X, so the defining property of ordinals holds.

[2.0.7] Example: If X is an ordinal, then Y = X {X}, with ordering

a b a b
is an ordinal, the successor of X. To see this, first note that, for all y Y we have y X, that is (by
definition of the ordering) y X. Thus, for y Y , if y 6= X, then y X and (since X is an ordinal) is the
set of its predecessors in X. And since y < X in Y , X is not among ys predecessors in Y , so y really is the
set of its predecessors in Y . And X is the set of its predecessors in Y .
///
Since everything is to be a set, following J. von Neumann, define the initial (finite) ordinals by
0 = = {}
1 = {0} = {, {}} = {{}, {{}}}
2 = {0, 1} = {, {}, {, {}}} = {{}, {{}}, {{}, {{}}}}
3 = {0, 1, 2} = {, {}, {, {}}, {, {}, {, {}}}}
= {{}, {{}}, {{}, {{}}}, {{}, {{}}, {{}, {{}}}}}
and so on
The set of natural numbers is [1]
= {0, 1, 2, . . .}
Define an order on by
x y x y
It is not at all immediate (with the present definition of the symbols) that is an ordinal.

[2.0.8] Proposition: If X and Y are ordinals and Y X then there is x X such that Y is the initial
segment
Y = {y X : y < x} = x

Proof: Let Z be the set of elements of X that are not in Y but are below some element of Y . The claim is
that Z is empty. If not, let z be the least element in Z. Let y Y be such that z < y. Since y is the set of
its predecessors in X, x y. But also y is the set of its predecessors in Y , so x y, contradiction.
///

[2.0.9] Theorem: Any two ordinals X, Y are comparable, in the sense that either X = Y , or X is an
initial segment of Y , or Y is an initial segment of X.

Proof: The intersection X Y is an ordinal, since for z X Y

{w X Y : w < z} = {x X : x < z} {y Y : y < z} = z z = z
Suppose that X is not contained in Y , and Y is not contained in X. From above, X Y is an initial
segment
X Y = {z X : z < x} = x
[1] Shuddering at the casual formation of this ostensibly infinite set is reasonable, since its existence as a set is not
formally assured by the existence of the separate finite ordinals.

204

Naive set theory

in X for some x X, and also an initial segment

X Y = {w Y : z < y} = y
in Y for some y Y . But then x = y, contradiction.

///

[2.0.10] Corollary: Two ordinals admit an order-preserving bijection between them if and only if they
are identical, and in that case the only order-preserving bijection is the identity map.
Proof: We already saw that there is at most one order-preserving bijection between two well-ordered sets.
Thus, let X and Y be ordinals, and X =
6 Y . By the theorem, one is an initial segment of the other, so
assume without loss of generality that Y is an initial segment
Y = {y X : y < x}
for some x in X. Let f : X Y be an order-preserving bijection. We saw earlier that f (z) z for any
well-ordered sets in this situation. But then f (x) x, which is impossible.
///

[2.0.11] Corollary: The relation on ordinals defined by x < y if and only if x is an initial segment of y
is a total ordering.

///

[2.0.12] Corollary: Given an ordinal x, its successor ordinal y = x {x} has the property that x < y.
///

[2.0.13] Corollary: There is no largest ordinal.

///

[2.0.14] Theorem: The union of any set of ordinals is an ordinal.

Proof: Let F be a set of ordinals, and
E=

XF

is also a set of ordinals. Define a relation < on E by x < y if x is an initial segment in y, that is, is an
element of y. The transitivity of < follows (again) from the fact that every element of an ordinal is an
ordinal. The comparability of all ordinals (from above) says that this is a total ordering. To prove that < is
a well-ordering, let D be a non-empty subset of E, and let d be any element of D. If d is least in D, we are
done. If d is not least in D, then nevertheless c D with c < d are elements of d, since c < d only for c an
initial segment of d, that is an element of d. Since d is an ordinal, it is well-ordered, so
{c D : c < d} = D d
is well-ordered. Thus, D contains a least element. Finally, we must prove that any element e of E is the set
of its predecessors in E. Let X be an element of F such that e X. Since X is an ordinal, e is the set of its
predecessors d in X. Thus, all such predecessors d are elements of X, so are elements of the union E. Thus,
e = {d X : d < e} {d E : d < e}
On the other hand, for any d E, the definition of d < e is that d is an initial segment of e, that is, that
d e. In that case, d X for every ordinal containing e. That is, we have the opposite inclusion
e = {d X : d < e} {d E : d < e}
and e is exactly the set of its predecessors in the union E.

