Logic and Set Theory

Lectured by I. B. Leader, Lent Term 2005, 2010

Chapter 1 Propositional Logic 1

Chapter 2 Well-Orderings and Ordinals 7

Chapter 3 Posets and Zorn’s Lemma 16

Chapter 4 Predicate Logic 24

Chapter 5 Set Theory 34

Chapter 6 Cardinals 43

Bonus lecture Incompleteness

Examples Sheets

Prerequisites. Have met groups, fields, vector spaces (as examples); countability

There are four examples sheets.

Books. 1. P. T. Johnstone, ‘Notes on Logic & Set Theory’, CUP 1987

2. D. Van Dalen, ‘Logic and Structure’, Springer-Verlag 1980 (good for Chapter 4)

3. A. Hajnal & P. Hamburger, ‘Set Theory’, CUP 1999 (for cardinals and ordinals)

4. T. Forster, ‘Logic, Induction and Sets’, CUP 2003 (good bedtime read)

Last updated: Tue 28th Oct, 2014

Please let me know of any corrections: [email protected]
 Course schedule

LOGIC AND SET THEORY (D) 24 lectures, Lent term

No specific prerequisites.

Ordinals and cardinals

Well-orderings and order-types. Examples of countable ordinals. Uncountable ordinals and
Hartogs’ lemma. Induction and recursion for ordinals. Ordinal arithmetic. Cardinals; the
hierarchy of alephs. Cardinal arithmetic. [5]

Posets and Zorn’s lemma

Partially ordered sets; Hasse diagrams, chains, maximal elements. Lattices and Boolean
algebras. Complete and chain-complete posets; fixed-point theorems. The axiom of choice
and Zorn’s lemma. Applications of Zorn’s lemma in mathematics. The well-ordering princi-
ple. [5]

Propositional logic
The propositional calculus. Semantic and syntactic entailment. The deduction and com-
pleteness theorems. Applications: compactness and decidability. [3]

Predicate logic
The predicate calculus with equality. Examples of first-order languages and theories. State-
ment of the completeness theorem; *sketch of proof*. The compactness theorem and the
Löwenheim-Skolem theorems. Limitations of first-order logic. Model theory. [5]

Set theory
Set theory as a first-order theory; the axioms of ZF set theory. Transitive closures, epsilon-
induction and epsilon-recursion. Well-founded relations. Mostowski’s collapsing theorem.
The rank function and the von Neumann hierarchy. [5]

Consistency
*Problems of consistency and independence*. [1]

Appropriate books

B.A. Davey and H.A. Priestley Lattices and Order. Cambridge University Press 2002 (19.95
paperback).
T. Forster Logic, Induction and Sets. Cambridge University Press (50.00 hardback).
A. Hajnal and P. Hamburger Set Theory. LMS Student Texts number 48, CUP 1999 (55.00
hardback, 22.99 paperback).
A.G. Hamilton Logic for Mathematicians. Cambridge University Press 1988 (25.95 paper-
back).
P.T. Johnstone Notes on Logic and Set Theory. Cambridge University Press 1987 (15.95
paperback).
D. van Dalen Logic and Structure. Springer-Verlag 1994 (18.50 paperback).
 Chapter 1 : Propositional Logic

Let P be a set of primitive propositions. Unless otherwise stated, P = {p1 , p2 , p3 , . . .}.

The set of propositions, written L or L(P ), is defined inductively by:

(i) if p ∈ P then p ∈ L,

(ii) ⊥ ∈ L (‘⊥’ is read ‘false’),

(iii) if p, q ∈ L then (p ⇒ q) ∈ L.

Examples. (p1 ⇒ ⊥), ((p1 ⇒ p2 ) ⇒ (p1 ⇒ p3 )), ((p ⇒ ⊥) ⇒ ⊥).

Notes. 1. A proposition is a finite string of symbols from the alphabet: ⊥ ⇒ ( ) p1 p2 . . .

(Often omit the outer brackets, or use different brackets [ ] for clarity.)

2. ‘L defined inductively’ means, more precisely, that we set L1 = {⊥} ∪ P , and for
n > 1, Ln+1 = Ln ∪ {(p ⇒ q) : p, q ∈ Ln }, and then L = L1 ∪ L2 ∪ . . ..
(So Ln = ‘things born in time n’.)

3. Every proposition is built up from 1 and 2 using 3 in a unique way.

E.g. [(p1 ⇒ p2 ) ⇒ (p1 ⇒ p3 )] came from (p1 ⇒ p2 ) and (p1 ⇒ p3 ).

Can now define, for example,

¬p (‘not p’) as an abbreviation for (p ⇒ ⊥)

p∨q (‘p or q’) as an abbreviation for (¬p) ⇒ q
p∧q (‘p and q’) as an abbreviation for ¬(p ⇒ (¬q))

Semantic Implication

A valuation on L is a function v : L → {0, 1} such that:

(i) v(⊥) = 0,

0 if v(p) = 1, v(q) = 0,
(ii) v(p ⇒ q) = for all p, q ∈ L.
1 otherwise,

Remark. On {0, 1}, can define a constant ⊥ by ⊥ = 0, and an operation ⇒ by


0 if a = 1, b = 0,
(a ⇒ b) =
1 otherwise.

Then a valuation is precisely a map v : L → {0, 1} that preserves the structure (⊥ and ⇒),
i.e. a homomorphism.

Proposition 1. (i) If v, v ′ are valuations with v(p) = v ′ (p) for all p ∈ P , then v = v ′ .

(ii) For any function w : P → {0, 1}, there exists a valuation v such that v(p) = w(p)
for all p ∈ L.

‘A valuation is determined by its values on P , and any values will do.’

1
Proof. (i) We have v(p) = v ′ (p) for all p ∈ L1 . But if v(p) = v ′ (p) and v(q) = v ′ (q) then
v(p ⇒ q) = v ′ (p ⇒ q), so v(p) = v ′ (p) for all p ∈ L2 .
Continuing inductively, we get v(p) = v ′ (p) for all p ∈ Ln , all n.

(ii) Set v(p) = w(p) for each p ∈ P , and v(⊥) = 0; this defines v on L1 . Having defined
v on Ln , use 
0 if v(p) = 1, v(q) = 0
v(p ⇒ q) =
1 otherwise
to define v on Ln+1 . 

E.g. Let v be the valuation given by: v(p1 ) = v(p2 ) = 1, and v(pn ) = 0 for n > 3. Then

v( ( p1 ⇒ p2 ) ⇒ p3 ) = 0.
| {z } |{z}
1 0

Say t is a tautology, written |= t, if v(t) = 1 for all v.

Examples.

1. p ⇒ (q ⇒ p) (‘a true statement is implied by anything’)

v(p) v(q) v(q ⇒ p) v(p ⇒ (q ⇒ p))

1 1 1 1
so a tautology, as last
1 0 1 1 ←−
column identically 1
0 1 0 1
0 0 1 1

2. (¬¬p) ⇒ p, i.e. ((p ⇒ ⊥) ⇒ ⊥) ⇒ p (‘law of the excluded middle’)

v(p) v(p ⇒ ⊥) v((p ⇒ ⊥) ⇒ ⊥) v(((p ⇒ ⊥) ⇒ ⊥) ⇒ p)

so a
1 0 1 1 ←−
tautology
0 1 0 1



3. p ⇒ (q ⇒ r) ⇒ (p ⇒ q) ⇒ (p ⇒ r)



Suppose not a tautology. Have v with v p ⇒ (q ⇒ r) = 1, v (p ⇒ q) ⇒ (p ⇒ r) = 0.
Then v(p ⇒ q) = 1 and v(p ⇒ r) = 0. Thus v(p) = 1, v(r) = 0, so v(q) = 1, so


v(q ⇒ r) = 0, and so v p ⇒ (q ⇒ r) = 0. /\/\

For S ⊂ L, t ∈ L, say S entails t, or semantically implies t, written S |= t, if

v(s) = 1 for all s ∈ S implies v(t) = 1.

‘Whenever all of S is true, t is true as well.’

E.g. {p ⇒ q, q ⇒ r} |= (p ⇒ r).
Need: any valuation v with v(p ⇒ q) = 1, v(q ⇒ r) = 1 has v(p ⇒ r) = 1.
If not, then v(p ⇒ r) = 0, whence v(p) = 1, v(r) = 0, so v(q) = 0, as v(q ⇒ r) = 1.
So v(p ⇒ q) = 0. /\/\

If v(t) = 1, say t is true in v, or v is a model of t. For S ⊂ L, a valuation v is a model of

S if v(s) = 1 for all s ∈ S. Thus S |= t says ‘every model of S is a model of t’.

Note. |= t means exactly ∅ |= t.

2
Syntactic Implication

For a notion of proof, we shall need some axioms and deduction rules. As axioms, we shall
take all propositions of the following form.

1. p ⇒ (q ⇒ p) (all p, q ∈ L)

2. [p ⇒ (q ⇒ r)] ⇒ [(p ⇒ q) ⇒ (p ⇒ r)] (all p, q, r ∈ L)

3. (¬¬p) ⇒ p (all p ∈ L)

Note. These are all tautologies. Sometimes, they are called ‘axiom-schemes’, as each is
infinitely many axioms.

As our deduction rule, we shall use only modus ponens: ‘from p and p ⇒ q, can deduce q’.

For S ⊂ L, t ∈ L, a proof of t from S is a finite sequence t1 , t2 , . . ., tn of propositions, with

tn = t, such that each ti is either

(i) an axiom,

(ii) a member of S,

(iii) such that there exist j, k < i with tj = (tk ⇒ ti ).

If there exists a proof of t from S, say S proves t, or syntactically implies t, written

S ⊢ t. If ∅ ⊢ t, say t is a theorem, written ⊢ t. In a proof, S consists of the hypotheses or
premises, and t is the conclusion.

Example. {p ⇒ q, q ⇒ r} ⊢ (p ⇒ r) ( ‘go for (p ⇒ q) ⇒ (p ⇒ r)’ )

1. [p ⇒ (q ⇒ r)] ⇒ [(p ⇒ q) ⇒ (p ⇒ r)] (axiom 2)

2. q⇒r (hypothesis)
3. (q ⇒ r) ⇒ [p ⇒ (q ⇒ r)] (axiom 1)
4. p ⇒ (q ⇒ r) (modus ponens on 2, 3)
5. (p ⇒ q) ⇒ (p ⇒ r) (modus ponens on 1, 4)
6. p⇒q (hypothesis)
7. p⇒r (modus ponens on 5,6)

Example. ⊢ (p ⇒ p) ( ‘go for (p ⇒ (p ⇒ p)) ⇒ (p ⇒ p), using axiom 2’ )

1. p ⇒ [(p ⇒ p) ⇒ p] ⇒ [p ⇒ (p ⇒ p)] ⇒ (p ⇒ p) (axiom 2)
2. p ⇒ [(p ⇒ p) ⇒ p] (axiom 1)
3. [p ⇒ (p ⇒ p)] ⇒ (p ⇒ p) (modus ponens on 1, 2)
4. p ⇒ (p ⇒ p) (axiom 1)
5. p⇒p (modus ponens on 3, 4)

In showing S ⊢ p, often helpful to use the following.

Proposition 2 (Deduction Theorem). Let S ⊂ L, p, q ∈ L. Then S ⊢ (p ⇒ q) if and

only if S ∪ {p} ⊢ q.

‘Provability corresponds to the connective ‘⇒’ in the language.’

3
Proof. (⇒) Given a proof of p ⇒ q from S, write down the lines
p (hypothesis)
q (modus ponens)
to obtain a proof of q from S ∪ {p}.

(⇐) Let t1 , . . ., tn be a proof of q from S ∪ {p}. We’ll show that S ⊢ (p ⇒ ti ) for all i.

1. If ti is an axiom, write down

ti ⇒ (p ⇒ ti ) (axiom 1)
ti (axiom)
p ⇒ ti (modus ponens)

2. If ti ∈ S, write down

ti ⇒ (p ⇒ ti ) (axiom 1)
ti (hypothesis)
p ⇒ ti (modus ponens)
showing S ⊢ (p ⇒ ti ).
3. If ti = p, then certainly S ⊢ (p ⇒ p), as ⊢ (p ⇒ p).
4. If ti obtained by modus ponens, then we have earlier lines tj and tk = (tj ⇒ ti ).
By induction, may assume S ⊢ (p ⇒ tj ) and S ⊢ (p ⇒ (tj ⇒ ti )), so write down

[p ⇒ (tj ⇒ ti )] ⇒ [(p ⇒ tj ) ⇒ (p ⇒ ti )] (axiom 2)

(p ⇒ tj ) ⇒ (p ⇒ ti ) (modus ponens)
p ⇒ ti (modus ponens)
(‘This is the real reason why axiom 2 is the way it is.’) 

For example, to show {p ⇒ q, q ⇒ r} ⊢ (p ⇒ r), enough by the Deduction Theorem to show

that {p ⇒ q, q ⇒ r, p} ⊢ r, which is trival by modus ponens twice.

Question. How are ⊢ and |= related?

Aim. Completeness Theorem: S ⊢ t ⇐⇒ S |= t.

This is made up of soundness (if S ⊢ t then S |= t) and adequacy (if S |= t then S ⊢ t).

Soundness says ‘our axioms are not absurd’, and adequacy says ‘our set of axioms is strong
enough to prove, from S, every semantic consequence of S’.

Proposition 3 (Soundness Theorem). Let S ⊂ L, t ∈ L. Then S ⊢ t ⇒ S |= t.

Proof. We have a proof of t from S. We must show that if v is a valuation with v(s) = 1
for all s ∈ S (i.e. v is a model of S) then v(t) = 1 (i.e. v is a model of t).

But v(p) = 1 for all p ∈ S (as v is a model of S) and v(p) = 1 for every axiom p (as
each axiom is a tautology), and if v(p) = 1, v(p ⇒ q) = 1 then v(q) = 1.

Hence, each line ti of a proof of t from S has v(ti ) = 1. 

Say S is consistent if S 6⊢ ⊥.

4
A special case of adequacy is: S |= ⊥ ⇒ S ⊢ ⊥ (i.e. ‘S has no model’ ⇒ ‘S is inconsistent’).
Or, in other words: S is consistent ⇒ S has a model.

In fact, this would imply adequacy in general. Indeed, given S |= t, have that S ∪ {¬t} has
no model, so we should know S ∪ {¬t} ⊢ ⊥, whence:

S ⊢ ((¬t) ⇒ ⊥) (by deduction theorem)

i.e. S ⊢ (¬¬t)
but S ⊢ ((¬¬t) ⇒ t) (axiom 3)
so S ⊢t (modus ponens)

So, for adequacy, we must show: S is consistent ⇒ S has a model. How might we show this?
Given a consistent set S, how can we ‘build’ a valuation v with v(s) = 1 ∀s ∈ S? We cannot
just set v(p) = 1 if p ∈ S and v(p) = 0 if not – since if for example p3 is not mentioned in S,
then we would be setting v(p3 ) = 0, v(¬p3 ) = 0. /\/\

Theorem 4 (Model existence lemma). Let S ⊂ L be consistent. Then S has a model.

Idea. We want to define v(p) = 1 if p ∈ S, and v(p) = 0 if not. But this fails if some p has
p∈/ S, ¬p ∈
/ S. So we shall try to extend S, keeping it consistent, to ‘swallow up’ one
of p and ¬p, for each p.

