THE ELEMENTARY THEORY OF FINITE FIELDS

by
Moshe Jarden, Tel Aviv University
April 1994
March 2004
* Notes of a course given at Tel Aviv University, Spring 1994 and Spring 2004.
Introduction
Transfer theorem: Let be an elementary statement about elds. Set
A() = p P(Q) [ is true in F
p

S() = Gal(Q) [ is true in

Q()
Then A() has a Dirichlet density (A()), S() is Haar measurable and (A()) =
(S()). Moreover, (A()) is a rational number which is positive exactly if A() is
innite.
An elementary statement is a statement which is equivalent to a rst order
sentence in the language of rings (examples follow).
Q = the eld of rational numbers;
P(Q) = the set of rational primes;

Q = the eld of all algebraic numbers;

Gal(Q) = the absolute Galois group of Q, i.e., Gal(

Q/Q); if Gal(Q), then

Q() = x

Q[ x = x.
The Dirichlet density of a subset A P(Q) is (if the limit exists)
(A) = lim
s1
+

pA
1
p
s
_

pP(Q)
1
p
s
.
In particular (P(Q)) = 1 and the density of each nite set is zero. However, there exist
innite set of primes of density zero.
The Haar measure of Gal(Q) is the unique one dened with respect to the
Krull topology of Gal(Q) such that (Gal(Q)) = 1.
The proof of the transfer theorem involves ingredients from several areas: Alge-
braic number theory, the theory of pronite groups, measure theory, algebraic geometry,
and model theory.
Credit: Ax [Ax] 1967, Jarden [Ja1] 1969, Fried [FrS] 1974.
References: Field Arithmetic [FrJ].
1
Example 1: is the sentence X: X
2
= 1.
F
p
[= there exists x F
p
: x
2
= 1
there exists x Z: x
2
1 mod p
p = 2 or
_
1
p
_
= 1 (Legendre symbol)
p = 2 or p 1 mod 4
By Dirichlets theorem on primes in arithmetic progressions (A()) =
1
2
.

Q() [= there exists x

Q(): x
2
= 1

1

Q()
res
Q(

1)
= 1
As Gal(Q(

1)/Q) has two elements, (A()) =

1
2
.
Example 2: f Z[X] is a monic polynomial of degree 10: f(X) = X
10
+

9
i=0
a
i
X
i
.
is the statement f factors into the product of a monic polynomial g of degree 4 and
a monic polynomial h of degree 6. Put a
10
= b
4
= c
6
= 1. For each eld K we have:
K [= g, h K[X]: g monic of degree 4 and h monic of degree 6 and f = gh
b
0
, b
1
, b
2
, b
3
Kc
0
, c
1
, c
2
, c
3
, c
4
, c
5
K:
10

i=0
a
i
X
i
=
4

j=0
b
j
X
j
6

k=0
c
k
X
k
b
0
, b
1
, b
2
, b
3
Kc
0
, c
1
, c
2
, c
3
, c
4
, c
5
K:
9

i=0
_
a
i
=

j+k=i
b
j
c
k

Thus is an elementary statement.

If L is the splitting eld of f(X) over Q, then [L : Q] 10!. If Gal(Q) and
res
L
= 1, then

Q() [= . Hence, (A()) = (A())
1
10!
. In particular, there are
innitely many primes p such that F
p
[= .
Example 3: f Z[X] is a monic irreducible polynomial of degree n > 1, is the
statement f(X) has no root, i.e., X: f(X) = 0. Let f(X) =

n
i=1
(X x
i
)
and L = Q(x
1
, . . . , x
n
). Then Q(x
1
) . . . , Q(x
n
) are conjugate to each another. Hence,
Gal(L/Q(x
1
)), . . . , Gal(L/Q(x
n
)) are proper subgroups of Gal(L/Q) which are conju-
gate to each other. By group theory, there exists
0
Gal(L/Q) which belogs to no
2
Gal(L/Q(x
i
)). Thus
0
x
i
,= x
i
for i = 1, . . . , n. If Gal(Q) and res
L
=
0
, then
x
i
,= x
i
for i = 1, . . . , n. Hence (A()) = (A())
1
n!
. In particular, there exists
innitely many primes p such that f(X) 0 mod p has no solution.
Remark: The irreduciblility of f is essential: f(X) = (X
2
2)(X
2
3)(X
2
6) is
a counter example. Another example: f(X) = (X
2
+ 3)(X
3
2). Here note that
=
3
satises
2
+ + 1 = 0. Hence =
1+

3
2
and therefore Q() = Q(

3). If

2 =
3

2 and (
3

2) =
2
3

2, then = , hence

3 =

3.
Example 4: Chevalleys theorem. Every homogeneous polynomial f F
p
[X
0
, . . . , X
n
]
of degree n has a non-trivial zero. If f(X) =

i
0
+...+i
n
=n
a
i
X
i
0
0
X
i
1
1
X
i
n
n
, then the
corresponding sentence is:
aX[
n

i=0
X
i
,= 0 & f(X) = 0].
Thus F
p
is a C
1
-eld. Hence, almost all

Q() is a C
1
-eld.
A proof of Chevalleys theorem: Source: Borevich and Shafarevich [BoS].
Motivation: If f K[X] vanishes at all x K and K is innite, then f = 0. For
K = F
p
we have X
p
X as a counter example.
Denition: A polynomial f F
p
[X
1
, . . . , X
n
] is reduced if deg
X
j
f < p for j =
1, . . . , n. Two polynomials f, g F
p
[X
1
, . . . , X
n
] ae equivalent if f(x) = g(x) for each
x F
n
p
.
Recall: x
p
= x for each x F
p
and
x
p1
=
_
1 if x ,= 0
0 if x = 0
Lemma 1: Every polynomial f F
p
[X
1
, . . . , X
n
] is equivalent to a reduced one, f

such that deg(f

) deg(f).
Proof: Replace each occurance of X
k
j
with k p by X
k(p1)
j
and proceed by induc-
tion.
3
Lemma 2: If f F
p
[X
1
, . . . , X
n
] is reduced and equivalent to 0, then f = 0.
Proof: Induction on n.
Lemma 3: if f, g F
p
[X
1
, . . . , X
n
] are reduced and equivalent, then f = g.
Proof: Apply Lemma 2 on f g.
Lemma 4: If f, g F
p
[X
1
, . . . , X
n
] are equivalent, and f is reduced, then deg
X
j
f
deg
X
j
g for j = 1, . . . , n. Hence deg(f) deg(g).
Proof: By Lemma 1, g is equivalent to a reduced polynomial g

with deg
X
j
(g

)
deg
X
j
(g) for j = 1, . . . , n. By Lemma 3, f = g

. Hence deg(f) deg(g).

Lemma 5 (Waring): Let f F
p
[X
1
, . . . , X
n
] be a polynomial of degree < n. Then
#a F
n
p
[ f(a) = 0 0 mod p.
Proof: Let A = x F
n
p
[ f(x) = 0 and let g(X) = 1 f(X)
p1
. Then
g(x) =
_
1 if x A
0 if x / A
and deg(g(X)) = (p 1) deg(f(X)) < (p 1)n. For each a A consider
h
a
(X) =
n

j=1
_
1 (X
j
a
j
)
p1
_
= (1)
n
X
p1
1
X
p1
n
+ lower terms.
Then, h
a
(X) is a reduced polynomial and deg(h
a
(X)) = (p 1)n. Also,
h
a
(x) =
_
1 if x = a
0 if x ,= a
It follows that
(1) h(X) =

aA
h
a
(X) = (1)
n
[A[X
p1
1
X
p1
n
+ lower terms
is a reduced polynomial and
h(x) =
_
1 if x A
0 if x / A.
4
Thus h(X) is equivalent to g(X). By Lemma 4, deg(h(X)) deg(g(X)) < (p 1)n.
Hence, the highest term in the right hand side of (1) vanishes. That is [A[ = 0 in F
p
.
Equivalently [A[ 0 mod p in Z.
Conclusion of the proof of Chevalleys theorem: If f(X
0
, . . . , X
n
) is a form of degree
n, then the number of its zeros is a multiple of p. Since f has the zero (0, . . . , 0) it must
have at least another one.
Example 5: Let G be a nite group, K a eld, t a transcendental element over K.
Then K(t) has a Galois extension F such that Gal(F/K(t))

= G is an elementary
statement about K. A special case of a theorem of Fried and Volklein [FrV] says that
is true in

Q() for almost all . Hence:
For each nite group G there exists a nite set S of primes such that for each
p / S, F
p
(t) has a Galois extension F with Galois group G.
Example 6: Let f Z[X
1
, . . . , X
n
] be an absolutely irreducible polynomial. Then, by
a theorem of Weil (= Riemann hypothesis for function elds) f(X) has a zero in F
p
for
almost all p. It follows that f(X) has a zero in

Q() for almost all .
The proof of the transfer theorem travels along a rectagle each of its verticis
consists of another family of elds:

F
p
/T ultra product of
nite elds

Q() xed elds

of Gal(K)
F
p
nite elds Q
Three basic theorems make this travel possible:
The Riemann hypothesis for function fields (= Weils theorem): For each d
there exists p
0
such that for all primes p p
0
, each absolutely irreducible polynomial
f F
p
[X, Y ] of degree d has a zero (x, y) F
2
p
.
Hilbert irreduciblity theorem: For each irreducible polynomialf Q[X, Y ] there
exist innitely many x Q such that f(x, Y ) is irreducible in Q[Y ].
5
Chebotarev density theorem: Let L be a nite Galois extension of Q and let (
be a conjugacy class in Gal(L/Q). Then the Dirichlet density of the set of all primes p
whose Artin symbol
_
L/Q
p
_
= ( is [([/[L : Q].
Examples for the Artin symbol: (a) L = Q(

a) (a nonsquare), p an odd prime,

_
L/Q
p
_
=
_
a
p
_
=
_
1 x: x
2
0 mod p
1 otherwise.
(b) L = Q(
n
), p n. If
_
L/Q
p
_
= , then

n
=
p
n
.
6
1. Innite Galois theory
The basic theorem of finite Galois theory: Let N/K be a nite Galois exten-
sion.
G = Gal(N/K) its Galois group;
L = L(N/K) is the collection of all intermediate elds K N L;
Sub(G) is the collection of all subgroups of G
For each L L we have Gal(N/L) Sub(G);
For each H Sub(G) we have N(H) = x N [ x = x for all H L.
The map L Gal(N/L) is a bijection L(N/K) Sub(G) whose inverse is
H N(H).
Moreover: L/K is Galois if and only if Gal(N/L) G. In this case we have the following
short exact sequence
1
-
Gal(N/L)
-
Gal(N/K)
-
res
L
Gal(L/K)
-
1
The Galois correspondence satises the following rules:
(1a) L
1
L
2
if and only if Gal(N/L
1
) Gal(N/L
2
);
(1b) H
1
H
2
if and only if N(H
1
) N(H
2
);
(1c) N(H
1
) N(H
2
) = N(H
1
, H
2
));
(1d) Gal(N/L
1
L
2
) = Gal(N/L
1
), Gal(N/L
2
));
(1e) N(H
1
H
2
) = N(H
1
)N(H
2
);
(1f) N(H

