Basic Set Theory

Daniel Murfet
May 16, 2006
Throughout we work with the foundation provided by standard ZFC (FCT,Section 3). In
particular we do not assume we are working inside any xed grothendieck universe. It seems to
me that some basic proofs about ordinals in standard references are awed, so since the end result
is the same we adopt a slightly dierent denition to make life easier.
Contents
1 Ordinal Numbers 1
2 Transnite Pain 4
3 Cardinal Numbers 7
4 Cardinal Operations 10
5 Regular Cardinals 11
1 Ordinal Numbers
Denition 1. Given a set x, we let x
+
denote the set x x.
Denition 2. A set x is transitive if whenever y x we have y x. A set x is urtransitive if it is
transitive, and if in addition whenever a proper subset y x is transitive we have y x. We say
x is an ordinal if it is urtransitive, and if every element of x is urtransitive. We tend to use lower
case greek letters , , , . . . to represent ordinals. Given ordinals , we write for .
One checks that this is a transitive irreexive relation on ordinals. We write _ if or
= , and this denes a partial order on ordinals.
Remark 1. The following observations are immediate
The empty set is an ordinal. The empty set is a member of any nonempty ordinal.
Any element of an ordinal is an ordinal.
If , are ordinals then if and only if .
Lemma 1. If is an ordinal, then so is
+
.
Proof. The set
+
is clearly transitive. To see that it is urtransitive, let y be a transitive
proper subset. We want to show that y . Suppose that y ,= . First we claim that / y.
For if this were the case, transitivity of y implies y. Suppose t y . Then t y
so we must have t = . But then y , which is a contradiction. This shows that / y.
But then y must be a proper subset of , and since is an ordinal this means y , as required.
This shows that
+
is urtransitive, and it is clear that every element of
+
is urtransitive, so
+
is an ordinal.
1
Lemma 2. If a set contains an ordinal, then it contains a minimal ordinal.
Proof. Let X be a set and suppose X for some ordinal . Then the set Z of elements of X
which are ordinals is nonempty, and applying the Axiom of Foundation to this set we obtain an
ordinal X with the property that no ordinal in is an element of X.
Proposition 3 (Minimal element). Let B(x) be a wf with x free, and suppose there exists an
ordinal with B(). Then there a minimal such ordinal. That is, there exists an ordinal with
B() but not B() for any ordinal .
Proof. Given an ordinal with B() we can apply Lemma 2 to the set of all ordinals with
B() (of course if this set is empty, then already has the right property).
Theorem 4 (Trichotomy). Given two ordinals , we have , or = .
Proof. Suppose not, and let , be two ordinals satisfying none of these conditions. We x , and
let

be an ordinal minimal with the property that

, ,=

. By minimality for
any

we must , = or . The rst two alternatives would imply , which

contradicts our hypothesis. We deduce that , and since was arbitrary we have

. By
denition of an ordinal this implies

, a contradiction. This shows that for any two ordinals

, trichotomy holds.
Remark 2. Most references would dene an ordinal as a transitive set whose elements are all
transitive. But it seems to me there is a subtle diculty in proving trichotomy for this denition.
In any case, if you believe trichotomy for these ordinals, then you can easily show that a set is an
ordinal (in this sense) if and only if it is an ordinal in our sense.
Corollary 5. Let , be ordinals. Then if and only if
+

+
. If then
+
_ .
Proposition 6. If X is a set of ordinals, the union set

X is also an ordinal.
Proof. Let us treat some easy cases rst
If

X is empty then it is trivially an ordinal, so we can assume that X is nonempty and
also that X contains a nonempty set.
If X contains an ordinal which is maximal, that is, there is no ordinal X with ~ ,
then by trichotomy every ordinal in X belongs to (or is ) and so

X = is certainly
an ordinal. So we can assume that for every ordinal X there is another ordinal X
with ~ . One easy consequence of this assumption is that

X =

X; that is,

X is
the union of all its elements.
The set

X is clearly transitive, and every one of its elements is urtransitive, so to complete
the proof it suces to show that any proper transitive subset y