Proof. First, for any consistent S ⊂ L and p ∈ L, either S ∪ {p} or S ∪ {¬p} is consistent.
For if not, then S ∪ {p} ⊢ ⊥ and S ∪ {¬p} ⊢ ⊥. But then S ⊢ (p ⇒ ⊥) (deduction
theorem), i.e. S ⊢ ¬p, so S ⊢ ⊥. /\/\

Now, L is countable (e.g. as each Ln is countable), so we can list L as {t1 , t2 , t3 , . . .}.

Set S0 = S. Then set S1 = S0 ∪ {t1 } or S0 ∪ {¬t1 } such that S1 is consistent, then set
S2 = S1 ∪ {t2 } or S1 ∪ {¬t2 } such that S2 is consistent, and continue inductively.

Let S = S0 ∪ S1 ∪ S2 ∪ . . .. Then, for each p ∈ L, we have p ∈ S or ¬p ∈ S. Also, S is

consistent: if S ⊢ ⊥ then, as proofs are finite, we have Sn ⊢ ⊥ for some n. /\/\

Note. S is deductively closed, meaning that if S ⊢ p then p ∈ S. Indeed, if p ∈

/S
then ¬p ∈ S, so S ⊢ (p ⇒ ⊥) and S ⊢ p, whence S ⊢ ⊥. /\/\

1 if p ∈ S
Define v : L → {0, 1}, p 7→
0 if p ∈
/S

Claim. v is a valuation.

Proof of claim. v(⊥) = 0 as ⊥ ∈

/ S (since S is consistent). For v(p ⇒ q):

1. If v(p) = 1, v(q) = 0, we have p ∈ S, q ∈ / S, and we want (p ⇒ q) ∈/ S

(i.e. v(p ⇒ q) = 0). But if (p ⇒ q) ∈ S, then S ⊢ q (modus ponens), whence
q ∈ S (since S is deductively closed). /\/\
2. If v(q) = 1, we have q ∈ S, and we want (p ⇒ q) ∈ S. But ⊢ q ⇒ (p ⇒ q)
(axiom 1), so S ⊢ (p ⇒ q), whence (p ⇒ q) ∈ S (as S is deductively closed).
3. If v(p) = 0, we have ¬p ∈ S, and we want (p ⇒ q) ∈ S. So it is enough to
show that (p ⇒ ⊥) ⊢ (p ⇒ q). By the deduction theorem, it is enough to show
that {p ⇒ ⊥, p} ⊢ q. So it is enough to show that ⊥ ⊢ q. But ⊢ (⊥ ⇒ ¬¬q)
(axiom 1), and ⊢ (¬¬q ⇒ q) (axiom 3), so ⊢ ⊥ ⇒ q. 

5
Remarks. 1. Sometimes Theorem 4 is also called the ‘Completeness Theorem’.

2. The proof used that P is countable (to get L countable). In fact, Theorem 4 remains
true for any P . We shall see this later – it needs Zorn’s Lemma.

By the remark before theorem 4, we now have:

Corollary 5 (Adequacy Theorem). Let S ⊂ L, t ∈ L. Then S |= t implies S ⊢ t. 

Putting Proposition 3 and Corollary 5 together, we get:

Theorem 6 (Completeness Theorem). Let S ⊂ L, t ∈ L. Then S ⊢ t ⇔ S |= t.

Proof. (⇒) Soundness. 

(⇐) Adequacy.

Corollary 7 (Compactness Theorem). Let S ⊂ L, t ∈ L. Then if S |= t then some

finite S ′ ⊂ S has S ′ |= t.

Proof. Trivial if we replace |= by ⊢ (as proofs are finite). 

In particular, if S |= ⊥ (S has no model) then some finite S ′ ⊂ S has S ′ |= ⊥ (S ′ has no

model). Equivalently, if every finite subset of S has a model, then S has a model. (Useful
form of compactness.)

This is actually equivalent to Corollary 7, because S |= t is the same as ‘S ∪ {¬t} has no

model’, and S ′ |= t is the same as ‘S ′ ∪ {¬t} has no model’.

Corollary 7′ (Compactness Theorem, equivalent form). Let S ⊂ L. If every finite

subset of S has a model, then so does S.

Another consequence of completeness is:

Corollary 8 (Decidability Theorem). For finite S ⊂ L and t ∈ L, there is an algorithm

to determine in finite time whether or not S ⊢ t.

Remark. Highly non-obvious.

Proof. Obvious if we replace ⊢ by |=. To check if S |= t, just do a truth table. 

6
 Chapter 2 : Well-Orderings and Ordinals

A total order or a linear order is a pair (X, <), where X is a set and < is a relation on
X that is:

(i) irreflexive : not x < x (∀x ∈ X),

(ii) transitive : x < y, y < z ⇒ x < z (∀x, y, z ∈ X),

(iii) trichotomous : x < y or x = y or y < x (∀x, y ∈ X).

(Note: in (iii), we cannot have more than one: for if x < y, y < x then x < x /\/\.)

Examples. 1. N, usual < . (Note: N = {0, 1, 2, . . .}, so 0 ∈ N. Write N+ for N \ {0}.)

2. Q, usual < .
3. R, usual < .
4. N+ , ‘a < b’ if a|b (and a 6= b) : not trichotomous.
5. P(S), ‘a < b’ if a ⊂ b (and a 6= b) : not trichotomous (if |S| > 1).

Write y > x for x < y, and x 6 y for ‘x < y or x = y’. In terms of 6, a total ordering is:

(i) reflexive : x 6 x (∀x ∈ X),

(ii) transitive : x 6 y, y 6 z ⇒ x 6 z (∀x, y, z ∈ X),

(iii) antisymmetric : x 6 y, y 6 x ⇒ x = y (∀x, y ∈ X),

(iv) trichotomous : x 6 y or y 6 x (∀x, y ∈ X).

A total order (X, <) is a well-ordering if every (non-empty) subset of X has a least element:

∀S ⊂ X, S 6= ∅ ⇒ ∃ x ∈ S such that y > x ∀y ∈ S.

Examples. 1. N (usual order).

2. Z : not a well-ordering.
3. Q : not a well-ordering.
4. R : not a well-ordering.

5. {x ∈ Q : x > 0} : not a well-ordering (consider {x ∈ Q : x > 0}).

6. { 21 , 23 , 43 , . . .}.
1
7. {1 − n : n = 2, 3, . . .} ∪ {1}.
1
8. {1 − n : n = 2, 3, . . .} ∪ {2}.
1 1
9. {1 − n : n = 2, 3, . . .} ∪ {2 − n : n = 2, 3, . . .}.

Remark. A total order X is a well-ordering ⇔ X has no infinite strictly-decreasing sequence.

Indeed, if we have such a sequence x1 > x2 > x3 > . . ., then {x1 , x2 , . . .} has no least
element. Conversely, if S ⊂ X has no least element, then for each x ∈ S there is an
x′ ∈ S with x′ < x. But then x > x′ > x′′ > x′′′ > . . . .

7
Say total orders x, y are isomorphic if there exists a bijection f : X → Y that is order-
preserving (x < y ⇔ f (x) < f (y)).

For example, in the preceding list:

– examples 1 and 6 are isomorphic,

– examples 7 and 8 are isomorphic,

– examples 6 and 7 are not isomorphic (only one has a greatest element).

Proposition 1 (Proof by induction). Let X be well-ordered, and let S ⊂ X such that

for all x ∈ X, the following holds: if y ∈ S for all y < x then x ∈ S. Then S = X.


Equivalently: given some property p(x), if ∀y < x : p(y) ⇒ p(x) (each x), then p(x) ∀x.

Proof. If S 6= X, let x be the least member of X \ S. Then y ∈ S ∀y < x (by choice of x),
whence x ∈ S. /\/\ 

An example of the proof by induction:

Proposition 2. Let X, Y be isomorphic well-orderings. Then there is a unique isomorphism

from X to Y .

Note. False for general total orders. For example: from Z → Z, could take identity or
x → x − 5; from R → R, could take identity or x → x3 .

Proof. Let f, g : X → Y be isomorphisms. We shall show that f (x) = g(x) ∀x ∈ X by

induction on X.

So, given f (y) = g(y) ∀y < x, we need f (x) = g(x). We must have that f (x) = a, the
least element of Y \ {f (y) : y < x} (which 6= ∅, as f (x) ∈ it), because if f (x) > a then
some x′ > x has f (x′ ) = a (f bijective), contradicting f order-preserving.

Similarly, g(x) = a, so f (x) = g(x) as required. 

A subset I of a total order X is an initial segment if x ∈ I, y < x ⇒ y ∈ I.

For example, for any x ∈ X, the set Ix = {y ∈ X : y < x} is an initial segment. Not every
proper initial segment of X need be of this form. For example: in R, could take {x : x 6 3};
in Q, could take {x : x 6 0 or x2 < 2}.

Remark. In a well-ordering X, every proper initial segment I is of the form Ix , some x ∈ X.

Indeed, let x be the least member of X \ I. Then y < x implies y ∈ I (by choice of x).
Also, y ∈ I implies y < x (if y = x or y > x we should get x ∈ I).

Aim. Every subset of X (well-ordered) is isomorphic to some initial segment of X.

Note. False for general total orders. For example: could take {1, 3, 4} in Z; or could take
Q in R.

For f : A → B and C ⊂ A, the restriction of f to C is f |C = {(x, f (x)) : x ∈ C}.

8
Theorem 3 (Definition by recursion). Let X be a well-ordered set, and Y any set. Then
for any G : P(X × Y ) → Y, there exists f : X → Y such that f (x) = G(f |Ix ) for all
x ∈ X. Moreover, f is unique.

Note. In defining f (x), we make use of f on Ix = {y : y < x}.

Proof. (Existence) Define ‘h is an attempt’ to mean the following (‘the clever bit’).

h : I → Y, some initial segment I of X, and x ∈ I ⇒ h(x) = G(h|Ix ).

Note that if h, h′ are attempts, both defined at x, then h(x) = h′ (x), by induction: for
if h(y) = h′ (y) ∀y < x, then certainly h(x) = h′ (x).

Also, for each x, there exists an attempt defined at x, again by induction. Indeed,
suppose that for all y < x, there is an attempt defined at y. Then there exists a unique
S
attempt hy defined on {z : z 6 y}. Put h = y<x hy : an attempt defined on Ix .

Then h ∪ {(x, G(h))} is an attempt defined at x (single-valued, by uniqueness).

Now define f by: f (x) = y if there is an attempt h with h defined at x and h(x) = y.

(Uniqueness) If f, f ′ satisfy the conditions then f (x) = f ′ (x) ∀x, by induction on x.

(As if f (y) = f ′ (y) for all y < x then f (x) = f ′ (x).) 

Proposition 4 (Subset collapse). Let X be a well-ordered set, and Y ⊂ X. Then Y is

isomorphic to an initial segment of X. Moreover, this initial segment is unique.

Proof. To obtain f : Y → X that is order-preserving and having image an initial segment,

we need f (x) = min X \ {f (y) : y ∈ Y, y < x} for all x ∈ Y .

Note that we cannot have X \ {f (y) : y < x, y ∈ Y } = ∅ – for example, because

f (y) 6 y for all y (by induction), so that x ∈
/ {f (y) : y < x, y ∈ Y }.

So done (existence and uniqueness) by recursion (Theorem 3) applied to Y . 

In particular, X cannot be isomorphic to a proper initial segment of X – by uniqueness in

subset collapse, since X is isomorphic to X.

How do different well-orderings relate to each other?

For well-orderings X, Y , write X 6 Y if X is isomorphic to an initial segment of Y .

E.g. If X = N, Y = { 21 , 32 , 43 , . . ., } ∪ {2}, then X 6 Y .

Theorem 5. Let X, Y be well-orderings. Then X 6 Y or Y 6 X.

Proof. Suppose Y 66 X. We need X 6 Y . For f : X → Y to be an isomorphism with an

initial segment of Y , need: for all x ∈ X, f (x) = min Y \ {f (y) : y < x}.

Note that we cannot have Y \ {f (y) : y < x} = ∅ as then Y is isomorphic to Ix . /\/\.

So done by recursion. 

9
Proposition 6. Let X, Y be well-orderings with X 6 Y , Y 6 X. Then X, Y are isomorphic.
(‘The best we could ever hope for.’)

Proof. Let f : X → Y be an isomorphism from X to an initial segment of Y , and g : Y → X

be an isomorphism from Y to an initial segment of X. Then g ◦ f is an isomorphism
from X to an initial segment of X (as an initial segment of an initial segment is an
initial segment), whence g ◦ f : X → X is the identity on X (by uniqueness).

Similarly, f ◦ g is the identity of Y . So f : X → Y is a bijection. 

New well-orderings from old

Say X < Y if X 6 Y but X is not isomorphic to Y . Equivalently, X < Y ⇒ X isomorphic

to a proper initial segment of Y .

Find a bigger one. Given a well-ordering X, choose x ∈ / X, and define a well-ordering of

X ∪ {x} by setting y < x for all y ∈ X. This is the successor of X, written X + . Clearly
X < X +.

Put some together. Given a set {Xi : i ∈ I} of well-orderings, seek well-ordering X with
X > Xi for all i. Given well-orderings (X, <X ) and (Y, <Y ), say X extends Y if Y ⊂ X, and
<X and <Y agree on Y , and Y is an initial segment of X. Say {Xi : i ∈ I} are nested if for
all i, j, either Xi extends Xj or Xj extends Xi .

Proposition 7. Let {Xi : i ∈ I} be a nested set of well-orderings. Then there exists a

well-ordering X with X > Xi for all i.
S
Proof. Let X = i Xi . Put x < y if for some i we have x, y ∈ Xi and x <i y (where <i is
S
the well-ordering on Xi ). [ Equivalently, < = i <i . ]

Clearly < is a total order on X, with each Xi an initial segment (by nestedness). Also,
given non-empty S ⊂ X, have S ∩ Xi 6= ∅, some i. Then S ∩ Xi has a minimal member,
x say (as Xi well-ordered). So x is minimal in S (as Xi an initial segment of X). Thus
< is a well-ordering on X, and X > Xi for all i. 

Remark. Proposition 7 also holds when the Xi are not nested.

Ordinals

‘Is the collection of all well-orderings itself a well-ordering?’

An ordinal is a well-ordered set, with two being regarded as the same if they are isomorphic.

(Just as the rationals consist of all symbols m

n (m, n ∈ Z, n 6= 0) with two regarded as the
same if mn′ = m′ n. But we cannot formalise ordinals using equivalence classes – see later.)

If X is a well-ordered set, corresponding to an ordinal α, say X has order-type α.

Examples. For k ∈ N, write k for the order-type of the (unique) well-ordering of size k.
Write ω for the order-type of N. Then in R, {1, 3, 4, 7} has order-type 4, while { 21 , 23 , 34 , . . .}
has order-type ω.

10
Write α 6 β if there exists X of order-type α and Y of order-type β with X 6 Y . (Note:
this does not depend on the choice of X, Y .) Similarly for α < β, α+ , etc.

Thus, for all α, β, we have α 6 β or β 6 α, and α 6 β, β 6 α ⇒ α = β.

Theorem 8. Let α be an ordinal. Then the ordinals < α form a well-ordered set, of order-
type α.