) = N(H)

;
(1g) Gal(N/L

) = Gal(N/L)

;
(1h) If L/K is Galois and M/K is an arbitrary extension, then LM/M is Galois and
res
L
: Gal(LM/M) Gal(L/L M) is an isomorphism;
(1i) If in addition M/K is Galois, then (L M)/K is Galois and N = LM is Galois
over K; in this case
Gal(N/K)

= (, ) Gal(L/K) Gal(M/K) [ res
LM
= res
LM
,
where (res
L
, res
L
). In particular, if L M = K, then Gal(N/K)

=
Gal(L/K) Gal(M/K).
7
The Krull topology. Now N/K is an arbitrary Galois extension. Let L
0
be the
collection of all intermediate elds K L N such that L/K is nite Galois. Basic
open neighborhoods of 1 in G = Gal(N/K) are Gal(N/L) with L L
0
. If L
1
, L
2
L
0
,
then L
1
L
2
L
0
and Gal(N/L
1
) Gal(N/L
2
) = Gal(N/L
1
L
2
).
Basic open neighborhood of G is Gal(N/L) with L L
0
.
If
1
, . . . ,
n
G represent G modulo Gal(N/L), then
G =
n
_

i=1

i
Gal(N/L).
Hence each Gal(N/L) is both open and closed. Thus G is a totally disconnected
topological group. In particular, G is Hausdor. To prove that G is compact, we give
another presentation to the Krull topology.
To each L/L

L
0
with L L

we associate the restriction map

res
L

,L
: Gal(L

/K) Gal(L/K)
and note that if L

L
0
, then res
L

,L
= res
L

,L
res
L

,L
. Then L, res
L

,L
)
L,L

L
0
is an inverse system of nite groups.
Profinite groups. In general we consider a set I with a partial ordering such
that
(2) for all i, j I there exists k I with i, j k.
Suppose that for each i I there is a set S
i
and for all i, j I with i j there is a
map
ji
: S
j
S
i
such that
(3a) if i j k, then
ki
=
ji

kj
;
(3b)
ii
is the identity map of S
i
.
We call S
i
,
ij
)
i,jI
and inverse system. Then inverse limit of S
i
,
ij
) is the set
S = s

iI
S
i
[

ij

ji
(s
j
) = s
i

together with the maps

i
: S S
i
given by
i
(s) = s
i
. We denote it by S = lim

S
i
.
If all S
i
belong to a certain category, then we demand that the
ij
are morphisms
in that category. Thus, if S
i
are topological spaces, then
ji
are continuous maps. In
8
this case we endow S with the topology induced by that of

S
i
. Then
i
are continuous.
If S
i
are groups, then
ji
are homomorphisms. In this case, S is a subgroup of

S
i
. If
S
i
are nite groups, then we consider them also as discrete topological spaces. In this
case we say that lim

S
i
is a pronite group.
Example 1.1: Gal(L/K), res
L

L
)
L,L

L
0
is an inverse system of nite groups in which
each res
L

L
is surjective
Lemma 1.2: If each S
i
is a Hausdor space, then S is a closed subset of

S
i
.
Proof: Let s

S and j i. We have to prove that
ji
(s
j
) = s
i
. Assume that

ji
(s
j
) ,= s
i
. Choose open disjoint subsets U
i
and U

i
of S
i
such that s
i
U
i
and

ji
(s
j
) U

i
. Then U = U
i

1
ji
(U

i
)

k=i,j
S
k
is an open neighborhood of s in

S
i
. So, there exists t U S. In particular, t
i
U
i
and t
j

1
ji
(U

i
). Hence
t
i
=
ji
(t
j
) U

i
and therefore U
i
U

i
,= , a contradiction. Conclude that s S.
Lemma 1.3: The inverse limit of a nonempty compact Hausdor spaces is a nonempty
compact Hausdor space.
Proof: By Lemma 1.2, we have only to prove that S = lim

S
i
is nonempty. Indeed, let
k j. As S
j
and S
k
are Hausdor and
kj
is continuous, its graph

kj
= (s
k
, s
j
)
S
k
S
j
[
kj
(s
k
) = s
j
is nonempty and closed. Hence,
R
kj
= s

iI
S
i
[
kj
(s
k
) = s
j
= (
k

j
)
1
(

kj
)
is nonempty and closed.
If k
1
j
1
, . . . , k
m
j
m
, we choose j j
1
, . . . , j
m
and k k
1
, . . . , k
m
, j. Then
R
kj
R
k
1
j
1
R
k
m
j
m
. It follows from the compactness of

iI
S
i
that S =

kj
R
kj
is nonempty.
Lemma 1.4: The inverse limit S = lim

S
i
of totally disconnected spaces is totally
disconnected.
Proof: Assume that the connected component C(x) of a point x S contains another
point y. Then there exists i such that
i
(x) ,=
i
(y). Since
i
(C(x)) is connected and
9
contains both
i
(x) and
i
(y) we must have
i
(x) =
i
(C(x)) =
i
(C(y)) =
i
(y),
a contradiction.
Corollary 1.5: The inverse limit S = lim

S
i
of nite nonempty discrete spaces is
nonempty, compact, Hausdor, totally disconnected space. A basis for the open neigh-
borhoods of a point x S is the collection
1
i
(
i
(x)) [ i I. It consists of open and
closed sets. We call S a pronite space.
Remark 1.6: In each topological space X the connected component C(x) of a point
x X is contained in each open and closed set that contains x. So, if X is Hausdor
and has a basis for its topology which consists of open and closed sets, then C(x) = x.
Conversely, it can be shown, that if X is a totally disconnected, compact Hausdor
space, then X is a pronite space.
Corollary 1.7: A pronite group G = lim

G
i
is a totally disconnected, compact
Hausdor group. A basis for the open neighborhoods of 1 consist of all normal open
subgroups. Each open subgroup H of G is closed and has a nite index. Each closed
subgroup H of nite index is open. Each closed subgroup H of G is an intersection of
open subgroups.
Proof: We prove only the latter statement. Suppose that x G belongs to every open
subgroup that contains H. Then x NH for each open normal subgroup N of G. Thus
xN H ,= . By compactness, there exists y

N
xN H. Hence yx
1

N = 1.
Conclude that x = y H.
Example 1.8: Let N/K be a Galois extension and denote, as before, the collection of
all nite Galois extensions L of K contained in N by L
0
. Then
Gal(N/K)

= lim

LL
0
Gal(L/K) ([
L
)
LL
0
as topological groups. In particular, Gal(N/K) is a pronite group, hence compact.
Proposition 1.9 (The main theorem of Galois theory): Let N/K be an arbitrary
Galois extension. Then, the map L Gal(N/L) maps the family of all intermediate
10
elds between K and N onto the set of all closed subgroups of Gal(N/K). The inverse
map is given by H N(H). This correspondence satises the rule (1)
Proof: We prove here only that the correspondence is bijective.
Let L be an intermediate eld. Then L N(Gal(N/L)). Conversely, each x
N(H) is contained in a nite Galois extension M of K contained in N. Since the map
res: Gal(N/L) Gal(M/M L) is surjective, x = x for each Gal(M/M L).
Conclude from nite Galois theory that x M L. Hence, N(Gal(N/L)) = L.
Conversely, let H be a closed subgroup of Gal(N/L) and let L = N(H). Then
H Gal(N/L). To show that Gal(N/L) belongs to H it suces to show that
belongs to the closure of H. Indeed, let M N be a nite Galois extension of K. Then
ML = M(res
M
H). Hence, by nite Galois theory, res
M
Gal(M/ML) = res
M
H.
Therefore H Gal(N/M) is nonempty. Conclude that Gal(N/N(H)) = H.
Remark 1.10: Artin, Leptin, Waterhouse. For each pronite group G there exists a
Galois extension N/K such that Gal(N/K)

= G.
The group (ring) Z
p
of p-adic integers. It is dened as the inverse limit Z
p
=
lim

Z/p
i
Z with respect to the canonical maps Z/p
j
Z Z/p
i
Z, z +p
j
Z z +p
i
Z. Z
p
is both a pronite group and a pronite ring. The ring Z naturally embeds into Z
p
, by
z (z +p
i
Z)
i=1,2,3,...
. The ring Z is dense in Z
p
but is not equal to Z
p
:
(
n1

i=0
p
i
+p
n
Z)
n=1,2,...
Z
p

Z,
(except for p = 2; change the example in this case.) Each element of Z
p
can be uniquely
written as a formal power series a =

i=0
a
i
p
i
, with 0 a
i
p 1. The element a is
invertible in Z
p
if and only if a
0
,= 0.
The lattice of subgroups of Z
p
.
(4a) p
i
Z
p
is an open subgroup (ideal) of Z
p
of index p
i
. It is isomorphic to Z
p
(as
groups).
Proof: The subgroup p
i
Z
p
is contained in the kernel of
i
: Z
p
Z/p
i
Z. Conversely
let x = (x
1
, x
2
, . . .) Z
p
be in the kernel of
i
. Then x
j+i
x
i
0 mod p
i
and
11
hence x
j+i
= p
i
y
j
for each j 1. If k j, then x
k+i
x
j+i
mod p
j+i
and hence
y
k
y
j
mod p
j
. So, y = (y
1
, y
2
, . . .) Z
p
and satises p
i
y = x. Indeed, p
i
y
j
= x
j+i

x
j
mod p
j
for j = 1, 2, . . . . It follows that p
i
Z
p
= Ker(
i
). Hence Z
p
/p
i
Z