X is actually an element of

X. We assume otherwise, and produce a contradiction. Of course trivially

X, so we may
assume y is nonempty.
Suppose

X is an ordinal with / y. We must also have y / , since otherwise y

X
which contradicts our hypothesis. Since is urtransitive, we also deduce that y . It follows
that there exists an ordinal y . By trichotomy we must have and therefore y
by transitivity of y. This shows that any element of

X not in y is a subset of y. But of course
by transitivity any element of

X which is in y is also a subset of y, so as

X is the union
of its elements we deduce that

X y, a contradiction. This shows that y

X, so

X is
urtransitive and therefore an ordinal.
Corollary 7. Any transitive set X of ordinals is an ordinal.
Proof. If X is empty this is trivial, so assume otherwise. Since X is transitive we have

X X.
If this is an equality then by Proposition 6 we are done, so suppose to the contrary that there
exists X

X. That is, for every X. By trichotomy this means for every

,= in X. In particular X = =
+
is an ordinal.
2
Denition 3. An ordinal is a successor ordinal if there is an ordinal with
+
= . An
nonempty ordinal which is not a successor ordinal is called a limit ordinal. An ordinal is a nite
ordinal if = or is a successor ordinal and every element of is either empty or a successor
ordinal. If is a nite ordinal and then is a nite ordinal. Obviously the successor of
any nite ordinal is a nite ordinal.
Lemma 8. Let be a nite ordinal and W a set with the property that W and x
+
W
whenever x W. Then W.
Proof. Suppose / W and let be minimal with / W. Since is a nite ordinal it must
be a successor ordinal (since W). But if =
+
then is a nite ordinal not belonging to
W, contradicting minimality of .
Denition 4. By the Axiom of Innity at least one set of the form described in Lemma 8 exists,
so we can dene a set by
= [ is a nite ordinal
If we write 0 = , 1 = 0
+
, 2 = 1
+
and so on, then intuitively is the set 0, 1, 2, 3, . . ..
Proposition 9. A nonempty ordinal is a limit ordinal if and only if =

.
Proof. Let be a nonempty ordinal and suppose =

. Let be an ordinal. Then by
trichotomy we have , or = . If = then certainly ,=
+
. If then

+
so the two ordinals cannot be equal. Finally if =

then there is with
. Then
+
_ so ,=
+
. This shows that is a limit ordinal.
Conversely if is a limit ordinal, then of course

. If then by trichotomy
+

and therefore

. This shows that

and completes the proof.

Denition 5. A relation < on a set Y is a well-ordering if it is irreexive and if for every nonempty
subset Z Y there exists y Z such that whenever v Z with v ,= y we have y < v and not
v < y. It is clear that if < is a well-ordering, it is also a total order. A morphism of well-ordered
sets (Y, <) (Y

, <

) is a function f : Y Y

with the property that whenever x < y in Y we

have f(x) < f(y) in Y

. We say f is an isomorphism of well-ordered sets if there is a morphism of

well-ordered sets g : (Y

, <

) (Y, <) such that fg = 1, gf = 1. Since any well-ordering is total,

it is easy to check that f is an isomorphism of well-ordered sets if and only if it is a bijection.
Lemma 10. Any ordinal is well-ordered by the relation on its elements.
Proof. If is empty this is trivial, so suppose otherwise. Then every element of is an ordinal,
so is certainly an irreexive relation on the set . The fact that is a well-ordering follows
from Lemma 2 and trichotomy.
Lemma 11. There is no largest ordinal, and there is no set of all ordinals.
Proof. For any ordinal we have
+
, so there can be no largest ordinal. If there existed a
set X of all ordinals, the set