Proof. Let X be a well-ordering of order-type α. Then the well-orderings < X are precisely
(up to isomorphism) the proper initial segments of X. But these are the Ix , x ∈ X,
and hence biject with X in an order-preserving way (x ↔ Ix ). 

Sometimes write Iα for {β : β < α}. (‘A nice set of order-type α.’)

Proposition 9. Let S be a non-empty set of ordinals. Then S has a least element.

Proof. Choose α ∈ S. If α is minimal, done. If α is not minimal, i.e. S ∩ Iα 6= ∅, take a

minimal β in S ∩ Iα (Iα is well-ordered, by Theorem 8). Then β is minimal in S. 

Theorem 10 (Burali-Forti paradox). The ordinals do not form a set.

Proof. Suppose not: let X be the set of all ordinals. Then X is a well-ordering, say of
order-type α. But then X is isomorphic to Iα , a proper initial segment of X. /\/\ 

Given a set S = {αi : i ∈ I} of ordinals, it has an upper bound α (i.e. α > αi for all i) by
applying Proposition 7 to the nested family {Iαi : i ∈ I}. Hence, by Proposition 9, S has a
least upper bound, written sup S. E.g., sup{2, 4, 6, 8, . . .} = ω.

On the following page are some ordinals.

Every ordinal in that picture is countable (as a countable union of countable sets is countable).
Is there an uncountable ordinal? I.e., is there an uncountable well-ordered set?

We can well-order: N, with the usual order;

Q, by bijecting with N;
R..? Not obvious.

Amazingly, we can show:

Theorem 11. There is an uncountable ordinal.

Idea. Look at {α ∈ ON : α countable}, where ‘α ∈ ON ’ means ‘α is an ordinal’. (Is this a

set? See Burali-Forti.)

Proof. Let A = {R ∈ P(N × N) : R is a well-ordering of a subset of N}, and let B = {order-

type (R) : R ∈ A}. So B is precisely the set of all countable ordinals.

Let ω1 = sup B. If ω1 is countable, then it is the greatest countable ordinal (definition

of B), contradicting ω1 < ω1+ . So ω1 is uncountable. 

11
Some ordinals.
:
0 ω ω2
1 :
2 ω ω3
3 :
: ω ω4
ω :
ω + 1 ← officially ω + 2
ω ω = ω (ω )
2

ω+2 :
ω+3 ω2 2
: ω
ω + ω = ω2 ← officially sup{ω, ω + 1, ω + 2, . . .} :
2
ω2 + 1 ωω 3
ω2 + 2 :
3
ω2 + 3 ωω
: :
4
ω3 ωω
: :ω
ω4 ωω
: :
ω2
ω5 ωω
: :
ωω = ω 2 ← officially sup{ω, ω2, ω3, . . .} ω ωω
3

ω2 + 1 :ω
ω2 + 2 ωω
ω
:
2
ω +w : ..
.
2 ω ωω
ω +w+1 w ω
officially sup{ω, ω , ω , . . .} → ω ω
= ǫ0
: ǫ0 + 1
ω 2 + w2 ǫ0 + 2
: :
ω2 + ω2 = ω22 ǫ0 + ω
ω22 + 1 :
: ǫ0 + ǫ0 = ǫ0 2
ω22 + w :
: ǫ0 3
ω23 :
: ǫ0 ω
ω24 :
: ǫ0 ǫ0 = ǫ20
ω3 :
3
: ǫ 0
ω 23 :
: ǫω 0
:
ω33 2

: ǫω0
ω4 :
ω3
: ǫ 0
ω5 :
ωω
: ǫ 0
: ..
ωω ← officially sup{ω, ω 2 , ω 3 , . . .} .
ωω
: ǫ0ω = ǫǫ00
ωω 2 :ǫ
: ǫ 0
ω
ǫ00
ω 3 :ǫ
: ǫ
ǫ0 0

ω ω ω = ω ω+1 ǫ0 0
: : ..
ω+2 ǫ.
ω ǫ 0
ǫ0 0
: ǫ0 = ǫ1
ω ω+3
‘ ... and so on ... ’
12
Note that by construction ω1 is the least uncountable ordinal (definition of B). Note also
that every ordinal in the picture is < ω1 . Two surprising properties of the total order ω1 are:

• it is an uncountable ordering in which for each x, the set {y : y < x} is countable;

• every sequence α1 , α2 , . . . in ω1 is bounded above, e.g. by sup{α1 , α2 , . . .}.

Theorem 11′ (Hartogs’ Lemma). For any set X, there is an ordinal α that does not
inject into X.

Proof. The same as in Theorem 11, with ‘X’ in place of ‘N’. 

The least such α is denoted γ(X). E.g., ω1 = γ(ω).

Successors and Limits

Let α be an ordinal. Does α have a greatest element? (That is, any set of order-type α, such
as Iα = {β : β < α}.)

If yes: let β be the greatest element, then γ < α ⇒ γ < β or γ = β (and converse trivially),
so α = β + . Say α is a successor.

If no: for all β < α, there is some γ < α such that γ > β, so α = sup{β : β < α}. Say α is
a limit.

E.g. 5 is a successor (5 = 4+ ),
ω + is a successor,
ω is a limit (no greatest element of {γ : γ < ω}),
(0 is a limit).

Ordinal Arithmetic

Define α + β (for ordinals α and β) by recursion on β (α fixed) as follows:

α+0 = α
α + β + = (α + β)+
α+λ = sup{α + γ : γ < λ} for λ a (non-zero) limit

E.g. ω + 1 = ω + 0+ = (ω + 0)+ = ω + ,
ω + 2 = (ω + 1)+ = ω ++ ,
1 + ω = sup{1 + γ : γ < ω} = ω

So + is not commutative.

Remark. Officially, since the ordinals are not a set, ‘recursion on the ordinals’ would mean:
to define α + β (given α, β), define α + γ, all γ 6 β, by recursion on the set {γ : γ 6 β}
(plus uniqueness). Similarly for proof by induction on the ordinals: if p(α) false for
some α, then p is not everywhere true on {γ : γ 6 α}.

13
Notes. 1. β 6 γ ⇒ α + β 6 α + γ (induction on γ)

2. β < γ ⇒ α + β < α + γ, because:

β < γ ⇒ β + 6 γ ⇒ α + β + 6 α + γ ⇒ (α + β)+ 6 α + γ ⇒ α + β < α + γ.

3. But 1 < 2 and yet 1 + ω = 2 + ω (= ω).

Proposition 12. For all α, β, γ ∈ ON , α + (β + γ) = (α + β) + γ.

Proof. By induction on γ (α, β fixed).

0. α + (β + 0) = α + β = (α + β) + 0.

Successors. (α+β)+γ + = ((α+β)+γ)+ = (α+(β+γ))+ = α+(β+γ)+ = α+(β+γ +).

Limits. For λ a non-zero limit,

(α + β) + λ = sup{(α + β) + γ : γ < λ} = sup{α + (β + γ) : γ < λ}.

What about α + (β + λ) ?

Claim. β + λ a limit.

Proof of claim. Have β + λ = sup{β + γ : γ < λ}, but γ < λ ⇒ ∃ γ ′ < λ with
γ ′ > γ ⇒ β + γ ′ > β + γ, so there is no greatest member of {β + γ : γ < λ}.
So sup{β + γ : γ < λ} is a limit.

So α + (β + λ) = sup{α + δ : δ < β + λ}. So our task is:

sup{α + (β + γ) : γ < λ} = sup{α + δ : δ < β + λ}.

6: For γ < λ, have β + γ < β + λ, so set on left ⊂ set on right.

>: For δ < β + λ, have δ 6 β + γ, some γ < λ (definition of β + λ), whence

α + δ 6 α + (β + γ). Thus each element of the RHS is 6 some element of the
LHS. 

Another viewpoint

The definition of ordinal addition given above is called the ‘inductive’ definition. There is
also a ‘synthetic’ definition: α + β is defined to be the order of α ⊔ β (disjoint union of α and
β, e.g. (α × {0}) ∪ (β × {1})) with all of α coming before all of β.

E.g. ω+1 = ω • = ω+,

←→
1+ω = • ω = ω.
←→

Armed with the synthetic definition, it is easy to see that α + (β + γ) = (α + β) + γ: each is

the order-type of α β γ .
←→ ←→ ←→

We must verify:

14
Proposition 13. The two definitions of ordinal addition coincide.

Proof. Write + for inductive, +′ for synthetic. We shall show α + β = α +′ β for all α, β,
by induction on β.

0. α + 0 = α = α +′ 0

Successors. α +′ β + = order-type of α β • = (α +′ β)+ = (α + β)+ = α + β +

←→ ←→

λ a non-zero limit. α + λ = sup{α + γ : γ < λ} = sup{α +′ γ : γ < λ} = α +′ λ.

(Nested union, so sup = union.) 

Moral. Synthetic is easier to use, if it is available.

Ordinal Multiplication

Define αβ by recursion on β as follows:

α0 = 0
α(β + ) = αβ + α
αλ = sup{αγ : γ < λ} for λ a (non-zero) limit

E.g. ω2 = ω1 + ω = (ω0 + ω) + ω = ω + ω,
2ω = sup{2γ : γ < ω} = ω (so multiplication is not commutative),
ωω = sup{ωγ : γ < ω} = sup{0, ω, ω + ω, ω + ω + ω, . . .} (as in our big picture).

Or synthetically: αβ is order-type of α × β, with (x, y) < (z, t) if y < t or y = t, x < z.

 .
.
 .


‘β copies of α – go up in rows’ β ←→α


 α
←→

 ω
←→
E.g. ω2 = order-type of = ω+ω
 ω
←→
 ..


 .
2ω = order-type of •• = ω.
 ←→

 ••
←→

Can check that definitions coincide, and (αβ)γ = α(βγ), etc.

Could also define exponentiation, towers, and so on. For example, ordinal exponentiation:
define αβ by recursion on β:
α0 = 1
(β + )
α = αβ α
αλ = sup{αγ : γ < λ} for λ a (non-zero) limit

E.g. ω 2 = ω 1 ω = (ω 0 ω)ω = (1ω)ω = ωω,

2ω = sup{2γ : γ < ω} = ω – note, this is countable.

15
 Chapter 3 : Posets and Zorn’s Lemma

A partially ordered set, or poset, is a pair (X, 6), where 6 is a relation on X that is:

(i) reflexive : x 6 x (∀x ∈ X),

(ii) transitive : x 6 y, y 6 z ⇒ x 6 z (∀x, y, z ∈ X),

(iii) antisymmetric : x 6 y, y 6 x ⇒ x = y (∀x, y ∈ X).

Equivalently, writing x < y to mean ‘x 6 y and x 6= y’, the conditions are:

(i) irreflexive : not x < x (∀x ∈ X),

(ii) transitive : x < y, y < z ⇒ x < z (∀x, y, z ∈ X).

Examples. 1. Any total order.

2. (N+ , ‘divides’). (Not a total order: 3 and 5 are incomparable.)

3. For S any set, take P(S), with A 6 B if A ⊂ B. (Very important.)

4. Take X = any subset of P(S), same 6 .

E.g. V a vector space, X = all subspaces of V .

5.
c❜❜s ✧
s s
s e
✧
b ❜❜✧s
✧d
a
Meaning: a 6 b, b 6 c, a 6 d, d 6 e, and everything following by transitivity.
E.g. a 6 c, but b and d are not related.

In general, a Hasse diagram of a poset consists of a drawing of the points of X with an

upward line from x to y if y covers x (meaning y > x and no z has y > z > x).

Hasse diagrams can be useful, e.g. (N, 6), or useless, e.g. (Q, 6) – no rational covers another!

Examples (continued).

6. c s
❩
✡✡ ❩se
b s✡ (so no notion of ‘height’ or ‘rank’)
❏❏ sd
s✚
❏✚
a

7. e s
s ✚❩❩sd
c ✚
❩ ✚
❩✚
✚ ❩
a s✚ ❩sb

8. • • • • •
a b c d e

16
A subset S of a poset X is a chain if it is a total order (∀x, y ∈ S : x 6 y or y 6 x).

E.g. – In a total order, every subset is a chain.

– In example 2, {1, 2, 4, 8, 16, . . .} is a chain.

– In example 5, {a, b, c} is a chain, as is {a, b}.

– Chains can be uncountable, e.g. (R, 6).

Say S is an antichain if no two members of S are related (∀x, y ∈ S, not x < y.)

E.g. – In example 2, {p : p prime} is an antichain.

– In example 5, {b, e} is an antichain.

– In example 7, {e} is an antichain, as is {a, b}.

– In example 8, the whole of X is an antichain.

For S ⊂ X, say x ∈ X is an upper bound for S if y 6 x for all y ∈ S, and say x is a least
upper bound or supremum for S if x is an upper bound for S, and every upper bound y
for S has y > x.
√
E.g. – In R, S = {x : x2 < 2} has 7 as an upper bound, and has a least upper bound, 2.
√ W √ W
Write sup S = 2, or S = 2 – ‘join of S’. (‘ is suggestive of union’)

– In Q, S = {x : x2 < 2} has 7 as an upper bound, but no supremum.

– In example 5, sup{a, b, c} is c (the only upper bound), and {b, d} has no upper bound.

– In example 7, {a, b} has upper bounds c, d, e, so no least upper bound.

W
Note. S may or may not belong to S. E.g. in R, sup ∈ {x : x 6 1}, but sup ∈
/ {x : x < 1}.

A poset X is complete if every set S ⊂ X has a supremum.

E.g. – (R, 6) is not complete – e.g. Z has no upper bound. (Note: different from ‘complete-
ness’ in ‘metric space’ sense.)

– [0, 1] is complete, but (0, 1) is not complete – e.g. (0, 1) itself has no supremum.

– Q is not complete – e.g. {x : x2 < 2} or Q itself.

S
– P(S) is always complete: given sets Ai for i ∈ I, take i∈I Ai .

Note. In any complete poset X, there is a greatest element (an x with x > y for all y),
namely sup X, and also a least element (an x with x 6 y for all y), namely sup ∅.

For a poset X, a function f : X → X is order-preserving if x 6 y ⇒ f (x) 6 f (y).

E.g. – On N, f (n) = n + 1.

– On [0, 1], f (x) = 1 − 21 (1 − x) (‘halve the distance to 1’).

– On P(S), f (A) = A ∪ {j}, some fixed j ∈ S – clearly A ⊂ B ⇒ f (A) ⊂ f (B).

17
Say x ∈ X is a fixed point of f if f (x) = x. Not every order-preserving f has a fixed point
– e.g. f (n) = n + 1 on N.

Theorem 1 (Knaster-Tarski fixed point theorem). Let X be a complete poset, and

f : X → X order-preserving. Then f has a fixed point.

Proof. Let E = {x ∈ X : x 6 f (x)}, and let s = sup E. We’ll show that f (s) = s.

To show s 6 f (s), enough to show that f (s) an upper bound for E (then s 6 f (s) as
s is the least upper bound). But x ∈ E ⇒ x 6 s ⇒ f (x) 6 f (s) ⇒ x 6 f (x) 6 f (s).

To show f (s) 6 s, enough to show that f (s) ∈ E (as s is an upper bound for E). But
s 6 f (s), so f (s) 6 f (f (s)) (as f order-preserving), i.e. f (s) ∈ E. 

An application of Knaster-Tarski:

Corollary 2 (Schröder-Bernstein Theorem). Let A, B be sets, and let f : A → B and

g : B → A be injections. Then there exists a bijection from A to B.