[2.0.15] Theorem: Every well-ordered set has an order-preserving bijection to exactly one ordinal.

///

Garrett: Abstract Algebra

205

Proof: First, let X be a well-ordered set with each initial segment

X <x = {y X : y < x}
for x X isomorphic [2] to an ordinal x . We claim that X is isomorphic to an ordinal. From above, since
no two distinct ordinals are isomorphic, and since an ordinal admits no non-trivial maps to itself, for each
x X the ordinal x is uniquely determined and the order-preserving map fx : x x is unique. We view
F : x x as an ordinal-valued function F on X.
Consider x < y in X. Since x and y are distinct initial segments of X, they are not isomorphic as ordered sets
(indeed, there is no order-preserving injection of y to x). Thus, F (x) = x is not isomorphic to F (y) = y .
Thus, since any two ordinals can be compared, either F (x) = x is an initial segment of F (y) = y or vice
versa. Unsurprisingly, if y < x then
y y x x
would give an isomorphism of y to a proper initial segment x, but (again) this is impossible. Thus, F is an
order-preserving bijection of X to a set = {x = F (x) : x X} of ordinals. Since = F (X) is the image
of the well-ordered set X, is well-ordered. To show that is an ordinal, by definition, we must show that
for the initial segment
< = { 0 : 0 < }
is equal to . Indeed, the hypothesis is exactly this, so is an ordinal, and X is an ordinal (being isomorphic
to ).
Now we prove the theorem. First we prove that every element of a (non-empty) well-ordered set X is
isomorphic to an ordinal. The least element of X is isomorphic to the ordinal . Given in X with
isomorphic to an ordinal for all < , then apply the claim to (in place of X) to conclude that is
isomorphic to an ordinal. And then the claim implies that X is isomorphic to an ordinal. Since two distinct
ordinals are not isomorphic, there is exactly one ordinal to which X is isomorphic.
///
The following corollary is sometimes recast as a paradox:

[2.0.16] Corollary: (Burali-Forti) The collection of all ordinals is not a set.

Proof: Suppose the collection F of all ordinals were a set. Then (by the theorem) the union
E=

SF

would be an ordinal. Thus, E would be an element of itself, contradiction.

///

3. Transfinite induction
[3.0.1] Theorem: Let P () be a property that may or may not hold of ordinals . Suppose that for any
ordinal if P () for all ordinals < then P () holds. The P () holds for all ordinals .

Proof: Let = {}, so is an ordinal containing . Then we can do induction on the set : prove
that P () holds for all (including ). If P () failed for some in , then there would be a least in
for which it failed. But P () holds for all < , and the hypothesis assures that P () does hold, after all.
This contradiction shows that P () holds for all , in particular, for .
///
In some situations the induction step, namely, proving that P () holds if P () holds for all < , must be
broken into cases, depending on the nature of .
The initial ordinal, .
[2] As ordered set, of course.

206

Naive set theory

Successor ordinals S
= {} for some .
Limit ordinals = < .

[3.0.2] Remark: First, contrast the definition of limit ordinal with the property enjoyed by every ordinal,
namely
= { : } = { : < }
A successor ordinal is not a limit ordinal, since if = {} then all predecessors of are subsets of ,
and likewise their union, which cannot contain as an element.

