NUMBER RINGS

P. Stevenhagen

Universiteit Leiden
2019
Date of this online version: November 14, 2020
Please send any comments on this text to psh@[Link].

Mail address of the author:

P. Stevenhagen
Mathematisch Instituut
Universiteit Leiden
Postbus 9512
2300 RA Leiden
Netherlands
CONTENTS

1. Introduction to number rings 4

The Pell equation • Gaussian integers • Exploiting number rings • The arithmetic of number rings
• Exercises

2. Ideal arithmetic 15
Ideals • Invertible ideals • Localization • Ideals in number rings • Primary decomposition • Local
number rings • Exercises

3. Explicit ideal factorization 30

The Kummer-Dedekind theorem • Singular primes • Quadratic and cyclotomic number rings • In-
tegral closure • Exercises

4. Linear algebra for number rings 41

Norm and trace • Discriminant • Computational techniques • Ramification • Exercises

5. Geometry of numbers 52
Minkowski’s theorem • Number rings as lattices • Finiteness of Picard groups • Minkowski’s constant
• Hermite’s theorem • Dirichlet’s unit theorem • Exercises

6. Zeta functions 66
Dedekind zeta function • Analytic class number formula • Functional equation

7. Computing units and class groups 69

Pell revisited • A cubic example • A larger cubic example • An example with unit rank 2 • Exercises

8. Galois theory for number fields 81

Galois action on extension primes • Decomposition and inertia groups • The Frobenius symbol
• The non-Galois case • Exercises

Literature 86
Index 88
1 Introduction to number rings
A number field is a finite field extension of the field of rational numbers Q, and a number ring
is a subring of a number field. This introduction shows how number rings arise naturally
when solving equations in ordinary integers.
Many basic questions about integers can be phrased as a problem of finding rational or
integral solutions to equations in several indeterminates with rational coefficients. Starting
for instance from the empirical observation that all small positive numbers can be written as
the sum of at most 4 integral squares, one might guess that the equation n = w2 +x2 +y 2 +z 2
admits integral solutions for all positive n. Similarly, the abundance of ‘essentially different’
integral solutions to the Pythagorean equation a2 +b2 = c2 makes one think that the equation
x2 + y 2 = 1, which describes the unit circle in the Euclidean plane, possesses infinitely many
rational solutions. For higher exponents, a deep theorem of Wiles [28] confirms Fermat’s
belief that for k > 2, the curve xk + y k = 1 admits only the trivial rational solutions with
xy = 0. All these equations, in which one restricts to integral or rational rather than real or
complex solutions, are examples of Diophantine equations. They are named after Diophantus
of Alexandria, who treated several of them in a work in 13 books called Arithmetica (± 250
AD). Only six books have survived [14]. They have been a source of inspiration to later
mathematicians such as Fermat (1601–1665) and Euler (1707–1783).
This introduction treats four classical examples of Diophantine equations. The reader
who is interested in the history of these examples or the history of number theory in general
should consult Weil’s account [27] (for the period before 1800) or the appropriate chapters
in [25, 13, 10].

I The Pell equation

We start with two special cases of a problem that was posed by Fermat as a challenge
problem to the English mathematicians in 1657. It has become known as the Pell equation
because of an erroneous attribution of its solution by Euler to the English mathematician
John Pell (1611–1685).

1.1. Problem. Find all integral solutions to the equation x2 − dy 2 = 1 for d = 3 and for
d = 1141.

Problem 1.1 requires the determination of all integers y for which 1+dy 2 is a square. If we do
not know how to approach such a problem, there is always the possibility to try a few values
of y and see whether we find solutions. Obviously, y gives a solution if and only if −y does,
so we may assume that y is non-negative. In case d = 3, it is not hard to see that the first few
values of y for which 1 + 3y 2 is a square are y = 0, 1, 4, and 15. Using a computer, one finds
the next few values to be 56, 209, 780, 2911 and 10864. The corresponding non-negative
x-values are x = 1, 2, 7, 26, 97, 362, 1351, 5042 and 18817.
A moment’s reflection shows that the sequences of x and y-values that we obtained both
satisfy the second order recurrence relation sn = 4sn−1 − sn−2 . Provided that we are able to
prove that we always obtain non-trivial solutions to our equation, we have experimentally
found a way to generate infinitely many solutions. This leaves us with a statement to prove

4
 §1: Introduction to number rings

and a completeness question: do we obtain all solutions by our procedure?

For d = 1141, we can try to follow the same procedure, but here something different
happens: after the obvious value y = 0, we find no small values of y that yield a solution.
Unlike Fermat’s contemporaries, the modern reader can use a computer to look for a small
solution. In this case a computer does not help: it will only tell us that there are no other
solutions with y < 106 or y < 1012 . More computer time can only raise the exponent to 15
or 18, it will not change the situation. Of course, such a huge number is likely to convince
us that there are no further solutions. It shouldn’t:

1 + 1141 · 306933853227656571973972082 = 10367823941572239632371252152

turns out to be the smallest solution.

At this point, it is clear that we need something more conceptual to approach the Pell
equation. For any integer d > 1 that √ is not a square,
√ the left hand side of the equation can
be factored in the √ number ring√ Z[ d] = {a + b d : a, b ∈ Z}. This is a subring of the real
quadratic field Q( d). In Z[ d], the Pell equation takes the form
√ √
(x + y d)(x − y d) = 1,
√
so every solution to the equation gives rise to a unit in the number ring Z[√ d]. The converse
is almost true, as can be seen with the help of the norm function N : Z[ d] → Z, which is
defined by √ √ √
N (x + y d) = (x + y d)(x − y d) = x2 − dy 2 .
√
We have√ ∗N (αβ) = N (α)N (β) for α, β ∈ Z[ d], so the norm gives rise to a homomorphism
N : Z[ d] → {±1} on the unit groups. The solutions to Pell’s equation correspond to the
elements in the
√ kernel of this homomorphism. Clearly, the kernel is a subgroup of index at
∗
most 2 in Z[ d] .
Let us first consider√the case d = 3. As x2 − 3y 2 is never congruent to −1 mod 4 for
x, y ∈ Z, all units in Z[ 3] have norm 1 and we conclude that the integral √ solutions to
2 2
the equation x − 3y = 1 correspond bijectively to the units in the √ ring Z[ 3]. Apart
from the ‘trivial solutions’ (x, y) = (±1, 0) corresponding √ to ±1 ∈ Z[ 3]∗ , there is the small
solution
√ ∗ (x, y) = (2, 1) corresponding to the unit ε3 = 2 + 3. This unit has infinite order in
Z[ 3] , and we obtain
√ ∗ infinitely many solutions to our equation √ coming from the subgroup
h−1i × hε3√
i ⊂ Z[ 3] . One can show (cf. exercises
√ 9–11) that 2 + 3 is in√fact a fundamental
unit in Z[ 3], meaning that −1 and 2 + 3 generate the full group Z[ 3]∗ . This implies
that the complete set of integral solutions to the equation x2 − 3y 2 = 1 is

{(xn , yn ) : n ∈ Z} ∪ {(−xn , −yn ) : n ∈ Z},

√ √ n
where the
√ numbers x n and y n are defined by x n + y n 3 = (2 + 3) . In this way, the number
ring Z[ 3] enables us to give a transparent description of the set of solutions of the Pell
equation for d = 3.
Exercise 1. Show that {xn }n and {yn }n satisfy the second order recurrence sn = 4sn−1 − sn−2 .

5
§1: Introduction to number rings

We√now turn to the case d = 1141 = 7 · 163. As before, we have to determine the unit group
Z[ 1141]∗ . Note that 2 2
√ all units have norm 1, since x −1141y is never √ congruent to −1 mod 7.
As in the case of Z[ 3], one shows that if there exists a √ unit in Z[ 1141] different from ±1,
then there exists a fundamental unit ε1141 such that Z[ 1141]∗ = h−1i × hε1141 i. Once we
have found the unit ε1141 , it is easy to describe the solutions of the equation x2 − 1141y 2 = 1.
Thus we are faced with the problem of finding ε1141 , if it √ exists. The Dirichlet unit theorem
(theorem 5.13) implies that every real quadratic ring Z[ d] has a fundamental unit εd . This
guarantees that our computer is bound to find a positive integer y for which 1 + 1141y 2 is
a square after a finite amount of time. However, one has to be extremely patient as the
smallest solution given above shows that the fundamental unit equals
√
ε1141 = 1036782394157223963237125215 + 30693385322765657197397208 1141.

The situation can even be worse: for d = 1000099, the smallest number y > 0 giving a
solution to the Pell equation has 1115 decimal digits.
Exercise 2. Making any reasonable hypothesis on computer equipment, show that no implementation of
our simple trial and error method will ever find this solution.

Examples as those above make clear that we need better ways to solve the Pell equation if
we want to do it in practice. Several approaches exist, and it will come as no surprise that
quadratic irrationalities play an essential role in them. We will exhibit a solution, in example
7.1, by general methods for finding units, Using such algorithms, a modern computer can
find a 1000-digit solution in less than a second.

I Gaussian integers
The second problem we will treat is very well known. It goes back to Fermat (1640), and
the first solution known to us occurs in a 1749 letter of Euler to Goldbach.
1.2. Problem. Determine which prime numbers can be written as the sum of two squares.
Solution. By looking at x2 + y 2 mod 4, it is easy to see that no prime p ≡ 3 mod 4 is a sum
of two squares. The prime 2 = 12 + 12 and the first few primes congruent to 1 mod 4 are
sums of squares in an essentially unique way:

5 = 22 + 12 13 = 32 + 22 17 = 42 + 12 29 = 52 + 22 .

Proving that this is a general phenomenon is done most easily by regarding such identities
√
as decompositions of prime numbers in the ring Z[i] = Z[ −1] of Gaussian integers:

5 = (2 + i)(2 − i) 13 = (3 + 2i)(3 − 2i) 17 = (4 + i)(4 − i) 29 = (5 + 2i)(5 − 2i).

The number ring Z[i] is in many ways similar to the ring Z of ordinary integers. Just like
in Z, we have a Euclidean algorithm in Z[i]. This means that, given any two elements
α, β ∈ Z[i] with β 6= 0, there exist elements q, r ∈ Z[i] such that α = qβ + r holds and
the (complex) absolute value |r| of the remainder of the division is strictly smaller than |β|.
Writing the identity above as

α/β = q + r/β with |r/β| < 1,

6
 §1: Introduction to number rings

we see that this amounts to saying that every element α/β ∈ Q(i) can be approximated by
a Gaussian integer q ∈ Z[i] in such a way that we have |α/β − q| < 1. A picture shows that
this is indeed the case: the open discs with radius 1 centered at the elements of Z[i] cover
the entire complex plane.

Any integral domain admitting a Euclidean algorithm is a principal ideal domain and admits
unique factorization.
Exercise 3. Check (or look up) these statements in case you have not seen them before.

Let now p ≡ 1 mod 4 be a prime number. Then −1 is the square of an element of order
4 in the cyclic group F∗p of order p − 1, so we can find an integer x for which p divides
x2 + 1 = (x + i)(x − i). As it is clear that p divides neither x + i nor x − i in Z[i], we find
that p is not a prime element in Z[i].
Exercise 4. Show that one can take x = ( p−1
2 )! in the argument above.

As in 1.1, we have a multiplicative norm N : Z[i] → Z that maps a + bi to (a + bi)(a − bi) =

a2 + b2 . As a2 + b2 = 1 has only four integral solutions, the only units in Z[i] are the four
powers of i.
As p is not prime in Z[i], there exist non-units π, π 0 ∈ Z[i] such that p = ππ 0 . The
identity p2 = N (p) = N (π)N (π 0 ) shows that π and π 0 both have norm p. Writing π = x + yi,
we find p = ππ = x2 + y 2 , as was to be shown. As π and π = π 0 are up to multiplication
by powers of i the unique divisors of p, we even find that the squares in the representation
p = x2 + y 2 are uniquely determined. As in the previous example, the introduction of a
suitable number ring leads to a complete answer.
We give another application of the Gaussian integers. In geometric language, the prob-
lem to be solved deals with the integral points on a plane cubic curve.

7
§1: Introduction to number rings

1.3. Problem. Find all integral solutions to the equation x2 + 1 = y 3 .

Solution. If x is odd, then x2 + 1 is never a cube as it is congruent to 2 mod 4. Suppose

that (x, y) is a solution to our equation. We pass again to the number ring Z[i], where we
have an equality
(x + i)(x − i) = y 3 .
A prime element in Z[i] that divides both x + i and x − i divides their difference 2i, so it is
up to units equal to 1 + i. However, 1 + i does not divide x + i if x is even, so we conclude
that x + i and x − i are coprime in Z[i]. Their product is a cube, so unique factorization
in Z[i] shows that each of them is the product of a unit and a cube in Z[i]. As all units in
Z[i] are cubes, there must be integers a, b ∈ Z such that x + i = (a + bi)3 . This yields the
equations
x = a(a2 − 3b2 ) and 1 = (3a2 − b2 )b.
It follows that we have b = ±1, and an inspection of both cases shows that the only solution
is (a, b) = (0, −1). This implies that the only solution to our original equation is x = 0,
y = 1. In other words: a non-zero square is never followed by a cube. 

I Exploiting number rings

Slightly changing the cubic curve from the previous problem, we run into number rings
different from Z[i].

1.4. Problem. Find all integral solutions to the equation x2 + 19 = y 3 .

Led by the similarity with the previous problem, we try to adapt the argument given there
to our present needs. This time, we factor the left hand side of our equation in the number
√
ring Z[ −19] as √ √
(x + −19)(x − −19) = y 3 .
We need to check that for a hypothetical solution (x, y), the two factors on the left hand
√ √
side are coprime elements in Z[ −19]. By this, we mean that there exist α, β ∈ Z[ −19]
√ √
such that we have α(x + −19) + β(x − y −19) = 1. Equivalently, one may show that
√ √
x + −19 is coprime to the difference 2 −19 of the two factors.
If x is odd, then x2 + 19 is congruent to 4 mod 8, so it cannot be a cube. If x is divisible
by 19, then x2 + 19 is congruent to 19 mod 192 , so it cannot be a cube. Thus x is even and
not divisible by 19, and this implies that x2 + 19 and 38 are coprime. Choose a, b ∈ Z such
√ √
that a(x2 + 19) + 38b = 1. Then α = a(x − −19) and β = −b −19 achieve what we want.
√
By the same argument as for Z[i], we deduce that x + −19 is the product of a unit
√ √
and a cube in Z[ −19]. This time the norm function N : Z[ −19] → Z is defined by
√ √
N (a+b −19) = a2 +19b2 , and we have Z[ −19]∗ = {±1} as the norm equation a2 +19b2 = 1
only has the trivial solutions (±1, 0). As before, we find that there are integers a and b such
that
x = a(a2 − 57b2 ) and 1 = (3a2 − 19b2 )b.
It is immediately obvious that this time the values b = ±1 do not lead to a solution. We
conclude that the equation from problem 1.4 does not admit integral solutions.

8
 §1: Introduction to number rings

This is a nice argument, and one would be inclined to believe it – if it weren’t true that
the equality
182 + 19 = 324 + 19 = 343 = 73
convincingly shows that there do exist solutions. It turns out that our argument is fallacious,
and a closer inspection reveals that there is a doubtful step in our argument: we applied the
‘obvious fact’ that a product of two coprime elements in a number ring can only be a cube
if each of the factors is the product of a unit and a cube. This is correct for the ring Z[i], as
√
one sees by invoking the uniqueness of factorization in Z[i]. For the number ring Z[ −19]
however, this is no longer an obvious fact. More precisely, it is easy to find factorizations
like √ √
20 = (1 + −19)(1 − −19) = 2 · 2 · 5
that are non-unique. These are factorizations into irreducible elements, as one can easily
√
check with the help of the norm function. Apparently, the number ring Z[ −19] is rather
different from Z[i], and we cannot indiscriminately take over arguments that are correct in
√
Z[i] to Z[ −19].
√ √ √
Exercise 5. Find elements α and β in Z[ −19] such that α(18 + −19) + β(18 − −19) = 1.

I The arithmetic of number rings

In this text, we will focus on the arithmetic of arbitrary number rings. As shown by our
examples, we need to control the units in these rings, and to develop a sound theory of
unique factorization.
Our first step will be the introduction of a suitable notion of ideals in number rings, for
which we will prove a form of unique factorization. Although ideals and elements are not the
same objects, the ‘obstruction’ to ideals having a generating element can be measured by the
Picard group of the number ring. This is an abelian group, and for number rings, which we
can embed in Euclidean spaces and study with methods from the geometry of numbers, we
will show that the Picard group is always finite. For the units, techniques from the geometry
of numbers also yield a finiteness theorem, the Dirichlet unit theorem, which describes the
structure of the unit group of a number ring.
These two basic finiteness theorems are at the heart of what one might call classical
algebraic number theory, a powerful instrument that enables us to solve many questions that
cannot easily be settled in other ways. They are the content of the sections 2 through 7
of this text. Section 8 is concerned with Galois theoretic aspects of number rings and the
behaviour of primes in number field extensions.
The main difference between our approach and those found in most textbooks on alge-
braic number theory is that we do not focus exclusively on the rings of integers in number
fields, which have the nicest theoretical properties
√ but may be algorithmically inaccessible.
Already in the case of the quadratic rings Z[ d] occurring in this section, one may not be
able to decide whether a given large integer d > 0 is squarefree. √ This is however not an
essential obstruction
√ to studying arithmetical properties of Z[ d]. In fact, Picard groups
of rings Z[ d] were already studied by Gauss in terms of class groups of binary quadratic
forms, and for d > 0 the solvability of the Pell equation – by continued fractions or other

9
§1: Introduction to number rings

means – is not essentially different if d has square factors.

Sometimes, the applicability of algebraic number theory is not at all obvious from the
problem. Nowadays, the most promising approach to the fundamental problem of finding
the prime factorization of a given large integer, for which straightforward methods are far
too slow to be of any practical help, is the number field sieve that we will treat in due
time. Somewhat surprisingly, it is exactly this kind of ‘practical application’ that forces us
to develop the theory in more generality than is customary for texts on this level. Thus, the
extra amount of abstract commutative algebra that goes into our treatment is caused by the
desire to obtain a theory that works in practical situations.
One should not get the impression from what we said before that classical algebraic
number theory is in any sense a complete theory. First of all, there are many basic problems
of both algorithmic [21] and theoretical nature in the theory that are still open. Furthermore,
there are many questions of elementary number theoretic nature that nobody can currently
answer. Even the development of modern algebraic number theory, which we will touch
upon in the final part of these notes, has not changed this situation.
In recent years, considerable progress has been realized in arithmetical questions using
methods from arithmetic algebraic geometry. In this area, one uses suitable algebraic gen-
eralizations of concepts that were originally developed by geometers working over the field
of complex numbers. The most striking example of this phenomenon is without doubt the
proof of Fermat’s last theorem by Wiles in 1993-94 [28]. Besides algebraic number theory,
this proof uses the arithmetic of elliptic curves, which is currently a topic of intensive re-
search in arithmetic algebraic geometry. It has given rise to several fundamental results and
conjectures, see [24, 18]. We refer to [12] for methods in arithmetic of an even more geo-
metrical nature, such as those leading to Faltings’s celebrated finiteness result (previously
Mordell’s conjecture) for the number of solutions to large classes of Diophantine equations.
Like any text in this area, it requires a strong background in abstract algebra, cf. [15].
We finally mention that there is no such thing as a universal theory of Diophantine
equations. This is the content of a theorem from mathematical logic (proved by Matiyasevich
in 1970, see [22]), which states that there is no general algorithm that will decide in a finite
number of steps whether a Diophantine equation admits an integral solution. This provides
a negative answer to a question commonly known as Hilbert’s tenth problem. The problem
had been the tenth in the list of open problems presented by Hilbert in his famous lecture
at the international congress of mathematics in Paris in 1900 [17].

Exercises.
2
6. Show that the set of rational solutions to the equation x2 + y 2 = 1 is {(1, 0)} ∪ {( tt2 −1 , 2t ) :
+1 t2 +1
t ∈ Q}.
[Hint: intersect the unit circle with the line y = t(x − 1).]
7. A Euclidean function on a ring R is a function g : R \ {0} → Z≥0 such that for any two
elements a and b 6= 0 in R, we can write a = qb + r with q, r ∈ R in such a way that we have
either r = 0 or g(r) < g(b). Show that a domain R is a principal ideal domain if it admits a
Euclidean function.
√
8. Show that the ring Z[ 3] is a principal ideal domain.

10
 §1: Introduction to number rings
√ √
9. Let d > 1 be a non-square integer with square root d ∈ R>0 , and define L : Z[ d]∗ → R
by L(u) = log |u|. Prove the following.
a. L is a homomorphism with kernel ker L √ = {±1} √ and im L is a discrete subgroup of R.
[Hint: im L√ u = x + y d ∈ Z[ d]∗ with x, y ∈ Z>0 }.] √
∩ R>0 = {log(u) : √
b. One has Z[ d]∗ = {±1} or Z[ d]∗ = h−1i × hεd i for some unit εd ∈ Z[ d]∗ of infinite
order . √
c. Suppose there exists a fundamental unit εd as in b. Then the group P ⊂ Z[ d]∗ of units
of norm 1 equals P = h−1i × hεkd i with k = 1 if εd is in P and k = 2 otherwise.
[The Dirichlet unit theorem 5.13 implies that the hypothesis is always satisfied.]
√ √
10. Let d be a non-square integer and α = a + b d an element of √ Q( d) of norm u = a2 − db2 .
Define the sequences {xn }∞ ∞ n
n=0 and {yn }n=0 by α = xn + yn d. Show that these sequences
have initial values x0 = 1, x1 = a, y0 = 0 and y1 = b and satisfy the recurrence sn+1 =
2asn − usn−1 for all n ∈ Z>0 .
11. Let d > 1 be a non-square integer and let y√be the smallest positive integer
√ for which dy 2 is
of the form x2 ± 1. Show that εd = x + y d is a fundamental unit in Z[ d]. Compute εd
and the norm N (εd ) for all non-square d ≤ 20.
12. Let d > 1 be an integer congruent to 1 mod 4 that is not a square, and let α ∈ R be a zero √of
2 d−1
the polynomial X − X − 4 ∈ Z[X]. Show √ that Z[α] is a subring of R that contains Z[ d]
as a subring of index 2. Show also that Z[ d] is of finite index in Z[α]∗ , and that this index
∗

is equal to 1 or 3. Give examples showing that both possibilities occur.

[If you want a large example: for d = 1141 the element 618715978 + 37751109α is a funda-
mental unit in Z[α].]
√
−19
13. Let R = Z[α] = Z[ 1+ 2 ] be the ring obtained by taking d = −19 in the previous exercise.
a. Show that we have R∗= {±1}.
b. We will see in exercise 36 that R is a unique factorization domain. Using this, show
that the only integral solutions to the equation x2 + 19 = y 3 are (x, y) = (±18, 7).
√
14. Show that the ring R = Z[ 1+ 2−19 ] does not admit a Euclidean function.
[Hint: pick x ∈ R \ {0, ±1} with minimal function value, show that R/xR has order 2 or 3,
and derive a contradiction.]
√
15. Let d ∈ Z6=1 be a squarefree integer and K = Q( d) the corresponding quadratic field. Show
that the subset OK ⊂ K of elements x ∈ K that have irreducible polynomial fQ x ∈ Z[X]

forms a subring of K, and that we have

( √
Z[ d]
√
if d ≡ 2 or 3 mod 4;
OK = 1+ d .
Z[ 2 ] if d ≡ 1 mod 4.
√ ∗ = {±1} unless
16. Show that for K = Q( d) as in the previous exercise and d < 0, we have OK
∗
d ∈ {−1, −3}. What is OK in these two exceptional cases?
17. Show that for every α ∈ Z[i] \ {0}, the norm N (α) is the cardinality of the residue class ring
Z[i]/αZ[i].
18. Show that an integer k > 0 can be written as a sum of two squares if and only if for every
prime p ≡ 3 mod 4, the number of factors p in k is even. Describe the number of solutions
(a, b) ∈ Z2 to k = a2 + b2 .

11
§1: Introduction to number rings

19. Determine which rational numbers can be written as the sum of two rational squares.
20. Let a, b, c be coprime integers satisfying a2 + b2 = c2 . Show that there exist integers m, n ∈ Z
such that we have, after possibly interchanging a and b,

a = m2 − n2 b = 2mn c = ±(m2 + n2 ).

[Hint: First method: as a and b have different parity, c2 = (a + bi)(a − bi) is the product
of coprime integers in Z[i]. Second method: if b is even, (b/2)2 = c−a c+a
2 · 2 is a product of
coprime integers. Third method: exercise 6.]
21. Let a, b ∈ Z be coprime integers satisfying ab = cn for some n ≥ 1. Show that one has
a = ±sn for s = gcd(a, c).
[This shows that it is not necessary to factor c in order to write a and b as n-th powers.]
22. Let p 6= 2 be a prime number.
a. Prove: −2 is a square in F∗p ⇐⇒ p ≡ 1, 3 mod 8.
[Hint: every primitive 8-th root of unity ζ8 ∈ Fp2 satisfies (ζ8 + ζ83 )2 = −2.]
b. Prove:
p ≡ 1, 3 mod 8 ⇐⇒ p = x2 + 2y 2 for x, y ∈ Z.
√
[Hint: Imitate the solution of 1.2 and write p = ππ̄ in the ring Z[ −2].]
23. Let p 6= 3 be a prime number. Prove:

p ≡ 1 mod 3 ⇐⇒ p = x2 + 3y 2 for x, y ∈ Z.
√ √
[Hint: The ring Z[ −3] is not Euclidean, but the ring Z[ −1+2 −3 ] is.]
24. Find all integral solutions to the equation x2 + 4 = y 3 .
25. Find all integral solutions to the equation x2 + 2 = y 3 .
[In 1659, Fermat claimed he could do this exercise and the previous one.]
√
26. Show that the norm function N : Z[ d] → Z for d ∈ Z not a square is never surjective.
Describe the prime factorizations of the elements in the image for d ∈ {−1, −2, −3, −4}.
√
27. Show that the equation x2 + 61 = y 3 has integral solutions. Deduce that Z[ −61] is not a
unique factorization domain.
√
28. Let d < −2 be an integer. Show that Z[ d] is not a unique factorization domain.
29. An integer T is called a triangular number if it is of the form
1
T = 1 + 2 + 3 + . . . + m = m(m + 1)
2
for some m ≥ 1. Check that m-th triangular number is a square for the values

m = 1, 8, 49, 288, 1681, 9800, 57121, 332928.

Prove that there are infinitely many m with this property, and that the values above form a
complete list of such m below 1,000,000.
30. Let q be a rational number. Prove that the following are equivalent:
(i) there are a triangular number T and a non-zero square S with T /S = q;

12
 §1: Introduction to number rings

(ii) there are infinitely many pairs (T, S) as in (i);

(iii) for each positive integer k, there is a pair (T, S) as in (i) for which k divides S;
(iv) the number 2q is not a rational square.
31. Let A be a commutative ring, and let B ⊂ A be a subring for which the abelian group A/B
is finite.
a. Prove that there is an ideal I of A with I ⊂ B for which A/I is finite.
b. Prove that the unit groups of A and B satisfy A∗ ∩ B = B ∗ , and that the abelian group
A∗ /B ∗ is finite.
32. Let K be an algebraic extension of Q, and R ⊂ K a subring.
a. Show that if I is a non-zero ideal of R, then I ∩ Z is a non-zero ideal of Z.
b. Suppose that K is the field of fractions of R. Show that every x ∈ K can be written as
x = r/k, with r ∈ R and k ∈ Z6=0 .
33. Let P denote the set of prime numbers. Show that there is a bijection

{subrings of Q} −→ {subsets of P}
R 7−→ R∗ ∩ P.

34. Let K be a number field of degree n > 1 over Q. Show that there are infinitely many subrings
R ⊂ K for which the additive group of R is free of rank n.
*35. Let n ≥ 2 be an integer. Show that the equation x2 + 1 = y n has no integral solutions with
x 6= 0.
√
*36. Let R = Z[α] with α = 1+ 2−19 be as in exercises 13 and 14, and consider R as a subring
of C. We say that (a, b) ∈ R × (R \ {0}) allows division with remainder in R if there exist
q, r ∈ R with a = qb + r and |r| < |b|.
a. Show that (a, b) allows division with remainder in R if and only if a/b ∈ C lies in the
union U = {x + d : x ∈ Z[α] and |d| < 1} of the open disks of radius 1 in the figure
below.

13
§1: Introduction to number rings

b. Show that the sum of two complex numbers in C \ U lies in U . Deduce that if (a, b)
does not allow division with remainder in R, then (2a, b) and at least one of the pairs
(αa, b) and ((1 − α)a, b) allow division with remainder in R.
c. Show that 2 is coprime to both α and 1 − α in R.
d. Prove that R is a principal ideal domain.

14
2 Ideal arithmetic
We have seen in section 1 that number rings are not in general unique factorization domains.
√ √
More specifically, we saw that in the ring Z[ −19], there are coprime elements 18 + −19
√
and 18 − −19 for which 5 the product is a cube, whereas neither of the elements is up to
√
units equal to the cube of an element in Z[ −19]. This somewhat puzzling phenomenon
was discovered around 1850 by the German mathematician Kummer (1820–1889), who en-
countered analogous phenomena in the cyclotomic number rings that arise naturally in the
study of the Fermat equation xp + y p = z p . The solution offered by Kummer consisted
of the introduction of ideale Zahlen, and was initially regarded as extremely mysterious by
his contemporaries. Only later in the 19-th century, when Kronecker and Dedekind polished
Kummer’s theory and generalized it to arbitrary number fields, his ideal theory found general
acceptance. Abstract ideal theory became common algebraic knowledge after 1930, when
the contents of the algebra courses of Hasse and E. Noether were published as a textbook
by Van der Waerden [26].

