A Course in Model Theory

Katin Tent & Martin Ziegler

September 22, 2022

Contents
1 The Basics 1
1.1 Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Language . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Elementary Extensions and Compactness 7

2.1 Elementary substructures . . . . . . . . . . . . . . . . . . . . 7
2.2 The Compactness Theorem . . . . . . . . . . . . . . . . . . . 9
2.3 The Löwenheim-Skolem Theorem . . . . . . . . . . . . . . . 13

3 Quantifier Elimination 14
3.1 Preservation theorems . . . . . . . . . . . . . . . . . . . . . . 14
3.2 Quantifier elimination . . . . . . . . . . . . . . . . . . . . . . 20
3.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4 Countable Models 32
4.1 The omitting types theorem . . . . . . . . . . . . . . . . . . . 32
4.2 The space of types . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3 ℵ0 -categorical theories . . . . . . . . . . . . . . . . . . . . . . 37
4.4 The amalgamation method . . . . . . . . . . . . . . . . . . . 41
4.5 Prime Models . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

5 ℵ1 -categorical Theories 47
5.1 Indiscernibles . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5.2 𝜔-stable theories . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.3 Prime extensions . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.4 Lachlan’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . 60

1
 5.5 Vaughtian pairs . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.6 Algebraic formulas . . . . . . . . . . . . . . . . . . . . . . . . 66
5.7 Strongly minimal sets . . . . . . . . . . . . . . . . . . . . . . . 68
5.8 The Baldwin-Lachlan Theorem . . . . . . . . . . . . . . . . . 75

6 Morley Rank 75
6.1 Saturated models and the monster . . . . . . . . . . . . . . . 75
6.2 Morley rank . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.3 Countable models of ℵ1 -categorical theories . . . . . . . . . . 90
6.4 Computation of Morley Rank . . . . . . . . . . . . . . . . . . 92

7 Simple Theories 92
7.1 Dividing and forking . . . . . . . . . . . . . . . . . . . . . . . 92
7.2 Simplicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

8 Stable Theories 103

8.1 Heirs and coheirs . . . . . . . . . . . . . . . . . . . . . . . . . 103
8.2 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
8.3 Definable types . . . . . . . . . . . . . . . . . . . . . . . . . . 103
8.4 Elimination of imaginaries and 𝑇 eq . . . . . . . . . . . . . . . 103

A Set Theory 103

A.1 Sets and classes . . . . . . . . . . . . . . . . . . . . . . . . . . 103
A.2 Cardinals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

B Fields 105
B.1 Ordered fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

C Combinatorics 106
C.1 Pregeometris . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
C.2 The Erdős-Rado Theorem . . . . . . . . . . . . . . . . . . . . 108

D Index 108

109

E TODO Don’t understand 109

2
1 The Basics
1.1 Structures
Definition 1.1. A language 𝐿 is a set of constants, function symbols and
relation symbols
Definition 1.2. Let 𝐿 be a language. An 𝐿-structure is a pair 𝔄 = (𝐴, (𝑍 𝔄 )𝑍∈𝐿 )
where
𝐴 if a non-empty set, the domain or universe of 𝔄
𝑧𝔄 ∈ 𝐴 if 𝑍 is a constant
𝑍 𝔄 ∶ 𝐴𝑛 → 𝐴 if 𝑍 is an 𝑛-ary function symbol
𝑍 𝔄 ⊆ 𝐴𝑛 if 𝑍 is an 𝑛-ary relation symbol
Definition 1.3. Let 𝔄, 𝔅 be 𝐿-structures. A map ℎ ∶ 𝐴 → 𝐵 is called a
homomorphism if for all 𝑎1 , … , 𝑎𝑛 ∈ 𝐴

ℎ(𝑐𝔄 ) = 𝑐𝔅
ℎ(𝑓 𝔄 (𝑎1 , … , 𝑎𝑛 )) = 𝑓 𝔅 (ℎ(𝑎1 ), … , ℎ(𝑎𝑛 ))
𝑅𝔄 (𝑎1 , … , 𝑎𝑛 ) ⇒ 𝑅𝔅 (ℎ(𝑎1 ), … , ℎ(𝑎𝑛 ))

We denote this by
ℎ∶𝔄→𝔅
If in addition ℎ is injective and

𝑅𝔄 (𝑎1 , … , 𝑎𝑛 ) ⇔ 𝑅𝔅 (ℎ(𝑎1 ), … , ℎ(𝑎𝑛 ))

for all 𝑎1 , … , 𝑎𝑛 ∈ 𝐴, then ℎ is called an (isomorphic) embedding. An

isomorphism is a surjective embedding
Definition 1.4. We call 𝔄 a substructure of 𝔅 if 𝐴 ⊆ 𝐵 and if the inclusion
map is an embedding from 𝔄 to 𝔅. We denote this by

𝔄⊆𝔅

We say 𝔅 is an extension of 𝔄 if 𝔄 is a substructure of 𝔅

Lemma 1.5. Let ℎ ∶ 𝔄 ⟶ ∼ 𝔄′ be an isomorphism and 𝔅 an extension of 𝔄. Then
∼ 𝔅′ extending ℎ
there exists an extension 𝔅′ of 𝔄′ and an isomorphism 𝑔 ∶ 𝔅 ⟶
For any family 𝔄𝑖 of substructures of 𝔅, the intersection of the 𝐴𝑖 is
either empty or a substructure of 𝔅. Therefore if 𝑆 is any non-empty subset
of 𝔅, then there exists a smallest substructure 𝔄 = ⟨𝑆⟩𝔅 which contains 𝑆.
We call the 𝔄 the substructure generated by 𝑆

3
Lemma 1.6. If 𝔞 = ⟨𝑆⟩, then every homomorphism ℎ ∶ 𝔄 → 𝔅 is determined by
its values on 𝑆
Definition 1.7. Let (𝐼, ≤) be a directed partial order. This means that for
all 𝑖, 𝑗 ∈ 𝐼 there exists a 𝑘 ∈ 𝐼 s.t. 𝑖 ≤ 𝑘 and 𝑗 ≤ 𝑘. A family (𝔄𝑖 )𝑖∈𝐼 of
𝐿-structures is called directed if

𝑖 ≤ 𝑗 ⇒ 𝔄𝑖 ⊆ 𝔄 𝑗

If 𝐼 is linearly ordered, we call (𝔄𝑖 )𝑖∈𝐼 a chain

If a structure 𝔄1 is isomorphic to a substructure 𝔄0 of itself,
∼ 𝔄
ℎ0 ∶ 𝔄0 ⟶ 1

then Lemma 1.5 gives an extension

∼ 𝔄
ℎ1 ∶ 𝔄1 ⟶ 2

Continuing in this way we obtain a chain 𝔄0 ⊆ 𝔄1 ⊆ 𝔄2 ⊆ … and an in-

∼ 𝔄
creasing sequence ℎ𝑖 ∶ 𝔄𝑖 ⟶ 𝑖+1 of isomorphism

Lemma 1.8. Let (𝔄𝑖 )𝑖∈𝐼 be a directed family of 𝐿-structures. Then 𝐴 = ⋃𝑖∈𝐼 𝐴𝑖
is the universe of a (uniquely determined) 𝐿-structure

𝔄 = ⋃ 𝔄𝑖
𝑖∈𝐼

which is an extension of all 𝔄𝑖

A subset 𝐾 of 𝐿 is called a sublanguage. An 𝐿-structure becomes a 𝐾-
structure, the reduct.
𝔄 ↾ 𝐾 = (𝐴, (𝑍 𝔄 )𝑍∈𝐾 )
Conversely we call 𝔄 an expansion of 𝔄↾𝐾.
1. Let 𝐵 ⊆ 𝐴 , we obtain a new language

𝐿(𝐵) = 𝐿 ∪ 𝐵

and the 𝐿(𝐵)-structure

𝔄𝐵 = (𝔄, 𝑏)𝑏∈𝐵

Note that Aut(𝔄𝐵 ) is the group of automorphisms of 𝔄 fixing 𝐵 ele-

mentwise. We denote this group by Aut(𝔄/𝐵)

4
 Let 𝑆 be a set, which we call the set of sorts. An 𝑆-sorted language 𝐿
is given by a set of constants for each sort in 𝑆, and typed function and
relations. For any tuple (𝑠1 , … , 𝑠𝑛 ) and (𝑠1 , … , 𝑠𝑛 , 𝑡) there is a set of relation
symbols and function symbols respectively. An 𝑆-sorted structure is a pair
𝔄 = (𝐴, (𝑍 𝔄 )𝑍∈𝐿 ), where

𝐴 if a family (𝐴𝑠 )𝑠∈𝑆 of non-empty sets

𝔄
𝑍 ∈ 𝐴𝑠 if 𝑍 is a constant of sort 𝑠 ∈ 𝑆
𝑍 𝔄 ∶ 𝐴𝑠1 × ⋯ × 𝐴𝑠𝑛 → 𝐴𝑡 if 𝑍 is a function symbol of type (𝑠1 , … , 𝑠𝑛 , 𝑡)
𝑍 𝔄 ⊆ 𝐴𝑠1 × ⋯ × 𝐴𝑠𝑛 if 𝑍 is a relation symbol of type (𝑠1 , … , 𝑠𝑛 )

Example 1.1. Consider the two-sorted language 𝐿𝑃 𝑒𝑟𝑚 for permutation groups
with a sort 𝑥 for the set and a sort 𝑔 for the group. The constants and func-
tion symbols for 𝐿𝑃 𝑒𝑟𝑚 are those of 𝐿𝐺𝑟𝑜𝑢𝑝 restricted to the sort 𝑔 and an
additional function symbol 𝜑 of type (𝑥, 𝑔, 𝑥). Thus an 𝐿𝑃 𝑒𝑟𝑚 -structure
(𝑋, 𝐺) is given by a set 𝑋 and an 𝐿𝐺𝑟𝑜𝑢𝑝 -structure 𝐺 together with a func-
tion 𝑋 × 𝐺 → 𝑋

1.2 Language
#—
Lemma 1.9. Suppose 𝑏 and #—
𝑐 agree on all variables which are free in 𝜑. Then
#—
𝔄 ⊨ 𝜑[ 𝑏 ] ⇔ 𝔄 ⊨ 𝜑[ #—
𝑐]

We define
𝔄 ⊨ 𝜑[𝑎1 , … , 𝑎𝑛 ]
#— #— #—
by 𝔄 ⊨ 𝜑[ 𝑏 ], where 𝑏 is an assignment satisfying 𝑏 (𝑥𝑖 ) = 𝑎𝑖 . Because of
Lemma 1.9 this is well defined.
Thus 𝜑(𝑥1 , … , 𝑥𝑛 ) defines an 𝑛-ary relation

𝜑(𝔄) = {𝑎 ∣ 𝔄 ⊨ 𝜑[𝑎]}

on 𝐴, the realisation set of 𝜑. Such realisation sets are called 0-definable

subsets of 𝐴𝑛 , or 0-definable relations
Let 𝐵 be a subset of 𝐴. A 𝐵-definable subset of 𝔄 is a set of the form
𝜑(𝔄) for an 𝐿(𝐵)-formula 𝜑(𝑥). We also say that 𝜑 are defined over 𝐵 and
that the set 𝜑(𝔄) is defined by 𝜑. We call two formulas equivalent if in every
structure they define the same set.
Atomic formulas and their negations are called basic. Formulas with-
out quantifiers are Boolean combinations of basic formulas. It is convenient

5
to allow the empty conjunction and the empty disjunction. For that we in-
troduce two new formulas: the formula ⊤, which is always true, and the
formula ⊥, which is always false. We define

⋀ 𝜋𝑖 = ⊤
𝑖<0

⋁ 𝜋𝑖 = ⊥
𝑖<0

A formula is in negation normal form if it is built from basic formulas

using ∧, ∨, ∃, ∀

Lemma 1.10. Every formula can be transformed into an equivalent formula which
is in negation normal form

Proof. Let ∼ denote equivalence of formulas. We consider formulas which

are built using ∧, ∨, ∃, ∀, ¬ and move the negation symbols in front of atomic
formulas using

¬(𝜑 ∧ 𝜓) ∼ (¬𝜑 ∨ ¬𝜓)

¬(𝜑 ∨ 𝜓) ∼ (¬𝜑 ∧ ¬𝜓)
¬∃𝑥𝜑 ∼ ∀𝑥¬𝜑
¬∀𝑥𝜑 ∼ ∃𝑥¬𝜑
¬¬𝜑 ∼ 𝜑

Definition 1.11. A formula in negation normal form which does not contain
any existential quantifier is called universal. Formulas in negation normal
form without universal quantifiers are called existential

Lemma 1.12. Let ℎ ∶ 𝔄 → 𝔅 be an embedding. Then for all existential formulas

𝜑(𝑥1 , … , 𝑥𝑛 ) and all 𝑎1 , … , 𝑎𝑛 ∈ 𝐴 we have

𝔄 ⊨ 𝜑[𝑎1 , …, 𝑎𝑛 ] ⇒ 𝔅 ⊨ 𝜑[ℎ(𝑎1 ), … , ℎ(𝑎𝑛 )]

For universal 𝜑, the dual holds

𝔅 ⊨ 𝜑[ℎ(𝑎1 ), … , ℎ(𝑎𝑛 )] ⇒ 𝔄 ⊨ 𝜑[𝑎1 , … , 𝑎𝑛 ]

Let 𝔄 be an 𝐿-structure. The atomic diagram of 𝔄 is

Diag(𝔄) = {𝜑 basic 𝐿(𝐴)-sentence ∣ 𝔄𝐴 ⊨ 𝜑}

6
Lemma 1.13. The models of Diag(𝔄) are precisely those structures (𝔅, ℎ(𝑎))𝑎∈𝐴
for embeddings ℎ ∶ 𝔄 → 𝔅

Proof. The structures (𝔅, ℎ(𝑎))𝑎∈𝐴 are models of the atomic diagram by
Lemma ??. For the converse, note that a map ℎ is an embedding iff it pre-
serves the validity of all formulas of the form

(¬)𝑥1 =𝑥
̇ 2
𝑐=𝑥
̇ 1
𝑓(𝑥1 , … , 𝑥𝑛 )=𝑥
̇ 0
(¬)𝑅(𝑥1 , … , 𝑥𝑛 )

Exercise 1.2.1. Every formula is equivalent to a formula in prenex normal

form:
𝑄1 𝑥1 … 𝑄𝑛 𝑥𝑛 𝜑
The 𝑄𝑖 are quantifiers and 𝜑 is quantifier-free

Proof.

(∀𝑥)𝜙 ∧ 𝜓 ⊨⊧ ∀𝑥(𝜙 ∧ 𝜓) if ∃𝑥⊤(at least one individual exists)

(∀𝑥𝜙) ∨ 𝜓 ⊨⊧ ∀𝑥(𝜙 ∨ 𝜓)
(∃𝑥𝜙) ∧ 𝜓 ⊨⊧ ∃𝑥(𝜙 ∧ 𝜓)
(∃𝑥𝜙) ∨ 𝜓 ⊨⊧ ∃𝑥(𝜙 ∨ 𝜓) if ∃𝑥⊤
¬∃𝑥𝜙 ⊨⊧ ∀𝑥¬𝜙
¬∀𝑥𝜙 ⊨⊧ ∃𝑥¬𝜙
(∀𝑥𝜙) → 𝜓 ⊨⊧ ∃𝑥(𝜙 → 𝜓) if ∃𝑥⊤
(∃𝑥𝜙) → 𝜓 ⊨⊧ ∀𝑥(𝜙 → 𝜓)
𝜙 → (∃𝑥𝜓) ⊨⊧ ∃𝑥(𝜙 → 𝜓) if ∃𝑥⊤
𝜙 → (∀𝑥𝜓) ⊨⊧ ∀𝑥(𝜙 → 𝜓)

1.3 Theories
Definition 1.14. An 𝐿-theory 𝑇 is a set of 𝐿-sentences

7
 A theory which has a model is a consistent theory. We call a set Σ of
#—
𝐿-formulas consistent if there is an 𝐿-structure and an assignment 𝑏 s.t.
#—
𝔄 ⊨ Σ[ 𝑏 ] for all 𝜑 ∈ Σ
Lemma 1.15. Let 𝑇 be an 𝐿-theory and 𝐿′ be an extension of 𝐿. Then 𝑇 is con-
sistent as an 𝐿-theory iff 𝑇 is consistent as a 𝐿′ -theory
Lemma 1.16. 1. If 𝑇 ⊨ 𝜑 and 𝑇 ⊨ (𝜑 → 𝜓), then 𝑇 ⊨ 𝜓
2. If 𝑇 ⊨ 𝜑(𝑐1 , … , 𝑐𝑛 ) and the constants 𝑐1 , … , 𝑐𝑛 occur neither in 𝑇 nor in
𝜑(𝑥1 , … , 𝑥𝑛 ), then 𝑇 ⊨ ∀𝑥1 … 𝑥𝑛 𝜑(𝑥1 , … , 𝑥𝑛 )
Proof. 2. Let 𝐿′ = 𝐿 ⧵ {𝑐1 , … , 𝑐𝑛 }. If the 𝐿′ -structure is a model of 𝑇 and
𝑎1 , … , 𝑎𝑛 are arbitrary elements, then (𝔄, 𝑎1 , … , 𝑎𝑛 ) ⊨ 𝜑(𝑐1 , … , 𝑐𝑛 ).
This means 𝔄 ⊨ ∀𝑥1 … 𝑥𝑛 𝜑(𝑥1 , … , 𝑥𝑛 ).

𝑆 and 𝑇 are called equivalent, 𝑆 ≡ 𝑇 , if 𝑆 and 𝑇 have the same models

Definition 1.17. A consistent 𝐿-theory 𝑇 is called complete if for all 𝐿-
sentences 𝜑
𝑇 ⊨ 𝜑 or 𝑇 ⊨ ¬𝜑
Definition 1.18. For a complete theory 𝑇 we define
|𝑇 | = max(|𝐿|, ℵ0 )
The typical example of a complete theory is the theory of a structure 𝔄
Th(𝔄) = {𝜑 ∣ 𝔄 ⊨ 𝜑}
Lemma 1.19. A consistent theory is complete iff it is maximal consistent, i.e., if it
is equivalent to every consistent extension
Definition 1.20. Two 𝐿-structures 𝔄 and 𝔅 are called elementary equiva-
lent
𝔄≡𝔅
if they have the same theory
Lemma 1.21. Let 𝑇 be a consistent theory. Then the following are equivalent
1. 𝑇 is complete
2. All models of 𝑇 are elemantarily equivalent
3. There exists a structure 𝔄 with 𝑇 ≡ Th(𝔄)
Proof. 1 → 3 → 2 → 1

8
2 Elementary Extensions and Compactness
2.1 Elementary substructures
Let 𝔄, 𝔅 be two 𝐿-structures. A map ℎ ∶ 𝐴 → 𝐵 is called elementary if for
all 𝑎1 , … , 𝑎𝑛 ∈ 𝐴 we have
𝔄 ⊨ 𝜑[𝑎1 , … , 𝑎𝑛 ] ⇔ 𝔅 ⊨ 𝜑[ℎ(𝑎1 ), … , ℎ(𝑎𝑛 )]
which is actually saying (𝔄, 𝑎)𝑎∈𝐴 ≡ (𝔅, 𝑎)𝑎∈𝐴 . We write
≺ 𝔅
ℎ∶𝔄⟶
Lemma 2.1. The models of Th(𝔄𝐴 ) are exactly the structures of the form (𝔅, ℎ(𝑎))𝑎∈𝐴
≺ 𝔅
for elementary embeddings ℎ ∶ 𝔄 ⟶
We call Th(𝔄𝐴 ) the elemantary diagram of 𝔄
A substructure 𝔄 of 𝔅 is called elementary if the inclusion map is ele-
mentary. In this case we write
𝔄≺𝔅
Theorem 2.2 (Tarski’s Test). Let 𝔅 be an 𝐿-structure and 𝐴 a subset of 𝐵. Then
𝐴 is the universe of an elementary substructure iff every 𝐿(𝐴)-formula 𝜑(𝑥) which
is satisfiable in 𝔅 can be satisfied by an element of 𝐴
Proof. If 𝔄 ≺ 𝔅 and 𝔅 ⊨ ∃𝑥𝜑(𝑥), we also have 𝔄 ⊨ ∃𝑥𝜑(𝑥) and there exists
𝑎 ∈ 𝐴 s.t. 𝔄 ⊨ 𝜑(𝑎). Thus 𝔅 ⊨ 𝜑(𝑎)
Conversely, suppose that the condition of Tarski’test is satisfied. First
we show that 𝐴 is the universe of a substructure 𝔄. The 𝐿(𝐴)-formula 𝑥=𝑥 ̇
is satisfiable in 𝔄, so 𝐴 is not empty. If 𝑓 ∈ 𝐿 is an 𝑛-ary function symbol
(𝑛 ≥ 0) and 𝑎1 , … , 𝑎𝑛 is from 𝐴, we consider the formula
𝜑(𝑥) = 𝑓(𝑎1 , … , 𝑎𝑛 )=𝑥
̇
Since 𝜑(𝑥) is always satisfied by an element of 𝐴, it follows that 𝐴 is closed
under 𝑓 ℬ
Now we show, by induction on 𝜓, that
𝔄⊨𝜓⇔𝔅⊨𝜓
for all 𝐿(𝐴)-sentences 𝜓.
For 𝜓 = ∃𝑥𝜑(𝑥). If 𝜓 holds in 𝔄, there exists 𝑎 ∈ 𝐴 s.t. 𝔄 ⊨ 𝜑(𝑎).
The induction hypothesis yields 𝔅 ⊨ 𝜑(𝑥), thus 𝔅 ⊨ 𝜓. For the converse
suppose 𝜓 holds in 𝔅. Then 𝜑(𝑥) is satisfied in ℬ and by Tarski’s test we
find 𝑎 ∈ 𝐴 s.t. 𝔅 ⊨ 𝜑(𝑎). By induction 𝔄 ⊨ 𝜑(𝑎) and 𝔄 ⊨ 𝜓

9
 We use Tarski’s Test to construct small elementary substructures
Corollary 2.3. Suppose 𝑆 is a subset of the 𝐿-structure 𝔅. Then 𝔅 has a elemen-
tary substructure 𝔄 containing 𝑆 and of cardinality at most
max(|𝑆|, |𝐿|, ℵ0 )
Proof. We construct 𝐴 as the union of an ascending sequence 𝑆0 ⊆ 𝑆1 ⊆ …
of subsets of 𝐵. We start with 𝑆0 = 𝑆. If 𝑆𝑖 is already defined, we choose an
element 𝑎𝜑 ∈ 𝐵 for every 𝐿(𝑆𝑖 )-formula 𝜑(𝑥) which is satisfiable in 𝔅 and
define 𝑆𝑖+1 to be 𝑆𝑖 together with these 𝑎𝜑 .
An 𝐿-formula is a finite sequence of symbols from 𝐿, auxiliary symbols
and logical symbols. These are |𝐿| + ℵ0 = max(|𝐿|, ℵ0 ) many symbols and
there are exactlymax(|𝐿|, ℵ0 ) many 𝐿-formulas
Let 𝜅 = max(|𝑆|, |𝐿|, ℵ0 ). There are 𝜅 many 𝐿(𝑆)-formulas: therefore
|𝑆1 | ≤ 𝜅. Inductively it follows for every 𝑖 that |𝑆𝑖 | ≤ 𝜅. Finally we have
|𝐴| ≤ 𝜅 ⋅ ℵ0 = 𝜅

A directed family (𝔄𝑖 )𝑖∈𝐼 of structures is elementary if 𝔄𝑖 ≺ 𝔄𝑗 for all

𝑖≤𝑗
Theorem 2.4 (Tarski’s Chain Lemma). The union of an elementary directed
family is an elementary extension of all its members
Proof. Let 𝔄 = ⋃𝑖∈𝐼 (𝔄𝑖 )𝑖∈𝐼 . We prove by induction on 𝜑(𝑥) that for all 𝑖 and
𝑎 ∈ 𝔄𝑖
𝔄𝑖 ⊨ 𝜑(𝑎) ⇔ 𝔄 ⊨ 𝜑(𝑎)

Exercise 2.1.1. Let 𝔄 be an 𝐿-structure and (𝔄𝑖 )𝑖∈𝐼 a chain of elementary

substructures of 𝔄. Show that ⋃𝑖∈𝐼 𝐴𝑖 is an elementary substructure of 𝔄.
Exercise 2.1.2. Consider a class 𝒞 of 𝐿-structures. Prove
1. Let Th(𝒞) = {𝜑 ∣ 𝔄 ⊨ 𝜑 for all 𝔄 ∈ 𝒞} be the theory of 𝒞. Then 𝔐 is a
model of Th(𝐶) iff 𝔐 is elementary equivalent to an ultraproduct of
elements of 𝒞
2. Show that 𝒞 is an elementary class iff 𝒞 is closed under ultraproduct
and elementary equivalence
3. Assume that 𝒞 is a class of finite structures containing only finitely
many structures of size 𝑛 for each 𝑛 ∈ 𝜔. Then the infinite models of
Th(𝒞) are exactly the models of
Th𝑎 (𝒞) = {𝜑 ∣ 𝔄 ⊨ 𝜑 for all but finitely many 𝔄 ∈ 𝒞}

10
Proof. Chang&Keisler p220

2.2 The Compactness Theorem

We call a theory 𝑇 finitely satisfiable if every finite subset of 𝑇 is consistent
Theorem 2.5 (Compactness Theorem). Finitely satisfiable theories are consis-
tent
Let 𝐿 be a language and 𝐶 a set of new constants. An 𝐿(𝐶)-theory 𝑇 ′ is
called a Henkin theory if for every 𝐿(𝐶)-formula 𝜑(𝑥) there is a constant
𝑐 ∈ 𝐶 s.t.
∃𝑥𝜑(𝑥) → 𝜑(𝑐) ∈ 𝑇 ′
The elements of 𝐶 are called Henkin constants of 𝑇 ′
An 𝐿-theory 𝑇 is finitely complete if it is finitely satisfiable and if every
𝐿-sentence 𝜑 satisfies 𝜑 ∈ 𝑇 or ¬𝜑 ∈ 𝑇
Lemma 2.6. Every finitely satisfiable 𝐿-theory 𝑇 can be extended to a finitely com-
plete Henkin Theory 𝑇 ∗
Note that conversely the lemma follows directly from the Compactness
Theorem. Choose a model 𝔄 of 𝑇 . Then Th(𝔄𝐴 ) is a finitely complete
Henkin theory with 𝐴 as a set of Henkin constants

Proof. We define an increasing sequence ∅ = 𝐶0 ⊆ 𝐶1 ⊆ ⋯ of new constants

by assigning to every 𝐿(𝐶𝑖 )-formula 𝜑(𝑥) a constant 𝑐𝜑(𝑥) and

𝐶𝑖+1 = {𝑐𝜑(𝑥) ∣ 𝜑(𝑥) a 𝐿(𝐶𝑖 )-formula}

Let 𝐶 be the union of the 𝐶𝑖 and 𝑇 𝐻 the set of all Henkin axioms

∃𝑥𝜑(𝑥) → 𝜑(𝑐𝜑(𝑥) )

for 𝐿(𝐶)-formulas 𝜑(𝑥). It is easy to see that one can expand every 𝐿-
structure to a model of 𝑇 𝐻 . Hence 𝑇 ∪ 𝑇 𝐻 is a finitely satisfiable Henkin
theory. Using the fact that the union of a chain of finitely satisfiable theo-
ries is also finite satisfiable, we can apply Zorn’s Lemma and get a maximal
finitely satisfiable 𝐿(𝐶)-theory 𝑇 ∗ which contains 𝑇 ∪ 𝑇 𝐻 . As in Lemma
1.19 we show that 𝑇 ∗ is finitely complete: if neither 𝜑 nor ¬𝜑 belongs to 𝑇 ∗ ,
neither 𝑇 ∗ ∪ {𝜑} nor 𝑇 ∗ ∪ {¬𝜑} would be finitely satisfiable. Hence there
would be a finite subset Δ of 𝑇 ∗ which would be consistent neither with 𝜑
nor with ¬𝜑. Then Δ itself would be inconsistent and 𝑇 ∗ would not be finite
satisfiable. This proves the lemma.

11
Lemma 2.7. Every finitely satisfiable 𝐿-theory 𝑇 can be extended to a finitely com-
plete Henkin theory 𝑇 ∗

Lemma 2.8. Every finitely complete Henkin theory 𝑇 ∗ has a model 𝔄 (unique up
to isomorphism) consisting of constants; i.e.,

(𝔄, 𝑎𝑐 )𝑐∈𝐶 ⊨ 𝑇 ∗

with 𝐴 = {𝑎𝑐 ∣ 𝑐 ∈ 𝐶}

Proof. Since 𝑇 ∗ is finite complete, every sentence which follows from a finite
subset of 𝑇 ∗ belongs to 𝑇 ∗
Define for 𝑐, 𝑑 ∈ 𝐶
𝑐 ≃ 𝑑 ⇔ 𝑐=𝑑 ̇ ∈ 𝑇∗
≃ is an equivalence relation. We denote the equivalence class of 𝑐 by 𝑎𝑐 , and
set
𝐴 = {𝑎𝑐 ∣ 𝑐 ∈ 𝐶}
We expand 𝐴 to an 𝐿-structure 𝔄 by defining

𝑅𝔄 (𝑎𝑐1 , … , 𝑎𝑐𝑛 ) ⇔ 𝑅(𝑐1 , … , 𝑐𝑛 ) ∈ 𝑇 ∗ (⋆)

𝑓 𝔄 (𝑎𝑐1 , … , 𝑎𝑐𝑛 ) ⇔ 𝑓(𝑐1 , … , 𝑐𝑛 )=𝑐
̇ 0 ∈ 𝑇∗ (⋆⋆)

We have to show that this is well-defined. For (⋆) we have to show that

𝑎𝑐1 = 𝑎𝑑1 , … , 𝑎𝑐𝑛 = 𝑎𝑑𝑛 , 𝑅(𝑐1 , … , 𝑐𝑛 ) ∈ 𝑇 ∗

implies 𝑅(𝑑1 , … , 𝑑𝑛 ) ∈ 𝑇 ∗ , which is obvious.

For (⋆⋆), we have to show that for all 𝑐1 , … , 𝑐𝑛 there exists 𝑐0 with 𝑓(𝑐1 , … , 𝑐𝑛 )=𝑐
̇ 0∈
∗
𝑇 .
Let 𝔄∗ be the 𝐿(𝐶)-structure (𝔄, 𝑎𝑐 )𝑐∈𝐶 . We show by induction on the
complexity of 𝜑 that for every 𝐿(𝐶)-sentence 𝜑

𝔄∗ ⊨ 𝜑 ⇔ 𝜑 ∈ 𝑇 ∗

Corollary 2.9. We have 𝑇 ⊨ 𝜑 iff Δ ⊨ 𝜑 for a finite subset Δ of 𝑇

Corollary 2.10. A set of formulas Σ(𝑥1 , … , 𝑥𝑛 ) is consistent with 𝑇 if and only

if every finite subset of Σ is consistent with 𝑇

12
Proof. Introduce new constants 𝑐1 , … , 𝑐𝑛 . Then Σ is consistent with 𝑇 is and
only if 𝑇 ∪ Σ(𝑐1 , … , 𝑐𝑛 ) is consistent. Now apply the Compactness Theorem

Definition 2.11. Let 𝔄 be an 𝐿-structure and 𝐵 ⊆ 𝐴. Then 𝑎 ∈ 𝐴 realises a

set of 𝐿(𝐵)-formulas Σ(𝑥) if 𝑎 satisfied all formulas from Σ. We write

𝔄 ⊨ Σ(𝑎)

We call Σ(𝑥) finitely satisfiable in 𝔄 if every finite subset of Σ is realised

in 𝔄
Lemma 2.12. The set Σ(𝑥) is finitely satisfiable in 𝔄 iff there is an elementary
extension of 𝔄 in which Σ(𝑥) is realised
Proof. By Lemma 2.1 Σ is realised in an elementary extension of 𝔄 iff Σ is
consistent with Th(𝔄𝐴 ). So the lemma follows from the observation that a
finite set of 𝐿(𝐴)-formulas is consistent with Th(𝔄𝐴 ) iff it is realised in 𝔄

Definition 2.13. Let 𝔄 be an 𝐿-structure and 𝐵 a subset of 𝐴. A set 𝑝(𝑥)

of 𝐿(𝐵)-formulas is a type over 𝐵 if 𝑝(𝑥) is maximal finitely satisfiable in 𝔄
(satisfiable in an elementary extension of 𝔄). We call 𝐵 the domain of 𝑝.
Let
𝑆(𝐵) = 𝑆 𝔄 (𝐵)
denote the set of types over 𝐵.
Every element 𝑎 of 𝔄 determines a type

tp(𝑎/𝐵) = 𝑡𝑝𝔄 (𝑎/𝐵) = {𝜑(𝑥) ∣ 𝔄 ⊨ 𝜑(𝑎), 𝜑 an 𝐿(𝐵)-formula}

So an element 𝑎 realises the type 𝑝 ∈ 𝑆(𝐵) exactly if 𝑝 = tp(𝑎/𝐵). If 𝔄′ is

an elementary extension of 𝔄, then
′ ′
𝑆 𝔄 (𝐵) = 𝑆 𝔄 (𝐵) and tp𝔄 (𝑎/𝐵) = tp𝔄 (𝑎/𝐵)

If 𝔄′ ⊨ 𝑝(𝑥) then 𝔄′ ⊨ ∃𝑥𝑝(𝑥), so 𝔄 ⊨ ∃𝑥𝑝(𝑥).

We use the notation tp(𝑎) for tp(𝑎/∅)
Maximal finitely satisfiable sets of formulas in 𝑥1 , … , 𝑥𝑛 are called 𝑛-
types and
𝔄
𝑆𝑛 (𝐵) = 𝑆𝑁 (𝐵)
denotes the set of 𝑛-types over 𝐵.

tp(𝐶/𝐵) = {𝜑(𝑥𝑐1 , … , 𝑥𝑐𝑛 ) ∣ 𝔄 ⊨ 𝜑(𝑐1 , … , 𝑐𝑛 ), 𝜑 an 𝐿(𝐵)-formula}

13
Corollary 2.14. Every structure 𝔄 has an elementary extension 𝔅 in which all
types over 𝐴 are realised

Proof. We choose for every 𝑝 ∈ 𝑆(𝐴) a new constant 𝑐𝑝 . We have to find a

model of
Th(𝔄𝐴 ) ∪ ⋃ 𝑝(𝑐𝑝 )
𝑝∈𝑆(𝐴)

This theory is finitely satisfiable since every 𝑝 is finitely satisfiable in 𝔄.

Or use Lemma 2.12. Let (𝑝𝛼 )𝛼<𝜆 be an enumeration of 𝑆(𝐴). Construct
an elementary chain

𝔄 = 𝔄0 ≺ 𝔄1 ≺ ⋯ ≺ 𝔄𝛽 ≺ … (𝛽 ≤ 𝜆)

s.t. each 𝑝𝛼 is realised in 𝔄𝛼+1 (by recursion theorem on ordinal numbers)

Suppose that the elementary chain (𝔄𝛼′ )𝛼′ <𝛽 is already constructed. If 𝛽
is a limit ordinal, we let 𝔄𝛽 = ⋃𝛼<𝛽 𝔄𝛼 , which is elementary by Lemma 2.4.
If 𝛽 = 𝛼+1 we first note that 𝑝𝛼 is also finitely satisfiable in 𝔄𝛼 , therefore we
can realise 𝑝𝛼 in a suitable elementary extension 𝔄𝛽 ≻ 𝔄𝛼 by Lemma 2.12.
Then 𝔅 = 𝔄𝜆 is the model we were looking for

2.3 The Löwenheim-Skolem Theorem

Theorem 2.15 (Löwenheim-Skolem). Let 𝔅 be an 𝐿-structure, 𝑆 a subset of 𝐵
and 𝜅 an infinite cardinal

1. If
max(|𝑆|, |𝐿|) ≤ 𝜅 ≤ |𝐵|
then 𝔅 has an elementary substructure of cardinality 𝜅 containing 𝑆

2. If 𝔅 is infinite and
max(|𝔅|, |𝐿|) ≤ 𝜅
then 𝔅 has an elementary extension of cardinality 𝜅

Proof. 1. Choose a set 𝑆 ⊆ 𝑆 ′ ⊆ 𝐵 of cardinality 𝜅 and apply Corollary

2.3

2. We first construct an elementary extension 𝔅′ of cardinality at least 𝜅.

Choose a set 𝐶 of new constants of cardinality 𝜅. As 𝔅 is infinite, the
theory
Th(𝔅𝐵 ) ∪ {¬𝑐=𝑑̇ ∣ 𝑐, 𝑑 ∈ 𝐶, 𝑐 ≠ 𝑑}

14
 is finitely satisfiable. By Lemma 2.1 any model (𝔅′𝐵 , 𝑏𝑐 )𝑐∈𝐶 is an ele-
mentary extension of ℬ with 𝜅 many different elements (𝑏𝑐 )
Finally we apply the first part of the theorem to ℬ′ and 𝑆 = 𝐵

Corollary 2.16. A theory which has an infinite model has a model in every cardi-
nality 𝜅 ≥ max(|𝐿|, ℵ0 )
Definition 2.17. Let 𝜅 be an infinite cardinal. A theory 𝑇 is called 𝜅-categorical
if for all models of 𝑇 of cardinality 𝜅 are isomorphic
Theorem 2.18 (Vaught’s Test). A 𝜅-categorical theory 𝑇 is complete if the fol-
lowing conditions are satisfied
1. 𝑇 is consistent
2. 𝑇 has no finite model
3. |𝐿| ≤ 𝜅
Proof. We have to show that all models 𝔄 and 𝔅 of 𝑇 are elemantarily equiv-
alent. As 𝔄 and 𝔅 are infinite, Th(𝔄) and Th(𝔅) have models 𝔄′ and 𝔅′ of
cardinality 𝜅. By assumption 𝔄′ and 𝔅′ are isomorphic, and it follows that
𝔄 ≡ 𝔄 ′ ≡ 𝔅′ ≡ 𝔅

Example 2.1. 1. The theory DLO of dense linear orders without endpoints
is ℵ0 -categorical and by Vaught’s test complete. Let 𝐴 = {𝑎𝑖 ∣ 𝑖 ∈ 𝜔},
𝐵 = {𝑏𝑖 ∣ 𝑖 ∈ 𝜔}. We inductively define sequences (𝑐𝑖 )𝑖<𝜔 , (𝑑𝑖 )𝑖<𝜔 ex-
hausting 𝐴 and 𝐵. Assume that (𝑐𝑖 )𝑖<𝑚 , (𝑑𝑖 )𝑖<𝑚 have defined so that
𝑐𝑖 ↦ 𝑑𝑖 , 𝑖 < 𝑚 is an order isomorphism. If 𝑚 = 2𝑘 let 𝑐𝑚 = 𝑎𝑗 where
𝑎𝑗 is the element with minimal index in {𝑎𝑖 ∣ 𝑖 ∈ 𝜔} not occurring
in (𝑐𝑖 )𝑖<𝑚 . Since 𝔅 is a dense linear order without endpoints there
is some element 𝑑𝑚 ∈ {𝑏𝑖 ∣ 𝑖 ∈ 𝜔} s.t. (𝑐𝑖 )𝑖≤𝑚 and (𝑑𝑖 )𝑖≤𝑚 are order
isomorphic. If 𝑚 = 2𝑘 + 1 we interchange the roles of 𝔄 and 𝔅
2. For any prime 𝑝 or 𝑝 = 0, the theory ACF𝑝 of algebraically closed fields
of characteristic 𝑝 is 𝜅-categorical for any 𝜅 > ℵ0
Consider the Theorem 2.18 we strengthen our definition
Definition 2.19. Let 𝜅 be an infinite cardinal. A theory 𝑇 is called 𝜅-categorical
if it is complete, |𝑇 | ≤ 𝜅 and, up to isomorphism, has exactly one model of
cardinality 𝜅

15
3 Quantifier Elimination
3.1 Preservation theorems
Lemma 3.1 (Separation Lemma). Let 𝑇1 , 𝑇2 be two theories. Assume ℋ is a set
of sentences which is closed under ∧, ∨ and contains ⊥ and ⊤. Then the following
are equivalent

1. There is a sentence 𝜑 ∈ ℋ which separates 𝑇1 from 𝑇2 . This means

𝑇1 ⊨ 𝜑 and 𝑇2 ⊨ ¬𝜑

2. All models 𝔄1 of 𝑇1 can be separated from all models 𝔄2 of 𝑇2 by a sentence

𝜑 ∈ ℋ. This means

𝔄1 ⊨ 𝜑 and 𝔄2 ⊨ ¬𝜑

For 1, suppose 𝑇1 = 𝑇 ∪ {𝜓} and 𝑇2 = 𝑇 ∪ {¬𝜓}. If 𝑇1 ⊨ 𝜑 and 𝑇2 ⊨ ¬𝜑,

then 𝑇 ⊨ 𝜓 → 𝜑 and 𝑇 ⊨ ¬𝜓 → ¬𝜑 which is equivalent to 𝑇 ⊨ 𝜑 → 𝜓. Thus
we have 𝑇 ⊨ 𝜑 ↔ 𝜓.