Proof. Want to write A = P ⊔ Q, B = R ⊔ S such that f |P bijects P with R, and g|S bijects
S with Q. (Then done: define h : A → B by taking h = f on P and g −1 on Q.)

So we want P ⊂ A such that A \ g(B \ f (P )) = P .

Let X = P(A), and define θ : X → X, P 7→ A \ g(B \ f (P )). Need a fixed point of θ.

But X is complete and θ is order-preserving: P ⊂ P ′ ⇒ θ(P ) ⊂ θ(P ′ ), so done by
Knaster-Tarski. 

Zorn’s Lemma

For X a poset, x ∈ X, say x is maximal if no y ∈ X has y > x.

E.g. – In example 5 earlier, c and e are maximal.

– N, Q, R have no maximal elements – but then, they have chains without upper bounds.

Theorem 3 (Zorn’s Lemma). Let X be a (non-empty) poset in which every chain has an
upper bound. Then X has a maximal element.

Proof. Suppose X has no maximal element. So for each x ∈ X there is x′ ∈ X with x′ > x.
We know that every chain C has some upper bound u(C).

Let γ = γ(X) (as guaranteed by Hartogs’ Lemma).

Pick some x ∈ X, and define xα , α < γ recursively by:

x0 = x
xα+ = x′α
xλ = u({xα : α < λ}) for λ a non-zero limit
(note {xα : α < λ} is a chain, by induction)

Then the xα , α < γ, are distinct, so we have injected γ into X. /\/\ 

18
Remarks. 1. We could define xλ = u({xα : α < λ})′ to avoid thinking about injectivity.

2. Proof was easy given well-orderings, definition by recursion, etc, from chapter 2.

A typical application of Zorn is: does every vector space have a basis?

Recall. A basis is a linearly independent (‘no finite linear combination = 0’) spanning set
(‘everything is a finite sum from the set’).

Examples. 1. V = set of all real polynomials. Then {1, x, x2 , x3 , . . .} is a basis.

2. V = set of all real sequences. Let en = (0, 0, . . ., 0, 1, 0, 0, . . .). These are linearly
independent, but they are not spanning: e.g., (1, 1, 1, . . .) is not in the span. So they
do not form a basis. In fact, there is no countable basis (easy exercise). Even more: it
is impossible to give an ‘explicit’ basis.

3. R as a vector space over Q..? A basis here is called a Hamel basis.

Theorem 4. Every vector space V has a basis.

Proof. (‘Go for a maximal linearly independent set.’)

Let X = {A ⊂ V : A is linearly independent}, ordered by ⊂. Seek a maximal element

A ∈ X. (Then done: if A does not span then choose x not in the span of A – then
A ∪ {x} is linearly independent. /\/\)
S
Given chain {Ai : i ∈ I}, let A = i∈I Ai . Then A ⊃ Ai for all i, so just need A ∈ X,
i.e. A linearly independent.

Suppose we have a linear dependence in A, say λ1 x1 +. . .+λn xn = 0, where x1 , . . ., xn ∈

A and λ1 , . . ., λn scalars, not all 0. Have x1 ∈ Ai1 , . . ., xn ∈ Ain , some i1 , . . ., in ∈ I.
But some Aik has Ai1 , . . ., Ain ⊂ Aik (as the Ai are a chain), contradicting Aik linearly
independent. /\/\

So, by Zorn, there is a maximal A ∈ X. 

Remarks. 1. The only actual ‘maths’ (i.e. linear algebra) we did was in the final check.
This is very typical of Zorn.

2. ‘(non-empty)’ is not strictly needed in statement of Zorn (as ∅ has an upper bound,
so X 6= ∅). However, it’s often safe to check X 6= ∅, as our chains should be non-empty.

Another application of Zorn: completeness theorem for propositional logic, without the as-
sumption that the primitive propositions are countable.

Theorem 5. Let S ⊂ L(P ), any set P . Then S consistent ⇒ S has a model.

Proof. Seek consistent S ⊃ S such that ∀t ∈ L(P ) have t ∈ S or (¬t) ∈ S.

(Then done, by setting v(p) = 1 if p ∈ S and v(p) = 0 if not – as in chapter 1.)

We seek a maximal consistent S ⊃ S. (If t ∈/ S then S ∪ {t} ⊢ ⊥, so S ⊢ (t ⇒ ⊥)

(deduction theorem), so (t ⇒ ⊥) ∈ S by maximality of S, i.e. (¬t) ∈ S.)

19
 Let X = {T ⊂ L(P ) : T consistent, T ⊃ S}, ordered by ⊂.
S
First, X 6= ∅, as S ∈ X. Given a non-empty chain {Ti : i ∈ I} in X, let T = i∈I Ti .
Then T ⊃ Ti ∀i, so just need T ∈ X.

Have T ⊃ S (as chain non-empty). If T inconsistent, have t1 , . . ., tn ∈ T with

{t1 , . . ., tn } ⊢ ⊥ (proofs are finite). Then t1 ∈ Ti1 , . . ., tn ∈ Tin , some i1 , . . ., in ∈ I,
whence t1 , . . ., tn ∈ Tik for some Tik (as the Ti form a chain), contradicting Tik consis-
tent. /\/\

So, by Zorn, X has a maximal element. 

One more application of Zorn.

Theorem 6 (Well-ordering Principle). Every set S can be well-ordered.

Remark. Very surprising for S = R, for example.

Proof. Let X = {(A, R) : A ⊂ S, R is a well-ordering of A}, ordered by extension. (That

is, (A, R) > (A′ , R′ ) if A′ ⊂ A, R and R′ agree on A′ , and A′ is an initial segment of
A in the ordering R.)

First, X 6= ∅ since ∅ well-orderable, i.e. (∅, ∅) ∈ X.

S S

Given a chain {(Ai , Ri ) : i ∈ I}, the (Ai , Ri ) are a nested family, so i∈I Ai , i∈I Ri
is an upper bound (as in chapter 2).

So, by Zorn, X has a maximal element, say (A, R).

Claim. A = S

Proof of claim. Suppose A 6= S. Choose x ∈ S \ A, and take the successor of A :

define a well-ordering on A ∪ {x} by setting x > y ∀y ∈ A.

This contradicts the maximality of A. /\/\ 

Zorn’s Lemma and the Axiom of Choice

In our proof of Zorn’s Lemma, we chose, for each x, an x′ greater than it. This is making
infinitely many arbitrary choices. We did the same in Part IA, when proving that a countable
union of countable sets is countable: given sets A1 , A2 , A3 , . . ., each listable, we chose a listing
of each.

In terms of ‘rules for building sets’, we are appealing to the axiom of choice, which states:
‘can choose an element from each of a family {Ai : i ∈ I} of non-empty sets’, or more
precisely:

every family {Ai : i ∈ I} of non-empty sets has a choice function, meaning an

S
f : I → i∈I Ai such that f (i) ∈ Ai for all i.

20
This is of a different character from the other rules for building sets (e.g. ‘given A and B,
can form A ∪ B’, or ‘given A, can form P(A)’), in that the object whose existence is asserted
is not uniquely specified by its properties (as opposed to, e.g., A ∪ B).

So it is often of interest to know: does a given proof involve AC or not?

Remark. AC trivial if |I| = 1 (A 6= ∅ means that there is x such that x ∈ A)

Similarly, for |I| finite, by induction on |I|.

However, for general I it turns out that AC cannot be deduced from the other set-
building rules.

Does the proof of Zorn’s Lemma need AC?

Yes, because we can deduce AC from Zorn’s Lemma (using only the other set-building rules):

Given {Ai : i ∈ I}, each Ai 6= ∅, a partial choice function is a function f : J →

S
i∈I Ai , some J ⊂ I, such that f (j) ∈ Aj for all j ∈ J.

S
Let X = {(J, f ) : J ⊂ I, f a partial choice function J → i∈I Ai }, ordered by extension:
(J, f ) 6 (J ′ , f ′ ) if J ⊂ J ′ and f ′ |J = f .

First, X 6= ∅ since (∅, ∅) ∈ X.

S S
Given a chain {(Jq , fq ) : q ∈ Q} has upper bound ( q∈Q Jq , q∈Q fq ). So, by Zorn,
have maximal (J, f ) ∈ X.

Want J = I. If J 6= I, choose i ∈ I \J, choose x ∈ Ai , and consider (J ∪{i}, f ∪{(i, x)})

– contradicts maximality of (J, f ). /\/\

Conclusion. ZL ⇔ AC (given other set-building rules)

Actually, we had well-ordering principle implied by Zorn, and trivially WO ⇒ AC (well-order

S
i∈I Ai , and let f (i) = least element of Ai ).

Also, AC ⇒ WO (without going via ZL, as in Theorem 6):

Let f be a choice function for {A ⊂ X : A =6 ∅}. Define xα , α < γ(X), recursively

by: having defined xβ , each β < α, if {xβ : β < α} = X then stop, otherwise set
xα = f (X − {xβ : β < α}).

This must stop, else we have injected γ(X) into X – contradiction. So we have an
injection from X to some well-ordered set (an initial segment of γ(X)).

Conclusion. ZL ⇔ AC ⇔ WO (given the other set-building rules).

Remark. Zorn is hard to prove from first principles because we need ordinals, recursion,
Hartogs, etc., and not because ZL ⇔ AC.

21
** Non-examinable section **

Some notions related to completeness

1. Chain-completeness and Bourbaki-Witt

A poset X is chain-complete if X 6= ∅ and every non-empty chain has a least upper bound.

E.g. – Any complete poset.

– Any finite (6= ∅) poset X, each chain has a greatest element.

– X = {A ⊂ V : A is linearly independent}, any vector space V .

Say f : X → X is inflationary if f (x) > x ∀x ∈ X.

Bourbaki-Witt Theorem: every inflationary f on a chain-complete X has a fixed point.

Follows instantly from Zorn: X has a maximal element x, and x 6 f (x) implies x = f (x).

f f f
Can prove Bourbaki-Witt without AC: x0 → x1 → . . . → xω , . . .. (Note: we did not use AC
in any ordinal theory, except a remark that ω1 is not a countable sup.)

In fact, easy to get from B-W to ZL (using AC). So can view B-W as ‘choice-free part of ZL’.

2. Lattices and Boolean Algebras

A lattice is a poset X in which every finite subset has a least upper bound and a greatest
lower bound.

E.g. – P(S), any set S (as complete).

– {A ⊂ N : A finite or N \ A finite}.

For a, b ∈ X, write a ∨ b (‘a join b’), for the least upper bound of {a, b}; and write a ∧ b
(‘a meet b’), for the greatest lower bound of {a, b}

E.g. In P(S), A ∨ B = A ∪ B, A ∧ B = A ∩ B.

A lattice X, say with greatest element 1 and least element 0, is a Boolean algebra if ‘X
behaves like P(S)’ :

(i) a ∧ (b ∨ c) = (a ∧ b) ∨ (a ∧ c) (∀a, b, c, ∈ X)

(ii) a ∨ (b ∧ c) = (a ∨ b) ∧ (a ∨ c) (∀a, b, c, ∈ X)

(iii) ∀a ∈ X, ∃ a′ ∈ X such that a ∨ a′ = 1, a ∧ a′ = 0.

E.g. P(S), any S.

Fact. Every finite Boolean algebra is isomorphic to P(S), some finite S.

Not true in general, e.g. {A ⊂ N : A finite or N \ A finite}. (Countably infinite, so not P(S).)

22
Example: Lindenbaum algebra

Take propositional language L, with ∼ defined by: p ∼ q if ⊢ (p ⇔ q).

Have 6 on L/ ∼, defined by [p] 6 [q] ⇔ ⊢ (p ⇒ q).

Then L/ ∼ is a Boolean algebra: [p] ∧ [q] = [p ∧ q]

[p] ∨ [q] = [p ∨ q]
[p]′ = [¬p]

** End of non-examinable section **

23
 Chapter 4 : Predicate Logic

Overview of the set-up

Recall that a group is a set A, equipped with functions m : A2 → A (of ‘arity’ 2), i : A1 → A
(arity 1), and a constant e ∈ A (arity 0, i.e. e : A0 → A), satisfying:


(∀x, y, z) m(x, m(y, z)) = m(m(x, y), z


(∀x) m(x, e) = x ∧ m(e, x) = x


(∀x) m(x, i(x)) = e ∧ m(i(x), x) = e

and a poset is a set A, equipped with a predicate 6 ⊂ A2 (arity 2), satisfying:

(∀x) (x 6 x) (i.e., (x, x) ∈ 6)

(∀x, y, z) ((x 6 y ∧ y 6 z) ⇒ x 6 z)
(∀x, y) ((x 6 y ∧ y 6 x) ⇒ x = y)

Propositional Logic Predicate Logic

E.g. language of groups

Language −→
(things like (1), (2), (3) above)

Structure: a set equipped with

Valuation −→
functions, relations of right arities

A model for S A model for S

−→
(valuation in which each s ∈ S holds) (structure in which each s ∈ S holds)

S |= t Same (e.g. should have

−→
(every model for S is a model for t) {group axioms} |= m(e, e) = e)

Same
S⊢t −→
(but a bit more complicated)

Let Ω (function symbols) and Π (relation symbols) be disjoint sets, and let α (‘arity’) be
α : Ω ∪ Π → N.

The language L = L(Ω, Π, α) is the set of formulae, defined as follows.

• Variables. Have variables x1 , x2 , x3 , . . .. (We may use x, y, . . ..)

• Terms. Defined inductively by:

(i) Every variable is a term.

(ii) If f ∈ Ω, α(f ) = n and t1 , . . ., tn terms, then so is f t1 . . .tn .
(Can insert brackets and commas if desired.)

E.g. In language of groups:

Ω = {m, i, e}, Π = ∅, α(m) = 2, α(i) = 1, α(e) = 0.

Some terms: x1 , m(x1 , x2 ), i(m(x1 , x2 )), e, m(x1 , e).

24
 • Atomic formulae.

(i) ⊥ is an atomic formula.

(ii) If s, t terms then (s = t) is an atomic formula.
(iii) If φ ∈ Π, α(φ) = n, and t1 , . . ., tn terms, then φ(t1 t2 . . .tn ) is an atomic formula.

E.g. In language of groups: x1 = x2 , m(x1 , x1 ) = e.

In language of posets, (with Ω = ∅, Π = {6}, α(6) = 2):
x1 = x1 , x1 6 x2 (officially ‘6 (x1 , x2 )’).

• Formulae. Defined inductively by:

(i) Every atomic formula is a formula.

(ii) If p, q formulae, then so is (p ⇒ q).
(iii) If p a formula and x a variable, then (∀x)p is a formula.

E.g. In language of groups:

(∀x)(m(x, e) = x), (∀x)(i(x) = x), (∀x) m(x, x) = e ⇒ (∃y)(m(y, y) = e) .

Notes. 1. A formula is a string of symbols.

2. Can define ‘¬p’ to mean ‘p ⇒ ⊥’, and similarly p ∨ q, p ∧ q, etc, and ‘(∃ x)p’ to mean
‘¬(∀x)(¬p)’.

Closed terms. A term is closed if it contains no variables.

E.g. In language of groups: e, m(e, e), m(e, i(e)), but not m(x1 , i(x1 )).