= Z/p
i
Z and
therefore p
i
Z
p
has index i in Z
p
.
(4b) If H is a subgroup of Z
p
of a nite index, then H = p
i
Z
p
for some i.
Proof: Let (Z
p
: H) = p
i
k with p k. Then p
i
Z
p
= p
i
kZ
p
H. Thus p
i
= (Z
p
:
p
i
Z
p
) (Z
p
: H) = p
i
k p
i
. It follows that p
i
Z
p
= H.
(4c) If H is a closed subgroup of Z
p
of innite index, then H = 0.
The Pr

ufer group (ring)

Z. It is dened as the inverse limit

Z = lim

Z/nZ
with respect to the natural maps x + nZ x + mZ if m[n. So,

Z is both a pronite
group and a pronite ring.
(5a) Z naturally embeds as a dense subset of

Z.
(5b) Each n

Z is an open subgroup of

Z of index n. In fact, we have a short exact
sequence 0
-
n

Z
-
Z
-
Z/nZ
-
0.
(5c) If H is a subgroup of

Z of index n, then H = n

Z.
(5d)

Z

Z
p
.
Proof: The canonical maps

Z
p
-

p|n
Z/p
k(p)
Z
-
Z/nZ with n =

p
k(p)
(Chinese remainder theorem) are compatible and therefore dene a continuous homo-
morphism f:

Z
p


Z, which is injective. It is the identity on Z. So, f(

Z
p
) is a
compact subset of

Z which contains a dense subset. Thus f is also surjective. Finally,
both spaces are Hausdor and compact. Hence f is a homeomorphism.
The absolute Galois group of a finite field.
F
q
is the eld with q elements.
F
q
= x

F
q
[ x
q
= x.
F
q
n is the unique extension of F
q
of degree n
12

n
is the Frobenius generator of Gal(F
q
n/F
q
),
n
(x) = x
q
.
Commutative diagram of isomorphisms
Z/nZ
-
Z/mZ
? ?
Gal(F
q
n/F
q
)
-
res
Gal(F
q
m/F
q
)
1 +nZ
-
1 +mZ
? ?

n
-

m
It follows that Gal(F
q
)

=

Z, the Frobenius automorphism is mapped onto 1 and
generates Gal(F
q
).
In particular the discrete subgroup
n
[ n Z has the same xed eld, namely
F
q
, as the whole group Gal(F
q
).
13
2. The Riemann Hypothesis over nite elds
A place of a eld F is a function : F K such that (a + b) = (a) + (b)
and (ab) = (a) +(b), whenever the right hand side is dened;
A Prime divisor is an equivalence class of places;
is an ordered abelian group;
A valuation of a eld is a map v: F such that v(ab) = v(a) + v(b),
v(a +b) minv(a), v(b) and v(a) = if and only if a = 0;
A valuation ring R of a eld, maximal ideal, residue eld.
Correspondence between equivalence classes of places, equivalence classes of valuations
and valuation rings of F.
Example 2.1: Z
p
is a valuation ring of Q
p
;
R a unique factorization domain with quotient eld F. Each prime element p R
corresponds to a prime divisor of F:
v
p
_
a
b
p
i
_
= i, a, b R, p a, b;

p
_
a
b
p
i
_
=
_
0 i0
a/

b i = 0; R/pR is the residue eld and a = a +pR

i0
R
p
=
a
b
[ a, b R, p b
R can be Z with the primes 2, 3, 5, 7, . . . . This gives all valuations of Q. In addition
there is the innite prime associated with the absolute value.
Alternatively, R can be K[t]; and the primes are irreducible polynomials p(t) =
t
n
+a
n1
t
n1
+ +a
0
. The eld F = K(t) has one more valuation:
v

_
f(t)
g(t)
_
= deg(g) deg(f).
All these are example of discrete valuations.
Chevalleys extension theorem: Let
0
: R
0
K be a homomorphism of an
integral domain R
0
into a eld K. Let F be a eld containing R
0
. Then
0
can be
extended to a place : G

K .
Let F be a nite extension of K(t) such that K is algebraically closed in F. Then
F is a function eld of one variable over K. Each prime p of K(t)/K extends to
14
nitely many primes p
1
, . . . , p
r
of F. The residue eld

F
p
i
is a nite extension of K and
deg(p
i
) = [

F
p
i
: K].
We denote the space of all prime divisors of F/K by 1(F/K). It is the Riemann
space of F/K.
A divisor of F is a formal sum a =

s
i=1
a
i
q
i
with a
i
Z. We write v
q
i
(a) = a
i
.
Degree of a divisor: deg(a) =

s
i=1
a
i
deg(q
i
).
To each f F

there corresponds a principal divisor

(f) =

p
v
p
(f)p
where p ranges over all primes of F. However, v
p
(f) ,= 0 only for nitely many primes.
The zero divisor of f: (f)
0
=

v
p
(f)>0
v
p
(f)p;
The pole divisor of f: (f)

v
p
(f)<0
v
p
(f)p;
Then (f) = (f)
0
(f)

is the divisor of f. We have (fg) = (f) + (g). By Chevalleys

extension theorem, (f) = 0 if and only if f K

.
If f / K, then deg(f)
0
= deg(f)

= [F : K(f)] and deg(f) = 0.

Partial order among divisors: a b if and only if v
p
(a) v
p
(b) for all p.
Vector space over K attached to a divisor a:
L(a) = f F [ a + (f) 0
dim(a) = dim
K
L(a) < .
Adele of F is a function : 1(F/K) F such that v
p
(
p
) 0 for almost all p.
We denote the F-algebra of adeles of F/K by A.
Another vector space over K attached to a divisor a:
(a) = A[ v
p
(a) +v
p
(
p
) 0 for all p 1(F/K)
(a) = dim
K
(A/(a) +F) < .
Riemann-Roch theorem: There exists a unique nonnegative integer g (called the
genus of F/K) such that for each divisor a:
dim(a) = deg(a) + 1 g +(a).
15
Moreover:
(1a) dim(0) = 1 (f F

implies dim((f)) = 1);

(2) deg(a) < 0 implies dim(a) = 0;
(3) deg(a) > 2g 2 implies dim(a) = deg(a) + 1 g.
Example: The genus of K(t) is 0. Conversely, if g(F) = 0 and if F has a prime divisor
of degree 1, then there exists u F such that F = K(u).
The function field of a curve f(X, Y ) = 0. Let f K[X, Y ] be an absolutely
irreducible polynomial of degree d over a eld K. Thus f is irreducible in the ring

K[X, Y ]. Choose a transcendental element x over K and an element y

K(x) such
that f(x, y) = 0. Then (x, y) is a generic point of the curve : f(X, Y ) = 0.
By assumption f(x, Y ) is irreducible in

K[x][Y ] and primitive (i.e., the greatest
common divisor in K[x] of the coecients of f(x, Y ) is 1. Hence, by Gau lemma,
f(x, Y ) is irreducible in

K(x)[Y ]. It follows that [K(x, y) : K(x)] = deg
Y
f(x, Y ) =
[

K(x, y) :

K(x)]. Hence K(x, y)

K(x) = K(x) and since K(x)

K = K, also K(x, y)

K = K. Indeed, if K(x) K(x, y)

K(x), then [

K(x, y) :

K(x)] [K(x, y) : K(x, y)

K(x)] < [K(x, y) : K(x)].

Thus F = K(x, y) is a function eld of one variable over K.
Proposition ([FrJ, Cor. 4.8]): The genus of F/K is at most
1
2
(d 1)(d 2).
Theorem (Weil): Let F be a function eld of genus g over F
q
. Denote the number of
prime divisors of F of degree 1 by N. Then
(3) [N (q + 1)[ 2g

q.
Remark: One associates zeta functions to F:

F
(s) =

a0
q
deg(a)s
Z
F
(t) =

a0
t
deg(a)
where a ranges over all nonnegative divisors of F/F
q
. Then Z
F
(t) has the following 2g
zeros w
1
i
, i = 1, . . . , 2g. They satisfy
(4) N (q + 1) = w
1
+ +w
2g
.
16
The Riemann Hypothesis for F, which Weil proved, states that all those zeros lie on the
line Re(s) =
1
2
. For each i there exists a zero s
i
of
F
(s) such that q
s
i
= w
1
i
. Hence,
[w
i
[ = q
Re(s
i
)
=

q. So, (2) is a consequence of (3).
Combine Weils theorem with the estimate for the genus to get
(4) [N (q + 1)[ (d 1)(d 2)

q.
Theorem ([FrJ, Thm. 4.9]): Let f F
q
[X, Y ] be an absolutely irreducible polynomial
of degree d. Let N

= (a, b) F
2
q
[ f(a, b) = 0. Then
(q + 1) (d 1)(d 2)

q d N

q + 1 + (d 1)(d 2)

q.
Corollary:
(a) If q > (d 1)
4
, then N

> 0.
(b) For each m there exist q
0
= q
0
(d, m) such that for all q q
0
we have N

m.
17
3. Elements of model theory
First order predicate calculus. A signature is a triple (, , K) consisting of
two functions: : I N, : J N, and a set K.
We associate a language L(, , K) with this signature. The letters of L(, , K)
are:
Variable symbols X
1
, X
2
, X
3
, . . .
Constant symbols c
k
for each k K

i
-ary relation symbol R
i
, i I

j
-ary function symbol F
j
, j J
equality symbol =
negative symbol , disjunction symbol , existential quantier symbol
parenthesis ( ) and brackets [ ].
Example: We denote the language of ring theory by L(ring). It has of two binary
function symbols: + and , and two constant symbols 0 and 1.
A string is a nite sequence of letters. words are terms and formulas.
The collection of terms is the smallest set of strings that contains
X
i
, i = 1, 2, 3, . . .
C
k
, k K
and satises the following rule:
For each j J, if t
1
, . . . , t

j
are terms, then so is F
j
(t
1
, . . . , t

j
).
For example, in L(ring), polynomials in several variables are terms.
Atomic formulas are:
t = t