Proof. 2 → 1. For any model 𝔄1 of 𝑇1 let ℋ𝔄1 be the set of all sentences
from ℋ which are true in 𝔄1 . (2) implies that ℋ𝔄1 and 𝑇2 cannot have a
common model. By the Compactness Theorem there is a finite conjunction
𝜑𝔄1 of sentences from ℋ𝔄1 inconsistent with 𝑇2 . Clearly

𝑇1 ∪ {¬𝜑𝔄1 ∣ 𝔄1 ⊨ 𝑇1 }

is inconsistent. Again by compactness 𝑇1 implies a disjunction 𝜑 of finitely

many of the 𝜑𝔄1 (Corollary 2.10) and

𝑇1 ⊨ 𝜑 and 𝑇2 ⊨ ¬𝜑

For structures 𝔄, 𝔅 and a map 𝑓 ∶ 𝐴 → 𝐵 preserving all formulas from

a set of formulas Δ, we use the notation

𝑓 ∶ 𝔄 →Δ 𝔅

We also write
𝔄 ⇒Δ 𝔅
to express that all sentences from Δ true in 𝔄 are also true in 𝔅

16
Lemma 3.2. Let 𝑇 be a theory, 𝔄 a structure and Δ a set of formulas, closed un-
der existential quantification, conjunction and substitution of variables. Then the
following are equivalent

1. All sentences 𝜑 ∈ Δ which are true in 𝔄 are consistent with 𝑇

2. There is a model 𝔅 ⊨ 𝑇 and a map 𝑓 ∶ 𝔄 →Δ 𝔅

Proof. 2 → 1. Assume 𝑓 ∶ 𝔄 →Δ 𝔅 ⊨ 𝑇 . If 𝜑 ∈ Δ is true in 𝔄, it is also true

in 𝔅 and therefore consistent with 𝑇 .
1 → 2. Consider ThΔ (𝔄𝐴 ), the set of all sentences 𝛿(𝑎) (𝛿(𝑥) ∈ Δ),
which are true in 𝔄𝐴 . The models (𝔅, 𝑓(𝑎)𝑎∈𝐴 ) of this theory correspond
to maps 𝑓 ∶ 𝔄 →Δ 𝔅. This means that we have to find a model of 𝑇 ∪
ThΔ (𝔄𝐴 ). To show finite satisfiability it is enough to show that 𝑇 ∪ 𝐷 is
consistent for every finite subset 𝐷 of ThΔ (𝔄𝐴 ). Let 𝛿(𝑎) be the conjunction
of the elements of 𝐷. Then 𝔄 is a model of 𝜑 = ∃𝑥𝛿(𝑥),
̄ so by assumption 𝑇
has a model 𝔅 which is also a model of 𝜑. This means that there is a tuple
𝑏 s.t. (𝔅, 𝑏) ⊨ 𝛿(𝑎)

Lemma 3.2 applied to 𝑇 = Th(𝔅) shows that 𝔄 ⇒Δ 𝔅 iff there exists a

map 𝑓 and a structure 𝔅′ ≡ 𝔅 s.t. 𝑓 ∶ 𝔄 →Δ 𝔅′

Theorem 3.3. Let 𝑇1 and 𝑇2 be two theories. Then the following are equivalent

1. There is a universal sentence which separates 𝑇1 from 𝑇2

2. No model of 𝑇2 is a substructure of a model of 𝑇1

Proof. 1 → 2. Let 𝜑 be a universal sentence which separates 𝑇1 and 𝑇2 . Let

𝔄1 be a model of 𝑇1 and 𝔄2 a substructure of 𝔄1 . Since 𝔄1 is a model of 𝜑,
𝔄2 is also a model of 𝜑. Therefore 𝔄2 cannot be a model of 𝑇2
2 → 1. Here we add some details for the proof 2 → 1. If 𝑇1 and 𝑇2 can-
not be separated by a universal sentence, then they have models 𝔄1 and 𝔄2
which cannot be separated by a universal sentence. That is, for all universal
sentence 𝜑, if 𝔄1 ⊨ 𝜑 then 𝔄2 ⊨ 𝜑. Thus 𝔄1 ⇒∀ 𝔄2 , here ⇒∀ means for all
universal sentence.
Now note that

𝔄1 ⊨ 𝜑 → 𝔄 2 ⊨ 𝜑 ⇔ 𝔄2 ⊨ ¬𝜑 → 𝔄2 ⊨ ¬𝜑

and ¬𝜑 is an existential sentence. Hence we have

𝔄2 ⇒∃ 𝔄1

17
 The reason that we want to use ∃ is that it holds in the substructure case
and we could imagine that 𝔄2 ⊆ 𝔄1 (I guess this is our intuition). Now
by Lemma 3.2 we have 𝔄1′ ≡ 𝔄1 and a map 𝑓 ∶ 𝔄2 →∃ 𝔄1′ . Apparently
𝔄1′ ⊨ Diag(𝔄2 ) and 𝑓 is an embedding. Hence 𝔄1′ is a model of 𝑇1 and
𝑇2

Definition 3.4. For any 𝐿-theory 𝑇 , the formulas 𝜑(𝑥), 𝜓(𝑥) are said to be
equivalent modulo 𝑇 (or relative to 𝑇 ) if 𝑇 ⊨ ∀𝑥(𝜑(𝑥) ↔ 𝜓(𝑥))

Corollary 3.5. Let 𝑇 be a theory

1. Consider a formula 𝜑(𝑥1 , … , 𝑥𝑛 ). The following are equivalent

(a) 𝜑(𝑥1 , … , 𝑥𝑛 ) is, modulo 𝑇 , equivalent to a universal formula

(b) If 𝔄 ⊆ 𝔅 are models of 𝑇 and 𝑎1 , … , 𝑎𝑛 ∈ 𝐴, then 𝔅 ⊨ 𝜑(𝑎1 , … , 𝑎𝑛 )
implies 𝔄 ⊨ 𝜑(𝑎1 , … , 𝑎𝑛 )

2. We say that a theory which consists of universal sentences is universal. Then

𝑇 is equivalent to a universal theory iff all substructures of models of 𝑇 are
again models of 𝑇
Proof. 1. Assume (2). We extend 𝐿 by an 𝑛-tuple 𝑐 of new constants
𝑐1 , … , 𝑐𝑛 and consider theory

𝑇1 = 𝑇 ∪ {𝜑(𝑐)} and 𝑇2 = 𝑇 ∪ {¬𝜑(𝑐)}

Then (2) says the substructures of models of 𝑇1 cannot be models of

𝑇2 . By Theorem 3.3 𝑇1 and 𝑇2 can be separated by a universal 𝐿(𝑐)-
sentence 𝜓(𝑐). By Lemma 1.16, 𝑇1 ⊨ 𝜓(𝑐) implies

𝑇 ⊨ ∀𝑥(𝜑(𝑥) → 𝜓(𝑥))

and from 𝑇2 ⊨ ¬𝜓(𝑐) we see

𝑇 ⊨ ∀𝑥(¬𝜑(𝑥) → ¬𝜓(𝑥))

2. Suppose a theory 𝑇 has this property. Let 𝜑 be an axiom of 𝑇 . If 𝔄

is a substructure of 𝔅, it is not possible for 𝔅 to be a model of 𝑇 and
for 𝔄 to be a model of ¬𝜑 at the same time. By Theorem 3.3 there is a
universal sentence 𝜓 with 𝑇 ⊨ 𝜓 and ¬𝜑 ⊨ ¬𝜓. Hence all axioms of 𝑇
follow from
𝑇∀ = {𝜓 ∣ 𝑇 ⊨ 𝜓, 𝜓 universal}

18
 An ∀∃-formula is of the form

∀𝑥1 … 𝑥𝑛 𝜓

where 𝜓 is existential

Lemma 3.6. Suppose 𝜑 is an ∀∃-sentence, (𝔄𝑖 )𝑖∈𝐼 is a directed family of models

of 𝜑 and 𝔅 the union of the 𝔄𝑖 . Then 𝔅 is also a model of 𝜑.

Proof. Write
𝜑 = ∀𝑥𝜓(𝑥)
where 𝜓 is existential. For any 𝑎 ∈ 𝐵 there is an 𝐴𝑖 containing 𝑎, clearly 𝜓(𝑎)
holds in 𝔄𝑖 . As 𝜓(𝑎) is existential it must also hold in 𝔅

Definition 3.7. We call a theory 𝑇 inductive if the union of any directed

family of models of 𝑇 is again a model

Theorem 3.8. Let 𝑇1 and 𝑇2 be two theories. Then the following are equivalent

1. there is an ∀∃-sentence which separates 𝑇1 and 𝑇2

2. No model of 𝑇2 is the union of a chain (or of a directed family) of models of

𝑇1

Proof. 1 → 2. Assume 𝜑 is a ∀∃-sentence which separates 𝑇1 from 𝑇2 , (𝔄𝑖 )𝑖∈𝐼

is a directed family of models of 𝜑, by Lemma 3.6 𝔅 is also a model of 𝜑.
Since 𝔅 ⊨ 𝜑, 𝔅 cannot be a model of 𝑇2
2 → 1. If (1) is not true, Suppose 𝔄 ⊨ 𝑇1 and 𝔅0 ⊨ 𝑇2 . Then

𝔄 ⇒∀∃ 𝔅0

Again we have
𝔅0 ⇒∃∀ 𝔄
we have a map
𝑓 ′ ∶ 𝔅0 →∃∀ 𝔄0
where 𝔄0 ≡ 𝔄. Since ∀-sentences are also ∃∀-sentences, we thus have a
map 𝑓 ∶ 𝔅0 →∀ 𝔄0 .
Here we need to prove that 𝔅0 is isomorphic to a substructure of 𝔄0 ,
which is clear since 𝑓 is an embedding. Then we can assume that 𝔅0 ⊆ 𝔄0
and 𝑓 is the inclusion map. Then
0
𝔄𝐵 ⇒∃ 𝔅0𝐵

19
(Here we are talking about existential sentences in the original language. If
𝔅0 ⊨ ∃𝑥𝜑(𝑥) for some 𝜑(𝑥), then 𝔅0 ⊨ 𝜑(𝑏). So we can use constants 𝐵
to talk about existential sentences) Applying Lemma 3.2 again, we obtain
an extension 𝔅1𝐵 of 𝔄𝐵 0
with 𝔅1𝐵 ≡ 𝔅0𝐵 , i.e. 𝔅0 ≺ 𝔅1 . Hence we have an
infinite chain

𝔅0 ⊆ 𝔄0 ⊆1 𝔅1 ⊆ 𝔄1 ⊆ 𝔅2 ⊆ ⋯
𝔅0 ≺ 𝔅1 ≺ 𝔅2 ≺ ⋯
𝔄𝑖 ≡ 𝔄

Let 𝔅 be the union of the 𝔄𝑖 . Since 𝔅 is also the union of the elementary
chain of the 𝔅𝑖 , it is an elementary extension of 𝔅0 and hence a model of
𝑇2 . But the 𝔄𝑖 are models of 𝑇1 , so (2) does not hold

Corollary 3.9. Let 𝑇 be a theory

1. For each sentence 𝜑 the following are equivalent

(a) 𝜑 is, modulo 𝑇 , equivalent to an ∀∃-sentence

(b) If
𝔄0 ⊆ 𝔄1 ⊆ ⋯
and their union 𝔅 are models of 𝑇 , then 𝜑 holds in 𝔅 if it is true in all
the 𝔄𝑖

2. 𝑇 is inductive iff it can be axiomatised by ∀∃-sentences

Proof. 1. Theorem 3.6 shows that ∀∃-formulas are preserved by unions
of chains. Hence (a)⇒(b). For the converse consider the theories

𝑇1 = 𝑇 ∪ {𝜑} and 𝑇2 = 𝑇 ∪ {¬𝜑}

Part (b) says that the union of a chain of models of 𝑇1 cannot be a

model of 𝑇2 . By Theorem 3.8 we can separate 𝑇1 and 𝑇2 by an ∀∃-
sentence 𝜓. Hence 𝑇 ∪ {𝜑} ⊨ 𝜓 and 𝑇 ∪ {¬𝜑} ⊨ ¬𝜓

2. Clearly ∀∃-axiomatised theories are inductive. For the converse as-

sume that 𝑇 is inductive and 𝜑 is an axiom of 𝑇 . Ifpp 𝔅 is a union
of models of 𝑇 , it cannot be a model of ¬𝜑. By Theorem 3.8 there is
an ∀∃-sentence 𝜓 with 𝑇 ⊨ 𝜓 and ¬𝜑 ⊨ ¬𝜓. Hence all axioms of 𝑇
follows from
𝑇∀∃ = {𝜓 ∣ 𝑇 ⊨ 𝜓, 𝜓 ∀∃-formula}

20
Exercise 3.1.1. Let 𝑋 be a topological space, 𝑌1 and 𝑌2 quasi-compact (com-
pact but not necessarily Hausdorff) subsets, and ℋ a set of clopen subsets.
Then the following are equivalent
1. There is a positive Boolean combination 𝐵 of elements from ℋ s.t.
𝑌1 ⊆ 𝐵 and 𝑌2 ∩ 𝐵 = ∅
2. For all 𝑦1 ∈ 𝑌1 and 𝑦2 ∈ 𝑌2 there is an 𝐻 ∈ ℋ s.t. 𝑦1 ∈ 𝐻 and 𝑦2 ∉ 𝐻
Proof. 2 → 1. Consider an element 𝑦1 ∈ 𝑌1 and ℋ𝑦1 , the set of all elements
of ℋ containing 𝑦1 . 2 implies that the intersection of the sets in ℋ𝑦1 is dis-
joint from 𝑌2 . So a finite intersection ℎ𝑦1 of elements of ℋ𝑦1 is disjoint from
𝑌2 . The ℎ𝑦𝑖 , 𝑦1 ∈ 𝑌1 , cover 𝑌1 . So 𝑌1 is contained in the union 𝐻 of finitely
many of the ℎ𝑦𝑖 . Hence 𝐻 separates 𝑌1 from 𝑌2

3.2 Quantifier elimination

Definition 3.10. A theory 𝑇 has quantifier elimination if every 𝐿-formula
𝜑(𝑥1 , … , 𝑥𝑛 ) in the theory is equivalent modulo 𝑇 to some quantifier-free
formula 𝜌(𝑥1 , … , 𝑥𝑛 )
For 𝑛 = 0, this means that modulo 𝑇 every sentence is equivalent to
a quantifier-free sentence. If 𝐿 has no constants, ⊤ and ⊥ are the only
quantifier free sentences. Then 𝑇 is either inconsistent or complete.
It’s easy to transform any theory 𝑇 into a theory with quantifier elim-
ination if one is willing to expand the language: just enlarge 𝐿 by adding
an 𝑛-place relation symbol 𝑅𝜑 for every 𝐿-formula 𝜑(𝑥1 , … , 𝑥𝑛 ) and 𝑇 by
adding all axioms
∀𝑥1 , … , 𝑥𝑛 (𝑅𝜑 (𝑥1 , … , 𝑥𝑛 ) ↔ 𝜑(𝑥1 , … , 𝑥𝑛 ))
The resulting theory, the Morleyisation 𝑇 𝑚 of 𝑇 , has quantifier elimination
A prime structure of 𝑇 is a structure which embeds into all models of 𝑇
Lemma 3.11. A consistent theory 𝑇 with quantifier elimination which possess a
prime structure is complete
Proof. If 𝔐, 𝔑 ⊨ 𝑇 and 𝔐 ⊨ 𝜑 and 𝔑 ⊨ ¬𝜑. Suppose prime structure is ℌ,
then ℌ ⊨ 𝜑 and ℌ ⊨ ¬𝜑 since we have quantifier elimination

Definition 3.12. A simple existential formula has the form

𝜑 = ∃𝑦𝜌
for a quantifier-free formula 𝜌. If 𝜌 is a conjunction of basic formulas, 𝜑 is
called primitive existential

21
Lemma 3.13. The theory 𝑇 has quantifier elimination iff every primitive existential
formula is, modulo 𝑇 , equivalent to a quantifier-free formula

Proof. We can write every simple existential formula in the form ∃𝑦 ⋁𝑖<𝑛 𝜌𝑖
for 𝜌𝑖 which are conjunctions of basic formulas. This shows that every sim-
ple existential formula is equivalent to a disjunction of primitive existential
formulas, namely to ⋁𝑖<𝑛 (∃𝑦𝜌𝑖 ). We can therefore assume that every simple
existential formula is, modulo 𝑇 , equivalent to a quantifier-free formula
We are now able to eliminate the quantifiers in arbitrary formulas in
prenex normal form (Exercise 1.2.1)

𝑄1 𝑥1 … 𝑄𝑛 𝑥𝑛 𝜌

if 𝑄𝑛 = ∃, we choose a quantifier-free formula 𝜌0 which, modulo 𝑇 , is equiv-

alent to ∃𝑥𝑛 𝜌 and proceed with the formula 𝑄1 𝑥1 … 𝑄𝑛−1 𝑥𝑛−1 𝜌0 . If 𝑄𝑛 = ∀,
we find a quantifier-free 𝜌1 which is, modulo 𝑇 , equivalent to ∃𝑥𝑛 ¬𝜌 and
proceed with 𝑄1 𝑥1 … 𝑄𝑛−1 𝑥𝑛−1 ¬𝜌1

Theorem 3.14. For a theory 𝑇 the following are equivalent

1. 𝑇 has quantifier elimination

2. For all models 𝔐1 and 𝔐2 of 𝑇 with a common substructure 𝔄 we have

1 2
𝔐𝐴 ≡ 𝔐𝐴

3. For all models 𝔐1 and 𝔐2 of 𝑇 with a common substructure 𝔄 and for all
primitive existential formulas 𝜑(𝑥1 , … , 𝑥𝑛 ) and parameter 𝑎1 , … , 𝑎𝑛 from
𝐴 we have
𝔐1 ⊨ 𝜑(𝑎1 , … , 𝑎𝑛 ) ⇒ 𝔐2 ⊨ 𝜑(𝑎1 , … , 𝑎𝑛 )
(this is exactly the equivalence relation)

If 𝐿 has no constants, 𝔄 is allowed to be the empty “structure”

Proof. 1 → 2. Let 𝜑(𝑎) be an 𝐿(𝐴)-sentence which holds in 𝔐1 . Choose a

quantifier-free 𝜌(𝑥) which is, modulo 𝑇 , equivalent to 𝜑(𝑥). Then
𝔐1 ⊨ 𝜑(𝑎) ⇒ 𝔐1 ⊨ 𝜌(𝑎)
⇒ 𝔄 ⊨ 𝜌(𝑎)) ⇒
𝔐2 ⊨ 𝜌(𝑎)) ⇒ 𝔐2 ⊨ 𝜑(𝑎)

22
 3 → 1. Let 𝜑(𝑥) be a primitive existential formula. In order to show that
𝜑(𝑥) is equivalent, modulo 𝑇 , to a quantifier-free formula 𝜌(𝑥) we extend
𝐿 by an 𝑛-tuple 𝑐 of new constants 𝑐1 , … , 𝑐𝑛 . We have to show that we can
separate 𝑇 ∪ {𝜑(𝑐)} and 𝑇 ∪ {¬𝜑(𝑐)} by a quantifier free sentence 𝜌(𝑐).
Then 𝑇 ⊨ 𝜑(𝑐) → 𝜌(𝑐) and 𝑇 ⊨ ¬𝜑(𝑐) → ¬𝜌(𝑐). Hence 𝑇 ⊨ 𝜑(𝑐) ↔ 𝜌(𝑐).
We apply the Separation Lemma (ℋ hear is the set of quantifier-free sen-
tence). Let 𝔐1 and 𝔐2 be two models of 𝑇 with two distinguished 𝑛-tuples
𝑎1 and 𝑎2 . Suppose that (𝔐1 , 𝑎1 ) and (𝔐2 , 𝑎2 ) satisfy the same quantifier-
free 𝐿(𝑐)-sentences. We have to show that

𝔐1 ⊨ 𝜑(𝑎1 ) ⇒ 𝔐2 ⊨ 𝜑(𝑎2 ) (⋆)

which says that if 𝑇 ’s model 𝔄1 , 𝔄2 satisfies the same quantifier-free sen-

tences, then 𝔐1 ⇒∃ 𝔐2 . If 𝔐1 ⊨ 𝑇 ∪ {𝜑(𝑐)} and 𝔐2 ⊨ 𝑇 ∪ {¬𝜑(𝑐)} and
satisfy the same quantifier-free 𝐿(𝑐) sentence, then 𝔐1 ⊆ 𝔐2 , a contradic-
tion. Thus we finish the proof
𝑖
Consider the substructure 𝔄𝑖 = ⟨𝑎𝑖 ⟩𝔐 , generated by 𝑎𝑖 . If we can show
that there is an isomorphism

𝑓 ∶ 𝔄1 → 𝔄2

taking 𝑎 to 𝑎, we may assume that 𝔄1 = 𝔄2 = 𝔄 and 𝑎1 = 𝑎2 = 𝑎. Then ⋆

follows directly from 3.
1
Every element of 𝔄1 has the form 𝑡𝔐 [𝑎1 ] for an 𝐿-term 𝑡(𝑥). The iso-
morphism 𝑓to be constructed must satisfy
1 2
𝑓(𝑡𝔐 [𝑎1 ]) = 𝑡𝔐 [𝑎2 ]

We define 𝑓 by this equation and have to check that 𝑓 is well defined and
injective. Assume
1 1
𝑠𝔐 [𝑎1 ] = 𝑡𝔐 [𝑎𝑓 1 ]
Then 𝔐1 , 𝑎1 ⊨ 𝑠(𝑐)=𝑡(𝑐),
̇ and by our assumption, 𝔐1 and 𝔐2 satisfy the
same quantifier-free 𝐿(𝑐)-sentence, it also holds in (𝔐2 , 𝑎2 ), which means
2 2
𝑠𝔐 [𝑎2 ] = 𝑡𝔐 [𝑎2 ]

Swapping the two sides yields injectivity.

Surjectivity is clear. It remains to show that 𝑓 commutes with the inter-
pretation of the relation symbols. Now
1 1
𝔐1 ⊨ 𝑅 [𝑡𝔐 1 𝔐 1
1 [𝑎 ], … , 𝑡𝑚 [𝑎 ]]

23
is equivalent to (𝔐1 , 𝑎1 ) ⊨ 𝑅(𝑡1 (𝑐), … , 𝑡𝑚 (𝑐)), which is equivalent to (𝔐2 , 𝑎2 ) ⊨
𝑅(𝑡1 (𝑐), … , 𝑡𝑚 (𝑐)), which in turn is equivalent to
2 2
𝔐2 ⊨ 𝑅 [𝑡𝔐 2 𝔐 2
1 [𝑎 ], … , 𝑡𝑚 [𝑎 ]]

Note that (2) of Theorem 3.14 is saying that 𝑇 is substructure complete;

i.e., for any model 𝔐 ⊨ 𝑇 and substructure 𝔄 ⊆ 𝔐 the theory 𝑇 ∪ Diag(𝔄)
is complete

Definition 3.15. We call 𝑇 model complete if for all models 𝔐1 and 𝔐2 of

𝑇
𝔐1 ⊆ 𝔐2 ⇒ 𝔐1 ≺ 𝔐2

𝑇 is model complete iff for any 𝔐 ⊨ 𝑇 the theory 𝑇 ∪ Diag(𝔐) is com-

plete
Note that if 𝔐1 ⊨ Diag(𝔐), then there is an embedding ℎ ∶ 𝔐 → 𝔐1
and 𝔐1 is isomorphic to an extension 𝔐1′ of 𝔐. Then we have 𝔐 ⊆ 𝔐1′ .
So here we are actually saying that all embeddings are elementary

Lemma 3.16 (Robinson’s Test). Let 𝑇 be a theory. Then the following are equiv-
alent

1. 𝑇 is model complete

2. For all models 𝔐1 ⊆ 𝔐2 of 𝑇 and all existential sentences 𝜑 from 𝐿(𝑀 1 )

𝔐2 ⊨ 𝜑 ⇒ 𝔐1 ⊨ 𝜑

3. Each formula is, modulo 𝑇 , equivalent to a universal formula

Proof. 1 ↔ 3. Corollary 3.5

(2) and Corollary 3.5 shows that all existential sentences are, modulo
𝑇 , equivalent to a universal sentence. Then by induction we can show 3.
(Details)

If 𝔐1 ⊆ 𝔐2 satisfies (2), we call 𝔐1 existentially closed in 𝔐2 . We

denote this by
𝔐1 ≺1 𝔐2

Definition 3.17. Let 𝑇 be a theory. A theory 𝑇 ∗ is a model companion of 𝑇

if the following three conditions are satisfied

24
 1. Each model of 𝑇 can be extended to a model of 𝑇 ∗

2. Each model of 𝑇 ∗ can be extended to a model of 𝑇

3. 𝑇 ∗ is model complete

Theorem 3.18. A theory 𝑇 has, up to equivalence, at most one model companion

𝑇∗

Proof. If 𝑇 + is another model companion of 𝑇 , every model of 𝑇 + is con-

tained in a model of 𝑇 ∗ and conversely. Let 𝔄0 ⊨ 𝑇 + . Then 𝔄0 can be
embedded in a model 𝔅0 of 𝑇 ∗ . In turn 𝔅0 is contained in a model 𝔄1 of
𝑇 + . In this way we find two elementary chains (𝔄𝑖 ) and (𝔅𝑖 ), which have
a common union ℭ. Then 𝔄0 ≺ ℭ and 𝔅0 ≺ ℭ implies 𝔄0 ≡ 𝔅0 since 𝑇 are
all sentences. Thus 𝔄0 is a model of 𝑇 ∗

Existentially closed structures and the Kaiser hull

Let 𝑇 be an 𝐿-theory. It follows from 3.3 that the models of 𝑇∀ = {𝜑 ∣ 𝑇 ⊨
𝜑 where 𝜑 is universal} are the substructures of models of 𝑇 . The condi-
tions (1) and (2) in the definition of “model companion” can therefore be
expressed as
𝑇∀ = 𝑇∀∗
(1 and 2 says Mod(𝑇∀ ) = Mod(𝑇∀∗ )) Hence the model companion of a the-
ory 𝑇 depends only on 𝑇∀ .

Definition 3.19. An 𝐿-structure 𝔄 is called 𝑇 -existentiallay closed (or 𝑇 -

ec) if

1. 𝔄 can be embedded in a model of 𝑇

2. 𝔄 is existentially closed in every extension which is a model of 𝑇

A structure 𝔄 is 𝑇 -ec exactly if it is 𝑇∀ -ec. Since every model of 𝔅 of 𝑇∀

can be embedded in a model 𝔐 of 𝑇 and 𝔄 ⊆ 𝔅 ⊆ 𝔐 and 𝔄 ≺1 𝔐 implies
𝔄 ≺1 𝔅

Lemma 3.20. Every model of a theory 𝑇 can be embedded in a 𝑇 -ec structure

Proof. Let 𝔄 be a model of 𝑇∀ . We choose an enumeration (𝜑𝛼 )𝛼<𝜅 of all exis-

tential 𝐿(𝐴)-sentences and construct an ascending chain (𝔄𝛼 )𝛼≤𝜅 of models
of 𝑇∀ . We begin with 𝔄0 = 𝔄. Let 𝔄𝛼 be constructed. If 𝜑𝛼 holds in an ex-
tension of 𝔄𝛼 which is a model of 𝑇 we let 𝔄𝛼+1 be such a model. Otherwise

25
we set 𝔄𝛼+1 = 𝔄𝛼 . For limit ordinals 𝜆 we define 𝔄𝜆 to be the union of all
𝔄𝛼 . 𝔄𝜆 is again a model of 𝑇∀
The structure 𝔄1 = 𝔄𝜅 has the following property: every existential
𝐿(𝐴)-sentence which holds in an extension of 𝔄1 that is a model of 𝑇 holds
in 𝔄1 . Now in the same manner, we construct 𝔄2 from 𝔄1 , etc. The union
𝔐 of the chain 𝔄0 ⊆ 𝔄1 ⊆ 𝔄2 ⊆ … is the desired 𝑇 -ec structure

Every elementary substructure 𝔑 of a 𝑇 -ec structure 𝔐 is again 𝑇 -ec:

Let 𝔑 ⊆ 𝔄 be a model of 𝑇 . Since 𝔐𝑁 ⇒∃ 𝔄𝑁 , there is an embedding of
𝔐 in an elementary extension 𝔅 of 𝔄 which is the identity on 𝑁 . Since 𝔐
is existentially closed in 𝔅, it follows that 𝔑 is existentially closed in 𝔅 and
therefore also in 𝔄

𝔅
≺ ≺1

𝔄 𝔐

≺1 ≺
𝔑

Lemma 3.21. Let 𝑇 be a theory. Then there is a biggest inductive theory 𝑇 KH with
𝑇∀ = 𝑇∀KH . We call 𝑇 KH the Kaiser hull of 𝑇

Proof. Let 𝑇 1 and 𝑇 2 be two inductive theories with 𝑇∀1 = 𝑇∀2 = 𝑇∀ . We

have to show that (𝑇 1 ∪ 𝑇 2 )∀ = 𝑇∀ . Note that for every model 𝔄 ⊨ 𝑇 1 and
𝔅 ⊨ 𝑇 2 we have 𝔄 ⇒∀ 𝔅 and vice versa. Then we have the embeddings
just like model companions. Let 𝔐 be a model of 𝑇 , as in the proof of 3.18
we extend 𝔐 by a chain 𝔄0 ⊆ 𝔅0 ⊆ 𝔄1 ⊆ 𝔅1 ⊆ ⋯ of models of 𝑇 1 and 𝑇 2 .
The union of this chain is a model of 𝑇 1 ∪ 𝑇 2

Lemma 3.22. The Kaiser hull 𝑇 𝐾𝐻 is the ∀∃-part of the theory of all 𝑇 -ec struc-
tures

Proof. Let 𝑇 ∗ be the ∀∃-part of the theory of all 𝑇 -ec structures. Since 𝑇 -
ec structures are models of 𝑇∀ , we have 𝑇∀ ⊆ 𝑇∀∗ . It follows from 3.20 that
𝑇∀∗ ⊆ 𝑇∀ . Hence 𝑇 ∗ is contained in the Kaiser Hull.
It remains to show that every 𝑇 -ec structure 𝔐 is a model of the Kaiser
hull. Choose a model 𝔑 of 𝑇 𝐾𝐻 which contains 𝔐. Then 𝔐 ≺1 𝔑. This
implies 𝔑 ⇒∀∃ 𝔐 and therefore 𝔐 ⊨ 𝑇 𝐾𝐻

This implies that 𝑇 -ec strctures are models of 𝑇∀∃

26
Theorem 3.23. For any theory 𝑇 the following are equivalent
1. 𝑇 has a model companion 𝑇 ∗

2. All models of 𝐾 KH are 𝑇 -ec

3. The 𝑇 -ec structures form an elementary class.

If 𝑇 ∗ exists, we have

𝑇 ∗ = 𝑇 KH = theory of all 𝑇 -ec structures

Proof. 1 → 2: let 𝑇 ∗ be the model companion of 𝑇 . As a model complete

theory
3 → 1: Assume that the 𝑇 -ec structures are exactly the models of the
theory 𝑇 + . By 3.20 we have 𝑇∀ = 𝑇∀+ . Criterion 3.16 implies that 𝑇 + is
model complete. So 𝑇 + is the model companion of 𝑇 .

Exercise 3.2.1. Let 𝐿 be the language containing a unary function 𝑓 and a

binary relation symbol 𝑅 and consider the 𝐿-theory 𝑇 = {∀𝑥∀𝑦(𝑅(𝑥, 𝑦) →
(𝑅(𝑥, 𝑓(𝑦))))}. Showing the following
1. For any 𝑇 -structure 𝔐 and 𝑎, 𝑏 ∈ 𝑀 with 𝑏 ∉ {𝑎, 𝑓 𝔐 (𝑎), (𝑓 𝔐 )2 (𝑎), … }
we have 𝔐 ⊨ ∃𝑧(𝑅(𝑧, 𝑎) ∧ ¬𝑅(𝑧, 𝑏))

2. Let 𝔐 be a model of 𝑇 and 𝑎 an element of 𝑀 s.t. {𝑎, 𝑓 𝔐 (𝑎), (𝑓 𝔐 )2 (𝑎), … }

is infinite. Then in an elementary extension 𝔐′ there is an element 𝑏
with 𝔐′ ⊨ ∀𝑧(𝑅(𝑧, 𝑎) → 𝑅(𝑧, 𝑏))

3. The class of 𝑇 -ec structures is not elementary, so 𝑇 does not have a

model companion
Exercise 3.2.2. A theory 𝑇 with quantifier elimination is axiomatisable by
sentences of the form
∀𝑥1 … 𝑥𝑛 𝜓
where 𝜓 is primitive existential formula

3.3 Examples
Infinite sets. The models of the theory Infset of infinite sets are all infinite
sets without additional structure. The language 𝐿∅ is empty, the axioms are
(for 𝑛 = 1, 2, …)
• ∃𝑥0 … 𝑥𝑛−1 ⋀𝑖<𝑗<𝑛 ¬𝑥𝑖 =𝑥
̇ 𝑗

27
Theorem 3.24. The theory Infset of infinite sets has quantifier elimination and is
complete

Proof. Since the language is empty, the only basic formula is 𝑥𝑖 = 𝑥𝑗 and
¬(𝑥𝑖 = 𝑥𝑗 ). By Lemma 3.13 we only need to consider primitive existential
formulas. Then for any 𝔐1 , 𝔐2 ⊨ Infset, they have a common substructure
𝔄 with 𝜔 different elements. Suppose 𝔐1 ⊨ ∃𝑥𝜑(𝑥),

Dense linear orderings.

∀𝑎, 𝑏(𝑎 ≤ 𝑏 ∧ 𝑏 ≤ 𝑎 → 𝑎=𝑏)

̇
∀𝑎, 𝑏, 𝑐(𝑎 ≤ 𝑏 ∧ 𝑏 ≤ 𝑐 → 𝑎 ≤ 𝑐)
∀𝑎, 𝑏(𝑎 ≤ 𝑏 ∨ 𝑏 ≤ 𝑎)
∀𝑎, 𝑏∃𝑐(𝑎 < 𝑏 → 𝑎 < 𝑐 < 𝑏)

Theorem 3.25. DLO has quantifier elimination

Proof. Let 𝐴 be a finite common substructure of the two models 𝑂1 and

𝑂2 . We choose an ascending enumeration 𝐴 = {𝑎1 , … , 𝑎𝑛 }. Let ∃𝑦𝜌(𝑦) be a
simple existential 𝐿(𝐴)-sentence, which is true in 𝑂1 and assume 𝑂1 ⊨ 𝜌(𝑏1 ).
We want to extend the order preserving map 𝑎𝑖 ↦ 𝑎𝑖 to an order preserving
map 𝐴 ∪ {𝑏1 } → 𝑂2 . For this we have an image 𝑏2 of 𝑏1 . There are four cases

1. 𝑏1 ∈ 𝐴, we set 𝑏2 = 𝑏1

2. 𝑏1 ∈ (𝑎𝑖 , 𝑎𝑖+1 ). We choose 𝑏2 in 𝑂2 with the same property

3. 𝑏1 is smaller than all elements of 𝐴. We choose a 𝑏2 ∈ 𝑂2 of the same

kind

4. 𝑏1 is bigger than all 𝑎𝑖 . Choose 𝑏2 in the same manner

This defines an isomorphism 𝐴 ∪ {𝑏1 } → 𝐴 ∪ {𝑏2 }, which show that

𝑂2 ⊨ 𝜌(𝑏2 )

Modules. Let 𝑅 be a (possibly non-commutative) ring with 1. An 𝑅-

module
𝔐 = (, 0, +, −, 𝑟)𝑟∈𝑅
is an abelian group (𝑀 , 0, +, −) together with operations 𝑟 ∶ 𝑀 → 𝑀
for every ring element 𝑟 ∈ 𝑅. We formulate the axioms in the language

28
𝐿𝑀𝑜𝑑 (𝑅) = 𝐿𝐴𝑏𝐺 ∪ {𝑟 ∣ 𝑟 ∈ 𝑅}. The theory Mod(𝑅) of 𝑅-modules consists
of

AbG
∀𝑥, 𝑦 𝑟(𝑥 + 𝑦)=𝑟𝑥
̇ + 𝑟𝑦
∀𝑥 (𝑟 + 𝑠)𝑥=𝑟𝑥
̇ + 𝑠𝑥
∀𝑥 (𝑟𝑠)𝑥=𝑟(𝑠𝑥)
̇
∀𝑥 1𝑥=𝑥
̇

for all 𝑟, 𝑠 ∈ 𝑅. Then Infset ∪ Mod(𝑅) is the theory of all infinite 𝑅-modules
A module over fields is a vector space

Theorem 3.26. Let 𝐾 be a field. Then the theory of all infinite 𝐾-vector spaces
has quantifier elimination and is complete

Proof. Let 𝐴 be a common finitely generated substructure (i.e., a subspace)

of the two infinite 𝐾-vector spaces 𝑉1 and 𝑉2 . Let ∃𝑦𝜌(𝑦) be a simple exis-
tential 𝐿(𝐴)-sentence which holds in 𝑉1 . Choose a 𝑏1 from 𝑉1 which satisfies
𝜌(𝑦). If 𝑏1 belongs to 𝐴, we finished. If not, we choose a 𝑏2 ∈ 𝑉2 ⧵ 𝐴. Pos-
sibly we have to replace 𝑉2 by an elementary extension. The vector spaces
𝐴 + 𝐾𝑏1 and 𝐴 + 𝐾𝑏2 are isomorphic by an isomophism which maps 𝑏1 to
𝑏2 and fixes 𝐴 elementwise. Hence 𝑉2 ⊨ 𝜌(𝑏2 )

Definition 3.27. An equation is an 𝐿𝑀𝑜𝑑 (𝑅)-formula 𝛾(𝑥) of the form

𝑟1 𝑥1 + ⋯ + 𝑟𝑚 𝑥𝑚 = 0

A positive primitive formula (pp-formula) is of the form

∃𝑦(𝛾1 ∧ ⋯ ∧ 𝛾𝑛 )

where the 𝛾𝑖 (𝑥𝑦) are equations

Theorem 3.28. For every ring 𝑅 and any 𝑅-module 𝑀 , every 𝐿𝑀𝑜𝑑 (𝑅)-formula
is equivalent (modulo the theory of 𝑀 ) to a Boolean combination of positive prim-
itive formulas

Remark. 1. We assume the class of positive primitive formulas to be closed

under ∧

2. A pp-formula 𝜑(𝑥1 , … , 𝑥𝑛 ) defines a subgroup 𝜑(𝑀 𝑛 ) of 𝑀 𝑛 :

𝑀 ⊨ 𝜑(0) and 𝑀 ⊨ 𝜑(𝑥) ∧ 𝜑(𝑦) → 𝜑(𝑥 − 𝑦)

29
Lemma 3.29. Let 𝜑(𝑥, 𝑦) be a pp-formula and 𝑎 ∈ 𝑀 . Then 𝜑(𝑀 , 𝑎) is empty or
a coset of 𝜑(𝑀 , 0)

Proof. 𝑀 ⊨ 𝜑(𝑥, 𝑎) → (𝜑(𝑦, 0) ↔ 𝜑(𝑥 + 𝑦, 𝑎))

Or, if 𝑥, 𝑦 ∈ 𝜑(𝑀 , 𝑎), then 𝜑(𝑥 − 𝑦, 0).

Corollary 3.30. Let 𝑎, 𝑏 ∈ 𝑀 , 𝜑(𝑥, 𝑦) a pp-formula. Then (in 𝑀 ) 𝜑(𝑥, 𝑎) and

𝜑(𝑥, 𝑏) are equivalent or contradictory

Lemma 3.31 (B. H. Neumann). Let 𝐻𝑖 denote subgroups of some abelian group.
𝑛
If 𝐻0 + 𝑎0 ⊆ ⋃𝑖=1 𝐻𝑖 + 𝑎𝑖 and 𝐻0 /(𝐻0 ∩ 𝐻𝑖 ) is infinite for 𝑖 > 𝑘, then 𝐻0 + 𝑎0 ⊆
𝑘
⋃𝑖=1 𝐻𝑖 + 𝑎𝑖
𝑘
Lemma 3.32. Let 𝐴𝑖 , 𝑖 ≤ 𝑘, be any sets. If 𝐴0 is finite, then 𝐴0 ⊆ ⋃𝑖=1 𝐴𝑖 iff

∑ (−1)|Δ| ∣𝐴0 ∩ ⋂ 𝐴𝑖 ∣ = 0
Δ⊆{1,…,𝑘} 𝑖∈Δ

Algebraically closed fields.

Theorem 3.33 (Tarski). The theory ACF of algebraically closed fields has quanti-
fier elimination

1#+BEGINproof Let 𝐾1 and 𝐾2 be two algebraically closed fields and 𝑅

a common subring. There may not is such a thing since, e.g., ℚ doesn’t have
a subfield Let ∃𝑦𝜌(𝑦) be a simple existential sentence with parameters in 𝑅
which hold in 𝐾1 . We have to show that ∃𝑦𝜌(𝑦) is also true in 𝐾2 .
Let 𝐹1 and 𝐹2 be the quotient fields of 𝑅 in 𝐾1 and 𝐾2 , and let 𝑓 ∶ 𝐹1 →
𝐹2 be an isomorphism which is the identity on 𝑅. Then 𝑓 extends to an
isomorphism 𝑔 ∶ 𝐺1 → 𝐺2 between the relative algebraic closures 𝐺𝑖 of 𝐹𝑖
in 𝐾𝑖 . Choose an element 𝑏1 ∈ 𝐾1 which satisfies 𝜌(𝑦)

𝐾1 𝐾2
𝐺1 (𝑏1 )
𝑔
𝐺1 𝐺2

𝑓
𝐹1 𝐹2

𝑖𝑑
𝑅 𝑅

30
There are two cases
Case 1: 𝑏1 ∈ 𝐺1 . Then 𝑏2 = 𝑔(𝑏1 ) satisfies the formula 𝜌(𝑦) in 𝐾2
Case 2: 𝑏1 ∉ 𝐺1 . Then 𝑏1 is transcendental over 𝐺 and the field extension
𝐺1 (𝑏1 ) is isomorphic to the rational function field 𝐺1 (𝑋). If 𝐾2 is a proper
extension of 𝐺2 , we choose any element from 𝐾2 ⧵ 𝐺2 for 𝑏2 . Then 𝑔 extends
to an isomorphism between 𝐺1 (𝑏1 ) and 𝐺2 (𝑏2 ) which maps 𝑏1 to 𝑏2 . Hence
𝑏2 satisfies 𝜌(𝑦) in 𝐾2 . In case that 𝐾2 = 𝐺2 we take a proper elementary
extension 𝐾2′ of 𝐾2 (Such a 𝐾2′ exists by 2.15 since 𝐾2 is infinite). Then
∃𝑦𝜌(𝑦) holds in 𝐾2′ and therefore in 𝐾2 #+ENDproof

Corollary 3.34. ACF is model complete

ACF is not complete: for prime numbers 𝑝 let

ACF𝑝 = ACF ∪ {𝑝 ⋅ 1=0}

̇

be the theory of algebraically closed fields of characteristic 𝑝 and

ACF0 = ACF ∪ {¬𝑛 ⋅ 1=0

̇ ∣ 𝑛 = 1, 2, … }

be the theory of algebraically closed fields of characteristic 0.

Corollary 3.35. The theories ACF𝑝 and ACF0 are complete

Proof. This follows from Lemma 3.11 since the prime fields are prime struc-
tures for these theories

Corollary 3.36 (Hilbert’s Nullstellensatz). Let 𝐾 be a field. Then any proper

ideal 𝐼 in 𝐾[𝑋1 , … , 𝑋𝑛 ] has a zero in the algebraic closure acl(𝐾)

Proof. As a proper ideal, 𝐼 is contained in a maximal ideal 𝑃 . Then 𝐿 =

𝐾[𝑋1 , … , 𝑋𝑛 ]/𝑃 is an extension field of 𝐾 in which the cosets of the 𝑋𝑖 are
a zero of 𝐼.