Free and bound variables. An occurrence of a variable x in a formula p is called bound

if it is inside the brackets of a ‘∀x’ quantifier. Otherwise, it is free.



E.g. m(x, x) = e ⇒ (∃ y) m(y, y) = x , (∀x) m(x, x) = e
↑ ↑ տ րր ↑ տ րր
free bound free bound

m(x, x) = e ⇒ (∀x)(∀y) m(x, y) = m(y, x) ←− unhelpful
↑ ↑ ↑ ↑ ↑
free bound bound bound

Sentences. A sentence is a formula with no free variables.

E.g. (∀x)(m(x, x) = e).

Substitution. For p a formula, x a variable, t a term, write p[t/x] for the formula obtained
by substituting t for each free occurence of x.

E.g. If p is (∃ y)(m(y, y) = x), then p[e/x] is (∃ y)(m(y, y) = e).

25
Semantic Entailment

Let L = L(Ω, Π, α) be a language. An L-structure is a non-empty (see later for why) set
A, together with

(i) for each f ∈ Ω, a function fA : An → A, where n = α(f );

(ii) for each φ ∈ Π, a set Aφ ⊂ An , where n = α(φ).

E.g. For L = language of groups, an L-structure is a set A, with functions mA : A2 → A,

iA : A → A, eA ∈ A. (Note: need not be a group.)

For L = language of posets, an L-structure is a set A, with a relation 6A ⊂ A2 .

For L-structure A, sentence p, want to define ‘p holds in A’.

For example, want ‘(∀x)(m(x, x) = e)’ to hold if and only if each a ∈ A has mA (a, a) = eA .
So: ‘add in ∈ A and subscript-A and read it aloud’. (Not a definition.)

Formal bit!

Define the interpretation of closed term t in L-structure to be tA ∈ A inductively by:

for f ∈ Ω, α(f ) = n, and t1 , . . ., tn closed terms: (f t1 . . .tn )A = fA (t1 )A . . .(tn )A
Note. For c a constant, cA already defined.

E.g. m(e, m(e, e))A = mA (eA , mA (eA , eA )).

Define the interpretation of sentence p in L-structure A to be pA ∈ {0, 1} inductively by:

Atomic formulae
⊥A = 0

1 if sA = tA
(s = t)A = (any closed terms s, t)
0 if not

1 if (t1A . . .tnA ) ∈ φA (each φ ∈ Π, α(φ) = n,
φ(t1 . . .tn )A =
0 if not closed terms t1 , . . ., tn )

Sentences

0 if pA = 1, qA = 0
(p ⇒ q)A = (p, q sentences)
1 if not

1 if p[a/x]A = 1 ∀a ∈ A
((∀x)p)A =
0 if not
(where, for any a ∈ A, we form a new language L′ by adding a constant symbol a,
and make A into an L′ -structure by setting aA = a).

End of formal bit!

Remark. If formula p has free variables, can define pA ⊂ A#free variables .

E.g. If p = (∃y) m(y, y) = x , then pA = {a ∈ A : ∃b ∈ A with mA (b, b) = a}.

26
If pA = 1, say p holds in A, or p is true in A, or A is a model of p.

If T is a theory (set of sentences), say A is a model of T if A is a model of p for all p ∈ T

(i.e. every p ∈ T holds in A).

For T a theory, p a sentence, say T entails p, written T |= p, if every model of T is also a

model of p.

Say p is a tautology if p holds in all L-structures, written |= p. (Equivalently, ∅ |= p.)

E.g. (∀x) (x = x) is a tautology.

Example: theory of groups

L = language of groups: Ω = {m, i, e}, Π = ∅

arities: 2 1 0


 (∀x)(∀y)(∀z) m(x, m(y, z)) = m(m(x,
y), z , 
Let T = (∀x) m(x, e) = x ∧ m(e, x) = x ,
.
 
(∀x) m(x, i(x)) = e ∧ m(i(x), x) = e

Then an L-structure A is a model for T ⇔ A is a group. (Two assertions.)

Say T axiomatises the class of groups, or ‘axiomatises the theory of groups’. Some-
times the elements of T are called the ‘axioms’ of T .

Example: theory of fields

L = language of fields: Ω = {+, ×, 0, 1, −}, Π = ∅

arities: 2 2 0 0 1

Let T consist of: abelian group under (+, −, 0)

× commutative, and distributive over +
(∀x)(1x = x)
¬(0 = 1)

(∀x) (¬(x = 0)) ⇒ (∃ y)(xy = 1)

Then an L-structure A is a model of T ⇔ A is a field.

So T axiomatises the theory of fields.

E.g. T |= ‘inverses are unique’ : (∀x) x 6= 0) ⇒ (∀y)(∀z)([xy = 1 ∧ xz = 1] ⇒ y = z) .

Example: theory of posets

L = language of posets: Ω = ∅, Π = {6} (arity 2)

T : (∀x) (x 6 x)

(∀x)(∀y)(∀z) (x 6 y ∧ y 6 z) ⇒ x
6 z
(∀x)(∀y) (x 6 y ∧ y 6 x) ⇒ x = y

Example: theory of graphs

L = language of graphs: Ω = ∅, Π = {a} (a = ‘is adjacent to’, arity 2)

T : (∀x) (¬a(x, x))

(∀x, y) (a(x, y) ⇒ a(y, x))

27
Proofs

Logical axioms. (3 usual, 2 for ‘=’, 2 for ‘∀’)

1. p ⇒ (q ⇒ p) (any formulae p, q)
2. (p ⇒ (q ⇒ r)) ⇒ ((p ⇒ q) ⇒ (p ⇒ r)) (any formulae p, q, r)
3. (¬¬p) ⇒ p (any formula p)
4. (∀x) (x = x) (any variable x)
5. (∀x)(∀y) ((x = y) ⇒ (p ⇒ p[y/x])) (any variables x, y, formula p in which y does
not occur bound)
6. ((∀x) p) ⇒ p[t/x] (any variable x, formula p, term t with no free variable of t
occurring bound in p)
7. ((∀x) (p ⇒ q)) ⇒ (p ⇒ (∀x) q) (any variable x, formulae p, q with x not occurring
free in p)

Rules of deduction

Modus ponens: from p, p ⇒ q, can deduce q.

Generalisation: from p, can deduce (∀x) p, provided x does not occur free in any
premise used in the proof of p.

For S ⊂ L, p ∈ L, a proof of p in S is a finite sequence of formulae, ending with p, such

that each one is a logical axiom or a hypothesis (member of S) or obtained from earlier lines
by a rule of deduction.

Say S proves p, or p is a theorem of S, if there is a proof of p from S – written S ⊢ p.

Note. Each logical axiom is a tautology.

Note on ∅. Suppose we are allowed ∅ as a structure (for a language L with no constants).

Then (∀x) ⊥ holds in ∅, and ⊥ does not hold in S, so ((∀x)⊥) ⇒ ⊥ does not hold in ∅

But this is an instance of axiom 6.

Example. {x = y, x = z} ⊢ y = z (‘use axiom 5 with p being x = z’)

1. (∀x)(∀y) ((x = y) ⇒ (x = z ⇒ y = z)) (axiom 5)

2. (∀x)(∀y) ((x = y) ⇒ (x = z ⇒ y = z))
(axiom 6)
⇒ (∀y) ((x = y) ⇒ (x = z ⇒ y = z))
3. (∀y) ((x = y) ⇒ (x = z ⇒ y = z)) (modus ponens on 1, 2)


4. (∀y) ((x = y) ⇒ (x = z ⇒ y = z)) (axiom 6)
⇒ ((x = y) ⇒ (x = z ⇒ y = z))
5. (x = y) ⇒ (x = z ⇒ y = z) (modus ponens on 3, 4)
6. x = y (hypothesis)
7. x = z ⇒ y = z (modus ponens on 5, 6)
8. x = z (hypothesis)
9. y = z (modus ponens on 7, 8)

28
Proposition 1 (Deduction Theorem). Let S be a set of formulae, and p, q formulae.
Then S ⊢ (p ⇒ q) if and only if S ∪ {p} ⊢ q

Proof. (⇒) As for propositional logic: have a proof of p ⇒ q from S. So add lines ‘p’ and
‘q (modus ponens)’ to obtain a proof of q from S ∪ {p}.

(⇐) Much as for propositional logic: only new case is generalisation. In other words,
in a proof of q from S ∪ {p}, we have written down
r
(∀x) r (generalisation)
and we have a proof of S ⊢ (p ⇒ r) (by induction), and we seek a proof of S ⊢ (p ⇒
(∀x) r). In deduction of r from S ∪ {p}, no hypothesis had x as a free variable, hence
also in deduction of p ⇒ r from S, no hypothesis had x as a free variable. Thus
S ⊢ (∀x) (p ⇒ r) by generalisation.

If x not free in p : from S ⊢ (∀x) (p ⇒ r), get S ⊢ (p ⇒ (∀x) r) by axiom 7 and

modus ponens.

If x free in p : our deduction of r from S ∪ {p} cannot have used p, so actually S ⊢ r,

so S ⊢ (∀x) r (generalisation), so S ⊢ (p ⇒ (∀x) r), by axiom 1. 

Aim. S ⊢ p ⇔ S |= p. For example, if a sentence holds in all groups then it may be deduced
from the group theory axioms.

** Non-examinable section **

Proposition 2 (Soundness Theorem). Let S be a set of sentences, and p a sentence.

Then S ⊢ p ⇒ S |= p.

Proof. Have a proof of p from S, and need to show that every model for S is a model for p.
This is an easy induction on the lines of the proof. 

For adequacy: want S |= p ⇒ S ⊢ p, i.e. S ∪ (¬p) |= ⊥ ⇒ S ∪ (¬p) ⊢ ⊥.

Theorem 3 (Model existence lemma, or completeness theorem). Let S be a consis-

tent set of sentences in a language L. Then S has a model.

Ideas. 1. Build model out of language itself: take a set of closed terms, with operations,
e.g. (1 + 1) +A (1 + 1) = (1 + 1) + (1 + 1)
2. E.g. for fields, 1 + 0 = 1 for any model, but closed terms 1 + 0 and 1 are distinct.
So we would quotient out A by s ∼ t if S ⊢ (s = t) and use equivalence classes.
3. For ‘fields of characteristic 2 or 3’ : S = field axioms, with ‘1+1 = 0 ∨ 1+1+1 = 0’.
Then S 6⊢ (1 + 1 = 0), and S 6⊢ (1 + 1 + 1 = 0), so [1 + 1] 6= [0] and [1 + 1 + 1] 6= [0].
So we do not get a field of characteristic 2 or 3.
So need to extend S to a maximal consistent set first.
4. For ‘fields with a square root of 2’ : S = field axioms, with ‘(∃ x) (xx = 1 + 1)’.
Then no closed term t has [tt] = [1 + 1].
S lacks ‘witnesses’ : so add constant symbol c, and add axiom ‘cc = 1 + 1’ to S.
But this has added to the language, hence it is no longer maximal consistent, so
we must loop back to Idea 3. Problem: doesn’t terminate!

29
Proof. Have consistent S in language L = L(Ω, Π). Extend S to maximal consistent S1
(by Zorn). Then each sentence p ∈ L has p ∈ S1 or (¬p) ∈ S1 , so S1 is complete
(i.e. ∀p ∈ L : S1 ⊢ p or S1 ⊢ ¬p).

For each (∃ x) p ∈ S1 , add a new constant t to the language, and add p[t/x] to S1 .
We obtain T1 , in language L1 = L(Ω ∪ C1 , Π), that has witnesses for S1 : for each
(∃ x) p ∈ S, have p[t/x] ∈ T1 , some closed term t. Easy to check T1 consistent.

Extend T1 to maximal consistent S2 (in language L1 ), then add witnesses for S2 to

obtain T2 in language L2 = L(Ω ∪ C1 ∪ C2 , Π). Continue inductively.

Let S = S1 ∪ S2 ∪ . . ., in language L = L(Ω ∪ C1 ∪ C2 ∪ . . ., Π).

Claim. S consistent, complete, and has witnesses (for itself).

Consistent. If S ⊢ ⊥, then some Sn ⊢ ⊥ (as proofs are finite). /\/\

Complete. For sentence p ∈ L, have p ∈ Ln for some n (since p can only mention
finitely many constants). So Sn+1 ⊢ p or Sn+1 ⊢ (¬p) (as Sn+1 complete in
language Ln ), so S ⊢ p or S ⊢ (¬p).

Witnesses. Given (∃ x) p ∈ S, have (∃ x) p ∈ Sn , some n. So there exists a

closed term t, in language Ln+1 , with p[t/x] ∈ Tn (definition of Tn ). Then
p[t/x] ∈ S (and t ∈ L).

On the set of closed terms of L, define s ∼ t to mean S ⊢ (s = t). Clearly, ∼ is an

equivalence relation. Let A be the set of equivalence classes, made an L-structure by:
– for f ∈ Ω, let fA ([t1 ], . . ., [tn ]) = [f (t1 , . . ., tn )].
– for φ ∈ Π, let φA = {([t1 ], . . ., [tn ]) : S ⊢ φ(t1 , . . ., tn )}.
Claim. For any sentence p ∈ L, have S ⊢ p ⇔ p holds in A, i.e. pA = 1.

Then done: A is a model of S, hence of S.

Proof of claim. An easy induction.

Atomic sentences

⊥ : S 6⊢ ⊥, and ⊥A = 0
s = t : S ⊢ (s = t) ⇔ [s] = [t] (definition of ∼)
⇔ s A = tA (definition of operations on A)
⇔ (s = t)A = 1 (i.e. s = t holds in A)
φ(t1 . . .tn ) : same

Induction step

p ⇒ q : S ⊢ (p ⇒ q) ⇔ S ⊢ ¬p or S ⊢ q
(⇒: if S 6⊢ ¬p, S 6⊢ q, then S ⊢ p, S ⊢ ¬q, by completeness of S./\/\)
⇔ pA = 0 or qA = 1(induction hypothesis)
⇔ (p ⇒ q)A = 1

(∃ x) p : S ⊢ (∃ x) p ⇔ S ⊢ p[t/x], some closed term t (⇒: S has witnesses)

⇔ p[t/x]A = 1, some closed term t (induction)
⇔ (∃ x) p holds in A (as A is the set of
(equivalence classes of) closed terms)

So, in particular, A is a model for S (as S ⊂ S). 

30
Hence, by remarks before Theorem 3, we have:

Corollary 4 (Adequacy Theorem). Let S be a set of sentences, and p a sentence. Then

S |= p ⇒ S ⊢ p. 

Theorem 5 (Gödel’s Completeness Theorem for First-Order Logic). Let S be a set

of sentences, and p a sentence. Then S |= p ⇔ S ⊢ p.

Proof. (⇐) Soundness.

(⇒) Adequacy. 

Remarks. 1. If L countable (Ω, Π countable) then Zorn not needed.

2. ‘First-order’ means: our quantifiers/variables ranged over elements of the L-structure

(not subsets).

** End of non-examinable section **

Corollary 6 (Compactness Theorem). Let S be a set of sentences. Then if every finite

subset of S has a model then S has a model.

Proof. Trivial if we replace ‘has a model’ with ‘is consistent’ (as proofs are finite). 

Note. No decidability theorem – not so obvious to check S |= p.

Typical applications of completeness (compactness)

Can we axiomatise the theory of finite groups? In other words, is there a set T of sentences
(in languages of group theory) such that a group G is a model of T ⇔ G finite?

