Algebraic Number Theory 2010/2011

1. Algebraic prerequisites
1.1. General
1.1.1.
Denition. For a eld F dene the ring homomorphism Z F by n n 1
F
. Its
kernel I is an ideal of Z such that Z/I is isomorphic to the image of Z in F. The
latter is an integral domain, so I is a prime ideal of Z, i.e. I = 0 or I = pZ for a
prime number p. In the rst case F is said to have characteristic 0, in the second
characteristic p .
DenitionLemma. Let F be a subeld of a eld L. An element a L is called
algebraic over F if one of the following equivalent conditions is satised:
(i) f(a) = 0 for a non-zero polynomial f(X) F[X];
(ii) elements 1, a, a
2
, . . . are linearly dependent over F;
(iii) F-vector space F[a] =

a
i
a
i
: a
i
F is of nite dimension over F;
(iv) F[a] = F(a).
Proof. (i) implies (ii): if f(X) =

n
i=0
c
i
X
i
, c
0
, c
n
,= 0, then

c
i
a
i
= 0.
(ii) implies (iii): if

n
i=0
c
i
a
i
= 0, c
n
,= 0, then a
n
=

n1
i=0
c
1
n
c
i
a
i
, a
n+1
=
a a
n
=

n1
i=0
c
1
n
c
i
a
i+1
=

n2
i=0
c
1
n
c
i
a
i+1
+ c
1
n
c
n1

n1
i=0
c
1
n
c
i
a
i
, etc.
(iii) implies (iv): for every b F[a] we have F[b] F[a], hence F[b] is of nite
dimension over F. So if b , F, there are d
i
such that

d
i
b
i
= 0, and d
0
,= 0. Then
1/b = d
1
0

n
i=1
d
i
b
i1
and hence 1/b F[b] F[a].
(iv) implies (i): if 1/a is equal to

e
i
a
i
, then a is a root of

e
i
X
i+1
1.
For an element a algebraic over F denote by
f
a
(X) F[X]
the monic polynomial of minimal degree such that f
a
(a) = 0.
This polynomial is irreducible: if f
a
= gh, then g(a)h(a) = 0, so g(a) = 0 or
h(a) = 0, contradiction. It is called the monic irreducible polynomial of a over F .
For example, f
a
(X) is a linear polynomial iff a F.
2 Alg number theory
Lemma. Dene a ring homomorphism F[X] L, g(X) g(a). Its kernel is the
principal ideal generated by f
a
(X) and its image is F(a), so
F[X]/(f
a
(X)) F(a).
Proof. The kernel consists of those polynomials g over F which vanish at a. Using
the division algorithm write g = f
a
h + k where k = 0 or the degree of k is smaller
than that of f
a
. Now k(a) = g(a) f
a
(a)h(a) = 0, so the denition of f
a
implies
k = 0 which means that f
a
divides g.
Denition. A eld L is called algebraic over its subeld F if every element of L is
algebraic over F. The extension L/F is called algebraic.
Denition. Let F be a subeld of a eld L. The dimension of L as a vector space
over F is called the degree [L : F[ of the extension L/F.
If a is algebraic over F then [F(a) : F[ is nite and it equals the degree of the
monic irreducible polynomial f
a
of a over F.
Transitivity of the degree [L : F[ = [L : M[[M : F[ follows from the observation:
if
i
form a basis of M over F and
j
form a basis of L over M then
i

j
form a
basis of L over F.
Every extension L/F of nite degree is algebraic: if L, then [F() : F[
[L : F[ is nite, so by (iii) above is algebraic over F. In particular, if is algebraic
over F then F() is algebraic over F. If , are algebraic over F then the degree of
F(, ) over F does not exceed the product of nite degrees of F()/F and F()/F
and hence is nite. Thus all elements of F(, ) are algebraic over F.
An algebraic extension F(a
i
) of F is is the composite of extensions F(a
i
),
and since a
i
is algebraic [F(a
i
) : F[ is nite, thus every algebraic extension is the
composite of nite extensions.
1.1.2. Denition. An extension F of Q of nite degree is called an algebraic number
eld, the degree [F : Q[ is called the degree of F .
Examples. 1. Every quadratic extension L of Q can be written as Q(

e) for a
square-free integer e. Indeed, if 1, is a basis of L over Q, then
2
= a
1
+ a
2
with
rational a
i
, so is a root of the polynomial X
2
a
2
X a
1
whose roots are of the
form a
2
/2

d/2 where d Q is the discriminant. Write d = f/g with integer f, g

and notice that Q(

d) = Q(
_
dg
2
) = Q(

fg). Obviously we can get rid of all square

divisors of fg without changing the extension Q(

fg).
2. Cyclotomic extensions Q
m
= Q(
m
) of Q where
m
is a primitive mth
root of unity. If p is prime then the monic irreducible polynomial of
p
over Q is
X
p1
+ + 1 = (X
p
1)/(X 1) of degree p 1.
2010/2011 3
1.1.3. Denition. Let two elds L, L

contain a eld F. A homo(iso)morphism

: L L

such that [
F
is the identity map is called a F-homo(iso)morphism of L
into L

.
The set of all F-homomorphisms from L to L

is denoted by Hom
F
(L, L

). Notice
that every F-homomorphism is injective: its kernel is an ideal of F and 1
F
does not
belong to it, so the ideal is the zero ideal. In particular, (L) is isomorphic to L.
The set of all F-isomorphisms from L to L

is denoted by Iso
F
(L, L

).
Two elements a L, a

are called conjugate over F if there is a F-homomorphism

such that (a) = a

. If L, L

are algebraic over F and isomorphic over F, they are

called conjugate over F.
Lemma. (i) Any two roots of an irreducible polynomial over F are conjugate over F.
(ii) An element a

is conjugate to a over F iff f

a
= f
a
.
(iii) The polynomial f
a
(X) is divisible by

(X a
i
) in L[X], where a
i
are all
distinct conjugate to a elements over F, L is the eld F(a
i
) generated by a
i
over
F.
Proof. (i) Let f(X) be an irreducible polynomial over F and a, b be its roots in a
eld extension of F. Then f
a
= f
b
= f and we have an F-isomorphism
F(a) F[X]/(f
a
(X)) = F[X]/(f
b
(X)) F(b), a b
and therefore a is conjugate to b over F.
(ii) 0 = f
a
(a) = f
a
(a) = f
a
(a

), hence f
a
= f
a
. If f
a
= f
a
, use (i).
(iii) If a
i
is a root of f
a
then by the division algorithm f
a
(X) is divisible by Xa
i
in L[X].
1.1.4. Denition. A eld is called algebraically closed if it does not have algebraic
extensions.
Theorem (without proof). Every eld F has an algebraic extension C which is
algebraically closed. The eld C is called an algebraic closure of F. Every two
algebraic closures of F are isomorphic over F.
Example. The eld of rational numbers Q is contained in algebraically closed eld
C. The maximal algebraic extension Q
a
of Q is obtained as the subeld of complex
numbers which contains all algebraic elements over Q. The eld Q
a
is algebraically
closed: if C is algebraic over Q
a
then it is a root of a non-zero polynomial with
nitely many coefcients, each of which is algebraic over Q. Therefore is algebraic
over the eld M generated by the coefcients. Then M()/M and M/Q are of nite
degree, and hence is algebraic over Q, i.e. belongs to Q
a
. The degree [Q
a
: Q[ is
innite, since [Q
a
: Q[ [Q(
p
) : Q[ = p 1 for every prime p.
The eld Q
a
is is much smaller than C, since its cardinality is countable whereas
the cardinality of complex numbers is uncountable).
4 Alg number theory
Everywhere below we denote by C an algebraically closed eld containing F.
Elements of Hom
F
(F(a), C) are in one-to-one correspondence with distinct roots
of f
a
(X) F[X]: for each such root a
i
, as in the proof of (i) above we have
: F(a) C, a a
i
; and conversely each such Hom
F
(F(a), C) maps a to one
of the roots a
i
.
1.2. Galois extensions
1.2.1. Denition. A polynomial f(X) F[X] is called separable if all its roots in C
are distinct.
Recall that if a is a multiple root of f(X), then f

(a) = 0. So a polynomial f is
separable iff the polynomials f and f

dont have common roots.

Examples of separable polynomials. Irreducible polynomials over elds of charac-
teristic zero, irreducible polynomials over nite elds.
Proof: if f is an irreducible polynomial over a eld of characteristic zero, then
its derivative f

is non-zero and has degree strictly smaller than f; and so if f has a

multiple root, than a g.c.d. of f and f

would be of positive degree strictly smaller than

f which contradicts the irreducibility of f. For the case of irreducible polynomials
over nite elds see section 1.3.
Denition. Let L be a eld extension of F. An element a L is called separable
over F if f
a
(X) is separable. The extension L/F is called separable if every element
of L is separable over F.
Example. Every algebraic extension of a eld of characteristic zero or a nite eld is
separable.
1.2.2. Lemma. Let M be a eld extension of F and L be a nite extension of M.
Then every F-homomorphism : M C can be extended to an F-homomorphism

: L C.
Proof. Let a L M and f
a
(X) =

c
i
X
i
be the minimal polynomial of a over
M. Then (f
a
)(X) =

(c
i
)X
i
is irreducible over M. Let b be its root. Then
f
a
= f
b
. Consider an F-homomorphism : M[X] C, (

a
i
X
i
) =

(a
i
)b
i
.
Its image is (M)(b) and its kernel is generated by f
a
. Since M[X]/(f
a
(X)) M(a),
determines an extension

: M(a) C of . Since [L : M(a)[ < [L : M[, by

induction

can be extended to an F-homomorphism

: L C such that

[
M
= .
1.2.3. Theorem. Let L be a nite separable extension of F of degree n. Then there
exist exactly n distinct F-homomorphisms of L into C, i.e. [ Hom
F
(L, C)[ = [L : F[.
2010/2011 5
Proof. The number of distinct F-homomorphisms of L into C is n is valid for
any extension of degree n. To prove this, argue by induction on [L : F[ and use the
fact that every F-homomorphism : F(a) C sends a to one of roots of f
a
(X) and
that root determines completely.
To show that there are n distinct F-homomorphisms for separable L/F consider
rst the case of L = F(a). Fromseparability we deduce that the polynomial f
a
(X) has
n distinct roots a
i
which give n distinct F-homomorphisms of L into C: a a
i
.
Now argue by induction on degree. For a L F consider M = F(a). There
are m = [M : F[ distinct F-homomorphisms
i
of M into C. Let

i
: L C
be an extension of
i
which exists according to 1.2.2. By induction there are n/m
distinct F(
i
(a))-homomorphisms
ij
of

i
(L) into C. Now
ij

i
are distinct
F-homomorphisms of L into C.
1.2.4. Proposition. Every nite subgroup of the multiplicative group F

of a eld F
is cyclic.
Proof. Denote this subgroup by G, it is an abelian group of nite order. From the
standard theorem on the stucture of nitely generated abelian groups we deduce that
G Z/m
1
Z Z/m
r
Z
where m
1
divides m
2
, etc. We need to show that r = 1 (then G is cyclic). If r > 1,
then let a prime p be a divisor of m
1
. The cyclic group Z/m
1
Z has p elements
of order p and similarly, Z/m
2
Z has p elements of order p, so G has at least p
2
elements of order p. However, all elements of order p in G are roots of the polynomial
X
p
1 which over the eld F cannot have more than p roots, a contradiction. Thus,
r = 1.
1.2.5. Theorem. Let F be a eld of characteristic zero or a nite eld. Let L be a nite
eld extension of F. Then there exists an element a L such that L = F(a) = F[a].
Proof. If F is of characteristic 0, then F is innite. By 1.2.3 there are n = [L : F[
distinct F-homomorphisms
i
: L C. Put V
ij
= a L :
i
(a) =
j
(a). Then
V
ij
are proper F-vector subspaces of L for i ,= j of dimension < n, and since F is
innite, there union
i=j
V
ij
is different from L. Then there is a L (V
ij
). Since
the set
i
(a) is of cardinality n, the minimal polynomial of a over F has at least
n distinct roots. Then [F(a) : F[ n = [L : F[ and hence L = F(a).
If L is nite, then L

is cyclic by 1.2.4. Let a be any of its generators. Then

L = F(a).
1.2.6. Denition. An algebraic extension L of F (inside C ) is called the splitting
eld of polynomials f
i
if L = F(a
ij
) where a
ij
are all the roots of f
i
.
An algebraic extension L of F is called a Galois extension if L is the splitting
eld of some separable polynomials f
i
over F.
6 Alg number theory
Example. Let L be a nite extension of F such that L = F(a). Then L/F is a
Galois extension if the polynomial f
a
(X) of a over F has deg f
a
distinct roots in L.
So quadratic extensions of Q and cyclotomic extensions of Q are Galois extensions.
1.2.7. Lemma. Let L be the splitting eld of an irreducible polynomial f(X) F[X].
Then (L) = L for every Hom
F
(L, C).
Proof. permutes the roots of f(X). Thus, (L) = F((a
1
), . . . , (a
n
)) = L.
1.2.8. Theorem. A nite extension L of F is a Galois extension iff
(L) = L for every Hom
F
(L, C) and [ Hom
F
(L, L)[ = [L : F[.
The set Hom
F
(L, L) equals to the set Iso
F
(L, L) which is a nite group with
respect to the composite of eld isomorphisms. This group is called the Galois group
Gal(L/F) of the extension L/F.
Sketch of the proof. Let L be a Galois extension of F. The right arrow follows from
the previous proposition and properties of separable extensions. On the other hand, if
L = F(b
i
) and (L) = L for every Hom
F
(L, C) then (b
i
) belong to L and
L is the splitting eld of polynomials f
b
i
(X). If [ Hom
F
(L, L)[ = [L : F[ then one
can show by induction that each of f
b
i
(X) is separable.
Now suppose we are in the situation of 1.2.5. Then L = F(a) for some a L.
L is the splitting eld of some polynomials f
i
over F, and hence L is the splitting
eld of their product. By 1.2.7 and induction we have L = L. Then L = F(a
i
) for
any root a
i
of f
a
, and elements of Hom
F
(L, L) correspond to a a
i
. Therefore
Hom
F
(L, L) = Iso
F
(L, L). Its elements correspond to some permutations of the set
a
i
of all roots of f
a
(X).
1.2.9. Theorem (without proof). Let L/F be a nite Galois extension and M be an
intermediate eld between F and L.
Then L/M is a Galois extension with the Galois group
Gal(L/M) = Gal(L/F) : [
M
= id
M
.
For a subgroup H of Gal(L/F) denote
L
H
= x L : (x) = x for all H.
This set is an intermediate eld between L and F.
1.2.10. Main theorem of Galois theory (without proof). Let L/F be a nite Galois
extension with Galois group G = Gal(L/F).
Then H L
H
is a one-to-one correspondence between subgroups H of G and
subelds of L which contain F; the inverse map is given by M Gal(L/M). We
have Gal(L/M) = H.
2010/2011 7
Normal subgroups H of G correspond to Galois extensions M/F and
Gal(M/F) G/H.
1.3. Finite elds
Every nite eld F has positive characteristic, since the homomorphism Z F is
not injective. Let F be of prime characteristic p. Then the image of Z in F can
be identied with the nite eld F
p
consisting of p elements. If the degree of F/F
p
is n, then the number of elements in F is p
n
. By 1.2.4 the group F