, where t, t

are terms
R
i
(t
1
, . . . , t

i
), where i I and t
1
, . . . , t

i
are terms.
The set of formulas is the smallest collection of strings that contains all atomic
formulas and satisfy the following rules:
is a formula implies that [] is a formula;

1
,
2
are formulas imply that
1

2
is a formula;
is a formula implies that (X
i
)[] is a formula.
18
Induction by structure for terms and formulas.
Free occurrence of X in a formulas :
Any occurrence of X in an atomic formula is free.
If an occurrence of X in is free, is any formula and Y ,= X, then the occurrence
of X in [], , and (Y )[] is free.
If X has a free occurrence in , then X is a free variable of . We write (X
1
, . . . , X
n
)
to indicate that the free variables of belong to the set X
1
, . . . , X
n
. A formula without
a free variables is a sentence.
Examples: (X)[X
2
= 2] is a sentence. X
n
1
+ X
n
2
= X
n
3
is a formula with three free
variables.
Abbreviations:
for [ ] conjunction
for implication
for [ ] [ ] double implication
(X
i
)[] for (X
i
)[] universal quantier
_
n
i=1

i
for
1

2

n
_
n
i=1

i
for
1

2

n
.
A formula in prenex normal form:
(Q
1
X
1
) (Q
m
X
m
)
_
m

i=1
n

j=1
[f
ij
(X, Y) = 0 g
ij
(X, Y) ,= 0]

where each Q
i
is either or and f
ij
, G
ij
are polynomials with coecients in a eld
K.
Structures. A structure for a language L(, , K) is a system
/ = A,

R
i
,

F
j
, c
k
[ i I, j J, k K)
A is a nonempty set the domain of /;

R
i
A

i
is a
i
-ary relation on A;

F
j
: A

j
A is a
j
-ary function on A;
c
k
A is a constant.
19
Substitution f: X
1
, X
2
, . . . A, f(X
i
) = x
i
, extends to terms:
f(c
k
) = c
k
;
f(F
j
(t
1
, . . . , t

j
)) =

F
j
(f(t
1
), . . . , f(t

j
).
Truth value of a formula under the substitution f is dened to be either true
or false by induction on the structure:
t = t

is true if and only if f(t) = f(t

);
R
i
(t
1
, . . . , t

i
) is true if and only if (f(t
1
), . . . , f(t

i
))

R
i
.
Suppose that the truth values of ,
1
,
2
have been dened for all possible sub-
stitutions. Then under f
is true if and only if is false;

1

2
is true if and only if
1
is true or
2
is true;
(X
l
)[] is true if and only if there exists x A such that is true under the substitution
g which is dened by: g(X
l
) = x and g(X
m
) = f(X
m
) if m ,= l.
One proves that the truth value of a formula (X
1
, . . . , X
n
) depends only on
f(X
1
) = x
1
, . . . , f(X
n
) = x
n
. We write then / [= (x
1
, . . . , x
n
). If is a sentence,
then its truth value is independent of f. Either / [= of / ,[= .
Models. A theory for a language L = L(, , K) is a set of sentences of L. A
structure / for L is a model of T is / [= for each T. We then write / [= T.
A theory is a set of axioms for T if [= T. That is, each model of is also a
model of T.
Example: The theory of elds has the following set of axioms:
(X)(Y )(Z)[(X +Y ) +Z = X + (Y +Z)]
(X)(Y )[X +Y = Y +X]
(X)[X + 0 = X]
(Y )(X)[X +Y = 0]
(X)(Y )(Z)[(XY )Z = X(Y Z)]
(X)(Y )[XY = Y X]
(X)(X 1 = X)
20
(X)[X ,= 0 (Y )[XY = 1]]
0 ,= 1
(X)(Y )(Z)[X(Y +Z) = XY +XZ]
Let R be an integral domain. Add the elements of R to the constant symbols
of L(ring) to obtain the language L(ring, R). Add the positive diagram of R to the
axioms :
(R): All equalities a +b = c and a

= c

that hold in R.
A model of (R) is a eld that contains the set

R = a [ a R and satises
a +

b = c and a

b = c whenever a + b = c and ab = c, respectively. Thus

R is a
homomorphic image of R (e.g., R = Z and

R is a eld of characteristic p).
An elementary statement about models of (R) is a mathematical statement
that applies to each model of (R) and for which there exists a sentence of L(ring, R)
which is true in a model if and only if the statement is true in that model.
Example: f(X
1
, . . . , X
n
) is a polynomial of degree d with coecients in R. Then f(X)
is irreducible is equivalent to
_
l+m=n
there exist no polynomials g(X) and h(X) of
degrees d and e respectively such that f(X) = g(X)h(X). The phrase there exists a
polynomial g(X) of degree d should be replaced by (u
1
) (u
k
), where u
1
, . . . , u
k
are the coecients of g. The equality f(X) = g(X)h(X) should be replaced by a
conjunction of equalities between the coecients of monomials of the same degrees on
both sides.
Elementary equivalent structures. / B
Extension of structures. Let / = A,

R
i
,

F
j
, c
k
), B = B, R

i
, F

j
, c

k
). Then
/ B, if
A B and for a
1
, a
2
, a
3
, . . . A we have
(a
1
, . . . , a

i
) R
i
if and only if (a
1
, . . . , a

i
) R

F
j
(a
1
, . . . , a

j
) = F

j
(a
1
, . . . , a

j
)
c
k
= c

k
.
21
Elementary extensions. / B: / B and for all a
1
, . . . , a
n
A we have
/ [= (a
1
, . . . , a
n
) if and only if B [= (a
1
, . . . , a
n
).
Example: If a eld K is an elementary subeld of a eld F, then K is algebracially
closed in F.
Indeed, take x

K F and f = irr(x, K) = X
n
+a
1
X
n1
+ +a
n
.
F [= (X)[X
n
+a
1
X
n1
+ +a
n
= 0].
Hence
K [= (X)[X
n
+a
1
X
n1
+ +a
n
= 0].
Hence, deg(f) = 1 and x K.
The Skolem-L

owenheim theorem: Let L be a countable language. Consider a

structure B = B, S
i
, G
j
, d
k
) for L, and let A
0
a countable subset of B. Then B has a
countable elementary substructure / = A, R
i
, F
j
, c
k
) such that A
0
A.
Proof: By induction on n dene sets A
0
A
1
A
2
B. If A
n
has been dened,
then A
n+1
consists of all d
k
with k K. In addition, if (X
1
, . . . , X
m
) is a formula,
x
1
, . . . , x
m1
A
n
and B [= (X
m
)(x
1
, . . . , x
m1
, X
m
), then choose one x
m
B such
that B [= (x
1
, . . . , x
m1
, x
m
) and add it to A
n+1
.
Dene A =

i=0
A
i
, R
i
= S
i
A

i
, F
j
= G
j
[
A

j . Then prove by induction on

formulas (X
1
, . . . , X
n
) and for x A
n
:
/ [= (x) if and only if B [= (x).
Note that each term t(X
1
, . . . , X
n
) has the same values on A

in both / and B. Check

for atomic formulas and then for compound formulas.
Ultrafilters. An ultralter of a set S is a family T of subsets of S with the
following property:
(1a) / T
(1b) A T and A B S imply B T
(1c) A T and B T imply A B T
22
(1d) For each A S, either A T or S A T.
It follows
(1e) S T
(1f) A B T implies A T or B T.
We refer to sets in T as big. If a property of elements of S is true for all a S
that belong to a set in T, we say that it holds for almost all a S.
Example: For each a S, the family
T = A S [ a A
is the principal ultralter generated by a.
If T is a nonprincipal ultralter of S, then T contains all conite subsets of S.
A family T of subsets of S is a lter if it satises only (1a), (1b), and (1c).
Lemma: a lter T of S is an ultralter if and only it is maximal lter.
Proof: If S A / T, then
T

= T B D[ A B S and D T
is a lter that contains T. So, if T is maximal, then T

= T and A T.
A family T
0
of subsets of S has the nite intersection property if it satises
(1a) and (1c).
Lemma:
(a) Each family T
0
that satises the nite intersection property is contained in a lter
T
1
.
(b) Each lter T
1
is contained in an ultralter T.
Proof of (a): T
1
= E S [ D
1
, . . . , D
m
T
0
: D
1
D
m
F.
Proof of (b): Use Zorns lemma to choose a maximal lter T that contains T
1
.
23
Lemma: Let T
0
be a family of sets with the following property:
A
1
, . . . , A
m
T
0
implies A
1
A
m
is an innite set.
Then there exists a nonprincipal ultralter T of S that contains T
0
.
Proof: The family T
1
that consists of T
0
and all conite sets of S has the nite
intersection property. Take T to be an ultralter that contains T
1
.
Ultraproducts. Consider a language L = L(, , K), a set S and an ultralter T
on S. For each s S, let /
s
= A
s
, R
is
, F
js
, c
ks
)
iI, jJ, kK
be a structure of the
language L.
Construct the ultraproduct of the structures o
s
modulo T: / =

sS
/
s
/T.
It is the structure / = A, R
i
, F
j
, c
k
), where
A is the cartesian product

sS
A
s
modulo the equivalent relation:
a b s S [ a
s
= b
s
T.
Use representatives instead of classes:
(a
1
, . . . , a

i
) R
i
s S [ (a
1s
, . . . , a

i
s
) R
is
T
F
j
(b
1
, . . . , b

j
) is the equivalence class of F
js
(b
1s
, . . . , b

j
s
)
c
k
is the equivalence class of c
ks
.
Claim: If t(X
1
, . . . , X
n
) is a term of L and x
1
, . . . , x
n
A, then
s S [ t(x
1
, . . . , x
n
)
s
= t(x
1s
, . . . , x
ns
) T.
The fundamental property of ultraproduct ( Los):
For each formula (X
1
, . . . , X
n
) and all x
1
, . . . , x
n
A we have:
/ [= (x
1
, . . . , x
n
) s S [ /
s
[= (x
1s
, . . . , x
ns
) T.
Corollary: If is a sentence of L, then
/ [= s S [ /
s
[= T.
24
Example: Ultraproduct of elds. If /
s
is a eld, so is / =