Real closed fields. It is axiomatised in the language 𝐿𝑂𝑅𝑖𝑛𝑔 of ordered

rings

Theorem 3.37 (Tarski-Seidenberg). RCF has quantifier elimination and is com-

plete

Proof. Let (𝐾1 , <) and (𝐾2 , <) be two real closed field with a common sub-
ring 𝑅. Consider an 𝐿𝑂𝑅𝑖𝑛𝑔 (𝑅)-sentence ∃𝑦𝜌(𝑦) (for a quantifier-free 𝜌),
which holds in (𝐾1 , <). We have to show ∃𝑦𝜌(𝑦) also holds in (𝐾2 , <)

31
 We build first the quotient fields 𝐹1 and 𝐹2 of 𝑅 in 𝐾1 and 𝐾2 . By ??
there is an isomorphism 𝑓 ∶ (𝐹1 , <) → (𝐹2 , <) which fixes 𝑅. The relative
algebraic closure 𝐺𝑖 of 𝐹𝑖 in 𝐾𝑖 is a real closure of (𝐹𝑖 , <). By ?? 𝑓 extends
to an isomorphism 𝑔 ∶ (𝐺1 , <) → (𝐺2 , <)
Let 𝑏1 ∈ 𝐾1 which satisfies 𝜌(𝑦). There are two cases
Case 1: 𝑏1 ∈ 𝐺1 : Then 𝑏2 = 𝑔(𝑏1 ) satisfies 𝜌(𝑦) in 𝐾2
Case 2: 𝑏1 ∉ 𝐺1 . Then 𝑏1 is transcendental over 𝐺1 and the field exten-
sion 𝐺1 (𝑏1 ) is isomorphic to the rational function field 𝐺1 (𝑋). Let 𝐺𝑙1 be the
set of all elements of 𝐺1 which are smaller than 𝑏1 , and 𝐺𝑟1 be the set of all
elements of 𝐺1 which are larger than 𝑏1 . Then all elements of 𝐺𝑙2 = 𝑔(𝐺𝑙1 )
are smaller than all elements of 𝐺𝑟2 = 𝑔(𝐺𝑟1 ). Since fields are densely or-
dered, we find in an elementary extension (𝐾2′ , <) of (𝐾2 , <) an element 𝑏2
which lies between the elements of 𝐺𝑙2 and the elements of 𝐺𝑟2 . Since 𝑏2 is
not in 𝐺2 , it is transcendental over 𝐺2 . Hence 𝑔 extends to an isomorphism
ℎ ∶ 𝐺1 (𝑏1 ) → 𝐺2 (𝑏2 ) which maps 𝑏1 to 𝑏2
In order to how that ℎ is order preserving it suffices to show that ℎ is
order preserving on 𝐺1 [𝑏1 ]. Let 𝑝(𝑏1 ) be an element of 𝐺1 [𝑏1 ]. Corollary ??
gives us a decomposition

𝑝(𝑋) = 𝜖 ∏ (𝑋 − 𝑎𝑖 ) ∏ ((𝑋 − 𝑐𝑗 )2 + 𝑑𝑗 )
𝑖<𝑚 𝑗<𝑛

with positive 𝑑𝑗 . The sign of 𝑝(𝑏1 ) depends only on the signs of the factors
𝜖, 𝑏1 − 𝑎0 , … , 𝑏1 − 𝑎𝑚−1 . The sign of ℎ(𝑝(𝑏1 )) depends in the same way on
the signs of 𝑔(𝜖), 𝑏2 − 𝑔(𝑎0 ), … , 𝑏2 − 𝑔(𝑎𝑚−1 ). But 𝑏2 was chosen in such a
way that
𝑏1 < 𝑎𝑖 ⟺ 𝑏2 < 𝑔(𝑎𝑖 )
Hence 𝑝(𝑏1 ) is positive iff ℎ(𝑝(𝑏1 )) is positive
Finally we have

(𝐾1 , <) ⊨ 𝜌(𝑏1 ) ⇒ (𝐺1 (𝑏1 ), <) ⊨ 𝜌(𝑏1 ) ⇒ (𝐺2 (𝑏2 ), <) ⊨ 𝜌(𝑏2 ) ⇒
⇒ (𝐾2′ , <) ⊨ ∃𝑦𝜌(𝑦) ⇒ (𝐾2 , <) ⊨ ∃𝑦𝜌(𝑦)

RCF is complete since the ordered field of the rationals is a prime structure

Corollary 3.38 (Hilbert’s 17th Problem). Let (𝐾, <) be a real closed field. A
polynomial 𝑓 ∈ 𝐾[𝑋1 , … , 𝑋𝑛 ] is a sum of squares

𝑓 = 𝑔12 + ⋯ + 𝑔𝑘2

32
of rational functions 𝑔𝑖 ∈ 𝐾(𝑋1 , … , 𝑋𝑛 ) iff

𝑓(𝑎1 , … , 𝑎𝑛 ) ≥ 0

for all 𝑎1 , … , 𝑎𝑛 ∈ 𝐾

Proof. Clearly a sum of squares cannot have negative values. For the con-
verse, assume that 𝑓 is not a sum of squares. Then by Corollary ??, 𝐾(𝑋1 , … , 𝑋𝑛 )
has an ordering in which 𝑓 is negative. Since in 𝐾 the positive elements are
squares, this ordering , which we denote by <, extends the ordering of 𝐾.
Let (𝐿, <) be the real closure of (𝐾(𝑋1 , … , 𝑋𝑛 ), <). In (𝐿, <), the sentence

∃𝑥1 , … , 𝑥𝑛 𝑓(𝑥1 , … , 𝑥𝑛 ) < 0

is true. Hence it is also true in (𝐾, <)

Exercise 3.3.1. Let Graph be the theory of graphs. The theory RG of the ran-
dom graph is the extension of Graph by the following axiom scheme

∀𝑥0 … 𝑥𝑚−1 𝑦1 … 𝑦𝑛−1 ( ⋀ ¬𝑥𝑖 =𝑦

̇ 𝑗→
𝑖≠𝑗

∃𝑧( ⋀ 𝑧𝑅𝑥𝑖 ) ∧ ( ⋀ ¬𝑧𝑅𝑦𝑗 ∧ ¬𝑧 =𝑦

̇ 𝑗 ))
𝑖<𝑚 𝑗<𝑛

From here, some definitions of random graphs

Let 𝑝 ∈ [0, 1] denote the probability with which a given pair is included.
We assume all the edges have the same probability of occurrence. We de-
note the set of graphs constructed in this manner by 𝒢(𝑛, 𝑝), where 𝑛 is the
number of elements in the vertex set.

Definition 3.39. A graph 𝐺 has property 𝒫𝑖,𝑗 with 𝑖, 𝑗 = 0, 1, 2, 3, … if, for

any disjoint vertex sets 𝑉1 and 𝑉2 with |𝑉1 | ≤ 𝑖 and |𝑉2 | ≤ 𝑗, there exists a
vertex 𝑣 ∈ 𝐺 that satisfies three conditions

1. 𝑣 ∉ 𝑉1 ∪ 𝑉2

2. 𝑣 ↔ 𝑥 for every 𝑥 ∈ 𝑉1 and

3. 𝑣 ↮ 𝑦 for every 𝑦 ∈ 𝑉2

Lemma 3.40. An infinite graph 𝐺 ∈ 𝒢(ℵ0 , 𝑝) has all the properties 𝒫𝑖,𝑗 with
probability 1

33
4 Countable Models
4.1 The omitting types theorem
Definition 4.1. Let 𝑇 be an 𝐿-theory and Σ(𝑥) a set of 𝐿-formulas. A model
𝔄 of 𝑇 not realizing Σ(𝑥) is said to omit Σ(𝑥). A formula 𝜑(𝑥) isolates Σ(𝑥)
if

1. 𝜑(𝑥) is consistent with 𝑇

2. 𝑇 ⊨ ∀𝑥(𝜑(𝑥) → 𝜎(𝑥)) for all 𝜎(𝑥) ∈ Σ(𝑥)

A set of formulas is often called a partial type.

Theorem 4.2 (Omitting Types). If 𝑇 is countable and consistent and if Σ(𝑥) is

not isolated in 𝑇 , then 𝑇 has a model which omits Σ(𝑥)

If Σ(𝑥) is isolated by 𝜑(𝑥) and 𝔄 is a model of 𝑇 , then Σ(𝑥) is realised

in 𝔄 by all realisations 𝜑(𝑥). Therefore the converse of the theorem is true
for complete theories 𝑇 : if Σ(𝑥) is isolated in 𝑇 , then it is realised in every
model of 𝑇

Proof. We choose a countable set 𝐶 of new constants and extend 𝑇 to a the-

ory 𝑇 ∗ with the following properties

1. 𝑇 ∗ is a Henkin theory: for all 𝐿(𝐶)-formulas 𝜓(𝑥) there exists a con-

stant 𝑐 ∈ 𝐶 with ∃𝑥𝜓(𝑥) → 𝜓(𝑐) ∈ 𝑇 ∗

2. for all 𝑐 ∈ 𝐶 there is a 𝜎(𝑥) ∈ Σ(𝑥) with ¬𝜎(𝑐) ∈ 𝑇 ∗

We construct 𝑇 ∗ inductively as the union of an ascending chain

𝑇 = 𝑇0 ⊆ 𝑇 1 ⊆ 𝑇 1 ⊆ …

of consistent extensions of 𝑇 by finitely many axioms from 𝐿(𝐶), in each

step making an instance of (1) or (2) true.
Enumerate 𝐶 = {𝑐𝑖 ∣ 𝑖 < 𝜔} and let {𝜓𝑖 (𝑥) ∣ 𝑖 < 𝜔} be an enumeration
of the 𝐿(𝐶)-formulas
Assume that 𝑇2𝑖 is the already constructed. Choose some 𝑐 ∈ 𝐶 which
doesn’t occur in 𝑇2𝑖 ∪ {𝜓𝑖 (𝑥)} and set 𝑇2𝑖+1 = 𝑇2𝑖 ∪ {∃𝑥𝜓𝑖 (𝑥) → 𝜓𝑖 (𝑐)}.
Up to equivalence 𝑇2𝑖+1 has the form 𝑇 ∪ {𝛿(𝑐𝑖 , 𝑐)} for an 𝐿-formula
𝛿(𝑥, 𝑦) and a tuple 𝑐 ∈ 𝐶 which doesn’t contain 𝑐𝑖 . Since ∃𝑦𝛿(𝑥, 𝑦) doesn’t
isolate Σ(𝑥), for some 𝜎 ∈ Σ the formula ∃𝑦𝛿(𝑥, 𝑦) ∧ ¬𝜎(𝑥) is consistent with
𝑇 . Thus 𝑇2𝑖+2 = 𝑇2𝑖+1 ∪ {¬𝜎(𝑐𝑖 )} is consistent

34
 Take a model (𝔄′ , 𝑎𝑐 )𝑐∈𝐶 of 𝑇 ∗ . Since 𝑇 ∗ is a Henkin theory, Tarski’s
Test 2.2 shows that 𝐴 = {𝑎𝑐 ∣ 𝑐 ∈ 𝐶} is the universe of an elementary
substructure 𝔄 (Lemma 2.7). By property (2), Σ(𝑥) is omitted in 𝔄

Corollary 4.3. Let 𝑇 be countable and consistent and let

Σ0 (𝑥0 , … , 𝑥𝑛0 ), Σ1 (𝑥1 , … , 𝑥𝑛1 ), …

be a sequence of partial types. If all Σ𝑖 are not isolated, then 𝑇 has a model which
omits all Σ𝑖

Proof. If Σ0 (𝑥), Σ1 (𝑥), …. Then 𝑇2𝑖+2 = 𝑇2𝑖+1 ∪ {¬𝜎𝑚 (𝑐𝑚𝑛 )}

If Σ(𝑥1 , … , 𝑥𝑛 ), then 𝑇2𝑖+1 = 𝑇2𝑖 ∪ {∃𝑥𝜓𝑖 (𝑥) → 𝜓𝑖 (𝑐)}.
Combine the two case

4.2 The space of types

Fix a theory 𝑇 . An 𝑛-type is a maximal set of formulas 𝑝(𝑥1 , … , 𝑥𝑛 ) consis-
tent with 𝑇 . We denote by 𝑆𝑛 (𝑇 ) the set of all 𝑛-types of 𝑇 . We also write
𝑆(𝑇 ) for 𝑆1 (𝑇 ). 𝑆0 (𝑇 ) is all complete extensions of 𝑇
If 𝐵 is a subset of an 𝐿-structure 𝔄, we recover 𝑆𝑛𝔄 (𝐵) as 𝑆𝑛 (Th(𝔄𝐵 )). In
particular, if 𝑇 is complete and 𝔄 is any model of 𝑇 , we have 𝑆 𝔄 (∅) = 𝑆(𝑇 )
For any 𝐿-formula 𝜑(𝑥1 , … , 𝑥𝑛 ), let [𝜑] denote the set of all types con-
taining 𝜑.

Lemma 4.4. 1. [𝜑] = [𝜓] iff 𝜑 and 𝜓 are equivalent modulo 𝑇

2. The sets [𝜑] are closed under Boolean operations. In fact [𝜑] ∩ [𝜓] = [𝜑 ∧ 𝜓],
[𝜑] ∪ [𝜓] = [𝜑 ∨ 𝜓], 𝑆𝑛 (𝑇 ) ⧵ [𝜑] = [¬𝜑], 𝑆𝑛 (𝑇 ) = [⊤] and ∅ = [⊥]

It follows that the collection of sets of the form [𝜑] is closed under finite
intersection and includes 𝑆𝑛 (𝑇 ). So these sets form a basis of a topology on
𝑆𝑛 (𝑇 )
In this book, compact means finite cover and Hausdorff

Lemma 4.5. The space 𝑆𝑛 (𝑇 ) is 0-dimensional and compact

Proof. Being 0-dimensional means having a basis of clopen sets. Our basic
open sets are clopen since their complements are also basic open
If 𝑝 and 𝑞 are two different types, there is a formula 𝜑 contained in 𝑝 but
not in 𝑞. It follows that [𝜑] and [¬𝜑] are open sets which separate 𝑝 and 𝑞.
This shows that 𝑆𝑛 (𝑇 ) is Hausdorff

35
 To prove compactness, we need to show that any collection of closed
subsets of 𝑋 with the finite intersection property has nonempty intersection.
Could check this
Consider a family [𝜑𝑖 ] (𝑖 ∈ 𝐼), with the finite intersection property.This
means that 𝜑𝑖𝑖 ∧ ⋯ ∧ 𝜑𝑖𝑘 are consistent with 𝑇 . So Corollary 2.10 {𝜑𝑖 ∣ 𝑖 ∈ 𝐼}
is consistent with 𝑇 and can be extended to a type 𝑝, which then belongs to
all [𝜑𝑖 ].

Lemma 4.6. All clopen subsets of 𝑆𝑛 (𝑇 ) has the form [𝜑]

𝑘
Proof. For any open sets 𝑂, 𝑂 = ⋃[𝜑𝑖 ]. As 𝑆𝑛 (𝑇 ) is compact, 𝑂 = ⋃𝑖=1 [𝜑𝑖 ] =
𝑘
[⋁𝑖=1 𝜑𝑖 ]
Closed subset of a compact space is compact. It follows from Exercise
3.1.1 that we can separate any two disjoint closed subsets of 𝑆𝑛 (𝑇 ) by a basic
open set.

The Stone duality theorem asserts that the map

𝑋 ↦ {𝐶 ∣ 𝐶 clopen subset of 𝑋}

yields an equivalence between the category of 0-dimensional compact spaces

and the category of Boolean algebras. The inverse map assigns to every
Boolean algebra to its Stone space
Definition 4.7. A map 𝑓 from a subset of a structure 𝔄 to a structure 𝔅 is el-
ementary if it preserves the truth of formulas; i.e., 𝑓 ∶ 𝐴0 → 𝐵 is elementary
if for every formula 𝜑(𝑥1 , … , 𝑥𝑛 ) and 𝑎 ∈ 𝐴0 we have

𝔄 ⊨ 𝜑(𝑎) ⇒ 𝔅 ⊨ 𝜑(𝑓(𝑎))

Lemma 4.8. Let 𝔄 and 𝔅 be 𝐿-structures, 𝐴0 and 𝐵0 subsets of 𝐴 and 𝐵, re-

spectively. Any elementary map 𝐴0 → 𝐵0 induces a continuous surjective map
𝑆𝑛 (𝐵0 ) → 𝑆𝑛 (𝐴0 )
Proof. If 𝑞(𝑥) ∈ 𝑆𝑛 (𝐵0 ), we define

𝑆(𝑓)(𝑞) = {𝜑(𝑥1 , … , 𝑥𝑛 , 𝑎) ∣ 𝑎 ∈ 𝐴0 , 𝜑(𝑥1 , … , 𝑥𝑛 , 𝑓(𝑎)) ∈ 𝑞(𝑥)}

If 𝜑(𝑥, 𝑓(𝑎)) ∉ 𝑞(𝑥), then 𝔅 ⊭ 𝜑(𝑥, 𝑎). Therefore 𝔄 ⊭ 𝜑(𝑥, 𝑎). 𝑆(𝑓) defines a
map from 𝑆𝑛 (𝐵0 ) to 𝑆𝑛 (𝐴0 ). Moreover, it is surjective since {𝜑(𝑥1 , … , 𝑥𝑛 , 𝑓(𝑎)) ∣
𝜑(𝑥1 , … , 𝑥𝑛 , 𝑎) ∈ 𝑝} is finitely satisfiable for all 𝑝 ∈ 𝑆𝑛 (𝐴0 ). And 𝑆(𝑓) is con-
tinuous since [𝜑(𝑥1 , … , 𝑥𝑛 , 𝑓(𝑎))] is the preimage of [𝜑(𝑥1 , … , 𝑥𝑛 , 𝑎)] under
𝑆(𝑓)

36
 There are two main cases

1. An elementary bijection 𝑓 ∶ 𝐴0 → 𝐵0 defines a homeomorphism

𝑆𝑛 (𝐴0 ) → 𝑆𝑛 (𝐵0 ). We write 𝑓(𝑝) for the image of 𝑝

2. If 𝔄 = 𝔅 and 𝐴0 ⊆ 𝐵0 , the inclusion map induces the restriction

𝑆𝑛 (𝐵0 ) → 𝑆𝑛 (𝐴0 ). We write 𝑞↾𝐴0 for the restriction of 𝑞 to 𝐴0 . We
call 𝑞 an extension of 𝑞↾𝐴0 )

Lemma 4.9. A type 𝑝 is isolated in 𝑇 iff 𝑝 is an isolated point in 𝑆𝑛 (𝑇 ). In fact,

𝜑 isolates 𝑝 iff [𝜑] = {𝑝}. That is, [𝜑] is an atom in the Boolean algebra of clopen
subsets of 𝑆𝑛 (𝑇 )

Proof. 𝑝 being an isolated point means that {𝑝} is open, that is, {𝑝} = [𝜑].
The set [𝜑] is a singleton iff [𝜑] is non-empty and cannot be divided into
two non-empty clopen subsets [𝜑∧𝜓] and [𝜑 ∧¬𝜓]. This means that for all 𝜓
either 𝜓 or ¬𝜓 follows from 𝜑 modulo 𝑇 . So [𝜑] is a singleton iff 𝜑 generates
the type
⟨𝜑⟩ = {𝜓(𝑥) ∣ 𝑇 ⊨ ∀𝑥(𝜑(𝑥) → 𝜓(𝑥))}
which is the only element of [𝜑]
This shows that [𝜑] = {𝑝} implies that 𝜑 isolates 𝑝.
Conversely, 𝜑 isolates 𝑝, this means that ⟨𝜑⟩ is consistent with 𝑇 and
contains 𝑝. Since 𝑝 is a type, we have 𝑝 = ⟨𝜑⟩

We call a formula 𝜑(𝑥) complete if

{𝜓(𝑥) ∣ 𝑇 ⊨ ∀𝑥(𝜑(𝑥) → 𝜓(𝑥))}

is a type.

Corollary 4.10. A formula isolates a type iff it is complete

Exercise 4.2.1. 1. Closed subsets of 𝑆𝑛 (𝑇 ) have the form {𝑝 ∈ 𝑆𝑛 (𝑇 ) ∣

Σ ⊆ 𝑝}, where Σ is any set of formulas

2. Let 𝑇 be countable and consistent. Then any meagre1 subset 𝑋 of

𝑆𝑛 (𝑇 ) can be omitted, i.e., there is a model which omits all 𝑝 ∈ 𝑋

Proof. 1. The sets [𝜑] are a basis for the closed subsets of 𝑆𝑛 (𝑇 ). So the
closed sets of 𝑆𝑛 (𝑇 ) are exactly the intersections ⋂𝜑∈Σ [𝜑] = {𝑝 ∈
𝑆𝑛 (𝑇 ) ∣ Σ ⊆ 𝑝}
1
A subset of a topological space is nowhere dense if its closure has no interior. A count-
able union of nowhere dense sets is meagre

37
 2. The set 𝑋 is the union of a sequence of countable nowhere dense sets
𝑋𝑖 . We may assume that 𝑋𝑖 are closed, i.e., of the form {𝑝 ∈ 𝑆𝑛 (𝑇 ) ∣
Σ𝑖 ⊆ 𝑝}. That 𝑋𝑖 has no interior means that Σ𝑖 is not isolated. The
claim follows now from Corollary 4.3

Exercise 4.2.2. Consider the space 𝑆𝜔 (𝑇 ) of all complete types in variables

𝑣0 , 𝑣1 , …. Note that 𝑆𝜔 (𝑇 ) is again a compact space and therefore not meagre
by Baire’s theorem

1. Show that {tp(𝑎0 , 𝑎1 , … ) ∣ the 𝑎𝑖 enumerate a model of 𝑇 } is comea-

gre in 𝑆𝜔 (𝑇 )

Exercise 4.2.3. Let 𝐵 be a subset of 𝔄. Show that the restriction (restriction

of variables) map 𝑆𝑚+𝑛 (𝐵) → 𝑆𝑛 (𝐵) is open, continuous and surjective.
Let 𝑎 be an 𝑛-tuple in 𝐴. Show that the fibre over tp(𝑎/𝐵) is canonically
homeomorphic to 𝑆𝑚 (𝑎𝐵).
Consider the restriction map 𝜋 ∶ 𝑆𝑚+1 (𝐵) → 𝑆1 (𝐵). Then 𝜋−1 (tp(𝑎/𝐵)) ≅
𝑆𝑚 (𝑎𝐵)

Proof. We define the restriction map 𝑓 ∶ 𝑆𝑚+𝑛 (𝐵) → 𝑆𝑛 (𝐵) as: for 𝑞(𝑥, 𝑦) ∈
𝑆𝑚+𝑛 (𝐵), we let 𝑓(𝑞(𝑥, 𝑦)) = {𝜑(𝑦) ∶ 𝜑(𝑦) ∈ 𝑞(𝑥, 𝑦)}, where 𝑥 and 𝑦 are of
size 𝑚 and 𝑛 respectively.
continuous is easy
Now given an open set [𝜙(𝑣, 𝑤)] ⊆ 𝑆𝑚+𝑛 (𝐵). We need to prove 𝑓([𝜙(𝑣, 𝑤)]) =
[∃𝑣𝜙(𝑣, 𝑤)] which is clear
[𝜙(𝑥, 𝑦)] ∈ 𝜋−1 (tp(𝑎/𝐵)) iff 𝔐 ⊨ ∃𝑥𝜙(𝑥, 𝑎). Thus define 𝑔 ∶ [𝜙(𝑥, 𝑦)] ↦
[𝜙(𝑥, 𝑎)]. If [𝜙(𝑥, 𝑎)] = [𝜓(𝑥, 𝑎)], then ⊨ 𝜙(𝑥, 𝑎) ↔ 𝜓(𝑥, 𝑎).

Exercise 4.2.4. A theory 𝑇 has quantifier elimination iff every type is implied
by its quantifier-free part
Exercise 4.2.5. Consider the structure 𝔐 = (ℚ, <). Determine all types in
𝑆1 (ℚ). Which of these types are realised in ℝ? Which extensions does a
type over ℚ have to a type over ℝ?

Proof.

4.3 ℵ0 -categorical theories

Theorem 4.11 (Ryll-Nardzewski). Let 𝑇 be a countable complete theory. Then 𝑇
is ℵ0 -categorical iff for every 𝑛 there are only finitely many formulas 𝜑(𝑥1 , … , 𝑥𝑛 )
up to equivalence relative to 𝑇

38
Definition 4.12. An 𝐿-structure 𝔄 is 𝜔-saturated if all types over finite sub-
sets of 𝐴 are realised in 𝔄

The types in the definition are meant to be 1-types. On the other hand, it
is not hard to see that an 𝜔-saturated structure realises all 𝑛-types over finite
sets (Exercise 4.3.3) for all 𝑛 ≥ 1. The following lemma is a generalisation
of the ℵ0 -categoricity of DLO.

Lemma 4.13. Two elementarily equivalent, countable and 𝜔-saturated structures

are isomorphic

Proof. Suppose 𝔄 and 𝔅 are as in the lemma. We choose enumerations 𝐴 =

{𝑎0 , 𝑎1 , … } and 𝐵 = {𝑏0 , 𝑏1 , … }. Then we construct an ascending sequence
𝑓0 ⊆ 𝑓1 ⊆ ⋯ of finite elementary maps

𝑓𝑖 ∶ 𝐴 𝑖 → 𝐵 𝑖

between finite subsets of 𝔄 and 𝔅. We will choose the 𝑓𝑖 in such a way that
𝐴 is the union of the 𝐴𝑖 and 𝐵 the union of the 𝐵𝑖 . The union of the 𝑓𝑖 is
then the desired isomorphism between 𝔄 and 𝔅
The empty map 𝑓0 = ∅ is elementary since 𝔄 and 𝔅 are elementarily
equivalent. Assume that 𝑓𝑖 is already constructed. There are two cases:
𝑖 = 2𝑛; We will extend 𝑓𝑖 to 𝐴𝑖+1 = 𝐴𝑖 ∪ {𝑎𝑛 }. Consider the type

𝑝(𝑥) = tp(𝑎𝑛 /𝐴𝑖 ) = {𝜑(𝑥) ∣ 𝔄 ⊨ 𝜑(𝑎𝑛 ), 𝜑(𝑥) a 𝐿(𝐴𝑖 )-formula}

Since 𝑓𝑖 is elemantarily, 𝑓𝑖 (𝑝)(𝑥) is in 𝔅 a type over 𝐵𝑖 . (note that 𝑓𝑖 is el-

ementary iff 𝔄𝐴𝑖 ≡ 𝔅𝐵𝑖 ) Since 𝔅 is 𝜔-saturated, there is a realisation 𝑏′ of
this type. So for 𝑎 ∈ 𝐴𝑖

𝔄 ⊨ 𝜑(𝑎𝑛 , 𝑎) ⇒ 𝔅 ⊨ 𝜑(𝑏′ , 𝑓𝑖 (𝑎))

Given 𝑏′ , then the type that it realises is fixed. Hence

𝔅 ⊨ 𝜑(𝑏′ , 𝑓𝑖 (𝑎)) ⇒ 𝔄 ⊨ 𝜑(𝑎𝑛 , 𝑎)

This shows that 𝑓𝑖+1 (𝑎𝑛 ) = 𝑏′ defines an elementary extension of 𝑓𝑖

𝑖 = 2𝑛 + 1; we exchange 𝔄 and 𝔅

Proof of Theorem 4.11. Assume that there are only finitely many 𝜑(𝑥1 , … , 𝑥𝑛 )
relative to 𝑇 for every 𝑛. By Lemma 4.13 it suffices to show that all models of
𝑇 are 𝜔-saturated. Let 𝔐 be a model of 𝑇 and 𝐴 an 𝑛-element subset. If there
are only 𝑁 many formulas, up to equivalence, in the variable 𝑥1 , … , 𝑥𝑛+1 ,

39
there are, up to equivalence in 𝔐, at most 𝑁 many 𝐿(𝐴)-formulas 𝜑(𝑥).
Thus, each type 𝜑(𝑥) ∈ 𝑆(𝐴) is isolated (w.r.t. Th(𝔐𝐴 )) by a smallest for-
mula 𝜑𝑝 (𝑥) (⋀ 𝑝(𝑥)). Each element of 𝑀 which realises 𝜑𝑝 (𝑥) also realises
𝑝(𝑥), so 𝔐 is 𝜔-saturated.
Conversely, if there are infinitely many 𝜑(𝑥1 , … , 𝑥𝑛 ) modulo 𝑇 for some
𝑛, then - as the type space 𝑆𝑛 (𝑇 ) is compact - there must be some non-
isolated type 𝑝 (if 𝑝 is isolated, then {𝑝} is open). Then by Lemma 4.9 𝑝 is
not isolated in 𝑇 . By the Omitting Types Theorem there is a countable model
of 𝑇 in which this type is not realised. On the other hand, there also exists
a countable model of 𝑇 realizing this type. So 𝑇 is not ℵ0 -categorical

The proof shows that a countable complete theory with infinite models
is ℵ0 -categorical iff all countable models are 𝜔-saturated
given a variables 𝜑𝑖 (𝑎𝑖 ) where 𝑎𝑖 ∈ 𝐴, we can consider ⋀ ∃𝑥𝑖 𝜑𝑖 (𝑥𝑖 ).
Definition 4.14. An 𝐿-structure 𝔐 is 𝜔-homogeneous if for every elemen-
tary map 𝑓0 defined on a finite subset 𝐴 of 𝑀 and for any 𝑎 ∈ 𝑀 there is
some element 𝑏 ∈ 𝑀 s.t.
𝑓 = 𝑓0 ∪ {⟨𝑎, 𝑏⟩}
is elementary
𝑓 = 𝑓0 ∪ {⟨𝑎, 𝑏⟩} is elementary iff 𝑏 realises 𝑓0 (tp(𝑎/𝐴))
Corollary 4.15. Let 𝔄 be a structure and 𝑎1 , … , 𝑎𝑛 elements of 𝔄. Then Th(𝔄) is
ℵ0 -categorical iff Th(𝔄, 𝑎1 , … , 𝑎𝑛 ) is ℵ0 -categorical
Proof. If Th(𝔄) is ℵ0 -categorical, then for any 𝑚 + 𝑛 there is only finitely
many formulas 𝜑(𝑥1 , … , 𝑥𝑚+𝑛 ) up to equivalence relative to Th(𝔄), hence
there is only finitely many 𝜑(𝑥1 , … , 𝑥𝑚 , 𝑎1 , … , 𝑎𝑛 ) up to equivalence relative
to Th(𝔄, 𝑎1 , … , 𝑎𝑛 )
For the converse, Th(𝔄) ⊂ Th(𝔄, 𝑎1 , … , 𝑎𝑛 )

Example 4.1. The following theories and ℵ0 -categorical

1. Infset (saturated)

2. For every finite field 𝔽𝑞 , the theory of infinite 𝔽𝑞 -vector spaces. (Vector
spaces over the same field and of the same dimension are isomorphic)

3. The theory DLO of dense linear orders without endpoints. This fol-
lows from Theorem 4.11 since DLO has quantifier elimination: for
every 𝑛 there are only finitely many (say 𝑁𝑛 ) ways to order 𝑛 ele-
ments. Each of these possibility corresponds to a complete formula

40
 𝜓(𝑥1 , … , 𝑥𝑛 ). Hence there are up to equivalence, exactly 2𝑁𝑛 many
formulas 𝜑(𝑥1 , … , 𝑥𝑛 )

Definition 4.16. A theory 𝑇 is small if 𝑆𝑛 (𝑇 ) are at most countable for all

𝑛<𝜔

Lemma 4.17. A countable complete theory is small iff it has a countable 𝜔-saturated
model

Proof. If 𝑇 has a finite model 𝔄, 𝑇 is small and 𝔄 is 𝜔-saturated: since 𝑇 is

complete, for any type 𝑝(𝑥) ∈ 𝑆𝑛 (𝑇 ), 𝑇 ⊨ 𝑝(𝑥). For finite model 𝔄, there are
only finitely many assignments. If we have two distinct types 𝑝(𝑥), 𝑞(𝑥) ∈
𝑆𝑛 (𝑇 ), then there is 𝜙(𝑥) ∈ 𝑝(𝑥) and 𝜙(𝑥) ∉ 𝑞(𝑥). Since they are maximally
consistent, 𝑞(𝑥) ⊨ ¬𝜙(𝑥) hence 𝑝(𝑥) and 𝑞(𝑥) cannot be realised by the same
element. So we may assume that 𝑇 has infinite models
If all types can be realised in a single countable model, there can be at
most countably many types.
if conversely all 𝑆𝑛+1 (𝑇 ) are at most countable, then over any 𝑛-element
subset of a model of 𝑇 there are at most countably many types. We construct
an elementary chain
𝔄0 ≺ 𝔄1 ≺ …
of models of 𝑇 . For 𝔄0 we take any countable model. if 𝔄𝑖 is already con-
structed, we use Corollary 2.14 and Theorem 2.15 to construct a countable
model 𝔄𝑖+1 in such a way that all types over finite subsets of 𝐴𝑖 are realised
in 𝔄𝑖+1 . This can be done since there are only countable many such types.
The union 𝔄 = ⋃𝑖∈𝜔 𝔄𝑖 is countable and 𝜔-saturated since every type over
a finite subset 𝐵 of 𝔄 is realised in 𝔄𝑖+1 if 𝐵 ⊆ 𝐴𝑖

Theorem 4.18 (Vaught). A countable complete theory cannot have exactly two
countably models

Proof. We can assume that 𝑇 is small and not ℵ0 -categorical (if 𝑇 is not
small, then it has no countable model). We will show that 𝑇 has at least
three non-isomorphic countable models. First, 𝑇 has an 𝜔-saturated count-
able model 𝔄 and there is a non-isolated type 𝑝(𝑥) which can be omitted in
a countable model 𝔅. Let 𝑝(𝑥) be realised in 𝔄 by 𝑎. Since Th(𝔄, 𝑎) is not
ℵ0 -categorical as 𝑇 ⊂ Th(𝔄, 𝑎), Th(𝔄, 𝑎) has a countable model (ℭ, 𝑐) which
is not 𝜔-saturated. Then ℭ is not 𝜔-saturated and therefore not isomorphic
to 𝔄. But ℭ realises 𝑝(𝑥) and is therefore not isomorphic to 𝔅

Exercise 4.3.1. Show that 𝑇 is ℵ0 -categorical iff 𝑆𝑛 (𝑇 ) is finite for all 𝑛

41
Exercise 4.3.2. Show that for every 𝑛 > 2 there is a countable complete theory
with exactly 𝑛 countable models

Proof. StackExchange

Exercise 4.3.3. If 𝔄 is 𝜔-saturated, all 𝑛-types over finite sets are realised.

Proof. Assume that 𝔄 is 𝜅-saturated, 𝐵 a subset of 𝐴 of smaller cardinality

than 𝜅 and 𝑝(𝑥, 𝑦) a (𝑛 + 1)-type over 𝐵. Let 𝑏 ∈ 𝐴 be a realisation of 𝑞(𝑦) =
𝑝↾𝑦 and 𝑎 ∈ 𝐴 a realisation of 𝑝(𝑥, 𝑏). Then (𝑎, 𝑏) realises 𝑝.

4.4 The amalgamation method

Definition 4.19. For any language 𝐿, the skeleton 𝒦 of an 𝐿-structure 𝔐
is the class of all finitely-generated 𝐿-structures which are isomorphic to
a substructure of 𝔐. We say that an 𝐿-structure 𝔐 is 𝒦-saturated if its
skeleton is 𝒦 and if for all 𝔄, 𝔅 in 𝒦 and all embeddings 𝑓0 ∶ 𝔄 → 𝔐 and
𝑓1 ∶ 𝔄 → 𝔅 there is an embedding 𝑔1 ∶ 𝔅 → 𝔐 with 𝑓0 = 𝑔1 ∘ 𝑓1

𝑓0
𝔄 𝔐

𝑓1 𝑔

Theorem 4.20. Let 𝐿 be a countable language. Any two countable 𝒦-saturated

structures are isomorphic
Proof. Let 𝔐 and 𝔑 be countable 𝐿-structures with the same skeleton 𝒦,
and assume that 𝔐 and 𝔑 are 𝒦-saturated. As in the proof of Lemma 4.13
we construct an isomorphisms between 𝔐 and 𝔑 as the union of an ascend-
ing sequence of isomorphisms between finitely-generated substructures of
𝑀 and 𝑁 .
If 𝑓1 ∶ 𝔄 → 𝔑 is an embedding of a finitely-generated substructure of 𝔄
of 𝔐 into 𝔑, and 𝑎 is an element of 𝔐, then by 𝒦-saturation 𝑓1 can be ex-
tended to an embedding 𝑔1 ∶ 𝔄′ → 𝔑 where 𝔄′ = ⟨𝐴𝑎⟩𝔐 . Now interchange
the roles of 𝔐 and 𝔑.

The proof shows that any countable 𝒦-saturated structure 𝔐 is ultraho-

mogeneous i.e., any isomorphism between finitely generated substructure
extends to an automorphism of 𝔐.
Theorem 4.21. Let 𝐿 be a countable language and 𝒦 a countable class of finitely-
generated 𝐿-structures. There is a countable 𝒦-saturated 𝐿-structure 𝔐 iff

42
 1. (Heredity) if 𝔄0 ∈ 𝒦, then all elements of the skeleton of 𝔄0 also belongs to
𝒦

2. (Joint Embedding) for 𝔅0 , 𝔅1 ∈ 𝒦 there are some 𝒟 ∈ 𝒦 and embeddings

𝑔𝑖 ∶ 𝔅 𝑖 → 𝔇

3. (Amalgamation) if 𝔄, 𝔅0 , 𝔅1 ∈ 𝒦 and 𝑓𝑖 ∶ 𝔄 → 𝔅𝑖 , (𝑖 = 0, 1) are

embeddings, there is some 𝒟 ∈ 𝒦 and two embeddings 𝑔𝑖 ∶ 𝔅𝑖 → 𝔇 s.t.
𝑔0 ∘ 𝑓 0 = 𝑔 1 ∘ 𝑓 1

𝔇
𝑔0 𝑔1

𝔅0 𝔅1

𝑓0 𝑓1
𝔄

in this case, 𝔐 is unique up to isomorphism and is called the Fraïssé limit of

𝒦

Proof. Let 𝒦 be the skeleton of a countably 𝒦-saturated structure 𝔐. Clearly,

𝒦 has the hereditary property (substructure of a substructure is still a sub-
structure). To see that 𝒦 has the Amalgamation Property, let 𝔄, 𝔅0 , 𝔅1 , 𝑓0
and 𝑓1 be as in 3. We may assume that 𝔅0 ⊆ 𝔐 and 𝑓0 is the inclusion
map. Furthermore we can assume 𝔄 ⊆ 𝔅1 and that 𝑓1 is the inclusion map.
Now the embedding 𝑔1 ∶ 𝔅1 → 𝔐 is the extension of the isomorphism
𝑓0 ∶ 𝔄 → 𝑓0 (𝔄) to 𝔅1 and satisfies 𝑓0 = 𝑔1 ∘ 𝑓1 . For 𝔇 we choose a finitely-
generated substructure of 𝔐 which contains 𝔅0 and the image of 𝑔1 . For
𝑔0 ∶ 𝔅0 → 𝔇 take the inclusion map. For Joint Embedding Property take
⟨𝐵0 𝐵1 ⟩𝔐
For the converse assume that 𝒦 has properties 1, 2 and 3. Choose an
enumeration (𝔅𝑖 )𝑖∈𝜔 of all isomorphism types in 𝒦 (they are not isomor-
phic). We construct 𝔐 as the union of an ascending chain

𝔐0 ⊆ 𝔐 1 ⊆ ⋯ ⊆ 𝔐

of elements of 𝒦. Suppose that 𝔐𝑖 is already constructed. If 𝑖 = 2𝑛, we

choose 𝔐𝑖+1 as the top of a diagram

43
 𝔐𝑖+1
𝑔0 𝑔1

𝔐𝑖 𝔅𝑛

where we can assume that 𝑔0 is the inclusion map. if 𝑖 = 2𝑛 + 1, let 𝔄 and

𝔅 from 𝒦 and two embeddings 𝑓0 ∶ 𝔄 → 𝔐𝑖 and 𝑓1 ∶ 𝔄 → 𝔅 be given.

𝔐𝑖+1
𝑔0 𝑔1

𝔐𝑖 𝔅

𝑓0 𝑓1
𝔄
To ensure that 𝔐 is 𝒦-saturated we have in the odd steps to make the right
choice of 𝔄, 𝔅, 𝑓0 and 𝑓1 . Assume that we have 𝔄, 𝔅 ∈ 𝒦 and embeddings
𝑓0 ∶ 𝔄 → 𝔐 and 𝑓1 ∶ 𝔄 → 𝔅. For large 𝑗 the image of 𝑓0 will be contained
in 𝔐𝑗 . During the construction of the 𝔐𝑖 , in order to guarantee the 𝒦-
saturation of 𝔐, we have to ensure that eventually, for some odd 𝑖 ≥ 𝑗, the
embeddings 𝑓0 ∶ 𝔄 → 𝔐𝑖 and 𝑓1 ∶ 𝔄 → 𝔅 were used in the construction
of 𝔐𝑖+1 . This can be done since for each 𝑗 there are - up to isomorphism -
at most countably many possibilities. Thus there exists an embedding 𝑔1 ∶
𝔅 → 𝔐𝑖+1 with 𝑓0 = 𝑔1 ∘ 𝑓1 .
𝒦 is the skeleton of 𝔐: the finitely-generated substructure are the sub-
structures of the 𝔐1 . Since 𝔐𝑖 ∈ 𝒦, their finitely-generated substructure
also belong to 𝒦. On the other hand each 𝐵𝑛 is isomorphic to a substructure
of 𝔐2𝑛+1
Uniqueness follows from Theorem 4.20

For finite relational languages 𝐿, any non-empty finite subset is itself

a (finitely-generated) substructure. For such languages, the construction
yields ℵ0 -categorical structures. We now take a look at ℵ0 -categorical theo-
ries with quantifier elimination in a finite relational language
Remark. A complete theory 𝑇 in a finite relational language with quantifier
elimination is ℵ0 -categorical. So all its models are 𝜔-homogeneous

Proof. For every 𝑛 there is only a finite number of non-equivalent quanti-

fier free formulas 𝜌(𝑥1 , … , 𝑥𝑛 ). If 𝑇 has quantifier elimination, this number
is also the number of all formulas 𝜑(𝑥1 , … , 𝑥𝑛 ) modulo 𝑇 and so 𝑇 is ℵ0 -
categorical by Theorem 4.11

44
Lemma 4.22. Let 𝑇 be a complete theory in a finite relational language and 𝔐 an
infinite model of 𝑇 . TFAE

1. 𝑇 has quantifier elimination

2. Any isomorphism between finite substructures is elementary

3. the domain of any isomorphism between finite substructures can be extended

to any further element

Proof. 2 → 1. if any isomorphism between finite substructure of 𝔐 is ele-

mentary, all 𝑛-tuples 𝑎 which satisfy in 𝔐 the same quantifier-free type

tpqf (𝑎) = {𝜌(𝑥) ∣ 𝔐 ⊨ 𝜌(𝑎), 𝜌(𝑥) quantifier-free}

satisfy the same simple existential formulas. We will show from this that
every simple existential formula 𝜑(𝑥1 , … , 𝑥𝑛 ) = ∃𝑦𝜌(𝑥1 , … , 𝑥𝑛 , 𝑦) is, mod-
ulo 𝑇 , equivalent to a quantifier-free formula. Let 𝑟1 (𝑥), … , 𝑟𝑘−1 (𝑥) be the
quantifier-free types of all 𝑛-tuples in 𝔐 which satisfy 𝜑(𝑥). Let 𝜌𝑖 (𝑥) be
equivalent to the conjunction of all formulas from 𝑟𝑖 (𝑥). Then

𝑇 ⊨ ∀𝑥(𝜑(𝑥) ↔ ⋁ 𝜌𝑖 (𝑥))
𝑖<𝑘

1 → 3 the theory 𝑇 is ℵ0 -categorical and hence all models are 𝜔-homogeneous.

Since any isomorphism between finite substructures is elementary, 3 fol-
lows.
3 → 2. If the domain of any finite isomorphism can be extended to any
further element, it is easy to see that every finite isomorphism is elementary.
Here we can only consider ∃𝑥𝜑(𝑥).