is cyclic of
order p
n
1, so every non-zero element of F is a root of the polynomial X
p
n
1
1.
Therefore, all p
n
elements of F are all p
n
roots of the polynomial f
n
(X) = X
p
n
X.
The polynomial f
n
is separable, since its derivative in characteristic p is equal to
p
n
X
p
n
1
1 = 1. Thus, F is the splitting eld of f
n
over F
p
. We conclude that
F/F
p
is a Galois extension of degree n = [F : F
p
[.
Lemma. The Galois group of F/F
p
is cyclic of order n: it is generated by an auto-
morphism of F called the Frobenius automorphism:
(x) = x
p
for all x F .
Proof.
m
(x) = x
p
m
= x for all x F iff n[m.
On the other hand, for every n 1 the splitting eld of f
n
over F
p
is a nite eld
consisiting of p
n
elements.
Thus,
Theorem. For every n there is a unique (up to isomorphism) nite eld F
p
n consisting
of p
n
elements; it is the splitting eld of the polynomial f
n
(X) = X
p
n
X. The nite
extension F
p
nm/F
p
n is a Galois extension with cyclic group of degree m generated
by the Frobenius automorphism
n
: x x
p
n
.
Lemma. Let g(X) be an irreducible polynomial of degree m over a nite eld F
p
n.
Then g(X) divides f
nm
(X) and therefore is a separable polynomial.
Proof. Let a be a root of g(X). Then F
p
n(a)/F
p
n is of degree m, so F
p
n(a) = F
p
nm.
Since a is a root of f
nm
(X), g divides f
nm
. The latter is separable and so is g.
8 Alg number theory
2. Integrality
2.1. Integrality over rings
2.1.1. Proposition Denition. Let B be a ring and A its subring.
An element b B is called integral over A if it satises one of the following
equivalent conditions:
(i) there exist a
i
A such that f(b) = 0 where f(X) = X
n
+a
n1
X
n1
+ +a
0
;
(ii) the subring of B generated by A and b is an A-module of nite type;
(iii) there exists a subring C of B which contains A and b and which is an
A-module of nite type.
Proof. (i) (ii): note that the subring A[b] of B generated by A and b coincides
with the A-module M generated by 1, . . . , b
n1
. Indeed,
b
n+j
= a
0
b
j
b
n+j1
and by induction b
j
M.
(ii) (iii): obvious.
(iii) (i): let C = c
1
A+ +c
m
A. Then bc
i
=

j
a
ij
c
j
, so

j
(
ij
ba
ij
)c
j
= 0.
Denote by d the determinant of M = (
ij
b a
ij
). Note that d = f(b) where
f(X) A[X] is a monic polynomial. From linear algebra we know that dE = M

M
where M

is the adjugate matrix to M and E is the identity matrix of the same order
of that of M. Denote by C the column consisting of c
j
. Now we get MC = 0 implies
M

MC = 0 implies dEC = 0 implies dC = 0. Thus dc

j
= 0 for all 1 j m.
Every c C is a linear combination of c
j
. Hence dc = 0 for all c C. In particular,
d1 = 0, so f(b) = d = 0.
Examples. 1. Every element of A is integral over A.
2. If A, B are elds, then an element b B is integral over A iff b is algebraic
over A.
3. Let A = Z, B = Q. A rational number r/s with relatively prime r and
s is integral over Z iff (r/s)
n
+ a
n1
(r/s)
n1
+ + a
0
= 0 for some integer a
i
.
Multiplying by s
n
we deduce that s divides r
n
, hence s = 1 and r/s Z. Hence
integral in Q elements over Z are just all integers.
4. If B is a eld, then it contains the eld of fractions F of A. Let
Hom
F
(B, C) where C is an algebraically closed eld containing B. If b B is
integral over A, then (b) (B) is integral over A.
5. If b B is a root of a non-zero polynomial f(X) = a
n
X
n
+ A[X],
then a
n1
n
f(b) = 0 and g(a
n
b) = 0 for g(X) = X
n
+ a
n1
X
n1
+ + a
n1
n
a
0
,
2010/2011 9
g(a
n
X) = a
n1
n
f(X). Hence a
n
b is integral over A. Thus, for every algebraic over
A element b of B there is a non-zero a A such that ab is integral over A.
2.1.2. Corollary. Let A be a subring of an integral domain B. Let I be a non-zero
A-module of nite type, I B. Let b B satisfy the property bI I. Then b is
integral over A.
Proof. Indeed, as in the proof of (iii) (i) we deduce that dc = 0 for all c I.
Since B is an integral domain, we deduce that d = 0, so d = f(b) = 0.
2.1.3. Proposition. Let A be a subring of a ring B, and let b
i
B be such that b
i
is
integral over A[b
1
, . . . , b
i1
] for all i. Then A[b
1
, . . . , b
n
] is an A-module of nite
type.
Proof. Induction on n. n = 1 is the previous proposition. If C = A[b
1
, . . . , b
n1
]
is an A-module of nite type, then C =

m
i=1
c
i
A. Now by the previous proposition
C[b
n
] is a C-module of nite type, so C[b
n
] =

l
j=1
d
j
C. Thus, C[b
n
] =

i,j
d
j
c
i
A
is an A-module of nite type.
2.1.4. Corollary 1. If b
1
, b
2
B are integral over A, then b
1
+ b
2
, b
1
b
2
, b
1
b
2
are
integral over A.
Certainly b
1
/b
2
isnt necessarily integral over A.
Corollary 2. The set B

of elements of B which are integral over A is a subring of

B containing A.
Denition. B

is called the integral closure of A in B. If A is an integral domain

and B is its eld of fractions, B

is called the integral closure of A.

A ring A is called integrally closed if A is an integral domain and A coincides
with its integral closure in its eld of fractions.
Let F be an algebraic number eld. The integral closure of Z in F is called the
ring O
F
of (algebraic) integers of F .
Examples. 1. A UFD is integrally closed. Indeed, if x = a/b with relatively prime
a, b A is a root of polynomial f(X) = X
n
+ + a
0
A[X], then b divides a
n
,
so b is a unit of A and x A.
In particular, the integral closure of Z in Q is Z.
2. O
F
is integrally closed (see below in 2.1.6).
2.1.5. Lemma. Let A be integrally closed. Let B be a eld. Then an element b B is
integral over A iff the monic irreducible polynomial f
b
(X) F[X] over the fraction
eld F of A has coefcients in A.
10 Alg number theory
Proof. Let L be a nite extension of F which contains B and all (b) for all
F-homomorphisms from B to an algebraically closed eld C. Since b L is integral
over A, (b) L is integral over A for every . Then f
b
(X) =

(X (b)) has
coefcients in F which belong to the ring generated by A and all (b) and therefore
are integral over A. Since A is integrally closed, f
b
(X) A[X].
If f
b
(X) A[X] then b is integral over A by 2.1.1.
Examples. 1. Let F be an algebraic number eld. Then an element b F is integral
iff its monic irreducible polynomial has integer coefcients.
For example,

d for integer d is integral.

If d 1 mod 4 then the monic irreducible polynomial of (1 +

d)/2 over Q is
X
2
X + (1 d)/4 Z[X], so (1 +

d)/2 is integral. Note that

d belongs to
Z[(1 +

d)/2], and hence Z[

d] is a subring of Z[(1 +

d)/2].
Thus, the integral closure of Z in Q(

d) contains the subring Z[

d] and the
subring Z[(1 +

d)/2] if d 1 mod 4. We show that there are no other integral

elements.
An element a + b

d with rational a and b ,= 0 is integral iff its monic irreducible

polynomial X
2
2aX+(a
2
db
2
) belongs to Z[X]. Therefore 2a, 2b are integers. If
a = (2k+1)/2 for an integer k, then it is easy to see that a
2
db
2
Z iff b = (2l +1)/2
with integer l and (2k + 1)
2
d(2l + 1)
2
is divisible by 4. The latter implies that d
is a quadratic residue mod 4, i.e. d 1 mod 4. In turn, if d 1 mod 4 then every
element (2k + 1)/2 + (2l + 1)

d/2 is integral.
Thus, integral elements of Q(

d) are equal to
_
Z[

d] if d , 1 mod 4
Z[(1 +

d)/2] if d 1 mod 4
2. O
Q
m is equal to Z[
m
] (see section 2.4).
2.1.6. Denition. B is said to be integral over A if every element of B is integral
over A. If B is of characteristic zero, its elements integral over Z are called integral
elements of B.
Lemma. If B is integral over A and C is integral over B, then C is integral over
A.
Proof. Let c C be a root of the polynomial f(X) = X
n
+b
n1
X
n1
+ +b
0
with
b
i
B. Then c is integral over A[b
0
, . . . , b
n1
]. Since b
i
B are integral over A,
proposition 2.1.3 implies that A[b
0
, . . . , b
n1
, c] is an A-module of nite type. From
2.1.1 we conclude that c is integral over A.
Corollary. O
F
is integrally closed
2010/2011 11
Proof. An element of F integral over O
F
is integral over Z due to the previous
lemma.
2.1.7. Proposition. Let B be an integral domain and A be its subring such that B is
integral over A. Then B is a eld iff A is a eld.
Proof. If A is a eld, then A[b] for b B0 is a vector space of nite dimension over
A, and the A-linear map : A[b] A[b], (c) = bc is injective, therefore surjective,
so b is invertible in B.
If B is a eld and a A0, then the inverse a
1
B satises a
n
+a
n1
a
n+1
+
+ a
0
= 0 with some a
i
A. Then a
1
= a
n1
a
0
a
n1
, so a
1
A.
2.2. Norms and traces
2.2.1. Denition. Let A be a subring of a ring B such that B is a free A-module
of nite rank n. For b B its trace Tr
B/A
(b), norm N
B/A
(b) and characteristic
polynomial g
b
(X) are the trace, the norm and the characteristic polynomial of the
linear operator m
b
: B B, m
b
(c) = bc. In other words, if M
b
is a matrix of the
operator m
b
with respect to a basis of B over A, then g
b
(X) = det(XE M
b
),
Tr
B/A
(b) = Tr M
b
, N
B/A
= det M
b
.
If g
b
(X) = X
n
+a
n1
X
n1
+ +a
0
then fromthe denition a
n1
= Tr
B/A
(b),
a
0
= (1)
n
N
B/A
(b).
2.2.2. First properties.
Tr(b + b

) = Tr(b) + Tr(b

), Tr(ab) = a Tr(b), Tr(a) = na,

N(bb

) = N(b)N(b

), N(ab) = a
n
N(b), N(a) = a
n
for a A.
2.2.3. Everywhere below in this section F is either a nite eld of a eld of charac-
teristic zero. Then every nite extension of F is separable.
Proposition. Let L be an algebraic extension of F of degree n. Let b L and
b
1
, . . . , b
n
be roots of the monic irreducible polynomial of b over F each one repeated
[L : F(b)[ times. Then the characteristic polynomial g
b
(X) of b with respect to L/F
is

(X b
i
), and Tr
L/F
(b) =

b
i
, N
L/F
(b) =

b
i
.
Proof. If L = F(b), then use the basis 1, b, . . . , b
n1
to calculate g
b
. Let f
b
(X) =
X
n
+ c
n1
X
n1
+ + c
0
be the monic irreducible polynomial of b over F, then the
12 Alg number theory
matrix of m
b
is
M
b
=
_
_
_
_
0 1 0 . . . 0
0 0 1 . . . 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
c
0
c
1
c
2
. . . c
n1
_
_
_
_
.
Hence g
b
(X) = det(XE M
b
) = f
b
(X) and det M
b
=

b
i
, Tr M
b
=

b
i
.
In the general case when [F(b) : F[ = m < n choose a basis
1
, . . . ,
n/m
of
L over F(b) and take
1
, . . . ,
1
b
m1
,
2
, . . . ,
2
b
m1
, . . . as a basis of L over
F. The matrix M
b
is a block matrix with the same block repeated n/m times on
the diagonal and everything else being zero. Therefore, g
b
(X) = f
b
(X)
|L:F(b)|
where
f
b
(X) is the monic irreducible polynomial of b over F.
Example. Let F = Q, L = Q(

d) with square-free integer d. Then

g
a+b

d
(X) = (X a b

d)(X a + b

d) = X
2
2aX + (a
2
db
2
),
so
Tr
Q(

d)/Q
(a + b

d) = 2a, N
Q(

d)/Q
(a + b

d) = a
2
db
2
.
In particular, an integer number c is a sumof two squares iff c N
Q(

