Algebraic Number Theory Fall 2023

Jingren Chi
Morningside Center of Mathematics,
Academy of Mathematics and Systems Science, Chinese Academy of Sciences
[email protected]

January 9, 2024
 Abstract

The current document consists of lecture notes for the undergraduate course “Algebraic Number
Theory I” at Qiuzhen College, Tsinghua University in Fall 2023.
Introduction

We discuss some classical problems that motivate the development of algebraic number theory.

0.1 Fermat’s equation

An important historical motivation for algebraic number theory is the problem of solving Diophan-
tine equations. This basically means to find integral or rational solutions to polynomial equations.
One of the most famous such problem is to find integers X, Y, Z satisfying the equation

Xn + Y n = Zn

where n ≥ 2 is an integer. The case n = 2 is easy.

Theorem 0.1.1. Every integer solution of the equation X 2 +Y 2 = Z 2 has the form (up to permuting
X and Y ) 
X = ±(a2 − b2 )c,

Y = ±2abc, a, b, c ∈ Z
 2 2
Z = ±(a + b )c.


This can be proved in an elementary way. Here we outline an argument in the style of algebraic
number theory to have a first taste of the subject. We may assume that gcd(X, Y, Z) = 1. Then Z
must be odd and Z is coprime to X and Y . We factorize the equation in the form

(X + iY )(X − iY ) = Z 2 (0.1.1)

This leads us to consider the ring of Gaussian integers

Z[i] := {a + bi|a, b ∈ Z}.

It suffices to show that for any x, y, z ∈ Z satisfying (0.1.1), we have x + iy = uα2 for some α ∈ Z[i]
and unit u ∈ Z[i]× .
Exercise 0.1.1. Show that Z[i] is a Euclidean domain, deduce that it is a PID and hence a UFD.
Show that the unit group of Z[i] is Z[i]× = {±1, ±i}.
By the unique factorization, we just need to show that the order of any prime element π ∈ Z[i] in
the factorization of x + yi is even. Let e = ordπ (x + yi). If e is odd, by unique factorization, we must
have π|(x − yi), which implies that π|2x and π|z. Since z is odd, we must have π|x. This implies
that N π := ππ̄ is a common divisor of x, y, z and contradicts the assumption that gcd(x, y, z) = 1.
This settles the case n = 2. Regarding the general n ≥ 3 case, we have the famous
Fermat’s Last Theorem. For any integer n ≥ 3, any integer solution of X n + Y n = Z n satisfies
XY Z = 0.
Some special cases are elementary.

1
Theorem 0.1.2. The only integer solutions of the equation X 4 + Y 4 = Z 2 are the trivial ones, i.e.
the ones with XY Z = 0. In particular, Fermat’s Last Theorem is true for n = 4.

Exercise 0.1.2. Use Theorem 0.1.1 and Fermat’s method of infinite descent to prove this.
Corollary 0.1.3. Fermat’s Last Theorem is true if and only if for any prime p > 2, all integer
solutions of X p + Y p = Z p satisfies XY Z = 0.
Our goal will be to learn various tools developed towards Kummer’s solution of many special
cases of Fermat’s Last Theorem.
Theorem 0.1.4 (Kummer). Let p > 2 be a regular prime, then Fermat’s Last Theorem is true for
p, i.e. any integer solution of X p + Y p = Z p satisfies XY Z = 0.
The notion of regular prime is defined as follows. Let Bn ∈ Q be the Bernoulli numbers defined
by the Taylor series expansion
∞
t X Bn n
= t . (0.1.2)
et − 1 n=0 n!

It is easy to see that B1 = − 21 , B2k+1 = 0 for k ≥ 1. The first few even Bernolli numbers are

1 1 1 1
B0 = 1, B2 = , B4 = − , B6 = , B8 = − ,
6 30 42 30
5 691 7 3617
B10 = , B12 = − , B14 = , B16 = −
66 2730 6 510
Definition 0.1.1. A prime number p > 2 is regular if it does not divide the numerators of the
Bernoulli numbers B2 , B4 , . . . , Bp−3 .
This definition seems bizarre at first. Later we will give an equivalent definition of regular primes
in terms of ideal class groups of cyclotomic fields, from which we will see their relevance to Kummer’s
attempt to solve Fermat’s Last Theorem. Just as n = 2 case, Kummer’s idea is to factorize the
equation X p + Y p = Z p as
p−1
Y
(X + ζpi Y ) = Z p
i=0
2πi
where ζp = e p is a primitive p-th root of unit. One is then led to consider the ring
(p−1 )
X
Z[ζp ] := ai ζpi |a0 , . . . , ap−1 ∈ Z .
i=0

If Z[ζp ] is a UFD, then Kummer is able to finish the proof. Unfortunately, this is not true in general
but Kummer’s argument still work under the weaker hypothesis that p is a regular prime.

0.2 Representing prime numbers by quadratic forms

To finish the introduction, let us mention another well-known and classical result of Fermat related
to the ring Z[i].
Theorem 0.2.1 (Fermat). An odd prime number p > 0 can be written as p = x2 + y 2 for some
x, y ∈ Z if and only if p ≡ 1 mod 4.

2
Proof. First we show that p = x2 + y 2 for some x, y ∈ Z if and only if p is not prime in Z[i]. If
p = x2 + y 2 = (x + iy)(x − iy) then we have x, y ̸= 0 so that x + iy and x − iy are non-units and
hence p is not prime in Z[i]. Conversely if p is not prime in Z[i], then since Z[i] is a UFD we can
factorize p = αβ for some α, β ∈ Z[i] that are not units.1 Write α = x + yi for x, y ∈ Z. Then from
the equality p2 = N (p) = N (α)N (β) we deduce that p = N (α) = x2 + y 2 .
From the isomorphism Z[i]/pZ[i] = Fp [X]/(X 2 + 1) we see that pZ[i] is not a prime ideal if and
only if −1 is a square in Fp . Finally let ϵ be a generator of the (cyclic) multiplicative group F×
p.
p−1 p−1
Then we see that −1 = ϵ 2 is a square iff 2 is even and we are done.
Many attempts have been made to generalize this theorem and solve the problem of representing
integers by a binary quadratic form ax2 + bxy + cy 2 where a, b, c ∈ Z. This is another important
driving force of the development of algebraic number theory.
Fix a constant integer n ∈ Z. We would like to have a criterion for which prime number p can
be written as p = x2 + ny 2 for some integers x, y ∈ Z. This lead us to study the ring
√ √
Z[ −n] = {a + b −n, a, b ∈ Z}
√
However, in general the ring Z[ −n] is not a unique factorization domain.
√
Example 0.2.1. When n = 4 we have Z[ −4] = {a+ 2bi, a, b ∈ Z}. We have the following different
factorizations
√ of 8 into irreducible elements (2 + 2i)(2 − 2i) = 23 . Note that 2 + 2i is irreducible in
Z[ −4] but is not prime since
√ Z[X]
Z[ −4]/(2 + 2i) ∼
= ∼
= Z/8Z
(X 2
+ 4, 2 + X)
√
is not an integral domain. The problem here is that √ 1+i ∈ / Z[ −4] but we should add it in to have
unique factorization property. In other words, Z[ −4] is not integrally closed and we should study
its integral closure which is Z[i].
√
Example 0.2.2. When n = 5 one can show that√Z[ −5] is√integrallly closed. But unique factor-
ization still fails as one can see from 2 · 3 = (1 + −5)(1 − −5). The problem
√ with this example
is more serious and is related to the fact that the ideal class group of Z[ −5] is nontrivial.

0.3 Notations and conventions

The basic objects we study in algebraic number theory are number fields, which are defined as finite
extensions of Q. In fact we will more often work with ring of integers in number fields (for example
the ring of Gaussian integers Z[i] in Q(i)) that we introduce in the next section.
We will assume standard facts from commutative algebra. Our convention is that all rings are
commutative and contain multiplicative unit 1.

1 This is the place where we use that Z[i] is a UFD. It might be false for a general integral domain. For example,
√
the element 2 + 2i in Z[ −4] is irreducible but not prime. See Example 0.2.1.

3
Chapter 1

Dedekind domain

1.1 Ring of integers

1.1.1 Integrality
Integrality is a kind of finiteness condition that generalize the notion of algebraicity in the context
of field extensions.
Definition 1.1.1. Let A be a ring and B an A-algebra. An element b ∈ B is integral over A if it
satisfies a polynomial equation

bn + an−1 bn−1 + · · · + a1 b + a0 = 01

where a0 , . . . , an ∈ A.
The integral closure of A in B, sometimes denoted A, is the set of all elements in B that are
integral over A. We say B is integral over A if A = B, i.e. all elements in B are integral over A. We
say A is integrally closed in B if A = A. An integral domain A is integrally closed if it is integrally
closed in its fraction field.
Lemma 1.1.1. Let A ⊂ B be ring extension. For an element b ∈ B, the following are equivalent:
1. b is integral over A;
2. A[b] is a finitely generated A-module;
3. There exists a finitely generated A-submodule M ⊂ B such that bM ⊂ M and for any β ∈ B,
βM = 0 implies β = 0.
Remark 1.1.2. The last condition is satisfied if either 1 ∈ M or B is an integral domain.
Proof. (3)⇒(1): Let m1 , . . . , mn be a set of generators of the A-module
Pn M . By assumption, there
exists n × n matrix U = (uij ) with entries in A such that bmi = j=1 uij mj . Let f (T ) ∈ A[T ]
be the characteristic polynomial of U . Then by Caley-Hamilton theorem f (b)M = 0. Thus by
assumption f (b) = 0 which implies that b is integral over A.
Corollary 1.1.3. A is a subring of B.
Proof. For any b1 , b2 ∈ A, the algebra A[b1 , b2 ] is a finite A-module that is stable under multiplication
by b1 + b2 and b1 b2 . By Lemma 1.1.1 we get that b1 + b2 ∈ A and b1 b2 ∈ A. Thus A is a subring of
B.
1 Here we are abusing notation. The a ’s in this equation should be understood as their images in B under the
i
natural homomorphism A → B.

4
Corollary 1.1.4. Let A ⊂ B ⊂ C be ring extensions. If B is integral over A and C is integral over
B, then C is integral over A.

Proof. For any x ∈ C, by assumption there exists b1 , . . . , bn ∈ B such that

xn + b1 xn−1 + · · · bn = 0.

Therefore x is integral over R := A[b1 , . . . , bn ] so that R[x] is a finite R-module. By assumption

each bi is integral over A and hence we deduce that R[x] is a finitely generated A-module. It is clear
stable under multiplication by x. By Lemma 1.1.1(3) we deduce that x is integral over A.
Example 1.1.1. Let K be a number field. The ring of integers of K, usually denoted by OK , is
the integral closure of Z in K. Then OK is an integrally closed integral domain. To see this, let
OK be the integral closure of OK in K. Then by Corollary 1.1.4 OK is integral over Z and thus
OK = OK . We will simply refer to the ring of integers OK as number rings.

In general it might be hard to determine OK explicitly. As a first step we introduce a hands-

on criterion to determine whether a specific element in K is integral in terms of its characteristic
polynomial.

1.1.2 Trace and norm

Let L/F be a finite field extension. For any α ∈ L, let µα be the K-linear endomorphism of L
defined by multiplication by α. In other words, µα (x) = α · x for any x ∈ L. Let fL/F,α (T ) ∈ K[T ]
(resp. mL/F,α (T ) ∈ F [T ]) be the characteristic (resp. minimal) polynomial of µα .
Define the trace and the norm

TrL/F (α) := Tr(µα : L → L), NL/F (α) := det(µα : L → L)

Proposition 1.1.5. Suppose moreover that the extension L/F is separable. Let C be an algebraic
closed field containing F (for example we could take C = F , the algebraic closure of F ). Then we
have Y
fL/F,α (T ) = (T − τ (α)) = mL/F,α (T )[L:F (α)]
τ ∈HomF (L,C)

Proof. Let n = [L : F ] and lable the field embeddings as

HomF (L, C) = {τ1 , . . . , τn }.

To compute the invariants of µα we can base change to C and diagonalize µα . More precisely we
consider the isomorphism of C-algebras L ⊗F C ∼
= C n defined by

α ⊗ β 7→ (τ1 (α)β, . . . , τn (α)β)

for any α ∈ L and β ∈ C. Then the matrix for µα on C n becomes the diagonal matrix

diag(τ1 (α), . . . , τn (α))

in which there are only [F (α) : F ] distinct values and each occurs [L : F (α)] times. From this we
deduce the desired identities.
In particular we get
X Y
TrL/F (α) = τ (α), NL/F (α) = τ (α)
τ ∈HomF (L,C) τ ∈HomF (L,C)

5
Proposition 1.1.6. Let K be a number field and OK its ring of integers. For any α ∈ K, the
following are equivalent:

1. x ∈ OK
2. mK/Q,α (T ) ∈ Z[T ]
3. fK/Q,α (T ) ∈ Z[T ]
Proof. (1)⇒(2) follows from Gauss’s lemma on primitive polynomials. (2)⇒(3) follows from Propo-
sition 1.1.5. (3)⇒(1) follows from the Caley-Hamilton theorem.
As an application we can determine the ring of integers of quadratic fields.
√
Example 1.1.2. Any quadratic extension K of Q has the form K = Q( D) where D is a square
free integer and D ̸= 0, 1. Then we have
( √
Z[ D], D ≡ 2, 3 mod 4
OK = √
1+ D
Z[ 2 ], D ≡ 1 mod 4

1.1.3 Finiteness conditions

Lemma-Definition 1.1.7. An A-module is noetherian if any of the following equivalent conditions
are satisfied:

1. Any A-submodule of M is finitely generated;

2. M satisfies the ascending chain condition, i.e. for any sequence of submodules N1 ⊂ N2 ⊂ · · ·
of M we have Nn = Nn+1 for all n sufficiently large;
3. Any nonempty set S of submodules of M has a maximal element. (An element N ∈ S is
maximal if for any N ′ ∈ S with N ⊂ N ′ we have N = N ′ . In general maximal elements are
not unique)
Proof. (1)⇒(2): Let N1 ⊂ N2 ⊂ · · · be a chain of submodules. Let N = ∪∞ n=1 Ni . Then N is finitely
generated by (1) so that there exists an index m such that Nm contains all the generators of N .
Thus Nn = Nn+1 = N for all n ≥ m.
(2)⇒(3): Suppose there is a nonempty set S of submodules in M that does not have any maximal
element. Then we can find a chain of submodules in S that does not stabilize and this contradicts
(2).
(3)⇒(1): Given a nonzero A-submodule N ⊂ M . Let S be the set of finitely generated A-
submodules of N . Then S is nonempty since N ̸= 0. Let N ′ be a maximal element of S. If N ′ ̸= N
we can find an element x ∈ N \ N ′ . Then N ′′ := N ′ + Ax is a finitely generated A-submodule of N
so that N ′′ ∈ S. On the other hand, since x ∈
/ N ′ we N ′ ⊊ N ′′ and this contradicts the maximality
′ ′
of N . Therefore N = N is finitely generated.
Standard results in commutative algebra gaurantees that taking integral closure in finite field
extensions preserves noetherian property. In our setting we can give a direct proof of a more general
result.

Theorem 1.1.8. Let K be a number field and let R ⊂ K be a subring. Then R is noetherian.
Proof. After replacing K by Frac(R), which is still a number field, we may assume that K = Frac(R).
Let d := [K : Q] be the Q-dimension of K.
First we show that for any prime number p, R/pR is an Fp -vector space of dimension ≤ d. For any
element x1 , . . . , xd+1 ∈ R/pR, lift them to x̃1 , . . . , x̃d+1 ∈ R. Then the x̃i ’s are linearly dependent

6
 Pd+1
in K so that we can find ai ∈ Z with i=1 ai x̃i = 0 and gcd(a1 , . . . , ad+1 ) = 1 . The image of ai in
Fp are not all zero and gives a linear relation among xi ’s. This shows that dimFp R/pR ≤ d.
For any integer r ≥ 1, pr R/pr+1 R is a cyclic R/pR-module generated by pr and hence is a finite
dimensional Fp -vector space. Then for any integer m ≥ 1, R/pm R has a composition series whose
successive quotients are finite dimensional Fp vector spaces. Hence R/pm R is a finite abelian group.
By the Chinese Remainder Theorem, we conclude that R/nR is a finite abelian group for any integer
n.
For any ideal 0 ̸= I ⊂ R, we have I ⊗Z Q = K so that I ∩ Z ̸= 0. Choose 0 ̸= n ∈ I ∩ Z. Then
I/nR is a subgroup of the finite abelian group R/nR, hence is also finite. This implies that I is a
finitely generated R module. In fact, the lift of any set of generators of I/nR to I, together with n
generate the R-module I. Therefore R is noetherian.
Remark 1.1.9. This result is a special case of Krull-Akizuki theorem. In particular, the ring of
integers OK and any of its localizations are noetherian. In fact, we will see later that any ideal of
OK is a finitely generated free abelian group.2
On the other hand, non-noetherian rings occur quite often in modern number theory. For
example, OQ , the integral closure of Z in Q, is not noetherian. (For any integer n > 1, the chain of
1 1
ideals (n) ⊂ (n 2 ) ⊂ (n 4 ) · · · do not stabilize). Other typical examples are perfectoid rings.
The proof of the following result is a nice illustration of the usage of noetherian condition.
Lemma 1.1.10. Let R be a noetherian ring and a ⊂ R a nonzero ideal. Then there exists nonzero
prime ideals p1 , . . . , pr of R such that p1 · · · pr ⊂ a
Proof. Let S be the set of nonzero ideals of R that do not contain any products of nonzero prime
ideals and suppose S is nonempty. Since R is noetherian, the set S has a maximal element I. Then
I is not a prime ideal and hence we can find elements a, b ∈ R such that ab ∈ I but a ∈/ I, b ∈
/ I.
Let I1 := I + (a) and I2 := I + (b). Then we have the following inclusions

I ⊊ I1 , I ⊊ I2 , I1 I2 ⊂ I.

By maximality of I we have I1 , I2 ∈
/ S so that they both contain products of nonzero prime ideals.
Then I ⊃ I1 I2 also contains products of nonzero prime ideals and we get a contradiction. Therefore
S is empty which means that any ideal contains a product of nonzero prime ideals.
We will see later that in number rings OK any nonzero ideal is equal to a product of prime
ideals, in an essentially unique way. This fact holds in more general rings sharing some common
features with number rings.
Definition 1.1.2. A Dedekind domain is a noetherian integrally closed integral domain in which
any nonzero prime ideal is maximal.
Basically a Dedekind domain should be thought as a smooth curve. Let us give some common
examples of Dedekind domains.
Lemma 1.1.11. Let R be a UFD. Then R is integrally closed.
a
Proof. Suppose α = b ∈ K = Frac(R) is integral over R where a, b ∈ R are coprime. Then α
satisfies an equation
αn + a1 αn−1 + . . . + an = 0
Then we deduce that
an = b(−a1 an−1 − · · · − an bn−1 ).
Since R is a UFD and a, b are coprime we must have b ∈ R× and hence α = ab−1 ∈ R.
2 this may not be true for other subrings of K, e.g. localizations of O : The ring Z[ 1 ] is noetherian but it is not
K 2
a finitely generated Z-module.

7
Corollary 1.1.12. Let R be a PID (principal ideal domain). Then R is a Dedekind domain.
Proof. Any PID is noetherian and also a UFD. Then by Lemma 1.1.11 R is integrally closed. Let
p ⊂ R be a nonzero prime ideal. Let a ⊂ R be an ideal with p ⊂ a. Suppose p = (x) and a = (a).
Then we can write x = ab for some b ∈ R. If p ̸= a, then a ∈
/ p and we must have b ∈ p so that
b = xc for some c ∈ R. Then we deduce that ac = 1. This means that a ∈ R× so that a = R. Thus
p is a maximal ideal.
Example 1.1.3. For any field k, the polynomial ring k[X1 , . . . , Xn ] is noetherian, integrally closed
(in fact a UFD). But it is Dedekind if and only if n = 1.
Proposition 1.1.13. The ring of integers OK of a number field K is a Dedekind domain.
Proof. It remains to check the last condition in the definition. Let p ⊂ OK be a nonzero prime ideal.
Then p ∩ Z is nonzero since any nonzero α ∈ p satisfies an equation αn + a1 αn−1 + . . . + an = 0
with ai ∈ Z and 0 ̸= an ∈ p ∩ Z. Hence p ∩ Z = pZ for a prime number p. Thus OK /p is an integral
domain that is an Fp -algebra. Since OK is integral over Z we get that OK /p is integral over Fp .
Then we conclude that p is a maximal ideal by the following Lemma.
Lemma 1.1.14. Let k be a field and let R be a k-algebra. Suppose that R is an integral domain
and is integral over k. Then R is a field.

Proof. Let x ∈ R be a nonzero element. By assumption x is integral over k. Since R is an

integral domain, the minimal polynomial of x over k has nonzero constant term3 . Hence there
exists a1 , . . . , an ∈ k such that an ̸= 0 and

xn + a1 xn−1 + · · · + an = 0

Then we deduce that x(xn−1 + a1 xn−2 + . . . + an−1 ) = −an ̸= 0 which implies that x is invertible.
Therefore R is a field.

1.2 Factorization
1.2.1 Failure of unique factorization
A central issue in number theory in the 19th √ century is the failure of unique factorization
√ in a
general number
√ ring.√ For example if K = Q( −5) with ring of integers O K = Z[ −5]. Then
2 · 3 = (1 + −5)(1 − −5) are two different factorizations of 6 into irreducible elements. Kummer’s
idea to remedy this failure is to try to factorize further in a larger ring of “ideal numbers”
√ √
2 = p21 , 3 = p2 p3 , 1 + −5 = p1 p2 , 1 − −5 = p1 p3

where p1 , p2 , p3 are what Kummer called “ideal numbers” in a larger ring. Then one hope to obtain
essentially unique factorization
√ √
2 · 3 = p21 · p2 p3 = (1 + −5)(1 − −5).

One is then left with the hard problem of determining the mysterious ring of “ideal numbers”.
3 This might be false if R is not an integral domain. For example, in the ring R = k[x]/x2 the minimal polynomial

of x is x2 .

8
1.2.2 Unique factorization of ideals
Dedekind’s observation is that instead of knowing the hypothetical “ideal numbers” p1 , p2 , p3 , it
would be enough to know which elements in OK should be divisible by these “ideal numbers”. This
leads to the notion of ideals in modern algebra. Just as elements in a number ring, one can add or
multiply ideals. After the conceptual leap from elements to ideals, one obtain unique factorization.
Theorem 1.2.1. Every nonzero ideal a in a Dedekind domain R can be written as a product of
maximal ideals a = p1 · · · pn unique up to permutation.
We start with some preparations.
Lemma-Definition 1.2.2. Let R be a noetherian integral domain with fraction field K = Frac(R).
A fractional ideal of R is a nonzero R-submodule 0 ̸= M ⊂ K that satisfies either of the following
equivalent conditions:
1. aM ⊂ R for some nonzero a ∈ R.
2. M is a finitely generated R-module.
Proof. (1)⇒(2): Since R is noetherian, aM ⊂ R is a finitely generated R-module and hence M =
a−1 (aM ) is also finitely generated.
(2)⇒(1): Suppose M is a finitely generated R-module. Let x1 , . . . , xn ∈ M be a set of generators.
For each 1 ≤ i ≤ n, write xi = ai b−1
i with ai , bi ∈ R and bi ̸= 0. Then (b1 · · · bn )M ⊂ R.
Example 1.2.1. For any nonzero α ∈ K × , (α) = αR is a fractional ideal. Such fractional ideals
are called principal.
Lemma-Definition 1.2.3. Let R be a noetherian integral domain. For any fractional ideal a of R,
define
a−1 := {x ∈ K = Frac(R), xa ⊂ R}.
Then a−1 is also a fractional ideal of R.
Proof. From the definition we see immediately that a−1 is an R-submodule of K. Take any nonzero
element a ∈ a we get that a−1 ⊂ a−1 R. It remains to show that a−1 ̸= 0. Let {ai b−1 i , 1 ≤ i ≤ r} be
a set of generators of a where ai , bi ∈ R and bi ̸= 0. For any x ∈ K we have x ∈ a−1 if and only if
xai b−1
i ∈ R for all 1 ≤ i ≤ r. In particular we have 0 ̸= b1 · · · br ∈ a−1 so that a−1 ̸= 0.

Example 1.2.2. For any positive integer n > 1, n1 Z is a fractional ideal of Z but Z[ n1 ] and Q are
not fractional ideals of Z since they are not finitely generated Z-module.
Lemma 1.2.4. Let p be a nonzero prime ideal of a Dedekind domain R. Then p · p−1 = R and for
any nonzero ideal a ⊂ R, we have a ⊂ ap−1 and a ̸= ap−1 .

Proof. By definition we have pp−1 ⊂ R ⊂ p−1 .

We first claim that p−1 ̸= R. Take a nonzero element a ∈ p. Then by Lemma 1.1.10 there exists
nonzero prime ideals p1 , . . . , pr such that (a) ⊃ p1 · · · pr . We assume that r is smallest possible.
Since p1 · · · pr ⊂ (a) ⊂ p and p is prime, there exists 1 ≤ i ≤ r such that pi ⊂ p. After renumbering,
we may assume that i = 1. Then p1 = p since R is Dedekind. By the minimality of r, there exists
/ (a). Thus a−1 b ∈
b ∈ p2 · · · pr with b ∈ / R. On the other hand, we have bp ⊂ p1 · · · pr ⊂ (a) so that
a b ∈ p . This shows that p−1 ̸= R.
−1 −1

For any nonzero ideal a ⊂ R, we have a ⊂ ap−1 . If a = ap−1 , then since R is integrally closed we
get p−1 ⊂ R by Lemma 1.1.1. But this contradicts to what we have just proved. Thus a ̸= ap−1 .
In particular, we have p ⊊ pp−1 . Since R is Dedekind, the ideal p is maximal and we must have
pp−1 = R.

9
Proof of Theorem 1.2.1. Existence. Suppose the set S of nonzero proper ideals not having prime
factorization is nonempty. Then S has a maximal element a since R is noetherian. Then a is not
prime and hence a ⊊ p for some prime ideal p. By Lemma 1.2.4 we get

a ⊊ ap−1 ⊂ pp−1 = R

and hence ap−1 ∈ / S. Then ap−1 = p1 · · · pr is a product of prime ideals and by Lemma 1.2.4 a is
also a product of prime ideals, contradicting the assumption that a ∈ S.
Uniqueness. Suppose that p1 · · · pr = q1 · · · qs where pi , qj are nonzero prime ideals. From the
inclusion q1 · · · qs ⊂ p1 we deduce that there exists j such that qj ⊂ p1 . After permuting indices we
may assume j = 1 so that q1 ⊂ p1 . Since R is Dedekind, q1 is a maximal ideal and thus q1 = p1 .
Multiplying by p−1 1 and using Lemma 1.2.4 we deduce that p2 · · · pr = q2 · · · qs . Then we proceed
inductively to deduce that r = s and pi = qi for all 1 ≤ i ≤ r after permuting indices if necessary.

1.2.3 Ideal class group

Theorem 1.2.5. The set of fractional ideals of a Dedekind domain R form an abelian group under
products of ideals. The unit element is R and the inverse of a fractional ideal a is a−1 defined as
above.
Proof. By Lemma 1.2.4, any nonzero prime ideal p is invertible and its inverse is p−1 . Combined
with Theorem 1.2.1 we see that any nonzero ideal a ⊂ R is invertible. Let b be its inverse so that
ab = R. (If a = p1 · · · pr is the prime factorization, then b = p−1 −1
1 · · · pr ). It remains to show that

b = a−1 := {x ∈ K, xa ⊂ R}.

The inclusion b ⊂ a−1 is clear and it remains to show a−1 ⊂ b. For any x ∈ a−1 , we have xa ⊂ R,
which implies that (x) = xab ⊂ b.
Denote by JR the group of fractional ideals and PR ⊂ JR the subgroup of principal fractional
ideals.
Definition 1.2.1. The ideal class group of R is ClR := JR /PR . For a number field K, we denote
ClK := ClOK and call it the ideal class group of K.
By definition We have an exact sequence

1 → R× → K × → JR → ClR → 1.

The following result shows that the ideal class group precisely measures the failure of unique fac-
torization of elements.
Proposition 1.2.6. For a Dedekind domain R, the following are equivalent
1. ClR = {1};
2. R is a PID;
3. R is a UFD;
4. Every prime ideal of R is principal.
Proof. (1)⇒(2) by definition. (2)⇒(3) is a standard result. (4)⇒(1) by Theorem 1.2.1.
(3)⇒(4): Let p ⊂ R be a nonzero prime ideal. Take a nonzero x ∈ p. Then we can write
x = a1 · · · ar where a1 , . . . , ar are prime elements. Then there exists 1 ≤ i ≤ r such that ai ∈ p so
that (ai ) ⊂ p. Since R is Dedekind and (ai ) is a nonzero prime ideal, it is maximal and therefore
p = (ai ) is principal.

10
Remark 1.2.7. Compared to a general Dedekind domain, the number rings OK are “small” in
many aspects. By this we mean the following fundamental finiteness properties:
• OK and all its fractional ideals are finitely generated abelian group.
• Finiteness of ideal class groups: ClK is finite.
×
• Dirichlet’s unit theorem: The unit group OK is a finitely generated abelian group whose
rank is precisely determined.
These are the key results in the first part of our course. We start by proving finiteness of OK by
utilizing trace pairing.

1.3 Duality of lattices in number fields

1.3.1 Trace pairing
Let L/F be a finite field extension.
Definition 1.3.1. The trace pairing on L is the symmetric F -bilinear form L × L → F defined by

(α, β) 7→ TrL/K (αβ)

for all α, β ∈ L.
In general, a symmetric bilinear form (·, ·) on a finite dimensional vector space V over F induces
a symmetric bilinear form on the exterior powers ∧k V by the formula

(v1 ∧ · · · ∧ vk , w1 ∧ · · · ∧ wk ) = det((vi , wj ))

where we are taking the determinant of a k ×k matrix whose (i, j)-th entry is as displayed. The form
(·, ·) is nondegenerate if and only if the induced form on the top exterior power is nondegenerate (this
boils down to the fact that a square matrix is invertible if and only if its determinant is nonzero).
Because of this we are especially interested in the top exterior power in the current setting of L
equipped with the trace pairing.
Definition 1.3.2. Let n = [L : F ]. The discriminant of n-elements x1 , . . . , xn ∈ L is defined by

d(x1 , . . . , xn ) := (x1 ∧ · · · ∧ xn , x1 ∧ · · · ∧ xn ) = det(TrL/K (xi xj ))

For convenience we also define the discriminant of a single element α ∈ L to be

d(α) := d(1, α, . . . , αn−1 ).

By the general remark above, the trace pairing is nondegenerate if and only if the discriminant
is not identically zero and in this case we have moreover:
• For any n = [L : F ] elements x1 , . . . , xn ∈ L, the discriminant d(x1 , . . . , xn ) ̸= 0 if and only if
x1 , . . . , xn form an F basis of L.
• For any α ∈ L, d(α) ̸= 0 if and only if L = F (α) (i.e. α is a primitive element for the extension
L/F .)
Example 1.3.1. Suppose L/F is separable. We do some explicit computations of the discriminant.
Let F be a separable closure of F . Let HomF (L, F ) = {τ1 , . . . , τn }. Then for any n-tuple of
elements ω1 , . . . , ωn in L we get

(TrL/K (ωi ωj ))1≤i,j≤n = t ΩΩ

11
where Ω is the n × n matrix whose (i, j)-th entry is τi (ωj ).
Take a primitive element θ ∈ L so that L = F (θ) = F [T ]/f (T ) where f (T ) is the minimal
polynomial of θ. Let θi := τi (θ) ∈ F . Take ωi = θi−1 . Then the matrix Ω becomes the Vandermonde
matrix and we get that
Y
d(1, θ, . . . , θn−1 ) = det(TrL/K (θi+j−2 )) = (det Ω)2 = (θi − θj )2 = disc(f )
1≤i<j≤n

This is the usual discriminant of the polynomial f (T ), hence the terminology in general. We also
mention the identity
n(n−1)
disc(f ) = (−1) 2 NL/F (f ′ (θ))
which is clear from definitions.
In particular, since f is a separable polynomial we have disc(f ) ̸= 0 and therefore we have
proved:
Proposition 1.3.1. If L/F is a finite separable extension, then the trace pairing on L is nonde-
generate.

1.3.2 Discriminant and different

Let K be a number field of degree n = [K : Q]. The trace pairing leads to the notion of duality
for lattices in K (i.e. free Z-submodule of full rank n containing a Q-basis of K). For any lattice
M ⊂ K, we let
M ∨ := {x ∈ K|TrK/Q (xM ) ⊂ Z}
Then M ∨ is also a lattice in K. Indeed, if x1 , · · · , xn is a Z-basis of M , then M ∨ is the free abelian
group of rank n spanned by the dual basis x∨ ∨ ∨
1 , . . . , xn ∈ L (which satisfy TrK/Q (xi xj ) = δij ).

Definition 1.3.3. The discriminant of a lattice M ⊂ K is the integer d(M ) := d(ω1 , . . . , ωn ) where
ω1 , . . . , ωn is a Z-basis of M .

Exercise 1.3.1. Show that the discriminant of a lattice is independent of the choice of basis.
Exercise 1.3.2. Suppose M ⊂ K is a lattice such that M ⊂ M ∨ . Show that [M ∨ : M ] = |d(M )|.
Deduce that M ∨ ⊂ d(M 1
)M.
The equality [M : M ] = |d(M )| is true for any lattice M as long as we define [M ∨ : M ] to be
∨

their relative index. Try to formulate this definition and verify the claim.

Proposition 1.3.2. Any fractional ideal a ⊂ K is a free abelian group of rank n = [K : Q].
Moreover, its dual a∨ is also a fractional ideal.
Proof. It suffices to prove this for nonzero ideals 0 ̸= a ⊂ OK . We observe that

a ⊗Z Q = a ⊗OK OK ⊗Z Q = a ⊗OK K = K

where the last equality is seen as follows: fix a nonzero 0 ̸= α ∈ a, then for any x = ab−1 ∈ K with
a, b ∈ OK we have (αb)x = aα ∈ a.
Consequently a ∩ Z ̸= 0. Take any 0 ̸= n ∈ Z ∩ a then we have nOK ⊂ a ⊂ OK . Thus it suffices
to show that OK is a free abelian group of rank n.
Let x1 , . . . , xn ∈ OK be a Q-basis of K and let M := Zx1 +· · ·+Zxn be the lattice they generate.
Then M ⊂ OK ⊂ M ∨ , where the second inclusion follows from the fact that TrK/Q (OK ) ⊂ Z. Since
M and M ∨ are both free abelian groups of rank n, OK is also a free abelian group of rank n.

12
 ∨
In particular OK is a fractional ideal, called the inverse different (with respect to the extension
∨ ∨ −1
OK /Z). By Proposition 1.1.6 we have OK ⊂ OK and therefore its inverse DK := (OK ) ⊂ OK is
an ideal of OK which we call the different ideal. Beware that except in the trivial case K = Q, we
always have DK ̸= OK (as a consequence of Corollary 2.3.7 and Corollary 4.1.8).
Definition 1.3.4. The discriminant of the number field K is the discriminant of the lattice OK .
We denote it by dK := d(OK ).
By Exercise 1.3.2 we get
∨
Lemma 1.3.3. |dK | = |OK /OK | = |OK /DK | = N DK
Example
√ 1.3.2. Let D be a square-free integer. Then the discriminant of the quadratic field
Q( D) is given by (
4D, if D ≡ 2, 3 mod 4
dQ(√D) =
D, if D ≡ 1 mod 4
It is in general a hard problem to determine the ring of integers OK explicitly. However, the
tool of lattice duality give some control on OK and help us determine it in many cases.
Example 1.3.3. Let θ ∈ K be a primitive element. After multiplying by an integer we may assume
that θ ∈ OK and let f (T ) ∈ Z[T ] be its minimal polynomial so that K = Q(θ) = Q[T ]/f (T ). The
ring Z[θ] = Z[T ]/f (T ) is a lattice in K and we have the inclusions
∨
Z[θ] ⊂ OK ⊂ OK ⊂ Z[θ]∨ .
By definition we have d(θ) = d(Z[θ]). By Exercise 1.3.2 we get
|d(θ)| = |Z[θ]∨ /Z[θ]| = |dK | · |OK /Z[θ]|2 .
If we can show that the number d(θ) is square free, then we conclude that OK = Z[θ]. From Example
1.3.1 we have Y
d(θ) = d(1, θ, . . . , θn−1 ) = (θi − θj )2 (1.3.1)
1≤i<j≤n

where θ1 , . . . , θn are the distinct roots of f (T ) in Q.

In fact, we have the following more precise result.
Proposition 1.3.4. With notations as above, we have
1
Z[θ]∨ = Z[θ]
f ′ (θ)
This is a consequence of
Lemma 1.3.5 (Euler). (
θm 1, m=n−1
TrK/Q ( ′ ) =
f (θ) 0, 0≤m<n−1
Proof. We have
n n
θm X θm
i
X θim
TrK/Q ( ′
)= ′
= Q
f (θ) i=1
f (θi ) i=1 j̸=i (θi − θj )

Then the lemma follows from the identity

n
X Y T − θj
θim ( ) = Tm
i=1
θi − θj
j̸=i

by comparing the coefficients of T n−1 terms.

13
 Let ω1 , . . . , ωn be the basis of Z[θ]∨ dual to 1, θ, . . . , θn−1 under trace pairing. By Euler’s lemma,
n−1
the transition matrix between f ′1(θ) , f ′θ(θ) , . . . , θf ′ (θ) and ω1 , . . . , ωn is in GLn (Z). Thus we get that
Z[θ]∨ = f ′1(θ) Z[θ].

Exercise 1.3.3. Fill in the details of the sketch above.

′
Remark 1.3.6.
′
QnIf OK = Z[θ], then by the above computation f (θ)OK is the different ideal. Note
that f (θ) = i=2 (θ1 − θi ) is a product of differents of roots. Maybe this is the reason for the
terminology different?
From Proposition 1.3.4, we deduce that

|d(θ)| = |d(Z[θ])| = |Z[θ]∨ /Z[θ]| = | det(f ′ (θ)|K)| = |NK/Q (f ′ (θ))| (1.3.2)

This is consistent with Exercise 1.3.2 and (1.3.1). Here we have used the following:
Exercise 1.3.4. Let g ∈ Mn (Z)∩GLn (Q) be an n×n matrix with integral entries and with nonzero
determinant (so that it is invertible in Mn (Q)). Then |Zn /g(Zn )| = | det(g)|.
Exercise 1.3.5. 1. Let f (T ) = T 3 + aT + b with a, b ∈ Z. Assume that f is irreducible in Q[T ].
Show that Disc(f ) = −(4a3 + 27b2 ).
2. Let α be a root of T 3 = T + 1 and K = Q(α), show that OK = Z[α].

1.4 Ring of integers in cyclotomic fields

1.4.1 Generalities on cyclotomic fields
We start by recalling the notion of roots of unity.
Definition 1.4.1. Let n be a positive integer. An n-th root of unity in a ring R is an element of
the group
µn (R) := {x ∈ R, xn = 1}.
An n-th root of unity ζ ∈ µn (R) is called primitive if ζ m ̸= 1 for any integer 0 < m < n. Let
µn (R)prim denote the set of primitive n-th root of unity in R.
By definition we have a disjoint union decomposition
G
µn (R) = µd (R)prim
d|n

where the index set runs over all positive divisors of n.

Lemma 1.4.1. Let F be a field.
1. Any finite subgroup of F × is cyclic. In particular µn (F ) is cyclic for any positive integer n.

2. Suppose F has a primitive n-th root of unity ζ. Then F has characteristic 0 or characteristic
p > 0 not dividing n. Moreover we have

µn (F )prim = {ζ m | m ∈ (Z/nZ)× }

and an isomorphism of cyclic groups Z/nZ ∼

= µn (F ) sending 1 to ζ.

14
Proof. (1) Suppose F × contains a non-cyclic finite subgroup. Then there exists an integer n > 1
and a subgroup of F × isomorphic to (Z/nZ)2 . But this contradicts the fact that a polynomial of
degree n (in our case X n − 1) has at most n distinct roots in any field.
(2) Suppose F has characteristic p > 0 and n = pm for some integer m ≥ 1. Then we have
X n − 1 = (X m − 1)p in F [X] so that ζ is a an m-th root of unity. This contradicts to the fact that
ζ is primitive. Hence we must have p ∤ n.
By (1) the group µn (F ) is cyclic and ζ is a generator, then the rest of the statement follows.
Suppose F is a field of characteristic p > 0 and F Then we must have p ∤ n since xp −1 = (x−1)p
for all x ∈ F . Moreover we have
Proposition 1.4.2. Let F be a field and let n > 1 be an integer. Suppose that the characteristic
of F is either 0 or does not divide n. Let E be the splitting field of the polynomial X n − 1 over F .
Then

1. E contains a primitive n-th root of unity;

2. For any primitive n-th root of unity ζ ∈ µn (E) we have E = F (ζ);
3. E/F is a Galois extension and for any σ ∈ Gal(E/F ) there exists a unique m ∈ (Z/nZ)×
such that σ(ξ) = ξ m for any ξ ∈ µn (E). Moreover the map thus obtained

Gal(E/F ) → (Z/nZ)× (1.4.1)

is an injective group homomorphism.

Proof. (1) The assumption on the characteristic of F implies that X n − 1 is a separable polynomial
in E[X]. In particular we have |µn (E)| = n. Since µn (E) is cyclic, any of its generator is a primitive
n-th root of unity.
(2) If ζ ∈ µn (E) is primitive, then any root of X n − 1 is a power of ζ and hence E = F (ζ).
(3) Let ζ be a primitive n-th root of unity. Then σ(ζ) is also a primitive n-th root of unity
(since σ is a field automorphism) and we deduce that σ(ζ) = ζ m for some integer m prime to n
whose congruence class mod n is uniquely determined. Thus we obtain the map (1.4.1). One checks
immediately that it is an injective group homomorphism.

The homomorphism (1.4.1) is in general not surjective4 . Next we will explain a criterion on
when it is surjective (hence an isomorphism) in terms of cyclotomic polynomials and then show that
it is an isomorphism when F = Q.

1.4.2 Cyclotomic polynomials

Definition 1.4.2. The n-th cyclotomic polynomial Φn (X) is defined by
Y Y
Φn (X) := (X − ζ) = (X − e2πmi/n ).
ζ∈µn (C)prim m∈(Z/nZ)×

From the definition we see immediately that:

• deg Φn (X) = |(Z/nZ)× | = ϕ(n).

• Φn (X) ∈ Z[ζn ][X] and it is monic.

• Φn (X) is invariant under Gal(Q(ζn )/Q).
4 For example if F is algebraically closed.

15
Thus we deduce that the cyclotomic polynomials Φn (X) are monic polynomials in Z[X].
From the disjoint union decomposition
G
µn (C) = µd (C)prim
d|n

we get the factorization Y

Xn − 1 = Φd (X).
d|n

For instance, we have Φ1 (X) = X − 1, Φ2 (X) = X + 1 and for any prime number p,

Xp − 1
Φp (X) = = X p−1 + · · · + X + 1.
X −1
More generally, for any prime power n = pr we have
r
pr−1 Xp − 1
Φpr (X) = Φp (X ) = pr−1
X −1
Exercise 1.4.1. Use Eisenstein’s criterion to show that Φpr (X) is irreducible in Q[X].
More generally we have
Lemma 1.4.3. For any integer n ≥ 1, the cyclotomic polynomial Φn (X) is irreducible in Q[X].

Proof. Let g(X) ∈ Q[X] be a monic irreducible factor of Φn (X) and write Φn (X) = g(X)h(X).
We may assume that n > 2 and deg g(X) > 1. Since Φn (X) ∈ Z[X] and is monic we see that
g(X) ∈ Z[X] and h(X) ∈ Z[X] by Gauss’s Lemma. Let α ∈ C be a root of g(X). For any prime
p ∤ n, αp is also a root of Φn (X). Suppose g(αp ) ̸= 0. Then we must have h(αp ) = 0. Since g(X) is
irreducible, we deduce that g(X) divides h(X p ) in Z[X] and therefore g(X) divides h(X)p in Fp [X].
But then Φn (X) = g(X)h(X) would not be separable in Fp (X) and this contradicts the assumption
that p ∤ n. Thus we have g(αp ) = 0. Consequently we deduce that for any j ∈ (Z/nZ)× , g(αj ) = 0
and this implies that g(X) = Φn (T ).
Lemma 1.4.4. Let F be a field and let n > 1 be an integer. Suppose that the characteristic of
F is either 0 or does not divide n. Let E be the splitting field of the polynomial X n − 1 over F .
Then the natural injective homomorphism Gal(E/F ) ,→ (Z/nZ)× is surjective if and only if Φn (T )
is irreducible in F [T ].
Proof. Let ζ be a primitive n-th root of unity in E. The minimal polynomial of ζ over F divides
Φn (X) and they are equal if and only if [E : F ] = deg Φn (X) = |(Z/nZ)× | if and only if the map
Gal(E/F ) ,→ (Z/nZ)× is surjective.
Corollary 1.4.5. Let ζn ∈ C be a primitive root of unity. Then there is a canonical isomorphism
of groups
Gal(Q(ζn )/Q) ∼ = (Z/nZ)× .
In particular we have [Q(ζn ) : Q] = ϕ(n).

1.4.3 Cyclotomic integers

Theorem 1.4.6. The ring of integers in Q(ζn ) is Z[ζn ].

Lemma 1.4.7. For any integer n ≥ 1, the discriminant d(ζn ) divides n[Q(ζn ):Q] = nϕ(n) .

16
Proof. The minimal polynomial of ζ = ζn is Φn (X) by Lemma 1.4.3. Let f (X) = X n − 1 and write
f (X) = Φn (X)g(X) where g(X) ∈ Z[X]. We have
nζ n−1 = f ′ (ζ) = Φ′n (ζ)g(ζ)
and therefore
NQ(ζn )/Q (Φ′n (ζ))NQ(ζn )/Q (g(ζ)) = ±nϕ(n) .
Since g(ζ) ∈ Z[ζn ] we have NQ(ζn )/Q (g(ζ)) ∈ Z and this proves the result since
d(ζn ) = ±NQ(ζn )/Q (Φ′n (ζ))
by (1.3.2).
We first treat the case where n = pr is a power of a prime number p.
Proposition 1.4.8. The ring of integers in Q(ζpr ) equals to Z[ζpr ]
Proof. To ease notation we denote ζ := ζpr , K := Q(ζpr ) and let O := OK . Let f (X) = Φpr (X) be
the minimal polynomial of ζ. Then we have
r
Xp − 1 r−1 r−1 Y
f (X) = pr−1 = (X p )p−1 + (X p )p−2 + · · · + 1 = (X − ζ m )
X −1
m∈(Z/pr Z)×

Let X = 1 we get that Y

(1 − ζ m ) = p. (1.4.2)
m∈(Z/pr Z)×
1−ζ m
For any positive integer m coprime to n we have 1−ζ ∈ Z[ζ] and conversely write ζ = (ζ m )a where
1−ζ
(a, m) = 1 we get that 1−ζ m ∈ Z[ζ]. Consequently we see that

1 − ζ m ∈ (1 − ζ)Z[ζ]× , ∀m ∈ (Z/nZ)× . (1.4.3)

Combined with (1.4.2) we get that
p ∈ π N Z[ζ]× (1.4.4)
r−1
where N := p (p − 1) = [Q(ζ) : Q].
Denote π := 1 − ζ. Then Z[ζ] = Z[π]. Recall that we have the inclusions
Z[ζ] ⊂ O ⊂ O∨ ⊂ Z[ζ]∨
Let d := |disc(f )| = |Z[ζ]∨ /Z[ζ]|. Then we get the inclusion
1 1
Z[π] = Z[ζ] ⊂ O ⊂ Z[ζ] = Z[π].
d d
By Lemma 1.4.7, d is a power of p. Therefore if O =
̸ Z[π] there exists an element β ∈ O of the form
1
β= (ci π i + · · · + cN −1 π N −1 )
p
where ci , ci+1 , . . . , cN −1 are integers and p ∤ ci . By (1.4.4) we get pβ/π i+1 ∈ O and then we deduce
that ci /π ∈ O and hence NK/Q (ci /π) ∈ Z. But NK/Q (ci ) = cN i is coprime to p and NK/Q (π) = p by
(1.4.2). This is a contradiction and consequently the ring of integers in Q(ζpr ) equals to Z[ζpr ].
Proposition 1.4.9. Let K, K ′ be number fields that are linearly disjoint and let L = KK ′ ∼
= K⊗Q K ′
be their compositum. Let d be the greatest common divisor of the discriminants dK and dK ′ . Then
we have the following inclusions of lattices in L:
1
OK OK ′ ⊂ OL ⊂ OK OK ′ .
d

17
Proof. The inclusion OK OK ′ ⊂ OL is clear.
Let {ω1 , . . . , ωn } be a set of Z-basis of OK and let {ω1′ , . . . , ωm
′
} be a set of Z-basis of OK ′ , where
n = [K : Q] and m = [K : Q]. By assumption we have [L : Q] = mn and {ωi ωj′ , 1 ≤ i ≤ n, 1 ≤ j ≤
′

m} form a set of Q-basis of L. We label the set of field embeddings as

HomQ (K, K ′ ) = HomK ′ (L, K ′ ) = {τ1 , . . . , τn }.

For any x ∈ OL we can write

X n
X
x= aij ωi ωj′ = ωi yi
1≤i≤n i=1
1≤j≤m

where aij ∈ Q and

m
X
yi := aij ωj′ ∈ K ′ .
j=1

Then for any 1 ≤ k ≤ n we get

n
X
τk (x) = τk (ωi )yi
i=1

Let Ω be the n × n matrix whose (i, j)-th entry is τi (ωj ). Then we have
t
(τ1 (x), . . . , τn (x)) = Ω · t (y1 , . . . , yn )

Let Ω∗ be the cofactor matrix of Ω so that ΩΩ∗ = det(Ω). Then we get

det(Ω) · t (y1 , . . . , yn ) = Ω∗t (τ1 (x), . . . , τn (x)).