Theorem 4.23. Let 𝐿 be a finite relational language and 𝒦 a class of finite 𝐿-

structures. If the Fraïssé limit of 𝒦 exists, its theory is ℵ0 -categorical and has
quantifier elimination

4.5 Prime Models

Suppose 𝑇 is countable and has infinite models, 𝔐 ⊨ 𝑇

45
 𝔐 is prime isolated types are dense

𝔐 is atomic,countable prime models are isomorphic

𝔐 is 𝜔-homogeneous

Let 𝑇 be a countable complete theory with infinite models

Definition 4.24. Let 𝑇 be a countable theory with infinite models, not nec-
essarily complete
1. We call 𝔄0 a prime model of 𝑇 if 𝔄0 can be elementarily embedded
into all models of 𝑇
2. A structure 𝔄 is called atomic if all 𝑛-tuples 𝑎 of elements of 𝔄 are
atomic. This means that the types tp(𝑎) are isolated in 𝑆𝑛𝔄 (∅) = 𝑆𝑛 (𝑇 )
Prime models need not exists. By Corollary 4.10, a tuple 𝑎 is atomic iff
it satisfies a complete formula.
Since 𝑇 has countable models, prime models must be countable and
since non-isolated types can be omitted in suitable models by Theorem 4.2,
only isolated types can be realised in prime models.
Theorem 4.25. A model of 𝑇 is prime iff it is countable and atomic
Proof. As just noted, a prime model has to be countable and atomic.
Let 𝔐0 be a countable and atomic model of 𝑇 and 𝔐 any model of 𝑇 . We
construct an elementary embedding of 𝔐0 to 𝔐 as a union of an ascending
sequence of elementary maps

𝑓 ∶𝐴→𝐵

between finite subsets 𝐴 of 𝑀0 and 𝐵 of 𝑀 . The empty map is elementary

since 𝑇 is complete and 𝔐0 ≡ 𝔐
We show that 𝑓 can be extended to any given 𝐴∪{𝑎}. Let 𝑝(𝑥) = tp(𝑎/𝐴)
and 𝑓(𝑝) = 𝑓(𝑝(𝑥)). We show that 𝑓(𝑝) has a realisation 𝑏 ∈ 𝑀
Let 𝑎 be a tuple which enumerates the elements of 𝐴 and 𝜑(𝑥, 𝑥) an
𝐿-formula which isolates the tp(𝑎𝑎) since 𝔐0 is atomic. Then 𝑝(𝑥) is iso-
lated by 𝜑(𝑥, 𝑎): clearly 𝜑(𝑥, 𝑎) ∈ tp(𝑎/𝑎) and if 𝜌(𝑥, 𝑎) ∈ tp(𝑎/𝑎) we have

46
𝜌(𝑥, 𝑦) ∈ tp(𝑎, 𝑎). This implies that 𝔐0 ⊨ ∀𝑥(𝜑(𝑥, 𝑎) → 𝜌(𝑥, 𝑎)) and 𝔐 ⊨
∀𝑥(𝜑(𝑥, 𝑓(𝑎)) → 𝜌(𝑥, 𝑓(𝑎))). Thus 𝑓(𝑝) is isolated by 𝜑(𝑥, 𝑓(𝑎)) and since
𝜑(𝑥, 𝑓(𝑎)) can be realised in 𝔐, so can be 𝑓(𝑝). Now we prove 𝑓(𝑝) is in-
deed a type. If there are 𝜑(𝑥, 𝑥) ∈ tp(𝑏𝑏) ⧵ 𝑓(𝑝). Then 𝔐0 ⊭ 𝜑(𝑎, 𝑎) and thus
¬𝜑(𝑥, 𝑥) ∈ 𝑓(𝑝) ⊆ tp(𝑏𝑏), a contradiction.

Theorem 4.26. All prime models of 𝑇 are isomorphic

Proof. Let 𝔐0 and 𝔐0′ be two prime models. Since prime models are atomic,
elementary maps between finite subsets of 𝔐0 and 𝔐0′ can be extended to
all finite extensions. Since 𝔐0 and 𝔐0′ are countable, it follows as Lemma
4.13 that 𝔐0 ≅ 𝔐0′ ;”

Corollary 4.27. Prime models are 𝜔-homogeneous

Proof. Let 𝔐0 be prime and 𝑎 any tuple of elements from 𝑀0 . By Theorem
4.25, (𝔐0 , 𝑎) is a prime model of its theory as it’s still countable and atomic
. The claim follows now from Theorem 4.26

Definition 4.28. The isolated types are dense in 𝑇 if every consistent 𝐿-

formulas 𝜓(𝑥1 , … , 𝑥𝑛 ) belongs to an isolated type 𝑝(𝑥1 , … , 𝑥𝑛 ) ∈ 𝑆𝑛 (𝑇 )
Theorem 4.29. 𝑇 has a prime model iff the isolated types are dense
Proof. Suppose 𝑇 has a prime model 𝔐 (so 𝔐 is atomic by Theorem 4.25).
Since consistent formulas 𝜓(𝑥) are realised in all models of 𝑇 , 𝜓(𝑥) is re-
alised by an atomic tuple 𝑎 and 𝜓(𝑥) belongs to the isolated type tp(𝑎)
For the other direction notice that a structure 𝔐0 is atomic iff for all 𝑛
the set

Σ𝑛 (𝑥1 , … , 𝑥𝑛 ) = {¬𝜑(𝑥1 , … , 𝑥𝑛 ) ∣ 𝜑(𝑥1 , … , 𝑥𝑛 ) complete}

is not realised in 𝔐0 . 𝑎 is atomic iff it realise at least one complete formula.

Hence by Corollary 4.3, it’s enough to show that the Σ𝑛 are not isolated in
𝑇 . This is the case iff for every consistent 𝜓(𝑥1 , … , 𝑥𝑛 ) there is a complete
formula 𝜑(𝑥1 , … , 𝑥𝑛 ) with 𝑇 ⊭ ∀𝑥(𝜓(𝑥) → ¬𝜑(𝑥)). Σ𝑛 is not isolated iff for
every complete formula 𝜃 there exists 𝛾 ∈ Σ𝑛 s.t. 𝑇 ⊭ 𝜃 → 𝛾. 𝛾 here is of
the form ¬𝜑 and we loose the condition of 𝜃. Since 𝜑(𝑥) is complete, this is
equivalent to 𝑇 ⊨ ∀𝑥(𝜑(𝑥) → 𝜓(𝑥)). We conclude that Σ𝑛 is not isolated iff
the isolated 𝑛-types are dense

Example 4.2. Let 𝑇 be the language having a unary predicate 𝑃𝑠 for every
finite 0-1-sequence 𝑠 ∈ 2<𝜔 . The axioms of Tree say that the 𝑃𝑠 , 𝑠 ∈ 2<𝜔 ,
form a binary decomposition of the universe

47
 • ∀𝑥 𝑃∅ (𝑥)

• ∃𝑥 𝑃𝑠 (𝑥)

• ∀𝑥 ((𝑃𝑠0 (𝑥) ∨ 𝑃𝑠1 (𝑥)) ↔ 𝑃𝑠 (𝑥))

• ∀𝑥 ¬(𝑃𝑠0 (𝑥) ∧ 𝑃𝑠1 (𝑥))

Tree is complete and has quantifier elimination. There are no complete

formulas and no prime model

See Marker to see the full content

Definition 4.30. A family of formulas 𝜑𝑠 (𝑥), 𝑠 ∈ 2<𝜔 is a binary tree if for

all 𝑠 ∈ 2<𝜔 the following holds

1. 𝑇 ⊨ ∀𝑥 ((𝜑𝑠0 (𝑥) ∨ 𝜑𝑠1 (𝑥)) → 𝜑𝑠 (𝑥))

2. 𝑇 ⊨ ∀𝑥 ¬(𝜑𝑠0 (𝑥) ∧ 𝜑𝑠1 (𝑥))

Theorem 4.31. Let 𝑇 be a complete theory

1. If 𝑇 is small, it has no binary tree of consistent 𝐿-formulas. If 𝑇 is countable,

the converse holds as well

2. If 𝑇 has no binary tree of consistent 𝐿-formulas, the isolated types are dense

Proof. 1. Let (𝜑𝑠 (𝑥1 , … , 𝑥𝑛 )) be a binary tree of consistent formulas. Then,

for all 𝜂 ∈ 2𝜔 , the set
{𝜑𝑠 (𝑥) ∣ 𝑠 ⊆ 𝜂}
is consistent and therefore is contained in some type 𝑝𝜂 (𝑥) ∈ 𝑆𝑛 (𝑇 ).
The 𝑝𝜂 (𝑥) are all different showing that 𝑇 is not small.

Exercise 4.5.1. Countable theories without a binary tree of consistent formu-

las are small

Proof. If countable theory 𝑇 is not small.

Exercise 4.5.2. Show that isolated types being dense is equivalent to isolated
types being (topologically) dense in the Stone space 𝑆𝑛 (𝑇 ).

Proof. Let 𝑆 = {the isolated types in 𝑆𝑛 (𝑇 )}. 𝑆 is dense in 𝑆𝑛 (𝑇 ) iff 𝑆 =

𝑆𝑛 (𝑇 ). For any 𝑝 ∈ 𝑆𝑛 (𝑇 ) ⧵ 𝑆, 𝑝 is non-isolated. For any 𝑝 ∈ [𝜙], 𝜙 belongs
to an isolated type 𝑞. Thus 𝑞 ∈ 𝑆𝑛 (𝑇 ) ∩ 𝑆. Hence 𝑆 = 𝑆𝑛 (𝑇 ).

48
5 ℵ1 -categorical Theories
5.1 Indiscernibles
Definition 5.1. Let 𝐼 be a linear order and 𝔄 an 𝐿-structure. A family (𝑎𝑖 )𝑖∈𝐼
of elements of 𝐴 is called a sequence of indiscernibles if for all 𝐿-formulas
𝜑(𝑥1 , … , 𝑥𝑛 ) and all 𝑖1 < ⋯ < 𝑖𝑛 and 𝑗1 < ⋯ < 𝑗𝑛 from 𝐼

𝔄 ⊨ 𝜑(𝑎𝑖1 , … , 𝑎𝑖𝑛 ) ↔ 𝜑(𝑎𝑗1 , … , 𝑎𝑗𝑛 )

if two of the 𝑎𝑖 are equal, all 𝑎𝑖 are the same. Therefore it is often as-
sumed that the 𝑎𝑖 are distinct
Sometimes sequences of indiscernibles are also called order indiscernible
to distinguish them from totally indiscernible sequences in which the or-
dering of the index set does not matter.

Definition 5.2. Let 𝐼 be an infinite linear order and ℐ = (𝑎𝑖 )𝑖∈𝐼 a sequence
of 𝑘-tuples in 𝔐, 𝐴 ⊆ 𝑀 . The Ehrenfeucht-Mostowski type EM(ℐ/𝐴) of ℐ
over 𝐴 is the set of 𝐿(𝐴)-formulas 𝜑(𝑥1 , … , 𝑥𝑛 ) with 𝔐 ⊨ 𝜑(𝑎𝑖1 , … , 𝑎𝑖𝑛 ) for
all 𝑖1 < ⋯ < 𝑖𝑛 ∈ 𝐼, 𝑛 < 𝜔

Lemma 5.3 (The Standard Lemma). Let 𝐼 and 𝐽 be two infinite linear orders and
ℐ = (𝑎𝑖 )𝑖∈𝐼 a sequence of elements of a structure 𝔐. Then there is structure 𝔑 ≡
𝔐 with an indiscernible sequences (𝑏𝑗 )𝑗∈𝐽 realizing the Ehrenfeucht-Mostowski
type EM(ℐ) of ℐ

We can also find a model 𝔑 ≻ 𝔐

Corollary 5.4. Assume that 𝑇 has an infinite model. Then for any linear order 𝐼,
𝑇 has a model with a sequence (𝑎𝑖 )𝑖∈𝐼 of distinct indiscernibles

Let [𝐴]𝑛 denote the set of all 𝑛-element subsets of 𝐴

Theorem 5.5 (Ramsey). Let 𝐴 be infinite and 𝑛 ∈ 𝜔. Partition the set of 𝑛-

elements subsets [𝐴]𝑛 into subsets 𝐶1 , … , 𝐶𝑘 . Then there is an infinite subset of 𝐴
whose 𝑛-element subsets all belong to the same subset 𝐶𝑖

Proof. Thinking of the partition as a colouring on [𝐴]𝑛 , we are looking for

an infinite subset 𝐵 of 𝐴 s.t. [𝐵]𝑛 is monochromatic. We prove the theorem
by induction on 𝑛. For 𝑛 = 1, the statement is evident from the pigeonhole
principle since there are infinite elements and finite colors.
Assuming the theorem is true for 𝑛, we now prove it for 𝑛+1. Let 𝑎0 ∈ 𝐴.
Then any colouring of [𝐴]𝑛+1 induces a colouring of the 𝑛-element subsets

49
of 𝐴′ = 𝐴 ⧵ {𝑎0 }: just colour 𝑥 ∈ [𝐴′ ]𝑛 by the colour of {𝑎0 } ∪ 𝑥 ∈ [𝐴]𝑛+1 .
By the induction hypothesis, there exists an infinite monochromatic subset
𝐵1 of 𝐴′ in the induced colouring. Thus, all the (𝑛 + 1)-element subsets of
𝐴 consisting of 𝑎0 and 𝑛 elements of 𝐵1 have the same colour but {𝑎0 } ∪ 𝐵
is not our desired set.
Now pick any 𝑎1 ∈ 𝐵1 . By the same argument we obtain an infinite
subset 𝐵2 ⊆ 𝐵1 with the same properties. We thus construct an infinite
sequence 𝐴 = 𝐵0 ⊃ 𝐵1 ⊃ 𝐵2 ⊃ … and elements 𝑎𝑖 ∈ 𝐵𝑖 ⧵𝐵𝑖+1 s.t. the colour
of each (𝑛 + 1)-element subset {𝑎𝑖(0) , … , 𝑎𝑖(𝑛) } with 𝑖(0) < 𝑖(1) < ⋯ < 𝑖(𝑛)
depends only on the value of 𝑖(0).

𝑎0 , 𝑎 1 , 𝑎 2 , … , 𝑎 𝑛 , …

Again by the pigeonhole principle there are infinitely many values of 𝑖(0)
for which this colour will be the same and we take {𝑎𝑖(0) }. These 𝑎𝑖(0) then
yields the desired monochromatic set.

Proof of Lemma ??. Choose a set 𝐶 of new constants with an ordering iso-
morphic to 𝐽 . Consider the theories

𝑇 ′ = {𝜑(𝑐) ∣ 𝜑(𝑥) ∈ EM(ℐ)}

𝑇 ″ = {𝜑(𝑐) ↔ 𝜑(𝑑) ∣ 𝑐, 𝑑 ∈ 𝐶}

Here the 𝜑(𝑥) are 𝐿-formulas and 𝑐, 𝑑 tuples in increasing order. We have
to show that 𝑇 ∪ 𝑇 ′ ∪ 𝑇 ″ is consistent. It is enough to show that

𝑇𝐶0 ,Δ = 𝑇 ∪ {𝜑(𝑐) ∈ 𝑇 ′ ∣ 𝑐 ∈ 𝐶0 } ∪ {𝜑(𝑐) ↔ 𝜑(𝑑) ∣ 𝜑(𝑥) ∈ Δ, 𝑐, 𝑑 ∈ 𝐶0 }

is consistent for finite sets 𝐶0 and Δ. Note that Diagel (𝔐) ⊆ 𝑇 .

We can assume that the elements of Δ are formulas with free variables
𝑥1 , … , 𝑥𝑛 and that all tuples 𝑐 and 𝑑 have the same length
for notational simplicity we assume that all 𝑎𝑖 are different. So we may
consider 𝐴 = {𝑎𝑖 ∣ 𝑖 ∈ 𝐼} as an ordered set, which is the interpretation of
𝐶. We define an equivalence relation on [𝐴]𝑛 by

𝑎 ∼ 𝑏 ⟺ 𝔐 ⊨ 𝜑(𝑎) ↔ 𝜑(𝑏) for all 𝜑(𝑥1 , … , 𝑥𝑛 ) ∈ Δ

where 𝑎, 𝑏 are tuples in increasing order. Since this equivalence has at most
2|Δ| many classes, by Ramsey’s Theorem there is an infinite subset 𝐵 ⊆ 𝐴
with all 𝑛-element subsets in the same equivalence class. We interpret the
constants 𝑐 ∈ 𝐶0 by elements 𝑏𝑐 in 𝐵 ordered in the same way as the 𝑐. Then
(𝔐, 𝑏𝑐 )𝑐∈𝐶0 is a model of 𝑇𝐶0 ,Δ .

50
Lemma 5.6. Assume 𝐿 is countable. If the 𝐿-structure 𝔐 is generated by a well-
ordered sequence (𝑎𝑖 ) of indiscernibles, then 𝔐 realises only countably many types
over every countable subset of 𝑀

Proof. : need more time to think

If 𝐴 = {𝑎𝑖 ∣ 𝑖 ∈ 𝐼}, then every element 𝑏 ∈ 𝑀 has the form 𝑏 = 𝑡(𝑎),
where 𝑡 is an 𝐿-term and 𝑎 is a tuple from 𝐴 since 𝔐 is generated by (𝑎𝑖 )
Consider a countable subset 𝑆 of 𝑀 . Write

𝑆 = {𝑡ℳ 𝑛
𝑛 (𝑎 ) ∣ 𝑛 ∈ 𝜔}

Let 𝐴0 = {𝑎𝑖 ∣ 𝑖 ∈ 𝐼0 } be the (countable) set of elements of 𝐴 which occur

in the 𝑎𝑛 . Then every type tp(𝑏/𝑆) is determined by tp(𝑏/𝐴0 ) since every
𝐿(𝑆)-formula
𝜑(𝑥, 𝑡ℳ 𝑛
𝑛1 (𝑎 ), … )

can be replaced by the 𝐿(𝐴0 )-formula 𝜑(𝑥, 𝑡𝑛1 (𝑎𝑛1 ), … )

tp(𝑏/𝐴0 ) = tp(𝑡(𝑎)/𝐴0 ) = {𝜑(𝑥) ℒ𝐴0 -formula ∶ 𝔐 ⊨ 𝜑(𝑡(𝑎))}.
Now the type of 𝑏 = 𝑡(𝑎) over 𝐴0 depends only on 𝑡(𝑥) (countably many
possibilities) and the type tp(𝑎/𝐴0 ) (really?). Write 𝑎 = 𝑎𝑖 for a tuple 𝑖 from
𝐼. Since the 𝑎𝑖 are indiscernible, the type depends only on the quantifier-free
type tpqf (𝑖/𝐼0 ) in the structure (𝐼, <) since it has quantifier elimination. This
type again depends on tpqf (𝑖) (finitely many possibilities) and on the types
𝑝(𝑥) = tpqf (𝑖/𝐼0 ) of the elements 𝑖 (Note the quantifier elimination, then we
only need to Booleanly combine these things to get tpqf (𝑖/𝐼0 )) of 𝑖. There
are three kinds of such types:

1. 𝑖 is bigger than all elements of 𝐼0

2. 𝑖 is an element 𝑖0 of 𝐼0

3. For some 𝑖0 ∈ 𝐼0 , 𝑖 is smaller than 𝑖0 but bigger than all elements of

{𝑗 ∈ 𝐼0 ∣ 𝑗 < 𝑖0 }

There is only one type in the first case, in the other case the type is de-
termined by 𝑖0 . This results in countably many possibilities for each com-
ponent of 𝑖

Definition 5.7. Let 𝐿 be a language. A Skolem theory Skolem(𝐿) is a theory

in a bigger language 𝐿Skolem with the following properties

1. Skolem(𝐿) has quantifier elimination

51
 2. Skolem(𝐿) is universal

3. Every 𝐿-structure can be expanded to a model of Skolem(𝐿)

4. |𝐿Skolem | ≤ max(|𝐿|, ℵ0 )

Theorem 5.8. Every language 𝐿 has a Skolem theory.

Proof. Nice slide. We have

1. ∃𝑥𝑃 (𝑥) is a consequence of 𝑃 (𝑎)

2. 𝑃 (𝑎) is not a consequence of ∃𝑥𝑃 (𝑥), but a model of ∃𝑥𝑃 (𝑥) provides
a model of 𝑃 (𝑎)

Skolemization eliminates existential quantifiers and transforms a closed

formula 𝐴 to a formula 𝐵 such that :

• 𝐴 is a consequence of 𝐵, 𝐵 ⊨ 𝐴

• every model of A provides a model of B

Hence, 𝐴 has a model if and only if 𝐵 has a model : skolemization preserves

the existence of a model, in other words it preserves satisfiability.
We define an ascending sequence of languages

𝐿 = 𝐿0 ⊆ 𝐿1 ⊆ 𝐿2 ⊆ ⋯

by introducing for every quantifier-free 𝐿𝑖 -formula 𝜑(𝑥1 , … , 𝑥𝑛 , 𝑦) a new 𝑛-

place Skolem function 𝑓𝜑 (if 𝑛 = 0, 𝑓𝜑 is a constant) and defining 𝐿𝑖+1 as
the union of 𝐿𝑖 and the set of these function symbols. The language 𝐿Skolem
is the union of all 𝐿𝑖 . We now define the Skolem theory as

Skolem = {∀𝑥(∃𝑦𝜑(𝑥, 𝑦) → 𝜑(𝑥, 𝑓𝜑 (𝑥))) ∣ 𝜑(𝑥, 𝑦) q.f. 𝐿Skolem -formula}

Corollary 5.9. Let 𝑇 be a countable theory with an infinite model and let 𝜅 be an
infinite cardinal. Then 𝑇 has a model of cardinality 𝜅 which realises only countably
many types over every countable subset.

Proof. Consider the theory 𝑇 ∗ = 𝑇 ∪ Skolem(𝐿). Then 𝑇 ∗ is countable, has

an infinite model and quantifier elimination
Claim. 𝑇 ∗ is equivalent to a universal theory

52
 Proof. Modulo Skolem(𝐿) every axiom 𝜑 of 𝑇 is equivalent to a quantifier-
free 𝐿Skolem -sentence 𝜑∗ . Therefore 𝑇 ∗ is equivalent to the universal theory
Let 𝐼 be a well-ordering of cardinality 𝜅 and 𝔑∗ a model of 𝑇 ∗ with indis-
cernibles (𝑎𝑖 )𝑖∈𝐼 (Existence by the Standard Lemma 5.3). The claim implies
that the substructure 𝔐∗ generated by the 𝑎𝑖 is a model of 𝑇 ∗ and 𝔐∗ has
cardinality 𝜅 (As we can’t control the size of an elementary extension and
Corollary 3.5). Since 𝑇 ∗ has quantifier elimination, 𝔐∗ is an elementary
substructure of 𝔑∗ and (𝑎𝑖 ) is indiscernible in 𝔐∗ . By Lemma 5.6, there are
only countably many types over every countable set realised in 𝔐∗ . The
same is then true for the reduct 𝔐 = 𝔐∗ ∣𝐿
Exercise 5.1.1. A sequence of elements in (ℚ, <) is indiscernible iff it is either
constant, strictly increasing or strictly decreasing
Proof. For any formula 𝜑(𝑥1 , 𝑥2 , … , 𝑥𝑛 ),
ℚ ⊨ 𝜑(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) ↔

5.2 𝜔-stable theories

In this section we fix a complete theory 𝑇 with infinite models
Our goal is theorem 5.20
Theorem 5.10. A countable theory 𝑇 is 𝜅-categorical iff all models of cardinality
𝜅 are saturated
∀𝔐 ⊨ 𝑇 ∧ |𝔐| = 𝜅 is saturated

elementarily equivalence saturated structures

of the same cardinality is isomorphic

𝑇 is 𝜅-categorical

𝑇 is 𝜔-stable

𝑇 is totally transcendental

𝑇 is 𝜅-stable

∀regular 𝜆 ≤ 𝜅,∃|𝔐| = 𝜅 𝜆-saturated

53
 In the previous section we saw that we may add indiscernible elements
to a model without changing the number of realised types. We will now
use this to show that ℵ1 -categorical theories a small number of types, i.e.,
they are 𝜔-stable. Conversely, with few types it is easier to be saturated and
since saturated structures are unique we find the connection to categorical
theories.

Definition 5.11. Let 𝜅 be an infinite cardinal. We say 𝑇 is 𝜅-stable if in each

model of 𝑇 , over every set of parameters of size at most 𝜅, and for each 𝑛,
there are at most 𝜅 many 𝑛-types, i.e.,

|𝐴| ≤ 𝜅 ⇒ |𝑆𝑛 (𝐴)| ≤ 𝜅

Note that if 𝑇 is 𝜅-stable, then - up to logical equivalence - we have |𝑇 | ≤

𝜅 (Exercise 5.2.3)

Lemma 5.12. 𝑇 is 𝜅-stable iff 𝑇 is 𝜅-stable for 1-types, i.e.,

|𝐴| ≤ 𝜅 ⇒ |𝑆(𝐴)| ≤ 𝜅

Proof. Assume that 𝑇 is 𝜅-stable for 1-types. We show that 𝑇 is 𝜅-stable for
𝑛-types by induction on 𝑛. Let 𝐴 be a subset of the model 𝔐 and |𝐴| ≤ 𝜅.
We may assume that all types over 𝐴 are realised in 𝔐 (otherwise we take
some elementary extensions by Corollary 2.14). Consider the restriction
map 𝜋 ∶ 𝑆𝑛 (𝐴) → 𝑆1 (𝐴). By assumption the image 𝑆1 (𝐴) has cardinality
at most 𝜅. Every 𝑝 ∈ 𝑆1 (𝐴) has the form tp(𝑎/𝐴) for some 𝑎 ∈ 𝑀 since
all types over 𝐴 are realized in 𝔐. By Exercise 4.2.3, the fibre 𝜋−1 (𝑝) is in
bijection with 𝑆𝑛−1 (𝑎𝐴) and so has cardinality at most 𝜅 by induction. This
shows |𝑆𝑛 (𝐴)| ≤ 𝜅 ⋅ 𝜅 = 𝜅.

Example 5.1 (Algebraically closed fields). The theories ACF𝑝 for 𝑝 a prime
or 0 are 𝜅-stable for all 𝜅

Note that by Theorem 5.15 below it would suffice to prove that the the-
ories ACF𝑝 are 𝜔-stable

Proof. Let 𝐾 be a subfield of an algebraically closed field. By quantifier elim-

ination, the type of an element 𝑎 over 𝐾 is determined by the isomorphism
type of the extension 𝐾[𝑎]/𝐾. If 𝑎 is transcendental over 𝐾, 𝐾[𝑎] is isomor-
phic to the polynomial ring 𝐾[𝑋]. If 𝑎 is algebraic with minimal polynomial
𝑓 ∈ 𝐾[𝑋], then 𝐾[𝑎] is isomorphic to 𝐾[𝑋]/(𝑓). So there is one more 1-type
over 𝐾 than there are irreducible polynomials

54
 That ACF𝑝 is 𝜅-stable for 𝑛-types has a direct algebraic proof: the iso-
morphism type of 𝐾[𝑎1 , … , 𝑎𝑛 ]/𝐾 is determined by the vanishing ideal 𝑃
of 𝑎1 , … , 𝑎𝑛 . By :((((
Theorem 5.13. A countable theory 𝑇 which is categorical in an uncountable car-
dinal 𝜅 is 𝜔-stable
Proof. Let 𝔑 be a model and 𝐴 ⊆ 𝑁 countable with 𝑆(𝐴) uncountable. Let
(𝑏𝑖 )𝑖∈𝐼 be a sequence of ℵ1 many elements with pairwise distinct types over
𝐴. (Note that we can assume that all types over 𝐴 are realised in 𝔑) We
choose first an elementary substructure 𝔐0 of cardinality ℵ1 which con-
tains 𝐴 and all 𝑏𝑖 . Then we choose an elementary extension 𝔐 of 𝔐0 . The
model 𝔐 is of cardinality 𝜅 and realises uncountably many types over the
countable set 𝐴. By Corollary 5.9, 𝑇 has another model where this is not the
case. So 𝑇 cannot be 𝜅-categorical

Definition 5.14. A theory 𝑇 is totally transcendental if it has no model 𝔐

with a binary tree of consistent 𝐿(𝑀 )-formulas
Theorem 5.15. 1. 𝜔-stable theories are totally transcendental
2. Totally transcendental theories are 𝜅-stable for all 𝜅 ≥ |𝑇 |
It follows that a countable theory 𝑇 is 𝜔-stable iff it is totally transcen-
dental

Proof. 1. Let 𝔐 be a model with a binary tree of consistent 𝐿(𝑀 )-formulas

with free variables among 𝑥1 , … , 𝑥𝑛 . The set 𝐴 of parameters which
occur in the tree’s formulas is countable but 𝑆𝑛 (𝐴) has cardinality 2ℵ0
2. Assume that there are there are more than 𝜅 many 𝑛-types over some
set 𝐴 of cardinality 𝜅. Let us call an 𝐿(𝐴)-formula big if it belongs
to more than 𝜅 many types over 𝐴 (|[𝜙]| > 𝜅) and thin otherwise.
By assumption the true formula is big. If we can show that each big
formula decomposes into two big formulas, we can construct a binary
tree of big formulas, which finishes the proof.
So assume that 𝜑 is big. Since each thin formula belongs to at most 𝜅
types and since there are at most 𝜅 formulas, there are at most 𝜅 types
which contain thin formulas. Therefore 𝜑 belongs to two distinct types
𝑝 and 𝑞 which contain only big formulas. If we separate 𝑝 and 𝑞 by
𝜓 ∈ 𝑝 and ¬𝜓 ∈ 𝑞, we decompose 𝜑 into the big formulas 𝜑 ∧ 𝜓 and
𝜑 ∧ ¬𝜓.

55
 The proof and Lemma 5.12 show that 𝑇 is totally transcendental iff there
is no binary tree of consistent formulas in one free variables
The general case follows from Exercise 5.2.2

Definition 5.16. Let 𝜅 be an infinite cardinal. An 𝐿-structure 𝔄 is 𝜅-saturated

if in 𝔄 all types over sets of cardinality less than 𝜅 are realised. An infinite
structure 𝔄 is saturated if it is |𝔄|-saturated

Lemma 4.13 generalises to sets

Lemma 5.17. Elementarily equivalent saturated structures of the same cardinality

are isomorphic

Proof. Let 𝔄 and 𝔅 be elementary equivalent saturated structures each of

cardinality 𝜅. We choose enumerations (𝑎𝛼 )𝛼<𝜅 and (𝑏𝛼 )𝛼<𝜅 of 𝐴 and 𝐵
and construct an increasing sequence of elementary maps 𝑓𝛼 ∶ 𝐴𝛼 → 𝐵𝛼 .
Assume that the 𝑓𝛽 are constructed for all 𝛽 < 𝛼. The union of the 𝑓𝛽 is an
elementary map 𝑓𝛼∗ ∶ 𝐴∗𝛼 → 𝐵𝛼∗ . The construction will imply that 𝐴∗𝛼 and
𝐵𝛼∗ have cardinality at most |𝛼|, which is smaller than 𝜅
We write 𝛼 = 𝜆 + 𝑛, and distinguish two cases
𝑛 = 2𝑖: In this case, we consider 𝑝(𝑥) = tp(𝑎𝜆+𝑖 /𝐴∗𝛼 ). Realise 𝑓𝛼∗ (𝑝) by
𝑏 ∈ 𝐵 and define
𝑓𝛼 = 𝑓𝛼∗ ∪ {⟨𝑎𝜆+𝑖 , 𝑏⟩}
𝑛 = 2𝑖 + 1: Similarly, we find an extension

𝑓𝛼 = 𝑓𝛼∗ ∪ {⟨𝑎, 𝑏𝜆+𝑖 ⟩}

Thus ⋃𝛼<𝜅 𝑓𝛼 is the desired isomorphism

Lemma 5.18. Let 𝑆0 ⊆ 𝑆1 ⊆ ⋯ ⊆ 𝑆𝛼 ⊆ ⋯ be an increasing chain of sets indexed

by 𝛼 < 𝜅 for some regular cardinal 𝜅. If 𝐴 ⊆ ⋃𝛼<𝜅 𝑆𝛼 and |𝐴| < 𝜅, then 𝐴 ⊆ 𝑆𝛼
for some 𝛼 < 𝜅

Proof. Define 𝑓 ∶ 𝐴 → 𝜅 by 𝑓(𝑥) = min{𝛼 ∶ 𝑥 ∈ 𝑆𝛼 }. Then |𝑓(𝐴)| ≤ |𝐴| < 𝜅,

so 𝛼 ∶= sup 𝑓(𝐴) < 𝜅. For any 𝑥 ∈ 𝐴, we have 𝑓(𝑥) ≤ 𝛼, and so 𝑥 ∈ 𝑆𝑓(𝑥) ⊆
𝑆𝛼 and thus 𝐴 ⊆ 𝑆𝛼

Lemma 5.19. If 𝑇 is 𝜅-stable, then for all regular 𝜆 ≤ 𝜅, there is a model of

cardinality 𝜅 which is 𝜆-saturated

Proof. By Exercise 5.2.3 we may assume that |𝑇 | ≤ 𝜅. Consider a model

𝔐 of cardinality 𝜅. Since 𝑆(𝑀𝛼 ) has cardinality 𝜅, Corollary 2.14 and the
Löwenheim–Skolem theorem give an elementary extension of cardinality

56
𝜅 in which all types over 𝔐 are realised. So can construct a continuous
elementary chain
𝔐0 ≺ 𝔐1 ≺ ⋯ ≺ 𝔐𝛼 ≺ ⋯ (𝛼 < 𝜆)
of models of 𝑇 with cardinality 𝜅 s.t. all 𝑝 ∈ 𝑆(𝑀𝛼 ) are realised in 𝔐𝛼+1 .
Then 𝔐 is 𝜆-saturated. In fact, if |𝐴| < 𝜆 and if 𝑎 ∈ 𝐴 is contained in 𝑀𝛼(𝑎)
then Λ = ⋃𝑎∈𝐴 𝛼(𝑎) is an initial segment of 𝜆 of smaller cardinality than 𝜆.
We can find since 𝜅 is regular iff cf(𝜅) = 𝜅 iff ∀𝛼 < 𝜅, ⋃𝛽<𝛼 𝑀𝛼 ⊊ 𝑀 . Thus
there is 𝛾 < 𝜅 s.t. ⋃𝛽<𝛼 𝑆𝛼 So Λ has an upper bound 𝜇 < 𝜆. It follows that
𝐴 ⊆ 𝔐𝜇 and all types over 𝐴 are realised in 𝔐𝜇+1
Remark. If 𝑇 is 𝜅-stable for a regular cardinal 𝜅, the previous lemma yields
a saturated model of cardinality 𝜅.
Theorem 5.20. A countable theory 𝑇 is 𝜅-categorical iff all models of cardinality
𝜅 are saturated
Proof. If all models of cardinality 𝜅 are saturated, it follows from Lemma
5.17 that 𝑇 is 𝜅-categorical
Assume, for the converse that 𝑇 is 𝜅-categorical. For 𝜅 = ℵ0 the theorem
follows from Theorem 4.11. So we may assume that 𝜅 is uncountable. Then
𝑇 is totally transcendental by Theorem 5.13 and 5.15 and therefore 𝜅-stable
by Theorem 5.15.
By Lemma 5.19, all models of 𝑇 of cardinality 𝜅 are 𝜇+ -saturated for all
𝜇 < 𝜅. i.e., 𝜅-saturated
Exercise 5.2.1. Show that the theory of an equivalence relation with two in-
finite classes has quantifier elimination and is 𝜔-stable. Is it ℵ1 -categorical?
Exercise 5.2.2. If 𝑇 is an 𝐿-theory and 𝐾 is a sublanguage of 𝐿, the reduct
𝑇 ↾𝐾 is the set of all 𝐾-sentences which follow from 𝑇 . Show that 𝑇 is totally
transcendental iff 𝑇 ↾𝐾 is 𝜔-stable for all at most countable 𝐾 ⊆ 𝐿
Proof.
Exercise 5.2.3. If 𝑇 is 𝜅-stable, then essentially (i.e., up to logical equivalence)
|𝑇 | ≤ 𝜅
Proof. First for any 𝜑, 𝜓 ∈ 𝑇 , define 𝜑 ∼ 𝜓 iff 𝑇 ⊨ 𝜑 ↔ 𝜓. If |𝑇 / ∼| > 𝜅.
If 𝑇 ⊭ 𝜑 ↔ 𝜓, then 𝑇 ⊨ (𝜑∧¬𝜓)∨(¬𝜑∧𝜓). Thus for any non-equivalent
𝜑 and 𝜓, they belong to different types. Thus 𝑆𝑛 (𝑇 ) > 𝜅.
If 𝑇 is 𝜅-stable, then |𝑆𝑛 (∅)| ≤ 𝜅. Choose for any two 𝑛-types over the
empty set a separating formula 𝜑. Then any formula is logically equivalent
to a finite Boolean combination of these 𝜅-many formulas.

57
5.3 Prime extensions
For any model 𝔐 ⊨ 𝑇 and any 𝐴 ⊆ 𝔐

𝑇 is totally transcendental

isolated types are dense over 𝐴

∃𝔐0 constructible extension of 𝐴

Constructible over 𝐴 ⇒ prime over 𝐴 𝔐0 is atomic

𝔐0 prime ∀𝔑 prime, 𝔑 is atomic

Definition 5.21. Let 𝔐 be a model of 𝑇 and 𝐴 ⊆ 𝑀 .

1. 𝔐 is a prime extension of 𝐴 (or prime over 𝐴) if every elementary
map 𝐴 → 𝔑 extends to an elementary map 𝔐 → 𝔑

𝔐 𝔑
𝑖𝑑𝐴

2. 𝐵 ⊆ 𝑀 is constructible over 𝐴 if 𝐵 has an enumeration

𝐵 = {𝑏𝛼 ∣ 𝛼 < 𝜆}
where each 𝑏𝛼 is atomic over 𝐴 ∪ 𝐵𝛼 (tp(𝑏𝛼 /𝐴 ∪ 𝐵𝛼 ) is isolated), with
𝐵𝛼 = {𝑏𝜇 ∣ 𝜇 < 𝛼}
So 𝔐 is a prime extension of 𝐴 iff 𝔐𝐴 is a prime model of Th(𝔐𝐴 )
Lemma 5.22. If a model 𝑀 is constructible over 𝐴, then 𝔐 is prime over 𝐴
Proof. Let (𝑚𝛼 )𝛼<𝜆 an enumeration of 𝑀 , s.t. each 𝑚𝛼 is atomic over 𝐴∪𝑀𝛼 .
Let 𝑓 ∶ 𝐴 → 𝔑 be an elementary map. We define inductively an increasing
sequence of elementary maps 𝑓𝛼 ∶ 𝐴∪𝑀𝛼+1 → 𝔑 with 𝑓0 = 𝑓. Assume that
𝑓𝛽 is defined for all 𝛽 < 𝛼. The union of these 𝑓𝛽 is an elementary map 𝑓𝛼′ ∶
𝐴∪𝑀𝛼 → 𝔑. Since 𝑝(𝑥) = tp(𝑎𝛼 /𝐴∪𝑀𝛼 ) is isolated, 𝑓𝛼′ (𝑝) ∈ 𝑆(𝑓𝛼′ (𝐴∪𝑀𝛼 ))
is also isolated and has a realisation 𝑏 in 𝔑. We set 𝑓𝛼 = 𝑓𝛼′ ∪ {⟨𝑎𝛼 , 𝑏⟩}
Finally, the union of all 𝑓𝛼 (𝛼 < 𝜆) is an elementary embedding 𝔐 →
𝔑.

58
Theorem 5.23. If 𝑇 is totally transcendental, every subset of a model of 𝑇 has a
constructible prime extension

Marker’s Theorem 4.2.11

Lemma 5.24. If 𝑇 is totally transcendental, the isolated types are dense over every
subset of any model

Proof. Consider a subset 𝐴 of a model 𝔐. Then Th(𝔐𝐴 ) ⊃ 𝑇 has no binary

tree of consistent formulas. By Theorem 4.31, the isolated types in Th(𝔐𝐴 )
are dense

Proof of Theorem 5.23. By Lemma 5.22 it suffices to construct an elementary

substructure 𝔐0 ≺ 𝔐 which contains 𝐴 and is constructible over 𝐴. An ap-
plication of Zorn’s Lemma gives us a maximal construction (𝑎𝛼 )𝛼<𝜆 , which
cannot be prolonged by an element 𝑎𝜆 ∈ 𝑀 ⧵ 𝐴𝜆 . We need to show first that
we can find 𝑎 s.t. 𝑎 is atomic over 𝐴. But as the isolated types are dense over
𝐴, pick an 𝐿(𝐴)-formula 𝜑 s.t. 𝔐 ⊨ 𝜑(𝑎). Then 𝑎 is atomic over 𝐴. Clearly
𝐴 is contained in 𝐴𝜆 . We show that 𝐴𝜆 is the universe of an elementary sub-
structure 𝔐0 using Tarski’s Test. So assume that 𝜑(𝑥) is an 𝐿(𝐴𝜆 )-formula
and 𝔐 ⊨ ∃𝑥𝜑(𝑥). Since isolated types over 𝐴𝜆 are dense by Lemma 5.24,
there is an isolated 𝑝(𝑥) ∈ 𝑆(𝐴𝜆 ) containing 𝜑(𝑥). Let 𝑏 be a realisation of
𝑝(𝑥) in 𝔐. We can prolong our construction by 𝑎𝜆 = 𝑏; thus 𝑏 ∈ 𝐴𝜆 by
maximality and 𝜑(𝑥) is realised in 𝐴𝜆 .

Lemma 5.25. Let 𝑎 and 𝑏 be two finite tuples of elements of a structure 𝔐. Then
tp(𝑎𝑏) is atomic iff tp(𝑎/𝑏) and tp(𝑏) are atomic

Proof. If 𝜑(𝑥, 𝑦) isolates tp(𝑎, 𝑏). As in the proof of Theorem 4.25, 𝜑(𝑥, 𝑏)
isolates tp(𝑎/𝑏) and we claim that ∃𝑥𝜑(𝑥, 𝑦) isolates 𝑝(𝑦) = tp(𝑏): we have
∃𝑥𝜑(𝑥, 𝑦) ∈ 𝑝(𝑦) and if 𝜎(𝑦) ∈ 𝑝(𝑦), then

𝔐 ⊨ ∀𝑥, 𝑦(𝜑(𝑥, 𝑦) → 𝜎(𝑦))

Hence 𝔐 ⊨ ∀𝑦(∃𝑥𝜑(𝑥, 𝑦) → 𝜎(𝑦)).

Now conversely, assume that 𝜌(𝑥, 𝑏) isolates tp(𝑎/𝑏) and that 𝜎(𝑦) iso-
lates 𝑝(𝑦) = tp(𝑏). Then 𝜌(𝑥, 𝑦) ∧ 𝜎(𝑦) isolates. Firstly, 𝜌(𝑥, 𝑦) ∧ 𝜎(𝑦) ∈
tp(𝑎, 𝑏). If 𝜑(𝑥, 𝑦) ∈ tp(𝑎, 𝑏), then 𝜑(𝑥, 𝑏) ∈ tp(𝑎/𝑏) and

𝔐 ⊨ ∀𝑥(𝜌(𝑥, 𝑏) → 𝜑(𝑥, 𝑏))

Hence
∀𝑥(𝜌(𝑥, 𝑦) → 𝜑(𝑥, 𝑦)) ∈ 𝑝(𝑦)

59
and it follows that

𝔐 ⊨ ∀𝑦(𝜎(𝑦) → ∀𝑥(𝜌(𝑥, 𝑦) → 𝜑(𝑥, 𝑦)))

Thus 𝔐 ⊨ ∀𝑥, 𝑦(𝜌(𝑥, 𝑦) ∧ 𝜎(𝑦) → 𝜑(𝑥, 𝑦))

Corollary 5.26. Constructible extensions are atomic

Proof. Let 𝔐0 be a constructible extension of 𝐴 and let 𝑎 be a tuple from
𝑀0 . We have to show that 𝑎 is atomic over 𝐴. We can clearly assume that
the elements of 𝑎 are pairwise distinct and do not belong to 𝐴. We can also
permute the elements of 𝑎 so that

𝑎 = 𝑎𝛼 𝑏

for some tuple 𝑏 ∈ 𝐴𝛼 . Let 𝜑(𝑥, 𝑐) be an 𝐿(𝐴𝛼 )-formula which is complete

over 𝐴𝛼 and satisfied by 𝑎𝛼 𝑎𝛼 is also atomic over 𝐴∪{𝑏𝑐}. Using induction,
we know that 𝑏𝑐 is atomic over 𝐴. Note that 𝑏𝑐 ∈ 𝐴𝛼 , then we find a smaller
ordinal. This process will end as there is no infinite decreasing sequence.
By Lemma 5.25 applied to (𝔐0 )𝐴 , 𝑎𝛼 𝑏𝑐 is atomic over 𝐴, which implies that
𝑎 = 𝑎𝛼 𝑏 is atomic over 𝐴.