/
s
/T.
If each /
s
is algebraically closed, so is /.
If K
p
is of characteristic p, and T is nonprincipal, then

K
p
/T is of characteristic 0.
An ultraproduct of perfect elds of characteristic p is perfect.
Saturation theorem: Let / =

sN
/
s
/T be a nonprincipal ultraproduct. Suppose
that r(1) < r(2) < r(3) < is an increasing sequence of positive integers and for each
n N,
n
(X
1
, . . . , X
r(n)
) is a formula of L such that
(2) / [= (X
1
) (X
r(n)
)
_
n

i=1

i
(X
1
, . . . , X
r(i)
)

Then there exist x

1
, x
2
, x
3
, . . . such that
(3) / [=
n
(x
1
, . . . , x
r(n)
) for n = 1, 2, 3, . . .
We say that / is
1
-saturated.
Proof: To simplify notation suppose that r(n) = n. Then
D
n
= s N[ /
s
[= (X
1
) (X
n
)
_
n

k=1

k
(X
1
, . . . , X
r(i)
)

belongs to T and D
1
D
2
D
3
. Since T is nonprincipal D

n
= D
n

1, 2, . . . , n
also belongs to T, D

1
D

2
D

3
, and

n=1
D

n
= .
For each s D

n+1
choose x
1s
, . . . , x
ns
A
s
such that
/
s
[=
n

k=1

k
(x
1s
, . . . , x
ks
).
Since,

n=1
D

n
= , this denes x
ks
for each s D

k
. For s N

k
dene x
ks
arbitrarily. Then x
k
satises (3).
Example: If K
n
/K is separable of degree > n, then

K
n
/T is not algebraic over
K
N
/T.
Let K
n
is the eld generated over Q by all Galois extensions of order n. The
order of each element of Gal(K
n
/K) is at most n. Hence K
n
,=

Q. But

K
n
/T
contains

Q.
25
4. Ultraproducts of nite elds
Ultraproducts of nite elds have the elementary properties which almost all nite elds
have. For example absolute irreducibility is an elementary property.
Lemma: Let f K[X
0
, . . . , X
n1
] be a polynomial of degree < d. If f factors over

K,
then it factors over an extension of K of degree < d
n
!.
Proof: Suppose that f(X) = g(X)h(X) with g, h

K[X]. Make the Kronecker
substitution
X
0
T, X
1
T
d
, . . . , X
n1
T
d
n1
to get polynomials f

(T), g

(T), h

(T) with f

(T) = g

(T)h

(T). Thus, if
f(X) =

a
i
X
i
0
0
X
i
1
1
X
i
n1
n1
,
then
f

(T) =

a
i
T
i
0
+i
2
d++i
n
d
n1
.
Note that (i
0
, . . . , i
n1
) ,= (j
0
, . . . , j
n1
) implies
i
0
+i
2
d + +i
n1
d
n1
,= j
0
+j
2
d + +j
n1
d
n1
.
Hence, the set of coecients of f(X) (resp., g(X), h(X)) is the same as that of f

(T)
(resp., g

(T), h

(T)).
Since deg(f

(T)) < d
n
, the coecients of g

(T) and h

(T), hence those of g(T)

and h(T), belong to an extension of degree < d
n
! of K, namely, the splitting eld of
f

(T) over K.
Proposition: The close each absolutely irreducible polynomial of degree < d in
K[X
1
, . . . , X
n
] is elementary.
Proof: Let f K[X
1
, . . . , X
n
] be a polynomial of degree d. Then
(1) f is absolutely irreducible
is equivalent to
(2a) f is irreducible over each separable extension of K of degree < d
n
! and
26
(2b)
_
n
i=1
f
X
i
,= 0 (which means that f is not a pth power of a polynomial over

K,
where p = char(K)).
If L is a separable extension of degree < d
n
!, then L = K(y), where h = irr(y, K)
has degree < d
n
!. Thus (2a) is equivalent to
(3) There exist no polynomials g
1
, g
2
, g
3
K[X, Y ] such that
(a) deg
Y
(g
i
) < d
n
!, deg
X
(g
i
) < d
(b) deg
Y
(g
3
) < 2(d
n
!) deg
X
(g
3
) < d
(c) f(X) = g
1
(X, Y )g
2
(X, Y ) +g
3
(X, Y )h(Y ).
Corollary: Let F =

sS
F
s
/T be a nonprincipal ultraproduct of nite elds such
that s S [ [F
s
[ n is nite for each n. Then each absolutely irreducible polynomial
f F[X, Y ] has innitely many zeros in F.
Corollary: Let F =

sS
F
s
/T be a nonprincipal ultraproduct of nite elds such
that s S [ [F
s
[ n is nite for each n. Then each absolutely irreducible polynomial
f F[X
1
, . . . , X
n
] has innitely many zeros in F.
Proof: Use Bertini-Noether to reduce to the case where n = 2.
To investigate the absolute Galois group of ultraproducts of nite elds, we have
to prove that realizability of a nite group over a eld is an elementary statement on
this eld.
Lemma: Let L = K(x) be a Galois extension of degree n of a eld K and let f =
irr(x, K). Let p
i
K[X] be polynomials of degree at most n 1 such that p
1
(X) = X
and x
i
= p
i
(x), i = 1, . . . , n are the n roots of f. Let G be a subgroup of S
n
. Then
Gal(f, K) = G if and only if
(4)

G
n

i=1
_
p
(i)
(x) = p
i
(p
(1)
(x))

S
n

G
n

i=1
_
p
(i)
(x) ,= p
i
(p
(1)
(x))

.
Proof: We identify Gal(f, K) with a subgroup of S
n
through its action on x
1
, . . . , x
n
.
Thus, each Gal(f, K) satises
(5) x
i
= x
(i)
, i = 1, . . . , n.
27
Since is an automorphism of L over K, Condition (5) is equivalent to
(6) p
i
(p
(1)
(x)) = p
(i)
(x) i = 1, . . . , n.
Conversely, suppose that an element S
n
satises (6). As f is irreducible, there
exists a unique Gal(f, K) such that x = x
(1)
. By (5) and (6),
x
(i)
= x
i
= p
i
(x) = p
i
(x) = p
i
(x
(1)
) = p
i
(p
(1)
(x)) = p
(i)
(x) = x
(i)
,
for i = 1, . . . , n. Hence = Gal(f, K).
Conclude that Gal(f, K) = G if and only if (4) holds.
Corollary: For each monic Galois polynomial f K[X] of degree n and each G S
n
,
the statement Gal(f, K)

= G is elementary.
Corollary: Let F be an ultraproduct of nite elds. Then
(a) every nite extension of F is cyclic;
(b) each Z/nZ occurs as a Galois group over F;
(c) Gal(F)

=

Z.
The proof of (c) includes depends on the following results:
Lemma: For each pronite group G and each G there exists a unique homomor-
phism h:

Z G such that h(1) = .
Proof: First for G nite. Uniqueness allows G to be innite.
Lemma:
(a) If each nite quotient of G is cyclic, then G is generated by one element (G is
pro-cyclic).
(b) If in addition, each Z/nZ is a quotient of G, then G

=

Z.
Proof of (a): If G = lim

G
i
, then the inverse limit of S
i
= G
i
[ G
i
= ) is
nonempty.
Proof of (b): There exists an epimorphism :

Z G. For each n, G has an open
normal subgroup G
n
such that G/G
n

= Z/nZ. Hence
1
(G
n
) = n

Z. So, Ker()

n=1

1
(G
n
) =

n=1
n

Z = 0.
28
The proof of (b) gives:
Corollary: If :

Z

Z is an epimorphism, then it is an isomorphism. If

Z
-

G
-

Z are epimorphisms, then they are isomorphisms.

29
5 Linear disjointness of elds
Central to eld theory is the concept linear disjointness of elds, an analog of linear
independence of vectors.
Lemma 5.1: Let E and F be extensions of a eld K. The following conditions are
equivalent:
(a) Each m-tuple (x
1
, . . . , x
m
) of elements of E which is linearly independent over K
is also linearly independent over F.
(b) Each n-tuple (y
1
, . . . , y
n
) of elements of F which is linearly independent over K is
also linearly independent over E.
Proof: Obviously it suces to prove that (a) implies (b). Let y
1
, . . . , y
m
be elements
of F for which there exist a
1
, . . . , a
m
E with a
1
y
1
+ +a
m
y
m
= 0. Let x
j
[ j J
be a linear basis for E over K and write a
i
=

jJ
a
ij
x
j
with a
ij
elements of K, only
nitely many dierent from 0. By (a), x
j
[ j J is linearly independent over F.
Therefore

a
ij
y
i
= 0 for every j. If y
1
, . . . , y
m
are linearly independent over K, then
a
ij
= 0 for every i and j, and therefore a
i
= 0, i = 1, . . . , m. Thus y
1
, . . . , y
m
are
linearly independent over E. This proves (b).
Denition: With E and F eld extensions of a eld K, refer to E and F as linearly
disjoint over K if (a) (or (b)) of Lemma 9.1 holds.
Corollary 5.2: Let E and F be extensions of a eld K such that [E : K] < . Then
E and F are linearly disjoint over K if and only if [E : K] = [EF : F]. If in addition
[F : K] < , then this is equivalent to [EF : K] = [E : K][F : K].
Proof: E and F are linearly disjoint over K and w
1
, . . . , w
n
is a basis for E/K, then
w
1
, . . . , w
n
is also a basis for EF over F. Hence [EF : F] = n = [E : K]. Conversely,
suppose that [E : K] = [EF : F] and let x
1
, . . . , x
m
E be linearly independent over
K. Extend x
1
, . . . , x
m
to a basis x
1
, . . . , x
n
of E/K. Since x
1
, . . . , x
n
generates
EF over K and n = [EF : F], x
1
, . . . , x
n
is a basis of EF/F. In particular x
1
, . . . , x
m
are linearly independent over F.
If E is a Galois extension of K, then E and F are linearly disjoint over K if and
30
only if E F = K. For arbitrary extensions this is clearly necessary, but not sucient.
Let L be a degree n > 1 extension of K for which L

is conjugate to L over K, but

L

L = K. Then [LL

: K] n(n 1). Thus, according to Corollary 5.2, L and L

are not linearly disjoint over K.