1)/Q
O
Q(

1)
.
More generally, c is in the form a
2
db
2
with integer a, b and square-free d not
congruent to 1 mod 4 iff
c N
Q(

d)/Q
Z[

d]
2.2.4. Corollary 1. Let
i
be distinct F-homomorphisms of L into C. Then Tr
L/F
(b) =

i
b, N
L/F
(b) =

i
(b).
Proof. In the previous proposition b
i
=
i
(b).
Corollary 2. Let A be an integral domain, F be its eld of fractions. Let L be an
extension of F of nite degree. Let A

be the integral closure of A in F. Then for

an integral element b L over A g
b
(X) A

[X] and Tr
L/F
(b), N
L/F
(b) belong to
A

.
Proof. All b
i
are integral over A.
Corollary 3. If, in addition, A is integrally closed, then Tr
L/F
(b), N
L/F
(b) A.
Proof. Since A is integrally closed, A

F = A.
2.2.5. Lemma. Let F be a nite eld of a eld of characteristic zero. If L is a nite
extension of F and M/F is a subextension of L/F, then the following transitivity
2010/2011 13
property holds
Tr
L/F
= Tr
M/F
Tr
L/M
, N
L/F
= N
M/F
N
L/M
.
Proof. Let
1
, . . . ,
m
be all distinct F-homomorphisms of M into C ( m = [M :
F[ ). Let
1
, . . . ,
n/m
be all distinct M-homomorphisms of L into C ( n/m = [L :
M[ ). The eld
j
(L) is a nite extension of F, and by 1.2.5 there is an element a
j
C
such that
j
(L) = F(a
j
). Let E be the minimal subeld of C containing M and all
a
j
. Using 1.2.3 extend
i
to

i
: E C. Then the composition

i

j
: L C is
dened. Note that

i

j
=

i
1

j
1
implies
i
=

i

j
[
M
=

i
1

j
1
[
M
=
i
1
, so
i = i
1
, and then j = j
1
. Hence

i

j
for 1 i m, 1 j n/m are all n distinct
F-homomorphisms of L into C. By Corollary 3 in 2.2.4
N
M/F
(N
L/M
(b)) = N
M/F
(

j
(b)) =

i
(

j
(b)) =

i

j
)(b) = N
L/F
(b).
Similar arguments work for the trace.
2.3. Integral basis
2.3.1. Denition. Let A be a subring of a ring B such that B is a free A-module
of rank n. Let b
1
, . . . , b
n
B. Then the discriminant D(b
1
, . . . , b
n
) is dened as
det(Tr
B/A
(b
i
b
j
)).
2.3.2. Proposition. If c
i
B and c
i
=

a
ij
b
j
, a
ij
A, then D(c
1
, . . . , c
n
) =
(det(a
ij
))
2
D(b
1
, . . . , b
n
).
Proof. (c
i
)
t
= (a
ij
)(b
j
)
t
, (c
k
c
l
) = (c
k
)
t
(c
l
) = (a
ki
)(b
i
b
j
)(a
lj
)
t
,
(Tr(c
k
c
l
)) = (a
ki
)(Tr(b
i
b
j
))(a
lj
)
t
.
2.3.3. Denition. The discriminant D
B/A
of B over A is the principal ideal of A
generated by the discriminant of any basis of B over A.
2.3.4. Proposition. Let D
B/A
,= 0. Let B be an integral domain. Then a set b
1
, . . . , b
n
is a basis of B over A iff D(b
1
, . . . , b
n
)A = D
B/A
.
Proof. Let D(b
1
, . . . , b
n
)A = D
B/A
. Let c
1
, . . . , c
n
be a basis of B over A and
let b
i
=

j
a
ij
c
j
. Then D(b
1
, . . . , b
n
) = det(a
ij
)
2
D(c
1
, . . . , c
n
). Denote d =
D(c
1
, . . . , c
n
).
Since D(b
1
, . . . , b
n
)A = D(c
1
, . . . , c
n
)A, we get aD(b
1
, . . . , b
n
) = d for some
a A. Then d(1a det(a
ij
)
2
) = 0 and det(a
ij
) is invertible in A, so the matrix (a
ij
)
is invertible in the ring of matrices over A. Thus b
1
, . . . , b
n
is a basis of B over A.
14 Alg number theory
2.3.5. Proposition. Let F be a nite eld or a eld of characteristic zero. Let L be
an extension of F of degree n and let
1
, . . . ,
n
be distinct F-homomorphisms of
L into C. Let b
1
, . . . , b
n
be a basis of L over F. Then
D(b
1
, . . . , b
n
) = det(
i
(b
j
))
2
,= 0.
Proof. det(Tr(b
i
b
j
)) = det(

k

k
(b
i
)
k
(b
j
)) = det((
k
(b
i
))
t
(
k
(b
j
))) = det(
i
(b
j
))
2
.
If det(
i
(b
j
)) = 0, then there exist a
i
L not all zero such that

i
a
i

i
(b
j
) = 0 for
all j. Then

i
a
i

i
(b) = 0 for every b L.
Let

a

i
(b) = 0 for all b L with the minimal number of non-zero a

i
A.
Assume a

1
,= 0.
Let c L be such that L = F(c) (see 1.2.5), then
1
(c) ,=
i
(c) for i > 1.
We nowhave

i
(bc) =

i
(b)
i
(c) = 0. Hence
1
(c)(

i
(b))

i
(b)
i
(c) =

i>1
a

i
(
1
(c)
i
(c))
i
(b) = 0. Put a

i
= a

i
(
1
(c)
i
(c)), so

a

i

i
(b) = 0 with
smaller number of non-zero a

i
than in a

i
, a contradiction.
Corollary. Under the assumptions of the proposition the linear map L Hom
F
(L, F):
b (c Tr
L/F
(bc)) between n-dimensional F-vector spaces is injective, and hence
bijective. Therefore for a basis b
1
, . . . , b
n
of L/F there is a dual basis c
1
, . . . , c
n
of
L/F, i.e. Tr
L/F
(b
i
c
j
) =
ij
.
Proof. If b =

a
i
b
i
, a
i
F and Tr
L/F
(bc) = 0 for all c L, then we get equations

a
i
Tr
L/F
(b
i
b
j
) = 0 this is a system of linear equations in a
i
with nondegenerate
matrix Tr
L/F
(b
i
b
j
), so the only solution is a
i
= 0. Elements of the dual basis c
j
correspond to f
j
Hom
F
(L, F), f
j
(b
i
) =
ij
.
2.3.6. Theorem. Let A be an integrally closed ring and F be its eld of fractions. Let
L be an extension of F of degree n and A

be the integral closure of A in L. Let F

be of characteristic 0. Then A

is an A-submodule of a free A-module of rank n.

Proof. Let e
1
, . . . , e
n
be a basis of F-vector space L. Then due to Example 5 in 2.1.1
there is 0 ,= a
i
A such that a
i
e
i
A

. Then for a =

a
i
we get b
i
= ae
i
A

form a basis of L/F.

Let c
1
, . . . , c
n
be the dual basis for b
1
, . . . , b
n
. Claim: A

c
i
A. Indeed, let
c =

a
i
c
i
A

. Then
Tr
L/F
(cb
i
) =

j
a
j
Tr
L/F
(c
j
b
i
) = a
i
A
by 2.2.5. Now

c
i
A = c
i
A, since c
i
is a basis of L/F.
2.3.7. Theorem (on integral basis). Let A be a principal ideal ring and F be its
eld of fractions of characteristic 0. Let L be an extension of F of degree n. Then
the integral closure A

of A in L is a free A-module of rank n.

2010/2011 15
In particular, the ring of integers O
F
of a number eld F is a free Z-module of
rank equal to the degree of F.
Proof. The description of modules of nite type over PID and the previous theorem
imply that A

is a free A-module of rank m n. On the other hand, by the rst part

of the proof of the previous theorem A

contains n A-linear independent elements

over A. Thus, m = n.
Denition. The discriminant d
F
of any integral basis of O
F
is called the discriminant
of F . Since every two integral bases are related via an invertible matrix with integer
coefcients (whose determinant is therefore 1 ), 2.3.2 implies that d
F
is uniquely
determined.
2.3.8. Examples. 1. Let d be a square-free integer. By 2.1.5 the ring of integers of
Q(

d) has an integral basis 1, where =

d if D , 1 mod 4 and = (1 +

d)/2
if d 1 mod 4.
The discriminant of Q(

d) is equal to
4d if d , 1 mod 4 , and d if d 1 mod 4 .
To prove this calculate directly D(1, ) using the denitions, or use 2.3.9.
2. Let F be an algebraic number eld of degree n and let a F be an inte-
gral element over Z. Assume that D(1, a, . . . , a
n1
) is a square free integer. Then
1, a, . . . , a
n1
is a basis of O
F
over Z, so O
F
= Z[a]. Indeed: choose a ba-
sis b
1
, . . . , b
n
of O
F
over Z and let c
1
, . . . , c
n
= 1, a, . . . , a
n1
. Let c
i
=

a
ij
b
j
. By 2.3.2 we have D(1, a, . . . , a
n1
) = (det(a
ij
)
2
D(b
1
, . . . , b
n
). Since
D(1, a, . . . , a
n1
) is a square free integer, we get det(a
ij
) = 1, so (a
ij
) is invertible
in M
n
(Z), and hence 1, a, . . . , a
n1
is a basis of O
F
over Z.
2.3.9. Example. Let F be of characteristic zero and L = F(b) be an extension of
degree n over F. Let f(X) be the minimal polynomial of b over F whose roots are
b
i
. Then
f(X) =

(X b
j
), f

(b
i
) =

j=i
(b
i
b
j
),
N
L/F
f

(b) =

i
f

(
i
b) =

i
f

(b
i
).
Then
D(1, b, . . . , b
n1
) = det(b
j
i
)
2
= (1)
n(n1)/2

i=j
(b
i
b
j
) = (1)
n(n1)/2
N
L/F
(f

(b)).
Let f(X) = X
n
+ aX + c. Then
b
n
= ab c, b
n1
= a cb
1
16 Alg number theory
and
e = f

(b) = nb
n1
+ a = n(a cb
1
) + a,
so
b = nc(e + (n 1)a)
1
.
The minimal polynomial g(Y ) of e over F corresponds to the minimal polynomial
f(X) of b; it is the numerator of c
1
f(nc(y + (n 1)a)
1
), i.e.
g(Y ) = (Y + (n 1)a)
n
na(Y + (n 1)a)
n1
+ (1)
n
n
n
c
n1
.
Hence
N
L/F
(f

(b)) = g(0)(1)
n
= n
n
c
n1
+ (1)
n1
(n 1)
n1
a
n
,
so
D(1, b, . . . , b
n1
)
= (1)
n(n1)/2
(n
n
c
n1
+ (1)
n1
(n 1)
n1
a
n
).
For n = 2 one has a
2
4c, for n = 3 one has 27c
2
4a
3
.
For example, let f(X) = X
3
+ X + 1. It is irreducible over Q. Its discriminant is
equal to (31), so according to example 2.5.3 O
F
= Z[a] where a is a root of f(X)
and F = Q[a].
2.4. Cyclotomic elds
2.4.1. Denition. Let
n
be a primitive nth root of unity. The eld Q(
n
) is called
the ( nth) cyclotomic eld.
2.4.2. Theorem. Let p be a prime number and z be a primitive p th root of unity. The
cyclotomic eld Q(
p
) is of degree p 1 over Q. Its ring of integers coincides with
Z[
p
].
Proof. Denote z =
p
. Let f(X) = (X
p
1)/(X 1) = X
p1
+ + 1. Recall that
z 1 is a root of the polynomial g(Y ) = f(1 + Y ) = Y
p1
+ + p is a p-Eisenstein
polynomial, so f(X) is irreducible over Q, [Q(z) : Q[ = p 1 and 1, z, . . . , z
p2
is
a basis of the Q-vector space Q(z).
Let O be the ring of integers of Q(z). Since the monic irreducible polynomial
of z over Q has integer coefcients, z O. Since z
1
is a primitive root of unity,
z
1
O. Thus, z is a unit of O.
Then z
i
O for all i Z ( z
1
= z
p1
). We have 1z
i
= (1z)(1+ +z
i1
)
(1 z)O.
2010/2011 17
Denote by Tr and N the trace and norm for Q(z)/Q. Note that Tr(z) = 1 and
since z
i
for 1 i p1 are primitive p th roots of unity, Tr(z
i
) = 1; Tr(1) = p1.
Hence
Tr(1 z
i
) = p for 1 i p 1 .
Furthermore, N(z 1) is equal to the free term of g(Y ) times (1)
p1
, so N(z 1) =
(1)
p1
p and
N(1 z) =

1ip1
(1 z
i
) = p,
since 1 z
i
are conjugate to 1 z over Q. Therefore pZ is contained in the ideal
I = (1 z)O Z.
If I = Z, then 1 z would be a unit of O and so would be its conjugates 1 z
i
,
which then implies that p as their product would be a unit of O. Then p
1
OQ = Z,
a contradiction. Thus,
I = (1 z)O Z = pZ.
Now we prove another auxiliary result:
Tr((1 z)O) pZ.
Indeed, every conjugate of y(1z) for y O is of the type y
i
(1z
i
) with appropriate
y
i
O, so Tr(y(1 z)) =

y
i
(1 z
i
) I = pZ.
Now let x =

0ip2
a
i
z
i
O with a
i
Q. We aim to show that all a
i
belong
to Z. Fromthe calculation of the traces of z
i
it follows that Tr((1z)x) = a
0
Tr(1z)+