By assumption all entries of the matrices Ω, Ω∗ are integral over Z. Combined with the assumption
x ∈ OL and the fact that dK = (det Ω)2 we see that
m
X
dK aij ωj′ = dK yi ∈ OK ′ , ∀1 ≤ i ≤ n.
j=1

This implies that aij ∈ d1K Z. By the same argument (and switch the role of K and K ′ ) we get
aij ∈ d 1 ′ Z. Thus we have aij ∈ d1K Z ∩ d 1 ′ Z = d1 Z which implies the inclusion OL ⊂ d1 OK OK ′ .
K K

Proof of Theorem 1.4.6. We prove by induction. When n is a prime power the result is clear by
Proposition 1.4.8.
In general we factorize n = n1 n2 where n1 , n2 are coprime to each other and 1 < n1 , n2 < n.
We first claim that Q(ζn ) = Q(ζn1 )Q(ζn2 ). Clearly we have Q(ζn1 )Q(ζn2 ) ⊂ Q(ζn ) and Q(ζn1 ) ∩
Q(ζn2 ) = Q. Thus

[Q(ζn1 )Q(ζn2 ) : Q] = [Q(ζn1 ) : Q][Q(ζn2 ) : Q] = ϕ(n1 )ϕ(n2 ) = ϕ(n).

By Corollary 1.4.5 this is equal to [Q(ζn ) : Q] and thus Q(ζn ) = Q(ζn1 )Q(ζn2 ). By Lemma 1.4.7,
the discriminants dQ(ζn1 ) and dQ(ζn2 ) are coprime to each other. Then by Proposition 1.4.9 and the
induction hypothesis we get that Z[ζn ] = OQ(ζn ) .

1.5 Local characterization of Dedekind domains

1.5.1 Localization of rings
Notation. For any ring homomorphism f : A → B and any ideal a ⊂ A we denote by aB the ideal
of B generated by f (a). For any ideal b ⊂ B we denote b ∩ A := f −1 (b).

18
 Let R be a ring. Let S ⊂ R be a multiplicative subset, i.e. a subset of R such that 1 ∈ S and
for any x, y ∈ S we have xy ∈ S. The localization of R with respect to S is the ring S −1 R whose
underlying set consists of equivalence classes of formal symbols as for a ∈ R and s ∈ S, where as11 is
equivalent to as22 if (a1 s2 − a2 s1 )s = 0 for some s ∈ S. The ring operations are naturally induced
from R:
a1 a2 a1 s2 + a2 s1 a1 a2 a1 a2
+ = , · = , ∀a1 , a2 ∈ R, s1 , s2 ∈ S.
s1 s2 s1 s2 s1 s2 s1 s2
There is a natural ring homomorphism R → S −1 R sending r to 1r . It is injective if S does not
contain any zero-divisors. Any element of S becomes invertible in S −1 R.
More generally for any R-module M we can define the localization S −1 M in the similar way and
it becomes a S −1 R-module. In fact we have S −1 M = M ⊗R S −1 R.
Lemma 1.5.1. S −1 R is a flat R-algebra.
Proof. Since tensor product is a right exact functor, we only need to show that for any injective
homomorphism of R-modules f : M → N , the induced map S −1 M → S −1 N is also injective.
Suppose there exists m ∈ M and s ∈ S such that f (m) s = 0 in S −1 N . Then there exists t ∈ S such
−1
that f (tm) = tf (m) = 0. Since f is injective we get tm = 0 and therefore ms = 0 in S M

In particular for any ideal I ⊂ R, the induced map S −1 I = I ⊗R S −1 R → S −1 R is injective.

Therefore we can identify S −1 I with its image I · S −1 R, which is the ideal of S −1 R generated by I.
Example 1.5.1. Let p ⊂ R be a prime ideal. Then the set R − p is multiplicative. We denote
Rp := (R − p)−1 R and call it the localization of R at p.
In particular, if R is an integral domain, the localization at the ideal (0) is the fraction field of
K = Frac(R) and any other localizations of R are subrings of K.
Lemma 1.5.2. Let R be an integral domain and let Ω be a set of prime ideals of R that contains
all the maximal ideals. Then we have \
R= Rp
p∈Ω

where the intersection is inside the fraction field of R.

a
Proof. The inclusion “⊂” is clear. Take any b ∈ ∩p∈Ω Rp where a, b ∈ R, b ̸= 0. Consider the set

I := {x ∈ R | xa/b ∈ R} = {x ∈ R | xa ∈ (b)}

This is clearly an ideal of R. For any maximal ideal m ⊂ R, by assumption m ∈ Ω and hence
a/b ∈ Rm . Therefore we can find s ∈ R \ m and r ∈ R such that a/b = r/s. Then sa = rb ∈ (b) so
that s ∈ I. This shows that I is not contained in any maximal ideal of R and we must have I = R.
In particular we have 1 ∈ I which implies that a/b ∈ R.
Lemma 1.5.3. Let R be a ring and let S ⊂ R be a multiplicative subset.

1. For any ideal J ⊂ S −1 R we have S −1 (J ∩ R) = J.

2. For any ideal I ⊂ R we have I ⊂ S −1 I ∩ R.
3. Let p ⊂ R be a prime ideal. If p ∩ S ̸= ∅, then S −1 p = S −1 R. If p ∩ S = ∅ then S −1 p is a
prime ideal of S −1 R and S −1 p ∩ R = p.

Consequently the maps p 7→ S −1 p and q 7→ q ∩ R define mutually inverse bijections between the set
of prime ideals p ⊂ R such that p ∩ S = ∅ and the set of prime ideals in S −1 R.

19
Proof. (1) Let I := J ∩ R. Then clearly S −1 I ⊂ J. Conversely for any xs ∈ J where x ∈ R and
s ∈ S we have x ∈ J ∩ R = I. Thus J ⊂ S −1 I.
(2) is tautological.
(3) Suppose S ∩ p ̸= ∅. Let s ∈ p ∩ S then 1 = ss ∈ S −1 p which means that S −1 p = S −1 R.
Now assume that p ∩ S = ∅. For any as11 , as22 ∈ S −1 R such that as11 sa22 ∈ S −1 p, there exists b ∈ p
and s ∈ S such that as11 sa22 = sb . This means that there exists t ∈ S such that t(a1 a2 s − s1 s2 b) = 0.
Therefore a1 a2 st ∈ p. Since p ∩ S = ∅ we have st ∈ / p and hence a1 a2 ∈ p so that a1 ∈ p or a2 ∈ p.
Thus S −1 p is a prime ideal.
By (2) it remains to show that S −1 p ∩ R ⊂ p. For any x ∈ S −1 p ∩ R we have x1 = ys for some
y ∈ p and s ∈ S. Then there exists t ∈ S such that tsx = ty ∈ p. Since p is prime and ts ∈ / p by
assumption, we must have x ∈ p. This shows that S −1 p ∩ R ⊂ p.

Corollary 1.5.4. For any prime ideal p ⊂ R, the localization Rp is a local ring with maximal ideal
pRp and prime ideals of Rp are in bijection with prime ideals of R contained in p.
Lemma 1.5.5. Let R be a ring and S ⊂ R be a multiplicative subset.
1. If R is noetherian, then S −1 R is noetherian.

2. Suppose R is an integral domain with fraction field K. Let L/K be a finite extension and let
R′ be the integral closure of R in L. Then S −1 R′ is the integral closure of S −1 R in L. In
particular if R is integrally closed, then S −1 R is also integrally closed.
Proof. (1) For any ideal J ⊂ S −1 R, by assumption J ∩ R is finitely generated and thus by the
previous Lemma J = (J ∩ R) · S −1 R is also finitely generated.
(2) Let y ∈ L be integral over S −1 R. Then there exists ai ∈ R and si ∈ S such that
a1 n−1 an
yn + y + ... + =0
s1 sn

Let s = s1 · · · sn and multiply the equation by sn we see that sy is integral over R. Thus sy ∈ R′
and y ∈ S −1 R′ Thus S −1 R′ is the integral closure of S −1 R in L.

Proposition 1.5.6. Let R be a Dedekind domain and S ⊂ R a multiplicative subset.

1. S −1 R is a Dedekind domain.
2. For any fractional ideal a of R, a · S −1 R = S −1 a is a fractional ideal of S −1 R and we have

a−1 · S −1 R = (S −1 a)−1

Proof. (1) By Lemma 1.5.5 S −1 R is noetherian and integrally closed. Let q ⊂ S −1 R be a nonzero
prime ideal. Suppose q ⊂ m for some maximal ideal m. Then q ∩ R = m ∩ R by the assumption that
R is Dedekind. Then we get that

q = S −1 (q ∩ R) = S −1 (m ∩ R) = m.

(2) Let K be the fraction field of R. The inclusion a−1 · S −1 R ⊂ (S −1 a)−1 is clear. It remains to
show the inverse inclusion. Let x1 , . . . , xn ∈ a be a set of generators. For any α ∈ (S −1 a)−1 we have
αxi ∈ S −1 R for any i. Thus there exists s ∈ S such that sαxi ∈ R for all i and hence sα ∈ a−1 .

20
1.5.2 Discrete valuation rings
Let R be a Dedekind domain. Any fractional ideal a of R can be factorized uniquely as
Y
a= pvp (a)
p

where the product runs over all nonzero prime ideals of R and vp (a) ∈ Z for all p vanishes for all
but finitely many p. If a ⊂ R, then vp (a) ≥ 0 for all p. Then we get a group homomorphism

vp : JR → Z

which induces a group isomorphism JR ∼

= ⊕p Z. If we only restricts to principal fractional ideals, we
obtain a group homomorphism (denoted by the same symbol):

vp : K × → Z

sending α to the exponent of p in the factorization of αR.

The homomorphism vp defined above depends on the pair (R, p) consisting of a subring R ⊂ K
which is a Dedekind domain and a nonzero prime ideal p ⊂ R. Different such pairs can give rise
to the same homomorphism. Indeed, by Proposition Q 1.5.6, for any multiplicative subset S ⊂ R and
any fractional ideal a of R with factorization a = p pvp (a) we have
Y
S −1 a = (S −1 p)vp (a) .
p,p∩S=∅

Thus if p ∩ S = ∅, then
vS −1 p (S −1 a) = vp (a).
Therefore the pairs (S −1 R, S −1 p) with S ∩ p = ∅ give rise to the same homomorphism as vp . The
largest among all such pairs is (Rp , pRp ) and it is uniquely determined by vp :

Rp = {x ∈ K | vp (x) ≥ 0} ∪ {0},

pRp = {x ∈ K | vp (x) > 0} ∪ {0}.

We give a characterization of such homomorphism vp that is independent of the choice of the pair
(R, p).
Definition 1.5.1. A discrete (additive) valuation on a field K is a surjective map

v : K → Z ∪ {∞}

that satisfies the following conditions:

1. v(x) = ∞ if and only if x = 0;
2. v(xy) = v(x) + v(y) for any x, y ∈ K;
3. v(x + y) ≥ min{v(x), v(y)} for any x, y ∈ K
By convention ∞ > n and ∞ + n = ∞ for any n ∈ Z.
Lemma-Definition 1.5.7. Let v be a discrete valuation on a field K.
1. The subset
Ov := {x ∈ K | v(x) ≥ 0}
is a subring of K, called the valuation ring of v.

21
 2. Ov is a local PID with only one nonzero prime ideal

mv := {x ∈ K | v(x) > 0}.

3. v induces an isomorphism of groups K × /Ov× ∼

= Z.
Proof. The conditions on v implies that Ov is a subring of K and moreover

Ov× = {x ∈ K | v(x) = 0}.

Therefore Ov is a local integral domain with maximal ideal mv . Any ideal a ⊂ R is generated by
an element x ∈ a with v(x) minimal among elements in a. Thus Ov is a PID with only one nonzero
prime ideal mv .
Definition 1.5.2. A discrete valuation ring (“DVR” for short) is an integral domain that equals
to the valuation ring of some discrete valuation of its fraction field.
Theorem 1.5.8. Let R be a ring which is not a field. The following are equivalent:
1. R is a discrete valuation ring;
2. R is a PID with only one nonzero prime ideal;

3. R is a local PID;
4. R is a local Dedekind domain;
Proof. (1)⇒(2) by Lemma 1.5.7.
(2)⇒(3) is tautological.
(3)⇒(4) by Corollary 1.1.12.
(4)⇒(1): Let m be the maximal ideal of R and let K be the fraction field of R. Then v = vm
defines a discrete valuation on K by Theorem 1.2.1. Let Ov be the valuation ring of v. Then clearly
R ⊂ Ov . For any nonzero x ∈ K we have xR = mv(x) by Theorem 1.2.1. If moreover x ∈ Ov then
v(x) ≥ 0 and we get x ∈ mn ⊂ R. Therefore R = Ov is a discrete valuation ring.

Definition 1.5.3. Let R be a discrete valuation ring. Any generator π of the maximal ideal of R
is called a uniformizer of R.
Theorem 1.5.9. Let R be a noetherian integral domain. Then R is a Dedekind domain if and only
if for all nonzero prime ideal p, the localization Rp is a discrete valuation ring.
Proof. If R is a Dedekind domain, then Rp is a DVR for any nonzero prime p by Theorem 1.5.8 and
Lemma 1.5.7.
Now suppose Rp is a DVR for any nonzero prime p. Let p ⊂ R be a nonzero prime ideal. Suppose
p ⊂ m where m is a maximal ideal. By assumption Rm is DVR so that it has only one nonzero
prime ideal. Therefore we have pRm = mRm and after intersecting with R we get p = m. Thus any
nonzero prime ideal of R is maximal.
It remains to show that R is integrally closed. Let K be the fraction field of R. Then it is also
the fraction field of Rp for any prime p. Let x ∈ K be integral over R. Then it is also integral
over Rp since R ⊂ Rp . By assumption Rp is Dedekind, hence in particular integrally closed and we
deduce that x ∈ ∩p Rp . By Lemma 1.5.2 we get x ∈ R and therefore R is integrally closed.

22
1.5.3 Chinese remainder theorem and application
Many results about a general Dedekind domain could be reduced to the case of discrete valuation
rings after localization, which are often much simpler since these are PIDs.
Theorem 1.5.10 (Chinese Remainder Theorem). Let R be a ring and let a1 , . . . , an be ideals in R
such that ai + aj = R for any i ̸= j. Let a = ∩ni=1 ai be their intersection. Then
Qn
1. We have a = i=1 ai ;
2. For any R-module M , the product of the natural quotient maps M → M/ai M induces an
isomorphism of R-modules
n
M/aM ∼
Y
= M/ai M.
i=1

If more over M is an R-algebra, then this is an isomorphism of R-algebras. In particular when

M = R we have a ring isomorphism
n
R/a ∼
Y
= R/ai
i=1

Q
Proof. For each 1 ≤ i ≤ n, let bi := j̸=i aj . We first claim that ai + bi = R. Indeed, by assumption
for each j ̸= i there exists aj ∈ ai and xj ∈ aj such that aj + xj = 1. Then we have
Y Y
xj = (1 − aj ) ∈ (1 + ai ) ∩ bi
j̸=i j̸=i

which proves the claim.

(1) First consider the case n = 2. The assumption a1 + a2 = R implies that there exists x1 ∈ a1
and x2 ∈ a2 with x1 + x2 = 1. Then for any r ∈ a1 ∩ a2 we have r = ra1 + ra2 ∈ a1 a2 which shows
that a1 ∩ a2 ⊂ a1 a2 . The inclusion in the other direction is obvious.
Suppose the statement is proved for n − 1. Let a′n−1 = an−1 ∩ an . Then a′n−1 ⊃ aj for any
1 ≤ j < n − 1 and the claim in the beginning of the proof shows that a′n−1 + aj = R for any
j ̸= n − 1 and the n = 2 case shows that a′n−1 = an−1 an . Then by the induction hypothesis we get

a = a1 · · · an−2 a′n−1 = a1 · · · an .
Q
(2) We have shown in the beginning of the proof that there exists ei ∈ bi = j̸=i aj with 1 − ei ∈ ai .
Pn
Let e = i=1 ei . Then we have ei ej ∈ a for any i ̸= j which implies that
n
Y
1−e∈ (1 − ei ) + a = a.
i=1

Denote the product of the natural homomorphisms M → M/ai M by

n
Y
f :M → M/ai M.
i=1

Then clearly aM ⊂ ker(f ). For any m ∈ ker(f ) we have m ∈ ai M so that ei m ∈ bi ai M = aM for

all 1 ≤ i ≤ n. Then we get that m + (e − 1)m = em ∈ aM . Since e − 1 ∈ a we get that m ∈ aM .
Thus ker(f ) = aM . Pn
For any m1 , . . . , mn ∈ M , let m = i=1 ei mi . Then we see that f (m) ∈ mi + ai M for all i.
This shows the surjectivity of f and therefore f is an isomorphism of R-modules. When M is an
R-algebra, each homomorphism M → M/ai M is an R-algebra homomorphism so that f is also an
R-algebra isomorphism.

23
Corollary 1.5.11. Let R be a Dedekind domain and a ⊂ R a nonzero ideal. For any R-module M
we have canonical isomorphisms of R-modules

M/aM ∼ M/pvp (a) M ∼

Y Y
= = Mp /pvp (a) Mp .
p p

If moreover M is an R-algebra, then these are isomorphisms of R-algebras

Proof. To apply Theorem 1.5.10, we need to check that for two different nonzero prime ideals p, q
of R and any integers m, n ≥ 0 we have pm + qn = R. If this is not true then pm + qn ⊂ m for some
maximal ideal m. Thus we have pm ⊂ m which implies that p ⊂ m and we must have p = m since
R is Dedekind. But similarly we should also have q = m and this contradicts to the assumption
that p ̸= q. Then the first isomorphism follows from Theorem 1.5.10. The second isomorphism is
deduced from the following Lemma.
Lemma 1.5.12. Let R be a ring and m ⊂ R a maximal ideal. Then for any R-module M and any
integer n ≥ 0 we have an isomorphism of R-modules M/mn M ∼
= Mm /mn Mm .
Proof. The ring R/mn is a local ring with maximal ideal m/mn . Therefore any element in R − m
becomes invertible in R/mn .
For any m ∈ M whose image in Mm lies in mn Mm there exists a ∈ mn , x ∈ M and s ∈ R − m
such that m = axs in Mm . This means that u(ms − ax) = 0 for some u ∈ R − m. Let t = us. Then
t ∈ R − m and tm ∈ mn M . As t is invertible in R/mn and tm vanishes in M/mn M = M ⊗R R/mn
we deduce that m ∈ mn M . Thus the natural map M/mn M → Mm /mn Mm is injective.
Finally we show surjectivity. For any element of Mm written in the form xs where x ∈ M and
s ∈ R − m. Choose an element t ∈ R − m such that st = 1 in R/mn . Then (ts − 1)x ∈ mn M and
hence tx − xs ∈ mn Mm . This shows surjectivity of the map M/mn M → Mm /mn Mm .
Corollary 1.5.13. Let R be a Dedekind domain and let a ⊂ R be a nonzero ideal. Then for any
fractional ideal b there is an isomorphism of R-modules b/ab ∼
= R/a.
Proof. For any nonzero prime ideal p of R, since Rp is PID we deduce that there is an isomorphism
bRp /abRp ∼= Rp /aRp . Then we conclude by Corollary 1.5.11.
∨
Let K be a number field with ring of integers OK . We have defined the inverse different OK
∨ −1
and the different DK = (OK ) ideal and as a special case of the Corollary above we obtain an
∨
isomorphism of OK -modules OK /OK ∼= OK /DK . In particular we get
∨
|dK | = |OK /OK | = |OK /DK |.

Definition 1.5.4. The absolute norm of a nonzero ideal a ⊂ OK is the number N a := |OK /a|.
This is compatible with the norm of elements since for any nonzero α ∈ OK we have

|NK/Q (α)| = |OK /αOK | = N (α).

Corollary 1.5.14. For any nonzero ideals a, b ⊂ OK we have

N (ab) = N a · N b.

Proof. By the previous Corollary we see that b/ab ∼

= OK /a as OK -module. Then we deduce that

N (ab) = |OK /ab| = |OK /b| · |b/ab| = |OK /b| · |OK /a| = N (a)N (b).

24
 We extend the notion of absolute norm to any fractional ideal a by setting

N a := |NK/Q (α)|−1 N (αa)

for any α ∈ K × with αa ⊂ OK . As a consequence of Corollary 1.5.14 we get a group homomorphism

N : JOK → Q>0

from the group of fractional ideals of OK to the multiplicative group of positive rational numbers.

25
Chapter 2

Extensions of Dedekind domains

2.1 Basic set-up

Proposition 2.1.1. Let R be integral domain such that the localization Rm is UFD for any maximal
ideal m.1 Let F be the fraction field of R and let L/F be a finite separable extension. Let α ∈ L be an
element with minimal polynomial m(T ) ∈ F [T ] and characteristic polynomial f (T ) ∈ F [T ]. Then
the following are equivalent:
1. α is integral over R;
2. m(T ) ∈ R[T ];
3. f (T ) ∈ R[T ]
Proof. The proof is the same as Proposition 1.1.6, except that when showing (1)⇒(2) we need the
following generalization of Gausss’s Lemma on primitive polynomials.
Lemma 2.1.2. Let R be integral domain such that the localization Rm is UFD for any maximal
ideal m. Let F be the fraction field of R. Suppose we have a factorization of monic polynomials
f (T ) = g(T )h(T ) in F [T ]. If f (T ) ∈ R[T ], then g(T ), h(T ) ∈ R[T ]
Proof. By assumption and Gauss’s lemma (which holds for any UFD) we have g(T ), h(T ) ∈ Rm [T ]
for any maximal ideal m of R. Then we conclude by Lemma 1.5.2.
Definition 2.1.1. By a degree n extension of Dedekind domains we mean a quadruple (A ⊂ B,
F ⊂ L) where
• A is a Dedekind domain with fraction field F = Frac(A);
• L/F is a finite extension of degree n = [L : F ] and B is the integral closure of A in L.
The extension is separable if moreover L/F is a separable extension.
The definition is justified by the following result.
Proposition 2.1.3. Let (A ⊂ B, F ⊂ L) be a degree n extension of Dedekind domains. Then we
have:
1. B is a Dedekind domain.
2. If L/F is separable, then any fractional ideal b of B is a finitely generated A-module. If
moreover A is PID, then b is a free A-module of rank n.
1 For example a Dedekind domain satisfies this condition.

26
Proof. The second part is proved in the same way as Proposition 1.3.2, using nondegeneracy of trace
pairing for finite separable extensions.
We prove (1) under the extra assumption that L/F is separable. The general case is Krull-
Akizuki theorem. See [Neu99, Chapter I, Proposition 12.8].
Since A is noetherian and B is a finite A-module, we deduce that B is noetherian A-module and
hence a noetherian ring. By Corollary 1.1.4, B is integrally closed.
Let q ⊂ B be a nonzero prime ideal and let p := q ∩ A. Take any 0 ̸= β ∈ q. It satisfies an
equation β m + am−1 β m−1 + · · · + a0 = 0 with ai ∈ A. Then 0 ̸= a0 ∈ p. Since A is Dedekind and
p ̸= 0, we deduce that A/p is a field. Since B is integral over A we see that B/q is integral over
A/p. By Lemma 1.1.14 we see that B/q is a field and thus q is maximal. This shows that B is a
Dedekind domain.
Remark 2.1.4. Part (2) may be false if L/F is not separable. See [Kap74, Theorem 100] for such
an example. This is related to the notion of “Japanese rings”, “Nagata rings”.

In the proof we have used trace pairing and duality of A-submodules of L that generalize the
construction in the case A = Z. More precisely, for any A-submodule M ⊂ L we define its dual to
be the A-module
M ∨ := {x ∈ L|TrL/F (xM ) ⊂ A}.
Lemma 2.1.5. Suppose L/F is a finite separable extension. Let M ⊂ L be an A-submodule.

1. If M contains an F -basis of L so that M ⊗A F = L, then M ∨ is a finitely generated A-module.

2. If M is finitely generated, then M ∨ contains an F -basis of L and hence M ∨ ⊗A F = L.
Pn
Proof. (1) Let x1 , . . . , xn be an F -basis of L contained in M and let N = i=1 Axi be the A-
submodule of M that they generate. By assumption the trace pairing TrL/F is non-degenerate and
hence there exists dual basis x∨ ∨ ∨
1 , . . . , xn that satisfies TrL/F (xi xj ) = δij for any 1 ≤ i, j ≤ n. Then
n
the dual A-module N ∨ = ∨ ∨
⊂ N ∨ we
P
i=1 Axi is finitely generated and from the inclusion M
∨
deduce that M is also a finitely generated A-module (since A is noetherian).
(2) Let y1 , . . . , ym be a set of generaters of the A-module M . FixPan F -basis v1 , . . . , vn of L.
n
For each 1 ≤ i ≤ m, choose a nonzero element ai ∈ A such that ai yi ∈ j=1 Avj and consider their
Pn
product a := a1 · · · am ∈ A. After replacing vj by a−1 vj we may assume that M ⊂ j=1 Avj . Then
the dual basis v1∨ , . . . , vn∨ are contained in M ∨ .

2.2 Factorization of prime ideals in extensions

Let (A ⊂ B, F ⊂ L) be a degree n extension of Dedekind domains in the sense of Definition 2.1.1.
We are interested in the factorization of the ideal pB for a nonzero prime ideal p ⊂ A.
Lemma 2.2.1. Let p ⊂ A and q ⊂ B be nonzero prime ideals. Show that the following are
equivalent:
1. q ∩ A = p;
2. pB ⊂ q
3. q occurs in the unique factorization of pB

We simply say q is above p or p is below q if these equivalent conditions are satisfied. Assume
this is the case and we introduce more terminologies.
Definition 2.2.1. The ramification index e(q|p) is the exponent of q in the factorization of pB.

27
Lemma-Definition 2.2.2. The residue field κ(q) = B/q is a finite extension of κ(p) = A/p of
degree at most [L : F ]. The inertia degree is f (q|p) := [B/q : A/p].
Let κ(p)s be the separable closure of κ(p) in κ(q). Define the separable inertia degree and the
inseparable inertia degree to be

fs (q|p) := [κ(p)s : κ(p)], fi (q|p) := [κ(q) : κ(p)s ]

so that f (P|p) = fs (P|p)fi (P|p). If char(κ(p)) = 0 we always have fi (q|p) = 1 and if char(κ(p)) =
p > 0 then fi (q|p) is a power of p.
Proof. Let x1 , . . . , xm be a set of κ(p)-linearly independent elements in κ(q). For each 1 ≤ i ≤ m
let x̃i ∈ B be a lift of xi . We claim that x̃1 , . . . , P
x̃m are F -linearly independent in L. If not, then
m
there exists a1 , . . . , am ∈ F , not all 0 such that i=1 ai x̃i = 0. After multiplying by an element
in Ap we may assume that ai ∈ Ap for all 1 ≤ i ≤ m and at least one ai lies in A× p . Then there
image in κ(p) would give a nontrivial linear dependence relation among x1 , . . . , xm and we get a
contradiction. Hence x̃1 , . . . , x̃m are F -linearly independent and hence m ≤ [L : F ]. Therefore κ(q)
is a finite extension of κ(p) of degree at most [L : F ].
Lemma 2.2.3. The ramification index e(q|p), inertia degree f (q|p) and its separable and inseparable
parts fs (q|p), fi (q|p) are all multiplicative in towers.
Proof. The fact that e(q|p) is multiplicative in towers is a consequence of Theorem 1.2.1 (the unique
factorization of ideals). The fact that f (q|p) is multiplicative in towers is clear from definition.
For the fact that fs (q|p) and fi (q|p) are multiplicative in towers, see for example [Mor96, Lemma
8.11].
Definition 2.2.2. We say q is unramified over A if e(q|p) = fi (q|p) = 1 (in other words, e(q|p) = 1
and the residue field extension κ(q)/κ(p) is separable). Otherwise we say q is ramified over A.
We say p is unramified in B if all primes of B above p are unramifed over A, otherwise we say
p is ramified in B.
e
Lemma 2.2.4. Let p ⊂ A be a nonzero prime ideal and let pB = Pe11 · · · Pgg be the unique factor-
ization into prime ideals Pi ⊂ B. Let fi := f (Pi |p) = [B/Pi : A/p] be the inertia degree. Then we
have
Xg
ei fi = [L : F ]
i=1

Proof. By the Chinese remainder theorem we have

g
B/pB ∼
Y
= B/Pei i
i=1

First we show that dimκ(p) B/Pei i = ei fi . The B-module B/Pei i has a composition series of length
m+1 m+1
ei whose associated grades are Pm i /Pi . For any x ∈ Pm
i \ Pi we have (x) + Pm+1
i = Pm i .
∼ m
From this we see that multiplication by x induces an isomorphism B/Pi = Pi /Pi m+1
(one can also
deduce this as a special case of Corollary 1.5.13). Consequently

i −1
eX
dimκ(p) B/Pei i = dimκ(p) Pm m+1
i /Pi = ei fi .
m=0

It remains to show that dimκ(p) B/pB = [L : F ]. If A is PID, this is clear since then B is a free
A-module of rank [L : F ]. In general we localize at p and apply Lemma 1.5.12 to reduce to the case
A = Ap is PID.

28
Definition 2.2.3. We say that p splits completely in B if it is unramified and all inertia degrees
are equal to 1. In this case we have pB = P1 · · · Pn where n = [L : F ] and P1 , . . . , Pn are distinct
prime ideals of B.
The following result provides an explicit factorization of primes in many cases.
Theorem 2.2.5 (Kummer). Let θ ∈ B be such that L = F (θ) and let f (T ) ∈ A[T ] be the minimal
polynomial of θ. Let p ⊂ A be a nonzero prime ideal with residue field κ(p) = Q A/p. Assume that2
g
Bp = Ap [θ]. Let f (T ) ∈ κ(p)[T ] be the reduction of f mod p. Let f (T ) = i=1 f¯i (T )ei be the
¯ ¯
¯
factorization into irreducible polynomials fi (T ) ∈ κ(p)[T ].
For each 1 ≤ i ≤ g let fi (T ) ∈ A[T ] be a monic lift of f¯i (T ) and let Pi := pB + fi (θ)B. Then
P1 , . . . , Pg are all the prime ideals of B above p. The inertia degrees are f (Pi |p) = deg f¯i (T ) and
e
the ramification indices are e(Pi |p) = ei . In particular pB = Pe11 · · · Pgg .
Proof. By assumption we have Bp ∼
= Ap [T ]/f (T ) and hence
g
B/pB ∼
= κ(p)[T ]/f¯(T ) ∼
Y
= κ(p)[T ]/f¯i (T )ei
i=1

Thus the primes in B containing pB are given by

ker(B → κ(p)[T ]/f¯i (T )) = pB + fi (θ)B =: Pi

for 1 ≤ i ≤ g. Moreover κ(Pi ) = B/Pi = κ(p)[T ]/f¯i (T ) so that the inertia degree is f (Pi |p) =
deg(f¯i (T )). Qg
Next we determine the ramification indices. Since i=1 f¯i (θ)ei = f¯(θ) = 0 in B/pB we get the
inclusions
g g g
e(P |p)
Y Y Y
ei ei
Pi ⊂ pB + ( fi (θ) )B ⊂ pB = Pi i
i=1 i=1 i=1

which implies that ei ≥ e(Pi |p). Then we get

g
X g
X
n= e(Pi |p)f (Pi |p) ≤ ei deg(f¯i (T )) = dimκ(p) B/pB = n.
i=1 i=1

Thus equality holds and we get e(Pi |p) = ei for all 1 ≤ i ≤ g.

Remark 2.2.6. Associate to the subring A[θ] ⊂ B its conductor

Cθ := {x ∈ B | xB ⊂ A[θ]}

which is a nonzero ideal in B. Let cθ := Cθ ∩A. We claim that if p does not divide cθ (or equivalently
cθ ⊈ p), then Ap [θ] = Bp . Clearly Ap [θ] ⊂ Bp . So it remains to show that Bp ⊂ Ap [θ]. Take an
element s ∈ cθ \ p. Then we have sB ⊂ A[θ] so that B ⊂ s−1 A[θ]. Since s ∈ / p, we get that
Bp ⊂ Ap [θ].
√ √
Example √ 2.2.1. Consider the quadratic field K2 = Q( −5) with ring of integers OK = Z[ −5].
Let θ = −5 with minimal polynomial f (T ) = T + 5.
From f (T ) ≡ (T + 1)2 ∈ F2 [T ] we get the factorization
√
2OK = (2, 1 + −5)2 .

From f (T ) ≡ (T + 1)(T − 1) ∈ F3 [T ] we get the factorization

√ √
3OK = (3, −5 + 1)(3, −5 − 1).
2 For example, this is satisfied if B = A[θ]. But see Example 3.6.2 for a case where this condition is not satisfied

29
 √ √ √ √
Let p1 = (2, 1 + −5), p2 = (3, −5 + 1) and p3 = (3, −5 − 1) = (3, 1 − −5). Then one checks
that these are prime ideals and
√ √
2OK = p21 , (1 + −5)OK = p1 p2 , (1 − −5)OK = p1 p3
√ √
Therefore the two different factorizations 2 · 3 = (1 + −5)(1 − −5) of the element 6 in OK refines
to the same factorization of the ideal 6OK = p21 p2 p3 .

2.3 Ramification
2.3.1 Relative index of lattices
Let R be a Dedekind domain and M an R-module of finite length. Choose a composition series

0 = M0 ⊂ M1 ⊂ · · · ⊂ Mr = M

such that the successive quotients Mi /Mi−1 ∼

= R/pi for nonzero prime ideals p1 , . . . , pr . Define the
characteristic ideal of M to be the product
r
Y
χR (M ) := pi .
i=1

If M = 0 we put χR (0) = R. By Jordan-Hölder theorem, this is independent of the choice of

composition series.
Example 2.3.1. Let p ⊂ R be a nonzero prime and n ≥ 0 be an integer. Then M = R/pn is a
finite length R-module with composition series

0 ⊂ pn−1 /pn ⊂ pn−2 /pn ⊂ · · · ⊂ R/pn .

Thus we have χR (R/pn ) = pn .

One easily verifies the following properties:
• For any short exact sequence of finite length R-modules 0 → M ′ → M → M ′′ → 0 we have
χR (M ) = χR (M ′ )χR (M ′′ ).
• For any multiplicative subset S ⊂ R and any finite length R-module M we have

χS −1 R (S −1 M ) = S −1 χR (M )

• For any nonzero ideal I ⊂ R by Theorem 1.5.10 we get

Y r
Y r
Y
χR (R/I) = χR ( R/pvp (I) ) = χR (R/pvp (I) ) = pvp (I) = I
p i=1 i=1

Let K = Frac(R) be the fraction field of R and let V be a finite dimensional K-vector space.
Definition 2.3.1. An R-lattice in V is a finitely generated R-submodule M ⊂ V such that M ⊗R
K =V.
Lemma 2.3.1. Let M1 , M2 be R-lattices in V . Then
1. There exists an R-lattice M3 in V such that M3 ⊂ M1 ∩ M2 .
2. If M1 ⊂ M2 then their quotient M2 /M1 is an R-module of finite length.

30
Proof. Choose generators x1 , . . . , xm of M2 . Since M1 ⊗R K = V there exists nonzero elements
a1 , . . . , am ∈ R such that ai xi ∈ M1 for all 1 ≤ i ≤ m. Let M3 be the R submodule of V generated
by a1 x1 , . . . , am xm . Then M3 ⊂ M1 ∩ M2 is an R-lattice in V .
If moreover M1 ⊂ M2 , then we have a surjective morphism of R-modules
m
(R/ai R) ∼
Y
= M1 /M3 ↠ M1 /M2
i=1

which shows that M1 /M2 is of finite length.

Definition 2.3.2. For any two R-lattices M1 , M2 in V , define the relative index of M1 , M2 to be
the fractional ideal
χR (M1 , M2 ) := χR (M1 /M3 )χR (M2 /M3 )−1
where M3 is any R-lattice contained in M1 ∩ M2 .
From the definition we easily deduce the following:

• χR (M1 , M2 ) is well-defined, i.e. independent of the choice of the lattice M3 ⊂ M1 ∩ M2 .

• χR (M1 , M2 ) = χR (M2 , M1 )−1 and if M1 ⊃ M2 , then χR (M1 , M2 ) = χR (M1 /M2 ) is an ideal
of R.
• For any multiplicative subset S ⊂ R we have

χS −1 R (S −1 M1 , S −1 M2 ) = S −1 χR (M1 , M2 ).

2.3.2 Different and discriminant

Let (A ⊂ B, F ⊂ L) be a degree n separable extension of Dedekind domains. For any fractional
B-ideal b ⊂ L, consider the dual

b∨ := {x ∈ L | TrL/F (xb) ⊂ A}.

Then b∨ is also a fractional B-ideal by Lemma 2.1.5.

Definition 2.3.3. The inverse different for the extension B/A is the fractional B-ideal B ∨ . The
different for the extension B/A is the ideal

DB/A := (B ∨ )−1 = {x ∈ L, xB ∨ ⊂ B}

Since B ⊂ B ∨ , we see that DB/A ⊂ B is indeed an ideal of B.

Definition 2.3.4. The relative discriminant of an A-lattice M ⊂ L is the following fractional ideal
of A:
∆B/A (M ) := χA (M ∨ , M )
When M = B we get an ideal of A, called the relative discriminant ideal for the extension B/A:

∆B/A := ∆B/A (B) = χA (B ∨ , B) ⊂ A.

By Lemma 2.1.5, for each A-lattice M ⊂ L the dual M ∨ is also an A-lattice in L and hence
∆B/A (M ) is well-defined.
Lemma 2.3.2. Let θ ∈ B such that L = F (θ). Let f (T ) ∈ A[T ] be a polynomial such that f (θ) = 0.
Then we have f ′ (θ) ∈ DB/A . If moreover B = A[θ] and f is the minimal polynomial of θ, then
DB/A = f ′ (θ)B.

31
Proof. First suppose that f is the minimal poylnomial of θ. The same argument as Proposition 1.3.4
shows that A[θ]∨ = f ′1(θ) A[θ]. Then we have the inclusions

1
A[θ] ⊂ B ⊂ B ∨ ⊂ A[θ]∨ = A[θ]
f ′ (θ)
which implies that f ′ (θ) ∈ (B ∨ )−1 = DB/A .
In general let g(T ) ∈ A[T ] be the minimal polynomial of θ. Then f (T ) = g(T )h(T ) for some
h(T ) ∈ F [T ]. By Lemma 2.1.2 we get h(T ) ∈ A[T ]. Then we have
f ′ (θ) = g ′ (θ)h(θ) ∈ g ′ (θ)B ⊂ DB/A .

The different and discriminant are related via relative norms of ideals.
Definition 2.3.5. The relative norm of a fractional B-ideal b ⊂ L is the fractional A-ideal defined
by
NL/F b = χA (B, b)
where we view B and b as A-lattices in L using Proposition 2.1.3.
Proposition 2.3.3. The relative norm NL/F has the following properties:
1. Let P ⊂ B be a nonzero prime ideal and let p = P ∩ A. Let f (P|p) = [B/P : A/p] be the
inertia degree. Then we have
NL/F P = pf (P|p) .

2. For any β ∈ L× , we have NL/F (βB) = (NL/F β)A.

3. NL/K defines a group homomorphism JB → JA between the group of fractional ideals.
4. Let a be a fractional ideal of A. Then we have
NL/F (aB) = a[L:F ]

Proof. 1. By definition we have NL/F P = χA (B/P). From the isomorphism of A-modules B/P = ∼
f (P|p) f (P|p)
(A/p) we get that NL/F P = p .
2. It suffices to show that they are equal after localization at any prime ideal p ⊂ A. Then we
are reduced to the case where A is DVR in which the result follows from the theorem of elementary
divisors.
3. Let b1 , b2 be fractional ideals of B. We need to show that NL/F (b1 b2 ) = NL/F b1 NL/F b2 .
After localization we reduce to the case where A is DVR and the statement follows from the previous
part.
e
4. By 3 we may assume that a = p is a nonzero prime ideal of A. Let pB = Pe11 · · · Pgg be the
unique factorization of prime ideals of B and let fi := [B/Pi : A/p] be the inertia degrees. By 3, 1
and Lemma 2.2.4 we get
Yg
NL/F (pB) = pei fi = p[L:F ]
i=1

Theorem 2.3.4. The relative discriminant and relative different are related by ∆B/A = NL/F DB/A
Proof. By Corollary 1.5.13 there is an isomorphism of B-modules B/DB/A ∼
= B ∨ /B and hence
∆B/A = χA (B ∨ , B) = χA (B ∨ /B) = χA (B/DB/A ) = NL/K DB/A .
where we use Corollary 1.5.13 to get an isomorphism of B-modules B ∨ /B ∼
= B/DB/A .

32
2.3.3 Ramification criterions
Theorem 2.3.5. Let P ⊂ B be a nonzero prime ideal and let p = P ∩ A. Let e = e(P|p) be the
ramification index and p be the characteristic of the residue field κ(p) = A/p.
1. Pe−1 | DB/A (equivalently, Pe−1 ⊃ DB/A );
2. Pe |DB/A if and only if p|e or the residue field extension κ(P)/κ(p) is inseparable.
Definition 2.3.6. A nonzero prime ideal P ⊂ B with p := P ∩ B is wildly ramified over A if either
char(A/p)|e(P|p) or B/P is not separable over A/p. Otherwise we say P is tamely ramified over A.
Remark 2.3.6. If P is tamely ramified over A, then Theorem 2.3.5 says that vP (DB/A ) = e(P|p)−1.
On the other hand, if P is wildly ramified over A, then Theorem 2.3.5 only give us a lower bound
vP (DB/A ) ≥ e. Using completion one can also get an upper bound (this is one of the first applications
of p-adic numbers in algebraic number theory). The basic idea is to study the localizations DB/A BP
for each prime P of B. One would hope that DB/A BP is determined by the extension BP /Ap .
However, if there are more than one prime above p, then the extension BP /Ap is not integral
and BP is not a finite Ap -module. Thus the “local different DBP /Ap ” does not quite make sense.
However we’ll see that after completion this becomes a finite extension of Dedekind domains and
we can define the local different that determines vP (DB/A ).
Corollary 2.3.7. A nonzero prime ideal P ⊂ B is ramified over A if and only if P divides DB/A ;
A nonzero prime ideal p ⊂ A is ramified in B if and only if p divides the discriminant ∆B/A
Proof. The first part follows directly from Theorem 2.3.5. Then the second part follows from
Theorem 2.3.4.
The following Corollary is quite useful in practice, especially when we do not have an explicit
description of B.
Corollary 2.3.8. Let θ ∈ B be such that L = F (θ). Let f (T ) ∈ A[T ] be a polynomial such that
f (θ) = 0.
1. If a nonzero prime ideal P ⊂ B does not divide f ′ (θ) (or equivalently if f ′ (θ) ∈
/ P), then P is
unramified over A.
2. If a nonzero prime ideal p ⊂ A does not divide NL/F f ′ (θ), then p is unramified in B
Proof. If f (T ) is the minimal polynomial of θ, then this follows from Lemma 2.3.2 and Corol-
lary 2.3.7.
In general let g(T ) ∈ A[T ] be the minimal polynomial of θ. Then there exists h(T ) ∈ F [T ] with
f (T ) = g(T )h(T ). By Lemma 2.1.2 we get h(T ) ∈ A[T ] and therefore f ′ (θ) = g ′ (θ)h(θ) ∈ g ′ (θ)B.
If P does not divide f ′ (θ) then it also does not divide g ′ (θ) and we are reduced to the previous
case.
Example 2.3.2. For any integer n > 1, we consider the n-th cyclotomic field K = Q(ζn ) where
ζn is a primitive n-th root of unit. Take θ = ζn and f (T ) = T n − 1. Note that f (T ) is not
the minimal polynomial of θ but the previous Corollary still applies. Then f ′ (θ) = nζnn−1 and
NK/Q f ′ (θ) = ±n[K:Q] . So we conclude that any prime p ∤ n is unramified in OQ(ζn ) = Z[ζn ].
In the case n = pr is a power of the prime p, we know from the identity
Y
p= (1 − ζ m )
m∈(Z/pr Z)×

that pOK = (πOK )[K:Q] . This implies that πOK is a prime ideal of OK and p is totally ramified in
OK , in other words, πOK is the only prime above p and e(πOK |pZ) = [K : Q].

33
Example 2.3.3. √ Let n > 1 be an integer not
√ divisible by the characteristic of F . Consider the
1
extension L = F ( n a) where a ∈ A. Let θ = n a and let f (T ) = T n − a. Then f ′ (θ) = na1− n and
NL/F f ′ (θ) = ±nn an−1 . Therefore any nonzero prime p of A that does not divide na is unramified
in B.
Let us illustrate Theorem 2.3.5 and Corollary 2.3.7 in the special case where B = A[θ] ∼ =
A[T ]/f (T ) for some θ ∈ B with minimal polynomial f (T ) ∈ A[T ]. In this case we have ∆B/A =
disc(f )A by Lemma 2.3.2.
By Kummer’s Theorem 2.2.5 we see that
p ramifies in B ⇔ f¯(T ) ∈ κ(p)[T ] is not separable ⇔ disc(f¯) = 0 ⇔ p divides ∆B/A = disc(f )A.
In general for a prime p ⊂ A that ramifies in B, only some of the primes above p ramifies over
A. To determine these, we consider the relative different DB/A = f ′ (θ)B. Keep the notation from
Theorem 2.2.5. Then we have
g
X Y e Y e X e Y e
f ′ (θ) ≡ (fkek )′ (θ) fj j (θ) = ei fiei −1 fi′ (θ)(θ) fj j (θ) + (fk k )′ (θ) fj j (θ) mod pB
k=1 j̸=k j̸=i k̸=i j̸=k

Since the second term in the sum is divisible by fiei (θ), we deduce that f ′ (θ) ∈ Piei −1 where we
recall that Pi = pB + fi (θ)B. Moreover, for any j ̸= i, since f¯j (T ) is relatively prime to f¯i (T ) we
have fj (θ) ∈
/ Pi and hence
f ′ (θ) ∈ Pei i ⇔ ei fi′ (θ) ∈ Pi ⇔ ei ∈ Pi ∩ Z = pZ or fi′ (θ) ∈ Pi
Finally fi′ (θ) ∈ Pi if and only if f¯i (T ) and f¯i′ (T ) are not relatively prime in κ(p)[T ]. This happens
precisely when the extension κ(Pi )/κ(p) is not separable.
Lemma 2.3.9. For any nonzero ideal b ⊂ B, we have b divides the different ideal DB/A (equivalently
b ⊃ DB/A ) if and only if TrL/K (b−1 ) ⊂ A.
Proof. Since DB/A = (B ∨ )−1 we have b ⊃ DB/A if and only if b−1 ⊂ B ∨ .
If b−1 ⊂ B ∨ , then TrL/F (b−1 ) ⊂ A by the definiton of B ∨ . Conversely if TrL/F (b−1 ) ⊂ A,
we have TrL/F (xB) ⊂ A for all x ∈ b−1 since b−1 is a B-module. Thus x ∈ B ∨ and we see that
b−1 ⊂ B ∨ .
Proof of Theorem 2.3.5. (1) We can write pB = Pe−1 b where b ⊂ B is an ideal which is contained
in any prime of B lying over p. Then b/pB is contained in the nilpotent radical of B/pB and hence
Tr(B/pB)/A/p (b/pB) = 0. From the proof of Lemma 2.2.4 we see that dimA/p B/pB = [L : F ]. Then
any basis of B/pB lifts to an F basis of L (contained in Bp , here we use that Bp is free Ap -module
of rank [L : F ]) and we deduce that TrL/F (b) ⊂ p. From this we see that
TrL/F (P1−e ) = TrL/F ((pB)−1 b) = TrL/F (p−1 b) = p−1 TrL/F (b) ⊂ A.
Hence Pe−1 |DB/A by Lemma 2.3.9.
(2) Write pB = Pe I, then I + P = B. By Lemma 2.3.9 we have the following equivalences
Pe |DB/A ⇔ TrL/F (P−e ) ⊂ A ⇔ TrL/F (I) ⊂ p ⇔ Tr(B/pB)/(A/p) (I/pB) ≡ 0.
By Corollary 1.5.13 there is an isomorphism of B-modules I/p ∼
= B/Pe and thus
Tr(B/pB)/(A/p) (I/pB) ≡ 0 ⇔ Tr(B/Pe )/(A/p) (B/Pe ) ≡ 0
For any x ∈ B/Pe we have
e
X
Tr(B/Pe )/(A/p) (x) = Tr(x|Pi−1 /Pi ) = eTrκ(P)/κ(p) (x).
i=1

Note that Trκ(P)/κ(p) ≡ 0 if and only if the extension κ(P)/κ(p) is inseparable. Combining all the
equivalences above we are done.

34
 √ √
Example 2.3.4.√Consider K = Q( 3 2) and we will show that OK = Z[ 3 2].
Denote θ := 2. Its minimal polynomial is f (T ) = T 3 − 2. We compute that
3

|dK | · [OK : Z[θ]]2 = |d(Z[θ])| = |NK/Q f ′ (θ)| = 22 · 33

Since 2 = θ3 we see that p := θOK is a prime ideal with inertia degree f (p|2Z) = 1 and 2OK = p3
is totally (tamely) ramified with e(p|2Z) = 3. Therefore p2 | DOK /Z .
×
From the identity 1 = θ3 − 1 = (θ − 1)(θ2 + θ + 1) we see that (θ − 1) ∈ OK . Then the identity
3 3 3
(θ + 1) = θ−1 implies that 3OK = (θ + 1) OK . Thus q := (θ + 1)OK is a prime ideal of OK with
inertia degree f (q|3Z) = 1 and 3OK = q3 is totally wildly ramified. Therefore q3 | DOK /Z .
In conclusion we have p2 q3 | DOK /Z and after taking norm we get 22 · 33 |dK . This implies that
√
OK = Z[ 3 2]

Exercise 2.3.1. Let K = Q(θ) where θ3 − θ − 1 = 0.

1. Calculate the discriminant dK . Use the result to show that OK = Z[θ] and only the prime 23
ramifies in OK .
2. Show that 23OK = pq2 for prime ideals p and q of OK . (Hint: use the main theorem on the
relation between different and ramification).
3. Factorize the polynomial f (T ) = T 3 − T − 1 mod 23 to determine explicitly the primes p and
q. (Hint: first find the zero of f ′ (T ) mod 23 to get the repeated root.)

