Professional Documents
Culture Documents
We start with some definitions and rapidly reach the notion of a well-ordered set.
Definition. For any X and any binary relation < on X, (X, <) is a partially ordered set (p.o. set)
if < is a transitive and anti-symmetric relation on X:
(X, <) is a linearly ordered set (l.o. set) ou a totally ordered set (t.o. set) if (X, <) is a partially
ordered set and ∀x ∈ X ∀y ∈ X (x < y or x = y or y < x) as well.
Definition. Now let < be any binary relation on X. If Y ⊆ X, then (Y, <) is an initial segment
of (X, <) if ∀x ∈ X ∀y ∈ X (x < y & y ∈ Y −→ x ∈ Y ). Often we will just say that Y is an initial
segment of X when it is clear about which relation < we are talking. (Y, <) is a proper initial
segment of (X, <) if Y 6= X as well. For all s ∈ X and Y ⊆ X, we say that s is a successor of Y
if s ∈
/ Y and { y ∈ X | y < s } ⊆ Y . And if Y ⊆ X and y ∈ X we say that y is <-minimal in Y if
y ∈ Y and ∀x ∈ X (x < y −→ x ∈ / Y ).
Theorem. The following conditions are equivalent for any X and any binary relation on X.
(1) Each proper initial segment of X has a successor.
(2) Each non-empty subset of X has a <-minimal element.
(3) The principle of induction on < is true:
∀x ∈ X ∀y ∈ X (y < x −→ φ(y)) −→ φ(x) −→ ∀x ∈ X φ(x).
Proof. (2) =⇒ (3). Suppose that ∀x ∈ X ∀y ∈ X (y < x −→ φ(y)) −→ φ(x) but
¬∀x ∈ X φ(x). Then {x ∈ X | ¬φ(x)} is non-empty and, by (2), has a member, let us say x0 ,
which is <-minimal. But then we have that ∀y < x0 φ(y) and hence φ(x), a contradiction.
(3) =⇒ (2). If Y ⊆ X and Y does not have a <-minimal element, then we have that
(1) =⇒ (2). For all x, y ∈ X let x <∗ y if there are x0 , . . . , xn−1 ∈ X and x < x0 < . . . < xn−1 < y.
<∗ is the transitive closure of <: it is transitive, <⊆<∗ and if R is a transitive binary relation with
<⊆ R, then <∗ ⊆ R.
(In order to see that <∗ is transitive take x <∗ y and y <∗ z and suppose that we have this because
x < x0 < . . . < xn−1 < y and y < y0 < . . . < ym−1 < z for some x0 , . . . , xn−1 , y0 , . . . , ym−1 ∈ X.
Thus, x < x0 < . . . < xn−1 < y < y0 < . . . < ym−1 < z, and so x < z. It is clear that
x < y =⇒ x <∗ y: take n = 0 in the definition of <∗ . And if x <∗ y there are x0 < . . . < xn−1 with
∗
version 1.1, 12/2/06
1
x < x0 < . . . < xn−1 < y, so if <⊆ R we have x R x0 R . . . Rxn−1 Ry, and hence, if R is transitive,
we have x R y.)
Definition. A binary relation < on a set X is well-founded if satisfies the conditions (1)-(3) of the
above theorem. A well-founded total order is called well-ordered or a well-ordering. (X, <) is then
a well-ordered set (a w.o. set).
2.2 Ordinals.
Thus, a well-ordered set is a set along which we can “count” one step at a time. The ordinals are
well-ordered sets that will be crucial for us. In fact we will see later that any well-ordered set is
isomorphic to some ordinal via an isomorphism which is order-preserving, hence one can say that
if we want to “count” we have to use the ordinals.
Definition. An ordinal is a transitive set that is well-ordered by ∈. (Frequently we will write <
instead of ∈ when we are dealing with ordinals.)
Examples. 0, 1, 2, 3, . . . , ω, . . . are (the first) ordinals. Exercise: if this is not apparent to you
then check it.
Proposition. If α is an ordinal and z ∈ α, then z is a transitive set and hence is a ordinal as well.
Proposition. If α, β are ordinals and α ⊆ β, then (a) α is an initial segment of β, and (b)
α 6= β =⇒ α ∈ β.
(b) If α 6= β, then, by the equivalent of being well-founded, there is a sucessor, let us call it s, of
α as a subset of β. Note that s is unique because ∈ is a l.o. of β. Also, because ∈ is a l.o. of β,
γ ∈ α implies that γ ∈ s, thus α ⊆ s. On the other hand, S = α ∪ { s} is an initial segment of β
and, thus, it is transitive and, so, for any γ ∈ s(∈ S) we have γ ∈ S. Because s ∈/ s (by the axiom
of foundation), γ ∈ s implies γ ∈ α and s ⊆ α. Hence α = s, ∈ β.