I Ideals
Let R be a commutative ring. Then an R-ideal is an additive subgroup of R satisfying
rI ⊂ I for every r ∈ R, or, more concisely, a subset of R that is an R-module. We say that
an ideal I divides an ideal J if I contains J. In particular, ideals ‘divide’ elements in the
sense that I divides the principal ideal xR for all x ∈ I. Kummer originally defined R-ideals
in a more restrictive way, as R-ideals I with the property that II 0 = xR 6= 0 for for some
R-ideal I 0 . Such ideals, which will play an important role in the sequel, are nowadays said
to be invertible.
If I and J are R-ideals, the sum I + J = {i + j : i ∈ I, j ∈ J}, the intersection I ∩ J
and the product
IJ = { ni=1 xi yi : xi ∈ I, yi ∈ J, n ∈ Z≥0 }
P

are again R-ideals. Note that, whereas the sum I + J of the R-ideals I and J consists of
elements i + j with i ∈ I and j ∈ J, the product ideal IJ is the R-ideal generated by the
elements ij. The ideal I + J is the smallest ideal containing both I and J, and we have the
distributive law H(I + J) = HI + HJ.
Ideals I and J are said to be coprime if we have I + J = R. For principal ideals I = Ra
and J = Rb this amounts to saying that a and b are coprime in the sense of section 1: one
has ax + by = 1 for certain x, y ∈ R.
Exercise 1. Show that a, b ∈ R are coprime if and only if there does not exist a ring homomorphism
f : R → F with F a field and a, b ∈ ker f .

If I and J are coprime, then so are I n and J n for every n ≥ 1; one simply observes that
R = (I + J)2n is contained in I n + J n . For coprime ideals I and J, we have IJ = I ∩ J and
the Chinese remainder theorem gives a natural isomorphism R/(IJ) ∼ = (R/I) × (R/J).
Before we develop any ideal arithmetic at all, we show that the fallacious argument in
the previous section becomes correct if we look at ideals rather than elements.

15
§2: Ideal arithmetic

2.1. Theorem. Let R be a ring and I and J coprime ideals of R such that IJ is the n-th
power of some ideal in R. Then I and J are both n-th powers of an ideal in R.

Proof. Suppose that IJ = Z n for some ideal Z. Using the standard multiplication rules for
ideals and the coprimality of the ideals I n−1 and J, we obtain

(I + Z)n = I n + I n−1 Z + · · · + IZ n−1 + IJ = I(I n−1 + · · · + Z n−1 + J) = IR = I.

√
2.2. Example. We have seen that in the ring R = Z[ −19], the product of the coprime
√ √
ideals I = (18 + −19) and J = (18 − −19) equals the cube of the principal ideal (7). It
follows that we have √ √
(18 + −19) = (18 + −19, 7)3 .
√
Our computation in 1.4 implies that the ideal P = (18 + −19, 7) is a non-principal ideal,
even though its cube P 3 is principal.
√ √
Exercise 2. Show that P = (18 + −19, 7) is a maximal ideal of index 7 in Z[ −19].

Ideals can be added and multiplied in any ring R. For a domain R, unrestricted division by
non-zero elements can be performed inside the field of fractions K = Q(R) of R. In order
to divide the non-zero ideals of a domain R, one is naturally led to the following extension
of the ideal concept.

2.3. Definition. Let R be a domain with field of fractions K. Then a fractional R-ideal I
is a non-zero R-submodule of K such that xI is contained in R for some x ∈ K ∗ .

Saying that I is an R-submodule of K means that I is an additive subgroup of K that is

mapped into itself under multiplication by elements of R. The element x in the definition
can be chosen to lie in R, and if we have xI ⊂ R then xI is an R-ideal. We say that I is
integral if it is contained in R. Every fractional R-ideal contains non-zero elements of R.
For every finite subset {x1 , x2 , . . . , xn } of K ∗ , the R-submodule Rx1 + Rx2 + . . . + Rxn of K
is a fractional R-ideal. A principal fractional ideal is a fractional ideal of the form Rx with
x ∈ K ∗ . If R is a principal ideal domain, then every fractional ideal is of this form. The
definition of fractional ideals shows that we can divide them in an obvious way by elements
of K ∗ . For arbitrary fractional ideals I and J we define the ideal quotient as

I : J = {x ∈ K : xJ ⊂ I}.

It is an immediate consequence of the definition that this is an R-submodule of K. If we

choose a ∈ I and b ∈ K ∗ such that bJ ⊂ R, then we have abJ ⊂ I, so I : J 6= 0. If we take
c, d ∈ K ∗ such that cI ⊂ R and d ∈ J, it follows that cd(I : J) ⊂ R. We have proved the
only non-obvious case of the following proposition.

2.4. Proposition. If I and J are fractional R-ideals, then so are the sum I + J, the product
IJ, the intersection I ∩ J and the quotient I : J. 

If we take R equal to Z, then all fractional ideals are of the form I = (q) = qZ with q ∈ Q∗ .
The product and the quotient of such ideals are simply given by (q1 )(q2 ) = (q1 q2 ) and

16
 §2: Ideal arithmetic

(q1 ) : (q2 ) = (q1 /q2 ). In this case, we see that the fractional ideals form a group isomorphic
to Q∗ /Z∗ = Q∗ /{±1}. For the sum and intersection of fractional ideals in Z (or an arbitrary
principal ideal domain) see exercise 11.
Two ideal quotients coming with a fractional R-ideal I for which we introduce a special
notation are the “inverse ideal”

I −1 = R : I = {x ∈ K : xI ⊂ R}

of I and the multiplier ring

r(I) = I : I = {x ∈ K : xI ⊂ I}

of I. Note that r(I) is indeed a subring of K, and that it contains R. If it is equal to R,

than I is said to be proper for the ring R. This is not to be confused with the word ‘proper’
to indicate that an ideal I ⊂ R is a proper subset of R, i.e., not equal to the full ring R.

I Invertible ideals
A fractional R-ideal I is said to be invertible if one has IJ = R for some fractional R-
ideal J. If such a J exists, it is contained in I −1 , and actually equal to I −1 , as we also have
I −1 = I −1 · IJ ⊂ RJ = J. Thus the invertible ideals are exactly those I for which we have
II −1 = R. As I −1 is again a fractional R-ideal, we find that an ideal I ⊂ R is invertible
if and only if there exists an ideal J ⊂ R such that IJ is a non-zero principal ideal. This
means that such I occur as the “divisors of principal ideals”, i.e., ideals in Kummer’s original
sense. Invertible ideals are not always principal, but something weaker is true.
2.5. Lemma. Any invertible R-ideal is finitely generated.
Proof. The equation IJ = R implies that there exist xi ∈ I and yi ∈ J such that
Pn Pn
i=1 xi yi = 1. Now any x ∈ I Pcan be written in the form x = i=1 (xyi )xi . As we
n
have xyi ∈ IJ = R, we obtain I = i=1 Rxi , and we are done.
The set I(R) of invertible fractional R-ideals forms a group under ideal multiplication.
Principal fractional ideals, which are obviously invertible, form a subgroup P(R) ⊂ I(R)
that is isomorphic to K ∗ /R∗ . The obstruction group measuring to which extent invertible
R-ideals are principal is the Picard group of R.
2.6. Definition. The Picard group of a domain R is defined as Pic(R) = I(R)/P(R).
In other words, the Picard group of a domain R is defined by the long exact sequence
f
0 −→ R∗ −→ K ∗ −→ I(R) −→ Pic(R) −→ 0,

where f (x) = xR for x ∈ K ∗ . If R is a principal ideal domain, then we have I(R) = f [K ∗ ] =

P(R) and Pic(R) = 0. As the condition that an ideal be invertible is rather restrictive, it
is not generally true that a domain with trivial Picard group is a principal ideal domain
(exercise 15).
√
The ring R = Z[ −19] is an example of a ring with non-trivial Picard group. As we
√
saw in example 2.2, the class of the ideal (18 + −19, 7) is an element of order 3 in Pic(R).

17
§2: Ideal arithmetic
√
This explains why our method to solve 1.4 failed: 18 + −19 generates the cube of an ideal,
but it is not the cube of an element. We will show later that in this case, Pic(R) is actually
of order 3.
The invertible ideals of a domain R can be neatly characterized as those finitely gener-
ated R-ideals that are locally principal, making the Picard group into an obstruction group
to a ‘local-global-principle’. The precise statement is the following.

2.7. Theorem. Let R be a domain and I a fractional R-ideal. Then I is invertible if and
only if the following two conditions hold:
(1) I is finitely generated;
(2) the localization Im at each maximal ideal m of R is a principal fractional Rm -ideal.

I Localization
Before proving theorem 2.7, we spend some time on the important concept of localization
from commutative algebra that occurs in condition (2). As we are working with domains R,
we can localize inside the field of fractions Q(R) of R. We will therefore avoid discussing
the localization concept for arbitrary commutative rings or the geometric explanation of its
‘local nature’. These are found in [9, section 3] or [15, chapter 2].
Let R be a domain and S a multiplicative subset of R, i.e., a subset of R \ {0} that
contains 1 and is closed under multiplication. Then we can define the localized ring

S −1 R = { rs ∈ K : r ∈ R, s ∈ S}.

This is a subring of K = Q(R) that contains R. There is the localization K = Q(R)

corresponding to S = R \ {0}. More generally, by taking S = R \ p, we have localizations

Rp = { rs ∈ K : r ∈ R, s ∈
/ p}

at the prime ideals p of R. The rings Rp are local rings in the sense that they have a unique
maximal ideal
pRp = { rs ∈ K : r ∈ p, s ∈/ p}
consisting of the complement of the unit group Rp∗ = { rs ∈ K : r, s 6∈ p} consisting of all
x ∈ K ∗ that are ‘invertible modulo p’. Conversely, a local domain R coincides with the
localization Rm at its maximal ideal m.
When dealing with the localizations of a domain R having more than one maximal
ideal, one often refers to R as the global ring. Localization enables one to pass from R to a
‘less complicated’ ring S −1 R without losing information on the ideals outside S. Informally,
one can say that localizing at S maps the ideals disjoint from S to the ideals of the localized
ring, and all ideals that meet S to the unit ideal of the localized ring.

2.8. Proposition. Let R be a domain and S ⊂ R \ {0} a multiplicative subset. Then for
every R-ideal I, the image of I under the localization map R → S −1 R generates an ideal

S −1 I = { si : i ∈ I, s ∈ S} ⊂ S −1 R.

We have S −1 I 6= S −1 R if and only if I ∩ S = ∅, and all ideals of S −1 R are of the form S −1 I.

18
 §2: Ideal arithmetic

Proof. It is clear that S −1 I is an ideal of S −1 R that contains 1 if and only if there exists
an element s ∈ I ∩ S. For every ideal J ⊂ S −1 R, the contraction R ∩ J is an R-ideal with
localization J.
Exercise 3. Show that the map I 7→ S −1 I induces a bijection between the prime ideals of R that do not
meet S and the prime ideals of S −1 R, and that the local rings at corresponding primes are isomorphic.

If I is a fractional R-ideal and S ⊂ R \ {0} a multiplicative subset, then

S −1 I = { si : i ∈ I, s ∈ S} ⊂ K

is a fractional S −1 R-ideal. In the case S = R \ p we denote this fractional Rp -ideal by Ip .

An ideal I ∈ I(R) is determined by its localizations Ip , as follows.
2.9. Lemma. Let R be a domain and I a fractional R-ideal. Then we have
T
I = m Im ,

where the intersection is taken over the localizations at all maximal ideals m of R.
T T
Proof. It is clear that I is contained in the intersection m Im . Given an element x ∈ m Im ,
we consider J = {r ∈ R : rx ∈ I} ⊂ R. This is an ideal of R, and since we can write x = ab
with a ∈ I and b ∈ R \ m for each maximal ideal m, the ideal J ⊂ R contains, for each m,
an element b 6∈ m. It follows that J is not contained in any maximal ideal of R, so we have
J = R and x = x · 1 ∈ I.
Proof of 2.7. Let I be an invertible R-ideal. Then I is finitely generated by 2.5, and there
exist xi ∈ I and yi ∈ I −1 such that ni=1 xi yi = 1. Let p ⊂ R be any prime. All terms xi yi
P

are in R ⊂ Rp , and they cannot all be in the maximal ideal of Rp . Suppose that we have
x1 y1 ∈ Rp∗ = Rp \ pRp . Then any x ∈ I can be written as

x = x1 · xy1 · (x1 y1 )−1 ∈ x1 R · R · Rp = x1 Rp .

It follows that Ip is principal with generator x1 .

Conversely, let I = Rx1 + Rx2 + . . . + Rxn be a finitely generated R-ideal with the
property that x ∈ I generates the localization Ip = xRp at a prime p. Then we can write
xi = x(ri /s) ∈ Rp , with ri ∈ R and s ∈ R \ p independent of i. From sx−1 · xi = ri ∈ R we
obtain sx−1 ∈ I −1 , so I · I −1 contains x · sx−1 = s ∈ R \ p.
If I is locally principal at all maximal ideals of R, then I · I −1 is not contained in any
maximal ideal of R. This means that we have I · I −1 = R, and that I is invertible.
Exercise 4. Show that we may replace ‘maximal ideal’ by ‘prime ideal’ in the second condition of 2.7.
√
2.10. Example. We saw in 2.2 that the principal ideal (18 + −19) is the cube of the
√ √
maximal ideal I = (7, 3 − −19) in the ring R = Z[ −19]. The ideal I is not principal,
but as it is invertible our theorem implies that all its localizations Ip are. For primes p not
containing I, we trivially find Ip = Rp as I contains elements outside p. The only prime
√
p ⊃ I is p = I, and here we have Ip = 7Rp = (3 − −19)Rp : the quotient
√
7 3 + −19
√ =
3 − −19 4

19
§2: Ideal arithmetic
√
is a unit in Rp since 3 + −19 and 4 are in R \ p ⊂ Rp∗ .
√
The ideal J = (2, 1 + −19) of index 2 is also not principal, but here the identity
√
J 2 = (4, 2 + 2 −19) = 2J shows that J is not invertible: if it were, the identity could be
multiplied by J −1 to prove that J is principal, and generated by 2. At the only prime p = J
containing J, the ideal Jp needs both generators.

I Ideals in number rings

Suppose now that R is a number ring. Then theorem 2.7 can be simplified a bit as number
rings are noetherian: all their ideals are finitely generated. Even more is true.
2.11. Theorem. Every non-zero ideal in a number ring is of finite index.
Proof. Let K = Q(R) be a number field of degree t ≥ 1 over Q, and suppose we are given a
non-zero ideal I ⊂ R. Pick any non-zero element x ∈ I. As x is algebraic over Q, it satisfies
an equation an xn + an−1 xn−1 + . . . + a1 x + a0 = 0 with ai ∈ Z, and we have a0 6= 0 as x is
non-zero. This shows that I ⊃ (x) contains the positive integer a = |a0 |.
Clearly R/aR maps surjectively to R/I, so it is sufficient to show that R/aR is a finite
ring. To show this, let M ⊂ R be any finitely generated subgroup of the additive group of R.
As M has no elements of finite order, M is a free abelian group. Any set of more than t
elements in R ⊂ K is linearly dependent over Q, so we see that M has to be of finite rank
k ≤ t. The natural map M → R/aR factors via the group M/aM , which is finite of order
ak ≤ at . This implies that every finitely generated subgroup of R/aR has order at most at ,
so R/aR itself is finite of order at most at .
2.12. Corollary. A number ring is noetherian, and all of its non-zero prime ideals are
maximal.
Proof. If I ⊂ R is a non-zero ideal, then 2.11 shows that every non-zero principal ideal
(x) ⊂ I is of finite index in I. It follows that I itself is finitely generated, so R is noetherian.
The second statement is immediate from 2.11 as every finite domain R/p is a field.
The proof of 2.11 shows that for R a number ring and n ≥ 1 any integer, the additive group
of R/nR can be generated by t = [Q(R) : Q] elements. This does not imply that the additive
group of R itself is finitely generated: the number ring R = Z[ 12 ] ⊂ Q consisting of rational
fractions with 2-power denominator is not finitely generated as an abelian group, but R/I
is finite cyclic for all I 6= 0.
Exercise 5. What is the cardinality of R/nR for R = Z[ 12 ] and n ∈ Z>0 ?

A number ring for which the additive group is finitely generated is called an order in its
field of fractions. As a number ring has no additive torsion elements, every order is in fact
free of finite rank over Z. The rank of an order R in K = Q(R) is bounded by the degree
t = [K : Q], and as we have R ⊗Z Q = K it has to be equal to t. This implies that we have

R = Z · ω1 ⊕ Z · ω2 ⊕ . . . ⊕ Z · ωt

for some Q-basis {ω1 , ω2 , . . . , ωt } of K. For every monic irreducible polynomial f ∈ Z[X]
of degree n, the associated order Z[α] = Z[X]/(f ) obtained by adjoining a root α of f has

20
 §2: Ideal arithmetic

rank n. All number rings in the introduction are orders of the form Z[α]. If an order R is of
the form R = Z[α] for some α ∈ R, then R is said to be monogenic.
A domain R which is not a field and in which every non-zero prime ideal of R is
maximal is said to be one-dimensional. More generally, one defines the Krull-dimension of
a commutative ring R as the supremum of the lengths of all chains of prime ideals in R.
Here a chain of prime ideals in R of length n is a strictly increasing sequence of prime ideals
p0 ( p1 ( . . . ( pn .
Exercise 6. Given an example of a domain R having infinite Krull dimension.

Fields and finite rings are zero-dimensional: all their prime ideals are maximal. By 2.12,
number rings that are not number fields are one-dimensional. Their non-zero prime ideals,
which are all maximal, are usually referred to as the primes of the number ring.

I Primary decomposition
It is not in general true that the invertible ideals of a number ring can uniquely be written
as powers of prime ideals. The study of invertible ideals I of a number ring R can however
be reduced to the study of its localizations Ip at the primes p, which are principal and only
differ from Rp for finitely many p.

2.13. Lemma. Let Rp be a local number ring. Then every non-zero ideal of Rp contains
some power of its maximal ideal.

Proof. As Rp is noetherian, we can apply noetherian induction: if there are counterexamples

to the lemma, the set of such ideals has a maximal element I (cf. exercise 26). Then I is
not prime, as the maximal ideal p is the only non-zero prime ideal of Rp . Let x, y ∈ R \ I
satisfy xy ∈ I. Then I + (x) and I + (y) strictly contain I, so they do satisfy the conclusion
of the lemma and contain a power of p. The same then holds for (I + (x))(I + (y)) ⊂ I:
contradiction.

2.14. Theorem. Let R be a number ring. Then we have an isomorphism

∼
M
φ : I(R) −→ P(Rp ).
p prime

that maps an invertible ideal I to its vector of localizations (Ip )p at the primes of R.

Proof. The map φ is well-defined by 2.7 and the fact that, by 2.11, every non-zero ideal
I ⊂ R is contained in only finitely many primes of R. It is a homomorphism by the basic
property S −1 I · S −1 J = S −1 (IJ) of localizations, and it is injective by 2.9.
In order to show φ is surjective it suffices to construct, for p a prime of R and Jp ∈ P(Rp )
an integral Rp -ideal, some global ideal I that has localizations Iq = Rq at q 6= p and Ip = Jp .
T
There is no choice for I by 2.9, so we let I = Jp ∩ q6=p Rq = Jp ∩ R. It is clear that I is
an integral R-ideal with localization Jp at p. By 2.13, the ideal Ip contains a power pn Rp of
the maximal ideal of Rp . It follows that we have pn ⊂ I. If q is a prime different from p, we
cannot have I ⊂ q as this would imply that q contains pn and therefore p. We find that we
have Iq = Rq for such q, as was to be shown.

21
§2: Ideal arithmetic

Theorem 2.14 states that for a number ring R, giving an invertible R-ideal is ‘the same’ as
giving a principal Rp -ideal Ip for each prime p of R, provided that we take Ip = Rp for almost
all p. Note that I is an integral R-ideal if and only if all Ip are integral Rp -ideals, and that
the isomorphism φ ‘preserves’ inclusions in an obvious way.

2.15. Example. For R = Z, the theorem reduces to unique factorization for rational num-
bers. In this case, the local ring Z(p) = { ab ∈ Q : p - b} at a prime p has unit group
Z∗(p) = { ab ∈ Q : p - a and p - b}. Any non-zero rational number can uniquely be writ-
ten as x = pordp (x) · u with ordp (x) ∈ Z and u ∈ Z∗(p) , so a principal fractional Z(p) -ideal
is of the form pk Z(p) for some unique k ∈ Z. This yields P(Z(p) ) ∼ = Z, and as we have
∗ ∗ ∗
I(Z) = Q /Z = Q /{±1}, theorem 2.14 yields an isomorphism

Q∗ /{±1} −→
∼ L
p prime Z with x 7−→ (ordp (x))p .

Looking at the definition of the function ordp , this can be stated as: every non-zero rational
number can uniquely be written as a product ± p pnp , where np = 0 for almost all p.
Q

There are more classical formulations of theorem 2.14 that work with integral ideals only,
and replace the localization Ip of an integral invertible R-ideal at a prime p by the p-primary
part
I(p) = Ip ∩ R
of I. This is an integral R-ideal with localization Ip , and we have I(p) = R if p does not
divide I. At the primes p dividing I, we have pn ⊂ I(p) ⊂ p for some n ≥ 0 as Ip contains
some power of the maximal ideal in Rp by 2.13. In particular, no prime q 6= p divides the
p-primary part of I, and the primary parts of I at distinct primes are coprime. As I is the
intersection of its localizations, we have
\ \ \
I =R∩ Ip = I(p) = I(p) .
p p p⊃I

The coprimality of the various p-primary parts shows that the last intersection can be written
as a product, where the convention is that empty ideal products and empty ideal intersections
are taken equal to R. This yields the following.

2.16. Theorem. Let R be a number ring. Then every integral ideal I ∈ I(R) has a primary
Q
decomposition I = p⊃I I(p) . 

The study of p-primary ideals in R and ideals in the local ring Rp is essentially the same, as
we have a natural isomorphism
∼
R/I(p) −→ Rp /Ip .
Indeed, injectivity of the natural map is clear from the definition of I(p) . For the surjectivity,
one needs to show that every s ∈ R \ p is a unit in R/I(p) . To see this, one observes that
by the maximality of p, there exists an element s0 ∈ R \ p such that ss0 − 1 is in p. As I(p)
contains pn for some n, the element ss0 − 1 is nilpotent in R/I(p) . It follows that ss0 and
therefore s are in (R/I(p) )∗ .

22
 §2: Ideal arithmetic

I Local number rings

Whether one prefers 2.14 or the integral version 2.16, either of these theorems reduces the
study of invertible R-ideals to the study of principal ideals in local number rings. Unique
prime ideal factorization for the number ring R is obtained exactly when all primes p of R
are invertible.
2.17. Theorem. Let p be a prime of a number ring R. Then the following are equivalent:
(1) p is an invertible R-ideal;
(2) Rp is a principal ideal domain, and every Rp -ideal is a power of pRp ;
(3) There exists π ∈ Rp such that every x ∈ K ∗ can uniquely be written as x = u · π k with
u ∈ Rp∗ and k ∈ Z.
Proof. For (1) ⇒ (2), we use 2.7.(2) to write pRp = πRp and observe that all inclusions in
the chain of principal Rp -ideals

Rp ⊃ pRp = (π) ⊃ (π 2 ) ⊃ (π 3 ) ⊃ . . .

are strict: an equality (π n ) = (π n+1 ) would imply π n = rπ n+1 for some r ∈ Rp , whence
rπ = 1 and π ∈ Rp∗ . Let now I 6= 0 be an integral Rp -ideal. As I contains all sufficiently
large powers of (π) by 2.13, there is a largest value n ≥ 0 for which we have (π n ) ⊃ I. Take
any r ∈ I \ (π n+1 ), then we have r = aπ n with a 6∈ (π). This implies that a is a unit in Rp ,
so we have (r) = (π n ) ⊂ I ⊂ (π n ) and I = (π n ). We conclude that Rp is a principal ideal
domain, and that every Rp -ideal is a power of (π).
For (2) ⇒ (3), we take for π a generator of pRp . For every x ∈ R we have xRp = π k Rp
for some uniquely determined integer k ≥ 0, and x = u · π k with u ∈ Rp∗ . Taking quotients,
this yields (3).
For (3) ⇒ (1), we note that we have π ∈ / Rp∗ and therefore

Rp = {u · π k : u ∈ Rp∗ and k ≥ 0}.

This shows that Rp is a local ring with principal maximal ideal (π), so by 2.7 we see that p
is invertible.
A local ring Rp satisfying the equivalent conditions (2) and (3) of 2.17 is called a discrete
valuation ring. The terminology is explained by the existence of the homomorphism v :
K ∗ → Z in (3) that sends x to ordp (x). As the notation suggests, this homomorphism does
not depend on the choice of π, and it satisfies

(∗) v(x + y) ≥ min(v(x), v(y))

for any two elements x, y ∈ K ∗ with x + y 6= 0. A non-zero homomorphism v : K ∗ → Z

satisfying (∗) is called a discrete valuation. One usually extends v to K by setting v(0) =
+∞. With this convention, inequality (∗) holds unrestrictedly for x, y ∈ K. Every discrete
valuation v on K gives rise to a discrete valuation ring Rv ⊂ K with maximal ideal mv given
by
Rv = {x ∈ K : v(x) ≥ 0} and mv = {x ∈ K : v(x) > 0}.
∼
We have a canonical isomorphism I(Rv ) = P(Rv ) −→ Z that maps mv to 1.

23
§2: Ideal arithmetic

Exercise 7. Show that for x, y ∈ K with valuations v(x) 6= v(y), one has v(x + y) = min(v(x), v(y)).

A number ring R that is not a number field, or, more generally, a one-dimensional noetherian
domain R is called a Dedekind domain if for every prime p of R, the local ring Rp is a discrete
valuation ring. In number rings R that are Dedekind, every non-zero ideal is invertible, and
we obtain unique prime ideal factorization.

2.18. Theorem. Let R be a number ring that is Dedekind. Then there is an isomorphism
∼
M
I(R) −→ Z
p

I 7−→ (ordp (I))p ,

pordp (I) .
Q
and every I ∈ I(R) factors uniquely as a product I = p 

If R is Dedekind, Pic(R) is often called the class group of R and denoted by Cl(R).
It may seem hard to check in practice whether a number ring is Dedekind, since this
imposes a condition at each prime of R. In the next section, we will see how to do this for
orders of the form R = Z[α]. More generally, we will see that every number ring R is a
subring of finite index of its normalization Re ⊃ R. The ring R e is a Dedekind domain, and
Rp is a discrete valuation ring at all primes p of R that do not divide the index [R
e : R].

Exercises.

8. Let R be a principal ideal domain different from a field, and x, y ∈ R. Prove that x and
y are coprime if and only if there does not exist a prime element of R that divides both
x and y. Show that this statement is incorrect if “principal ideal domain” is replaced by
“unique factorization domain”.
9. Let R be a domain. Call an ideal I ⊂ R invertible if there exists an ideal J ⊂ R such that
IJ is a non-zero principal R-ideal. Show that the relation

I ∼ I0 ⇐⇒ there exist non-zero elements x, y ∈ R such that xI = yI 0

is an equivalence relation on the set D of invertible R-ideals, and that the set of equivalence
classes can be identified with Pic(R).
[This shows that Pic(R) can be defined in terms of ‘ordinary’ R-ideals.]
10. Let R be a domain. Prove that the ideal quotient of fractional R-ideals satisfies the following
properties:
T T P T
H : (I · J) = (H : I) : J ( k Ik ) : J = k (Ik : J) I : ( k Jk ) = k (I : Jk ).

11. Let R be a principal ideal domain with field of fractions K, and let a, b ∈ K ∗ . Show that we
can write a = u p pn(p) and b = v p pm(p) , where u, v are units of R, the elements p range
Q Q

over a finite set of pairwise non-associate prime elements of R, and n(p), m(p) are integers.
Prove that Ra ∩ Rb = Rc, where c = p pmax{n(p),m(p)} , and that Ra + Rb = Rd, where
Q
Q min{n(p),m(p)}
d = pp . Are these statements correct for unique factorization domains?

24
 §2: Ideal arithmetic

12. Let R be a domain and I an invertible R-ideal. Show that I is proper, i.e. it has multiplier
ring r(I) = R. Deduce that an additive subgroup I ⊂ K = Q(R) is an invertible ideal for at
most one subring of K.
√ √
13. Consider the integral ideal I = (2, 1 + −19) ⊂ R = Z[ −19]. Show that I is a maximal
R-ideal with multiplier ring r(I) 6= R, and that it satisfies I 2 = 2I. Conclude that I is not
an invertible R-ideal, and that 2R is not a product of prime ideals in R.
14. Let R be a local domain. Show that a fractional R-ideal is invertible if and only if it is
principal. Deduce that Pic(R) is trivial.
√ √
15. Denote by −3 the complex number i 3.
√
a. Prove that for every x ∈ C there exists r ∈ Z[ −3] with |x − r| ≤ 1, and determine for
which x the equality sign is needed.
√ √
b. Prove that every fractional Z[ −3]-ideal is either of the form Z[ −3]a or of the form
√ √
Z[(1 + −3)/2]a, with a ∈ Q( −3).
√
c. Prove that Z[ −3] is not a principal ideal domain, and that its Picard group is trivial.
√
d. Prove that Pic Z[(1 + −3)/2] is trivial.
16. Let R be a noetherian domain. Show that every non-zero element of R \ R∗ can be written
as a product of irreducible elements.
√ √ √
17. Show that (2, 3 + −61) and (5, 3 + −61) are invertible ideals in Z[ −61], and determine
√
the order of their classes in Pic(Z[ −61]).
√
18. Let it be given that Pic(Z[ −19]) is a finite group of order 3. Use this to find all integral
solutions of the equation x2 + 19 = y 5 .
19. Let τ be a zero of an irreducible polynomial
√ aX 2 + bX + c ∈ Z[X]. Show that R = Z[aτ ] is
an order in the quadratic field Q( b2 − 4ac), and that I = Z + Zτ is an√invertible R-ideal.
[Hint: compute I · σ(I), where σ is the non-trivial automorphism of Q( b2 − 4ac).]
20. Let f ∈ Z[X] be a monic irreducible polynomial of degree 3, and A = Z[α] the cubic order
generated by a root α of f .
a. For p a prime number, show that R = Z + pA is a subring of A.
b. Prove that I = R + Rα is a fractional R-ideal with multiplier ring r(I) = R.
c. Show that we have r(I 2 ) = A, and deduce that the R-ideal I is proper but not invertible.
21. Let R be a number ring with field of fractions K. Show that for every prime ideal I 6= 0 of R,
the order of R/I equals pk for a prime number p and an integer k ≤ [K : Q].
22. Let B > 0 be a real number. Show that a number ring has only finitely many ideals of index
at most B.
Q
23. Let R be a ring and p a prime ideal that contains a finite product of ideals i ai . Show that
p contains some ai . Do you recognize the statement for R = Z?
S
24. Let R be a ring and a an ideal that is contained in a finite union of prime ideals i pi . Show
that a is contained in some pi .
[This is called prime avoidance: if a contains an element outside pi for i = 1, 2, . . . , n, then it
contains an element that does not lie in any pi .]
25. Let R be a number ring and I ⊂ R an invertible R-ideal. Show that I is a product of prime
ideals if and only if all primes p ⊃ I are invertible.