2.4 The case of Galois extensions

2.4.1 Decomposition groups and inertia groups
Now we assume moreover that L/F is a finite Galois extension with Galois group G = Gal(L/F ).
The action of G on L restricts to automorphisms of the A-algebra B and induces action on the set
of prime ideals of B. For a prime ideal P ⊂ B, any σ ∈ G induces an isomorphism of residue fields
∼
σ : B/P − → B/σ(P).

Lemma 2.4.1. Let p, q be prime ideals of B such that p ∩ A = q ∩ A. Then there exists σ ∈ G such
that σ(p) = q.
Proof. Suppose p ̸= σ(q) for all σ ∈ Gal(L/F ). By the Chinese Remainder Theorem,
Q there exists x ∈
B that x ∈ p but x ∈
/ σ(q) for any σ ∈ Gal(L/F ). Consider the norm NL/F (x) = σ∈Gal(L/F ) σ(x).
Then we have NL/F (x) ∈ p ∩ A but NL/F (x) ∈ / q ∩ A. This contradicts the hypothesis that
p ∩ A = q ∩ A.
Consequently in the factorization of primes pB = Pe11 · · · Perr , all ramification indices ei are the
same (denote it by e) and all inertia degrees fi are the same (denoted f ) and we have ef g = [L : K].
The factorization has the simple form

pB = (P1 · · · Pg )e

where the factors Pi ’s are mutually distinct.

Definition 2.4.1. The decomposition group of a nonzero prime ideal P ⊂ B (with respect to L/F )
is its stabilizer in G:
L/F
DP = DP := {σ ∈ G|σ(P) = P}
The decomposition field of P (with respect to L/F ) is the subfield ZP := LDP .

35
 Any σ ∈ DP induces an automorphism of the residue field κ(P) = B/P, leaving stable the
subfield κ(p) = A/p where p = P ∩ A. Thus we get a homomorphism DP → Autκ(p) (κ(P)).
Lemma 2.4.2. The residue field extension κ(P)/κ(p) is normal and the homomorphism DP →
Autκ(p) (κ(P)) is surjective.
Proof. Take any element θ ∈ κ(P) and let g(T ) ∈ κ(p)[T ] be its minimal polynomial. Choose β ∈ B
lifting θ. Let f (T ) ∈ A[T ] be the minimal polynomial of β and let f¯(T ) ∈ κ(p)[T ] be its reduction
mod p. Then we have g|f¯. Since the extension L/F is normal, f (T ) splits into linear factors in
B[T ]. Hence g(T ) splits into linear factors in κ(P)[T ]. This shows that the extension κ(P)/κ(p) is
normal.
Now take θ ∈ κ(P) so that κ(p)(θ) is the separable closure of κ(p) in κ(P). Then we have
Autκ(p) (κ(P)) ∼
= Gal(κ(p)(θ)/κ(p)).
The minimal polynomial of θ is factorized as
Y
g(T ) = (T − τ (θ)).
τ ∈Autκ(p) (κ(P))

By the Chinese Remainder Theorem, there exists α ∈ B satisfying the following congruences:
(
α ≡ θ mod P
α ≡ 0 mod σ −1 (P), ∀σ ∈ G \ DP

Let f (T ) ∈ A[T ] be the characteristic polynomial of α. Let m := |G \ DP |. For each σ ∈ DP we let

σ̄ denote its image in Autκ(p) (κ(P)). Then we have
Y Y
f (T ) = (T − σ(α)) ≡ T m (T − σ̄(θ)) mod P.
σ∈G σ∈DP

From the divisibility g|f¯ we deduce that for any τ ∈ Autκ(p) (κ(P)) there exists σ ∈ DP such that
τ (θ) = σ̄(θ). Since θ generates the separable closure of κ(p) in κ(P), we get that τ = σ̄. Therefore
the homomorphism DP → Autκ(p) (κ(P)) is surjective.
Remark 2.4.3. In general the extension κ(P)/κ(p) might be non-separable when L/F is separable.
See [Conb] for typical examples. However in our case of interest where L and F are number fields
(or finite extensions of Fp (t), the so-called function fields over finite fields), the residue fields are
finite fields and all their algebraic extensions are separable.
Definition 2.4.2. The inertia group at P (with respect to L/F ) is the subgroup
L/F
IP = IP := ker(DP → Autκ(p) (κ(P)))

The inertia field of P (with respect to L/F ) is the subfield TP := LIP .

Recall from Definition 2.2.2 that we factorize the inertia degree as f (P|p) = fs (P|p)fi (P|p) in
which
fs (P|p) := [κ(p)s : κ(p)], fi (P|p) := [κ(P) : κ(p)s ]
where κ(p)s is the separable closure of κ(p) in κ(P). In particular we have
fs (P|p) = |Autκ(p) (κ(P))|.
Lemma 2.4.4. Let fs := fs (P|p) and fi := fi (P|p) and f = f (P|p) = fs fi . Then we have
|DP | = ef, |IP | = efi
and IP = {1} if and only if P is unramified above A.

36
Proof. Let p := P ∩ A. Let e = e(P|p), f = f (P|p) and let g be the number of distinct primes of B
above A so that ef g = n = |G|. By Lemma 2.4.1 we get |DP | = n/g = ef . Then by Lemma 2.4.2
we get
ef = |IP | · |Autκ(p) (κ(P))| = |IP |fs
Then we deduce all the statements easily.

Lemma 2.4.5. Let (A ⊂ B, F ⊂ L) be a finite Galois extension of Dedekind domains. Let F ⊂

K ⊂ L be a sub-extension and let C be the integral closure of A in K. Let P ⊂ B be a nonzero
prime ideal. Let p := P ∩ F (resp. q := P ∩ K) be the prime ideal of A (resp. C) below P. Denote

G := Gal(L/F ) ⊃ H := Gal(L/K).

1. The decomposition and inertia groups of P for the extensions L/F and L/K are related by
L/K L/F L/K L/F
DP = DP ∩ H, IP = IP ∩ H.

2. Suppose H is a normal subgroup of G so that K/F is a Galois extension with Gal(K/F ) ∼ =

G/H. Then the decomposition and inertia groups of q for the extension K/F are described by
L/F L/F L/F
K/F DP H DP DP
Dq = = L/F
= L/K
,
H DP ∩H DP

L/F L/F L/F

K/F IP H IP IP
Iq = = L/F
= L/K
.
H IP ∩H IP

Proof. 1. This follows directly from the definitions.

2. The natural homomorphism G → Gal(K/F ) = G/H defined by σ 7→ σ|K induces homomor-
L/F K/F L/F K/F
phisms DP → Dq and IP → Iq . It remains to show that both maps are surjective.
K/F
Let σ ∈ G with σ|K ∈ Dq . Then σ(P) is a prime of B above q and by Lemma 2.4.1 there
L/F
exists τ ∈ H = Gal(L/K) with τ σ(P) = P. This shows that σ ∈ DP H and hence the statement
K/F
about Dq follows.
K/F
Assume moreover that σ|K ∈ Iq . By what we have just proved, after multiplying by an
L/F
element in H we may assume that σ ∈ DP . Let σ̄ be the image of σ in Autκ(p) (κ(P)). Then
K/F
σ̄ lies in the subgroup Autκ(q) (κ(P)) ⊂ Autκ(p) (κ(P)) since σ|K ∈ Iq . By Lemma 2.4.2 there
L/K L/F
exists τ ∈ DP ⊂ H such that σ̄τ̄ = 1. Thus στ ∈ IP and this proves the statement about
inertia groups.

2.4.2 Decomposition fields and inertia fields

Theorem 2.4.6. Let p = P ∩ F = P ∩ A, PZ := P ∩ ZP and PT := P ∩ TP .
1. P is the only prime of L above PZ and PT .
2. We have e(P|PZ ) = e, f (P|PZ ) = f and e(PZ |p) = f (PZ |p) = 1.

3. We have e(P|PT ) = e, fs (P|PT ) = 1, fi (P|PT ) = fi . Consequently e(PT |p) = fi (PT |p) = 1

so that PT is unramified over F .

37
 We summarize this in the following diagram (in which we have assumed κ(P)/κ(p) is separable):

F
1 / ZP 1 / TP e /L
1 f 1
p / PZ / PT /P

where the upper row indicates the ramification indices and the lower row indicates the inertia
degrees.
Proof. (1) By Lemma 2.4.1, any prime ideal of L above PZ is of the form σ(P) for σ ∈ Gal(L/ZP ) =
DP . The only such prime is P itself. Since ZP ⊂ TP ⊂ L, P is also the only prime of L above PT .
L/Z L/F
(2) By Lemma 2.4.5 we have DP P = DP ∩ Gal(L/ZP ) = DP and by Lemma 2.4.4 we get
L/ZP
ef = |DP | = |DP | = e(P|PZ )f (P|PZ ).
Using the fact that ramification indices and inertia degrees are multiplicative in towers we get
e(P|PZ ) = e, f (P|PZ ) = f
and
e(PZ |p) = 1, f (PZ |p) = 1.
L/T L/F
(3) By Lemma 2.4.5 we have IP P = IP ∩ Gal(L/TP ) = IP and by Lemma 2.4.4 we get
L/TP
e(P|PT )fi (P|PT ) = |IP | = |IP | = efi .
Then we deduce that
e(P|PT ) = e, fi (P|PT ) = fi
and hence e(PT |p) = 1 and fi (PT |p) = 1 so that p is unramified in TP .
L/T
Moreover since IP P = Gal(L/TP ) we get fs (P|PT ) = 1 by Lemma 2.4.2.
Theorem 2.4.7. Let K ⊂ L be a subextension of L/F and let q := P ∩ K.
1. If q is unramified over F , i.e. e(q|p) = fi (q|p) = 1, then K ⊂ TP (the inertia field).
2. If e(q|p) = f (q|p) = 1, then K ⊂ ZP (the decomposition field).
3. If P is the only prime of L above q, then ZP ⊂ K;
4. If q is totally ramified in L, i.e. e(P|q)fi (P|q) = [L : K], then TP ⊂ K.
Proof. Let H = Gal(L/K) be the subgroup of G corresponding to K.
(1) Suppose q is unramified over F so that e(q|p) = fi (q|p) = 1. Then we get e(P|q) = e(P|p)
and fi (P|q) = fi (P|p). On the other hand we have
L/K
Gal(L/KTP ) = H ∩ IP = IP
where the second equality follows from Lemma 2.4.5. Combined with Lemma 2.4.4 we deduce
L/K
[L : KTP ] = |IP | = e(P|q)fi (P|q) = e(P|p)fi (P|p) = [L : TP ].
Since TP ⊂ KTP , this equality implies that K ⊂ TP .
(2) The proof is similar to (1). We just need to replace fi (resp. TP ) by f (resp. ZP )
(3) If P is the only prime of L above q, then H ⊂ DP and hence ZP ⊂ K.
L/K
(4) If q is totally ramified in L, then H = IP = H ∩ IP and hence H ⊂ IP . Then we get that
IP H
TP = L ⊂ K = L .
Corollary 2.4.8. Let K1 , K2 be extensions of F inside L. Then p is unramified (resp. splits
completely) in the compositum K1 K2 if and only if it is unramified (resp. splits completely) in both
K1 and K2 .

38
2.4.3 Frobenius elements
Now we assume moreover that κ(p) = A/p is a finite field with cardinality #κ(p) = N p = |A/p|.
This is the case when F is a number field or a finite extension of Fp (T ) for some prime number
p. Let P ⊂ B be a nonzero prime ideal with p = P ∩ A. The residue field κ(P) is a finite field
with cardinality #κ(P) = N P = |B/P|. The extension κ(P)/κ(p) is Galois and Gal(κ(P)/κ(p)) is
cyclic with a canonical generator (called simply the Frobenius) given by N p-th power map on κ(P).
Any element in the decomposition group DP that maps to the Frobenius in Gal(κ(P)/κ(p)) will be
called a Frobenius at P. They form a coset under
 the inertia IP . By abuse of notation any element
L/F
in this coset will be denoted by the symbol P and is characterized by the formula
 
L/F
(β) ≡ β N p mod P, ∀β ∈ B.
P
If P is unramified over A (or equivalently p is unramified in B, since
 theextension is Galois), then
∼
IP = {1} and DP = Gal(κ(P)/κ(p)). In this case the element L/F is uniquely determined.
P
Moreover, for any σ ∈ G we have
   
L/F L/F
=σ σ −1 .
σ(P) P
Since
 G acts transitively on the set of primes of B above p, we  see that
 the conjugacy class of
L/F L/F
P in G only depends on p. We denote this conjugacy class by p and call it the Frobenius
 
conjugacy class at p. In the special case when G is abelian, any Frobenius conjugacy class L/F p
consists of a single element which coincides with the Frobenius at P for any P of B above p.
 
Lemma 2.4.9. p splits completely in L if and only if p is unramified in L and L/F p = {1}.
 
Proof. p splits completely if and only if DP = {1} for any P above p, if and only if L/F P = {1}
 
L/F
since P is a generator of DP .

Lemma 2.4.10. Let the notations be the same as in Lemma 2.4.5. Assume that P is unramified
over F .
1. The Frobenius elements of P for the extensions L/F and L/K are related by
   f (q|p)
L/K L/F
= .
P P

2. The Frobenius elements of P and q are related by

   
K/F L/F
= .
q P K

Proof. 1. For any β ∈ B we have

 
L/K
(β) ≡ β N q = (β N p )f (q|p) mod P.
P
This implies the identity.
2. For any γ ∈ C, where C is the integral closure of A in K, we have
 
L/F
(γ) − γ N p ∈ P ∩ K = q
P

39
or in other words,  
L/F
(γ) ≡ γ N p mod q.
P
 
K/F
Then we get the identity since this equality uniquely characterizes the element q .

Example 2.4.1. We consider the case of cyclotomic field K = Q(ζn  integer n ≥ 1. By

) for any
K/Q
Corollary 2.3.8, any prime ℓ ∤ n is unramified in K. We denote Frobℓ := ℓZ . We claim that under
the canonical isomorphism Gal(Q(ζn )/Q) ∼ = (Z/nZ)× , Frobℓ is mapped to ℓ. Let m ∈ (Z/nZ)× be
m
such that Frobℓ (ζn ) = ζn . Choose any prime ideal q ⊂ OK = Z[ζn ] above ℓ. Then we have

ζnm = Frobℓ (ζn ) ≡ ζnℓ mod q.

If ζnm ̸= ζnℓ , then (ζnm − ζnℓ ) divides the discriminant of T n − 1 which is a power of n. Then we deduce
that q divides n, contradicting the assumption that ℓ ∤ n.
Consequently we have Frobℓ (ζn ) = ζnℓ for any prime ℓ ̸= p. We deduce that the decomposition
group at ℓ is the the subgroup of (Z/nZ)× generated by ℓ and the inertia degree is the order of ℓ in
(Z/nZ)× .

2.4.4 Gauss’s Quadratic Reciprocity Law

√
Lemma 2.4.11. Let K = Q( D) where D ̸= 0, 1 is a square-free integer. Then
1. An odd prime p > 2 with p ∤ D splits in OK if and only if ( D
p ) = 1;

2. 2 splits in OK if D ≡ 1 mod 8, 2 is inert in OK if D ≡ 5 mod 8 and 2 ramifies in OK if

D ≡ 2, 3 mod 4
3. If we identify Gal(K/Q) = {±1}, then for any odd prime p ∤ D we have
   
K/Q D
= .
pZ p
√
Proof. When D ≡ 2, 3 mod 4, we have OK = Z[ D] ∼ = Z[T ]/(T 2 − D) and dK = 4D and hence 2
is ramified in OK . Let p be an odd prime with p ∤ D. Then we have

OK /pOK ∼
= Fp [T ]/(T 2 − D)

Therefore we get that p splits in OK if and only if T 2 −D is reducible in Fp [T ] if and only if ( D

p ) = 1.
√
1+ D ∼ 2 1−D
When D ≡ 1 mod 4, we have OK = Z[ 2 ] = Z[T ]/(T − T + 4 ) and dK = D. Thus 2
and any odd prime p ∤ D are unramified in OK . We have
1−D
OK /pOK ∼
= Fp [T ]/(T 2 − T + ).
4
From the equality T 2 − T + 1−D 4 = 14 ((2T − 1)2 − D) we get that an odd prime p splits in OK if
and only if ( D
p ) = 1.
Finally in F2 [T ] we have T 2 + T = T (T + 1) and T 2 + T + 1 is irreducible. Thus 2 splits in OK
if and only if 1−D
4 is even, if and only if D ≡ 1 mod 8, while 2 is inert in OK if and only if 1−D4 is
odd, if and only if D ≡ 5 mod 8. This proves 1. and 2.
3. This follows from Lemma 2.4.9.
p−1
Let p > 2 be an odd prime number and √ let p∗ = (−1) 2 p. Then we have p∗ ≡ 1 mod 4 and
the discriminant of√the quadratic field Q( p ) is dQ(√p∗ ) = p∗ . In particular p is the only prime
∗

that ramifies in Q( p∗ ).

40
Lemma 2.4.12. Let H ⊂ Gal(Q(ζp )/Q) be the unique subgroup of index 2, corresponding to the sub-
group of quadratic residues
√ ∗ in (Z/pZ)× under the canonical isomorphism Gal(Q(ζp )/Q) ∼
= (Z/pZ)× .
H
Then Q(ζp ) = Q( p ) and it is the unique quadratic subfield of Q(ζp ).
Proof. Since the Galois group Gal(Q(ζp )/Q) ∼ = (Z/pZ)× is a cyclic group of order p − 1, H is its
H
unique subgroup of index 2. Hence Q(ζp ) is the unique quadratic subfield of Q(ζp ). Since any
√ ∗ℓ ̸= p is unramified
prime √ ∗ in Q(ζp ) and hence also unramified in Q(ζp )H , we must have Q(ζp )H =
Q( p ) since Q( p ) is the only quadratic extension of Q in which all primes different from p is
unramified.

Let q ̸= p be another prime number. Then we have the following equivalences

     √ ∗ 
q Q(ζp )/Q Q( p )/Q
= 1 ⇔ q(modp) ∈ H ⇔ √
= 1 ⇔ =1
p qZ Q( p∗ ) qZ

where the first equivalence is by definition of H as the subgroup of quadratic  residues, the second
Q(ζp )/Q
equivalence follows from the previous Lemma and the description of qZ from Example 2.4.1,
while the last equivalence follows from Lemma 2.4.10. √
When q is odd, the last condition above is equivalent to that q splits in Q( p∗ ), which is
∗
equivalent to ( pq ) = 1 by Lemma 2.4.11.
When q = 2, the last condition above is equivalent to that p∗ ≡ 1 mod 8, or equivalently p ≡ ±1
q−1
mod 8. Combined with the fact that ( −1 q ) = (−1)
2 for any odd prime q (which is shown in the
proof of Theorem 0.2.1) we finally get:
Theorem 2.4.13. Let p > 2 be an odd prime number.
1. (Quadratic Reciprocity) For any odd prime q > 2 different from p, we have
  
p q p−1 q−1
= (−1) 2 · 2 .
q p

2. (First supplementary law)  

−1 p−1
= (−1) 2 .
p

3. (Second supplementary law)

  (
2 p2 −1 1 if p ≡ ±1 mod 8
= (−1) 8 = .
p −1 if p ≡ ±3 mod 8

41
Chapter 3

Valuation theory

3.1 p-adic numbers

Fix a prime number p > 0. The following simple exercise is used very frequently when working with
p-adic numbers.
Exercise 3.1.1. Let R be any commutative ring. Suppose x, y ∈ R satisfies x ≡ y(mod pR). Show
that for any integer n ≥ 0, we have
n n
xp ≡ y p mod pn+1 R
p


(Hint: prove by induction. Use the fact that p divides the binomial coefficients i for 0 < i < p.)
The idea is that we want to have Taylor expansion of any integer (or rational number) at a fixed
prime number p (more precisely, at the maximal ideal pZ ∈ SpecZ). Then we get the ring of p-adic
integers Zp and field of p-adic numbers Qp whose underlying sets can be described by

Definition 3.1.1 (Hensel’s original definition). The set of p-adic integers consists of formal series:

X∞
Zp = { ai pi | ai ∈ {0, 1, . . . , p − 1} ∀i ≥ 0}.
i=0

The set of p-adic numbers consists of formal series:

∞
X
Qp = { ai pi | m ∈ Z, ai ∈ {0, 1, . . . , p − 1} ∀i ≥ m}
i=m
P∞
At the moment we understood the formal series s = i=m ai pi ∈ Qp as a sequence of partial
sums X
sn := ai pi ∈ Q, ∀n ≥ 1.
i<n

In particular, a series with only finitely many nonzero terms (so that the sequence of partial sums
sn eventually stabilizes) determines a positive rational number. Conversely, we can expand any
rational number α ∈ Q as a formal series in Qp as follows:
After multiplying by a non-negative power of p (to clear factors of p in the denominator), we
may assume that α ∈ Z(p) . Since Z(p) is the localization of Z at the maximal ideal pZ, we have ring
isomorphisms
Z(p) /pn Z(p) ∼
= Z/pn Z, ∀n ≥ 0.

42
Then there exists a unique a0 ∈ {0, 1, . . . , p − 1} such that α ≡ a0 mod p, or equivalently α−a
p
0
∈
α−a0
Z(p) . Then we find a unique a1 ∈ {0, 1, . . . , p − 1} such that p ≡ a1 mod p, or equivalently

α − a0 − a1 p ≡ 0 mod p2 .
Repeating this procedure we obtain a sequence an ∈ {0, 1, . . . , p − 1}, n ≥ 0, uniquely determined
by α, such that
n−1
X
α− ai pi ≡ 0 mod pn , ∀n ≥ 1.
i=0
This defines a map Q → Qp which restricts to a map Z(p) → Zp . These maps are injective since
∩n≥1 pn Z(p) = Z(p) . In this way we view Q as a subset of Qp and Z(p) as a subset of Zp .
We would like to define addition and multiplication on the set Qp so that Q → Qp becomes a
field embedding and Zp becomes a subring of Qp . We can do this directly with the sequence of
partial sums sn , in the same way as how we add or multiply integers in decimals with the rule of
carrying. Though straightforward, the ring structure on Qp defined in this way is a bit opaque.
There are two approaches to understand the ring structure on Qp conceptually more transparent.
One algebraic and the other analytic.

3.1.1 The algebraic approach

P∞
In the algebraic approach we start with a construction of Zp . For a formal series s = i=0 ai pi ∈ Zp
Pn−1
with associated partial sum {sn = i=0 ai pi }n≥1 . We have sn+1 ≡ sn mod pn and therefore the
sequence of congruence classes {sn ∈ Z/pn Z, n ≥ 1} defines an element in the inverse limit:
∞
Y
lim Z/pn Z := {(an ) ∈ Z/pn Z | an+1 ≡ an mod pn , ∀n ≥ 1}.
← −
n n=1

Lemma 3.1.1. Let S ⊂ Z be a subset mapping bijectively to Fp . For any integer n ≥ 1, any
Pn−1
congruence class in Z/pn Z has a unique representative of the form i=0 ai pi where ai ∈ S for each
0 ≤ i ≤ n − 1.
P∞
Lemma 3.1.2. The map Zp → limn Z/pn Z defined by sending s = i=0 ai pi ∈ Zp to the congruence
←−Pn−1
classes of the partial sums sn = i=0 ai pi mod pn is bijective and restricts to the canonical ring
homomorphism Z(p) → limn Z/pn Z defined by the product of the natural maps
←−
∼ Z/pn Z.
Z(p) → Z(p) /pn Z(p) =
Pn−1
Proof. Any congruence class in Z/pn Z has a unique representative of the form i=0 ai pi where
ai ∈ {0, 1, . . . , p − 1} for each 0 ≤ i ≤ n − 1. Then the Lemma follows immediately.
We will identify Zp = limn Z/pn Z. The inverse limit is a subring of the direct product ring and
←−
therefore we get a description of the ring structure on Zp .
Lemma 3.1.3. Zp = limn Z/pn Z is a discrete valuation ring with field of fractions Qp and maximal
←−
ideal pZp . Moreover, for each n ≥ 1 the projection Zp → Z/pn Z induces a ring isomorphism
Zp /pn Zp ∼
= Z/pn Z.
Proof. It follows from the definition of inverse limit that pn Zp = ker(Zp → Z/pn Z). Also, since
the transition maps Z/pn+1 Z → Z/pn Z are surjective, the natural homomorphism Zp → Z/pn Z is
surjective and hence induces an isomorphism Zp /pn Zp ∼ = Z/pn Z. In particular, pZp is a maximal
×
ideal of Zp and Zp \pZp = Zp . Then we deduce that Zp is a local ring with maximal ideal pZp and
any nonzero element of Zp can be written in the form pn u for u ∈ Z× p and n ∈ Z≥0 . Clearly p is
not nilpotent in Zp . Therefore Zp is a discrete valuation ring and Qp = Zp [ p1 ] is the fraction field of
Zp .

43
3.1.2 The analytic approach
In the analytic approach we start by a constructionP∞of Qp . The idea is that we would like a p-adic
number number, written as a formal series s = i=m ai pi , to really converge to something. For
this purpose the sequence of partial sums has to be a Cauchy sequence and in particular pn has to
converge to 0 when n increases. The way to achieve this is to take the completion of Q under a
metric that makes pn arbitrarily close to 0 when n increases, just like we define R to be equivalence
classes of Cauchy sequence in Q for the usual metric.
Recall that we have defined a discrete valuation vp on Q such that vp (pn ) = n and vp (0) = +∞.
Define the p-adic metric on Q by

|x − y|p := p−vp (x−y) , ∀x, y ∈ Q.

Note that we have |pn |p = p−n and |0|p = 0. In particular |pn |p → 0 when n → +∞.
Then Qp is defined to be the completion of Q with respect to the p-adic metric | · |p , consisting of
equivalence classes of Cauchy sequences in Q under the p-adic metric. Two such Cauchy sequences
{xn }∞ ∞
n=1 and {yn }n=1 are equivalent if lim |xn − yn |p = 0. By construction Q embeds in Qp as a
n→∞
subfield consisting of constant Cauchy sequences.
The absolute value | · |p and valuation vp extends by continuity to Qp , where the value group
R is equipped with its usual topology. For x = {xn }∞ n=1 ∈ Qp , define |x|p := lim |xn |p and
n→∞
log |x|
vp (x) = − log p p (which is +∞ if x = 0). Since vp (Q× ) = Z is discrete, by continuity we also have
vp (Q×p ) = Z. In particular, vp is a discrete additive valuation on Qp . Then se define Zp to be the
valuation ring
Zp = {x ∈ Qp , |x|p ≤ 1}.
Therefore in the analytic approach Zp is a discrete valuation ring by construction. Moreover, we
have Zp ∩ Q = Z(p) .
Lemma 3.1.4. Zp is the closure of Z and also the closure of Z(p) in Qp . For any n ≥ 1 the inclusion
Z ,→ Zp induces ring isomorphism Z/pn Z ∼ = Zp /pn Znp and they combine into a ring isomorphism
Zp ∼
= lim n
Z/p Z.
←−n
Proof. For an element x ∈ Zp represented by a Cauchy sequence {xn }∞ n=1 in Q we have |x| = |xn |
for n sufficiently large. After modifying finitely many terms we may assume xn ∈ Z(p) of all n and
hence Zp is the closure of Z(p) in Qp . Write xn = an /bn with an , bn ∈ Z and p ∤ bn . Find yn ∈ Z
such that bn yn ≡ an mod pn . Then we have |xn − yn | ≤ p−n and hence limn yn = limn xn = x.
This shows that Zp is the closure of Z in Qp .
Then we deduce that Z/pn Z ∼ = Zp /pn Zp for any n ≥ 1. Thus we get a natural homomorphism
Zp → limn Z/pn Z. In jectivity is clear. Surjectivity follows from Lemma 3.1.1.
←−
Therefore the algebraic and analytic approaches both agree with Hensel’s original definition of
Zp and Qp . The following illustrate an important property of p-adic numbers.
Proposition 3.1.5 (Hensel’s lemma: special case). Let f (T ) ∈ Zp [T ] be a monic polynomial and
let f¯ ∈ Fp [T ] be its reduction mod p. Suppose there exists α ∈ Fp such that f¯(α) = 0 but f¯′ (α) ̸= 0.
In other words, we assume α is a simple root of f¯. Then there exists a unique x ∈ Zp such that
x ≡ α(mod p) and f (x) = 0.
Proof. The solution x is constructed by Newton approximation method.
Start by choosing any x0 ∈ Zp such that x0 ≡ α(mod p). Let x1 = x0 − ff′(x 0)
(x0 ) . By assumption
we have f (x0 ) ∈ pZp and f ′ (x0 ) ∈ Z× 2 ′ ×
p , from which we deduce that f (x1 ) ∈ p Zp and f (x1 ) ∈ Zp .
Repeat this procedure, we get a sequence {xn , n ≥ 0} in Zp such that
f (xn−1 )
• xn = xn−1 − f ′ (xn−1 ) ∈ xn−1 + pn Zp ;

44
 • f (xn ) ∈ pn+1 Zp and f ′ (xn ) ∈ Z×
p.

Then x := lim xn exists in Zp and satisfies the requirements. It is unique since x̄ = α is a simple
n→∞
root of f¯.
 
Example 3.1.1. Suppose p is odd. Let a ∈ Z with p ∤ a and let f (T ) = T 2 − a. Suppose ap = 1
so that f¯(T ) has a simple root α ∈ Fp . Then α lifts to a root α̃ ∈ Zp of f . In particular if p ≡ 1
mod 4 there exists α̃ ∈ Zp with α̃2 + 1 = 0. Note however that T 2 + 1 does not have any root in
Z(p) ⊂ Q.
Corollary 3.1.6. There is a unique group homomorphism ω : F× ×
p → Zp such that ω(a) ≡ a(mod p)
×
for all a ∈ Fp .
Proof. Recall that F×p is a cyclic group of order p − 1. Therefore f (T ) := T
p−1
− 1 is a separable
×
polynomial that splits into linear factors in Fp [T ]. Therefore each a ∈ Fp lifts uniquely to a (p−1)-th
root of unity ω(a) ∈ Z×p . This defines a group homomorphism by uniqueness of the lifting.

The homomorphism ω : F× ×
p → Zp is called the Teichmuller character. It is the canonical
generator of the group of Qp -valued Dirichlet characters mod p:
×
Hom(F× i
p , Qp ) = {ω , 0 ≤ i ≤ p − 2}.

As a comparison, we note that the group Hom(F× ×

p , C ) is also cyclic of order p − 1, but there is no
canonical generator.
By Exercise 3.1.1 we see that for any a ∈ Zp with image ā ∈ Fp and any integer n ≥ 0,
n
ω(ā) ≡ ap mod pn+1 (3.1.1)

and therefore n
ω(ā) = lim ap .
n→∞

Unlike R or C, the arithmetic of Qp is more complicated (but still much simpler than Q). Fix an
algebraic closure Qp of Qp . Then Qp is an infinite extension of Qp . This can be seen easily from the
discreteness of the value groups:

vp (Q×
p ) = Z, |Q× Z
p |p = p .

The p-adic valuation vp and metric | · |p extends uniquely to Qp . But Qp is no longer 1 complete
under | · |p . We let Cp be the p-adic completion of Qp . Using Krasner’s lemma one can show that Cp
is algebraically closed (see Corollary 3.4.15). It is the p-adic analogue of C and is the natural place
to define p-adic meromorphic functions. The p-adic valuation vp and metric | · |p extends naturally
to Cp . The extended value groups are
× ×
vp (C×
p ) = vp (Qp ) = Q, |C× Q
p |p = |Qp |p = p .

3.1.3 p-adic meromorphic functions

We will encounter some common p-adic functions that are defined essentially by the same power
series as their complex analytic analogues. But one has to pay extra attention to their radius of
convergence since the topology of Cp and C are quite different. The following lemma shows that in
many aspects analysis in Cp is simpler than in C.
1 It is a general fact that infinite algebraic extensions of a complete nonarchimedean fields are never complete under

the extended absolute value. (By Theorem 3.3.10, which is a consequence of Hensel’s lemma, the valuation extends
uniquely to any algebraic extension.)

45
 P∞
Lemma 3.1.7. An infinite series n=1 an in Qp (or any complete nonarchimedean field, like Cp )
converges if and only if limn→∞ an = 0.
This is an immediate consequence of the ultrametric inequality. With this in mind let’s introduce
some common p-adic functions.
Recall that the exponential and logarithm are power series in Q[[X]] defined by
∞ ∞
X Xn X (−1)n−1 (X − 1)n
exp(X) = , log X = .
n=0
n! n=1
n

They satisfy familiar identities

log(exp(X)) = X, exp(log(1 + X)) = 1 + X in Q[[X]],

log((1 + X)(1 + Y )) = log(1 + X) + log(1 + Y ), exp(X + Y ) = exp(X) exp(Y ) in Q[[X, Y ]].

These can be checked by viewing them as power series in C[[X]] or C[[X, Y ]] and then using complex
analysis.
P∞
Lemma 3.1.8. For any integer n ≥ 1 we have vp (n!) = i=1 ⌊ pni ⌋ and

n − p log n n
− < vp (n!) <
p−1 log p p−1
Proof. Since ⌊ pni ⌋ counts the number of positive multiples of pi in {1, 2, . . . , n}, the first equality is
log n
clear. There is a unique integer a ≥ 0 such that pa ≤ n < pa+1 , then we have a ≤ log p and
a a
X n X n n − np−a n − p log n
vp (n!) = ⌊ i⌋ ≥ ( i − 1) = −a> − .
i=1
p i=1
p p−1 p−1 log p

On the other hand we have

a ∞
X n X n n
vp (n!) = ⌊ i⌋ < i
= .
i=1
p i=1
p p−1

1
Proposition 3.1.9. For any x ∈ Cp , the series exp(x) converges if and only if |x|p < p− p−1 , or
1
equivalently vp (x) > p−1 .
Proof. For any x ∈ Cp , by Lemma 3.1.8 we have
n xn n − p log n
nvp (x) − < vp ( ) < nvp (x) − + .
p−1 n! p−1 log p
1 n n
1
If vp (x) > p−1 , or equivalently if |x|p < p− p−1 we get vp ( xn! ) → +∞ as n → ∞. Then limn→∞ xn! =
0 so that exp(x) converges.
1 n n
1
If vp (x) < p−1 , or equivalently if |x|p > p− p−1 we get vp ( xn! ) → −∞. Then limn→∞ xn! = +∞
so that exp(x) diverges.
1
If vp (x) = p−11
, or equivalently if |x|p = p− p−1 , we cannot directly conclude from the above
pr −1 n−1
inequalities. However notice that for the subsequence n = pr we have vp (pr !) = p−1 = p−1 and
therefore  pr 
x r pr − 1 1
vp r
= p v p (x) − = .
p ! p−1 p−1
xn
Thus the sequence n! does not converge to 0 and hence exp(x) also diverges.

46
 Next we study the p-adic logarithm.
Proposition 3.1.10. For any x ∈ Cp , the series log(x) converges if and only if |x − 1|p < 1, or
equivalently vp (x − 1) > 0.
Proof. For an integer n ≥ 1, write n = pvp (n) m where p ∤ m we get that
log n − log m log n
1 ≤ vp (n) = ≤
log p log p
Then we have
(x − 1)n
 
log n
nvp (x − 1) − ≤ vp ≤ nvp (x − 1)
log p n
 n

from which we deduce that lim vp (x−1) n = +∞ if and only if vp (x − 1) > 1.
n→+∞

Lemma 3.1.11. Let m ≥ 1 be an integer. If p = 2 we require furthermore that m ≥ 2. Then exp

and logp induces mutually inverse isomorphisms of the additive and multiplicative groups
pm Zp ∼
= 1 + pm Zp .
Proposition 3.1.12. 1. There is a canonical isomorphism of topological groups
(
× ∼ Z × F×p × Zp if p > 2
Qp =
Z × Z/2Z × Z2 if p = 2

2. Let µ(p) be the group of roots of unity of order prime to p in Cp . Let U1 = {x ∈ Cp , |x−1|p < 1}
be the open subgroup of C× p . Then there is a (noncanonical) isomorphism of topological groups

C× ∼ (p)
p =Q×µ × U1 .
With these preparation, we can extend logp to all C×
p.

Proposition 3.1.13. There is a unique group homomorphism

logp : C×
p → Cp

such that
1. For any x ∈ Cp with |x − 1|p < 1 we have
∞
X (−1)n+1 (x − 1)n
logp (x) = ;
n=1
n

2. logp (p) = 0;
One basically requires that ker logp = pQ × µ(p) . This and the usual power series on U1 uniquely
determines logp . Beware that unlike µ(p) , any pn -th root of unity lies in U1 .
For each x ∈ Z× p , let ⟨x⟩ := ω(x)
−1
x. Then we have ⟨x⟩ ∈ 1 + pZp and
∞
X (−1)n+1
logp x = logp ⟨x⟩ = (⟨x⟩ − 1)n .
n=1
n

For any a ∈ Z×
p , we define
ax = ⟨a⟩x := exp(x logp ⟨a⟩).
Since logp a = logp ⟨a⟩ ∈ 2pZp we see that ax converges if
( 1
p1− p−1 if p > 2
|x|p <
2 if p = 2

47
3.2 Valuations
3.2.1 General definitions
Definition 3.2.1. An absolute value, or multiplicative valuation on a field K is a map

|·|:K →R

that satisfies the following properties:

1. |x| ≥ 0 for all x ∈ K and |x| = 0 if and only if x = 0;
2. |xy| = |x| · |y| for all x, y ∈ K,
3. |x + y| ≤ |x| + |y| for all x, y ∈ K.

The absolute value | · | is non-archimedean if moreover the following stronger condition holds:
3’. |x + y| ≤ max{|x|, |y|} for all x, y ∈ K.
The absolute value | · | is archimedean if 3’ does not hold.

The following observations are immediate:

• An absolute value | · | induces a homomorphism of multiplicative groups K × → R>0 . We will
always assume that | · | nontrivial, i.e. |K × | =
̸ {1}.
• The absolute value |·| is called discrete if |K × | is a discrete nontrivial subgroup of R>0 . In this
case there exists c > 0, c ̸= 1 such that |K × | = cZ . Then | · | determines a discrete (additive)
|x|
valuation by setting v(x) := log log c . Conversely any discrete additive valuation defines an
absolute value.
• For any root of unity ζ ∈ K we have |ζ| = 1. In particular | − 1| = 1 and hence | − x| = x for
any x ∈ K. Another consequence is that there is no non-trivial absolute value on a finite field
or the algebraic closure of a finite field.

• An absolute value | · | induces a metric, hence a topology on K, by d(x, y) := |x − y| for any

x, y ∈ K.
Example 3.2.1. • Any field embedding K ,→ C induces an archimedean valuation on K;
• The p-adic absolute value | · |p on Q for any prime number p;

• Let K = k(T P)∞or k((T )) where k is any field. Then any element in K can be written as
f (T ) = T n ( i=0 ai T i ) where n ∈ Z, ai ∈ k for all i ≥ 0 and a0 ̸= 0. Define vT (f ) = n and
|f |T = cvT (f ) where 0 < c < 1 is a fixed constant. Then vT is a discrete valuation on K and
| · |T is a discrete non-archimedean absolute value on K.

Definition 3.2.2. Two absolute values | · |1 and | · |2 are equivalent if they induce the same topology
on K.
Lemma 3.2.1. Let | · |1 and | · |2 be two absolute values on a field K and suppose that | · |1 nontrivial.
Then the following are equivalent:
1. | · |1 and | · |2 are equivalent;

2. For any x ∈ K, |x|1 < 1 implies |x|2 < 1;

3. There exists a > 0 such that |x|2 = |x|a1 for all x ∈ K.

48
Proof. 1.⇒2. For any absolute value | · | on K and any x ∈ K we have |x| < 1 if and only if
limn |x|n = 0.
2.⇒3. Since | · |1 is nontrivial, there exists y ∈ K with |y|1 > 1. By 2. we get |y|2 > 1. Let
|y|2 log |y|1
a := log a
log |y|1 > 0. Then we get |y|2 = |y|1 . For any x ∈ K
×
let b := log b
|x|1 so that |x|1 = |y|1 . It
remains to show that |x|2 = |y|b2 .
For any rational number m n > b where m, n ∈ Z and n > 0, since |y|1 > 1 we have |x|1 = |y|1 <
b
m
m/n
|y|1n and hence |xn y −m |1 < 1. By 2. this implies that |xn y −m |2 < 1 and hence |x|2 < |y|2 . Then
we get that |x|2 ≤ |y|b2 . Similarly we have |x|2 ≥ |y|b2 and we are done.
3.⇒1. This is clear since | · |1 and | · |a1 define the same open balls.
Lemma 3.2.2. An absolute value | · | on a field K is nonarchimedean if and only if it is bounded
on its prime ring (i.e. subring generated by 1, or smallest subring of K).
Proof. Suppose | · | is nonarchimedean. Then for any n ∈ Z>0 we have |n| = |1 + · · · + 1| ≤ max{|1|}.
Suppose | · | is bounded no the prime ring. Then there exists C > 0 such that |m| ≤ C for all
m ∈ Z. Then for any n ≥ 0 and any x, y ∈ K we have
n   n  
n
X n i n−i
X n
|x + y| = xy ≤ |x|i |y|n−i ≤ (n + 1)C max{|x|, |y|}n
i=0
i i=0
i

and hence
1 1
|x + y| ≤ (n + 1) n C n max{|x|, |y|}.
Let n → ∞ we deduce that |x + y| ≤ max{|x|, |y|}.
Recall that the p-adic absolute value | · |p on Q is constructed from the p-adic valuation vp . In
general, all non-archimedean valuations arises in this way.
Definition 3.2.3. An additive valuation on K is a map v : K → R ∪ {∞} satisfying the following
conditions:
1. For all x ∈ K, v(x) = ∞ if and only if x = 0;

2. For any x, y ∈ K, v(xy) = v(x) + v(y);

3. For any x, y ∈ K, v(x + y) ≥ min{v(x), v(y)}.
where we took the convention that ∞ + ∞ = ∞, r + ∞ = ∞ and r < ∞ for any r ∈ R.
An additive valuation v induces an absolute value | · |v by setting |x|v := cv(x) for any x ∈
×
K where 0 < c < 1 is any constant. In particular v induces a topology on K. Two additive
valuations are equivalent if they induce the same topology on K. Then we get a bijection between
the equivalence classes of non-archimedean absolute values on K and equivalence classes of additive
valuations on K.
We record the following simple yet important observation.

Lemma 3.2.3. Let | · |v be a non-archimdean absolute value on a field K associated with an additive
valuation v. For any x, y ∈ K, if |x|v ̸= |y|v , then we have

|x + y|v = max{|x|v , |y|v }, v(x + y) = min{v(x), v(y)}.

Proof. We may assume that |x|v < |y|v . Suppose |x + y|v < max{|x|v , |y|v } = |y|v . Then we would
get
|y|v = |(x + y) − x|v ≤ max{|x + y|v , |x|v } < |y|v
which is absurd. Therefore we must have |x + y|v = max{|x|v , |y|v }.

49
 Let (K, | · |v ) be a nonarchimedean valued field. Consider the closed and open unit balls

Ov := {x ∈ K, |x|v ≤ 1}, mv := {x ∈ K, |x|v < 1}

When | · |v is discrete, we have seen that Ov is a DVR. In general one has a weaker result:
Lemma-Definition 3.2.4. The set Ov is a local integrally closed integral domain with fraction field
K, maximal ideal mv and unit group

Ov× = {x ∈ K, |x|v = 1}.

Moreover, any finitely generated ideal of Ov is principal. If | · |v is discrete, then Ov is a discrete

valuation ring and {mnv }n≥0 form a system of neighborhoods of 0 in Ov and K.
On the other hand if | · |v is non-discrete, then mv is not finitely generated and mv = mnv for all
integer n ≥ 1. In particular, Ov is non-noetherian.
We call the pair (Ov , mv ) the valuation ring of | · |v and κv := Ov /mv the residue field of | · |.
Proof. Since Ov is a subring of K, it is an integral domain. By definition, for any x ∈ K × we have
either x ∈ Ov or x−1 ∈ Ov and hence K is the fraction field of Ov . Clearly the unit group Ov× has
the given description and mv = Ov \Ov× is an ideal. Thus Ov is local with maximal ideal mv . Let
x ∈ K × be integral over Ov . Then there exists a1 , . . . , an ∈ Ov such that xn + a1 xn−1 + · · · + a0 = 0.
/ Ov , then x−1 ∈ Ov . Multiply the equation by x1−n ∈ Ov we get
If x ∈

x + (a1 + · · · + a0 x1−n ) = 0.

Since the expression in the bracket lies in Ov by assumption we get that x ∈ Ov , a contradiction.
Thus x ∈ Ov and Ov is integrally closed.
For any finitely generated ideal I = (x1 , . . . , xr ) ⊂ Ov . Let xi has the maximal absolute value
among the generators. Then I = (xi ) is principal.
When | · |v is discrete, then |K × |v = cZ for some 0 < c < 1. Then mnv = {x ∈ K, |x|v ≤ cn } and
they form a system of neighborhood of 0 in Ov and K.
Now assume | · |v is non-discrete. Then the value group |K × |v contains elements that are smaller
than and arbitrarily close to 1 and hence mv cannot be finitely generated. By the same reason
for any x ∈ mv there exists y ∈ mv such that |x|v < |y 2 |v . Then we get |xy −2 |v < 1 and hence
xy −2 ∈ Ov . But this implies that x = xy −2 · y 2 ∈ m2v . Therefore we have mv = m2v , which implies
that mnv = mn+1
v and hence mnv = mv for all n ≥ 1.
Remark 3.2.5. Examples of non-discrete non-archimedean valued fields include
∞ ∞
1 [ 1 [
Qp , , Cp , Qp (p p∞ ) = Qp (p pn ), Qp (ζp∞ ) := Qp (ζpn )
n=1 n=1

each equipped with the unique extension of the p-adic absolute value on Qp .

3.2.2 Valuations on the rational numbers

We let | · |∞ be the usual archimedean absolute value on Q. For each prime number p, let | · |p
be the p-adic absolute value on Q defined by |x| = p−vp (x) for any x ∈ Q× . These are mutually
non-equivalent: for two primes p ̸= q we have |p|p = p−1 < 1 while |p|q = 1 and |p|∞ = p > 1.
Theorem 3.2.6 (Ostrowski). Let | · | be a nontrivial absolute value on Q.
1. If |n| ≥ 1 for any integer n ≥ 1, then | · | is equivalent to the archimedean absolute value | · |∞ .
2. If there exists an integer n > 1 with |n| ≤ 1, then |m| ≤ 1 for all m ∈ Z and there exists a
unique prime number p such that |p| < 1 and | · | is equivalent to the p-adic absolute value | · |p .

50
In particular, any archimedean absolute value of Q is equivalent to | · |∞ and any nontrivial nonar-
chimedean absolute value on Q is equivalent to | · |p for some prime number p.
Proof. Let | · | be a nontrivial absolute value on Q. For any two natural numbers m > 1, n > 1,
there exists a unique integer r ≥ 0 such that nr ≤ m < nr+1 and hence r ≤ log m
log n . Then we can
write
m = a0 + a1 n + · · · ar nr , ai ∈ {0, 1, . . . , n − 1}, ∀0 ≤ i ≤ r.
By the triangle inequality we have

|ai | = |1 + · · · + 1| ≤ ai ≤ n, ∀0 ≤ i ≤ r.

Let N := max{1, |n|}. Then we get

r
X log m log m
|m| ≤ |ai | · |n|r ≤ (r + 1)nN r ≤ (1 + )nN log n .
i=0
log n

Replace m by mt for any integer t > 0 we get that

t log m log m
|m|t ≤ (1 + )nN t· log n
log n
and therefore
t log m 1 1 log m
|m| ≤ (1 + ) t n t N log n .
log n
Let t → +∞ we get that
log m
|m| ≤ N log n .
log m
If |n| ≥ 1 for all n ∈ Z>1 then N = max{1, |n|} = |n| and hence |m| ≤ |n| log n . Since the role of
m, n are symmetric this is always an equality and we deduce that
1 1
|m| log m = |n| log n , ∀m, n ∈ Z>1 .

Let c be this constant number. Since |n| ≥ 1 for all n ∈ Z we have c ≥ 1 and since | · | is nontrivial
we have furthermore c > 1 and hence log c > 0. Then we get that

|m| = clog m = mlog c = |m|log

∞ ,
c
∀m ∈ Z>1 .

By multiplicativity we deduce that for any α ∈ Q we have |α| = |α|log ∞

c
where log c > 0 is a constant.
Thus | · | is equivalent to the archimedean absolute value | · |∞ .
On the other hand, suppose |n| ≤ 1 for some n ∈ Z>1 . Then N = max{1, |n|} = 1 and we get
that |m| ≤ 1 for all m ∈ Z. In particular | · | is non-archimedean by Lemma 3.2.2. Then there exists
a prime number p with |p| < 1 since otherwise by unique factorization we would get |x| = 1 for all
x ∈ Q× . Let I := {m ∈ Z, |m| < 1}. Then one checks immediately that I is an ideal of Z and
pZ ⊂ I. Thus I = pZ since pZ is maximal and I ̸= Z (as 1 ∈ / I).
For any nonzero integer a ∈ Z we can write a = pvp (a) b where p ∤ b. Then b ∈ / I so that |b| = 1
and we get
|a| = |p|vp (a) = p−svp (a) = |a|sp
|p|
where s := − log
log p . This shows that | · | is equivalent to the p-adic absolute value | · |p .