0<ip2
a
i
Tr(z
i
z
i+1
) = a
0
p and so a
0
p Tr((1z)O) pZ; therefore, a
0
Z.
Since z is a unit of O, we deduce that x
1
= z
1
(xa
0
) = a
1
+a
2
z+ +a
p2
z
p3
O.
By the same arguments a
1
Z. Looking at x
i
= z
1
(x
i1
a
i1
) O we conclude
a
i
Z for all i. Thus O = Z[z].
2.4.3. The discriminant of O/Z is the ideal of Z generated by D(1, z, . . . , z
p2
)
which by 2.3.9 is equal (1)
(p1)(p2)/2
N(f

(z)). We have f

(z) = pz
p1
/(z 1) =
pz
1
/(z 1) and N(f

(z)) = N(p)N(z)
1
/N(z 1) = p
p1
(1)
p1
/((1)
p1
p) =
p
p2
. Thus, the discriminant of OZ is the principal ideal (1)
(p1)(p2)/2
p
p2
Z =
p
p2
Z.
2.4.4. In general, the extension Q(
m
)/Q is a Galois extension and elements of the
Galois group Gal(Q(
m
)/Q) are determined by their action on the primitive mth root

m
of unity:
i : (
m
) =
i
m
, (i, m) = 1.
This map induces a group isomorphism
Gal(Q(
m
)/Q) (Z/mZ)

.
18 Alg number theory
One can prove that the ring of integers of Q(
m
) is Z(
m
).
2010/2011 19
3. Dedekind rings
3.1. Noetherian rings
3.1.1. Recall that a module M over a ring is called a Noetherian module if one of the
following equivalent properties is satised:
(i) every submodule of M is of nite type;
(ii) every increasing sequence of submodules stabilizes;
(iii) every nonempty family of submodules contains a maximal element with respect
to inclusion.
A ring A is called Noetherian if it is a Noetherian A-module.
Example. A PID is a Noetherian ring, since every ideal of it is generated by one
element.
Lemma. Let M be an A-module and N is a submodule of M. Then M is a Noethe-
rian A-module iff N and M/N are.
Corollary 1. If N
i
are Noetherian A-modules, so is
n
i=1
N
i
.
Corollary 2. Let A be a Noetherian ring and let M be an A-module of nite type.
Then M is a Noetherian A-module.
3.1.2. Proposition. Let A be a Noetherian integrally closed ring. Let K be its eld of
fractions and let L be a nite extension of K. Let A

be the integral closure of A in

L. Suppose that K is of characteristic 0. Then A

is a Noetherian ring.
Proof. According to 2.3.6 A

is a submodule of a free A-module of nite rank. Hence

A

is a Noetherian A-module. Every ideal of A

is in particular an A-submodule of
A

. Hence every increasing sequence ideals of A

stabilizes and A

is a Noetherian
ring.
3.1.3. Example. The ring of integers O
F
of a number eld F is a Noetherian ring.
It is a Z-module of rank n where n is the degree of F.
Every nonzero element of O
F
0 factorizes into a product of prime elements and
units (not uniquely in general).
Indeed, assume the family of principal ideals (a) which are generated by elements
O
F
which are not products of prime elements and units is nonempty and then choose a
maximal element (a) in this family. The element a is not a unit, and a is not prime.
Hence there is a factorization a = bc with both b, c , O

F
. Then (b), (c) are strictly
larger than (a), so b and c are products of prime elements. Then a is, a contradiction.
20 Alg number theory
3.2. Dedekind rings
3.2.1. Denition. An integral domain A is called a Dedekind ring if
(i) A is a Noetherian ring;
(ii) A is integrally closed;
(iii) every non-zero prime ideal of A is maximal.
Example. Every principal ideal domain A is a Dedekind ring.
Proof: for (i) see 3.1.1 and for (ii) see 2.1.4. If (a) is a prime ideal and (a) (b) ,= A,
then b isnt a unit of A and b divides a. Write a = bc. Since (a) is prime, either b
or c belongs to (a). If b does then (a) = (b). If b doesnt, then c must belong to (a),
so c = ad for some d A, and a = bc = bda which means that b is a unit of A, a
contradiction. Thus, property (iii) is satised as well.
3.2.2. Lemma. Let A be an integral domain. Let K be its eld of fractions and let
L be a nite extension of K. Let B be the integral closure of A in L. Let P be a
non-zero prime ideal of B. Then P A is a non-zero prime ideal of A.
Proof. Let P be a non-zero prime ideal of B. Then P A ,= A, since otherwise
1 P A and hence P = B.
If c, d A and cd P A, then either c P A or d P A. Hence P A
is a prime ideal of A.
Let b P, b ,= 0. Then b satises a polynomial relation b
n
+a
n1
b
n1
+ +a
0
= 0
with a
i
A. We can assume that a
0
,= 0. Then a
0
= (b
n
+ + a
1
b) A P, so
P A is a non-zero prime ideal of A.
3.2.3. Theorem. Let A be a Dedekind ring. Let K be its eld of fractions and let L
be a nite extension of K. Let B be the integral closure of A in L. Suppose that K
is of characteristic 0. Then B is a Dedekind ring.
Proof. B is Noetherian by 3.1.2. It is integrally closed due to 2.1.6. By 3.2.2 if P is
a non-zero proper prime ideal of B, then P A is a non-zero prime ideal of A. Since
A is a Dedeking ring, it is a maximal ideal of A. The quotient ring B/P is integral
over the eld A/(P A). Hence by 2.1.7 B/P is a eld and P is a maximal ideal of
B.
3.2.4. Example. The ring of integers O
F
of a number eld F is a Dedekind ring.
2010/2011 21
3.3. Factorization in Dedekind rings
3.3.1. Lemma. Every non-zero ideal in a Dedekind ring A contains some product of
maximal ideals.
Proof. If not, then the set of non-zero ideals which do not contain products of maximal
ideals is non-empty. Let I be a maximal element with this property. The ideal I is not A
and is not a maximal ideal, since it doesnt contain a product of maximal ideals. Hence I
is not a prime ideal. Therefore there are a, b A such that ab I and a, b , I. Since
I +aA and I +bA are strictly greater than I, there are maximal ideals P
i
and Q
j
such
that

P
i
I + aA and

Q
j
I + bA. Then

P
i

Q
j
(I + aA)(I + bA) I,
a contradiction.
3.3.2. Lemma. Let a prime ideal P of A contain I
1
. . . I
m
, where I
j
are ideals of
A. Then P contains one of I
j
.
Proof. If I
k
, P for all 1 k m, then take a
k
I
k
P and consider the product
a
1
. . . a
m
. It belongs to P, therefore one of a
i
belongs to P, a contradiction.
3.3.3. The next proposition shows that for every non-zero ideal I of a Dedekind ring
A there is an ideal J such that IJ is a principal non-zero ideal of A. Moreover, the
proposition gives an explicit description of J.
Proposition. Let I be a non-zero ideal of a Dedekind ring A and b be a non-zero
element of I. Let K be the eld of fractions of A. Dene
J = a K : aI bA.
Then J is an ideal of A and IJ = bA.
Proof. Since b I, we get bA I.
If a J then aI bA I, so aI I. Nowwe use the Noetherian and integrality
property of Dedekind rings: Since I is an A-module of nite type, by Remark in 2.1.1
a is integral over A. Since A is integrally closed, a A. Thus, J A.
The set J is closed with respect to addition and multiplication by elements of A,
so J is an ideal of A. It is clear that IJ bA. Assume that IJ ,= bA and get a
contradiction.
The ideal b
1
IJ is a proper ideal of A, and hence it is contained in a maximal
ideal P. Note that b J, since bI bA. So b
2
IJ and b b
1
IJ, bA b
1
IJ.
By 3.3.1 there are non-zero prime ideals P
i
such that P
1
. . . P
m
bA. Let m be the
minimal number with this property.
22 Alg number theory
We have
P
1
. . . P
m
bA b
1
IJ P.
By 3.3.2 P contains one of P
i
. Without loss of generality we can assume that P
1
P.
Since P
1
is maximal, P
1
= P.
If m = 1, then P bA b
1
IJ P, so P = bA. Since bA I we get P I.
Since P is maximal, either I = P or I = A. The denition of J implies in the rst
case J = a K : aI = aP bA = P = A and IJ = bA and in the second case
b J implies bA J = a K : aA bA a K : a bA = bA and so
J = bA and IJ = bA.
Let m > 1. Note that P
2
. . . P
m
, bA due to the denition of m. Therefore,
there is d P
2
. . . P
m
such that d , bA. Since b
1
IJ P, db
1
IJ dP
PP
2
. . . P
m
bA. So (db
1
J)I bA, and the dening property of J implies that
db
1
J J. Since J is an A-module of nite type, by 2.1.1 db
1
belongs to A, i.e.
d bA, a contradiction.
3.3.4. Corollary 1 (Cancellation property). Let I, J, H be non-zero ideals of A,
then IH = JH implies I = J.
Proof. Let H

be an ideal such that HH

= aA is a principal ideal. Then aI = aJ

and I = J.
3.3.5. Corollary 2 (Factorization property). Let I and J be ideals of A. Then
I J if and only if I = JH for an ideal H.
Proof. If I J and J is non-zero, then let J

be an ideal of A such that JJ

= aA
is a principal ideal. Then IJ

aA, so H = a
1
IJ

is an ideal of A. Now
JH = Ja
1
IJ

= a
1
IJJ

= a
1
aI = I.
3.3.6. Theorem. Every proper ideal of a Dedekind ring factorizes into a product of
maximal ideals whose collection is uniquely determined.
Proof. Let I be a non-zero ideal of A. There is a maximal ideal P
1
which contains
I. Then by the factorization property 3.3.5 I = P
1
Q
1
for some ideal Q
1
. Note
that I Q
1
is a proper inclusion, since otherwise AQ
1
= Q
1
= I = P
1
Q
1
and by
the cancellation property 3.3.4 P
1
= A, a contradiction. If Q
1
,= A, then there is a
maximal ideal P
2
such that Q
1
= P
2
Q
2
. Continue the same argument: eventually we
have I = P
1
. . . P
n
Q
n
and I Q
1
Q
n
are all proper inclusions. Since A is
Noetherian, Q
m
= A for some m and then I = P
1
. . . P
m
.
If P
1
. . . P
m
= Q
1
. . . Q
n
, then P
1
Q
1
. . . Q
n
and by 3.3.2 P
1
being a prime
ideal contains one of Q
i
, so P
1
= Q
i
. Using 3.3.4 cancel P
1
on both sides and use
induction.
2010/2011 23
3.3.7. Remark. A maximal ideal P of A is involved in the factorization of I iff
I P.
Indeed, if I P, then I = PQ by 3.3.5.
3.3.8. Example. Let A = Z[

5]. This is a Dedekind ring, since 5 , 1 mod 4,

and A is the ring of integers of Q(

5).
We have the norm map N(a + b

5) = a
2
+ 5b
2
. If an element u is a unit of A
then uv = 1 for some v A, and the product of two integers N(u) and N(v) is 1,
thus N(u) = 1. Conversely, if N(u) = 1 then u times its conjugate u

is one, and so
u is a unit of A. Thus, u A

iff N(u) Z

.
The norms of 2, 3, 1

5 are 4, 9, 6. It is easy to see that 2, 3 are not in the

image N(A).
If, say, 2 were not a prime element in A, then 2 =
1

2
and 4 = N(
1
)N(
2
)
with both norms being proper divisors of 4, a contradiction. Hence 2 is a prime element
of A, and similarly 3, 1

5 are.
Now 2, 3, 1

5 are prime elements of A and

6 = 2 3 = (1 +

5)(1

5).
Note that 2, 3, 1

5 are not associated with each other (the quotient is not a unit)
since their norms differ not by a unit of Z. Thus A isnt a UFD.
The ideals
(2, 1 +

5), (3, 1 +

5), (3, 1

5)
are maximal.
For instance, [A/(2)[ = 4, and it is easy to show that A ,= (2, 1 +

5) ,= (2), so
[A/(2, 1 +

5)[ = 2, therefore A/(2, 1 +

5) is isomorphic to Z/2Z, i.e. is a eld.