2
2.3. Trichotomy.
Theorem. If α and β are ordinals one and exactly one of the following holds: α ∈ β or α = β or
β < α.
(ii) α ∩ β = α and α ∩ β ⊂ β and α ∩ β 6= β. Then the last proposition of §2.2 immediately gives
that α ∈ β.
(iii) α ∩ β = β and α ∩ β ⊂ α and α ∩ β 6= α. Then the last proposition of §2.2 gives that β ∈ α.
(iv) α∩β ⊂ α, β, α∩β 6= α and α∩β 6= β. Note that we have already shown that α∩β is an ordinal.
Then the last proposition of §2.2 gives that α ∩ β ∈ α and that α ∩ β ∈ β. Henceα ∩ β ∈ α ∩ β, a
contradiction! (To the axiom of foundation.)
Notation. We write α ≤ β if α ∈ β or α = β.
Proof. It is easy to see that the collection is partially ordered by ∈ by the transitivity of ordinals.
The trichotomy theorem gives that ∈ totally orders the collection. Hence what remains to be shown
is that ∈ is well-founded on the collection.
Suppose that Z is a subset of the collection of all ordinals and Z 6= ∅. Let α ∈ Z. Either α is
∈-minimal (when we are finished), or Y = α ∩ Z 6= ∅. Y is a non-empty subset of α, hence there is
some β ∈ Y that is ∈-minimal. But β must be ∈-minimal in Z as well, because if γ ∈ β we have
that β ∈ α and, by the transitity of α, γ ∈ α. Hence γ ∈ Z.
Proof. Suppose the collection was a set, x, let us say. Then we would have that x ∈ x by the
previous corollary and the definition of ordinal. This would contradict the axiom of foundation
(again).
This corollary is sometimes called the Burali-Forti paradox. As you’ve already see, as with Cantor’s
paradox, this is not in fact a true paradox. Again, as with V , it is convenient to have a name for
the (proper) class of all ordinals, and so we write On for this class.
S
Corollary. If x is a set of ordinals, then x is also an ordinal.
S S
Proof.SIt is clear that x is transitive.SAnd because x ⊆ On, trichotomy gives that ∈ totally
orders x. So, suppose that ∅ 6= Z ⊆ x. Take α ∈ Z. As in the corollary two ago, either α
already is ∈-minimal, or α ∩ Z has a member that is ∈-minimal and that is ∈-minimal in Z as well.
2.4. Successors and limits. Let α be an ordinal. There are two things that can happen.
3
Either there is a ∈-maximal element of α, i.e., a β ∈ α such that ∀γ ∈ α γ ≤ β. In this case
α = {β } ∪ {γ | γ < β }, = {β } ∪ β, = s(β). If there is some β such that α = s(β) we say that α is
a successor ordinal .
2.5. Recursion Theorems. Now we want to define + and . on the ordinals, in a similar manner
to that in which we did on the natural numbers. But before we can do this we need to know that we
can make definitions using recursion and how to make these defintions. So this section is devoted
to that theme. Perhaps at first glance it will seem very technical, but it will be highly useful in the
end.
Weak recursion theorem. Let X, Y be two sets. Let < be a well-founded relation on X, and
let g : X × P (Y ) −→ Y be a function. Then there is a unique function f : X −→ Y such that
f (x) = g(x, { f (z) | z < x}) for all x ∈ X.
But if dom(f ) 6= X there is some x0 that is <-minimal in {x ∈ X | x ∈ / dom(f )}. For such an x
¯ ¯ ¯
define f : dom(f ) ∪ { x0 } −→ Y by f (x) = f (x) if x ∈ dom(f ) and f (x0 ) = g(x0 , {f (x) | x < x0 }).
Then one would clearly have that f¯ is an attempt, but is not a subset of f . contradiction! Hence
f : X −→ Y as we want.
Now we shall see a stronger recursion theorem which deals with classes.
Let A be a class. < is a well-founded Relation (note the capital “R”) on A if < is a subclass do
A × A, ∀a ∈ A { b ∈ A | b < a} is a set, and any B with ∅ =
6 B ⊆ A has <-minimal elements.
4
and for each x ∈ A there is a unique y ∈ B such that (x, y) ∈ G. If x ∈ A, y ∈ B and (x, y) ∈ G
we shall write (naturally) G(x) = y. (Note that to say that G is a (sub)class means that there is a
formula χ(z, c̄) in LTC such that x ∈ G if and only if χ(x, c̄) is true.)