Proposition 3.2.7. For any x ∈ Q× we have

Y
|x|v = 1
v≤∞

where the product runs over all the prime numbers, together with ∞

51
3.3 Completion
Definition 3.3.1. A valued field (K, | · |) is complete if any Cauchy sequence in K converges to a
(necessarily unique) element in K.
From any valued field (K, | · |), we construct its completion K̂ consisting of equivalence classes of
Cauchy sequences in K. Then K embeds canonically as a subfield of K̂ and |·| extends by continuity
to an absolute value on K̂, under which K̂ becomes a complete valued field.
Theorem 3.3.1 (Ostrowski). Let K be a field that is complete with respect to an archimedean
valuation | · |. The there exists an isomorphism σ from K to either R or C and a real number
0 < s ≤ 1 such that |x| = |σ(x)|s for all x ∈ K.
See [Neu99, II.4.2] for the proof. In the following we study complete non-archimedean valued
fields.
Lemma 3.3.2. Let (K, | · |v ) be a nonarchimedean valued field and let K̂ be its completion. For
any Cauchy sequence {xn }∞ n=1 in K we have |xn | = |xn+1 | for all n sufficiently large. Moreover any
element x ∈ K̂ can be represented by (i.e. is the limit of ) a Cauchy sequence {xn }∞ n=1 in K such
that |xn |v = |x|v for all n ≥ 1.
In particular, (K, | · |) and (K̂, | · |) have the same value group: |K × | = |K̂ × |.
Proof. Let x = {xn }∞n=1 ∈ K̂ be a nonzero element where xn ∈ K. For all n sufficiently large we
have |xn+1 − xn | < |xn | and then by Lemma 3.2.3 we get

|xn+1 | = |(xn+1 − xn ) + xn | = |xn |

Therefore for all n sufficiently large we have |xn |v = |x|v and after modifying finitely many terms
to get an equivalent Cauchy sequence also converging to x, we may assume that |xn |v = |x|v for all
n.
Lemma 3.3.3. Let (K, | · |v ) be a nonarchimedean valued field with valuation ring (Ov , mv ) and
residue field κv = Ov /mv . Let K̂ be its completion and let (Ôv , m̂v ) be the valuation ring of K̂.
Then the following holds:
1. mnv = m̂nv ∩ K for all integer n ≥ 0.
2. Ôv is the closure of Ov in K̂ and more generally m̂nv is the closure of mnv in K̂ for all integer
n ≥ 0.
3. The natural embedding Ov ,→ Ôv induces ring isomorphisms Ov /mnv ∼ = Ôv /m̂nv for all n. In
particular the residue fields of Ov and Ôv are canonically isomorphic.
4. If | · |v is discrete then we have a natural ring isomorphism

Ôv ∼
= lim O /mn
←− v v
n

while if | · |v is non-discrete then limn Ov /mnv ∼

= κv = Ov /mv
←−
Proof. By Lemma 3.2.4 we see that for all n ≥ 0, mnv and m̂nv are open balls in K (resp. K̂) of the
same radius. Then (1) follows since the metric on K̂ restricts to the metric on K and (2) follows
from Lemma 3.3.2.
From (2) we get that Ov + m̂nv = Ôv for all n. Combined with (1) we deduce (3).
Finally we prove (4). If |·|v is non-discrete then we conclude by (3) and Lemma 3.2.4. Now assume
that | · |v is discrete. By (3) it suffices to show that the natural homomorphism Ôv → limn Ôv /m̂nv
←−
is an isomorphism. The injectivity is clear since ∩∞ n
n=1 m̂v = {0} by the axiom of an absolute value,
while the surjectivity follows from the following Lemma.

52
Lemma 3.3.4. Let (K, |·|v ) be a non-archimedean discrete valued field with valuation ring (Ov , mv ).
Let S ⊂ Ov be a set of representatives of the residue field κv = Ov /mv . Let ϖ ∈ mv be a generator.
Pn−1
1. For any n ≥ 1, any class in Ov /mnv has a unique representative in Ov of the form i=0 ai ϖi ,
in which ai ∈ S for all 0 ≤ i ≤ n − 1.
2. If moreover (K, | · |v ) is complete, then any x ∈ K̂ × can be written uniquely as
∞
X
x = ϖm ai ϖi
i=0

where m ∈ Z, ai ∈ S for all i and a0 ̸= 0.

Proof. 1. We prove by induction on n. The case n = 1 is clear. Assume the statement is true
for n − 1. For any class in Ov /mnv represented by an x ∈ Ov , there exists a unique a0 ∈ S such
that x ≡ a0 mod mv . Let y = ϖ−1 (x − a0 ) ∈ Ov . By induction hypothesis there exists unique
a1 , . . . , an−1 ∈ S such that

y ≡ a1 + a2 ϖ + · · · + an−1 ϖn−2 mod mn−1

v
Pn−1
Then we get that x ≡ i=0 ai ϖi mod mv and a0 , . . . , an−1 ∈ S are uniquely determined by the
class of x in Ov /mnv .
2. There exists a unique m ∈ Z such that ϖ−m x ∈ Ô× . Then the statement follows from the
first part.

3.3.1 Factorizing polynomials in complete nonarchimedean fields

Theorem 3.3.5. Let (K, | · |) be a complete non-archimedean valued field with valuation ring (O, m)
and residue field κ := O/m. Let f (X) ∈ O[T ] be a polynomial. Suppose there exists a0 ∈ O such
that
|f (a0 )| < |f ′ (a0 )|2
Then there exists a unique root a ∈ O of f such that |a − a0 | < |f ′ (a0 )| ≤ 1. Moreover we have

f (a0 )
|a − a0 | = .
f ′ (a0 )

Proof. First we prove uniqueness. Suppose a, a′ ∈ O both satisfies the requirements. Then we have
|a − a0 | < |f ′ (a0 )| and |a′ − a0 | < |f ′ (a0 )| which implies that

|a − a′ | < |f ′ (a0 )|.

Since f ′ (a) − f ′ (a0 ) ∈ (a − a0 )O we have |f ′ (a) − f ′ (a0 )| ≤ |a − a0 | < |f ′ (a0 )| and hence |f ′ (a)| =
|f ′ (a0 )| by Lemma 3.2.3. By Taylor expansion we have

0 = f (a′ ) − f (a) = f ′ (a)(a′ − a) + c(a′ − a)2 , c ∈ O.

Suppose a′ ̸= a, then we deduce that

|f ′ (a0 )| · |a′ − a| = |f ′ (a)| · |a′ − a| = |c(a′ − a)2 | ≤ |a′ − a|2 < |f ′ (a0 )| · |a′ − a|.

This is a contradiction and thus we must have a′ = a and this shows uniqueness.
To show the existence we first consider the special case where |f ′ (a0 )| = 1 so that |f (a0 )| < 1.
f (an )
We define a sequence {an }∞ n=0 inductively by an+1 := an − f ′ (an ) and we will show by induction
that for all n ≥ 0 n
|f (an )| ≤ |f (a0 )|2 and |f ′ (an )| = |f ′ (a0 )| = 1.

53
The case n = 0 is clear by assumption. Suppose the identities are true for n. By Taylor expansion
(more precisely, binomial expansion for polynomials) we get that
   2  2
′ f (an ) f (an ) f (an )
f (an+1 ) ∈ f (an ) + f (an ) − ′ + − ′ O= O
f (an ) f (an ) f ′ (an )
Then by induction hypothesis we get
2
f (an ) n+1
|f (an+1 )| ≤ = |f (an )|2 ≤ |f (a0 )|2 .
f ′ (an )
f (an )
Since f ′ (an+1 ) − f ′ (an ) ∈ (an+1 − an )O = f ′ (an ) O, we get by induction hypothesis that

f (an ) n
|f ′ (an+1 ) − f ′ (an )| ≤ = |f (an )| ≤ |f (a0 )|2 < 1.
f ′ (an )

Therefore by the induction hypothesis |f ′ (an )| = 1 and Lemma 3.2.3 we get |f ′ (an+1 )| = 1. This
finishes the proof of the identities above. As a consequence we get
|f (an )| n
|an+1 − an | = ′
= |f (an )| ≤ |f (a0 )|2
|f (an )|

which decreases and goes to 0 as n → ∞ since |f (a0 )| < 1 by assumption.

Therefore {an }∞
n=1 is a Cauchy sequence in O and since O is complete it converges to an element
a ∈ O. We have n
|f (a)| = lim |f (an )| ≤ lim |f (a0 )|2 = 0
n→∞ n→∞

and thus f (a) = 0. Moreover, for each n ≥ 1 we have

n−1
X
|an − a0 | = (ai+1 − ai ) ≤ max {|ai+1 − ai |} ≤ |f (a0 )|.
0≤i≤n−1
i=0

After taking limit when n → ∞ we get |a − a0 | ≤ |f (a0 )| < 1. This proves the special case where
|f ′ (a0 )| = 1, except that we only get |a − a0 | ≤ |f (a0 )| rather than equality.
f (a0 )
Finally we prove existence in the general. By assumption we have f ′ (a0 ) < |f ′ (a0 )| ≤ 1. Then
by Taylor expansion (or actually binomial expansion) we have an identities of polynomials in O[X]:
 
f (a0 ) f (a0 ) 2
f a0 − ′ X = f (a0 )(1 − X + ′ X h(X)) = f (a0 )g(X)
f (a0 ) f (a0 )2
f (a0 ) 2
where h(X) ∈ O[X] and g(X) = 1 − X + f ′ (a0 )2 X h(X) ∈ O[X].
We observe that |g(1)| ≤ | ff′ (a
(a0 )
2|
′
< 1 and |g (1)| = 1. By the special case we just proved there
0)
exists b ∈ O such that g(b) = 0 and |b − 1| ≤ |g(1)| < 1. In particular we get |b| = 1 by Lemma
3.2.3. Let a := a0 − ff′(a0)
(a0 ) b. Then we get that f (a) = 0 and

f (a0 ) f (a0 )
|a − a0 | = b = ′ < |f ′ (a0 )| ≤ 1.
f ′ (a0 ) f (a0 )

Example 3.3.1. Consider the case O = Z2 and f (X) = X 2 + 7. Let a0 = 1. Then |f (a0 )|2 = 2−3
and |f ′ (a0 )|2 = 2−1 . Then the condition of the theorem is satisfied and we get an element a ∈ Z2
such that a2 + 7 = 0 and |a − 1|2 = 2−2 . In other words, a − 1 ∈ 4Z× 2.

54
Definition 3.3.2. A polynomial f (X) ∈ O[X] is primitive if the ideal generated by its coefficients
is the unit ideal O.
For example, a monic polynomial in O[X] is primitive.
Theorem 3.3.6 (Hensel’s lemma). Let (K, | · |) be a complete non-archimedean valued field with
valuation ring (O, m) and residue field κ := O/m. Let f (X) ∈ O[X] be a polynomial. Suppose there
is a factorization
f (X) ≡ ḡ(X)h̄(X) mod m
where ḡ(X), h̄(X) ∈ κ[X] are relatively prime and ḡ ̸= 0. Then there exists g(X), h(X) ∈ O[X] with
g(X) ≡ ḡ(X) mod m and h(X) ≡ h̄(X) mod m such that f (X) = g(X)h(X) and deg g = deg ḡ.
Proof. If f is not primitive, then f ≡ 0 mod m and the hypothesis implies that h̄ = 0 and ḡ is a
nonzero constant. Then we take g = λ ∈ O× lifting ḡ and h = λ−1 f .
Now assume that f is primitive so that ḡ ̸= 0 and h̄ ̸= 0. Let d = deg(f ) and m = deg(ḡ). Then
we have deg(h̄) ≤ d − m. Choose g0 , h0 ∈ O[X] such that
g0 ≡ ḡ mod m, h0 ≡ h̄ mod m
and deg g0 = deg ḡ = m, deg h0 = deg h̄ ≤ d − m. Since ḡ and h̄ are relatively prime, there exists
a(X), b(X) ∈ O[X] such that
ag0 + bh0 ≡ 1 mod m.
Then we have ag0 + bh0 − 1 ∈ m[X] and f − g0 h0 ∈ m[X]. By Lemma 3.2.4 the coefficients of
these two polynomials generate a principal ideal πO ⊂ m. To finish the proof, it suffices to find
g, h ∈ O[X] of the form
g = g0 + p1 π + p2 π 2 + · · · , h = h0 + q1 π + q2 π 2 + · · ·
such that deg pi < m, deg qi ≤ d − m and f = gh.
Suppose we have found pi , qi for all 1 ≤Pi ≤ n − 1 such that deg pi P
< m, deg qi ≤ d − m and
n−1 n−1
f ≡ gn−1 hn−1 mod π n where gn−1 = g0 + i=1 pi π i and hn−1 = h0 + i=1 qi π i . Then we need
to find pn , qn ∈ O[X] such that deg pn < m, deg qn ≤ d − m and
(gn−1 + pn π n )(hn−1 + qn π n ) ≡ f mod π n+1 .
Let fn := π −n (f − gn−1 hn−1 ) ∈ O[X]. Since hn−1 ≡ h0 mod π, gn−1 ≡ g0 mod π, the congruence
above is equivalent to
pn h0 + qn g0 ≡ fn mod π. (3.3.1)
We have the congruence
fn ag0 + fn bh0 ≡ fn mod π.
If we simply take qn = fn a, pn = bfn , the congruence (3.3.1) would be satisfied but we may lose
control of the degree so that the resulting series g, h might be a power series in X, rather than a
polynomial. To avoid this possibility we proceed as follows.
Write bfn = qg0 + pn where q, pn ∈ K[X] and deg pn < deg(g0 ) = m. Since the highest degree
term of g0 lies in O× by assumption, we have actually q, pn ∈ O[X]. Then we get
(afn + qh0 )g0 + pn h0 ≡ fn mod π.
Ignore all terms in afn + qh0 whose coefficient is divisible by π and let qn ∈ O[X] by the remaining
terms. Then (3.3.1) is satisfied. Moreover, we have deg(pn h0 ) < m + d − m = d, deg g0 = m and
deg fn = d. Then we get that deg qn = deg q̄n ≤ d − m.
Let gn := gn−1 + pn π n , hn := hn−1 + qn π n . Then {gn }∞ ∞
n=0 and {hn }n=0 are sequences of
polynomials with bounded degree and the coefficients of their terms of fixed degree form a Cauchy
sequence. Then they converge to polynomials g, h ∈ O[X]. Since gn hn ≡ f mod π n+1 we get
f = gh.

55
Remark 3.3.7. In general it may happen that deg(f¯) < deg f so we cannot expect deg g = deg ḡ
and deg h = deg h̄ to be both true. But the theorem says that at least one of them is true, say
deg g = deg ḡ. So in general we might get two essentially different factorization from the theorem
(except the possibility of multiplying by scalars), one in which deg ḡ = deg g, and the other in which
deg h̄ = deg h.
Consider the example K = Q3 and f (X) = 9X 4 + X 3 − 3. We have f¯ = X 3 = ḡ h̄ where ḡ(X) =
3
X and h̄(X) = 1. One possible lifting is h(X) = 1 and g(X) = f (X) so that deg h = deg h̄ = 0.
In the other possible lifting we have deg g = deg ḡ = 3 and deg h(X) = 1. From h ≡ h̄ mod 3 one
can check that h(X) = αX + 1 for some α ∈ Z3 with v3 (α) = 2. In particular we get that f (X) has
a root −α−1 ∈ Q3 . Note however that f does not have any root in Z3 since X 3 − 3 does not have
any root in Z/9Z.
Definition 3.3.3. Let (K, | · |) be a non-archimedean valued field. For all polynomial f (X) =
a0 + a1 X + · · · + an X n ∈ K[X], define its norm to be

∥f ∥ = max {|ai |}.

0≤i≤n

In particular, f (X) ∈ O[X] is primitive if and only if ∥f ∥ = 1.

Corollary 3.3.8. Let (K, | · |) be a complete nonarchimedean valued field. Let f (X) = a0 + a1 X +
· · ·+an X n ∈ K[X] be an irreducible polynomial with a0 an ̸= 0. Then we have ∥f ∥ = max{|a0 |, |an |}.
Proof. We may assume that f (X) ∈ O[X] and ∥f ∥ = 1. Let r be the minimal index such that
|ar | = ∥f ∥ = 1. Then we have

f (X) ≡ X r (ar + ar+1 X + · · · + an X n−r ) mod m.

Apply Hensel’s Lemma (Theorem 3.3.6) to ḡ(X) = X r and h̄(X) = (ar + · · · + an X n−r ) mod m
(they are relatively prime since |ar | = 1) to get g, h ∈ O[X] such that deg g = deg ḡ = r and
f (X) = g(X)h(X). If 0 < r < n, then f would not be irreducible. Then we must have r = 0 or
r = n.
As a consequence, we obtain the following result that generalize and improves Proposition 2.1.1
in the setting of complete non-archimedean field.
Corollary 3.3.9. Let (K, | · |) be a complete nonarchimedean valued field with valuation ring O.
Let L/K be a finite extension and let O
e be the integral closure of O in L. Then we have

O
e = {α ∈ L | NL/K (α) ∈ O}.

Proof. First we show the inclusion “⊂”. This part is true without the completeness assumption. But
the current setting is not exactly the same as in Proposition 2.1.1 (since | · | might be non-discrete
so that O might be non-noetherian, hence not Dedekind), we cannot directly apply Proposition
2.1.1. Instead we argue directly as follows. Let K sep be a separable closure of K and let M be the
separable closure of K in L, then for any α ∈ L we have α[L:M ] ∈ M and
 
Y
NL/K (α) =  τ (α[L:M ] ) .
τ ∈HomK (M,K sep )

If α is integral over O, then each τ (α[L:M ] ) is integral over O and hence NL/K (α) is integral over
O. But we know that O is integral closed by Lemma 3.2.4, therefore we have NL/K (α) ∈ O.
Next we show the reverse inclusion “⊃”. Take any α ∈ L with NL/K (α) ∈ O. Consider its
minimal polynomial f (X) = X m + am−1 X m−1 + · · · + a0 ∈ K[X]. Then f (X) is irreducible
and NL/K (α) is a power of a0 . Hence |a0 | ≤ 1 and by Corollary 3.3.8 we get ∥f ∥ ≤ 1 so that
f (X) ∈ O[X]. This shows that α ∈ O. e

56
Theorem 3.3.10. Let (K, | · |) be a complete valued field. Then | · | can be extended uniquely to
any algebraic extension L/K. If the extension L/K is finite of degree n, then L is complete and the
extended absolute value is given by
1
|α| := |NL/K (α)| n , ∀α ∈ L.

Proof. If | · | is archimedean, we conclude by Ostrowski’s Theorem 3.3.1. Now we assume that | · | is

nonarchimedean and it suffices to consider the case where L/K is a finite extension of degree n since
any algebraic extension is a union of finite subextensions. Let O be the valuation ring of (K, | · |)
and let Oe be the integral closure of O in L.
1
First we prove existence of extension by showing that the map α 7→ |NL/K α| n defines a non-
archimedean absolute value on L. It clearly extends the absolute value | · | on K (because of the
power n1 ). The only non-trivial property to check is the ultra-metric inequality. Take any α, β ∈ L
with
1 1
|NL/K α| n ≤ |NL/K β| n ̸= 0.

Then we have NL/K (αβ −1 ) ∈ O and by Corollary 3.3.9 we get that αβ −1 ∈ O. e Then we get
1
1 + αβ −1 ∈ O
e and by Corollary 3.3.9 again we get |NL/K (1 + αβ −1 )| n ≤ 1 and hence

1 1 1 1
|NL/K (α) + NL/K (β)| n ≤ |NL/K β| n = max{|NL/K α| n , |NL/K β| n }.

Next we prove uniqueness. Let | · |′ be any absolute value on L that extends | · | on K. Then | · |′
is non-archimedean by Lemma 3.2.2 (since L and K have the same prime ring). Let O′ be the
valuation ring of (L, | · |′ ). Then O′ is integrally closed by Lemma 3.2.4 and therefore O
e ⊂ O′ . Then
for any α ∈ L× with |α|′ < 1 we have α−1 ∈ / O′ and hence α−1 ∈/ O.
e By Corollary 3.3.9 we get
1
|NL/K (α)| < 1 and hence |NL/K (α)| n < 1. Therefore by Lemma 3.2.1 there exists c > 0 such that
c
|α|′ = |NL/K (α)| n ∀α ∈ L.

If | · | is nontrivial, take an α ∈ K × with |α| ̸= 1 we see that c = 1. If | · | is trivial then | · |′ is also

trivial. This shows uniqueness of extension.
Finally L is complete under the extended absolute value by the following general result in func-
tional analysis.
Proposition 3.3.11. Let (K, | · |) be a complete valued field and let V be finite dimensional normed
K-vector space. Then V is complete and for any K-basis {v1 , . . . , vn } the norm on V is equivalent
to the maximal norm defined by
n
X
xi vi := max {|xi |}, ∀x1 , . . . , xn ∈ K.
1≤i≤n
i=1

Definition 3.3.4. A norm on a vector space V over a valued field (K, |·|) is a function ∥·∥ : V → R≥0
that satisfies the following conditions:

• ∥v∥ = 0 if and only if v = 0;

• For any λ ∈ K and v ∈ V we have ∥λv∥ = |λ| · ∥v∥;
• For any v1 , v2 ∈ V we have ∥v1 + v2 ∥ ≤ ∥v1 ∥ + ∥v2 ∥.
Two norms ∥ · ∥ and ∥ · ∥′ on V are equivalent if there exists c1 , c2 > 0 such that

c1 ∥v∥ ≤ ∥v∥′ ≤ c2 ∥v∥, ∀v ∈ V.

57
Proof. We prove by induction on the dimension n = dimK V . The case n = 1 is clear and we show
the induction step. Let {v1 , . . . , vn } be a K-basis of PVn and let ∥ · ∥ be the associated maximal norm
on V . Let ∥ · ∥′ be any norm on V and set c2 := i=1 ∥vi ∥′ . For any v = x1 v1 + · · · xn vn ∈ V we
have
X n
∥v∥′ ≤ |xi | · ∥vi ∥′ ≤ c2 max {|xi |} = c2 ∥v∥.
1≤i≤n
i=1

Suppose there does not exist any c1 > 0 such that ∥v∥′ ≥ c1 ∥v∥ for all v ∈ V . Then we can find a
Pn (j)
sequence {v (j) = i=1 xi vi }∞
j=1 in V such that ∥v
(j) ′
∥ < 1j ∥v (j) ∥ for all j ≥ 1. After extracting a
(j)
subsequence and reordering the basis if necessary, we may assume that ∥v (j) ∥ = |xn | for all j ≥ 1.
(j) (j)
Dividing each v (j) by xn we may assume further that xn = 1 for all j ≥ 1 and also
1
∥v (j) ∥ = 1 ∥v (j) ∥′ < , ∀j ≥ 1.
j
Let W ⊂ V be the subspace spanned by v1 , . . . , vn−1 . For each j ≥ 1, consider the vector
n−1
(j)
X
w(j) := xi vi = v (j) − vn ∈ W.
i=1

When j, ℓ → ∞ we have
∥w(j) − w(ℓ) ∥′ = ∥v (j) − v (ℓ) ∥′ → 0.
By induction hypothesis there exists c > 0 such that
(j) (ℓ)
max {|xi − xi |} ≤ c∥w(j) − w(ℓ) ∥′
1≤i≤n−1

(j)
Therefore {xi }∞
j=1 is a Cauchy sequence in K for all 1 ≤ i ≤ n − 1 and by completeness of K they
(j)
have limits xi = limj→∞ xi ∈ K. Then we get that
n−1 ′ n−1 n−1 ′
(j) (j)
X X X
x i vi + vn = (xi − xi )vi + x i v i + vn
i=1 i=1 i=1
n−1
(j)
X
≤ |xi − xi | · ∥vi ∥′ + ∥v (j) ∥′ → 0, when j → ∞.
i=1

Pn−1
Then we must have i=1 xi vi + vn = 0, contradicting the fact that {v1 , . . . , vn } form a K-basis of
V.
Exercise 3.3.1. Let (K, | · |) be a complete nontrivially valued field and let L/K be an infinite
algebraic extension. Show that L is not complete under the unique extension of | · |.

3.4 Local and global fields

3.4.1 Characterizing local fields
We take the following as our definition of a “local field”2 .
2 Insome literature (e.g. [Ser79]), the terminology “local fields” means complete discrete valued field (and some
authors require moreover that the residue field is perfect). These excludes the archimedean fields, but include more
general non-archimedean fields, e.g. C((T )). We follow the convention in [Wei95]. The convention in [Neu99] is close
to ours but Neukirch excludes archimedean local fields

58
Definition 3.4.1. A local field is a locally compact topological field whose topology is induced by
a nontrivial absolute value. A local field is non-archimedean (resp. archimedean) if its topology is
induced by a non-archimedean (resp. archimedean) absolute value.
In particular, the underlying additive group of a local field is a locally compact Hausdorff topo-
logical group.
Lemma 3.4.1. Let K be a local field and | · | an absolute value on K that induces its topology. Then
(K, | · |) is complete and non-compact.
Proof. Let (K̂, | · |) be the completion of (K, | · |). Then K̂ is Hausdorff since it is a metric space.
By Proposition B.1.5 we get that K is closed in K̂ and therefore K = K̂.
Since | · | : K → R≥0 is continuous, if K is compact then |K| is bounded but this is impossible
when | · | is nontrivial. Thus K is non-compact.
Remark 3.4.2. “Completeness” is a property of the metric while “local compactness” is property
of the underlying topology. In general there is no relation between these properties. For example
the open interval (0, 1) in R is locally compact but not complete under the usual metric. But on the
other hand (0, 1) is homeomorphic to R, which is complete. Also Cp is complete under the p-adic
metric but not locally compact (by Theorem 3.4.5).
Corollary 3.4.3. A local field whose topology is induced by an arhimedean absolute value is iso-
morphic (as a topological field) to either R or C.
Proof. Combine Lemma 3.4.1 and Theorem 3.3.1.
Lemma 3.4.4. Let (K, | · |) be a complete nontrivially valued field. Let V be a normed vector space
over K. If V is locally compact, then V is finite dimensional and complete.
Proof. Let C be a compact neighborhood of 0 in V and let α ∈ K be an element with 0 < |α| < 1.
Then there exists x1 , . . . , xn ∈ C such that C ⊂ ∪ni=1 (xi + αC). Let V ′ be the finite dimensional
subspace of V spanned by x1 , . . . , xn . Then we have C + V ′ ⊂ αC + V ′ and we get inductively that
C + V ′ ⊂ αn C + V ′ for all n ≥ 1. For any element v ∈ C we get a sequence {vn }∞ ′
n=1 in V such
n ∞
that v − vn ∈ α C for all n ≥ 1. In particular ∥v − vn ∥ → 0 as n → ∞. Then {vn }n=1 is a Cauchy
sequence in V ′ . Since V ′ is finite dimensional, it is complete by Propotision 3.3.11. Therefore
{vn }∞ ′ ′
n=1 converges in V ⊂ V and the limit must be v. This shows that C ⊂ V . In general for any
v ∈ V we have α v ∈ C for m sufficiently large and hence v ∈ α C ⊂ α V = V ′ . Thus V = V ′
m −m −m ′

is finite dimensional and by Proposition 3.3.11 it is complete.

Theorem 3.4.5. Let (K, | · |) be a non-archimedean valued field. The following are equivalent:
1. K is a nonarchimedean local field in the sense of Definition 3.4.1;
2. K is complete, | · | is discrete nontrivial and the residue field is a finite field.
3. K is a finite extension of Qp or Fp ((T )) for some prime number p.
Proof. (1)⇒(3): Let K be a non-archimedean local field. Then K is complete by Lemma 3.4.1. If
char(K) = 0, then Q ⊂ K and the restriction of | · | to Q is equivalent to the p-adic absolute value
for some prime number p by Theorem 3.2.6. By completeness of K we get that Qp ⊂ K and hence
K is finite a finite extension of Qp by Lemma 3.4.4.
If char(K) = p > 0. Then Fp ⊂ K. Take an element t ∈ K with 0 < |t| < 1. Then t is not
algebraic over Fp since otherwise Fp (t) would be a finite field and t would be a root of unity so that
|t| = 1. Then we get a field embedding Fp (T ) ⊂ K sending the indeterminate T to the element t that
is transendental over Fp . The restriction of | · | to Fp (T ) is equivalent to the absolute value defined
by the T -adic valuation. Since K is complete we get a field embedding Fp ((T )) ⊂ K. Therefore K
is a finite extension of Fp ((T )) by Lemma 3.4.4.

59
 (3)⇒(2): Let F = Qp or Fp ((t)) and let K be finite extension of F . By Theorem 3.3.10 the
absolute value on F extends uniquely to a discrete nontrivial absolute value on K and K is complete
under the extended valuation by Proposition 3.3.11. Let (O, m) be the valuation ring of K and let
κ = O/m be the residue field. Let a1 , . . . , am be a subset of κ that is linearly independent over Fp .
Let xi ∈ O be a lift of ai for each i. Then x1 , . . . , xm is linearly independent over F . Therefore
m ≤ [K : F ] and hence κ is a finite extension of Fp . In particular κ is a finite field.
(2)⇒(1): By assumption we have isomorphism of topological rings O = Ô ∼ = lim O/mn and
n
←−n
O/m is a finite set for any n. Therefore O is a profinite set and hence compact.
This finishes the proof. We also give alternative proofs of some implications that have indepen-
dent interest.
(1)⇒(2): Suppose (K, | · |) is a non-archimedean local field. By Lemma 3.4.1 it is complete. Let
(O, m) be its valuation ring. Choose a nonzero element π ∈ m. Then π n O form a system of open
neighborhood of 0 in K. Let C be a compact neighborhood of 0. Then π n O ⊂ C for n sufficiently
large. Since π n O is also closed subset of C it is compact. Therefore O is also compact (since it
is homeomorphic to π n O for any n). Let S ⊂ O be a set of representatives of the residue field
κ = O/m. Then we have open cover O = ⊔x∈S (x + m) and since O is compact κ must be finite.
Since m is an open subgroup of O it is also closed and hence compact. If | · | is non-discrete, we
can find a sequence {xn }∞ n=1 in m such that |xn | < |xn+1 | for all n ≥ 1 and limn→∞ |xn | = 1.
Then we get an open cover m = ∪∞ n=1 xn O that does not have any finite subcover, contradicting the
compactness of m. Therefore (K, | · |) is a complete discrete valued field with finite residue field.
(2)⇒(3): Suppose (K, | · |) is a complete discrete nontrivially valued field with valuation ring
(O, m) such that the residue field κ = O/m is a finite field.
If char(K) = 0 then Q ⊂ K and therefore Qp ⊂ K for some prime p. Let π ∈ m be a generator.
There exists an integer e ≥ 1 such that pO = π e O. Let f = [κ : Fp ]. Then O/pO = O/ϖe O is a finite
dimensional Fp vector space of dimension n := ef . Choose x1 , . . . , xn ∈ O such that their image in
O/pO form an Fp -basis. Let M ⊂ O be the Zp submodule generated by x1 , . . . , xn . Then we have
O = M + pO from which we deduce that O = M + pn O for any n ≥ 1. In particular M is dense
in O. On the other hand M is complete by Proposition 3.3.11. Thus we have O = M = ⊕ni=1 Zp xi
and hence K = O[ p1 ] has dimension n over Qp .
Now assume char(K) = p > 0. By assumption the residue field κ is a finite extension of Fp .
Write κ = Fp (a) and let f (T ) ∈ Fp [T ] ⊂ K[T ] be the minimal polynomial of a. Then f splits in
κ[T ] and by Hensel’s lemma it also splits in K[T ]. Therefore we get a field embedding κ ⊂ K and
by Lemma 3.3.4 we get an isomorphism of topological fields K ∼ = κ((T )) and in particular K is a
finite extension of Fp ((T )).

3.4.2 Global fields

Definition 3.4.2. A place (or prime) of a field K is an equivalence class of nontrivial absolute
values on K. A place is archimedean (resp. non-archimedean) if it consists of archimedean (resp.
non-archimedean) absolute values.
Definition 3.4.3. A global field is a finite extension of either Q or Fp (T ) for some prime number p.
Lemma 3.4.6. Let K be a global field. Then the completion of K under any nontrivial absolute
value is a local field. In particular, any nontrivial non-archimedean absolute value on K is discrete.
Moreover, there is a natural bijection between places of K and equivalence classes of field embed-
dings ι : K ,→ L into a local field L such that ι(K) is dense in L. Two such embeddings ι : K ,→ L
and ι′ : K ,→ L′ are equivalent if there is an isomorphism of topological fields φ : L ∼
= L′ such that
′
ι = φ ◦ ι. The map sends any nontrivial absolute value | · | on K to the natural embedding K ,→ K̂
where K̂ is the completion of K under | · |. The inverse map sends an embedding ι : K ,→ L to the
restriction of the absolute value of L on K.
Proof. Let | · | be any nontrivial absolute value on K and let K̂ be the completion of K under | · |.
Let F = Q or Fp (t) so that K is a finite extension of F . Then | · | restricts to an absolute value | · |v

60
on F and the completion Fv of F under | · |v is a local field3 . Then the composition KFv in K̂ is
a finite extension of Fv and therefore complete by Proposition 3.3.11. Thus we get K̂ = KFv is a
finite extension of Fv and hence it is a local field.
The map described is clearly injective. Next we check that it is also surjective. Let ι : K ,→ L
be an embedding into a local field L with dense image. Let | · |L be an absolute value that induces
its topology and let | · | be its restriction to K. Let K̂ be the completion of K under | · |. Then ι
extends to an isomorphism of topological fields K̂ ∼ = L and this shows surjectivity.
Let us give a more down to earth description of places of a global field.

Lemma-Definition 3.4.7. Let K be a number field. There is a bijection between archimedean

places of K and the set of Gal(C/R)-orbits on HomQ (K, C). The orbits that consists of singletons
are elements in HomQ (K, R) and these are called the real places. The orbits that consist of conjugate
pairs of non-real embeddings K ,→ C are called complex places.

Proof. The archimedean local fields are either R or C and to determine the equivalences of embed-
dings we have to figure out the automorphisms of R and C as topological fields (not just abstract
fields).
The only automorphism of R as a topological field is the identity (actually the only automorphism
of R as abstract field is also identity). Any field automorphism of C fixes elements in Q and if it is
an automorphism of topological field, then it is also continuous and hence fixes elements in R. Thus
the automorphisms of C as topological fields are Gal(C/R).
Remark 3.4.8. The group Aut(C/Q) of field automorphisms of C is extremely huge. It surjects
onto the absolute Galois group Gal(Q/Q) where Q is the algebraic closure of Q in C. On the other
hand, Aut(R/Q) = {1} as one can see by noting that any field automorphism of R must preserve the
order on R (since the order is determined by the algebra structure: x > y0 if and only if x − y = a2
for some a ∈ R× ).
Lemma 3.4.9. Let K be a number field with ring of integers OK . For any nonzero prime ideal
p ⊂ OK and any α ∈ K × , let
|α|p := (N p)−vp (α)
where vp (α) is the exponent of p in the unique factorization of the fractional ideal αOK . Then | · |p
defines a non-archimedean absolute value on K whose restriction to Q is equivalent to | · |p where
p = char(A/p). This defines a bijection between non-archimedean places of K and nonzero prime
ideals of OK , where the inverse map is defined by sending a nonarchimedean absolute value | · | on
K to {x ∈ OK , |x| < 1}.
Proof. By definition, for any α ∈ K we have |α|p ≤ 1 if and only if α ∈ OK,p , the localization of
OK at p. From this one easily deduce that | · |p is a discrete nontrivial non-archimedean absolute
value on K with valuation ring OK,p . Since OK,p ∩ Z = Z(p) where p = char(OK /p) we see that the
restriction of | · |p to Q is equivalent to | · |p .
Conversely let | · | be a nontrivial non-archimedean absolute value on K and let (O, m) be its
valuation ring. Then the restriction of | · | to Q is equivalent to the p-adic absolute value | · |p for
some prime number p and in particular Z ⊂ O. Since O is integrally closed by Lemma 3.2.4 we have
OK ⊂ O. Then p := m ∩ OK is a nonzero prime ideal of OK containing pZ and we have OK,p = O,
pOK,p = m. Therefore | · | is equivalent to | · |p by Lemma 3.2.1.
Proposition 3.4.10. Let p be a prime number and let K be a finite extension of Fp (T ). Then K is
the rational function field of a smooth projective geometrically connected curve X over a finite field
κ of characteristic p. All places of K are non-archimedean and there is a natural bijection between
the set of places of K and the set of closed points of X.
3 For F = Fp (T ) we take this fact for granted. Maybe add the proof somewhere later

61
 For each closed point x ∈ X, let | · |x be a corresponding absolute value on K. Then | · |x is
discrete nontrivial and the valuation ring can be identified with the stalk OX,x of the structure sheaf
OX at x.

3.4.3 Local fields are completions of global fields

We have seen that the completions of global fields under any nontrivial absolute values are local
fields. Now we address the question whether any local field arise as the completion of a global field.
This is clear for R, C or κ((t)) where κ is a finite field. So we consider the case where F is a finite
extension of Qp .
Theorem 3.4.11 (Krasner’s Lemma). Let (K, | · |) be a non-archimedean valued field and let K be
the algebraic closure of K. Suppose that4 the absolute value | · | on K extends uniquely to an absolute
value on K, denoted also by | · |. Let α, β ∈ K be two elements such that
• α is separable over K(β) and
• For any σ ∈ Aut(K/K) with σ(α) ̸= α we have |β − α| < |α − σ(α)|.
Then we have α ∈ K(β).
Proof. Since α is separable over K(β) it suffices to show that τ (α) = α for any τ ∈ Aut(K/K(β)).
Suppose on the contrary that τ (α) ̸= α. Then by uniqueness of the extension of | · | we have
|β − α| = |τ (β) − τ (α)| = |β − τ (α)| ≤ max{|β − α|, |α − τ (α)|}
By assumption we have |β − α| < |α − τ (α)|. But then by Lemma 3.2.3 the equality holds and we
get |β − α| = |α − τ (α)| which is a contradiction. Therefore α ∈ K(β).
Proposition 3.4.12. Let f ∈ K[X] be a monic separable polynomial of degree n. Fix ε0 > 0 such
that for any distinct roots α, α′ ∈ K of f we have |α′ − α| ≥ ε0 (if there is only one root we take any
ε0 ). Then for all 0 < ε < ε0 there exists δ > 0, depending only on f and ε, such that for any monic
degree n polynomial g ∈ K[X] with ∥g − f ∥ < δ and any root β ∈ K of g there exists a unique root
α ∈ K of f such that |α − β| < ε. Furthermore we have α ∈ K(β). If moreover f is irreducible,
then g is irreducible and K(α) = K(β) (in particular, g is separable).
Qn
Proof. Let α1 , . . . , αn be the distinct roots of f in K so that f (X) = i=1 (X − αi ). By assumption
ε0 ≤ |αi − αj | for all i ̸= j. Let C := max{1, ∥f ∥n−1 }. For any 0 < ε < ϵ0 , let δ = min{∥f ∥, C −1 εn }.
Let g ∈ K[X] be monic degree n such that ∥f − g∥ < δ and let β ∈ K be a root of g. Since
∥f − g∥ < δ ≤ ∥f ∥ we have ∥g∥ = ∥f ∥ by Lemma 3.2.3. Write g(X) = X n + b1 X n−1 + . . . bn . From
g(β) = 0 we get
|β|n ≤ max {|bi | · |β|n−i } = |bj | · |β|n−j
1≤i≤n
j
for some 1 ≤ j ≤ n and hence |β| ≤ |bj | ≤ ∥g∥ = ∥f ∥. Then we deduce that
n
Y
(β − αi ) = |f (β)| = |f (β) − g(β)| ≤ ∥f − g∥ max{1, |β|n−1 }
i=1
n−1
≤ ∥f − g∥ max{1, ∥f ∥ j } < Cδ ≤ εn
Thus there exists i such that |β − αi | < ε < ε0 . Then for each j ̸= i we have |β − αi | < |αj − αi |
and therefore |β − αj | = |αj − αi | > ε by Lemma 3.2.3. In other words, αi is uniquely determined
by β. By Krasner’s lemma we get K(αi ) ⊂ K(β). If moreover f is irreducible we have
deg(f ) = [K(αi ) : K] ≤ [K(β) : K] ≤ deg(g) = n
and hence K(αi ) = K(β) and g is irreducible.
4 By Theorem 3.3.10 this assumption is satisfied if (K, | · |) is complete

62
Remark 3.4.13. See [Cona] for a more general result.
Corollary 3.4.14. Let L/Qp be a finite extension. Then there exists a monic irreducible polynomial
g(T ) ∈ Q[T ] such that L ∼
= Qp [T ]/g(T ). In particular L is isomorphic to a completion of the number
field K = Q[T ]/g(T ).
Proof. Since the extension L/Qp is separable, there exists α ∈ L such that L = Qp (α). Let
f (T ) ∈ Qp [T ] be the minimal polynomial of α. Then f is irreducible. By Theorem there exists
monic g ∈ Q[T ] with deg(g) = deg(f ) such that L ∼= Qp [T ]/g(T ). In particular g is irreducible in
Qp [T ] hence also irreducible in Q[T ]. Then K = Q[T ]/g(T ) is a number field and we have a field
embedding K ,→ L with dense image, which determines a place of K. The completion of K under
the corresponding absolute value is isomorphic to L.
Corollary 3.4.15. Let (K, | · |) be an algebraically closed non-archimedean valued field and let K̂
be its completion. Then K̂ is algebraically closed.
Proof. We prove the weaker result that K̂ is separably closed (when char(K) = 0 this implies that
K̂ is algebraically closed). See [Cona] for the proof in general.
Let f (T ) ∈ K̂[T ] be a monic separable polynomial of degree n > 0. By Proposition 3.4.12 for
any g(T ) ∈ K[T ] sufficiently close to f and any root β ∈ K = K of g, there exists a root α of f in
the separable closure of K̂ such that K̂(α) ⊂ K̂(β) = K̂. Therefore α ∈ K̂ and hence K̂ is separably
closed.
In particular this shows that Cp , the p-adic completion of Qp , is algebraically closed.

3.5 Extensions of valuations

For a place v of a field K (i.e. an equivalence class of nontrivial absolute values), we let | · |v denote
an absolute value in the equivalence class of v. In case v is non-archimedean, the same letter v also
denotes an additive valuation corresponding to | · |v . We denote by Kv the completion of K under
| · |v . The place v and absolute value | · |v uniquely extends to Kv , which we denote by the same
symbol. Let K v be an algebraic closure of Kv and let v̄ (resp. | · |v̄ ) be the unique extension of v
(resp. | · |v ) to K v (cf. Theorem 3.3.10). Let Kvsep be the separable closure of Kv in K v so that
we have AutKv (K v ) ∼ = Gal(Kvsep /Kv ). In case v is non-archimedean, we denote by (Ov , mv ) the
valuation ring of Kv and κv := Ov /mv the residue field.
Definition 3.5.1. Let L/K be an extension field and let w be a place of L. Then the restriction
of | · |w to K determines a place v of K. In this case we say that w extends v, or w is above v, or v
is below w and indicate this by writing w|v.
If v is a non-archimedean place of K and w|v is an extension, we define the ramification index
to be
ew = e(w|v) = [w(L× ) : v(K × )] ∈ Z>0 ∪ {∞}
and the inertia degree to be
fw = f (w|v) := [λw : κv ] ∈ Z>0 ∪ {∞}
where λw (resp. κv ) is the residue field of (L, | · |w ) (resp. (K, | · |v )).
Proposition 3.5.1. Let v be a nonarchimedean place of a field of K. Let L/K be a field extension
and let w be a place of L above v. Let Ow , (resp. Ov ) be the valuation ring and λ (resp. κ) be the
residue fields of L (resp. K).
×
For any elements ω1 , . . . , ωr ∈ Ow whose images in the residue field λ are κ-linearly independent
and any nonzero elements π1 , . . . , πm ∈ Ow whose valuations are mutually distinct in the quotient
group w(L× )/v(K × ), the elements {πi ωj , 1 ≤ i ≤ m, 1 ≤ j ≤ r} are K-linearly independent in L.
In particular when L/K is a finite extension, the ramification index ew = e(w|v) and the inertia
degree fw = f (w|v) are both finite and we have ew fw ≤ [L : K].

63
Proof. By the same argument as in the proof of Lemma 2.2.2 we see that ω1 , . . . , ωr are K-linearly
independent. Let ω̄1 , . . . , ω̄r be their images in λ. P
Suppose there exists aij ∈ K for all 1 ≤ i ≤ m, 1 ≤ j ≤ r such that i,j aij ωj πi = 0.
Pr
Let si := j=1 aij ωj . If si ̸
= 0 then the coefficients aij , 1 ≤ j ≤ r are not all 0. Let d =
× −1
min1≤j≤f v(aij ) ∈ v(K ) and let 1 ≤ k ≤ r be an index with v(aik ) = d. Then aik si ∈ Ow and its
image in λ is a nontrivial linear combination of {ω̄j , 1 ≤ j ≤ r} and hence nonzero by assumption.
Therefore a−1 × ×
ik si ∈ Ow and hence w(si ) = w(aik ) = v(aik ) ∈ v(K ). In conclusion we have shown
that for all 1 ≤ i ≤ m, either si = 0 or w(si ) ∈ v(K × ). By Pm assumption {w(πi ), 1 ≤ i ≤ m} are
mutually distinct modulo v(K × ) and therefore the equality i=1 si πi = 0 forces si = 0 for all i and
hence aij = 0 for all i, j by the K-linear independence of {ωj , 1 ≤ j ≤ r}.
In the case of fraction fields of Dedekind domains, let us spell out the dictionary between the
language of valuations and the language of ideals.
Proposition 3.5.2. Let (A ⊂ B, F ⊂ L) be a finite extension of Dedekind domains in the sense of
Definition 2.1.1. Let p ⊂ A be a nonzero prime ideal and let vp be the associated discrete additive
valuation on K. Then there is a natural bijection between the set of prime ideals of B above p and
the set of places of L above vp , sending a prime P to its associated discrete valuation vP and the
inverse map sends a place w to {b ∈ B, |b|w ≤ 1}. Moreover if P corresponds to w = vP under this
bijection, we have the identity between the corresponding ramification indices e(P|p) = e(w|v) and
inertia degrees f (P|p) = f (w|v).
Theorem 3.5.3. Let L/K be an algebraic extension and let v be a place of K. Then there is
a bijection between the set of AutKv (K v )-orbits on HomK (L, K v ) and the set of places of L that
extends v. The map sends any field embedding τ ∈ HomK (L, K v ) to the place w of L defined by
|x|w := |τ (x)|v̄ for any x ∈ L. In particular, when L/K is finite extension, there are only finitely
many places w of L above v.
Proof. By Theorem 3.3.10, any automorphism in AutKv (K v ) fixes v̄ and hence the map is well-
defined.
Next we show surjectivity. Let w be a place of L above v. We write L = ∪i Li as a union of
finite extensions Li /K. Let Li,w be the completion of Li under | · |w . The compositum Li Kv (inside
Li,w ) is a finite extension of Kv and therefore complete by Proposition 3.3.11. Since Li is dense in
Li,w we have Li,w = Li Kv and therefore Li,w is a finite extension of Kv and the absolute value | · |w
naturally extends to Li,w . Then we can choose a compatible system of embeddings Li,w ,→ K v for
each i and they induce an embedding τ ∈ HomK (L, K v ). Since Kv is complete, the restriction of v̄
to Li,w equals to | · |w (which is the unique extension of | · |v to Li,w ). Therefore |τ (x)|v̄ = |x|w for
all x ∈ L. This shows that the map is surjective.
Finally we prove injectivity. Let τ, τ ′ ∈ HomK (L, K v ) be two embeddings that induce the same
absolute value | · |w on L. Then τ (L) and τ ′ (L) are two subfields of K v and the isomorphism
∼
τ ′ ◦ τ −1 : τ (L) −
→ τ ′ (L) preserves the absolute value | · |v̄ . Thus it extends by continuity to a field
∼
isomorphism τ (L)Kv − → τ ′ (L)Kv . This isomorphism extends to an automrophism of their algebraic
closure σ ∈ AutKv (K v ) so that we have σ ◦ τ = τ ′ . This means that τ and τ ′ are in the same
AutKv (K v )-orbit and therefore the map is injective.
Example 3.5.1. In the case K = Q and L is a number field, let v = ∞ be the (unique) archimedean
place of Q, this recovers Lemma 3.4.7. Let v = p a prime number. Then we get a bijection between
Gal(Qp /Qp )-orbits on HomQ (L, Qp ) and the set of prime ideals p ⊂ OL such that p ∩ Z = pZ.
In the case of a finite separable extension we have L = K(θ) where θ has separable minimal
polynomial f (X) ∈ K[X]. Then we have a concrete description of the places w|v in terms of the
factorization of f (X) in Kv [X]. Suppose that in Kv [X] we have the factorization
r
Y
f (X) = fi (X)
i=1

64
where fi (X) ∈ Kv [X] is monic irreducible for all 1 ≤ i ≤ r. Let di = deg fi and let αi,1 , . . . , αi,di
be all the distinct roots of fi (X) in K v . Then {αi,j , 1 ≤ j ≤ di , 1 ≤ i ≤ r} are all the distinct roots
of f (X) in K v and they are in bijection with HomK (L, K v ). Under this bijection the AutKv (K v )-
orbits HomK (L, K v ) are the subsets {αi,1 , . . . , αi,di } for 1 ≤ i ≤ r. Therefore the places of L above
v are in bijection with the irreducible factors {fi (X)}1≤i≤r of f (X) in Kv [X]. Let wi be the place
corresponding to fi (X). Then for any g(X) ∈ K[X] we have |g(θ)|wi := |g(αi,j )|v̄ for any 1 ≤ j ≤ di .
By the Chinese Remainder Theorem we have natural ring isomorphisms
r
= Kv [X]/f (X) ∼
L ⊗K Kv ∼
Y
= Kv [X]/fi (X).
i=1

The i-th factor of the isomorphism induces a field embedding L ,→ Kv [X]/fi (X) which extends to
an isomorphism of topological fields Lwi ∼
= Kv [X]/fi (X). In summary we have:
Proposition 3.5.4. Let L/K be a finite separable extension and let v be a place of K. For each
place w|v of L, the embeddings K ⊂ L ⊂ Lw uniquely extends to a field embedding Kv ⊂ Lw and
therefore an algebra homomorphism L ⊗K Kv → Lw . The product of these homomorphisms over all
places w of L above v induces a ring isomorphism

L ⊗K K v ∼
Y
= Lw .
w|v

In particular for each w|v, the extension Lw /Kv is finite separable and we have
X
[L : K] = [Lw : Kv ].
w|v

Moreover, for any element α ∈ L we have

X Y
TrL/K (α) = TrLw /Kv (α), NL/K (α) = NLw /Kv (α).
w|v w|v

Next we consider the case of (possibly infinite) Galois extensions. Let L/K be a Galois extension
with Galois group G = Gal(L/K). For any place w of L above a place v of K and any σ ∈ G, we
get another place σ(w) of L above v which is defined by |x|σ(w) := |σ −1 (x)|w . We say that σ(w)
and w are conjugate (under G).
Lemma 3.5.5. G acts transitively on the set of places w of L above v. Consequently the decompo-
sition groups of any two places above v are conjugate in G.
Proof. We prove the case where L/K is finite. See [Neu99, II.9.1] for how to reduce to the finite
extension case.
Suppose there are two places w, w′ of L above v that are not conjugate under G. Then by the
approximation theorem below, we can find an x ∈ L such that
|σ(x)|w < 1, |σ(x)|w′ > 1, ∀σ ∈ G.
Let α = NL/K (x). Then we have
Y Y
|α|v = |σ(x)|w = |σ(x)|w′
σ∈G σ∈G

and the two inequalities above result in a contradiction.