X would be a largest ordinal.
Lemma 12. If , are ordinals and f : a function preserving the relation , then for
every we have _ f().
Proof. When we say that f preserves the relation we mean that for ordinals we have
f() f(). Suppose to the contrary that there exists some ordinal with f() . We
may assume is minimal with this property. Then for every we must have _ f() f()
and consequently f(). This shows that f(). But this is impossible, since f() , and
from this contradiction we deduce the result.
Lemma 13. If two ordinals , are isomorphic as sets ordered by , then = .
Proof. That is, if there exist functions f : , g : preserving with fg = 1 and
gf = 1 then we claim = . This follows easily from the previous result.
3
2 Transnite Pain
Proposition 14 (Transnite Induction, First Form). Let B(x) be a wf with x free. Suppose that
for every ordinal , whenever we have B() for every ordinal we have B(). Then B()
for every ordinal .
Proof. Suppose that the inductive condition is satised. In particular we have B() vacuously.
Suppose for a contradiction that there exists an ordinal for which B() does not hold. Then
the set of ordinals with B() is nonempty (since otherwise the inductive condition would
force B()) and we can let

be minimal among these ordinals. This minimality together with

the induction hypothesis clearly implies B(

), which is the required contradiction.

Proposition 15 (Transnite Induction, Second Form). Let B(x) be a wf with x free. Suppose
(i) B().
(ii) B() implies B(
+
) for any ordinal .
(iii) If is a limit ordinal and B() for every ordinal , then B().
Then B() for every ordinal .
Proof. Follows trivially from Proposition 14.
Remark 3. Let B(x, y) be a wf with x, y free. Suppose that the wf x
1
yB(x, y) is a theorem in
ZFC (here
1
means exists a unique). Then as usual we can introduce a new function symbol
(x) and axiomxB(x, (x)). Formally this is how one introduces the notation AB, AB, AB
and so on for sets. Intuitively B(x, y) expresses that y is some construction starting from x, and
we have introduced the notation (x) to denote the end result of this construction. Throughout
the remainder of this section, if we say is a construction on sets we mean that has the
meaning just elaborated.
Theorem 16 (Construction by Transnite Recursion). Let be a construction on sets. For any
ordinal there is a unique function f dened on
+
with the property that
f() =
_
f() [
_
for every ordinal _ .
Proof. By transnite induction it suces to show that if the theorem is true for every ordinal
, it is true for . So let an ordinal be given, and assume that the theorem is true for
each ordinal . Let f

be the function dened on

+
. We dene a function f on
+
by rst
dening it on ordinals by
f() =
_
f

() [
_
Then dening
f() =
_
f() [
_
This denes a function f on
+
, but we have yet to show it has the required property or that it
is unique. We prove each statement separately:
Proof of existence. The rst step is to show that for for ordinals and we
have f

() = f

(). If an ordinal existed without this property, there would be a minimal such
. Then for every we would have f

() = f

() and consequently
f

() =
_
f

() [
_
=
_
f

() [
_
= f

()
4
which is a contradiction. Hence f

() = f

() for every . Consequently

f() =
_
f

() [
_
=
_
f

() [
_
= f

()
and f() = (f

() [ ) = (f() [ ), so f is a function on
+
with the correct
property. It remains to prove uniqueness.
Proof of uniqueness. Supose f

were another function dened on

+
with the stated property.
If there existed an ordinal with f

() ,= f() then there would exist a minimal such .

But then
f

() =
_
f

() [
_
=
_
f() [
_
= f()
a contradiction. Hence f

() = f() for every , and it follows immediately that f

() =
f() as well, completing the proof.
Remark 4. Let be a construction on sets, ordinals and f

, f

the induced functions on

+
,
+
respectively. Then
+

+
so f

restricts to a function on
+
. One checks easily that in
fact f

+ = f

.
Theorem 17. Let , be constructions on sets. Then for any ordinal and set z there is a
unique function f dened on
+
with the property that
(i) f() = z.
(ii) f(
+
) = (f()) for any ordinal .
(iii) f() =
_
f() [
_
for any limit ordinal _ .
Proof. Just to make the exposition clearer, we say a set q is groovy if it is a pair (x, y) with x an
ordinal. A set x is awesome if it is nonempty, all its elements are groovy, and the same ordinal
never occurs twice as the rst element of a pair in x. Given an awesome set x, let x
1
denote all
the ordinals occurring in the rst position of some pair in x, and similarly let x
2
be all the second
coordinates in x. We dene a new construction on sets as follows (the careful reader may write
out the corresponding explicit wf B(x, y) of ZFC if they desire)
If x is the empty set then (x) is the set (0, z).
If x is nonempty but not awesome, then (x) is the set (0, 0).
Assume that x is awesome. There are two cases
(a) The set of ordinals x
1
contains a maximal element. That is, there is some ordinal
x
1
with _ for every x
1
. If the pair (, m) is the (unique) pair in x with
rst coordinate then we dene (x) to be (
+
, (m)).
(b) The set of ordinals x
1
contains no maximal element. Then