Corollary 5.27. If 𝑇 is totally transcendental, prime extensions are atomic

Proof. Let 𝔐 be a model of 𝑇 and 𝐴 ⊆ 𝑀 . Since 𝐴 has at least one con-
structible extension 𝔐0 and since all prime extensions of 𝐴 are contained
in 𝔐0 (isomorphic over 𝐴 to elementary substructure of 𝔐0 ), all prime ex-
tensions are atomic

A structure 𝔐 is called a minimal extension of the subset 𝐴 if 𝑀 has no

proper elementary substructure which contains 𝐴
Lemma 5.28. Let 𝔐 be a model of 𝑇 and 𝐴 ⊆ 𝑀 . If 𝐴 has a prime extension
and a minimal extension, they are isomorphic over 𝐴, i.e., there is an isomorphism
fixing 𝐴 elementwise
Proof. A prime extension embeds elementarily in the minimal extension.
This embedding must be surjective by minimality

Exercise 5.3.1. For every countable 𝑇 the following are equivalent

1. Every parameter set has a prime extension (We say that 𝑇 has prime
extensions)

2. Over every countable parameter set the isolated types are dense

60
 3. Over every parameter set the isolated types are dense

Proof. 3 → 2 → 1 is from the text.

1 → 3 from Theorem 4.29

5.4 Lachlan’s Theorem

Theorem 5.29 (Lachlan). Let 𝑇 be totally transcendental and 𝔐 an uncountable
model of 𝑇 . Then 𝔐 has arbitrary large elementary extensions which omit every
countable set of 𝐿(𝑀 )-formulas that is omitted in 𝔐.

Proof. We call an 𝐿(𝑀 )-formula large if its realisation set 𝜑(𝔐) is uncount-
able. Since there is no infinite binary tree of large formulas, there exists a
minimal large formula 𝜑0 (𝑥) in the sense that for every 𝐿(𝑀 )-formula 𝜓(𝑥)
either 𝜑0 (𝑥) ∧ 𝜓(𝑥) or 𝜑0 (𝑥) ∧ ¬𝜓(𝑥) is at most countable. Now it’s easy to
see that
𝑝(𝑥) = {𝜓(𝑥) ∣ 𝜑0 (𝑥) ∧ 𝜓(𝑥) large}
is a type in 𝑆(𝑀 ). For any formula 𝜓, if 𝜑(𝔐) = (𝜑(𝔐) ∧ 𝜓(𝔐)) ∪ (𝜑(𝔐) ∧
¬𝜓(𝔐)). So exactly one of it belongs to 𝑝(𝑥).
Clearly 𝑝(𝑥) contains no formula of the form 𝑥=𝑎 ̇ for 𝑎 ∈ 𝑀 , so 𝑝(𝑥) is
not realised in 𝑀 . On the other hand, every countable subset Π(𝑥) ⊆ 𝑝(𝑥) is
realised in 𝔐: since 𝜑0 (𝔐) ⧵ 𝜓(𝔐) = 𝜑0 (𝔐) ∧ ¬𝜓(𝔐) is at most countable
for every 𝜓(𝑥) ∈ Π(𝑥), the elements of 𝜑0 (𝔐) which do not belong to the
union of these sets realised Π(𝑥).
Let 𝑎 be a realisation of 𝑝(𝑥) in a (proper) elementary extension 𝔑. By
Theorem 5.23, we can assume that 𝔑 is atomic over 𝔐 ∪ {𝑎}.
Fix 𝑏 ∈ 𝑁 . We have to show that every countable subset Σ(𝑦) ⊂ tp(𝑏/𝑀 )
is realised in 𝑀 . If the countable set is omitted in 𝔐, then it is omitted in
𝔑.
Let 𝜒(𝑥, 𝑦) be an 𝐿(𝑀 )-formula s.t. 𝜒(𝑎, 𝑦) isolates 𝑞(𝑦) = tp(𝑏/𝑀 ∪{𝑎}).
If 𝑏 realised an 𝐿(𝑀 )-formula 𝜎(𝑦), we have 𝔑 ⊨ ∀𝑦(𝜒(𝑎, 𝑦) → 𝜎(𝑦)). Hence
the formula
𝜎∗ (𝑥) = ∀𝑦(𝜒(𝑥, 𝑦) → 𝜎(𝑦))
belongs to 𝑝(𝑥) 𝑝(𝑥) = tp𝔑 (𝑎/𝑀 ). Note that ∃𝑦𝜒(𝑥, 𝑦) belongs also to 𝑝(𝑥).
Choose an element 𝑎′ ∈ 𝑀 which satisfies

{𝜎∗ (𝑥) ∣ 𝜎 ∈ Σ} ∪ {∃𝑦𝜒(𝑥, 𝑦)}

and choose 𝑏′ ∈ 𝑀 with 𝔐 ⊨ 𝜒(𝑎′ , 𝑏′ ). Since 𝔐 ⊨ 𝜎∗ (𝑎′ ), 𝔐 ⊨ 𝜎(𝑏′ ). So 𝑏′

realises Σ(𝑦).

61
 We have shown that 𝔐 has a proper elementary extension which realises
no new countable set of 𝐿(𝑀 )-formulas. By iteration we obtain arbitrarily
long chains of elementary extensions with the same property

Corollary 5.30. A countable theory which is 𝜅-categorical for some uncountable

𝜅, is ℵ1 -categorical
Proof. Let 𝑇 be 𝜅-categorical and assume that 𝑇 is not ℵ1 -categorical. Then
𝑇 has a model 𝔐 of cardinality ℵ1 which is not saturated by Theorem 5.20.
≇
𝔐 𝔑
≺ ≺

𝔐′ ≅
𝔑′

For any sentence 𝜙, 𝔐 ⊨ 𝜙 ⇒ 𝔐′ ⊨ 𝜙 ⇒ 𝔑′ ⊨ 𝜙 ⇒ 𝔑 ⊨ 𝜙. Thus 𝔐 ≡

𝔑. Then we can use Lemma 5.17. So there is a type 𝑝 over a countable
subset of 𝑀 which is not realised in 𝔐. By Theorem 5.13 and 5.15 𝑇 is
totally transcendental and we have a atomic constructible prime extension.
Theorem 5.29 gives an elementary extension 𝔑 of 𝔐 of cardinality 𝜅 which
omits all countable sets of formulas which are omitted in 𝔐. Thus also 𝑝 is
also omitted. Since 𝔑 is not saturated, 𝑇 is not 𝜅-categorical, a contradiction.

Exercise 5.4.1. Prove in a similar way: if a countable theory 𝑇 is 𝜅-categorical

for some uncountable 𝜅, it is 𝜆-categorical for every uncountable 𝜆 ≤ 𝜅

5.5 Vaughtian pairs

A crucial fact about uncountably categorical theories is the absence of de-
finable sets whose size is independent of the size of the model in which they
live
In this section, 𝑇 is a countable complete theory with infinite models

𝑇 dosen’t have a Vaughtian pair 𝑇 eliminates ∃∞

𝑇 is 𝜅-categorical for 𝜅 > ℵ0

Prime extension of 𝐴 ∪ 𝜑(𝔐) is unique

Definition 5.31. We say that 𝑇 has a Vaughtian pair if there are two models
𝔐 ≺ 𝔑 and an 𝐿(𝑀 )-formula 𝜑(𝑥) s.t.

62
 1. 𝔐 ≠ 𝔑

2. 𝜑(𝔐) is infinite

3. 𝜑(𝔐) = 𝜑(𝔑)
If 𝜑(𝑥) doesn’t contain parameters, we say that 𝑇 has a Vaughtian pair
for 𝜑(𝑥)
Remark. Notice that 𝑇 does not have a Vaughtian pair iff every model 𝔐 is
a minimal extension of 𝜑(𝔐) ∪ 𝐴 for any formula 𝜑(𝑥) with parameters in
𝐴 ⊆ 𝑀 which defines an infinite set in 𝔐. If 𝔐 ≺ 𝔑 is a Vaughtian pair and
𝜑(𝔐) = 𝜑(𝔑). Then as 𝔑 is the minimal extension of 𝜑(𝔐) ∪ 𝐴, 𝔑 ≺ 𝔐
and thus we have an isomorphism
Let 𝔑 be a model of 𝑇 where 𝜑(𝔑) is infinite but has smaller cardinality
than 𝔑. The Löwenheim–Skolem Theorem yields an elementary substruc-
ture 𝔐 of 𝔑 which contains 𝜑(𝔑) and has the same cardinality as 𝜑(𝔑).
Then 𝔐 ≺ 𝔑 is a Vaughtian pair for 𝜑(𝑥). The next theorem shows that a
converse of this observation is also true
Theorem 5.32 (Vaught’s Two-cardinal Theorem). If 𝑇 has a Vaughtian pair,
it has a model 𝔐 of cardinality ℵ1 with 𝜑(𝔐) countable for some formula 𝜑(𝑥) ∈
𝐿(𝑀 )
Lemma 5.33. Let 𝑇 be complete, countable and with infinite models
1. Every countable model of 𝑇 has a countable 𝜔-homogeneous elementary ex-
tension

2. The union of an elementary chain of 𝜔-homogeneous models is 𝜔-homogeneous

3. Two 𝜔-homogeneous countable models of 𝑇 realizing the same 𝑛-types for all
𝑛 < 𝜔 are isomorphic
Proof. 1. Let 𝔐0 be a countable model of 𝑇 . We realise the countably
many types

{𝑓(tp(𝑎/𝐴)) ∣ 𝑎, 𝐴 ⊆ 𝑀0 , 𝐴 finite, 𝑓 ∶ 𝐴 → 𝑀0 elementary}

in a countable elementary extension 𝔐1 . By iterating this process we

obtain an elementary chain

𝔐0 ≺ 𝔐1 ≺ ⋯

whose union is 𝜔-homogeneous

63
 2. Clear

3. Suppose 𝔄 and 𝔅 are 𝜔-homogeneous, countable and realise the same

𝑛-types. We show that we can extend any finite elementary map 𝑓 ∶
{𝑎1 , … , 𝑎𝑖 } → {𝑏1 , … , 𝑏𝑖 }; 𝑎𝑗 ↦ 𝑏𝑗 to any 𝑎 ∈ 𝐴 ⧵ 𝐴𝑖 . Realise the type
tp(𝑎1 , … , 𝑎𝑖 , 𝑎) by some tuples 𝑏′ = 𝑏1′ , … , 𝑏𝑖+1
′
in 𝐵. tp𝔄 (𝑎1 , … , 𝑎𝑖 , 𝑎) =
𝔅 ′
tp (𝑏 ) ⇒ tp (𝑏1 , … , 𝑏𝑖 ) = tp (𝑎1 , … , 𝑎𝑖 ) = tp𝔅 (𝑏1′ , … , 𝑏𝑖′ ).
𝔅 𝔄

Using the 𝜔-homogeneity of 𝐵 we may extend the finite partial iso-

′
morphism 𝑔 = {(𝑏𝑗′ , 𝑏𝑗 ) ∣ 1 ≤ 𝑗 ≤ 𝑖} by (𝑏𝑖+1 , 𝑏) for some 𝑏 ∈ 𝐵. Then
𝑓𝑖+1 = 𝑓𝑖 ∪ {(𝑎, 𝑏)} is the required extension. Reverse the roles of 𝐵
and 𝐴 we construct the desired isomorphism.

Proof of Theorem 5.32. Suppose that the Vaughtian pair is witnessed (in cer-
tain models) by some formula 𝜑(𝑥). For simplicity we assume that 𝜑(𝑥)
does not contain parameters (see Exercise 5.5.2). Let 𝑃 be a new unary
predicate. It is easy to find an 𝐿(𝑃 )-theory 𝑇VP whose models (𝔑, 𝑀 ) con-
sist of a model 𝔑 ⊨ 𝑇 and a subset 𝑀 defined by the new predicate 𝑃 which
is the universe of an elementary substructure 𝔐 which together with 𝔑
forms a Vaughtian pair for 𝜑(𝑥). We can express the condition for Vaugh-
tian pair in first-order language with 𝑃 :

1. ∃𝑥(¬𝑃 (𝑥))

2. For each 𝑘 > 0, ∃𝑣1 … 𝑣𝑘 (⋀𝑖<𝑗 𝑣𝑖 ≠ 𝑣𝑗 ∧ ⋀ 𝜑(𝑣𝑖 ))

3. ∀𝑥(𝜑(𝑥) → 𝑃 (𝑥))

And in addition, the elementary substructure

𝑘
4. ∀𝑣((⋀𝑖=1 𝑃 (𝑣𝑖 ) ∧ 𝜓(𝑣)) → 𝜓𝑃 (𝑣))

As in Marker’s p152. Let 𝔐 be the elementary substructure of 𝔑 by

Löwenheim–Skolem Theorem . The Löwenheim–Skolem Theorem applied
to 𝑇VP yields a Vaughtian pair 𝔐0 ≺ 𝔑0 for 𝜑(𝑥) with 𝔐0 , 𝔑0 countable
We first construct an elementary chain

(𝔑0 , 𝑀0 ) ≺ (𝔑1 , 𝑀1 ) ≺ ⋯

of countable Vaughtian pairs, with the aim that both components of the
union pair
(𝔑, 𝑀 )

64
are 𝜔-homogeneous and realise the same 𝑛-types. If (𝔑𝑖 , 𝑀𝑖 ) is given, we
first choose a countable elementary extension (𝔑′ , 𝑀 ′ ) s.t. 𝔐′ realises all
𝑛-types which are realised in 𝔑𝑖 . We only need to consider the 1-type. Then
for each 1-type in 𝔑𝑖 , add a constant. Then we choose as in the proof of
Lemma 5.33 a countable elementary extension (𝔑𝑖+1 , 𝔐𝑖+1 ) of (𝔑′ , 𝑀 ′ ) for
which 𝔑𝑖+1 and 𝔐𝑖+1 are 𝜔-homogeneous Prove: If (𝔑, 𝑀 ) is a countable
𝜔-homogeneous elementary extension of (𝔑′ , 𝑀 ′ ), then both 𝔑 and 𝔐 are
homogeneous
Suppose tp𝔐 (𝑎) = tp𝔐 (𝑏) where 𝑎, 𝑏 ∈ 𝑀 𝑛 and take 𝑎 ∈ 𝔐. For any
𝜑(𝑥) ∈ tp𝔐 (𝑎), 𝜑(𝑥) ∈ tp(𝔑,𝑀) (𝑎) and so tp(𝔑,𝑀) (𝑎) = tp(𝔑,𝑀) (𝑏). And
𝑛+1
there is 𝑏 ∈ 𝔑 s.t. tp(𝔑,𝑀) (𝑎, 𝑎) = tp(𝔑,𝑀) (𝑏, 𝑏). But note that ⋀𝑖=1 𝑃 (𝑥𝑖 ) ∈
tp(𝔑,𝑀) (𝑏, 𝑏) and hence 𝑏 ∈ 𝑀 .
It follows from Lemma 5.33 (3) that 𝔐 and 𝔑 are isomorphic since 𝔐 ≺
𝔑.
Next we construct a continuous elementary chain
𝔐0 ≺ 𝔐1 ≺ ⋯ ≺ 𝔐𝛼 ≺ ⋯ (𝛼 < 𝜔1 )
with (𝔐𝛼+1 , 𝔐𝛼 ) ≅ (𝔑, 𝑀 ) for all 𝛼. We start with 𝔐0 = 𝔐. If 𝔐𝛼 is
constructed, we choose an isomorphism 𝔐 → 𝔐𝛼 and extend it to an iso-
morphism 𝔑 → 𝔐𝛼+1 (Lemma 1.5). For a countable limit ordinal 𝜆, 𝔐𝜆 is
the union of the 𝔐𝛼 (𝛼 < 𝜆). So 𝔐𝜆 is isomorphic to 𝔐 by Lemma 5.33 (2)
and 5.33 (3)
Finally we set
𝔐 = ⋃ 𝔐𝛼
𝛼<𝜔1

𝔐 has cardinality ℵ1 while 𝜑(𝔐) = 𝜑(𝔐𝛼 ) = 𝜑(𝔐0 ). If 𝔐 ⊨ 𝜑(𝑎), then

there is some 𝛼 < 𝜔1 s.t. 𝑎 ∈ 𝑀 𝛼 ≅ 𝑀 0 .
Corollary 5.34. If 𝑇 is categorical in an uncountable cardinality, it does not have
a Vaughtian pair
Proof. If 𝑇 has a Vaughtian pair, then by Theorem 5.32 it has a model 𝔐 of
cardinality ℵ1 s.t. for some 𝜑(𝑥) ∈ 𝐿(𝑀 ) the set 𝜑(𝔐) is countable. On the
other hand, if 𝑇 is categorical in an uncountable cardinal, it is ℵ1 -categorical
by Corollary 5.30 and by Theorem 5.20, all models of 𝑇 of cardinality ℵ1 are
saturated. In particular, each formula is either satisfied by a finite number
or by ℵ1 many elements, a contradiction.
Corollary 5.35. Let 𝑇 be categorical in an uncountable cardinal, 𝔐 a model, and
𝜑(𝔐) infinite and definable over 𝐴 ⊆ 𝑀 . Then 𝔐 is the unique prime extension
of 𝐴 ∪ 𝜑(𝔐)

65
Proof. By Corollary 5.34, 𝑇 does not have a Vaughtian pair, so 𝔐 is minimal
over 𝐴 ∪ 𝜑(𝔐). If 𝔑 is a prime extension

Definition 5.36. We say that 𝑇 eliminates the quantifier ∃∞ 𝑥 (there are

infinitely many 𝑥), if for every 𝐿-formula 𝜑(𝑥, 𝑦) there is a finite bound 𝑛𝜑
s.t. in all models 𝔐 of 𝑇 and for all parameters 𝑎 ∈ 𝑀

𝜑(𝔐, 𝑎)

is either infinite or has at most 𝑛𝜑 elements

Remark. This means that for all 𝜑(𝑥, 𝑦) there is a 𝜓(𝑦) s.t. in all models 𝔐 of
𝑇 and for all 𝑎 ∈ 𝑀

𝔐 ⊨ ∃∞ 𝑥𝜑(𝑥, 𝑎) ⟺ 𝔐 ⊨ 𝜓(𝑎)

We denote this by
𝑇 ⊨ ∀𝑦 (∃∞ 𝑥𝜑(𝑥, 𝑦) ↔ 𝜓(𝑦))

Proof. If 𝑛𝜑 exists, we can use 𝜓(𝑦) = ∃>𝑛𝜑 𝑥𝜑(𝑥, 𝑦). If conversely 𝜓(𝑦) is
a formula which is implied by ∃∞ 𝑥𝜑(𝑥, 𝑦), a compactness argument shows
that there must be a bound 𝑛𝜑 s.t.

𝑇 ⊨ ∃>𝑛𝜑 𝑥𝜑(𝑥, 𝑦) → 𝜓(𝑦)

First note that 𝑇 is complete. If there is no such bound, then for any 𝑛 ∈ ℕ,
𝑇 ⊭ ∃>𝑛 𝑥𝜑(𝑥, 𝑦) → 𝜓(𝑦), which is 𝑇 ⊨ ∃>𝑛 𝑥𝜑(𝑥, 𝑦) ∧ ¬𝜓(𝑦). Thus by
compactness 𝑇 ⊨ ∃∞ 𝑥𝜑(𝑥, 𝑦) ∧ ¬𝜓(𝑦), a contradiction.

Lemma 5.37. A theory 𝑇 without Vaughtian pair eliminates the quantifier ∃∞ 𝑥

Check Marker’s Lemma 4.3.37 and Lemma 6.1.14

Proof. Let 𝑃 be a new unary predicate and 𝑐1 , … , 𝑐𝑛 new constants. Let 𝑇 ∗

be the theory Check Marker’s Lemma 6.1.14 to see the the formal version
of all 𝐿 ∪ {𝑃 , 𝑐1 , … , 𝑐𝑛 }-structures

(𝔐, 𝑁 , 𝑎1 , … , 𝑎𝑛 )

where 𝔐 is a model of 𝑇 , 𝑁 is the universe of a proper elementary substruc-

ture, 𝑎1 , … , 𝑎𝑛 elements of 𝑁 and 𝜑(𝔐, 𝑎) ⊆ 𝑁 . Suppose that the bound 𝑛𝜑
does not exists. Then, for any 𝑛, there is a model 𝔑 of 𝑇 and 𝑎 ∈ 𝑁 s.t.
𝜑(𝔑, 𝑎) is finite, but has more than 𝑛 elements. Let 𝔐 be a proper elemen-
tary extension of 𝔑. Then 𝜑(𝔐, 𝑎) = 𝜑(𝔑, 𝑎) (as 𝜑(𝔑, 𝑎) is finite, we can

66
add formulas to ensure this) and the pair (𝔐, 𝑁 , 𝑎) is a model of 𝑇 ∗ . This
shows that the theory

𝑇 ∗ ∪ {∃>𝑛 𝑥𝜑(𝑥, 𝑐) ∣ 𝑛 = 1, 2, … }

is finitely satisfiable. A model of this theory gives a Vaughtian pair for 𝑇 .

Exercise 5.5.1. If 𝑇 is totally transcendental and has a Vaughtian pair for

𝜑(𝑥), then it has, for all uncountable 𝜅, a model of cardinality 𝜅 with count-
able 𝜑(𝔐).

Proof. Marker’s Theorem 4.3.41

Exercise 5.5.2. Let 𝑇 be a theory, 𝔐 a model of 𝑇 and 𝑎 ⊆ 𝑀 a finite tuple

of parameters. Let 𝑞(𝑥) be the type of 𝑎 in 𝔐. Then for new constants 𝑐, the
𝐿(𝑐)-theory

𝑇 (𝑞) = Th(𝔐, 𝑎) = 𝑇 ∪ {𝜑(𝑐) ∣ 𝜑(𝑥) ∈ 𝑞(𝑥)}

is complete. Show that 𝑇 is 𝜆-stable (or without Vaughtian pair etc.) iff
𝑇 (𝑞) is. For countable languages this implies that 𝑇 is categorical in some
uncountable cardinal iff 𝑇 (𝑞) is.

Proof. If 𝑇 is 𝜆-stable and 𝔑, 𝑏 ⊨ 𝑇 (𝑞), then there is an partial elementary

map 𝑓 ∶ 𝑎 → 𝑏 from 𝔐 to 𝔑. By Marker’s Corollary 4.1.7, we can extend 𝑓
to an elementary map 𝑓 ′ ∶ 𝔐 → 𝔑′ where 𝔑 ≺ 𝔑′ .

5.6 Algebraic formulas

Definition 5.38. Let 𝔐 be a structure and 𝐴 a subset of 𝑀 . A formula 𝜑(𝑥) ∈
𝐿(𝐴) is called algebraic if 𝜑(𝔐) is finite. An element 𝑎 ∈ 𝑀 is algebraic over
𝐴 if it realizes an algebraic 𝐿(𝐴)-formula. We call an element algebraic if it
is algebraic over the empty set. The algebraic closure of 𝐴, acl(𝐴), is the set
of all elements of 𝔐 algebraic over 𝐴, and 𝐴 is called algebraically closed
if it equals its algebraic closure

Remark. Note that the algebraic closure of 𝐴 does not grow in elementary
extensions of 𝔐 because an 𝐿(𝐴)-formula which defines a finite set in 𝔐
defines the same set in every elementary extension We can express there
are exactly 𝑚 solutions in formula.

67
 By Theorem 2.15
|acl(𝐴)| ≤ max(|𝑇 |, |𝐴|)
In algebraically closed fields, an element 𝑎 is algebraic over 𝐴 precisely if
𝑎 is algebraic (in the field-theoretical sense) over the field generated by 𝐴.
This follows from quantifier elimination in ACF
We call a type 𝑝(𝑥) ∈ 𝑆(𝐴) algebraic iff 𝑝 contains an algebraic for-
mula. Any algebraic type 𝑝 is isolated by an algebraic formula 𝜑(𝑥) ∈ 𝐿(𝐴),
namely by any 𝜑 ∈ 𝑝 having the minimal number of solutions in 𝔐. Sup-
pose 𝜓 ∈ 𝑝 is algebraic. If 𝜓 doesn’t isolate 𝑝. Then there is 𝜙 ∈ 𝑝 s.t.
𝜙 ∧ 𝜓(𝔐) is a proper subset of 𝜓(𝔐). This process will end since 𝜓 is alge-
braic This number is called the degree deg(𝑝) of 𝑝. As isolated types are
realised in every model, the algebraic types over 𝐴 are exactly of the form
tp(𝑎/𝐴) where 𝑎 is algebraic over 𝐴. The degree of 𝑎 over 𝐴 deg(𝑎/𝐴) is the
degree of tp(𝑎/𝐴).

Lemma 5.39. Let 𝑝 ∈ 𝑆(𝐴) be non-algebraic and 𝐴 ⊆ 𝐵. Then 𝑝 has a non-

algebraic extension 𝑞 ∈ 𝑆(𝐵).

Proof. The extension 𝑞0 (𝑥) = 𝑝(𝑥) ∪ {¬𝜓(𝑥) ∣ 𝜓(𝑥) algebraic 𝐿(𝐵)-formula}

is finitely satisfiable. For otherwise there are 𝜑(𝑥) ∈ 𝑝(𝑥) ( 𝑝 is a type and
is closed under conjunction) and algebraic 𝐿(𝐵)-formulas 𝜓1 (𝑥), … , 𝜓𝑛 (𝑥)
with
𝔐 ⊨ ∀𝑥(𝜑(𝑥) → 𝜓1 (𝑥) ∨ ⋯ ∨ 𝜓𝑛 (𝑥))
But then 𝜑(𝑥) (𝜑(𝑥) has finitely many solutions) and hence 𝑝(𝑥) is algebraic.
So we can take for 𝑞 any type containing 𝑞0 .

Remark. Since algebraic types are isolated by algebraic formulas, an easy

compactness argument shows that a type 𝑝 ∈ 𝑆(𝐴) is algebraic iff 𝑝 has
only finitely many realisations (namely deg(𝑝) many) in all elementary ex-
tensions of 𝔐.

Proof. ⇒. Obvious.
⇐. Suppose 𝑝 ∈ 𝑆(𝐴) is not algebraic in 𝔐. Add infinitely many con-
stants 𝐶, for any 𝜑 ∈ 𝑝, let Φ = {𝜑(𝑐) ∶ 𝑐 ∈ 𝐶} and

Γ = Diagel (𝔐) ∪ {𝑐 ≠ 𝑑 ∶ 𝑐, 𝑑 ∈ 𝐶} ∪ ⋃{Φ ∶ 𝜑 ∈ 𝑝}

Then Γ is finitely satisfied by 𝔐 and we have a model where 𝑝 has infinitely

many realisations

68
Lemma 5.40. Let 𝔐 and 𝔑 be two structures and 𝑓 ∶ 𝐴 → 𝐵 an elementary
bijection between two subsets. Then 𝑓 extends to an elementary bijection between
acl(𝐴) and acl(𝐵)

Proof. Let 𝑔 ∶ 𝐴′ → 𝐵′ a maximal extension of 𝑓 to two subsets of acl(𝐴) and

acl(𝐵). Let 𝑎 ∈ acl(𝐴). Since 𝑎 is algebraic over 𝐴′ , 𝑎 is atomic over 𝐴′ . We
can therefore realise the type 𝑔(tp(𝑎/𝐴′ )) in 𝔑 - by an element 𝑏 ∈ acl(𝐵) -
and obtain an extension 𝑔∪{⟨𝑎, 𝑏⟩} of 𝑔. It follows that 𝑎 ∈ 𝐴′ . So 𝑔 is defined
on the whole acl(𝐴). Interchanging 𝐴 and 𝐵 shows that 𝑔 is surjective

Definition 5.41. A pregeometry (or matroid) (𝑋, cl) is a set 𝑋 with a clo-
sure operator cl ∶ 𝒫(𝑋) → 𝒫(𝑋) where 𝒫 denotes the power set, s.t. for all
𝐴 ⊆ 𝑋 and 𝑎, 𝑏 ∈ 𝑋

1. (REFLEXIVITY) 𝐴 ⊆ cl(𝐴)

2. (FINITE CHARACTER) cl(𝐴) is the union of all cl(𝐴′ ), where the 𝐴′

range over all finite subsets of 𝐴

3. (TRANSITIVITY) cl(cl(𝐴)) = cl(𝐴)

4. (EXCHANGE) 𝑎 ∈ cl(𝐴𝑏) ⧵ cl(𝐴) ⇒ 𝑏 ∈ cl(𝐴𝑎)

A set 𝐴 is called closed if 𝐴 = cl(𝐴).

Lemma 5.42. If 𝑋 is the universe of a structure, acl satisfies REFLEXIVITY, FI-

NITE CHARACTER and TRANSITIVITY

5.7 Strongly minimal sets

We fix a complete theory 𝑇 with infinite models.

Definition 5.43. Let 𝔐 be a model of 𝑇 and 𝜑(𝑥) a non-algebraic 𝐿(𝑀 )-

formula

1. The set 𝜑(𝔐) is called minimal in 𝔐 if for all 𝐿(𝑀 )-formulas 𝜓(𝑥)
the intersection 𝜑(𝔐) ∧ 𝜓(𝔐) is either finite or cofinite in 𝜑(𝔐)

2. The formula 𝜑(𝑥)p is strongly minimal if 𝜑(𝑥) defines a minimal set

in all elementary extensions of 𝔐. In this case, we also call the de-
finable set 𝜑(𝔐) strongly minimal. A non-algebraic type containing a
strongly minimal formula is called strongly minimal

3. A theory 𝑇 is strongly minimal if the formula 𝑥=𝑥

̇ is strongly minimal

69
 Strong minimality is preserved under definable bijections; i.e., if 𝐴 and
𝐵 are definable subsets of 𝔐𝑘 , 𝔐𝑚 defined by 𝜑 and 𝜓, respectively, s.t.
there is a definable bijection between 𝐴 and 𝐵, then if 𝜑 is strongly minimal
so is 𝜓. Suppose bijection 𝑓(𝑎) = 𝑏 iff 𝛾(𝑎, 𝑏). Then for any 𝜃(𝑥), we have
𝜃′ (𝑦) = ∃𝑥(𝜃(𝑥) ∧ 𝛾(𝑥, 𝑦))

Example 5.2. 1. The following theories are strongly minimal, which is

easily seen in each case using quantifier elimination

• Infset. The sets which are definable over a parameter set 𝐴 in a

model 𝑀 are the finite subsets 𝑆 of 𝐴 and their completements
𝑀 ⧵𝑆
• For a field 𝐾, the theory of infinite 𝐾-vector spaces. The sets de-
finable over a set 𝐴 are the finite subsets of the subspace spanned
by 𝐴 and their complements 𝐾 is divided by the subspace spanned
by 𝐴 and its complement.
• The theories ACF𝑝 . The definable sets of any model 𝐾 are Boolean
combinations of zero-sets

{𝑎 ∈ 𝐾 ∣ 𝑓(𝑎) = 0}

of polynomials 𝑓(𝑋) ∈ 𝐾[𝑋]. Zero-sets are finite, or if 𝑓 = 0, all

of 𝐾. 𝑓(𝑥) ≠ 0 is cofinite .

2. If 𝐾 ⊨ ACF𝑝 , for any 𝑎, 𝑏 ∈ 𝐾, the formula 𝑎𝑥1 + 𝑏 = 𝑥2 defining an

affine line 𝐴 in 𝐾 2 is strongly minimal as there is a definable bijection
between 𝐴 and 𝐾. The formula defines a map

3. For any strongly minimal formula 𝜑(𝑥1 , … , 𝑥𝑛 ), the induced theory

𝑇 ↾𝜑 is strongly minimal. Here, for any 𝔐 ⊨ 𝑇 , the induced theory
is the theory of 𝜑(𝔐) with the structure given by all intersections of
0-definable subsets of 𝑀 𝑛𝑚 with 𝜑(𝔐)𝑚 for all 𝑚 ∈ 𝜔. This theory
depends only on 𝑇 and 𝜑, not on 𝔐.

Whether p𝜑(𝑥, 𝑎) is strongly minimal depends only on the type of the

parameter tuple 𝑎 and not on the actual model: observe that 𝜑(𝑥, 𝑎) is strongly
minimal iff for all 𝐿-formulas 𝜓(𝑥, 𝑧) the set

Σ𝜓 (𝑧, 𝑎) = {∃>𝑘 𝑥(𝜑(𝑥, 𝑎)∧𝜓(𝑥, 𝑧))∧

∃>𝑘 𝑥(𝜑(𝑥, 𝑎) ∧ ¬𝜓(𝑥, 𝑧)) ∣ 𝑘 = 1, 2, … }

70
cannot be realised in any elementary extension. This means that for all
𝜓(𝑥, 𝑧) there is a bound 𝑘𝜓 s.t.

𝔐 ⊨ ∀𝑧(∃≤𝑘𝜓 𝑥(𝜑(𝑥, 𝑎) ∧ 𝜓(𝑥, 𝑧)) ∨ ∃≤𝑘𝜓 (𝜑(𝑥, 𝑎) ∧ ¬𝜓(𝑥, 𝑧)))

This is an elementary property of 𝑎, i.e., expressible by a first-order for-

mula. So it makes sense to call 𝜑(𝑥, 𝑎) a strongly minimal formula without
specifying a model
Note that in our definition, 𝜑(𝑥) a non-algebraic 𝐿(𝑀 )-formula. Thus
from our definition, 𝜑(𝑥, 𝑎) ∈ 𝐿(𝐴) is strongly minimal as long as it has such
elementary which is required for all elementary extensions of 𝔄. Guess this
is the BASE model of all elementary extensions.
One consequence is, we only need to say |𝜑(ℭ)| > 𝑘𝜑 to say |𝜑(ℭ)| is
infinite. Just like we eliminate the ∃∞

Lemma 5.44. If 𝔐 is 𝜔-saturated, or if 𝑇 eliminates the quantifier ∃∞ , any min-

imal formula is strongly minimal. If 𝑇 is totally transcendental, every infinite de-
finable subset of 𝔐𝑛 contains a minimal set 𝜑(𝔐).

Proof. If 𝔐 is 𝜔-saturated and 𝜑(𝑥, 𝑎) not strongly minimal, then for some
𝐿-formula 𝜓(𝑥, 𝑧) the set Σ𝜓 (𝑧, 𝑎) is realised in 𝔐, so 𝜑 is not minimal.
If on the other hand 𝜑(𝑥, 𝑎) is minimal and 𝑇 eliminates the quantifier
∃∞ , then for all 𝐿-formulas 𝜓(𝑥, 𝑧)

¬(∃∞ 𝑥(𝜑(𝑥, 𝑎) ∧ 𝜓(𝑥, 𝑧)) ∧ ∃∞ 𝑥(𝜑(𝑥, 𝑎) ∧ ¬𝜓(𝑥, 𝑧)))

is an elementary property of 𝑧. If we can eliminate ∃∞ , then we can express

minimality by a first-order sentence. Thus it’s strongly minimal. Guess the
power of infinitary disjunction 😅
If 𝜑0 (𝔐) does not contain a minimal set, one can construct from 𝜑0 (𝑥)
a binary tree of 𝐿(𝑀 )-formulas defining infinite subsets of 𝔐. As 𝜑0 (𝔐)
does not contain a minimal set, its not minimal?
If 𝜑(𝔐) is not minimal, then there is an 𝐿(𝑀 )-formula 𝜓(𝑥) and both
𝜑(𝔐) ∧ 𝜓(𝔐) and 𝜑(𝔐) ∧ ¬𝜓(𝔐) are infinite and not minimal. Thus we
can construct a binary tree from this.

From now on we will only consider strongly minimal formulas in one

variable.

Lemma 5.45. The formula 𝜑(𝔐) is minimal iff there is a unique non-algebraic
type 𝑝 ∈ 𝑆(𝑀 ) containing 𝜑(𝑥)

71
Proof. If 𝜑(𝔐) is minimal, then clearly

𝑝 = {𝜓 ∣ 𝜓(𝑥) ∈ 𝐿(𝑀 ) s.t. 𝜑 ∧ ¬𝜓 is algebraic}

is the unique non-algebraic type in 𝑆(𝑀 ) containing it. 𝜑(𝑥) is minimal iff
for any 𝜓(𝑥) ∈ 𝐿(𝑀 ), 𝜑 ∧ 𝜓 or 𝜑 ∧ ¬𝜓 is algebraic.. Guess algebraic requires
|𝜑(𝔐)| > 0.
if there is another non-algebraic type 𝑞 ∈ 𝑆(𝑀 ) and we take 𝛾(𝑥) ∈ 𝑞 ⧵ 𝑝.
Then 𝛾 ∧ 𝜑 ∈ 𝑞 and hence 𝜑 ∧ ¬𝛾 is algebraic. Thus 𝛾 ∈ 𝑝
If 𝜑(𝔐) is not minimal, there is some 𝐿-formula 𝜓 with both 𝜑 ∧ 𝜓 and
𝜑 ∧ ¬𝜓 non-algebraic. By Lemma 5.39, there are at least two non-algebraic
types in 𝑆(𝑀 ) containing 𝜑.

Corollary 5.46. A strongly minimal type 𝑝 ∈ 𝑆(𝐴) has a unique non-algebraic

extension to all supersets 𝐵 of 𝐴 in an elementary extensions of 𝔐. Consequently,
the type of 𝑚 realisations 𝑎1 , … , 𝑎𝑚 of 𝑝 with 𝑎𝑖 ∉ acl(𝑎1 , … , 𝑎𝑖−1 𝐴), 𝑖 = 1, … , 𝑚
is uniquely determined.
Proof. Existence of non-algebraic extensions follows from Lemma 5.39, which
also allows us to assume that 𝐵 is a model.
Uniqueness follows from Lemma 5.45 applied to any strongly minimal
formula of 𝑝. The last sentence follows by induction.
See Marker’s Lemma 6.1.6. Thus 𝑝 is strongly minimal in the problem.
First we need to prove that for any 𝑎, 𝑏 ∉ acl(𝐴), tp(𝑎/𝐴) = tp(𝑏/𝐴). As
𝑝 is strongly minimal, there is a strongly minimal formula 𝜙(𝑥) ∈ 𝑝(𝑥). For
any 𝔐 ⊨ 𝜓(𝑎), as 𝜙(𝑎) ∧ 𝜓(𝑎), 𝜙(𝔐) ∧ 𝜓(𝔐) is infinite, and thus 𝜙(𝔐) ∧
¬𝜓(𝔐) is finite. As 𝔐 ⊨ 𝜙(𝑏), we have 𝑏 ∉ 𝜙(𝔐) ∧ ¬𝜓(𝔐) and hence
𝔐 ⊨ 𝜙(𝑏) ∧ 𝜓(𝑏). Consequently, tp(𝑎/𝐴) = tp(𝑏/𝐴).
Inductive step is similar

Theorem 5.47. If 𝜑(𝑥) is strongly minimal formula in 𝔐 without parameters, the

operation
cl ∶ 𝔓(𝜑(𝔐)) → 𝔓(𝜑(𝔐))
defined by
𝑀
cl(𝐴) = acl (𝐴) ∩ 𝜑(𝔐)
is a pregeometry (𝜑(𝔐), cl)
Proof. We have to verify EXCHANGE. Prove 𝑎 ∈ cl(𝐴𝑏)⧵cl(𝐴) ⇒ 𝑏 ∈ cl(𝐴𝑎).
For notational simplicity we assume 𝐴 = ∅. Now we prove 𝑎 ∈ cl(𝑏) ⧵
cl(∅) ⇒ 𝑏 ∈ cl(𝑎). Let 𝑎 ∈ 𝜑(𝔐) be not algebraic over ∅ and 𝑏 ∈ 𝜑(𝔐) not

72
algebraic over 𝑎. (Prove by contradiction) By Corollary 5.46, all such pairs
𝑎, 𝑏 have the same type 𝑝(𝑥, 𝑦). Let 𝐴′ be an infinite set of non-algebraic
elements realising 𝜑 (which exists in an elementary extension of 𝔐) 𝑎 is
non-algebraic that realising 𝜑 iff for any 𝜓 that cofinite in 𝜑, 𝑎 ∈ ⋂ 𝜑(𝔐) ∧
𝜓(𝔐).
If 𝜑(𝔐) ∧ 𝜓(𝔐) and 𝜑(𝔐) ∧ 𝜃(𝔐) are infinite, then either (𝜑 ∧ 𝜓 ∧ 𝜃)(𝔐)
is infinite or (𝜑 ∧ ¬(𝜓 ∧ 𝜃))(𝔐) is infinite. But 𝜑 ∧ ¬(𝜓 ∧ 𝜃) = (𝜑 ∧ ¬𝜓) ∨
(𝜑 ∧ ¬𝜃), thus it’s finite and (𝜑 ∧ 𝜓 ∧ 𝜃)(𝔐) is infinite. Hence {𝜑} ∪ {𝜓 ∶
(𝜑 ∧ 𝜓)(𝔐) infinite} is finitely satisfiable and thus satisfiable.
Then we just add constants satisfying these formulas. This is an elemen-
tary extension by Tarski test. and 𝑏′ non-algebraic over 𝐴′ . 𝜑 is strongly
minimal and we can view 𝐴′ as some extensions Since all 𝑎′ ∈ 𝐴′ have
the same type 𝑝(𝑥, 𝑏′ ) over 𝑏′ , no 𝑎′ is algebraic over 𝑏′ . 𝑎′ is algebraic over
𝑏′ iff there is 𝔐 ⊨ 𝜑(𝑎′ , 𝑏′ ) s.t. 𝜑(𝔐, 𝑏′ ) is finite. But for all 𝑎′ , 𝑎″ ∈ 𝐴′ ,
tp(𝑎′ , 𝑏′ ) = tp(𝑎″ , 𝑏′ ). Thus |𝜑(𝔐, 𝑏′ )| ≥ |𝜑(𝐴′ )|. Thus also 𝑎 is not alge-
braic over 𝑏.
The same proof shows that algebraic closure defines a pregeometry on
the set of realizations of a minimal type, i.e., a non-algebraic type 𝑝 ∈ 𝑆1 (𝐴)
having a unique non-algebraic extension to all supersets 𝐵 of 𝐴 in elemen-
tary extensions of 𝔐. Here is an example to show that a minimal type need
not be strongly minimal
Let 𝑇 be the theory of 𝔐 = (𝑀 , 𝑃𝑖 )𝑖<𝜔 in which the 𝑃𝑖 form a proper
descending sequence of subsets. The type 𝑝 = {𝑥 ∈ 𝑃𝑖 ∣ 𝑖 < 𝜔} ∈ 𝑆1 (∅)
is minimal. If all 𝑃𝑖+1 are cofinite in 𝑃𝑖 , then 𝑝 does not contain a minimal
formula and is not strongly minimal
In pregeometries there is a natural notion of independence and dimen-
sion, so in light of Theorem 5.47 , we may define the following
If 𝜑(𝑥) is strongly minimal without parameters, the 𝜑-dimension of a
model 𝔐 of 𝑇 is the dimension of the pregeometry (𝜑(𝔐), cl)
dim𝜑 (𝔐)
If 𝔐 is the model of a strongly minimal theory, we just write dim(𝔐)
If 𝜑(𝑥) is defined over 𝐴0 ⊆ 𝑀 , the closure operator of the pregeometry
𝜑(𝔐𝐴0 ) is given by
𝑀
cl(𝐴) = acl (𝐴0 ∪ 𝐴) ∩ 𝜑(𝔐)
and
dim𝜑 (𝔐/𝐴0 ) ∶= dim𝜑 (𝔐𝐴0 )
is called the 𝜑-dimension of 𝔐 over 𝐴0 .