25
§2: Ideal arithmetic

26. Show that the following properties of a ring R are equivalent.

(1) R is noetherian;
(2) every ascending chain of ideals I1 ⊂ I2 ⊂ I3 ⊂ . . . in R stabilizes;
(3) every non-empty collection of ideals of R has a maximal element with respect to inclu-
sion.
27. Let R be a noetherian ring and I ( R an R-ideal. Show that there exist prime ideals
p1 , p2 , . . . , pn of R that contain I and satisfy ni=1 pi ⊂ I. Show also that the ideals pi can
Q

be chosen such that no strict inclusions between them occur, and that with this choice, the
set {p1 , p2 , . . . , pn } is uniquely determined by I and consists of the minimal elements in the
collection of prime ideals containing I.
[Hint: noetherian induction.]
28. A ring is said to be reduced if its nilradical N = {x ∈ R : xn = 0 for some n > 0} is the
zero-ideal. Show that a reduced noetherian ring can be embedded as a subring in a finite
product of noetherian domains.
[Hint: apply the previous exercise to I = 0.]
*29. Let R be a domain, and suppose that every prime ideal of R is principal. Prove that R is a
principal ideal domain.
[Hint: show that the collection of non-principal ideals in any ring has maximal elements if it
is non-empty, and that these are prime ideals.]
30. Let S be a multiplicatively closed subset of a domain A. Show that a ring homomorphism
φ : A → B factors via S −1 A if and only if φ[S] is contained in B ∗ .
[This is the characterizing universal property of localizations.]
31. Let R be a number ring. Prove that the localization of fractional R-ideals satisfies the
following properties:

Ip + Jp = (I + J)p Ip Jp = (IJ)p Ip ∩ Jp = (I ∩ J)p .

32. Let R be a domain with fractional ideals I and J, and suppose that J is finitely generated.
Show that we have
Ip : Jp = (I : J)p
for every prime ideal p of R. Deduce that a finitely generated R-ideal is invertible if and only
if it is locally everywhere invertible.
[This shows that for noetherian domains R, invertibility of R-ideals is a local property.]
33. Define M ⊂ Q by M = { ab ∈ Q : b is squarefree}. Show that M is a sub-Z-module of
Q, and that for every prime p of Z, the localization Sp−1 M is a principal Sp−1 Z-ideal, with
Sp = Z \ pZ. Is M an invertible Z-ideal? If not, is there a subring R ⊂ Q for which M is an
invertible R-ideal?
In a similar way, determine for q a prime number and Mq = { ab ∈ Q : b is a power of q} ⊂ Q
whether each localization Sp−1 Mq is a principal Sp−1 Z-ideal. Is there a subring R ⊂ Q such
that Mq is an invertible R-ideal?
34. Show that S = 1 + pZ is a multiplicative subset of Z, and that S −1 Z is the local ring Z(p)
occurring in 2.15. Thus different S can yield the same localization. *Can you describe when
two multiplicative subsets of a domain give rise to the same localization?
[Hint: [9, exercise 3.7]]

26
 §2: Ideal arithmetic
√
35. Find for each prime p in R = Z[ −61] a local generator at p of the non-principal R-ideals
√ √
(2, 3 + −61) and (5, 3 + −61).
√ √
36. For R = Z[ −3] and x = (1 + −3)/2, show that I = Rx is a fractional R-ideal that is
non-integral and satisfies I 3 = R. Deduce that I(R) has a non-trivial torsion subgroup, and
that R is not a Dedekind domain. Find a non-invertible R-ideal.
37. Show that the polynomial ring C[X] is a Dedekind domain and identify its primes. Make
theorem 2.14 as explicit for this ring as we did for the ring Z. Same questions for the poly-
nomial ring K[X] over an arbitrary field K.
38. Show that every localization S −1 R of a Dedekind domain R different from Q(R) is again a
Dedekind domain. Is the same true for quotient rings R/I with I 6= R?
39. Show that R is a principal ideal domain if and only if R is a Dedekind domain with Pic(R) = 0.
40. Let R be a discrete valuation ring for which the residue class field is finite. Prove that the
function N : I 7→ [R : I] is a function on the set of non-zero R-ideals with values in Z>0 that
satisfies N (I · J) = N (I)N (J).
41. Let R be a number ring that is Dedekind. Show that the norm function N : I 7→ [R : I] on
R-ideals extends to a homomorphism N : I(R) → Q∗ . Is this also true for arbitrary number
rings?
42. (Semi-local rings.) A semi-local ring is by definition a ring with only finitely many maximal
ideals. Show that for p1 , p2 , . . . , pn prime ideals of a ring R, the set S = R \ ni=1 pi is
S

multiplicatively closed and the localization S −1 R is a semi-local ring. Show also that for R a
number ring and I 6= 0 any ideal, the quotient ring R/I is semi-local.
[Hint: exercise 23.]
43. Show that Pic(R) is trivial if R is a semi-local domain.
[Hint: By the Chinese remainder theorem there exist em ∈ R such that em ≡ 1 mod m and
P
em ∈ n if m 6= n. Choose xm ∈ I \ mI for each maximal m. Now show that m xm em generates
I.]
44. Let R be a finite ring. Show that the unit group R∗ has order
Y 1
#R∗ = #R ·


1− .
p
#(R/p)

Here p ranges over the prime ideals of R.

45. (Approximation theorem.) Let X be a finite set of primes of a Dedekind domain R, and
suppose we are given an integer np for each p ∈ X. Show there exists x ∈ K = Q(R) such
that
ordp (x) = np if p ∈ X
ordp (x) ≥ 0 if p 6∈ X.
[Hint: use the Chinese remainder theorem if all np are positive. For the general case, apply
this special case twice.]
46. Show that every ideal in a Dedekind domain can be generated by 2 elements.
47. Let X be a finite set of primes of a Dedekind domain R. Show that Cl(R) is generated by
the classes of the primes of R that are not in X. (In particular, no generators are needed if
R is semi-local.)

27
§2: Ideal arithmetic

48. An ideal I 6= R of a ring R is called primary if it satisfies the implication

xy ∈ I ⇒ x ∈ I or y n ∈ I for some n ≥ 0.

Show that the radical of a primary ideal is a prime ideal, and that the p-primary part I(p) of
an ideal I in 2.16 is a primary ideal with radical p.
49. (Valuation rings.) A subring R of a field K is said to be a valuation ring of K if for each
x ∈ K ∗ , we have either x ∈ R or x−1 ∈ R (or both). Show that every valuation ring is a local
ring, and that a discrete valuation ring is indeed a valuation ring.
50. Let C(X) be the field of rational functions with complex coefficients. For α ∈ C, define
ordα (f ) for a non-zero polynomial f ∈ C[X] as the order of the zero of f at α and set
ord∞ (f ) = − deg(f ). Show that these maps extend to normalized valuations on C(X), and
T T
determine the corresponding valuation rings Rα and R∞ . Describe α∈C Rα and ( α∈C Rα )∩
R∞ ?
51. Show that every normalized valuation v on C(X) for which v[C∗ ] = 0 is equal to one of
the valuations from the previous exercise. Deduce that every f ∈ C(X)∗ satisfies the sum
P
formula v v(f ) = 0, where the sum is taken over all v.
[Hint: if v(X) ≥ 0 one has C[X] ⊂ Rv and mv ∩ C[X] is a prime ideal of C[X]. Otherwise
look at 1/X.]
52. Show that every normalized valuation v on Q is of the form v(x) = vp (x) = ordp (x) for some
prime number p.
53. Let v be a discrete valuation on a field K and Rv the corresponding discrete valuation ring.
Show that the v-adic metric
d(x, y) = 2−v(x−y)
on K × K defines a distance function on K, and that the induced v-adic topology makes K
into a totally disconnected Hausdorff space. Is Rv closed in K?
54. Let K and v be as in the previous exercise. Show that with respect to the v-adic topology,
addition and multiplication yield continuous maps K × K → K, and x 7→ x−1 is a continuous
map on K ∗ . Use this to construct a v-adic completion Kv of K and show that Kv is a
complete topological field, i.e. a field that is a complete metric space and in which all field
operations are continuous.
[Hint: as when constructing R from Q, let Kv be the set of equivalence classes of Cauchy
sequences in K with respect to the v-adic metric.]
55. Describe the completions of C(X) with respect to the valuations ordα and ord∞ .
[Hint: one obtains Laurent series in X − α and X −1 .]
56. (Approximation theorem revisited.) Let X be a finite set of primes of a Dedekind domain R
with field of fractions K, and suppose that we are given an integer np and an element xp ∈ K ∗
for each p ∈ X. Show that there exists x ∈ K ∗ such that
ordp (x − xp ) = np if p ∈ X;
ordp (x) ≥ 0 if p 6∈ X.
Q
Deduce that the image of K in p∈X K under the diagonal embedding is dense if the topology
on the product is obtained by giving the p-th component the topology coming from the p-adic
valuation.

28
 §2: Ideal arithmetic

*57. (p-adic numbers.) If the completion process in the previous exercises is carried through for
the field Q and the p-adic valuation v = ordp associated to a prime number p, one obtains
the p-adic number field Qp . The closure of Z in Qp is the ring of p-adic integers Zp . Show
that Zp is a closed subring of Qp , and that we have a (topological) isomorphism

Q∗p ∼
= hpi × Z∗p

in which hpi ∼
= Z has the discrete topology. Show that every element x ∈ Zp has a unique
representation
x= ∞ k
P
k=0 ak p

in which the ‘p-adic digits’ ak are chosen from the set {0, 1, 2, . . . , p − 1}.

29
3 Explicit ideal factorization
In order to factor an ideal I in a number ring R in the sense of 2.16, we have to determine
for all prime ideals p ⊃ I the p-primary part I(p) of I. As I is of finite index in R, a prime
p of R divides the integer [R : I], hence a prime number p. We have p ∩ Z = pZ for this
prime number p, and one says that p extends p or lies above or over p. The first step in
factoring I consists of determining the primes p of R lying over the primes p dividing [R : I],
and deciding which of them divide I. At the regular primes of R, i.e., the primes p that are
invertible, I(p) is a power of p. If p is non-invertible or singular, the situation can be more
complicated.
For given p, there are only finitely many primes p of R that extend p, and these p
correspond to the maximal ideals of the finite ring R/pR. If p extends p, the degree f (p) of
the field extension Fp ⊂ R/p is the residue class degree of p. One also says that p is an ideal
of norm pf (p) .

I The Kummer-Dedekind theorem

A simple integral extension of Z is a number ring of the form Z[α], with α a zero of a monic
irreducible polynomial f ∈ Z[X]. Every number ring R contains subrings of this nature, and
if there is a subring Z[α] ⊂ R that maps surjectively to R/pR, it suffices to determine the
primes over p in Z[α] in order to obtain them in R.
Exercise 1. Show that for such Z[α], the local rings at the primes over p in Z[α] and R are ‘the same’.

The Kummer-Dedekind theorem determines the explicit form and the regularity of the primes
lying over p in number rings Z[α].
3.1. Theorem (Kummer-Dedekind). Let f ∈ Z[X] be a monic irreducible polynomial,
α ∈ Q a zero of f and p a prime number. Let R be the ring Z[α] = Z[X]/(f ), and choose
monic polynomials gi ∈ Z[X] such that the factorization of f modulo p is
s
Y
f= g ei i ∈ Fp [X]
i=1

with ei ∈ Z≥1 and the irreducible polynomials g i = (gi mod p) ∈ Fp [X] pairwise distinct.
Then the following holds:
(1) the prime ideals of R that lie above p are the ideals pi = pR + gi (α)R, and we have an
inclusion si=1 pei i ⊂ pR;
Q

(2) the equality pR = si=1 pei i holds if and only if every prime pi is invertible;
Q

(3) writing ri ∈ Z[X] for the remainder of f upon division by gi , one has
pi is singular ⇐⇒ ei > 1 and p2 divides ri ∈ Z[X].

Proof. (1) The ring Fp [X] is a principal ideal domain, so we see from the isomorphism
R/pR ∼= Fp [X]/(f ) that the primes of R that lie above p correspond bijectively to the monic
irreducible divisors g|f in Fp [X]. The prime pi |p corresponding to a factor g i is
pi = ker[ψi : R −→ Fp [X]/(g i )] = pR + gi (α)R.

30
 §3: Explicit ideal factorization

Here ψi maps t(α) ∈ Z[α] to t mod g i , so we have

t(α) ∈ pi ⇐⇒ g i |t ∈ Fp [X].
∼
The isomorphism R/pi −→ Fp [X]/(g i ) induced by ψi shows that the residue class degree
f (pi /p) is equal to the degree of gi . As si=1 gi (α)ei is in f (α) + pR = pR, we get the desired
Q

inclusion si=1 pei i ⊂ pR + si=1 gi (α)ei R ⊂ pR.

Q Q

(2) If the inclusion in (1) is an equality, every pi is invertible as it divides the principal
ideal pR in the strict sense of Kummer: we have pi · I = pR for some ideal I. As R is an
order of rank deg f , the index of pR in R equals pdeg f . Conversely, suppose that all primes pi
over p are invertible, i.e. that the rings Rpi are discrete valuation rings. As we already have
an inclusion, it suffices to show that I = si=1 pei i also has index pdeg f in R. By the Chinese
Q

remainder theorem, we have #(R/I) = si=1 #(Rpi /pei i Rpi ). From the description in 2.17 it
Q

is clear that all quotients pki Rpi /pk+1

i Rpi in the discrete valuation ring Rpi are isomorphic (as
R-modules) to the residue class field Rpi /pi Rpi of order pdeg gi , so the quotient ring Rpi /pei i Rpi
has order pei deg gi . It follows that R/I has order pn with n = si=1 ei deg(gi ). Comparing
P

degrees in the factorisation f = si=1 g ei i shows that we have n = deg(f ), as was to be shown.
Q

(3) The remainder ri of f upon division by gi in Z[X] is divisible by p, so there are

polynomials qi , si ∈ Z[X] satisfying f = qi · gi + psi and deg(si ) < deg(gi ). Substitution of
α yields the relation
psi (α) = −qi (α)gi (α) ∈ pi
between the two R-generators p and gi (α) of pi .
Suppose that g i occurs with exponent ei = 1 in f . Then q i is not divisible by g i in
Fp [X], and this means that qi (α) is not in pi . It follows that qi (α) is in Rp∗i , so we have
gi (α) ∈ pRpi , and Rpi is a discrete valuation ring with maximal ideal generated by p.
Similarly, the hypothesis ri = psi ∈ / p2 Z[X] means that si ∈ Fp [X] is non-zero of degree
deg(si ) < deg(g i ). This implies g i - si and that si (α) is a unit in Rpi . In this situation Rpi
is a discrete valuation ring with discrete valuation ringring!discrete valuationmaximal ideal
generated by gi (α).
If we have both ei ≥ 2 and ri ∈ p2 Z[X], then qi (α) and si (α) are both in pi , and pi is
not invertible as it is not locally principal. In fact, it is not a proper R-ideal as the element
p−1 qi (α) is in r(pi ) but not in R: it satisfies
p−1 qi (α)pi = p−1 qi (α) · pR + p−1 qi (α) · gi (α)R = qi (α)R + si (α)R ⊂ pi .
This finishes the proof of 3.1.
3.2. Corollary. Suppose we have f = qi · gi + ri in 3.1.(c), and that pi is singular. Then
1
q (α) is in the multiplier ring of pi but not in Z[α].
p i

We call a number ring R singular above a rational prime p if p has a singular extension in R,
and regular above p if all extensions are regular.
Suppose that R is regular above p. Then p is called inert in R if p is a prime element
in R, i.e. if pR is the unique prime of R lying above p. If there are different extensions of p,
then p is said to split in R. If there is an extension p|p that occurs with multiplicity e > 1,
we say that p ramifies in R, or that R is ramified above p. The multiplicity with which a
prime p|p occurs in p is the ramification index of p over p.

31
§3: Explicit ideal factorization

Exercise 2. Show that Z[α] is regular above p if and only if the semi-local ring Z(p) [α] is Dedekind.

3.3. Example. Take R = Z[α] with α a zero of the polynomial f = X 3 + X + 1. The

factorizations
f = X 3 + X + 1 (mod 2)
f = (X − 1)(X 2 + X − 1) (mod 3)
show that 2 is inert in R, whereas 3 splits into prime ideals (3, α − 1) and (3, α2 + α − 1) of
norm 3 and 9. If f has multiple factors modulo a prime p > 3, then f and f 0 = 3X 2 + 1 have
a common factor modulo this prime p, and this is the linear factor f − (X/3)f 0 = 32 X + 1
0
with zero X = −3/2. As we have f (−3/2) = 31/4 = 0 only for p = 31, this is the unique
prime for which f has multiple factors. Modulo 31, we find

f = (X − 14)2 (X − 3) (mod 31),

and the remainder of f upon division by X − 14 is f (14) = 2759 = 31 · 89. As 89 is not

divisible by 31, the prime (31, α − 14) is regular. It follows that all primes of R are regular,
so R is a Dedekind domain. The prime 31, which has factorization

31R = (31, α − 14)2 (31, α − 3)

in R, is the only rational prime that ramifies in R. The reader may check that p = 47 is the
smallest rational prime that splits into 3 primes in Z[α].

Example 3.3 shows that in number rings Z[α], there is a simple relation between the residue
class degrees and ramification indices of the primes above a regular rational prime. The
same relation holds for arbitrary orders.

3.4. Theorem. Let R be an order in a number field K. If R is regular above a rational

prime p, we have X
e(p)f (p) = [K : Q].
p|p

Here the sum ranges over the primes of R extending p, and e(p) and f (p) denote the rami-
fication index and residue class degree of p over p.

Proof. Write n = [K : Q], then R is of rank n and R/pR has order pn . Factoring pR as
pR = p|p pe(p) , we can evaluate the order of R/pR = p|p R/pe(p) as in the proof of 3.1 as
Q Q
e(p)f (p)
Q
p|p p . The result follows.

It follows from 3.4 that in an order R of rank n, a regular rational prime p has at most
n extensions. If it has n extensions, we say that p is totally split in R. These extensions
necessarily have e(p) = f (p) = 1. If a regular prime p has a unique extension p in R with
e(p) = n, it is said to be totally ramified in R.

32
 §3: Explicit ideal factorization

I Singular primes
3.5. Lemma. There are only finitely many rational primes p for which an order Z[α] in 3.1
is ramified or singular above p.
α
Proof. A prime p is ramified or singular in Z[α] exactly when f = (fQ mod p) has a multiple
0
factor in Fp [X]. This factor also divides the derivative f of f . As f and its derivative f 0 are
coprime polynomials in Q[X], we can find an integer k 6= 0 and polynomials g1 , g2 ∈ Z[X]
0
satisfying g1 f + g2 f 0 = k. Reducing modulo p, we find that f and f are coprime in Fp [X]
unless p is one of the prime divisors of the non-zero integer k.
Exercise 3. Show that one has Z ∩ (f Z[X] + f 0 Z[X]) = 31Z for f = X 3 + X + 1.

In cases where theorem 3.1 tells us that a ring R = Z[α] is not Dedekind, 3.2 provides us
with elements outside R that occur in the multiplier rings of these singular primes. Such
elements can be used to enlarge the ring R into a ring with fewer singular primes.
√
3.6. Example. The ring R = Z[ −19] from section 1 is of the form Z[α], with α a zero
of the polynomial f = X 2 + 19. In this case f and f 0 = 2X are coprime modulo all primes
√
p∈ / {2, 19}, so R is regular above these primes. The identity 19R = ( −19)2 shows that R
is regular and totally ramified above 19. As we have f ≡ (X + 1)2 mod 2 and the remainder
√
of f upon division by X + 1 is f (−1) = 20 ∈ 22 Z[X], the prime p2 = (2, −19 + 1) of norm
2 is the unique singular prime of R. Note that p2 = 2p ( 2R is of index 8 in R, and that
2R is not a product of prime ideals in R.
The identity f = (X + 1)(X − 1) + 20 shows that the multiplier ring of p2 contains an
√
element β = ( −19 − 1)/2 ∈ / R. The extension ring R e = Z[β], which contains R = Z[α]
as a subring of index 2, is regular above 2 as the irreducible polynomial X 2 + X + 5 of β
is irreducible modulo 2 and 2 is inert in R. e It is also regular above all primes p 6= 2, as we
have Z(p) [α] = Z(p) [β] for these p. It follows that R
e is a Dedekind domain.
√
3.7. Example. The ring R = Z[ 3 −19] is of the form Z[α] with α a zero of the polynomial
f = X 3 + 19. This time f and f 0 = 3X 2 are coprime modulo all primes p ∈ / {3, 19}, so R
3
is regular above these primes. The factorization 19R = (α) shows that R is regular and
totally ramified above 19. Modulo 3, we find
f = (X + 1)3 (mod 3),
and the remainder of f upon division by X + 1 is f (−1) = 18 ∈ 32 Z[X]. It follows that the
unique prime p3 = (3, α + 1) above 3 is not invertible, and the identity
f = (X 2 − X + 1)(X + 1) + 18
shows that the multiplier ring of p3 contains an element β = (α2 − α + 1)/3 outside R. Note
that we have (α + 1)β = −18/3 = −6.
We claim that the ring R e = R[β] = Z[α, β], which contains R as a subring of index 3,
is a Dedekind domain. As we have Z(p) [α] = Z(p) [α, β] at all primes p 6= 3, we know that
Re is regular at all primes p 6= 3. By elementary linear algebra, we compute the minimal
polynomial of β as
β
fQ = X 3 − X 2 − 6X − 12.

33
§3: Explicit ideal factorization

β
The factorization fQ = X 2 (X − 1) mod 3 shows that the primes over 3 in Z[β] are (3, β) and
β
(3, β − 1). The prime (3, β) is invertible as fQ has remainder −12 ∈ / 32 Z[X] upon division
2
by X, so Z[β] is regular above 3. The relation α + 1 = −6/β = (−β + β + 6)/2 shows that
Z[β] is of index 2 in Z[α, β], so locally at 3 we find Z(3) [β] = Z(3) [α, β]. It follows that R
e is
regular above 3 as well.
Exercise 4. Show that Z[β] is regular above all primes p 6= 2, and that above p = 2, it has a a regular
prime (2, β + 1) and a singular prime (2, β) with multiplier ring R.
e

The examples 3.6 and 3.7 show that singular primes of a number ring R may ‘disappear’ in
suitable small extensions of R. Such extensions never introduce new singular primes.

3.8. Lemma. Let S ⊂ T be an extension of number rings such that T is contained in the
field of fractions of S. Let q be a prime of T for which p = S ∩ q is regular in S. Then q is
a regular prime of T , and it is the unique extension of p to T .

Proof. By assumption, the local ring Tq at q is a proper subring of K = Q(S) that contains
the discrete valuation ring Sp . If Tq is strictly larger than Sp , then it is equal to K as we
have Sp [x] = K for every x ∈ K \ Sp . As the local ring at a prime of a number ring is not
a field, we must have Sp = Tq , so q is regular in T . This also shows that the prime q ⊂ T is
uniquely determined by p, as q = T ∩ qTq = T ∩ pSp .

3.9. Theorem. A number ring has only finitely many singular primes.

Proof. Pick an element α ∈ R such that α generates the number field Q(R) over Q. Re-
placing α when necessary by an integral multiple kα, we may assume that its irreducible
α
polynomial fQ is in Z[X]. Applied to the extension Z[α] ⊂ R, lemma 3.8 shows that if R
is singular above a rational prime p, then the same is true for Z[α]. There are only finitely
many such p by 3.5, and each of them has only finitely many extensions to R.

I Quadratic and cyclotomic number rings

Before going into the general ‘desingularization’ of number rings, we apply the Kummer-
Dedekind theorem to give two important examples of Dedekind domains.
√
3.10. Theorem. Let d ∈ Z6=1 be squarefree and R = Z[ d] the corresponding quadratic
order. Then R is a Dedekind domain for√d ≡ 2, 3 mod 4. For d ≡ 1 mod 4, it is singular
only above 2, and the extension ring Z[ 1+2 d ] ⊃ R is a Dedekind domain.

Proof. We take f = X 2 − d in 3.1. If f has irreducible factors modulo p that occur

with multiplicity greater than 1, then it has these factors in common with its derivative
f¯0 = 2X ∈ Fp [X]. Such factors can only exist when p divides 2d. If p divides d, the multiple
factor is X. The primes lying over p are then regular since the remainder −d of f upon
division by X is squarefree and therefore not in p2 Z[X].
If p divides 2d but not d, i.e., if p = 2 and d is odd, we have f = (X + 1)2 ∈ F2 [X].
From X 2 − d = (X − 1)(X + 1) + (1 −√ d) we see that the remainder 1 − d is in 4Z[X] exactly
√
when d ≡ 1 mod 4. We find that Z[ d] is Dedekind for d 6≡ 1 mod 4, and that Z[ d] is
singular only above 2 for d ≡ 1 mod 4.

34
 §3: Explicit ideal factorization
√
In the case √d ≡ 1 mod 4 the multiplier ring of the singular prime (2, 1 + d) contains
an element α = d−1 2
/ R. The irreducible polynomial (X + 21 )2 − d4 = X 2 + X + 1−d
∈ 4
∈ Z[X]
of α has no multiple roots modulo 2, so the extension ring Z[α] ⊃ R is regular above 2. As
R is already regular above all odd primes, we find that Z[α] is a Dedekind domain.
√
By 3.1 and 3.10, the factorisation of an odd prime p in the ring Z[ d] can
be found from
the factorisation of X 2 − d in Fp [X]. It depends on the Legendre symbol dp .
√
3.11. Corollary. Let d 6
= 1 be squarefree and p an odd prime. Then p is split in Z[ d] for
d d d




p
= 1, inert for p = −1 and ramified for p = 0. 
√ √
Exercise 5. Show that 2 splits in Z[ 1+2 d ] for d ≡ 1 mod 8, and remains inert in Z[ 1+2 d ] for d ≡ 5 mod 8.

Let Z[ζ] be the cyclotomic number ring obtained by adjoining a root of unity ζ ∈ Q to Z.
If ζ = ζn is a primitive n-th root of unity, its irreducible polynomial over Q is the n-th
cyclotomic polynomial Φn , which may be defined inductively for n ≥ 1 by
Y
Xn − 1 = Φd (X).
d|n

The degree of Φn is φ(n), where φ is the Euler φ-function, and the factorization of Φn in the
n-th cyclotomic field Q(ζ) is
Y
Φn (X) = ∗
(X − ζ i ).
i∈(Z/nZ)

As X n − 1 and its derivative nX n−1 are coprime in Fp [X] for all primes p - n, the polynomial
Φn is separable modulo all primes p - n. This implies that Z[ζ] is regular above all primes
that do not divide n. This leaves a single prime to consider when n is a prime power.

3.12. Theorem. Let p be prime and ζ ∈ Q a primitive pk -th root of unity for some k ≥ 1.
Then R = Z[ζ] is a Dedekind domain that is unramified above all q 6= p and totally ramified
above p.
Denote for q 6= p by fq and gq the order and the index of hq mod pk i ⊂ (Z/pk Z)∗ . Then
there are exactly gq extensions of q to Z[ζ], and each of them has residue class degree fq .
k
X p −1
Pp−1 k−1
Proof. Apply 3.1 with f equal to Φpk (X) = X pk−1 −1
= i=0 X ip . Modulo p, we have
pk −pk−1
Φpk (X) = (X − 1) ∈ Fp [X]. The remainder of Φpk upon division by X − 1 equals
Φpk (1) = p, which is not in p2 Z[X]. It follows that the unique prime p = (p, ζ − 1) over p is
k k−1
invertible in Z[ζ] and that (p) = pp −p is totally ramified.
We have already seen that Φpk is separable modulo q 6= p, so it suffices to show that
the irreducible factors of Φpk in Fq [X] are of degree fq . A finite field Fqt of characteristic
q contains a primitive pk -th root of unity if and only if pk divides the order q t − 1 of the
cyclic group F∗qt . This happens exactly when fq divides t, so all irreducible factors of Φpk
have degree fq .
√
The ring Z[ζ4 ] = Z[i] of Gaussian integers and the ring Z[ζ3 ] = Z[(−1 + −3)/2] of Eisen-
stein are cyclotomic rings that are also quadratic. Comparing the description of the splitting

35
§3: Explicit ideal factorization

behavior of rational primes in them given by 3.12 (for pk = 3, 4) and 3.11 (for d = −1, −3),
we find once more the solution to problem 1.2, and deduce that −3 is a square modulo an
odd prime q if and only if we have q ≡ 1 mod 3. This is a special case of the quadratic
reciprocity law.

I Integral closure
For every number ring R, there exists a smallest extension ring R ⊂ R e inside K = Q(R)
that is regular at all primes, and therefore Dedekind. It is this integral closure of R in K
that we will discuss in the rest of this section.
3.13. Definition. Let A ⊂ B be an extension of rings. An element b ∈ B is called integral
over A if there exists a monic polynomial f ∈ A[X] with f (b) = 0. We say that A is integrally
closed in B if all x ∈ B that are integral over A are contained in A.
The most important case for us is the inclusion R ⊂ K of a number ring R in its field
of fractions. In this case, we simply say that R is integrally closed if it is integrally closed
in K. The rough idea of the integrality condition for R ⊂ K is that for an element x ∈ K,
Pn−1
an integrality relation xn = k
k=0 rk x with rk ∈ R implies that x cannot have a true
‘denominator’ when written as a quotient of elements of R: the ‘denominator’ of xn would
be ‘worse’ than that of n−1 k
P
k=0 rk x . The following lemma makes this idea more precise.

