Simpson
Math 563: Model Theory
Stephen G. Simpson
May 2, 1998
Department of Mathematics
The Pennsylvania State University
University Park, State College PA 16802
simpson@math.psu.edu
www.math.psu.edu/simpson/courses/math563/
Note: Chapters 12 and 13 are not nished.
2
Contents
1 Sentences and models 7
1.1 Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Truth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5 Models and theories . . . . . . . . . . . . . . . . . . . . . . . . 10
2 Complete theories 13
2.1 Denitions and examples . . . . . . . . . . . . . . . . . . . . . 13
2.2 Vaughts test . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Applications of Vaughts test . . . . . . . . . . . . . . . . . . . 16
3 The compactness theorem 21
3.1 Proof of the compactness theorem . . . . . . . . . . . . . . . . 21
3.2 Some applications to eld theory . . . . . . . . . . . . . . . . 23
3.3 The LowenheimSkolemTarski theorem . . . . . . . . . . . . . 24
4 Decidability 27
4.1 Recursively axiomatizable theories . . . . . . . . . . . . . . . . 27
4.2 Decidable theories . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.3 Decidable models . . . . . . . . . . . . . . . . . . . . . . . . . 31
5 Elementary extensions 35
5.1 Denition and examples . . . . . . . . . . . . . . . . . . . . . 35
5.2 Existence of elementary extensions . . . . . . . . . . . . . . . 38
5.3 Elementary monomorphisms . . . . . . . . . . . . . . . . . . . 40
3
4 CONTENTS
6 Algebraically closed elds 43
6.1 Simple eld extensions . . . . . . . . . . . . . . . . . . . . . . 43
6.2 Algebraic closure . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.3 Completeness and model completeness . . . . . . . . . . . . . 49
6.4 Hilberts Nullstellensatz . . . . . . . . . . . . . . . . . . . . . 52
7 Saturated models 55
7.1 Element types . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
7.2 Saturated models . . . . . . . . . . . . . . . . . . . . . . . . . 58
7.3 Existence of saturated models . . . . . . . . . . . . . . . . . . 60
7.4 Preservation theorems . . . . . . . . . . . . . . . . . . . . . . 63
8 Elimination of quantiers 71
8.1 The model completion of a theory . . . . . . . . . . . . . . . . 71
8.2 Substructure completeness . . . . . . . . . . . . . . . . . . . . 73
8.3 The role of simple extensions . . . . . . . . . . . . . . . . . . . 76
9 Real closed ordered elds 79
9.1 Ordered elds . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
9.2 Uniqueness of real closure . . . . . . . . . . . . . . . . . . . . 83
9.3 Quantier elimination for RCOF . . . . . . . . . . . . . . . . . 86
9.4 The solution of Hilberts 17th problem . . . . . . . . . . . . . 89
10 Prime models (countable case) 93
10.1 The omitting types theorem . . . . . . . . . . . . . . . . . . . 93
10.2 Prime models . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
10.3 The number of countable models . . . . . . . . . . . . . . . . 101
10.4 Decidable prime models . . . . . . . . . . . . . . . . . . . . . 104
11 Dierentially closed elds of characteristic 0 109
11.1 Simple extensions . . . . . . . . . . . . . . . . . . . . . . . . . 109
11.2 Dierentially closed elds . . . . . . . . . . . . . . . . . . . . 115
11.3 Dierential closure (countable case) . . . . . . . . . . . . . . . 117
11.4 Ritts Nullstellensatz . . . . . . . . . . . . . . . . . . . . . . . 120
12 Totally transcendental theories 125
12.1 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
12.2 Rank of an element type . . . . . . . . . . . . . . . . . . . . . 125
12.3 Indiscernibles . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
CONTENTS 5
12.4 Existence of saturated models . . . . . . . . . . . . . . . . . . 125
13 Prime models (uncountable case) 127
13.1 Strongly atomic models . . . . . . . . . . . . . . . . . . . . . . 127
13.2 Normal sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
13.3 Uniqueness and characterization of prime models . . . . . . . 127
6 CONTENTS
Chapter 1
Sentences and models
1.1 Symbols
1.
We assume the availability of a large supply of nonlogical symbols of the
following kinds:
1. nary relation symbols R( , . . . , ), n 1;
2. nary operation symbols o( , . . . , ), n 1;
3. constant symbols c.
These collections of symbols are assumed to be disjoint.
2.
We make use of the following logical symbols:
1. propositional connectives (negation), , (conjunction, disjunction),
, (implication, biimplication);
2. quantiers , (universal, existential);
3. equality =;
4. variables v
0
, v
1
, . . . , v
n
, . . . .
Note that = is a logical symbol although syntactically it behaves as a
binary relation symbol.
7
8 CHAPTER 1. SENTENCES AND MODELS
1.2 Formulas
1.
The notion of a term is dened inductively as follows. A constant symbol
is a term. A variable is a term. If t
1
, . . . , t
n
are terms and o is an nary
operation symbol, then o(t
1
, . . . , t
n
) is a term.
2.
The notion of atomic formula is dened as follows. If t
1
and t
2
are terms,
then t
1
= t
2
is an atomic formula. If t
1
, . . . , t
n
are terms and R is an nary
relation symbol, then R(t
1
, . . . , t
n
) is an atomic formula.
3.
The notion of a formula is dened inductively as follows. An atomic formula
is a formula. If and are formulas then so are , , , ,
. If is a formula and v is a variable, then v and v are formulas.
We assume familiarity with the concept of a free variable, i.e. one not
bound by a quantier. We assume unique readability of formulas.
4.
If S is a set of formulas and/or terms, the signature of S is the set of all
nonlogical symbols occurring in it. This is sometimes called in the literature
the similarity type of S. Note that = never bolongs to the signature since it
is a logical symbol. We write sig(S) = signature of S.
5.
A sentence is a formula with no free variables.
Examples: The formula xy(x+y = 0) is a sentence. Here + is a binary
operation symbol, 0 is a constant symbol, and = is a logical symbol. The
formula x + y = y + x is not a sentence.
If we write x, y, . . . in the same formula, we tacitly assume that x, y, . . .
are distinct variables.
1.3. STRUCTURES 9
Examples, continued: The formula x(y(y y = x) z(z z = x)) is a
sentence. It is logically equivalent to the sentence xy(y y = x y y =
x) but these two sentences are not identical. We asume that the student
has some previous acquaintance with the syntactical and semantical notions
of logical equivalence. These notions will be dened later.
1.3 Structures
1.
A structure is an ordered pair / = ([/[, ) where [/[ is a nonempty set,
called the universe of /, and is a function whose domain is a set of non
logical symbols. The domain of is called the signature of /. To each nary
relation symbol R sig(/) we assume that assigns an nary relation
R [/[
n
= [/[ [/[
. .
n times
.
To each nary operation symbol o sig(/) we assume that assigns an
nary operation
o : [/[
n
[/[ .
To each constant symbol c sig(/) we assume that assigns an individual
constant c [/[.
Example: the structure of the reals
1 = ([1[, +, , , 0, 1, <)
where [1[ = R. We cannot include because it is not an operation on [1[
(because not everywhere dened). Here the universe is [1[ = R = (, );
+, are binary operations; is a unary operation; 0, 1 are constants; < is a
binary relation.
10 CHAPTER 1. SENTENCES AND MODELS
1.4 Truth
1.
Given a structure / and a sentence such that sig() sig(/), we assume
known the meaning of
/ [= (/ satises , is true in /) .
For example, 1 [= xy(y y = x y y = x) expresses the fact that
every real number or its negative is a square. Note that the structure
Z = (Z, +, , , 0, 1, <) ,
where [Z[ = Z = . . . , 2, 1, 0, 1, 2, . . . , has the same signature as 1 but
satises the negation of the above sentence.
In general, / [= means that is true in / when the variables are
interpreted as ranging over [/[, the other symbols in being given their
obvious interpretation.
Another example:
Z [= x(x > 0 y
1
y
2
y
3
y
4
(x = y
2
1
+ y
2
2
+ y
2
3
+ y
2
4
))
and this expresses the fact that every positive integer is the sum of four
squares.
1.5 Models and theories
1.
Let S be a set of sentences. A model of S is a structure /such that /[=
for all S, and sig(/) = sig(S).
For example, a group can be described as a model ( = ([([, ,
1
, 1) of the
axioms of group theory:
xyz((x y) z = x (y z))
x(x 1 = 1 x = x)
x(x x
1
= x
1
x = 1)
1.5. MODELS AND THEORIES 11
2.
The class of all models of S is denoted Mod(S). A sentence is said to be a
logical consequence
1
of S (written S [= ) if sig() sig(S), and /[= for
all / Mod(S).
A theory is a set T of sentences which is consistent and closed under logical
consequence; in other words, T has at least one model, and T whenever
is a sentence such that sig() sig(T) and /[= for all / Mod(T).
For example, the theory of groups is the set of all logical consequences
of the axioms of group theory. These axioms have many nonobvious logical
consequences, e.g. the Jacobi identity
((x, y), z
x
) ((y, z), x
y
) ((z, x), y
z
) = 1
where we use abbreviations (x, y) = x
1
y
1
x y and x
y
= y
1
x y.
3.
A model class is a nonempty class of structures all having the same signature.
An elementary model class is a model class of the form Mod(S) where S is
a consistent set of sentences. There are lots of nonelementary model classes,
e.g. the class of nite groups.
If K is a model class, we write Th(K) for the theory of K, i.e. the set of
sentences such that sig() sig(K) and /[= for all / K. Note that
Th(K) is a theory and for any theory T we have T = Th(Mod(T)). There is
a natural 11 correspondence between theories and elementary model classes.
4.
If T is a theory and S T, we say that S is a set of axioms for T if
T = Th(Mod(S)). If there exists a nite set of axioms for T, we say that T
is nitely axiomatizable.
For example, the theory of groups is nitely axiomatizable (a nite set of
axioms for it is displayed above). We shall see later that the theory of elds
of characteristic 0 is not nitely axiomatizable.
1
Note: This denition is somewhat unusual because, for instance, x(x < 0 x =
0 x > 0) is not a logical consequence of xy(x < y x = y x > y), owing to the
restriction on signature.
12 CHAPTER 1. SENTENCES AND MODELS
Chapter 2
Complete theories
2.1 Denitions and examples
1.
A theory T is complete if for all sentences , either T or T,
provided sig() sig(T).
Examples: The theory of groups is not complete. The theory of elds
of characteristic 0 is not complete (e.g. x(x x = 1 + 1) is true in R, false
in Q). We shall see later that the theory of algebraically closed elds of
characteristic 0 is complete.
2.
Two structures / and B are elementarily equivalent (written / B) if they
have the same signature and satisfy the same sentences. In other words,
Th(/) = Th(B).
3.
Proposition. For a theory T the following are equivalent.
1. T is complete;
2. all models of T are elementarily equivalent;
13
14 CHAPTER 2. COMPLETE THEORIES
3. T = Th(/) for some structure /.
Proof. Trivial.
4.
Exercise. Prove that the theory of dense linear ordreings without end points
is complete. Use the method of elimination of quantiers, described below.
axioms for linear orderings:
xyz(x < y y < z x < z)
xy(x < y y < x)
xy(x < y x = y y < x)
dense : xy(x < y z(x < z < y))
without end points : xyz(y < x < z) .
5.
We use the following notational convention: (x
1
, . . . , x
n
) denotes a formula
whose free variables are among x
1
, . . . , x
n
.
A theory T is said to admit elimination of quantiers if for all formulas
(x
1
, . . . , x
n
), n 1, there exists a quantier free formula
(x
1
, . . . , x
n
)
such that
T [= x
1
. . . x
n
((x
1
, . . . , x
n
)
(x
1
, . . . , x
n
)) .
For the above exercise, show that the theory of dense linear orderings
without end points admits elimination of quantiers. (Do this by induction
on the number of quantiers, working from the inside out.) Once this has
been done, completeness follows easily.
Historically, the method of elimination of quantiers came very early. It
is also the most versatile method for showing that algebraic theories are com
plete and/or decidable. The drawback is that the method is syntactical. We
shall develop various modeltheoretic methods for proving the same results.
2.2. VAUGHTS TEST 15
2.2 Vaughts test
Vaughts test is another method for establishing completeness of theories. It
is less versatile than quantier elimination, but much easier to use.
1.
Two structures / and B are said to be isomorphic (written /
= B) if
sig(/) = sig(B) and there exists an isomorphic map of / onto B, i.e.
i : [/[ [B[ such that
1. i is oneone and onto;
2. R
A
(a
1
, . . . , a
n
) if and only if R
B
(a
1
, . . . , a
n
);
3. i(o
A
(a
1
, ..., a
n
)) = o
B
(i(a
1
), . . . , i(a
n
));
4. i(c
A
) = c
B
.
Note that isomorphic structures are essentially identical. In particular they
satisfy the same sentences, i.e. /
= B implies / B.
2.
The cardinality or power of a structure / is the cardinality of its universe.
For instance, we say that / is countable if and only if [/[ is countable, etc.
The power of / is denoted / or card(/).
3.
A theory T is said to be categorical if
(i) T has at least one model of power ;
(ii) any two such models are isomorphic.
16 CHAPTER 2. COMPLETE THEORIES
4.
Theorem (Vaughts test for completeness). Let T be a countable theory
such that
(i) T has no nite models;
(ii) for some innite cardinal , T is categorical.
Then T is complete.
We shall prove this later as an application of the L owenheimSkolem
Tarski theorem. For now we content ourselves with giving various examples
of how the theorem is used to prove completeness of specic theories.
2.3 Applications of Vaughts test
1.
We can use Vaughts test to show that the theory of dense linear orderings
without end points is complete.
Clearly the theory has no nite models. We shall show that the theory is
0
categorical. Let / = ([/[, <
A
) and and B = ([B[, <
B
) be countable dense
linear orderings without end points. To show /
= B we use a backandforth
argument. Let [/[ = a
n
: n , [B[ = b
n
: n . The construction
of an isomorphism proceeds in stages. At each stage we have constructed a
nite partial isomorphism of / onto B. Stage 2n. Pick the least k so that
a
k
= a
, b
). Stage 2n+1.
Proceed as in stage 2n replacing as by bs and vice versa. In this way we
build up an isomorphism of / onto B.
2.
We use Vaughts test to show that the theory of nontrivial torsion free divis
ible Abelian groups is complete.
2.3. APPLICATIONS OF VAUGHTS TEST 17
Abelian groups:
xyz((x + y) + z = x + (y + z))
xy(x + y = y + x)
x(x + 0 = x)
x(x + (x) = 0)
An Abelian group is said to be divisible if it satises
xy(ny = x)
for n = 1, 2, 3, . . . . This is an innite set of axioms. Note that ny is an
abbreviation for
y + + y
. .
n times
.
An Abelian group is said to be torsion free if it satises
x(nx = 0 x = 0)
for n = 1, 2, . . . . This is again an innite set of axioms.
An Abelian group is nontrivial if it contains a nonzero element. Clearly
any nontrivial torsion free Abelian group is innite. Thus our theory has
no nite models. To prove completeness, we shall apply Vaughts test by
showing that our theory is categorical for any uncountable cardinal .
An example of a torsion free divisible Abelian group is the additive group
of the rationals (Q, +, , 0). More generally, if V is any vector space over
the eld of rational numbers, then the additive group of V is torsion free and
divisible.
Fact: Any torsion free divisible Abelian group is of this form.
Proof. Let / = ([/[, +, , 0) be a torsion free divisible Abelian group.
For a [/[ and r Q we want to dene ra. Let r = m/n, m, n Z, n > 0.
Dene
ra = some b with ma = nb .
We need to show that ra is well dened. So suppose ma = nb
1
= nb
2
. Then
n(b
1
b
2
) = 0. hence b
1
b
2
= 0 by torsion freeness of /. The vector space
18 CHAPTER 2. COMPLETE THEORIES
axioms
(r
1
+ r
2
)a = r
1
a + r
2
a
r(a
1
+ a
2
) = ra
1
+ ra
2
1a = a
(r)a = ra
are easily veried. Thus / is the additive group of a vector space over Q.
We shall need the following facts from the theory of vector spaces. (1)
Any vector space V over a eld F has a basis, i.e. a set U V such that
every element of V is uniquely expressible in the form
r
1
u
1
+ + r
n
u
n
, r
i
F ,
where the u
i
are distinct elements of U, 1 i n. (Proof: use Zorns
lemma.) (2) Any two bases of V over F have the same cardinality, the
dimension of V . (3) If V
1
and V
2
are vector spaces over F of the same
dimension, then they are isomorphic over F.
Let / and B be torsion free divisible Abelian groups of power where
is uncountable. As vector spaces over Q they have bases U
A
and U
B
respectively. Put = card(U
A
). Since every element of [/[ is of the form
r
1
u
1
+ + r
n
u
n
, r
i
Q, u
i
U
A
, we have [/[
0
. From this we
easily deduce card(U
A
) = = . Similarly card(U
B
) = . So by fact (3),
/
0
categorical, since there are countable vector spaces over Q of dimensions
1, 2, 3, . . . . The theory of dense linear orderings without end points is not
categorical, >
0
.
2.3. APPLICATIONS OF VAUGHTS TEST 19
4.
Remark. Let T
p
be the theory of algebraically closed elds of characterstic
p, where p is a prime or p = 0. We shall see later that Vaughts test can be
used to show that T
p
is complete. Namely, T
p
is categorical for all >
0
.
The eld axioms are as follows:
axioms for a commutative ring:
xyz((x + y) + z = x + (y + z))
xy(x + y = y + x)
x(x + 0 = x)
x(x + (x) = 0)
xyz((x y) z = x (y z))
xy(x y = y x)
xyz(x (y + z) = x y + x z)
0 ,= 1
eld axiom: x(x ,= 0 y(x y = 1)
If we weaken the last axiom to
xy((x ,= 0 y ,= 0) x y ,= 0)
we get the axioms for a domain (usually called integral domain). A eld (or
domain) is said to be of characteristic p if p is the least n such that
n = 1 + + 1
. .
n times
= 0 .
Here p will have to be a prime number (since m n = 0 implies m = 0 or
n = 0). If there is no such p, then we say that the eld is of characteristic 0.
For example Q, R, C, Z are of characteristic 0.