x
1
is an ordinal, and we
dene (x) to be (

x
1
, (x
2
)).
By Theorem 16 for any given ordinal there exists a unique function g on
+
with
g() =
_
g() [
_
for every ordinal _ . By denition of its output is always a pair, so we can dene a new
function f on
+
by dening f() to be the second component of the pair g(). We claim that
this is the desired function.
Clearly f() = z. Given an ordinal we have
g(
+
) =
_
g() [
+

_
=
_
(, f()) [
+

_
5
But occurs as a maximal ordinal in these pairs, so by denition of we have g(
+
) =
(
+
, (f())) and so f(
+
) = (f()) as required. If _ is a limit ordinal, then the set
of ordinals occurring in the rst position of pairs in (, f()) [ has no maximal element,
so we have
g() =
_
(, f()) [
_
=
_
_
,
_
f() [
_
_
and therefore f() =
_
f() [
_
as required. This proves that a function f with the desired
properties exists, and uniqueness follows from a simple transnite induction.
Theorem 18. Let A be a set well-ordered by some relation <. Then there exists a unique ordinal
such that (A, <) and (, ) are isomorphic as well-ordered sets.
Proof. Uniqueness is immediate by Lemma 13, so it suces to show existence. If A is empty we
may take the empty ordinal, so assume A ,= . Let q be an arbitrary set not in A. Let z be a
least member of A. That is, since A is well-ordered by <, we can in particular nd z A with
the property that for every v A with v ,= z we have z < v and not v < z. Dene a construction
on sets as follows:
If A x is empty then (x) is the set q.
If A x is nonempty then (x) is the least member of this set.
By transnite recursion (i.e. Theorem 16) for any ordinal there is a function f

on
+
with the
following property for _
f

() =
_
least member of A f() [ A f

() [ , =
q otherwise
In particular f

(0) = z. Clearly either f

() = q or f

() A, with these two possibilities being

mutually exclusive. So for every ordinal we have the set f

(). Suppose that , are distinct

ordinals with f

(), f

() A. We can assume wlog that . Then f

() is an element of
A f

() [ and in particular f

() ,= f

(). But by Remark 4 we have f

() = f

(),
so we have shown that f

() ,= f

(). Since an element of A is therefore of the form f

() for at
most one ordinal , we can form the following set
= [ f

() A
To show that is an ordinal, it suces by Corollary 7 to show that it is transitive. So take
ordinals with . Since f

() A the set A f

() [ must have been nonempty,

so then clearly A f

() [ must also be nonempty. Therefore f

() A and , and
we have shown that is an ordinal. We claim that the set
F = (, f

()) [
is an order isomorphism A. It is certainly a function, which is injective by our earlier
comments. If this map were not surjective then the set A f

() [ would be nonempty,
from which we deduce the contradiction . So our map is a bijection. To see that this is an
isomorphism of well-ordered sets, it suces to show that it preserves the ordering on . So let
be ordinals. We already know A = f

() [ so
A f

() [ = f

() [ _
and by denition f

() is the least member of this set. From this we deduce f

() < f

() which
is what we wanted to show. We have produced an ordinal and an isomorphism of well-ordered
sets A, so the proof is complete.
6
3 Cardinal Numbers
Denition 6. Two sets x, y have the same cardinality if there is a bijection f : x y, and we
write x

= y. This is an equivalence relation on sets. We write x y if there is an injective map
f : x y. We write x < y if x y and not x

= y.
Lemma 19. Let S be a set and f : T(S) T(S) a function with the property that X Y
implies f(X) f(Y ). Then f has a xed point; that is, f(T) = T for some T T(S).
Proof. The set A = X T(S) [ X f(X) is nonempty, and if X A then also f(X) A. We
set T =