Theorem 3.5.6 (Approximation Theorem). Let | · |1 , . . . , | · |n be pairwise inequivalent nontrivial
absolute values on a field K. For each 1 ≤ i ≤ n, let K̂i be the completion of K under | · |i and
extend the absolute value | · |i to Ki by continuity. Choose an element ai ∈ K̂i for any 1 ≤ i ≤ n.
Then for any ε > 0 there exists x ∈ K such that |x − ai |i < ε for all 1 ≤ i ≤ n.

65
Proof. We can choose bi ∈ K such that |bi − ai |i < 12 ε for all 1 ≤ i ≤ n. If we could find ξ ∈ L such
that |ξ − bi |i < 21 ε for all 1 ≤ i ≤ n, then |ξ − ai | ≤ |ξ − bi |i + |bi − ai |i < ε. Thus we may assume
that ai ∈ K for all 1 ≤ i ≤ n.
First we prove by induction on n that there exists z ∈ K such that |z|1 > 1 and |z|j < 1 for all
j ̸= 1. Since | · |1 and | · |2 are inequivalent, by Lemma 3.2.1 there exists α ∈ K with |α|1 < 1 and
|α|2 ≥ 1. Similarly there exists β ∈ K with |β|2 < 1 and |β|1 ≥ 1. Let z = α−1 β. Then |z|1 > 1
and |z|2 < 1. This proves the case n = 2. Now we prove the induction step. Assume we have found
z ∈ K such that |z|1 > 1 and |z|j < 1 for all 2 ≤ j ≤ n − 1. By the n = 2 case there exists y ∈ L
with |y|1 > 1 and |y|n < 1. If |z|n ≤ 1, then z m y will satisfy the requirement for m large enough.
zm
If |z|n > 1, the sequence tm = 1+z m converges to 1 under | · |1 and | · |n , and converges to 0 under

| · |j for any 2 ≤ j ≤ n − 1. Then tm y will satisfy the requirement for m sufficiently large.
zm
For z ∈ K such that |z|1 > 1 and |z|j < 1 for all j ̸= 1, the sequence 1+z m converges to 1 under

| · |1 and converges to 0 under | · |j for any j ̸= 1. Thus for every 1 ≤ i ≤ n we can construct an
element zi which can be arbitrarily close to 1 under | · |i and arbitrarily close to 0 under | · |j for any
j ̸= i. The element x = a1 z1 + · · · + an zn will satisfy the requirement.
Definition 3.5.2. The decomposition group of a place w of L is the subgroup

Gw = Gw (L/K) = {σ ∈ G | σ(w) = w}.

The decomposition field of w is the subfield Zw = LGw .

From the definition we see immediately that Gσ(w) = σGw σ −1 for any σ ∈ G.
For places w|v, if L/K is finite we let Lw be the completion of L under | · |w . Then Lw /Kv
is a finite extension. If L/K is infinite algebraic extension we write L = ∪i Li as a union of finite
extensions Li /K and let Lw be the union of the completions Li,w (either take the union inside K v ,
or define it abstractly as an inductive limit). Following [Neu99], we call Lw the localization of L at
w (this does not seem to be a standard terminology). In any case Lw /Kv is an algebraic extension.
(In the case where L/K is infinite, if we simply take the completion of L, it will not be an algebraic
extension of Kv .)
Proposition 3.5.7. Let L/K be a Galois extension with G = Gal(L/K). For any place w of L
above a place v of K, the localization Lw is a Galois extension of Kv . There is an isomorphism
Gal(Lw /Kv ) ∼
= Gw with the decomposition group of w defined by restricting an automorphism of
Lw along the natural embedding L ⊂ Lw . In particular the decomposition field of w is Zw = L ∩ Kv
where the intersection is taken inside Lw .
Proof. Choose an embedding ιw : L ,→ K v such that | · |v̄ restricts to | · |w , then Lw is identified
with the compositum ιw (L)Kv inside K v . Since L/K is Galois, L is the splitting field of a set
of separable polynomials in K[X] and hence Lw is the splitting field of the image of the same set
of polynomials in Kv [X]. Thus Lw /Kv is a Galois extension and we get a group homomorphism
Gal(Lw /Kv ) → G = Gal(L/K) by restricting and automorphism of Lw to L. It is injective since Lw
is generated by L over Kv . Since the natural extension of | · |w to Lw is the unique absolute value on
Lw extending the absolute value | · |v on Kv by Theorem 3.3.10, any element in Gal(Lw /Kv ) fixes
the place w. Thus the image of the embedding Gal(Lw /Kv ) ,→ G lies in the decomposition group
Gw . By definition any σ ∈ Gw defines an isometry of (L, | · |w ) and by continuity extends uniquely to
an automorphism of Lw . Thus the map is surjective and we get an isomorphism Gal(Lw /Kv ) ∼ = Gw .
In particular we get Kv ∩ L = Zw .
The study of a general extension L/K is thus reduced to the case where K is complete. We will
study this case in a more general context of Henselian fields, emphasizing the important role of the
unique extension property for absolute values.

66
3.6 Henselian fields
Definition 3.6.1. A Henselian field is a non-archimedean valued field (K, |·|) such that the absolute
value extends uniquely to any algebraic extension of K.
The Henselization of a field K at a nonarchimedean place v is the algebraic closure of K in the
completion K̂v of K under | · |v .
From the definition we see that any algebraic extension of a Henselian field is a Henselian. Also,
by Theorem 3.3.10 any complete non-archimedean field is Henselian. To explain the terminology
we will relate the property of being Henselian to Hensel’s lemma in the form of Theorem 3.3.6. For
this we need some preparation on Newton polygons of a polynomial.

3.6.1 Newton polygons

Definition 3.6.2. Let v be an additive valuation on a field K. Let

f (X) = a0 + a1 X + · · · + an X n ∈ K[X]

be a polynomial with a0 an ̸= 0. The Newton polygon of f is the lower convex envelope of the points

{(i, v(ai )) ∈ R2 | 0 ≤ i ≤ n, ai ̸= 0}.

The Newton polygon of f consists of line segments and by convexity their slopes increase from
left to right.
Proposition 3.6.1. Let v be an additive valuation on a field K and let f (X) = a0 + a1 X + · · · +
an X n ∈ K[X] be a polynomial with a0 an ̸= 0. Let L be the splitting field of f and let w be an
additive valuation of L extending v on K. If (r, v(ar )) ↔ (s, v(as )) is a line segment of slope −m
in the Newton polygon of f , then f has precisely (s − r) roots α1 , . . . , αs−r in L with valuation

w(α1 ) = · · · = w(αs−r ) = m.

This follows from the fact that the coefficients are elementary symmetric polynomials in the
roots. See [Neu99, §II.6.3] for the proof.
Example 3.6.1. Let f (X) = a0 + a1 X + X 2 be a monic quadratic polynomial and let α1 , α2 ∈ L
be its roots. Then we have a0 = α1 α2 and −a1 = α1 + α2 . Suppose w(α1 ) ≤ w(α2 ).
If w(α1 ) < w(α2 ), then v(a0 ) = w(α1 ) + w(α2 ) and v(a1 ) = w(α1 ) < 21 v(a0 ) = w(α1 )+w(α
2
2)
. In
this case the Newton polygon of f consists of two segments (0, v(a0 )) ↔ (1, v(a1 )) and (1, v(a1 )) ↔
(2, 0) with slopes v(a1 ) − v(a0 ) = −w(α2 ) and −v(a1 ) = −w(α1 ).
If w(α1 ) = w(α2 ), then v(a0 ) = 2w(α1 ) and v(a1 ) ≥ w(α1 ) = 21 v(a0 ). In this case the Newton
polygon consists of a single segment (0, v(a0 )) ↔ (2, 0) with slope − 12 v(a0 ) = −w(α1 ) = −w(α2 ).
Definition 3.6.3. Let v be an additive valuation on a field K and let f (X) = a0 +a1 X+· · ·+an X n ∈
K[X] be a polynomial with a0 an ̸= 0. Let L be the splitting field of f and let w be an additive
valuation of L extending v on K. Let −mr < · · · < −m1 be the slopes of the Newton polygon of f .
Let α1 , . . . , αn ∈ L be the roots of f , countedQwith multiplicity. The slope decomposition of f (with
r
respect to w) is the factorization f (X) = an j=1 fj (X) in L[X] where for each 1 ≤ j ≤ r,
Y
fj (X) := (X − αi ) ∈ L[X].
1≤i≤n
w(αi )=mj

Proposition 3.6.2. With notations as above. Suppose moreover that the valuation w on L is
the unique extension of v (for example, this is satisfied if (K, v) is Henselian). Then the slope
decomposition is defined over K, i.e. fj (X) ∈ K[X] for all 1 ≤ j ≤ r. Moreover, if f is irreducible,
then f = f1 and its Newton polygon is a single line segment.

67
Proof. We may assume f is monic and let f = f1 · · · fr ∈ L[X] be the slope decomposition. Let
G = AutK (L) If f is irreducible in K[X] then G acts transitively on the set of roots and by
uniqueness of extension we get w(αi ) = w(αj ) for all 1 ≤ i, j ≤ n. Thus f (X) = f1 (X) ∈ K[X].
In general we prove by induction on n = deg(f ). If n = 1 the statement is obvious. Let
h(X) ∈ K[X] be the minimal polynomial of α1 . Then h is irreducible and hence h|f1 by what
we already proved. Let g(X) = f (X)/h(X) ∈ K[X] and let g1 (X) = f1 (X)/h(X) ∈ L[X]. Then
g(X) = g1 (X)f2 (X) · · · fr (X) is the slope decomposition of g and since deg g < deg f , by induction
hypothesis we get that g1 , f2 , . . . , fr ∈ K[X] and therefore fj ∈ K[X] for all 1 ≤ j ≤ r.
Example 3.6.2. Let f (X) = X 3 − X 2 − 2X − 8. One checks easily that f has a unique root α in
R. Let K = Q(α).
For an odd prime number p > 2, the Newton polygon of f ∈ Qp [X] is a single line segment of
slope 0, while the Newton polygon of f ∈ Q2 [X] consists of 3 segments of slopes −2, −1, 0. If α ∈ Q,
then by Proposition 3.6.1 we get α ∈ Z×(p) for any odd p and α ∈ Z(2) . Then we must have α = 2
n
n
for some n ∈ Z≥0 but one can easily see that f (2 ) ̸= 0 for any n ∈ Z≥0 . Therefore f is irreducible
in Q[X] and it is the minimal polynomial of α. Then one calculates that

dK · [OK : Z[α]]2 = disc(f ) = −2012 = −4 · 503.

From Proposition 3.6.2 we deduce that f splits completely in Q2 [X] and thus 2 splits completely in
OK by Proposition 3.5.4. In particular OK ⊗ Z(2) ̸= Z(2) [θ] for any θ ∈ K (and hence OK ̸= Z[θ] for
any θ ∈ K) since otherwise the reduction mod 2 of the minimal polynomial of θ would have 3 distinct
roots in the field F2 with only 2 elements, which is absurd! Consequently we get [OK : Z[α]] = 2
and dK = −503.
In this example the condition in Kummer’s Theorem 2.2.5 is not satisfied at 2 so we cannot
directly apply it to get the factorization of 2OK .

3.6.2 Characterizing Henselian fields

Now we can give some equivalent characterizations of Henselian fields.
Theorem 3.6.3. Let (K, | · |) be a non-archimedean valued field. Then the following are equivalent:
1. (K, | · |) is Henselian (in the sense of Definition 3.6.1).
2. The Newton polygon of any irreducible polynomial in K[X] consists of a single line segment.
3. For any irreducible polynomial f (X) = a0 + a1 X + · · · + an X n ∈ K[X] with an ̸= 0. Let
∥f ∥ := max0≤i≤n {|ai |}. Then ∥f ∥ = max{|a0 |, |an |}.
4. (K, | · |) satisfies Hensel’s Lemma in the form of Theorem 3.3.6.
5. (K, | · |) satisfies the weaker version of Hensel’s Lemma (Theorem 3.3.6) in which we require
furthermore that the polynomial f is monic.
Proof. (1)⇒(2) by Proposition 3.6.2.
(2)⇔(3) by definition of Newton polygon.
(3)⇒(1): same proof as in Theorem 3.3.10.
(1)+(3)⇒(4): See [Neu99, §II.6.6].
(4)⇒(5) is tautological.
(5)⇒(2): See [Neu99, §II.6.7].
Definition 3.6.4. Let (K, v) be a Henselian field. Let L/K be a separable algebraic extension and
let w be the unique valuation on L extending v. The ramification index of the extension L/K is the
number
e(L/K) := [w(L× ) : v(K × )] ∈ Z>0 ∪ {∞}

68
Let κ (resp. λ) be the residue fields of v (resp. w). The inertia degree of the extension L/K is the
number
f (L/K) := [λ : κ] ∈ Z>0 ∪ {∞}.
The extension L/K is

• tamely ramified if char(κ) does not divide e(L/K) and the residue field extension λ/κ is
separable;
• unramified if it is tamely ramified and e(L/K) = 1;
• wildly ramified if it is not tamely ramified;

• totally ramified if f (L/K) = 1.

Corollary 3.6.4. Let (K, v) be a Henselian field and let L/K be a finite extension. Let w be the
unique extension of v to L. Let Ow (resp. Ov ) be the valuation rings of w (resp. v). Then Ow is
the integral closure of Ov in L and we have
1
w(α) = v(NL/K (α)), ∀α ∈ L× .
[L : K]

In particular if v is discrete then w is also discrete.

Proof. Once we know Theorem 3.6.3, the same argument as in the proof of Corollary 3.3.9 shows
that the integral closure of Ov in L equals to

O
e = {α ∈ L | NL/K (α) ∈ Ov }.

From this we deduce the formula w(α) = 1

for all α ∈ L× by the same argument
[L:K] v(NL/K (α))
as in the proof of Theorem 3.3.10. In particular we get that Ow = O e and thus Ow is indeed the
integral closure of Ov in L.
Let us summarize a hierarchy of fields that we have encountered:

{nonarchimedean
local fields } ⊂ {complete discrete
valued fields
} ⊂ complete
{nonarchimedean fields
}
⊂ ⊂
⊂

{discrete valued fields} ⊃ {Henselian discrete}

valued field ⊂ {Henselian fields}


{global fields} ⊂ fraction fields of }
{Dedekind domains

We give some examples of fields in each category that does not belong to smaller subcategories:
• Complete discrete valued fields that are not local fields: C((T )), Fp ((T )), Q̆p :=completion of
the maximal unramified extension of Qp in Qp .

\ 1
• Complete non-archimedean fields that are not discrete valued: Cp , Fp ((t p∞ )), p-adic comple-
1
tions of Qp (ζp∞ ) and Qp (p p∞ ).
• Discrete valued fields that are not Henselian: (Q, | · |p ), Fp (T ) with T -adic absolute value;
• Henselian discrete valued fields that are not complete: Hensilization of Q at p (i.e. algebraic
closure of Q in Qp ), Qur
p (the maximal unramified extension of Qp in Qp ).

69
 • Henselian fields that are not complete and not discrete valued: Qp , C((t)), Fp ((t)), any infinite
algebraic extension of Qp with infinite ramification index.
• For a Dedekind domain A with fraction field K, any nonzero prime ideal p ⊂ A defines a
discrete valuation vp on K and (K, vp ) becomes a discrete valued field. Fraction fields of
Dedekind domains that are not global fields or Henselian fields include: C(T ), Fp (T ), rational
function fields of Riemann surfaces.

3.7 Extensions of Henselian discrete valued fields

We study extensions of a Henselian discrete valued fields (K, v). We will normalize the additive
valuation v so that v(K × ) = Z. Then the value group of a finite extension (L, w) (where w is the
unique valuation of L extending v) will be w(L× ) = 1e Z where e = e(L|K) is the ramification index.

3.7.1 The fundamental identity

Theorem 3.7.1. Let (K, v) be a Henselian discrete valued field. Let L/K be a finite extension and
let w be the unique valuation on L extending v. Let Ov (resp. Ow ) be the valuation ring of v (resp.
w) with residue field κ (resp. λ). Let e := e(L/K) and f := f (L/K). Suppose that either L/K
is separable or K is complete. Then Ow is a free Ov -module of rank ef . In particular we have
ef = [L : K].
Moreover, for any ω1 , . . . , ωf ∈ Ow whose image in λ form a κ-basis and any πw ∈ Ow with
w(πw ) = 1e (i.e. a uniformizer of Ow ), the elements {πw i−1
ωj , 1 ≤ i ≤ e, 1 ≤ j ≤ f } form an
Ov -basis of Ow .
Proof. Since v is discrete, w is also discrete by Corollary 3.6.4. Let πw ∈ Ow and πv ∈ Ov be
e ×
uniformizers so that πv ∈ πw Ow . Let ω1 , . . . , ωf ∈ Ow be elements whose image in λ form a κ-basis.
Consider the following Ov -submodules in Ow :
X f
X
i−1
M := O v πw ωj , N= Ov ωj .
1≤i≤e j=1
1≤j≤f

By Proposition 3.5.1, M is a free Ov -module of rank ef . Moreover, we have M = N + πw N + · · · +

e−1
πw N and Ow = N + πw Ow (since N surjects onto λ = Ow /πw Ow ). Then we deduce inductively
that
2 e
Ow = N + πw Ow = N + πw N + πw O w = · · · = M + πw Ow = M + πv Ow
and therefore
Ow = M + πv Ow = M + πv (M + πv Ow ) = M + πv2 Ow = · · · = M + πvm Ow , ∀m ∈ Z≥1 .
Suppose that L/K is finite separable. By Corollary 3.6.4, Ow is the integral closure of Ov in L.
Then by Proposition 2.1.3 Ow is a finitely generated Ov -module. From Ow = M +πv Ow we get that
πv Ow ⊂ M and apply Nakayama’s Lemma to the quotient Ov -module Ow /M we get that Ow = M .
Suppose that K is complete. From the above reasoning we see that M is dense in Ow . Since M
is complete by Proposition 3.3.11, we get that M = Ow .
In any case we conclude that Ow is a free Ov -module of rank ef so that ef = [L : K].
Lemma 3.7.2 (Nakayama). Let R be a local ring with maximal ideal m and let M be a finitely
generated R-module. If mM = M , then M = 0.
Proof. Suppose M ̸= 0. Let x1 , . . . , xP
n be a set of generators of M so that n is minimum possible.
n Pn−1
By assumption we can write xn = i=1 ai xi for some ai ∈ m. Then (1 − an )xn = i=1 a i xi .
Pn−1
Since (1 − an ) ∈ 1 + m ⊂ R× we see that xn ∈ i=1 Rxi so that x1 , . . . , xn−1 generate M . This
contradicts with minimality of n and we are done.

70
 Combining Proposition 3.5.4 and Theorem 3.7.1 we get:
Corollary 3.7.3 (Fundamental identity of valuation theory). Let L/K be a finite separable field
extension and let v be a discrete non-archimedean place of K. Then we have
X
e(w|v)f (w|v) = [L : K]
w|v

where the sum runs over places w of L above v.

3.7.2 Unramified and tamely ramified extensions

Theorem 3.7.4. Let (K, v) be a Henselian discrete valued field and let (L, w) be a finite extension
of (K, v). Let (Ov , mv ) and (Ow , mw ) be the valuation rings of v and w. Assume that the residue
field extension is separable. Assume moreover that either K is complete or the extension L/K is
separable. Then

1. There exists x ∈ Ow such that Ow = Ov [x].

2. For any x ∈ Ow such that Ow = Ov [x] and any y ∈ Ow with y − x ∈ m2w , we have Ov [y] = Ow
Proof. (1) Let mv (resp. mw ) be the maximal ideal of Ov (resp. Ow ). Fix a uniformizer πw ∈ mw .
Since the residue field extension λ/κ is separable we can choose an element x ∈ Ow with image x̄ ∈ λ
such that λ = κ(x̄). Let φ(T ) ∈ Ow [T ] be a monic polynomial such that its image φ̄(T ) ∈ κ[T ] is
the minimal polynomial of x̄. Then we have φ(x) ∈ mv and

φ(x + πw ) = φ(x) + φ′ (x)πw + yπw

2
, y ∈ Ow .

Since φ̄ is separable we have φ̄′ (x̄) ̸= 0 and hence φ′ (x) ∈ Ow . Thus if φ(x) ∈ m2w we get that
×
φ(x + πw ) ∈ πw Ow . Then replacing x by x + πw if necessary we may assume that φ(x) is a
uniformizer in Ow . By Theorem 3.7.1 we get that {xj φ(x)i , 1 ≤ i ≤ e(L|K), 1 ≤ j ≤ f (L|K)} form
an Ov -basis of Ow and therefore Ow = Ov [x].
(2) Suppose Ow = Ov [x] and y ∈ x + m2w . Then πw = g(x) for some g(T ) ∈ Ov [T ] and moreover
g(y) − g(x) ∈ m2w . Therefore g(y) is a uniformizer. On the other hand since y ≡ x mod mw , the
powers 1, y, . . . , y f lifts a κ-basis of the residue field λ. Therefore by the same reasoning as the
previous part we get that {y j g(y)i , 1 ≤ i ≤ e, 1 ≤ j ≤ f } form and Ov -basis of Ow and hence
Ow = Ov [y].
With notations and assumptions as in Theorem 3.7.4. Let f (T ) = T n + an−1 T n−1 + · · · + a0 ∈
Ov [T ] be the minimal polynomial of x and let f¯(T ) ∈ κ[T ] be its reduction mod mv . Since f is
irreducible, its Newton polygon consists of the single line segment (0, v(a0 )) ↔ (n, 0) by Theorem
3.6.3 (except in the trivial case where L = K and x = 0, in which the Newton polygon is a
single point). In particular we have w(x) = n1 v(a0 ). We can also see this by noting that a0 =
(−1)n NL/K (x) and use Corollary 3.6.4.
If w(x) > 0 so that x̄ = 0, then f (L|K) = 1 and e(L|K) = [L : K] = n. In this case the extension
L|K is totally ramified. Let πw ∈ Ow be a uniformizer so that w(πw ) = n1 and w(x) ∈ n1 Z>0 . Since
Ow = Ov [x] and in particular πw can be expressed as a polynomial of x, we must have w(x) = n1
and hence v(a0 ) = 1. Therefore x itself is a uniformizer in Ow and its minimal polynomial f (T ) is
an Eisenstein polynomial defined as follows:
Definition 3.7.1. Let (K, v) be a discrete valued field with valuation ring Ov and let n ≥ 1 be an
integer. Suppose v is normalized so that v(K × ) = Z. An Eisenstein polynomial of degree n is a
monic polynomial f (X) = X n + an−1 X n−1 + · · · a1 X + a0 ∈ K[X] such that
• v(ai ) > 0 for all 0 ≤ i ≤ n − 1 and

71
 • v(a0 ) = 1.
In particular f (X) ∈ Ov [X] and its Newton polygon consists of a single line segment with slope − n1
(and vice versa).
An Eisenstein poynomial f (X) is irreducible. Indeed, if f (X) = g(X)h(X) where g, h are monic.
Then g, h ∈ Ov [X] by Gauss’s lemma. The constant term of one of g, h, say g has to be in Ov× .
If deg g ≥ 1, then the Newton polygon of g would be a horizontal line segment and its roots in
the splitting field would have valuation 0, but the Newton polygon of f is a single line segment
with negative slope and all its root should have positive valuation by Proposition 3.6.1. This is a
contradiction and hence f is irreducible.
Proposition 3.7.5. Let (K, v) be a Henselian discrete valued field. Each Eisenstein polynomial
f (X) ∈ K[X] of degree n generates a totally ramified extension K[X]/f (X) of degree n. Conversely,
let L/K be a totally ramified extension of degree n and let πw ∈ L be a uniformizer. Then Ow =
Ov [πw ] (and hence L = K(πw )) and the minimal polynomial of πw is an Eisenstein polynomial of
degree n.
Proof. Let f (X) ∈ K[X] be an Eisenstein polynomial of degree n. We have shown above that
f (X) is irreducibe and hence L := K[T ]/f (T ) is a field of degree [L : K] = n. Let w be the
unique extension of v to L. By Proposition 3.6.1 the roots of f in L has valuation n1 . Therefore
e(L/K) ≥ n = [L : K] which forces [L : K] = e(L/K) and hene L/K is totally ramified.
Suppose L/K is totally ramified of degree n and let πw ∈ L be a uniformizer. By Theorem 3.7.1
we have Ow = Ov [πw ]. The Newton polygon of the minimal polynomial of πw consists of a single
1
segment by Proposition 3.6.2. Since w(πw ) = e(L/K) = n1 , it has slope − n1 by Proposition 3.6.1.
Thus it is an Eisenstein polynomial of degree n.
Proposition 3.7.6. Let (K, v) be a Henselian discrete valued field and let (L, w) be a separable
algebraic extension of K where w is the unique additive valuation on L extending v. Let λ (resp.
κ) be the residue field of L (resp. K). There is a canonical inclusion-preserving bijection between
finite unramified extensions of K contained in L and finite separable extensions of κ contained in
λ, sending any such extension L′ to its residue field. In particular, the separable closure κs of κ in
λ corresponds to the maximal unramified subextension L0 of K in L, and L0 is the compositum of
the all finite unramified extensions of K in L.
Proof. Let λ′ = κ(x̄) be a finite separable extension of κ. Let φ̄(T ) ∈ κv [T ] be the minimal
polynomial of x̄ and let φ(T ) ∈ Ov [T ] be a monic polynomial lifting φ̄. Since φ̄ is irreducible, we
deduce that φ(T ) is irreducible in K[T ] by Gauss’s lemma.
By Theorem 3.6.3 (that Hensel’s lemma holds for K) there exists a root x′ ∈ Ow ⊂ L of φ(T )
such that x′ ≡ x̄ mod mw . Let L′ = K(x′ ). Then Ov [x′ ] is contained in the valuation ring of L′
and hence λ′ is contained in the residue field of L′ . On the other hand we have [L′ : K] = deg φ =
deg φ̄ = [λ′ : κ]. Therefore L′ /K is unramified and the residue field of L′ is λ′ .
Suppose L′′ is another finite extension of K contained in L with residue field λ′ = κ(x̄), by the
same reasoning we see that φ(T ) has a root x′′ ∈ L′′ lifting x̄. Now x′ , x′′ are two roots in Ow ⊂ L
lifting x̄, by uniqueness of lifting (see Theorem 3.3.5, which also holds for Henselian fields) we must
have x′ = x′′ and thus L′ ⊂ L′′ . Since L′′ /L′ is both unramified and totally ramified (as they have
the same residue field), we have L′′ = L′ . This shows that the map is bijective on the subset of finite
extensions. Then we get bijectivity for all extensions by writing a general extension as a union of
finite subextensions.
In the case where the residue field κ is finite (for example, when K is a nonarchimedean local
field), there is a unique extension of κ of any given degree. Hence there is a unique unramified
extension of K of any given degree and it is generated by roots of unity of order prime to char(κ)
by Hensel’s lemma.

72
Corollary 3.7.7. Let F be a local field. For any positive integer n there are only finitely many
degree n extensions of F up to isomorphism.

Proof. Any finite extension L/F is constructed from an unramified extension L0 /F followed by a
totally ramified extension L/L0 . Since there is a unique unramified extension of any given degree,
it suffices to show that there are only finitely many totally ramified extension of degree n.
Let O be the valuation ring of F and let π ∈ O be a uniformizer. We have seen above that
any totally ramified degree n extension of F is obtained by adjoining the roots of a degree n
Eisenstein polynomial. The set of degree n Eisenstein polynomials in O[T ] can be identified with
(πO)n−1 × πO× (where the coefficients of their constant terms lie in πO× ). By Theorem 3.4.12
there is an open cover of this set so that the Eisenstein polynomials in each open set give rise to
the same extension of F . Since F is a local field, (πO)n−1 × πO× is a compact set and there is a
finite subcover. Thus up to isomorphism there are only finitely many totally ramified extensions of
degree n.

Next we study tamely ramified extensions.

Proposition 3.7.8. Let (K, v) be a Henselian discrete valued field whose residue field κ has char-
acteristic p. Let L/K be a finite separable extension and let L0 ⊂ L be the maximal unramified
extension of K in L. Then the extension L/K is tamely ramified √ if and only if there exists a ∈ L0
and a positive integer m not divisible by p such that L = L0 ( m a).

Proof. Let e = e(L/K) = e(L/L0 ) be the ramification index. Since the residue field of L0 is the
separable closure of κ in the residue field λ of L, the extension L/K is tamely ramified if and only
if L/L0 is tamely ramified. Thus we may assume that K = L0 so that the residue field extension
λ/κ is purely inseparable.
Suppose L/K is tamely ramified. Then λ = κ and p does not divide e = [L : K]. Let w be the
discrete additive valuation on L extending v. Choose uniformizers πw ∈ Ow and πv ∈ Ov . Then
e −1 × ×
ε := πw πv ∈ Ow . Since L and K has the same residue field κ by assumption, there exists u ∈ Ow
× e
with u ≡ 1 mod πw Ow and b ∈ Ov such that ε = ub. Since p ∤ e, the polynomial X −1 is separable
×
in κ[T ]. By Theorem 3.6.3 we can apply Hensel’s lemma to get an element β ∈ Ow with β e = u.
−1
Let a = bπv ∈ Ov and α := β πw . Then a ∈ Ov and √ α ∈ Ow are also uniformizers and we have
αe = a. Therefore we have L = K(πw √) = K(α) = K( e
a) wherer a ∈ K.
Conversely suppose that L = K( m a) where a ∈ K and p ∤ m. Write a = πvd u where d = v(a) ∈ Z
and u ∈ Ov× . Let ū ∈ κ× be the image of u. Let κ′ (resp. λ′ ) be the splitting field of T m − ū over κ
(resp. λ). Then λ′ /λ and κ′ /κ are finite separable extensions since p ∤ m. Let K ′ (resp. L′ ) be the
unramified extensions of K (resp. L) with residue field κ′ (resp. λ′ ). It suffices to show that L′ /K ′
is a tamely ramified extension of K ′ .
We have LK ′ ⊂ L′ and hence the residue field of LK ′ is contained in λ′ (which is the residue
field of L′ ). On the other hand the residue field of LK ′ contains λκ′ since the valuation ring of LK ′
contains the compositum of the valuation rings of L and K ′ . Since λ′ and λκ′ are both the splitting
field of T m − ū over λ we have λ′ = λκ′ . Therefore the fields LK ′ ⊂ L′ have the same
√ residue field λ′
′ ′ ′ m m
and since they are both unramified extension of L we see that L = LK = K ( a). Since T − ū
has a root in κ′ , by Hensel’s lemma there exists ε ∈ K ′ with εm = u. Then we have a = εm πvd and
√ d
hence L′ = K ′ ( m a) = K ′ (πvm ). After factoring out the greatest common divisor we may assume
d 1
that (d, m) = 1. Then there exists x, y ∈ Z such that xd + ym = 1 and hence x m +y = m . Thus
d 1 1
(πvm )x πvy = πvm and we get that L′ = K ′ (πvm ). Since T m − πv is an Eisenstein polynomial, it is
irreducible and hence L′ /K ′ is a totally ramified extension of degree m. Since p ∤ m it is tamely
ramified.
Remark 3.7.9. See [Neu99, §II.7.7] for a more general result without assuming that v is discrete.
Corollary 3.7.10. The composites of tamely ramified extensions are tamely ramified.

73
Example 3.7.1. Let F be a finite extension of Qp . The maximal unramified extension F ur of F
in the algebraic closure F is obtained by adjoining all roots of unity of order prime to p, since Fp
is obtained from Fp in this way. Indeed, F× n
pn is a cyclic group of order p − 1 and hence it is the
pn −1
splitting field of T − 1 over Fp . For any integer m coprime to p, let n be the order of p in
(Z/mZ)× , then m|(pn − 1) and hence all the m-th root of unity in Fp is contained in Fpn .
The maximal tamely ramified extension of Qp is obtained from the maximal unramified extension
1
Qur
p by adjoining p
n for all positive integer n coprime to p.

3.8 Applications to ramification theory

3.8.1 Local and global different
Let (A ⊂ B, K ⊂ L) be a finite separable extension of Dedekind domains (cf. Definition 2.1.1).
Lemma 3.8.1. For any multiplicative subset S ⊂ A we have DS −1 B/S −1 A = S −1 DB/A and ∆S −1 B/S −1 A =
S −1 ∆B/A .
Corollary 3.8.2. For each nonzero prime ideal p ⊂ A, let Bp := B ⊗A ApQand ∆p := ∆Bp /Ap ∩ A.
Then we have a factorization of the relative discriminant ideal ∆B/A = p ∆p where the product
runs over all nonzero prime ideals of A.
Lemma 3.8.3. Let P be a nonzero prime ideal of B and let p := P ∩ A. Let B̂P (resp. Âp ) be the
completion of the localization BP (resp. Ap ). Then we have DB/A B̂P = DB̂P /Âp .

Proof. By Lemma 3.8.1 we may localize at p and assume that A is a discrete valuation ring.
Let v = vp be the discrete valuation of K associated with p. Let Kv be the completion of K
under v. Let w be the valuation on L extending v and equivalent to the valuation vP associated
1
with P (more precisely we have w = e(P|p) vP ). Let Lw be the completion of L under w. Denote the
valuation rings by Av = Âp , Bw = B̂P . Then Bw is the integral closure of Av in Lw by Corollary
3.6.4. Therefore the quadruple (Av ⊂ Bw , Kv ⊂ Lw ) is a finite separable extension of Dedekind
domains in the sense of Definition 2.1.1.
Recall that the inverse differents are defined by
−1
DB/A = B ∨ = {x ∈ L | TrL/K (xB) ⊂ A},

−1 ∨
DB w /Av
= Bw = {x ∈ Lw | TrLw /Kv (xBw ) ⊂ Av }.
Let x ∈ B ∨ be a nonzero element. For any y ∈ Bw , by Theorem 3.5.6 there exists η ∈ L such that

|η − y|w < min{|x−1 |w , 1}, |η|w′ < min{|x−1 |w′ , 1}, ∀w′ |v, w′ ̸= w.

Then for all w′ |v we have η ∈ Bw′ , the valuation ring of Lw′ . By assumption these are all the
nonzero prime ideals of B, so we get η ∈ B and hence TrL/K (xη) ∈ A ⊂ Av . For any w′ |v, w′ ̸= w
we have |xη|w′ < 1 and hence xη ∈ Bw′ . This implies that TrLw′ /Kv (xη) ∈ Av since Bw′ is integral
over Av by Corollary 3.6.4. Then from the identity
X
TrL/K (xη) = TrLw /Kv (xη) + TrLw′ /Kv (xη)
w′ |v,w′ ̸=w

we get that TrLw /Kv (xη) ∈ Av .

∨
The inequality |xη−xy|w < 1 implies that xη−xy ∈ Bw ⊂ Bw . Therefore TrLw /Kv (xη−xy) ∈ Av
and hence
TrLw /Kv (xy) = TrLw /Kv (xη) − TrLw /Kv (xη − xy) ∈ Av .

74
This shows that B ∨ ⊂ Bw
∨
.
∨
Now take a nonzero element z ∈ Bw . For any 0 < ε < 1, by Theorem 3.5.6 there exists ξ ∈ L
such that
|ξ − z|w < ε, |ξ|w′ < ε ∀w′ |v, w′ ̸= w.
Then for any b ∈ B we have |ξb − zb|w < 1 and |ξb|w′ < 1 for all w′ |v, w′ ̸= w. Then we get
X
TrL/K (ξb) = TrLw /Kv (ξb) + TrLw′ /Kv (ξb) ∈ K ∩ Av = A
w′ |v,w′ ̸=w

and hence ξ ∈ B ∨ . This shows that B ∨ is dense in Bw ∨

and thus Bw ∨
= B ∨ ⊗B Bw . Let b1 , . . . , bn
be nonzero generators for the B-module B ∨ . Then they also generate the Bw -module Bw ∨
. Thus we
have DB/A = ∩ni=1 b−1
i B and DBw /Av = ∩n
b
i=1 i
−1
B w and therefore DBw /Av = D B
B/A w .
Corollary 3.8.4. For each nonzero prime ideal P ⊂ B above p = P ∩ A, let DP := DB̂P /Âp ∩ B
where the intersection is inside the completion B̂P . Then we have
Y
DB/A = DP
P

where the product ranges over nonzero prime ideals of B.

Proof. By Lemma 3.8.3 the exponents of any prime in the factorization of both sides coincide.
Definition 3.8.1. For an element α ∈ L, let f (T ) ∈ K[T ] be the minimal polynomial of α over K.
Define its different with respect to L/K to be
(
f ′ (α) if L = K(α);
δL/K (α) =
0 if L ̸= K(α)

Note that if α ∈ B, then δL/K (α) ∈ A

Theorem 3.8.5. Suppose all residue field extensions of B/A are separable. Then DB/A is the ideal
generated by the elements δL/K (α) for all α ∈ B.

Proof. By Lemma 2.3.2 we have δL/K (α) ∈ DB/A for all α ∈ B. Thus it suffices to show that for
any nonzero prime P ⊂ B, there exists α ∈ B such that vP (δL/K (α)) = vP (DB/A ).
Fix a nonzero prime ideal P ⊂ B above p ⊂ A. Let v = vp be the discrete valuation of K
associated with p. Let Kv be the completion of K under v. Let w be the valuation on L extending
1
v and equivalent to the valuation vP associated with P (more precisely we have w = e(P|p) vP ). Let
Lw be the completion of L under w. Let K v be the algebraic closure of K and let v̄ be the unique
valuation of K v extending v. The group Aut(K v /Kv ) acts naturally on the set of field embedding
E := HomK (L, K v ). Let E = ⊔gi=1 Ei be the decomposition into Aut(K v /Kv )-orbits. Pick an
embedding τi ∈ Ei in each orbit and let Pi := {x ∈ B, |τi (x)|v̄ < 1}. Then P1 , . . . , Pg are the prime
ideals of B above p. After reordering if necessary, we may assume that P = P1 . By continuity τ1
uniquely extends to an embedding τ1 : Lw ,→ K v .
By Theorem 3.7.4 there exists x ∈ B̂P such that B̂P = Âp [x]. Take an element a ∈ A such that
×
x − a ∈ B̂P . By the Chinese Remainder Theorem we can find an element α ∈ B that satisfies the
following conditions:
• α − x ∈ P2 B̂P ;
• α − a ∈ Pi for all i = 2, . . . , g

75
By Theorem 3.7.4, the first condition implies that B̂P = Âp [α] and then by Lemma 2.3.2 we get

δLw /Kv (α)B̂P = DB̂P /Âp .

The first condition also implies that |τ1 (α) − a|v̄ = |α − a|w = |x − a|w = 1. By the second condition,
for any 2 ≤ i ≤ g and any τ ∈ Ei we have |τ (α) − a|v̄ < 1. Together these implies that
|τ1 (α) − τ (α)|v̄ = |(τ1 (α) − a) − (τ (α) − a)|v̄ = 1, ∀τ ∈ Ei , i > 1.
By definition we have
Y g Y
Y g Y
Y
τ1 (δL/K (α)) = (τ1 (α)−τ (α)) (τ1 (α)−τ (α)) = τ1 (δLw /Kv (α)) (τ1 (α)−τ (α))
τ ∈E1 ,τ ̸=τ1 i=2 τ ∈Ei i=2 τ ∈Ei

Combining the discussion above we get that |τ1 (δL/K (α))|v̄ = |τ1 (δLw /Kv (α))|v̄ and hence
vP (δL/K (α)) = vP (δLw /Kv (α)) = vP (DB̂P /Âp )

By Lemma 3.8.3 this implies that vP (δL/K (α)) = vP (DB/A ) and we are done.
We have a similar description of the relative discriminant ideal ∆B/A in terms of generating
elements, whose proof is much easier and does not require completion or valuation theory.
Proposition 3.8.6. The relative discriminant ideal ∆B/A is generated by the elements
d(ω1 , . . . , ωn ) = det(TrL/K ((ωi ωj ))1≤i,j≤n )
for any tuple of n-elements ω1 , . . . , ωn ∈ B where n = [L : K].
Proof. Let p ⊂ A be a nonzero prime ideal. Then Ap is a PID and Bp is a free Ap -module.
Take an Ap -basis x1 , . . . , xn of Bp . Multiplying by an element in A − p to clear denominators,
we may assume that x1 , . . . , xn ∈ B. Then we have ∆B/A Ap = d(x1 , . . . , xn )Ap and in particular
vp (∆B/A ) = vp (d(x1 , . . . , xn )).
Lemma 3.8.7. Let (A ⊂ B ⊂ C, F ⊂ K ⊂ L) be a tower of finite separable extensions of Dedekind
domains. Then we have
1. DC/A = DC/B DB/A .
[L:K]
2. ∆C/A = ∆B/A NK/F ∆C/B .
Proof. (1) First note that for any ideal b ⊂ B we have b−1 C = (bC)−1 by unique factorization of
−1
ideals. In particular we have DB/A C = (DB/A C)−1 and it suffices to prove the identity of inverse
differents
−1 −1 −1
DC/A = DC/B DB/A .
−1 −1 −1 −1
By definition we have TrL/K (DC/A ) ⊂ DB/A which implies that DB/A DC/A ⊂ DC/B and hence
−1 −1 −1
DC/A ⊂ DC/B DB/A . It remains to show the reverse inclusion.
−1
For any x ∈ DC/B we have TrL/K (xC) ⊂ B and hence TrL/F (xC) ⊂ A, which shows that
−1 −1 −1
DC/B ⊂ DC/A . For any x ∈ DB/A ⊂ K we have

TrL/F (xC) = TrK/F (xTrL/K (C)) ⊂ TrK/F (xB) ⊂ A

−1 −1 −1 −1 −1
and this shows that DB/A ⊂ DC/A . Therefore DC/B DB/A ⊂ DC/A and we are done.
(2) By (1), Theorem 2.3.4 and Proposition 2.3.3 we get
[L:K]
∆C/A = (NK/F ∆C/B )NL/F (DB/A C) = (NK/F ∆C/B )∆B/A .

76
Theorem 3.8.8. Let P ⊂ B be a nonzero prime ideal and let s = vP (DB/A ) be the exponent of P
in the factorization of the relative different ideal DB/A . Assume the residue field extension at P is
separable and let e = e(P|p) be the ramification index where p = P ∩ A. Then we have s = e − 1 if
P is tamely ramified over A and e ≤ s ≤ e − 1 + vP (e) if P is wildly ramified over A.
We note that all statements except the inequality s ≤ e − 1 + vP (e) have been proved in Theorem
2.3.5, without any assumption on the residue field extension.
Proof. By Lemma 3.8.3 we may assume that A and B are both complete discrete valuation rings. Let
L0 be the maximal unramified extension of K in L and let B0 be the valuation ring of L0 . We have
DB0 /A = B0 (either by a simple special case of Theorem 2.3.5, or argue directly as follows: B0 = A[θ]
where θ has minimal polynomial φ(T ) ∈ A[T ] whose image in κ[T ] is irreducible and separable,
therefore DB0 /A = φ′ (θ)B0 = B0 ). Then by Lemma 3.8.7 we have DB/A = DB/B0 DB0 /A = DB/B0 .
Therefore after replacing K by L0 , we may assume that L/K is totally ramified and hence
[L : K] = e. Then B = A[α] and the minimal polynomial of α is a degree e Eisenstein polynomial

f (X) = a0 + a1 X + · · · + ae X e ∈ A[X]

/ p2 . Then different ideal DB/A is generated by

where ae = 1, ai ∈ p for all 0 ≤ i ≤ e − 1 and a0 ∈
e
X
f ′ (α) = iai αi−1 .
i=1

Note that vP (a) = evp (a) for any a ∈ A and vP (α) = 1 by Proposition 3.6.1 (the unique extension
of vp to L is w = 1e vP ). Then we see that for any 1 ≤ i ≤ e,

vP (iai αi−1 ) = evp (iai ) + i − 1 ≡ i − 1 mod eZ.

In particular {vP (iai αi−1 ), 1 ≤ i ≤ e} are mutually distinct and we get

vP (f ′ (α)) = min (evp (iai ) + i − 1)

1≤i≤e

For each 1 ≤ i ≤ e − 1 we have vp (ai ) ≥ 1 and hence

evp (iai ) + i − 1 ≥ e + i − 1 ≥ e

while the value of the i = e term is

evp (e) + e − 1 = e − 1 + vP (e).

In particular we always have

vP (f ′ (α)) = min (evp (iai ) + i − 1) ≤ e − 1 + vP (e).

1≤i≤e

When L/K is tamely ramified, we have vP (e) = 0 and thus vP (f ′ (α)) = e−1. When L/K is wildely
ramified, we have vP (e) = evp (e) ≥ e and therefore vP (f ′ (α)) ≥ e.

3.8.2 Ramification groups

Let (K, v) be a Henselian discrete valued field and let (L, w) be a Galois extension of K with Galois
group G = Gal(L/K). Let (A, p) (resp. (B, q)) be the valuation ring and κ = A/p (resp. λ = B/q)
be the residue field of v (resp. w). Here w is the unique valuation on L extending v. Then for
any σ ∈ G and any x ∈ L we have |σ(x)|w = |x|w by uniqueness of the extension. In other words,
for any x ∈ L× we have x−1 σ(x) ∈ B × . We will define certain subgroups of G by requiring that
x−1 σ(x) lies in smaller subgroups of B × .

77
Lemma-Definition 3.8.9. The residue field extension λ/κ is normal and there is a natural sur-
jective homomorphism G → Autκ (λ) whose kernel is defined to be the inertia group of L/K:
σ(x)
I = I(L/K) = {σ ∈ G | σ(x) − x ∈ q, ∀x ∈ B} = {σ ∈ G | ∈ 1 + q, ∀x ∈ B × }.
x
Moreover, the inertia field LI coincides with the maximal unramified extension L0 of K in L and
hence we have a canonical isomorphism Gal(L0 /K) ∼ = Autκ (λ).
Proof. We can reduce all statements to finite subextensions of L and hence we may assume that
L/K is a finite extension. Then the first statement is a special case of Lemma 2.4.2. For any σ ∈ G,
σ(L0 ) is also an unramified extension of K and hence σ(L0 ) = L0 . Therefore L0 /K is a Galois
extension. Since the residue field of L0 is the separable closure of κ in λ, the natural surjective
homomorphism G → Autκ (λ) factors through a surjective homomorphism Gal(L0 /K) → Autκ (λ).
By Lemma 2.4.4 it is also injective since e(L0 /K) = fs (L0 /K) = 1. Thus we have L0 = LI .
Definition 3.8.2. The wild inertia group of L/K is the subgroup of the inertia group I(I/K)
defined by
σ(x)
P = P (L/K) = {σ ∈ G | ∈ 1 + q, ∀x ∈ L× }.
x
Assume from now on that the residue field extension λ/κ is separable and the valuation w on
L is also discrete. Let vL = vq = ew be the normalized discrete valuation equivalent to w so that
vL (L× ) = Z.
Lemma 3.8.10. Let σ ∈ G and let i ≥ −1 be an integer. Let x ∈ B be an element such that
B = A[x] (By Theorem 3.7.4 such an x exists when λ/κ is separable). Then the following are
equivalent:
1. σ acts trivially on B/qi+1 .
2. vL (σ(β) − β) ≥ i + 1 for all β ∈ B.
3. vL (σ(x) − x) ≥ i + 1.
Proof. (1)⇔(2) and (2)⇒(3) are obvious.
(3)⇒(2): Any β ∈ B can be written as β = g(x) for some polynomial g(T ) ∈ A[T ] we have
σ(g(x)) − g(x) = g(σ(x)) − g(x) ∈ (σ(x) − x)B and hence vL (σ(β) − β ≥ vL (σ(x) − x) ≥ i + 1.
Fix x ∈ B such that B = A[x]. Define a function iL/K : G → Z≥0 ∪ {∞} by the formula

iL/K (σ) = vL (σ(x) − x).

Then iL/K (σ) = 1 if and only if σ = 1 and by the Lemma it is independent of the choice of x.
Definition 3.8.3. For each integer i ≥ −1, define the i-th ramification group for L/K to be the
following subgroup of G = Gal(L/K):

Gi := {σ ∈ G | iL/K (σ) ≥ i + 1}

For any subgroup H ⊂ G we can define its ramification subgroup (for the extension L/LH ) and
from the definition we check immediately that

Hi = H ∩ G i , ∀i ∈ Z≥−1 .

If moreover H ⊂ G is a normal subgroup, then we have the ramification subgroups (G/H)i for the
extension LH /K. However the relations between ramification subgroups of G and of the quotient
G/H are subtle and to clarify them one need the “upper ramification groups” defined by shifting
indices.

78
Proposition 3.8.11. {Gi , i ∈ Z≥−1 } form a decresing sequence of normal subgroups of G. We
have G−1 = G, G0 = I is the inertia group, G1 = P is the wild inertia group and Gi = {1} for i
sufficiently large.
Proof. From condition (2) in the lemma we see that the function iL/K takes the same value on any
conjugacy class of G and this implies that Gi are normal subgroups of G. By definition it is clear
that G−1 = G and G0 = I is the inertia group. Fix a uniformizer πL ∈ L. For any σ ∈ G0 = I, we
have σ ∈ P if and only if σ(π L)
πL ∈ 1 + q. This is equivalent to vL (σ(πL ) − πL ) ≥ 2 and then we get
G1 ⊂ P .
Let L0 = LI be the maximal unramified extension of K in L. For any α ∈ B × there exists
α ∈ (B ∩ L0 )× such that α − α′ ∈ πL B. If σ ∈ P then vL (σ(α − α′ ) − (α − α′ )) ≥ 2 and since
′

σ(α′ ) = α′ we get that vL (σ(α) − α) ≥ 2. This shows that σ acts trivially on B/q2 and σ ∈ G1 .
This shows P ⊂ G1 and combined with the previous paragraph we get P = G1 .
Proposition 3.8.12. For each σ ∈ G0 , we have iL/K (σ) = vL (σ(πL ) − πL ) and σ ∈ Gi if and only
i
if σ(πL )/πL ∈ 1 + πL B.
Proof. Let H = G0 . Then L0 = LH be the maximal unramified extension of K in L. Then for each
i ≥ 0 we have Hi = H ∩ Gi = Gi since Gi ⊂ G0 = H. Thus we may replace K by L0 and assume
that L/K is totally ramified. Then we have B = A[πL ] and the result follows.
−1
Proposition 3.8.13. The map σ 7→ πL σ(πL ) induces natural injective group homomorphisms
i
×
G0 /G1 = I/P ,→ λ and Gi /Gi+1 ,→
1+π LB
i+1
∼
= qi /qi+1 for each i ≥ 1, that are independent of the
1+πL B
choice of the uniformizer πL . In particular if char(κ) = p > 0, then the order of G0 /G1 is prime to
p and P = G1 is the Sylow p-subgroup of G0 .