A eld T = ([T[, +, , 0, , 1) is said to be algebraically closed if it satises
x
0
x
1
. . . x
n
(x
n
,= 0 y(x
n
y
n
+ + x
1
y + x
0
= 0))
for n = 1, 2, 3, . . . . For example, the fundamental theorem of algebra asserts
that the complex eld ( = (C, +, , 0, , 1) is algebraically closed.
20 CHAPTER 2. COMPLETE THEORIES
5.
Remark. A dicult theorem of Morley says that if a theory T is countable
and categorical for some uncountable cardinal , then it is categorical for
all uncountable cardinals . This shows that the class of theories to which
Vaughts test applies is rather limited. Many important complete theories
are not categorical for any . This limitation of Vaughts test will be partly
overcome by means of saturated models. (See theorem 7.3.6 below.)
Chapter 3
The compactness theorem
In this chapter and in chapter 4 it will be convenient to assume that formulae
have been dened in terms of , , only; then , , , can be introduced
by denition as usual.
3.1 Proof of the compactness theorem
1.
A set S of sentences is said to be consistent if it has a model. The following
theorem is due to G odel 1929 in the countable case and Malcev 1936 in the
uncountable case.
2.
Theorem (compactness theorem). If every nite subset of S is consistent,
then S is consistent.
Proof. We give a proof by transnite induction in the style of Henkin 1949.
Assume that S is nitely consistent, i.e. every nite subset of S has a model.
Let = max(
0
, card(S)). Thus is a cardinal; we identify cardinals with
initial ordinals. Let c
: < . Let
, , and a function
h : as follows.
Stage 0: S
0
= S.
Stage 2 + 1: Let h() be the least such that c
(x). Dene
S
2+1
= S
2
(x
(x))
(c
h()
) .
The sentence (x
(x))
(c
h()
) is called a Henkin sentence. We claim
that S
2+1
is nitely consistent (assuming S
2
was). Let / be a model of
some nite subset of S
2+1
. We want to expand / to a model of those
sentences plus the Henkin sentence. First, interpret in [/[ arbitrarily the
nonlogical symbols in
. Let /
<
S
.
If o is an nary operation symbol dene
o([t
1
], . . . , [t
n
]) = [o(t
1
, . . . , t
n
)] .
For c a constant symbol put c = [c]. Our model /= ([/[, ) is dened by
[/[ = T/, (R) = R, (o) = o, (c) = c.
We claim that, for all sentences with sig() W, / [= if and
only if S
2
or =
1
it is easy to check. Suppose = y(y). If / [= then
/ [= (t) for some t T. Hence by induction (t) S
. Therefore
y(y) S
. By
3.2. SOME APPLICATIONS TO FIELD THEORY 23
relabeling variables we may assume that y = x. Also (x) =
) S
) S
), hence / [= y(y).
This completes the proof.
3.2 Some applications to eld theory
1.
Theorem (A. Robinson). Suppose is a sentence which is true in all elds
of characteristic 0. Then is true in all elds of characteristic p, for all but
nitely many primes p (the exceptional primes).
Proof. If the conclusion fails, then is consistent with the eld ax
ioms plus 1 + + 1
. .
p
= 0, for innitely many primes p. But, for primes p,
1 + + 1
. .
p
= 0 implies 1 + + 1
. .
n
,= 0 for all n < p. Hence is nitely
consistent with the theory of elds of characteristic 0. Hence, by the com
pactness theorem, holds in some eld of characteristic 0, contradiction.
2.
Corollary. The theory of elds of characteristic 0 is not nitely axiomatiz
able.
3.
An ane variety is a set X C
n
such that
X = (z
1
, . . . , z
n
) : f
i
(z
1
, . . . , z
n
) = 0, 1 i k
where f
i
, 1 i k are polynomials in n variables with coecients from C.
A mapping G : X C
n
is said to be polynomial if there exist polynomials
g
j
, 1 j n in n variables with coecients from C, such that for all
(z
1
, . . . , z
n
) X,
G(z
1
, . . . , z
n
) = (g
1
(z
1
, . . . , z
n
), . . . , g
n
(z
1
, . . . , z
n
)) .
24 CHAPTER 3. THE COMPACTNESS THEOREM
Theorem (Ax). Let G : X X be a polynomial map of an ane variety
in C
n
into itself. If G is oneone, then G is onto.
Proof. More generally, we shall prove that the theorem holds for an
arbitrary algebraically closed eld in place of C. Besides the compactness
theorem, we shall use completeness of the theory T
p
of algebraically closed
elds of characteristic p where p is 0 or a prime.
For prime p let F
p
be the algebraic closure of the prime eld F
p
= Z/pZ.
Note that any nite set of elements in F
p
is contained in a nite subeld of
F
p
. Hence the theorem holds for F
p
.
Observe that the theorem holds for a particular eld T if and only if T
satises a certain collection of sentences. If one of these sentences is false
in C, then by completeness of T
0
we have T
0
[= . Hence T
p
[= for all
suciently large primes p. Hence F
p
[= . This is a contradiction.
For more applications of the compactness theorem in eld theory, see
Robinsons book and more recent literature. For applications to local prop
erties in group theory, see Malcevs book.
3.3 The L owenheimSkolemTarski theorem
1.
Note that the proof of the compactness theorem also established the fol
lowing result: If S is (nitely) consistent then S has a model of power
max(
0
, card(S)).
2.
Theorem (L owenheimSkolemTarski). Let S be a set of sentences which
has an innite model. Let be a cardinal max(
0
, card(S)). Then S has
a model of power .
Proof. Let c
= S c
,= c
: < < .
Since S has an innite model, S
OWENHEIMSKOLEMTARSKI THEOREM 25
S
, B
,
B B
. Then /
, B
. Hence /
= B
so T is not categorical.
26 CHAPTER 3. THE COMPACTNESS THEOREM
Chapter 4
Decidability
4.1 Recursively axiomatizable theories
1.
Let V be a xed recursive
1
signature. The formulas with sig() V can
be Godel numbered in the usual way. This permits notions from recursion
theory to be introduced. For instance, a set F of formulas is said to be
recursive if its set of Godel numbers () : F is recursive, etc.
2.
Denition. A theory T is said to be recursively axiomatizable if (i) sig(T)
is recursive, and (ii) T has a recursive set of axioms.
Remark. The notion of recursive axiomatizability is an important general
ization of that of nite axiomatizability. Most theories which arise in prac
tice have nite signature and are recursively axiomatizable, e.g. the theory of
groups, the theory of elds of characteristic 0, rst order Peano arithmetic,
ZF set theory. Examples of theories which are not recursively axiomatizable
are: the theory of nite groups; Th(Z) where Z = (Z, +, , , 0, 1, <).
1
i.e. V is a recursive set, and we can recursively identify the elements of V as nary
relation or operation symbols or constant symbols. This is the case if V is nite.
27
28 CHAPTER 4. DECIDABILITY
3.
A very important general result about recursively axiomatizable theories is
the following, which is a reformulation of G odels completeness theorem.
Theorem (essentially G odel 1929). Let T be a recursively axiomatizable
theory. Then T is recursively enumerable.
This theorem would usually be proved, following G odel, by showing that
T can be generated from its axioms and a certain explicit recursive set of
logical axioms by means of a certain explicit recursive set of logical rules.
One would then apply Churchs thesis to conclude that T is recursively enu
merable. Such a proof would give more or less information depending on the
choice of logical axioms and rules. We shall give a bare bones proof which
avoids these concepts entirely.
4.
Denition. A sentence is logically valid if / [= for all structures / with
sig(/) sig().
5.
Lemma. Let V be a xed recursive signature. The set of all logically valid
sentences with sig() V is recursively enumerable.
Proof. We may safely assume that V contains no operation symbols. We
may also assume that the only logical symbols are , , . (Thus we are
dispensing with =.)
Let c
n
: n be a recursive set of new constant symbols and put
W = V c
n
: n . Let
n
(v
in
) : n be a recursive enumeration of
all formulas (v
i
) with exactly one free variable and with sig() W. We
can choose
2
our enumeration so that
sig(
n
(v
in
)) V c
0
, . . . , c
2n1
.
We shall work with Henkin sentences (v
in
n
(v
in
)) (c
2n
). Let S =
n
:
n be a recursive enumeration of all sentences with sig() W.
2
This amounts to arranging things so that, in the proof of the compactness theorem
3.1, the function h is given by h(n) = 2n.
4.1. RECURSIVELY AXIOMATIZABLE THEORIES 29
Given a sentence with sig() V , we have
is logically valid
has no model
has no countable model (L owenheimSkolem theorem)
there is no structure / such that [/[ = c
n
: n and sig(/) = W
and / [= and, for all n, / [= (v
in
n
(v
in
))
n
(c
2n
)
there is no function f : S 0, 1 such that for all n, f(
n
) =
1 f(
n
), and for all m and n, f(
m
m
) = max(f(
m
), f(
n
)) and
f(v
in
n
(v
in
)) = f(
n
(c
2n
) f(
n
(c
m
)).
f : S 0, 1 nR(f S
n
, ) where S
n
=
0
, . . . ,
n1
and R is a
xed primitive recursive relation
N f : S 0, 1 (n N) R(f S
n
, ) (by K onigs lemma)
N P(N, ) where P is a xed primitive recursive relation.
This proves the lemma.
Remark. The set of logically valid sentences is not recursively enumerable
(Churchs theorem).
6.
Proof of the theorem. Given a recursively axiomatizable theory T, let A =
n
: n be a recursive set of axioms for T. Put A
n
=
0
, . . . ,
n1
. If
is a sentence with sig() sig(T), we have
T
A [=
nA
n
[= (compactness theorem)
n
_ _
n1
i=0
i
is logically valid
.
By the lemma, the expression in square brackets is recursively enumerable.
Hence so is T.
30 CHAPTER 4. DECIDABILITY
7.
Exercise (Craig). Prove the converse of the above theorem: If a theory T
with sig(T) recursive is recursively enumerable, then it is recursively axiom
atizable.
4.2 Decidable theories
1.
Denition. A theory T is decidable if (i) sig(T) is recursive, (ii) T is recur
sive.
Remark. Every decidable theory is recursively axiomatizable, but the con
verse fails, e.g. rst order Peano arithmetic. there even exist nitely axiom
atizable undecidable theories, e.g. Robinsons Q, the theory of groups, the
theory of elds.
2.
Remark. The rst order theories which are important in algebra tend to be
decidable. We make a short table.
Decidable Undecidable
Abelian groups (nite) groups
Boolean algebras (nite) distributive lattices
linear orderings
. It should be easy to
decide whether
T.
4.3 Decidable models
1.
Denition. A countable structure / is decidable if (i) sig(/) is nite; (ii)
there exists an enumeration of the universe of /, [/[ = a
k
: k , such
that
(((v
1
, . . . , v
n
)), k
1
, . . . , k
n
)) : / [= (a
k
1
, . . . , a
kn
)
32 CHAPTER 4. DECIDABILITY
is recursive.
2.
Denition. A countable structure / is computable if (i) sig(/) is nite, and
(ii) as above but restricted to atomic formulas .
For example, the structure Z = (Z, +, , , 0, 1, <) is computable but not
decidable. The structure (Q, <) is computable and hence decidable in view
of the following proposition.
3.
Proposition. Let T be a recursively axiomatizable theory which admits
elimination of quantiers. If / Mod(T) is computable, then it is decid
able.
Proof. To decide whether / [= (a
1
, . . . , a
n
) nd a quantier free
such
that T [= x
1
. . . x
n
(
recursively. So /[= (a
1
, . . . , a
n
) if and only if /[=
(a
1
, . . . , a
n
).
Since
(a
1
, . . . , a
n
).
Example. It is fairly easy to see that the eld Q of algebraic numbers is re
cursive. However, it is also a model of the theory of algebraically closed elds
of characteristic 0. We shall see later that this theory admits elimination of
quantiers. Hence Qis decidable. Similarly, all countable algebraically closed
elds are computable and hence decidable.
4.
Theorem. Let T be a decidable theory such that sig(T) is nite. Then T
has a decidable model.
Proof. We imitate the proof of the compactness theorem 3.1.2. Put V =
sig(T) and let c
m
: m be a recursive list of new constant symbols.
Put W = V c
m
: m . Fix a variable x and let
n
(x) : n be
a recursive enumeration of all formulas (x) with only one free variable x
4.3. DECIDABLE MODELS 33
and with sig() W. Let
n
: n be a recursive enumeration of all
sentences with sig() W. Perform the following recursive construction.
Stage 0: Let S
0
= T.
Stage 2n + 1: Let h(n) be the least m such that c
m
does not occur in
S
2n
n
(x). Put S
2n+1
= S
2n
(x
n
(x))
n
(c
h(n)
).
State 2n+2: Put S
2n+2
= S
2n+1
n
if this is consistent; S
2n+2
= S
2n+1
n
otherwise. To see that we can make this decision recursively, note that
S
2n+1
= T
0
, . . . ,
2n
for some nitely many sentences
0
, . . . ,
2n
. Let
0
, . . . ,
2n
,
n
be the result of replacing the new constant symbols c
m
by new
variables z
m
. Then
S
2n+1
n
is consistent
T z
0
. . . z
j
__
2n
i=0
i
n
_
is consistent
z
0
. . . z
j
__
2n
i=0
i
n
_
/ T,
and we can decide this recursively.
At the end of the construction, S
n
S
n
is a complete recursive
theory. As in the proof of 3.1.2 we can build a model / = ([/[, ) where
[/[ = T/, T = Godel numbers of variable free terms, t
1
t
2
if and
only if t
1
t
2
S
.
5.
Remark. There exist examples of recursively (even nitely) axiomatizable
theories with no decidable (even computable) model. For example, we may
take as axioms a nite fragment of Peano arithmetic together with a false
0
1
sentence.
34 CHAPTER 4. DECIDABILITY
Chapter 5
Elementary extensions
5.1 Denition and examples
1.
Denition. Let /, B be structures with sig(/) = sig(B) and [/[ [B[.
We say that /
e
B if for all formulas (x
1
, . . . , x
n
) with only the free
variables shown, and all a
1
, . . . , a
n
[/[, / [= (a
1
, . . . , a
n
) if and only if
B [= (a
1
, . . . , a
n
). We then say that / is an elementary substructure of B,
or B is an elementary extension of /.
2.
Let /, B be as above. We say that / B (/ is a substructure of B, B is an
extension of /) if the above condition holds for atomic formulas .
3.
Example. Algebra is full of examples of substructures. E.g. if / and B are
groups, / B if and only if / is a subgroup of B. Similarly for rings, elds,
linear orderings, etc.
4.
Example. Let Q = (Q, +, , , 0, 1, <) and 1 = (R, +, , , 0, 1, <). Then
clearly Q 1, but Q ,
e
1 since, for example, 1 [= x(x x = 1 + 1) while
35
36 CHAPTER 5. ELEMENTARY EXTENSIONS
Q does not. This example actually shows that Q , 1, i.e. Q and 1 are not
elementarily equivalent (denition 2.1.2).
5.
Proposition. If /
e
B then / B.
Proof. Obvious.
The converse does not hold, e.g.
6.
Example. Let N = 0, 1, 2, . . . , P = 1, 2, . . . . Then (P, <) (N, <), and
(P, <) (N, <) since they are isomorphic. However (P, <) [= x(1 = x1 <
x) and (N, <) [= x(x < 1) so (P, <) ,
e
(N, <) .
However, this phenomenon does not occur in theories which admit elim
ination of quantiers:
7.
Proposition. If /, B Mod(T) where T admits elimination of quantiers,
then / B implies /
e
B.
Proof. Obvious.
8.
Examples. Let X be any densely ordered subset of the real numbers such
that X has no rst or last element. E.g. X = (0, 1) (1, 2) or X = (0, 1) Q.
Then (X, <)
e
(R, <) because the theory of dense linear ordering without
endpoints admits elimination of quantiers.
9.
Example. The fact that the theory of algebraically closed elds admits
elimination of quantiers yields the following result:
5.1. DEFINITION AND EXAMPLES 37
Let Q be the eld of algebraic numbers. Let
f
1
(x
1
, . . . , x
n
) = = f
k
(x
1
, . . . , x
n
) = 0 ,= g(x
1
, . . . , x
n
)
be a nite system of equations and inequations with coecients
in Q. If the system has a solution in some extension eld of Q,
then it has one in Q.
Proof. We may assume that the extension eld is algebraically closed.
Hence it is an extension of Q, and by quantier elimination the extension is
elementary. This gives the result immediately if we look at the formula
x
1
. . . x
n
_
k
i=1
f
i
(x
1
, . . . , x
n
) = 0 g(x
1
, . . . , x
n
) ,= 0
_
.
In other words, if a variety is nonempty, then it has a point with coordinates
in Q. This is closely related to Hilberts Nullstellensatz. (See chapter 6.)
9.
Denition (A. Robinson). A theory T is said to be model complete if for all
/, B Mod(T), / B implies /
e
B.
The previous proposition says that if T admints elimination of quan
tiers, then T is model complete. The converse is not true, e.g. T =
Th(Z, +, , 0, 1, <) = Presburger arithmetic. This theory is model com
plete but does not admint elimination of quantiers. However, it admits
elimination of quantiers if we add dened relations
m
, m 2.
10.
Example. We shall show later that the theory of algebraically closed elds
is model complete and admits elimination of quantiers. These concepts
have algebraic meaning: model completeness expresses the Hilbert Nullstel
lensatz, while quantier elimination is an abstract formulation of the theory
of resultants.
38 CHAPTER 5. ELEMENTARY EXTENSIONS
5.2 Existence of elementary extensions
1.
Theorem. Any innite structure / has a proper elementary extension /
.
Proof. Let / = ([/[, ). Introduce new constant symbols a for each
a [/[ and form the elementary diagram of /, i.e. the set of all sen
tences (a
1
, . . . , a
n
), sig((x
1
, . . . , x
n
)) sig(/), a
1
, . . . , a
n
[/[, such
1
that / [= (a
1
, . . . , a
n
). Note that an elementary extension of / is virtually
the same thing as a model of the elementary diagram of /.
Let c be a new constant symbol and let S = (elementary diagram of
/) c ,= a : a [/[. Since [/[ is innite, S is nitely consistent. [ Details:
Let S
0
be a nite subset of S. Let c
0
[/[, c
0
,= a for all a mentioned in S
0
.