3.14. Lemma. A unique factorization domain is integrally closed.

Proof. Suppose that we have r, s ∈ R such that x = rs ∈ K satisfies an integral relation
xn = n−1 k n
P
k=0 ck x with ck ∈ R. Multiplying by s , we obtain

n−1
X n−1
X
n k n−k
r = ck r s =s· ck rk sn−1−k .
k=0 k=0

If R is a unique factorization domain, this relation shows that every prime element dividing
s divides rn and therefore r. Removing common prime elements from r and s, we find s ∈ R∗
and x ∈ R.
The argument in the preceding proof indicates that integrality is a ‘local property’: it can
be checked locally at the prime ideals of the ring.
3.15. Proposition. A domain R is integrally closed if and only if for every prime ideal
p ⊂ R, the localization Rp is integrally closed.
Proof. Note first that R and its localizations all have the same field of fractions K. If an
element x ∈ K is integral over R, then it is obviously integral over all localizations Rp . If all
T
Rp are integrally closed, we have x ∈ p Rp = R by 2.9, so R is integrally closed.
Pn−1
Conversely, suppose that x ∈ K satisfies an integrality relation xn = k=0 rk xk with
rk ∈ Rp for some p. If s ∈ R\p is chosen such that we have srk ∈ R for all k, multiplication by
Pn−1
sn yields an integrality relation (sx)n = k=0 rk sn−k (sx)k for sx with coefficients rk sn−k ∈ R.
If R is integrally closed we have sx ∈ R and therefore x ∈ Rp . Thus Rp is integrally
closed.

36
 §3: Explicit ideal factorization

The definition of integrality in terms of monic polynomials is often replaced by one of the
following equivalent formulations.

3.16. Lemma. Let R ⊂ R0 be an extension of domains. Then the following are equivalent
for an element x ∈ R0 :
(1) x is integral over R;
(2) R[x] is a finitely generated R-module;
(3) there exists a finitely generated R-module M ⊂ Q(R0 ) with M 6= 0 and xM ⊂ M .

Proof. Suppose x satisfies an integrality relation xn = rn−1 xn−1 + rn−2 xn−2 + . . . + r1 x + r0 .

One can then express all powers of x as R-linear combinations of the elements 1, x, x2 ,. . .,
xn−1 . This yields R[x] = R + R · x + . . . + R · xn−1 and the implication (1) ⇒ (2). The
implication (2) ⇒ (3) follows trivially by taking M = R[x]. For (3) ⇒ (1), one writes
M = Rm1 + . . . + Rmn . The inclusion xM ⊂ M means that we have identities

xmi = nj=1 rij mj

P
(i = 1, 2, . . . , n).

Now the n × n-matrix A = x · idn −(rij )ni,j=1 with entries in K = Q(R0 ) maps the non-zero
vector (mi )i ∈ K n to zero, so we have det(A) = 0. This yields the integrality relation
xn + n−1 k
P
k=0 rk x = 0 for x.

3.17. Proposition. If R ⊂ R0 is an extension of domains and x, y ∈ R0 are integral over R,

then so are xy and x + y.

Proof. If we have xM ⊂ M and yN ⊂ N for finitely generated R-modules M and N as in

3.16.(3), we can form the finitely generated R-module

M N = { i mi ni : mi ∈ M, ni ∈ N } ⊂ Q(R0 ).
P

It is mapped into itself by both x and y, hence also by x + y and xy.

It follows from 3.17 that for every extension of domains R ⊂ R0 , the set of elements x ∈ R0
that are integral over R is a subring of R0 . It is the normalization or integral closure of R in
R0 . For R0 = Q(R) the field of fractions of R, we obtain what is called the normalization or
integral closure of R.

3.18. Proposition. Let R ⊂ R0 be an extension of domains and R

e the normalization of R
in R0 . Then R
e is integrally closed in R0 .

Proof. It suffices to show that every x ∈ R0 that is integral over R e is also integral over R.
n−1
Suppose we have an integrality relation xn = k
P
k=0 ck x with ck ∈ R. By a repeated
e
application of 3.16.(2), the ring R0 = R[c0 , c1 , . . . , cn−1 ] is finitely generated as an R-module,
and M = R0 [x] is finitely generated over R0 . This implies that M is finitely generated
over R. As we have xM ⊂ M , it follows from 3.16.(3) that x is integral.

3.19. Theorem. Let R be a number ring. Then the following holds.

(1) R is Dedekind if and only if it is integrally closed.
(2) The normalization Re of R is a Dedekind domain.

37
§3: Explicit ideal factorization

Proof. As the normalization of a number ring is integrally closed by 3.18, (1) clearly implies
(2). For (1), we note that R is by definition Dedekind when its localizations Rp are discrete
valuation rings, and that it is integrally closed if and only if all Rp are integrally closed (3.15).
It therefore suffices to show the local version of (1): a local number ring Rp is a discrete
valuation ring if and only if discrete valuation ringring!discrete valuationit is integrally closed.
If Rp is a discrete valuation ring, it is a unique factorization domain, hence integrally
closed by 3.14. Conversely, let Rp be integrally closed. By 2.17, it suffices to show that its
maximal ideal p is principal. Take a non-zero element a ∈ p. Then there exists by 2.13 a
smallest positive integer n for which pn is contained in aR. Choose b ∈ pn−1 \ aR, and take
π = a/b. By construction, we have π −1 = b/a ∈ / R and π −1 p ⊂ R. As p is a finitely generated
R-module and π −1 = b/a is not integral over R, we see from 3.16.(3) that we cannot have
π −1 p ⊂ p. It follows that π −1 p equals R, so we have p = πR.

Clearly, the normalization of a number ring is the smallest Dedekind domain containing it.
More generally, we can form the integral closure of a number ring in any number field that
contains it. This yields an integrally closed number ring, whence a Dedekind domain.
The integral closure of Z in a number field K is the ring of integers OK of K. It is
the smallest Dedekind domain with field of fractions K. Its explicit determination is a key
problem in algorithmic number theory.

3.20. Theorem. Let R be a number ring with field of fractions K, and OK the ring of
integers of K. Then the following holds:
x
(1) OK = {x ∈ K : fQ ∈ Z[X]};
(2) the normalization of R equals R e = ROK ;
(3) R is Dedekind if and only if it contains OK .

Proof. It is clear that x ∈ K is in OK if its irreducible polynomial is in Z[X]. Conversely,

x
if x ∈ K is the zero of some monic polynomial g ∈ Z[X], then fQ is a monic polynomial
x
dividing g, so fQ is in Z[X] by the Gauss lemma. This proves (1).
For (2), we note first that all primes of the ring ROK are regular by 3.8, applied to the
extension OK ⊂ ROK . It is therefore Dedekind, and integrally closed. As R e contains both
R and, since it is integrally closed, OK , we have Re = ROK . This proves (2), and (3) follows
immediately.
e = OK .
Exercise 6. Show that for R an order in K, we have R

3.21. Examples. If we take K = Q(α) with α3 + α + 1 = 0, then α is integral over Z and

OK contains Z[α]. As Z[α] is Dedekind by 3.3, we have OK = Z[α].
Similarly,
√ we can use 3.10 to deduce that the ring of integers of the quadratic field
K = Q( d) for a squarefree integer d 6= 1 equals
( √
Z[ d] if d ≡ 2 or 3 mod 4;
OK = √
1+ d
Z[ 2 ] if d ≡ 1 mod 4.

From this point of view, the natural ring to exploit in solving the equation x2 + 19 = y 3 from
√ √
problem 1.4 is not Z[ −19] but Z[ 1+ 2−19 ] (cf. exercise 1.13).

38
 §3: Explicit ideal factorization
√ x
Exercise 7. Find OK for K = Q( d) directly from 3.20 by looking at fQ for x ∈ K.

Exercises.

8. Let A ⊂ B be an extension of number rings. Show that for every prime p of A, the ring
Bp = (A \ p)−1 B is a semi-local number ring with primes corresponding to the extensions of
p in B.
√ √
9. Let p = (2, 1 + −19) be the singular prime of R = Z[ −19]. Compute the index of pk in R
for k = 1, 2, 3. Conclude that R/p and p/p2 are not isomorphic as R-modules, and that the
principal ideal 2Rp is not a power of the maximal ideal in Rp .
10. Let α be a zero of the polynomial X 3 − X − 1, and R = Z[α]. Show that R is a Dedekind
ring, and determine all prime ideals of norm at most 30 in R. Show also that the unit group
R∗ is infinite.
√
11. Same questions as in the previous problem for the number ring Z[ 3 2].
12. Same questions for the cyclotomic ring Z[ζ7 ] and its subring Z[ζ7 + ζ7−1 ].
13. Let R be a number ring with field of fractions K. Show that the number of extensions of a
rational prime to R is at most [K : Q].
√
14. Determine the singular primes of the quadratic ring Z[ d] when d is a non-square (but not
necessarily squarefree) integer.
√
15. Let d be an integer that is not a cube. Show that Z[ 3 d] is a Dedekind domain if and only
√ if
d is squarefree and satisfies d 6≡ ±1 mod 9. Determine the singular primes in case Z[ 3 d] is
not Dedekind.
√
16. Generalize the previous exercise to the rings Z[ p d], where p is an odd prime number.
17. Show that the unique prime p|p in Z[ζpk ] is principal with generator 1 − ζpk .
18. Show that the cyclotomic ring Z[ζ20 ] is a Dedekind domain, and determine for each possible
residue class p modulo 20 of a prime number p the number of primes p of Z[ζ20 ] extending
p, as well as their residue class field degrees f (p) and ramification indices e(p). Deduce that
the irreducible polynomial Φ20 ∈ Z[X] is reducible modulo every prime p.
√
19. Show that every quadratic ring Z[ d] is contained in some cyclotomic ring Z[ζ]. Given d,
can you find the minimal order of ζ for which there is an inclusion?
[Hint: recall that Q(ζp ) contains a Gauss sum with square equal to (−1)(p−1)/2 p.]
20. Show that a prime p of a number ring is regular if and only if it is proper, i.e., its multiplier
ring r(p) is equal to R.
[Hint: if p is singular and a ∈ p is non-zero, we have p1 p2 . . . pt ⊂ (a) ⊂ p by exercise 2.27.
Assuming p1 = p and minimality of t, there exists b ∈ p2 . . . pt \ (a) with bp ⊂ (a).]
21. A domain B is said to be integral over a subring A ⊂ B if every b ∈ B is the zero of a monic
polynomial in A[X]. Given inclusions of domains A ⊂ B ⊂ C, show that C is integral over
A if and only if C is integral over B and B is integral over A.
22. Let R be a domain with normalization Re ⊂ K. Show that for every multiplicative subset
S ⊂ R, the normalization of S R equals S −1 R.
−1 e

39
§3: Explicit ideal factorization

23. Show that every valuation ring (as in exercise 2.49) is integrally closed.
24. Let R be a number ring that is a unique factorization domain. Show that R contains OK ,
with K the field of fractions of R.
25. Show that a number ring is a unique factorization domain if and only if it is a principal ideal
domain. More generally, show that a ring R is a principal ideal domain if and only if it is a
unique factorization domain of dimension at most 1.
[Hint: use exercise 2.29.]
26. (Algebraic integers.) Let A be the integral closure of Z in an algebraic closure Q of Q. Show
that A ∩ K = OK for every number field K ⊂ Q, and determine which of the characterizing
properties of a Dedekind domain (integrally closed, dimension 1, noetherian) hold for A.
27. Let R be a number ring with field of fractions K, and suppose that R is Dedekind. Show
that there exists a set of primes S of O = OK such that
T
R= / Op
p∈S = {x ∈ K : ordp (x) ≥ 0 for all p ∈
/ S}.

[This shows that any Dedekind domain with field of fractions K arises from OK by ‘inverting’
some of its primes.]
28. Show that, in the situation of the previous exercise, we have a long exact sequence of abelian
groups
φ
1 −→ O∗ −→ R∗ −→ p∈S Z −→ Pic(O) −→ Pic(R) −→ 1
L

in which φ maps the generator corresponding to p ∈ S to the class [p] ∈ Pic(O).

[Hint: compare the exact sequences after 2.6 defining Pic(R) and Pic(O) using the snake
lemma from commutative algebra.]
√
29. Compute the singular primes and the normalization of the order Z[ 3 37].
√ √ √
30. Show that Z[ −5] is a Dedekind domain, and that the identities 21 = (4+ −5)(4− −5) and
21 = 3 · 7 represent two factorizations of 21 into pairwise non-associate irreducible elements.
√
How does the ideal (21) factor into prime ideals in Z[ −5]? Determine the order of the
√
subgroup of Cl(Z[ −5]) that is generated by the classes of the primes dividing (21). Can
√
you find an ideal in Z[ −5] for which the class is not in this subgroup?
*31. This exercise gives an example of a noetherian one-dimensional domain with infinitely many
prime ideals, all of which are singular.
a. Prove that for every prime number p there exists a monic polynomial fp ∈ Z[X] of
degree p with the property that for each prime number q ≤ p the polynomial (fp mod
q) ∈ Fq [X] is irreducible.
In the rest of this exercise we let fp be as above, we denote by αp a zero of fp in some given
algebraic closure of Q, and we write βp = pαp . We let R = Z[β2 , β3 , β5 , . . .] be the ring
generated by all βp .
a. Let p be a prime ideal of the subring Z[β2 , β3 , . . . , βp ] of R generated by all βq with
q ≤ p. Suppose that p ∈ p. Prove that the R-ideal generated by p is a finitely generated
prime ideal of R.
c. Prove that R is a one-dimensional noetherian domain, that R has infinitely many non-
zero prime ideals, and that none of them is invertible.

40
4 Linear algebra for number rings
This section provides the basic tools from linear algebra that are used in the study of number
rings. They will enable us to compute the normalization of a number ring and to study its
singular and ramifying primes.

I Norm and trace

Let A ⊂ B be an extension of rings such that B is free of finite rank n as an A-algebra. This
means that there exist x1 , x2 , . . . , xn ∈ B that form an A-basis for B:
B = A · x1 ⊕ A · x2 ⊕ . . . ⊕ A · xn .
For x ∈ B, let Mx : B → B denote the A-linear multiplication map b 7→ xb. If we choose
an A-basis for B, this map can be described by an n × n-matrix with coefficients in A. We
define the norm and the trace from B to A by
NB/A (x) = det Mx and TrB/A (x) = trace Mx .
It is immediate from this definition that the norm is a multiplicative map, whereas the trace
TrB/A : B → A is a homomorphism of the additive groups. The characteristic polynomial
x
fB/A of x ∈ B is the characteristic polynomial of the map Mx , i.e.,
x
fB/A (X) = det(X · idB −Mx ).
x
= nk=0 ak X k ∈ A[X] is monic and of degree n. Its constant coefficient
P
The polynomial fB/A
is a0 = (−1)n NB/A (x), and its second highest coefficient is an−1 = − TrB/A (x).
In our examples, A will be a domain such as a number ring or a field. If A is a field,
classical linear algebra tells us that norms and traces do not depend on the choice of a basis
x
for B over A, and that we have fB/A (x) = 0 by the Cayley-Hamilton theorem. The same
is true for arbitrary domains A as the notions of norm, trace and discriminant are stable
under base changes such as the inclusion map A ⊂ Q(A) of A in its field of fractions. By
this we mean that for any ring homomorphism f : A → A0 , we can form the A0 -algebra
B 0 = B ⊗A A0 , which is again free of rank n as the induced map f∗ : B → B 0 maps a basis
of B over A to a basis of B 0 over A0 . For x ∈ B, the multiplication matrix for f∗ (x) in
the A0 -algebra B 0 is obtained by applying f to the entries of Mx , so we have commutative
diagrams
f∗
B B 0 = B ⊗A A0
NB/A NB 0 /A0
TrB/A TrB 0 /A0
f
A A0

for norm and trace. In particular, the absolute norm and the absolute trace R → Z for an
order R of rank n can be viewed as restrictions of the norm and trace maps for the field
extension Q ⊂ Q(R) of degree n. Similarly, their reduction modulo a prime number p yield
the norm and trace maps R/pR → Fp .

41
§4: Linear algebra for number rings

4.1. Lemma. Let R be an order and x ∈ R. Then the index of xR in R equals |NR/Z (x)|.

Proof. As R is free of finite rank as an abelian group, the index of the image xR = Mx [R]
of R under the Z-linear map Mx equals | det Mx | = |NR/Z (x)| = |NK/Q (x)|, with K the field
of fractions of R.

We already used the word norm for an ideal I ⊂ R to indicate the index of I in R. For
orders, the absolute norm of x ∈ R and the ideal norm of xR coincide as Z-ideals.
x
For a finite field extension K ⊂ K(x), the Cayley-Hamilton identity fK(x)/K (x) = 0
x
implies that the characteristic polynomial fK(x)/K is divisible by the irreducible polynomial
fKx . As they are both of degree [K(x) : K], they coincide. In case x is some element of
a finite extension L of K, one can write L as a direct sum of K(x)-vector spaces to find
x
fL/K = (fKx )[L:K(x)] .
Exercise 1. Deduce: TrL/K (x) = [L : K(x)] · TrK(x)/K (x) and NL/K (x) = NK(x)/K (x)[L:K(x)] .

There is a different characterization of norms and traces in separable field extensions. Re-
call that a finite field extension K ⊂ L is called separable if the set HomK (L, K) of K-
homomorphisms of L into an algebraic closure K of K has cardinality [L : K]. Extensions
in characteristic zero and extensions of finite fields are always separable. Every finite sep-
arable field extension K ⊂ L is primitive, i.e., it is of the form L = K(α). The irreducible
polynomial of an element α generating a separable extension is itself separable in the sense
that its roots in K are simple roots.

4.2. Lemma. Let L/K be a separable field extension of degree n. Then we have
Y
x
fL/K = (X − σ(x)).
σ∈HomK (L,K)

Q P
for x ∈ L. In particular, we have NL/K (x) = σ σ(x) and TrL/K (x) = σ σ(x).

Proof. For L = K(x), the K-homomorphisms L → K map x to the various roots of the
irreducible polynomial of x, so we have σ (X − σ(x)) = fKx = fL/K x
Q
, as desired. In the
general case, every K-homomorphism σ : K(x) → K has exactly [L : K(x)] extensions to a
K-homomorphism L → K and one finds
Y
(X − σ(x)) = (fKx )[L:K(x)] = fL/K
x
.
σ∈HomK (L,K)

x
The final statements follow by inspection of the appropriate coefficients of fL/K .
√ √
4.3. Example. For the quadratic√field Q( d) and the order Z[ d], we already √ defined√the
norm in various cases by N (x + y d) = x2 − dy 2 . This is the product (x + y d)(x − y d)
of the two zeroes
of the corresponding irreducible polynomial,
√ and also the determinant
√ of
x dy
the matrix y x describing multiplication by x + y d with respect to the basis {1, d}.

42
 §4: Linear algebra for number rings

I Discriminant
It follows from 4.1 that the absolute norm of an integral element in a number field K is
an integer measuring in some sense the ‘size’ of the element. For orders R ⊂ K there is a
similar notion of size that can be defined using the trace map.

4.4. Definition. The discriminant of an order R of rank n is defined as

∆(R) = det(TrR/Z (xi xj ))ni,j=1 ,

where x1 , x2 , . . . , xn is some Z-basis for R.

More generally, one defines the discriminant of n elements x1 , x2 , . . . , xn ∈ B in a free A-

algebra B of rank n as

∆(x1 , x2 , . . . , xn ) = det(TrB/A (xi xj ))ni,j=1 .

If x , x , . . . , xn is an A-basis for B and y1 , y2 , . . . , yn some other basis defined by yi =

Pn 1 2
j=1 aij xj for some T = (aij ) ∈ GLn (A), we find

∆(y1 , y2 , . . . , yn ) = (det T )2 ∆(x1 , x2 , . . . , xn )

from the relation (TrB/A (yi yj ))ni,j=1 = T ·(TrB/A (xi xj ))ni,j=1 ·T t . Here T t denotes the transpose
of the transformation matrix T . It follows that the discriminant of a basis depends on the
choice of the basis, but only up to the square of a unit in A. For A = Z, we find that the
discriminant of an order is independent of the basis chosen in 4.4.
If R ⊂ O is an inclusion of orders of the same rank n, we have R = T [O] for some Z-
linear map T that maps a Z-basis of O to a Z-basis of R. By the theory of finitely generated
abelian groups, the index [O : R] is in this situation finite and equal to |det T |. This yields
the useful relation

(4.5) ∆(R) = [O : R]2 · ∆(O)

between the discriminants of orders in the same number field.

4.6. Proposition. Let L/K be a separable field extension of degree n and σ1 , σ2 , . . . , σn ∈

HomK (L, K) the set of embeddings of L in K. Then one has
2
∆(x1 , x2 , . . . , xn ) = det(σi (xj ))ni,j=1


for every n elements x1 , x2 , . . . , xn ∈ L. If α ∈ L generates L over K, the power basis

1, α, α2 , . . . , αn−1 has discriminant

∆(1, α, α2 , . . . , αn−1 ) = ∆(fKα ),

where ∆(fKα ) denotes the discriminant of the polynomial fKα . One has ∆(x1 , x2 , . . . , xn ) 6= 0
if and only if x1 , x2 , . . . , xn is a K-basis for L.

43
§4: Linear algebra for number rings

Proof. If we multiply the matrix X = (σi (xj ))ni,j=1 by its transpose and use the description
of the trace map from 4.1, we find

n
n Pn
n
n
X t · X = σk (xi ) i,k=1 · σk (xj ) k,j=1 = k=1 σk (xi xj ) i,j=1 = TrL/K (xi xj ) i,j=1 .

Taking determinants, we obtain the first statement. If we have L = K(α), the elements
αi = σi (α) are the roots of the polynomial fKα . For the basis 1, α, α2 , . . . , αn−1 of L/K, the
associated discriminant [det(σi (αj−1 ))ni,j=1 ]2 is the square of the Vandermonde determinant
det(αij−1 )ni,j=1 = i>j (αi − αj ). This yields
Q

Y
∆(1, α, α2 , . . . , αn−1 ) = (αi − αj )2 = ∆(fKα ).
i>j

As fKα has distinct roots, the discriminant ∆(fKα ) is non-zero.

Every basis of L/K is obtained from the power basis 1, α, α2 , . . . , αn−1 by an invertible
coordinate transformation, so the discriminant ∆(x1 , x2 , . . . , xn ) is non-zero for every basis
of L/K. For a K-linearly dependent n-tuple, the discriminant clearly vanishes.
4.7. Corollary. Let f ∈ Z[X] be a monic irreducible polynomial with root α. Then the
order Z[α] has discriminant ∆(f ).
Proof. We have ∆(Z[α]) = ∆Z[α]/Z (1, α, . . . , αn−1 ) = ∆Q(α)/Q (1, α, . . . , αn−1 ) = ∆(f ).
One can reformulate the non-vanishing of the discriminant for finite separable field extensions
L/K by saying that the trace form
TrL/K : L × L −→ K
(x, y) 7−→ TrL/K (xy)

is a non-degenerate bilinear form, or that the map L → L b = Hom(L, K) to the dual vector
space of L over K that sends x to the map (y 7→ TrL/K (xy)) is an isomorphism. This implies
that for every basis x1 , x2 , . . . , xn of L over K, there is a dual basis y1 , y2 , . . . yn of L over K
such that
TrL/K (xi yj ) = δij .
Here δij denotes the Kronecker delta.
4.8. Theorem. If K is a number field of degree n, then there exists a basis ω1 , ω2 ,. . ., ωn
for K over Q for which we have

OK = Zω1 ⊕ Zω2 ⊕ . . . ⊕ Zωn .

Proof. Choose an arbitrary basis x1 , x2 , . . . , xn of K over Q consisting of integral elements,

and let y1 , y2 , . . . , yn be the dual basis. Then the abelian group M † = Zy1 ⊕ Zy2 ⊕ . . . Zyn
is the trace dual of the abelian group M = Zx1 ⊕ Zx2 ⊕ . . . Zxn , i.e., it can be given as

M † = {x ∈ K : TrK/Q (xM ) ⊂ Z}.

If x ∈ OK is integral, then xM consists of integral elements and we have TrK/Q (xM ) ⊂ Z.

This shows that we have M ⊂ OK ⊂ M † . As M and M † are free abelian groups of rank n,
the same holds for OK .

44
 §4: Linear algebra for number rings

A basis as in 4.8 is called an integral basis for OK . Its existence shows that OK is an order
in K. As every order R ⊂ K is contained in OK (cf. 3.16, or exercise 3.6), the ring of
integers is the maximal order in K. It is the only order in K that is a Dedekind domain. Its
discriminant ∆(OK ) is usually referred to as the discriminant ∆K of the number field K.

4.9. Theorem. Every number ring R is of finite index in its normalization R.

e

Proof. Writing O = OK for the ring of integers of K = Q(R), we have R e = RO by 3.20.

The ring R ∩ O is an order with field of fractions K, so its additive group is free of the same
rank [K : Q] as O. It follows that R ∩ O is of finite index k ∈ Z≥1 in O, and that k Re = kOR
is contained in R. By 2.11, we conclude that R is of finite index in R as it contains the
e
non-zero R-ideal
e k R.
e

The rational primes p dividing the index [R

e : R] are the p that have singular extensions in
R. For orders R = Z[α], we have Re = OK and we can detect such p using 4.7 and 4.5.

4.10. Theorem. Let Z[α] be an order in a number field K. Then one has
α
∆(fQ ) = [OK : Z[α]]2 · ∆K .

It follows that if f ∈ Z[X] is a monic irreducible polynomial with squarefree discriminant, the
order Z[α] is the ring of integers in Q(α). The theorem also shows the need for computational
techniques to find discriminants of polynomials.

I Computational techniques
Let A be a domain and f ∈ A[X] a monic polynomial. Then B = A[α] = A[X]/(f ) with
α = X mod f is a free A-algebra of rank n. In this algebra, one can compute the norm of
elements of the form a0 + a1 α as
−a0
NB/A (a0 + a1 α) = (−a1 )n f ( ).
a1
For the general case, one can use resultants. The resultant of two non-zero polynomials
g = b ri=1 (X − βi ) and h = c sj=1 (X − γj ) with coefficients and zeroes in some field F is
Q Q

defined as r Y s
Y
s r
R(g, h) = b c (βi − γj ).
i=1 j=1

One directly derives from this definition that R(g, h) satisfies the following properties:
(R1) R(g, h) = (−1)rs R(h, g);
(R2) R(g, h) = bs ri=1 h(βi );
Q

(R3) R(g, h) = bs−s1 R(g, h1 ) if h1 6= 0 satisfies h1 ≡ h mod gF [X] and s1 = deg h1 .

It is immediate from property (R2) that for x = g(α) ∈ B = A[α] in our situation above,
one has
NB/A (g(α)) = R(f, g).

45
§4: Linear algebra for number rings

If f is separable with zeroes α1 , α2 , . . . , αn in F ⊃ A, one has f 0 (α1 ) = i≥2 (α1 −αi ). Taking
Q

for g the derivative f 0 of f in the formula above, one finds that the discriminant of f can be
written as
Y
∆(f ) = (αi − αj )2 = (−1)n(n−1)/2 NB/A (f 0 (α)) = (−1)n(n−1)/2 R(f, f 0 ).
i<j

This reduces the computation of norms and polynomial discriminants to the computation
of resultants, which can be performed inside the field containing the coefficients of the
polynomials.
4.11. Example. Let f = X 3 − X 2 − 6X − 12 be the irreducible polynomial of the element β
occurring in 3.7. Long division shows that the remainder of f upon division by its derivative
f 0 = 3X 2 − 2X − 6 equals (X 3 − X 2 − 6X − 12) − 91 (3X − 1)(3X 2 − 2X − 6) = − 38
9
(X + 3).
This is a linear polynomial with zero −3, so we can apply (R1)–(R3) to find
∆(f ) = −R(f, f 0 ) = −R(f 0 , f ) = 32 · R(f 0 , − 38
9
(X + 3))
= 32 · R(− 38
9
(X + 3), f 0 ) = −32 · (− 38
9
)2 · f 0 (−3) = −22 · 3 · 192 .
√ √
As Z[β] has index 2 in the ring of integers of its field of fractions K = Q( 3 −19) = Q( 3 19),
we find from 4.10 that K has discriminant ∆K = −3 · 192 .
√
Exercise 2. Derive the same conclusion starting from the order Z[ 3 −19] of discriminant ∆(X 3 + 19).

4.12. Example. Let p > 2 be a prime and K = Q(ζp ) the p-th cyclotomic field. Then we
have ∆K = ∆(Φp ) = (−1)(p−1)/2 NK/Q (Φ0p (ζp )). Every difference of two distinct p-th roots of
unity is a product of a root of unity and a conjugate of 1 − ζp . As we have NK/Q (ζp ) = 1
and NK/Q (1 − ζp ) = Φp (1) = p, we find ∆Q(ζp ) = (−1)(p−1)/2 pp−2 .
Example 4.11 indicates how one can find OK for a small number field K = Q(α). After
replacing α when necessary by a suitable multiple kα, we may suppose that α is integral. One
α
computes the discriminant ∆(fQ ) of the order Z[α] using either the resultant or the values
of the power sums of α (exercises 19– 21). For each prime number p for which p2 divides
α
∆(fQ ), one has to check whether p divides the index [OK : Z[α]]. The primes p that divide
the index are exactly the primes for which the semi-local ring Z(p) [α] is not integrally closed,
i.e. the primes p for which there is a prime ideal p|p in Z[α] that is not invertible. These
primes can be found using the Kummer-Dedekind theorem 3.1, and for each of them we find
an element of x ∈ OK outside Z[α]. After adjoining all these elements to Z[α] we obtain a
subring R ⊂ OK with fewer singular primes than Z[α]. If there are primes p|[R : Z[α]] for
which p2 divides ∆(fQ α
)/[R : Z[α]]2 , one still has to check whether R(p) is integrally closed.
If it is not, then for one of the finitely many non-zero elements in x ∈ p1 R/R the lift x ∈ p1 R
is integral over Z. This shows that we find OK after a finite computation.
√ √
4.13. Example. Consider the order R = Z[ 4 −19] ⊂ K = Q( 4 −19). As we have R = Z[α]
α
with fQ = X 4 + 19, we can apply 3.1. Just as in 3.7, we easily find that R is regular and
unramified above p 6= 2, 19, and that R is regular and totally ramified above 19. The unique
prime (2, α − 1) in R above 2 is singular: from
X 4 + 19 = (X − 1)(X 3 + X 2 + X + 1) + 22 · 5

46
 §4: Linear algebra for number rings

we find an integral element β = (α3 + α2 + α + 1)/2 and an integral extension R ⊂ R[β]

of index 2. As R has discriminant ∆(X 4 + 19) = 28 · 193 , the discriminant of R[β] is
2−2 ∆(R) = 26 · 193 . In order to see whether R[β] is regular above 2, we have to check
whether there are integral elements in 21 R[β] \ R[β]. In this case, such an element is easily
√
found: the generator γ = (α2 + 1)/2 = ( −19 + 1)/2 of the ring of integers of the subfield
√
Q( −19) ⊂ K is not in R[β]. Note that we have (α + 1)γ = β. This gives us an order
B = Z[α, β, γ] = Z[α, γ] in K of discriminant 4−2 ∆(R) = 24 · 193 . One can check that none
of the 15 non-zero elements in 12 B/B has an integral lift in K, so we have B = OK and
∆K = 24 · 193 . A more efficient way to see this proceeds by applying a relative version of
the Kummer-Dedekind theorem 3.1 to the extension A ⊂ B = A[α] of the Dedekind domain
A = Z[γ] – see exercises 31 and 32.