Remember that sets are classes as well and if A and B are sets, then A × B is a set, < is a
well-founded relation and G is a function, respectively.
Proof. (Part 1.) The first part is just as in the weak recursion theorem. Define φ to be an attempt
if φ : B −→ V , where B is a set that is an initial segment of A and φ(x) = G(x, {φ(y) | y < x}).
Then any two attempts agree on the intersection of their initial segments.
S
(Part 2.) Let F = { φ | φ is an attempt } (that is (x, z) ∈ F if there is an attempt φ with φ(x) = z).
Note that the last time this was a set, but here we merely have that F ⊆ P (A × V ) and we do
not know that F is a set, since it will not be one if A is not. On the other hand we have that
F (x) = G(x, { F (y) | y < x}) for all x ∈ dom(F ). Suppose that dom(F ) 6= A, and choose a0 that is
<-minimal in the set A \ dom(F ).
Now we have that ∀b < a0 ∃φ (φ is an attempt and b ∈ dom(φ). By this we can apply the Ax-
iom of Collection which gives that ∃v ∀b < a0 ∃φ ∈ v (φ is an attempt and b ∈ dom(fS ). Hence
w = v ∩ { φ | φ is anattempt} is a set by the Axiom of Comprehension. Clearly f = w is an
attempt and we can consider f¯ = f ∪ {(a0 , G(a0 , {f (b) | b < a0 }))} which is an attempt but is not
a subset do f . Thus, as in the weak recursion theorem we have a contradiction! Hencedom(F ) = A.
Why are there differences in the proof from that of the weak recursion theory? Well, in order to
imitate that proof you would have to know that G(a0 , {F (a) | a < a0 }) makes sense before defining
F̄ = F ∪ { (a0 , G(a0 , { F (a) | a < a0 }))}. And in order to know this you would have to know that
{ F (a) | a < a0 } is a set, which is a consequence of applying Replacement to the set {a | a < a0 }.
Hence one sees that it is not possible to avoid the use of Collection or Replacement.
I ought to warn that one cannot avoid the use of Collection in the strong recursion theorem even
if A is a set if G is not one, and this is also so even if < is a well-order A and rather than just a
well-founded relation.
Definition. A function f : (X, <X ) −→ (Y, <Y ) is an isomorphism which preserves order if f is a
bijetion between X and Y and ∀x, y ∈ X (x <X y ←→ f (x) <Y f (y)).
Note. If f : (X, <X ) −→ (Y, <Y ) is a function that preserves order f is injective, and so you only
need to know that f is also a surjection in order to have that f is an isomorphism.
Theorem. For any well-ordered set (X, <) there is a unique ordinal α for which there is an (unique)
isomorphism from (X, <) to α which preserves order.
Proof. Define, by the (weak) recursion theorem, a function f : X −→ V (On) which satisfies
f (x) = { f (y) | y < x}.
(i) ∀x ∈ X f (x) ∈ On. If not choose some x0 which is <-minimal in the set { x ∈ X | f (x) ∈/ On}.
Suppose that x ∈ X and for all y < x we have that f (y) ∈ On. Then f (x) ⊆ On by the definition
of f , and f (x) is well-ordered by ∈. But also, if z ∈ y ∈ f (x), we have that y = f (v) for some
5
v < x. Thus z ∈ {f (u) | u < v } and there is some u < v < x such that z = f (u). Consequently,
z ∈ f (x), and f (x) is transitive. Hence f (x) ∈ On. So, by the principle of indution on <, we have
that ∀x ∈ X f (x) ∈ On.
(ii) im(f ) is a set (by the axioms of collection and comprehension or by replacement), and (i) shows
that im(f ) ⊆ On, thus im(f ) is well-ordered by ∈. Suppose that z ∈ y ∈ im(f ), hence there is
some x ∈ X such that z ∈ y = f (x), and so there is some v ∈ X such that v < x and z = f (v).
This gives that z ∈ im(f ) and im(f ) ∈ On.
Comparison Theorem. Let (X, <X ) and (Y, <Y ) be two w.o.sets. Then there is an isomorphism
that preserves order between one of them and an initial segment of the other.
Proof. The previous theorem says that there are α, β ∈ On and isomorphisms f : (X, <X ) −→ α
and g : (Y, <Y ) −→ β which preserve order. Now we apply trichotomy. If α = β we have that g −1 ·f
is an order-preserving isomorphism between (X, <X ) and (Y, <Y ). If α ∈ β we have that g −1 · f is
an order-presrving isomorphism between (X, <X ) and a proper initial segment of (Y, <Y ) (of which
g −1 (α) is a successor). And if β ∈ α we have that f −1 · g is an isomorphism which preserves order
between (Y, <Y ) and a proper initial segment of (X, <X ) (of which f −1 (β) is a successor).