Proof. Fix an index i ∈ Z≥0 . For any σ ∈ Gi and u ∈ B × we have vL ( σ(u)

u −1) = vL (σ(u)−u) ≥ i+1
and hence σ(u)
u ∈ 1 + π i+1
L B.
For any σ, τ ∈ Gi , we have
στ (πL ) τ (πL ) σ(πL )
= σ( ) .
πL πL πL
Since u := τ (π L) i × i+1
πL ∈ 1 + πL B ⊂ B we have σ(u) ∈ u(1 + πL B) by the observation at the beginning
of the proof. Then we get that

στ (πL ) σ(πL ) τ (πL ) i+1

∈ (1 + πL B).
πL πL πL
−1
This shows that the map σ 7→ πL σ(πL ) induces injective group homomorphisms
i
1 + πL B
Gi /Gi+1 ,→ i+1
, ∀i ∈ Z≥0 .
1 + πL B

For any u ∈ B × and σ ∈ Gi , by the observation at the beginning of the proof we have
−1 −1
(uπL )−1 σ(uπL ) = u−1 σ(u)πL σ(πL ) ∈ πL i+1
σ(πL )(1 + πL B).

Therefore the homomorphisms are independent of the choice of the uniformizer πL .

When i = 0 the target is B × /(1 + πL B) = λ× . When i ≥ 1 we have a canonical group
i
1+πL B ∼ i i+1
isomorphism 1+πi+1 B
= q /q sending u to u − 1.
L
Note that for any i ∈ Z≥1 , the quotient qi /qi+1 is a λ-vector space of dimension 1. In particular
when char(κ) = p > 0, the quotients Gi /Gi+1 have p-power order when i ≥ 1 and G0 /G1 is a
subgroup of λ× and thus its order is prime to p by Lemma 1.4.1.

79
Theorem 3.8.14. Suppose G is finite. Let gi := |Gi | be the cardinality of the ramification groups
for each i ≥ −1. Then we have
X ∞
X
vL (DB/A ) = iL/K (σ) = (gi − 1).
σ∈G,σ̸=1 i=0

Proof. We may assume that L/K is totally ramified. Let f (T ) ∈ A[T ] be the minimal polynomial
of x. Since B = A[x] we have DB/A = f ′ (x)B and hence
X X
vL (DB/A ) = vL (f ′ (x)) = vL (x − σ(x)) = iL/K (σ).
σ∈G,σ̸=1 σ∈G,σ̸=1

By definition, the function iL/K takes constant value i on Gi−1 \Gi for all i ≥ 0 such that Gi−1 ̸= {1}.
Then we get
X ∞
X ∞
X
iL/K (σ) = i(gi−1 − gi ) = (gi − 1).
σ∈G,σ̸=1 i=0 i=0

Example 3.8.1. Let K = Qp and L = Qp (ζpn ). Recall that the minimal polynomial of π := 1 − ζpn
over Q becomes an Eisenstein polynomial in Qp [T ] and in particular remain irreducible in Qp [T ].
Then L/K is a totally ramified Galois extension with G = Gal(L/K) ∼ = (Z/pn Z)× . For any integer
n
a coprime to p, let σa ∈ G be the image of (the mod p class of) a under this isomorphism. The
element π is a uniformizer in L. We have A = Zp and B = A[π]. Recall from (1.4.4) that
Y Y
NL/K (π) = σ(π) = (1 − ζ i ) = p.
σ∈G i∈(Z/pn Z)×

Let a be an integer coprime to p and let m = vp (a − 1) ∈ Z≥0 . Then we have

m
p
iL/K (σa ) = vL (π − σa (π)) = vL (ζpn − ζpan ) = vL (1 − ζpa−1
n ) = vL (1 − ζpn ).

Thus if m < n we have iL/K (σa ) = [Qp (ζpn ) : Qp (ζpn−m )] = pm , while if m ≥ n we have σa = 1 and
iL/K (σa ) = ∞. The ramification subgroups of G are as follows:

G0 = G−1 = G ∼ = (Z/pn Z)× ,

∼ 1 + pZ ,
G1 = · · · = Gp−1 =
1 + pn Z
...
1 + pm Z
Gpm−1 = · · · = Gpm −1 ∼
= ,
1 + pn Z
Gi = {1}, ∀i ≥ pn−1

In particular if n = 1 we have G1 = {1} and L/K is tamely ramified. When p = 2 we always have
G0 = G1 . The filtrations {Gi , i ∈ Z≥−1 } on G is thus essentially the natural filtration on (Z/pn Z)× ,
one introduces the upper numbering ramification groups Gv which
except a shift of index. In general m
m ∼ 1+p Z
in this special case are G = 1+pn Z .

80
Chapter 4

Ideal class groups and unit groups

4.1 Finiteness of ideal class groups

We let K be a number field of degree n = [K : Q].
Lemma 4.1.1. For any constant C > 0, there are only finitely many nonzero ideals a ⊂ OK whose
absolute norm satisfies N a := |OK /a| ≤ C.
Proof. Let N := (⌊C⌋ + 1)! ∈ Z>0 . Then for any ideal a ⊂ OK with |OK /a| ≤ C we have
N OK ⊂ a ⊂ OK . Since OK is a free abelian group of rank n = [K : Q] the quotient OK /N OK is a
finite group and hence there are only finitely many possibilities for a.
The basic idea to prove finiteness of ClK is to find a representative in any ideal class whose norm
is bounded by some constant.
Theorem 4.1.2. For a number field K, the ideal class group ClK is finite.
Proof. Choose a Z-basis ω1 , . . . , ωn of OK . For a class of ClK we first choose a fractional ideal in it
of the form b−1 for some nonzero ideal b ⊂ OK . Consider the finite set
n
1
X
S = {α = mi ωi ∈ OK | mi ∈ Z, 0 ≤ mi < (N b) n + 1, ∀i}
i=1

whose cardinality satisfies |S| > N b = [OK : b]. By pigeonhole principle, there exists two different
elements α, β ∈ S such that α − β ∈ b. Then a := (α − β)b−1 is a nonzero ideal of OK in the same
Pn 1
class of b−1 . Write α − β = i=1 ri ωi . Then |ri | < (N b) n + 1 and we get
n
Y n
X
|NK/Q (α − β)| = | ri τj (ωi )|
j=1 i=1
n
" n
#
Y 1
X (4.1.1)
< ((N b) + 1)
n |τj (ωi )|
j=1 i=1
1
n
< C((N b) + 1) n

Qn Pn
where C = j=1 ( i=1 |τj (ωi )|) is a constant independent of b. Then we get that
1
N a = |NK/Q (α − β)|(N b)−1 < C(1 + (N b)− n )n
We can replace b by xb for any 0 ̸= x ∈ OK without changing the ideal class. Then N b can be
arbitrarily large and we conclude that each ideal class contains an ideal a whose norm is bounded
by a constant. Since there are only finitely many ideals with bounded norm, we conclude that ClK
is finite.

81
 The upper bound C in this proof is usually quite large and it is hard to use it to determine the
structure of ClK in practice. Next we introduce Minkowski’s analytic method to achieve a better
upper bound.

4.1.1 Minkowski’s theory

Let KR := K ⊗Q R. For each complex embedding τ ∈ HomQ (K, C), let τ̄ be its complex conjugate,
defined as the composition of τ with complex conjugation on C. If τ = τ̄ , then τ is a real embedding
K ,→ R; otherwise we say τ is a complex embedding. Let {ρi : K ,→ R, 1 ≤ i ≤ r1 } be all the real
embeddings and choose complex embeddings {σj : K ,→ C, 1 ≤ j ≤ r2 } so that {σj , σ̄j , 1 ≤ j ≤ r2 }
is the set of all complex embeddings. Then we have r1 + 2r2 = n = [K : Q] and the embeddings
ρi , σj induces isomorphisms of R-algebras

KR = K ⊗Q R x ∈_ K (4.1.2)
∼
=
 
R r 1 × Cr 2 (ρ1 (x), . . . , ρr1 (x), σ1 (x), . . . , σr2 (x))
_
∼
=
 
Rn (ρ1 (x), . . . , ρr1 (x), Re(σ1 (x)), Im(σ1 (x)), . . . , Re(σr2 (x)), Im(σr2 (x))

The starting point of Minkowski’s theory is to view OK and any fractional ideal a as lattices in the
Euclidean space KR ∼= Rn .
Definition 4.1.1. A subgroup Γ ⊂ Rn is

• a lattice if it is free of rank n and contains a basis of Rn ;

• discrete if any γ ∈ Γ has an open neighborhood U ⊂ Rn such that U ∩ Γ = {γ};
• cocompact if there exists a compact subset Ω ⊂ Rn whose translations by Γ cover Rn .
Example 4.1.1. Any fractional ideal a ⊂ K of OK is a lattice in KR . Indeed, by Proposition 1.3.2
a is a free abelian group of rank n = [K : Q]. There exists a positive integer m ∈ Z>0 such that
ma ⊂ OK is a finite index subgroup. Since OK contains a Q-basis of K and hence an R-basis of
KR , ma (and hence a) also contains an R-basis of KR . Therefore a is a lattice in KR .
√
subgroup of√Rn of rank n may
In general, a √ √ not be a lattice. For example, if we embed Z[ 2]
into R2 by a + b 2 7→ (a + b 2, 0). Then Z[ 2] has rank 2 but √ it is not a lattice since it does
√ not
2 2
contain
√ any basis
√ of R . However,
√ if we use the embedding Z[ 2] ,→ R defined by a + b 2 7→
(a + b 2, a − b 2), then Z[ 2] becomes a lattice in R2 .
Lemma 4.1.3. A subgroup Λ ⊂ Rn is a lattice if and only if it is a discrete cocompact subgroup.
Proof. See [Neu99, Chapter I.§4]. Note that a lattice in our definition is called a “complete lattice”
in loc. cit.
Fix a Z-basis v1 , . . . , vn of a lattice Λ ⊂ Rn . The compact subset
n
X
E := { xi vi | 0 ≤ xi ≤ 1, ∀i}
i=1

is a fundamental domain for Λ and by definition we have

vol(Rn /Λ) = vol(E) = | det[v1 v2 · · · vn ]|.

82
Theorem 4.1.4. Let Λ ⊂ Rn be a lattice and X ⊂ Rn a convex, centrally symmetric subset.
Suppose that vol(X) > 2n vol(Rn /Λ). Then X ∩ Λ contains a nonzero element.
Proof. Let Y = 12 X. Then vol(Y ) = 1
> vol(Rn /Λ) and we get
2n vol(X)
X X
vol(E) = vol(Rn /Λ) < vol(Y ) = vol(Y ∩ (λ + E)) = vol(E ∩ (Y − λ))
λ∈Λ λ∈Λ

where the equality follows from the translation invariance of Lebesgue measure (i.e. it is a Haar
measure on the additive group Rn ). Hence there exists λ1 , λ2 ∈ Λ with λ1 ̸= λ2 such that

E ∩ (Y − λ1 ) ∩ (Y − λ2 ) ̸= ∅.

This means that there exists y1 , y2 ∈ Y with

y1 − y2 = λ1 − λ2 ∈ Λ − {0}.

To finish the proof, it remains to show that y1 − y2 ∈ X. After writing y1 − y2 = 21 (2y1 − 2y2 ), this
follows from the assumption that X is convex and centrally symmetric.
Lemma 4.1.5. For any ideal 0 ̸= a ⊂ OK , we have
p
vol(KR /a) = 2−r2 |dK |N a.

Proof. Let ω1 , . . . , ωn be a Z-basis of OK . Let ι : OK ,→ Rn be the embedding induced by (4.1.2).

For simplicity we rename the elements in the set HomQ (K, C) by

{τ1 , . . . , τn } = {ρ1 , . . . , ρr1 , σ1 , σ̄1 , . . . , σr2 , σ̄r2 }

Then we have an identity of matrices in Mn (C):

[ι(ω1 ), . . . , ι(ωn )] = C −1 [τi (ωj )]

where C is the
 block
 diagonal matrix whose blocks are Ir1 (identity matrix of size r1 ) followed by
1 i
r2 copies of . Thus we get
1 −i
p
vol(KR /OK ) = | det[ι(ω1 ), . . . , ι(ωn )]| = 2−r2 | det(τi (ωj ))| = 2−r2 |dK |.

(See Example 1.3.1 for the last equality.) From this we deduce that for any ideal 0 ̸= a ⊂ OK ,
p
vol(KR /a) = vol(KR /OK )|OK /a| = 2−r2 |dK |N a.

We extend the norm map NK/Q : K → Q to a map N : Rr1 × Cr2 ∼

= KR → R by

N (x1 , . . . , xr1 , z1 , . . . , zr2 ) = x1 · · · xr1 |z1 · · · zr2 |2 . (4.1.3)

For any t > 0, define the following subset
r1
X r2
X
Xt = {(x1 , . . . , xr1 , z1 , . . . , zr2 ) ∈ KR | |xi | + 2 |zj | ≤ t}
i=1 j=1

Then for any x = (x1 , . . . , xr1 , z1 , . . . , zr2 ) ∈ Xt , we have

n
tn
P P
Y Y |xi | + 2 |zj |
|N (x)| = |xj | |zj | ≤ ≤ n
n n

83
Exercise 4.1.1. Let Vt (r1 , r2 ) be the volume of Xt under the Lebesgue measure on Rr1 × Cr2 ∼
= Rn
∼ 2
where the identification comes from the usual isomorphism C = R by taking real and imaginary
parts.
1. Show that Xt is convex and centrally symmetric.
2r2
2. Show that Vt (0, r2 ) = ( π2 )r2 (2r
t
2 )!
.
tn r1 π r2
3. Show that vol(Xt ) = Vt (r1 , r2 ) = n! 2 ( 2 ) .
Hint: Use the homogeneous property Vt (r1 , r2 ) = tr1 +2r2 V1 (r1 , r2 ) to deduce that
2
V1 (r1 , r2 ) = V1 (r1 − 1, r2 )
r1 + 2r2

Theorem 4.1.6 (Minkowski bound). Any ideal class in ClK is represented by a nonzero ideal
a ⊂ OK such that   r2
4 n! p
N a := |OK /a| ≤ |dK |
π nn
This number is called the Minkowski bound. In particular since N a ≥ 1 we get that
p  π  r2 n n  π  n2 nn
|dK | ≥ ≥ .
4 n! 4 n!

r2 n! p
Proof. Denote the right hand side by MK := π4 nn |dK |. In a fixed ideal class, we pick a
fractional ideal of the form b−1 for some nonzero ideal 0 ̸= b ⊂ OK . For any t > 0 such that
t n r1 π r 2 p
vol(Xt ) = 2 ( ) > 2n vol(KR /b) = 2r1 +r2 |dK |N b (4.1.4)
n! 2
there exists nonzero α ∈ b ∩ Xt . Then the ideal a := αb−1 is in the class of b−1 we start with and
it satisfies
tn
N a = |NK/Q (α)|(N b)−1 ≤ n (N b)−1
n
The inequality (4.1.4) is equivalent to
  r2
4 p
tn > n! |dK |N b.
π
Let tn approach the right hand side. Then we see that for any C > MK there exists an ideal a in the
given class with N a ≤ C. Since the norm is always a positive integer, take C to be in the interval
(MK , ⌊MK ⌋ + 1) we see that there exists an ideal a ⊂ OK in the given class such that N a ≤ MK .
The last inequality follows from n = r1 + 2r2 ≥ 2r2 .
Corollary 4.1.7. ClK is finite.
Corollary 4.1.8. For any number field K ̸= Q, we have |dK | > 1. In particular, the only finite
extension of Q in which every prime is unramified is Q itself.
[K:Q]
Moreover, log |dK | is bounded for all number fields K ̸= Q. In particular, |dK | → ∞ as [K :
Q] → ∞.

n n
Proof. One shows that the sequence an = π4 2 nn! is strictly increasing and a2 > 1. Thus |dK | > 1
when K ̸= Q. Then the claim about unramified extensions follows from Corollary 2.3.7. Using
Stirling’s formula:
√  n n θ
n! = 2πn e 12n , 0 < θ < 1
e
one gets an upper bound of log n|dK | .

84
 √ √
Example 4.1.2. Let K = Q( −5) so that O√ K = Z[ −5] and dK = −20. In this case r1 = 0 and
r2 = 1. The Minkowski bound in this case is 4 5/π < 3 and hence any ideal class contains an ideal
of norm 1√or 2. If an ideal a has norm 2, then it must be maximal and above 2, the only such ideal
is (2, 1 + −5) and it is non-principal. Thus we have ClK ∼ = Z/2Z.

4.1.2 Hermite-Minkowski theorem

Theorem 4.1.9. There are only finitely many number fields with given discriminant.
Proof. It suffices to show that there are only finitely many number fields with given discriminant that
are not totally real. Indeed, suppose this has been proved. If a totally real √ field K has discriminant
d, then there are only finitely many possibilities for√ the discrimant of K( −1) by Lemma 3.8.7 and
hence finitely many possibilities for the field K( −1). This implies that there are finitely many
totally real field with any given discriminant.
Let K be a number field with discriminant d and a complex embedding τ0 : K ,→ C (so that
τ0 ̸= τ̄0 ). Then the degree n = [K : Q] is bounded in terms of d by Theorem 4.1.6. Consider the
following convex, centrally symmetric subset of KR :
p
X := {(zτ ) ∈ KR | |Im(zτ0 )| < C |d|, |Re(zτ0 )| < 1, |zτ | < 1∀τ ̸= τ0 , τ̄0 }
where the constant C is chosen such that vol(X) > 2n vol(KR /OK ). By Theorem 4.1.4 there exists
a nonzero element α ∈ OK ∩ X. The fact that α ∈ OK − {0} implies that |NK/Q (α)| ≥ 1 and
combined with the fact that α ∈ X we get that |τ0 (α)| > 1 and hence Im(τ0 (α)) ̸= 0. Thus we have
τ0 (α) ̸= τ¯0 (α) and τ0 (α) ̸= τ (α) for any τ ̸= τ0 , τ̄0 (since for these τ we have |τ (α)| < 1). Then we
deduce that K = Q(α).
The condition α ∈ X implies that the coefficients of the minimal polynomial of α are bounded
in terms of d. Therefore there are finitely many such polynomials and hence finitely many possible
K.
Theorem 4.1.10 (Hermite-Minkowski). Let K be a number field and let S be a finite set of primes
of OK . Then there exists only finitely many extensions L/K of given degree n which are unramified
outside S.
Proof. By Corollary 2.3.7, the factorization of the relative discriminant ideal ∆L/K only involves
the finitely many prime ideals in S. For any p ∈ S and any prime P of L above p, the valuation
vP (DOL /OK ) is bounded above by Corollary 3.7.7 (or by Theorem 3.8.8). Therefore vp (∆OL /OK ) is
also bounded above since ∆OL /OK = NL/K DOL /OK . Then by Lemma 3.8.7, the absolute discrimi-
nant
|dL | = |dK |[L:K] · N ∆OL /OK
is bounded above and hence there are finitely many such extension L/K by Theorem 4.1.9.

4.2 Dirichlet’s unit theorem

We consider a number field K with ring of integers OK . Let µ(K) be the group of roots of unit in
×
K. Then µ(K) is the torsion subgroup of both OK and K × .
We fix the following identification of multiplicative groups induced by the isomorphism in (4.1.2):
KR× ∼
= (R× )r1 × (C× )r2
×
We would like to apply Minkowski’s method to study the multiplicative group OK . First we have
to embed it as a lattice in a real vector space. This is done by the following logarithm map:

Log : KR× / Rr1 +r2 (4.2.1)

(x1 , . . . , xr1 , z1 , . . . , zr2 )  / (log |x1 |, . . . , log |xr1 |, 2 log |z1 |, . . . , 2 log |zr2 |)

85
By definition we see immediately that

ker(Log) ∼
= (µ2 )r1 × U (1)r2

where µ2 = {±1} and U (1) = {z ∈ C, |z| = 1} is the unitary group in one variable.
Exercise 4.2.1. Prove the following equalities:
×
• OK = {x ∈ OK , |NK/Q (x)| = 1};
• µ(K) = {α ∈ OK , |τ (α)| = 1 for any embedding τ : K ,→ C}.
Then give counterexamples to the following claims that are false in general:
×
• OK = {α ∈ K, |NK/Q (α)| = 1};
• µ(K) = {α ∈ K, |τ (α)| = 1 for any embedding τ : K ,→ C}.
× × ×
From this we see that ker(Log) ∩ OK = µ(K). Let Λ := Log(OK ) be the image of OK under
Log.
Recall the extended norm homomorphism N : KR× → R× defined by (4.1.3). We let |N | : KR× →
R>0 be the composition of N with the absolute value map. Let

KR±1 := ker |N | = {x ∈ KR , N (x) = ±1}.

×
Then we have OK ⊂ KR±1 by Exercise 4.2.1. Let H ⊂ Rr1 +r2 be the (r1 + r2 − 1)-dimensional
hyperplane consisting of vectors whose sum of coordinates equal to 0. Then we have the following
commutative diagram in which the rows are short exact sequences:

|N |
1 / K ±1 / K× / R>0 /1
R R

Log Log log

  
0 /H / Rr1 +r2 Σ /R /0

×
where Σ is the summation of coordinates map. Then we have Λ = Log(OK ) ⊂ H.
×
Lemma 4.2.1. The group of roots of unit µ(K) is a finite cyclic group and Λ = Log(OK ) is a
discrete subgroup of H.
Proof. Since ker(Log) is a compact subset of KR and OK is a discrete subgroup of KR (by Lemma
4.1.3), we see that µ(K) = ker(Log) ∩ OK is finite. Then we see that µ(K) is cyclic by Lemma 1.4.1.
The preimage of any bounded subset of Rr1 +r2 under Log is a bounded subset of KR× and thus
×
contains only finitely many elements of OK (since OK is discrete in KR ). So any bounded open
neighborhood of an element λ ∈ Λ in Rr1 +r2 contains only finitely many elements of Λ. After
shrinking the neighborhood, it only contains the single element λ ∈ Λ. Thus Λ is a discrete subset
of H.
× ∼ ×
Theorem 4.2.2 (Dirichlet’s unit theorem). Λ = Log(OK ) = OK /µ(K) is a free abelian group of
rank r1 + r2 − 1. In particular it is a lattice in H.
Proof. By Lemma 4.1.3 and the previous lemma, it remains to show that H/Λ is compact, which is
equivalent to show that KR±1 /OK
×
is compact.
We take a bounded convex, centrally symmetric region X ⊂ KR ∼ = Rn such that vol(X) >
n ±1 −1 −1
2 vol(KR /OK ). For any y ∈ KR we have vol(y X) = vol(X) and y X is also convex centrally
symmetric. By Minkowski’s lattice point theorem 4.1.4 there exists a nonzero α ∈ y −1 X ∩ OK .
Since αy ∈ X lies in a bounded region, we get that |NK/Q (α)| = |N (αy)| ≤ C for some constant C
that only depend on X.

86
 There are only finitely many principal ideals (a1 ), . . . , (am ) of OK whose norms are bounded by
×
C. Hence there exists i such that α = ai u for some u ∈ OK . Then from αy = ai uy ∈ X ∩ ai KR±1
we get that
uy ∈ (a−1
i X) ∩ KR .
±1

−1 ±1 ±1 × ±1
Consider the bounded subset S := ∪m i=1 ((ai X) ∩ KR ) of KR . We have shown that OK · S = KR
× ×
and this implies that KR /OK is compact.
√
Example 4.2.1. For an imaginary quadratic field K = Q( D) with D < 0. We have r1 = 0, r2 = 1
×
and OK = µ(K). √
For real quadratic field K = Q( D) with D > 0. We have r1 = 2, r2 = 0 so that r1 + r2 − 1 = 1.
× ∼
In this case µ(K) = µ2 and OK = µ2 × Z.
×
Definition 4.2.1. A system of fundamental units of K is a set of r1 +r2 −1 elements {ui ∈ OK ,1 ≤
∼ ×
i ≤ r1 + r2 − 1} that projects to a Z-basis of the free abelian group Λ = OK /µ(K).
√ √ ×
Example 4.2.2. Let K = Q( 2). Then ϵ = 1 + 2 is a fundamental unit and OK = ±ϵZ . These
2 2
give all integer solutions of Pell’s equation x − 2y = ±1 by the formula
√ √
x + y 2 = ±(1 + 2)n , ∀n ∈ Z.

Exercise 4.2.2. We outline an alternative approach to (the main step of) Dirichlet’s unit theorem.
Same notation as above.
1. Show that for any index 1 ≤ k ≤ r1 + r2 , for any nonzero α ∈ OK , there exists a nonzero
β ∈ OK with
2 p
|NK/Q (β)| ≤ ( )r2 |dK |
π
such that |τi (β)| < |τi (α)| for all i ̸= k. (Hint: apply Minkowski’s lattice point theorem to a
region in Rr1 +r2 defined by inequalities |xi | ≤ ci , 1 ≤ i ≤ r1 + r2 . )
×
2. For any index 1 ≤ k ≤ r1 + r2 , show that there exists uk ∈ OK such that for any i ̸= k, the
i-th coordinate of Log(uk ) is negative. (Hint: apply part (1) repeatedly, use that there are
only finitely many ideals with bounded norm.)
3. Show that the subgroup of Rr1 +r2 generated by the elements {Log(uk )}1≤k≤r1 +r2 from previ-
ous part has rank r1 + r2 − 1. (Reduce this to the following statement: Let M be an m × m
matrix whose diagonal entries are positive and off diagonal entries are negative. If moreover
the entries in each row sum to 0, then M has rank m − 1. Then prove this statement either
directly or by induction.)

Exercise√ 4.2.3. Let D ̸= 0, 1 be a square free integer. Show that the roots of unit in the quadratic
field Q( D) has the following description
 2πi 4πi

√ µ6 = {±1, ±e 3 , ±e 3 } if D = −3

µ(Q( D)) = µ4 = {±1, ±i} if D = −1

µ2 = {±1} else


Here µn denotes the cyclic group of n-th roots of unit.

Exercise 4.2.4. Let K be a cubic field (i.e. [K : Q] = 3) with only one real embedding. Then
× ×
r1 = r2 = 1 and by Dirichlet unit theorem we know that OK has free rank one. Let u ∈ OK be the
×
normalized fundamental unit (i.e. the unique generator of the free part of OK whose image under
the unique real embedding is > 1). In this exercise we obtain a lower bound for u that can help us
find u in many cases.

87
 1. Identify u with its image under the unique real embedding. Let u, ρeiθ , ρe−iθ be all the
conjugates of u in C. Show that u = ρ−2 and the discriminant du of Z[u] is given by

du = −4 sin2 θ(ρ3 + ρ−3 − 2 cos θ)2

2. Show that
|du | < 4(u3 + u−3 + 6)
(Hint: View this as an inequality between functions in the variable x = ρ3 + ρ−3 .)
3. Conclude that
|dK |
u3 > −7
4
where dK is the discriminant of OK .
√ √
Exercise 4.2.5. Let K = Q( 3 2). Recall from Example 2.3.4 that OK = Z[ 3 2].
1. Show that OK is a principal ideal domain.
2. Let u be the normalized fundamental unit of OK as in the previous Exercise. Show that
u3 > 20.
√
3. Let β := ( 3 2 − 1)−1 . Show that β is a unit between 1 and u2 , conclude that β = u.

4.3 Counting ideals

Let n = [K : Q] be as before. We apply Minkowski’s theory to solve the problem of aymptotic
counts of ideals in OK , which will pave the way for the class number formula.
For any ideal class C ∈ ClK and any t > 0, we denote P by iC (t) the number of nonzero ideals
of OK in C whose absolute norm is ≤ t. We let i(t) := C∈ClK be the number of nonzero ideals
whose norm is ≤ t. Our goal is to establish an estimate of the form
1
iC (t) = κt + O(t1− n )

for an explicit constant κ that is independent of the class C and consequently also get
1
i(t) = κhK t + O(t1− n )

where hK = |ClK | is the class number of OK . We will see that the coefficient of the leading term of
i(t) is
2r1 (2π)r2 RK hK
κhK = p
wK |dK |
where wK := |µ(K)| is the number of roots of unity in K, p RK is the regulator of the number field
K which is roughly the volume of KR±1 /OK ×
. (Compare to |dK | = 2r2 vol(KR /OK ))
Fix a nonzero ideal b ⊂ OK such that [b−1 ] = C. Then for any ideal a ⊂ OK in the class of C,
we have ab = αOK for some nonzero element α ∈ b whose absolute norm is |NK/Q (α)| = N a · N b.
Therefore iC (t) is the cardinality of the following sets that are in bijection:

{Nonzero ideals a ⊂ OK | [a] = C, N a ≤ t}

1:1 / {0 ̸= α ∈ b | |NK/Q (α)| ≤ tN b}/O× (4.3.1)
K
 / (α) = ab
a

a = αb−1 o α
×
In other words, iC (t) is counting the number of OK -orbits in a certain subset of the lattice
b ⊂ KR . Our general strategy will be to find a fundamental domain for this action and count the
number of lattice points in it.

88
 × ∼ ×
Let r := r1 + r2 − 1 be the rank of Λ = Log(OK ) = OK /µ(K). Choose a Z-basis λ1 , . . . , λr of
×
Λ and lift them to a system of fundamental units ϵ1 , . . . , ϵr ∈ OK which induces an isomorphism
× ∼ ∼
OK = µ(K) = Λ.
Consider the following fundamental domain for the lattice Λ in the hyperplane H:
Xr
E := { xi λi | 0 ≤ xi < 1, ∀i} (4.3.2)
i=1

We fix the vector v = (1(r1 ) , 2(r2 ) ) ∈ Rr1 +r2 \H where the first r1 -coordinates equal to 1 and the
remaining r2 coordinates equal to 2.
Then E × Rv is a fundamental domain for the translation action of Λ in Rr1 +r2 and

D := Log−1 (E × Rv) = {x ∈ KR× | Log(x) ∈ E × Rv}

×
is a fundamental domain for the OK action on KR× .
For each ρ > 0, let Dρ := {x ∈ D, |N x| ≤ ρ} and let Dρ+ ⊂ Dρ be the subset of totally positive
elements. In other words,

Dρ+ := {x = (x1 , . . . , xr1 , z1 , . . . , zr2 ) ∈ Dρ | xi > 0, ∀1 ≤ i ≤ r1 }.

1
We notice that Dρ = ρ n D1 and vol(Dρ ) = 2r1 vol(Dρ+ ) for all ρ > 0. From (4.3.1) we see that

1 1 1
iC (t) = |Dt·N b ∩ b \ {0}| = (|b ∩ (tN b) n D1 | − 1)
wK wK
Lemma 4.3.1. Let Λ ⊂ Rn be a complete lattice. Let B ⊂ Rn be a bounded subset with “nice”
boundary. Then for any t > 0, we have the asymptotic estimate

vol(B) n
|Λ ∩ tB| = t + O(tn−1 )
vol(Rn /Λ)

For a precise meaning of the “nice” boundary condition and the proof, see [Mar18, Chapter 6,
vol(tB)
Lemma 2]. Heuristically, it is clear that |Λ ∩ tB| should roughly be equal to the ratio vol(R n /Λ) =
vol(B) n
The point of the Lemma is to show that the error term has lower order O(tn−1 ).
vol(Rn /Λ) t .
By the Lemma, we get the asymptotic estimate

vol(D1 )tN b 1 2r2 vol(D1 ) 1 2r1 +r2 vol(D1+ ) 1

iC (t) = + O(t1− n ) = p t + O(t1− n ) = p t + O(t1− n ) (4.3.3)
wK vol(KR /b) wK |dK | wK |dK |

It remains to compute vol(D1+ ). From the definition of the map Log in (4.2.1) we see that for
x = (x1 , . . . , xr1 , z1 , . . . , zr2 ) ∈ D ⊂ KR× , we have |N (x)| ≤ 1 if and only if Log(x) ∈ E × R≤0 v.
Using polar coordinates for the complex variables zj = ρj eiθj , 1 ≤ j ≤ r2 , we get
Z
vol(D1+ ) = (2π)r2 ρ1 · · · ρr2 dx1 · · · dxr1 dρ1 · · · dρr2
D1+

Next we do a change of variables

(y1 , . . . , yr1 +r2 ) = (log x1 , . . . , log xr1 , 2 log ρ1 , . . . , 2 log ρr2 )

to get
Z rX
1 +r2

vol(D1+ ) =π r2
exp ( yj )dy1 · · · dyr1 +r2
E×R≤0 v j=1

89
Interpret the variables as column vectors, we do a further linear change of variables
r
X
t
(y1 , . . . , yr1 +r2 ) = tj λj + tr+1 v = M · t (t1 , . . . , tr1 +r2 )
j=1

where
M = [λ1 · · · λr v]
is the square matrix of size r1 +r2 = r +1 whose column’s are the vectors λ1 , . . . , λr , v. The ranges of
integration for the new variables become: tj ∈ [0, 1] for all 1 ≤ j ≤ r and tr+1 ∈ (−∞, 0]. Moreover
Pr1 +r2
since λj ∈ H we have j=1 yj = (r1 + 2r2 )tr+1 = ntr+1 . Therefore for the integral is converted to
Z 0
+ r2 π r2 | det(M )|
vol(D1 ) = π | det(M )| entr+1 dtr+1 = (4.3.4)
−∞ n
Let us discuss a bit more about the meaning of | det(M )|.
Definition 4.3.1. The regulator of K, denoted by RK , is the absolute value of any r × r minor
×
of the (r + 1) × r matrix [λ1 , . . . , λr ] = [Logϵ1 , . . . , Logϵr ] where ϵ1 , . . . , ϵr ∈ OK is a fundamental
system of units.
Lemma 4.3.2. Let Ev be the r + 1 = r1 + r2 dimensional cube spanned by E and the vector
1 1
n v = n (1 2 ) whose first r1 coordinates are n1 and the remaining r2 coordinates are n2 . Then
(r1 ) (r2 )

we have
1 1 1
RK = | det(M )| = vol(Ev ) = √ vol(E) = √ vol(H/Λ).
n r+1 r+1
In particular the regulator RK is well-defined and independent of the choice of fundamental system
of units.
Proof. We perform the following row operations on M that does not change det(M ): choose any row
of M , add all the other rows to the chosen one while keeping the other rows unchanged. Then the
chosen row becomes (0, . . . , 0, n) after these operations since the coordinates of the column vectors
λi add up to zero for any 1 ≤ i ≤ r. Therefore we get that
1 v
RK = | det(M )| = | det[λ1 · · · λr ]| = vol(Ev ).
n n
Note that the vector nv lies in the hypersurface H1 ⊂ Rr+1 consisting of vectors whose coordinate
sum equal to one. Therefore we can replace nv by any vector in H1 without changing the determinant
of the (r + 1) × (r + 1) matrix [λ1 · · · λr nv ]. We choose the vector w = r+1
1
(1, . . . , 1) ∈ H1 . Then we
get that RK = | det[λ1 · · · λr w]|. On the other hand, w is orthogonal to H. Therefore we get
1 1
RK = | det[λ1 · · · λr w]| = ∥w∥ · vol(E) = √ vol(E) = √ vol(H/Λ).
r+1 r+1
(By the way, we can also replace w by the special vectors in H1 which is 1 in one entry and 0 at all
other entries and then deduce that vol(Ev ) = RK .)
Combined with (4.3.4) and (4.3.3) we get
Theorem 4.3.3. For any ideal class C ∈ ClK and any t > 0, the number of ideals a ⊂ OK in the
class C whose norm is ≤ t satisfies
2r1 (2π)r2 RK 1
iC (t) = p t + O(t1− n )
wK |dK |
and the number of all ideals a ⊂ OK whose norm is ≤ t satisfies
2r1 (2π)r2 RK hK 1
i(t) = p t + O(t1− n )
wK |dK |

90
 √ ×
Example 4.3.1. Consider a real quadratic field K = Q( D) with D > 1. Let ε ∈ OK be a
fundamental unit. We may assume
√ ε > 1 under a chosen real embedding K ,→ R. (For example
2 log ε
when D = 2 we can choose ε = 2+1.) Then the regulator is RK = log ε. Then we have κ = √ .
|dK |

4.4 Dedekind zeta function and Dirichlet L functions

The asymptotic counting formula for ideals in a number field K (Theorem 4.3.3) can be interpreted
by the Dedekind zeta function of K, which is a kind of weighted count of all ideals in the ring of
integers OK , instead of counting only those with bounded norm as we did in §4.3. The Dedekind
zeta function contains more information than the asymptotic counting formula. For example, it
has an Euler product expansion which leads to, at least in the case of abelian extensions of Q, a
factorization into Dirichlet L-functions, which are purely analytic objects.

4.4.1 Initial definition

Definition 4.4.1. The Dedekind zeta function of K is defined by the series
∞
X
−s
X an
ζK (s) = (N a) =
n=1
ns
0̸=a⊂OK

where the first sum runs over all nonzero ideals of OK and an is the number of ideals of norm n.
Lemma 4.4.1. Let {an }∞ n=1 be a sequence of complex numbers. Suppose there exists σ ≥ 0 such
that for all t > 0 we have the asymptotic estimate
X
an = O(tσ ).
n≤t
P∞
Then the series n=1 anns converges uniformly on any compact subset of the half plane Re(s) > σ
and hence defines a holomorphic function on this half plane.
Pn
Proof. Let A0 = 0 and An := k=1 ak for all n ∈ Z≥1 . The heuristic idea is that from the
σ
assumption
P∞ A n = O(n ) we hope to get an = O(nσ−1 ) and then the series becomes approximately
1
n=1 ns−σ+1 which converges absolutely when Re(s) > σ.
By assumption there exists a constant C > 0 such that An ≤ Cnσ for all n ≥ 0. Fix a compact
subset Ω of the half plane Re(s) > σ there exists σ1 > σ such that Re(s) ≥ σ1 for any s ∈ Ω.
For any positive integers n ≥ m and any s ∈ Ω we have
n n n−1  
X ak X Ak − Ak−1 An Am−1 X 1 1
= = − + A k −
ks ks ns ms ks (k + 1)s
k=m k=m k=m
n−1 Z k+1
An Am−1 X 1
= s − s
+ s A k s+1
dx.
n m k x
k=m

From the bounds Re(s) ≥ σ1 and Ak ≤ Ck σ we get that

k+1 k+1 k+1
kσ
Z Z Z
1 1
|Ak dx| ≤ C dx ≤ C dx
k x s+1
k x σ1 +1
k xσ1 −σ+1

and hence
n−1 Z k+1 Z ∞
X 1 1 C
| Ak dx| ≤ C dx = .
k x s+1
m xσ1 −σ+1 (σ1 − σ)mσ1 −σ
k=m

91
Therefore we get
n
X ak 1 1 C
| | ≤ C( σ1 −σ + σ1 −σ ) + |s| .
ks n m (σ1 − σ)mσ1 −σ
k=m
P∞
This implies the uniform convergence of n=1 anns on the compact set Ω.
By the estimate in Theorem 4.3.3 we see that ζK (s) defines a meromorphic function on the half
plane Res(s) > 1.

4.4.2 Residue formula

We will see that ζK (s) has meromorphic continuation to the whole complex plane with a simple pole
at s = 1 whose residue is determined by the leading term coefficient in the estimate of Theorem 4.3.3.
Let us first give a short proof of a weaker statement which would be sufficient to derive the class
number formula.
We start with a weaker statement in the case of Riemann zeta function ζ(s) = ζQ (s).
P∞
Lemma 4.4.2. Let ζ(s) = n=1 n1s . Then ζ(s) − s−1 1
extends to a holomorphic function on the
half plane Re(s) > 0.
Proof. Although we will prove the stronger statement that ζ(s) has meromorphic continuation to
the whole complex plane later, let us give a short proof of the current weaker result. Consier the
function
∞
X (−1)n
η(s) = s
= 1 − 2−s + 3−s − 4−s + · · ·
n=1
n

By Lemma 4.4.1, the series η(s) defines a holomorphic function on the half plane Re(s) > 0.
Moreover it is easy to see that when Re(s) > 1 we have ζ(s) = (1 − 21−s )−1 η(s). Consequently
this expression gives a meromorphic continuation of ζ(s) to the half plane Re(s) > 0 whose possible
poles are among the points where 21−s = 1 listed as follows:
2πi
sk = 1 + k, k ∈ Z.
log 2
To see that the only pole is at s = s0 = 1, we consider the function
1 2 1 1 2
g(s) = (1 − 31−s )ζ(s) = 1 + − s + s + s − s + ···
2s 3 4 5 6
By Lemma 4.4.1 again we see that g(s) is holomorphic on the half plane Re(s) > 0 so that (1 −
31−s )−1 g(s) also gives the meromorphic continuation of ζ(s) to Re(s) > 0. Comparing the two
expressions
(1 − 31−s )−1 g(s) = (1 − 21−s )−1 η(s)
we see that the only possible pole is at s = 1 (basically because log 2 and log 3 are Z-linearly
independent in C). Finally from the identity η(1) = log 2 we deduce that ζ(s) has a simple pole at
s = 1 with residue lim (s − 1)ζ(s) = 1.
s→1

Corollary 4.4.3. Let {an }∞n=1 be a sequence of complex numbers. Suppose there exsits 0 < ε ≤ 1
and α ∈ C such that for any t > 0 we have the asymptotic estimate
X
an = αt + O(t1−ε ),
n≤t
P∞
Then the series f (s) = n=1 anns extends to a meromorphic function on the half plane Re(s) > 1 − ε
with only one pole at s = 1 of order 1 and residue α.

92
 P∞ bn P
Proof. Let bn := an − α. Then f (s) − αζ(s) = n=1 ns . The assumption implies that n≤t bn =
O(t1−ε ). Then the result follows from Lemma 4.4.1.
Theorem 4.4.4 (Class number formula). For any number field K, the Dedekind zeta function ζK (s)
1
has meromorphic continuation to Re(s) > 1 − [K:Q] . Its only pole is a simple pole at s = 1 with
residue
2r1 (2π)r2 RK hK
Ress=1 ζK (s) = lim (s − 1)ζK (s) = p .
s→1 wK |dK |

4.4.3 Euler product and factorization of the Dedekind zeta function

Theorem 4.4.5. When Re(s) > 1, the Dedekind zeta function can be expressed as an infinite
product over all nonzero prime ideals p ⊂ OK :
Y 1
ζK (s) = .
p
1 − N p−s

Proof. Fix a positive number t > 0. Let p1 , . . . , pr be all the nonzero prime ideals of OK whose
absolute norm if ≤ t. Then for any s ∈ C with Re(s) > 1 we have
r
Y X 1
|ζK (s) − (1 − N p−s
i )
−1
|≤
i=1
|N a−s |
0̸=a⊂OK
N a≤t

When t → ∞, this converges uniformly to 0 on any compact subset of the half plane Re(s) > 1.
Example 4.4.1. Let K = Q(i) so that OK = Z[i] is the ring of Gaussian integers. Recall that the
nonzero prime ideals of OK are classified according to the splitting behaviour over Z as follows:
• Split primes: (a + bi)Z[i] where a2 + b2 = p is a prime number with p ≡ 1(mod4). In this case
the inertia degree is 1 and N ((a + bi)Z[i]) = p
• Inert primes: pZ[i] for (positive) prime numbers p ≡ 3(mod4). In this case inertia degree is 2
and N (pZ[i]) = p2 ;
• Ramified prime: (1 + i)Z[i] with inertia degree 1 and ramification index 2. In this case
N ((1 + i)Z[i]) = 2
In the Euler product
Y 1
ζQ(i) (s) =
1 − N p−s
p⊂Z[i]

the contribution of split primes is Y

(1 − p−s )−2
p≡1(mod4)

and the contribution of inert primes is

Y Y
(1 − p−2s )−1 = (1 + p−s )−1 (1 − p−s )−1
p≡3(mod4) p≡3(mod4)

while the contribution of the ramified prime is (1 − 2−s )−1 .

From this we see that we can factor out a copy of Riemann zeta function ζ(s) from ζQ(i) (s) and
the quotient is
Y Y 1 1 1
(1 − p−s ) (1 + p−s )−s = 1 − + s − s + ···
3s 5 7
p≡1(mod4) p≡3(mod4)

93
Let χ : (Z/4Z)× → C× be the unique nontrivial homomorphism (so that χ(1) = 1 and χ(3) = −1).
Extend χ to a multiplicative function on Z by requiring χ(m) = 0 for any even number. Then the
above quotient is
∞
Y 1 X χ(n)
L(s, χ) := −s
=
p
1 − χ(p)p n=1
ns

Then we have
ζQ(i) (s) = ζ(s)L(s, χ)
Exercise 4.4.1. Use the class number formula for Q(i) to deduce that L(1, χ) = π4 . This is the
famous Leibniz formula
1 1 1 π
1 − + − + ··· =
3 5 7 4
hiding behind which is the fact that Z[i] has class number 1 (or equivalently, it is a UFD).
This factorization of ζQ(i) can be generalized to any cyclotomic field Q(ζn ) (and hence any finite
abelian extension of Q by the Kronecker-Weber theorem). Let us first introduce all the possible
factors.
A Dirichlet character is a group homomorphism χ : (Z/nZ)× → C× for some integer n ≥ 1.
The parity of χ is the number ϵ = ϵ(χ) ∈ {0, 1} such that χ(−1) = (−1)ϵ . We say χ is even if ϵ = 0
and odd if ϵ = 1.
We follow the convention that (Z/1Z)× = {1}. If n|m, then there is a natural homomorphism
(Z/mZ)× → (Z/nZ)× and any character of (Z/nZ)× induces a character of (Z/mZ)× .
The conductor of the Dirichlet character χ is the minimal integer f = fχ ≥ 1 such that χ factors
through (Z/f Z)× . In particular χ(a) is defined for any integer a coprime to the conductor fχ . We
view any Dirichlet character χ also as a multiplicative function χ : Z → C by requiring χ(a) = 0 if
a is not coprime to the conductor fχ . With this convention we introduce:
Definition 4.4.2. The Dirichlet L-functions associated to a Dirichlet character χ is defined by the
series
∞
X χ(n)
L(s, χ) =
n=1
ns

By Lemma 4.4.1 and Lemma 4.4.2, the series L(s, χ) defines a meromorphic function on the half
plane Re(s) > 0 and is holomorphic if χ is nontrivial (if χ is the trivial character one recovers the
Riemann zeta function).
Theorem 4.4.6. Let K/Q be a finite abelian extension and let XK := Hom(Gal(K/Q), C× ) be the
group of characters of Gal(K/Q). Then the Dedekind zeta function of K can be factorized as
Y
ζK (s) = L(s, χ).
χ∈XK

In the case K = Q(i) = Q(ζ4 ), we get (as already explained above) that ζQ(i) = ζ(s)L(s, χ)
where χ : (Z/4Z)× ∼= {±1} is the canonical isomorphism viewed as a Dirichlet character.
Before the proof in general let’s explain the special case of K = Q(ζp ) where p > 2 is an odd
prime. Since pOK = (πOK )p−1 where π = 1 − ζp , the factor at p on both sides are (1 − p−s )−1 .
For any prime ℓ ̸= p, we know that ℓ is unramified in OK , say it factors as ℓOK = p1 · · · pg . Each
prime ideal pi has the same inertia degree f , which is also the order of the decomposition group
Dℓ ∼
= ⟨ℓ⟩ ⊂ (Z/pZ)× . Then we have f g = [K : Q] = p − 1 and also N pi = ℓf for all i = 1, . . . , g.
Hence the contribution of ℓ to ζK (s) is
g
Y
(1 − N p−s
i ) = (1 − ℓ
−f s −g
) .
i=1

94
 2πi
The possible values of χ(ℓ) are ζfa , a = 1, . . . , f where ζf = e f . We group the characters in
XK ∼
= Hom((Z/pZ)× , C× ) into cosets of

⟨ℓ⟩⊥ := {χ ∈ XK , χ(ℓ) = 1}.

The characters from the same coset takes on the same value. We have a canonical isomorphism
×
(Z/pZ)
X/⟨ℓ⟩⊥ ∼
= Hom( , C× )
⟨ℓ⟩⊥

from which we see that

p−1 p−1
|⟨ℓ⟩⊥ | = = = g.
|⟨ℓ⟩| f
Therefore the contribution of ℓ to the right hand side is

Y f
Y
(1 − χ(ℓ)ℓ−s )−1 = ( (1 − ζfa ℓ−s )−1 )g = (1 − ℓ−f s )−g .
χ∈XK a=1

Proof of Thoerem 4.4.6. Let p be a prime number. We have a factorize pOK = (P1 · · · Pg )e and
N Pi = pf for each 1 ≤ i ≤ g. By the Kronecker-Weber Theorem we have an embedding K ⊂ Q(ζN )
for some integer N > 1. Write N = pm N ′ where m ∈ Z≥0 and N ′ is coprime to p. Then
G = Gal(K/Q) is a quotient of (Z/N Z)× ∼ = (Z/pm Z)× × (Z/N ′ Z)× and we view XK as a subset
of the set of Dirichlet characters modulo N . Let Dp ⊂ G be the decomposition group of any Pi
and Ip ⊂ Dp the inertia group. Then Ip is the image of (Z/pm Z)× in G. For any χ ∈ XK , viewed
as a character of (Z/N Z)× , we have χ(p) ̸= 0 if and only if χ factors through (Z/N ′ Z)× , if and
only if χ is trivial on Ip . Let K ′ = K Ip be the inertia field at p. Then we have K ′ = K ∩ Q(ζN ′ )
and G′ := Gal(K ′ /Q) ∼ = G/Ip . Therefore the subset of characters χ ∈ XK such that χ(p) ̸= 0 is
precisely the set XK ′ of characters of G′ . Note that G′ is a quotient of (Z/N ′ Z)× and Dp /Ip ⊂ G′
is the cyclic subgroup of order f generated by the image of p. Then the possible values of χ(p) for
χ ∈ XK ′ are ζfa where 1 ≤ a ≤ f . The number of characters χ ∈ XK ′ that take the same value at p
equals to |G/Dp | = g. Therefore we get

Y Y f
Y
(1 − χ(p)p−s )−1 = (1 − χ(p)p−1 )−1 = (1 − ζfa p−s )−g = (1 − p−f s )−g
χ∈XK χ∈XK ′ a=1

On the other hand, the factor at p of ζK (s) is

g
Y
(1 − N P−s
i )
−1
= (1 − p−f s )−g .
i=1

Corollary 4.4.7. Let K/Q be an abelian extension and let XK := Hom(Gal(K/Q), C× ) be the
group of characters of Gal(K/Q). Then we have
Y 2r1 (2π)r2 RK hK
L(1, χ) = p
χ∈XK ,χ̸=1
wK |dK |

4.4.4 Values at 1
In this section we find explicit formulas for L(1, χ) for nontrivial Dirichlet characters χ that occur
in class number formula of Corollary 4.4.7.