[3.0.3] Proposition: Every ordinal is either the initial ordinal , a successor ordinal, or a limit ordinal.
Proof: Suppose is not and is not a successor. Let be the union of the predecessors of . Since a
union of ordinals is an ordinal, is an ordinal, and . If < then is among s predecessors, so
is in the union of predecessors, so is the largest among the predecessors of . The assumption < gives
{} . It cannot be that {} since otherwise {} would be a predecessor of , and thus
{}, which is false. So, then, the successor {} of is , contradiction to the hypothesis that
is not a successor. Thus, = .
///
Thus, we can rewrite the first theorem in a manner that refers explicitly to the types of ordinals: to prove a
property P () holds for all ordinals :
Prove P () holds.
Prove (for all ) that if P () holds then P ( {}) holds.
Prove for every limit ordinal that if P () holds for all < then P () holds.

Garrett: Abstract Algebra

207

4. Finiteness, infiniteness
A set S is Peano finite if there is some n such that there is a bijection of S to
n = {0, 1, 2, . . . , n 1}
The set is Peano infinite if it is not Peano finite.
A set S is Dedekind infinite if there is an injection from S to a proper subset of S. It is Dedekind finite
if it is not Dedekind infinite.

[4.0.1] Theorem: (Granting the Axiom of Choice) The two notions of infinite are the same.
[4.0.2] Remark: To avoid circularity, we should not presume arithmetic at this point.
Proof: Let f : S S be an injection of S to a proper subset of itself. Choose s1 S but not lying in the
image f (S). Claim f (f (S)) is a proper subset of f (S). Indeed, f (s1 ) cannot be in f (f (S)), or there would
be t f (S) such that f (t) = f (s1 ), and then by injectivity of f we would have t = s1 , contradicting the fact
that s1 6 f (S). Certainly f restricted to f (S) is still injective.
Thus, f (f (f (S))) is strictly smaller than f (f (S)) By induction, we can find s1 , s2 , . . . such that s1 6 f (S),
s2 f (S) but s2 6 f (f (S)), s3 f (f (S)) but s3 6 f (f (f (S))), etc. In particular, all these si are distinct,
so we have an injection
{1, 2, 3, . . .} S
Thus, Dedekind infinite implies Peano infinite.

///

5. Comparison of infinities
The Cantor-Schroeder-Bernstein Theorem proven here is the key result that allows comparison of
infinities. Perhaps it is the first serious theorem in set theory after Cantors diagonalization argument.
Apparently Cantor conjectured this result, and it was proven independently by F. Bernstein and E. Schr
oder
in the 1890s. The proof given below is a natural proof that one might find after sufficient experimentation
and reflection.
It is noteworthy that there is no invocation of the Axiom of Choice, since one can imagine that it would
have been needed.
The argument below is not the most succinct possible, but is intended to lend a greater sense of inevitability
to the conclusion than would the shortest possible version.

[5.0.1] Theorem: (Cantor-Schroeder-Bernstein) Let A and B be sets, with injections f : A B and

g : B A. Then there exists a canonical bijection F : A B.

Proof: Let
Ao = {a A : a 6 g(B)}

Bo = {b B : b 6 f (A)}

The sets
A2n = (g f )n (Ao )

are disjoint. Let A

A2n+1 = (g f )n g(Bo )
S
be the complement in A to the union n An . Define F by

(for a An , n 2Z)
f (a)
F (a) = g 1 (a) (for a An , n 1 + 2Z)

f (a)
(for a A )

208

Naive set theory

We must verify that this moderately clever apparent definition really gives a well-defined F , and that F is
a bijection. For n 1, let
Bn = f (An1 )
and also let B = f (A ).
The underlying fact is that AB (disjoint union) is partitioned into one-sided or two-sided maximal sequences
of elements that map to each other under f and g: we have three patterns. First, one may have
f

ao b1 a1 b2 a2 . . . bn an . . .
beginning with ao Ao , all ai A and bi B. Second, one may have
g

bo a1 b1 a2 b2 . . . an bn . . .
with bo Bo , and ai A and bi B. The third and last possibility is that none of the elements involved is
an image of Ao or Bo under any number of iterations of f g or g f . Such elements fit into pictures of the
form
g

. . . a2 b1 a1 bo ao b1 . . .
where ai A and bi B. The fundamental point is that any two distinct such sequences of elements are
disjoint. And any element certainly lies in such a sequence.
The one-sided sequences of the form
f

ao b1 a1 b2 a2 . . . bn an . . .
beginning with ao Ao , can be broken up to give part of the definition of F by
f

F : ao b1

F : a1 b2 . . .