Now we can rigorously define (from the axioms) the Vα hierarchy. Define, bySthe strong recursion
theorem, a Function V : On −→ V (the collection of all sets) such that Vα = { P (Vβ ) | β < α }.
Proposition. ∀α Vα is transitive.
Proof. If not choose α (<-)minimal such that Vα not transitive. We have already seen (in the first
proposition
S of §1.5) that P (X) is transitive if X is transitive. So ∀β < α P (Vβ ) is transitive. But
Vα = { P (Vβ ) | β < α } is a union of transitive sets and so is also transitive.
Proof. If α = β it is trivial. If α < β, then Vα ⊆ P (Vα ), because Vα and, so, P (Vα ) are transitive,
and P (Vα ) ⊆ Vβ .
6
and so that x0 ∈ Vs(sup(u)) . contradiction!
S
“⇐=”. Let A 6= ∅ be a class with A ⊆ { Vα | α ∈ On}. Take α minimal such that A ∩ Vα 6= ∅.
Then any a ∈ A ∩ Vα is ∈-minimal in A.
S
Definition. Define a function rk : V −→ V (On) by rk(x) = { s(rk(y)) | y ∈ x}, using the
(strong) recursion theorem.
Proposition. (i) ∀x ∈ V rk(x) ∈ On. (ii) rk(x) = the minimal α such that x ⊆ Vα . (iii) rk(x)+1 =
the minimal α such that x ∈ Vα .
For each set define, by recursion over (ω, ≤), a function t(n, x) such that
[
t(0, x) = x and t(n + 1, x) = t(n, x) ∪ t(n, x).
S
At the end, define TC(x) = { t(n, x) | n ∈ ω }. TC(x) is the transitive closure of x
Proposition. For every x, TC(x) is the minimal transitive set z with x ⊆ z. (That is, if y is
transitive and x ⊆ y, then TC(x) ⊆ y.)
S Let u ∈ v ∈ TC(x). Then there is some n < ω such that u ∈ v ∈ t(n, x).
Proof. So
u ∈ t(n, x) ⊆ t(n + 1, x) ⊆ TC(x). Thus TC(x) is transitive.
Let y be transitive x ⊆ y. We shall show by induction on ω that each t(n, x) ⊆ y. This is clear
whenS n = 0. Suppose that t(n, x) ⊆ y. If z ∈ t(n + 1, x) we either have z ∈ t(n, x), when z ∈ y, or
z ∈ t(n, x). So there is some v ∈ t(n, x) such that z ∈ v. But if z ∈ v ∈ t(n,Sx) ⊆ y, we have that
z ∈ y by the transitivity of y. So we have t(n + 1, x) ⊆ y, and thus TC(x) = t(n, x) ⊆ y.
Proposition. The axiom of foundation for sets (each non-empty set has a ∈-minimal element)
implies the axiom of foundation for classes (each non-empty class has a ∈-minimal element).
Now we will show that α + β0 is transitive, and so is an element of On, giving a contradiction to
the supposition. Let z ∈ y ∈ α ∪ { α + γ | γ < β0 }. Then, either y ∈ α, when z ∈ α, because α is
7
an ordinal, or there is some γ < β such that y = α + γ ∈ On, hence, again, z ∈ { α + δ | δ < γ }. In
each case we have that z ∈ α + { α + γ | γ < β0 }, giving the desired contradiction. So α + β0 ∈ On.
Proposition.
S (i) α + 0 = α; (ii) α + s(β) = s(α + β); (iii) if λ is a limit ordinal, then
α + λ = { α + γ | γ < λ}.
If i < j then i = 0 and j = 1, f (γ, i) = γ and f (δ, j) = α + δ, and we have that γ < α ≤ α + δ.
If i = j = 0, then f (γ, i) = γ and f (δ, j) = δ and we have γ < δ. And if i = j = 1 we have that
α + γ < α + δ by the previous lemma.
S S
Proposition. If x ⊆ On is a non-empty set then α + x= {α + β | β ∈ x}. (+ is continuous
in the second place/variable.)
S S S S
Proof. If β ∈ x then β ≤ x, and so ∀β ∈ x (α+β ≤ α+ x). Hence {α+β | β ∈ x} ⊆ α+ x.