73
Lemma 5.48. Let 𝜑(𝑥) be defined over 𝐴0 and strongly minimal, and let 𝔐 and 𝔑
be models containing 𝐴0 . Then there exists an 𝐴0 -elementary map between 𝜑(𝔐)
and 𝜑(𝔑) iff 𝔐 and 𝔑 have the same 𝜑-dimension over 𝐴0

Proof. An 𝐴0 -elementary map between 𝜑(𝔐) and 𝜑(𝔑) maps bases to bases,
so one direction is clear
For the other direction we use Corollary 5.46: if 𝜑(𝔐) and 𝜑(𝔑) have
the same dimension over 𝐴0 , let 𝑈 and 𝑉 be bases of 𝜑(𝔐) and 𝜑(𝔑), re-
spectively, and let 𝑓 ∶ 𝑈 → 𝑉 be a bijection. By Corollary 5.46, 𝑓 is 𝐴0 -
elementary The are indiscernibles. and by Lemma 5.40 𝑓 extends to an
elementary bijection 𝑔 ∶ acl(𝐴0 𝑈 ) → acl(𝐴0 𝑉 ). Thus 𝑔 ↾ 𝜑(𝔐) is an 𝐴0 -
elementary map from 𝜑(𝔐) to 𝜑(𝔑)

Corollary 5.49. 1. A theory 𝑇 is strongly minimal iff over every parameter set
there is exactly one non-algebraic type

2. In models of a strongly minimal theory the algebraic closure defines a prege-

ometry

3. Bijections between independent subsets of models of a strongly minimal the-

ory are elementary. In particular, the type of 𝑛 independent elements is
uniquely determined

Proof. 1. Lemma 5.45

2. From Theorem 5.47

If 𝑇 is strongly minimal, by the preceding we have

|𝑆(𝐴)| ≤ |acl(𝐴)| + 1

1 is for the unique non-algebraic type. Every algebraic type 𝑝 is isolated

by 𝜑𝑝 (𝑥). Also 𝜑(𝔐) ⊆ acl(𝐴). For different 𝜑𝑝 and 𝜑𝑞 , 𝜑𝑝 (𝔐) ≠ 𝜑𝑞 (𝔐).
So different algebraic 𝑝 is at least realized by one unique element. Strongly
minimal theories are therefore 𝜆-stable for all 𝜆 ≥ |𝑇 | as |acl(𝐴)| ≤ max(|𝑇 |, |𝐴|).
Also there can be no binary tree of finite or cofinite sets. So by the remark
after the proof of Theorem 5.15 𝑇 is totally transcendental as we restrict 𝜑 to
one variable. . If 𝜑(𝔐) is cofinite and 𝔑 a proper elementary extension of
𝔐, then 𝜑(𝔑) is a proper extension of 𝜑(𝔐) as |𝔐 − 𝜑(𝔐)| is fixed. Thus
strongly minimal theories have no Vaughtian pairs.

74
Theorem 5.50. Let 𝑇 be strongly minimal. Models of 𝑇 are uniquely determined
by their dimensions. The set of possible dimensions is an end segment of the cardi-
nals. A model 𝔐 is 𝜔-saturated iff dim(𝔐) ≥ ℵ0 . All models are 𝜔-homogeneous

Proof. Let 𝔐0 , 𝔐1 be models of the same dimension, and let 𝐵0 , 𝐵1 be bases

for 𝔐0 and 𝔐1 , respectively. Then any bijection 𝑓 ∶ 𝐵0 → 𝐵1 is an el-
ementary map by Corollary 5.49, which extends to an isomophism of the
algebraic closure 𝔐0 and 𝔐1 by Lemma 5.40
Claim. Every infinite algebraically closed subset 𝑆 of 𝑀 is the universe
of an elementary substructure
Proof. By Theorem 2.2 it suffices to show that every consistent 𝐿(𝑆)-
formula 𝜑(𝑥) can be realised in 𝑆. If 𝜑(𝔐) is finite, all realisations are al-
gebraic over 𝑆 and belong to 𝑆. If 𝜑(𝔐) is cofinite, 𝜑(𝔐) meets all infinite
sets.
Let 𝐴 be a finite subset of 𝔐 and 𝑝 the non-algebraic type in 𝑆(𝐴). Sup-
pose there are i guess. Since all algebraic types are isolated. Thus 𝑝 is
realised in 𝔐 exactly if 𝑀 ≠ acl(𝐴) Note that for any 𝜑 ∈ 𝑝, 𝜑(𝔐) is cofi-
nite. If 𝑝 is not realised, then ⋃𝜑∈𝑝 (¬𝜑)(𝔐) = 𝔐 and every element of 𝔐
is algebraic over 𝐴. i.e., iff dim(𝔐) > dim(𝐴). Since all algebraic types
over 𝐴 are always realised in 𝔐, this shows that 𝔐 is 𝜔-saturated iff 𝔐 has
infinite dimension.
Let 𝑓 ∶ 𝐴 → 𝐵 be an elementary bijection between two finite subsets
of 𝑀 . By Lemma 5.40, 𝑓 extends to an elementary bijection between acl(𝐴)
and acl(𝐵). If 𝑎 ∈ 𝑀 ⧵ acl(𝐴), then 𝑝 = tp(𝑎/𝐴) is the unique Corollary 5.46
non-algebraic type over 𝐴 and 𝑓(𝑝) is the unique non-algebraic type over
𝐵. Since dim(𝐴) = dim(𝐵), the argument in the previous paragraph shows
that 𝑓(𝑝) is realised in 𝔐 As 𝑀 ⧵ acl(𝐴) ≠ ∅, dim(𝔐) > dim(𝐴)

Corollary 5.51. If 𝑇 is countable and strongly minimal, it is categorical in all

uncountable cardinalities

Proof. Let 𝔐1 and 𝔐2 be two models of cardinality 𝜅 > ℵ0 . Choose two

bases 𝐵1 and 𝐵2 of 𝔐1 and 𝔐2 respectively. Then 𝐵1 and 𝐵2 both have
cardinality 𝜅 as |acl(𝐴)| ≤ max(|𝑇 |, |𝐴|). Then any bijection 𝑓 ∶ 𝐵1 → 𝐵2 is
an elementary map by Corollary 5.49, which extends to an isomophism of
the algebraic closures 𝑀1 and 𝑀2 by Lemma 5.40

Exercise 5.7.1. If 𝔐 is minimal and 𝜔-saturated, then Th(𝔐) is strongly min-

imal

75
5.8 The Baldwin-Lachlan Theorem
Theorem 5.52 (Baldwin-Lachlan). Let 𝜅 be an uncountable cardinal. A count-
able theory 𝑇 is 𝜅-categorical iff 𝑇 is 𝜔-stable and has no Vaughtian pairs

Proof. If 𝑇 is categorical in some uncountable cardinal, then 𝑇 is 𝜔-stable by

Theorem 5.13 and has no Vaughtian pair by Corollary 5.34.
For the other direction we first obtain a strongly minimal formula: since
𝑇 is totally transcendental, it has a prime model 𝔐0 . (This follows from The-
orems 4.29 and 4.31 or from Theorem 5.23) Let 𝜑(𝑥) be a minimal formula
in 𝐿(𝑀0 ), which exists by Lemma 5.44. Since 𝑇 has no Vaughtian pairs, ∃∞
can be eliminated by Lemma 5.37 and hence 𝜑(𝑥) is strongly minimal by
Lemma 5.44.
Let 𝔐1 , 𝔐2 be models of cardinality 𝜅. We may assume that 𝔐0 is an
elementary submodel of both 𝔐1 and 𝔐2 as 𝔐0 is prime. Since 𝑇 has no
Vaughtian pair, 𝔐𝑖 is a minimal extension of 𝑀0 ∪ 𝜑(𝔐𝑖 ), 𝑖 = 1, 2. There-
fore 𝜑(𝔐𝑖 ) has cardinality 𝜅 𝑀0 is prime model and thus is countable since
𝑇 . and hence we conclude that dim𝜑 (𝔐1 /𝑀0 ) = 𝜅 = dim𝜑 (𝔐2 /𝑀0 ).
By Lemma 5.48 there exists an 𝑀0 -equivalent map from 𝜑(𝔐0 ) to 𝜑(𝔐1 ),
which by Lemma 5.28 extends to an isomorphism from 𝔐1 to 𝔐2

Corollary 5.53. Let 𝜅 be an uncountable cardinal. Then 𝑇 is ℵ1 -categorical iff 𝑇

is 𝜅-categorical

Corollary 5.54. Suppose 𝑇 is ℵ1 -categorical, 𝔐1 , 𝔐2 are models of 𝑇 , 𝑎𝑖 ∈ 𝔐𝑖

and 𝜑(𝑥, 𝑎𝑖 ) strongly minimal, 𝑖 = 1, 2, with tp(𝑎1 ) = tp(𝑎2 ). If 𝔐1 and 𝔐2
have the same respective 𝜑-dimension, then they are isomorphic

6 Morley Rank
6.1 Saturated models and the monster
Lemma 6.1. Let 𝑆0 ⊆ 𝑆1 ⊆ ⋯ ⊆ 𝑆𝛼 ⊆ ⋯ be an increasing chain of sets indexed
by 𝛼 < 𝜅 for some regular cardinal 𝜅. If 𝐴 ⊆ ⋃𝛼<𝜅 𝑆𝛼 and |𝐴| < 𝜅, then 𝐴 ⊆ 𝑆𝛼
for some 𝛼 < 𝜅
Or more generally if 𝜅 is not regular, 𝐴 ⊆ ⋃𝛼<cf(𝜅) 𝑆𝛼 and |𝐴| < cf(𝜅) implies
that 𝐴 ⊆ 𝑆𝛼

Proof. Define 𝑓 ∶ 𝐴 → 𝜅 by 𝑓(𝑥) = min{𝛼 ∶ 𝑥 ∈ 𝑆𝛼 }. Then |𝑓(𝐴)| ≤ |𝐴| <

cf(𝜅), so 𝛼 ∶= sup 𝑓(𝐴) < 𝜅. For any 𝑥 ∈ 𝐴, we have 𝑓(𝑥) ≤ 𝛼 and so
𝑥 ∈ 𝑆𝑓(𝑥) ⊆ 𝛼. Thus 𝐴 ⊆ 𝑆𝛼

76
Definition 6.2. A structure 𝔐 of cardinality 𝜅 ≥ 𝜔 is special if 𝔐 is the
union of an elementary chain 𝔐𝜆 where 𝜆 runs over all cardinals less than
𝜅 and each 𝔐𝜆 is 𝜆+ -saturated.
𝔐𝜆 is 𝜆+ -saturated implies that |𝔐𝜆 | ≥ 𝜆.
We call (𝔐𝜆 ) a specialising chain
Theorem 6.3. If 𝔐 is a structure and 𝜅 is a cardinal, there is a 𝜅-saturated 𝔑 ⪰ 𝔐
Proof. Build an elementary chain
𝔐 = 𝔐 0 ⪯ 𝔐1 ⪯ ⋯ ⪯ 𝔐𝛼 ⪯ ⋯
of length 𝜅+ , where
1. 𝔐𝛼+1 is an elementary extension of 𝔐𝛼 realizing every type in 𝑆1 (𝑀𝛼 )
2. If 𝛼 is a limit ordinal, then 𝔐𝛼 = ⋃𝛽<𝛼 𝔐𝛽
Let 𝔑 = ⋃𝛼<𝜅+ 𝔐𝛼 . Then 𝔑 ⪰ 𝔐. If 𝐴 ⊆ 𝑁 and |𝐴| < 𝜅, then 𝐴 ⊆ 𝑀𝛼
for some 𝛼 < 𝜅+ since 𝜅+ is regular! Any 𝑝 ∈ 𝑆1 (𝐴) extends to a 𝑝′ ∈
𝑆1 (𝑀𝛼 ) which is realized by 𝔐𝛼+1 ⊆ 𝔑

Remark. Saturated structures are special. If |𝔐| is regular, the converse is

true

Proof. From a base model, we build for each 𝜆+ -saturated model

Lemma 6.4. Let 𝜆 be an infinite cardinal ≥ |𝐿|. Then every 𝐿-structure 𝔐 of

cardinality 2𝜆 has a 𝜆+ -saturated elementary extension of cardinality 2𝜆 .
Marker’s Theorem 4.3.12

Proof. Every set of cardinality 2𝜆 has 2𝜆 many subsets of cardinality at most

𝜆. Equivalently to see the number of functions 𝜆 → 2𝜆 , which is equal to
∣(2𝜆 )𝜆 ∣ = 2𝜆 . This allows us to construct a continuous elementary chain
𝔐 = 𝔐 0 ≺ 𝔐1 ≺ ⋯ ≺ 𝔐𝛼 ≺ ⋯ (𝛼 < 𝜆+ )
of structures of cardinality 2𝜆 s.t. all 𝑝 ∈ 𝑆(𝐴), for 𝐴 ⊆ 𝑀𝛼 , |𝐴| ≤ 𝜆, are
realised in 𝔐𝛼+1 . The union of this chain has the desired properties.

Corollary 6.5. Let 𝜅 > |𝐿| be an uncountable cardinal. Assume that

𝜆 < 𝜅 ⇒ 2𝜆 ≤ 𝜅
Then every infinite 𝐿-structure 𝔐 of cardinality smaller than 𝜅 has a special ex-
tension of cardinality 𝜅.

77
 Inaccessible cardinal 𝜅
Let 𝛼 be a limit ordinal. Then for any cardinal 𝜇, 𝜅 = ℶ𝛼 (𝜇) satisfies
(6.5) and we have cf(𝜅) = cf(𝛼).

Proposition 6.6. Let 𝔐, 𝔑 be 𝐿-structures. Suppose 𝐴 ⊆ 𝐴1 ⊆ 𝑀 and 𝐵 ⊆ 𝑁

and 𝑓 ∶ 𝐴 → 𝐵 is a partial elementary map. Suppose 𝑁 is 𝜅-saturated, |𝐴| < 𝜅
and |𝐴1 | ≤ 𝜅. Then there is a partial elementary map 𝑔 ∶ 𝐴1 → 𝐵1 extending 𝑓

Proof. Let 𝐴1 = {𝑎𝛼 ∶ 𝛼 < 𝜆} where |𝐴1 | = 𝜆. Then at each step, dom(𝑓𝛼 ) ≤
|𝐴| + |𝛼| < 𝜆 ≤ 𝜅

Theorem 6.7. Two elementarily equivalent special structure of the same cardinal-
ity are isomorphic

Proof. Let 𝔄 and 𝔅 be two elementarily equivalent special structures of car-

dinality 𝜅 with specialising chains (𝔄𝜆 ) and (𝔅𝜆 ), respectively. The well-
ordering defined in the proof of Lemma A.5 can be used to find enumera-
tions (𝑎𝛼 )𝛼<𝜅 and (𝑏𝛼 )𝛼<𝜅 of 𝐴 and 𝐵 s.t. 𝑎𝛼 ∈ 𝐴|𝛼| and 𝑏𝛼 ∈ 𝐵|𝛼| . Do we
really need this enumeration We construct an increasing sequence of ele-
mentary maps 𝑓 𝛼 ∶ 𝐴𝛼 → 𝐵𝛼 s.t. for all 𝛼 which are zero or limit ordinals
we have 𝑎𝛼+𝑖 ∈ 𝐴𝛼+2𝑖 , 𝑏𝛼+1 ∈ 𝐵𝛼+2𝑖+1 , and also |𝐴𝛼 | ≤ |𝛼|, 𝐴𝛼 ⊆ 𝐴|𝛼| ,
+
|𝐵𝛼 | ≤ |𝛼|, 𝐵𝛼 ⊆ 𝐵|𝛼| . This is doable since each 𝐴|𝛼| is |𝛼| -saturated

Definition 6.8. A structure 𝔐 is

• 𝜅-universal if every structure of cardinality < 𝜅 which is elementarily

equivalent to 𝔐 can be elementarily embedded into 𝔐

• 𝜅-homogeneous if for every subset 𝐴 of 𝑀 of cardinality smaller than

𝜅 and for every 𝑎 ∈ 𝑀 , every elementary map 𝐴 → 𝑀 can be extended
to an elementary map 𝐴 ∪ {𝑎} → 𝑀

• strongly 𝜅-homogeneous if for every subset 𝐴 of 𝑀 of cardinality less

than 𝜅, every elementary map 𝐴 → 𝑀 can be extended to an automor-
phism of 𝔐.

Theorem 6.9. Special structures of cardinality 𝜅 are 𝜅+ -universal and strongly

cf(𝜅)-homogeneous

Proof. Let 𝔐 be a special structure of cardinality 𝜅.

Fix a specialising chain (𝔐𝜆 )𝜆<𝜅 . For any 𝔑 ≡ 𝔐 with |𝔑| ≤ 𝜅, ∅ is the
partial elementary map from 𝑁 to 𝑀 . For each 𝑎 ∈ 𝑁 and 𝑓𝛼 , we map to
some 𝑏 ∈ 𝔐|𝑓𝛼 | . Thus we get a elementary map from 𝔑 to 𝔐

78
 Let 𝐴 be a subset of 𝑀 of cardinality less than cf(𝜅) and let 𝑓 ∶ 𝐴 → 𝑀
an elementary map. Fix a specialising sequence (𝔐𝜆 ). For 𝜆0 sufficiently
large, 𝔐𝜆0 contains 𝐴. The sequence

(𝔐𝜆 , 𝑎)𝑎∈𝐴 if 𝜆0 ≤ 𝜆
𝑀𝜆∗ = {
(𝔐𝜆0 , 𝑎)𝑎∈𝐴 if 𝜆 < 𝜆0

is then a specialising sequence of (𝔐, 𝑎)𝑎∈𝐴 . For the same reason (𝔐, 𝑓(𝑎))𝑎∈𝐴
is special. By Theorem 6.7 these two structures are isomorphic under an au-
tomorphism of 𝔐 which extends 𝑓

Let 𝑇 be a complete theory with infinite models. For convenience, we

would like to work in a very large saturated structure, large enough so that
any model of 𝑇 can be considered as an elementary substructure. If 𝑇 is
totally transcendental, by Remark 5.2 we can choose such a monster model
as a saturated model of cardinality 𝜅 where 𝜅 is a regular cardinal greater
than all the models we ever consider otherwise. Using Exercise ?? this also
works for stable theories and regular 𝜅 with 𝜅|𝑇 | = 𝜅. For any infinite 𝜆,
𝜅 = (𝜆|𝑇 | )+ has this property.
In order to construct the monster model ℭ for an arbitrary theory 𝑇
we will work in BGC. This is a convervative extension of ZFC which adds
classes to ZFC. Then being a model of 𝑇 interpreted as being the union of
an elementary chain of (set-size) models of 𝑇 . The universe of our monster
model ℭ will be a proper class.
Theorem 6.10 (BGC). There is a class-size model ℭ of 𝑇 s.t. all types over all
subsets of 𝐶 are realised in ℭ. Moreover ℭ is uniquely determined up to isomor-
phism
Proof. Global choice allows us to construct a long continuous elementary
chain (𝑀𝛼 )𝛼∈𝑂𝑛 of models of 𝑇 s.t. all types over 𝑀𝛼 are realised in 𝑀𝛼+1 .
Let ℭ be the union of this chain. The uniqueness is proved as in Lemma
5.17.

We call ℭ the monster model of 𝑇 . Note that Global Choice implies that
ℭ can be well-ordered.
Corollary 6.11. • ℭ is 𝜅-saturated for all cardinals 𝜅

• Any model of 𝑇 is elementarily embeddable in ℭ

• Any elementary bijection between two subsets of ℭ can be extended to an

automorphism of ℭ

79
 We say that two elements are conjugate over some parameter set 𝐴 if
there is an automorphism of ℭ fixing 𝐴 elementwise and taking one to the
other. Note that 𝑎, 𝑏 ∈ ℭ are conjugate over 𝐴 iff they have the same type
over 𝐴. We call types 𝑝 ∈ 𝑆(𝐴), 𝑞 ∈ 𝑆(𝐵) conjugate over 𝐷 if there is an
automorphism 𝑓 of ℭ fixing 𝐷 and taking 𝐴 to 𝐵 and s.t. 𝑞 = {𝜑(𝑥, 𝑓(𝑎)) ∣
𝜑(𝑥, 𝑎) ∈ 𝑝}. Note that strictly speaking Aut(ℭ) is not an object in Bernays-
Gödel Set Theory but we will nevertheless use this term as a way of talking
about automorphisms
Readers who mistrust set theory can fix a regular cardinal 𝛾 bigger than
the cardinality of all models and parameter sets they want to consider. For
ℭ they may then use a special model of cardinality 𝜅 = ℶ𝛾 (ℵ0 ). This is 𝜅+ -
universal and strongly 𝛾-homogeneous
We will use the following convention throughout the rest of this book

• Any model of 𝑇 is an elementary substructure of ℭ. We identify mod-

els with their universes and denote them by 𝑀 , 𝑁 , …

• Parameter sets 𝐴, 𝐵, … are subsets of ℭ

• Formulas 𝜑(𝑥) with parameters define a subclass 𝔽 = 𝜑(ℭ) of ℭ. Two

formulas are equivalent if they define the same class

• We write ⊨ 𝜑 for ℭ ⊨ 𝜑

• A set of formulas with parameters from ℭ is consistent if it is realised

in ℭ

• If 𝜋(𝑥) and 𝜎(𝑥) are partial types we write 𝜋 ⊨ 𝜎 for 𝜋(ℭ) ⊆ 𝜎(ℭ)

• A global type is a type 𝑝 over ℭ; we denote this by 𝑝 ∈ 𝑆(ℭ)

Lemma 6.12. An elementary bijection 𝑓 ∶ 𝐴 → 𝐵 extends to an elementary

bijection between acl(𝐴) → acl(𝐵).

Proof. Extend 𝑓 to an automorphism 𝑓 ′ of ℭ. Clearly 𝑓 ′ maps acl(𝐴) to

acl(𝐵) Strongly homogeneous

This implies Lemma 5.40 and the second claim in the proof of Theorem
5.48
Note that by the remark over Lemma 5.39 for any model 𝑀 and any
𝐴 ⊆ 𝑀 the algebraic closure of 𝐴 in the sense of 𝑀 equals the algebraic
closure in the sense of ℭ.

Lemma 6.13. Let 𝔻 be a definable class and 𝐴 a set of parameters. T.F.A.E.

80
 1. 𝔻 is definable over 𝐴

2. 𝔻 is invariant under all automorphisms of ℭ which fix 𝐴 pointwise

Proof. ⇒ is easy as for any 𝐹 ∈ Aut(ℭ/𝐴) and 𝔻 = 𝜑(ℭ, 𝑎), ℭ ⊨ 𝜑(𝑠, 𝑎) iff
ℭ ⊨ 𝜑(𝐹 (𝑠), 𝑎). StackExchange
𝑥 ∈ 𝔻 ⇔⊨ 𝜑(𝑥, 𝑎) ⇔ 𝜑(𝐹 (𝑥), 𝐹 (𝑎)) ↔ 𝜑(𝐹 (𝑥), 𝑎) ⇔ 𝐹 (𝑥) ∈ 𝔻
⇐. Another proof from Chernikov. Assume that 𝔻 = 𝜑(ℭ, 𝑏) where
𝑏 ∈ ℭ, and let 𝑝(𝑦) = tp(𝑏/𝐴)
Claim 1. 𝑝(𝑦) ⊢ ∀𝑥(𝜑(𝑥, 𝑦) ↔ 𝜑(𝑥, 𝑏)), which says that for any realisa-
tions 𝑏′ , 𝜑(ℭ, 𝑏) = 𝜑(ℭ, 𝑏′ )
Indeed, let 𝑏′ ⊨ 𝑝(𝑦) be arbitrary. Then tp(𝑏/𝐴) = tp(𝑏′ /𝐴) so there is
some 𝜎 ∈ Aut(ℭ/𝐴) with 𝜎(𝑏) = 𝑏′ . Then 𝜎(𝑋) = 𝜑(ℭ, 𝑏′ ) and by assump-
tion 𝜎(𝑋) = 𝑋, thus 𝜑(ℭ, 𝑏) = 𝑋 = 𝜑(ℭ, 𝑏′ ).
There is some 𝜓(𝑦) ∈ 𝑝 (there is a finite subset of 𝑝(𝑦) that does the job
and we take the conjunction) s.t.

𝜓(𝑦) ⊨ ∀𝑥(𝜑(𝑥, 𝑦) ↔ 𝜑(𝑥, 𝑏))

Let 𝜃(𝑥) be the formula ∃𝑦(𝜓(𝑦)∧𝜑(𝑥, 𝑦)). Note that 𝜃(𝑥) is an 𝐿(𝐴)-formula,
as 𝜓(𝑦) is
Claim 2. 𝑋 = 𝜃(ℭ)
If 𝑎 ∈ 𝑋, then ⊨ 𝜑(𝑎, 𝑏), and as 𝜓(𝑦) ∈ tp(𝑏/𝐴) we have ⊨ 𝜃(𝑎). Con-
versely, if ⊨ 𝜃(𝑎), let 𝑏′ be s.t. ⊨ 𝜓(𝑏′ ) ∧ 𝜑(𝑎, 𝑏′ ). But by the choice of 𝜓 this
implies that ⊨ 𝜑(𝑎, 𝑏)
⇐ Let 𝔻 be defined by 𝜑, defined over 𝐵 ⊃ 𝐴. Consider the maps
𝜏 𝜋
ℭ−
→ 𝑆(𝐵) −
→ 𝑆(𝐴)

where 𝜏 (𝑐) = tp(𝑐/𝐵) and 𝜋 is the restriction map. Let 𝑌 be the image of
𝔻 in 𝑆(𝐴). Since 𝑌 = 𝜋[𝜑]. 𝑌 is closed. Note that 𝜏 (𝔻) = [𝜑]. 𝜏 (𝔻) =
{tp(𝑐/𝐵) ∶ ℭ ⊨ 𝜑(𝑐)} ⊆ [𝜑]. For any 𝑞(𝑥) ∈ [𝜑], as ℭ is saturated, ℭ ⊨ 𝑞(𝑑)
and 𝑑 ∈ 𝔻. Thus 𝑞 ∈ 𝜏 (𝔻). 𝜋 is continuous
Assume that 𝔻 is invariant under all automorphisms of ℭ which fix 𝐴
pointwise. Since elements which have the same type over 𝐴 are conjugate
by an automorphism of ℭ, this means that 𝔻-membership depends only on
the type over 𝐴, i.e., 𝔻 = (𝜋𝜏 )−1 (𝑌 ). For any tp(𝑐/𝐴) = tp(𝑑/𝐴) and 𝑐 ∈ 𝔻,
as 𝑐 and 𝑑 are conjugate, 𝑑 ∈ 𝔻.
For any 𝑐 ∉ 𝔻, 𝜋𝜏 (𝑐) ∈ 𝑌 iff tp(𝑐/𝐴) ∈ 𝜋[𝜑] iff there is 𝑑 ∈ 𝔻 s.t.
tp(𝑐/𝐴) = tp(𝑑/𝐴) but then 𝑐 ∈ 𝔻.
This implies that [𝜑] = 𝜋−1 (𝑌 ) 𝜏 (𝔻) = [𝜑] = 𝜏 (𝜏 −1 𝜋−1 )(𝑌 ) = 𝜋−1 (𝑌 ) ,
or 𝑆(𝐴) ⧵ 𝑌 = 𝜋[¬𝜑]; hence 𝑆(𝐴) ⧵ 𝑌 is also closed and we conclude that 𝑌

81
is clopen. By Lemma 4.6 𝑌 = [𝜓] for some 𝐿(𝐴)-formula 𝜓. This 𝜓 defines
𝔻. For any 𝑑 ∈ ℭ
⊨ 𝜓(𝑑) ⇔ tp(𝑑/𝐴) ⇔ 𝑑 ∈ 𝔻

The same proof shows that the same is true for definable relations 𝑅 ⊆
ℭ𝑛 ; namely, 𝑅 is 𝐴-definable iff it is invariant under all 𝛼 ∈ Aut(ℭ/𝐴)

Definition 6.14. 𝑋 ⊆ ℭ𝑛 is definable almost over 𝐴 if there is an 𝐴-definable

equivalence relation 𝐸 on ℭ𝑛 with finitely many classes and 𝑋 is a union of
some 𝐸-classes

A slight generalization of the previous lemma

Lemma 6.15. Let 𝑋 ⊆ ℭ𝑛 be definable. TFAE

1. 𝑋 is almost 𝐴-definable, i.e., there is an 𝐴-definable equivalence relation 𝐸

on ℭ𝑛 with finitely many classes, s.t. 𝑋 is a union of 𝐸-classes

2. The set {𝜎(𝑋) ∶ 𝜎 ∈ Aut(ℭ/𝐴)} is finite

3. The set {𝜎(𝑋) ∶ 𝜎 ∈ Aut(ℭ/𝐴)} is small

Proof. 1 → 2. Let 𝜑(𝑥1 , 𝑥2 ) ∈ 𝐿(𝐴) be the 𝐴-definable equivalence relation

𝐸, and let 𝑏1 , … , 𝑏𝑛 ∈ 𝑀 be representatives in each equivalence class so
that each class can be written as [𝑏𝑖 ] = 𝜑(ℭ, 𝑏𝑖 ). Given 𝜎 ∈ Aut(ℭ/𝐴), since
𝜑(𝑥1 , 𝑥2 ) ↔ 𝜑(𝜎(𝑥1 ), 𝜎(𝑥2 )), the image of each [𝑏𝑖 ] under 𝜎 will be

𝜎([𝑏𝑖 ]) = {𝜎(𝑥) ∶ 𝜑(𝑥, 𝑏𝑖 )} = {𝑥′ ∶ 𝜑(𝑥′ , 𝜎(𝑏𝑖 ))} = {𝑥 ∶ 𝜑(𝑥, 𝑏𝑗𝑖 )} = [𝑏𝑗1 ]

for some 𝑗𝑖 ≤ 𝑛. Now 𝑋 is a disjoint union of some [𝑏𝑖 ]’s, so 𝜎(𝑋) is a

disjoint union of some [𝑏𝑗 ]’s. Since there are only finitely many equivalence
classes, there can only be finitely many possibilities for disjoint unions of
these classes
2 → 1. Let 𝑋 = 𝜑(ℭ, 𝑏) and 𝑝(𝑦) = tp(𝑏/𝐴). Given 𝜎 ∈ Aut(ℭ/𝐴),
we have 𝜎(𝑋) = 𝜑(ℭ, 𝜎(𝑏)). Then from assumption, there must be distinct
𝑏1 , … , 𝑏𝑛 s.t.

{𝜎(𝑋) ∶ 𝜎 ∈ Aut(ℭ/𝐴)} = {𝜑(ℭ, 𝑏𝑖 ) ∶ 𝑖 ≤ 𝑛}

Now if tp(𝑏′ /𝐴) = tp(𝑏/𝐴), then strong homogeneity yields some 𝜎 ∈

Aut(ℭ/𝐴) s.t. 𝜎(𝑏) = 𝑏′ . Then the above argument again shows that 𝜑(𝑥, 𝑏′ )
defines 𝜎(𝑋) for some 𝜎 ∈ Aut(ℭ/𝐴). Thus 𝜎(𝑋) = 𝜑(ℭ, 𝑏′ ) = 𝜑(ℭ, 𝑏𝑖 ) for

82
some 𝑖 ≤ 𝑘. Therefore 𝑝(𝑦) ⊢ ⋁𝑖≤𝑘 ∀𝑥(𝜙(𝑥, 𝑦) ↔ 𝜙(𝑥, 𝑏𝑖 )). By compactness
there is some 𝜓(𝑦) ∈ 𝑝 s.t. 𝜓(𝑦) ⊢ ⋁𝑖≤𝑘 ∀(𝜙(𝑥, 𝑦) ↔ 𝜙(𝑥, 𝑏𝑖 )). Now define
𝐸(𝑥1 , 𝑥2 ) as
∀𝑦(𝜓(𝑦) → (𝜙(𝑥1 , 𝑦) ↔ 𝜙(𝑥2 , 𝑦)))
so it is 𝐴-definable. It is easy to check that 𝐸 is an equivalence relation with
finitely many classes, and that 𝑋 is a union of 𝐸-classes (𝑎1 𝐸𝑎2 iff they
agree on 𝜙(𝑥, 𝑏𝑖 ) for all 𝑖 ≤ 𝑘, and so 𝑋 = 𝜙(ℭ, 𝑏0 ) is given by the union of
all possible combinations intersected with it)
3 → 1 Assume for contradiction that

|{𝜎(𝑋) ∶ 𝜎 ∈ Aut(ℭ/𝐴)}| = 𝜆 ≥ 𝜔

we can find 𝜆-many elements (𝑏𝑖 ∶ 𝑖 < 𝜆) ⊂ ℭ to represent the distinct

images under automorphisms. Then the set

𝑞(𝑦) = 𝑝(𝑦) ∪ {¬∀𝑥(𝜑(𝑥, 𝑦) ↔ 𝜑(𝑥, 𝑏𝑖 )) ∶ 𝑖 < 𝜆}

will be finitely satisfiable. Thus 𝑞(𝑦) is realised by some 𝑏′ . But such 𝑏′

has the same type as 𝑏 over 𝐴 and so strong homogeneity yields some 𝜎 ∈
Aut(ℭ/𝐴) s.t. 𝜎(𝑏) = 𝑏′ . Applying such 𝜎 on 𝑋 gives the image 𝜑(ℭ, 𝑏′ ) =
𝜑(ℭ, 𝑏𝑖 ) for some 𝑖 < 𝜆, a contradiction

Definition 6.16. The definable closure dcl(𝐴) of 𝐴 is the set of elements 𝑐 for
which there is an 𝐿(𝐴)-formula 𝜑(𝑥) s.t. 𝑐 is the unique element satisfying
𝜑. Elements or tuples 𝑎 and 𝑏 are said to be interdefinable if 𝑎 ∈ dcl(𝑏) and
𝑏 ∈ dcl(𝑎).
Both acl(𝐴) and dcl(𝐴) are preserved by Aut(ℭ/𝐴)
Corollary 6.17. 1. 𝑎 ∈ dcl(𝐴) iff 𝑎 has only one conjugate over 𝐴.

2. 𝑎 ∈ acl(𝐴) iff 𝑎 has finitely many conjugates

Proof. 1. 𝑎 ∈ acl(𝐴) iff {𝑎} is 𝐴-definable iff {𝑎} is invariant under all
automorphisms of ℭ which fix 𝐴 pointwise

2. Follows from Remark 5.6 since the realisations of tp(𝑎/𝐴) are exactly
the conjugates of 𝑎 over 𝐴.
𝑎 ∈ acl(𝐴) iff tpℭ (𝑎/𝐴) is algebraic iff tp(𝑎/𝐴) has finitely many reali-
sations

Example 6.1. If 𝑇 is a theory of a set, then 𝑎 ∈ acl(𝐵) = dcl(𝐵) iff 𝑎 ∈ 𝐵

83
Exercise 6.1.1. Finite structures are saturated

Proof. Suppose |𝔄| = 𝑛 and for any |𝐴| < 𝑛. Suppose 𝑝(𝑥) ∈ 𝑆 𝔄 (𝐴) is not
realised, then for any 𝑎 ∈ 𝔄, there is a 𝜑𝑎 ∈ 𝑝(𝑥) s.t. 𝔄 ⊭ 𝜑𝑎 (𝑎). Hence
𝔄 ⊭ ⋀𝑎∈𝔄 𝜑𝑎 (𝑎). 𝔄 has a elementary extension 𝔅 s.t. 𝔅 ⊨ 𝑝(𝑏). Then
𝔅 ⊨ ⋀𝑎∈𝔄 𝜑𝑎 and hence 𝔄 ⊨ ∃𝑎 ⋀𝑎∈𝔄 𝜑𝑎 . A contradiction

Exercise 6.1.2. acl(𝐴) is the intersection of all models which contain 𝐴

Proof. acl(𝐴) = {𝑎 ∈ ℭ ∣ ∃𝜑 ∈ 𝐿(𝐴) s.t. |𝜑(ℭ)| < 𝜔∧ ⊨ 𝜑(𝑎)}. For any

models 𝑀 contains 𝐴, 𝜑(𝑀 ) = 𝜑(ℭ) since 𝑀 ≺ ℭ. Thus acl(𝐴) ⊆ 𝑀 for any
𝑀.
Fix any model 𝑀 which contains 𝐴. If 𝑏 is not algebraic over 𝐴, then 𝑏
has infinitely many conjugates over 𝐴.
then 𝑏 has a conjugate over 𝐴 which does not belong to 𝑀 . This implies
that 𝑀 has a conjugate 𝑀 ′ over 𝐴 which does not contain 𝑏.

Exercise 6.1.3 (Robinson’s Joint Consistency Lemma). Extend the complete

𝐿-theory 𝑇 to an 𝐿1 -theory 𝑇1 and an 𝐿2 -theory 𝑇2 s.t. 𝐿 = 𝐿1 ∩ 𝐿2 . If 𝑇1
and 𝑇2 are both consistent, show that 𝑇1 ∪ 𝑇2 is consistent.

Proof. Choose special models 𝔄𝑖 of 𝑇𝑖 of the same cardinality and observe

that a reduct of a special model is again special

Exercise 6.1.4 (Beth’s Interpolation Theorem). If ⊨ 𝜑1 → 𝜑2 for 𝐿𝑖 -sentences

𝜑𝑖 , there is an 𝐿 = 𝐿1 ∩ 𝐿2 -sentence 𝜃 s.t. ⊨ 𝜑1 → 𝜃 and ⊨ 𝜃 → 𝜑2
Exercise 6.1.5. If 𝑀 is 𝜅-saturated, then over every set of cardinality smaller
than 𝜅, every type in 𝜅 many variables is realised in 𝑀

6.2 Morley rank

Let 𝑇 be a complete (possibly uncountable) theory
We now define the Morley rank MR for formulas 𝜑(𝑥) with parameters
in the monster model.
MR 𝜑 ≥ 0 if 𝜑 is consistent
MR 𝜑 ≥ 𝛽 + 1 if there is an infinite family (𝜑𝑖 (𝑥) ∣ 𝑖 < 𝜔) of formulas
(in the same variable 𝑥) which imply 𝜑, are pairwise
inconsistent and s.t. MR 𝜑𝑖 ≥ 𝛽 for all 𝑖
MR 𝜑 ≥ 𝜆 (for a limit ordinal 𝜆) if MR 𝜑 ≥ 𝛽 for all 𝛽 < 𝜆

Remark. 𝜑𝑖 implies 𝜑 means 𝜑𝑖 (ℭ) ⊆ 𝜑(ℭ). Thus we are finding infinite

definable disjoint subsets i guess

84
Definition 6.18. To define MR 𝜑 we distinguish three cases
1. If there is no 𝛼 with MR 𝜑 ≥ 𝛼, we put MR 𝜑 = −∞

2. MR 𝜑 ≥ 𝛼 for all 𝛼, we put MR 𝜑 = ∞

3. Otherwise, by the definition of MR 𝜑 ≥ 𝜆 for limit ordinals, there is a

maximal 𝛼 with MR 𝜑 ≥ 𝛼, and we set MR 𝜑 ≥ 𝛼, and we set MR 𝜑 =
max{𝛼 ∣ MR 𝜑 ≥ 𝛼}.
Note that

MR 𝜑 = −∞ ⇔ 𝜑 is inconsistent
MR 𝜑 = 0 ⇔ 𝜑 is consistent and algebraic

If a formula has ordinal-valued Morley rank, we also say that this for-
mula has Morley rank. The Morley rank MR(𝑇 ) of 𝑇 is the Morley rank of
the formula 𝑥=𝑥. ̇ The Morley rank of a formula 𝜑(𝑥, 𝑎) only depends on
𝜑(𝑥, 𝑦) and the type of 𝑎. It follows that if a formula has Morley rank, then
+
it is less than (2|𝑇 | ) .
Remark. If 𝜑 implies 𝜓, then MR 𝜑 ≤ MR 𝜓. If 𝜑 has rank 𝛼 < ∞, then for
every 𝛽 < 𝛼 there is a formula 𝜓 which implies 𝜑 and has rank 𝛽
⋅
Example 6.2. In Infset the formula 𝑥1 = 𝑎 has Morley rank 0. It has quan-
tifier elimination. If considered as a formula in two variables, 𝜑(𝑥1 , 𝑥2 ) =
⋅
𝑥1 = 𝑎, it has Morley rank 1
The next lemma expresses the fact that the formulas of rank less than 𝛼
form an ideal in the Boolean algebra of equivalence classes of formulas
Lemma 6.19.
MR(𝜑 ∨ 𝜓) = max(MR 𝜑, MR 𝜓)
Proof. By the previous remark, we have MR(𝜑 ∨ 𝜓) ≥ max(MR 𝜑, MR 𝜓).
For the other inequality we show by induction on 𝛼 that

MR(𝜑 ∨ 𝜓) ≥ 𝛼 + 1 implies max(MR 𝜑, MR 𝜓) ≥ 𝛼 + 1

Let MR(𝜑 ∨ 𝜓) ≥ 𝛼 + 1. Then there is an infinite family of formulas (𝜑𝑖 )

that imply 𝜑 ∨ 𝜓, are pairwise inconsistent and s.t. MR 𝜑𝑖 ≥ 𝛼. By the
induction hypothesis, for each 𝑖 we have MR(𝜑𝑖 ∧ 𝜑) ≥ 𝛼 or MR(𝜑𝑖 ∧ 𝜓) ≥ 𝛼
as MR((𝜑 ∧ 𝜓) ∧ 𝜑𝑖 ) ≥ 𝛼 and ((𝜑 ∧ 𝜓) ∧ 𝜑𝑖 )(𝔐) = 𝜑𝑖 (𝔐). If the first case
holds for infinite many 𝑖, then MR 𝜑 ≥ 𝛼 + 1. Otherwise MR 𝜓 ≥ 𝛼 + 1

85
 We call 𝜑 and 𝜓 𝛼-equivalent

𝜑 ∼𝛼 𝜓

if their symmetric difference 𝜑△𝜓 has rank less than 𝛼.Then 𝛼-equivalence
is an equivalence relation. 𝜑△𝜓 = (𝜑 ∧ ¬𝜓) ∨ (¬𝜑 ∧ 𝜓) = ¬(𝜑 → 𝜓) ∨ ¬(𝜓 →
𝜑) = ¬((𝜑 → 𝜓) ∧ (𝜓 → 𝜑))
Suppose 𝜑 ∼𝛼 𝜓 and 𝜓 ∼𝛼 𝜃. Note that 𝜑△𝜃 = (𝜑△𝜓)△(𝜓△𝜃).
As MR((𝜑△𝜓) ∨ (𝜓△𝜃)) < 𝛼 and (𝜑△𝜓)△(𝜓△𝜃) ⊂ (𝜑△𝜓) ∨ (𝜓△𝜃),
MR((𝜑△𝜓)△(𝜓△𝜃)) < 𝛼 and thus MR(𝜑△𝜃) < 𝛼.
We call a formula 𝜑 𝛼-strongly minimal if it has rank 𝛼 and for any
formula 𝜓 implying 𝜑 either 𝜓 or 𝜑 ∧ ¬𝜓 has rank less than 𝛼, (equivalently,
if every 𝜓 ⊆ 𝜑 is 𝛼-equivalent to ∅ or to 𝜑). Thus we are actually talking
about for any formula 𝜓, either 𝜑 ∧ 𝜓 or 𝜑 ∧ ¬𝜓 has rank less than 𝛼, which
is natural for building a tree In particular, 𝜑 is 0-strongly minimal iff 𝜑 is
realised by a single element and 𝜑 is 1-strongly minimal iff 𝜑 is strongly
minimal

Lemma 6.20. Each formula 𝜑 of rank 𝛼 < ∞ is equivalent to a disjunction of

finitely many pairwise disjoint 𝛼-strongly minimal formulas 𝜑1 , … , 𝜑𝑑 , the 𝛼-
strongly minimal components (or just components) of 𝜑. The components are
uniquely determined up to 𝛼-equivalence

Proof. Suppose 𝜑 is a formula of rank 𝛼 without such a decomposition.