We get factorization of ideals
(2) = (2, 1 +

5)
2
,
(3) = (3, 1 +

5)(3, 1

5),
(1 +

5) = (2, 1 +

5)(3, 1 +

5),
(1

5) = (2, 1 +

5)(3, 1

5).
To prove the rst equality note that (1 +

5)
2
= 4 + 2

5 (2), so the
RHSLHS; we also have 2 = 2(1 +

5) 2
2
(1 +

5)
2
RHS, so LHS=RHS.
For the second equality use (1+

5)(1

5) = 6 (3), 3 = 3
2
(1+

5)(1

5) RHS.
For the third equality use 6 (1+

5), 1+

5 = 3(1+

5)2(1+

5) RHS.
For the fourth equality use conjugate the third equality and use (2, 1 +

5) =
(2, 1

5).
24 Alg number theory
Thus
(2) (3) = (2, 1 +

5)
2
(3, 1 +

5)(3, 1

5)
= (2, 1 +

5)(3, 1 +

5)(2, 1 +

5)(3, 1

5)
= (1 +

5)(1

5).
3.3.9. Lemma. Let I + J = A. Then I
n
+ J
m
= A for every n, m 1.
Proof. We have A = (I + J) . . . (I + J) = I(...) + J
m
I + J
m
, so I + J
m
= A.
Similarly I
n
+ J
m
= A.
Proposition. Let P be a maximal ideal of A. Then there is an element P such
that
P = A+ P
n
for every n 2.
Hence the ideal P/P
n
is a principal ideal of the factor ring A/P
n
. Moreover, it
is the only maximal ideal of that ring.
Every ideal of the ring A/P
n
is principal of the form P
m
/P
n
= (
m
A+ P
n
)/P
n
for some m n.
Proof. If P = P
2
, then P = A by cancellation property, a contradiction. Let
P P
2
. Since A+ P
n
P, factorization property implies that A+ P
n
= PQ for
an ideal Q.
Note that Q , P, since otherwise P
2
, a contradiction.
Therefore, P + Q = A. The Lemma implies P
n1
+ Q = A. Then
P = P(Q+ P
n1
) PQ+ P
n
= A+ P
n
P,
so P = A+ P
n
.
For m n we deduce P
m

m
A+ P
n
P
m
, so P
m
=
m
A+ P
n
.
Let I be a proper ideal of A containing P
n
. Then by factorization property
P
n
= IK with some ideal K. Hence the factorization of I involves powers of P
only, so I = P
m
, 0 < m n. Hence ideals of A/P
n
are P
m
/P
n
with m n.
3.3.10. Corollary. Every ideal in a Dedekind ring is generated by 2 elements.
Proof. Let I be a non-zero ideal, and let a be a non-zero element of I. Then
aA = P
n
1
1
. . . P
n
m
m
with distinct maximal ideals P
i
.
By Lemma 3.3.9 we have P
n
1
1
+ P
n
k
k
= A if l ,= k, so we can apply the Chinese
remainder theorem which gives
A/aA A/P
n
1
1
A/P
n
m
m
.
2010/2011 25
For the ideal I/aA of A/aA we get
I/aA (I + P
n
1
1
)/P
n
1
1
(I + P
n
m
m
)/P
n
m
m
.
Each of ideals (I + P
n
i
i
)/P
n
i
i
is of the form (
l
i
i
A+ P
n
i
i
)/P
n
i
i
by 3.3.9. Hence I/aA
is isomorphic to

(
l
i
i
A + P
n
i
i
)/P
n
i
i
. Using the Chinese remainder theorem nd
b A such that b
l
i
i
belongs to P
n
i
i
for all i. Then I/aA = (aA + bA)/aA and
I = aA+ bA.
3.3.11. Theorem. A Dedekind ring A is a UFD if and only if A is a PID.
Proof. Let A be not a PID. Since every proper ideal is a product of maximal ideals,
there is a maximal ideal P which isnt principal. Consider the family F of non-zero
ideals I such that PI is principal. It is nonempty by 3.3.5. Let I be a maximal
element of this family and PI = aA, a ,= 0.
Note that I isnt principal, because otherwise I = xA and PI = xP = aA, so
a is divisible by x. Put y = ax
1
, then (x)P = (x)(y) and by 3.3.4 P = (y), a
contradiction.
Claim: a is a prime element of A. First, a is not a unit of A: otherwise
P PI = aA = A, a contradiction. Now, if a = bc, then bc P, so either b P or
c P. By 3.3.5 then either bA = PJ or cA = PJ for an appropriate ideal J of A.
Since PI PJ, we get aI = IPI IPJ = aJ and I J. Note that J F. Due
to maximality of I we deduce that I = J, and hence either bA or cA is equal to aA.
Then one of b, c is asociated to a, so a is a prime element.
P , aA, since otherwise aA = PI aI, so A = I, a contradiction.
I , aA, since otherwise aA I implies aA = I, I is principal, a contradiction.
Thus, there are d P and e I not divisible by a. We also have ed PI = aA
is divisible by the prime element a. This can never happen in UFD. Thus, A isnt a
UFD.
Using this theorem, to establish that the ring Z[

5] of 3.3.8 is not a unique

factorization domain it is sufcient to indicate a non-principal ideal of it.
3.4. The norm of an ideal
In this subsection F is a number eld of degree n, O
F
is the ring of integers of F.
3.4.1. Proposition. For a non-zero element a O
F
[O
F
: aO
F
[ = [N
F/Q
(a)[.
26 Alg number theory
Proof. We know that O
F
is a free Z-module of rank n. The ideal aO
F
is a free
submodule of O
F
of rank n, since if x
1
, . . . , x
m
are generators of aO
F
, then
a
1
x
1
, . . . , a
1
x
m
are generators of O
F
, so m = n. By the theorem on the structure
of modules over principal ideal domains, there is a basis a
1
, . . . , a
n
of O
F
such that
e
1
a
1
, . . . , e
n
a
n
is a basis of aO
F
with appropriate e
1
[ . . . [e
n
. Then O
F
/aO
F
is
isomorphic to

Z/e
i
Z, so [O
F
: aO
F
[ =

[e
i
[. By the denition N
F/Q
(a) is equal
to the determinant of the matrix of the linear operator f : O
F
O
F
, b ab. Note
that aO
F
has another basis: aa
1
, . . . , aa
n
, so (aa
1
, . . . , aa
n
) = (e
1
a
1
, . . . , e
n
a
n
)M
with an invertible matrix M with integer entries. Thus, the determinant of M is 1
and N
F/Q
(a) is equal to

e
i
.
3.4.2. Corollary. [O
F
: aO
F
[ = [a[
n
for every non-zero a Z.
Proof. N
F/Q
(a) = a
n
.
3.4.3. Denition. The norm N(I) of a non-zero ideal I of O
F
is its index [O
F
: I[.
Note that if I ,= 0 then N(I) is a nite number.
Indeed, by 3.4.1 N(aO
F
) = [N
F/Q
(a)[ for a non-zero a which belongs to I. Then
aO
F
I and N(I) N(aO
F
) = [N
F/Q
(a)[.
3.4.4. Proposition. If I, J are non-zero ideals of O
F
, then N(IJ) = N(I)N(J).
Proof. Since every ideal factors into a product of maximal ideals by 3.3.6, it is sufcient
to show that N(IP) = N(I)N(P) for a maximal ideal P of O
F
.
The LHS = [O
F
: IP[ = [O
F
: I[[I : IP[. Recall that P is a maximal ideal of
O
F
, so O
F
/P is a eld.
The quotient I/IP can be viewed as a vector space over O
F
/P. Its subspaces
correspond to ideals between IP and I according to the description of ideals of the
factor ring. If IP J I, then by 3.3.5 J = IQ for an ideal Q of O
F
.
By 3.3.3 there is a non-zero ideal I

such that II

is a principal non-zero ideal

aO
F
. Then IP IQ implies aP aQ implies P Q. Therefore either Q = P
and then J = IP or Q = O
F
and then J = I. Thus, the only subspaces of the vector
space I/IP are itself and the zero subspace IP/IP. Hence I/IP is of dimension
one over O
F
/P and therefore [I : IP[ = [O
F
: P[.
3.4.5. Corollary. If I is a non-zero ideal of O
F
and N(I) is prime, then I is a
maximal ideal.
Proof. If I = JK, then N(J)N(K) is prime, so, say, N(J) = 1 and J = O
F
. So I
has no proper prime divisors, and therefore is a maximal ideal.
2010/2011 27
3.5. Splitting of prime ideals in eld extensions
In this subsection F is a number eld and L is a nite extension of F. Let O
F
and
O
L
be their rings of integers.
3.5.1. Proposition-Denition. Let P be a maximal ideal of O
F
and Q a maximal
ideal of O
L
. Then Q is said to lie over P and P is said to lie under Q if one of the
following equivalent conditions is satised:
(i) PO
L
Q;
(ii) P Q;
(iii) Q O
F
= P.
Proof. (i) is equivalent to (ii), since 1 O
L
. (ii) implies Q O
F
contains P, so
either Q O
F
= P or Q O
F
= O
F
, the latter is impossible since 1 , Q. (iii)
implies (ii).
3.5.2. Proposition. Every maximal ideal of O
L
lies over a unique maximal ideal P of
O
F
. For a maximal ideal P of O
F
the ideal PO
L
is a proper non-zero ideal of O
L
.
Let PO
L
=

Q
i
be the factorization into a product of prime ideals of O
L
. Then Q
i
are exactly those maximal ideals of O
L
which lie over P.
Proof. The rst assertion follows from 3.2.2.
Note that by 3.3.3 for b P P
2
there is an ideal J of O
F
such that PJ = bO
F
.
Then J , P, since otherwise b P
2
, a contradiction. Take an element c J P.
Then cP bO
F
.
If PO
L
= O
L
, then cO
L
= cPO
L
bO
L
, so cb
1
O
L
F = O
F
and
c bO
F
P, a contradiction. Thus, PO
L
is a proper ideal of O
L
.
According to 3.5.1 a prime ideal Q of O
L
lies over P iff PO
L
Q which is
equivalent by 3.3.7 to the fact that Q is involved in the factorization of PO
L
.
3.5.3. Lemma. Let P be a maximal ideal of O
F
which lie under a maximal ideal Q
of O
L
. Then the nite eld O
F
/P is a subeld of the nite eld O
L
/Q.
Proof. O
L
/Q is nite by 3.4.3. The kernel of the homomorphism O
F
O
L
/Q is
equal to Q O
F
= P, so O
F
/P can be identied with a subeld of O
L
/Q.
3.5.4. Corollary. Let P be a maximal ideal of O
F
. Then P Z = pZ for a prime
number p and N(P) is a positive power of p.
Proof. P Z = pZ for a prime number p by 3.2.2. Then O
F
/P is a vector space
over Z/pZ of nite positive dimension, therefore [O
F
: P[ is a power of p.
28 Alg number theory
3.5.5. Denition. Let a maximal ideal P of O
F
lie under a maximal ideal Q of
O
L
. The degree of O
L
/Q over O
F
/P is called the inertia degree f(Q[P). If
PO
L
=

Q
e
i
i
is the factorization of PO
L
with distinct prime ideals Q
i
of O
L
, then
e
i
is called the ramication index e(Q
i
[P).
3.5.6. Lemma. Let M be a nite extension of L and P Q R be maximal
ideals of O
F
, O
L
and O
M
correspondingly. Then f(R[P) = f(Q[P)f(R[Q) and
e(R[P) = e(Q[P)e(R[Q).
Proof. The rst assertion follows from 1.1.1. Since PO
L
= Q
e(Q|P)
. . . , we get
PO
M
= Q
e(Q|P)
O
M
= (QO
M
)
e(Q|P)
= (R
e(R|Q)
)
e(Q|P)
. . . , so the second
assertion follows.
3.5.7. Theorem. Let Q
1
, . . . Q
m
be different maximal ideals of O
L
which lie over a
maximal ideal P of O
F
. Let n = [L : F[. Then
m

i=1
e(Q
i
[P)f(Q
i
[P) = n.
Proof. We consider only the case F = Q. Apply the norm to the equality pO
L
=

Q
e
i
i
. Then by 3.4.2, 3.4.4
p
n
= N(pO
L
) =

N(Q
i
)
e
i
=

p
f(Q
i
|P)e(Q
i
|P)
.
3.5.8. Example. One can describe in certain situations howa prime ideal (p) factorizes
in nite extensions of Q, provided the factorization of the monic irreducible polynomial
of an integral generator (if it exists) modulo p is known.
Let the ring of integers O
F
of an algebraic number eld F be generated by one
element : O
F
= Z[], and f(X) Z[X] be the monic irreducible polynomial of
over Q.
Let f
i
(X) Z[X] be monic polynomials such that
f(X) =
m

i=1
f
i
(X)
e
i
F
p
[X]
is the factorization of f(X) where f
i
(X) is an irreducible polynomial over F
p
. Since
O
F
Z[X]/(f(X)), we have
O
F
/(p) Z[X]/(p, f(X)) F
p
[X]/(f(X)),
and
O
F
/(p, f
i
()) Z[X]/(p, f(X), f
i
(X)) F
p
[X]/(f
i
(X)).
2010/2011 29
Putting P
i
= (p, f
i
()) we see that O
F
/P
i
is isomorphic to the eld F
p
[X]/(f
i
(X)),
hence P
i
is a maximal ideal of O
F
dividing (p). We also deduce that
N(P
i
) = p
|F
p
[X]/(f
i
(X)):F
p
|
= p
deg f
i
.
Now

P
e
i
i
=

(p, f
i
())
e
i
pO
F
, since

f
i
()
e
i
f() pO
F
. We also
get N(

P
e
i
i
) = p

e
i
deg f
i
= p
n
= N(pO
F
). Therefore from 3.5.7 we deduce that
(p) =

m
i=1
P
e
i
i
is the factorization of (p).
So we have proved
Theorem. Let the ring of integers O
F
of an algebraic number eld F be generated by
one element : O
F
= Z[], and f(X) Z[X] be the monic irreducible polynomial
of over Q. Let f
i
(X) Z[X] be irreducible polynomials such that
f(X) =
m

i=1
f
i
(X)
e
i
F
p
[X]
is the factorization of f(X) where f
i
(X) is an irreducible polynomial over F
p
.
Then in O
F
(p) =
m

i=1
P
e
i
i
where P
i
= (p, f
i
()) is a maximal ideal of O
F
with norm p
deg f
i
.
DenitionExample. Let F = Q and L = Q(

d) with a square free integer d.