I Ramification
α α
If a prime p divides the discriminant ∆(fQ ) of an order R = Z[α], then fQ mod p has multiple
factors and it follows from 3.1 that either p is regular and ramified in R, or p is singular in R
and a divisor of the index [OK : Z[α]]. In view of 4.10, this strongly suggests the following
result.
4.14. Theorem. A rational prime p ramifies in the ring of integers OK of K if and only if
it divides the discriminant ∆K .
As not every ring of integers is of the form Z[α], we cannot derive this result from 3.1. The
proof is based on the following elementary lemma.
4.15. Lemma. Let M = M1 × M2 × . . . Mt be a product of free R-algebras of finite rank.
Then we have TrM/R ((mi )ti=1 ) = ti=1 TrMi /R (mi ) and ∆(M/R) = ti=1 ∆(Mi /R).
P Q

Proof. It suffices to prove the lemma for M = M1 × M2 , the general case then follows by
induction. View the R-algebras M1 and M2 as R-submodules of M and choose R-bases X of
M1 and Y of M2 . Then X ∪Y is a basis for M , and we have xy = 0 for x ∈ X and y ∈ Y . The
trace of an element mi ∈ Mi as an element of Mi and as an element of M ⊃ Mi × {0} = Mi
coincides, so we have the general identity TrM/R (m1 , m2 ) = TrM1 /R (m1 ) + TrM2 /R (m2 ). One
deduces that the discriminant of the basis X ∪ Y of M is the determinant of a block matrix
in which the blocks are the discriminant matrices for M1 and M2 . This yields ∆(M/R) =
∆(M1 /R)∆(M2 /R).
Proof of 4.14. In order to study ∆K mod p, we apply the base extension Z → Fp . This
simply means that we reduce everything modulo p and look at the Fp -algebra O/pO =
O ⊗Z Fp . This algebra has discriminant ∆(OK ) · Fp = (∆K mod p) over Fp . We have
pO = p|p pe(p/p) in O, so the Chinese remainder theorem for O/pO and 4.15 yield
Q

Y
∆K mod p = ∆((O/pe(p/p) )/Fp ).
p|p

If p is unramified in O, all discriminants ∆((O/pe(p/p) )/Fp ) = ∆((O/p)/Fp ) are non-zero by

4.6, since Fp ⊂ O/p is a separable field extension. In this case we find that ∆K mod p is
non-zero, so p does not divide ∆K .

47
§4: Linear algebra for number rings

If p is ramified in O, there exists p|p in O with e = e(p/p) ≥ 2. In this case we can

choose a basis for O/pe over Fp that contains an element x ∈ p \ p2 . As the element x is
nilpotent in O/pe , we have Tr(xy) = 0 ∈ Fp for all y ∈ O/pe . It follows that the discriminant
of every basis containing x vanishes, so we have ∆((O/pe )/Fp ) = 0 and ∆K = 0 mod p.
√ √
The examples we have treated so far, such as K = Q(ζp ), Q( 3 −19) or Q( 4 −19), show that
the exponent with which p divides ∆K tends to be high if the ramification indices of the
primes over p are high. The precise relation involves the different of K.

4.16. Definition. The different DK of a number field K is the integral OK -ideal whose
inverse
D−1
K = {x ∈ K : TrK/Q (xOK ) ⊂ Z}.

is the trace dual of OK .

Note that D−1 K is indeed a fractional OK -ideal, and that its inverse is integral as we
have DK ⊃ OK . If ω1 , ω2 , . . . , ωn is a Z-basis for OK , the co-different D−1
−1
K has Z-basis
ω1∗ , ω2∗ , . . . , ωn∗ .

4.17. Theorem. The different DK of K is an OK -ideal of norm [OK : DK ] = |∆K | that is

divisible only by the ramified primes of OK . For every prime p of OK one has pe(p/p)−1 |DK .

Proof. Choose a basis ω1 , ω2 , . . . , ωn of OK , and let T : D−1

K → OK be a Z-linear map
−1
sending the Z-basis ω1 , ω2 , . . . , ωn of DK to the basis of OK by T ωi∗ = ωi . Then OK has
∗ ∗ ∗

index |det T | in D−1

K , and we find

1 = det(TrK/Q (ωi ωj∗ ))ni,j=1 = det(T −1 ) · det(TrK/Q (ωi ωj )ni,j=1 ) = det(T −1 ) · ∆K .

It follows that we have [OK : DK ] = [D−1

K : OK ] = |det(T
−1
)| = |∆K |.
Suppose that p|p is a prime of O = OK that ramifies with index e > 1. Then every
element x in the integral O-ideal p1−e p reduces modulo p to an element in O/pO that is
contained in all prime ideals of O/pO. It follows that x is a nilpotent element in O/pO, so
we have 0 = Tr(O/pO)/Fp (x) = TrK/Q (x) mod p. This proves the inclusion

TrK/Q (p1−e p) ⊂ pZ.

As TrK/Q is Z-linear, we obtain TrK/Q (p1−e ) ⊂ Z, whence p1−e ⊂ D−1

K and p
e−1
|DK .
The preceding argument shows that for every prime p above p dividing DK , we have
TrK/Q (p−1 p) ⊂ pZ and, reducing modulo p, the identity Tr(O/pO)/Fp (p−1 p/pO) = 0. If p is
unramified, the ideal p−1 p/pO ⊂ O/pO is the summand
Y
O/p ⊂ O/pO = O/p × O/qe(q/p) ,
q|p,q6=p

and Tr(O/pO)/Fp is by 4.15 the field trace Tr : O/p → Fp . As the trace does not vanish in a
separable field extension, we find p - DK for unramified p.

48
 §4: Linear algebra for number rings

One can show that pe(p/p)−1 is the exact power of p dividing DK if and only if the ramification
index e(p/p) is not divisible by p. In this case one says that p is tamely ramified over p. If p
does divide the ramification e(p/p), the ramification is said to be wild.
Theorem 4.17 shows that the different is a finer measure for ramification than the
discriminant as it detects the primes in OK that are ramified, not just the rational primes
lying below them. As a consequence, it lives in OK and not in Z.
In a similar way, the index [Re : R] has as its prime divisors the primes p above which
R is singular. For the conductor of R measuring the singular primes themselves, we refer to
exercise 25.

Exercises.

3. Show that the norm and the trace are transitive in towers of separable field extensions, i.e.,
for a tower K ⊂ L ⊂ M we have NL/K ◦ NM/L = NM/K and TrL/K ◦ TrM/L = TrM/K .
4. Let OK be the ring of integers of K and NK/Q : I(OK ) → Q∗ the ideal norm, i.e., the
homomorphism that maps a prime p of OK to NK/Q p = pf (p/p) . Show the compatiblity

|NK/Q (x)| = NK/Q (xOK )

of the ideal norm of a principal ideal xOK with the element norm NK/Q (x) of x.
5. Let R be a number ring. Show by an example that the norm map I 7→ N (I) = [R : I] on the
set of integral R-ideals is not necessarily multiplicative.
6. Let K be a field and f = ni=0 ai X i ∈ K[X] a monic irreducible polynomial. Define L =
P

K[X]/(f ) and x = X mod (f ). Write the multiplication map Mx : L → L as a matrix with

respect to the basis 1, x, x2 , . . . , xn−1 of L/K and verify that Mx has characteristic polynomial
f.
7. Let K be a number field of degree n. Show that there is an isomorphism K ⊗Q C ∼
= Cn
Q P
mapping k⊗z to (zσ(k))σ∈Hom(K,C) . Deduce: NK/Q (x) = σ σ(x) and TrK/Q (x) = σ σ(x).
8. Let {x1 , x2 , . . . , xn } be a basis for the separable field extension L/K and {x∗1 , x∗2 , . . . , x∗n } the
dual basis. Prove: ∆(x1 , x2 , . . . , xn ) · ∆(x∗1 , x∗2 , . . . , x∗n ) = 1.
√
9. Show that for a squarefree integer d 6= 1, the corresponding quadratic field K = Q( d) has
discriminant 
4d if d ≡ 2 or 3 mod 4;
∆K =
d if d ≡ 1 mod 4.

10. A free A-algebra of finite rank B is separable over A if ∆(B/A) is a unit in A.

a. When is a finite field extension separable in this sense?
b. Show that B is separable over A if and only if the natural map B → HomA (B, A)
sending b to the homomorphism x 7→ TrB/A (xb) is an isomorphism of A-modules.
c. Show that if B is separable over A, then B 0 = B ⊗A A0 is separable over A0 for any base
change A → A0 . Does the converse hold?
11. Let K be a number field and α ∈ OK an element such that α 6∈ kOK for any k ∈ Z>1 . Show
that there exists an integral basis for OK containing α. Deduce that we can choose ω1 = 1
in every integral basis.

49
§4: Linear algebra for number rings

12. Let K be a number field and 2s the number of embeddings σ : K → C with σ[K] 6⊂ R. Show
that we have sign(∆K ) = (−1)s .
13. (Stickelberger’s criterion.) Show that the discriminant of an order R satisfies

∆(R) ≡ 0 or 1 mod 4.
P Qn
[Hint: the discriminant is of the form (P − N )2 , where P = π i=1 σi (ωπ(i) ) with π ranging
over the even permutations of {1, 2, . . . , n}.]
14. Let K = Q(α) be of degree n with α ∈ OK integral. Show the following:
a. Z[α] is regular above p if and only if the natural map Z[α] → OK /pOK is surjective;
b. Z[α] is not regular above a prime p < n that splits completely in OK ;
c. OK is not of the form Z[x] for any x ∈ K if some prime p < n splits completely in OK .
2
15. Let K = Q(α) with α3 +α2 −2α+8 = 0. Show that we have OK = Z[α, α+α 2 ] and OK 6= Z[β]
for every β ∈ OK . (This example is due to Dedekind.)
√
16. Determine the ring of integers and the discriminant of the field K = Q( 3 20). Show that
there is no α ∈ OK such that OK = Z[α].
[Hint: the equation 2a3 − 5b3 = ±1 has no solutions modulo 7.]
√
17. Let d be an integer that is not a cube. Compute the ring of integers of K = Q( 3 d).
[You may want to start with the easier cases 3 - d or d squarefree.]
18. Prove the identities
1
∆(X 3 − aX − b) = 4a3 − 27b2 and ∆(X n + a) = (−1) 2 n(n−1) nn an−1

for n ∈ Z>0 and a, b in some field K.

19. (Power sums.) Let K be a field and f = ni=1 (X − αi ) ∈ K[X] a separable polynomial.
Q

Define the power sums pk of f for k ∈ Z by pk = ni=1 αik . Show that the discriminant of f
P

equals
∆(f ) = det(pi+j−2 )ni,j=1 .

20. (Newton’s formulas.) Let K be a field and f = ni=1 (X − αi ) = nk=0 ak X n−k ∈ K[X] a
Q P

monic polynomial of degree n with zeroes α1 , α2 , . . . , αn ∈ K. Put ak = 0 whenever k > n,

and define the power sums of f by pk = ni=1 αik as in the previous exercise. Show that for
P

every k ≥ 1 one has

k−1
X
kak + aj pk−j = 0.
j=0
Qn
[Hint: Compute the logarithmic derivative of i=1 (1 − αi T ).]
21. Set f = X 4 +X 2 +X +1. Compute the power sums pk of f for 0 ≤ k ≤ 6 and the discriminant
∆(f ). Show also that f is irreducible over Q, and that Z[X]/(f ) is a Dedekind domain.
22. Let p be a prime number. Compute the discriminant of the cyclotomic field Q(ζpk ).
23. Compute the different DK for a quadratic number field.
√
24. Same question for Q( 3 −19).

50
 §4: Linear algebra for number rings

25. Let R be a number ring with normalization R.

e The conductor of R is defined as

fR = {x ∈ R e ⊂ R}.
e : xR

Show that fR is the largest R-ideal

e that is contained in R, and that a prime p of R divides
fR if and only if it is singular.
√
26. Compute the singular primes, the conductor and the normalization of the order Z[ 3 37].
27. Let R be an order in a quadratic field K. Show that we have R = Z + f OK for some unique
non-negative integer f ∈ Z. Express the conductor fR in terms of f .
28. Let O be the ring of integers of a real quadratic field K of odd discriminant, and R ⊂ O√the
order of index 2 in O. Suppose that we have R∗ 6= O∗ . Prove: [O∗ : R∗ ] = 3 and K = Q( d)
for some positive squarefree integer d ≡ 5 mod 8.
[Compare with exercise 1.12.]
29. (Euler’s identities.) Let K = Q(α) be a number field of degree n.
n−1
a. Show that the Q-basis of K that is trace dual to the power basis {αi }i=0 equals {bi }n−1
i=0 ,
with bi ∈ K defined by

f (X)
= b0 + b1 X + . . . + bn−1 X n−1 ∈ K[X].
f 0 (α)(X
− α)

b. Suppose that α is integral. Show that we have Z[α]† = f 0 (α)−1 Z[α]. Deduce that we
have DK = f 0 (α)OK if Z[α] is the ring of integers of K.
r
α f (X)
[Hint: Show that the polynomial g = f 0 (α)(X−α) ∈ K[X] satisfies TrK/Q g = X r , where the
trace of the polynomial is taken coefficientwise.]
30. (Linear independence of group characters, Artin-Dedekind.) Show that distinct group homo-
morphisms σi : G → C ∗ from a group G to the unit group of a field C are linearly independent
over C, i.e. the expression

a1 σ1 (g) + a2 σ2 (g) + . . . + an σn (g)

with ai ∈ C is non-zero for some g ∈ G unless a1 = a2 = . . . = an = 0. Deduce that the

discriminant of a basis in a finite separable extension is non-zero.
[Hint: Assume we have a dependence relation with n minimal. Then n > 1, and for g, g 0 ∈ G
we can subtract σ1 (g 0 ) times the relation for g from the relation for gg 0 to obtain a relation
without σ1 .]
31. Let A be a number ring and B = A[α] a simple integral extension obtained by adjoining a
zero in an extension of K = Q(A) of a polynomial f ∈ A[X] that is irreducible in K[X].
Let p be a regular prime of A. Formulate and prove the analogue of 3.1 for the primes in B
extending p.
√ √
32. Show that the normalization of R = Z[ 4 −19] is R
e = R[γ] with γ = ( −19 + 1)/2.
√
[Hint: put A = Z[γ] and B = A[ 4 −19] and apply the theorem in the previous exercise.]

51
5 Geometry of numbers
In this section, we prove the classical finiteness theorems for a number ring R: the Picard
group Pic(R) is a finite group, and the unit group R∗ is in many cases finitely generated.
These are not properties of arbitrary Dedekind domains, and the proofs rely on the special
fact that number rings can be embedded in a natural way as lattices in a finite dimensional
real vector space. The key ingredient in the proofs is non-algebraic: it is the theorem of
Minkowski on the existence of lattice points in symmetric convex bodies given in 5.1.

I Minkowski’s theorem
Let V be a vector space of finite dimension n over the field R of real numbers, and h·, ·i :
V × V → R a scalar product, i.e., a positive definite bilinear form on V × V . The scalar
product induces a notion of volume on V , which is also known as the Haar measure on V .
For a parallelepiped
B = {r1 x1 + r2 x2 + . . . + rn xn : 0 ≤ ri < 1}
spanned by x1 , x2 , . . . , xn , the volume is defined by
vol(B) = | det(hxi , xj i)ni,j=1 |1/2 .
This definition shows that the ‘unit cube’ spanned by an orthonormal basis for V has volume
1, and that the image of this cube under a linear map T has volume | det(T )|. If the vectors
xi are written with respect to an orthonormal basis for V as xi = (xik )nk=1 , then we have
| det(hxi , xj i)ni,j=1 |1/2 = | det(M · M t )|1/2 = | det(M )|
for M = (xij )ni,j=1 .
The volume function on parallelepipeds can be uniquely extended to a measure on V .
Under the identification V ∼ = Rn via an orthonormal basis for V , this is the Lebesgue
measure on Rn . We usually summarize these properties by saying that V is an n-dimensional
Euclidean space.
A lattice in V is a subgroup of V of the form
L = Z · x1 + Z · x2 + . . . + Z · x k ,
with x1 , x2 , . . . , xk ∈ V linearly independent. The integer k is the rank of L. It cannot
exceed n = dim V , and we say that L is complete or has maximal rank if it is equal to n.
For a complete lattice L ⊂ V , the co-volume vol(V /L) of L is defined as the volume of the
parallelepiped F spanned by a basis of L. Such a parallelepiped is a fundamental domain
for L as every x ∈ V has a unique representation x = f + l with f ∈ F and l ∈ L. In fact,
vol(V /L) is the volume of V /L under the induced Haar measure on the factor group V /L.
A subset X ⊂ V is said to be symmetric if it satisfies −X = {−x : x ∈ X} = X.
5.1. Minkowski’s theorem. Let L be a complete lattice in an n-dimensional Euclidean
space V and X ⊂ V a bounded, convex, symmetric subset satisfying
vol(X) > 2n · vol(V /L).

52
 §5: Geometry of numbers

Then X contains a non-zero lattice point. If X is closed, the same is true under the weaker
assumption vol(X) ≥ 2n · vol(V /L).

Proof. By assumption, the set 12 X = { 21 x : x ∈ X} has volume vol( 21 X) = 2−n vol(X) >
vol(V /L). This implies that the map 12 X → V /L cannot be injective, so there are distinct
points x1 , x2 ∈ X with 12 x1 − 12 x2 = ω ∈ L. As X is symmetric, −x2 is contained in X. By
convexity, we find that the convex combination ω of x1 and −x2 ∈ X is in X ∩ L.
Under the weaker assumption volume vol(X) ≥ 2n vol(V /L), each of the sets Xk =
(1 + 1/k)X with k ∈ Z≥1 contains a non-zero lattice point ωk ∈ L. As all ωk are contained
in the bounded set 2X, there are only finitely many different possibilities for ωk . It follows
T T
that there is a lattice element ω ∈ k Xk , and for closed X we have k Xk = X.

I Number rings as lattices

Let K be a number field of degree n. Then K is an n-dimensional Q-vector space, and by
base extension we can map K into the complex vector space

KC = K ⊗Q C ∼
Y
= C = Cn
σ:K→C

by the canonical map ΦK : x 7→ (σ(x))σ . Note that ΦK is a ring homomorphism, and that
the norm and trace on the free C-algebra KC extend the norm and the trace of the field
extension K/Q. The image ΦK [K] of K under the embedding lies in the R-algebra

KR = {(zσ )σ ∈ KC : zσ = z σ }

consisting of the elements of KC invariant under the involution F : (zσ )σ −→ (z σ )σ . Here σ

denotes the embedding of K in C that is obtained by composition of σ with complex conju-
gation.
On KC ∼ = Cn , we have the standard hermitian scalar product h · , · i. It satisfies
hF z1 , F z2 i = hz1 , z2 i, so its restriction to KR is a real scalar product that equips KR with a
Euclidean structure. In particular, we have a canonical volume function on KR . It naturally
leads us to the following fundamental observation.

5.2. Lemma. Let R be an order in a number field K. Then ΦK [R] is a lattice of co-volume
|∆(R)|1/2 in KR .

Proof. Choose a Z-basis {x1 , x2 , . . . , xn } for R. Then ΦK [R] is spanned by the vectors
(σxi )σ ∈ KR , and using the matrix X = (σi (xj ))ni,j=1 from the proof of 4.6, we see that the
co-volume of ΦK [R] equals

| det(h(σxi )σ , (σxj )σ i)ni,j=1 |1/2 = | det(X t · X)|1/2 = |∆(R)|1/2 .

If I ⊂ R is a non-zero ideal of norm N (I) = [R : I] ∈ Z, then 5.2 implies that ΦK [I] is a

lattice of co-volume N (I) · |∆(R)|1/2 in KR . To this lattice in KR we will apply Minkowski’s
theorem 5.1, which states that every sufficiently large symmetric box in KR contains a
non-zero element of ΦK [I].

53
§5: Geometry of numbers

In order to compute volumes in KR , we have a closer look at its Euclidean structure.

Denote the real embeddings of K in C by σ1 , σ2 , . . . , σr and the pairs of complex embeddings
of K by σr+1 , σr+1 , σr+2 , σr+2 , . . . , σr+s , σr+s . Then we have r + 2s = n = [K : Q], and an
isomorphism of R-algebras

KR −→ Rr × Cs
(5.3)
(zσ )σ 7−→ (zσi )r+s
i=1 .

The inner product on KR is taken componentwise, with the understanding that at a complex
component, the inner product of elements z1 = x1 + iy1 and z2 = x2 + iy2 ∈ C equals

h zz11 , zz22 i = z1 z 2 + z 1 z2 = 2 Re(z1 z 2 ) = 2(x1 x2 + y1 y2 ).

Note that this differs by a factor 2 from the inner product under the ‘standard identification’
C = R2 of C as the ‘complex plane’. For this reason, volumes in KR are 2s times larger
than they are in Rr × Cs with the ‘standard’ Euclidean structure.

I Finiteness of Picard groups

As a direct application of Minkowski’s theorem, we can show that in a number ring R that
is an order, every ideal I ⊂ R contains an element x 6= 0 that is not much larger than N (I).
As a consequence, we deduce that the Picard group Pic(R) is finite.

5.4. Theorem. Let R be an order in a number field K with s pairs of complex embeddings.
Then

every 1/2
ideal class in the Picard group Pic(R) contains an integral ideal of norm at most
2 s
π
|∆(R)| , and Pic(R) is a finite abelian group.

Proof. Let I ⊂ R be an ideal and Xt ⊂ KR the closed box consisting of elements (zσ )σ
with |zσ | ≤ t for all σ. Then Xt ⊂ Rr × Cs is a product of r real intervals [−t, t] and s disks
in C of radius t, so its canonical volume in KR equals

vol(Xt ) = 2s · (2t)r (πt2 )s = 2r+s π s tn .

Minkowski’s theorem 5.1 implies that Xt contains a non-zero element ΦK (x) ∈ ΦK [I] ∩ Xt ,
provided that we have 2r+s π s tn ≥ 2n N (I)|∆(R)|1/2 . We therefore choose t ∈ R>0 to satisfy

s
tn = π2 N (I) · |∆(R)|1/2 .

The norm of the element x ∈ I obtained satisfies |NK/Q (x)| = σ |σ(x)| ≤ tn since all |σ(x)|
Q

are bounded by t, so we conclude that every integral R-ideal I contains a non-zero element
2 s


x of norm |NK/Q (x)| ≤ π N (I) · |∆(R)|1/2 .
If I is invertible, the element x ∈ I we just found gives rise to an integral ideal xI −1
s
in the ideal class [I −1 ] ∈ Pic(R) of norm at most π |∆(R)|1/2 . As I was arbitrary, this
2

implies that every ideal class in Pic(R) contains an integral ideal satisfying this norm bound.
There are only finitely many ideals in R having norm below a given bound, so we find that
Pic(R) is finite.

54
 §5: Geometry of numbers

For R = OK the ring of integers of a number field K, the Picard group Pic(R) is the class
group
Cl(K) = Cl(OK )
of K, and its order is known as the class number hK of K. It is a fundamental invariant of
the number field K.
Exercise 1. Show that Pic(R) is finite for every Dedekind domain R ⊂ K. [Hint: use exercise 3.28.]

As the primary ideals occurring in the factorization of an invertible integral ideal I ⊂ R have
norm at most N (I), it follows from
5.4 that for an order R, the Picard group is generated by
s
primary ideals of norm at most π2 |∆(R)|1/2 . For the ring of integers of K, we find that
2 s


Cl(OK ) is generated by prime ideals of norm at most π |∆K |1/2 . This enables us to find
an explicit set of generators of the class group by factoring the rational primes up to this
bound.
√
5.5. Examples. For the Dedekind domain R = Z[ −5] we have ∆(R)
√ = −20, so Pic(R)
2
is generated by the ideal classes of the primes of norm at most π 20 < 3. The unique
√
prime p = (2, 1 + −5) of norm 2 in R is non-principal, but its square is generated by 2. We
√
find that Pic(R) ∼= Z/2Z is generated by [p2 ], and that Q( −5) is a field of class number 2.
√
The ring R = Z[ −19] is not a Dedekind domain.
√ By 3.11 the prime 3 is inert in R,
so the only primary R-ideals of norm at most π2 4 · 19 < 6 are the non-invertible prime
√ √
p2 = (2, 1 + −19), the principal ideal (2) and the prime ideals p5 = (5, 1 + −19) and
√
q5 = (5, 1 − −19) lying over 5. As p5 q5 = (5) is principal, we find that Pic(R) is generated
√
by p5 . As p5 is non-principal with cube p35 = (7 + 2 −19), the Picard group Pic(R) is cyclic
of order 3.
√ √
Let O = Z[(1 + −19)/2] be the integral closure of the ring Z[ −19] √ just considered.
2
Then 2 is inert in O, so there are no primes in O of norm smaller than π 19 < 3. It follows
√
that no primes are needed to generate the class group, so Pic(O) is trivial and Q( −19)
has class number 1.

I Minkowski’s constant

s
The constant π2 in 5.4 can be improved if we can replace Xt by a box of the same volume
containing elements of smaller norm. A look at the real quadratic case in the pictures below
suggests that we should take Xt ⊂ KR of the form
X
(5.6) Xt = {(zσ )σ ∈ KR : |zσ | ≤ t}.
σ

Computing the volume of the box Xt in (5.6) is an exercise in integration.

n
5.7. Lemma. The box Xt ⊂ KR in (5.6) has canonical volume vol(Xt ) = 2r π s tn! .

Proof. Inside Rr × Cs , our box takes the shape

X X
Xt = {((xi )i , (zj )j ) ∈ Rr × Cs : |xi | + 2 |zj | ≤ t},
i j

55
§5: Geometry of numbers

s
and we need to show that this box has ‘standard volume’ Vr,s (t) = 2r π2 tn n!. As the
volume function is a real valued function depending on two integral parameters r and s, we
can apply induction with respect to each of these parameters.
We note first that V1,0 (t) = vol([−t, t]) = 2t and V0,1 (t) = vol({|z| ≤ t/2}) = πt2 /4 have
the required value. For Vr+1,s (t) we use the induction hypothesis to obtain
Z t  s Z t (t − x)n  π s tn+1
r π
Vr+1,s (t) = Vr,s (t − |x|)dx = 2 ·2 dx = 2r+1 .
x=−t 2 x=0 n! 2 (n + 1)!
For Vr,s+1 (t) the method is similar, but we have to integrate with respect to a complex
variable z = x + iy = ρeiθ to obtain its value
Z t/2
(t − 2ρ)n  π s+1 tn+2
Z
r π
 s
Vr,s+1 (t) = Vr,s (t − 2|z|)dxdy = 2 · 2π ρdρ = 2r .
|z|<t/2 2 ρ=0 n! 2 (n + 2)!
This establishes the result.
5.8. Theorem. Let R be an order in a number field K of degree n with s pairs of complex
embeddings. Then every ideal class of Pic(R) contains an integral ideal of norm not exceeding
the Minkowski constant  s
4 n!
MR = · |∆(R)|1/2
π nn
of the order R.
Proof. As in the proof of 5.4, we need to bound the absolute value of the norm of an
element x = (zσ )σ ∈ Xt , where t is chosen such that vol(Xt ) = 2n N (I)|∆K |1/2 . Using the
n
value vol(Xt ) = 2r π s tn! from 5.7 and the fact that the geometric mean ( σ |zσ |)1/n does not
Q

exceed the arithmetic mean n1 σ |zσ |, we obtain

P

Y 1X n tn
s n!
|N (x)| = |zσ | ≤ ( |zσ |) ≤ n = π4 n
· |∆(R)|1/2 .
σ
n σ n n
The result follows as in 5.4.

56
 §5: Geometry of numbers

5.9. Corollary. The class group of a
number field K is generated by the classes of the
s
prime ideals of norm at most MK = π4 nn!n |∆K |1/2 . 

A second corollary of 5.8 is the existence of a lower bound for discriminants of orders in
terms of their rank.

5.10. Corollary. Let R be an order of rank n over Z. Then we have

 π 2s n2n  π n n2n
def
|∆(R)| ≥ ≥ b n = .
4 (n!)2 4 (n!)2
1/n
One has bn ≥ π n /4 for all n, and limn→∞ bn = πe2 /4 ≈ 5.803.

Proof. As every integral R-ideal has integral norm, we must have MR ≥ 1 in 5.8. This
yields the inequality. From the identity bn+1 /bn = (π/4) · (1 + n1 )2n ≥ π we obtain both the
1/n
lower bound for bn and the limit behavior limn→∞ bn = limn→∞ bn+1 bn
= πe2 /4.

Better lower bounds for discriminants have been derived around 1975 by complex analytic
methods [23]. Our bounds already imply the famous result of Minkowski that there are no
number fields K 6= Q of discriminant 1.