Corollary 7. The class of finite groups is not axiomatisable (in the language of groups).

Remark. Extraordinary that we can prove this, as opposed to just being convinced it is
true!

Proof. Suppose T axiomatises finite groups. Form T ′ by adding to T the sentences

(∃ x1 )(∃ x2 ) (x1 6= x2 )
‘|G| > 2’

(∃ x1 )(∃ x2 )(∃ x3 ) (x1 6= x2 ) ∧ (x1 6= x3 ) ∧ (x2 6= x3 ) ‘|G| > 3’

Etc. Then any finite subset of T ′ has a model (Zm , some m big enough), and so by
compactness, T ′ itself has a model. /\/\ 

Note. We used compactness, which came from completeness – so we are using the full
strength of the model existence lemma.

Corollary 7′ Let S be a theory with arbitrarily large finite models. Then S has an infinite
model.

Proof. Add sentences as in Corollary 7, and apply compactness as in Corollary 7. 

‘Finiteness is not a first-order property.’

31
Corollary 8 (Upward Löwenheim-Skolem Theorem). Let S be a theory with an infi-
nite model. Then S has an uncountable model.

Proof. Add to the language an uncountable family {ci : i ∈ I} of constants, and let S ′ =
S ∪ {‘ci 6= cj ’ : i, j ∈ I, i 6= j}. Seek a model for S ′ .

But any finite subset of S ′ has a model (as it can mention only finitely many of the ci
– so any infinite model of S will do). So by compactness, S ′ has a model. 

Remark. Similarly, we can ensure our model does not inject into X, for any fixed X – e.g.
add γ(X) constants, or P(X) constants.

For example, there exists an infinite field (e.g. Q), so there is an uncountable field (e.g. R),
and so also a field that does not inject into P(P(R)), say.

Corollary 9 (Downward Löwenheim-Skolem Theorem). Let S be a theory in a count-

able language (i.e. Ω, Π countable). If S has a model, then S has a countable model.

Proof. The model constucted in Theorem 3 is countable. 

Peano Arithmetic

We try to make the usual axioms for N into a first-order theory.

Language: Ω = {0, s, +, ×}, Π = ∅

↑ ↑ ↑ ↑
arities: 0 1 2 2

Axioms:
1. (∀x)(s(x) 6= 0)
2. (∀x)(∀y)(s(x) = s(y) ⇒ x = y)


3. (∀y1 ). . .(∀yn ) (p[0/x] ∧ (∀x)(p ⇒ p[s(x)/x])) ⇒ (∀x)p
(one such for each formula p, free variables y1 , . . ., yn , x – an axiom-scheme, mean-
ing an infinite set of axioms)
4. (∀x)(x + 0 = x)
5. (∀x)(∀y)(x + s(y) = s(x + y))
6. (∀x)(x × 0 = 0)
7. (∀x)(∀y)(x × s(y) = (x × y) + x)

These axioms are called Peano arithmetic (PA), or sometimes formal number theory.

Note on axiom 3. For induction, our first guess would be

(p[0/x] ∧ (∀x)(p ⇒ p[s(x)/x])) ⇒ (∀x)p

But then we would be missing sets such as {x : x > y}, for a given y. Hence we add
parameters as above.

32
Then PA has an infinite model (N), and hence, by Upward Löwenheim-Skolem, an uncount-
able model. Thus PA has a model not isomorphic to N – doesn’t this contradict the fact that
N is uniquely defined by the usual axioms?

Answer. Axiom 3 is only ‘first-order induction’: it is not true induction (over all subsets of
the structure). E.g., even in N itself, axiom 3 only refers to countably many subsets.

For S ⊂ N, say S definable (or definable in PA) if there exists a formula p (in language
of PA) with free variable x such that

∀m ∈ N : m ∈ S ⇐⇒ (p[m/x] holds in N) (m = s(s(. . .s(0). . .)))

So only countably many sets are definable.

E.g. – Set of squares. p(x) : (∃ y)(yy = x).

– Set of primes. p(x) : (∀y)(y|x ⇒ [y = 1 ∨ y = x]), where 1 = s(0) and y|x means
(∃t)(yt = x).

– Powers of 2. p(x) : (∀y)([y prime ∧ y divides x] ⇒ y = 2).

Exercise. {x : x is a power of 4}.

Challenge. {x : x is a power of 10}.

Is PA a complete theory? I.e. for every sentence p, either PA ⊢ p or PA ⊢ ¬p ?

Gödel’s Incompleteness Theorem says: PA is not complete (and a bit more).

So there exists p: PA 6⊢ p, PA 6⊢ ¬p. But one of p and ¬p is true in N.

Conclusion. There exists a sentence p which is true in N, but PA 6⊢ p.

This does not contradict the completeness theorem, which tells us that if p holds in every
model of PA, then PA ⊢ p.

33
 Chapter 5 : Set Theory

Aim. What does the ‘universe of sets’ look like?

We shall view set theory as ‘just’ another first-order theory. (‘A liberating viewpoint.’)

Zermelo-Fraenkel Set Theory

Language of ZF: Ω = ∅, Π = {∈} (∈ of arity 2).

Axioms of ZF: 2 to get started, 4 to build things, 3 one might not think of at first.

Then a ‘universe of sets’ will mean a model (V, ∈) of the ZF axioms.

Our question is: what does (V, ∈) look like?

❈❈ ✛ ✄✄ P(the natural numbers)

❈ ✄
❈ ✛✄ the natural numbers
❈ ✄
❈ ✛✄
❈ ✄ {∅, {∅}}
❈✄✛ ∅

(‘Does V really look like this?’)

We could view chapter 5 as a worked example from chapter 4, but very scary, as (hopefully)
every model of ZF should contain a copy of all of mathematics, and therefore will be incredibly
complicated.

Axioms of ZF

1. Axiom of extension. ‘If two sets have the same members, then they are equal.’

(∀x)(∀y)[(∀z)(z ∈ x ⇔ z ∈ y) ⇒ x = y]

Note. Converse is an instance of a logical axiom.

2. Axiom of separation. ‘Can form a subset of a set.’

More precisely, for any x and property p, can form {z ∈ x : p(z)}.

(∀t1 ). . .(∀tn )(∀x)(∃ y) (∀z) [z ∈ y ⇔ (z ∈ x ∧ p)]

| {z } | {z }
parameters ‘the elements of y are those z such that. . . ’

– one for each formula p, free variables t1 , . . ., tn , z.

Note. Need parameters t1 , . . ., tn , e.g. to allow {z ∈ x : t ∈ z}, for fixed t (parameter).

Remark. Really, this is an axiom-scheme (an infinite set of axioms).

34
3. Empty-set axiom. ‘There is a set with no members.’

(∃ x)(∀y)(¬y ∈ x)

We write ∅ for this (unique by extension) set. This is an abbreviation – so p(∅) is

short for (∃ x)((∀y)(¬y ∈ x) ∧ p(x)).

Similarly, write {z ∈ x : p(z)} for the set guaranteed by separation.

4. Pair-set axiom. ‘Given x and y, can form {x, y}.’

(∀x)(∀y)(∃ z)(∀t)[t ∈ z ⇔ (t = x ∨ t = y)]

We write {x, y} for this set, and write {x} for {x, x}.

Can now define (as an abbreviation) the ordered pair (x, y) to be {x}, {x, y} .

Easy to check that (x, y) = (z, t) ⇔ (x = z ∧ y = t). (Follows from the axioms so far.)

Say x is an ordered pair to mean (∃ y)(∃ z)(x = (y, z)).

Then f is a function means

(∀x)(x ∈ f ⇒ x is an ordered pair) ∧ (∀x)(∀y)(∀z) [(x, y) ∈ f ∧ (x, z) ∈ f ] ⇒ y = z ,

and x is the domain of f , written x = dom f , means

(f is a function) ∧ (∀y) y ∈ x ⇔ (∃ z)[(y, z) ∈ f ] ,

and f : x → y means


(f is a function) ∧ (x = dom f ) ∧ (∀z) (∃ t)[(t, z) ∈ f ] ⇒ z ∈ y .

5. Union axiom. ‘Can form unions.’

(∀x)(∃ y)(∀z)[z ∈ y ⇔ (∃ t)(t ∈ x ∧ z ∈ t)]

S
E.g., A ∪ B ∪ C = {A, B, C } ←− z ∈ this ⇐⇒ z ∈ A or z ∈ B or z ∈ C.
| {z }
x

6. Power-set axiom. ‘Can form power-sets.’

(∀x)(∃ y)(∀z)(z ∈ y ⇔ z ⊂ x)

where we have written x ⊂ y for (∀z)(z ∈ x ⇒ z ∈ y).

S
Notes. 1. Write x and P(x) for the sets given by the union and power set axioms.
S
Often write x ∪ y for {x, y}, etc.
T
2. No new axiom needed for intersections: can form x (6= ∅) as a subset of y,
T
for any y ∈ x, so x obtainable from axiom of separation.

3. Can now form x × y, as a subset of P(P(x ∪ y)) – since if t ∈ x, u ∈ y then


(t, u) = {t}, {t, u} ∈ P(P(x ∪ y)).

Similarly, can form the set of all functions from x to y, as a subset of P(x × y).

35
7. Axiom of infinity.

So far, any model (V, ∈) must be infinite. For example, for any set x, let the successor
of x be x+ = x ∪ {x}, and then ∅, ∅+ , ∅++ , ∅+++ , . . . are distinct.

∅+ = {∅}, ∅++ = {∅, {∅}}, ∅+++ = {∅, {∅}, {∅, {∅}}}, ...

Often write 0 for ∅, 1 for ∅+ , 2 for ∅++ , etc. Then

0 = ∅, 1 = {0}, 2 = {0, 1}, 3 = {0, 1, 2}, . . .

Does V have an infinite set?

From outside (real world of maths): V is an infinite set.

From inside V : V is not a set – meaning ¬(∃ x)(∀y)(y ∈ x) (Russell’s paradox)

We want V to have an infinite set: an x ∈ V such that ∅ ∈ x, ∅+ ∈ x, ∅++ ∈ x, . . ..

Say x is a successor set iff (∅ ∈ x) ∧ (∀y)(y ∈ x ⇒ y + ∈ x).

(‘A good idea – it is a legal (finite) formula.’)

Axiom of infinity: ‘There is an infinite set’ / ‘There is a successor set’

(∃ x)(x is a successor set)

Note that any intersection of successor sets is a successor set, so there is a smallest
successor set (namely the intersection of all successor sets): call it ω. (This will be our
version, in V , of the natural numbers.)


Thus (∀x) x ∈ ω ⇔ (∀y)(y is a successor set ⇒ x ∈ y) , so any subset of ω that is a
successor set must equal ω (by definition of ω):


(∀x) [(x ⊂ ω) ∧ (∅ ∈ x) ∧ (∀y)(y ∈ x ⇒ y + ∈ x)] ⇒ x = ω

(Genuine induction, applying to all x ⊂ ω – not like PA in chapter 4.)

Can check (∀x)(x ∈ ω ⇒ x+ 6= ∅), and (∀x)(∀y) [(x ∈ ω) ∧ (y ∈ ω) ∧ (x+ = y + )] ⇒

x = y , so ω satisfies (in V ) our usual rules for the natural numbers.


Can now define x is finite to mean (∃ y) (y bijects with x) ∧ (y ∈ ω) , and x is
countable to mean (x is finite) ∨ (x bijects with ω).

8. Axiom of foundation. ‘Sets are built up from simpler sets’.

Want to disallow things like x ∈ x (i.e. ‘{x} has no ∈-minimal member’), and x ∈ y ∈ x
(i.e. ‘{x, y} has no ∈-minimal member’), etc.

Similarly, want to disallow x1 ∈ x0 , x2 ∈ x1 , x3 ∈ x2 , . . . (i.e. ‘{x0 , x1 , x2 , . . .} has no

∈-minimal member’).

Axiom of foundation: ‘Every (non-empty) set has an ∈-minimal member.’

(∀x) x 6= ∅ ⇒ (∃ y)[y ∈ x ∧ (∀z)(z ∈ x ⇒ z ∈
/ y)]

36
 9. Axiom of replacement

Often say: ‘for each i ∈ I, have Ai : take {Ai : i ∈ I}’. But there is a problem: why
should i 7→ Ai be a function? (I.e., why should {(i, Ai ) : i ∈ I} be a set?)

Want ‘the image of a set, under something looking like a function, is a set’. This is
going from I out into the universe.

Digression on classes

Idea. x 7→ {x} looks like a function, but it is not, as it is not even a set. (Every
SS
function f has a domain, defined for example as a suitable subset of f , and this f
would have domain V /\/\.)

Let (V, ∈) be an L-structure. A class is a collection C of members of V such that for

some formula p, free variables x (and maybe more), x ∈ C ⇔ p holds.

E.g. V is a class – take p to be ‘x = x’.

For t ∈ V , {x ∈ V : t ∈ x} is a class – take p to be ‘t ∈ x’. (t is a parameter.)
Clearly every set is a class – take p to be ‘x ∈ y’.

If a class C is not a set, i.e. ¬(∃y)(∀x) (x ∈ y ⇔ p(x)), say C is a proper class.

E.g. V is a proper class, as is {x ∈ V : x infinite}.

Similarly, a function-class is a collection F of ordered pairs such that for some formula
p, free variables x, y (and maybe more), we have: (x, y) belongs to F ⇔ p holds, and if
(x, y) ∈ F, (x, z) ∈ F , then y = z.

E.g. ‘x 7→ {x}’ is a function-class: take p = ‘y = {x}’.

End of digression

Axiom of replacement: ‘The image of a set under a function-class is a set.’



(∀t1 ). . .(∀tn ) (∀x)(∀y)(∀z) (p ∧ p[z/y]) ⇒ z = y
| {z } | {z }
parameters p is a function-class

 
⇒ (∀x)(∃ y)(∀z) z ∈ y ⇔ (∃ t)(t ∈ x ∧ p[t/x, z/y]) ,
| {z }
‘F (t)=z’
| {z }
image of x under p is a set

– one for each formula p, free variables x, y, t1 , . . ., tn .

Note. This is an axiom-scheme.

For example, taking F (x) = {x}, we have that for any set x we can form {{t} : t ∈ x},
by replacement with p being ‘y = {x}’. This is a bad example, as we can deduce it from
earlier axioms (like power-set) instead. See later for a good example.

The above axioms are called the ZF axioms.

37
Remarks. 1. Sometimes ‘foundation’ is called regularity. Sometimes ‘separation’ is called
comprehension.

2. Axiom of choice is not included – can have ZF+AC, denoted ZFC, where:

Axiom of choice: ‘Every family of non-empty sets has a choice function.’

 
(∀f ) (f is a function) ∧ (∀x)(x ∈ dom f ⇒ f (x) 6= ∅) ⇒
 
(∃g) (g a function) ∧ (dom g = dom f ) ∧ (∀x)(x ∈ dom g ⇒ g(x) ∈ f (x))

Goal. ‘What does V look like?’

Say x is transitive if every member of a member of x is a member of x:

S
(∀y) [(∃ z)(y ∈ z ∧ z ∈ x)] ⇒ y ∈ x , or equivalently x ⊂ x.

E.g. ω is transitive, as n = {0, 1, 2, . . ., n − 1} for each n ∈ ω.