Then / [= S
0
if we interpret a as a, c as c
0
. ] Therefore by the compactness
theorem, S is consistent, i.e. has a model B = ([B[, ). Put /
= ([/
[,
)
where [/
[ = [B[,
= sig(/). Clearly /
e
/
since B [= elementary
diagram of /. Also [/[ , = [/
[ since c = (c) [/
[ [/[.
2.
Example. Let 1 = (R, +, , , 0, 1, <, ) where represents other struc
ture on R. Let 1
will
contain innitesimals, i.e. quantities such that 0 < < r for all r R,
r > 0. Elements of 1
)
<
where is a limit ordinal, such
that /
<
/
[ =
<
[/
[;
2. /
<
R
A
, o
A
<
o
A
, and c
A
= c
A
0
.
6.
Theorem (Tarski/Vaught elementary chain principle). Given an elementary
chain, i.e. a chain (/
)
<
such that /
e
/
<
/
. Then /
e
/
[, /
[= (a
1
, . . . , a
n
) if and only if
/
[= (a
1
, . . . , a
n
). This is trivial if is atomic or a Boolean combination of
shorter formulas. Assume that (x
1
, . . . , x
n
) = y(x
1
, . . . , x
n
, y). If /
[=
y(a
1
, . . . , a
n
, y), let a
n+1
[/
[ be such that /
[= (a
1
, . . . , a
n
, a
n+1
).
Then /
[= (a
1
, . . . , a
n
, a
n+1
) by induction, so /
[= y(a
1
, . . . , a
n
, y).
Conversely, suppose /
[= y(a
1
, . . . , a
n
, y). Then /
[= (a
1
, . . . , a
n
, b)
for some b [/
[ =
<
[/
[. Then
/
[= (a
1
, . . . , a
n
, b) by induction. Hence /
[= y(a
1
, . . . , a
n
, y). Hence
/
[= y(a
1
, . . . , a
n
, y) since /
e
/
.
Later we shall use this theorem plus compactness to construct saturated
models.
40 CHAPTER 5. ELEMENTARY EXTENSIONS
5.3 Elementary monomorphisms
1.
Let /, B be structures with sig(/) = sig(B). We say that f : [/[ [B[ is
an elementary embedding or elementary monomorphism if f is oneone and,
for all formulae (x
1
, . . . , x
n
) and all a
1
, . . . , a
n
[/[, / [= (a
1
, . . . , a
n
)
B [= (f(a
1
), . . . , f(a
n
)).
2.
If the above equivalence holds for atomic , we say that f is an embedding
or monomorphism of / into B, or an isomorphism of / into (not necessarily
onto) B. For example, what is usually called a monomorphism of groups,
rings, etc. is a monomorphism in this sense.
The concepts of monomorphism and elementary monomorphism are es
sentially just trivial variants of the concepts of extension and elementary
extension.
3.
Theorem. If / B then there exists a structure ( such that both / and
B are elementarily embeddable into (. (The converse is obvious.)
Proof. Let S = (elementary diagram of /) (elementary diagram of B).
[ We assume that a : a [/[ b : b [B[ = . ] Clearly any
model of S yields a ( as desired. So by the compactness theorem it suf
ces to show that S is nitely consistent. Let S
0
be a nite subset of S.
Say S
0
= (a
1
, . . . , a
m
), (b
1
, . . . , b
n
) where sig( ) sig(/) = sig(B),
a
1
, . . . , a
m
[/[, b
1
, . . . , b
n
[B[. Since / [= x
1
. . . x
m
(x
1
, . . . , x
m
) and
B [= y
1
. . . y
n
(y
1
, . . . , y
n
) and / B, we see that x
1
. . . x
m
(x
1
, . . . , x
m
)
y
1
. . . y
n
(y
1
, . . . , y
n
) is consistent. Hence S
0
is consistent, Q.E.D.
4.
From the above we can derive a criterion for model completeness:
Theorem(Robinsons test). A theory T is model complete for any /, B
Mod(T) and monomorphism / B, we can nd an elementary extension
5.3. ELEMENTARY MONOMORPHISMS 41
/
of / and a monomorphism B /
B
B
B
B
B
B
B
B
/
O
e /
commutes.
Proof. : Since T is model complete, the monomorphism / B is elemen
tary, so we may take /
= B.
: Given a monomorphism / B where /, B Mod(T). Use the
hypothesis repeatedly to construct a pair of elementary chains
B
=
B
0
/
e
!
B
B
B
B
B
B
B
B
B
1
/
e
!
B
B
B
B
B
B
B
B
/
e
!
B
B
B
B
B
B
B
B
B
n
/
e
!
B
B
B
B
B
B
B
B
/
=
O
/
0
/
e
/
1
/
e
O
/
e
/
n
/
e
O
where all triangles commute. Hence /
n
/
n
=
n
B
n
= B
so we
have by Tarski/Vaught
/
= B
/
O
e
/
B
O
e
whence / B is elementary. So T is model complete.
42 CHAPTER 5. ELEMENTARY EXTENSIONS
Chapter 6
Algebraically closed elds
6.1 Simple eld extensions
1.
Let / be any structure. For each a [/[ introduce a new constant symbol
a. The diagram of / is the set of atomic sentences (a
1
, . . . , a
n
) which hold
in /. Recall that an extension of / is any structure B /. Note that an
extension of / is virtually the same thing as a model of the diagram of /.
2.
Let / be a eld. A eld extension of / is an extension B / such that B
is a eld. A eld extension of / is thus virtually the same thing as a model
of (eld axioms) (diagram of /).
3.
If /, B are elds, / B, and b [B[, let /[b] be the smallest substructure
of B containing [/[ b, i.e. the subring generated by [/[ b. Let /(b)
be the smallest subeld of B containing [/[ b. A eld extension of / is
said to be simple if it is of the form /(b).
We want to survey all possible simple eld extensions of /. This survey is
the main piece of algebraic information that we need for our modeltheoretic
analysis.
43
44 CHAPTER 6. ALGEBRAICALLY CLOSED FIELDS
4.
Let / be a eld and consider terms t(x) such that sig(t(x)) sig(diagram of
/)) and x is the only free variable in t(x). Two such terms t
1
(x) and t
2
(x)
are said to be equivalent if
(eld axioms) (diagram of /) [= x(t
1
(x) = t
2
(x)) .
The set of equivalence classes is naturally a commutative ring and is denoted
/[x]. We shall see that /[x] is a domain.
5.
A polynomial is a term f(x) as above with either f(x) 0 or f(x) =
a
n
x
n
+ + a
1
x + a
0
, a
i
[/[. Clearly each equivalence class in /[x]
contains a polynomial. We claim that each equivalence class contains only
one polynomial. Thus the polynomials are a system of representatives for
the equivalence classes in /[x].
Obviously the polynomials over / form a domain, since if f(x) = a
m
x
m
+
+a
1
x +a
0
, g(x) = b
n
x
n
+ +b
1
x +b
0
, a
m
,= 0 ,= b
n
, then f(x) g(x) =
a
m
b
n
x
m+n
+ , = 0. We now use the following fact: any domain is embed
dable in a eld, its eld of quotients. In particular, the set of polynomials
over / is embedded in a eld extension of /. Hence distinct polynomials
represent distinct equivalence classes in /[x], as claimed above.
6.
It follows that /[x] is a domain. The quotient eld of /[x] is denoted /(x).
This is in agreement with our earlier notation for simple extensions. /(x) is
our rst example of a simple extension of /.
7.
The degree of a nonzero polynomial f(x) /[x] is deg(f) = n where f(x) =
a
n
x
n
+ + a
1
x + a
0
, a
n
,= 0. The degree of the zero polynomial, deg(0), is
undened.
Lemma (division algorithm). If f(x), g(x) /[x], g ,= 0, there exist
q(x), r(x) /[x] such that f(x) = g(x) q(x) + r(x) and either r 0 or
deg(r) < deg(g).
6.1. SIMPLE FIELD EXTENSIONS 45
Proof. As usual.
In the above lemma, if r 0 we say that g divides f.
8.
Denition. For f(x) /[x] we say that f(x) is nonconstant if deg(f) > 0.
We say that f(x) is irreducible if f is nonconstant and there is no nonconstant
g(x) /[x] of lower degree than f(x) such that g(x) divides f(x).
9.
Let f(x) /[x] be nonconstant. For g
1
, g
2
/[x] say g
1
= g
2
mod f if
f divides g
1
g
2
. Let /[x]/(f(x)) be the set of equivalence classes mod
f. Clearly /[x]/(f(x)) has the structure of a commutative ring with /
as a subeld. Clearly each equivalence class mod f contains one and only
one polynomial g(x) /[x] such that deg(g) < deg(f) or g 0. Thus
/[x]/(f(x)) may be viewed as a vector space of dimension deg(f) over /.
10.
Lemma. If f(x) is irreducible then /[x]/(f(x)) is a eld.
Proof. Given g /[x], deg(g) < deg(f), g ,= 0 mod f, to nd an inverse
mod f. Put
I = sf + tg : s, t /[x] .
Let h I, h , 0 be of least possible degree. Apply the division algorithm
to get f(x) = g(x) q(x) + r(x). Note that r I, hence deg(r) < deg(h) is
impossible. Hence r 0, i.e. h divides f. Also deg(h) deg(g) < deg(f).
Since f is irreducible it follows that h is a constant. h ,= 0 since f ,= 0. We
now have sf + tg = h [/[, so h
1
tg = 1 mod f, so h
1
t is the desired
inverse, Q.E.D.
11.
In the above lemma, let b be the equivalence class of x in /[x]/(f(x)). Then
/(b) = /[x]/(f(x)) is a simple eld extension of /. Note also that /[b] =
46 CHAPTER 6. ALGEBRAICALLY CLOSED FIELDS
/(b) in this case, so the dimension of /(b) over / as a vector space is nite,
being just deg(f).
We now have two kinds of examples of simple eld extensions of /. The
next theorem says that there are no others.
12.
Theorem. Let /(b) be any simple eld extension of /. Then either
1. /(b)
= /(x).
In both cases the isomorphism is the identity on / and sends b to x.
Proof. Case 1: b is algebraic over /, i.e. f(b) = 0 for some nonzero f /[x].
Let f be of least degree such that this holds. Clearly f is irreducible.
We claim that g(b) = 0 if and only if f divides g. One direction is
obvious. For the other direction, suppose g(b) = 0. The division algorithm
gives g(x) = f(x) q(x) + r(x). Hence r(b) = f(b) q(b) + r(b) = g(b) = 0.
Hence we cannot have deg(r) < deg(f). Hence r 0, i.e. f divides g. This
proves the claim.
It follows that g
1
(b) = g
2
(b) if and only if f divides g
1
g
2
. This matches
the denition of /[x]/f, so we have /[b]
= /[x]/f. But by lemma 10 the
latter is a eld, hence so is the former, i.e. /[b] = /(b).
Case 2: not case 1, i.e. b is transcendental over /. Thus g
1
(b) = g
2
(b)
if and only if g
1
(x) g
2
(x). Thus /[b]
= /[x] as rings. Hence the quotient
elds are also isomorphic, i.e. /(b)
= /(x).
13.
Corollary. Let /(b
1
) and /(b
2
) be two simple extensions of /. If f(b
1
) =
f(b
2
) = 0 for some ireducible f /[x], then /(b
1
)
= /(b
2
).
14.
Corollary. b is algebraic over / if and only if /(b) is nite dimensional over
/.
6.2. ALGEBRAIC CLOSURE 47
6.2 Algebraic closure
1.
A eld extension B / is algebraic over / if each b [B[ is algebraic over
/.
Lemma. If /
alg
B
alg
( then /
alg
(.
Proof. Given c [([, to show that [/(c) : /[ (the dimension of /(c) over /
as a vector space) is nite. Let g B[x] be nonzero such that g(c) = 0. Put
B
0
= /(b
0
)(b
1
) (b
n
) where g(x) = b
n
x
n
+ +b
1
x +b
0
. Hence [B
0
(c) : /[
is nite. Hence [/(c) : /[ is nite.
B
/
nite
B(c)
/
O
/
nite
/(c)
O
2.
Denition. A eld B is algebraically closed if every nonconstant g(x) B[x]
has a root in B, i.e. g(b) = 0 for some b [B[. For any eld /, an algebraic
closure of / is a eld extension B / such that (i) B is algebraic over /,
and (ii) B is algebraically closed.
3.
Proposition. Every eld / has an algebraic closure.
Proof. Note rst that for any nonconstant g(x) /[x] we can adjoin a root,
i.e. we can nd a simple algebraic eld extension /(b) such that g(b) = 0.
(Just factor g(x) into irreducibles and apply the construction of Lemma 6.1.10
to one of these irreducible factors.)
Now let g
(b
) where g
(b
) = 0, and /
<
/
for
limit ordinals . Finally put B
1
= /
. By Lemma 6.2.1 B
1
is algebraic
over /, and by construction every nonconstant g(x) /[x] has a root in B
1
.
48 CHAPTER 6. ALGEBRAICALLY CLOSED FIELDS
Repeating this construction times, we get
/ = B
0
alg
B
1
alg
alg
B
n
alg
,
n , such that each nonconstant g(x) B
n
[x] has a root in B
n+1
. Thus
B =
n
B
n
is an algebraic closure of /.
4.
Lemma. If / is algebraically closed and /
alg
B, then / = B.
Proof. Since / is algebraically closed, the only irreducible polynomials in
/[x] are linear, i.e. of degree 1. Hence by Theorem 6.1.12 every element of
an algebraic extension of / actually belongs to /.
5.
Proposition. Any two algebraic closures of a eld / are isomorphic over
/.
Proof. Let B
1
and B
2
be two algebraic closures of /.
B
1
B
2
/
`A
A
A
A
A
A
A
>
}
}
}
}
}
}
}
If / is algebraically closed, then B
1
= B
2
= / by the previous lemma.
By Zorns lemma or transnite induction, it suces to nd b
i
[B
i
[ [/[,
i = 1, 2, such that /(b
1
)
= /(b
2
) over /. Since / is not algebraically closed,
let f /[x] be nonconstant with no root in /. We may assume that f is
irreducible. Take b
i
[B
i
[ so that f(b
i
) = 0, i = 1, 2. Then /(b
1
)
= /(b
2
)
over / by Corollary 6.1.13. This completes the proof.
By the above proposition we are justied in writing / = the algebraic
closure of / (unique up to isomorphism over /). By the proof of Proposition
6.2.3 we have / = max(/,
0
).
6.3. COMPLETENESS AND MODEL COMPLETENESS 49
6.
Corollary. If B / is algebraically closed, then / is isomorphic over / to
a subeld of B.
B
/
=




/
O
Proof. Let ( be the subeld of B with [([ = b [B[ : b is algebraic over /.
By Lemma 6.2.4 ( is algebraically closed. Hence ( is an algebraic closure of
/. Hence by Proposition 6.2.5 we have that / is isomorphic to ( over /.
6.3 Completeness and model completeness
In this section we show that the theory T
p
of algebraically closed elds of
characteristic p (p = 0 or prime) is complete and model complete.
1.
Lemma. Let / be a eld and let B
1
, B
2
be algebraically closed extensions
of / such that B
1
 = B
2
 = > max(
0
, /). Then B
1
and B
2
are
isomorphic over /.
B
1
B
2
/
`A
A
A
A
A
A
A
>
}
}
}
}
}
}
}
Proof. We use a backandforth argument. Let [B
i
[ = b
i
: < , i = 1, 2.
At stage of our construction we will have a partial isomorphism i
: /
=
50 CHAPTER 6. ALGEBRAICALLY CLOSED FIELDS
/
2
where / /
i
B
i
, /
i
 < , i = 1, 2.
B
1
B
2
i
: /
1
O
=
/
2
O
/
`A
A
A
A
A
A
A
>
}
}
}
}
}
}
}
Stage 0: /
0
1
= /
0
2
= /, i
0
= identity.
Stage 2 + 1: If b
1
is algebraic over /
2
1
, let f /
2
1
[x] be irreducible
such that f(b
1
) = 0, and take b [B
2
[ such that i
2
(f)(b) = 0. If b
1
is
transcendental over /
2
1
, take b [B
2
[
/
2
2
1
)
= /
2
2
(b) = /
2+1
2
by i
2+1
i
2
, i
2+1
(b
1
) = b.
Stage 2 +2: Same as stage 2 +1 except reverse the roles of B
1
and B
2
.
Stage limit : /
i
=
<
/
i
, i = 1, 2, i
<
i
. Finally we have
i
: B
1
= /
= /
2
= B
2
, Q.E.D.
2.
Theorem. T
p
is categorical for all >
0
.
Proof. Let /be the smallest eld of characteristic p, i.e. / = Q = (Q, +, , , 0, 1)
if p = 0, / = Z/p if p is prime. Note that / is a subeld of any eld of
characteristic p. Let B
1
, B
2
Mod(T
p
), B
1
 = B
2
 = . Then we have
that B
1
= B
2
over / by Lemma 6.3.1.
3.
Corollary. T
p
is complete.
Proof. Immediate by Vaughts test 2.2.4.
6.3. COMPLETENESS AND MODEL COMPLETENESS 51
4.
Corollary. T
p
is decidable.
Proof. Immediate from theorem 4.2.3.
5.
Exercise. Prove that the theory of algebraically closed elds (of all charac
teristics) is decidable.
6.
Theorem (A. Robinson). T
p
is model complete.
Proof. Given a monomorphism / B where /, B are algebraically closed,
we want to show that this monomorphism is elementary. Let be a cardinal
> B. By Exercise 5.2 we can get elementary extensions /
of / and B
of B such that /
 = B
 = . Then /
= B
/
O
e
/
B
O
e
Since the diagram commutes, the monomorphism / B is elementary,
Q.E.D.
7.
Corollary (Hilbert). Let / be a eld and let / be its algebraic closure. Let
g
1
(x
1
, . . . , x
n
) = = g
m
(x
1
, . . . , x
n
) = 0
f(x
1
, . . . , x
n
) ,= 0
_
()
be a nite system of polynomial equations and inequations with coecients
in /. If () has a solution in some eld extension of /, then it has a solution
in /.
(Algebraic reformulation: The algebraic points on a variety are dense with
respect to the Zariski topology on that variety.)
52 CHAPTER 6. ALGEBRAICALLY CLOSED FIELDS
Proof. This is immediate from model completeness of the theory of alge
braically closed elds. Consider the sentence
= x
1
x
n
_
m
i=0
g
i
= 0 f ,= 0
_
.