5.11. Corollary. Let K 6= Q be a number field. Then there exists a prime p that is ramified
in K.

Proof. For K 6= Q we have n ≥ 2 and |∆K | ≥ π 2 /4 > 2, so there is a prime dividing ∆K .

This prime ramifies in K/Q by 4.14.

I Hermite’s theorem
It follows from 5.11 and 4.5 that the discriminant of any order R 6= Z in a number field
satisfies |∆(R)| > 1. There are other restrictions on the values of discriminants, such as
the Stickelberger criterion ∆(R) ≡ 0, 1 mod 4 from exercise 4.13. If R is quadratic over Z,
it is uniquely determined by its discriminant. This is not true for orders of higher rank
(exercise 18), but the geometry of numbers yields the following finiteness result for orders
‘with bounded ramification’.

5.12. Hermite’s theorem. Let D ∈ Z be an integer. Then there are, up to isomorphism,

only finitely many orders R of discriminant ∆(R) = D.

Proof. As the ring of integers OK of a number field K contains only finitely many orders
of bounded index, it suffices in view of 4.10 to show that there are only finitely many
isomorphism classes of number fields of discriminant ∆K = D. The degree n of such K is
bounded by 5.10, so we may assume that both D and n are fixed.
Suppose K is a field of degree n with absolute discriminant D. Then we construct a
box X ⊂ KR that is bounded in terms of D and contains an element Φ(x) with K = Q(x) as
follows. If K has a real embedding τ : K → R we let X consist of the elements (zσ )σ ∈ KR
satisfying
|zτ | ≤ C and |zσ | < 1 if σ 6= τ,

57
§5: Geometry of numbers
p
where C ∈ R>0 is chosen to have vol(X) > 2n |D|. If K has only complex embeddings and
τ : K → C is one of them, we define X as consisting of the elements with

zτ , zτ ∈ [−1, 1] + [−C, C]i ⊂ C and |zσ | ≤ 1 if σ 6= τ, τ ,

p
with C ∈ R>0 such that again vol(X) = 2n |D|. In both cases, X is symmetric and convex
and C depends on D but not on K.
By 5.1 and 5.2, there exists a non-zero element x ∈ OK such that Φ(x) = (σ(x))σ is
contained in X. We will show that K = Q(x) for this x. As the embeddings σ : K → C are
the [K : Q(x)] extensions to K of each of the [Q(x) : Q] embeddings Q(x) → C, it suffices
to show that for our chosen τ , we have σ(x) 6= τ (x) whenever σ 6= τ . In the case that τ is a
real embedding, we cannot have τ (x) = σ(x) for any σ 6= τ as this would imply |σ(x)| < 1 for
Q
all embeddings σ : K → C and therefore |NK/Q (x)| = σ |σ(x)| < 1. This obviously cannot
happen for the non-zero integer NK/Q (x) ∈ Z. In the case that τ is a complex embedding,
the same argument shows that τ (x) 6= σ(x) if σ 6= τ, τ . We also cannot have τ (x) = τ (x),
since this would imply that τ (x) is a real element in (−1, 1) + (−C, C)i, which leads to the
same impossible estimate |τ (x)| < 1.
We conclude that K is generated over Q by an element x ∈ OK with Φ(x) ∈ X. This
x
means that the zeroes σ(x) ∈ C of the irreducible polynomial fQ ∈ Z[X] have an absolute
x
value that can be bounded in terms of D. As the degree n of fQ is fixed, we can also bound
x
the coefficients of fQ in Z. It follows that, given D, there are only finitely many possibilities
x
for fQ , and therefore the number of possible K ⊂ Q is also finite.
Given r, s ∈ Z≥0 , it is a challenging problem to count up to isomorphism the number Sr,s (X)
of number fields K having r real and 2s complex embeddings and satisfying |∆K | < X. One
can do this exactly for given X, or asymptotically for X → ∞. Hermite’s theorem, though
in principle constructive, does not provide a direct answer to these questions. Only recently
one has been able to obtain answers in cases with n = r + 2s > 3.
Exercise 2. Find asymptotic expressions for S2,0 (X) and S0,1 (X).

I Dirichlet’s unit theorem

As a final application of Minkowski’s theorem, we show that the unit group of an order R
in a number field with r real and 2s complex embeddings is finitely generated of free rank
r + s − 1. The classical formulation of this result is as follows.
5.13. Dirichlet unit theorem. Let R be an order admitting r real and 2s complex em-
beddings, and write µR for the group of roots of unity in R. Then µR is finite, and R∗ /µR
is a free abelian group of rank r + s − 1.
More explicitly but less canonically, 5.13 states that there exists a finite set η1 , η2 , . . ., ηr+s−1
of fundamental units such that we have

R∗ = µR × hη1 i × hη2 i × . . . × hηr+s−1 i.

Such a system of fundamental units, which forms a Z-basis for R∗ /µR , is only unique up to
coordinate transformations and multiplication by roots of unity. In the real quadratic case

58
 §5: Geometry of numbers
√
R = Z[ d], we saw already that the fundamental unit solving the Pell equation x2 −dy 2 = ±1
is only unique up to sign and inversion.
In case r > 0 we have µR = {±1} as there are no other roots of unity in R. In general
one has µR ⊂ µK = hζm i with m the largest integer for which the m-th cyclotomic field
Q(ζm ) can be embedded in K. For such m, the degree [Q(ζm ) : Q] = φ(m) divides [K : Q]
and the odd prime factors of m all divide ∆K . This severely restricts m, and in practice the
groups µK and µR are also easily determined in the totally complex case r = 0.
In order to prove 5.13, we reformulate it in terms of our R-algebra KR . As we are now
dealing with a subgroup the multiplicative group K ∗ , we consider the embedding

Φ : K ∗ → KR
∗ ∗
= {(zσ )σ ∈ KC : z σ = zσ }.

As the geometry of numbers with its lattices prefers to work with additive groups, we apply
∗
the logarithm componentwise on the unit group KC = (C∗ )n of KC to obtain a homo-
morphism Log : KC → R that sends (zσ )σ to (log |zσ |)σ . Write L : R∗ → Rn for the
∗ n

composition of Φ with this logarithmic map. With this notation, the Dirichlet unit theorem
can be phrased as follows.

5.14. Theorem. Let R be an order admitting r real and 2s complex embeddings. Then
the homomorphism L : R∗ → Rn that sends x ∈ R∗ to (log |σx|)σ has a finite cyclic kernel
µR consisting of the roots of unity in R∗ and maps R∗ to a lattice of rank r + s − 1 in Rn .
∗
Proof. For every bounded set B = [−M, M ]n ⊂ Rn , the inverse image in KR under the
−M M
logarithmic map is the bounded set {(zσ )σ ∈ KR : e ≤ |zσ | ≤ e }. The intersection of
this set with the lattice Φ(R) is finite, so the inverse image L−1 [B] ⊂ R∗ is also finite. This
implies first of all that L[R∗ ] is a discrete subgroup of Rn , i.e. a lattice in Rn . Secondly,
taking M = 0, we see that ker L is a finite subgroup of R∗ . The elements in this group have
finite order, so they are roots of unity. As every root of unity in R is clearly in ker L, we
have ker L = µR . As this is a finite subgroup of K ∗ , it is cyclic.
It is easy to see that the image L[R∗ ] is a lattice in an r + s − 1 dimensional subspace
of Rn . First of all, we have log |σ(x)| = log |σ(x)| for every x ∈ K ∗ , so L[R∗ ] lies in the
(r + s)-dimensional subspace
∗
Log[KR ] = {(xσ )σ ∈ Rn : xσ = xσ } ⊂ Rn .

Secondly, the absolute value |N (η)| of the norm of a unit η equals [R : ηR] = 1, so the sum
of the coordinates of L(η) equals
P Q
σ log |σ(x)| = log σ |σ(x)| = log |N (η)| = 0.

We can take the first restriction into account by composing L : R∗ → Rn = Rr+2s with the
surjection Rr+2s = Rr × (R2 )s → Rr+s that maps each component R2 corresponding to a
pair of complex conjugate embeddings to R by adding the components. The composition
L0 : R∗ → Rr+s has the same kernel as L, and its image lies again in the ‘trace-zero-
Pr+s
hyperplane’ H ⊂ Rr+s consisting of elements (xi )r+si=1 satisfying i=1 xi = 0. The difficult
0 ∗
part of the theorem consists in showing that L [R ] is a lattice in H of maximal rank.

59
§5: Geometry of numbers

∗
Let E ⊂ KR be defined as the ‘norm-1-subspace’
Q
E = {(zσ )σ : σ zσ = ±1}.

Under the composition Log0 of the logarithmic map Log : KR ∗

→ Rn with our surjection
Rn → Rr+s , we have Log0 [E] = H. We will construct a subset Y ⊂ E such that Log0 [Y ] ⊂ H
is bounded and satisfies L[R∗ ] + Log0 [Y ] = H. By exercise 5, this implies that L[R∗ ] has
maximal rank in H, as we want to show.
Let X = Xt = {(zσ )σ ∈ KR : |zσ | ≤ t for all σ} be our standard box in KR , and choose
t such that we have vol(X) = 2n · |∆K |1/2 . By Minkowski’s theorem, X contains a non-zero
element of the lattice Φ[R]. For every e = (eσ )σ ∈ E, the set

eX = {ex : x ∈ X} = {(zσ )σ ∈ KR : |zσ | < |eσ |t}

is a box around the origin with volume vol(eX) = vol(X), so it also contains an element
Φ(xe ) ∈ Φ[R]. The norm N (xe ) of xe ∈ R is bounded by σ |eσ |t = tn for each e, so the set
Q

of ideals {xe R : e ∈ E} is finite. Suppose that it consists of {ai R}ki=1 . We claim that
k
[
Φ(a−1


Y =E∩ i )X
i=1

is the required subset of E.

We note first that all boxes Φ(a−1
i )X are bounded in KR , so their union is bounded as
well. The absolute values |yσ | of the coordinates of an element y = (yσ )σ ∈ Y are bounded
Q
away from zero since each |yσ | is bounded from above and we have σ |yσ | = 1. This implies
that Log0 [Y ] is a bounded subset of Rr+s .
Now let e ∈ E be arbitrary. Then there exist a non-zero element a ∈ R such that Φ(a)
is contained in e−1 X and an element ai ∈ R as defined above satisfying ai a−1 = u ∈ R∗ .
It follows that e is contained in Φ(a−1 )X = Φ(u)Φ(a−1 i )X for some ai . This proves the
inclusion E ⊂ Φ[R ] · Y . Applying Log yields L[R ] + Log0 [Y ] = H.
∗ 0 ∗

As the proof of 5.14 relies on Minkowski’s theorem 5.1, it is not directly constructive and
cannot be used to explicitly find unit groups. Our explicit methods in section 7 will therefore
be of a different nature.
The L0 -image of R∗ is a complete lattice in the ‘trace-zero-hyperplane’ H ⊂ Rr+s . The
co-volume of the lattice obtained by any projection Rr+s → Rr+s−1 that leaves out one of
the coordinates is called the regulator Reg R of R. The proof of 5.14 yields the following
explicit definition.

5.15. Definition. Let K be a number field and {σi }r+s

i=1 a complete set of pairwise non-
conjugate embeddings of K in C. Then the regulator of a set {ε1 , ε2 , . . . , εr+s−1 } of r + s − 1
elements in K ∗ of norm ±1 is defined as

Reg(ε1 , ε2 , . . . , εr+s−1 ) = det(ni log |σi εj |)r+s−1

i,j=1 .

Here the integer ni ∈ {1, 2} equals 1 if σi is a real embedding and 2 otherwise.

60
 §5: Geometry of numbers

The regulator Reg(R) of an order R in K is the regulator of a system of fundamental units for
R∗ . We put Reg(R) = 1 if R∗ is finite, i.e., if R is either Z or an imaginary quadratic order.
The regulator RK = Reg(OK ) of the ring of integers of K is known as the regulator of the
number field K. By the Dirichlet unit theorem, regulators of orders do not vanish. Unlike
the discriminant of the order, which is an integer, the regulator of an order is a positive
real number that is usually transcendental as it is an expression in terms of logarithms of
algebraic numbers.
Exercise 3. Show that the unit group R∗ of an order R ⊂ K is a subgroup of finite index in OK
∗
, and that
∗ ∗
we have [OK : µK R ] = Reg(R)/ Reg(OK ).

To conclude this section, we briefly look at the case of number rings R that are not necessarily
orders. Any number ring R is by 4.9 of finite index in its normalization R, e and if k is this
index, then the kernel of the natural map

(5.16) e∗ −→ (R/k
R e R)e∗

is a subgroup of finite index of Re∗ that is contained in R∗ as we have 1 + k R e ⊂ R. This

implies that R∗ is of finite index in R
e∗ , and if R
e∗ is finitely generated abelian of free rank
t, then so is R∗ . It may however be that R e∗ is not finitely generated – think of the case
R = K.

61
§5: Geometry of numbers

Exercises.

4. Let R = ⊕ni=1 Z · ωi be an order in a number field K. This exercise shows how to prove Pic(R)
finite without any volume computations.
a. Show that there exists a constant C = C(K) such that

|NK/Q ( ni=1 ai ωi )| < C · (max |ai |)n .

P

b. Show that there exists a constant M = M (R) such that for every x ∈ K ∗ , we can find
ω ∈ R and t ∈ {1, 2, . . . , M } with

|NK/Q (tx − ω)| < 1.

[Hint: for large M , there exist 1 ≤ t1 < t2 < M for which t1 x and t2 x are ‘close’ in
K/R. Take t = t2 − t1 .]
c. Show that every non-zero ideal I ⊂ R contains an element α such that αI −1 divides
M ! · R. Deduce that Pic(R) is finite.
[Hint: take α ∈ I non-zero with |NK/Q (α)| minimal. Then every β ∈ I satisfies tβ =
αω ∈ I for some t and ω as in b.]
5. Prove that a subgroup L ⊂ Rn is a lattice if and only if it is discrete. For n = 1, show that
L is either a lattice or a dense subgroup.
6. Show that a lattice L ⊂ Rn has maximal rank if and only if the following equivalent conditions
hold
(i) the factor group Rn /L is compact in the quotient topology;
(ii) there exists a bounded subset B ⊂ Rn such that L + B = Rn .
√
7. Let V be the set of non-zero lattices L ⊂ C that satisfy x · L ⊂ L for every x ∈ Z[ 1+ 2−23 ].
P
Show that V becomes an abelian group if we set L1 · L2 = { xi√yi ∈ C : xi ∈ L1 , yi ∈ L2 }.
Let P ⊂ V be the set of lattices of the form Lz = Z · z + Z · (1+ 2−23)z with z ∈ C∗ . Show
that P is a subgroup of V and that V /P ∼= Z/3Z.
8. (Minkowski’s theorem on linear forms) Suppose that n linear forms on Rn with real coeffi-
cients are given by Li (x) = nj=1 aij xj , and that A = (aij )ni,j=1 has non-zero determinant.
P

Show that there exists a non-zero element x ∈ Zn satisfying

|Li (x)| < ci for i = 1, 2, . . . , n

if the constants ci ∈ R>0 satisfy ni=1 ci > | det A|.

Q

Can you find positive integers x, y satisfying

√ √ √ √ √ √
q
|x 7 − y 6| + |x 15 − y 13| < 2 91 − 90 = 0.458515 . . .?

9. Find all integral solutions to the equation X 2 = Y 3 − 13.

√
10. Determine the class number of Q( d) for d = −41, −47, −163.
11. Show that all real quadratic fields of discriminant ∆ < 40 have class number 1. What is the
√
class number of Q( 10)?

62
 §5: Geometry of numbers

12. (Lagrange’s four squares theorem.) Let p be a prime number.

a. Show that there exist u, v ∈ Z such that u2 + v 2 + 1 ≡ 0 mod p, and that the corres-
ponding lattice

Lu,v = {(a, b, c, d) ∈ Z4 : c ≡ ua + vb mod p and d ≡ ub − va mod p}

has co-volume p2 in R4 .
b. Show that every positive integer is the sum of four squares.
[Hint: for a prime number p this can be deduced from the fact that an open ball of ra-
√
dius 2p in R4 contains a non-zero lattice point from Lu,v . Then use the multiplicativity
of the norm N : R[i, j, k] → R on the quaternion-algebra.]
13. Apart from the involution F on KC defined in the text, there is also the ‘component-wise’
complex conjugation on KC sending (zσ )σ to (z σ )σ . We denote it by an overhead bar. Prove
that the following conditions are equivalent.
(i) The automorphism of KC maps K to itself;
(ii) There is a field automorphism τ of K such that for all x ∈ K and all embeddings
σ : K → C one has στ x = σx;
(iii) Either K is totally real, or K is a totally imaginary quadratic extension of a totally real
field.
An algebraic number field K is called a CM-field if the above three conditions are satisfied.
The abbreviation stands for “complex multiplication”. This has its origin in a connection
with endomorphism rings of elliptic curves and abelian varieties. Give an example of an
algebraic number field that is not a CM-field.
14. Let f ∈ Z[X] be a monic polynomial. Prove that |∆(f )| = 1 if and only if there exists k ∈ Z
such that f = X − k or f = X 2 − (2k + 1)X + k(k + 1).
15. Let L be a lattice in an n-dimensional Euclidean space V . Prove:
p
#{x ∈ L : hx, xi ≤ t} ωn
lim n
= ,
t→∞ t vol(V /L)

where ωn = π n/2 /Γ( n2 + 1) denotes the volume of the unit ball in V .

16. Let R be an order in a number field K. Show that we have R = OK if and only if all primes
p ⊂ R of index at most the Minkowski constant MR of R are invertible.
17. Let K be a number field with Minkowski constant M < 2.
a. Prove that OK is a principal ideal domain.
b. Prove that for each x ∈ K there exists ω ∈ OK with |N (x − ω)| < 1.
[Hint: let X be the box in 5.6 with t = n/2 = [K : Q]/2, and set Y = X ∪ (ΦK (x) + X).
Then the map Y → KR /ΦK [OK ] is not injective.]
c. Prove that OK is a Euclidean ring.
18. Show that the three cubic fields that are obtained by adjoining to Q a root of one of the
polynomials
X 3 − 18X − 6, X 3 − 36X − 78, X 3 − 54X − 150
all have the same discriminant, but that no two of them are isomorphic.

63
§5: Geometry of numbers

*19. Let K be a a number field of degree n.

a. Let m be a positive integer. Prove that the number of non-zero ideals I ⊂ OK of norm m
is less than or equal to the number of vectors (xi )ni=1 with xi ∈ Z, xi > 0, ni=1 xi = m.
Q

b. Let M be the Minkowski constant of K. Prove that the class number h of K satisfies

h ≤ vol{(xi )ni=1 ∈ Rn>0 : ni=1 max{1, xi } ≤ M }

Q

(n − 1 + log M )n−1
≤M· .
(n − 1)!

20. Let R be a subring of a ring A, and suppose that A is integral over R. Show that the unit
groups satisfy A∗ ∩ R = R∗ . Can the integrality condition be omitted?
21. (Artin.) Let K be a cubic field with a single real embedding σ : K → R. Show that the
group of units of OK that have positive image under σ is isomorphic to Z, and let u > 1 be
the σ-image of a generator. Show that ∆K is a negative integer that satisfies

|∆K | ≤ 4u3 + 24.

[Hint: write the conjugates of the generator as u = x2 and x−1 e±iy and estimate the function
φ(x, y) = |∆(1, u, u2 )|1/2 for fixed x.]
22. Let a ≥ 2 be an integer for which 4a3 + 27 is squarefree. Show that the unit group of the
ring of integers of the field Q(α) with α3 + aα − 1 = 0 has rank 1, and that α ∈ OK∗ is a

fundamental unit.
23. Let K = Q(α) with α3 = α + 1. Prove that the elements

α, α − 1, α2 − 1, α3 − 1, (α − 2)2 /(α + 3), (2α − 1)2 /(α + 5)

belong to OK ∗ and find a minimal set of generators for the subgroup U generated by these
∗ ?
units. Is U equal to OK
∗ for each of the following number fields K:
24. Determine the class number and the unit group OK
√ √ √ √ √
Q[X]/(X 3 − X + 2),
3
Q( 2), Q( 5, −5), Q( −7, 21).

∗ lies dense in R if and only if K is either totally

25. Let K ⊂ R be a number field. Show that OK
real cubic or of degree n ≥ 4.
26. Let K be a number field and Φ : K → KR the canonical embedding. Show that for any
non-zero x ∈ OK , the square of the length of Φ(x) ∈ KR satisfies hΦ(x), Φ(x)i ≥ [K : Q],
with equality if and only if x is in µK .
27. Let k > 0 and l ≥ 0 be integers and denote by ϕ the Euler function. Prove that there exists
an algebraic number field K for which OK∗ is isomorphic to (Z/kZ) ⊕ Zl if and only if k is

even and ϕ(k) divides 2(l + 1).

28. Let n be a positive integer. Suppose one has an equilateral polygon in the Euclidean plane
with the property that all angles, with the possible exception of two consecutive ones, are
integral multiples of π/n. Show that the remaining two angles are integral multiples of π/n
as well.

64
 §5: Geometry of numbers

29. Show that Z[ζ5 ] is a Euclidean ring with unit group Z[ζ5 ]∗ = hζ10 i × hηi, where η denotes a
√
fundamental unit in the ring of integers of Q( 5). Express 1+ζ5 , 1+ζ5 +ζ52 and 1+ζ5 +ζ52 +ζ53
on this basis.
30. Let K1 be a real quadratic field and K2 an imaginary quadratic field. Prove:
∗ ∗
[OK 1 K2
: µK1 K2 OK 1
] ∈ {1, 2}.

Do both values occur?

31. Let K ( L be an extension of number fields. Show that [OL ∗ : O ∗ ] is finite if and only if L
K
is a totally imaginary CM-field with maximal real subfield K. Show also that if the index is
finite, we have
∗ ∗
[OL : µL OK ] = 21−[K:Q] RK /RL .
∗ : O ∗ ] is finite. Show that there is a group homomorphism ψ : O ∗ → µ that
32. Suppose [OL K L L
∗ to u/σ(u), with σ the non-trivial K automorphism of L. Prove also that
maps u ∈ OL
∗ ∗ ∗
ker ψ = OK , # coker ψ · [OL : OK · µL ] = 2.

33. Let n > 1 be an integer, n 6≡ 2 mod 4, and ζn ∈ Q a primitive n-th root of unity.
a. Prove that the number of roots of unity in Z[ζn ] equals lcm(n, 2).
b. Prove that
Z[ζn ]∗ = hζn i · Z[ζn + ζn−1 ]∗
if n is a prime power, and that

[Z[ζn ]∗ : hζn i · Z[ζn + ζn−1 ]∗ ] = 2,

Z[ζn ]∗ = h1 ± ζn i · Z[ζn + ζn−1 ]∗

if n is not a prime power.

√ √ √
34. Let K1 = Q( d1 ) and K2 = Q( d2 ) be distinct real quadratic fields. Define K3 = Q( d1 d2 )
and K = K1 K2 .
a. Show that OK∗ O ∗ O ∗ is of finite index in O ∗ .
1 K2 K3 √ √ K
∗ O∗ O∗ ?
b. Take d1 = 2 and d2 = 3. Is 2 + 3 in OK 1 K2 K3

65
6 Zeta functions
In its present form, this section contains just the definition of the zeta function of a number
field and, without proof, in 6.3 and 6.5, two of its properties. All we need for the moment is
the identity (6.4), which will serve as an independent check to verify the correctness of our
computations in of Picard groups and unit groups in the next section.

I Dedekind zeta function

The Dedekind zeta-function ζK of a number field K is a complex analytic function which,
despite its simple definition, encodes a lot of fundamental information on the number field.
For K = Q, this zeta function is the well known Riemann zeta-function, which is defined
on the complex right half-plane Re(t) > 1 by the formula ζ(t) = ∞ −t
P
n=1 n . This sum can
be viewed as a sum over the non-zero ideals nZ of the ring of integers Z of Q, and for
arbitrary K, the function ζK is defined in a similar way as
X
(6.1) ζK (t) = (NK/Q (I))−t ,
I6=0

where the sum ranges over all non-zero ideals I ⊂ O of the ring of integers OK of K. The
convergence properties of the sum are just as for the Riemann zeta function, and it also has
a representation as an Euler product over prime ideals, as follows.
6.2. Lemma. Let t ∈ C be a complex number with Re(t) > 1. Then we have an identity
X Y
ζK (t) = (NK/Q (I))−t = (1 − (NK/Q (p))−t )−1
I6=0 p prime

in which the sum over I 6= 0 and the Euler product over p are absolutely convergent. The
function ζK (t) is a holomorphic function without zeroes in the half plane Re(t) > 1.
Proof. For each rational prime number p, there are at most n = [K : Q] primes p|p, and
these primes have NK/Q (p) ≥ p. The estimate
X X
|N (p)−t | ≤ n p− Re(t)
N (p)≤X p≤X

shows that p N (p)−t converges absolutely in the half plane Re(t) > 1, and this implies
P

that the product p prime (1 − (NK/Q (p))−t )−1 also converges absolutely for these t. From
Q

the geometric series

X∞
(1 − NK/Q (p)−t )−1 = NK/Q (p)−kt
k=0
and the fact that every ideal I has a unique factorization as a product of prime ideal powers,
it is easily seen that for t ∈ R>1 , the sum I6=0 NK/Q (I)−t has positive terms and is bounded
P

by p (1 − (NK/Q (p))−t )−1 . It follows that the sum converges absolutely whenever Re(t) > 1.
Q

The estimate
X Y X
| N (I)−t − (1 − N (p)−t )−1 | ≤ N (I)− Re(t) → 0
I6=0 N (p)≤X N (I)>X

66
 §6: Zeta functions

for X → ∞ implies that sum and product coincide. The convergence is uniform on compact
subsets of the half plane Re(t) > 1, so the limit function ζK is holomorphic on this half
plane. The product representation shows that there are no zeroes in this region.

I Analytic class number formula

The sum in (6.1) diverges for t = 1, but ζK does have a meromorphic continuation to the
left of the half space Re(t) > 1 such that t = 1 becomes a simple pole for this extended
function. Its residue at t = 1 involves all basic invariants of the number field, including the
class number of OK .

6.3. Theorem. Let K be a number field of degree n with r real and 2s complex embeddings.
Then the zeta-function ζK of K admits a meromorphic extension to the half-plane Re(t) >
1 − 1/n. It is holomorphic except for a single pole at t = 1 with residue

∗ 2r (2π)s hK RK
ζK (1) = .
wK |∆K |1/2

Here wK = #µK is the number of roots of unity in K.

Proof. See [5, Chapter 5, section 1].

∗
The formula for the residue ζK (1) at t = 1 is known as the analytic class number formula.
∗
It shows that ζK (1), and therefore the product hK RK , can be computed analytically from
the limit
∗
ζK (1) = lim(t − 1)ζK (t).
t→1

As the Riemann zeta function ζQ has a pole with residue 1 at t = 1, one can compute
limt→1 (t − 1)ζK (t) as the limit limt→1 ζK (t)/ζQ (t) of a quotient of zeta functions that is
holomorphic at t = 1. The quotient of their Euler products is convergent at t = 1, and one
obtains
2r (2π)s hK RK Y
(6.4) = E(p),
wK |∆K |1/2 p

where the Euler factor at the rational prime p is defined by

−1
Q
−1 p|p (1 − NK/Q (p) )
E(p) = .
1 − p−1
We will see in the following section how this formula can be applied.

I Functional equation
Just like the Riemann zeta function, the Dedekind zeta-function can be extended to a mero-
morphic function ζK on all of C which is holomorphic except for a simple pole at t = 1.
It has a functional equation relating its values at t and 1 − t which most clearly shows
its symmetries if we complete ζK with Euler factors ‘at infinity’ coming from the complex
embeddings of K.

67
§6: Zeta functions

6.5. Theorem. Let K be a number field of degree n with r real and 2s complex embeddings.
Then ζK can be extended to a holomorphic function on C\{1}. The completed zeta function

r
s
ZK (t) = |∆K |t/2 Γ(t/2)π −t/2 Γ(t)(2π)−t ζK (t)

satisfies the functional equation ZK (t) = ZK (1 − t).

The Γ-function occurring in the definition of the completed zeta function ZK was already
defined by Euler as Z ∞
Γ(t) = xt−1 e−x dx.
0

As it satisfies the functional equation Γ(t + 1) = tΓ(t) and has the easily computed value
Γ(1) = 1, we see that the Γ-function has simple poles at integers k ∈ Z≤0 , and satifies
Γ(k + 1) = k! for integers k ≥ 0. It then follows from the functional equation that ζK has
‘trivial zeroes’ at all negative integers k ∈ Z<0 . At odd integers k the multiplicity of the
zero equals s, at even k the multiplicity equals r + s. At k = 0, the zeta function ζK has a
∗
zero of order r + s − 1 with leading coefficient ζK (0) = −hK RK /wK in its Taylor expansion.
All other zeroes ρ of ζK satisfy 0 < Re(ρ) < 1.
The first proof of 6.5 was given by Hecke in 1910. A more elegant proof using harmonic
analysis on adelic groups, which shows how the Γ-factors in ZK can be seen as Euler factors
at the ‘infinite primes’ of K, was given by Tate in 1959. These proofs can both be found
in [7, Chapter XIII and XIV]. As indicated there, the techniques of Hecke have been used
recently by Zimmert and Skoruppa to show that not only the discriminant (cf. 5.10), but
also the regulator grows exponentially with the degree. An explicit lower bound is

(6.6) RK /wK ≥ .02 · exp(.46r + .1s).

The famous generalized Riemann hypothesis (GRH—seegeneralized Riemann hypothesis)

states that all zeroes ρ of ζK in the critical strip 0 < Re(t) < 1 satisfy Re(ρ) = 1/2. This
is the most important open conjecture in analytic number theory. There are many other
unproved conjectures concerning special values of the zeta function that go back to Stark,
Beilinson and others.