Lemma 5.3 (The tower property): Let K E and K L F be four elds. Then
E is linearly disjoint from F over K if and only if E is linearly disjoint from L over K
and EL is linearly disjoint from F over L.
Proof: The only nontrivial part is to show that if E and F are linearly disjoint over
K, then EL and F are linearly disjoint over L.
Apply Lemma 9.1. Suppose that y
1
, . . . , y
m
are elements of F which are linearly
independent over L, but a
1
, . . . , a
m
are elements of EL such that

m
i=1
a
i
y
i
= 0. Clear
denominators to assume that a
i
L[E], so that a
i
=

a
ij
x
i
with a
ij
L, where
x
j
[ j J is a linear basis for E over K. Then

j
(

i
a
ij
y
i
)x
j
= 0 and a
i
= 0,
i = 1, . . . , m.
Tensor products give an alternative approach to linear disjointness. Let E
1
, . . . , E
n
be n extensions of a eld K which are contained in a common eld. Then E
1
, . . . , E
n
are
linearly disjoint over K if and only if the canonical homomorphism of E
1

K

K
E
n
into E
1
E
n
that maps x
1
x
n
onto x
1
x
n
is injective. Associativity of the
tensor product construction shows that E
1
, . . . , E
n
are linearly disjoint over K if and
only if E
i
is linearly disjoint from E
1
E
i
1 over K, i = 2, . . . , n. Finally, for
any sequence E
1
, E
2
, E
3
, of eld extensions of K contained in a common eld,
dene E
1
, E
2
, E
3
, . . . to be linearly disjoint over K if every nite subfamily of
E
1
, E
2
, E
3
, . . . is linearly disjoint over K. The next lemma is an easy observation.
Lemma 9.4: Let E
1
, . . . , E
n
(resp., F
1
, . . . , F
n
) be linearly disjoint eld extensions of
K (resp., L). Let
i
: E
i
F
i
, be isomorphisms which coincide on K and for which

i
(K) = L, i = 1, . . . , n. Then

1
. . .
n
: E
1

K
. . .
K
E
n
F
1

L
. . .
L
F
n
is a well dened isomorphism. Therefore, there exists an isomorphism : E
1
. . . E
n

F
1
. . . F
n
that extends each of the
i
s.
31
6. Separable and regular extensions
We generalize the notion of separable algebraic extension to arbitrary eld extensions.
Denote the characteristic of a eld K by p. If p > 0, then K
1/p
is the eld
generated over K by the pth roots of all elements of K. If p = 0, then K
1/p
= K. Use
K
1/p

for the maximal purely inseparable extension of K. Let E be a nitely generated

extension of K. A collection t
1
, . . . , t
r
F of elements algebraically independent over K
is a separating transcendence basis if F/K(t
1
, . . . , t
r
) is a nite separable extension.
Lemma 6.1: An extension F of a eld K is said to be separable if it satises one of
the following equivalent conditions:
(a) F is linearly disjoint from K
1/p

over K;
(b) F is linearly disjoint from K
1/p
over K; or
(c) every nitely generated extension E of K which is contained in F has a separating
transcendence basis.
Moreover, a separating transcendence basis can be selected from a given set of generators
for F/K.
Proof: The implications (a) = (b) and (c) = (a) are easy. For (b) = (c) see
[L4, p. 54]. [FrJ, Lemma 17.7] gives a constructive proof.
Now apply the rules of linear disjointness.
Corollary 6.2:
(a) If E/K and F/E are separable extensions, then F/K is also separable.
(b) If F/K is a separable extension, then E/K is separable for every eld K E F.
(c) Every extension of a perfect eld is separable.
Example: A separable tower does not imply separable steps. Consider the tower of
elds F
p
F
p
(t
p
) F
p
(t), where t is transcendental over F
p
. The extension F
p
(t)/F
p
is separable, but F
p
(t)/F
p
(t
p
) is not.
Lemma 6.3: Call a eld extension F/K regular if it satises one of the following
equivalent conditions:
(a) the extension F/K is separable and K is algebraically closed in F; or
32
(b) the eld F is linearly disjoint from

K over K.
Proof: The implication (b) = (a) is immediate.
To prove (a) = (b) it suces to assume that F/K is nitely generated. Then
F/K has a separating transcendence basis, t
1
, . . . , t
r
, which is also a separating tran-
scendence basis for the extension FK
s
/K
s
. It follows that FK
s
is linearly disjoint from

K over K
s
. Also, K
s
/K is a Galois extension and F K
s
= K. Hence F is linearly
disjoint from K
s
over K. From Lemma 5.3 conclude that F is linearly disjoint from

K
over K.
Corollary 6.4:
(a) If E/K and F/E are regular extensions, then F/K is regular.
(b) If F/K is a regular extension, then E/K is regular for every eld K E F.
(c) Every extension of an algebraically closed eld is regular.
An extension E of K is said to be algebraically independent (or free) from an
extension F, if every nite set of elements of E algebraically independent over K remains
algebraically independent over F. By considering monomials in elements x
1
, . . . , x
n
of
E, it is clear that if E and F are linearly disjoint over K, then they are also algebraically
independnet over K. The converse, however, is false: Any two algebraic extensions of
K are algebraically independnet over K. But here is a partial converse.
Lemma 6.5: Let E be a regular extension of a eld K and let F be an extension of
K. If E is algebraically independnet from F over K, then E is linearly disjoint from F
over K.
Proof (Artin): Let x
1
, . . . , x
n
be elements of E for which there exist a
1
, . . . , a
n
F,
not all zero, such that

a
i
x
i
= 0. Let be a K-place of F into

K and let
T be a transcendence basis for E over K. Then the elements of T are algebraically
independent over F. Hence extends to a K(T)-place of F(T). Since E is an algebraic
extension of K(T), extends to an E-place of EF.
With no loss we may divide a
1
, . . . , a
n
by, say a
1
, to assume that a
1
= 1 and that
all the a
i
are nite under . Thus

(a
i
)x
i
= 0 is a nontrivial linear combination of
33
the x
i
over

K. But E is linearly disjoint from

K over K. Hence x
1
, . . . , x
n
are also
linearly dependent over K.
Corollary 6.6:
(a) Let F be regular extension of a eld K, algebraically independent from an extension
E of K. Then FE is a regular extension of E.
(b) If two regular extensions F and E of K are algebraically independent from each
other, then FE/K is regular.
Proof: For (a) note that F is also algebraically independent from

E over K. From
Lemma 6.5, F is linearly disjoint from

E over K. Therefore FE is linearly disjoint from

E over E. That is, FE/E is regular.

For (b) use (a) and Corollary 6.4(a).
34
7. Varieties
Concepts: Basic eld K, universal domain , Ane n-space A
n
, point in A
n
,
the polynomial ring K[X], algebraic set.
Hilberts basis theorem (Lan, p. 145): The polynomial ring K[X] in n variables
over a eld K is Noetherian. That is, it satises the following equivalent conditions:
(a) Every ideal in K[X] is nitely generated; or
(b) Every ascending sequence of ideals of K[X] is eventually stationary.
Concepts: Zariski topology, irreducible varieties, prime ideal, generic
point, coordinate ring:
0
-
p
-
K[X]
-
K[x]
-
0
K-specialization, function eld, dimension, K-curve, K-hypersurface, K-hy-
perplane.
Concepts: absolutely irreducible variety, eld of denition.
Lemma: Let W be a K-variety with a generic point x. For each extension L of K let
p
L
be the prime ideal of all f L[X] such that f(x) = 0. Then the following statements
are equivalent:
(a) K(x)/K is a regular extension;
(b) p

K
=

K p
K
;
(c) W is an absolutely irreducilbe variety dened over K.
Proof of (a) = (b): Let f p

K
and write f(X) =

iI
w
i
f
i
(X), with w
i
[ i I is
a basis for

K/K and f
i
K[X]. Then the w
i
are linearly independent over K(x) and
0 =

w
i
f
i
(x). Hence, f
i
(x) = 0 and therefore f
i
p
K
for each i I. Conclude that
p

K
=

K p
K
.
Proof (b) = (a): Suppose again that w
i
[ i I is a basis for

K/K. We have
to prove that it is linearly independent over K(x). So, suppose that g
i
are elements
of K[X] such that

w
i
g
i
(x) = 0. Then

w
k
g
i
(X) p

K
and therefore there exist
f
i
p
K
such that

w
i
g
i
(X) =

w
i
f
i
(X). Conclude that g
i
= f
i
p
K
and hence
g
i
(x) = 0 for each i I.
35
Proof (b) = (c): Observe that p

K
=

K p
K
implies that I

K
(W) = p

K
. Since p

K
is
prime, W is irreducible over

K, and dened over K.
Proof (c) =(b): Let x

be a generic point of W over

K. Then the map x

x extends
to a

K-homomorphism

K[x

]