If / B [= then B [= . By Corollary 6.2.6 we have / / B. Hence
/ [= by model completeness.
6.4 Hilberts Nullstellensatz
1.
Let / be a eld and let 1 = /[x
1
, . . . , x
n
] be the ring of polynomials in n
variables x
1
, . . . , x
n
over /. A zero of f 1 is an ntuple b
1
, . . . , b
n
[B[,
where B is a eld extension of /, such that f(b
1
, . . . , b
n
) = 0. An algebraic
zero of f is a zero whose coordinates b
1
, . . . , b
n
lie in /, the algebraic closure
of /.
2.
Theorem (Nullstellensatz). Suppose f, g
1
, . . . , g
m
1 such that every com
mon algebraic zero of g
1
, . . . , g
m
is a zero of f. Then there exists a nonnega
tive integer k such that
f
k
=
m
i=1
p
i
g
i
where p
1
, . . . , p
m
1. (The converse holds trivially.)
Proof. Let I be the set of f 1 for which the conclusion holds. Suppose
the conclusion fails, i.e. f / I. then we claim that I is an ideal in 1, i.e.
(i) g, h I g + h I;
(ii) g I, h 1 g h I;
(iii) 0 I and 1 / I.
6.4. HILBERTS NULLSTELLENSATZ 53
To prove (i), suppose g, h I, say g
k
=
p
i
g
i
and h
l
=
q
i
g
i
. Then
(g + h)
k+l
=
i+j=k+l
_
k + l
i
_
g
i
h
j
.
Since i +j = k +l, we must have either i k or j l. Therefore each term
on the right hand side contains g
k
or h
l
. Therefore each term is of the form
r
i
g
i
. So (g + h)
k+l
is also of this form. So g + h I. To prove (ii) just
note that if g
k
=
p
i
g
i
then (g h)
k
=
h
k
p
i
g
i
so g h I. Part (iii) is
obvious since f / I.
Thus I is an ideal containing no power of f. By Zorns lemma let J I
be an ideal containing no power of f and maximal with this property.
We claim that J is a prime ideal
1
, i.e. g / J, h / J g h / J. If g / J
then the ideal generated by J g is properly larger than J, so some power
of f belongs to it, say f
k
= sg + u where s 1, u J. Similarly, if h / J
then f
l
= th + v where t 1, v J. Hence
f
k+l
= (sg + u) (th + v) = stgh + sgv + thu + uv
. .
J
.
If gh J then f
k+l
J, contradiction. This proves the claim.
Put 1
1
= 1/J = the quotient ring of 1 by J. Since J is prime, 1
1
is a
domain. Let B be the quotient eld of 1
1
. We claim that / is canonically
isomorphic to a subeld of B. This is clear since [/[ J = 0 (since
[/[ J ,= 0 implies 1 J).
Let b
1
, . . . , b
n
[B[ correspond to x
1
, . . . , x
n
1, i.e. b
i
= [x
i
]
J
. Then
for any g 1, g(b
1
, . . . , b
n
) = 0 if and only if g J.
In particular, g
1
(b
1
, . . . , b
n
) = = g
m
(b
1
, . . . , b
n
) = 0 and f(b
1
, . . . , b
n
) ,=
0. Hence by model completeness (Corollary 6.3.7) we can nd a
1
, . . . , a
n
[/[ such that g
1
(a
1
, . . . , a
n
) = = g
m
(a
1
, . . . , a
n
) = 0 and f(a
1
, . . . , a
n
) ,=
0. Q.E.D.
3.
Given the statement of Hilberts Nullstellensatz, it is natural to ask whether a
bound on the exponent k can be computed. We can use G odels completeness
theorem 4.1.3 to show that this is the case:
1
The Nullstellensatz actually implies that J is a maximal ideal, i.e. 1/J is a eld.
54 CHAPTER 6. ALGEBRAICALLY CLOSED FIELDS
Theorem (eective bounds for Hilberts Nullstellensatz). We can nd a
recursive function K :
3
with the following property:
Let / be a eld and let f, g
1
, . . . , g
m
1 = /[x
1
, . . . , x
n
] be
such that every common algebraic zero of g
1
, . . . , g
m
is a zero of
f. If f, g
1
, . . . , g
m
are of degree d, then f
k
=
p
i
g
i
for some
k K(m, n, d) and some p
i
1 of degree K(m, n, d).
Proof. Let T be the theory of algebraically closed elds with a distinguished
subeld. Let
mndk
be a sentence in the language of T asserting that, for
all polynomials f, g
1
, . . . , g
m
of degree d in n variables x
1
, . . . , x
n
with
coecients in the distinguished subeld, if all common zeros of g
1
, . . . , g
m
are zeros of f, then there exists polynomials p
1
, . . . , p
m
of degree k in
variables x
1
, . . . , x
n
with coecients in the distinguished subeld, such that
f
l
=
p
i
g
i
for some l k.
We claim that m, n, d k
mndk
T. To see this, assume the contrary for
some xed m, n, d and consider a theory T
mndk
fails for this choice of f, g
1
, . . . , g
m
. By the compactness theorem, T
n
k=1
k
n
. For example [S
1
(T)[ = 1, [S
2
(T)[ = 3,
[S
3
(T)[ = 13, etc.
We shall see later that a theory T is
0
categorical if and only if n(S
n
(T)
is nite). Here T is assumed to be countable and complete and to have an
innite model.
7.1. ELEMENT TYPES 57
7.
Remark. Let T be a complete theory and let p be a complete ntype over
T, n 1. It is easy to see that every nite subset of p is realized in every
model of T. (Consider sentences of the form v
1
v
n
(
1
k
) where
1
, . . . ,
k
p.) However, p itself need not be realized in a given model of
T. For example, let T = T
0
= the theory of algebraically closed elds of
characteristic zero. Let p S
1
(T) be a 1type saying that v
1
is transcenden
tal, i.e. a
n
v
n
1
+ + a
1
v
1
+ a
0
,= 0 for a
i
Z, a
n
,= 0. Clearly every nite
subset of p is realized in every algebraically closed eld of characteristic 0.
But p itself is not realized in Q = the algebraic closure of the rational eld
Q = (Q, +, , , 0, 1).
8.
Proposition. Let / be an innite model of a complete theory T and assume
[T[ /.
(i) Given p S
n
(T) we can nd an elementary extension of / of the same
power as / in which p is realized.
(ii) / has an elementary extension of power max(/, [S
n
(T)[) in which
every p S
n
(T) is realized.
Proof. (i) Let c
1
, . . . , c
n
, a (a [/[) be new constant symbols. Since each
nite subset of p is realized n /, we have that
(elementary diagram of /) (c
1
, . . . , c
n
) : p
is nitely consistent. Hence by the compactness and LowenheimSkolem
theorems, this has a model of cardinality max(/, [p[) = /. This is an
elementary extension of / and the interpretation of c
1
, . . . , c
n
realizes p.
(ii) We use an elementary chain argument. Let S
n
(T) = p
: <
where = [S
n
(T)[. Put /
0
= /, /
+1
= an elementary extension of /
in which p
is realized; /
<
/
for limit
; nally B = /

max(/, [[). Hence B max(/, ). If need be, use the upward
L owenheimSkolem theorem to raise the cardinality of B to max(/, ).
58 CHAPTER 7. SATURATED MODELS
7.2 Saturated models
1.
Given a structure B = ([B[, ) and a set X [B[, introduce new constant
symbols a (a X) and let B
X
= ([B[,
X
) where
X
= (a, a) : a
X. Thus B
X
is just like B but with its signature expanded to include
constant symbols denoting the elements of X. Th(B
X
) is just the set of
all sentences (a
1
, . . . , a
n
) with a
1
, . . . , a
n
X, sig() sig(B), such that
B [= (a
1
, . . . , a
n
).
2.
Denition. Let be an innite cardinal. We say that B is saturated if
for all X [B[ of cardinality less than , every complete 1type over X is
realized in B.
(To be precise we should say: every complete 1type over Th(B
X
) is
realized in B
X
.)
3.
Exercise. Show that if B is saturated, then for all X [B[ of cardinality
< and all ntypes p over X (n 1), p is realized in B.
4.
Example. Let B = ([B[, <) be a linear ordering, and let be an innite
cardinal. We say that B is dense if for every pair of sets X
0
, X
1
[B[
of cardinality < , if X
0
< X
1
we can nd b [B[ so that X
0
< b < X
1
.
For example, (Q, <) is
0
dense (in our previous terminology, dense without
endpoints). Also, (R, <) is not
1
dense, as may be seen by taking X
0
= 0,
X
1
= 1/2, 1/4, 1/8, . . . .
Proposition. Let B be a dense linear ordering without end points. B is
dense if and only if B is saturated.
Proof. This is easy using the fact that the theory of dense linear orderings
without end points admits elimination of quantiers.
7.2. SATURATED MODELS 59
5.
We note that a nite structure is saturated for all innite . If an innite
structure B is saturated, then B .
6.
Denition. A structure B is saturated if it is Bsaturated.
For example, (Q, <) is
0
saturated, hence saturated.
7.
Proposition. (i) every uncountable algebraically closed eld is saturated.
(ii) Q and F
p
are not saturated. (iii) There exist countable saturated alge
braically closed elds of every characteristic.
Proof. Let X [B[ where B is an algebraically closed eld. A complete 1
type over X is virtually the same thing as a simple extension of the subeld
/ generated by X. We may therefore apply our classication of simple eld
extensions (see 6.1). The details are left to the reader.
8.
Theorem (uniqueness of saturated models). Let / and B be saturated
models of the same power. If / and B are elementarily equivalent, then they
are isomorphic.
Proof. We use a backandforth argument. Let / = B = and x
well orderings of [/[ and [B[ of order type . We shall dene enumerations
[/[ = a
)
<
(B, b
)
<
.
Stage < , even: We have inductively (/, a
)
<
(B, b
)
<
. Let a
be the least element of [/[ (with respect to the xed well ordering) dierent
from a
, < . Let
p
S
1
(Th((/, a
)
<
)) = S
1
(Th((B, b
)
<
))
60 CHAPTER 7. SATURATED MODELS
be the complete 1type over a
: < realized by a
. Since B is 
saturated we can nd b
[B[ realizing p
over b
: < . Thus
(/, a
(B, b
.
Stage < , odd: Reverse the roles of /, B.
Finally (/, a
)
<
(B, b
)
<
and these enumerations exhaust the uni
verses of [/[ and [B[ respectively since the elements were chosen by means
of xed well orderings of order type .
Thus a
is an isomorphism of / onto B.
9.
Theorem (universality of saturated models). Let B be a saturated model
and / B, / . Then there exists an elementary monomorphism of /
into B.
Proof. Similar to the previous proof.
10.
Example. For algebraically closed elds, theorems 8 and 9 have the fol
lowing signicance. Let p = 0 or a prime. There is exactly one countable
saturated algebraically closed eld. Every countable algebraically closed eld
of characteristic p is embeddable into this one.
7.3 Existence of saturated models
1.
We begin with an existence theorem for countable saturated models.
Theorem (Vaught). Let T be a complete countable theory. The following
are equivalent.
(1) T has a countable saturated model;
(2) [S
n
(T)[
0
for all n.
7.3. EXISTENCE OF SATURATED MODELS 61
Proof. (1) (2): Easy. Let / be the countable saturated model of T. Then
each p S
n
(T) is realized in /. Hence S
n
(T) is countable.
(2) (1): Note rst that (2) implies
(3) for all nite X [B[, B Mod(T), one has [S
1
(X)[ X
0
.
We shall prove (3) (1) by an elementary chain argument.
Stage 0. Let /
0
be any countable model of T.
Stage n + 1. We have a countable /
n
Mod(T). Let X
i
n
: i
be an enumeration of the nite subsets of [/
n
[. Put /
0
n
= /
n
, /
i+1
n
=
a countable elementary extension of /
i
n
in which all complete 1types over
X
i
n
are realized. (/
i+1
n
exists because S
1
(X
i
n
) is countable.) Finally put
/
n+1
=
i
/
i
n
. This completes stage n.
Finally put B =
n
/
n
. We claim that B is
0
saturated. Let X B
be nite. Then X [/
n
[ for some n. Hence X = X
i
n
for some i. Hence
every 1type over X is realized in /
i+1
n
. This is an elementary submodel of
B, so these types are realized in B, Q.E.D.
2.
Example. We give an example of a countable complete theory T with no
countable saturated model. Let T = Th(Q
Q
) where Q = (Q, <). Then
[S
1
(T)[ = 2
0
because each Dedekind cut in (Q, <) gives rise to a dierent
complete 1type over T. (There are other complete 1types over T which do
not correspond to Dedekind cuts.)
3.
The basic theorem on the existence of uncountable saturated models is the
following:
Theorem (Morley and Vaught). Let / be an innite model and let be an
innite cardinal such that sig(/). Then /has a
+
saturated elementary
extension of power /
.
Proof. We use an elementary chain argument as in the proof of the previous
theorem. Put = /
= . We construct an elementary
chain in which each model has power .
Stage 0: Let /
e
/
0
where /
0
 = .
62 CHAPTER 7. SATURATED MODELS
Stage + 1, <
+
: We have /
of power . Let X
: < be an
enumeration of all X [/
= /
; /
+1
an elementary extension of /
)[ 2
); and /
<
/
for
limit . Finally put /
+1
= /
.
Stage
+
, limit: Put /
<
/
.
Finally put B = /
+. Clearly /
e
B and B = . We claim B is
+
saturated. Suppose X [B[, [X[ . Since
+
is a regular cardinal,
X /
e
B. This completes the proof.
4.
Corollary. Let T be a complete theory of cardinality , and suppose that
T has an innite model. Then T has a
+
saturated model of power 2
.
Proof. By LowenheimSkolem let / be a model of T of power . Apply
the previous theorem to get a
+
saturated elementary extension of power
= 2
.
5.
Corollary. Assume the Generalized Continuum Hypothesis. Let T be a
complete theory of cardinality . Then T has a saturated model of power
+
. This model is unique up to isomorphism.
Proof. The G.C.H. tells us that 2
=
+
. Apply the previous corollary and
the uniqueness result for saturated models (7.2).
6.
The previous corollary may be applied to give the following useful variant of
Vaughts test. Unlike Vaughts test itself, the following provides a necessary
and sucient condition for completeness.
Theorem. Assume 2
0
=
1
. Let T be a countable complete theory with
no nite models. T is complete any two saturated models of T of power
1
are isomorphic.
Proof. Immediate from the previous corollary.
7.4. PRESERVATION THEOREMS 63
7.
Remark. The G.C.H. is needed for the above results. For example, let T
be the complete theory of dense linear ordering without end points. Then
any
1
saturated model of T has cardinality 2
0
. Thus we do not get
satisfactory results unless we assume 2
0
=
1
.
However, the G.C.H. can be eliminated from most applications of satu
rated models, by observing that the conclusions are usually absolute. Thus,
if we are trying to prove that a certain theory T is complete, we may as
sume G.C.H., apply the variant of Vaughts test referred to above, and then
eliminate G.C.H. by noting that completeness is an absolute property of T.
Actually, there are several ways to avoid the need to assume the G.C.H.
(1) Assume G.C.H. and then eliminate it by absoluteness arguments.
(2) Assume the existence of inaccessible cardinals (some people might con
sider this assumption more reasonable than G.C.H.).
(3) Use special models (MorleyVaught).
(4) Use recursively saturated models (BarwiseSchlipf).
We shall simply assume the G.C.H. when needed.
7.4 Preservation theorems
In this section we present a typical application of saturated models, namely
to the proofs of preservation theorems. Preservation theorems relate model
theoretic properties of T to syntactic properties of axioms for T.
1.
The following syntactical classication of formulas is occasionally useful.
Denition. A formula (y) is
k
if it is of the form
(y) = x
1
x
2
x
k
(x
1
, x
2
, . . . , x
k
, y)
where is quantier free and x
i
= x
i1
. . . x
in
i
. Thus a
k
formula consists of
k alternating blocks of quantiers followed by a quantier free matrix, and
the rst block is existential. The class of
k
formulas is dened similarly
except that the rst block is universal.
64 CHAPTER 7. SATURATED MODELS
2.
Let be a class of formulas (e.g. =
k
or
k
). A theory is a theory T
such that T is a set of axioms for T. We write /
3
B to mean that
sig(/) = sig(B) and Th(/) Th(B) .
3.
Lemma (localization lemma). Let be a class of formulas which is closed
under disjunction. Let T be a theory such that Mod(T) is closed under
3
, i.e. / Mod(T), /
3
i
for some
1
, . . . ,
k
. Hence
_
k
i=1
i
T . Hence B [=
_
k
i=1
i
, a contradiction.
4.
If sig(/) = sig(B), we write f : /
/
B (f is a morphism of / into B)
to mean that f : [/[ [B[ and f preserves satisfaction of formulas, i.e. for
all formulas (x
1
, . . . , x
n
) and a
1
, . . . , a
n
[/[, if / [= (a
1
, . . . , a
n
) then
B [= (f(a
1
), . . . , f(a
n
)).
Examples:
(1) If =
0
=
0
= quantier free formulas, then f : /
/
B means
that f is a monomorphism of / into B.
(2) If = atomic formulas, then f : /
/
1
/ , then / Mod(T). Let B B
where B
is saturated, =
/. We have /
3
1
B , hence /
3
1
B
.
Claim: There exists a monomorphism of / into B
[B
[,
< , so that the embedding a
is a
1
morphism. Stage < : Let
Y
be the set of
1
formulas (a
1
, . . . , a
n
, v
1
),
1
< <
n
< , such
that / [= (a
1
, . . . , a
n
, a
)
<
3
1
(B
, b
)
< .
Hence B
[= v
1
(b
1
, . . . , b
n
, v
1
) for each (a
1
, . . . , a
n
, v
1
) Y
. Hence
by saturation there exists b
[B
[ such that B
[= (b
1
, . . . , b
n
, b
) for
all (a
1
, . . . , a
n
, v
1
) Y
.