A and claim that f(T) = T. To begin with, if X A then X T so f(X) f(T),
and hence X f(X) f(T). It follows that T f(T), whence T A and f(T) A, showing
that f(T) T, which completes the proof.
Theorem 20 (Schroder-Bernstein). If a, b are two sets with a b and b a, then a

= b.
Proof. Suppose we are given injective maps f : a b and g : b a. We want to nd a subset
t a so that f on t pieced together with g
1
on at gives a bijection. To this end, for each x a
we dene a subset x

a by
x

= a g(b f(x))
Then x

measures how far we are from t. Now

x
0
x
1
f(x
0
) f(x
1
)
b f(x
0
) b f(x
1
)
g(b f(x
0
)) g(b f(x
1
))
a g(b f(x
0
)) a g(b f(x
1
))
In other words, x
0
x
1
implies x

0
x

1
. Hence by Lemma 19 there is t such that t = t

. That is,
a t = g(b f(t))
We dene a function h : a b piecewise by
h(x) =
_
f(x) x t
g
1
(x) x a t
Now h is onto since if y b f(t), let x = g(y) a t, then h(x) = y. And h is injective since if
h(x) = h(x

) and x, x

both belong to either t or a t then trivially x = x

. If one belongs to t and

one to its complement, then we would have an element in the intersection of f(t) and b f(t), a
contradiction. Therefore h is a bijection, as required.
Theorem 21 (Hartog). For any set a there is an ordinal which does not satisfy a.
Proof. That is, there is an ordinal which does not admit an injection into a. Suppose to the
contrary that every ordinal injects into a. In particular for every ordinal there is a subset b a
with

= b. In particular this induces a well-ordering < on b with (b, <)

= (, ) as well-ordered
sets. For each ordinal we dene a set
F

= (b, <) [ b a and < well-orders b and (b, <)

= (, ) as well-ordered sets
So our hypothesis implies that F

is nonempty. Using Lemma 13 we deduce that if F

= F

then
= . Since each F

is a subset of T(a) T(a a) and any subset of this set is of the form F

for at most one , we can form the following set

C = F

[ is an ordinal
But using the Axiom of Replacement we can replace each F

by and still have a set. This shows

that there is only a set of ordinals, which we already know to be false. This contradiction shows
that there must exist an ordinal not injecting into a.
7
Theorem 22. The following statements are true in ZFC
Zorns Lemma (ZL).
Given any set Y there exists a well-ordering on Y (WO).
Every set is in bijection with some ordinal (OT).
Proof. (ZL) Let (A, ) be a nonempty partially ordered set in which every chain has an upper
bound and let f be a choice function for T(A). Let q be a set not in A, and dene by transnite
recursion the following function G

on
+
for any ordinal (where we agree that f(0) = q)
G

() =
_
f
_
y A[ (G

() < y)
_
G

() A for every
q otherwise
The idea is that at each stage we produce an element of A strictly larger than all those previously
selected. By Remark 4 there is no harm in dropping the subscript from G and simply talking
about G as a function on ordinals. We say an ordinal is tame if G() A. If is tame, then
is tame for every . For an ordinal all of whose elements are tame, we dene
U

= y A[ (G() < y)
Then G() U

, provided U

,= . If U

is empty then G() = q so is not tame. So we

encounter untame ordinals when we run out of things in A. Of course G(0) = f(A) so the zero
ordinal is certainly tame.
We claim that not every ordinal is tame. Suppose otherwise. Then given ordinals
the ordinal has tame elements and by assumption U

,= so one deduces G() < G(). In

particular every ordinal corresponds to a distinct element of A, and by a standard argument this
would imply there is only a set of ordinals, which is false. Therefore some ordinal is not tame,
and we may assume is minimal. Then every element of is tame, so for to be not tame, we
must have U

= .
Clearly G() [ is a chain, and hence has an upper bound w in A. We claim that w is
maximal in A. For if not, we have v A with w < v. Then for all we have G() w < v
and so v U

which is a contradiction. This completes the proof of Zorns Lemma.