Let p be a prime in Z and let pO
L
=

m
i=1
Q
e
i
i
. Then there are three cases:
(i) m = 2, e
1
= e
2
= 1, f(Q
i
[P) = 1. Then pO
L
= Q
1
Q
2
, Q
1
,= Q
2
. We say that
p splits in L.
(ii) m = 1, e
1
= 2, f(Q
1
[P) = 1. Then pO
L
= Q
2
1
. We say that p ramies in L.
(iii) m = 1, e
1
= 1, f(Q
1
[P) = 2. Then pO
L
= Q
1
. We say that p remains prime
in L.
Using the previous theorem we see that p remains prime in O
F
iff f is irreducible
over F
p
; p splits ( pO
F
= P
1
. . . P
m
) iff f is separable and reducible, and p ramies
( pO
F
= P
e
) iff f is a positive power of an irreducible polynomial over F
p
.
3.5.9. In particular, if F = Q(

d) then one can take

d for d , 1 mod 4 and (1 +

d)/2 for d 1 mod 4 as . Then f(X) = X

2
d and f(X) = X
2
X +(1 d)/4
resp.
We have X
2
X + (1 d)/4 = 1/4(Y
2
d) where Y = 2X 1, so if p is
odd (so the image of 2 is invertible in F
p
), the factorization of f(X) corresponds to
the factorization of X
2
d independently of what d is. The factorization of X
2
d
certainly depends on whether d is a quadratic residue modulo p, or not.
30 Alg number theory
For p = 2 f(X) = (X d)
2
F
2
[X] and f(X) = X
2
+ X + (1 d)/4 F
2
[X]
resp.
Thus, we get
Theorem. If p is odd prime, then
p splits in L = Q(

d) iff d is a quadratic residue mod p.

p ramies in L iff d is divisible by p.
p remains prime in L iff d is a quadratic non-residue mod p.
If p = 2 then
if d 1 mod 8, then 2 splits in Q(

d),
if d , 1 mod 4 then 2 ramies in Q(

d);
if d 1 mod 4, d , 1 mod 8 then 2 remains prime in Q(

d).
3.5.10. Let p be an odd prime. Recall from 2.4.2 that the ring of integers of the
p th cyclotomic eld Q(
p
) is generated by
p
. Its irreducible monic polynomial is
f(X) = X
p1
+ + 1 = (X
p
1)/(X 1). Since X
p
1 (X 1)
p
mod p we
deduce that (f(X), p) = ((X 1)
p1
, p). Therefore by 3.5.8 p = (
p
1)
p1
ramies
in Q(
p
)/Q. For any other prime l one can show that the polynomial f(X) modulo l
is the product of distinct irreducible polynomials over F
l
. Thus, no other prime ramies
in Q(
p
)/Q.
3.6. Finiteness of the ideal class group
In this subsection O
F
is the ring of integers of a number eld F.
3.6.1. Denition. For two non-zero ideals I and J of O
F
dene the equivalence rela-
tion I J if there are non-zero a, b O
F
such that aI = bJ. Classes of equivalence
are called ideal classes. Dene the product of two classes with representatives I and J
as the class containing IJ. Then the class of O
F
(consisting of all nonzero principal
ideals) is the indentity element. By 3.3.3 for every non-zero I there is a non-zero J
such that IJ is a principal ideal, i.e. every ideal class is invertible. Thus ideal classes
form an abelian group which is called the ideal class group C
F
of the number eld F.
The ideal class group shows how far from PID the ring O
F
is. Note that C
F
consists of one element iff O
F
is a PID iff O
F
is a UFD.
3.6.2. Proposition. There is a positive real number c such that every non-zero ideal I
of O
F
contains a non-zero element a with
[N
F/Q
(a)[ cN(I).
2010/2011 31
Proof. Let n = [F : Q[. According to 2.3.7 there is a basis a
1
, . . . , a
n
of the
Z-module O
F
which is also a basis of the Q-vector space F. Let
1
, . . . ,
n
be all
distinct Q-homomorphisms of F into C. Put
c =
n

i=1
_
n

j=1
[
i
a
j
[
_
.
Then c > 0.
For a non-zero ideal I let m be the positive integer satisfying the inequality
m
n
N(I) < (m + 1)
n
. In particular, [O
F
: I[ < (m + 1)
n
. Consider (m + 1)
n
elements

n
j=1
m
j
a
j
with 0 m
j
m, m
j
Z. There are two of them which
have the same image in O
F
/I. Their difference 0 ,= a =

n
j=1
n
j
a
j
belongs to I and
satises [n
j
[ m.
Now
[N
F/Q
(a)[ =
n

i=1
[
i
a[ =
n

i=1
[
n

j=1
n
j

i
a
j
[
n

i=1
_
n

j=1
[n
j
[[
i
a
j
[
_
m
n
c cN(I).
3.6.3. Corollary. Every ideal class of O
F
contains an ideal J with N(J) c.
Proof. Given ideal class, consider an ideal I of the inverse ideal class. Let a I be as
in the theorem. By 3.3.3 there is an ideal J such that IJ = aO
F
, so (I)(J) = (aO
F
) = 1
in C
F
. Then J belongs to the given ideal class. Using 3.4.1 and 3.4.4 we deduce that
N(I)N(J) = N(IJ) = N(aO
F
) = [N
F/Q
(a)[ cN(I). Thus, N(J) c.
3.6.4. Theorem. The ideal class group C
F
is nite. The number [C
F
[ is called the
class number of F.
Proof. By 3.5.4 and 3.5.2 for each prime p there are nitely many maximal ideals P
lying over (p), and N(P) = p
m
for m 1. Hence there are nitely many ideals

P
e
i
i
satisfying N(

P
e
i
i
) c.
Example. The class number of Q(

19) is 1, i.e. every ideal of the ring of integers

of Q(

19) is principal.
Indeed, by 2.3.8 we can take a
1
= 1, a
2
= (1 +

19)/2 as an integral basis of the

ring of integers of Q(

19). Then
c =
_
1 + [(1 +

19)/2[
__
1 + [(1

19)/2[
_
= 10.4... .
So every ideal class of O
Q(

19)
contains an ideal J with N(J) 10. Let J =

P
e
i
i
be the factorization of J, then N(P
i
) 10 for every i.
By Corollary 3.5.4 we know that N(P
i
) is a positive power of a prime integer, say
p
i
. From3.5.2 we knowthat P
i
is a prime divisor of the ideal (p
i
) of O
Q(

19)
. So we
32 Alg number theory
need to look at prime integer numbers not greater than 7 and their prime ideal divisors
as potential candidates for non-principal ideals. Now prime number 3 has the property
that -19 is a quadratic non-residue modulo them, so by Theorem 3.5.9 it remains prime
in O
Q(

19)
.
Odd prime numbers 5, 7 have the property that -19 is a quadratic residue module
them, so by Theorem 3.5.9 they split in O
Q(

19)
. It is easy to check that
5 =
_
(1 +

19)/2
__
(1

19)/2
_
,
7 =
_
(3 +

19)/2
__
(3

19)/2
_
..
Each of ideals generated by a factor on the right hand side is prime by 3.4.5, since its
norm is a prime number. So prime ideal factors of (5), (7) are principal ideals.
Finally, 2 remains prime in O
Q(

19)
, as follows from 3.5.9.
Thus, O
Q(

19)
is a principal ideal domain.
Remark. The bound given by c is not good in practical applications. A more rened
estimation is given by Minkowskis Theorem 3.6.6.
3.6.5. Denition. Let F be of degree n over Q. Let
1
, . . . ,
n
be all Q-homo-
morphisms of F into C. Let
: C C
be the complex conjugation. Then
i
is a Q-homomorphism of F into C, so it
is equal to certain
j
. Note that
i
=
i
iff
i
(F) R. Let r
1
be the number
of Q-homomorphisms of this type, say, after renumeration,
1
, . . . ,
r
1
. For every
i > r
1
we have
j
,=
j
, so we can form couples (
j
,
j
). Then n r
1
is an
even number 2r
2
, and r
1
+ 2r
2
= n.
Renumerate the
j
s so that
i+r
2
=
i
for r
1
+ 1 i r
1
+ r
2
. Dene the
canonical embedding of F by
: a (
1
(a), . . . ,
r
1
+r
2
(a)) R
r
1
C
r
2
, a F.
The eld F is isomorphic to its image (F) R
r
1
C
r
2
. The image (F) is called
the geometric image of F and it can be partially studied by geometric tools.
3.6.6. Minkowskis Bound Theorem. Let F be an algebraic number eld of degree
n with parameters r
1
, r
2
. Then every class of C
F
contains an ideal I such that its
norm N(I) satises the inequality
N(I) (4/)
r
2
n!
_
[d
F
[/n
n
where d
F
is the discriminant of F.
Proof. Use the geometric image of F and some geometric combinatorial considera-
tions. In particular, one uses Minkowskis Lattice Point Theorem:
2010/2011 33
Let L be a free Z-module of rank n in an n-dimensional Euclidean vector space
V over R (then L is called a complete lattice in V ). Denote by Vol (L) the volume
of the set
a
1
e
1
+ + a
n
e
n
: 0 a
i
1,
where e
1
, . . . , e
n
is a basis of L. Notice that Vol (L) does not depend on the choice
of basis.
Let X be a centrally symmetric convex subset of V . Suppose that Vol (X) >
2
n
Vol (L). Then X contains at least one nonzero point of L.
3.6.7. Examples. 1. Let F = Q(

5). Then r
1
= 2, r
2
= 0, n = 2, [d
F
[ = 5.
(4/)
r
2
n!
_
[d
F
[/n
n
= 2!

5/2
2
= 1.1...,
so N(I) = 1 and therefore I = O
F
. Thus, every ideal of O
F
is principal and
C
F
= 1.
2. Let F = Q(

5). Then r
1
= 0, r
2
= 1, n = 2, [d
F
[ = 20, (2/)
_
[20[ < 3.
Hence, similar to Example in 3.6.4 we only need to look at prime numbers 2 ( < 3 ) and
prime ideal divisors of the ideal (2) as potential candidates for non-principal ideals.
From 3.3.8 we know that (2) = (2, 1 +

5)
2
and 2 = N(2, 1 +

5). So the ideal

(2, 1 +

5) is maximal by 3.4.5.
The ideal (2, 1 +

5) is not principal: Indeed, if (2, 1 +

5) = aO
L
then
2 = N(2, 1 +

5) = N(aO
L
) = [N
L/Q
(a)[. If a = c + d

5 with c, d Z we
deduce that c
2
+ 5d
2
= 2, a contradiction.
We conclude that C
Q(

5)
is a cyclic group of order 2.
3. Let F = Q(

14). Then r
1
= 2, r
2
= 0, n = 2, [d
F
[ = 56 and (1/2)

56 =
3.7... < 4. So we only need to inspect prime ideal divisors of (2) and of (3).
Now 2 = (4+

14)(4

14), so (2) = (4+

14)(4

14). Since N(4

14) = 2,
3.4.5 implies that the principal ideals (4 +

14), (4

14) are prime.

14 is quadratic non-residue modulo 3, so by Theorem 3.5.9 we deduce that 3
remains prime in O
F
. Thus, every ideal of the ring of integers of Q(

14) is principal,
C
Q(

14)
= 1.
4. It is known that for negative square-free d the only quadratic elds Q(

d) with
class number 1 are the following:
Q(

1), Q(

2), Q(

3), Q(

7), Q(

11),
Q(

19), Q(

43), Q(

67), Q(

163).
For d > 0 there are many more quadratic elds with class number 1. Gauss
conjectured that there are innitely many such elds, but this is still unproved.
34 Alg number theory
3.6.8. Now we can state one of the greatest achivements of Kummer.
Kummers Theorem. Let p be an odd prime. Let F = Q(
p
) be the p th cyclotomic
eld.
If p doesnt divide [C
F
[,
or, equivalently, p does not divide numerators of (rational) Bernoulli numbers
B
2
, B
4
, . . . , B
p3
given by
t
e
t
1
=

i=0
B
i
i!
t
i
,
then the Fermat equation
X
p
+ Y
p
= Z
p
does not have positive integer solutions.
Among primes < 100 only 37, 59 and 67 dont satisfy the condition that p does
not divide [C
F
[, so Kummers theoremimplies that for any other prime number smaller
100 the Fermat equation does not have positive integer solutions.
3.7. Units of rings of algebraic numbers
3.7.1. Denition. A subgroup Y of R
n
is called discrete if for every bounded closed
subset Z of R
n
the intersection Y Z is nite.
Example: points of R
n
with integer coordinates form a discrete subgroup.
3.7.2. Proposition. Let Y be a discrete subgroup of R
n
. Then there are m linearly
independent over R vectors y
1
, . . . y
m
Y such that y
1
, . . . , y
m
is a basis of the
Z-module Y .
Proof. Let x
1
, . . . , x
m
be a set of linearly independent elements in Y over R with
the maximal m. Denote
L = x R
n
: x =
m

i=1
c
i
x
i
: 0 c
i
1.
The set L is bounded and closed, so L Y is nite. For y Y write y =

m
i=1
b
i
x
i
with b
i
R. Dene
z = y

[b
i
]x
i
=

(b
i
[b
i
])x
i
L Y.
2010/2011 35
Hence the group Y is generated by the nite set L Y and x
i
, and Y is nitely
generated as a Z-module.
Since the torsion of Y is trivial, the main theorem on the structure of nitely
generated modules over principal ideal domains implies the assertion of the proposition.
3.7.3. Dirichlets Unit Theorem. Let F be a number eld of degree n, r
1
+ 2r
2
= n.
Let O
F
be its ring of integers and U be the group of units of O
F
. Then U is the
direct product of a nite cyclic group T consisting of all roots of unity in F and a free
abelian group U
1
of rank r
1
+ r
2
1 :
U T U
1
T Z
r
1
+r
2
1
.
A basis of the free abelian group U
1
is called a fundamental system of units in O
F
.
Proof. Consider the canonical embedding of F into R
r
1
C
r
2
. Dene
f: O
F
0 R
r
1
+r
2
,
f(x) =
_
log [
1
(x)[, . . . , log [
r
1
(x)[, log([
r
1
+1
(x)[
2
), . . . , log([
r
1
+r
2
(x)[
2
)
_
.
The map f induces a homomorphism g: U R
r
1
+r
2
.
We now show that g(U) is a discrete group. Let u g
1
(Z) and Z be a bounded
set. Then there is c such that [
i
(u)[ c for all i. The coefcients of the characteristic
polynomial g
u
(X) =

n
i=1
(X
i
(u)) of u over F being functions of
i
(u) are
integers bounded by max(c
n
, nc
n1
, . . . ), so the number of different characteristic
polynomials of g
1
(Z) is nite, and so is g
1
(Z).
Every nite subgroup of the multiplicative group of a eld is cyclic by 1.2.4. Hence
the kernel of g, being the preimage of 0, is a cyclic nite group. On the other hand,
every root of unity belongs to the kernel of g, since mg(z) = g(z
m
) = g(1) = 0 implies
g(z) = 0 for the vector g(z). We conclude that the kernel of g consists of all roots of
unity T in F.
Since for u U the norm N
F/Q
(u) =