[ Details: To show that Y
is realized in (B
, b
)
<
it suces to show
that each nite subset is, i.e. (B
, b
)
<
[= v
_
k
i=1
i
(b
1
, . . . , b
n
, v) for all
1
, . . . ,
k
Y
. But this is a
1
sentence, true in (/, a
)
<
, hence true in
(B
, b
)
<
. ]
Now clearly
(/, a
1
(B
, b
)
<
3
1
(B
, b
)
<
so in particular a
is a monomorphism of / into B
B and B is saturated,
/, then f : /
/
B .
We can use this lemma for =
2
to prove the following preservation
theorem.
7.
Theorem (Chang LosSusko). Let T be a theory such that Mod(T) is closed
under unions of chains, i.e. /
0
/
n
(n ), /
n
Mod(T)
implies
n
/
n
Mod(T). Then T is a
2
theory. (The converse is obvi
ous.)
Proof. By the localization lemma, in order to show that T is a
2
theory, it
suces to show that if / Mod(T) and /
3
2
B then B Mod(T).
We have B
3
2
/ Mod(T). Let / /
1
where /
1
is Bsaturated.
Then by the embedding lemma we can nd a
2
embedding f : B
/
2
/
1 .
Let B
e
B
1
, B
1
/
1
saturated. Then
(/
1
, f(b))
bB
3
2
(B
1
, b)
bB
so by the embedding lemma again we can nd an embedding g : /
1
/
1
B
1
such that the following diagram commutes:
/
1
1
B
?
2
~
~
~
~
~
~
~
~
e
B
1
Repeating times we get
/
1
1
/
2
1
/
n+1
1
B
=
B
0
>
2
}
}
}
}
}
}
}
}
e
B
1
>
2








e
B
2
e
e
B
n
<
2
y
y
y
y
y
y
y
y
e
B
n+1
e
.
7.4. PRESERVATION THEOREMS 67
So nally B
n
B
n
=
n
/
n
. Since /
n
Mod(T) we have by
assumption
n
/
n
Mod(T). Since B
e
B
, B B
where /
 = B
 =
and /
, B
. We claim that B
is a homomorphic image of /
.
The proof is similar to that of the embedding lemma (7.4.6), except we need
a backandforth argument.
[ Details: Choose well orderings of [/
[ and [B
[ of order type . We
construct enumerations [/
[ = a
: < , [B
[ = b
: < .
Stage , even: We have
(/
, a
)
<
3
pos
(B
, b
)
< .
Let a
[ not among a
: < . Let Y
be the set
of positive formulas (a
1
, . . . , a
n
, v) such that /
[= (a
1
, . . . , a
n
, a
). As
in the proof of the embedding lemma, we have /
[= v
_
k
i=1
i
(a
1
, . . . , a
n
, v),
hence B
[= v
_
k
i=1
i
(b
1
, . . . , b
n
, v) for each nite set of formulas
1
, . . . ,
k
Y
. Hence by saturation of B
it follows that Y
is realized in (B
, b
)
<
,
i.e. there exists b
[B
[ such that
(/
, a
3
pos
(B
, b
)
.
Stage , odd: We have
(/
, a
)
<
3
pos
(B
, b
)
< ,
hence
(B
, b
)
<
3
neg
(/
, a
)
< ,
where a negative formula is dened to be the negation of a positive one.
Let b
[ not among b
: < . According to
the DeMorgan laws, the class of negative formulas is closed under and .
Hence we can apply the same argument as before to nd a
[/
[ such that
(B
, b
3
neg
(/
, a
)
,
i.e.
(/
, a
3
pos
(B
, b
)
.
Note that a
,= a
, < , since v ,= a
is a negative formula.
7.4. PRESERVATION THEOREMS 69
Finally we get enumerations [/
[ = a
: < and [B
[ = b
: <
such that < < a
,= a
and
(/
, a
)
<
3
pos
(B
, b
)
< .
Hence in particular the mapping a
is a homomorphism of /
onto
B
. ]
Since / /
Mod(T).
Hence B Mod(T), Q.E.D.
For more on preservation theorems, see Chang and Keisler, Model Theory.
70 CHAPTER 7. SATURATED MODELS
Chapter 8
Elimination of quantiers
8.1 The model completion of a theory
1.
Denition (Robinson). Let T and T
is a model completion of T if
(i) T T
;
(ii) every model of T can be extended to a model of T
;
(iii) for any / Mod(T), T
(diagram of /) is complete.
(Recall that the diagram of / is the set of all quantier free sentences
(a
1
, . . . , a
n
), sig() sig(/), a
1
, . . . , a
n
[/[, / [= (a
1
, . . . , a
n
).)
Condition (iii) is equivalent to
(iii) if / Mod(T), / B
i
Mod(T
), i = 1, 2, then B
1
and B
2
are
elementarily equivalent over /.
B
1
B
2
/
`A
A
A
A
A
A
A
>
}
}
}
}
}
}
}
71
72 CHAPTER 8. ELIMINATION OF QUANTIFIERS
2.
Example. The theory of algebraically closed elds is a model completion of
the theory of elds.
Proof. (i) is trivial, and (ii) holds since every eld can be extended to an
algebraically closed eld (Proposition 6.2.3). For (iii), let / be a eld and
let B
1
and B
2
be two algebraically closed elds extending /. We want to
show that B
1
B
2
over /. By the upward L owenheimSkolem theorem let
B
i
e
B
i
, i = 1, 2, where B
1
 = B
2
 = and > max(
0
, /). By
Lemma 6.3.1 we have that B
= B
2
over /. Hence B
1
B
2
over /.
B
= B
2
B
1
O
e
B
2
O
e
/
`A
A
A
A
A
A
A
A
>
}
}
}
}
}
}
}
}
3.
Remarks. The following facts about model completions are obvious.
(a) If T
is model complete.
(b) T is model complete if and only if T is its own model completion.
4.
Theorem (Robinson). Let T
0
, T
1
be model completions of T. Then T
0
= T
1
.
Proof. Given / Mod(T
0
), we shall show that / Mod(T
1
). We use an
elementary chain argument. Let / = /
0
Mod(T
0
). By (i) and (ii) of the
denition, we can nd /
1
Mod(T
1
), /
0
/
1
. More generally, given /
2n
Mod(T
0
) let /
2n
/
2n+1
Mod(T
1
) and /
2n+1
/
2n+2
Mod(T
0
). Put
B =
n
/
n
. Since T
0
is model complete, we have /
2n
e
/
2n+2
so by the
elementary chain principle, / = /
0
e
B. But since T
1
is model complete,
we also have /
2n+1
e
/
2n+3
so B Mod(T
1
). Hence / Mod(T
1
), Q.E.D.
Thus the model completion of T is unique if it exists.
8.2. SUBSTRUCTURE COMPLETENESS 73
4.
Another example. The theory of dense linear order without end points is the
model completion of the theory of linear order. (This is an easy consequence
of quantier elimination.)
8.2 Substructure completeness
1.
Denition. A theory T is substructure complete if T (diagram of /) is
complete whenever / is a substructure of a model of T.
2.
Denition. T admits elimination of quantiers if for all n 1 and all
F
n
(T), there exists a quantier free formula
F
n
(T) such that
T [= v
1
. . . v
n
[(v
1
, . . . , v
n
)
(v
1
, . . . , v
n
)].
3.
Theorem. Let T be a theory. The following are equivalent.
(i) T is the model completion of a universal (i.e.
1
) theory.
(ii) T is substructure complete.
(iii) T admits elimination of quantiers.
Proof. (i) (ii). Let T be the model completion of a universal theory U.
Let / B Mod(T). Then / Mod(U) since U is universal. Hence
T (diagram of /) is complete, by the denition of model completion.
(ii) (iii). Assume that T is substructure complete. Given F
n
(T),
to nd a quantier free
F
n
(T) such that T
[=
. We proceed as
in the proof of the localization lemma 7.4.3. Let Y be the set of quantier
free F
n
(T) such that T [= . If
a
0
a
1
. . . a
m
a
0
a
1
. . . a
m
a
0
a
1
. . . a
m
b
0
b
1
. . . b
n
b
0
b
1
. . . b
n
b
0
b
1
. . . b
n
[b
], /
<
/
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
B
1
/
e
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
B
1
/
e
/
?
~
~
~
~
~
~
~
~
@
@
@
@
@
@
@
@
B
2
/
e
F
2
/
e
F
2
/
e
First choose B
2
e
B
2
suciently saturated, and apply (ii) to get B
1
B
2
over /. Then choose B
1
e
B
1
suciently saturated and apply (ii) to get
B
2
B
1
over /. Etc. Finally we get
B
1
/
e
B
1
=
n
B
(n)
1
/
?
>
>
>
>
>
>
>
>
=
B
2
/
e
B
2
=
n
B
(n)
2
and the diagram gives an isomorphism of B
1
onto B
2
over /. Hence B
1
B
2
over /.
78 CHAPTER 8. ELIMINATION OF QUANTIFIERS
3.
Example. We can use the above criterion to prove (using a minimum of
algebra) that the theory of algebraically closed elds is the model completion
of the theory of domains. Suppose / B, B algebraically closed and
+

saturated, where max(
0
, /). We must show that B contains every
simple extension /[b] of the domain /. All such extensions are contained
in simple eld extensions of the quotient eld of /. Using our classication
of simple eld extensions (6.1) we see immediately that B contains all such
extensions.
In the next chapter we shall use this criterion to show that the theory of
real closed ordered elds is the model completion of the theory of ordered
elds.
Chapter 9
Real closed ordered elds
9.1 Ordered elds
1.
The axioms for ordered elds are as follows:
(a) the eld axioms (as in example 2.3.4)
(b) the axioms for linear order:
xyz((x < y y < z) x < z)
xy(x < y x ,= y)
xy(x < y x = y y < x)
(c) special axioms:
xyz(x < y x + z < y + z)
xy((0 < x 0 < y) 0 < x y)
An ordered eld is a model of these axioms. An ordered domain is dened
similarly.
2.
Examples.
Q = (Q, +, , , 0, 1, <) = the ordered eld of rationals.
79
80 CHAPTER 9. REAL CLOSED ORDERED FIELDS
1 = (R, +, , , 0, 1, <) = the ordered eld of real numbers.
Z = (Z, +, , , 0, 1, <) = the ordered eld of rationals.
3.
The following facts are easily proved.
(a) In an ordered domain, we have
x < y y x > 0
x > 0 x < 0
x ,= 0 x
2
> 0
(x < y z > 0) x z < y z
(b) Any ordered eld (or ordered domain) has characteristic 0. (Because
1 + 1 + + 1
. .
n
> 0.)
(c) The ordering of an ordered eld is dense (because x < y implies x <
(x + y)/2 < y).
(d) Any ordered eld contains Q as an ordered subeld. Any ordered
domain contains Z as an ordered subdomain.
4.
Proposition. Let / be an ordered domain and let /
1
be its eld of quo
tients. Then there is one and only one way to order /
1
so that it becomes
an ordered eld with / as an ordered subdomain.
Proof. We are forced to order /
1
by
x
y
> 0 x y =
x
y
y
2
> 0 .
It is easy to check that this makes /
1
an ordered eld, etc.
For example, there is only one way to order the rational eld so as to
make it an ordered eld.
9.1. ORDERED FIELDS 81
5.
Denition. Let T be an ordered eld and let f T[x]. We say that
T has the intermediate value property (IVP) for f if a, b [/[, a < b,
f(a) < 0 < f(b) imply f(c) = 0 for some c [T[, a < c < b.
We say that T is real closed if it has the IVP for all f T[x].
6.
Examples. The rational eld Q does not have the IVP for f(x) = x
2
2.
The real eld 1 does have IVP for x
2
2. The real eld 1 is in fact real
closed. The complex eld C is not real closed, indeed it cannot be ordered
(since i
2
= 1).
7.
The property of being real closed is rst order, so we may speak of the theory
of real closed ordered elds, RCOF. The axioms of RCOF are those for ordered
elds plus the following, for each n 1:
w
0
w
n
xy [(x < y w
n
x
n
+ + w
0
< 0 < w
n
y
n
+ + w
0
)
z (x < z < y w
n
z
n
+ + w
0
= 0)] .
8.
Our main goal is to show that RCOF is the model completion of the theory
of ordered elds, and hence of ordered domains. From this it will follow
that RCOF admits elimination of quantiers, and is complete and decidable
(Tarski).
9.
Lemma. Let T be an ordered eld which is not real closed. Then there is
an ordered eld T
1
T such that T
1
is algebraic over T.
Proof. Pick f T[x] such that the IVP fails for f but holds for all g T[x] of
smaller degree. Let a, b T[x] be such that a < b, f(a) < 0 < f(b), f(c) ,= 0
for all c [a, b] = c [T[ : a c b. We employ a sort of Dedekind
cut construction. Put A = a
] and
82 CHAPTER 9. REAL CLOSED ORDERED FIELDS
B = [a, b] A. Obviously A, B ,= , [a, b] = A B, and a
< b
for all a
A,
b
B.
Claim: A has no greatest element and B has no least element.
Proof: This is clear since f is continuous (by an argument).
Claim: For every g T[x] of degree < deg(f) we can nd a
A, b
B
so that g has constant sign (positive, negative, or zero) on the interval [a
, b
].
Proof: If g 0 there is nothing to prove. Otherwise g has only nitely
many roots in [T[. Find a
A, b
B such that [a
, b
, b
].
Claim: f is irreducible. For suppose f = f
1
f
2
, deg(f
i
) < deg(f), i = 1, 2.
We must have f
i
(a)f
i
(b) < 0 for at least one i. Then by the IVP for f
i
we
must have f
i
(c) = 0 for some c [a
, b
, b
], a
A, b
, b
], a
A,
b
B; g() = 0 if g 0.
Claim: this makes T() an ordered eld. The only nontrivial axiom to
be checked is the product rule: the product of positive elements is positive.
Suppose g
i
() > 0, deg(g
i
) < deg(f), i = 1, 2. Let g
1
() g
2
() = g
3
(),
deg(g
3
) < deg(f). We must show g
3
() > 0. We have
g
1
(x) g
2
(x) g
3
(x) = f(x) h(x) .
Put m = deg(f). The left side has degree 2m2. Hence so does the right
side, so deg(h) m2. Let a
A, b
B be such that g
1
, g
2
, g
3
, h all have
constant sign on [a
, b
].
Case 1: h > 0 on [a
, b
]. Let c [a
, b
, b
]. Hence g
3
() > 0.
Case 2: h < 0 on [a
, b
]. Take c [a
, b
, b
]. Hence h 0 so g
3
= g
1
g
2
> 0 on [a
, b
], hence
g
3
() > 0. This completes the proof.
9.2. UNIQUENESS OF REAL CLOSURE 83
10.
Denition. Let T be an ordered eld. A real closure of T is an ordered
eld ( T so that (1) ( is algebraic over T; (2) ( is real closed.
11.
Proposition. Any ordered eld T has a real closure.
Proof. This is an easy consequence of the previous lemma plus Zorns lemma
(or transnite induction). Let B be the algebraic closure of T regarded as
a eld. Form a transnite sequence of extensions T = T
0
, T
+1
= proper
ordered extension of T
, T
<
T
.
(Here g
(x) = na
n
x
n1
+ + a
1
.)
Proof. Given a < b, g(a) = g(b) = 0. Say g(x) = (x a)
k
(x b)
l
p(x) where
p ([x] and p(a) ,= 0, p(b) ,= 0. We may safely assume that g(c) ,= 0 for
a < c < b. Hence p(c) ,= 0 for c [a, b]. Hence p has constant sign on [a, b].
We have
g(x) = (x a)
k1
(x b)
l1
q(x)
where
q(x) = [k(x b) + l(x a)]p(x) + (x a)(x b)p
(x) .
84 CHAPTER 9. REAL CLOSED ORDERED FIELDS
We have q(a) = k(ab)p(a), q(b) = l(ba)p(b) so q(a) and q(b) have opposite
sign, hence q(c) = 0 for some a < c < b. Hence g
(c) = 0.
2.
Lemma (maximum principle). Let ( be a real closed ordered eld, g ([x],
a, b (, a < b. Then g takes on a maximum value in the interval [a, b], say
at . If a < < b then g
() = 0.
Proof. If g is a linear function there is nothing to prove. Otherwise g
has
only nitely many roots in the interval [a, b]. Let
a = c
0
< c
1
< < c
k
< c
k+1
= b
be the set consisting of a, b, and all the roots of g
() = 0 for
some [([ with M < < M. By the induction hypothesis, g
() = 0
gives [T[. But then g() 0 contradicts the fact that g < 0 on [T[.
9.2. UNIQUENESS OF REAL CLOSURE 85
4.
Proposition. If ( is a RCOF and (
1
is an ordered eld extension of ( which
is algebraic over (, then (
1
= (.
Proof. Immediate from the previous lemma.
5.
Proposition. Let T be an ordered eld. Then any two real closures of T
are isomorphic over T.
Proof. Let (
1
and (
2
be real closures of T. If T is real closed, then (
1
= (
2
=
T by the previous proposition. Therefore, by Zorns lemma or transnite
induction, it suces to show that if T is not real closed, then we can nd
i
[(
i
[ [T[, i = 1, 2, such that T(
1
)
= T(
2
) over T as ordered elds.
In order to nd
1
,
2
we imitate the proof of lemma 9.1.9. Let f T[x]
be such that the IVP fails for f but holds for all g of lower degree. Let
a, b, A, B be as in the proof of lemma 9.1.9. Since (
1
is real closed, [(
1
[ [T[
contains
1
such that f(
1
) = 0 and a
<
1
< b
for all a
A, b
B.
We claim that these facts determine the structure of T(
1
) as an ordered
eld. First of all, the eld structure of T(
1
) is determined by the fact that
T(
1
)
, b
] in T where a
A, b
B. By
the previous lemma we know that any root of g lying in [(
1
[ already lies in
[T[. Hence g > 0 on [a
, b
] in (
1
. Hence in particular g(
1
) > 0. Since every
element of T(
1
) is of the form g(
1
), g T[x], deg(g) < deg(f), our claim
is proved. Similarly we can nd
2
[(
2
[ [T[ such that f(
2
) = 0 and
a
<
2
< b
for all a
A, b
= T(
2
) over T
as ordered elds. This completes the proof.
6.
Denition. If T is an ordered eld, we denote its real closure by T. (The
previous result shows that T is unique up to isomorphism over T.)
86 CHAPTER 9. REAL CLOSED ORDERED FIELDS
7.
Corollary. Let T be an ordered eld and let ( be a RCOF extending T.