68
7 Computing units and class groups
This section is devoted to the actual computation of Picard groups and unit groups of number
rings. This is actually one problem, since finding class group relations and units is done in
exactly the same way. Moreover, one needs information on the unit group of a number ring
in order to prove that the orders of elements in the Picard group are what they appear to
be.
Both from a theoretical and from an algorithmic point of view, the computations are
by no means easy. Calculations by hand or using only a simple calculator are necessarily
restricted to fields of small degree, usually not exceeding 5, and of fairly small discriminant.
Computer packages as Pari and Magma that have been developed over the last 20 years can
handle slightly larger fields, of degree up to about 20, if the discriminant is of moderate
size. One very soon runs into fundamental problems: the discriminant can be so large that
it becomes impossible to find its factorization, and the units can be so large that simply
writing them down is already not feasible. These are important problems in practice, and
we will have to deal with them when discussing the number field sieve for factoring integers
in section **, which employs number rings of moderate degree but with huge discriminant.
We refer to [21] for a more detailed discussion of these fundamental issues.

I Pell revisited
√
We start with the quadratic order Z[ 1141] occurring in problem 1.1.
√
7.1. Example. Find the unit group R∗ for the real quadratic order R = Z[ 1141].
In this
√
case, we√know by 3.21 that R is an order of index 2 in the ring of integers O =
Z[ 1+ 21141 ] of Q( 1141). It is usually easier to find unit groups and Picard groups of integral
extensions of a number ring, since these have smaller Minkowski constants. We therefore
compute the class group Cl = Cl(O) and the unit group O∗ √ of O first.
α
The ring O is of the form O = Z[α] with α = (1 + 1141)/2 a zero of f = fQ =
2
X −X√ − 285, and its discriminant equals ∆(O) = 1141 = 7 · 163. The Minkowski constant
1
MO = 2 1141 of O is smaller than 17, so Cl is generated by the primes of norm at most 13.
We easily deduce from 3.11 that 2 and 11 are inert in O, so we can restrict our attention to
the primes over 3, 5, 7 and 13. Applying 3.1, we find the following factorizations:

3O = p3 q3 = (3, α) · (3, α − 1)
5O = p5 q5 = (5, α) · (5, α − 1)
7O = p27 = (7, α − 4)2
13O = p13 q13 = (13, α − 4) · (13, α + 3).

In order to determine Cl, we have to find relations between these seven generators. In
line with the ‘linear algebraic’ nature of most class group computations, we write Cl as an
additive group. From the factorizations above we find relations [pi ] + [qi ] = 0 for i = 3, 5, 13
and 2[p7 ] = 0. In order to find additional relations, we factor suitable principal ideals in O
into prime ideals.

69
§7: Computing units and class groups

We usually try to factor ideals of the form (k − α) with k ∈ Z, since in that case the
norm of k − α is easily computed as NK/Q (k − α) = f (k). As we need relations between
prime ideals of small norm, we take values of k for which f (k) is not too large. From the
primes dividing f (k) it is easy to determine which prime ideals occur in the factorization of
(k − α). Note that no rational prime divides k − α, so only primes p of degree f (p/p) = 1 can
occur, and if such a prime occurs the other prime q|p does not occur. Moreover, if p|p divides
(k − α), then it divides all ideals (k 0 − α) with k 0 ≡ k mod p. We compile the following small
table of factorizations in O.
k 15 16 17 18 19 20 21
f (k) −75 −45 −13 21 57 95 135
(k − α) p3 p25 q23 q5 p13 p3 p7 q3 p19 p5 q19 p33 q5

We have written p19 = (19, α) and q19 = (19, α − 1) for the primes over 19. The entry
for k = 17 tells us that p13 is principal. The same is then true for its inverse q13 in Cl.
The entry for k = 18 shows that p7 and p3 are in ideal classes that are inverse to each
other. As we know already that p27 = (7) is principal, the square of p3 is also principal.
As we have [p19 ] + [q19 ] = 0, multiplication of the entries for k = 19 and k = 20 yields
[p5 ] = −[q3 ] = [p3 ]. We deduce that Cl is generated by [p3 ]. From the entry k = 15 we find
[p3 ] + 2[p5 ] = 3 · [p3 ] = 0, and as we know already that 2 · [p3 ] = 0 we conclude that Cl is the
trivial group, i.e., O is a principal ideal domain.
Exercise 1. Find explicit generators for all primes in O of norm at most 13.

In order to find a non-trivial unit in O, we employ the fact that our table has 3 entries
involving only primes over 3 and 5. As we have q3 = 3p−1 −1
3 and q5 = 5p5 , we can write all
factorizations in terms of p3 and p5 as
(15 − α) = p3 p25 (15 + α) = p23 p5 (21 − α)/5 = p33 p−1
5 .

The principal ideal generated by the element

η = (15 − α)a (15 + α)b (21 − α)c 5−c
factors as pa+2b+3c
3 p2a+b−c
5 , and this is the unit ideal if we choose a, b and c satisfying a +
2b + 3c = 2a + b − c = 0. We can take (a, b, c) = (−5, 7, −3) and compute the resulting unit
η = −618715978 − 37751109α. It remains to prove that η is fundamental, so that we have
O∗ = h−1i × h618715978 + 37751109αi.
If ε = 618715978 + 37751109α is not fundamental, there exists a unit ε0 = t + uα ∈ Z[α] and
k ∈ Z>1 such that εk0 = ε. Note that t and u are non-negative since ε is a unit with positive
coefficients on the basis {1, α}. In order to bound the index k, we take the embedding
O → R under which α is positive. For the fundamental unit ε0 = t + uα, we have a trivial
lower bound like ε0 > α or, after checking that t + uα of norm t2 − tu − 285u2 can not be a
unit for u = 1, 2, 3, the slightly better bound ε0 > 4α to find
Reg(ε) log(ε) log(ε)
k= = ≤ ≈ 4.94.
Reg(ε0 ) log(ε0 ) log(4α)

70
 §7: Computing units and class groups

This shows that we have ε = εk0 with k at most 4. It now suffices to show that ε is not a
square or a cube in O, and this can be shown most easily by reducing O modulo suitable
primes and checking that ε does not map to a square or cube. The unit ε is congruent to 1
modulo both p3 and q3 , but modulo p5 we find that ε = 3 ∈ O/p5 = F5 is not a square. Thus
ε is not a square. For cubes we have to look at splitting primes congruent to 1 mod 3, and
one finds that ε is a cube modulo the primes over 7 and 13, but that ε = 16 ∈ O/p19 = F19
is not a cube. This proves that ε is fundamental in O∗ .
The fundamental unit of O does not lie in R = Z + 2O, but (cf. 5.16) the index of R∗ in
O∗ is bounded
√ by the order of (O/2O)∗ = F∗4 . We conclude that ε3 is a fundamental unit in
R = Z[ 1141], and it is an easy calculator check to see that this corresponds to the smallest
solution to x2 − 1141y 2 = 1 that appeared out of the blue in section 1. 
We list the observation occurring in this example in slightly greater generality for the ease
of future reference.
7.2. Proposition. Let f ∈ Z[X] be a monic irreducible polynomial and R = Z[α] the order
obtained by adjoining a root α of f . Let p be a prime number and k ∈ Z an integer. Then
there is a prime ideal p|p dividing the principal ideal (k − α) if and only if p divides f (k). If
such a p exists, it is the ideal p = (p, k − α) of residue class degree f (p/p) = 1. In particular,
there is at most one prime over p dividing (k − α).
Proof. A homomorphism φ : R → F to a finite field F of characteristic p with kernel
ker φ ⊃ (k − α) maps α to k ∈ Fp ⊂ F , so it is uniquely determined and has image Fp .
There exists φ : R → Fp with φ(α) = k mod p if and only if k is a zero of f mod p.

I A cubic example
Our next example is the cubic number ring of the form Z[α, β] that we encountered in 3.7.
This forces us to apply the Kummer-Dedekind theorem 3.1 to more than one polynomial.
√
7.3. Example. Find the class group and the unit group of the maximal order in Q( 3 19).
We determined the maximal order O = Z[α, β] for this field in 3.7, and we√computed its
discriminant ∆(O) = −3 · 192 in 4.11. The Minkowski constant MO = π4 · 276
· 3 · 192 ≈ 9.3
shows that Cl(O) is generated by the primes of norm ≤ 7. From the irreducible polynomials
α β
fQ = X 3 + 19 and fQ = X 3 − X 2 − 6X − 12 one derives by 3.1 the factorizations

2O = p2 p4 = (2, α − 1) · (2, α2 + α + 1)
3O = p23 q3 = (3, β)2 · (3, β − 1)
5O = p5 p25 = (5, α − 1) · (5, α2 + α + 1)
7O = p343 = (7).

From our set of generators {p2 , p4 , p3 , q3 , p5 } we can discard p4 and q3 , since the factorizations
of 2 and 3 yield the relations [p4 ] = −[p2 ] and [q3 ] = −2[p3 ] in Cl(O). We have N (α − 1) =
α
−fQ (1) = −20 = −22 · 5, and it is clear that α − 1 is contained in p2 and p5 . We deduce
that there is a factorization (α − 1) = p22 p5 , and this allows us to discard the generator [p5 ].
β
The element β = (α2 − α + 1)/3 is contained in p3 and p4 and of norm N (β) = −fQ (0) = 12,

71
§7: Computing units and class groups

so we have (β) = p3 p4 and a resulting relation [p3 ] = −[p4 ] = [p2 ]. It follows that Cl(O)
is cyclic and generated by [p2 ]. The factorization of (α + 3) = p32 or (β + 1) = p32 show
that Cl(O) has order 1 or 3. Factoring additional elements yield no new relations: we have
(α − 2) = p3 q23 in the class of −3[p3 ] and (β − 3) = p22 p3 in the class of 3[p3 ]. This suggests
that Cl(O) has order 3.
Showing that an ideal is not principal involves the knowledge of the unit group. In this
case, finding a unit is easy as the 8 elements β + i with i ∈ Z between −4 and 3 and the 5
elements α + j with j ∈ {±1, −2, 3, 4} all factor into primes lying over 2, 3 and 5.
Exercise 2. Find these factorizations, and use them to produce a few units in O. Check how they are
related to the unit η we will now construct.

As α + 3 and β + 1 both generate p32 , the element η = (α + 3)/(β + 1) = 1 − α − β is a unit

in O∗ . Before showing that η is fundamental, let us see how we can apply this to prove that
the order of [p2 ] in the class group is 3.
Suppose that p2 = (x) is a principal ideal. Then x3 and α + 3 generate the same ideal
in O, so we have
α + 3 = ε · x3
for some unit ε ∈ O∗ . We have O∗ = h−1i × hηi by our assumption that η is fundamental,
so ε = ±η k for some k ∈ Z. In order to derive a contradiction, we reduce the equation
modulo a suitable prime of norm p ≡ 1 mod 3. Modulo p19 = (α), we find that β maps to
1/3 = −6 ∈ O/p19 = F19 and η = 1 − α − β to 7 ∈ F19 . As 7 is a cube in F19 , we find that
ε · x3 maps to a cube in O/p19 for all k. However, α + 3 maps to 3, which is not a cube in
F19 . This shows that p2 is not principal, and that Cl(O) has order 3.
It remains to show that η is a fundamental unit in O. In fact, the argument above only
uses that η generates the cyclic group O∗ /(O∗ )3 , and this can again be proved by exhibiting
a prime p in O for which η is not a cube in O/p.
Exercise 3. Check that η maps modulo p97 = (97, α − 60) to the non-cube 54 ∈ F∗97 .

In order to prove that η is fundamental, we need a lower bound on the regulator of O.

2
From problem 5.21, we have Reg(O) ≥ 31 log( 3·194−24 ) ≈ 1.86. Under the unique embedding
O → R we have log(η) ≈ −2.63, so the inequality Reg(η)/ Reg(O) < 2 shows that η is
fundamental.

The method employed in 7.3 to show that an ideal that appears to be non-principal is indeed
of some order k > 1 in the Picard group works in general number rings in the following way.
Suppose we know that ak = (x) is a principal ideal, and that we want to show that [a] has
order k in the Picard group. We need to show that ak/p is non-principal for all prime divisors
p of k. If ak/p = (y) is principal, we have x = ε · y p for some unit ε. As y can be changed
by an arbitrary unit, it suffices to show that this equation cannot hold for ε ranging over a
set of representatives of the cosets of (O∗ )p in O∗ . By the Dirichlet unit theorem, there are
only finitely many cosets, so this yields finitely many equations. Showing that x = ε · y p is
not solvable for fixed ε is done by reducing modulo sufficiently many primes of prime norm
congruent to 1 mod p. If xε−1 is not a p-th power, one is bound to find such a prime after a
finite amount of time. It is often possible to deal with many (or even all) cases at one time

72
 §7: Computing units and class groups

by exhibiting primes modulo which all units are p-th powers, but x is not. This is what we
did in 7.3.

I A larger cubic example

Any cubic field that is not totally real, and for which the unit rank is therefore equal√to 1,
can be dealt with in a way similar to the one we employed for the pure cubic field Q( 3 19).
As the Minkowski constant is larger in this example, we proceed in a more systematic way.

7.4. Example. Find the class group and the unit group of the maximal order in K = Q(α)
with α3 + α2 + 5α − 16 = 0.
Before anything else, we tabulate a few values of the polynomial f = X 3 + X 2 + 5X − 16 in
factored form.

n f (n) n f (n)
−10 −2 · 3 · 7 · 23 0 −24
−9 −709 1 −32
−8 −2 · 32 · 7
3
2 2·3
−7 −3 · 5 · 23 3 5·7
2
−6 −2 · 113 4 2 ·3·7
−5 −3 · 47 5 3 · 53
−4 −22 · 3 · 7 6 2 · 7 · 19
−3 −72 7 3 · 137
−2 −2 · 3 · 5 8 2 · 3 · 52
3

−1 −3 · 7 9 839

This table shows that f has no zeroes modulo 11, 13 and 17, so it is irreducible modulo these
primes. In particular, it is irreducible in Z[X].
Its discriminant can be computed from the resultant R(f, f 0 ) as

∆(f ) = −R(X 3 + X 2 + 5X − 16, 3X 2 + 2X + 5) = −32 R( 28 9

X − 149
9
, 3X 2 + 2X + 5)
= −32 · ( 28 )2 · 3( 149 )2 + 2( 149


9 28 28
) + 5 = −8763 = −3 · 23 · 127.

As ∆(f ) is squarefree, we have ∆K = −8763 and O = OK = Z[α]. As ∆(f ) is negative, the

cubic polynomial f has a single real root. This gives r = s = 1 for our field, so Minkowski’s
constant equals
3! 4 √
MK = 3 8763 ≈ 26.5,
3 π
and the class group is generated by the classes of the primes of norm at most 25. We can
factor the rational primes up to 23 by simply looking at the values of f in our table. Leaving

73
§7: Computing units and class groups

out the inert primes 11, 13 and 17, we obtain factorizations

2O = p2 p4 = (2, α) · (2, α2 + α + 1)
3O = p23 q3 = (3, α + 1)2 · (3, α − 1)
5O = p5 p25 = (5, α + 2) · (5, x5 )
7O = p7 q7 r7 = (7, α + 1)(7, α − 3)(7, α + 3)
19O = p19 p361 = (19, α − 4) · (19, x19 )
23O = p223 q23 = (23, α + 7)2 (23, α + 10)

in which x5 and x19 denote elements which we do not bother to compute. These factorizations
show that the class group is generated by the classes of the primes p2 , p3 , p5 , p19 , p23 and
two of the primes over 7. We can express the classes of the large primes in those of smaller
primes using the factorizations of principal ideals (k −α). In view of 7.2, these can be derived
from the values of f (k) in our table.
The entry k = −7 yields (−7 − α) = p3 p5 p23 , so we can omit [p23 ] from our list of
generators. Similarly, we can omit [p19 ] as the entry k = 6 gives (6 − α) = p2 p7 p19 . The
primes over 7 can be dealt with using the identities (−1 − α) = p3 p7 and (3 − α) = p5 q7 .
The relation (−2 − α) = p2 q3 p5 = 3p2 p−2 3 p5 takes care of [p5 ], and finally (2 − α) = p2 p3
shows that the class group of K is generated by [p2 ]. The order of this class divides 4 since
we have (α) = p42 .
It turns out that we do not find relations indicating that the order of [p2 ] is smaller than
4, so we try to prove this. This comes down to showing that the ideal p22 is not principal.
We need to know the group O∗ /(O∗ )2 in order to show this. As in 7.1, we can produce a
non-trivial unit from the fact that the factorizations of 3, (α), (α − 1) and (α − 2) involve
only p2 and the primes over 3. One deduces that

(α − 1)(α − 2)4
η= = 4α2 + α − 13.
9α
is a unit of norm N (η) = 1. From the Dirichlet unit theorem (with r = s = 1) we have
O∗ ∼= h−1i × P , where P ∼= Z can be taken to be the group of units of norm 1. In order to
prove that η generates P/P 2 , it suffices to show that η is not a square in O∗ . This is easy:
reducing modulo p3 we find η ≡ 4 − 1 − 13 ≡ −1, and −1 is not a square in O/p3 = F3 .
Suppose now that p22 = (y) is principal. Then y 2 and α are both generators of p42 , so
there exists a unit ε with y 2 = ε · α. As the norm N (ε · α) = 16N (ε) = N (y)2 is positive,
we have ε ∈ P . If ε is in P 2 , then α = ε−1 y 2 is a square, contradicting the fact that we
have α ≡ −2 mod p5 . If ε is in ηP 2 , then η · α is a square, and this is contradicted by the
congruence η · α ≡ (4(−2)2 + (−2) − 13) · −2 ≡ 3 mod p5 . We conclude that no unit ε exists,
and that p2 is not principal. It follows that Cl is cyclic of order 4 with generator [p2 ].
The question remains whether η is a fundamental unit. This can be decided as in
7.3. Under the unique real embedding K → R the image of η has absolute logarithm
Reg(η) ≈ 7.684. From the lower bound RK ≥ 31 log( |∆K4|−24 ) ≈ 2.563 we obtain [P : hηi] =
Reg(η)/RK < 3, so η is fundamental if it is not a square in O. We saw this already, so we
have indeed O∗ = h−1i × hηi. 

74
 §7: Computing units and class groups

I An example with unit rank 2

When dealing with number rings of unit rank 2 and higher, a more detailed administration
of units is necessary in order to keep track of the unit group that is generated by the units
we produce from factorizations of principal ideals.
7.5. Example. Find the class group and the unit group of the maximal order in K = Q(α)
with α4 − 2α2 + 3α − 7 = 0.
Note first that the polynomial f = X 4 −2X 2 +3X −7 ∈ Z[X] is irreducible as it is irreducible
modulo 2. We begin again by tabulating a list of small values of f in factored form.

n f (n) n f (n)
−15 50123 1 −5
−14 5 · 72 · 31
2
2 7
−13 19 · 1483 3 5 · 13
−12 5 · 7 · 11 · 53 4 229
−11 83 · 173 5 11 · 53
−10 13 · 751 6 5 · 13 · 19
−9 5 · 19 · 67 7 7 · 331
−8 31 · 127 8 5 · 797
2
−7 5 · 7 · 13 9 72 · 131
−6 11 · 109 10 11 · 19 · 47
−5 7 · 79 11 52 · 577
−4 5 · 41 12 20477
−3 47 13 5 · 5651
−2 −5 14 7 · 5437
−1 −11 15 149 · 337
0 −7 16 5 · 7 · 11 · 132

For a change, we compute the discriminant of f from the power sums pk of the roots of f .
We have p0 = 4 and p1 = 0, and from Newton’s formula’s (exercise 4.20) we find

p2 = −2a2 − p1 a1 = −2 · 2 + 0 = 4
p3 = −3a3 − p2 a1 − p1 a2 = 3 · (−3) + 0 + 0 = −9
p4 = 2p2 − 3p1 + 7p0 = 2 · 4 − 0 + 7 · 4 = 36
p5 = 2p3 − 3p2 + 7p1 = 2 · (−9) − 3 · 4 + 0 = −30
p6 = 2p4 − 3p3 + 7p2 = 2 · 36 − 3 · (−9) + 7 · 4 = 127.
The discriminant is then by exercise 4.19 equal to the determinant
−9
 
4 0 4
0 4 −9 36 
 4 −9 36 −30 = −98443,
∆(f ) = det  

−9 36 −30 127

75
§7: Computing units and class groups

which is a prime number. This implies ∆K = −98443 and O = OK = Z[α]. We deduce from
the sign of ∆K that K has r = 2 real embeddings and s = 1 pair of complex embeddings
(cf. exercise 4.12). Minkowski’s constant is then equal to
4! 4 √
98443 ≈ 37.4,
44 π
so the class group is generated by the classes of the primes of norm at most 37. The table
shows that f has no zeroes modulo p for the primes p = 2, 3, 17, 23 and 29. Computing a
few additional values we see that f has no zeroes modulo 37 either. This implies that there
are no prime ideals of norm p for these primes p. It is easily checked that f is irreducible
modulo 2 and 3 and that it factors modulo 5 as

f ≡ (X − 1)(X + 2)(X 2 − X + 1) (mod 5).

This implies that 2 and 3 are inert in K and that 5 factors as a product (5) = p5 q5 p25 , where
p5 and q5 have norm 5 and p25 is a prime of norm 25. All other primes of norm less than
the Minkowski bound have prime norm, so they can be found from our table.

p5 = (α − 1) q5 = (α + 2)
p7 = (α) q7 = (α − 2)
p11 = (α + 1) q11 = (11, α − 5)
p13 = (13, α − 3) q13 = (13, α − 6)
p19 = (19, α − 6) q19 = (19, α + 9)
p31 = (31, α + 8) q31 = (31, α + 14)

Here we use the fact that (p, α − k) is already generated by (α − k) when f (k) = ±p.
The class group is generated by the classes of the primes in this table and the class [p25 ].
The primes lying over 5 and 7 are all principal (note that we have p25 = 5p−1 −1
5 q5 ), and so
is p11 . This suggests strongly that Cl(O) is trivial. In order to prove this, we try to express
all primes in the table in terms of the principal ideals. From the entry k = 3 in our table we
obtain (3 − α) = p13 q5 , showing that p13 is principal. The relation (16 − α) = p5 q7 q11 p213 then
shows that q11 is also principal. Similarly, we have principality of q13 from (−7−α) = q25 p7 q13
and of p19 from (6 − α) = p5 q13 p19 . Finally, we use (−14 − α) = p25 p27 q31 to eliminate q31 .
This exploits all useful relations from our table, leaving us with the primes q19 and p31 .
In order to prove that these primes are also principal, we factor a small element in them.
Modulo q19 = (19, α + 9) we have α = −9 ∈ F19 and 1 − 2α is therefore a small element
in the ideal. Similarly, we have α = −8 ∈ F31 when working modulo p31 = (31, α + 8), so
1 + 4α is in p31 . The norms of these elements are N (1 − 2α) = 24 f (1/2) = −5 · 19 and
N (1 + 4α) = (−4)4 f (−1/4) = 5 · 13 · 31, which implies that q19 and p31 are principal. The
corresponding explicit factorizations are (1 − 2α) = q5 q19 and (1 + 4α) = p5 p13 p31 . This
proves that Cl(O) is trivial.
At this stage, we have produced explicit generating elements for all prime ideals of norm
below the Minkowski bound. Although we do not need all of these generators, we list them
for completeness sake.

76
 §7: Computing units and class groups

p5 = (α − 1) q5 = (α + 2)
p7 = (α) q7 = (α − 2)
p11 = (α + 1) q11 = (32α3 + 53α2 + 25α + 138)
p13 = (α3 − 2α2 − 2α − 2) q13 = (2α3 − 4α2 + 5α − 6)
p19 = (α3 + α2 + α + 8) q19 = (α3 − 2α2 + 2α − 3)
p31 = (2α3 + 3α2 + 2α + 11) q31 = (α3 + α2 + 6)

These generators are not necessarily the ‘smallest’ or ‘most obvious’ generators of the ideals
in question, they happen to come out of the arguments by which we eliminated all generators
of the class group. The search for units that is to follow will provide other generators, and one
can for instance check that the large coefficients of our generator for q11 are not ‘necessary’
as we have q11 = (α2 − 3).
From now on every further factorization of a principal ideal (x) as a product of primes
in this table will give us a unit in O, since both x and a product of generators from our
table generate (x). This means that their quotient is a unit. Trying some elements a + bα,
for which we can easily compute the norm, one quickly generates a large number of units.
The rank of the unit group O∗ for our field K equals r + s − 1 = 2, so some administration
is needed to keep track of the subgroup of O∗ generated by these units. As in the proof
of the Dirichlet unit theorem, one looks at the lattice in R2 generated by the ‘log-vectors’
L(u) = (log |σ1 (u)|, log |σ2 (u)|) for each unit u. Here σ1 and σ2 are taken to be the real
embeddings K → R, so they send α to the the real roots α1 ≈ −2.195 and α2 ≈ 1.656 of f .
The following units are obtained in this way.

relation u L(u)
(2α + 1) = q11 q13 α3 − 2α2 + 3α − 4 (3.4276, −3.7527)
(2α − 3) = q31 α3 − 2α2 + 3α − 4 (3.4276, −3.7527)
(2α + 3) = p25 q7 −3α3 − 5α2 − 2α − 12 (−3.4276, 3.7527)
(3α + 1) = q5 q7 p19 5α3 − 11α2 + 14α − 16 (5.0281, −1.2731)
(3α − 5) = q13 α3 − 4α + 2 (−1.6005, −2.4796)
(3α − 4) = q25 q11 −4743α3 + 10412α2 − 13371α + 15124 (11.8833, −8.7785)
(4α − 7) = q5 p7 p11 −α3 + 2α2 − 3α + 4 (3.4276, −3.7527)

From the table we see that

η1 = α3 − 2α2 + 3α − 4 and η2 = α3 − 4α + 2

are likely to be fundamental. This impression can be confirmed by factoring more elements
of small norm. Note that the log-vectors are very useful to make a relation like

5α3 − 11α2 + 14α − 16 = ±(α3 − 2α2 + 3α − 4)(α3 − 4α + 2)−1 = ±η1 η2−1

that is otherwise non-obvious immediately visible.

Exercise 4. Express the unit −4743α3 + 10412α2 − 13371α + 15124 in terms of η1 and η2 .

77
§7: Computing units and class groups

In order to prove that the unit group is indeed equal to O∗ = h−1i × hη1 i × hη2 i, we have to
check that the regulator of K is equal to
   
log |σ1 (η1 )| log |σ1 (η2 )| 3.4276 −1.6005
Reg(η1 , η2 ) = det ≈ det ≈ 14.506.
log |σ2 (η1 )| log |σ2 (η2 )| −3.7527 −2.4796
There are two ways to proceed. The first way is to bound the index [O∗ : h−1, η1 , η2 i] =
Reg(η1 , η2 )/RK as in the previous examples, and show modulo suitable primes of O that η1
and η2 generate O∗ /{±1} modulo p-th powers for all primes p below the index bound. As f
has degree 4, we have to use the lower bound (6.6) for RK to achieve this.
The second way, which is more efficient and generally shows that RK and hK equal
the values obtained, proceeds numerically by approximating the residue in t = 1 of the zeta
function ζK (t) of K. If η1 and η2 are fundamental, the residue should equal
2r (2π)s hK R(η1 , η2 ) 22 (2π) · 1 · 14.506
p ≈ √ ≈ 0.5810.
wK |∆| 2 · 98443
From the previous section, we know that this residue can be computed from the Euler
Q
product p E(p), where
−1
Q
p|p (1 − NK/Q (p) )
E(p)−1 = .
1 − p−1
The factor E(p)−1 is a polynomial expression in p−1 that depends only on the residue class
degrees of the primes p|p, i.e. on the degrees of the irreducible factors of the defining poly-
nomial f modulo p. In this case, it is not even necessary to factor f ∈ Fp [X] to determine
the degrees of the irreducible factors: it suffices to count the number of zeroes of f in Fp [X].
If we disregard the single ramified prime 98443, there are 5 possible factorization types of f
modulo p. If the number np of zeroes of f mod p equals 4, 2, or 1 we immediately know the
degree of all irreducible factors of f mod p. For np = 0 the polynomial f is either irreducible
modulo p or a product of two quadratic irreducibles. We can distinguish the two cases for
np = 0 using exercise 15, which states that the parity of the number of irreducible factors of
∆(f )

f mod p can be read off from the Legendre symbol p . It follows that we have
(1 − p−1 )3


 if np = 4;
−1 −2
 (1 − p )(1 − p ) if np = 2;



−1
E(p) = 1 − p−3 if np = 1;
−1 −2 ∆(f )

(1 + p )(1 − p ) if n = 0 and = 1;

p


 p
−1 −2 −3 ∆(f )


1+p +p +p if np = 0 and p = −1.


A simple computer program enables us to evaluate the Euler product with some precision.
The following data indicate the speed of convergence of this product.
Q Q
N p<N E(p) N p<N E(p)
100 0.625211 5000 0.579408
200 0.595521 10000 0.579750
500 0.581346 20000 0.581892
1000 0.584912 50000 0.581562
2000 0.585697 100000 0.581423

78
 §7: Computing units and class groups

We see that the convergence is non-monotonous and slow, but certainly close to the value
.5810 we found before. If our units η1 and η2 were not fundamental, the Euler product should
be at least twice as small as 0.5810, which is highly unlikely. This convinces us that we have
found the full unit group. A rigorous proof would involve a more detailed analysis on the
convergence of the Euler product.

Exercises.