i
(u), as the product of units, is a unit
in Z, it is equal to 1. Then

[
i
(u)[ = 1 and log [
1
(u)[ + + log [
r
1
(u)[ +
log([
r
1
+1
(u)[
2
) + + log([
r
1
+r
2
(u)[
2
) = 0. We deduce that the image g(U) is
contained in the hyperplane H R
r
1
+r
2
dened by the equation y
1
+ + y
r
1
+r
2
= 0.
Since g
1
(Z) is nite for every bounded set Z, the intersection g(U) Z is nite.
Hence by 3.7.2 g(U) has a Z-basis y
i
consisting of m r
1
+ r
2
1 linearly
independent vectors over Z. Denote by U
1
the subgroup of U generated by z
i
such
that g(z
i
) = y
i
; it is a free abelian group, since there are no nontrivial relations among
y
i
. From the main theorem on group homomorphisms we deduce that U/T g(U)
and hence U = TU
1
. Since U
1
has no nontrivial torsion, T U
1
= 1. Then U as a
Z-module is the direct product of the free abelian group U
1
of rank m and the cyclic
group T of roots of unity.
36 Alg number theory
It remains to show that m = r
1
+ r
2
1, i.e. g(U) contains r
1
+ r
2
1 linearly
independent vectors. Put l = r
1
+ r
2
. As an application of Minkowskis geometric
method one can show that
for every integer k between 1 and l there is c > 0 such that for every non-zero
a O
F
0 with g(a) = (
1
, . . . ,
l
) there is a non-zero b = h
k
(a) O
F
0
such that
[N
F/Q
(b)[ c and g(b) = (
1
, . . . ,
l
) with
i
<
i
for i ,= k .
(for the proof see Marcus, Number Fields, p.144145)
Fix k. Start with a
1
= a and construct the sequence a
j
= h
k
(a
j1
) O
F
for
j 2. Since N(a
j
O
F
) = [N
F/Q
(a
j
)[ c, in the same way as in the proof of 3.6.4
we deduce that there are only nitely many distinct ideals a
j
O
F
. So a
j
O
F
= a
q
O
F
for some j < q l. Then u
k
= a
q
a
1
j
is a unit and satises the property: the i th
coordinate of g(u
k
) = f(a
q
) f(a
j
) = (
(k)
1
, . . . ,
(k)
l
) is negative for i ,= k. Then

(k)
k
is positive, since

i

(k)
i
= 0.
This way we get l units u
1
, . . . , u
l
. We claim that there are l 1 linearly
independent vectors among the images g(u
i
). To verify the claim it sufces to check
that the rst l 1 columns of the matrix (
(k)
i
) are linearly independent.
If there were not, then there would be a non-zero vector (t
1
, . . . , t
l1
) such that

l1
i=1
t
i

(k)
i
= 0 for all 1 k l. Without loss of generality one can assume that
there is i
0
between 1 and l 1 such that t
i
0
= 1 and t
i
1 for i ,= i
0
, 1 i l 1.
Then t
i
0

(i
0
)
i
0
=
(i
0
)
i
0
and for i ,= i
0
t
i

(i
0
)
i

(i
0
)
i
since t
i
1 and
(i
0
)
i
< 0. Now
we would get
0 =
l1

i=1
t
i

(i
0
)
i

l1

k=1

(i
0
)
i
>
l

i=1

(i
0
)
i
= 0,
a contradiction.
Thus, m = r
1
+ r
2
1.
3.7.4. Example. Let F = Q(

d) with a square free non-zero integer d.

If d > 0, then the group of roots of 1 in F is 1, since F R and there are
only two roots of unity in R.
Let O
F
be the ring of integers of F. We have n = 2 and r
1
= 2, r
2
= 0 if d > 0;
r
1
= 0, r
2
= 1 if d < 0. If d < 0, then
U(O
F
) = T
is a nite cyclic group consisting of all roots of unity in F. It has order 4 for d = 1, 6
for d = 3, and one can show it has order 2 for all other negative square free integers.
If d > 0, U(O
F
) is the direct product of 1) and the innite group generated by
a unit u (fundamental unit of O
F
):
U(O
F
) 1) u) = u
k
: k Z.
2010/2011 37
Here is an algorithm how to nd a fundamental unit if d , 1 mod 4 (there is a
similar algorithm for an arbitrary square free positive d ):
Let b be the minimal positive integer such that either db
2
1 or db
2
+1 is a square
of a positive integer, say, a. Then N
F/Q
(a + b

d) = a
2
db
2
= 1, so a + b

d > 1
is a unit of O
F
.
Let u
0
= e + f

d be a fundamental unit. Changing the sign of e, f if necessary,

we can assume that e, f are positive. Due to the denition of u
0
there is an integer k
such that a + b

d = u
k
0
. The sign is +, since the left hand side is positive; k > 0,
since u
0
1 and the left hand side is > 1. From a + b

d = (e + f

d)
k
we deduce
that if k > 1 then b = f+ some positive integer > f, a contradiction. Thus, k = 1
and a + b

d > 1 is a fundamental unit of O

F
.
For example, 1 +

2 is a fundamental unit of Q(

2) and 2 +

3 is a fundamental
unit of Q(

3).
3.7.5. Now suppose that d > 0, and for simplicity, d , 1 mod 4. We already know
that if an element u = a + b

d of O
F
is a unit, then its norm N
F/Q
(u) = a
2
db
2
is
1. On the other hand, if a
2
db
2
= 1, then u
1
= a b

d is in O
F
, so u is a
unit. Thus, u = a + b

d is a unit iff a
2
db
2
= 1.
Let u
0
= e + f

d be a fundamental unit.
From the previous we deduce that all integer solutions (a, b) of the equation
X
2
dY
2
= 1
satisfy a + b

d = (e + f

d)
m
for some integer m, which gives formulas for a and
b as functions of e, f, m.
38 Alg number theory
4. p-adic numbers
4.1.1. p-adic valuation and p-adic norm. Fix a prime p.
For a non-zero integer m let
k = v
p
(m)
be the maximal integer such that p
k
divides m, i.e. k is the power of p in the
factorization of m. Then v
p
(m
1
m
2
) = v
p
(m
1
) + v
p
(m
2
).
Extend v
p
to rational numbers putting v
p
(0) := and
v
p
(m/n) = v
p
(m) v
p
(n),
this does not depend on the choice of a fractional representation: if m/n = m

/n

then mn

= m

n, hence v
p
(m) + v
p
(n

) = v
p
(m

) + v
p
(n) and v
p
(m) v
p
(n) =
v
p
(m

) v
p
(n

).
Thus we get the p-adic valuation v
p
: Q Z +. For non-zero rational
numbers a = m/n, b = m

/n

we get
v
p
(ab) = v
p
(mm

/(nn

)) = v
p
(mm

) v
p
(nn

)
= v
p
(m) + v
p
(m

) v
p
(n) v
p
(n

)
= v
p
(m) v
p
(n) + v
p
(m

) v
p
(n

)
= v
p
(m/n) + v
p
(m

/n

)
= v
p
(a) + v
p
(b).
Thus v
p
is a homomorphism from Q

to Z.
4.1.2. p-adic norm. Dene the p-adic norm of a rational number by
[[
p
= p
v
p
()
, [0[
p
= 0.
Then
[[
p
= [[
p
[[
p
.
If = m/n with integer m, n relatively prime to p, then v
p
(m) = v
p
(n) = 0 and
[[
p
= 1. In particular, [ 1[
p
= [1[
p
= 1 and so [ [
p
= [[
p
for every rational .
4.1.3. Ultrametric inequality. For two integers m, n let k = min(v
p
(m), v
p
(n)), so
both m and n are divisible by p
k
. Hence m+ n is divisible by p
k
, thus
v
p
(m+ n) min(v
p
(m), v
p
(m)).
2010/2011 39
For two nonzero rational numbers = m/n, = m

/n

v
p
( + ) = v
p
(mn

+ m

n) v
p
(nn

)
min(v
p
(m) + v
p
(n

), v
p
(m

) + v
p
(n)) v
p
(n) v
p
(n

)
min(v
p
(m) v
p
(n), v
p
(m

) v
p
(n

))
= min(v
p
(), v
p
()).
Hence for all rational , we get
v
p
( + ) min(v
p
(), v
p
()).
This implies
[ + [
p
max([[
p
, [[
p
).
This inequality is called an ultrametric inequality.
In particular, since max([[
p
, [[
p
) [[
p
+ [[
p
, we obtain
[ + [
p
[[
p
+ [[
p
,
so [ [
p
is a metric ( p-adic metric) on the set of rational numbers Q and
d
p
(, ) = [ [
p
gives the p-adic distance between rational , .
4.1.4. All norms on Q. In general, for a eld F a norm [ [: F R
0
is a map
which sends 0 to 0, which is a homomorphism from F

to R

>0
and which satises
the triangle inequality: [ + [ [[ + [[. In particular,
[1[ = 1, 1 = [1[ = [(1)(1)[ = [ 1[
2
,
so [ 1[ = 1, and hence
[ a[ = [ 1[[a[ = [a[.
A norm is called nontivial if there is a nonzero a F such that [a[ , = 1.
In addition to p-adic norms on Q we get the usual absolute value on Q which we
will denote by [ [

.
A complete description of norms on Q is supplied by the following result.
Theorem (Ostrowski). A nontrivial norm [ [ on Q is either a power of the absolute
value [ [
c

with positive real c, or is a power of the p-adic norm [ [

c
p
for some prime
p with positive real c.
Proof. For an integer a > 1 and an integer b > 0 write
b = b
n
a
n
+ b
n1
a
n1
+ + b
0
with 0 b
i
< a, a
n
b. Then
40 Alg number theory
[b[ ([b
n
[ + [b
n1
[ + + [b
0
[) max(1, [a[
n
)
and
[b[ (log
a
b + 1) d max(1, [a[
log
a
b
),
with d = max([0[, [1[, . . . , [a 1[).
Substituting b
s
instead of b in the last inequality, we get
[b
s
[ (s log
a
b + 1) d max(1, [a[
s log
a
b
),
hence
[b[ (s log
a
b + 1)
1/s
d
1/s
max(1, [a[
log
a
b
).
When s + we deduce
[b[ max(1, [a[
log
a
b
).
There are two cases to consider.
(1) Suppose there is an integer b such that [b[ > 1. We can assume b is positive.
Then
1 < [b[ max(1, [a[
log
a
b
),
and so [a[ > 1, [b[ [a[
log
a
b
for every integer a > 1. Swapping a and b we get
[a[ [b[
log
b
a
, thus,
[a[ = [b[
log
b
a
for every integer a and hence for every rational a.
Choose c > 0 such that [b[ = [b[
c

then we obtain [a[ = [a[

c

for every rational a.

(2) Suppose that [a[ 1 for all integer a. Since [ [ is nontrivial, let a
0
be the
minimal positive integer such that [a
0
[ < 1. If a
0
= a
1
a
2
with positive integers
a
1
, a
2
, then [a
1
[ [a
2
[ < 1 and either a
1
= 1 or a
2
= 1. This means that a
0
= p
is a prime. If q / pZ, then pp
1
+ qq
1
= 1 with some integers p
1
, q
1
and hence
1 = [1[ [p[ [p
1
[ +[q[ [q
1
[ [p[ +[q[. Writing q
s
instead of q we get [q[
s
1[p[ > 0
and [q[ (1 [p[)
1/s
. The right hand side tends to 1 when s tends to innity. So we
obtain [q[ = 1 for every q prime to p. Therefore, [[ = [p[
v
p
()
, and [ [ is a power of
the p-adic norm.
4.1.5. Lemma (reciprocity law for all [ [
p
). For every nonzero rational

i prime or
[[
i
= 1.
Proof. Due to the multiplicative property of the norms and factorization of integers it
is sufcient to consider the case of = p a prime number, then [p[
p
= p
1
, [p[

= p
and [p[
i
= 1 for all other i.
2010/2011 41
4.2. The eld of p-adic numbers Q
p
4.2.1. The denition. Similarly to the denition of real numbers as the completion
of Q with respect to the absolute value [ [

dene Q
p
as the completion of Q
with respect to the p-adic norm [ [
p
. So Q
p
consists of equivalences classes of all
fundamental sequences (with respect to the p-adic norm) (a
n
) of rational numbers a
n
:
two fundamental sequences (a
n
), (b
n
) are equivalent if and only if [a
n
b
n
[
p
tends
to 0.
The eld Q
p
is called the eld of p-adic numbers and its elements are called p-adic
numbers.
4.2.2. p-adic series presentation of p-adic numbers. As an analogue of the decimal
presentation of real numbers every element of Q
p
has a series representation: it can
be written as an innite convergent (with respect to the p-adic norm) series

i=n
a
i
p
i
with coefcients a
i
0, 1, . . . , p 1 and a
n
,= 0.
4.2.3. The p-adic norm and p-adic distance. We have an extension of the p-adic
norm from Q to Q
p
by continuity: if Q
p
is the limit of a fundamental sequence
(a
n
) of rational numbers, then [[
p
:= lim[a
n
[
p
. Since two fundamental sequences
(a
n
), (b
n
) are equivalent if and only if [a
n
b
n
[
p
tends to 0, the p-adic norm of is
well dened.
If we use the series representation =

i=n
a
i
p
i
with coefcients a
i
0, 1, . . . , p
1 and a
n
,= 0, then [[
p
= p
n
.
The p-adic norm on Q
p
satises the ultrametric inequality: let = lima
n
, =
limb
n
, (a
n
), (b
n
) are fundamental sequences of rational numbers, then + =
lim(a
n
+ b
n
). Suppose that [[
p
[[
p
, then [a
n
[
p
[b
n
[
p
for all sufciently large
n, and so
[ + [
p
= lim[a
n
+ b
n
[
p
limmax([a
n
[
p
, [b
n
[
p
) = lim[b
n
[
p
= [[
p
= max([[
p
, [[
p
).
For , such that [[
p
< [[
p
we obtain = + where = . By the
ultrametric inequality [[
p
max([[
p
, [[
p
), so [[
p
[[
p
and by the ultrametric
inequality [[
p
max([[
p
, [ [
p
) = max([[
p
, [[
p
) = [[
p
. Thus if [[
p
< [[
p
then [ [
p
= [[
p
.
Using the p-adic distance d
p
we have shown that for every triangle with vertices in
0, , if the p-adic length of its side connecting 0 and is smaller than the p-adic
length of its side connecting 0 and then the p-adic length of the third side connecting
and equals to the former. Thus, in every triangle two sides are of the same p-adic
length!
42 Alg number theory
4.2.4. The ring of p-adic integers Z
p
. Dene the set Z
p
of p-adic integers as
those p-adic numbers whose p-adic norm does not exceed 1, i.e. whose p-adic series
representation has n
0
0. For two elements , Z
p
we get [[
p
0, [[
p