Lemma 1. Every set x is contained in a transitive set.

Remarks. 1. Officially: ‘let (V, ∈) be a model of ZF. Then . . . holds in V ’, or ‘ZF ⊢ . . .’

2. Any intersection of transitive sets is transitive, so once we have proved lemma 1,

we shall know that every x is contained in a smallest transitive set, the transitive
closure of x, written T C(x).
S SS SSS
Proof. Want to form ‘ x ∪ ( x) ∪ ( x) ∪ ( x) ∪ . . . ’.

(Clearly transitive, and contains x.)

S SS
This will be by the union axiom, applied to the set {x, x, x, . . .}, which will
be a set, by axiom of replacement (‘a good example’) applied to ω and function-class
S SS
0 7→ x, 1 7→ x, 2 7→ x, . . ..

But why is this a function-class?

Define f is an attempt (‘the clever bit’) to mean

(f is a function) ∧ (dom f ∈ ω) ∧ (dom f 6= ∅) ∧ (f (0) = x)

S

∧ (∀n) [(n ∈ dom f ) ∧ (n 6= 0)] ⇒ f (n) = f (n − 1) .

Then it is easy to check that

(∀f )(∀g)(∀n) [(f is an attempt) ∧ (g is an attempt)

∧ (n ∈ dom f ) ∧ (n ∈ dom g)] ⇒ f (n) = g(n)
(by usual ω-induction) and also that


(∀n) n ∈ ω ⇒ (∃ f )[(f is an attempt) ∧ (n ∈ dom f )]

(also by ω-induction), so take function-class to be p(y, z), where

p(y, z) = (∃ f ) (f is an attempt) ∧ (y ∈ dom f ) ∧ (f (y) = z) . 

38
Want foundation to be telling us ‘sets are built up from simpler sets’. If this is correct, we
should want: if p(x) holds whenever (∀y ∈ x) p(y), then p(x) holds for all x.

Theorem 2 (Principle of ∈-induction). For each formula p, free variables t1 , . . ., tn , x,

 

(∀t1 ). . .(∀tn ) (∀x)(∀y)(y ∈ x ⇒ p(y)) ⇒ p(x) ⇒ (∀x)p(x) .
|{z}
officially p[y/x]

Proof. Given t1 , . . ., tn , suppose ¬(∀x)p(x). Then have ¬p(x), some x.

(Want to say: choose ∈-minimal x with ¬p(x), by foundation, and hence /\/\.
But {x : ¬p(x)} need not be a set! E.g., p could have been ‘x 6= x’.)

Let t = T C({x}), and u = {y ∈ t : ¬p(y)}. Then u 6= ∅ (as x ∈ u), so u has an

∈-minimal member, say y. Then ¬p(y) (as y ∈ u). But z ∈ y ⇒ z ∈ t (as t transitive),
so z ∈
/ u (as y minimal). I.e., (∀z ∈ y) p(z). /\/\ 

Remarks. 1. We are using the existence of transitive closures.

2. Foundation is actually equivalent to ∈-induction, as we can deduce foundation from

∈-induction (in presence of the other ZF axioms). Indeed, say ‘x is regular’ to mean
(∀y)(x ∈ y ⇒ y has a minimal member). So foundation says: every x is regular.

To prove this by ∈-induction, enough to show: if every y ∈ x is regular then x is regular.

Proof. For z with x ∈ z, if x minimal in z, done. If x not minimal in z, have y ∈ z,

some y ∈ x, so z has a minimal member (as y regular).

Note. Definition of ‘regular’ was a clever idea.

What about recursion? Want to be able to define f (x) in terms of the f (y), y ∈ x.

Theorem 3 (∈-recursion theorem). Let G be a function-class (x, y) ∈ G ⇔ p(x, y),

some formula p , everywhere defined.


Then there is a function-class F (x, y) ∈ F ⇔ q(x, y), some formula q , everywhere


defined, such that (∀x) F (x) = G(F |x ) . Moreover, F is unique.

Note. F |x = {(t, F (t)) : t ∈ x} is a set, by replacement.

Proof. (Existence) Say ‘f is an attempt’ if

(f is a function) ∧ ( dom f is transitive) ∧ (∀x) x ∈ dom f ⇒ f (x) = G( f |x ) .
|{z}
makes sense, as
dom f is transitive


Then (∀x)(∀f )(∀f ′ ) [(f, f ′ attempts) ∧ (x ∈ dom f ) ∧ (x ∈ dom f ′ )] ⇒ f (x) = f ′ (x) ,
by ∈-induction (as if f and f ′ agree at all y ∈ x then they agree at x).


Also, (∀x)(∃ f ) (f is an attempt) ∧ (x ∈ dom f ) , again by ∈-induction.

If each y ∈ x has an attempt defined at y, then for each y ∈ x there is a unique attempt
S
fy defined on T C({y}). Put f = {fy : y ∈ x} and then put f ′ = f ∪ {(x, G(f |x ))}.


So define F by: q(x, y) = ‘ (∃ f ) (f an attempt) ∧ (x ∈ dom f ) ∧ (f (x) = y) ’.

(Uniqueness) If we have suitable F, F ′ then (∀x)(F (x) = F ′ (x)), by ∈-induction. 

39
Note. Proofs of ∈-induction and ∈-recursion look similar to induction and recursion on a
well-ordered set (from chapter 2).

Which properties of the ‘relation’ ∈ (i.e. the formula p(x, y) = ‘x ∈ y’) have we used?

1. p is well-founded. (Every non-empty set has a p-minimal member.)

2. p is local. (For each y, {x : p(x, y)} is a set.) ← to build transitive closure.

So, for any p(x, y) that is well-founded and local, can prove p-induction and p-recursion. If
r is a relation on a set a, then trivially r is local, so to have r-induction and r-recursion, we
just need r well-founded. (So theorems in chapter 2 were a special case of this.)

‘Can we model a relation by ∈ ?’

Example. On {a, b, c} consider relation r = {(a, b), (b, c)}.


Take a′ = ∅, b′ = {∅}, c′ = {∅} . Then {a′ , b′ , c′ } is transitive, and x r y ⇔ x′ ∈ y ′ .

Say a relation r on a set a is extensional if

(∀x, y ∈ a) [(∀z ∈ a) (z r x ⇔ z r y)] ⇒ x = y .

E.g. The relation above.

The analogue of ‘subset collapse’ is:

Theorem 4 (Mostowski’s Collapsing Theorem). Let r be a relation on a set a that

is well-founded and extensional. Then there exists a transitive set b, and a bijection
f : a → b such that (∀x, y ∈ a)(x r y ⇔ f (x) ∈ f (y)). Moreover, b and f are unique.

Remark. ‘Well-founded’ and ‘extensional’ are trivially necessary.

Proof. (Existence) Define f (x) = {f (y) : y r x} – a definition by r-recursion on the set

a (‘the only sensible choice’). Note that f is a function (not just a function-class), by
replacement (it is an image of a). Let b = {f (x) : x ∈ a}, which is a set by replacement.

Then b transitive (definition of f ), and f surjective (definition of b), and so just need
to check that f is injective (then also have f (x) ∈ f (y) ⇔ x r y). We shall show that
(∀y ∈ a)(f (x) = f (y) ⇒ x = y) for each x ∈ a, by r-induction on x.

So suppose f (x) = f (y), and that (∀ t r x)(∀u ∈ a)(f (t) = f (u) ⇒ t = u).

Have {f (t) : t r x} = {f (u) : u r y} (by definition of f ), so {t : t r x} = {u : u r y} (by

induction hypothesis), so x = y by extensionality.

(Uniqueness) If f, f ′ suitable then f (x) = f ′ (x) ∀x ∈ a (by r-induction) 

An ordinal is a transitive set that is totally ordered by ∈. (Equivalently, ‘well-ordered’,

thanks to foundation.)

E.g. ∅, {∅}, any n ∈ ω (as n = {0, 1, . . ., n − 1}), ω itself.

40
So by Mostowski, a well-ordering is order-isomorphic to a unique ordinal – that ordinal is
the order-type of this well-ordering. (This was owed from chapter 2.)

So well-orderings x and y are order-isomorphic ⇔ they have the same order type.

Amusing remark. For a well-ordering x, Mostowski sends each initial segment Iy to its
order-type, so x is sent to {order-type of Iy : y ∈ x}. So, for each ordinal α, we have
α = {β : β < α}. Thus α < β ⇔ α ∈ β.
S
So, for example, α+ = α ∪ {α} and sup{αi : i ∈ I} = {αi : i ∈ I} ←− unhelpful.

Picture of the Universe

..
❈ . ✄
❈ ✄

.. 
‘Start with ∅ ; take P repeatedly (lots!).’ ❈ . ✄ 


❈ ✄ Vω+1



❈ ✄
 

 

Define sets Vα , α ∈ ON , by ∈-recursion:


 ❈ .. ✄ 


❈ ✄
 

 . 

❈ ✄

 

V0 = ∅

❈ Vω1
[ 
 ✄ 
Vα+ =SPVα Vω = Vn  ❈ ✄




 ❈ ✄
n<ω   

Vλ = γ<λ Vγ , for λ a non-zero limit 
  

 ❈ ✄ V2 = PP∅ 

 V7 

❈ ✄ V1 = P∅

  

  
❈ ✄

 
 

  
 V
❈✄ 0 = ∅ 

Want: each x ∈ V belongs to some Vα .

Lemma 5. Each Vα is transitive.

Proof. Use induction on α.

0. V0 transitive.

Successors. Vα transitive ⇒ PVα transitive, as if x ∈ y ∈ PVα , then y ⊂ Vα , so

x ∈ Vα , so x ⊂ Vα (as Vα transitive), so x ∈ PVα

Limits. Any union of transitive sets is transitive. 

Lemma 6. Vα ⊂ Vβ for all α 6 β.

Proof. Induction on β (α fixed).

If β = α. Done, as Vα ⊂ Vα .

Successors. Given Vα ⊂ Vβ , have Vβ ⊂ PVβ (as x ∈ Vβ ⇒ x ⊂ Vβ , as Vβ transitive),

so Vα ⊂ PVβ = Vβ + .

Limits. Done. 

41
 S
Theorem 7. Every x belongs to some Vα . ‘ V = α∈ON Vα ’.

Notes. 1. x ⊂ Vα ⇔ x ∈ Vα+1 .

2. If x ⊂ Vα then there exists a least such α, called rank of x.

E.g. rank(∅) = 0, rank({∅}) = 1, rank(ω) = ω. In general, rank(α) = α for all α ∈ ON .

Proof. We shall show (∀x)(∃ α)(x ⊂ Vα ) by ∈-induction.

So may assume that given x, each y ∈ x has y ⊂ Vrank(y) , so y ∈ Vrank(y)+ .

Let α = sup{rank(y)+ : y ∈ x}. Then y ∈ Vα , all y ∈ x, so x ⊂ Vα . 

Remarks. 1. The Vα are the von Neumann hierarchy.

2. Proof shows rank(x) = sup{rank(y)+ : y ∈ x}. (‘Best way to work out ranks.’)

42
 Chapter 6 : Cardinals

Looking at ‘sizes of sets’. Work in ZFC.

Say x ↔ y if (∃ f ) (f a bijection from x to y).

Want to define ‘card x’, such that card x = card y ⇔ (x ↔ y).

Cannot define card x = {y : y ↔ x}, as this might not be a set.

We know that x ↔ α, some ordinal α, so could define card x to be the least such α.

[ Just in ZF, use the ‘Scott trick’: let α(x) = least rank(y), over all y ↔ x (sometimes called
the ‘essential rank’ of x), and put card x = {y ⊂ Vα(x) : y ↔ x}. ]

Say m is a cardinal or cardinality if m = card x, some x.

The Alephs

What are the cardinalities of well-orderable sets / ordinals ?

Say α is initial if (∀β < α) (¬β ↔ α).

E.g. 0, 1, 2, . . ., ω, ω1 , and indeed γ(X) for any set X. But not ω 2 (countable, so ω 2 ↔ ω).

Define ωα , α ∈ ON , recursively by:

ω0 = ω
ωα+1 = γ(ωα )
ωλ = sup{ωβ : β < λ}

Then each ωα is initial (by induction). Also, every infinite initial δ is an ωα . Indeed, the ωα
are unbounded in the ordinals – e.g. we have ωα > α ∀α, by induction. Taking the least α
with δ 6 ωα , must have δ = ωα , by definition of the ωα .

Write ℵα (‘aleph-α’) for card ωα .

So the alephs are the cardinalities of the infinite well-orderable sets.

E.g. card ω = ℵ0 , card ω1 = ℵ1 .

For cardinals m, n, say m 6 n if there is an injection M → N , where M, N sets with card

M = m, card N = n. (This does not depend on choice of M, N ).

Similarly, m < n means (m 6 n and m 6= n).

E.g. card ω < card P(ω).

Notes. If m 6 n, n 6 m, then n = m (Schröder-Bernstein), so 6 is a partial order.

In ZFC, it is a total order. In fact, just in ZF it need not be a total order – the ℵα are
the cardinalities of the well-orderable sets.

43
Cardinal Arithmetic

For cardinals m, n, define:

m+n to be card (M ⊔ N )
mn to be card (M × N )
mn to be card (M N ), where M N = {f : f a function from N to M }

with M , N any sets with card M = m, card N = n. (Does not depend on choice of M , N .)
P
Similarly, i∈I mi = card (⊔i∈I Mi ). (Note: AC used here.) Etc.

E.g. 1. R ↔ Pω ↔ 2ω , so card R = card (Pω) = card (2ω ) = 2ℵ0

2. How many sequences of reals are there?

Want card (Rω ) = (2ℵ0 )ℵ0 = 2ℵ0 ℵ0 = 2ℵ0 . (Using (D) and (A) below.)

Note. Used obvious facts like:

(A) ℵ0 ℵ0 = ℵ0 (as ω × ω ↔ ω)
(B) m+n=n+m (as M ⊔ N ↔ N ⊔ M )
(C) mn = nm (as M × N ↔ N × M )
(D) (mn )p = mnp (as (M N )P ↔ M N ×P )

Warning. Cardinal exponentiation is not the same as ordinal exponentiation.

E.g. Ordinals: ω ω is countable (as ω ω = sup{ω β : β < ω})

Cardinals: ℵℵ0 0 > 2ℵ0 > ℵ0 , so uncountable.

We know ℵ0 ℵ0 = ℵ0 . What about ℵ1 ℵ1 ? All + and · are easy on the alephs, thanks to:

Theorem 1. For all α ∈ ON , we have ℵα ℵα = ℵα .

Proof. Induction on α.

Define a well-ordering of ωα × ωα by ‘going up in squares’:

(x, y) < (z, t) if either max(x, y) < max(z, t),

or max(x, y) = max(z, t) = β, say, and y < β, t = β,
or y = t = β, x < z,
or x = z = β, y < t.

For any δ ∈ ωα × ωα , consider an initial segment Iδ . Then Iδ ⊂ β × β, some β < ωα .

But card β < card ωα (as ωα initial), so β × β ↔ β (by the induction hypothesis) or β
is finite, so card Iδ 6 card (β × β) = card β < card ωα .

Thus every proper initial segment has order-type < ωα , whence our well-ordering has
order-type 6 ωα . Thus ωα × ωα injects into ωα , so ℵα ℵα 6 ℵα .