S S S
Let ρ = { α + β | β ∈ x} If ρ ∈ α + x, then either ρ ∈ α or there is some γ ∈ x such S that
ρ = α + γ. If ρ ∈ α, then ∀β ∈ x α + β ≤ ρ < α, a contradiction. And ifSρ = α + γ with γ ∈ x,
then there is some β ∈ x such that
S γ < β and so ρ < α + β and S ρ < { α + β | β ∈ x} = ρ, a
contradiction again. So, ρ ∈
/ α + x and we have that ρ = α + x.
We can define subtraction as well, but because addition is not commutativity it is not as useful as
addition.
8
if β > α and α − β = {α − γ | γ < β } if β ≤ α.
Proof. Suppose that α.γ ∈ On for each γ < β. So α.γ + δ ∈ On for each γ < β by the
proposition that ξ + ζ ∈ On if ξ, ζ ∈ On, and we have that α.β ⊆ On and hence is well-ordered
by ∈. Also, for each γ < β the function fγ : α −→ {α.γ + δ | δ < α } given by fγ (δ) = α.γ + δ has
fγ “α = { α.γ +δ | δ < α }, and hence { α.γ +δ | δ < α }, Aγ let’s call it, is a set by Replacement.
S Let g
S given by γ 7→ Aγ for all γ < β. Then g“β is again a set and so α.β = {Aγ | γ < β }
be the function
is the union g“β of a set and is also a set. Finally, if x ∈ y ∈ α.β, then there is some γ < β
and δ < α such that x ∈ y = α.γ + δ. Hence either x ∈ α.γ or there is some σ < δ such that
x = α.γ + σ. In the first case there are some < γ and τ < α such that x = α. + τ . So in each
case we have that x ∈ α.β. Because we now have that ∀β (∀γ < β (α.γ ∈ On) −→ α.β ∈ On), we
have that ∀β ∈ On α.β ∈ On.
Lemma. ∀α ∈ On (i) 0.α = 0; (ii) 1.α = α; (iii) ∀β, γ ∈ On (β < γ) =⇒ α.β < α.γ; and (iv) ∀δ
β ∈ On, (α ≤ δ =⇒ α.β ≤ δ.β).
Proof. (i)–(iii) are obvious. (iv) Suppose that α, δ ∈ On, β ∈ On and for all γ < β we have that
α.γ ≤ δ.γ. Using the lemma about addition from §2.8, we have for all γ < β and all < α that
α.γ + ≤ δ.γ + . Hence either α.γ + = δ.γ + , when α.γ + ∈ δ.β, or α.γ + ∈ δ.γ + ∈ δ.β,
and α.γ + ∈ δ.β, by the transitivity of δ.β. In each case we have that α.β ⊆ δ.β and so α.β ≤ δ.β.
But this shows that (∀γ < β α.β ≤ δ.γ) −→ α.β ≤ δ.β. Hence, because < is well-founded on the
ordinals, ∀β ∈ On α.β ≤ δ.β.
Proposition.
S (i) α.0 = 0; (ii) α.s(α) = α.β + α; and (iii) if λ ∈ On is a limit ordinal, then
α.λ = { α.γ | γ < λ}.
9
δ0 < δ1 , when α.γ0 + δ0 < α.γ0 + δ1 by the lemma about addition. Hence, f preserves order, and
so is an isomorphism.
Note. 2.ω = ω, 3.ω = ω, . . . , but ω.ω 6= ω. Note also that (ω + 1).ω = ω.ω
Theorem. Let α, β ∈ On, β > 0. Then there are unique σ, τ with τ < β such that α = β.σ + τ .
Definition. Let α, β ∈ On. Then αβ = {0 } ∪ {αγ .δ + | γ < β & δ < α & < αγ }.
Lemma. ∀α, β, γ δ ∈ On we have (i) αβ .αγ = αβ+γ ; (ii) (αβ )γ = αβ.γ ; (iii) β < γ =⇒ αβ < αγ ;
and (iv) α ≤ δ =⇒ αβ ≤ δ β .
α ∈ On. (i) α0 = 1; (ii) αs(β) = αβ .α for each β ∈ On; and (iii) if λ is a limit
Proposition. Let S
ordinal, then α = {αγ | γ < λ}.
λ
The proofs are similar to the proofs of the equivalents for ‘.’. You should check these if you desire.
ω
Note. ∀n ∈ ω nω = ω, but ω ω 6= ω, when ω.ω ω = ω ω . And if we define 0 = { ω, ω ω , ω ω , . . . },
S
then ω 0 = 0 .
When you come back to this section later you should remember that all of these facts are for
exponentiations of ordinals, that is very different from the theory of cardinal exponentiation that
we will see in Chapter (4).
10