(WO) We prove that ZL WO. If our set Y is empty this is trivial, so assume Y is nonempty
and let / be the set of all pairs (A, <) where A Y and < is a well-ordering on A (this nonempty
since it contains with the trivial ordering). We dene a relation (A, <) (B, <

) to mean that
A B, the inclusion is a morphism of well-ordered sets, and a <

b for every a A, b B A.
This is a partial order on /.
Given a chain ( /, let R be the union of all the sets A occurring in pairs (A, <) ( and
dene a relation on R as follows: given a, b R nd a pair (A, <) ( with a, b A and dene
a b in R i. a < b. This doesnt depend on the chosen (A, <) and denes an irreexive relation
on R. To see that this is a well-ordering on R, let Z R be a nonempty subset and let (A, <) (
be arbitrary with Z A ,= . A least element of Z A will also be a least element of Z, so we
can assume Z A. But then A is well-ordered by <, so Z certainly has a least element in A and
therefore also in (R, ). This shows that every chain in ( has an upper bound, so by ZL it must
have a maximal element (M, <
M
).
We claim that M = Y , in which case the proof is complete. If not, choose a Y M and
extend <
M
to M a by declaring that a is larger than everything else in M. One checks
that this new set is well-ordered, and is strictly larger than (M, <
M
) in /, which is impossible.
Therefore M = Y and we are done.
(OT) Since we have shown WO, this now follows immediately from Theorem 18.
Denition 7. An ordinal is a cardinal if there is no ordinal with

= . Intuitively,
is the smallest ordinal in its equivalence class under bijection of sets. Bijective cardinals are
equal. For any ordinal there is a unique cardinal _ with

= .
8
Lemma 23. Let x, y be sets. Then x
+
= y
+
if and only if x

= y.
Proof. Let f : x
+
y
+
be a bijection. If f(x) = y then the restriction of f to x gives a bijection
x y. If f(x) = g y then there is some d x with f(d) = y and we dene a bijection
f

: x y to be f on every ordinal in x except for d, which we map to g. This is clearly a

bijection.
Lemma 24. Every nite ordinal is a cardinal.
Proof. If there exist nite ordinals which are not cardinals, then there exists a minimal such
ordinal . Since the empty set is clearly a cardinal, is nonempty. Therefore it is of the form
+
for some ordinal , which by minimality must be a cardinal. Since is not a cardinal, there
exists with

= . Since is nonempty, so is , so =
+
is a successor ordinal. From
this we deduce
+
=
+
and therefore

= . But _ which contradicts the fact that is
a cardinal. Therefore every nite ordinal is a cardinal.
Lemma 25. There is a bijection

=
+
. In particular the ordinal
+
is not a cardinal.
Proof. Dene a function f :
+
by f() = 0 and f(n) = n + 1 (here of course we write
n + 1 for the ordinal n
+
). This is easily checked to be a bijection.
Lemma 26. The ordinal is a cardinal.
Proof. If not, there is some nonempty nite ordinal with

= . We deduce
+
=
+
=

=
which contradicts the fact that the nite ordinal n
+
is a cardinal.
Lemma 27. Given two cardinals , we have _ if and only if . Similarly if and
only if < .
Proof. Certainly _ implies . The converse follows from the denition of a cardinal and
Theorem 20.
Lemma 28. If X is a set of cardinals, the union set

X is also a cardinal.
Proof. We know from Proposition 6 that

X is an ordinal, which we may as well assume is
nonempty. Suppose for a contradiction that there is an ordinal

X with

=

X. By
denition we have for some cardinal X. This implies , and from

X we
deduce

X and therefore . By Theorem 20 we have

= which is impossible because
is an ordinal. This contradiction shows that

X must be a cardinal.
Denition 8. Let x be a set. By Theorem 22 the set x is in bijection with some ordinal, and
therefore also with a cardinal . The unique cardinal with x

= is called the cardinality of x,
and is denoted [x[ or card(x). Obviously any cardinal is its own cardinality.
Denition 9. A set x is nite if [x[ , otherwise it is innite. We say x is countable if [x[ _ .
An ordinal is a nite ordinal if and only if it is a nite set. An innite ordinal contains the
successor of each of its nite elements.
Lemma 29. Let x, y be sets. Then x y if and only if [x[ [y[ and x < y if and only if [x[ < [y[.
For any two sets x, y we have x