Then ( contains (an isomorphic copy of) T.
Proof. Let T
1
be the algebraic closure of T within (. Clearly T
1
is a real
closure of T. By the previous result it follows that T is isomorphic to T
1
over T.
9.3 Quantier elimination for RCOF
1.
Lemma. Let ( be an RCOF and let (() be a simple ordered eld extension
of (, / [([. Then is transcendental over (, and the ordering of ( is
completely determined by the sets A = a [([ : a < and B = b [([ :
< b.
Proof. The fact that is transcendental over ( follows immediately from
the uniqueness of the real closure of (. The ordering of (() is completely
determined by that of ([] which is completely determined by the following
claim.
Claim: Let g ([x].
(i) If A = (respectively B = ) then the sign of g() (respectively
g()) is the same as that of the leading coecient of g.
(ii) If A ,= , = B, then g() > 0 if and only if there exist a A, b B
such that g(c) > 0 for all c [([, a < c < b.
Part (i) is obvious, and part (ii) is an easy consequence of lemma 9.2.3. (One
may compare the proof of lemma 9.1.9, but note that in the present situation
it is possible that A has a greatest element.)
2.
Theorem. The theory of real closed ordered elds is the model completion
of the theory of ordered elds.
9.3. QUANTIFIER ELIMINATION FOR RCOF 87
Proof. It suces to show that if T T
1
are ordered elds and if T ( where
( is a real closed ordered eld and ( is saturated, = T
1
, then T
1
is
embeddable into ( over T. (Cf. theorem 8.3.2.) By transnite induction we
may safely assume that either (1) T
1
is algebraic over T, or (2) T
1
= T(),
transcendental over T, / [T[, and T is real closed.
In case (1), we have T
1
T and by corollary 9.2.7 T is embeddable into
( over T. In case (2), put A = a [T[ : a < and B = b [T[ : < b.
Consider the set of formulas
a < v : a A v < b : b B f(v) ,= 0 : f T[x], f , 0 .
This set of formulas is nitely realizable in T. Hence by saturation it is
realized in (, say by [([. By the previous lemma it follows that T() is
order isomorphic to T() over T. This completes the proof.
3.
Corollary (Tarski). The theory of real closed ordered elds is complete and
decidable.
Proof. Any RCOF has an ordered subeld isomorphic to
Q = (Q, +, , , 0, 1, <) .
By the previous theorem, any two RCOFs are elementary equivalent over Q.
In particular any two RCOFs are elementarily equivalent, so the theory RCOF
is complete. Decidability follows by theorem 4.2.3.
(Tarski actually gave a decision procedure based on a specic explicit
quantier elimination procedure.)
4.
Corollary (Tarskis transfer principle). Any rst order sentence which is
true in the real eld 1 = (R, +, , , 0, 1, <) is true in any real closed ordered
eld.
5.
Examples. (1) The fact that the complex eld ( is algebraically closed
(i.e. the fundamental theorem of algebra) is expressible as a set of rst order
88 CHAPTER 9. REAL CLOSED ORDERED FIELDS
sentences in the theory of RCOF. (We identify ab
1 C with an ordered
pair (a, b) R
2
.) It follows by Tarskis principle that, if ( is any RCOF, then
((
1) = ([x]/(x
2
+ 1) is algebraically closed.
(2) Bott and Milnor have used homotopy theory to prove that the only
nite dimensional nonassociative algebras over R are those of dimensions
1, 2, 4, and 8 respectively (i.e. R, C, quaternions, octonians). By Tarskis
transfer principle, the same holds for any RCOF.
6.
Corollary (Tarski). The rst order theory of the real eld
1 = (R, +, , , 0, 1, <)
is decidable.
(In other words, there is a decision procedure for high school mathematics,
including plane and solid geometry since the latter can be interpreted into
the rst order theory of 1 using Cartesian coordinates.)
7.
Corollary (Tarksi). Given a nite system S of equations and inequalities
in n variables with rational coecients, we can decide recursively whether S
has a solution in reals.
8.
Theorem (Tarksi). The theory of real closed ordered elds admits elimina
tion of quantiers.
Proof. From proposition 9.1.4 and theorem 9.3.3 it follows that RCOF is
the model completion of the theory of ordered domains. The latter theory
is universal so by theorem 8.2.3 it follows that RCOF admits elimination of
quantiers.
9.
Corollary (Tarski). The theory of real closed ordered elds is model com
plete.
9.4. THE SOLUTION OF HILBERTS 17TH PROBLEM 89
10.
Corollary (Artin). Let S be a nite system of equations and inequalities
in n unknowns with coecients in an ordered eld T. If S has a solution
in some ordered eld extension of T, then S has a solution in T, the real
closure of T.
Proof. Immediate from model completeness plus corollary 9.2.7.
In the next section we shall apply this result to obtain Artins solution of
Hilberts 17th problem.
9.4 The solution of Hilberts 17th problem
1.
Let T be an ordered eld, T its real closure. Put B = T(x
1
, . . . , x
n
) = the
eld of rational functions over T in n variables x
1
, . . . , x
n
. Hilberts 17th
problem is concerned with the representation of rational functions as sums
of squares.
Theorem (Artin). Let f [B[, f , 0. Suppose that f(a
1
, . . . , a
n
) 0 for
all a
1
, . . . , a
n
[T[ such that f(a
1
, . . . , a
n
) is dened. Then there exists a
positive integer k such that
f =
k
i=1
c
i
g
2
i
where c
i
[T[, c
i
> 0, g
i
[B[, g
i
, 0.
Proof. Let I [B[ be the set of all f of this form. We have
(i) g
2
I for all g [B[, g , 0
(ii) g, h I implies g + h, g h I
(iii) 0 / I.
90 CHAPTER 9. REAL CLOSED ORDERED FIELDS
A set I [B[ satisfying (i), (ii) and (iii) is called an order ideal.
The following lemma holds for any order ideal in any eld B.
Lemma. Let I be an order ideal, f / I, f , 0. Then there exists an order
ideal J I such that f J.
Proof of lemma. Assume f ,= 0. Let J be the set of elements of [B[ of the
form
p(f) = (f)
m
1
g
1
+ + (f)
m
l
g
l
where l 1, g
i
I, m
i
0. Clearly f J and J satises (i) and (ii).
Suppose (iii) fails for J, say p(f) = 0. We then have
p(f) = q(f
2
) f r(f
2
) = 0
where q(f
2
) and r(f
2
) I or = 0. If r(f
2
) = 0 then q(f
2
) = 0 so both are 0.
This is impossible since l 1. Hence r(f
2
) ,= 0. Hence
f =
q(f
2
)
r(f
2
)
= q(f
2
) r(f
2
)
_
1
r(f
2
)
_
2
I
since f ,= 0. This proves the lemma.
Now to prove the theorem, suppose f / I, f ,= 0. By the lemma, get an
order ideal J I containing f. By Zorns lemma let K J be a maximal
order ideal. Then K satises (i), (ii), (iii) and also
(iv) for all g ,= 0 either g K or g K.
This is immediate from the lemma, by maximality of K.
For g, h [B[ dene g < h if and only if h g K. It is easy to check
that this denition makes B an ordered eld. Furthermore, if c [T[ is
positive in T, then it is positive in B. So B is an ordered eld extension of
T.
We see now that there exist b
1
. . . , b
n
lying in an ordered eld extension of
T, such that f(b
1
, . . . , b
n
) < 0 (namely b
i
= x
i
lying in B). Hence by model
completeness of RCOF (see also corollary 9.3.10), it follows that there exist
a
1
, . . . , a
n
[T[ such that f(a
1
, . . . , a
n
) < 0. Q.E.D.
9.4. THE SOLUTION OF HILBERTS 17TH PROBLEM 91
2.
Corollary (Artins solution of Hilberts 17th problem). Let T be either the
real eld R or the rational eld Q. Suppose f(x
1
, . . . , x
n
) T(x
1
, . . . , x
n
)
is denite over T, i.e. f(a
1
, . . . , a
n
) 0 for all a
1
, . . . , a
n
[T[ such that
f(a
1
, . . . , a
n
) is dened. Then there exist g
1
, . . . , g
k
in T(x
1
, . . . , x
n
) such
that
f =
k
i=1
g
2
i
.
Proof. For the reals this is immediate from the previous theorem. For the
rationals, it is immediate from the theorem plus the following two observa
tions: (1) Q is dense in R; (2) every positive element of Q is the sum of 4
squares in Q.
3.
Remark. An ordered eld T is said to be Hilbertian if (1) T is dense in
T, (2) every positive element of T is a sum of squares in T. For example,
T = reals, T = rationals, T = any RCOF are Hilbertian. Clearly the above
corollary holds for any Hilbertian ordered eld. McKenna
1
has proved the
converse: If the corollary holds for T, then T is Hilbertian.
4.
We now derive a corollary concerning eective bounds, analogous to theorem
6.4.3.
Corollary. We can nd a recursive function K :
2
with the following
property:
Let T be an ordered eld, and let f T(x
1
, . . . , x
n
) be such that
f(a
1
, . . . , a
n
) 0 for all a
1
, . . . , a
n
T such that this is dened.
If f is of degree d then f =
k
i=1
c
i
g
2
i
for some k K(n, d),
c
i
T, c
i
0, g
i
T(x
1
, . . . , x
n
) of degree K(n, d).
1
SLNM 498, pp. 220230.
92 CHAPTER 9. REAL CLOSED ORDERED FIELDS
Proof. Let T be the theory of real closed ordered elds with a distinguished
subeld, T. Let
ndm
be the sentence asserting that the desired conclusion
holds with K(n, d) replaced by m. We have nd m
ndm
T (otherwise
we could use compactness to construct a counterexample to theorem 9.4.1).
So put K(n, d) = least m such that
ndm
T. This is recursive since T is
decidable.
5.
Remark. Ax (unpublished) and Pster (1967)
2
have shown that for real
closed T we can always get k 2
n
(but apparently with no bound on the
degrees). There are some open problems in this area.
6.
Exercise. Prove the following result which is known as the Nullstellensatz
for ordered elds.
Theorem (Dubois
3
). Let f, h
1
, . . . , h
m
T[x
1
, . . . , x
n
] be polynomials in
n variables over an ordered eld T. Suppose that every common zero of
h
1
, . . . , h
m
in T, the real closure of T, is a zero of f. Then there exist
nonnegative integers r and k such that
f
2r
+
k
i=1
c
i
g
2
i
=
m
j=1
p
j
h
j
where c
i
[T[, c
i
0, g
i
, p
j
T[x
1
, . . . , x
n
].
Deduce a version with eective bounds.
2
Proc. Symp. Pure Math. 28, pp. 483491.
3
Ark. Mat. 8 (1969), pp. 111114. See also Prestels monograph.
Chapter 10
Prime models (countable case)
10.1 The omitting types theorem
1.
Denition. Let p be an ntype over a theory T. A model / of T is said to
omit p if there is no ntuple a
1
, . . . , a
n
[/[ which realizes p.
We have already seen how to use the compactness theorem to construct a
model which realizes p. It is somewhat more dicult to construct a model
which omits p. Indeed, such a model may not even exist. The omitting types
theorem gives a sucient condition for the existence of a model of T which
omits p. (If T is complete, this sucient condition is also necessary.)
2.
Denition. We say that p is principal over T if it is generated by a single
formula, i.e. there exists F
n
(T) such that p = F
n
(T) : T [=
v
1
v
n
((v
1
, . . . , v
n
) (v
1
, . . . , v
n
)). Such a is called a generator of p.
We say that p is essentially nonprincipal over T if p is not included in any
principal ntype over T.
3.
Theorem (the omitting types theorem). Let T be a countable theory and
let p be an ntype over T. Suppose that p is essentially nonprincipal over T.
93
94 CHAPTER 10. PRIME MODELS (COUNTABLE CASE)
Then there exists a model of T which omits p.
Proof. We go back to the Henkin proof of the completeness theorem. Let
C = c
i
: i be a countable set of new constant symbols. Let
s
:
s be an enumeration of all sentences with sig() sig(T) C. Let
S
(v) : s be an enumeration of all formulas (v) with no free variables
other than v, such that sig((v)) sig(T) C. Let c
s
1
, . . . , c
s
n
) : s be
an enumeration of all ntuples of constant symbols from C.
Stage 0: T
0
= T.
Stage 3s + 1: Put T
3s+1
= T
3s
s
if this is consistent, otherwise
T
3s+1
= T
3s
s
.
Stage 3s + 2: Let h(s) be the least i such that c
i
does not appear in
T
3s+1
s
(v). Put T
3s+2
= T
3s+1
(v
s
(v))
s
(c
h(s)
).
Stage 3s + 3: Let q be the set of all F
n
(T) such that T
3s+2
(c
s
1
, . . . , c
s
n
) is inconsistent. Clearly q is a principal ntype over T.
Hence q does not include p. Pick a formula p q and put T
3s+3
=
T
3s+2
(c
s
1
, . . . , c
s
n
).
Finally put T
s
T
s
. Dene a Henkin model / as in the proof of
theorem 3.1.2. Each ntuple of elements of [/[ is denoted by an ntuple of
Henkin constants c
s
1
, . . . , c
s
n
. Thus / omits p.
4.
Remark. The hypothesis that T is countable cannot be omitted from the
omitting types theorem.
5.
We digress to present a typical application of the omitting types theorem.
Theorem. Let / = ([/[,
A
) be a countable model of ZF set theory. Then
/ has a proper elementary extension B such that
A
=
B
, i.e. / and B
have the same natural numbers.
Proof. Let T be the theory generated by
(elementary diagram of /) rank(c) > rank(a) : a [/[
where c is a new constant symbol. T is consistent by the compacttness
theorem, and every model of T gives rise to a proper elementary extension
10.1. THE OMITTING TYPES THEOREM 95
of /. Let p be the 1type over T generated by v
1
v
1
,= n : n
A
.
To prove the theorem, it suces to nd a model of T which omits p.
Suppose this were not possible. By the omitting types theorem, p is
included in a principal 1type over T. Let F
n
(T) be a generator of
such a principal 1type. In particular T [= (n) for all n
A
. Let
(v) = (c, v) where sig() sig(elementary diagram of /). Then for each
n
A
we have T [= (c, n). Hence by compactness, for each n
A
there
exists a [/[ such that the elementary diagram of / contains
x(rank(x) > rank(a) (x, n)) .
Hence
/ [= n y x(rank(x) > rank(y) (x, n)) .
By the replacement axiom in / it follows that
/ [= y n x(rank(x) > rank(y) (x, n)) .
Let a [/[ be such that
/ [= n x(rank(x) > rank(a) (x, n)) .
Since T [= rank(c) > rank(a) it follows that T [= n (c, n). In other
words, T [= v
1
((v
1
) v
1
/ ). This contradicts the assumption that the
principal 1type generated by includes p.
6.
The omitting types theorem can be generalized as follows:
Theorem. Let T be a countable theory. For each i , let p
i
be an
essentially nonprincipal n
i
type over T. Then T has a model which omits p
i
for each i .
The proof is a straightforward generalization of the proof of the omitting
types theorem.
96 CHAPTER 10. PRIME MODELS (COUNTABLE CASE)
7.
Exercise. Use the above generalization of the omitting types theorem to
prove the following result, which generalizes theorem 10.1.4 above.
Theorem (Keisler and Morley). Let / be a countable model of ZF set the
ory. Then / has a proper elementary end extension, i.e. a proper elementary
extension B such that rank(a) < rank(b) for all a [/[, b [B[ [/[.
In the next section, the omitting types theorem will be used to study
prime models.
10.2 Prime models
1.
Denition. Let T be a complete theory. A model / of T is said to be prime
if every model of T has an elementary submodel which is isomorphic to /.
2.
Examples.
1. Let T = ACF(0). This theory has a prime model Q = the algebraic
closure of the rational eld Q = (Q, +, , , 0, 1).
2. Similarly for T = ACF(p), the theory of algebraically closed elds of
prime characteristic p. The prime model is the algebraic closure of
T
p
= (F
p
, +, , , 0, 1).
3. Generalizing examples 1 and 2, let / be any eld and let T = ACF
(diagram of /). We know that T is complete because ACF admits
elimination of quantiers. The prime model of T is /, the algebraic
closure of /.
4. T = RCOF. The prime model is Q = the real closure of the ordered
eld Q = (Q, +, , , 0, 1, <).
5. T = RCOF (diagram of T) where T is any ordered eld. The prime
model is T, the real closure of T.
10.2. PRIME MODELS 97
6. T = theory of dense linear ordering without end points. The prime
model is (Q, <).
7. We give an example of a complete countable theory with no prime
model. Let T have 1place relations S
i
(x), i , and axioms saying
that the S
i
s are independent, i.e. any nontrivial Boolean combination of
them is nonempty. The completeness of T can be proved by quantier
elimination. It is easy to see that no model of T is prime.
3.
Remark. It is clear from the denition of prime model that any elementary
submodel of a prime model is prime. Example 6 shows that a prime model
may have proper elementary submodels.
4.
The purpose of this section is to establish necessary and sucient conditions
for a complete theory T to have a prime model, and to establish the unique
ness of prime models when they exist. Throughout this chapter we deal with
countable theories T; for the uncountable case, see chapter 13.
5.
The following lemma is just a restatement of the omitting types theorem in
the special case when T and p are complete.
Lemma. Let T be a countable complete theory, and let p S
n
(T) be a
complete ntype over T. The following are equivalent.
(i) p is principal over T.
(ii) p is realized in every model of T.
Proof. Assume that p is principal, say generated by F
n
(T). In particular
T v
1
v
n
(v
1
, . . . , v
n
) is consistent. Since we are assuming that T is
complete, it follows that T [= v
1
v
n
(v
1
, . . . , v
n
). Let / be any model of
T. Since / [= v
1
v
n
(v
1
, . . . , v
n
), there exist a
1
, . . . , a
n
[/[ such that
/ [= (a
1
, . . . , a
n
). Then clearly the ntuple a
1
, . . . , a
n
realizes p.
98 CHAPTER 10. PRIME MODELS (COUNTABLE CASE)
Conversely, suppose that p is nonprincipal over T. Since p is complete,
it follows that p is essentially nonprincipal over T. Hence by the omitting
types theorem, there exists a model of T which omits p.
6.