The one-sided sequences of the form

g

bo a1 b1 a2 b2 . . . an bn . . .
with bo Bo , beginning with bo Bo , can be broken up to give another part of the definition of F
g

bo a1

b1 a2 . . .

which is to say
F :

g 1
a1 bo

F :

g 1
a2 b1

...

For a double-sided sequence,

g

. . . a2 b1 a1 bo ao b1 . . .
there are two equally simple ways to break it up, and we choose
f

F : ai bi+1
Since the sequences partition A B, and since every element of B (and A) appears, F is surely a bijection
from A to B.
///

Garrett: Abstract Algebra

209

6. Example: transfinite Lagrange replacement

Let V be a vector space over a field k. Let E = {e : A} be a set of linearly independent elements, and
F = {f : B} be a basis for V .

[6.0.1] Theorem: We have an inequality of cardinalities: |A| |B|.

Proof: Well order [3] A. We prove by transfinite induction that there is an injection j : A B such that
{e : A} {f : B, 6 j(A)}
is a basis for V . That is, we can exchange (following Lagrange) every element in E for a basis element in F
and still have a basis. Thus, since E injects to F we have an inequality of cardinalities.
Fix A. Let
A< = { A : < }
For the induction step, suppose that we have an injection
j : A< B
such that
{e : < } {f : 6 j(A< )}
is a disjoint union, and is still a basis for V . Then, since these elements span V , there exist elements a and
b in the field such that
X
X
e =
a e +
b f
<

6j(A< )

Since the es were linearly independent, not all the b s can be 0. Pick 6 j(A< ) such that b 6= 0, and
extend j by defining j() = .
We must check that
{e : } {f : 6 j(A )}
is still a basis (and that the union is disjoint). For linear independence, since
{e : < } {f : 6 j(A< )}
is a basis, any linear relation must properly involve e , as
e =

c e +

<

d f

6j(A )

Replace e by its expression

e =

X
<

a e +

b f

6j(A< )

to obtain
X
<

a e +

X
6j(A )

b f + b f =

X
<

c e +

d f

6j(A )

But b 6= 0, f occurs only on the left-hand side, and the vectors involved in this sum are a basis, so this is
impossible. This proves the linear independence (and disjointness of the union).
[3] To well-order a set is, in effect, an invocation of the Axiom of Choice, and should not be taken lightly, even if it
is useful or necessary. See the last section in this chapter.

210

Naive set theory

To prove the spanning property, use the fact that

{e : < } {f : 6 j(A< )}
is a basis. That is, given v V , there are field elements x and y such that
v=

x e + +

<

y f

6j(A< )

Since b 6= 0 above, we can express f in terms of e , by

f = b1
e

a e +

<

b f

6j(A )

Thus, we can replace f by this expression to express v as a linear combination of

{e : } {f : 6 j(A )}
proving the spanning. By transfinite induction there exists an injection of A to B.

///

[6.0.2] Remark: We could make the invocation of Well-Ordering more explicit: if there were no injection
A B as indicated, by Well-Ordering let be the first element in A such that there is no such injection
on A< . Then the same discussion yields a contradiction.
We use the Axiom of Choice in the guise of the Well-Ordering Principle: we assume that any set can be wellordered. From the theory of ordinals and well-orderings any well-ordered set is isomorphic (as well-ordered
set) to a unique ordinal. From the theory of ordinals, any two ordinals are comparable, in the sense that one
is an initial segment of the other. Thus, putting these things together, any two sets A, B are comparable in
size, in the sense that either A injects to B, or B injects to A.