95
 Let χ be a nontrivial Dirichlet character of conductor f = fχ . Then the series
∞
X χ(n)
L(1, χ) =
n=1
n

converges conditionally. The starting point to calculate this sum is the Taylor expansion of the
logarithm function
∞
X zn
log(1 − z) = − .
n=1
n
We take the branch of the multivalued function log in which the values has imaginary part lying
2πi
in the interval (−π, π]. Denote ζf := e f . The series converges absolutely only when |z| < 1.
However, if z = ζfa where (a, f ) = 1 the series also converges to log(1 − ζfa ) by Lemma 4.4.1. Then
we get
f ∞ f
X X 1X
χ(a) log(1 − ζfa ) = − χ(a)ζfan
a=1 n=1
n a=1
(a,f )=1

Recall our convention that χ(a) = 0 if (a, f ) ̸= 1.

Definition 4.4.3. The Gauss sum of a Dirichlet character χ with conductor f is defined by
f f
X 2πia X 2πia
τ (χ) = χ(a)e f = χ(a)e f .
a=1 a=1
(a,f )=1

2πia
Lemma 4.4.8. Let χ be a Dirichlet character with conductor f . Denote ζf := e f .
1. For any n ∈ Z, we have
f
X
χ(a)ζfan = χ(n)τ (χ).
a=1
√
2. We have τ (χ)τ (χ) = χ(−1)f and therefore |τ (χ)| = f.
Proof. (1) When (n, f ) = 1, we can choose k ∈ Z such that kn ≡ 1 mod f . Then we have
χ(k) = χ(n) and hence
f
X X X
χ(a)ζfan = χ(ak)ζfakn = χ(k) χ(a)ζfa = χ(n)τ (χ).
a=1 a∈(Z/f Z)× a∈(Z/f Z)×

Now suppose d = (n, f ) > 1 and write f = dg and n = dm. Then we have χ(n) = 0 and
f
X X X
χ(a)ζfan = ζfdbm χ(a) = 0
a=1 b∈(Z/gZ)× a∈(Z/f Z) ×

a≡b mod g

where we used that the inner sum equals to zero since χ is a nontrivial character on the subgroup
ker((Z/f Z)× → (Z/gZ)× ).
(2) By (1) we have
f f f
(a+1)n (a+1)n
X X X X X
τ (χ)τ (χ̄) = χ(n)τ (χ)ζfn = χ(a)ζf = χ(a) ζf
n=1 n=1 a∈(Z/f Z)× a∈(Z/f Z)× n=1

96
The inner sum is nonzero only when a ≡ −1 mod f in which case it is equal to f . Thus we get
that τ (χ)τ (χ) = χ(−1)f and then we deduce that

|τ (χ)|2 = τ (χ)τ (χ) = χ(−1)τ (χ)τ (χ̄) = f

By Lemma 4.4.8 we deduce that

f
τ (χ) X
L(1, χ) = − χ(−a) log(1 − ζfa )
f a=1
(a,f )=1
(4.4.1)
f
τ (χ) X
=− χ(a)[χ(−1) log(1 − ζfa ) + log(1 − ζf−a )]
2f a=1
(a,f )=1

where the second equality follows by observing that the sum is invariant under the substitution
a → f − a and then taking average. From this identity we deduce:
Theorem 4.4.9. Let χ be a nontrivial Dirichlet character with conductor f .
1. If χ is odd, i.e. χ(−1) = −1, then
f f
τ (χ)πi X τ (χ)πi X
L(1, χ) = χ(a)a = χ(a)a.
f2 a=1
f 2 a=1
(a,f )=1

2. If χ is even, i.e. χ(−1) = 1, then

f
τ (χ) X
L(1, χ) = − χ(a) log |1 − ζfa |.
f a=1
(a,f )=1

Proof. We observe that 1 − ζfa = −ζfa (1 − ζf−a ). Then (1) follows from the identity

2πia
log(1 − ζfa ) − log(1 − ζf−a ) = log(−ζfa ) = − πi
f

while (2) follows from the identity

log(1 − ζfa ) + log(1 − ζf−a ) = 2 log |1 − ζfa |.

4.4.5 Density of primes

In this section we explore the applications of the following immediate consequence of Corollary 4.4.7:
Corollary 4.4.10. For any nontrivial Dirichlet character χ we have L(1, χ) ̸= 0.
In fact one can prove a stronger statement that L(s, χ) does not vanish on the line Re(s) = 1,
which would imply the prime number theorem. Of course this would be a consequence of the
Generalized Riemann Hypothesis, which predicts that all the nontrivial zeros of L(s, χ) are on the
line Re(s) = 21 .

97
Lemma 4.4.11. Let {un }∞ n=1 be a sequence of real numbers such that un ≥ 2 for all n ≥ 1.
Q∞
Suppose the infinite product f (s) = n=1 (1 − u−s
n )
−1
converges uniformly on any compact subset of
Re(s) > 1. Then for all Re(s) > 1 we have
∞
X
log f (s) = u−s
n + g(s)
n=1

where g(s) is bounded in a neighborhood of s = 1.

Proof. By the uniform convergence we have
∞ ∞ X
∞ ∞
X X 1 X
log f (s) = − log(1 − u−s
n )= sm
= u−s
n + g(s)
n=1 n=1 m=1
mu n n=1

where
∞ X
X 1
g(s) := .
n=1 m≥2
musm
n

Let σ := Re(s) > 1. Then we have

X 1 X 1 1 1
| sm
|≤ mσ
= 2σ (1 − u−1
n ) < 2σ .
mun 2un 2un un
m≥2 m≥2

1
The convergence of f (2σ) when σ > 2 implies that g(s) is a holomorphic function on Re(s) > 12 .
Proposition 4.4.12.
1 X X
log ζK (s) ∼ log ∼ N p−s ∼ N p−s
s−1 p p,deg p=1

where in the last summation, if p ∩ Z = pZ, then deg p := f (p|pZ) = [OK /p : Fp ] denotes the inertia
degree.
1
Proof. By Theorem 4.4.4 we get log ζK (s) ∼ log s−1 and by the previous lemma we get log ζK (s) ∼
−s
P
p N p . For any prime number p, there are at most [K : Q] primes of OK above p. Then we get
that
X [K : Q] ∞
X X 1
N p−s ≤ 2σ
≤ [K : Q] 2σ
p
p n=1
n
deg p≥2

and this is finite when σ = Re(s) > 12 .

Corollary 4.4.13. There are infinitely many prime ideals p ⊂ OK with deg(p) = 1.

Definition 4.4.4. Let X be a set of nonzero prime ideals of OK . We say that X has Dirichlet
density δ, denoted δ(X) = δ, if the limit
−s −s
P P
p∈X N p p∈X N p
lim+ 1 = lim P −s
s→1 log s−1 s→1+ p Np

exists and equals to δ. Here the summation in the denominator of the right hand side runs over all
nonzero prime ideals of OK .
Theorem 4.4.14. Let n > 1 be an integer. For any a ∈ Z with (a, n) = 1, the set of prime numbers
1
that are congruent to a modulo n has Dirichlet density φ(n) where φ(n) = |(Z/nZ)× |.

98
Proof. For
Pany nontrivial character χ : (Z/nZ)× → C× , since L(s, χ) is holomorphic at s = 1 we
−s
get that p χ(p)p is bounded in a neighborhood of s = 1. Therefore from Corollary 4.4.10 we
get that
X 1 X X 1 X −s 1 X X
p−s = χ(a)−1 χ(p)p−s = p + χ(a)−1 χ(p)p−s
φ(n) χ p
φ(n) p
φ(n) p
p≡a mod n χ̸=1

where the second term is bounded and the result follows.

This implies Dirichlet’s theorem on primes in arithmetic progressions:
Corollary 4.4.15. Let n > 1 be an integer and a ∈ Z with (a, n) = 1. Then there are infinitely
many prime numbers p such that p ≡ a(mod n).
We also have the following result of a similar flavour:
Theorem 4.4.16. For any ideal class C ∈ ClK , the set of prime ideals in C has Dirichlet density
1
hK .

Using class field theory, these can be viewed as special cases of the following
Theorem 4.4.17 (Chebatorev density theorem). Let L/F be a finite Galois extension of number
 class C ⊂ Gal(L/F ), the set of prime ideals p ⊂ OF that are unramified
fields. For any conjugacy

L/F |C|
in OL such that p = C has Dirichlet density [L:F ].

4.5 Adèles and idèles

4.5.1 Basic definitions
Let K be a number field. For each place v of K, let Kv be the completion of K under | · |v . If v
is a finite (i.e. non-archimedean) place we let pv ⊂ OK be the corresponding prime ideal and let
(Ôv , p̂v ) be the valuation ring of Kv . Choose a uniformizer ϖv ∈ Ov . Recall the exact sequence of
abelian groups
×
1 → OK → K × → JK → ClK → 1
where JK is the group of fractional ideals of OK . By the unique factorization of ideals we have an
isomorphism
Kv× /Ov× ∼
M
= JK
v finite

(ϖvnv )v nv
Q
sending to v pv where the product runs over the finite places of K.
Definition 4.5.1. The group of finite idèles of K is the multiplicative group
Y
Af,×
K := {(xv ) ∈ Kv× | xv ∈ Ôv× for all but finitely many v}.
v finite

The group of idèles of K is the product

f,×
A× ×
K = KR × AK

The group Af,×

Q
K has a topology generated by open sets of the form v finite Uv where Uv is an
open subset of Kv and Uv = Ov for all but finitely many v. Equip AK = KR× × Af,×
× × ×
K with the
product topology. Then Af,× K and A ×
K are locally compact (Hausdorff) abelian groups. Moreover,
×
:= v finite Ôv× is the maximal compact subgroup of Af,×
Q
ÔK K and we have canonical isomorphisms

× ∼
Kv× /Ov× ∼
f,×
M
AK /ÔK = = JK
v finite

99
We embed K × diagonally as subgroups of Af,× ×
K and AK . By the discussion above we have isomor-
phisms
× ∼ × × ∼
K × \Af,× × ×
K /ÔK = K \AK /KR ÔK = ClK

Definition 4.5.2. The idèle class group of K is the quotient group CK := K × \A×
K.

Definition 4.5.3. The ring of finite adèles of K is

Y
AfK := {(xv )v ∈ Kv | xv ∈ Ôv for all but finitely many v}.
v

The ring of adèles of K is the product ring AK = KR × AfK . As the notation suggests, the group of
(finite) idèles is precisely the group of invertible elements of the (finite) adèle ring by definition.

Similar to idèle groups, the adèle rings AK and AfK are locally compact topological rings. A
basis of open sets of AfK are v finite Ωv where Ωv is an open subset of Kv for each v and Ωv = Ov
Q

for all but finitely many v. Also ÔK := v finite Ôv is the maximal compact subring of AfK and we
Q

embed K diagonally as a subring of AfK and AK .

Remark 4.5.1. Beware that the topology of A× K is finer than but not the same as the subspace
×
topology induced from AK . Indeed, the compact open subgroup ÔK ⊂ A×K is not the intersection
×
of AK with any open subset of AK . In general for a topological ring R, the subspace topology on
the multiplicative group R× may not be fine enough to make the map x 7→ x−1 continuous. To
make R× into a topological group, we should equip R× with the subspace topology induced from
the embedding R× ,→ R × R defined by x 7→ (x, x−1 ). In the case where R = Kv is topological
field, this topology on Kv× the same as the subspace topology from Kv . In the case R = AK (resp.
R = AfK ), this gives precisely the topology on A× f,×
K (resp. AK ).

4.5.2 Structure of adèle quotient group

Lemma 4.5.2. The subgroup K ⊂ AK is discrete and the quotient K\AK is compact.

Proof. We have K ∩ (KR × ÔK ) = OK and this implies that K is discrete in AK . Indeed, since
OK is discrete in KR we can take a sufficiently small open neighborhood Ω of 0 in KR so that
(Ω × ÔK ) ∩ K = {0}.
Next we show that K + ÔK = AfK . For any x = (xv ) ∈ AfK , there exists a nonzero element
α ∈ OK such that αxv ∈ Ôv for all finite place v. Let S be the set of finite places v such that
×
v(α) > 0. Then S is a finite set and for any finite place w ∈
/ S we have α ∈ Ôw and xw ∈ Ôw . By
the Chinese Remainder Theorem, there exists β ∈ OK such that β − αxv ∈ αOv for all v ∈ S. Then
we get that α−1 β − xv ∈ Ôv for all finite place v.
Then we get that
K\AK ∼ = OK \(KR × ÔK ) (4.5.1)
Let E ⊂ KR be a compact subset such that E + OK = KR . By the isomorphism above the compact
set E × ÔK surjects onto K\AK and thus K\AK is compact.

The isomorphism (4.5.1) allows us to view K\AK as a fiber bundle over the compact manifold
OK \KR (which is homeomorphic
Q to a product of [K : Q] circles) whose fibers are homeomorphic to
the compact set ÔK = v Ôv . Let us describe this picture in more details.
For any nonzero ideal a = v pvnv ⊂ OK , the ideal aÔK = v (aÔv ) = v ϖvnv Ôv is compact
Q Q Q

open in ÔK and we have ÔK /aÔK ∼ = OK /a. From the isomorphism Ôv ∼ = lim O /pn we deduce
←−n K v
an isomorphism of topological rings
ÔK ∼
= lim O /a
← − K
a

100
where the limit is over the set of nonzero ideals of OK , ordered by inclusion (or divisibility). For any
nonzero ideals a ⊂ b the transition map in the system is the natural homomorphism OK /a → OK /b.
Consequently we get

K\AK ∼
= lim OK \(KR × (OK /a)) ∼
= lim a\KR
← −
a
← −
a

where in the middle term OK acts diagonally on the product KR × (OK /a) and the second isomor-
phism is induced by the embedding KR ,→ KR × (O/a) sending x to (x, 0).
In other words the topological group K\AK consists of a compatible system of points in the
tower of compact Lie groups KR /a, each isomorphic to a product of [K : Q] copies of S 1 . In
particular, K\AK is connected. We will see below that in contrast the idèle class group K × \A×
K
is non-compact and highly non-connected. In fact from class field theory we know that the group
of connected components of K × \A× ab
K can be identified with Gal(K /K) where K
ab
is the maximal
abelian extension of K.
Q
Example 4.5.1. Consider the case K = Q. Then ÔK = Ẑ = p Zp and we have

Q\AQ ∼
= lim Z\(R × (Z/nZ)) ∼
= lim R/nZ ∼
= lim C1
←− n
←− ←−n n

where C1 = {z ∈ C× , |z| = 1} is the unit circle group and the limit is taken over the set of positive
integers ordered by divisibility (not the usual order by absolute value). For any positive integers
m|n the transition map is the natural projection R/nZ → R/mZ. In the last isomorphism, for each
n ∈ Z≥1 we identify R/nZ ∼ = C1 by x 7→ e2πix/n . Thus for each m|n, the transition map in the last
term from the n-th copy of C1 to the m-th copy of C1 is z 7→ z n/m .

4.5.3 Structure of idèle class group

We start with the case of Q. Since the class group of Q is trivial we have Af,×
Q = Q× Ẑ× . Then we
get
× × ∼
CQ = Q× \A× × ×
Q = Q \(R × Q Ẑ ) = R>0 × Ẑ
×
(4.5.2)
More precisely, for any idèle (x∞ ; xp ) ∈ A× ×
Q there exists a unique r ∈ Q>0 such that rxp ∈ Zp for
all prime number p. Then the isomorphism sends the class of (x∞ ; xp ) to (r|x∞ |; sign(x∞ )rxp ).
In particular the idèle class group Q× \A×Q is noncompact. The failure of compactness is due to
the product formula v≤∞ |x|v = 1 for all x ∈ Q× , which means that Q× lies in the subgroup of
Q

norm 1 idèles A1Q = {(x∞ ; xp ) ∈ A×

Q
Q | v≤∞ |xv |v = 1} that sits in a short exact sequence

1 → A1Q → A×
Q → R>0 → 1.

The isomorphism above restricts to an isomorphism Q× \A1Q = ∼ Ẑ× .

To generalize these to all number fields K, we first fix normalized absolute values on the com-
pletions Kv as follows:
• If v is a finite place corresponding to a prime ideal p ⊂ OK , then we define

|x|v := (N p)−vp (x) , ∀x ∈ Kv×

where as usual N p = |OK /p| is the absolute norm and vp is the normalized additive valuation
on Kv such that vp (ϖv ) = 1 for any uniformizer ϖv ∈ Ov .
• If v is real, we let | · |v be the usual absolute value on R;

101
 • If v is complex, we let | · |v = | · |C be the square of the usual absolute value on C, i.e.
|z|C := z z̄ = |z|2 . (Strictly speaking it is not an “absolute value” since it does not satisfy
triangle inequality. But this is a minor issue of terminologies and will not matter us in the
following.)
Theorem 4.5.3 (Product formula). For any α ∈ K × we have
Y
|α|w = 1
w

where the product runs over all places w of K.

Proof. For any place v of Q and any place w of K above v, according to the definition of our
×
normalized absolute value | · |w we have |x|w = |NKw /Qv (x)|v for all x ∈ Kw . This is clear if w is
archimedean while if w is non-archimedean this follows from Theorem 3.3.10, keeping in mind that
the normalized absolute value | · |w differs from the unique extension of | · |v to Lw by a power of
[Lw : Kv ]. For any α ∈ K × and any place v of Q, by Proposition 3.5.4 we have
Y
NK/Q (x) = NKw /Qv (x) ∈ Qv
w|v

and thus we get that

Y YY YY Y
|α|w = |x|w = |NKw /Qv (x)|v = |NK/Q x|v = 1
w v w|v v w|v v

where the last equality follows from the product formula for Q.
Let A1K := {(xv ) ∈ AK | v |x|v = 1} be the group of norm 1 idèles that sits in a short exact
Q
sequence
∥·∥
1 → A1K → A×
K −
−→ R>0 → 1 (4.5.3)
where we use the norm map on adèles defined by
Y
∥(xv )v ∥ := |xv |v , ∀(xv )v ∈ AK .
v

By the product formula we have K × ⊂ A1K . After embedding R>0 into any archimedean factor of
× ∼ 1
A×
K we get a splitting AK = AK × R>0 .

Theorem 4.5.4. K × is a discrete subgroup of the group of norm one idèles A1K and the quotient
1
CK := K × \A1K is compact Hausdorff (when equipped with the quotient topology).

Proof. Note that for any (av )v<∞ ∈ Af,× × 1

K there exists a∞ ∈ KR such that (a∞ ; av ) ∈ AK . Therefore
we have a short exact sequence
×
1 → OK \(KR±1 × ÔK
×
) → K × \A1K → ClK → 1. (4.5.4)
×
Since OK is a discrete subgroup of KR±1 , we deduce that K × is discrete in A1K . Then by Proposition
1
B.1.5 and Lemma B.1.2 we see that CK is Hausdorff.
×
The compactness of K × \A1K is equivalent to the compactness of OK \KR±1 together with the
finiteness of ClK .
As a consequence the idèle class group CK = K × \A×
K is locally compact and Hausdorff since
CK ∼ 1
= CK × R>0 after embedding R>0 into an archimedean component of A× K.
0
Definition 4.5.4. Let DK := CK be the connected component of identity of the idèle class group
× ×
CK = K \AK .

102
 Let us analyze the structure of the idèle class group CK = K × \A×
K . Similar to the quotient
×
K \A1K in the proof above, we have a short exact equence
×
1 → OK \(KR× × ÔK
×
) → K × \A×
K → ClK → 1.
× ∼
Just like in the case of K\AK , we utilize the isomorphism ÔK = lim (O /a)× to get
←−a K
× × ∼
OK \(KR× × ÔK ×
) = lim OK \(KR× × (OK /a)× )
←−
a

However a crucial difference with the case of additive group is that in general the natural map
×
OK → (OK /a)× is far from being surjective, therefore the natural injection
×
OK (a)\KR× ,→ OK
×
\(KR× × (OK /a)× )
× ×
is not bijective in general and the latter has more connected components. Here OK (a) := ker(OK →
× ×
(OK /a)× ) and the map is induced by the embedding KR ,→ KR × (OK /a)× sending x to (x, 1).
Therefore it is better to separate the problem into two parts: understand the structure of the identity
component DK = CK 0
and then the structure of the group of connected components π0 (K × \A× K ).
×
First we discuss the identity component DK . Let KR,>0 := Rr>0
1
× (C× )r2 be the connected
component of identity of KR× and let KR,>0 ±1 ×
:= KR,>0 ∩ KR±1 . Let OK,>0
×
:= OK × ×
∩ KR,>0 be the
× ×
subgroup of totally positive units and for any nonzero ideal a ⊂ OK let OK (a)>0 := ker(OK,>0 →
×
(OK /a) ). Then we have
× ×
DK = lim OK (a)>0 \KR,>0 .
←−
a
Similarly the connected component of identity of the norm 1 idèle class group K × \A1K is
× ±1
(K × \A1K )0 = lim OK (a)>0 \KR,>0
← −
a
× ±1
where each factor OK (a)>0 \KR,>0 is a compact manifold that is homeomorphic to a product of r
1
copies of S .
To describe the group of connected components we introduce the following compact open sub-
groups of Af,×
K :
×
Definition 4.5.5. For any nonzero ideal a ⊂ OK let U (a) := ker(ÔK → (OK /a)× ).
Q
Then we have U (a) = v finite Uv (nv ) where nv = v(a) is the exponent of pv in the factorization
of a and Uv (nv ) = ker(Ôv× → (Ov /ϖvnv )× ).
Since CK1
is compact and CK ∼ = CK1
× R>0 , we see that π0 (CK ) is compact and totally discon-
nected and therefore it is a profinite group. When a ranges over all nonzero ideals of OK , the image
×
of KR,>0 × U (a) in π0 (CK ) form a fundamental system of neighborhood of 1. As a consequence we
get that
π0 (K × \A×
K) = ← lim K × \A× ×
K /(KR,>0 × U (a))
−
a
and it sits in a short exact sequence
×
1 → OK \(π0 (KR× ) × ÔK
×
) → π0 (K × \A×
K ) → ClK → 1

where π0 (KR× ) = {±1}r1 and the map from OK ×

to it sends an element α ∈ OK ×
to the signs of α
under the r1 real embeddings of K.
By class field theory, there is a canonical isomorphism π0 (K × \A× ∼ ab ab
K ) = Gal(K /K) where K is
the maximal abelian extension of K. Under this map, the quotient ClK is identified with Gal(H/K)
where H is the Hilbert class field of K.
Example 4.5.2. When K = Q we have Q× \A× ∼ × × ×
Q = R>0 × Ẑ . Thus DQ = R>0 and π0 (Q \AQ ) =
× ab
S
Ẑ . On the other hand, we know from the Kronecker-Weber theorem that Q = n≥1 Q(ζn ) and
therefore Gal(Qab /Q) = limn Gal(Q(ζn )/Q) ∼ = lim (Z/nZ)× = Ẑ× .
←− ←−n

103
Chapter 5

Meromorphic continuation and

functional equations

5.1 Hecke’s approach

5.1.1 The case of Riemann zeta function
Let φ(t) be a function on [0, ∞) that has Rrapid decay at infinity, extended to an even function on
∞
R. When Re(s) > 1, its Mellin transform 0 φ(t)ts dt t converges. For any n > 0 we have
Z ∞ Z ∞
dt dt
φ(nt)ts = n−s φ(t)ts
0 t 0 t

from which we get

Z ∞
∞X Z ∞
dt dt
ζ(s, φ) := φ(nt)ts = ζ(s) φ(t)ts (5.1.1)
0 n=1
t 0 t

Two of Riemann’s proof of the meromorphic continuation of ζ(s) is based on this integral. The
2
difference of two methods is in the choice of the function φ(t) to be either e−πt or e−t . The
integrals in both methods are closely related to the Gamma function, defined when Re(s) > 0 by
the integral Z ∞
dt
Γ(s) = e−t ts
0 t
We first summarize its basic properties. For more details, see for example https://mathworld.
wolfram.com/GammaFunction.html.
Proposition 5.1.1. The function Γ(s) has meromorphic continuation to the whole complex plane.
Denote the extended function also by Γ(s). Then it has the following properties:
1. Γ(n) = (n − 1)! for any positive integer n;

2. Γ(s) is holomorphic and everywhere nonzero on the open set C \ Z≤0 and has simple pole at
the non-positive integers 0, −1, −2, . . . whose residues are
1
lim (s + m)Γ(s) = (−1)m , ∀m ∈ Z≤0 .
s→−m m!
π
3. Γ(s + 1) = sΓ(s), Γ(s)Γ(1 − s) = sin(πs) .

104
 √
4. Γ( 2s )Γ( s+1
2 )=2
1−s
πΓ(s).
For later use, we also introduce the following variants of the Gamma function
s s
ΓR (s) := π − 2 Γ( ), ΓC (s) := (2π)1−s Γ(s), ∀s ∈ C. (5.1.2)
2
Then we have ΓR (1) = ΓC (1) = 1.
Now let us come back to the integral (5.1.1). We break it into two parts:
Z ∞ Z ∞X ∞
dt dt
ζ(s) φ(t)ts = φ(nt)ts = ζ0 (s, φ) + ζ1 (s, φ)
0 t 0 n=1
t

where Z ∞
1X Z ∞
∞X
s dt dt
ζ0 (s, φ) = φ(nt)t , ζ1 (s, φ) = φ(nt)ts .
0 n=1 t 1 n=1
t
Since φ(t) decays rapidly as t → +∞, the second integral ζ1 (s, φ) extends to an entire function of
−t
s.
P∞It remains to extend the first integral. In the method in which we take φ(t) = e , the sum
n=1 φ(nt) can be calculated explicitly in terms of the Bernoulli numbers and as a consequence we
also get formulas for special values of the extended function ζ(s). Here we will explain the other
2
method in which we take φ(t) = e−πt . In this method we do not explicitly calculate this infinite
sum but rather convert it to another infinite sum by Fourier transform and Poisson summation
formula. As a consequence we will get the functional equation for the extended function at the same
time.
Recall that the Fourier transform is the R-linear automorphism of the space of Schwartz1 func-
tions S(R) sending any φ ∈ S(R) to the function F(φ) ∈ S(R) defined by
Z
F(φ)(x) = φ(y)e−2πixy dy.
R

It satisfies the following basic properties:

2
1. F(F(φ))(x) = φ(−x) for all φ ∈ S(R) and x ∈ R. If φ(t) = e−πt , then F(φ) = φ.
2. For any t > 0, let φt (x) := φ(tx). Then F(φt ) = t−1 F(φ)1/t for all φ ∈ S(R).
3. (Poisson Summation Formula) For all φ ∈ S(R), we have
X X
φ(n) = F(φ)(n).
n∈Z n∈Z

Combining (2) and (3), we see that for any even function φ ∈ S(R) and any t > 0 there is an
identity
∞ ∞
X 1X F(φ)(0) φ(0)
φ(tn) = F(φ)(t−1 n) + − .
n=1
t n=1
2t 2
From this we deduce, by a change of variable t → t−1 , the following identity when Re(s) > 1:
Fφ(0) φ(0)
ζ0 (s, φ) = ζ1 (1 − s, F(φ)) − − .
2(1 − s) 2s
The right hand side clearly has meromorphic continuation to C with possible simple poles at 0 and
1. Consequently we get the meromorphic continuation of
Z ∞
dt F(φ)(0) φ(0)
ζ(s) φ(t)ts = ζ1 (s, φ) + ζ1 (1 − s, F(φ)) − − .
0 t 2(1 − s) 2s
1 see [SS03] for the definition

105
 2
We observe that the right hand side has some symmetry under s ↔ 1 − s. Take φ(t) = e−πt , then
F(φ) = φ and we get
s s 2 2 1 1
π − 2 Γ( )ζ(s) = 2ζ1 (s, e−πt ) + 2ζ1 (1 − s, e−πt ) − − .
2 1−s s
√
From the formula Γ( 12 ) = π we deduce that ζ has simple pole at s = 1 with residue 1.
s
Theorem 5.1.2. Let ξ(s) := π − 2 Γ( 2s )ζ(s) be the completed zeta function. Then ξ(s) = ξ(1 − s).

5.1.2 The case of Dirichlet L-functions

Next we use similar method to study Dirichlet L-functions.
Exercise 5.1.1. Let χ be a Dirichlet character with conductor f . For any φ ∈ S(R), use Poisson
summation formula to prove the identity
X τ (χ) X n
χ(n)φ(n) = χ(n)F(φ)( ).
f f
n∈Z n∈Z

For any φ ∈ S(R) and any χ ̸= 1 we have

Z ∞X
dt
ζ(s, φ, χ) := χ(n)φ(tn)ts
0 t
n∈Z
Z ∞X
dt τ (χ) ∞ X
Z
dt
= χ(n)φ(tn)ts + s χ(n)F(φ)(tn)t1−s
1 t f 1 t
n∈Z n∈Z

and similarly
Z ∞ Z ∞
1−s dt dt
X X
−1 −1 s−1
ζ(1 − s, F(φ), χ )= χ(n)F(φ)(tn)t + τ (χ )f χ(n)φ(−tn)ts
1 t 1 t
n∈Z n∈Z

If φ satisfies φ(−t) = χ(−1)φ(t), then by Lemma 4.4.8 we obtain

ζ(s, φ, χ) = τ (χ)f −s ζ(1 − s, F(φ), χ−1 ) (5.1.3)

Theorem 5.1.3. Let χ be a Dirichlet character of conductor f and parity ϵ ∈ {0, 1} so that χ(−1) =
(−1)ϵ . Then L(s, χ) satisfies a functional equation

ΓR (s + ϵ)L(s, χ) = τ (χ)f −s (−i)ϵ ΓR (1 − s + ϵ)L(1 − s, χ−1 ).

Moreover, this equation is equivalent to

   s
π(s − ϵ) τ (χ) 2π
Γ(s) cos L(s, χ) = L(1 − s, χ−1 ).
2 2iϵ f
R∞
Proof. For φ ∈ S(R) that satisfies φ(−t) = χ(−1)φ(t) we have ζ(s, φ, χ) = 2L(s, χ) 0
φ(t)ts dt
t .
2
Take φ(t) = tϵ e−πt . Then we get that

ζ(s, φ, χ) = ΓR (s + ϵ)L(s, χ).

Also one calculates that F(φ) = (−i)ϵ φ and thus

ζ(1 − s, F(φ), χ−1 ) = (−i)ϵ ΓR (1 − s + ϵ) L(1 − s, χ−1 )

106
The functional equation of L(s, χ) then follows from (5.1.3).
From Proposition 5.1.1 we get that
Γ(s) 1

1 = 21−2s π − 2 cos(πs)Γ(2s).
Γ( 2 − s)

From this one deduce the formula

ΓR (s + ϵ) π(s − ϵ)
= 2(2π)−s cos( )Γ(s)
ΓR (1 − s + ϵ) 2
which then implies the second equation.

5.1.3 Dedekind zeta functions

Proposition 5.1.4. Let V be a finite dimensional R-vector space and let V ∗ be its dual vector
space. Let L ⊂ V be a lattice and let L∨ := {ξ ∈ V ∗ | ξ(L) ⊂ Z} be its dual lattice in V ∗ . Let dv be
any Haar measure on the additive group V . For any Schwartz function φ ∈ S(V ) define its Fourier
transform (with respect to dv) to be
Z
F(φ)(ξ) := φ(v)e−2πiξ(v) dv, ∀ξ ∈ V ∗
V
∗
Then F(φ) is a Schwartz function on V and we have
X X
φ(v) = vol(V /L, dv)−1 F(φ)(ξ)
v∈L ξ∈L∨

Proof. The Pontryagin dual of V /L, i.e. the group of continuous P homomorphisms V /L → C1 , is
∨ ∞
isomorphic to the dual lattice L . For each v ∈ V let F (v) = λ∈L φ(x + λ). Then F ∈ C (V /L)
has Fourier coefficients
Z
F̂ (ξ) := vol(V /L, dv)−1 F (v)e−2πiξ(v) dv
V /L
Z
(5.1.4)
= vol(V /L, dv)−1 φ(v)e−2πiξ(v) dv
V
= vol(V /L, dv)−1 F(φ)(ξ), ∀ξ ∈ L∨

Then we have the Fourier expansion

X X
F (v) = F̂ (ξ)e2πiξ(v) = vol(V /L, dv)−1 F(φ)(ξ)e2πiξ(v) , ∀v ∈ V /L.
ξ∈L∨ ξ∈L∨

Take v = 0 we get the desired formula.

In the case V = KR for a number field K we identify V with its dual via the trace pairing on K.
×
Example 5.1.1. We consider the case K = Q(i) and OK = Z[i], OK = {±1, ±i}. Then OK is a
PID and any nonzero ideal in OK has a unique generator of the form m + ni with m ∈ Z≥1 and
n ∈ Z≥0 . Then we have
∞ X
∞
X 1 1 X 1
ζK (s) = = , Re(s) > 1.
m=1 n=0
(m2 2
+n )s 4 (m2 + n2 )s
(m,n)∈Z2 −{(0,0)}

We fix the measure on KR = C to be induced by the two form 2dx∧dy (i.e. twice the usual Lebesgue
measure) and denote it by “dz” (as is common in the literature, we are abusing notation here: it is a

107
measure on C, not a differential 1-form). For φ ∈ S(KR ) = S(C) we normalize its Fourier transform
to be Z Z
F(φ)(w) = φ(z)e−2πiTrC/R (wz) dz = φ(z)e−4πiRe(wz) dz
C C
2
+y 2 )
With this normalization we have F(F(φ))(z) = φ(−z) and for the function φ(x + yi) = e−2π(x
we have F(φ) = φ. Then we have
Z Z
dz 1 dz
φ((m + ni)z)(z z̄)s = 2 + n2 )s
φ(z)(z z̄)s
C × |z|C (m C× |z|C

and the analogue of (5.1.1) in this example becomes

Z Z
s dz dz
X
ζ(s, φ) := φ((m + ni)z)(z z̄) = 4ζK (s) φ(z)(z z̄)s .
C ×
2
|z|C C× |z|C
(m,n)∈Z −{(0,0)}

dz
We recall that |z|C := z z̄ and therefore |z| C
is a Haar measure on the multiplicative group
× iθ
C . In terms of polar coordinate z = re our additive Haar measure is dz = 2rdrdθ and the
dz
multiplicative Haar measure is |z|C
= 2 dr 1
r dθ. The integral against dθ is an averaging along S -orbits
∞
which produces a function ϕ ∈ C (R>0 ) such that
Z 2π
φ(zeiθ )dθ = ϕ(z z̄), ∀z ∈ C
0

We are basically summing over the lattice OK = Z[i] whose dual under the trace pairing is the
∨
inverse different OK = 12 Z[i]. The functional equation for the integral ζ(s, φ) turns out to be

ζ(s, φ) = 21−2s ζ(1 − s, F(φ)).

2 2
Then we take φ(x + yi) = e−2π(x +y ) to get ζ(s, φ) = 2(2π)−s Γ(s)ζK (s). Define the completed zeta
function to be ξK (s) = 2s ζ(s, φ). Then we get the functional equation ξK (s) = ξK (1 − s).
Theorem 5.1.5. Define the completed Dedekind zeta function for K by
s
ξK (s) := |dK | 2 ΓR (s)r1 ΓC (s)r2 ζK (s).

Then it satisfies a functional equation

ξK (s) = ξK (1 − s).

We will deduce this as a special case of a more general result later, following the adèlic approach of
Tate. Here we briefly indicate the classical approach of Hecke. For any Schwartz function φ ∈ S(KR )
and any ξ ∈ KR× we have
Z Z
φ(xξ)(N x)s d× x = (N ξ)−s φ(x)(N x)s d× x, ∀s ∈ C, Re(s) > 0.
KR× KR×
Q
Here N x = v archimedean |xv |v where | · |v is the normalized absolute value (when v is complex it
is the square of the usual absolute value) and d× x = v archimedean |xdxv v|v is a Haar measure on the
Q

multiplicative group KR× = (R× )r1 × (C× )r2 . P

We write the Dedekind zeta function of K as ζK (s) = C∈ClK ζK,C (s) where for each ideal class
C ∈ ClK represented by a−1 for an integral ideal a ⊂ OK ,
X X
ζK,C (s) := (N I)−s = (N a)s (N ξ)−s .
I⊂OK ,I∈C ×
ξ∈(a−{0})/OK

108
Then when Re(s) > 1 we have
Z X Z
ζK,C (s) φ(x)(N x)s d× x = (N a)s φ(xξ)(N x)s d× x
KR× × KR×
ξ∈(a−0)/OK
Z Z ∞ (5.1.5)
X dt
= (N a)s φ(tyξ)ts dy
×
OK \KR±1 0 t
ξ∈(a−0)

where we choose an isomorphism KR× ∼ = R>0 ×KR±1 to decompose the Haar measure into the product
× dt
measure d x = t · dy. Then we apply Poisson summation formula to the inner sum over the lattice
a ⊂ KR and choose a Gaussian function φ that (almost) equals to its own Fourier transform to get
the analytic continuation and functional equation at the same time.
Corollary 5.1.6. Let K/Q be a finite abelian extension and let XK := Hom(Gal(K/Q), C× ) be the
group of characters of Gal(K/Q). Then we have
1. Conductor-Discriminant Formula:
Y
dK = (−1)r2 fχ
χ∈XK

2. As K/Q is Galois we have either r1 = 0 or r2 = 0, in each case we have

(p
Y |dK |, if K is totally real, i.e. r2 = 0
τ (χ) = [K:Q]/2 p
χ∈X
i |dK | if K is totally imaginary, i.e. r1 = 0
K

Proof. If K is totally real, i.e. r2 = 0, then all χ ∈ XK are even and we have
Y Y τ (χ) Y
ΓR (s)r1 L(s, χ) = s
ΓR (1 − s)r1 L(1 − s, χ−1 )
fχ
χ∈XK χ∈XK χ∈XK

On the other hand, by Theorem 5.1.5 we have

1
ΓR (s)r1 ζK (s) = |dK | 2 −s ΓR (1 − s)r1 ζK (1 − s)
Then we get that
Y τ (χ) 1
= |dK | 2 −s .
fχs
χ∈XK

Then the result follows from Lemma 4.4.8.

If K is totally complex, i.e. r1 = 0, let K+ := K ∩ R. Then K/K+ is a quadratic extension
and the even characters in XK are precisely those coming from the natural embedding XK+ ⊂ XK .
Thus we have
Y Y τ (χ) Y
(ΓR (s)ΓR (s + 1))r2 L(s, χ) = s
(−i)r2 (ΓR (1 − s)ΓR (2 − s))r2 L(1 − s, χ−1 )
fχ
χ∈XK χ∈XK χ∈XK

From Proposition 5.1.1 we deduce that

ΓR (s)ΓR (s + 1) = π −1 ΓC (s).
On the other hand, by Theorem 5.1.5 we have
1
ΓC (s)r2 ζK (s) = |dK | 2 −s ΓC (1 − s)r2 ζK (1 − s).
Then we deduce that
Y τ (χ) 1
= ir2 |dK | 2 −s .
fχs
χ∈XK

Then the result in this case also follows from Lemma 4.4.8.

109
 Let us examine the case where K is a quadratic extension of Q. Let χK : Gal(K/Q) ∼ = {±1} be
the canonical isomorphism that we view as a Dirichlet character after embedding K into a cyclotomic
field. Then XK = {1, χK√ } and it is clear that f1 = 1 and τ (1) = 1. The Corollary implies that
fχK = |dK | and τ (χK ) = dK . Moreover, χK (−1) equals to the sign of dK . In other words, χK is
even (resp. odd) if K is real (resp. imaginary).
Then if dK > 0 (i.e. K is real quadratic field), the functional equation of L(s, χK ) simplifies to
1
−s
ΓR (s)L(s, χK ) = dK
2
ΓR (1 − s)L(1 − s, χK )
while if dK < 0 (i.e. K is imaginary quadratic field), the functional equation of L(s, χK ) simplifies
to
1
ΓR (s + 1)L(s, χK ) = |dK | 2 −s ΓR (2 − s)L(1 − s, χK )

5.1.4 Hecke characters

Hecke discovered that his method can be applied to study a more general class of L-functions that
include the Dedekind zeta functions and the Dirichlet L-functions as special cases. These L-functions
are associated to what Hecke called “Größencharaktere”, which are nowadays understood in terms
of characters of the idèle class group.
Definition 5.1.1. A (unitary) Hecke character of the number field K is a continuous homomor-
phism of topological groups ω : K × \A× 1 1 ×
K → C where C := {z ∈ C , |z| = 1}.

Clearly a Hecke character ω induces a continuous group homomorphism A× ×

K → C . For each
place v of K let iv : Kv× ,→ A×
K be the embedding such that i v (x) is the idèle whose v-component
is x and all other components are 1. We also use iv to denote its composition with A× × ×
K → K \AK .
Let ωv := ω ◦ iv : Kv× → C× be the local factor of ω at v. We first analyze the structure of these
local factors.
• If v is a real place, then ωv : R× → C1 is given by ωv (x) = sgn(x)nv |x|itv where nv ∈ {0, 1}
and tv ∈ R;
• If v is a complex place, then ωv : C× → C1 is given by ωv (z) = (z/|z|)nv |z|itv where nv ∈ Z
and tv ∈ R.
• If v is a finite place, then ker(ωv ) contains2 an open subgroup of Ôv× . Let nv ∈ Z≥0 be the
smallest integer such that ωv is trivial on Uv (nv ) := ker(Ôv× → (Ôv /ϖvnv )× ). The integer
nv (or rather the ideal ϖvnv Ov ) is called the conductor of ωv . If nv = 0 we say that ωv is
unramified and that ω is unramified at v. In this case ωv is uniquely determined by the value
ωv (ϖv ) ∈ C1 .
Let ωf (resp. ω∞ ) be the nonarchimedean (resp. archimedean) component of ω, defined as the
restriciton of ω to Af,×
K (resp. KR× ). Then ker ωf contains an open subgroup of ÔK×
of the form
× ×
U (m) := ker(ÔK → (OK /m) ) (since these subgroups form a fundamental system of neighborhoods
×
of 1 in ÔK ). We indicate this by saying that ω is a Hecke character mod m. The largest
Q nonzero
ideal m ⊂ OK such that U (m) ⊂ ker ωf is called the conductor of ω. Then we have m = v finite pnv v
where nv is the local conductor of ω at v. In particular we see that ωv is unramified for all but
finitely many v.
For any x = (xv ) ∈ A× K we have ωv (xv ) = 1 for all place v ∈/ S where S is a finite set of
places of K that includes all the archimedean places. We write x = xS xS where the components of
xS , xS ∈ A×K are ( (
xv if v ∈ S S 1 if v ∈ S
(xS )v = , (x )v =
1 if v ∈
/S xv if v ∈ /S
2 This follows from the fact that the open subgroups form a fundamental system of neighborhoods of 1 in Ô × and
v
that there is no nontrivial subgroup of C× contained in a sufficiently small open neighborhood of 1 in C× .

110
Then we have ω(xS ) = 1 and therefore
Y Y
ω(x) = ω(xS ) = ωv (xv ) = ωv (xv )
v∈S v

where the last product runs over all places v of K and for any v ∈ / S we have ωv (xv ) = 1. We
interpret this formula as a factorization of ω into local components ωv and write:

ω = ⊗v ωv

where the tensor product runs over all places of K. Similarly we have ω = ω∞ ⊗ ωf where

ω∞ = ⊗v archimedean ωv , ωf = ⊗v finite ωv .

Next we relate the modern definition of a “Hecke character” as above with Hecke’s original notion
of “Größencharaktere”.
For any nonzero ideal m ⊂ OK , let Im be the group of fractional ideals of OK whose factorization
does not involve any prime factor of m. Let ω : K × \A× ×
K → C be a Hecke character mod m. Since
Im is the free abelian group generated by pv for all finite place v ∤ m and ωv is unramified for any
such place, we can define a character χ : Im → C× by requiring

χ(pv ) = ωv (ϖv ), ∀v ∤ m.

This seems to depend only on the non-archimedean component ωf . However, the fact that ω is trivial
on the discrete subgroup K × impose further restrictions on χ. To explain it we first introduce more
notations.
For any elements x, y ∈ K × , we define the multiplicative congruence relation x ≡ y mod × m to
mean that xy −1 ∈ Uv (nv ) := ker(Ôv× → (Ôv /ϖvnv )× ) for all place v | m, where nv is the exponent
of pv in the unique factorization of m. We define the following multiplicative groups associated to
m:
× ×
• Let Km,1 := {α ∈ K × | α ≡ 1 mod × m} and let OK (m) := Km,1 ∩ OK = K × ∩ U (m) where
f,×
the intersection is inside AK (see Definition 4.5.5 for U (m)).
×
• Let Pm,1 := Km,1 /OK (m) be the group of principal fractional ideals generated by an element
in Km,1 .

• Let Af,× f,× × × f,×

K (m) := {x ∈ AK | xv ∈ Uv (nv ), ∀v | m} and let AK (m) := KR × AK (m). Then we
have U (m) ⊂ AK (m) and there is a canonical isomorphism AK (m)/U (m) ∼
f,× f,×
= Im .
By the approximation theorem K × is dense in v|m Kv× (more precisely we have K × /Km,1 ∼
Q
=
Q × × × × ∼
v|m Kv /Uv (nv )). Then the inclusion AK (m) ,→ AK induces an isomorphism Km,1 \AK (m) =
×
K × \AK . We summarize these groups in the following commutative diagram in which the rows and
columns are all short exact sequences:

1 / O× (m) / Km,1 / Pm,1 /1

K

  
1 / K × × U (m) / A× (m) / Im /1
R K

  
1 / O× (m)\(K × × U (m)) / K × \A× / Im /Pm,1 /1
K R K

In particular, from the compactness of K × \A1K we see that Im /Pm,1 is a finite abelian group.

111
 Then for any element α ∈ Km,1 , we get
Y
χ(αOK ) = ωv (α) = ωf (α) = ω∞ (α)−1 , ∀α ∈ Km,1
v finite,v∤m

where
• the first equality follows from the definition of χ and the fact that the factorization of α ∈ Km,1
does not involve primes dividing m;
• the second equality follows from the fact that ωv (α) = 1 for any place v|m since ωv is trivial
on Uv (nv ) by assumption.
• the third equality follows since ωf (α)ω∞ (α) = ω(α) = 1 by the definition of a Hecke character.
Conversely, suppose χ : Im → C× is a character such that there exists a continuous homomorphism
ω∞ : KR× → C× satisfying ω∞ (α) = χ(αOK )−1 for all α ∈ Km,1 . The characters ω∞ and χ combine
into a character of A×
K (m) defined by

ω∞ ⊗ χ : x = (x∞ , xf ) 7→ ω∞ (x∞ )χ(x̄f )

where x̄f ∈ Im is the image of xf ∈ Af,×

K (m). By assumption ω ⊗ χ is trivial on the subgroup Km,1
which is embedded diagonally in A× ×
K (m). Therefore ω∞ ⊗ χ defines a character of Km,1 \AK (m)
× ∼ × ×
and via the isomorphism Km,1 \AK (m) = K \AK it further defines a Hecke character ω. Then by
construction the archimedean component of ω is the given character ω∞ and for any finite place
v ∤ m, we have ωv (ϖv ) = χ(pv ).
Example 5.1.2. Let K = Q and m = nZ for some n ∈ Z≥1 . Then Im = { ab ∈ Q× | (a, n) =
(b, n) = 1, a, b ∈ Z}/{±1} and Qm,1 = { ab ∈ Q× | a, b ∈ Z, a ≡ b mod n}. Therefore Im /Pm,1 ∼
=
(Z/nZ)× /{±1} and any even Dirichlet character mod n determines a Hecke character mod m = nZ.
To get the Hecke character for all Dirichlet characters, we need to consider more general discrete
quotients of the idèle class groups by open subgroups whose archimedean P factor is R>0 . These are
given by an effective divisor (or modulus), defined as a formal sum m = v mv [v] over all places
of K, where mv ∈ Z≥0 for all v and mv = 0 for all complex places and all but finitely many
non-archimedean places, while mv ∈ {0, 1} for all real places.
Definition 5.1.2. Let ω : K × \A×
K → C
×
be a Hecke character of conductor m, with associated
×
character χ : Im → C described above. The Hecke L-series of ω is defined when Re(s) > 1 by
X χ(a)
L(s, ω) :=
(N a)s
a⊂OK
a+m=OK

where the sum runs over all nonzero ideals a ⊂ OK relatively prime to m.
Similar to Dedekind zeta function, we have a factorization
Y
L(s, ω) = (1 − ωv (ϖv )N p−s )−1 , Re(s) > 1.
v∤m

5.2 Tate’s approach

5.2.1 Schwartz functions
For each place v of K, define the space of local Schwartz functions S(Fv ) to be the space of usual
Schwartz functions if v is archimedean and if v is non-archimedean we let S(Fv ) = Cc∞ (Fv ) be

112
the space of locally constant functions with compact support. Using these we define the space
of Schwartz functions on AK and AfK as follows. For each finite place v of K, choose a function
fv ∈ S(Kv ) such that fv = 1Ôv (the characteristic function of Ôv ) for all but finitely many places
v. Define their tensor product ⊗v fv to be the function on AfK sending x = (xv ) to v fv (xv )
Q

(this product makes sense since for all but finitely many places we have xv ∈ Ôv and fv = 1Ôv so
that fv (xv ) = 1). Define the Schwartz space S(Af ) to be the C-linear span of all such factorizable
functions. Then we define S(AK ) := S(KR ) ⊗ S(Af ). In other words, a function in S(AK ) is a finite
C-linear combination of factorizable functions f = ⊗v fv where the product is over all places of K
and fv = 1Ôv for all but finitely many non-archimedan places.