Then 𝜑 can be written as the disjoint disjunction of a formula 𝜑1 of rank
𝛼 and another formula 𝜓1 of rank 𝛼 not having such a decomposition. 𝜑 is
not 𝛼-strongly minimal and MR 𝜑 = 𝛼 implies that there is a 𝜓 implying 𝜑
s.t. MR 𝜓 = MR(𝜑 ∧ ¬𝜓) = 𝛼. As 𝜑 ↔ 𝜓 ∨ (𝜑 ∧ ¬𝜓). At least one of them
is not 𝛼-strongly minimal. If one of them is 𝛼-strongly minimal, then the
other doesn’t have such decomposition. If both of them is not 𝛼-strongly
minimal, then we can continue this process for both of them. This will end
since otherwise MR 𝜑 will be greater than 𝛼. (ANOTHER TREE!) Induc-
tively there are formulas 𝜑 = 𝜑0 , 𝜑1 , … of rank 𝛼 and 𝜓1 , 𝜓2 , … so that 𝜑𝑖 is
the disjoint union of 𝜑𝑖+1 and 𝜓𝑖+1 . But then the rank of 𝜑 would be greater
than 𝛼
Let 𝜓 be an 𝛼-strongly minimal formula implying 𝜑 and let 𝜑1 , … , 𝜑𝑑 be
the 𝛼-strongly minimal components. Then 𝜓 can be decomposed into the
formulas 𝜓 ∧ 𝜑𝑖 , one of which must be 𝛼-equivalent to 𝜓. 𝜓△(𝜓 ∧ 𝜑𝑖 ) =
(𝜑 ∨ ¬𝜓𝑖 )… So up to 𝛼-equivalence the components of 𝜑 are exactly the
𝛼-strongly minimal formulas implying 𝜑

86
 First, note that 𝜑 is 𝛼-strongly minimal if it’s minimal over ℭ
Given any 𝐿(𝐴)-formula 𝜑 and MR 𝜑 = 𝛼, if 𝜑 is not 𝛼-minimal over
𝐴, then it can be decomposed into disjoint 𝐿(𝐴)-formulas 𝜑1 and 𝜓1 with
MR 𝜑1 = MR 𝜓1 = 𝛼. If one of them is not 𝛼-minimal, then we decompose
it. This process will end since otherwise MR 𝜑 > 𝛼

Definition 6.21. For a formula 𝜑 of Morley rank 𝛼 < ∞, the Morley degree
MD(𝜑) is the number of its 𝛼-strongly minimal components

The Morley degree is not defined for inconsistent formulas or formu-

las not having Morley rank. The Morley degree of a consistent algebraic
formula is the number of its realisations. Strongly minimal formulas are ex-
actly the formulas of Morley rank and Morley degree 1. As with strongly
minimal formulas it is easy to see that Morley rank and degree are preserved
under definable bijections
Defining MD𝛼 (𝜑) as the Morley degree for formulas 𝜑 of rank 𝛼

Lemma 6.22. If 𝜑 is the disjoint union of 𝜓1 and 𝜓2 , then

MD𝛼 (𝜑) = MD𝛼 (𝜓1 ) + MD𝛼 (𝜓2 )

Theorem 6.23. The theory 𝑇 is totally transcendental iff each formula has Morley
rank

Proof. Since there are no arbitrarily large ordinal Morley ranks, each for-
mula 𝜑(𝑥) without Morley rank can be decomposed into two disjoint formu-
las without Morley rank, yielding a binary tree of consistent formulas in the
free variable 𝑥 Let 𝛽 = sup{MR 𝜓 ∶ 𝜓 implies 𝜑 and MR 𝜓 < ∞}. Then as
MR 𝜑 = ∞ ≥ 𝛽 + 2, then there is an infinite family (𝜑𝑖 (𝑥) ∣ 𝑖 < 𝜔) of formu-
las which implies 𝜑, are pairwise inconsistent and s.t. MR 𝜑𝑖 ≥ 𝛽+1 for all 𝑖.
Then MR(𝜑 ∧ ¬𝜑𝑖 ) ≥ 𝛽 + 2 ≥ 𝛽 + 1. Hence MR(𝜑 ∧ 𝜑𝑖 ) = MR(𝜑 ∧ ¬𝜑𝑖 ) = ∞.

Let (𝜑𝑠 (𝑥) ∣ 𝑠 ∈ <𝜔2 ) be a binary tree of consistent formulas. Then non
of the 𝜑𝑠 has Morley rank. Otherwise there is a 𝜑𝑠 whose ordinal rank 𝛼 is
minimal and (among the formulas of rank 𝛼) of minimal degree. Then both
𝜑𝑠0 and 𝜑𝑠1 have rank 𝛼 and therefore smaller degree than 𝜑, a contradiction

A group is said to have the descending chain condition (dcc) on defin-

able subgroups, if there is no infinite properly descending chain 𝐻0 ⊃ 𝐻1 ⊃
𝐻2 ⊃ ⋯ of definable subgroups

87
Remark. A totally transcendental has the descending chain condition on de-
finable subgroups

Proof. If 𝐻 is a definable proper subgroup of a totally transcendental group

𝐺, then either the Morley rank or the Morley degree of 𝐻 must be smaller
than that of 𝐺 since any coset of 𝐻 has the same Morley rank and degree
as 𝐻. Therefore the claim follows from the fact that the ordinals are well-
ordered

Definition 6.24. The Morley rank MR(𝑝) of a type 𝑝 is the minimal rank of
any formula in 𝑝. If MR(𝑝) is an ordinal, then its Morley degree MD(𝑝) is
the minimal degree of a formula of 𝑝 having rank 𝛼. If 𝑝 = tp(𝑎/𝐴) we also
write MR(𝑎/𝐴) and MD(𝑎/𝐴)

Algebraic types have Morley rank 0 and

MD(𝑝) = deg(𝑝)

Strongly minimal types are exactly the types of Morley rank and Morley
degree 1.
Let 𝑝 ∈ 𝑆(𝐴) have Morley rank 𝛼 and Morley degree 𝑑. Then by defi-
nition there is some 𝜑 ∈ 𝑝 of rank 𝛼 and degree 𝑑. Clearly, 𝜑 is uniquely
determined up to 𝛼-equivalence since for all 𝜓 we have MR(𝜑 ∧ ¬𝜓) < 𝛼 iff
𝜓 ∈ 𝑝. Thus 𝑝 is uniquely determined by 𝜑:

𝑝 = {𝜓(𝑥) ∣ 𝜓 ∈ 𝐿(𝐴), MR(𝜑 ∧ ¬𝜓) < 𝛼} (1)

Obviously, 𝛼-equivalent formulas determines the same type (see Lemma

5.45) i guess here is an analogy
Thus 𝜑 ∈ 𝐿(𝐴) belongs to a unique type of rank 𝛼 iff 𝜑 is 𝛼-minimal
over 𝐴; i.e., if 𝜑 has rank 𝛼 and cannot be decomposed as the union of two
𝐿(𝐴)-formulas of rank 𝛼 If 𝜑 = 𝜓 ∨ 𝜃 and MR 𝜑 = MR 𝜓 = MR 𝜃 = 𝛼, then
MD(𝜑) = MD(𝜓) + MD(𝜃).
Strongly minimality is on ℭ but minimality is on 𝐴
𝑀 ℭ
If 𝑀 is a 𝜔-saturated model containing 𝐴. Then MR (𝜑) = MR (𝜑).

Lemma 6.25. Let 𝜑 be a consistent 𝐿(𝐴)-formula

1. MR 𝜑 = max{MR(𝑝) ∣ 𝜑 ∈ 𝑝 ∈ 𝑆(𝐴)}

2. Let MR 𝜑 = 𝛼. Then

MD 𝜑 = ∑ (MD(𝑝) ∣ 𝜑 ∈ 𝑝 ∈ 𝑆(𝐴), MR(𝑝) = 𝛼)

88
Proof. 1. If MR 𝜑 = ∞, then {𝜑} ∪ {¬𝜓 ∣ 𝜓 ∈ 𝐿(𝐴), MR 𝜓 < ∞} is
consistent. Suppose {𝜑} ∪ {¬𝜓1 , … , ¬𝜓𝑛 } is inconsistent, then ⊨ 𝜑 →
¬ ⋀ ¬𝜓𝑖 , which is equivalent to ⊨ 𝜑 → ⋁ 𝜓𝑖 . But MR(⋁ 𝜓𝑖 ) = max MR(𝜓𝑖 ) <
∞ and MR(𝜑) = ∞, a contradiction Any type over 𝐴 containing this
set of formulas has rank ∞
If MR 𝜑 = 𝛼, there is a decomposition of 𝜑 in 𝐿(𝐴)-formulas 𝜑1 , … , 𝜑𝑘 ,
𝛼-minimal over 𝐴. (Note that 𝑘 is bounded by MD 𝜑). By (1), the 𝜑𝑖
determine a type 𝑝𝑖 of rank 𝛼

2. The 𝑝𝑖 are exactly the types of rank 𝛼 containing 𝜑. Furthermore

MD 𝜑𝑖 = MD(𝑝𝑖 )

Corollary 6.26. If 𝑝 ∈ 𝑆(𝐴) has Morley rank and 𝐴 ⊆ 𝐵, then

MD(𝑝) = ∑ {MD(𝑞) ∣ 𝑝 ⊆ 𝑞 ∈ 𝑆(𝐵), MR(𝑝) = MR(𝑞)}

Corollary 6.27. Let 𝑝 ∈ 𝑆(𝐴) have Morley rank and 𝐴 ⊆ 𝐵. Then 𝑝 ∈ 𝑆(𝐴) has
at least one and at most 𝑀 𝐷(𝑝) many extension to 𝐵 of the same rank

Caveat: Set-theoretically we defined the Morley rank as a function which

maps each 𝛼 to a class of formulas. In Bernays-Gödel set theory one cannot
in general define functions from ordinals to classes by a recursive scheme.
The more conscientious reader should therefore use a different definition:
for each set 𝐴 define the relation MR𝐴 (𝜑) ≥ 𝛼 using only formulas with
parameters from 𝐴, and put MR 𝜑 ≥ 𝛼 if MR𝐴 (𝜑) ≥ 𝛼 for some (suffi-
ciently large) 𝐴. The following exercise shows that if 𝜑 is defined over an
𝜔-saturated model 𝑀 , we have MR 𝜑 = MR𝑀 𝜑
Exercise 6.2.1. Let 𝜑 be a formula with parameters in the 𝜔-saturated model
𝑀 . If MR 𝜑 > 𝛼, show that there is an infinite family of formulas with
parameters in 𝑀 which each imply 𝜑, are pairwise inconsistent and have
Morley rank ≥ 𝛼.

Proof. From Marker’s Lemma 6.2.2

There are an infinite family (𝜓𝑖 (𝑥, 𝑐𝑖 ) ∣ 𝑖 ∈ 𝜔), each of which implies 𝜑
and are pairwise inconsistent. Let 𝑎 be the elements of 𝑀 occurring in 𝜑.
As 𝑀 is 𝜔-saturated and each 𝑐𝑖 is finite, for each 𝑚 ∈ 𝜔 we have

tpℭ (𝑎, 𝑐1 , … , 𝑐𝑚 ) = tp𝑀 (𝑎, 𝑑1 , … , 𝑑𝑚 ) = tpℭ (𝑎, 𝑑1 , … , 𝑑𝑚 )

89
for 𝑑𝑖 ∈ 𝑀 .
Claim If tpℭ (𝑎) = tpℭ (𝑏), then MR(𝜃(𝑥, 𝑎)) = MR(𝜃(𝑥, 𝑏)) for any 𝜃(𝑥, 𝑦)
We prove that MR(𝜃(𝑥, 𝑎)) ≥ 𝛼 iff MR(𝜃(𝑥, 𝑏)) ≥ 𝛼
If MR(𝜃(𝑥, 𝑎)) = −∞, then ⊨ ¬∃𝑥𝜃(𝑥, 𝑎) and so ⊨ ¬∃𝑏𝑥𝜃(𝑏). So MR(𝜃(𝑥, 𝑏)) =
−∞. And vice versa. So MR(𝜃(𝑥, 𝑎)) ≥ 0 iff MR(𝜃(𝑥, 𝑏)) ≥ 0.
If 𝛼 = 𝛽 + 1. We have 𝜓1 (𝑥, 𝑐1 ), 𝜓2 (𝑥, 𝑐2 ), …. As ℭ is saturated, for each
𝑚 ∈ 𝜔, we have
tp(𝑎, 𝑐1 , … , 𝑐𝑚 ) = tp(𝑏, 𝑑1 , … , 𝑑𝑚 )
Then 𝜓1 (𝑥, 𝑑1 ), 𝜓2 (𝑥, 𝑑2 ), … is what we want
Hence MR(𝜓𝑖 (𝑣, 𝑑𝑖 )) = MR(𝜓𝑖 (𝑣, 𝑐𝑖 )) ≥ 𝛼 and we are done.

Exercise 6.2.2. Let 𝜑 be a formula of Morley rank 𝛼 < ∞ and 𝜓0 , 𝜓1 , … an

infinite sequence of formulas. Assume that there is a number 𝑘 s.t. the
conjunction any 𝑘 of the 𝜓𝑖 has Morley rank smaller than 𝛼. Then MR(𝜑 ∧
𝜓𝑖 ) < 𝛼 for almost all 𝑖
𝑛
Proof. Note that ⊨ 𝜑 ↔ ⋁𝑖=1 𝜑𝑖 . If we have an upper bound for the size of
each ∣[𝜑𝑖 ]∼𝛼 ∣, then the number of formulas of Morley rank 𝛼 is finite.
𝜑 ∧ 𝜓0 , 𝜑 ∧ 𝜓1 , … is just a infinite sequence of formulas implying 𝜑. We
can consider 𝜃0 , 𝜃1 , … for simplicity.
Claim If MR 𝜃𝑖 = MR 𝜃𝑗 = 𝛼 and MR(𝜃𝑖 △𝜃𝑗 ) < 𝛼, then MR(𝜃𝑖 ∧ 𝜃𝑗 ) = 𝛼.
Or more generally, if 𝜑 = 𝜑0 ∨ 𝜑1 and 𝜑0 and 𝜑1 are inconsistent. If MR 𝜑 =
𝛼 and MR 𝜑𝑖 < 𝛼, then MR 𝜑1−𝑖 = 𝛼.
Claim If 𝜃1 ∼𝛼 𝜃2 and 𝜃1 ∧ 𝜃2 consistent, then 𝜃1 ∧ 𝜃2 ∼𝛼 𝜃1 .
Hence if 𝜃1 ∼𝛼 𝜃2 ∼𝛼 𝜃3 , then MR(𝜃1 ∧ 𝜃2 ) = 𝛼 and MR(𝜃1 ∧ 𝜃2 ∧ 𝜃3 ) = 𝛼
Let 𝐼 be set of all 𝑖 for which 𝜃𝑖 has rank 𝛼. If there is a 𝑘-elements
subsets 𝐴 of 𝐼 whose elements are all 𝛼-equivalent. Then for any 𝑖, 𝑗 ∈ 𝐴,
MR(𝜃𝑖 ∧ 𝜃𝑗 ) = 𝛼. Then MR(⋀𝑖∈𝐴 𝜃𝑖 ) = 𝛼, a contradiction. Thus we get an
upper bound
So |𝐼| ≤ (𝑘 − 1) MD 𝜑.

Exercise 6.2.3. If 𝑇 is totally transcendental, then all types over 𝜔-saturated

models have Morley degree 1

Proof. Let 𝑀 be an 𝜔-saturated model and let 𝑝 be a type over 𝑀 of Mor-

ley rank 𝛼 and degree 𝑛, witnessed by 𝜑(𝑥, 𝑚) ∈ 𝑝. If 𝑛 > 1, there is a
formula 𝜓(𝑥, 𝑏) s.t. 𝜑(𝑥, 𝑚) ∧ 𝜓(𝑥, 𝑏) and 𝜑(𝑥, 𝑚) ∧ ¬𝜓(𝑥, 𝑏) both have Mor-
ley rank 𝛼. Choose 𝑎 ∈ 𝑀 with tp(𝑎/𝑚) = tp(𝑏/𝑚). Then both formulas
𝜑(𝑥, 𝑚) ∧ 𝜓(𝑥, 𝑎) and 𝜑(𝑥, 𝑚) ∧ ¬𝜓(𝑥, 𝑎) have rank 𝛼 and degree less than
𝑑, a contradiction

90
Exercise 6.2.4. If 𝑝 is a type over acl(𝐴), then 𝑝 and 𝑝 ↾ 𝐴 have the same
Morley rank

Proof. Let 𝑎 be in acl(𝐴) and 𝑎1 , … , 𝑎𝑛 the conjugates of 𝑎 over 𝐴. Then

𝜑(𝑥, 𝑎) and 𝜑(𝑥, 𝑎1 ) ∨ ⋯ ∨ 𝜑(𝑥, 𝑎𝑛 ) have the same Morley rank
Let 𝑎 be in acl(𝐴), then there is an 𝐿(𝐴)-formula 𝜑(𝑥) s.t. ⊨ 𝜑(𝑎) and let
𝜓(𝑥) asserts that |𝜑(ℭ)| = 𝑠. Then 𝜓(𝑥) ∈ tp(𝑎/𝐴) and hence there are only
finitely many conjugates of 𝑎 over 𝐴.
Let 𝑎1 , … , 𝑎𝑛 be the conjugates of 𝑎 over 𝐴. If MR(𝜑(𝑥, 𝑎)) ≥ 0 and
⋀ 𝜑(ℭ, 𝑎𝑖 ) = ∅, then, assume that ⊨ ∀𝑥 𝜑(𝑥, 𝑎𝑖 )∧¬𝜑(𝑥, 𝑎𝑗 ). Thus ⊨ ∃𝑦∀𝑥(𝜑(𝑥, 𝑦)∧
¬𝜑(𝑥, 𝑎𝑗 )) and ∃𝑦∀𝑧(𝜑(𝑧, 𝑦)∧¬𝜑(𝑧, 𝑥)) ∈ tp(𝑎𝑖 /𝐴) = tp(𝑎𝑗 /𝐴). But ∃𝑧𝜑(𝑧, 𝑥) ∈
tp(𝑎𝑗 /𝐴), a contradiction. Thus ⋀ 𝜑(ℭ, 𝑎𝑖 ) ≠ ∅ and MR(⋀ 𝜑(𝑥, 𝑎𝑖 )) ≥ 0.
Now suppose MR(𝜑(𝑥, 𝑎)) ≥ 𝛼 + 1 and MR(⋀ 𝜑(𝑥, 𝑎𝑖 )) ≥ 𝛼.

6.3 Countable models of ℵ1 -categorical theories

Fix a countable ℵ1 -categorical theory 𝑇 . For models 𝑀 ≺ 𝑁 of 𝑇 and
𝜑(𝑥) ∈ 𝐿(𝑀 ) a strongly minimal formula, we write dim𝜑 (𝑁 /𝑀 ) for the
𝜑-dimension of 𝑁 over 𝑀

Theorem 6.28. Let 𝑇 be a countable ℵ1 -categorical theory, 𝑀 ≺ 𝑁 be models of

𝑇 , 𝐴 ⊆ 𝑀 and 𝜑(𝑥) ∈ 𝐿(𝐴) a strongly minimal formula.

1. If 𝑏1 , … , 𝑏𝑛 ∈ 𝜑(𝑁 ) are independent over 𝑀 and 𝑁 is prime over 𝑀 ∪

{𝑏1 , … , 𝑏𝑛 }, then
dim𝜑 (𝑁 /𝑀 ) = 𝑛

2. dim𝜑 (𝑁 ) = dim𝜑 (𝑀 ) + dim𝜑 (𝑁 /𝑀 )

Proof. For ease of notation we assume 𝐴 = ∅

1. Let 𝑐 ∈ 𝜑(𝑁 ). We want to show that 𝑐 is algebraic over 𝑀 ∪{𝑏1 , … , 𝑏𝑛 }.

Assume the contrary. Then 𝑝(𝑥) = tp(𝑐/𝑀 ∪ {𝑏1 , … , 𝑏𝑛 }) is strongly
minimal tp(𝑐/𝑀 ∪ {𝑏1 , … , 𝑏𝑛 }) is not algebraic and contains strongly
minimal 𝜑 and is axiomatised by

{𝜑(𝑥)} ∪ {¬𝜑𝑖 (𝑥) ∣ 𝑖 ∈ 𝐼}

where the 𝜑𝑖 range over all algebraic formulas defined over 𝑀 ∪{𝑏1 , … , 𝑏𝑛 }.
Since 𝜑(𝑀 ) is infinite, any finite subset of 𝑝(𝑥) is realised by an el-
ement of 𝑀 . Since 𝑝(𝑥) is axiomatised by {𝜑(𝑥)} ∪ {¬𝜑𝑖 (𝑥) ∣ 𝑖 ∈
𝐼}. And every disjunction of finite subset of algebraic formulas has

91
 only finitely many realisations Thus 𝑝(𝑥) is not isolated Suppose that
𝑝(𝑥) is isolated by 𝜓(𝑥). Then we have ⊨ ∀𝑥(𝜓(𝑥) → ¬𝜑𝑖 (𝑥)) and
∀𝑥(𝜑𝑖 (𝑥) → 𝜓(𝑥)). Thus ¬𝜓(ℭ) ⊇ ⋃ 𝜑𝑖 (ℭ) = acl(𝑀 ∪ {𝑏1 , … , 𝑏𝑛 })
and so 𝜓(ℭ) is finite, a contradiction. But all elements of the prime
extension 𝑁 are atomic over 𝑀 ∪ {𝑏1 , … , 𝑏𝑛 } by Corollary 5.27.
2. This follows from Remark C.7 if we can show that a basis of 𝜑(𝑁 ) over
𝜑(𝑀 ) is also a basis of 𝜑(𝑁 ) over 𝑀 . So the proof is complete once we
have established the following lemma

Lemma 6.29. Let 𝑇 be 𝜔-stable, 𝑀 ≺ 𝑁 models of 𝑇 , 𝜑(𝑥) be strongly minimal

and 𝑏𝑖 ∈ 𝜑(𝑁 ). If the 𝑏𝑖 are independent over 𝜑(𝑀 ), they are independent over 𝑀
1#+BEGINproof Assume that 𝑏1 , … , 𝑏𝑛 are algebraically independent over
𝜑(𝑀 ) but dependent over 𝑎 ∈ 𝑀 . Put 𝑏 = 𝑏1 , … , 𝑏𝑛 . An argument as in the
proof of Theorem 5.32 shows that we may assume that 𝑀 is 𝜔-saturated.
assume 𝑀 has a 𝜔-saturated elementary extension 𝑀 ′ . We want to show
that we doesn’t change the result.
If 𝑏′ are algebraically independent over 𝜑(𝑀 ′ ) ⊇ 𝜑(𝑀 ), then tp(𝑏/𝜑(𝑀 )) =
tp(𝑏′ /𝜑(𝑀 ′ )) by Corollary 5.46
And if we prove that 𝑏′ are algebraically independent over 𝑀 ′ ,
Let 𝑝 = tp(𝑏/𝑀 ). We choose a sequence 𝑏0 , 𝑏1 , … in 𝜑(𝑀 ) s.t. 𝑏2𝑖 is an
𝑛-tuple of elements algebraically independent over 𝑎𝑏0 , … , 𝑏2𝑖−1 and 𝑏2𝑖+1
realises 𝑝 ↾ 𝑎𝑏0 … 𝑏2𝑖 . By Theorem 5.50, we always can find such 𝑏2𝑖 since
Guess 1: dim(𝜑(𝑀 )) ≥ 𝜔
Let 𝑞 = tp(𝑎/𝐵) where 𝐵 = ⋃(𝑏𝑖 ). Since the sequence (𝑏𝑖 ) is indis-
cernible, 𝑏 is algebraically independent over 𝜑(𝑀 ) and hence over ⋃𝑖∈𝜔 𝑏𝑖
every permutation 𝜋 of 𝜔 defines a type 𝜋(𝑞) over 𝐵 of the form 𝜑(𝑥, 𝑏𝜋(0) , 𝑏𝜋(1) , … ).
If {𝑖 ∣ 𝜋(2𝑖) even} ≠ {𝑖 ∣ 𝜋′ (2𝑖) even}, we have 𝜋(𝑞) ≠ 𝜋′ (𝑞). Suppose for
all 𝑖, 𝑗, 𝑓(𝑖) < 𝑓(𝑗) and 𝑔′ (𝑖) < 𝑔′ (𝑗). Then as 𝑓𝑓 −1 (0) < 𝑓𝑓 −1 (1) < …,
𝑔𝑓 −1 (0) < 𝑔𝑓 −1 (1) < …. But as 𝑔 and 𝑓 is surjective onto ℕ, 𝑔𝑓 −1 (𝑖) = 𝑖 and
𝑓𝑓 −1 (𝑖) = 𝑖. Thus 𝑓 = 𝑔. So there are uncountably many types over 𝐵 and
𝑇 is not 𝜔-stable #+ENDproof
Corollary 6.30. The dimension

dim(𝑁 /𝑀 ) = dim𝜑 (𝑁 /𝑀 )

of 𝑁 over 𝑀 does not depend on 𝜑: it is the maximal length of an elementary chain

𝑀 = 𝑁0 ⪱ 𝑁1 ⪱ ⋯ ⪱ 𝑁𝑛 = 𝑁

92
between 𝑀 and 𝑁

Proof. Follows from the previous theorem since 𝑇 has no Vaughtian pairs

6.4 Computation of Morley Rank

Lemma 6.31. If 𝑏 is algebraic over 𝑎𝐴, we have MR(𝑏/𝐴) ≤ MR(𝑎/𝐴)

Proof. Let MR(𝑎/𝐴) = 𝛼. We prove RM(𝑏/𝐴) ≤ 𝛼 by induction on 𝛼. Let

𝑑 = MD(𝑏/𝐴𝑎). Choose an 𝐿(𝐴)-formula 𝜑(𝑥, 𝑦) in tp(𝑎𝑏/𝐴) s.t. MR(∃𝑦𝜑(𝑥, 𝑦)) =
𝛼 and |𝜑(𝑎′ , ℭ)| ≤ 𝑑 for all 𝑎′ .
We show that the Morley rank of 𝜒(𝑦) = ∃𝑥𝜑(𝑥, 𝑦) is bounded by 𝛼. For
this consider an infinite family 𝜒𝑖 (ℭ) of disjoint subclasses of 𝜒(ℭ) defined
over some extension 𝐴′ of 𝐴. Put 𝜓𝑖 (𝑥) = ∃𝑦(𝜑(𝑥, 𝑦) ∧ 𝜒𝑖 (𝑦)). Since any
𝑑 + 1 of the 𝜓𝑖 has empty intersection

7 Simple Theories
7.1 Dividing and forking
We work in a countable complete theory 𝑇 with infinite models
supplement

Lemma 7.1 (The Standard Lemma 5.3). Let 𝐴 be a set of parameters, ℐ an

infinite sequence of tuples and 𝐽 a linear order. Then there is a sequence of indis-
cernibles over 𝐴 of order type 𝐽 realizing EM(ℐ/𝐴)

Note that we are working in a monster model

Definition 7.2. A family (𝜑𝑖 (𝑥))𝑥∈𝐼 is 𝑘-inconsistent if for every 𝑘-element

subset 𝐾 of 𝐼 the set {𝜑𝑖 ∣ 𝑖 ∈ 𝐾} is inconsistent

Definition 7.3. We say 𝜑(𝑥, 𝑏) divides over 𝐴 (w.r.t. 𝑘) if there is a sequence

(𝑏𝑖 )𝑖<𝜔 of realisations of tp(𝑏/𝐴) s.t. (𝜑(𝑥, 𝑏𝑖 ))𝑖<𝜔 is 𝑘-inconsistent. A set of
formulas 𝜋(𝑥) divides over 𝐴 if 𝜋(𝑥) implies some 𝜑(𝑥, 𝑏) which divides over
𝐴. There is no harm in allowing 𝜑(𝑥, 𝑦) to contain parameters from 𝐴

Note that although we don’t restrict our choice of realisations, but by

𝑘-inconsistency, the number of same realisations is less than 𝑘
If 𝜑(𝑥, 𝑎) implies 𝜓(𝑥, 𝑎′ ) and 𝜓(𝑥, 𝑎′ ) divides over 𝐴, then 𝜑(𝑥, 𝑎) di-
vides over 𝐴. Thus 𝜑 divides over 𝐴 iff {𝜑} divides over 𝐴. Also a set 𝜋

93
divides over 𝐴 iff a conjunction of formulas from 𝜋 divides over 𝐴. Note
that it makes sense to say that 𝜋(𝑥) divides over 𝐴 for 𝑥 an infinite sequence
of variables as we may use dummy variables without changing the meaning
of dividing
Check this. by adding dummy variables we have

⊨ ∀𝑥(𝜑(𝑥, 𝑏, 𝑏′ ) → 𝜓(𝑥, 𝑏, 𝑏′ ))

since 𝜓(𝑥, 𝑏, 𝑏′ ) divides over 𝐴, there is a sequence (𝑏𝑖 , 𝑏𝑖′ )𝑖∈ℕ realising tp(𝑏𝑏′ /𝐴)
and s.t. {𝜓(𝑥, 𝑏𝑖 , 𝑏𝑖′ ) ∣ 𝑖 ∈ ℕ} is 𝑘-inconsistent, so {𝜑(𝑥, 𝑏𝑖 , 𝑏𝑖′ ) ∣ 𝑖 ∈ ℕ} is 𝑘-
inconsistent

Example 7.1. 1. The formula 𝑥 = 𝑏 divides over 𝐴 iff there is infinitely

many different element realising tp(𝑏/𝐴), which means 𝑏 ∉ acl(𝐴).
𝑘=2
Hence if 𝑎 ∉ acl(𝐴), then tp(𝑎/𝐴𝑎) divides over 𝐴

2. If a set 𝜋(𝑥) of formulas is consistent and defined over acl(𝐴), then it

doesn’t divide over 𝐴

3. In the theory DLO, the formula 𝑏1 < 𝑥 < 𝑏2 divides over the empty
set (for 𝑘 = 2). The type 𝑝 = {𝑥 > 𝑎 ∣ 𝑎 ∈ ℚ} does not divide over the
empty set for any 𝑘

tp(𝑏3 , 𝑏4 ) = tp(𝑏1 , 𝑏2 ) iff 𝑏3 < 𝑏4 . For QE

Lemma 7.4. The set 𝜋(𝑥, 𝑏) divides over 𝐴 iff there is a sequence (𝑏𝑖 )𝑖<𝜔 of indis-
cernibles over 𝐴 with tp(𝑏0 /𝐴) = tp(𝑏/𝐴) and ⋃𝑖<𝜔 𝜋(𝑥, 𝑏𝑖 ) inconsistent

We may replace 𝜔 by any infinite linear order. Note also that 𝑏 may be a
tuple of infinite length

Proof. If (𝑏𝑖 )𝑖<𝜔 is a sequence of indiscernibles over 𝐴 with tp(𝑏0 /𝐴) = tp(𝑏/𝐴)
(Note that ⊨ 𝜑(𝑏𝑖 ) ↔ 𝜑(𝑏𝑗 ) for any 𝑖, 𝑗 ∈ 𝜔. Thus each of (𝑏𝑖 )𝑖<𝜔 realises
tp(𝑏/𝐴)) Since ⋃𝑖<𝜔 𝜋(𝑥, 𝑏𝑖 ) is inconsistent, there is 𝑖0 , … , 𝑖𝑁 and 𝜙𝑖𝑗 (𝑥, 𝑏𝑖𝑗 ) ∈
𝜋(𝑥, 𝑏𝑖𝑗 ) s.t. {𝜙𝑖0 (𝑥, 𝑏0 ), … , 𝜙𝑖𝑁 (𝑥, 𝑏𝑁 )} is inconsistent. Take 𝜑(𝑥, 𝑏) = ⋀0≤𝑗≤𝑁 𝜙𝑖𝑗 (𝑥, 𝑏).
Then Σ(𝑥) = {𝜑(𝑥, 𝑏𝑖 ) ∣ 𝑖 < 𝜔} is inconsistent. So Σ contains some 𝑘-
element inconsistent subset. This implies that (𝜑(𝑥, 𝑏𝑖 ))𝑖<𝜔 is 𝑘-inconsistent
by indiscernibility
Assume conversely that 𝜋(𝑥, 𝑏) divides over 𝐴. Then some finite con-
junction 𝜑(𝑥, 𝑏) of formulas from 𝜋(𝑥, 𝑏) divides. Let (𝑏𝑖 )𝑖<𝜔 be a sequence
of realisations of tp(𝑏/𝐴) s.t. (𝜑(𝑥, 𝑏𝑖 ) ∣ 𝑖 < 𝜔) is 𝑘-inconsistent. Then by

94
Lemma 7.1 there is a sequence (𝑐𝑖 )𝑖<𝜔 of indiscernibles realizing EM((𝑏𝑖 )𝑖<𝜔 /𝐴).
Therefore each of (𝑐𝑖 )𝑖<𝜔 realises tp(𝑏/𝐴) and (𝜑(𝑥, 𝑐𝑖 ) ∣ 𝑖 < 𝜔) is 𝑘-inconsistent.
Thus ⋃𝑖<𝜔 𝜋(𝑥, 𝑐𝑖 ) is inconsistent

Then there is a automorphism 𝜎 ∈ Aut(ℭ/𝐴) and 𝜎(𝑏0 ) = 𝑏. Take 𝜎(𝑏)

and we get a new sequence with 𝑏0 = 𝑏

Corollary 7.5. TFAE

1. tp(𝑎/𝐴𝑏) does not divide over 𝐴

2. For any infinite sequence of 𝐴-indiscernibles ℐ containing 𝑏, there exists some

𝑎′ with tp(𝑎′ /𝐴𝑏) = tp(𝑎/𝐴𝑏) and s.t. ℐ is indiscernible over 𝐴𝑎′

3. For any infinite sequence of 𝐴-indiscernibles ℐ containing 𝑏, there exists ℐ′

with tp(ℐ′ /𝐴𝑏) = tp(ℐ/𝐴𝑏) and s.t. ℐ′ is indiscernible over 𝐴𝑎

4. For any infinite sequence of 𝐴-indiscernibles ℐ containing 𝑏, there exists

a sequence ℐ′ and some 𝑎′ with tp(ℐ′ /𝐴𝑏) = tp(ℐ/𝐴𝑏), tp(𝑎′ /𝐴𝑏) =
tp(𝑎/𝐴𝑏) and s.t. ℐ′ is indiscernible over 𝐴𝑎′

Proof. 2 → 3: Since tp(𝑎′ /𝐴𝑏) = tp(𝑎/𝐴𝑏), there is a 𝜎 ∈ Aut(ℭ/𝐴𝑏) s.t.

𝜎(𝑎) = 𝑎′ . Then 𝜎(ℐ) is what we want
3 → 2: same
By 2 and 3 we have 4
1 → 4. Let ℐ = (𝑏𝑖 )𝑖∈𝐼 be an infinite sequence of indiscernibles with 𝑏𝑖0 =
𝑏. Let 𝑝(𝑥, 𝑦) = tp(𝑎𝑏/𝐴). Then ⋃𝑖∈𝐼 𝑝(𝑥, 𝑏𝑖 ) is consistent by Lemma 7.4. As
(𝑏𝑖 )𝑖∈𝐼 contains 𝑏, tp(𝑏/𝐴) = tp(𝑏𝑖 /𝐴) for all 𝑖 ∈ 𝜔. Let 𝑎′ be a realisation.
By Lemma 7.1, there is ℐ″ = (𝑏𝑖″ )𝑖∈𝐼 indiscernible over 𝐴𝑎′ and realising
EM(ℐ/𝐴𝑎′ ). Since ⊨ 𝑝(𝑎′ , 𝑏𝑖″0 ), there is an automorphism 𝛼 ∈ Aut(ℭ/𝐴𝑎′ )
taking 𝑏𝑖″0 to 𝑏. put ℐ′ = 𝛼(ℐ″ )
2 → 1. Let 𝑝(𝑥, 𝑦) = tp(𝑎𝑏/𝐴) and let (𝑏𝑖 )𝑖<𝜔 be a sequence of indis-
cernibles with tp(𝑏0 /𝐴) = tp(𝑏/𝐴). We have to show that ⋃𝑖<𝜔 𝑝(𝑥, 𝑏𝑖 ) is
consistent. By assumption there is 𝑎′ with tp(𝑎′ /𝐴𝑏) = tp(𝑎/𝐴𝑏) s.t. ℐ is
indiscernible over 𝐴𝑎′ . As ⊨ 𝑝(𝑎′ , 𝑏), 𝑎′ is a realisation of ⋃𝑖<𝜔 𝑝(𝑥, 𝑏𝑖 )

Proposition 7.6. If tp(𝑎/𝐵) does not divide over 𝐴 ⊆ 𝐵 and tp(𝑐/𝐵𝑎) does not
divide over 𝐴𝑎, then tp(𝑎𝑐/𝐵) does not divide over 𝐴

| 𝐵∧𝑐⌣
𝑐⌣ | 𝐵𝑎 ⇒ 𝑎𝑐 ⌣
| 𝐵
𝐴 𝐴𝑎 𝐴

95
Proof. Let 𝑏 ∈ 𝐵 be a finite tuple and ℐ an infinite sequence of 𝐴-indiscernible
containing 𝑏. If tp(𝑎/𝐵) does not divide over 𝐴, there is some ℐ′ with
tp(ℐ′ /𝐴𝑏) = tp(ℐ/𝐴𝑏) and indiscernible over 𝐴𝑎. If tp(𝑐/𝐵𝑎) does not di-
vide over 𝐴𝑎, there is ℐ″ with tp(ℐ″ /𝐴𝑎𝑏) = tp(ℐ′ /𝐴𝑎𝑏) and indiscernible
over 𝐴𝑎𝑐 proving the claim

Definition 7.7. The set of formulas 𝜋(𝑥) forks over 𝐴 if 𝜋(𝑥) implies a dis-
junction ⋁𝑙<𝑑 𝜑𝑙 (𝑥) of formulas 𝜑𝑙 (𝑥) each dividing over 𝐴

Thus if 𝜋(𝑥) divides over 𝐴, it forks over 𝐴

By definition and compactness, we immediately see the following
Remark (Non-forking is closed). If 𝑝 ∈ 𝑆(𝐵) forks over 𝐴, there is some
𝜑(𝑥) ∈ 𝑝 s.t. any type in 𝑆(𝐵) containing 𝜑(𝑥) forks over 𝐴

Corollary 7.8 (Finite character). If 𝑝 ∈ 𝑆(𝐵) forks over 𝐴, there is a finite subset
𝐵0 ⊆ 𝐵 s.t. 𝑝 ↾ 𝐴𝐵0 forks over 𝐴

Lemma 7.9. If 𝜋 is finitely satisfiable in 𝐴, then 𝜋 does not fork over 𝐴

Proof. If 𝜋(𝑥) implies the disjunction ⋁𝑙<𝑑 𝜑𝑙 (𝑥, 𝑏), then some 𝜑𝑙 has a real-
isation 𝑎 in 𝐴. If the 𝑏𝑖 , 𝑖 < 𝜔, realise tp(𝑏/𝐴), then {𝜑𝑙 (𝑥, 𝑏𝑖 ) ∶ 𝑖 < 𝜔} is
realised by 𝑎. So 𝜑𝑙 does not divide over 𝐴

Lemma 7.10. Let 𝐴 ⊆ 𝐵 and let 𝜋 be a partial type over 𝐵. If 𝜋 does not fork over
𝐴, it can be extended to some 𝑝 ∈ 𝑆(𝐵) which does not fork over 𝐴

Proof. Let 𝑝(𝑥) be a maximal set of 𝐿(𝐵)-formulas containing 𝜋(𝑥) which

doesn’t fork over 𝐴. Clearly, 𝑝 is consistent Since ⊥ implies everything. Let
𝜑(𝑥) ∈ 𝐿(𝐵). If neither 𝜑, ¬𝜑 ∉ 𝑝, then both 𝑝 ∪ {𝜑} and 𝑝 ∪ {¬𝜑} fork over
𝐴, and hence 𝑝 forks over 𝐴.

Exercise 7.1.1. 1. Let 𝜑(𝑥) be a formula over 𝐴 with Morley rank and let
𝜓(𝑥) define a subclass of 𝜑(ℭ). If 𝜓 forks over 𝐴, it has smaller Morley
rank than 𝜑

2. Let 𝑝 be a type with Morley rank and 𝑞 an extension of 𝑝. If 𝑞 forks

over 𝐴, it has smaller Morley rank than 𝑝

Proof. 1. 𝜑(𝑥, 𝑎) ∈ 𝐿 and 𝜓(𝑥, 𝑏) ∈ 𝐿. We have ⊨ ∀𝑥(𝜓(𝑥, 𝑏) → 𝜑(𝑥, 𝑎)).