K[x]. In particular, trans.deg

K(x

) trans.deg

K(x).
On the other hand, x

is a point of W and therefore the map x x

extends to a
K-homomorphism K[x] K[x

]. In particular trans.degK(x) trans.degK(x

) and
therefore also trans.deg

K(x) trans.deg

K(x

). So, the equality holds. It follows that

I

K
(W) = p

K
. Since W is dened over K, this implies that p

K
=

K p.
Lemma (Weil [Lan, p. 74]): Every absolutely irreducible K-variety V has a smallest
eld of denition L, which is a nite purely inseparable extension of K.
36
8. Pseudo algebraically closed eld
We say that a eld K is pseudo algebraically closed (abbreviated PAC) if each
absolutely irreducible variety V dened over K has a K-rational point.
Equivalently, for each absolutely irreducible polynomial f K[T
1
, . . . , T
r
, X] with
f
X
,= 0 and each 0 ,= g K[T
1
, . . . , T
r
], there exists (a, b) K
r+1
such that f(a, b) = 0
and g(a) ,= 0.
Equivalently (using intersection with general hyperplanes), each absolutely irre-
ducible polynomial f K[T, X] has innitely many K-rational points.
Equivalently, if R is a nitely generated integral domain over K with quotient
eld which is regular over K, then there exists a homomorphism : R K such that
(a) = a for each a K.
If K is PAC, V is an absolutely irreducible variety dened over K and U is a
nonempty Zariski open subset of K, then V (K) ,= 0.
Suppose that K is PAC and
1
-saturated. Let R be a countably generated in-
tegral domain over K and whose quotient eld is regular over K. Then there exists a
homomorphism : R K such that (a) = a for each a K.
Examples: (a) Every algebraically closed eld is PAC. (b) Every separably closed eld
is PAC. (c) (Ershov) Every innite algebraic extension of a PAC eld is PAC. (d)
Ultraproducts of PAC elds are PAC. (e) (Ax-Roquette) Every algebraic extension of a
PAC eld is PAC. (f) Finite elds are not PAC elds. (g) If a eld K has a valuation
with a nite residue eld, then K is not PAC. In particular, global and local elds are
not PAC. (h) If K has an ordering, then K is not PAC.
See [FrJ, Chap. 10] for proofs.
The embedding lemma ([FrJ, Lemma 18.2]): Let L and M be perfect elds. Let E
be a countable extension of L which is perfect. Let F be a perfect PAC
1
-saturated
eld which contains M. Suppose that there exists an isomorphism
0
:

L

M such
37
that
0
(L) = M. Assume also that there is a commutative diagram
(1)
Gal(E)

Gal(F)
?
res
?
res
Gal(L)

0
Gal(M)
where
0
is the isomorphism induced by
0
and is a homomorphism.
Then, there exists an extension of
0
to an embedding :

E

F such that
(()x) = (x) for each Gal(F) and each x

E.
Lemma ([FrJ, Lemma 18.3]): Let E/L and F/M be extensions of perfect elds. Sup-
pose that both L and M are countable and contain a eld K. Assume that E and
F are PAC eld,
1
-saturated, and there exists a K-isomorphism
0
:

L

M such
that
0
(L) = M. Assume also that (1) holds with an isomorphism. Then E is
K-elementarily equivalent to F.
Lemma ([FrJ, Corollary 18.5]): Let E and F be two elds and let
0
: Gal(F) Gal(E)
be an isomorphism. Let T be an ultralter on I. Let E

= E
I
/T and F

= F

/T.
Then there exists a commutative diagram
(1)
Gal(E

Gal(F

)
?
res
?
res
Gal(E)

0
Gal(F)
where is an isomorphism.
The elementary equivalence theorem for PAC fields: Let E/L and F/M be
perfect elds with both L and M containing a eld K. Assume that E and F are PAC,
that there exists a K-isomorphism
0
:

L

M such that
0
(L) = M, and there exists
a commutative diagram
(3)
Gal(E)

Gal(F)
?
res
?
res
Gal(L)

0
Gal(M)
38
with
0
induced by
0
and an isomorphism. Then E is K-elementarily equivalent to
F.
We call a eld F pseudo nite if it is perfect, PAC, and Gal(F)

=

Z.
The elementary equivalence theorem for pseudo finite fields: Let E and
F be two pseudo nite elds that contain a subeld K. If

K E

=
K

K F, then E is
K-elementarily equivalent to F.
39
9. Haar measure
Haar measure of a pronite group G:
B = -algebra of subsets of G that contains all closed subset of G
: B [0, 1]
0 (B) 1,
() = 0, (G) = 1
B
1
, B
2
, B
3
, . . . pairwise disjoint: (

i=1
B
i
) =

i=1
(B
i
) (-additivity)
(gB) = (Bg) = (B) (invariance under translations)
For all B B and for all > 0 there exist an open subset U and a closed subset C such
that C B U and (U

C) < (regularity)
B B and (B) = 0 and B
0
B imply B
0
B (completness).
Consequences:
A
1
A
2
A
3
imply (

i=1
A
i
) = lim
i
(A
i
)
A
1
A
2
A
3
A
3
imply (

i=1
A
i
) = lim
i
(A
i
)
If H < G is an open subgroup of index n, then (H) =
1
n
.
If : G

G is an epimorphism a nite group and

S

G, then (
1
(S)) =
|

S|
|

G|
.
Denition: A sequence A
1
, A
2
, A
3
, . . . are independent if (

iI
A
i
) =

iI
(A
i
)
for every nite set I. In this case the sequence of complements
G

A
1
, G

A
2
, G

A
3
, . . .
is also independent.
Lemma: Let A
1
, A
2
, A
3
, . . . be an independent sequence of subsets of G. Suppose that

i=1
(A
i
) = . Then
_

i=1
A
i
_
= 1.
Proof: (G

i=1
A
i
) = (

i=1
G

A
i
) =

i=1
(G

A
i
) =

i=1
(1 (A
i
))
-
0.
Lemma: Let H
1
, . . . , H
n
be open subgroup of G. They are independent if and only if
(G : H) =

n
i=1
(G : H
i
).
Proof: Consider K = H
1
H
m
with m n. The map G/K

m
i=1
G/H
i
,
gK (gH
1
, . . . , gH
m
) of cosets is injective. Hence (G : K)

m
i=1
(G : H
i
). Similarly,
40
(K : H)

n
i=m+1
(K : H
i
K)

n
i=m+1
(G : H
i
). Hence (G : H) = (G : K)(K : H)
and (G : H) =

n
i=1
(G : H
i
) gives (G : K) =

m
i=1
(G : H
i
).
Lemma: For each i let
i
: G G
i
be an epimorphism of nite group,

A
i
G
i
, A
i
=

1
i
(A
i
), and N
i
= Ker(
i
). Suppose that (G : N
1
N
n
) =

n
i=1
(G : N
i
) for each
n. Then A
1
, A
2
, A
3
, . . . are independent. If

i=1
[

A
i
[/[G
i
[ = , then (

i=1
A
i
) = 1.
Proof: Let N = N
1
N
n
and : G

n
i=1
G/N
i
, (g) = (gN
1
, . . . , gN
n
).
Then decomposes through G/N and is therefore surjective. Hence, A
1
A
n
=

1
(

A
1


A
n
) and therefore
(A
1
A
n
) =
[

A
1


A
n
[
[G/N[
=
n

i=1

A
i
[G
i
[
=
n

i=1
(A
i
).
Corollary: Let L
1
, L
2
, L
3
, . . . be a sequence of nite separable extensions of a eld
K. Let G = Gal(K).
(a) Gal(L
1
), Gal(L
2
), Gal(L
3
), . . . is independent if and only if L
1
, L
2
, L
3
, . . . is linearly
disjoint over K.
So, suppose that L
1
, L
2
, L
3
, . . . is linearly disjoint.
(b) If

i=1
1
[L
i
:K]
= , then (

i=1
Gal(L
i
)) = 1.
(c) Suppose that L
i
/K is Galois,

A
i
Gal(L
i
/K), A
i
= G[ [
L
i


A
i
, and

i=1
[

A
i
[/[L
i
: K] = , then (

i=1
A
i
) = 1.
41
10. Hilbertian elds
Let K be a eld. Consider irreducible polynomials
f
1
, . . . , f
m
K(T
1
, . . . , T
r
)[X
1
, . . . , X
n
]
and 0 ,= g K[T
1
, . . . , T
r
]. They dene a Hilbert subset of K
r
:
H
K
(f
1
, . . . , f
m
; g) = a K
r
[
m

i=1
f
i
(a, X) is dened and irreducible in K[X],
and g(a) ,= 0
K is Hilbertian if all its Hilbert sets are nonempty.
If K is Hilbertian, then every nite extension L of K is Hilbertian. Moreover, if
L/K is separable, then every Hilbert subset of L
r
contains a subset of K
r
.
Lemma: Let f K(T
1
, . . . , T
r
)[X] be a separable polynomial. Then
a K
r
[ Gal(f(a, X), K)

= Gal(f(T, X), K(T))
contains a Hilbert subset of K
r
.
Examples of Hilbertian elds: Q, K
0
(t) (K
0
any eld). Hence, every number eld.
Every nitely generated transcendental extension of K
0
.
K
0
((t
1
, t
2
, . . . , t
r
)), K
0
any eld, r 2.
Examples of innite extensions of a Hilbertian eld K which are Hilbertian:
If Gal(N/K) is nitely generated, then N is Hilbertian.
If Gal(N/K) is abelian, then N is Hilbertian.
If N/K is Galois and M is a proper nite separable extension, then M is Hilbertian
(Weissauer)
If N = N
1
N
2
, N
i
/K is Galois, i = 1, 2 and N
1
, N
2
, N
2
, N
1
, then N is Hilbertian.
Examples of non Hilbertian elds:
K nite
K separably closed
Gal(K) nitely generated
42
K Henselian, e.g., K = Q
p
or K = K
0
((t))
K is the maximal pro-p extension of a eld K
0
K is the maximal prosolvable extension of a eld K
0
Lemma: Let K be a Hilbertian eld, f K(T
1
, . . . , T
r
)[X] an absolutely irreducible
Galois polynomial, G = Gal(f(T, X), K(T)). Then K has a linearly disjoint sequence
L
1
, L
2
, L
3
, . . . of Galois extensions such that Gal(L
i
/K)

= G, i = 1, 2, . . .
Proposition: Let K be a Hilbertian eld. Then )

=

Z for almost all Gal(K).
Theorem: Let K be a countable Hilbertian eld. Then K
s
() and