= y, x < y or y < x.
Lemma 30. For any innite ordinal there is a bijection

=
+
.
Proof. We dene a map f :
+
by f() = 0, f(n) = n + 1 for nite ordinals n and
f() = for any innite ordinal . It is straightforward to check this is a bijection.
Lemma 31. Any innite cardinal is a limit ordinal.
Proof. Let be an innite cardinal. Then is nonempty, so it suces to show it is not a
successor. Suppose to the contrary that =
+
for some ordinal . If = then this would
imply is a nite ordinal, which it is not. So > and therefore _ implies is an
innite ordinal, so

=
+
= , but this is impossible because is an ordinal.
9
Let be an ordinal. By Theorem 21 there exists at least one ordinal which does not satisfy
(that is, it does not admit an injection into ). Let H() denote the unique ordinal which
is minimal with this property (that is, it does not admit an injection into , but all its members
do). Clearly H() by trichotomy. We make this into a construction on sets by declaring that
H(x) is the emptyset if x is not an ordinal. Then by Theorem 17 we can dene sets

for every
ordinal with the following properties

0
=

+ = H(

=
_

[
where is a limit ordinal. To be precise, for every ordinal we obtain a uniquely dened function
f

on
+
(using z = , = H, =

) and we dene

= f

().
Proposition 32. For any ordinal the set

is a cardinal.
Proof. Firstly, we observe that for any set x the set H(x) is a cardinal. Secondly, the union set
of any set of cardinals is a cardinal by Theorem 28, so in our denition by transnite recursion of
the alephs, everything in sight is a cardinal. Therefore the end product

must also be a cardinal

(more precisely, if

were not a cardinal we could assume was minimal with this property, and
derive a contradiction).
Corollary 33. For any ordinals we have

<

. In particular every aleph is an innite

cardinal.
Proof. Note that since alephs are cardinals, the statements

and

<

are equivalent.
We prove the result by the second form of transnite induction on the ordinal (with ranging
over all ordinals at each stage). The result holds vacuously for = 0. Now we suppose the result
is true for , and prove it for
+
. So let an ordinal
+
be given. If then by hypothesis

<

<

+, and if = then

<

+. So the result is true for

+
.
Now suppose that is a limit ordinal. In this case we dont even need to use our inductive
hypothesis. Since is a limit ordinal, for any we have
+
and so both

and

+
occur in the set Q =

[ . In particular

is an element of the following union

=
_

[
so

and therefore

<

, as required.
4 Cardinal Operations
Denition 10. Let x
i

iI
be a nonempty family of sets. We dene the disjoint union of this
indexed family to be the set

iI
x
i
= (i, y) [ i I and y x
i
. In particular we can dene the
disjoint union x Hy of any two sets x, y.
Denition 11. Let , be cardinals. We dene their cardinal sum and product as follows
+ = [ H[
= [ [
2

= [T()[
In fact one can check that if x, y are sets with [x[ = , [y[ = then = [x y[ and if x, y are in
addition disjoint then + = [xy[. More generally given a nonempty family
i

iI
of cardinals
we dene

iI

i
= card
_

iI

i
_

iI

i
= card
_

iI

i
_
10
Theorem 34. For any ordinal we have

. For any ordinals we have

.
Proposition 35. For any cardinal we have < 2

.
5 Regular Cardinals
Denition 12. Let be an ordinal. A subset A is conal in if for every ordinal
there exists A with _ .
Denition 13. For any ordinal there is a unique ordinal which is minimal with the property
that it admits a function whose image is a conal subset of . This ordinal is denoted
cf() and is called the conality of . By denition we have cf() _ .
Denition 14. We say that a cardinal is regular if cf() = , and singular otherwise. The
connality of a cardinal measures how accessible it is from below.
Example 1. If is a successor ordinal then cf() = 1 so the ordinal 1 is the only regular successor
ordinal. In particular the only nite ordinals which are regular are 0, 1. The ordinal is regular,
but

is not since cf(

) = (and since 0 we have =

0

).
Theorem 36 (Konig). Suppose we are given a nonempty set I and for each i I cardinals
1
i
<
i
. Then

iI

i
<

iI

i
Proposition 37. Any cardinal can be written as a sum of cf() cardinals all < . That is, we
can nd cardinals

< for cf() such that

=

Proposition 38. For any ordinal the cardinal

+ is regular.
Lemma 39. Any innite cardinal is of the form

for a unique ordinal .