5. Find the smallest integral solution y > 0 to the Pell equation x2 − 61y 2 = 1.
6. Find the smallest integral solution y > 0 to the Pell equation x2 − 109y 2 = 1.
√
7. Compute the class group and the unit group of the maximal order in Q( 229).
8. Find the fundamental units of the number rings Z[α] and Z[β] contained in the order O of
example 7.3.
9. Determine for the twelve prime ideals p in example 7.4 of norm at most 25 the class of
[p] ∈ Cl(O) = hp2 i.
√
10. Compute the Picard group and the unit group of the order Z[ 3 7].
√
11. Compute the Picard group and the unit group of the order Z[ 3 37].
√
12. Compute the Picard group and the unit group of the order Z[ 3 −19].
[Hint: use 4.13 and the fact that the quadratic subring A ⊂ R
e of discriminant −19 has class
number 1.]
13. Pick integers a, b, c and d of absolute value at most 4 such that f = X 4 + aX 3 + bX 2 + cX + d
is irreducible in Z[X]. Compute the class group and the unit group of the field Q[X]/(f ).
14. Let F be a field and f ∈ F [X] a separable polynomial. Write n = deg f , and denote by α1 ,
α2 , . . . , αn the zeroes of f in a splitting field of f over F . We put
n
X Y X n
Y
e(f ) = i−1
ασ(i) , e0 (f ) = i−1
ασ(i) ,
σ∈An i=1 σ∈Sn −An i=1

where Sn is the symmetric group of degree n and An ⊂ Sn the alternating group. We view
the Galois group Gal(f ) of f over F as a subgroup of Sn , via its action on α1 , α2 , . . . , αn .
a. Prove that (X −e(f ))(X −e0 (f )) ∈ F [X] and that (e(f )−e0 (f ))2 equals the discriminant
of f . Deduce that e(f ) 6= e0 (f ).
b. Show that e(f ) and e0 (f ) belong to F if and only if Gal(f ) ⊂ An .
c. Prove that for F of characteristic different from 2, the discriminant of f is a square in F
if and only if Gal(f ) ⊂ An .
d. Prove that for F = F2 , one has e(f )e0 (f ) = 0 if and only if Gal(f ) ⊂ An .
e. Suppose that F is finite, and denote the number of irreducible factors of f in F [X] by t.
Prove that Gal(f ) ⊂ An if and only if n ≡ t mod 2.
15. Let f ∈ Z[X] be a monic irreducible polynomial, n its degree, α a zero of f in some extension
field of Q, and p a prime number not dividing the discriminant ∆(f ) of f . Denote by t the
number of prime ideals p of Z[α] with p ∈ p. Prove that ∆(f )
n−t . (Here ∆(f )


p = (−1) p
denotes the Kronecker symbol if p = 2.)

79
§7: Computing units and class groups

16. Let O = Z[α] with α3 + α + 1 = 0 be the order of example 3.3, and p a prime number. Show
that p is the product of two prime ideals in O if −31


p = −1, and that pO is either inert or
the product of three prime ideals if p = 1. Do both possibilities for −31
−31



p = 1 occur?
Determine the 5 smallest primes that split completely in O.
√
17. Show that the maximal order in Q( −31) has class number 3. Show that a rational prime p
has a principal extension of norm p in this order if and only if p is of the form p = x2 + 31y 2 .
Determine the 5 smallest primes p of this form. *Can you explain the relation with the
previous exercise?

80
8 Galois theory for number fields
In the previous section, we have seen how to compute, starting from an irreducible polynomial
f ∈ Z[X], the fundamental arithmetic invariants of the number field K = Q[X]/(f ) defined
by f . A basic algorithmic tool here is the Kummer-Dedekind theorem, which tells us how
to derive the explicit splitting of a rational prime p in K (more precisely: in OK ) from the
splitting of f modulo p.
In Galois!theory, a theory (or theorem) of which we assume the basic notions to be
known, one learns that the polynomial f gives rise to a finite group Gal(f ), the automorphism
group of the splitting field ΩfQ of f over Q. This section discusses the interplay between the
Galois group Gal(f ) and the splitting of rational primes p in K and in ΩfQ . It will ultimately
lead us to various results describing the ‘average splitting behavior’ of f mod p if f is fixed
and p varies.

I Galois action on extension primes

Let K be a number field that is Galois over Q with group G = Gal(K/Q). Then G
acts naturally on all intrinsically defined objects and invariants related to K that we have
encountered in the previous sections.
Exercise 1. Check this for OK and Cl(K), and for the set of primes in OK extending a rational prime p.

For K as above, any choice of an embedding φ ∈ HomQ (K, Q) leads to an identification

G ↔ HomQ (K, Q) given by σ 7→ φ ◦ σ. In particular, we can use this identification in 4.2 to
rewrite the products and sums defining norms and traces as ranging over all σ ∈ G.

8.1. Theorem. Let K be a number field that is Galois over Q with group G, and p a
rational prime. Then G acts transitively on the primes p of K extending p.

Proof. Suppose there exist extensions p and p0 of p that are in different G-orbits. Then we
can use the Chinese remainder theorem to construct an element x ∈ p satisfying x ∈ / σp0 for
all σ ∈ G. The norm NK/Q (x) = σ∈G σ(x) is then by construction in p but not in p0 . As
Q

it lies in p ∩ Z = pZ = p0 ∩ Z, we arrive at a contradiction.

8.2. Corollary. All extensions p of p in K are isomorphic, and the residue class degree
fp = f (p/p) and the ramification index ep = e(p/p) only depend on p. If gp is the number
of extensions of p in K, we have
ep fp gp = [K : Q].
√
8.3. Example. Take K = Q(ζ3 , 3 −19) to be the splitting field of X 3 + 19. Then K is
Galois over Q with non-abelian Galois group of order 6. The prime 3 ramifies in the quadratic
√
subfield Q(ζ3 ) = Q( −3), so the primes in K over 3 have even ramification index e3 . As 3
√
has two extensions to Q( 3 −19) by 3.7, we have g3 ≥ 2. From e3 f3 g3 = 6 we find e3 = 2,
f3 = 1 and g3 = 3. This shows without any explicit computation that the primes occurring
√ √
in the factorization (3) = p2 q in Q( 3 −19) factor in the quadratic extension K of Q( 3 −19)
as pOK = P1 P2 and qOK = Q2 .

81
§8: Galois theory for number fields
√
The prime 2 is inert in Q(ζ3 ) and splits as (2) = p2 p4 in Q( 3 −19). For this prime f2 is
even and g2 is at least 2, so we have e2 = 1, f2 = 2 and g2 = 3. We conclude that p2 is inert
√
in Q( 3 −19) ⊂ K, giving rise to a prime P of norm 4, and that p4 splits into two primes Q
and R of norm 4 each.

I Decomposition and inertia groups

In the situation of 8.1, the stabilizer

Gp = {σ ∈ G : σp = p} ⊂ G

of a prime p of K extending p is known as the decomposition group of p. As G acts transitively

on the primes extending p, the G-set G/Gp of left cosets of Gp in G may be identified with
the set of extensions of p to K. By 8.2, the order of Gp equals ep fp .
If p0 is another extension of p to K, we have p0 = σp for some σ ∈ G and consequently
Gp0 = Gσp = σGp σ −1 . This shows that the various decomposition groups of the primes above
a rational prime p are all conjugate in G. If G is abelian, or, more generally, if Gp is normal
in G, the decomposition group is independent of the choice of the extension p|p and can be
denoted by Gp .
The decomposition group Gp acts naturally as a group of automorphisms of the residue
class field extension Fp = Z/pZ ⊂ kp = OK /p, which is an extension of finite fields of
degree f (p/p). In particular, Fp ⊂ kp is a cyclic Galois extension of order f (p/p) with group
generated by the Frobenius automorphism Frobp : x 7→ xp on kp .
8.4. Lemma. The natural map Gp → Gal(kp /Fp ) is surjective.
Proof. Choose an element x ∈ OK that is contained in all primes p0 6= p over p and for
which the reduction x = x mod p satisfies Fp (x) = kp . Then σ(x) is in p for all σ ∈ G\Gp , so
the characteristic polynomial f = σ∈G (X − σ(x)) of x factors in kp [X] as f = X #(G−Gp ) · hp
Q

for some polynomial hp ∈ kp [X]. As f is in Fp [X] we also have hp ∈ Fp [X], and the identity
f (x) = 0 = hp (x) implies that the irreducible polynomial of x over Fp divides hp . This shows
that all conjugates of x over Fp can be obtained as the reduction of an element σ(x) with
σ ∈ Gp , and that every element of Gal(kp /Fp ) comes from some σ ∈ Gp .
The kernel of the map Gp → Gal(kp /Fp ) is the inertia group Ip ⊂ Gp of p in G. It is a
normal subgroup of Gp , and we have an exact sequence

(8.5) 0 −→ Ip −→ Gp −→ Gal(kp /Fp ) −→ 0

showing that the group Gp of order e(p/p)f (p/p) has a normal subgroup Ip of order e(p/p)
for which the quotient Gp /Ip is cyclic of order f (p/p).
√
8.6. Example. Let K = Q(ζ3 , 3 19) be as in example 8.3. Then the decomposition groups
at the primes over 3 in Gal(K/Q) are the three subgroups of √
order 2. Note that these are
3
conjugate subgroups, and that we have GQ = IQ = Gal(K/Q( 19)).
√
The prime 19 has two extensions (4 ± −3) in Q(ζ3 ) that are both totally ramified in
K/Q(ζ3 ). Their decomposition group G19 = I19 is the normal subgroup of order 3. It does
not depend on the choice of the extension.

82
 §8: Galois theory for number fields

The decomposition groups of the primes

√ over 2 are again the three subgroups of order 2.
3
The prime P in 8.3 has GP = Gal(K/Q( 19)).

Everything we have done so far for the Galois extension Q ⊂ K can immediately be
generalized to an arbitrary Galois extension K ⊂ L of number fields. The Galois group
G = Gal(L/K) acts transitively on the set of primes q of L extending a prime p of K as in
8.1, and the stabilizer of q in G is the decomposition group Gq = Gq/p of q over p. It acts
on the residue class field extension kp ⊂ kq , and as in 8.5 we obtain an exact sequence

(8.7) 0 −→ Iq/p −→ Gq/p −→ Gal(kq /kp ) −→ 0

showing that Gq/p is the extension of a cyclic group of order f (q/p) by the inertia group Iq/p
of order e(q/p).
Exercise 2. Formulate and prove the analogues of 8.1 and 8.4 for the extension K ⊂ L.

If H is a subgroup of G = Gal(L/K) corresponding to the intermediate field E and qE is

the restriction of q to E, then it is immediate from the definitions that the decomposition
and inertia group of q in the Galois extension E ⊂ L equal

Hq = Gq ∩ H and Iq/qE = Iq ∩ H.

Moreover, if H is normal in G, the extension K ⊂ E is Galois with group G/H and the
natural map Gq/p → G/H induces isomorphisms
∼
Gq /Hq = Gq/p /(Gq/p ∩ H) −→ (G/H)qE /q
∼
Iq/p /Iq/qE = Iq/p /(Iq/p ∩ H) −→ IqE /q .

To see this, it suffices to observe that both maps are clearly injective, and therefore surjective
by the transitivity relations f (q/p) = f (q/qE )f (qE /p) and e(q/p) = e(q/qE )e(qE /p).
The subfields LGq and LIq corresponding to the decomposition and inertia groups of q
over p are the decomposition field (‘Zerlegungskörper’) and the inertia field (‘Trägheitskör-
per’) of q in K ⊂ L. By the functorial properties of decomposition and inertia groups, we
see that the inclusions G ⊃ Gq ⊃ Iq ⊃ 1 lead to a tower of fields that relates to the extension
behavior of q in the following way:

K ⊂ LGq ⊂ LIq ⊂ L
residue class degree 1 f (q/p) 1
ramification index 1 1 e(q/p)

One finds that the decomposition field LGq is the maximal subextension E of K ⊂ L for
which we have e(qE /p) = f (qE /p) = 1, and the inertia field LIq is the maximal subextension
E of K ⊂ L for which qE is unramified over p.
Exercise 3. Prove these statements, and describe the respective subgroups of G corresponding to the
maximal subextension of K ⊂ L for which p is totally split (unramified) in K ⊂ E.

83
§8: Galois theory for number fields

I The Frobenius symbol

If q is unramified over p in 8.7, then the inertia group Iq/p is trivial and we see that Gq/p itself
has a ‘Frobenius of q over p’ as its canonical generator. This automorphism is the Frobenius
symbol Frq ∈ Gq/p ⊂ G of p in G, and it is characterized by the identity Frq (x) = x#kp mod q
for all x ∈ OL . If q0 = σq is some other prime over p, we have Frσq = σ Frq σ −1 , so the
Frobenius elements at the primes over p are conjugate elements in G. If G is abelian, we can
speak of the Frobenius element of p in G. This element, which is also known as the Artin
symbol of p in G, plays a key role in class field theory.

8.8. Example. The cyclotomic field K = Q(ζn ) is Galois over Q with abelian Galois group
(Z/nZ)∗ , with a ∈ (Z/nZ)∗ corresponding to the automorphism of K defined by ζn 7→ ζna .
Let p be a prime number, and write n = pk m with p - m. Then Q(ζm ) is the inertia
field of p, since p is unramified in Q(ζm ) with extensions that are totally ramified in Q(ζm ) ⊂
Q(ζn ). In particular, we have

Ip = (Z/pk Z)∗ × (1 mod m) ⊂ (Z/pk Z)∗ × (Z/mZ)∗ = (Z/nZ)∗ .

If p does not divide n, we have m = n and p is unramified in Q(ζn ). In this case the Frobenius
element Frp in (Z/nZ)∗ satisfies Fr(ζn ) ≡ ζnp modulo all primes over p. As Φn is separable in
characteristic p - n, this shows that we have Frp (ζn ) = ζnp and Gp = hp mod ni ⊂ (Z/nZ)∗ .
The prime p splits completely in Q(ζn )Gp —this explains the name decomposition field—and
its extensions remain inert in the extension Q(ζn )Gp ⊂ Q(ζn ).
√
Exercise 4. Show that the prime 2 has decomposition field Q( −7) and inertia field Q(ζ7 ) in Q(ζ28 ).
Determine the decomposition fields of the primes 3, 5, 7, 13 and 29, and show that no prime is inert in
Q ⊂ Q(ζ28 ).

I The non-Galois case

Even if the extension Q ⊂ K one is dealing with is not Galois, one can effectively apply
Galois theory to obtain the splitting behavior of primes. This is because there is a Galois
action on the fundamental set XK = Hom(K, Q) of embeddings of K in an algebraic closure
Q of Q for every number field K. The set XK has n = [K : Q] elements, and the images
σ[K] ⊂ Q for σ ∈ XK generate the normal closure L of K in Q. The extension Q ⊂ L is
Galois, say with group G, and the natural left action of G given by composition is transitive.
If K = Q(α) is generated over Q by a root α of f ∈ Z[X], then L is the splitting field of
the polynomial f in Q, and the natural action of G = Gal(f ) = Gal(L/Q) on the n-element
set XK identifies G with a subgroup of Sn . As the map σ 7→ σ(α) identifies the elements
σ ∈ XK with the zeroes of f in Q, this is nothing but the classical Galois action of Gal(f )
on the n roots of f in Q.

8.9. Theorem. Let XK be the fundamental set of a number field K of degree n and
G = Gal(L/Q) the Galois group over Q of the normal closure L of K. Given integers
ei , fi > 0 for i = 1, 2, . . . , t such that ti=1 ei fi = n, the following are equivalent:
P

(1) the prime p has t distinct extensions p1 , p2 , . . . , pt in K with ramification indices

e(pi /p) = ei and residue class field degrees f (pi /p) = fi ;

84
 §8: Galois theory for number fields

(2) for every decomposition group Gq ⊂ G of a prime q above p in L, there are t different
Gq -orbits Xi ⊂ XK of length #Xi = ei fi . Under the action of the inertia group Iq ⊂ Gq
on Xi , there are fi orbits of length ei each.
Proof. **to be supplied**
8.10. Corollary. A prime number p is unramified (totally split) in K if and only if it is
unramified (totally split) in the normal closure of K.
Proof. A subgroup H ⊂ G = Gal(L/Q) that acts trivially on XK is necessarily trivial as
the normal closure L is generated by the subfields σ[K] with σ ∈ XK . Now apply this with
H the inertia (decomposition) group of a prime q over p in L.
8.11. Corollary. Let f ∈ Z[X] be a monic irreducible polynomial and p a prime num-
ber. Suppose that f mod p factors as a product of k distinct irreducible factors of degrees
d1 , d2 , . . . , dk . Then Gal(f ), viewed as a permutation group on the roots of f , contains a
permutation that is the product of k disjoint cycles of lengths d1 , d2 , . . . , dk .
Proof. The hypothesis means that f mod p is separable, and that p does not divide ∆(f ).
The number ring Z[α] = Z[X]/(f ) obtained by adjoining a root of f to Z is then regular
and unramified at p. By the Kummer-Dedekind theorem 3.1, the primes extending p in the
ring of integers of K = Q(α) have residue class degrees d1 , d2 , . . . , dk . If Frq is the Frobenius
element of a prime over p in G = Gal(f ), then the lengths of the disjoint cycles of Frq are
the lengths of the orbits under the action of Gq = hFrq i ⊂ G on XK or – equivalently – the
roots of f . By 8.9, these lengths are d1 , d2 , . . . , dk .
8.12. Example. Take f = X 4 + X + 1. Then f is irreducible modulo 2 and factors as a
linear times a cubic polynomial modulo 3. It follows that Gal(f ) is a permutation group on
4 elements containing a cycle of length 4 and a cycle of length 3. We immediately deduce
that Gal(f ) is the full symmetric group S4 of order 24.

Exercises.
5. For each residue class a ∈ (Z/20Z)∗ , determine the decomposition group Gp of a prime
p ≡ a mod 20 in Gal(Q(ζ20 )/Q) and the corresponding decomposition field.
6. Determine the decomposition and inertia fields for the primes p < 20 in the splitting field of
f = X 4 − 19.
7. Let f ∈ Z[X] be monic and irreducible of degree n ∈ Z≥1 . Show that Gal(f ), when viewed
as a permutation group on the roots of f , is contained in An if and only if ∆(f ) is a square
in Z.
8. Let Q ⊂ K be Galois with group G. Show that G is generated by the inertia groups of the
primes of OK .
9. Determine the Galois groups of the polynomials X 4 + 3X + 1 and X 4 + 3X 2 + 1.
10. Determine the isomorphism types of Galois groups of irreducible polynomials f ∈ Z[X] of
degree 4 that exist, and find a polynomial with that Galois group for each type.
11. Determine the isomorphism types of Galois groups of irreducible polynomials f ∈ Z[X] of
degree 5 that exist, and find a polynomial with that Galois group for each type.

85
Literature
There are several texts covering a large part of the material in these notes. We list a few of
them, roughly in ascending order of difficulty, and indicate their most striking features.
1. I. Stewart, D.O. Tall, Algebraic Number Theory, Chapman and Hall 1979. Second
edition 1987.
A very readable, somewhat elementary account. Many examples, exercises and moti-
vating remarks.
2. K. Ireland, M. Rosen, A classical introduction to modern number theory, Springer
GTM 84, 1982. Second edition, 1990.
Short and very readable chapters. Contains many attractive special topics not covered
in our notes.
3. P. Samuel, Algebraic theory of numbers, translation of Théorie algébrique des nombres
(1967), Hermann, 1970.
A very clear and logical presentation in the style of Bourbaki.
4. A. Fröhlich, M.J. Taylor, Algebraic Number Theory, Cambridge University Press, 1991.
Classical in spirit, with special attention to explicit computations in number fields of
low degree.
5. Z. I. Borevich, I. R. Shafarevich, Number theory, translation of Teoria čisel (1964),
Academic Press 1967.
An unconventional classic.
6. J. Neukirch, Algebraische Zahlentheorie, Springer, 1992.
A modern account stressing the analogy between algebraic number theory and algebraic
geometry that has given rise to the fruitful area of arithmetic algebraic geometry.
Contains class field theory in Neukirch’s own axiomatic set up and an extensive chapter
on zeta functions and L-series.
7. S. Lang, Algebraic number theory, Addison Wesley, 1970. Reprinted by Springer (GTM
110, 1986).
More succinct than all preceding texts, no exercises. Contains class field theory and a
lot of analytic number theory as well.
8. J. W. S. Cassels, A. Fröhlich (eds), Algebraic Number Theory, Academic Press, 1967.
Proceedings of a 1965 instructional conference. The first chapters provide a concise
introduction to algebraic number theory. Cohomological class field theory is found in
the later chapters.

References
9. M. F. Atiyah, I. G. Macdonald, Introduction to commutative algebra, Addison-Wesley,
1969.
10. N. Bourbaki, Elements of the history of mathematics, Springer, 1994?. Translation of
Éléments d’histoire des mathématiques, Masson, Paris, 1984.

86
 Literature

11. J. Buchmann, H. W. Lenstra, Jr., Computing maximal orders and decomposing primes
in number fields.
12. G. Cornell, J. H. Silverman, Arithmetic Geometry, Springer, 1986.
13. J. Dieudonné, Abrégé d’histoire des mathématiques, 1700–1900, Hermann, nouvelle
édition 1986.
14. Diophantus, Arithmetica.
15. D. Eisenbud, Commutative algebra with a viewpoint towards algebraic geometry,
Springer GTM 153, 1995.
16. R. Hartshorne, Algebraic Geometry, Springer GTM 52, 1977.
17. D. Hilbert, Mathematische Probleme, Gesammelte Abhandlungen Band III, 290–329,
Springer Verlag, 1970.
18. Anthony W. Knapp, Elliptic Curves, Princeton University Press 1992.
19. S. Lang, Algebra, Addison-Wesley. Third edition 1993.
20. A. K. Lenstra, H. W. Lenstra, Jr. (eds), The development of the number field sieve,
Springer LNM 1554, 1993.
21. H. W. Lenstra, Jr., Algorithms in algebraic number theory, Bull. AMS 26(2), 211–244
(1992).
22. Yu. I. Manin, A course in mathematical logic, Springer GTM 53, 1977.
23. J.-P. Serre, Minorations de discriminants, Collected Works, vol III, 240–243, Springer,
1986.
24. J. H. Silverman, The Arithmetic of Elliptic Curves, Springer GTM 106, 1986.
25. J. Stilwell, Mathematics and its history, Undergraduate Text in Mathematics, Sprin-
ger, 1989.
26. B. L. van der Waerden, Moderne Algebra, 1930-31. Later editions: Algebra I&II.
English translation edited by Springer, 1991.
27. A. Weil, Number Theory, an approach through history; from Hammurapi to Legendre,
Birkhäuser, 1984.
28. A. Wiles, Modular elliptic curves and Fermat’s last theorem, Annals of Mathematics
142, 443–551 (1995).

87
Index
Γ-function, 68 conductor, 49, 51
I(R), 17 contraction, 19
µR , 58 coprime
OK , 38 elements, 8, 9, 15
P(R), 17 ideals, 15, 16
p-adic numbers, 29 polynomials, 33
p-primary part, 22 critical strip, 68
R-ideal, 15 CRT, see Chinese remainder theorem
Rp , 18 curve
S −1 R, 18 elliptic, 10
v-adic plane cubic, 7
completion, 28 cyclotomic
metric, 28 field, 46
topology, 28 number ring, 15, 35
Z[i], 6 polynomial, 35

absolute norm and trace, 41 decomposition

algebraic integer, 40 field, 83, 84
analytic class number formula, 67 group, 82
approximation theorem, 27 Dedekind
Artin symbol, 84 domain, 24
Artin-Dedekind theorem, 51 zeta-function, 66
desingularization, 34
base change, 41 different, 48
basis Diophantine equation, 4, 10
dual, 44 universal theory of, 10
integral, 45 Diophantus of Alexandria, 4
power, 43, 44 Dirichlet unit theorem, 6, 9, 11, 58
bilinear form discrete valuation, 23
non-degenerate, 44 discrete valuation ring, 31, 38
positive definite, 52 discriminant, 43
binary quadratic form, 9 lower bound, 57
of number field, 45, 68
Cayley-Hamilton theorem, 41
of order, 43
chain of prime ideals, 21
of polynomial, 46
characteristic polynomial, 41
dual basis, 44
Chinese remainder theorem, 15, 47, 81
class field theory, 84 element
class group, 9, 24, 55 coprime, 8, 9, 15
relations, 69 Frobenius, 84
class number, 55 integral, 36
analytic formula, 67 irreducible, 9
closure elliptic curve, 10
integral, 37 Euclidean
normal, 84 algorithm, 6, 7
CM-field, 63 function, 10, 11
co-different, 48 ring, 12, 63, 65
co-volume, 52, 60 space, 9, 52
complete lattice, 52 Euler
computer φ-function, 35
equipment, 6 factor, 67, 78
packages, 69

88
 Number rings – Index

identities, 51 extending, 30
product, 66, 78 fractional, 16
extension principal, 16
integral, 39 integral, 16
integrally closed, 36 intersection of, 15
primitive, 42 inverse of, 17
separable, 42 invertible, 15, 17, 18
simple integral, 30, 51 locally principal, 18
lying over, 30
Faltings, G., 10 non-principal, 16, 72
Fermat norm of, 30
equation, 15 principal, 15
last theorem, 10 product of, 15
field extension proper, 17
primitive, 42 quotient of, 16
separable, 42 sum of, 15
fractional ideal, 16 ideale Zahlen, 15
principal, 16 inert prime, 31
Frobenius inertia
automorphism, 82 field, 83
element, 84 group, 82
symbol, 84 integral
fundamental basis, 45
domain, 52 closure, 37
problems, 69 in a ring, 37
set, 84 domain, 7
unit, 5, 11, 58, 71, 77 element, 36
extension, 39
Galois
ideal, 16
action, 84
point, 7
group, 81
integrality relation, 36, 37
theorem, 81
integrally closed, 36
theory, 81
inverse ideal, 17
Gamma function, 68
invertible ideal, 17, 18
Gauss
irreducible elements, 9
lemma, 38
sum, 39 Kronecker
Gaussian integer, 6 delta, 44
generalized Riemann hypothesis, 68 symbol, 79
geometry of numbers, 9, 57, 59 Krull-dimension, 21
global ring, 18 Kummer, E. E., 15, 17
GRH, see generalized Riemann hypothesis Kummer-Dedekind theorem, 30
Haar measure, 52 Lagrange’s four squares theorem, 63
Hermite’s theorem, 57 lattice, 52, 59
Hilbert’s tenth problem, 10 co-volume, 52
complete, 52
ideal, 9, 15
Lebesgue measure, 52
coprime, 15, 16
Legendre symbol, 35, 78
distributive law, 15
length of chain, 21
divisibility of, 15
linear algebra, 33, 41, 69
empty intersection of, 22
linear independence of characters, 51
empty product of, 22
local property, 36

89
Number rings – Index

local ring, 18 analytic, 68

local-global-principle, 18 history of, 4
localization, 18
universal property of, 26 obstruction group, 17
locally principal ideal, 18 order, 20, 54
log-vectors, 77 discriminant of, 43
logarithmic map, 59 maximal, 45
monogenic, 21
maximal order, 45
measure Pell equation, 4, 5, 9
Haar, 52 Picard group, 9, 17
Lebesgue, 52 PID, see principal ideal domain
Minkowski constant, 56 plane cubic curve, 7
Minkowski’s theorem, 53 polynomial
on linear forms, 62 characteristic, 41
monogenic order, 21 coprime, 33
Mordell’s conjecture, 10 cyclotomic, 35
multiplicative subset, 18 discriminant, 46
multiplier ring, 17 resultant, 45, 73
separable, 42
Newton’s formula’s, 75 power basis, 43, 44
Newton’s formulas, 50 power sum, 46, 50
noetherian practical application, 10
induction, 21 primary decomposition, 22
ring, 20 prime, 21
non-degenerate bilinear form, 44 avoidance, 25
norm, 5, 7, 8, 11, 12, 81 element, 7
absolute, 41 inert, 31
of extension, 41 non-invertible, 30
of ideal, 30 of number ring, 21
norm-1-subspace, 60 ramified, 31
normal closure, 84 tamely, 49
normalization, 24, 37, 45 totally, 32, 82
number field, 4 wildly, 49
class group of, 24 regular, 30
different of, 48 singular, 30
discriminant of, 45, 68 split, 31
Picard group of, 17 totally, 32
regulator of, 61, 68, 78 tamely ramified, 49
sieve, 10, 69 totally ramified, 32, 82
totally complex, 59 totally split, 32
zeta-function of, 66 wildly ramified, 49
number ring, 4 primitive
class group of, 24 n-th root of unity, 35
conductor of, 49, 51 field extension, 42
cyclotomic, 15, 35 principal ideal, 15
Picard group of, 17 fractional, 16
prime of, 21 principal ideal domain, 7, 10, 16, 17
regular, 31 product
regulator of, 60 Euler, 66
singular, 31 of ideals, 15
number theory proper ideal, 17
algorithmic, 38

90
 Number rings – Index

quadratic form, binary, 9 Legendre, 35, 78

quadratic reciprocity law, 36
quaternion-algebra, 63 tamely ramified prime, 49
theorem
ramification index, 31, 81 approximation, 27
ramified prime, 31 Artin-Dedekind, 51
rank Cayley-Hamilton, 41
of lattice, 52 Chinese remainder, 15, 47, 81
of order, 20 Dirichlet unit, 6, 9, 11, 58
recurrence, 4, 5, 11 Fermat’s last, 10
reduced ring, 26 Galois’s, 81
regular Hermite’s, 57
number ring, 31 Kummer-Dedekind, 30
prime, 30 Lagrange’s four squares, 63
regulator, 60 Minkowski’s, 53
lower bound, 68, 78 totally complex number field, 59
of number field, 61, 68, 78 totally ramified prime, 32, 82
of number ring, 60 totally split prime, 32
residue class degree, 30, 81 trace
resultant, 45, 73 absolute, 41
Riemann dual, 44, 48
hypothesis, generalized, 68 form, 44
zeta-function, 66 of extension, 41
ring trace-zero-hyperplane, 59, 60
discrete valuation, 23, 31, 38 triangular number, 12
Euclidean, 12, 63, 65
global, 18 UFD, see unique factorization domain
local, 18 unique factorization, 7, 9
noetherian, 20 unique factorization domain, 15
reduced, 26 universal property, 26
semi-local, 27 of localizations, 26
ring of integers, 9, 38
valuation
root of unity, 35, 58
ring, 28
group of, 58
valuation, discrete, 23
primitive n-th, 35
Vandermonde determinant, 44
semi-local ring, 27 volume, 52
separable canonical, 53, 54
algebra, 49 co-, 52
field extension, 42
wildly ramified prime, 49
polynomial, 42
simple integral extension, 30, 51 zeta-function
singular completed, 68
number ring, 31 Dedekind, 66
prime, 30 of number field, 66
snake lemma, 40 Riemann, 66
split prime, 31
Stickelberger’s criterion, 50, 57
sum of two squares, 6, 11
symbol
Artin, 84
Frobenius, 84
Kronecker, 79

91