Denition. A structure / is atomic if for every a
1
, . . . , a
n
[/[ the com
plete ntype realized by a
1
, . . . , a
n
is principal over Th(/).
7.
Theorem (Vaught). Let T be a countable complete theory. The following
are equivalent for / Mod(T).
(i) / is prime.
(ii) / is countable and atomic.
Proof. (i) (ii). Assume /is prime. Since T has a countable model B and /
is isomorphic to an elementary submodel of B, it follows that / is countable.
To show that / is atomic, let a
1
, . . . , a
n
[/[ and let p be the complete
ntype over T realized by a
1
, . . . , a
n
. Since / is prime, p is realized in every
model of T. Hence by the previous lemma it follows that p is principal. Thus
/ is atomic.
(ii) (i). Assume that / is countable and atomic. Let [/[ = a
n
:
n be an enumeration of [/[. Let B be a model of T. We want to
construct an elementary embedding of / into B. Assume inductively that
we have chosen b
0
, . . . , b
n1
[B[ so that (/, a
i
)
i<n
(B, b
i
)
i<n
. Since / is
atomic, so is (/, a
i
)
i<n
. [ For, let c
1
, . . . , c
k
[/[. The complete (n + k)
type realized by a
0
, . . . , a
n1
, c
1
, . . . , c
k
) is principal over T, say generated by
(u
0
, . . . , u
n1
, w
1
, . . . , w
k
). Put (w
1
, . . . , w
k
) (a
0
, . . . , a
n1
, w
1
, . . . , w
k
).
Then generates the complete ktype realized by c
1
, . . . , c
k
over Th((/, a
i
)
i<n
). ]
Let p
n
S
1
((/, a
i
)
i<n
) be the complete 1type realized by a
n
. Since p
n
is principal, it follows by the previous lemma that p
n
is realized in (B, b
i
)
i<n
,
say by b
n
[B[. Then (/, a
i
)
in
(B, b
i
)
in
. Finally (/, a
i
)
i
(B, b
i
)
i
.
Thus a
i
b
i
gives an elementary embedding of / into B.
10.2. PRIME MODELS 99
8.
Theorem (Vaught). Let T be a countable complete theory. If T has a prime
model, it is unique up to isomorphism.
Proof. By the previous theorem it suces to show that any two countable
atomic models of T are isomorphic. We use a backandforth argument. The
inductive step is as in the proof of the previous theorem. The details are
routine.
9.
Theorem (Vaught). Let T be a countable complete theory. The following
are equivalent.
(i) T has a prime model.
(ii) T has an atomic model.
(iii) Every principal ntype over T is included in a complete principal ntype
over T.
10.
Remark. A Boolean algebra B is said to be atomic if for all b B, b ,= 0,
there exists a b such that a is an atom, i.e. a ,= 0 and there is no a
1
such that 0 < a
1
< a. The condition (iii) in the theorem can be restated as
follows: for each n , the Boolean algebra B
n
(T) is atomic.
11.
Proof of theorem 9.
(i) (ii). Let / be a prime model of T. By 10.2.7 above, / is atomic.
(ii) (iii). Let / be an atomic model. Let q be a principal ntype
over T, say generated by F
n
(T). Since T is complete, we have T [=
v
1
. . . v
n
(v
1
, . . . , v
n
). Hence q is realized in /, say by a
1
, . . . , a
n
. Let
p S
n
(T) be the complete ntype realized by a
1
, . . . , a
n
. Then p q and p
is principal.
(iii) (i). Assuming (iii) we construct a countable atomic model of
T. We employ a Henkin construction as in the proof of the omitting types
100 CHAPTER 10. PRIME MODELS (COUNTABLE CASE)
theorem 10.1.3. Let c
s
1
, . . . , c
s
ns
) : s be an enumeration of all nite
sequences of Henkin constants. Stages 0, 3s + 1, and 3s + 2 are as in the
proof of theorem 10.1.3.
Stage 3s + 3: Put n = n
s
and let q be the set of F
n
(T) such that
T
3s+2
(c
s
1
, . . . , c
s
n
) is inconsistent. Let p q be a principal complete
ntype over T. Let F
n
(T) be a generator of p. Put T
3s+3
= T
3s+2
(c
s
1
, . . . , c
s
n
). Clearly T
3s+3
is consistent.
At the end of the construction we obtain a countable atomic model of T.
This is prime by theorem 10.2.7.
12.
Corollary (Vaught). Let T be a countable complete theory. Suppose S
n
(T)
is countable for each n . Then T has a prime model.
Proof. Assuming that S
n
(T) is countable, we shall show that B
n
(T) is atomic,
i.e. every principal ntype q over T is included in a complete principal ntype
over T. Suppose not, i.e. any principal ntype which includes q is incomplete.
We use a splitting argument. For each nite sequence s of 0s and 1s we
dene a principal ntype q
s
which includes q. Begin with q
= q. Given q
s
,
we know that q
s
is incomplete, so let
s
F
n
(T) be a formula such that
q
s
s
and q
s
s
are both consistent. Let q
s1
be the ntype generated
by q
s
s
and let q
s0
be the ntype generated by q
s
s
.
For each f 2
let q
f
=
q
fn
: n . Then q
f
is an ntype over
T, and f ,= g implies q
f
q
g
is inconsistent. Let p
f
be a complete ntype
over T which includes q
f
. Then f ,= g implies p
f
,= p
g
. Hence [S
n
(T)[ = 2
0
contradicting our assumption that S
n
(T) is countable.
13.
Remark. In terms of Boolean algebras, the previous argument shows that
if B is a countable Boolean algebra with S(B) countable, then B is atomic.
10.3. THE NUMBER OF COUNTABLE MODELS 101
10.3 The number of countable models
1.
Let T be a complete theory and let be a cardinal. We write (T, ) = the
number of nonisomorphic models of T of cardinality . This function is a
central object of study in pure model theory.
In this section, we study (T,
0
) where T is complete and countable. We
begin with the case when (T,
0
) = 1, i.e. T is
0
categorical.
2.
Theorem (RyllNardzewski). Let T be a complete countable theory with
no nite models. The following are equivalent:
(1) T is
0
categorical.
(2) S
n
(T) is nite for each n .
(3) Every countable model of T is prime.
(4) Every countable model of T is saturated.
(The equivalence of (1) and (2) is due to RyllNardzewski.)
Proof. (1) (2). Suppose that S
n
(T) is innite. We claim that there
exists a nonprincipal q S
n
(T). To see this, for each principal p S
n
(T) let
p
F
n
(T) be a generator of p. Put Y =
p
: p S
n
(T), p nonprincipal.
Since S
n
(T) is innite, it follows by compactness that Y is consistent over
T. Hence Y can be extended to a complete ntype q S
n
(T). Clearly q is
nonprincipal.
By lemma 10.2.5 there exists a countable model / Mod(T) such that
/ omits q. Also there exists a countable B Mod(T) such that B realizes q.
/ and B are nonisomorphic, so T is not
0
categorical.
(2) (3). Since S
n
(T) is nite, every p S
n
(T) is principal. Hence
every model of T is atomic. Hence by theorem 10.2.7 it follows that every
countable model of T is prime.
(2) (4). Let / be a countable model of T., and let X [/[ be nite.
From (2) it follows that S
1
(X) is nite. hence each p S
1
(X) is principal,
hence realized in /
X
. This shows that / is saturated.
102 CHAPTER 10. PRIME MODELS (COUNTABLE CASE)
(3) (4) (1). This is because of uniqueness of prime models (10.2.8)
or saturated models (7.2.8) respectively.
3.
Remark. In terms of Boolean algebras, the argument for (1) (2) above
shows that if B is a Boolean algebra with S(B) innite (equivalently B is
innite), then B has a nonprincipal ultralter.
4.
Next we consider the situation where 1 < (T,
0
) <
0
. We say that a
countable model / of T is weakly saturated if every p S
n
(T), n is
realized in /.
5.
Theorem (Rosenstein). Suppose 1 < (T,
0
) <
0
where T is a countable
complete theory. Then T has a weakly saturated countable model which is
not saturated.
Proof. Suppose not. Let /
1
, . . . , /
k
be the countable models of T which are
not saturated. By assumption these models are not weakly saturated, so let
p
i
S
n
i
(T) be omitted in /
i
, 1 i k.
Introduce new constant symbols c
i
1
, . . . , c
i
n
i
, 1 i k, and let T
be a
countable complete theory which includes T (c
i
1
, . . . , c
i
n
i
) : p
i
, 1
i k. Let /
= ([/
[,
. Then / =
([/
[,
is saturated. We
have now shown that any countable model of T
is saturated. Hence T
is
0
categorical. Hence by RyllNardzewskis theorem 10.3.2, it follows that
S
n
(T
0
. For instance (ACF(0),
0
) =
0
and (RCOF,
0
) =
2
0
.
104 CHAPTER 10. PRIME MODELS (COUNTABLE CASE)
8.
We omit the proof of the following theorem.
Theorem (Morley). Let T be a countable complete theory. If (T,
0
) is
innite, then it is equal to either
0
,
1
, or 2
0
.
9.
Remark. Vaught has conjectured that for all countable complete theories
T, if (T,
0
) >
0
then (T,
0
) = 2
0
. No counterexample to Vaughts
conjecture is known. Steel
1
has shown that Vaughts conjecture holds for
theories T whose models are trees. (A tree is a partially ordered set such
that the predecessors of any element are linearly ordered.) Shelah has shown
that Vaughts conjecture holds for superstable theories. Also, Lachlan
2
has
shown that for superstable theories, (T,
0
) <
0
implies (T,
0
) = 1.
10.4 Decidable prime models
1.
Let / be a countable model such that sig(/) is nite (or recursive). Recall
that / is said to be decidable if there exists an enumeration [/[ = a
n
: 1
n < of the elements of /, such that
_
_
n
F
n
(T) : / [= (a
1
, . . . , a
n
)
_
is recursive. (This is easily seen to be equivalent to denition 4.3.1.)
2.
Recall theorem 4.3.4 which said that if T is a decidable theory, then T has
a decidable model. However, it is not in general true that if T is a complete
decidable theory which has a prime model, then T has a decidable prime
model. We now present a theorem which gives a necessary and sucient
condition for T to have a decidable prime model.
1
SLNM 689
2
FM 81
10.4. DECIDABLE PRIME MODELS 105
3.
Theorem (Harrington). Let T be a complete decidable theory. The follow
ing are equivalent.
(a) T has a decidable prime model.
(b) Given n 1 and F
n
(T) such that is consistent with T, we can
recursively enumerate a complete ntype p
S
n
(T), such that p
and p
is principal.
Note. We do not assume in (b) that it is possible to pass recursively from
to a generator of p
.
Proof. (a) (b). Let / be a decidable prime model of T. Given F
n
(T)
consistent with T, since T is complete we have T [= v
1
v
n
so we can
recursively nd an ntuple a
1
, . . . , a
n
[/[ such that / [= (a
1
, . . . , a
n
). We
can then enumerate the complete ntype realized by a
1
, . . . , a
n
and dene p
, and p
n
(x) : 1 n < is an enumeration of all formulas (x) with one free
variable x such that sig((x)) sig(T)C, and we assume that sig(
n
(x))
sig(T) c
1
, . . . , c
n
.
We shall construct a complete decidable theory T
extending T
0
such
that, for each m 1, there will exist a formula
m
F
m
(T) such that
c
1
, . . . , c
m
realizes p
m
. This clearly implies that the Henkin model associated
with T (as in the proof of theorem 4.3.4) is decidable and prime.
If it were possible to pass recursively from to a generator of p
, we
could construct T
recursively in stages T
0
T
1
T
s
.
At each stage s 0, T
s
will consist of T
0
plus nitely many sentences.
106 CHAPTER 10. PRIME MODELS (COUNTABLE CASE)
Let
s
(c
1
, . . . , c
Ns
) be the conjunction of this nite set of sentences, where
s
F
Ns
(T). Put
s
m
(v
1
, . . . , v
m
) v
m+1
v
n
[ H
1
(v
1
) H
n
(v
1
, . . . , v
n
)
s
(v
1
, . . . , v
Ns
) ]
where n = max(m, N
s
). Thus
s
m
expresses all the assertions about c
1
, . . . , c
m
to which we are committed at the end of stage s. Note that
s
m
F
m
(T).
At each stage s, there will be a recursive sequence of formulas
s
m
F
m
(T), 1 m < . Here
s
m
is the stage s approximation of the desired
formula
m
.
We now present the construction.
Stage 0. Let T
0
be as above. Put
0
m
(v
1
, . . . , v
m
) =
0
m
(v
1
, . . . , v
m
).
Stage s + 1. Let s = (n 1, k, t) in some xed recursive enumeration
of . Let (v
1
, . . . , v
n
) be the kth formula in some xed recur
sive enumeration of F
n
(T). At this stage we shall consider whether to add
(c
1
, . . . , c
n
) to our theory.
If p
s
n
and T ,[=
s
n
and T [=
s
m
v
m+1
v
n
[
s
n
] for each
m, 1 m < n, then put
T
s+1
= T
s
(c
1
, . . . , c
n
)
and, for all m 1,
s+1
m
=
_
s
m
if 1 m n,
s+1
m
if m > n.
Otherwise do nothing, i.e. T
s+1
= T
s
and
s+1
m
=
s
m
for all m 1.
It is clear that the construction is recursive and hence produces a recur
sively axiomatizable theory T
s+1
m
.
Claim 2:
s
m
p
s
m
.
We prove this by induction on s. For s = 0 it is trivial since
0
m
=
0
m
p
0
m
. If no action is taken at stage s + 1, the induction step is trivial.
Consider a stage s + 1 at which action is taken. Let n and be as in the
construction. For m < n we have
s+1
m
v
1
v
n
[
s
n
] so by hypothesis
10.4. DECIDABLE PRIME MODELS 107
T [=
s
m
s+1
m
. Also
s+1
m
=
s
m
so by induction
s+1
m
p
s
m
= p
s+1
m
. For
m = n, we have
s+1
n
=
s
n
and
s
n
p
s
n
by induction, and p
s
n
by
hypothesis. Hence
s+1
n
p
s
n
= p
s+1
n
since
s+1
n
=
s
n
. Finally, for m > n,
we have
s+1
m
=
s+1
m
p
s+1
m
.
Claim 3: ns s
s
s
n
=
s
n
. (We then dene
n
= lim
s
s
n
.)
Claim 4: ns (
s
n
is a generator of p
n
).
We prove claims 3 and 4 simultaneously by induction on n. Assume that
3 nd 4 hold for all m < n. Let be a stage such that for all m < n and all
s ,
s
m
=
m
and
s
m
generates p
s
m
.
We rst prove 3 for n. Given s , let s = (m1, k, t) and let be the
kth formula in F
m
(T). If m n, then
s+1
n
=
s
n
. If m < n, then p
s
m
implies T [=
s
m
, so nothing was done at stage s. Hence again
s+1
n
=
s
n
.
Thus by induction we have s
s
n
=
n
.
Next we prove claim 4 for n. Let be a generator of p
n
= p
n
. Look
at what happens at a stage s such that s = (n 1, k, t) where is the
kth formula in F
n
(T). We have p
s
n
and, by claim 2,
s
n
p
s
n
. Hence
s
n
is consistent with T. Hence, for each m < n, since
s
m
generates p
s
m
,
it follows that
s
m
is an atom in B
m
(T), so
T [=
s
m
v
m+1
v
n
[
s
n
] .
Thus, unless T [=
s
n
already, we have T
s+1
= T
s
(c
1
, . . . , c
n
). In
any case T [=
s+1
n
so
s+1
n
is a generator of p
n
. This completes the
proof of claims 3 and 4.
From claim 4 it follows that T
s
T
s
generates a complete decidable
extension of T
0
whose asociated Henkin model is atomic. This completes the
proof of the theorem.
4.
Remark. In chapter 11 we shall apply the above theorem to show that
the dierential closure of a computable dierential eld of characteristic 0 is
computable. This was Harringtons original application.
5.
Exercise. Prove the following theorem of Morley
3
:
3
Israel J. Math. vol. 25 pp. 233240
108 CHAPTER 10. PRIME MODELS (COUNTABLE CASE)
Theorem (Morley). A complete decidable theory T has a decidable sat
urated model if and only if there is a uniform recursive enumeration of
n
S
n
(T).
For more information on decidable models, see e.g. Millar, Annals of
Math. Logic vol. 13, pp. 4572.
6.
Exercise. Use theorem 10.4.3 to prove that the algebraic closure of a com
putable eld is computable, and the real closure of a computable ordered
eld is computable.
Chapter 11
Dierentially closed elds of
characteristic 0
11.1 Simple extensions
1.
The theory of dierential elds, DF, has the following set of axioms:
(a) eld axioms
(b) xy (x + y)
= x
+ y
(c) xy (x y)
= x
y + x y
Here
is a 1place operation symbol.
A dierential eld is a model of these axioms. A dierential domain is a
model of the axioms for domains plus (b), (c). A dierential ring is dened
similarly. The characteristic of a dirential eld is the characteristic of the
underlying eld.
2.
Examples. (1) Let / be any eld. The polynomial ring /[x] can be made
into a dierential ring by interpreting
as the formal derivative, i.e. if f =
a
n
x
n
+ + a
1
x + a
0
, a
i
[/[, then f
= na
n
x
n1
+ + a
1
.
109
110 CHAPTER 11. DIFF. CLOSED FIELDS OF CHAR. 0
We can extend the derivative to the eld of rational functions /(x) =
eld of quotients of /[x] in the usual way:
_
f
g
_
=
f
g f g
g
2
.
Then /(x) becomes a dierential eld.
(2) Any eld / can be made into a dierential eld in a trivial way by
putting a
= derivative
of f = df/dz.
(4) Let ^ be the set of all functions which are meromorphic in some open
set containing 0 in the zplane. (Identify two such functions if they are equal
on some open set containing 0.) This is again a dierential eld in an obvious
way.
3.
Remark. Ritt and Seidenberg
1
have shown that every dierential eld which
is nitely generated over the reational eld is isomorphic to a dierential
subeld of ^. Thus everything we shall do in this chapter is meaningful for
analysis.
4.
We now study simple extensions of dierential elds of characteristic 0.
Lemma. Let T be a dierential domain and let T be its eld of quotients,
T = c/d : c, d [T[, d ,= 0. Then there is one and only one way to make
T into a dierential eld extension of T.