Denition 15. Let be a cardinal. The cardinal successor of is the unique cardinal >
with the property that there is no cardinal with < < . If is a nite cardinal then the
cardinal successor is the usual ordinal successor
+
. Otherwise =

for some ordinal , and

the cardinal successor to is

+.
Denition 16. A cardinal is called a successor cardinal if it is the cardinal successor of some
cardinal . A nonzero cardinal which is not a successor cardinal is called a limit cardinal. Every
nonzero nite cardinal is a successor cardinal. Given an innite cardinal , we can write =

for a unique ordinal . Clearly is a successor cardinal if and only if is a successor ordinal (resp.
is a limit cardinal i. is a limit ordinal or zero).
Lemma 40. Let be a nonzero cardinal. Then is a limit cardinal if and only if for every
cardinal < there exists a cardinal with < < .
Lemma 41. Any regular ordinal is a cardinal.
Proof. See p.134 of Azriel Levys Basic Set Theory.
Proposition 42. Let be an innite cardinal. Then the following conditions are equivalent
(a) is regular.
(b) cannot be written as the union of < cardinals all of which are < .
(c) cannot be written as the sum of < cardinals all of which are < .
11
(d) Any sum of < cardinals all of which are < is also < .
Proof. See p.135 of Azriel Levys Basic Set Theory.
We already know that the cardinal

is regular whenever is a successor ordinal. Are there

innite regular cardinals which are not successor cardinals? We give such cardinals a special name.
Denition 17. Let > be an uncountable cardinal. We say that is weakly inaccessible if it
is a regular limit cardinal. An uncountable cardinal is called strongly inaccessible if it is regular
and satises the following condition
For every cardinal < , we have 2

< .
Clearly a strongly inaccessible cardinal is also weakly inaccessible.
Remark 5. Assuming that ZFC is consistent, the existence of (strongly or weakly) inaccessible
cardinals cannot be proved in ZFC. In fact the existence of strongly inaccessible cardinals is often
adopted as an additional axiom in foundations for category theory (this being equivalent to the
existence of universes). See SGA4 for more details in this direction.
In the remainder of this section we discuss the interaction of grothendieck universes and large
cardinals. The reader should consult (FCT,Section 4) for the denition of a grothendieck universe.
In particular our convention is that all universes are innite (that is, contain N).
Denition 18. Let U be a universe and Q be the following set of cardinals
Q = [x[ [ x U
By (BST,Lemma 28) the union set c(U) =

Q is also a cardinal. Since any element of U is also

a subset, we have c(U) [U[ (in general this is not an equality).
Lemma 43. Let U be a universe and a cardinal. Then < c(U) if and only if there is x U
with = [x[. In particular if x U then [x[ < c(U).
Proof. Suppose that < c(U). Then by denition there exists y U with < [y[ c(U).
Therefore is in bijection with some subset x y, which must also belong to U. For the converse,
we need only observe that if x U then T(x) U, so [x[ < 2
|x|
= [T(x)[ c(U).
Proposition 44. Let U be a universe. Then c(U) is a strongly inaccessible cardinal.
Proof. By convention = N U, so < c(U) by Lemma 43. First we have to show that
c(U) is regular. By Proposition 42 it suces to show that if I U is nonempty and
i

iI
a
family of cardinals all < c(U) then

i

i
< c(U). But each
i
is in bijection with some x
i
U,
and therefore

i

i
is in bijection with the disjoint union

iI
x
i
which belongs to U. So by
Lemma 43 we have

i

i
< c(U), as required. Now suppose we are given a cardinal < c(U).
Find a set x U with = [x[. Then T(x) U so by Lemma 43 we have [T(x)[ < c(U), and
therefore 2

= 2
|x|
= [T(x)[ < c(U). This completes the proof that c(U) is a strongly inaccessible
cardinal.
12