K() are PAC for
almost all Gal(K).
43
11. Elements of algebraic number theory
An integral domain R is a Dedekind domain if
(1a) R is Noetherian (i.e., each ideal is nitely generated);
(1b) integrally closed (i.e., if an element x of the quotient eld K of R satises an
equation x
n
+ a
1
x
n1
+ + a
n
= 0 with a
i
R, then x R. We say that x is
integral over R.); and
(1c) every nonzero ideal p of R is maximal (i.e.,

K
p
= R/p is a eld).
We make the set of all nonzero ideals of R into a semigroup: ab =

a
i
b
i
[ a
i

a, b
i
b. The unit element of this semigroup is R. Every ideal a of R has a
unique presentation, a = p
m
1
1
p
m
2
2
p
m
r
r
, where p
1
, p
2
, . . . , p
r
are maximal ideals and
m
1
, m
2
, . . . , m
r
are positive integers. We say that p
i
divides a.
Examples: Every principal ideal domain is a Dedekind domain. In particular, Z,
F[t] (F a eld, t transcendental element over F) and every discrete valuation ring
are Dedekind domains.
Let R be a Dedekind domain with quotient eld K. Let L be a nite extension
of K. Then the set S of all elements x L which are integral over R is a Dedekind
domain (S is the integral closure of R in L.) In particular, if p is a prime ideal of R,
then Sp = P
e
1
P
e
2
2
P
e
g
g
with distinct prime ideals P
1
, P
2
, . . . , P
g
of S and positive
integers e
1
, e
2
, . . . , e
g
. If e
1
= e
2
= = e
g
= 1, then p is unramied in L. If, in
addition L/K is separable, then only nitely many ideals of R are ramied in L. Indeed,
R has an ideal d, called the discriminant of S/R such that p is ramied in L if and
only if p divides d.
Assume in addition, that L is a nite Galois extension of K. Let p be a prime
ideal of R which does not ramify in L and let P be a prime divisor of p in L that lies
above p (i.e, P R = p). The decomposition group of P over K is the following
subgroup of Gal(L/K):
D
P
= Gal(L/K) [ P

= P.
The xed eld L(P) of D(P) in L is the decomposition eld of P over K. In this
set up,

K
p
= R/p embeds into

L
P
= S/P by x + p x + P.
44
Assume that P is unramied over K, Then

L
P
/

K
p
is a Galois extension and
D(P)

= Gal(

L
P
/

K
p
). Under this isomorphism D(P) is mapped onto the element
Gal(

L
P
/

K
p
) such that x = x for each x S. Here x = x + P.
If P

is another prime ideal of S that lies above p, then there exists Gal(L/K)
such that P

= P

. We then have D(P

) = D(P)

.
A global eld K is either a nite extension of Q or a nite extension of F
q
(t) for
some prime power q and a transcendental element t. In the rst case K is a number
eld. Its ring of integers O
K
is the integral closure of Z in K. In the latter case, K
is a function eld. Its ring of integers O
K
is the integral closure of F
p
[t] in K. In
both cases O
K
is a Dedekind domain.
Also, if p is a maximal ideal of O
K
then

K
p
= O
K
/p is a nite eld. Indeed, in
the number eld case, it is a nite extension of F
p
, where pZ = p Z. In the number
eld case

K
p
is a nite extension of F
q
. In both cases we denote the number of elements
of

K
p
by Np. Denote the set of all nonzero prime ideals of K by P(K). For each m
there are only nitely many p P(K) such that Np m.
Suppose that L is a nite Galois extension of K. Suppose that p does not ramify
in L. Let P be a prime ideal of O
L
above p. Then D
P
is isomorphic to the group
Gal(

L
P
/

K
p
). The latter group is cyclic with a canonical generator Frob dened by
Frobx = x
Np
. The unique element of D(P) which is mapped onto Frob is the Frobenius
element of P/p. It is denoted by
_
L/K
P

and is the unique element of Gal(L/K) such

that
_
L/K
P
_
x x
Np
mod P
for each x O
L
. If Gal(L/K), then
_
L/K
P

=
_
L/K
P

. Thus, when ranges over all

elements of Gal(L/K),
_
L/K
P

ranges over a conjugacy class of Gal(L/K) known as the

Artin symbol of p in L:
_
L/K
p
_
The Chebotarev density theorem: Let L/K be a Galois extension of global eld
and let C be a conjugacy class of Gal(L/K). Then, the Dirichlet density of the set of
all prime ideals p of O
K
such that
_
L/K
p
_
= C is equal to [C[/[L : K].
45
Here the Dirichlet density of a set A of primes of O
K
is dened as the limit (if
it exists)
(A) = lim
s1
1

pA
1/Np
s

all p
1/Np
s
.
Lemma: Let L/K be a nite Galois extension of global elds. Suppose that f K[X]
is a polynomial which decomposes over L into linear factors. Then, for almost all
p P(K), f(X) has a root modulo p if and only if there exists (for all) P P(L) over
p such that f(X) has a root in L(P).
46
12. The transfer theorem
Regular ultraproducts of

K(). An ultralter T of Gal(K) is regular if each
subset of Gal(K) of measure 1 belongs to T.
Lemma: Let T
0
be a collection of subsets of G = Gal(K). Suppose that A
1
, . . . , A
n

T
0
implies A
1
A
n
is not a zero set. Then G has a regular ultralter T which
contains T
0
.
Proof: Let T
1
= T
0
1-sets. If A
i
T
0
, B
j
are 1-sets and A
1
A
m
B
1

B
n
= , then A
1
A
n
G

(B
1
B
n
), and hence A
1
, . . . , A
n
is a zero
set, a contradition. Take T T
1
.
Lemma: For each Gal(K) there exists a regular ultralter T of Gal(K) such that

K()/T

=

K().
Proof: For each nite Galois extension L of K let A
L
= Gal(K) [ [
L
= [
L
.
Take a regular ultralter T which contains all A
L
.
Proposition: Let K be a countable Hilbertian eld and let be a sentence in the
language L(ring, K). Equivalent:
(a)

K() [= for almost all Gal(K).
(b) F [= for all perfect pseudo nite elds that contain K.
Corollary: Let K be a global eld. If a sentence of L(ring, O
K
) is true in

K() for
almost all Gal(K), then is true in

K
p
for almost all p P(K).
Boolean algebras. Let K be a global eld and let G = Gal(K). For each sentence
of L(ring, K) let
S() = Gal(K) [

K() [= .
Then S(
1

2
) = S(
1
) S(
2
), S(
1

2
) = S(
1
) S(
2
), S() = G

S().
Denote the Boolean algebra generated by all basic test sets and all zero sets by o.
Here S(: f(X) = 0) is a basic test set if f K[X] is a monic separable polynomial.
A Boolean combination of basic test sets is a test set.
47
Lemma: Let A G, A / o. Then there exist regular ultralters T, T

of G such that
T o = T

o, A T, but A / T

.
Proof: Let o
0
be the collection of all B o such that either A

B or G

B.
Then
(1) B
1
, . . . , B
m
o
0
implies B
1
B
m
is not a zero set.
Indeed, suppose B
1
, . . . , B
m
o
1
and B
1
B
m
. Assume without loss
that A

B
i
, i = 1, . . . , l and G

B
i
, i = l +1, . . . , m. Then G = A

(G

A)

(B
1
B
l
) (B
l+1
B
m
). Hence, A B
1
B
l
, so A o
1
. This contradicts
our assumption on A.
Use Zorns lemma to choose a maximal subcollection o
1
of o that contains o
0
and
satises (1). In particular
(2) C
1
, . . . , C
m
o
1
implies C
1
C
m
o
1
.
Now let ( = o
1
A. Then ( satises (1). Indeed, let C
1
, . . . , C
m
o
1
and
C = C
1
C
m
. By (2), C o
1
. Assume that A C . Then A

C, so
G

C o
0
o
1
. This contradicts (1).
It follows that ( is contained in a regular ultralter T. Since o o T and o T
satises (1), we have o
1
= o T.
Similarly, there exists a regular ultralter T

which contains o
1
G

A. It also
satises o
1
= o T

.
For each test set let
A() = p P(K) [

K
p
[= .
Corollary: For each sentence there exists a test sentence such that S() S()
and A() A().
Theorem: For each sentence , (S()) = (S()). Moreover, (A()) is a rational
number and (A()) > 0 if and only if A() is innite.
Proof: By the corollary, it suces to take a test sentence which depends on
(X)[f
1
(X) = 0], , (X)[f
m
(X) = 0].
48
Let L be the splitting eld of f
1
, . . . , f
m
. Let C = Gal(L/K) [ L() [= . Then

K() [= if and only if [

L
C. Hence (S()) = [C[/[L : K]. C is a union of
conjugacy classes of Gal(L/K). By the Chebotarev density theorem, the set
A

() = p P(K) [
_
L/K
p
_
C
has density [C[/[L : K]. So, it suces to prove that A

() A().
Induction on the structure of reduces the proof to the case where has the form
(X)[f(X) = 0], where f K[X] monic irreducible polynomial.
Let x
1
, . . . , x
n
be the roots of f. Then
A

((X)[f(X) = 0]) = p P(K) [

_
L/K
p
_

n
_
i=1
Gal(L/K(x
i
))
p P(K) [

K
p
[= (X)[f(X) = 0.
49
References
[Ax] J. Ax, Solving diophantine problems modulo every prime, Annals of Mathematics 85
(1967), 161183.
[BoS] Z.I. Borevich and I.R. Shafarevich, Number Theorey, Academic Press, New York, 1966.
[FrJ] M.D. Fried and M. Jarden, Field Arithmetic, Ergebnisse der Mathematik (3) 11, Sprin-
ger, Heidelberg, 1986.
[FrV] M. D. Fried and H. Volklein, The inverse Galois problem and rational points on moduli
spaces, Mathemtische Annalen 290 (1991), 771800.
[Ja1] M. Jarden, Elementary statements over large algebraic elds, Transactions of AMS
164 (1972), 6791.
[Lan] S. Lang, Introduction to algebraic geometry, Interscience Publishers, New York, 1958.
[Mum] D. Mumford, The Red Book of Varieties and Schemes, Lecture Notes in Mathematics
1358, Springer, Berlin, 1988.
4 November, 2007
50