If 𝜓 forks over 𝐴, then there is an sequence (𝑏𝑖 )𝑖∈𝜔 each of which realiz-
ing tp(𝑏/𝐴) and {𝜓(𝑥, 𝑏𝑖 ) ∶ 𝑖 ∈ 𝜔} is 𝑘-inconsistent. Then for all 𝑖 ∈ 𝜔,
𝜓(ℭ, 𝑏𝑖 ) ⊆ 𝜑(𝑥, 𝑎). Let 𝜓𝑖 be the 𝑘-conjunction from {𝜓(𝑥, 𝑏𝑖 ) ∶ 𝑖 ∈ 𝜔}

96
Exercise 7.1.2. Let 𝑝 be a type over the model 𝑀 and 𝐴 ⊆ 𝑀 . Assume that
+
𝑀 is |𝐴| -saturated. Show that 𝑝 forks over 𝐴 iff 𝑝 divides over 𝐴

Proof. If 𝑝 forks over 𝐴, there is some 𝜑(𝑥, 𝑚) ∈ 𝑝 which implies a disjunc-

tion ⋁𝑙<𝑑 𝜑𝑙 (𝑥, 𝑏) of formulas each of which divides over 𝐴. Here 𝑏 is the
union of all the parameters 𝑏𝑙 . Choose a tuple 𝑏′ in 𝑀 which realises the
type of 𝑏 over 𝐴𝑚. The formulas 𝜑𝑙 (𝑥, 𝑏′ ) fork over 𝐴 and one of them be-
longs to 𝑝

Exercise 7.1.3. A global type which is 𝐴-invariant does not fork over 𝐴

Proof. For a global type 𝑝, 𝑝 forks over 𝐴 iff 𝑝 divides over 𝐴 by Exercise 7.1.2.
By 7.4, we For any sequence (𝑏𝑖 )𝑖<𝜔 of indiscernibles over 𝐴 with tp(𝑏𝑖 /𝐴) =
tp(𝑏/𝐴) for all 𝑖 ∈ 𝜔. Then for each 𝑖 ∈ 𝜔 there is 𝜎𝑖 ∈ Aut(ℭ/𝐴) with
𝜎(𝑏) = 𝑏𝑖 and hence 𝜙(𝑥, 𝑏𝑖 ) ∈ 𝑝. Thus {𝜙(𝑥, 𝑏𝑖 ) ∶ 𝑖 ∈ 𝜔} is consistent

7.2 Simplicity
𝑇 a countable complete theory with infinite models

Definition 7.11. 1. A formula 𝜑(𝑥, 𝑦) has the tree property w.r.t. 𝑘 is

there is a tree of parameters (𝑎𝑠 ∣ ∅ ≠ 𝑠 ∈ <𝜔𝜔 ) s.t.
<𝜔
(a) For all 𝑠 ∈ 𝜔 , (𝜑(𝑥, 𝑎𝑠𝑖 ) ∣ 𝑖 < 𝜔) is 𝑘-inconsistent
𝜔
(b) For all 𝜎 ∈ 𝜔 , {𝜑(𝑥, 𝑎𝑠 ) ∣ ∅ ≠ 𝑠 ⊆ 𝜎} is consistent

2. A theory 𝑇 is simple if there is no formula 𝜑(𝑥, 𝑦) with the tree prop-

erty

Definition 7.12. Let Δ be a finite set of formulas 𝜑(𝑥, 𝑦) without parameters.

A Δ-𝑘-dividing sequence over 𝐴 is a sequence (𝜑𝑖 (𝑥, 𝑎𝑖 ) ∣ 𝑖 < 𝛿) s.t.

1. 𝜑𝑖 (𝑥, 𝑦) ∈ Δ

2. 𝜑𝑖 (𝑥, 𝑎𝑖 ) divides over 𝐴 ∪ {𝑎𝑗 ∣ 𝑗 < 𝑖} w.r.t. 𝑘

3. {𝜑𝑖 (𝑥, 𝑎𝑖 ) ∣ 𝑖 < 𝛿} is consistent

Lemma 7.13. 1. If 𝜑 has the tree property w.r.t. 𝑘, then for every 𝐴 and 𝜇 there
exists a 𝜑-𝑘-dividing sequence over 𝐴 of length 𝜇

2. If no 𝜑 ∈ Δ has the tree property w.r.t. 𝑘, there is no infinite Δ-𝑘-dividing

sequence over ∅

97
Proof. 1. Note first that we may assume that 𝜇 is a limit ordinal. A com-
pactness argument shows that for every 𝜇 and 𝜅 there is a tree (𝑎𝑠 ∣
∅ ≠ 𝑠 ∈ <𝜇𝜅 ) s.t. all families (𝜑(𝑥, 𝑎𝑠𝑖 ) ∣ 𝑖 < 𝜅) are 𝑘-inconsistent and
for all 𝜎 ∈ 𝜇𝜅 , {𝜑(𝑥, 𝑎𝑠 ) ∣ ∅ ≠ 𝑠 ⊆ 𝜎} is consistent. If 𝜅 > 2max(|𝑇 |,|𝐴|,𝜇) ,
we recursively construct a path 𝜎 s.t. for all 𝑠 ∈ 𝜎, infinitely many 𝑎𝑠𝑖
have the same type over 𝐴 ∪ {𝑎𝑡 ∣ 𝑡 ≤ 𝑠} since 𝜅 is larger than possi-
ble numbers of types. Now (𝜑(𝑥, 𝑎𝜎↾𝑖+1 ) ∣ 𝑖 < 𝜇) is a 𝜑-𝑘-dividing
sequence over 𝐴

2. Suppose there is an infinite Δ-𝑘-dividing sequence over ∅. If 𝜑 appears

infinitely many times in this sequence, there is an infinite 𝜑-𝑘-dividing
sequence (𝜑(𝑥, 𝑎𝑖 ) ∣ 𝑖 < 𝜔). For each 𝑖 we choose a sequence (𝑎𝑛𝑖 ∣ 𝑛 <
𝜔) with tp(𝑎𝑛𝑖 /{𝑎𝑗 ∣ 𝑗 < 𝑖}) = tp(𝑎𝑖 /{𝑎𝑗 ∣ 𝑗 < 𝑖}) s.t. (𝜑(𝑥, 𝑎𝑛𝑖 ) ∣
𝑛 < 𝜔) is 𝑘-inconsistent. Then we find parameters 𝑏𝑠 showing that 𝜑
has the tree property w.r.t. 𝑘 as follows: assume 𝑠 ∈ 𝑖+1𝜔 and 𝑏 =
(𝑏𝑠↾1 , … , 𝑏𝑠↾𝑖 ) have been defined s.t. tp(𝑎0 , … , 𝑎𝑖−1 ) = tp(𝑏). Choose
𝑠(𝑖)
𝛼 ∈ Aut(ℭ) with 𝛼(𝑎0 , … , 𝑎𝑖−1 ) = 𝑏 and put 𝑏𝑠 = 𝛼(𝑎𝑖 )

It is easy to see that in simple theories for every finite set Δ and all 𝑘 there
exists a finite bound on the possible lengths of Δ-𝑘-dividing sequences

Proposition 7.14. Let 𝑇 be a complete theory. TFAE

1. 𝑇 is simple

2. (Local Character) For all 𝑝 ∈ 𝑆𝑛 (𝐵) there is some 𝐴 ⊆ 𝐵 with |𝐴| ≤ |𝑇 |

s.t. 𝑝 does not divide over 𝐴

3. There is some 𝜅 s.t. for all models 𝑀 and 𝑝 ∈ 𝑆𝑛 (𝑀 ) there is some 𝐴 ⊆ 𝑀

with |𝐴| ≤ 𝜅 s.t. 𝑝 does not divide over 𝐴

Check this. Use Skolemization to find suitable models

+
Proof. 1 → 2: If 2 doesn’t hold, there is a sequence (𝜑𝑖 (𝑥, 𝑏𝑖 ) ∣ 𝑖 < |𝑇 | ) of
formulas from 𝑝(𝑥) s.t. every 𝜑𝑖 (𝑥, 𝑏𝑖 ) divides over {𝑏𝑗 ∣ 𝑗 < 𝑖} w.r.t. 𝑘𝑖 .
There is an infinite subsequence for which all 𝜑𝑖 (𝑥, 𝑦) equal 𝜑(𝑥, 𝑦) and all
𝑘𝑖 = 𝑘 yielding a 𝜑-𝑘-dividing sequence
2 → 3: Clear
3 → 1: If 𝜑 has the tree property, there are 𝜑-𝑘-dividing sequences
(𝜑(𝑥, 𝑏𝑖 ) ∣ 𝑖 < 𝜅+ ). It is easy to construct an ascending sequence of models
𝑀𝑖 , (𝑖 < 𝜅+ ) s.t. 𝑏𝑗 ∈ 𝑀𝑖 for 𝑗 < 𝑖 and 𝜑(𝑥, 𝑏𝑖 ) divides over 𝑀𝑖 . Extend

98
the set of 𝜑(𝑥, 𝑏𝑖 ) to some type 𝑝(𝑥) ∈ 𝑆(𝑀 ) where 𝑀 = ⋃𝑖<𝜅+ 𝑀𝑖 . Then 𝑝
divides over each 𝑀𝑖
Let ℭ+ be an expansion of the monster model ℭ by Skolem functions, let
𝐿+ be the expanded language, and let 𝑇 + be Th𝐿+ (ℭ+ ). Then 𝑇 + has the
tree property, witnessed by the same 𝐿-formula 𝜑(𝑥; 𝑦) and the same tree
So there is a 𝜑-𝑘-dividing sequence (𝜑(𝑥, 𝑏𝛼 ))𝛼<𝜅+ . Let 𝑀 = ⟨{𝑏𝛼 ∣ 𝛼 <
𝜅+ }⟩. Since 𝑇 + has Skolem functions, 𝑀 ⊨ 𝑇 + . Similarly, for all 𝛽 < 𝜅+ ,
𝑀𝛽 = ⟨{𝑏𝛼 ∣ 𝛼 < 𝛽}⟩ is a model

Corollary 7.15. Let 𝑇 be simple and 𝑝 ∈ 𝑆(𝐴). Then 𝑝 does not fork over 𝐴

Proof. Suppose 𝑝 forks over 𝐴, so 𝑝 implies some disjunction ∨𝑙<𝑑 𝜑𝑙 (𝑥, 𝑏) of

formulas all of which divide over 𝐴 w.r.t. 𝑘. Put Δ = {𝜑𝑙 (𝑥, 𝑦) ∣ 𝑙 < 𝑑}
We show by induction that for all 𝑛 there is a Δ-𝑘-dividing sequence
over 𝐴 of length 𝑛. This contradicts the remark after Lemma 7.13. We will
assume also that the dividing sequence is consistent with 𝑝(𝑥)
Suppose that (𝜓𝑖 (𝑥, 𝑎𝑖 ) ∣ 𝑖 < 𝑛) is a Δ-𝑘-dividing sequence over 𝐴, con-
sistent with 𝑝(𝑥). By Exercise 7.2.1 we can replace 𝑏 with a conjugate 𝑏′ over
𝐴 s.t. (𝜓𝑖 (𝑥, 𝑎𝑖 ) ∣ 𝑖 < 𝑛) is a dividing sequence over 𝐴𝑏′ . Now one of the for-
mulas 𝜑𝑙 (𝑥, 𝑏′ ), say 𝜑0 (𝑥, 𝑏′ ), is consistent with 𝑝(𝑥) ∪ {𝜓𝑖 (𝑥, 𝑎𝑖 ) ∣ 𝑖 < 𝑛} as
⋁𝑙<𝑑 𝜑𝑙 (𝑥, 𝑏) is consistent with 𝑝. So 𝜑0 (𝑥, 𝑏′ ), 𝜓0 (𝑥, 𝑎0 ), … , 𝜓𝑛−1 (𝑥, 𝑎𝑛−1 ) is
a Δ-𝑘-dividing sequence over 𝐴 and consistent with 𝑝(𝑥)

Let 𝑝 be a type over 𝐴 and 𝑞 an extension of 𝑝. We call 𝑝 a forking

extension if 𝑞 forks over 𝐴

Corollary 7.16 (Existence). If 𝑇 is simple, every type over 𝐴 has a non-forking

extension to any 𝐵 containing 𝐴

Proof. Follows from Corollary 7.15 and Lemma 7.10

Definition 7.17. The set 𝐴 is independent from 𝐵 over 𝐶, written

| 𝐵
𝐴⌣
𝐶

if for every finite tuple 𝑎 from 𝐴, the type tp(𝑎/𝐵𝐶) does not fork over 𝐶. It
𝐶 is empty, we may omit it and write 𝐴 ⌣ | 𝐵

tp(𝑎/𝐵𝐶) forks over 𝐶 if the type of a subsequence of 𝑎 forks over 𝐶. So

this is the same as saying that tp(𝐴/𝐵𝐶) does not forks over 𝐶

Definition 7.18. Let 𝐼 be a linear order. A sequence (𝑎𝑖 )𝑖∈𝐼 is called

99
 | {𝑎𝑗 ∣ 𝑗 < 𝑖} for all 𝑖
1. independent over 𝐴 if 𝑎𝑖 ⌣𝐴

2. a Morley sequence over 𝐴 if it is independent and indiscernible over

𝐴

3. a Morley sequence in 𝑝(𝑥) over 𝐴 if it is a Morley sequence over 𝐴

consisting of realisations of 𝑝

Example 7.2. Let 𝑞 be a global type invariant over 𝐴. Then any sequence
(𝑏𝑖 )𝑖∈𝐼 where each 𝑏𝑖 realises 𝑞 ↾ 𝐴 ∪ {𝑏𝑗 ∣ 𝑗 < 𝑖} is a Morley sequence

Proof. Let us call such sequences good. Clearly a subsequence of a good

sequence is good again. So for indiscernibility it suffices to show that all
finite good sequences 𝑏0 , … , 𝑏𝑛 and 𝑏0′ , … , 𝑏𝑛′ have the same type over 𝐴.
Indeed, using induction, we may assume that 𝑏0 , … , 𝑏𝑛−1 and 𝑏0′ , … , 𝑏𝑛−1
′
′ ′
have the same type and so 𝛼(𝑏0 … 𝑏𝑛−1 ) = 𝑏0 … 𝑏𝑛−1 for some 𝛼 ∈ Aut(ℭ/𝐴).
Then

𝛼(tp(𝑏𝑛 /𝐴𝑏0 … 𝑏𝑛−1 )) = 𝛼(𝑞 ↾ 𝐴𝑏0 … 𝑏𝑛−1 ) = 𝑞 ↾ 𝐴𝑏0′ … 𝑏𝑛−1

′

= tp(𝑏𝑛′ /𝐴𝑏0′ … 𝑏𝑛−1

′
)

which proves our claim. Independence follows from Exercise 7.1.3: tp(𝑏𝑖 /𝐴𝑏0 … 𝑏𝑖−1 ) ⊆
𝑞 doesn’t fork over 𝐴

We call such a sequence (𝑏𝑖 )𝑖∈𝐼 a Morley sequence of 𝑞 over 𝐴. Note that
our proof shows that the type of a Morley sequence of 𝑞 over 𝐴 is uniquely
determined by its order type

Lemma 7.19. If (𝑎𝑖 )𝑖∈𝐼 is independent over 𝐴 and 𝐽 < 𝐾 are subsets of 𝐼, then
tp((𝑎𝑘 )𝑘∈𝐾 /𝐴{𝑎𝑗 ∣ 𝑗 ∈ 𝐽 }) does not divide over 𝐴

Stack

Proof. We may assume that 𝐾 is finite. The claim now follows from Propo-
sition 7.6 by induction on |𝐾|. If tp(𝑐/𝐵𝑎) doesn’t divide over 𝐴, then over
𝐴𝑎

Lemma 7.20 (Shelah). For all 𝐴 there is some 𝜆 s.t. for any linear order 𝐼 of car-
dinality 𝜆 and any family (𝑎𝑖 )𝑖∈𝐼 there exists an 𝐴-indiscernible sequence (𝑏𝑗 )𝑗∈𝜔
s.t. for all 𝑗1 < ⋯ < 𝑗𝑛 < 𝜔 there is a sequence 𝑖1 < ⋯ < 𝑖𝑛 in 𝐼 with
tp(𝑎𝑖1 … 𝑎𝑖𝑛 /𝐴) = tp(𝑏𝑗1 … 𝑏𝑗𝑛 /𝐴)

Proof. We only need that 𝜆 satisfies the following. Let 𝜏 = sup𝑛<𝜔 |𝑆𝑛 (𝐴)|

100
 1. cf(𝜆) > 𝜏

2. For all 𝜅 < 𝜆 and all 𝑛 < 𝜔 there is some 𝜅′ < 𝜆 with 𝜅′ → (𝜅)𝑛𝜏

By Erdős-Rado C.8 we may take 𝜆 = ℶ𝜏 + as ℶ𝑛 (𝜏 )+ < ℶ

We now construct a sequence of types 𝑝1 (𝑥1 ) ⊆ 𝑝2 (𝑥1 , 𝑥2 ) ⊆ ⋯ with
𝑝𝑛 ∈ 𝑆𝑛 (𝐴) s.t. for all 𝜅 < 𝜆 there is some 𝐼 ′ ⊆ 𝐼 with |𝐼 ′ | = 𝜅 s.t.
tp(𝑎𝑖1 , … , 𝑎𝑖𝑛 ) = 𝑝𝑛 for all 𝑖1 < ⋯ < 𝑖𝑛 from 𝐼 ′
Then we can choose the (𝑏𝑖 )𝑖<𝜔 as a realisation of ⋃𝑖<𝜔 𝑝𝑖
If 𝑝𝑛−1 has been constructed and we are given 𝜅 < 𝜆, we choose 𝜅′ < 𝜆
with 𝜅′ → (𝜅)𝑛𝜏 and some 𝐼 ′ ⊆ 𝐼 with |𝐼 ′ | = 𝜅′ s.t. tp(𝑎𝑖1 … 𝑎𝑖𝑛−1 /𝐴) =
𝑝𝑛−1 for all 𝑖1 < ⋯ < 𝑖𝑛−1 from 𝐼 ′ . Thus there are 𝐼 ″ ⊆ 𝐼 ′ and 𝑝𝑛𝜅 with
tp(𝑎𝑖1 , … , 𝑎𝑖𝑛 /𝐴) = 𝑝𝑛𝜅 for all 𝑖1 < ⋯ < 𝑖𝑛 from 𝐼 ″ . Since cf(𝜆) > 𝜏 , there is
some 𝑝𝑛 with 𝑝𝑛𝜅 = 𝑝𝑛 for cofinally many 𝜅

The existence of a Ramsey cardinal 𝜅 > 𝜏 would directly imply that any
sequence of order type 𝜅 contains a countable indiscernible subsequence

Lemma 7.21. If 𝑝 ∈ 𝑆(𝐵) does not fork over 𝐴, there is an infinite Morley sequence
in 𝑝 over 𝐴 which is indiscernible over 𝐵. In particular, if 𝑇 is simple, for every
𝑝 ∈ 𝑆(𝐴), there is an infinite Morley sequence in 𝑝 over 𝐴

Proof. Let 𝑎0 be a realisation of 𝑝. By Lemma 7.10 there is a non-forking

extension 𝑝′ of 𝑝 to 𝐵𝑎0 (or 𝐴𝐵𝑎0 ). Let 𝑎1 be a realisation of 𝑝′ . Continuing
in this way we obtain a sequence (𝑎𝑖 )𝑖<𝜆 with 𝑎𝑖 ⌣ | 𝐵(𝑎𝑗 )𝑗<𝑖 for arbitrary
𝐴
𝜆. By Lemma 7.20 we obtain a sequence of length 𝜔 with the same property
and indiscernible over 𝐵. The last sentence is immediate by Corollary 7.15

Proposition 7.22 (Kim’s lemma for simple theories). Let 𝑇 be simple and
𝜋(𝑥, 𝑦) be a partial type over 𝐴. Let (𝑏𝑖 )𝑖<𝜔 be an infinite Morley sequence over 𝐴
and ⋃𝑖<𝜔 𝜋(𝑥, 𝑏𝑖 ) is consistent. Then 𝜋(𝑥, 𝑏0 ) does not divide over 𝐴

Proof. By Lemma 7.1, for every linear order 𝐼 there is a Morley sequence
(𝑏𝑖 )𝑖∈𝐼 in tp(𝑏0 /𝐴) over 𝐴 s.t. Σ(𝑥) = ⋃𝑖∈𝐼 𝜋(𝑥, 𝑏𝑖 ) is consistent. Check this.
It’s because forking is always witnessed by a formula (“finite character”),
and whether a formula 𝜑(𝑥, 𝑐) forks over � just depends on tp(𝑐/𝐴) (“in-
variance”).
By Lemma 7.21, we have an infinite Morley sequence (𝑎𝑖 )𝑖∈𝐽 in tp(𝑏0 /𝐴)
over 𝐴.
Suppose that (𝑏𝑖 )𝑖∈𝐼 is not independent. Then there is some 𝑘 ∈ 𝐼 s.t.
tp(𝑏𝑘 /𝐴𝑏<𝑘 ) forks over 𝐴. This is witnessed by some formula 𝜑(𝑥, 𝑏𝑖1 , … , 𝑏𝑖𝑛 , 𝑐) ∈

101
tp(𝑏𝑘 /𝐴𝑏<𝑘 ) which forks over 𝐴, where 𝑐 is a tuple from 𝐴 and 𝑖1 < ⋯ <
𝑖𝑛 < 𝑘 ∈ 𝐼.
Now since (𝑏𝑖 )𝑖∈𝐼 satisfies EM((𝑎𝑗 )𝑗∈𝐽 /𝐴), pick any 𝑗1 < ⋯ < 𝑗𝑛 < 𝑘′ ∈
𝐽 and we have

1. 𝜑(𝑥, 𝑎𝑗1 , … , 𝑎𝑗𝑛 , 𝑐) forks over 𝐴 (since tp(𝑎𝑗1 , … , 𝑎𝑗𝑛 𝑐/𝐴) = tp(𝑏𝑖1 , … , 𝑏𝑖𝑛 𝑐/𝐴))
and

2. 𝜑(𝑥, 𝑎𝑗1 , … , 𝑎𝑗𝑛 , 𝑐) ∈ tp(𝑎𝑘′ /𝐴𝑎<𝑘′ ) (since tp(𝑎𝑗1 , … , 𝑎𝑗𝑛 𝑎𝑘′ 𝑐/𝐴) = tp(𝑏𝑖1 … 𝑏𝑖𝑛 𝑏𝑘 𝑐/𝐴))

so tp(𝑎𝑘′ /𝐴𝑎<𝑘′ ) forks over 𝐴, contradicting our assumption that (𝑎𝑗 )𝑗∈𝐽
+
is a Morley sequence Choose 𝐼 having the inverse order type of |𝑇 | . Let 𝑐
be a realisation of Σ. By Proposition 7.14 (2), there is some 𝑖0 s.t. tp(𝑐/𝐴 ∪
{𝑏𝑖 ∣ 𝑖 ∈ 𝐼}) does not divide over 𝐴 ∪ {𝑏𝑖 ∣ 𝑖 > 𝑖0 }. This implies that
tp(𝑐/𝐴 ∪ {𝑏𝑖 ∣ 𝑖 ≥ 𝑖0 }) does not divide over 𝐴 ∪ {𝑏𝑖 ∣ 𝑖 > 𝑖0 }. By Lemma 7.19,
tp((𝑏𝑖 ∣ 𝑖 > 𝑖0 )/𝐴𝑏𝑖0 ) does not divide over 𝐴. Hence tp(𝑐(𝑏𝑖 ∣ 𝑖 > 𝑖0 )/𝐴𝑏𝑖0 )
does not divide over 𝐴 by Proposition 7.6. This implies that 𝜋(𝑥, 𝑏𝑖0 ) does
not divide over 𝐴

Proposition 7.23. Let 𝑇 be simple. Then 𝜋(𝑥, 𝑏) divides over 𝐴 iff it forks over 𝐴

Proof. By definition, if 𝜋(𝑥, 𝑏) divides over 𝐴, it forks over 𝐴. For the con-
verse assume 𝜋(𝑥, 𝑏) does not divide over 𝐴. So if 𝜓(𝑥, 𝑏) = ⋁𝑙<𝑑 𝜑𝑙 (𝑥, 𝑏)
is implied by 𝜋(𝑥, 𝑏), it does not divide over 𝐴. Let (𝑏𝑖 )𝑖<𝜔 be a Morley se-
quence in tp(𝑏/𝐴) over 𝐴, which exists since 𝑇 is simple. So {𝜓(𝑥, 𝑏𝑖 ) ∣ 𝑖 ∈ 𝜔}
is consistent. By the pigeon-hole principle there must be some 𝑙 and some
infinite 𝐼 ⊆ 𝜔 s.t. {𝜑𝑙 (𝑥, 𝑏𝑖 ) ∣ 𝑖 ∈ 𝐼} is consistent. By Proposition 7.22,
𝜑𝑙 (𝑥, 𝑏) does not divide over 𝐴. Hence 𝜋(𝑥, 𝑏) does not fork over 𝐴

Proposition 7.24 (Symmetry). In simple theories, independence is symmetric

Proof. Assume 𝐴 ⌣ | 𝐵 and consider finite tuples 𝑎 ∈ 𝐴 and 𝑏 ∈ 𝐵. Since

𝐶
𝑎⌣ | 𝑏, Lemma 7.21 gives an infinite Morley sequence (𝑎𝑖 )𝑖∈𝜔 in tp(𝑎/𝐶𝑏)
𝐶
over 𝐶, indiscernible over 𝐶𝑏. Let 𝑝(𝑥, 𝑦) = tp(𝑎𝑏/𝐶). Then ⋃𝑖<𝜔 𝑝(𝑎𝑖 , 𝑦) is
consistent because it is realised by 𝑏. Thus, by Proposition 7.22, 𝑝(𝑎, 𝑦) does
not divide over 𝐶. This proves 𝑏 ⌣ | 𝑎. Since this holds for all 𝑎 ∈ 𝐴, 𝑏 ∈ 𝐵,
𝐶
| 𝐴 by Finite Character
it follows 𝐵 ⌣𝐶

Corollary 7.25 (Monotonicity and Transitivity). Let 𝑇 be simple, 𝐵 ⊆ 𝐶 ⊆ 𝐷.

Then we have 𝐴 ⌣| 𝐷 iff 𝐴 ⌣
| 𝐶 and 𝐴 ⌣ | 𝐷
𝐵 𝐵 𝐶

102
Proof. For transitivity, note by Proposition 7.23 we may read Proposition 7.6
after replacing finite tuples by infinite ones as

𝐴′ ⌣ | 𝐵 ⇒ 𝐶𝐴′ ⌣
| 𝐵 and 𝐶 ⌣ | 𝐵
𝐴 ′ 𝐴𝐴 𝐴

Swapping the left and the right hand sides, this 𝐴 ⊆ 𝐴′ ⊆ 𝐶

| 𝐴′ and 𝐵 ⌣
𝐵⌣ | 𝐶𝐴′
| 𝐶 ⇒ 𝐵⌣
𝐴 ′ 𝐴𝐴 𝐴

Corollary 7.26. That (𝑎𝑖 )𝑖∈𝐼 is independent over 𝐴 does not depend on the order-
ing of 𝐼

Proof. Let 𝑖 be an element of 𝐼 and 𝐽 , 𝐾 two subsets s.t. 𝐽 < 𝑖 < 𝐾.

Write 𝑎𝐽 = {𝑎𝑗 ∣ 𝑗 ∈ 𝐽 } and 𝑎𝐾 = {𝑎𝑘 ∣ 𝑘 ∈ 𝐾}. We have to show
| 𝑎𝐽 𝑎𝐾 . Now by Lemma 7.19 we have 𝑎𝐾 ⌣
that 𝑎𝑖 ⌣ | 𝑎𝐽 𝑎𝑖 . Monotonicity
𝐴 𝐴
| |
yields 𝑎𝐾 ⌣𝐴𝑎 𝑎𝑖 and by Symmetry we have 𝑎𝑖 ⌣𝐴𝑎 𝑎𝐾 . The claim follows
𝐽 𝐽
now from 𝑎𝑖 ⌣ | 𝑎𝐽 and Transitivity
𝐴

Exercise 7.2.1. If 𝜑(𝑥, 𝑏) divides over 𝐴 and 𝐴 ⊆ 𝐵, there is some 𝐴-conjugate

𝐵′ of 𝐵 s.t. 𝜑(𝑥, 𝑏) divides over 𝐵′

Proof. 𝜑(𝑥, 𝑏) divides over 𝐴 ⇔ there is 𝐴-indiscernible sequence (𝑏𝑖 )𝑖∈𝜔

with 𝑏0 = 𝑏 and {𝜑(𝑥, 𝑏𝑖 )} inconsistent. But such a sequence is also 𝐴𝑏-
indiscernible since
If (𝑏𝑖 ∶ 𝑖 < 𝜔) is indiscernible over 𝐴, there is an 𝐴-conjugate 𝐵′ of 𝐵 s.t.
(𝑏𝑖 ∣ 𝑖 < 𝜔) is indiscernible over 𝐵′
By adding a new set of constants 𝐶 with |𝐶| = |𝐵|, we only need to show
that

{𝜑(𝑏1 , 𝑐) ↔ 𝜑(𝑏2 , 𝑐) ∣ 𝑏1 , 𝑏2 ∈ 𝑏, 𝑐 ∈ 𝐶}∪{𝜑(𝑐, 𝑎) ↔ 𝜑(𝑏, 𝑎) ∣ 𝑎 ∈ 𝐴, 𝑏 ∈ 𝐵, 𝑐 ∈ 𝐶}

where 𝑏1 , 𝑏2 are in increasing order. This is easy in infinite case

For finite case, WLOG,

{𝜑(𝑏1 , 𝑥, 𝑎) ↔ 𝜑(𝑏2 , 𝑥, 𝑎) ∶ 𝑎 ∈ 𝐴, 𝑏1 , 𝑏2 ∈ 𝑏} ∪ {𝜑(𝑥, 𝑎) ↔ 𝜑(𝑥, 𝑎) ∶ 𝑎 ∈ 𝐴}

is consistent by compactness

103
8 Stable Theories
8.1 Heirs and coheirs
8.2 Stability
8.3 Definable types
8.4 Elimination of imaginaries and 𝑇 eq

A Set Theory
A.1 Sets and classes
Bernays-Gödel set theory is formulated in a two-sorted language, one type
of objects being sets and the other type of objects being classes, with the
element-relation defined between sets and sets and between sets and classes
only. We use lower case letters as variables for sets and capital letters for
classes. BG has the following axioms
1. (a) Extensionality: Sets containing the same elements are equal
(b) Empty set: The empty set exists
(c) Pairing: For any sets 𝑎 and 𝑏, {𝑎, 𝑏} is a set. This means that there
is a set which has exactly the elements 𝑎 and 𝑏
(d) Union: For every set 𝑎, the union ⋃ 𝑎 = {𝑧 ∣ ∃𝑦 𝑧 ∈ 𝑦 ∈ 𝑎} is a set
(e) Power set: For every set 𝑎, the power set 𝔓(𝑎) = {𝑦 ∣ 𝑦 ⊆ 𝑎} is a
set
(f) Infinity: There is an infinite set

2. (a) Class extensionality:

(b) Comprehension: If 𝜑(𝑥, 𝑦1 , … , 𝑦𝑚 , 𝑌1 , … , 𝑌𝑛 ) is a formula in which
only set-variables are quantified, and if 𝑏1 , … , 𝑏𝑚 , 𝐵1 , … , 𝐵𝑛 are
sets and classes, respectively, then

{𝑥 ∣ 𝜑(𝑥, 𝑏1 , … , 𝑏𝑚 , 𝐵1 , … , 𝐵𝑛 )}

is a class
(c) Replacement: If a class 𝐹 is a function, i.e., if for every set 𝑏 there
is a unique set 𝑐 = 𝐹 (𝑏) s.t. (𝑏, 𝑐) = {{𝑏}, {𝑏, 𝑐}} belongs to 𝐹 ,
then for every set 𝑎 the image {𝐹 (𝑧) ∣ 𝑧 ∈ 𝑎} is a set.

3. Regularity: Every nonempty set has an ∈-minimal element

104
 For BGC we add
4. Global Choice: There is a function 𝐹 s.t. 𝐹 (𝑎) ∈ 𝑎 for every nonempty
set 𝑎.
BGC is a conservative extension of ZFC

A.2 Cardinals
Theorem A.1 (Cantor’s Theorem). 1. If 𝜅 is infinite, then 𝜅 ⋅ 𝜅 = 𝜅
2. 2𝜅 > 𝜅
Corollary A.2. 1. If 𝜆 is infinite, then 𝜅 + 𝜆 = max(𝜅, 𝜆)
2. If 𝜅 > 0 and 𝜆 are infinite, then 𝜅 ⋅ 𝜆 = max(𝜅, 𝜆)
3. If 𝜅 is infinite, then 𝜅𝜅 = 2𝜅
Corollary A.3. The set
2<𝜔 = ⋃ 2𝑛
𝑛<𝜔

of all finite sequences of elements of a nonempty set 𝑥 has cardinality max(|𝑥|, ℵ0 )

Proof. Let 𝜅 be the cardinality of all finite sequences in 𝑥. Clearly |𝑥| ≤ 𝜅
and ℵ0 ≤ 𝜅. On the other hand
𝑛 𝑛
𝜅 = ∑|𝑥| ≤ (sup|𝑥| ) ⋅ ℵ0 = max(|𝑥|, ℵ0 )
𝑛∈ℕ 𝑛∈ℕ

because
⎧1 if |𝑥| = 1
𝑛 {
sup|𝑥| = ℵ0 if 2 ≤ |𝑥|ℵ0
⎨
𝑛∈ℕ {|𝑥| if ℵ0 ≤ |𝑥|
⎩

For every cardinal 𝜇 the beth function is defined as

⎧𝜇 if 𝛼 = 0
{
ℶ𝛼 (𝜇) = ⎨2ℶ𝛽 (𝜇) if 𝛼 = 𝛽 + 1
{sup ℶ (𝜇) if 𝛼 is a limit ordinal
⎩ 𝛽<𝛼 𝛽

For any linear linear order (𝑋, <) we can easily construct a well-ordered
cofinal subset, i.e., a subset 𝑌 s.t. for any 𝑥 ∈ 𝑋 there is some 𝑦 ∈ 𝑌 with
𝑥 ≤ 𝑦.

105
Definition A.4. The cofinality cf(𝑋) is the smallest order type of a well
ordered cofinal subset of 𝑋

cf(𝑋) is a regular cardinal where an infinite cardinal 𝜅 is regular if cf(𝜅) =

𝜅. Successor cardinals and 𝜔 are regular.

Lemma A.5 (The Gödel well-ordering). There is a bijection 𝑂𝑛 → 𝑂𝑛 × 𝑂𝑛

which induces a bijection 𝜅 → 𝜅 × 𝜅 for all infinite cardinals 𝜅

Proof. Define

(𝛼, 𝛽) < (𝛼′ , 𝛽 ′ ) ⇔ (max(𝛼, 𝛽), 𝛼, 𝛽) <lex (max(𝛼′ , 𝛽 ′ ), 𝛼′ , 𝛽 ′ )

Since this is a well-ordering, there is a unique order-preserving bijection

𝛾 ∶ 𝑂𝑛 × 𝑂𝑛 → 𝑂𝑛. We show by induction that 𝛾 maps 𝜅 × 𝜅 to 𝜅 for every
infinite cardinal 𝜅, which in turn implies 𝜅 ⋅ 𝜅 = 𝜅 Since the image of 𝜅 × 𝜅
is an initial segment, it suffices to show that the set 𝑋𝛼,𝛽 of predecessors of
(𝛼, 𝛽) has smaller cardinality than 𝜅 for every 𝛼, 𝛽 < 𝜅. We note first that
𝑋𝛼,𝛽 is contained in 𝛿 × 𝛿 with 𝛿 = max(𝛼, 𝛽) + 1. Since 𝜅 is infinite, we
have that the cardinality of 𝛿 is smaller than 𝜅. Hence by induction ∣𝑋𝛼,𝛽 ∣ ≤
|𝛿| ⋅ |𝛿| < 𝜅.

B Fields
B.1 Ordered fields
Let 𝑅 be an integral domain. A linear < ordering on 𝑅 is compatible with
the ring structure if for all 𝑥, 𝑦, 𝑧 ∈ 𝑅

𝑥<𝑦 →𝑥+𝑧 <𝑦+𝑧

𝑥 < 𝑦 ∧ 0 < 𝑧 → 𝑥𝑧 < 𝑦𝑧

A field (𝐾, <) together with a compatible ordering is an ordered field

Lemma B.1. Let 𝑅 be an integral domain and < a compatible ordering of 𝑅. Then
the ordering < can be uniquely extended to an ordering of the quotient field of 𝑅

It is easy to see that in an ordered field sums of squares can never be

negative. In particular, 1, 2, . . . are always positive and so the characteristic
of an ordered field is 0. A field K in which −1 is not a sum of squares is
called formally real.

106
C Combinatorics
C.1 Pregeometris
Definition C.1. A pregeometry (𝑋, cl) is a set 𝑋 with a closure operator
cl ∶ 𝔓(𝑋) → 𝔓(𝑋) s.t for all 𝐴 ⊆ 𝑋 and 𝑎, 𝑏 ∈ 𝑋

1. (REFLEXIVITY) 𝐴 ⊆ cl(𝐴)

2. (FINITE CHARACTER) cl(𝐴) is the union of all cl(𝐴′ ), where the 𝐴′

range over all finite subsets of 𝐴

3. (TRANSITIVITY) cl(cl(𝐴)) = cl(𝐴)

4. (EXCHANGE) 𝑎 ∈ cl(𝐴𝑏) ⧵ cl(𝐴) ⇒ 𝑏 ∈ cl(𝐴𝑎)

Remark. The following structures are pregeometries

1. A vector space 𝑉 with the linear closure operator

A pregeometry where points and the empty set are closed, i.e., where
′ ′
cl (∅) = ∅ and cl (𝑥) = {𝑥} for all 𝑥 ∈ 𝑋

is called geometry. For any pregeometry (𝑋, cl), there is an associated ge-
′ ′
ometry (𝑋 ′ , cl ) obtained by setting 𝑋 ′ = 𝑋 • / ∼ and cl (𝐴/ ∼) = cl(𝐴)• / ∼
where ∼ is the equivalence relation on 𝑋 • = 𝑋 ⧵ cl(∅) defined by cl(𝑥) =
cl(𝑦).

Definition C.2. Let (𝑋, cl) be a pregeometry. A subset 𝐴 of 𝑋 is called

1. independent if 𝑎 ∉ cl(𝐴 ⧵ {𝑎}) for all 𝑎 ∈ 𝐴

2. a generating set if 𝑋 = cl(𝐴)

3. a basis if 𝐴 is an independent generating set

Lemma C.3. Let (𝑋, cl) be a pregeometry with generating set 𝐸. Any indepen-
dent subset of 𝐸 can be extended to a basis contained in 𝐸. In particular, every
pregeometry has a basis

Proof. Let 𝐵 be an independence set. If 𝑥 ∈ 𝑋 ⧵ cl(𝐵), 𝐵 ∪ {𝑥} is again

independent. As for any 𝑏 ∈ 𝐵, 𝑏 ∉ cl(𝐵 ⧵ {𝑏}), whence 𝑏 ∉ cl(𝐵 ⧵ {𝑏} ∪ {𝑥}).
This implies that for a maximal independent subset 𝐵 of 𝐸, we have
𝐸 ⊆ cl(𝐵) and therefore 𝑋 = cl(𝐵)

107
Definition C.4. Let (𝑋, cl) be a pregeometry. Any subset 𝑆 gives rise to two
𝑆
new pregeometries, the restriction (𝑆, cl ) and the relativisation (𝑋, cl𝑆 )
where
𝑆
cl (𝐴) = cl(𝐴) ∩ 𝑆
cl𝑆 (𝐴) = cl(𝐴 ∪ 𝑆)
𝑆
Remark. Let 𝐴 be a basis of (𝑆, cl ) and 𝐵 a basis of (𝑋, cl𝑆 ). Then the (dis-
joint) union 𝐴 ∪ 𝐵 is a basis of (𝑋, cl)

Proof. Clearly 𝐴 ∪ 𝐵 is a generating set. Given cl(𝐴) = 𝑆, we show that

cl(𝐴 ∪ 𝐵) = cl(𝑆 ∪ 𝐵) = 𝑋. For this is clear since 𝑆 ∪ 𝐵 ⊆ cl(𝐴 ∪ 𝐵).
Since 𝐵 is independent over 𝑆, we have 𝑏 ∉ cl𝑆 (𝐵 ⧵ {𝑏}) = cl(𝐴 ∪ 𝐵 ⧵ {𝑏})
for all 𝑏 ∈ 𝐵. Consider an 𝑎 ∈ 𝐴. We have to show that 𝑎 ∉ cl(𝐴′ ∪ 𝐵),
where 𝐴′ = 𝐴 ⧵ {𝑎}. As 𝑎 ∉ cl(𝐴′ ), we let 𝐵′ be a maximal subset of 𝐵 with
𝑎 ∉ cl(𝐴′ ∪ 𝐵′ ). If 𝐵′ ≠ 𝐵 this would imply that 𝑎 ∈ cl(𝐴′ ∪ 𝐵′ ∪ {𝑏}) for any
𝑏 ∈ 𝐵 ⧵ 𝐵′ which would in turn imply 𝑏 ∈ cl(𝐴 ∪ 𝐵′ ), a contradiction

Lemma C.5. All bases of a pregeometry have the same cardinality

Proof. Let 𝐴 be independent and 𝐵 a generating subset of 𝑋. We show that

|𝐴| ≤ |𝐵|

Assume first that 𝐴 is infinite. Then we extend 𝐴 to a basis 𝐴′ . Choose

for every 𝑏 ∈ 𝐵 a finite subset 𝐴𝑏 of 𝐴′ with 𝑏 ∈ cl(𝐴𝑏 ) FINITE CHAR-
ACTER condition Since the union of the 𝐴𝑏 is a generating set, we have
𝐴′ = ⋃𝑏∈𝐵 𝐴𝑏 . This implies that 𝐵 is infinite and

|𝐴| ≤ |𝐴′ | ≤ |𝐵|

Now assume that 𝐴 is finite. That |𝐴| ≤ |𝐵| follows immediately from
the following exchange principle: Given any 𝑎 ∈ 𝐴 ⧵ 𝐵 there is some 𝑏 ∈ 𝐵 ⧵ 𝐴
s.t. 𝐴′ = {𝑏} ∪ 𝐴 ⧵ {𝑎} is independent. For, since 𝑎 ∈ cl(𝐵), 𝐵 cannot be
contained in cl(𝐴 ⧵ {𝑎}) 𝐴 ⧵ {𝑎} cannot be a generating set since 𝑎 ∉ cl(𝐴 ⧵
{𝑎}). Choose 𝑏 in 𝐵 but not in cl(𝐴 ⧵ {𝑎}). It follows from the exchange
property that 𝐴′ is independent For any 𝑎′ ∈ 𝐴′ , let 𝐴″ = 𝐴′ ⧵{𝑎′ }, we have,

𝑏 ∉ cl(𝐴″ 𝑎′ ) ⇒ 𝑎′ ∉ cl(𝐴″ 𝑏) ∨ 𝑎′ ∈ cl(𝐴″ )

Hence 𝑎′ ∉ cl(𝐴″ 𝑏), so 𝐴′ is independent

108
Definition C.6. The dimension dim(𝑋) of a pregeometry (𝑋, cl) is the car-
𝑆
dinality of a basis. For a subset 𝑆 of 𝑋 let dim(𝑆) be the dimension of (𝑆, cl )
and dim(𝑋/𝑆) the dimension of (𝑋, cl𝑆 )

By Remark C.1 we have

Lemma C.7. dim(𝑋) = dim(𝑆) + dim(𝑋/𝑆)

C.2 The Erdős-Rado Theorem

Theorem C.8 (Erdős-Rado). ℶ𝑛 (𝜇)+ → (𝜇+ )𝑛+1
𝜇

D Index

109
𝛼-minimal, 88 model complete, 24
𝛼-strongly minimal, 86
𝜔-homogeneous, 40 totally transcendental, 55
type, 13
prime model, 46
atomic model, 46 universal theory, 18

E TODO Don’t understand

Companion to this book Chernikov’s Lecture notes on stability theory
Lemma 3.22
Exercise 3.2.2
theorem 4.11 need to enhance my TOPOLOGY and ALGEBRA!!!

5.1 5.2 5.5:done 5.35 1 3

5.7:done 5.47 6.1:done 6.2 6.2:done 6.29
6.1 7.1

110