1
Proc. AMS 9 (1958) 159164; 23 (1969) 689691.
11.1. SIMPLE EXTENSIONS 111
Proof. We must have
c
=
_
c
d
d
_
=
_
c
d
_
d +
c
d
d
so
_
c
d
_
=
c
d c d
d
2
.
It is easy to check that this makes T a dierential eld, etc.
5.
Lemma. Let /be a dierential eld of characteristic 0. Let B = /(b
0
, . . . , b
n1
, b
n
)
be a nitely generated eld extension of / such that
(i) for each i < n, b
i
is transcendental over /(b
0
, . . . , b
i1
),
(ii) b
n
is algebraic over /(b
0
, . . . , b
n1
).
Then there is one and only one way to make B into a dierential eld exten
sion of / subject to the conditions b
i
= b
i+1
, i < n.
6.
Before proving this lemma, we give an example. Let b
0
be transcendental
over / and let b
1
satisfy b
2
1
b
0
= 0. Then /(b
0
, b
1
) can be made into a
dierential eld extension of / in which b
0
is a solution of the rst order
dierential equation (y
)
2
y = 0 and is not the solution of any algebraic
equation over /.
7.
Proof of lemma 5. Consider the ring /[y] = /[y
0
, . . . , y
n1
, y
n
]. A typical
element g of /[y
0
, . . . , y
n1
, y
n
] looks like
g(y) =
j
a
j
y
j
where a
j
= a
j
0
j
1
...jn
[/[ and y
j
= y
j
0
0
y
j
n1
n1
y
jn
n
. Let us write
g(y) =
j
a
j
y
j
112 CHAPTER 11. DIFF. CLOSED FIELDS OF CHAR. 0
and g/y
i
= the formal partial derivative of g(y
0
, . . . , y
n1
, y
n
) with respect
to y
i
. Given g(b
0
, . . . , b
n1
, b
n
) /[b
0
, . . . , b
n1
, b
n
] we are forced to dene
g(b
0
, . . . , b
n1
, b
n
)
= g(b
0
, . . . , b
n1
, b
n
) +
n
i=0
g
y
i
(b
0
, . . . , b
n1
, b
n
) b
i+1
where however b
n+1
remains to be determined.
Let f(b
0
, b
1
, . . . , b
n1
, y
n
) /(b
0
, . . . , b
n1
)[y
n
] be irreducible such that
B
= /(b
0
, . . . , b
n1
)[y
n
]/f (by theorem 6.1.12). Multiplying f by a suitable
element of /[b
0
, . . . , b
n1
], we may safely assume that f(y
0
, . . . , y
n1
, y
n
)
/[y
0
, . . . , y
n1
, y
n
]. As in the previous paragraph, we want
0 = f(b
0
, b
1
, . . . , b
n1
, b
n
)
=
f(b
0
, . . . , b
n1
, b
n
) +
n
i=0
f
y
i
(b
0
, . . . , b
n1
, b
n
) b
i+1
.
Since f/y
n
is of lower degree than f in y
n
, we must have
f
y
n
(b
0
, . . . , b
n1
, b
n
) ,= 0 .
Hence we are forced to dene
b
n+1
=
f(b
0
, . . . , b
n1
, b
n
) +
n1
i=0
f
y
i
(b
0
, . . . , b
n1
, b
n
) b
i+1
f
yn
(b
0
, . . . , b
n1
, b
n
)
.
We must check that
is well dened. We have f(b
0
, . . . , b
n1
, b
n
) = 0. Also
f(b
0
, . . . , b
n1
, b
n
)
= 0 by choice of b
n+1
. Suppose g
1
, g
2
/[y
0
, . . . , y
n1
, y
n
]
are such that g
1
(b
0
, . . . , b
n1
, b
n
) = g
2
(b
0
, . . . , b
n1
, b
n
). Then
g
1
(b
0
, . . . , b
n1
, y
n
) g
2
(b
0
, . . . , b
n1
, y
n
) mod f(b
0
, . . . , b
n1
, y
n
)
in the polynomial ring /(b
0
, . . . , b
n1
)[y
n
]. Hence g
1
g
2
= f h where
h /(y
0
, . . . , y
n1
)[y
n
]. Letting h = p/q, we have
q (g
1
g
2
) = f p
where p /[y
0
, . . . , y
n1
, y
n
] q /[y
0
, . . . , y
n1
], q , 0. Substituting b
i
for
y
i
and dierentiating both sides we have
q(
b) (g
1
(
b)
g
2
(
b)
) = f(
b)
p(
b) + f(
b p(
b)
= 0
11.1. SIMPLE EXTENSIONS 113
and q(
b) ,= 0, hence g
1
(b
0
, . . . , b
n1
, b
n
)
= g
2
(b
0
, . . . , b
n1
, b
n
)
. This shows
that
is well dened on the ring /[b
0
, . . . , b
n1
, b
n
]. It is easy to check the sum
and product rules, so we now know that there is one and only one way to make
/[b
0
, . . . , b
n1
, b
n
] into a dierential ring extension of /. By lemma 11.1.4
we get the same conclusion for the eld of quotients B = /(b
0
, . . . , b
n1
, b
n
).
This completes the proof.
8.
Remark. Let / be a dierential eld. We have considered four dierent
kinds of simple extensions of /, namely:
/[b] = ring generated by [/[ b,
/(b) = eld generated by [/[ b,
/b = dierential ring generated by [/[ b,
/b) = dierential eld generated by [/[ b.
Our goal is to classify the simple dierential eld extensions /b).
9.
Denition. Let / be a dierential eld. A dierential polynomial f(y) =
f(y, y
, . . . , y
(n)
) is an element of the dierential ring
/y = /[y, y
, . . . , y
(n)
, . . . ]
where of course y
(n)
= y
..
ntimes
and y is a dierential indeterminate. (Alter
natively, we could dene a dierential polynomial to be an equivalence class
of terms t(y) in the language of the diagram of /, as in 6.1.4.)
Note that /y is a dierential domain and /y), its eld of quotients,
is a dierential eld by lemma 11.1.4.
10.
Denition. If f(y) /y is a nonzero dierential polynomial over /, we
dene the order of f to be the largest n such that y
(n)
occurs in f, and the
114 CHAPTER 11. DIFF. CLOSED FIELDS OF CHAR. 0
degree of f to be the largest k such that (y
(n)
)
k
occurs in f, where n is the
order of f.
For example, the dierential polynomial (y
(3)
)
4
+(y
)
5
y
(3)
y has order
3 and degree 4.
11.
We now survey all simple dierential eld extensions /b) of a dierential
eld / of characteristic 0.
Case 1: f(b) = 0 for some nonzero f /y. Choose such an f of lowest
possible order n, and then of lowest possible degree k for that order. Note
that f/y
(n)
has lower order or else the same order and lower degree. Hence
f
y
(n)
(b) ,= 0 .
Hence, as in 11.1.7, we have
b
(n+1)
=
f(b) +
n1
i=0
f
y
(i)
(b) b
(i+1)
f
y
(n)
(b)
so in particular b
(n+1)
/(b, . . . , b
(n1)
, b
(n)
). From this it follows that b
(j)
/(b, . . . , b
(n1)
, b
(n)
) for all j, and hence the eld /(b, . . . , b
(n1)
, b
(n)
) is closed
under
. Hence /b) = /(b, . . . , b
(n1)
, b
(n)
) and its structure is as in lemma
11.1.5. In this case we say that b is dierentially algebraic over /.
Case 2: negation of case 1. In this case it is easy to see that /b
= /y,
hence the quotient elds /b)
= /y) by lemma 11.1.4. In this case we say
that b is dierentially transcendental over /.
This completes our survey of the simple dierential eld extensions of /.
12.
From the above survey of simple extensions we immediately obtain the fol
lowing result.
Theorem. Let / be a dierential eld of characteristic 0. Up to isomor
phism over /, there are exactly / distinct simple dierential eld exten
sions /b).
11.2. DIFFERENTIALLY CLOSED FIELDS 115
11.2 Dierentially closed elds
1.
Denition (Blum). Let B be a dierential eld of characteristic 0. We
say that B is dierentially closed if B is algebraically closed and, for all
f, g By with 0 order(g) < order(f), there exists b [B[ with f(b) = 0,
g(b) ,= 0.
2.
Lemma. Let / be a dierential eld of characteristic 0. Then there exists
a dierential eld B / such that (i) B is dierentially closed, (ii) B is
dierentially algebraic over /, i.e. every b [B[ is dierentially algebraic
over /.
Proof. By transnite induction, it suces to show:
3.
Sublemma. Given f /y of order > 0, there exists a simple dierential
eld extension /b) such that f(b) = 0 and g(b) ,= 0 for all g /y with
0 order(g) < order(f).
(In this event b is called a generic solution of the dierential equation f(y) =
0 over /.)
To prove the sublemma, let n be the order of f and let /
= /(b
0
, . . . , b
n1
)
be a eld extension of / in which each b
i
, i < n, is transcendental over
/(b
0
, . . . , b
i1
). Dene a polynomial p(x) /
(c) be a sim
ple algebraic eld extension of /
(c) = /(b
0
, . . . , b
n1
, b
n
) into a simple dif
ferential eld extension /b) of / where b = b
0
, b
i
= b
i+1
for i < n. We
then have f(b) = f(b
0
, . . . , b
n1
, b
n
) = p(b
n
) = p(c) = 0 and by construc
tion g(b
0
, . . . , b
n1
) ,= 0 for all nonzero g /[y
0
, . . . , y
n1
], in other words
g(b) ,= 0 for all g /y of order m, 0 m < n. This completes the proof.
116 CHAPTER 11. DIFF. CLOSED FIELDS OF CHAR. 0
4.
Remark. In the statement of the lemma B is not unique. See 11.3.4 below.
5.
Theorem (Robinson, Blum). The theory of dierentially closed elds of
characteristic 0 (DCF(0)) is the model completion of the theory of dierential
elds of characteristic 0 (DF(0)).
Proof. By the previous lemma plus theorem 8.3.2 it suces to show the
following:
Let / B where / is a DF(0), B a DCF(0), B
+
saturated
where = /. Then for any simple dierential eld extension
/c) of /, we can nd b [B[ such that /b)
= /c) over /.
To prove this, we consider two cases.
Case 1: c is dierential algebraic over /, i.e. f(c) = 0 for some nonzero
f /y. Take f to be of least order, and of least degree for that order.
Let p be the 1type over / generated by the dierential equation f(y) = 0
and the inequations g(y) ,= 0, g /y, 0 order(g) < order(f). Since
B is dierentially closed, every nite subset of p is realized in B. Hence by
saturation p is realized in B, say by b [B[. We claim that /b)
= /c) over
/. This is clear from 11.1.5 and 11.1.11.
Case 2: c is dierentially transcendental over /. Consider the 1type p
over / generated by g(y) ,= 0 for all g /y, 0 order(g). Again, since
B is dierentially closed, each nite subset of p is realized in B. Hence p is
realized in B, say by b [B[. Then b is dierentially transcendental over /,
so /b)
= 0. Put
B = /b). Then B is dierentially closed, and dierentially algebraic over
/, but we claim that B is not isomorphic to / over /. To see this, let p be
the complete 1type over T = DCF(0) (diagram of /) realized by b [B[.
Clearly p is nonprincipal. Hence by theorem 10.2.7 it follows that p is not
realized in /.
4.
In order to prove our next theorem, we need the following algebraic result
which we state without proof.
Lemma. Let / be a dierential eld of characteristic 0. Let S be an innite
set of algebraic dierential equations in nitely many unknowns y
1
, . . . , y
n
with coecients in [/[. Then there exists a nite subset S
0
of S such that
every solution of S
0
(in some dierential eld extension of /) is a solution of
S.
This is an immediate corollary of the Ritt basis theorem, i.e. theorem
11.4.4 below. For a proof, see Kaplansky, Introduction to Dierential Algebra.
11.3. DIFFERENTIAL CLOSURE (COUNTABLE CASE) 119
5.
Theorem (Harrington). Let / be a computable dierential eld of charac
teristic 0. Then /, the dierential closure of /, is computable.
Proof. / is by denition the prime model of the complete theory T whose
axioms are DCF(0) (diagram of /). Since / is computable, T is recursively
axiomatizable, and hence decidable. But theorem 10.4.3 gives a necessary
and sucient condition for a complete decidable theory to have a decidable
prime model. Thus, in order to show that / is computable, it suces to
show that T veries condition (b) of that theorem.
Given F
n
(T) consistent with T, by quantier elimination we may
safely assume that (y
1
, . . . , y
n
) consists of nitely many algebraic diferential
equations and inequations in n unknowns y
1
, . . . , y
n
with coecients in [/[.
Let
k
(y
1
, . . . , y
n
) : k be a recursive enumeration of all algebraic
dierential equations in n unknowns y
1
, . . . , y
n
with coecients in [/[. Let p
k
if this is consistent with T,
k+1
=
k
otherwise. The passage from to p
is complete.
It remains to show that p
i=1
l
j=1
p
ij
g
(j)
i
where p
ij
1 and g
(j)
i
is the jth derivative of g
i
.
Proof. Let I be the set of all f 1 for which the conclusion holds. Suppose
f / I. Then we have
(i) g, h I g + h I
(ii) g I, h 1 g h I
(iii) 0 I, 1 / I
(iv) g I g
I
i.e. I is a dierential ideal. Here (i), (ii), (iii) are proved as in the proof
of Hilberts Nullstellensatz 6.4.2. For (iv), suppose g I, say g
k
0
mod g
1
, . . . , g
m
. Dierentiate and multiply by 1/k to get
g
k1
g
0 mod g
1
, . . . , g
m
.
11.4. RITTS NULLSTELLENSATZ 121
This is the case i = 1 of the statement
g
ki
(g
)
2i1
0 mod g
1
, . . . , g
m
which we prove by induction on i, 1 i k. Dierentiating and multiplying
by g
we get
(k i)g
ki1
(g
)
2i+1
+ (2i 1)g
ki
(g
)
2i1
g
. .
0
0 .
Multiplying by 1/(k i) we get
g
ki1
(g
)
2i+1
0 mod g
1
, . . . , g
m
so the induction step is proved. Finally for i = k we get
(g
)
2k1
0 mod g
1
, . . . , g
m
so (iv) is proved.
It is also clear that I is a radical dierential ideal, i.e. I satises
(v) g
k
I g I ,
and f / I. By Zorns lemma let J I be a radical dierential ideal such
that f / J and maximal with this property.
We claim that J is prime, i.e. g / J, h / J imply g h / J. Suppose not.
Note that g h J implies (g
h + g h
) g
h J whence (g
h)
2
J,
whence g
h J. Similarly g
(i)
h
(j)
J for all i, j 0. If g / J then the
radical dierential ideal generated by J and g must contain f, say
f
k
= s +
i
p
i
g
(i)
where s J, p
i
1. Similarly if h / J then
f
l
= t +
j
q
j
h
(j)
where t J, q
j
1. Hence
f
k+l
= s f
l
+ t f
k
+
j
p
i
q
j
g
(i)
h
(j)
. .
J
122 CHAPTER 11. DIFF. CLOSED FIELDS OF CHAR. 0
contradicting the fact that J is radical and f / J. This proves the claim.
Now let 1
1
= 1/J = the quotient ring of 1 mod J. Clearly 1
1
is
a dierential domain. Let B be the eld of quotients of 1
1
. Then B is a
dierential eld extension of /. Furthermore, there exist b
1
, . . . , b
n
[B[
(corresponding to y
1
, . . . , y
n
[1[) such that for all g 1, g(b
1
, . . . , b
n
) = 0
if and only if g J. In particular f(b
1
, . . . , b
n
) ,= 0 while g
1
(b
1
, . . . , b
n
) =
= g
m
(b
1
, . . . , b
n
) = 0. Hence by model completeness of DCF(0) we can
nd a
1
, . . . , a
n
[/[ such that f(a
1
, . . . , a
n
) ,= 0 and g
1
(a
1
, . . . , a
n
) = =
g
m
(a
1
, . . . , a
n
) = 0. This completes the proof of the theorem.
3.
Remark. As in the case of Hilberts Nullstellensatz, we can deduce a version
of Ritts Nullstellensatz with eective bounds. We leave to the reader the
formulation and proof of this result. (Compare 6.4.3.)
4.
An interesting theorem related to Ritts Nullstellensatz is the following.
Theorem (Ritts basis theorem). Let 1 be as above and let I be a radical
dierential ideal in 1. Then I has a nite basis, i.e. there exists a nite
set g
1
, . . . , g
r
I such that I is the radical dierential ideal generated by
g
1
, . . . , g
r
, i.e. the set of all f 1 such that
f
k
=
r
i=0
l
j=0
p
ij
g
(j)
i
for some k, l and p
ij
1.
Proof. See Kaplansky, Introduction to Dierential Algebra.
5.
Remark. The Ritt Basis Theorem tends to clarify several aspects of the
proof of the Ritt Nullstellensatz. In the rst place, the Ritt Basis Theorem
shows that the use of Zorns lemma could have been eliminated. In the
second place, since the radical dierential ideal J has a nite basis g
1
, . . . , g
r
,
11.4. RITTS NULLSTELLENSATZ 123
it follows by quantier elimination that the complete ntype of b
1
, . . . , b
n
over [/[ is generated by the dierential equations g
1
(y
1
, . . . , y
n
) = =
g
r
(y
1
, . . . , y
n
) = 0 and the single inequation f(y
1
, . . . , y
n
) ,= 0. In particular
this ntype is principal
2
and hence already realized in /. In other words, the
dierential eld B = /b
1
, . . . , b
n
) is already embeddable into /. (We could
have used Hilberts Basis Theorem to make similar remarks about the proof
of the Hilbert Nullstellensatz.)
2
The algebraists express this fact by saying that J is constrained. See Wood, Israel J.
Math. vol. 25, p. 331.
124 CHAPTER 11. DIFF. CLOSED FIELDS OF CHAR. 0
Chapter 12
Totally transcendental theories
12.1 Stability
1.
12.2 Rank of an element type
12.3 Indiscernibles
12.4 Existence of saturated models
125
126 CHAPTER 12. TOTALLY TRANSCENDENTAL THEORIES
Chapter 13
Prime models (uncountable
case)
13.1 Strongly atomic models
1.
13.2 Normal sets
13.3 Uniqueness and characterization of prime
models
127