0. Hence Z
p
is a subring of Q
p
.
The units Z

p
of the ring Z
p
are those p-adic numbers u whose p-adic norm is 1.
Every nonzero p-adic number can be uniquely written as p
v
p
()
u with u Z

p
.
Thus
Q

p
p) Z

p
where p) is the innite cyclic group generated by p.
Let I be a non-zero ideal of Z
p
. Let n = minv
p
() : I. Then p
n
u
belongs to I for some unit u, and hence p
n
belongs to I, so p
n
Z
p
I p
n
Z
p
, i.e.
I = p
n
Z
p
. Thus Z
p
is a principal ideal domain and a Dedekind ring.
4.2.5. Note that Z
p
is the closed ball of radius 1 in the p-adic norm.
Let be its internal point, so [[
p
< 1. Then for every on the boundary of the
open ball, i.e. [[
p
= 1 we obtain, applying 4.2.3, we obtain [ [
p
= [[
p
= 1.
Thus, the p-adic distance from to every point on the boundary of the ball is 1, i.e.
every internal point of a p-adic ball is its centre!
5. On class eld theory
To describe some very basic things about it, we rst need to go through a very useful
notion of the projective limit of algebraic objects.
5.1.1. Projective limits of groups/rings. Let A
n
, n 1 be a set of groups/rings,
with group operation, in the case of groups, written additively. Suppose there are
group/ring homomorphisms
nm
: A
n
A
m
for all n m such that

nn
= id
A
n
,

nr
=
mr

nm
for all n m r.
The projective limit lim

A
n
of (A
n
,
nm
) is the set
(a
n
) : a
n
A
n
,
nm
(a
n
) = a
m
for all n m
with the group/ring operation(s) (a
n
) + (b
n
) = (a
n
+ b
n
) and (a
n
)(b
n
) = (a
n
b
n
)
For every m one has a group/ring homomorphism
n
: lim

A
n
A
m
, (a
n
) a
m
.
5.1.2. Examples.
1. If A
n
= A for all n and
nm
= id then lim

A
n
= A.
2. If A
n
= Z/p
n
Z and
nm
(a + p
n
Z) = a + p
m
Z then (a
n
) lim

Z/p
n
Z means
p
min(n,m)
[(a
n
a
m
) for all n, m.
2010/2011 43
The sequence (a
n
) as above is a fundamental sequence with respect to the p-adic
norm, and thus determines a p-adic number a = lima
n
Z
p
. For its description,
denote by r
m
the integer between 0 and p
m
1 such that r
m
a
m
mod p
m
. Then
r
m
a
n
mod p
m
for n m and r
n
r
m
mod p
m
for n m. Denote c
0
= r
0
and c
m
= (r
m
/r
m1
)p
m+1
, so c
m
0, 1, . . . , p1. Then a =

m0
c
m
p
m
=
limr
m
Z
p
.
We have a group and ring homomorphism
f: lim

Z/p
n
Z Z
p
, (a
n
) a = lima
n
Z
p
.
It is surjective: if a =

m0
c
m
p
m
then dene r
m
by the inverse procedure to the
above, then a is the image of (r
n
) lim

Z/p
n
; and its kernel is trivial, since a = 0
implies that for every k p
k
divides a
n
for all sufciently large n, and so p
k
divides
a
k
.
Thus,
lim

Z/p
n
Z Z
p
.
This can be used as another (algebraic) denition of the ring of p-adic integers.
In particular, we a surjective homomorphism Z
p
Z/p
n
Z whose kernel equals to
p
n
Z
p
.
Fromthe above we immediately deduce that if A
n
= (Z/p
n
Z)

and
nm
(a+p
n
Z) =
a + p
m
Z, (a, p) = 1, then similarly we have a homomorphism
f: lim

(Z/p
n
Z)

p
, (a
n
) limr
m
Z

p
(note that (r
m
, p) = 1 and hence limr
m
, pZ
p
). Thus, there is an isomorphism
lim

(Z/p
n
Z)

p
.
3. One can extend the denition of the projective limit to the case when the maps

nm
are dened for some specic pairs (n, m) and not necessarily all n m.
Let A
n
= Z/nZ and let
nm
: A
n
A
m
be dened only if m[n and then

nm
(a + nZ) = a + mZ. Dene, similarly to the above denition of the projective limit
the projective limit lim

A
n
.
By the Chinese remainder theorem
Z/nZ = Z/p
k
1
1
Z Z/p
k
r
r
Z
where n = p
k
1
1
. . . p
k
r
r
is the factorization of n. The maps
nm
induce the maps
already dened in 2 on Z/p
r
Z, and we deduce
lim

Z/nZ = lim

Z/2
r
Z lim

Z/3
r
Z Z
2
Z
3
=

Z
p
.
44 Alg number theory
The group lim

Z/nZ is denoted

Z and is called the procyclic group (topologically
it is generated by its unity 1). This group is uncountable. We have a surjective
homomorphism

Z Z/nZ whose kernel is n

Z.
Similarly we have

= lim

(Z/nZ)

= lim

(Z/2
r
Z)

lim

(Z/3
r
Z)

p
.
5.2.1. Innite Galois theory. As described in 1.3
Gal(F
q
m/F
q
) Z/mZ,
where q = p
n
and the isomorphism is given by
n
1 + mZ. The algebraic
closure F
a
q
of F
q
is the compositum of all F
q
m. From the point of view of innite
Galois theory and it is natural to dene the innite Galois group Gal(F
a
q
/F
q
) as the
projective limit lim

Gal(F
q
m/F
q
) with respect to the natural surjective homomorphisms
Gal(F
q
m/F
q
) Gal(F
q
r /F
q
), r[m. This corresponds to
mr
dened in Example 4
above.
Hence we get
Gal(F
a
q
/F
q
) lim

Z/nZ =

Z.
Similarly, using 2.4.3 for the maximal cyclotomic extension Q
cycl
, the composite
of all nite cyclotomic extensions Q(
m
) of Q, we have
Gal(Q
cycl
/Q) lim

(Z/nZ)

.
The main theorem of extended (to innite extensions) Galois theory (one has to
add a new notion of closed subgroup for an appropriate extension of the nite Galois
theory), can be stated as follows:
Let L/F be a (possibly innite) Galois extension, i.e. L is the compositum of split-
ting elds of separable polynomials over F. Denote G = Gal(L/F) = lim

Gal(E/F)
where E/F runs through all nite Galois subextensions in L/F. Call a subgroup H
of G closed if H = lim

Gal(E/K) where K runs through a subfamily of nite subex-

tensions in E/F, and the projective maps Gal(E

/K

) Gal(E

/K

) are induced
by Gal(E

/F) Gal(E

/F).
There is a one-to-one correspondence (H L
H
) between closed subgroups H of
G and elds M, F M L, the inverse map is given by M H = lim

Gal(E/K)
where K = E M. We have Gal(L/M) = H.
Normal closed subgroups H of G correspond to Galois extensions M/F and
Gal(M/F) G/H.
2010/2011 45
5.3.1. We have already seen the importance of cyclotomic elds in Kummers theorem
3.6.8.
Another very important property of cyclotomic elds is given by the following
theorem
Theorem (KroneckerWeber). Every nite abelian extension of Q is contained in
some cyclotomic eld Q(
n
). Therefore the maximal abelian extension Q
ab
of Q
coincides with the cyclotomic eld Q
cycl
which is the compositum of all cyclotomic
elds Q(
n
).
According to 2.4.3 the Galois group Gal(Q(
n
)/Q) is isomorphic to (Z/nZ)

. So
the innite group Gal(Q
ab
/Q) is isomorphic to the limit of (Z/nZ)

which by 5.1.2
coincides with the group of units of

Z = lim

Z/nZ.
The isomorphism
:

Gal(Q
ab
/Q)
can be described as follows: if a

is congruent to m modulo n via

Z/n

Z Z/nZ,
then (a)(
n
) =
m
n
.
Using 5.1.2 we have an isomorphism
:

Gal(Q
ab
/Q).
On the left hand side we have an object

Z

which is dened at the ground level of Q,

on the right hand side we have an object which incorporates information about all nite
abelian extensions of Q.
The restriction of the isomorphism to quadratic extensions of Q is related with the
Gauss quadratic reciprocity law, see below.
Abelian class eld theory generalizes the KroneckerWeber theoremfor an algebraic
number eld K to give a reciprocity homomorphismwhich relates an object (idele class
group) dened at level of K and the Galois group of the maximal abelian extension of
K over K.
5.3.2. Ideles. Recall (see 4.2.4) that Q

p
p) Z

p
, a (n, u) where n = v
p
(a)
and u = ap
n
,
v
p
is the p-adic valuation.
Denote Q

= R and include in the set of primes of Z. Form the so called

restricted product
I
Q
=

p
= (a

, a
2
, a
3
, . . . ) : a
p
Q

of R

= Q

, Q

2
, Q

3
, . . . such that almost all components a
p
are p-adic units.
Elements of I
Q
are called ideles over Q.
46 Alg number theory
Dene a homomorphism
f: I
Q
=

p
Q

p
,
(a

, a
2
, a
3
, . . . ) (a, a

a
1
, a
2
a
1
, a
3
a
1
, . . . )
where a = sgn(a

p
v
p
(a
p
)
Q

and sgn(a) is the sign of a.

It is easy to verify that f is an isomorphism.
5.3.3. Dene a homomorphism

Q
:

p
Gal(Q
ab
/Q)
by the following local-global formula:

Q
(a

, a
2
, a
3
, . . . ) =

Q
p
(a
p
).
Here the local reciprocity map
Q
p
is described as follows: if a
p
= p
n
u where
n = v
p
(a), then for a q
m
th primitive root of unity with prime q

Q
p
(a
p
)() =
_

p
n
, if p ,= q

u
1
, if p = q.
In particular, if p ,= q, then
Q
p
(p) sends to
p
, similar to the p th Frobenius
automorphism dened in 1.3. So one can say that the reciprocity map sends prime p to
the p th Frobenius automorphism.
For p = put

(a

)() =
sgn(a

)
.
The homomorphism
Q
is called the reciprocity map.
Theorem (class eld theory over Q).
1. Reciprocity Law: for a non-zero rational number a one has

Q
(a, a, a, . . . ) = 1.
2. For units u
p
Z

p
one has

Q
(1, u
2
, u
3
, . . . ) = (u
2
, u
3
, . . . )
1
.
3. Using f dene
g: I
Q
Q

p
,
(a, b, u
2
, u
3
, . . . ) (u
2
, u
3
, . . . ). Then

Q
()
1
= g().
4. The kernel of the reciprocity map
Q
equals to g
1
(1, 1, 1, . . . ) = the product
of the diagonal image of Q

in I
Q
and of the image of R

+
in I
Q
with respect to the
2010/2011 47
homomorphism (, 1, 1, . . . ). It induces an isomorphism
I
Q
/Q

+
Gal(Q
ab
/Q).
Proof. To verify the rst property, due to the multiplicativity of
Q
it is sufcient to
show that for a primitive q
m
th root of unity

Q
(p, p, . . . )() = for all positive prime numbers p

Q
(1, 1, . . . )() = .
From the denition of
Q
we deduce that

Q
l
(p)() =
_

_
, if l ,= q, l ,= p

p
, if l ,= q, l = p

p
1
, if l = q, l ,= p
, if l = q = p.
So (

Q
l
(p))() = for q ,= p and for q = p. Similarly one checks the second
assertion.
The second property is easy: due to multiplicativity it sufces to show that
(1, . . . , u
p
, 1, . . . )
1
=
Q
(1, . . . , u
p
, 1, . . . )
and this follows immediately from the denition of ,
Q
.
The third property follows from the denition of f and the rst and second proper-
ties. The fourth property follows from the third.
From this theorem one can deduce Gauss quadratic reciprocity law.
5.3.4. For an algebraic number eld F one can dene, in a similar way, the idele
group I
F
as a restricted product of the multiplicative groups F

P
of completions F
P
of F with respect to non-zero prime ideals P of the ring of integers of F, and of real
or complex completions of F with respect to real and complex imbeddings of F into
C.
Except the case of Q and imaginary quadratic elds one does not have an explicit
description of the maximal abelian extension as in KroneckerWeber theorem 4.2.3. So
one needs to directly dene a reciprocity map

F
: I
F
Gal(F
ab
/F)
and study its properties. This global reciprocity map is dened as the product of
composites of local reciprocity maps F

P
Gal(F
P
ab
/F
P
) and homomorphisms
Gal(F
P
ab
/F
P
) Gal(F
ab
/F).
The analog of the reciprocity law is that the kernel of
F
contains the image of
F

in I
F
.
48 Alg number theory
Part of class eld theory associates to every open subgroups N in I
F
/F

its class
eld L the unique nite abelian extension of F such that N
L/F
(I
L
)F

= N.
It also contains information on arithmetical properties of the behavior of prime
numbers in nite abelian extensions as a generalization of Theorem 3.5.9 and Gauss
quadratic reciprocity law.