7. Equivalents of the Axiom of Choice

There are several statements which are all logically equivalent to each other, and often used to prove existence
when only existence is required, and no object must be explicitly exhibited. These are Zorns Lemma,
Hausdorff Maximality Principle, Well-Ordering Principle, and Axiom of Choice. Here we describe
these assertions in the context of naive set theory, in the style of the discussion above, rather than formal
or axiomatic set theory. [4]
The Axiom of Choice or Zermelos postulate asserts that, given a set of sets
{Si : i I}
with (not necessarily mutually disjoint) non-empty sets Si (indexed by a set I), there exists a set of choices
si , one from each Si . That is, there exists a choice set
C = {si : i I}

with si Si for all indices i I

[4] In the late nineteenth and early twentieth centuries, it was unclear whether or not one could expect to prove these

assertions from first principles. Further, some mathematicians felt that one or more of these assertions was obviously
true, while others felt uneasy to varying degrees about invocation of them. In the early 1930s Kurt G
odel proved
that the Axiom of Choice is consistent (in the Zermelo-Frankel first-order axiomatization) with the other axioms of
set theory. In 1963, Paul Cohen proved that the Axiom of Choice was independent of the other axioms. In fact, G
odel
also proved that the Continuum Hypothesis is consistent. This is the hypothesis that there are no cardinals between
the countable and the cardinality of the reals. Cohen also proved that the Continuum Hypothesis is independent.

Garrett: Abstract Algebra

211

This is intuitively obvious for finite sets I, but less obviously clear for infinite sets of sets. Sometimes this is
stated in the form that there is a choice function f on the index set I such that f (i) Si .
The Well-ordering Principle asserts that every set can be well-ordered. More precisely, the assertion is that,
given a set S, there is a bijection of S to an ordinal.
To state Zorns lemma some preparation is needed. In a poset X, a chain is a totally ordered subset. An
upper bound for a totally ordered subset Y of a poset X is an element b X (not necessarily in Y ) such that
y b for all y Y . A maximal element m X is an element of X such that, for all x X, m x implies
m = x. Then Zorns lemma asserts that every poset in which every chain has an upper bound contains at
least one maximal element.
The Hausdorf maximality principle asserts that in any poset, every totally ordered subset is contained in
a maximal totally ordered subset. Here a maximal totally ordered subset is what it sounds like, namely, a
totally ordered subset such that any strictly larger subset fails to be totally ordered. A seemingly weaker,
but equivalent, form is the assertion that every poset contains a maximal totally ordered subset.
We give a representative proof.

Proof: (Axiom of Choice implies the Well-ordering Principle.) Fix a set X. Let c be a choice function on
the set of subsets of X. Try to define a function f on ordinals by transfinite induction, by


f () = c X {f () : ordinals < }
where for two sets X, A
X A = {x : x X, x 6 A}
This definition fails if-and-when
X {f () : ordinals < }

Let us show that each function so defined (as long as we have not run out of elements of X to hit) is injective.
Indeed, for ordinals > , consider the definition


f () = c X {f () : ordinals < }
The set of values removed from X to choose a value for f () includes f (), so necessarily f () 6= f (). If
at any point


X {f () : ordinals < } =
then f gives a surjection from {) : < } to X, which we have just shown is injective, giving a well-ordering
of X. Thus, it suffices to show that it is impossible that
X {f () : ordinals < }

6=

for all ordinals . Indeed, if this were so, then the transfinite induction proceeds uninterrupted, and we have
an injective map f from all ordinals to X. But the collection of all ordinals is a class, not a set, so cannot
be injected to any set, contradiction. That is, at some point the transfinite induction fails, and we have the
desired well-ordering.
///

Exercises
14.[7.0.1] Show that the Well-Ordering Principle implies the Axiom of Choice.
14.[7.0.2] Show that an arbitrary poset is isomorphic, as a poset, to a set of sets, partially ordered by set
inclusion.