Trivially ℵα 6 ℵα ℵα , so ℵα ℵα = ℵα . 

44
Corollary 2. Let α 6 β. Then ℵα + ℵβ = ℵα ℵβ = ℵβ .

Proof. ℵβ 6 ℵα + ℵβ 6 2ℵβ 6 ℵα ℵβ 6 ℵβ ℵβ = ℵβ . 

Example. In ZFC, an infinite set X has X ↔ X ⊔ X.

Remarks. However, exponentiation is hard. For example 2ℵ0 might not even be an aleph
(if no AC).

Even in ZFC, cannot prove or disprove 2ℵ0 = ℵ1 . (The continuum hypothesis.)

ℵ
Even today, not all implications about values of cardinal exponentiation (ℵαβ ) are
known.

45
 ** Bonus lecture : Incompleteness **

(PTJ chapters 4 and 9 for everything.)

Peano Arithmetic (PA): axioms (7 of them), in language 0, s, +, · .

Aim. PA incomplete: i.e., there exists a sentence p such that PA 6⊢ p and PA 6⊢ ¬p.

Equivalently: there exists a sentence p, true in N, such that PA 6⊢ p.

Here, ‘true’ = ‘true in N’, ‘provable’ = ‘PA proves it’. So we want p, true, but not provable.

Idea. Find p saying ‘I am not provable’, or, more precisely, p with p true ⇔ p not provable.

[ Then done: if p false then PA ⊢ p, whence p holds in every model of PA, and in particular
p holds in N /\/\. So p true, hence p not provable. ]

We shall ‘code up’ formulae, proofs, etc. inside PA (i.e. as natural numbers). But it still
looks as if, in any format, ‘p not provable’ has to be longer than p.

Recall. S ⊂ N is definable or definable in the language of PA if there exists a formula

p(x) (in the language of PA, x a free variable) such that for all m ∈ N, m ∈ S ⇔ p(m)
holds in N (where p(m) means p[ss.
| {z. ..s} 0/x]).
m

E.g. Set of primes: p(x) : (∀y)(∀z)([yz = x] ⇒ [y = 1 ∨ z = 1]) ∧ (x 6= 1).

Say ‘m is prime’ is definable.

Similarly, ‘m is a power of 2’ definable: p(x) : (∀y)([ y|x ∧ y prime] ⇒ y = 2).

|{z}
(∃ z)(yz = x)

Similarly, a function f : N → N is definable if there is a formula p(x, y) such that ∀ m, n ∈ N

f (m) = n ⇔ p(m, n) true.

Fact. Any function f : N → N given by some algorithm is definable.

E.g. f (n) = 2n definable, meaning: there is a formula p(x, y) such that for all m, n ∈ N,
m = 2n ⇔ p[n/x, m/y] holds.

Coding
′
Language symbols are: 0 s + · ⊥ ⇒ ( ) ∀ x =

Assign to each a distinct value: v(0) = 1, v(s) = 2, v(+) = 3, . . ., v( ′ ) = 11, v(=) = 12.

Now code a formula p by raising successive primes to the powers of successive symbols.

E.g. If p is ‘(∀x)(x = 0)’, have code c(p) = 27 39 510 78 117 1310 1712 191 238 .

Not every number codes a formula, e.g. 27 37 or 29 57 .

46
For any m ∈ N, ‘m codes a formula’ is definable (as there exists an algorithm).

Write Sm for the formula coded by m (and set Sm = ‘⊥’ if m does not code a formula).

Now code a finite sequence p1 , . . ., pn of formulae by:

S(p1 , . . ., pn ) = 2c(p1 ) 3c(p2 ) . . .(nth prime)c(pn ) .

Observe that ‘m codes an axiom’ is definable (where ‘axiom’ means logical or PA), as there
is an algorithm. (Easy check.)

Similarly, ‘l, m, n code formulae, with Sn obtained from Sl , Sm by modus ponens’ is definable,
and same for generalisation.

So θ(m, n) = ‘n codes a proof of Sm ’ is definable.

So φ(m) = ‘m codes a provable statement’ (i.e. ‘Sm is provable’) is definable, as φ(m) ⇔

(∃ n) θ(m, n).

Clever bit:

Consider χ(m) = ‘m codes a formula, with one free variable, and Sm (m) is a non-provable
statement’. Clearly definable, say by formula p(x). (I.e. p(m) holds in N ⇔ χ(m) true.)

Let the code for p be N (i.e. p(x) = SN ). So χ(N ) asserts: ‘N codes a formula, with one
free variable, and SN (N ) unprovable’. (I.e. ‘χ(N ) not provable’.)

Thus the sentence p(N ) will do!

We have shown:

Theorem 1 (Gödel’s Incompleteness Theorem.) PA is incomplete. 

Could we make PA complete by adding some clever sentence p (true in N) to it?

Answer: no. If PA′ = PA ∪ {p}, run the same proof.

However, we can certainly enlarge PA to a complete theory: e.g. set T = {p : p true in N}.

Why does the above proof not still work? (I.e. with PA replaced by T throughout.) It can
only be because:

Theorem 2. T is not definable. 

So ‘m codes a true statement’ not definable. ‘Truth is not definable.’

Why does the proof of Theorem 1 not formalise, in PA, into a proof of our true-but-unprovable
p from PA?

Answer. We used existence of a model of PA (namely N), i.e. we used con(PA) = ‘PA is
consistent’ = ‘(∀n)(n does not code a proof of ⊥)’.

Formalising Theorem 1 gives: PA ∪ con(PA) ⊢ p.

47
Hence:

Theorem 3. PA 6⊢ con(PA). 

Does ZF ⊢ con(PA)? (So, (∀x ∈ ω)(x does not code a proof of ⊥).)

Yes: as ZF ⊢ ‘PA has a model’ (namely ω).

However, running Theorem 1 of ZF language, we get:

Theorem 4. If ZF consistent, then ZF incomplete. (Get p as before, if ZF consistent.) 

And as for Theorem 1 ⇒ Theorem 3, get:

Theorem 5. If ZF consistent, then ZF 6⊢ con(ZF). 

48
 Lent 2010 LOGIC AND SET THEORY – EXAMPLES 1 IBL

1. Which of the following propositions are tautologies?

(i) (p1 ⇒ (p2 ⇒ p3 )) ⇒ (p2 ⇒ (p1 ⇒ p3 ))
(ii) ((p1 ∨ p2 ) ∧ (p1 ∨ p3 )) ⇒ (p2 ∨ p3 )
(iii) (p1 ⇒ (¬p2 )) ⇒ (p2 ⇒ (¬p1 ))

2. Write down a proof of ⊥ ⇒ q. Use this to write down a proof of p ⇒ q from ¬p.

3. Use the Deduction Theorem to show that p ⊢ ¬¬p.

4. Show that {p, q} ⊢ p ∧ q in three different ways: by writing down a proof, by using the
Deduction Theorem, and by using the Completeness Theorem.

5. Give propositions p and q for which (p ⇒ q) ⇒ ¬(q ⇒ p) is a tautology.

6. Explain carefully why the set of all propositions is countable.

7. Three people each have a set of beliefs: a consistent deductively closed set. Show
that the set of propositions that they all believe is also consistent and deductively closed.
Must the set of propositions that a majority believe be consistent? Must it be deductively
closed?

8. Can the third axiom be deduced from the first two? In other words, is there a proof of
(¬¬p) ⇒ p that uses only the first two axioms and modus ponens?

9. Let t1 , t2 , . . . be propositions such that, for every valuation v, there exists n with
v(tn ) = 1. Use the Compactness Theorem to show that in fact we may bound the values of
n: there must be an N such that, for every valuation v, there exists n ≤ N with v(tn ) = 1.

10. Two sets S, T of propositions are equivalent if S ⊢ t for every t ∈ T and T ⊢ s for every
s ∈ S. A set S of propositions is independent if for every s ∈ S we have S − {s} 6⊢ s. Show
that every finite set of propositions has an independent subset equivalent to it. Give an
infinite set of propositions that has no independent subset equivalent to it. Show, however,
that for every set of propositions there exists an independent set equivalent to it.

11. Give a direct proof of the Compactness Theorem (not making use of the notion of
syntactic implication).

12. Give an explicit function f from natural numbers to natural numbers such that every
tautology of length n has a proof that is at most f (n) lines long.

13. A set S of propositions is a chain if for any distinct p, q ∈ S we have p ⊢ q or q ⊢ p

but not both. Write down an infinite chain. If the set of primitive propositions is allowed
to be uncountable, can there exist an uncountable chain?
+
14. Suppose that the set of primitive propositions is allowed to be uncountable. Is it
true that for every set of propositions there exists an independent set equivalent to it?
 Lent 2010 LOGIC AND SET THEORY – EXAMPLES 2 IBL

1. Write down subsets of the reals that have order-types ω + ω, ω 2 and ω 3 .

2. Let α and β be non-zero ordinals. Must we have α + β > α? Must we have α + β > β?

3. Is there a non-zero ordinal α with αω = α? What about ωα = α?

4. Show that the inductive and the synthetic definitions of ordinal multiplication coincide.

5. Let α, β, γ be ordinals. Prove that (αβ)γ = α(βγ).

6. Let α, β, γ be ordinals. Must we have (α + β)γ = αγ + βγ? Must we have α(β + γ) =

αβ + αγ?

7. Let α and β be ordinals with α ≥ β. Show that there is a unique ordinal γ such that
β + γ = α. Must there exist an ordinal γ with γ + β = α?

8. An ordinal written as ω α1 n1 + . . . + ω αk nk , where α1 > . . . > αk are ordinals (and

k and n1 , . . . , nk are non-zero natural numbers), is said to be in Cantor Normal Form.
Show that every non-zero ordinal has a unique Cantor Normal Form. What is the Cantor
Normal Form for the ordinal ǫ0 ?

9. Is ω1 a successor or a limit?

10. Let α be a countable (non-zero) limit ordinal. Prove that there exists an increasing
sequence α1 < α2 < α3 < . . . with supremum equal to α. Is this result true for α = ω1 ?

11. Show that, for every countable ordinal α, there is a subset of Q of order-type α. Why
is there no subset of R of order-type ω1 ?

12. Let X be a totally ordered set such that every subset of X is isomorphic to some initial
segment of X. Prove that the ordering must be a well-ordering.

13. Is it possible to select for each countable (non-zero) limit ordinal α an ordinal xα < α
in such a way that the xα are distinct?

+
14. Let X be a totally ordered set such that the only order-preserving injection from X
to itself is the identity. Must X be finite?
 Lent 2010 LOGIC AND SET THEORY – EXAMPLES 3 IBL
1. How many different partial orders (up to isomorphism) are there on a set of 4 elements?
How many of these are complete?
2. Which of the following posets (ordered by inclusion) are complete?
(i) The set of all subsets of N that are finite or have finite complement
(ii) The set of all independent subsets of a vector space V
(iii) The set of all subspaces of a vector space V
3. Let X be a complete poset, and let f : X → X be order-reversing (meaning that x ≤ y
implies f (x) ≥ f (y) ). Give an example to show that f need not have a fixed point. Show,
however, that there must exist either a fixed point of f or two distinct points x and y with
f (x) = y and f (y) = x.
4. Use Zorn’s Lemma to show that every partial order on a set may be extended to a total
order.
5. Give a direct proof of Zorn’s Lemma (not using ordinals and not using the Axiom of
Choice) for countable posets.
6. Show that the statement ‘for any sets X and Y , either X injects into Y or Y injects
into X’ is equivalent to the Axiom of Choice (in the presence of the other rules for building
sets). [Hint for one direction: Hartogs’ Lemma.]
7. What is yellow and equivalent to the Axiom of Choice?
8. Formulate sets of axioms in suitable languages (to be specified) for the following theories.
(i) The theory of fields of characteristic 2
(ii) The theory of posets having no maximal element
(iii) The theory of bipartite graphs
(iv) The theory of algebraically closed fields
(v) The theory of groups of order 60
(vi) The theory of simple groups of order 60
(vii) The theory of real vector spaces
9. Write down axioms (in the language of groups) for the theory of groups that are either
infinite or have order a multiple of 100.
10. Show that the theory of fields of positive characteristic is not axiomatisable (in the
language of fields), and that the theory of fields of characteristic zero is axiomatisable but
not finitely axiomatisable.
11. Is every countable model of Peano Arithmetic isomorphic to N?
12. Write down axioms, in a suitable language, for the theory of groups that have an
element of infinite order. Can this be done in the language of groups?
13. Let L be the language consisting of a single function symbol f , of arity 1. Write
down a theory T that asserts that f is a bijection with no finite orbits, and describe the
countable models of T . Prove that T is a complete theory.
14. Show that the following theories are not axiomatisable.
(i) The theory of connected graphs (in the language of graphs)
(ii) The theory of simple groups (in the language of groups)
+
(iii) The theory of non-abelian simple groups (in the language of groups)
 Lent 2010 LOGIC AND SET THEORY – EXAMPLES 4 IBL

1. Show that the Empty-Set Axiom is deducible from the Axioms of Infinity and Separation
(or, if you prefer, just from the Axiom of Infinity), and that the Axiom of Separation is
deducible from the Axiom of Replacement.
2. Show that the Pair-Set Axiom is deducible from the Axioms of Empty-Set, Power-Set
and Replacement.
3. Write down sentences (in the language of ZF) to express the assertions that, for any two
sets A and B, the product A × B and the set of all functions from A to B exist. Indicate
how to deduce these sentences from the axioms of ZF.
4. Is it true that if x is a transitive set then the relation ∈ on x is a transitive relation?
Does the converse hold?
5. What is the rank of {2, 3, 6}? What is the rank of {{2, 3}, {6}}? Work out the ranks of
Z, Q and R, using your favourite constructions of these objects from ω.
6. A set x is called hereditarily finite if each member of T C({x}) is finite. Prove that
the class HF of hereditarily finite sets coincides with Vω . Which of the axioms of ZF are
satisfied in the structure HF (i.e. the set HF , with the relation ∈ |HF )?
7. Which of the axioms of ZF are satisfied in the structure Vω+ω ?

8. What is the cardinality of the set of all continuous functions from R to R?

9. Is there an ordinal α such that ωα = α?
10. Explain why, for each n ∈ ω, there is no surjection from ℵn to ℵn+1 . Use this fact to
show that there is no surjection from ℵω to ℵω ℵ0 , and deduce that 2ℵ0 6= ℵω .
11. If ZF is consistent then, by Downward Löwenheim-Skolem, it has a countable model.
Doesn’t this contradict the fact that, for example, the power-set of ω is uncountable?
12. Assume that ZF is consistent. We extend the language of ZF by adding new constants
α1 , α2 , . . ., and extend the axioms of ZF by adding (for each n) the assertions that αn is
an ordinal and that αn+1 < αn . Explain why this theory has a model. In this model of
ZF, haven’t we contradicted the fact that the ordinals are well-ordered?
13. Prove (in ZF ) that a countable union of countable sets cannot have cardinality ℵ2 .
14. The function-classes x + y = x △ y and xy = x ∩ y ‘make V into a ring’, in the sense
that all of the axioms for a ring hold in this structure. Is it possible to make V into a ring
with 1?
+
15. Show that the function f (n) = 2n is definable in the language of PA – in other words,
find a formula p(x, y) in the language of PA such that, in the natural numbers, p(m, n)
holds if and only if n = 2m .