5.2.2 Additive Haar measures and Fourier transform

For any place v of K, we let ιv be the natural embedding from Kv to AK (or AfK , depending on the
context) sending an element x ∈ Kv to the adèle (or finite adèle) whose v-component is x and all
other components are 0. For any character (i.e. continuous group homomorphism) ψ : AK → C1 ,
let ψv := ψ ◦ ιv : Kv → C1 be its local factors.
Lemma 5.2.1. Let A be the additive group of one of the following topological rings: a local field,
AfK , AK where K is a global field3 . Let ψ : A → C1 be a nontrivial character and if A is AK or AfK
we assume moreover that all the local factors ψv are nontrivial.
1. For each a ∈ A, let ψa (x) := ψ(ax) for any x ∈ A. Then the map A → A
b : a 7→ ψa is an
isomorphism of topological groups.
2. There exists a unique Haar measure dµ on A that is self-dual with respect to ψ in the following
sense: for any Schwartz function φ ∈ S(A), define its Fourier transform (with respect to dµ
and ψ) to be Z
F(φ)(x) := φ(y)ψ(xy)dµ(y), ∀x ∈ A.
A

Then we have the Fourier inversion formula: F(F(φ))(x) = φ(−x) for any x ∈ A.
Proof. The second part will be a consequence of the first part and the Fourier inversion formula in
the context of locally compact abelian groups (see [RV99, §3.3]). We only prove the first part.
The map is well-defined since multiplication by any a ∈ A defines a continuous endomorphism
on A and hence the map ψa : A → C× is a continuous homomorphism. We denote the map a 7→ ψa
by
ιψ : A → A.
b

Clearly ιψ is a group homomorphism. It is easily seen to be injective (when A is local field, injectivity
follows from the fact that ψ is nontrivial; when A is adèle ring, injectivity follows from the case
of local fields since all local factors of ψ are assumed to be nontrivial). By the same reasoning, if
an element a ∈ A satisfies ιψ (b)(a) = 0 for all b ∈ A, then a = 0. This shows that ιψ (A) is dense
in Ab by the Pontryagin duality theorem (that A ∼ = A).
bb
Since A is locally compact, we must have
ιψ (A) = A b by Proposition B.1.5.
Next we show that ιψ is continuous. Let C ⊂ A be a compact subset and U ⊂ C1 an open
subset. Let W (C, U ) be the open neighborhood of 1 ∈ A b consisting of characters ψ ′ such that
′ −1
ψ (K) ⊂ Ω. We need to show that ιψ (W (C, U )) is open. When A is a local field, this is clear since
a sufficiently small open ball would multiply the bounded set C into ψ −1 Q (U ), and hence contained in
ι−1
ψ (W (C, U )). When A is the ring of adèles, we may assume that C = v Cv is a product of compact
subsets Cv ⊂ Kv and there is a finite set S of places of K containing all archimedean places such
that for any v ∈ / S we have ker(ψv ) = Ôv = Cv . For any v ∈ S we choose an open neighborhood
3 We are mainly interested in the case of number field. When K is a function field we have AK = AfK

113
1 ∈ Uv ⊂ C1 such that v Uv ⊂ U . By the case of local field, there exists open neighborhood
Q

0 ∈ Ωv ⊂ Kv such that Ωv ⊂ ι−1

Q Q
ψv (W (Cv , Uv )). Then the open subset Ω := v∈S Ωv × / Ôv ⊂ A
v ∈S
−1
is contained in ιψ (W (C, U )).
Finally we show that the map ιψ is open. Since ιψ is bijective, it suffices to show that for
any open neighborhood 0 ∈ Ω ⊂ A, there exists a compact subset C ⊂ A and open neighborhood
1 ∈ U ⊂ C1 such that Ω contains W (C, U ) := {a ∈ A | ψ(aC) ⊂ U }. First we consider the case
where A is a local field. Let x0 ∈ A be an element such that ψ(x0 ) ̸= 1. We take U small enough
/ U . Choose 0 ̸= y0 ∈ A such that the open ball {x ∈ A, |x| < |x0 y0−1 |} is contained
so that ψ(x0 ) ∈
in Ω. Let C = {x ∈ A, |x| ≤ |y0 |}. Then for any a ∈ W (C, U ) we have ψ(x0 ) ∈ / ψ(aC) and hence
x0 ∈/ aC. This implies that |ay0 | < |x0 | and hence a ∈ Ω. Next we consider the case A = AfK or
A = AK . We may assume that there is a finite set S of places of K including all archimedean places
such that:
• Ω = v Ωv where Ωv ⊂ Kv is an open neighborhood of 0 for each place v and Ωv = Ôv for all
Q
v∈/ S;
• For any v ∈/ S, the kernel of the local factor ψv = ψ ◦ ιv : Kv → C1 equals to Ôv (this is
possible by continuity of ψ).
From the case of local field we just proved, for each v ∈ S we can find a compact subset Cv ⊂ Kv
and open
Q neighborhood
Q 1 ∈ Uv ⊂ C1 such that W (Cv , Uv ) ⊂ Ωv . Let U := ∩v∈S Uv and let
C := v∈S Cv × v∈S / Ôv . Then we have W (C, U ) ⊂ Ω.

We fix nontrivial additive characters on the completions of Q as follows:

• Let e∞ : R → C1 be the character defined by e∞ (x) = e−2πix for all x ∈ R. Then ker(e∞ ) = Z.
• For a prime number, define the character ep : Qp → C1 to be the composition of the following
maps
1
ep : Qp → Qp /Zp ∼ = Z[ ]/Z → C1
p
P i
where the isomorphism sends x = ci p ∈ Qp to the “negative part” of the Taylor expansion
i
for all i; and the last map sends α ∈ Z[ p1 ] to e2πiα .
P
i<0 c i p , where c i ∈ {0, 1, . . . , p − 1}
Then ker(ep ) = Zp .
Let e := ⊗p≤∞ ep : AQ → C1 be the nontrivial character defined by e(x) = p≤∞ ep (xp ). Similarly
Q

we have its finite part ef := ⊗p<∞ ep : AfQ → C1 . These are well defined since for all but finitely
many p we have xp ∈ Zp and hence ep (x) = 1. P Moreover we have Q ⊂ ker(e). Indeed, any rational
a
number α ∈ Q can be written as α = ⌊α⌋ + p pnpp where ⌊α⌋ ∈ Z is the integral part of α, while
for any prime number p we have np ∈ Z>0 , ap ∈ Z and ap = 0 for all but finitely many p. Then
for any prime number p, the image of α under the composition Q → Qp /Zp ∼ = Z[ p1 ]/Z equals to the
ap ap
class of pnp and hence ep (α) = exp(2πi pnp ). Then we get
Y Y ap
e(α) = ep (α) = exp(−2πiα) exp(2πi ) = exp(−2πi⌊α⌋) = 1.
p<∞
p np
p≤∞

Therefore e induces a character Q\A → C1 . In fact under the isomorphism Q\A ∼ = lim C1 from
←−n
Example 4.5.1, this character is just the projection to the factor of n = 1.
For a general number field K and any place v of K, we fix the nontrivial additive character ψv :=
ep ◦ TrKv /Qp : Kv → C1 where p ≤ ∞ is the place of Q below v. Let ψf := ⊗v finite ψv : AfK → C1
and ψ := ⊗v ψv : AK → C1 . Then for any α ∈ K we have
Y Y Y
ψ(α) = ep (TrKv /Qp (α)) = ep (TrK/Q α) = e(TrK/Q α) = 1.
p≤∞ v|p p≤∞

114
Therefore ψ factors through a character K\AK → C1 .
After fixing the nontrivial local (resp. global) characters ψv (resp. ψ), let us describe the
associated local and global self-dual Haar measures on the additive groups.
• If v is a real place, we let dxv be the usual Lebesgue measure on Kv = R;
• If v is a complex place, we let dxv be 2 times the usual Lebesgue measue on Kv = C;
• If v is a finite place of K above a rational prime p, we let dxv be the unique Haar measure
1
on Kv such that vol(Ôv , dxv ) = (N Dv )− 2 where Dv = (Ôv∨ )−1 is the different ideal of the
extension Ôv /Zp and N Dv := |Ôv /Dv | is its absolute norm.
Lemma 5.2.2. For any place v of K lying above a place p of Q, the Haar measure dxv described
above is self-dual with respect to the character ψv := ep ◦ TrKv /Qp .
Proof. When v is archimedean, this follows from the usual Fourier inversion formula. Now suppose
v is a finite place lying above a rational prime p. Let a ⊂ Kv be a fractional ideal of Ôv . For any
Haar measure dµ on Kv and any y ∈ Kv , the associated Fourier transform of the characteristic
function 1a ∈ S(Kv ) is given by
(
/ a−1 Dv−1
if y ∈
Z Z
0
F(1a )(y) := φ(x)ψv (xy)dµ(x) = ψ(xy)dµ(x) =
Kv a µ(a) if y ∈ a−1 Dv−1

The second equality is clear since for any y ∈ a−1 Dv−1 we have TrKv /Qp (ya) ⊂ Zp . The first equality
/ a−1 Dv−1 then there exists x ∈ a such that TrKv /Qp (xy) ∈
is seen as follows. If y ∈ / Zp and hence
ψ(xy) ̸= 1. On ther other hand we have a−1 D−1 ⊂ (y) and hence y −1 D−1 ⊂ a. For any x ∈ y −1 D−1
we have TrKv /Qp (xy) ∈ Zp so that ψ(xy) = 1. Thus ψ induces a nontrivial character on the quotient
group a/y −1 D−1 and we get
Z X
ψ(xy)dµ(x) = µ(y −1 D−1 ) ψ(x) = 0.
a x∈a/y −1 D −1

In summary we have F(1a ) = µ(a)1a−1 Dv−1 . Then we get that

F(F(1a )) = µ(a)F(1a−1 Dv−1 ) = µ(a)µ(a−1 Dv−1 )1a = (N Dv )µ(Ôv )1a

1
Therefore the self-dual Haar measure is the one for which µ(Ôv ) = (N Dv )− 2 .

On the adèle rings AK and AfK we take the additive Haar measure to be the product of the
local self-dual measures dxv fixed above. Then for any factorizable function φ = ⊗v φv ∈ S(AK )
its Fourier transform is also factorizable. In fact for any x ∈ AK there is a finite set of places S
including all archimedean places such that for any place v ∈
/ S the following conditions are satisfied:
1. x ∈ Ôv ;
2. φv = 1Ôv ;

3. The different is trivial: Dv = Ôv . In other words, the extension Kv /Qp is unramified (where
p is the prime below v). As a consequence we have vol(Ôv , dxv ) = 1 and F(1Ôv ) = 1Ôv .
Then we have
Z YZ Y
F(φ)(x) = φ(y)ψ(xy)dy = φ(y)ψv (xy)dy = F(φv )(xv )
AK v Kv v

where the factors in the product equals to 1 for any v ∈

/ S. In other words, we have
F(⊗v φv ) = ⊗v F(φv ).

115
Lemma 5.2.3. The isomorphism AK ∼
=Ab K determined by ψ induces isomorphisms of topological
groups K\AK ∼
=Kb and K ∼ \
= K\AK .
Proof. Since ψ is trivial on K we have inclusions

K ⊂ K\A bK ∼
\K ⊂ A = AK .

Moreover, K\A\K is a discrete subgroup of A \K /K

b K since K\AK is compact. Then the quotient K\A
\K
is discrete and compact (since it is closed in the compact group K\AK ) and hence finite. Since K\A
is a K-vector space and K is not a finite field, we must have K = K\A \K . Taking duals we get
K\AK = K.∼ b

Lemma 5.2.4. Under the product Haar measure dx on AK fixed above, we have vol(K\AK , dx) = 1.
Proof. From the exact sequence

0 → OK \(KR × O
bK ) → K\AK → OK \KR → 0

we see that vol(K\AK , dx) = vol(O bK )·vol(OK \KR ). From the definition of dxv at non-archimedean
bK ) = |dK | 21 . On the other hand, from the definition of dxv for archimedean
places we see that vol(O −

places
p v we see that vol(OK \KR ) is 2r2 times the Lebesgue measure of OK \KR and hence equal to
|dK | by Lemma 4.1.5.
Then following is the analogue of Proposition 5.1.4 in the setting of adèles.
Proposition 5.2.5 (Adèlic Poisson summation formula). For any φ ∈ S(AK ) we have F(φ) ∈
S(AK ) and X X
φ(α) = F(φ)(α)
α∈K α∈K

Proof. From φ we obtain a continuous function on the compact group K\AK defined by
X
F (x) = φ(α + x)
α∈K

From Lemma 5.2.3 the Pontryagin dual of K\AK is the discrete group K. Then by Lemma 5.2.4
the Fourier coefficients of F are
Z Z X
F̂ (ξ) := F (x)ψ(ξx)dx = φ(α + x)ψ(ξ(x + α))dx = F(φ)(ξ).
K\AK K\AK α∈K

By the Fourier inversion formula on K\AK we have

X X
F (x) = F̂ (ξ)ψ(ξx) = F(φ)(ξ)ψ(ξx), ∀x ∈ AK .
ξ∈K ξ∈K

Let x = 0 we get the result.

5.2.3 Multiplicative Haar measures and zeta integrals

We fix the Haar measures d× xv on the local multiplicative groups Kv× as follows
• If v is archimedean, we let d× xv := dxv
|xv |v .

• If v is nonarchimedean, we let d× xv := N pv dxv

N pv −1 |xv |v . Then we have
1
vol(Ôv× , d× xv ) = vol(Ôv , dxv ) = (N Dv )− 2 .

116
On the group of idèles A× d× x = v d× xv . Similarly on the group of
Q
K we take the product measure
finite idèles we also take the product measure d× x = v finite d× xv . In particular we have
Q

1 1
Y
×
vol(ÔK , d× x) = (N Dv )− 2 = |dK |− 2 .
v finite

As in Hecke’s classical approach, we will apply Poisson summation formula to the scalings of a
given Schwartz function.
Lemma 5.2.6. For any place v of K and any Haar measure µ on Kv we have dµ(ax) = |a|v dµv (x)
for all a ∈ Kv× . In the global case for any a ∈ A×
K we have d(ax) = ∥a∥x.

Corollary 5.2.7. For any φ ∈ S(AK ) and a ∈ A× K , let φa ∈ S(AK ) be the function defined by
φa (x) = φ(ax). Then we have F(φa )(x) = ∥a∥−1 F(φ)(a−1 x) for all x ∈ AK .
We may view this as the number field analogue of Serre duality for algebraic curves. Then the
following result provides a number field analogue of the Riemann-Roch theorem for algebraic curves.
Theorem 5.2.8 (Riemann-Roch). For any φ ∈ S(AK ) and any x ∈ A×
K we have
X X
φ(αx) = ∥x∥−1 F(φ)(x−1 α).
α∈K α∈K

For any character ω : A× ×

K → C and any φ ∈ S(AK ), define the zeta integral to be
Z
ζ(s, ω, φ) = φ(x)ω(x)∥x∥s d× x, Re(s) > 1. (5.2.1)
A×
K

The convergence of the zeta integral when Re(s) > 1 will follow from explicit computations in the
next section. Assuming this we first prove its meromorphic continuation and functional equation.
Theorem 5.2.9. For any Hecke character ω : K × \A× K → C
×
and any φ ∈ S(AK ), the zeta
integral ζ(s, ω, φ) has meromorphic continuation to the whole complex plane and satisfies a functional
equation
ζ(s, ω, φ) = ζ(1 − s, ω −1 , F(φ)).
If ω is nontrivial on K × \A1K , then ζ(s, ω, φ) is holomorphic on C. If ω is trivial on K × \A1K so that
ω(x) = ∥x∥itω where tω ∈ R, then the only possible poles of ζ(s, ω, φ) are simple poles at s = −itω
and s = 1 − itω with respective residues
−vol(K × \A1K , d1 x)φ(0), vol(K × \A1K , d1 x)F(φ)(0).
Proof. We break the global zeta integral (5.2.1) into two parts ζ(s, ω, φ) = ζ0 (s, ω, φ) + ζ1 (s, ω, φ)
where ζ0 is the integral on A≤1 × ≥1
K := {x ∈ AK , ∥x∥ ≤ 1} and ζ1 is the integral on AK := {x ∈
×
AK , ∥x∥ ≥ 1}. Then ζ1 (s, ω, φ) extends to an entire function of s.
When Re(s) > 1, using the assumption that ω is trivial on K × and Theorem 5.2.8 we get that:
Z X
ζ0 (s, ω, φ) = φ(αx)ω(x)∥x∥s d× x
≤1
K × \AK α∈K ×
Z !
X
−1 −1 −1
= ∥x∥ F(φ)(αx ) + ∥x∥ F(φ)(0) − φ(0) ω(x)∥x∥s d× x
≤1
K × \AK α∈K ×
 
F(φ)(0)
Z X
−1 1−s × × φ(0)
= Fφ(αx)ω(x) ∥x∥ d x − δω vol(K \A1K ) +
≥1
K × \AK α∈K × 1 − s − itω s + itω
 
−1 × F(φ)(0) φ(0)
= ζ1 (1 − s, ω , F(φ)) − δω vol(K \A1K ) +
1 − s − itω s + itω
(5.2.2)

117
where δω = 0 if ω is nontrivial on A1K , while if ω is trivial on A1K then δω = 1 and ω(x) = ∥x∥itω for
tω ∈ R. Note that the right hand side clearly has meromorphic continuation to the whole complex
plane with possible simple poles at s = −itω and s = 1 − itω . Therefore we get the meromorphic
continuation and functional equation:

ζ(s, ω, φ) = ζ(1 − s, ω −1 , F(φ)).

The volume term in (5.2.2) can be explicitly calculated. First we describe the measure on
K × \A1K . From the measure d× x on A×
K fixed above and the short exact sequence (4.5.3) we get a
×
Haar measure d1 x on A1K such that the quotient measure dd1 xx on R>0 is the measure dx
x . We also
×
denote d x the Haar measure on K \AK defined as the quotient of d1 x by the counting measure4
1 1

on the discrete subgroup K × .

Proposition 5.2.10. The volume of K × \A1K with respect to the measure d1 x described above equals
to
2r1 (2π)r2 RK hK
vol(K × \A1K ) = p
wK |dK |
Proof. From the short exact sequence (4.5.4) we get that
×
× hK vol(OK \KR±1 )
vol(K × \A1K ) = hK vol(OK \(KR±1 × ÔK
× ×
)) = hK vol(OK \KR±1 )vol(ÔK
×
, d× x) = p .
|dK |
×
To compute vol(OK \KR±1 , d1 x) we use the homomorphism Log defined in (4.2.1) to get a short exact
sequence
Log
1 → {±1}r1 × (C1 )r2 → KR× −−→ Rr1 +r2 → 1
Then the measure d× x = v archimedean d× xv decompose into the product of the Lebesgue measure
Q
on Rr1 +r2 and the measure on {±1}r1 × (C1 )r2 with total volume 2r1 (2π)r2 .
×
Recall that Λ := Log(OK ) is free Z-module of rank r = r1 + r2 − 1 and we have a commutative
diagram in which the rows are short exact sequences:

1 / {±1}r1 × (C1 )r2 / K ±1 Log

/H /0
R

  
1 / µ(K)\({±1}r1 × (C1 )r2 ) / O× \K ±1 / Λ\H /0
K R

Pr+1
where H := {(xi ) ∈ Rr+1 | i=1 xi = 0}. To determine the measure of Λ\HP(i.e. the measure of a
r+1 ∼
fundamental
P domain of Λ in H), we fix a splitting R = H × R of the map : Rr → R defined by
(xi ) 7→ i xi (this map is obtained from the absolute norm map by applying Log). This amounts
to choose a vector v1 ∈ H1 := {(xi ) ∈ Rr+1 |
P
xi = 1}. Then the induced measure on H is the
one whose product with the Lebesgue measure on R equals to the Lebesgue measure on Rr . Under
this measure, the volume of Λ\H equals to the Lebesgue measure of the parallelotope spanned by
v1 and the fundamental domain E for the action of Λ on H, which equals to the regulator RK by
1
Lemma 4.3.2 (note that this differs from the Lebesgue measure on H by a factor √r+1 , which is the
distance between the two hyperplanes H and H1 ). Then we get that

× 2r1 (2π)r2
vol(OK \KR±1 ) = · RK
wK
and the formula for vol(K × \A1K ) follows immediately.
4 i.e. any point has measure 1

118
5.2.4 Local computations
When φ = ⊗v φv ∈ S(AK ) is factorizable, we will see that the global zeta integral factorize into
products of local zeta integrals
Y
ζ(s, ω, φ) = ζ(s, ωv , φv ), Re(s) > 1. (5.2.3)
v

where ωv := ω ◦ iv : Kv× → C× is the local factor of ω at v and

Z
ζ(s, ωv , φv ) := φv (xv )ωv (xv )|xv |sv d× xv .
Kv×

Note that most factors in (5.2.3) are not equal to 1 and the formula should be understood as a
convergent limit, just like the Euler product of L-functions. More precisely, we have
Y Y Z
ζ(s, ωv , φv ) := lim ζ(s, ωv , φv ) = lim φ(x)ω(x)∥x∥s d× x, Re(s) > 1.
|S|→∞ |S|→∞ A×
v v∈S K,S

as long as we prove the convergence of the Q first limit. Here S runs over all sufficiently large increasing
finite sets of places of K and A× × ×
Q
K,S := v∈S Kv × / Ôv runs over a sequence of open subsets
v ∈S
× ×
of AK whose union is AK . See [Tat67, §3.3] for more details. The convergence of the first limit will
follow from the convergence of the Euler product for the L-function L(s, ω).
In fact we will choose a special test function φ◦ = ⊗v φ◦v ∈ S(AK ) and compute the local
zeta integrals ζ(s, ωv , φ◦v ), which will always converge when Re(s) > 0. From the result of this
computation we will define in each case a local L-factor L(s, ωv ) and local epsilon factor εv (s, ωv , ψv )
and verify that the following local functional equations are satisfied when 0 < Re(s) < 1:
ζ(1 − s, ωv−1 , F(φ◦v )) ζ(s, ωv , φ◦v )
−1 = εv (s, ωv , ψv ) (5.2.4)
L(1 − s, ωv ) L(s, ωv )
As a consequence we will get the convergence of the infinite product (5.2.3), together with the
meromorphic continuation and functional equations of the Hecke L-functions L(s, ω).
1. Suppose v is non-archimedean and ωv is unramified. Let φ◦v := 1Ôv . Then we have F(φ◦v ) =
1
(N Dv )− 2 1Dv−1 and when Re(s) > 0,
Z
ζ(s, ωv , φ◦v ) = φ◦v (x)ωv (x)|x|sv d× xv
Kv×
∞
=
X
ωv (ϖv )n (N pv )−ns vol(Ôv× ) (5.2.5)
n=0
1
= (N Dv )− 2 (1 − ωv (ϖv )(N pv )−s )−1
Similarly, let dv := v(Dv ), then when Re(s) < 1 we have
∞
1
X
ζ(1 − s, ωv−1 , F(φ◦v )) = (N Dv )− 2 ωv (ϖv )−n (N pv )n(s−1) vol(Ôv× , d× xv )
n=−dv (5.2.6)
= ωv (ϖvdv )(N Dv )−s (1 − ωv−1 (ϖv )(N pv )s−1 )−1
Define the local L-factor to be Lv (s, ωv ) := (1 − ωv (ϖv )N p−s
v )
−1
and the local ε-factor to be
1
dv −s
εv (s, ωv , ψv ) := ωv (ϖv )(N Dv ) 2 .
In particular we see that for any factorizable φ = ⊗v φv ∈ S(AK ) and for all but finitely
many places v we have Dv = Ôv , φv = 1Ôv and ζ(s, ωv , φv ) = L(s, ωv ). Since the product
Q
v∤m L(s, ωv ) (here m is the conductor of ω) converges to L(s, ω) when Re(s) > 1, we get the
infinite product expansion (5.2.3) of the global zeta integral.

119
2. Suppose v is non-archimedean and ωv is ramified of conductor mv ≥ 1. Let dv := v(Dv ) be
the normalized valuation of the local different ideal. Let φ◦v := ωv−1 1Ôv× . In other words,
(
ωv (x)−1 if x ∈ Ôv×
φ◦v (x) =
0 / Ôv×
if x ∈

Then for all s ∈ C we have

Z
1
◦
ζ(s, ωv , φv ) = φ◦v (x)ωv (x)|x|sv d× xv = vol(Ôv× , d× xv ) = (N Dv )− 2 .
Kv×

Next we compute F(φ◦v ). For any x ∈ Kv we have

Z Z
F(φ◦v )(x) = φ◦v (y)ψv (xy)dy = ωv (y)−1 ψv (xy)dy
Kv Ôv×

where the integral is using the additive Haar measure.

When x ∈ ϖv−dv −mv +1 Ôv , then the value ψv (xy) is constant when y ranges in the subgroup
Uv (mv − 1) = ker(Ôv× → (Ôv /ϖvmv −1 )× ). On the other hand, ωv is nontrivial on Uv (mv − 1),
therefore the integral on each coset of Uv (mv − 1) vanishes and hence F(φ◦v )(x) = 0.
To determine the other values of F(φ◦v ) we decompose the integral into integrals on cosets of
Uv (mv ):
X Z
F(φ◦v )(x) = ωv (y)−1 ψv (xyu)du
Uv (mv )
y∈(O/pmv ×
v )
X Z (5.2.7)
= (N pv )−mv ωv (y)−1 ψv (xy) ψv (xyϖvmv a)da
Ôv
y∈(O/pmv ×
v )

/ ϖv−dv −mv Ôv = ϖv−mv Dv−1 , then ψv is nontrivial on xϖvmv Ôv and the integral
If x ∈
Z Z
ψv (xyu)du = ψv (xy) ψv (xy(u − 1))du
Uv (mv ) 1+ϖvm Ôv

vanishes and we get F(φ◦v )(x) = 0.

Finally if x ∈ ϖv−dv −mv Ôv× , we have

F(φ◦v )(x) = ωv (x)τ (ωv , ψv )

where we define
Z
τ (ωv , ψv ) :=ωv (ϖvdv +mv ) ωv (y)−1 ψv (ϖv−dv −mv y)dy
Ôv×
X
=ωv (ϖvdv +mv )vol(Uv (mv ), dx) ωv (y)−1 ψv (ϖv−dv −mv y)
(5.2.8)
y∈(O/pm v ×
v )
X
=(N pv )−mv ωv (ϖvdv +mv ) ωv (y)−1 ψv (ϖv−dv −mv y)
y∈(O/pmv ×
v )

in which the integral is using the additive Haar measure. Note that τ (ωv , ψv ) does not depend
on the choice of the uniformizer ϖv .
In summary we have
F(φ◦v ) = τ (ωv , ψv ) · ωv 1ϖv−dv −mv Ôv×

120
 from which we get that for all s ∈ C,
Z
ζ(1 − s, ωv−1 , F(φ◦v )) = F(φ◦v )(x)ωv (x)−1 |x|1−s d× xv
Kv×
(5.2.9)
= τ (ωv , ψv )(N pv )(dv +mv )(1−s) vol(Ôv× , d× x)
1
= τ (ωv , ψv )(N pv )(dv +mv )(1−s) (N Dv )− 2
We define the local L-factor to be L(s, ωv ) = 1 and the local epsilon factor to be
εv (s, ωv , ψv ) = τ (ωv , ψv )(N pv )(dv +mv )(1−s)

3. Suppose v is a real place. Then ωv (x) = ( |x|xv )nv |x|it

v
v
where nv ∈ {0, 1} and tv ∈ R. Let
2
φ◦v (x) := xnv e−πx . Then F(φ◦v ) = (−i)nv φ◦v and when Re(s) > 0 we have
Z
ζ(s, ωv , φ◦v ) = φ◦v (x)ωv (x)|x|sv d× x
×
K
Z v
2 x nv itv s dx
= xnv e−πx ( ) |x|v |x| (5.2.10)
R × |x| v x
Z ∞
2 dx
=2 e−πx xs+nv +itv = ΓR (s + nv + itv ).
0 x
Similarly when Re(s) < 1,
ζ(1 − s, ωv−1 , F(φ◦v )) = (−i)nv ΓR (1 − s + nv − itv )
Define the local L-factor to be L(s, ωv ) = ΓR (s + nv + itv ) and the local epsilon factor to be
ε(s, ωv , ψv ) = (−i)nv .
4. Suppose v is a complex place. Then ωv (reiθ ) = r2itv einv θ where nv ∈ Z and tv ∈ R. Let
2
φ◦v (reiθ ) := r|nv | e−inv θ e−2πr . Then F(φ◦v ) = (−i)|nv | φ◦ v and when Re(s) > 0 we have
Z
◦
ζ(s, ωv , φv ) = φ◦v (x)ωv (x)|x|sv d× x
Kv×
Z ∞ Z 2π
2 dr
=2 r|nv | e−inv θ e−2πr r2itv einv θ r2s dθ
0 r
Z 0∞ (5.2.11)
−2πr 2 2s+|nv |+2itv dr
= 2(2π) e r
0 r
|nv |
= ΓC (s + itv + )
2
Similarly when Re(s) < 1 we have
|nv |
ζ(1 − s, ωv−1 , F(φ◦v )) = (−i)|nv | ΓC (1 − s − itv + ).
2
Define the local L-factor to be L(s, ωv ) := ΓC (s + itv + |n2v | ) and the local epsilon factor to be
ε(s, ωv , φ◦v ) = (−i)|nv | .
Definition 5.2.1. The completed L-function of a Hecke character ω : K × \A×
K → C
×
is defined
when Re(s) > 1 to be Y
Λ(s, ω) := Lv (s, ωv )
v
where the product runs over all places of K. In other words, we have
Y
Λ(s, ω) = L(s, ω) Lv (s, ωv ), Re(s) > 1
v|∞

121
Lemma-Definition 5.2.11. The global epsilon factor of ω is the product
Y
ε(s, ω) := εv (s, ωv , ψv )
v

of local epsilon factors over all places of K. It is an everywhere nonzero entire function of s ∈ C
and is independent of the choice of the additive character ψ = ⊗v ψv .
Theorem 5.2.12. The completed L-function of a Hecke character ω : K × \A×
K has meromorphic
continuation to C and satisfies a functional equation
Λ(s, ω) = ε(s, ω)Λ(1 − s, ω −1 )
Moreover, if ω is nontrivial on A1K , then Λ(s, ω) is holomorphic on C; while if ω is trivial on A1K
so that ω(x) = ∥x∥itωpfor tω ∈ R, then Λ(s, ω) has simple poles at s = −itω and s = 1 − itω with
respective residues − |dK |vol(K × \A1K , d1 x) and vol(K × \A1K , d1 x) where
2r1 (2π)r2 RK hK
vol(K × \A1K , d1 x) = p .
wK |dK |
Q ζ(s,ωv ,φ◦v )
Proof. From the computation above we see that v L(s,ω v)
extends to an entire function of C.
Then we obtain the meromorphic continuation and functional equation of Λ(s, ω) from Theorem
5.2.9.
When ω is nontrivial on A1K , then ζ(s, ω, φ◦ ) is holomorphic on C and hence Λ(s, ω) is holomor-
phic on C. Suppose ω(x) = ∥x∥itω for all x ∈ A× ◦
K . Then one checks immediately that φ (0) = 1,
1 1
F(φ◦ ) = |dK |− 2 and Λ(s, ω) = |dK | 2 ζ(s, ω, φ◦ ). The statement on residues then follow from Theo-
rem 5.1.5 and Proposition 5.2.10.
Example 5.2.1. Let ω = triv be the trivial Hecke character. Then the associated Hecke L-function
is the Dedekind zeta function ζK (s) and the completed L-function is
Λ(s, triv) = ΓR (s)r1 ΓC (s)r2 ζK (s).
1 1
The global epsilon factor is ε(s, ω) = v finite (N Dv ) 2 −s = |dK | 2 −s . Then we get Theorem 5.1.5 by
Q
s
noting that ξK (s) = |dK | 2 Λ(s, triv) in the notation of loc. cit. Moreover from ΓR (1) = ΓC (1) = 1
we recover the class number formula.
Example 5.2.2. Let χ : (Z/f Z)× → C× be a Dirichlet character of conductor f > 1 and parity
ϵ ∈ {0, 1} (so that χ(−1) = (−1)ϵ ). From χ we define a Hecke character ω : Q× \A× ×
Q → C to be
the composition
χ−1
Q× \A× ∼ = R>0 × Ẑ× → (Z/f Z)× −−→ C×
Q

in which the first isomorphism is (4.5.2), the middle map sends (λ, u) ∈ R>0 × Ẑ× to u mod f ∈
(Z/f Z)× . Then for any prime number p ∤ f the factor ωp is unramified and ωp (p) = χ(p). Moreover,
ω∞ (x) = χ(sign(x)) = sign(x)ϵ . Therefore we have L(s, ω) = L(s, χ), the completed L-function of
ω is
Λ(s, ω) = ΓR (s + ϵ)L(s, χ)
and the global epsilon factor is
Y
ε(s, ω) = (−i)ϵ f 1−s τ (ωp , ep ).
p|f

pmp where mp = vp (f ), then we have

Q
Write f =
X 2πiy
τ (ωp , ep ) = p−mp ωp (pmp ) ωp (y)−1 exp( ).
pmp
y∈(Z/pmp Z)×

122
To relate ε(s, ω) and the Gauss sum Qτ (χ) we need to describe the factors ωp for p|f . Via the
canonical isomorphism (Z/f Z)× ∼ = p|f (Z/pmp Z)× , the character χ corresponds to a family of
characters χp : (Z/pmp Z)× → C× for p|f . Using the canonical homomorphism Z× p → (Z/p
mp
Z)×
×
we view each χp also as a character of Zp . Then for any prime p|f we have:
Y Y
ωp |Z×
p
= χ−1
p , ωp (p) = χ−1
q (p
−1
)= χq (p).
q|f,q̸=p q|f,q̸=p

f mp
P f
For any prime p|f , choose ap ∈ Z such that ap · pm p ≡ 1( mod p ). Then we have p|f ap · pm p ≡ 1(

mod f ) and therefore

2πix Y 2πiap x
exp( )= exp( mp ), ∀x ∈ R.
f p
p|f

On the other hand we have

Y Y Y Y Y Y f Y
ωp (pmp ) = χq (pmp ) = χq (pmp ) = χq ( m q ) = χ−1
p (ap ).
q
p|f p|f q|f,q̸=p q|f p|f,p̸=q q|f p|f

Consequently we get that

X 2πix
τ (χ) = χ(x) exp( )
f
x∈(Z/f Z)×
Y X 2πiap x
= χp (x) exp( )
pmp
p|f x∈(Z/pmp Z)×
  
Y X 2πix 
=  χ−1
p (ap )
 χp (x) exp( mp ) (5.2.12)
p
p|f x∈(Z/pmp Z)×
 
Y X 2πix
= ωp (pmp ) ωp (x)−1 exp
pmp
p|f x∈(Z/pmp Z)×
Y Y
= pmp τ (ωp , ep ) = f τ (ωp , ep )
p|f p|f

Then we deduce that ε(s, ω) = (−i)ϵ f −s τ (χ) and this reproves Theorem 5.1.3.
Let ω : K × \A×
K → C
×
be a Hecke character of conductor m. Then ω∞ : KR× → C× is trivial
× ×
×
on K ∩ U (m) = OK (m) and induces a homomorphism OK (m)\KR× → C× . Let χ : Im → C× be
the character
Q associated to ω. Recall that
Q for any finite place v ∤ m weQ
have χ(pv ) = ωv (ϖv ). The
product v|m ω defines a character of v|m Kv× whose restriction to v|m Ôv× factors through a
character that we denote by
ωm : (OK /m)× → C× .
Then for any 0 ̸= α ∈ OK relatively prime to m we have
Y Y
χ(αOK ) = ωv (α) = ω∞ (α)−1 ωv (α)−1 = ω∞ (α)−1 ωm (α)−1 .
v∤m v|m

Example 5.2.3. Assume that OK is a PID. Then the set of nonzero ideals a ⊂ OK relatively
×
prime to m corresponds bijective to the set of OK orbits on {α ∈ OK | (α) + m = OK }. Choose a
×
representative β in each class of Clm = coker(OK → (OK /m)× ). Then we have a bijection
× ∼
G
×
{α ∈ OK | (α) + m = OK }/OK −→ {α ∈ OK | α ≡ β mod m}/OK (m).
β∈Clm

123
The L-series of ω is
X χ(a) X ω∞ (α)−1 ωm (α)−1
L(s, ω) = =
N as ×
(N α)s
0̸=a⊂OK {α∈OK |(α)+m=OK }/OK
a+m=OK
(5.2.13)
−1
X
−1
X ω∞ (α)
= ωm (β)
×
(N α)s
β∈Clm {α∈OK |α≡β mod m}/OK (m)

124
Appendix A

Field theory

The main reference is [Mor96].

A.1 Compositum and linear disjointness

We mainly follow [Mor96, §20].
Let K1 , K2 be two fields that are embedded in another field L. Denote φi : Ki ,→ L the field
embeddings for i = 1, 2.
Definition A.1.1. The compositum of K1 and K2 in L, denoted K1 K2 , is the smallest subfield of
L containing both φ1 (K1 ) and φ2 (K2 ). Let F be the prime field of L (either Q or Fp ). Then K1 K2
is the image of the homomorphism

φ1 ⊗ φ2 : K1 ⊗F K2 → L.

In general the√compositum K1 K2 depends on the field embeddings φ1 , φ2 . For example, take

K1 = K2 = Q( 3 2). Let φ1 : K1 ,→ C √ be the unique
√ embedding whose image is in R. Let
3 3 2πi
φ2 : K2 ,→ C be the embedding √ sending 2 to ω 2 where√ ω = e 3 . Then the compositum
K1 K2 under φ1 and φ2 is Q( 3 2, ω), the Galois closure of Q( √
3
2). But if we change φ2 to the same
3
embedding as φ1 , then the compositum K1 K2 would be Q( 2).
However such ambiguity disappears if K1 , K2 are extensions of a common field F and one of
them, say K1 /F , is a Galois extension. In this case K1 is the splitting field of a set of separable
polynomials in F [T ] and K1 K2 is the splitting field of the same set of polynomials, but viewed in
K2 [T ] (which are still separable). In particular K1 K2 /K2 is also Galois and we have an isomorphism

Gal(K1 K2 /K2 ) ∼
= Gal(K1 /K1 ∩ K2 ).

In such situations we can talk about the compositum K1 K2 without embedding them into any larger
field.
Suppose moreover that K1 , K2 are both Galois extensions of a field F . Then their compositum
K1 K2 is also a Galois extension of F and we have an injective group homomorphism

Gal(K1 K2 /F ) ,→ Gal(K1 /F ) × Gal(K2 /F )

whose image consists of pairs (σ1 , σ2 ) ∈ Gal(K1 /F ) × Gal(K2 /F ) such that σ1 |K1 ∩K2 = σ2 |K1 ∩K2 .
To determine whether this map is surjective we need the following notion.
Lemma-Definition A.1.1. Two finite extensions K1 , K2 of a field F are linearly disjoint over F
if any of the following equivalent conditions are satisfied:

125
 1. K1 ⊗F K2 is a field.
2. K1 ⊗F K2 is an integral domain.

3. For any extension L/F containing both K1 and K2 , let K1 K2 be their compositum in L. Then
[K1 K2 : F ] = [K1 : F ] · [K2 : F ].
4. For any extension L/F containing both K1 and K2 , let φi : Ki ,→ L be the F -algebra embed-
dings for i = 1, 2. Then for any F -linearly independent subset S ⊂ K1 , the subset φ1 (S) ⊂ L
is K2 -linearly independent.
These conditions imply, and when either K1 /F or K2 /F is Galois, are equivalent to the following
condition:
5. For any extension L/F containing both K1 and K2 , we have K1 ∩ K2 = F where the intersec-
tion is taken inside L.

Remark A.1.2. If K1 /F and K2 /F are not Galois, then √ the last condition does
√ not imply linearly
3 3 2πi
disjointness in general. For example, take K1 = Q( 2) and K2 = Q(ω 2) where ω = e 3 ,
both viewed√as subfields of C. Then K1 ∩ K2 = Q but they are not linearly disjoint over Q since
K1 K2 = Q( 3 2, ω) is the Galois closure
√ of K
√ 1 and we have [K1 K2 : Q] = 6 but [K1 : F ]·[K2 : F ] = 9.
In fact we have K1 ⊗Q K2 ∼ 3 3
= Q( 2) × Q( 2, ω).
When two extensions K1 and K2 of F are linearly disjoint over F , we can unambiguously define
their compositum K1 K2 without using any embedding into a larger extension, since under any such
embedding the compositum is isomorphic to K1 ⊗F K2 . When K1 , K2 are linearly disjoint Galois
extensions of F we have a canonical isomorphism

Gal(K1 K2 /F ) ∼
= Gal(K1 /F ) × Gal(K2 /F ).

126
Appendix B

Topological groups

A modern reference is [RV99, §1-§3].

B.1 Basic properties

Definition B.1.1. A topological group is a group G equipped with a topology such that the multi-
plication map G × G → G : (x, y) 7→ xy and the inverse map G → G : x 7→ x−1 are both continuous
(here G × G is equipped with the product topology).
Definition B.1.2. A topological ring is a ring R equipped with a topology such that the underlying
additive group is a topological group and the multiplication map A × A → A : (x, y) 7→ xy is
continuous.
A topological field is a topological ring K that is a field and such that the map K × → K × : x 7→
x is continous when K × = K − {0} is equipped with the subspace topology from K.
−1

Lemma B.1.1. Let G be a topological group and let e ∈ G denote the identity element. For any
neighborhood U of e in G,

1. There exists an open neighborhood e ∈ V ⊂ U such that V · V ⊂ U ;

2. There exists an open neighborhood e ∈ V ⊂ U such that V = V −1 ;
3. There exists an open neighborhood e ∈ V ⊂ U such that V · V ⊂ U and V = V −1
Proof. (1) Let m : G × G → G be the multiplication map. Since m−1 (U ) is an open neighborhood
of (e, e) in G × G, there exists open neighborhoods V1 , V2 of e in G such that V1 × V2 ⊂ m−1 (U ) by
the definition of product topology. Let V := V1 ∩ V2 . Then V is an open neighborhood of e in G
satisfying V · V ⊂ U .
(2) For any open neighborhood e ∈ U ⊂ G, the open neighborhood V := U ∩ U −1 of e satisfies
V = V −1 .
(3) Take V from (1) and then use (2) to get an open neighborhood e ∈ W ⊂ V such that
W = W −1 and W · W ⊂ V · V ⊂ U .
Let G be a topological group and H ⊂ G be a subgroup. We equip the quotient space G/H
with the quotient topology, i.e. the finest topology such that the quotient map π : G → G/H is
continuous. Then a subset V ⊂ G/H is open if and only if its π −1 (V ) is open. With this topology,
the quotient map π is open. Indeed, for any open subset U ⊂ G, π −1 (π(U )) = ∪h∈H U h is open and
hence π(U ) ⊂ G/H is open.

Lemma B.1.2. Let H be a subgroup of a topological group G.

127
 1. If H is open, then H is also closed in G.
2. The quotient space G/H is discrete if and only if H is open.

3. The quotient space G/H is Hausdorff if and only if H is closed.

Proof. (1) Since H is open, any right coset of H in G is also open. Thus H is closed since its
complement G \ H, being a union of right H-cosets, is open in G.
(2) If H is open, then any right coset of H is open. Since the quotient map G → G/H is open,
we get that any point in G/H is open so that G/H is discrete. Conversely if G/H is discrete, then
H is open since it is the inverse image of a point in G/H (which is open in G/H).
(3) If G/H is Hausdorff, then any point in it is closed and hence H is closed. Conversely assume
that H is a closed subgroup of G. For any x, y ∈ G such that xH ̸= yH, the set xHy −1 is closed in
G and does not contain the identity e. Then its complement U = G\xHy −1 is an open neighbood of
e in G. By the previous lemma, there exists an open neighborhood e ∈ V ⊂ U such that V · V ⊂ U .
Then π(V x) and π(V y) are open neighborhoods of xH and yH in G/H. It remains to show that
π(V x) ∩ π(V y) = ∅. Suppose not, then there exists z ∈ V xH ∩ V yH and hence

e = zz −1 ∈ (V xH)(V yH)−1 = V xHy −1 V −1

and we get that V · V ∩ xHy −1 ̸= ∅. But this contradicts with the fact that V · V ⊂ U = G \ xHy −1
and we are done.
Lemma B.1.3. Let G be a topological group and H ⊂ G a subgroup. Then the closure H of H in
G is also a subgroup of G.
Proof. Let m : G × G → G be the multiplication map. Suppose there exists x, y ∈ H such that
xy ∈ / H. Then m−1 (G − H) is an open neighborhood of (x, y) in G × G and hence there exists
open neighborhoods x ∈ U1 , y ∈ U2 in G such that U1 × U2 ⊂ m−1 (G − H). Since H is dense
in H, there exists x′ ∈ U1 ∩ H and y ′ ∈ U2 ∩ H. Then we have m(x′ , y ′ ) = x′ y ′ ∈ H and hence
(x′ , y ′ ) ∈ m(U1 × U2 ) ∩ H. But this is a contradiction with the fact that U1 × U2 ⊂ m−1 (G − H).
Thus H is a subgroup of G.
Lemma B.1.4. Let X be a Hausdorff topological space. Let Y ⊂ X be a subset and let Y be its
closure in X. If Y is locally compact for the subspace topology, then Y is open in Y .
Proof. Since X is Hausdorff, Y is also Hausdorff. Let y ∈ Y and let y ∈ U ⊂ Y be an open
neighborhood such that the closure C of U in Y is compact. Since Y is Hausdorff, C is closed in
Y . Let V ⊂ Y be an open subset such that V ∩ Y = U . Then V \ C is an open subset of Y and we
have Y ∩ (V \ C) ⊂ (Y ∩ V ) \ C = U \ C = ∅. Since Y is dense in Y we must have V \ C = ∅ and
hence V ⊂ C ⊂ Y . Then V is an open neighborhood of y in Y and hence Y is open in Y .

Proposition B.1.5. Let H be a subgroup of a Hausdorff topological group. Suppose H is locally

compact for the subspace topology. Then H is a closed subgroup of G. In particular, if H is discrete
then it is closed.
Proof. Let H be the closure of H in G. By Lemma B.1.3 H is a subgroup of G and by Lemma B.1.4
H is an open subgroup of H. Therefore H is also closed in H by Lemma B.1.2 and hence H = H
is a closed subgroup of G.
Since any discrete topological space is locally compact, the last statement then follows.

128
B.2 Pontryagin duality
Definition B.2.1. A (unitary) character (resp. quasi-character ) of a topological group G is a
continuous homomorphism of topological groups G → C1 (resp. G → C× ), where C1 := {z ∈
C× , |z| = 1}.

We denote the set of all characters of G by G.b It is a group under pointwise multiplication.
We equip G b with the compact open topology for which a neighborhood base of the trivial character
b : χ(K) ⊂ V } where K ⊂ G is a compact subset and V ⊂ C1 is an
consists of W (K, V ) = {χ ∈ G
open neighborhood of 1.
Lemma B.2.1. Let G be an abelian topological group.
1. If G is discrete, then G
b is compact.

2. If G is compact, then G
b is discrete.

3. If G is locally compact Hausdorff, then G

b is also locally compact Hausdorff.

Theorem B.2.2 (Pontryagin duality). Let G be a locally compact Hausdorff abelian group. Then
the map
G→G : g 7→ (χ 7→ χ(g)), ∀g ∈ G, χ ∈ G
bb b

is an isomorphism of topological groups.

Lemma B.2.3. Let G be a locally compact Hausdorff abelian group and let H ⊂ G be a closed
[ is a closed subgroup of G
subgroup. Then H and G/H are both locally compact Hausdorff and G/H b
with quotient isomorphic to H.
b In other words, we have a short exact sequence of abelian groups

[ →G
1 → G/H b→H
b →1

129
Bibliography

[Cona] Brian Conrad. Completion of algebraic closure. Expository note. http://virtualmath1.

stanford.edu/~conrad/248APage/handouts/algclosurecomp.pdf. 63
[Conb] Keith Conrad. A separable extension with inseparable residue field. Expository note. https:
//kconrad.math.uconn.edu/blurbs/gradnumthy/sepfield-and-insep-resfield.pdf.
36
[Kap74] Irving Kaplansky. Commutative rings. University of Chicago Press, Chicago, Ill.-London,
revised edition, 1974. 27
[Mar18] Daniel A. Marcus. Number fields. Universitext. Springer, Cham, second edition, 2018.
With a foreword by Barry Mazur. 89
[Mor96] Patrick Morandi. Field and Galois theory, volume 167 of Graduate Texts in Mathematics.
Springer-Verlag, New York, 1996. 28, 125
[Neu99] Jürgen Neukirch. Algebraic number theory, volume 322 of Grundlehren der mathematis-
chen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag,
Berlin, 1999. Translated from the 1992 German original and with a note by Norbert Schap-
pacher, With a foreword by G. Harder. 27, 52, 58, 65, 66, 67, 68, 73, 82
[RV99] Dinakar Ramakrishnan and Robert J. Valenza. Fourier analysis on number fields, volume
186 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1999. 113, 127

[Ser79] Jean-Pierre Serre. Local fields, volume 67 of Graduate Texts in Mathematics. Springer-
Verlag, New York-Berlin, 1979. Translated from the French by Marvin Jay Greenberg.
58
[SS03] Elias M. Stein and Rami Shakarchi. Fourier analysis, volume 1 of Princeton Lectures in
Analysis. Princeton University Press, Princeton, NJ, 2003. An introduction. 105

[Tat67] J. T. Tate. Fourier analysis in number fields, and Hecke’s zeta-functions. In Algebraic
Number Theory (Proc. Instructional Conf., Brighton, 1965), pages 305–347. Academic
Press, London, 1967. 119
[Wei95] André Weil. Basic number theory. Classics in Mathematics. Springer-Verlag, Berlin, 1995.
Reprint of the second (1973) edition. 58

130