FACTORISATION THEORY IN A NON-COMMUTATIVE

ALGEBRA
Stephen Ozvatic
Supervised by Dr Daniel Chan
School of Mathematics,
The University of New South Wales.
October 30, 2009
Submitted in partial fulfillment of the requirements of the degree of
Bachelor of Science with Honours
I hereby declare that this submission is my own work and to
the best of my knowledge it contains no materials previously
published or written by another person, nor material which to
a substantial extent has been accepted for the award of any
other degree or diploma at UNSW or any other educational
institution, except where due acknowledgement is made in
the thesis. Any contribution made to the research by others,
with whom I have worked at UNSW or elsewhere, is explicitly
acknowledged in the thesis.
I also declare that the intellectual content of this thesis is the
product of my own work, except to the extent that assistance
from others in the projects design and conception or in style,
presentation and linguistic expression is acknowledged.
Stephen Ozvatic
i
Acknowledgements
I dont really know what my mum wanted me to be when I grew up. She might say
a pilot among other things since she has always liked traveling, planes and Europe,
her birthplace. She was told by one of my primary school teachers many years ago
that I was good at maths, so perhaps she had hope for a few short years that it
would eventuate, but I quickly threw that idea out. Now Im not saying Im good
at maths (I found out Im probably the opposite), but Ive always been interested
in it and it was probably a surprise when my mum found out I decided I was going
to study it at university, since my family migrated here to Australia from a farm in
Europe. Now four years later Im here and have this thesis to show for it. It may
only be an honours thesis to some, but whether this thesis is good or terrible, trivial
or complicated, or if I get a mark of 50 (hopefully I wont get below this) or 95,
it wont impact on how proud I am to have completed this; though 95 would be nice.
Id like to thank my supervisor Daniel along with Hendrik, Gary, Ian and Anthony
who helped me along the way during my studies. My thanks also go to my support
and distraction Sonia, my colleagues Hugh, Oliver and Roland among others, my
friends, and my family: Ana, Maria and Joe.
Stephen Ozvatic
ii
Prerequisites
Though most denitions and the proofs of theorems are included within, at least
an elementary knowledge of ring theory, factorisation theory and module theory
is required. If you have not heard the words homomorphism, integral domain or
module, then refer to the references page for a list of suitable sources. For any
proofs that are not included here, there will be references pointing to proofs in
widely accepted sources; and for any material that is required conceptually, there
will be references given when required.
iii
Introduction
The theory of Dedekind domains comes from the studies of the factorisation prop-
erties of the ring of algebraic integers
1
in an algebraic number eld 1. Richard
Dedekind (1831-1916) in the latter half of his life, knowing that
1
was not always
a unique factorisation domain, looked at fractional ideals of
1
rather than ele-
ments and saw that there was always a unique factorisation of them. Through his
work, he provided the concepts of ideals, modules (though he wrote modul ), dened
prime ideals (that is, generalised prime numbers), introduced the word eld and
perhaps most importantly, the concept of a ring is due to him, though the term
ring rst appears due to Hilbert and our axiomatic denitions were not introduced
until the 1920s by Emmy Noether and Krull.
The work done on these rings by Dedekind were part of the third (1879) and
fourth (1894) editions of Vorlesungen ber Zahlentheorie and the notions created
eventually lead to be fundamental to ring theory, allowing him to provide an alge-
braic proof to the Riemann-Roch Theorem after only three years.
So what is a Dedekind domain? They are basically rings where each ideal is
able to be factorised into prime ideals. If this sounds familiar it probably is since
each integer, an element of , has that same property. That is, each integer can be
factorised into prime integers. It turns out that the integers are also a Dedekind
domain themselves, but this understates their usefulness. If we look at roots of
monic polynomials
r
a
+c
a1
r
a1
+ +c
1
r +c
0
with coecients in , then the set of these roots are called algebraic integers and
denoted 1. If we now look at a nite extension 1 of the eld , then the set 11
is actually a Dedekind domain denoted by
1
and called the ring of algebraic
integers of 1. This is the ring studied by Dedekind in which he found the unique
factorisation property of ideals, even though the unique factorisation of elements in
this ring does not always hold.
With a more modern theory available, this construction which is now often used
as an introduction to algebraic number theory, is mainly used as an example as
Dedekind domains are more powerful than this. For example, every principal ideal
domain is a Dedekind domain. So in Chapter 1 we will look at Dedekind domains
and see their remarkable property of ideal factorisation and furthermore we will
see that in fact the set of all these ideals, to be called fractional ideals, forms an
abelian group and is generated by the prime ideals of the Dedekind domain and
their inverses, which will also be looked at.
This theory will then act as background knowledge as in the next two chapters
we generalise it to a non-commutative setting. We do this by dening orders in
iv
semisimple algebras, which can be conceptually thought of as a subring that is a
lattice in space (the space here being a semisimple algebra). The theory of orders
here depends heavily on the requirement that they be maximal in the sense that
there is no other order larger than it (perhaps other than the algebra it is in). Using
this conceptual denition, a Dedekind domain also forms a lattice in space that is
a subring of its quotient eld. For example the integers form a lattice in the
rationals and are a Dedekind domain, as has been mentioned.
However when moving to orders, there is actually two kinds of ideal factorisation
to consider. The rst is two-sided ideals, the second one-sided ideals. Obviously
the one-sided case encompasses the two-sided case, but there are certain properties
in the two-sided case that can be related back to Dedekind domains, where these
fail in the one-sided case. More specically, the set of all two-sided ideals in a
maximal order is an abelian group generated by the prime ideals of the maximal
order. However, in the one-sided case no such nicety occurs. The set of one-sided
ideals of a single maximal order can not even be considered. It turns out that we
have to consider all ideals of all maximal orders and in that case it turns out to be
a groupoid, a generalisation of a group, which will be called the Brandt Groupoid
associated to the semisimple algebra.
Finally, we will look briey at some nice results due to the theory from Chapters
1,2 and 3. This will be in regard to seeing when the factorisation of elements in
Dedekind domains and orders occurs (and also when unique factorisation of elements
in an arbitrary ring implies the same for ideals), as well as looking at the rst case
of a Dedekind domain, the ring of algebraic integers. But nicest of all (for the
author of this thesis anyway), we will look at the rational quaternions, an extension
of the complex numbers, and then nish with a proof due to the preceding theory
in Chapters 1,2 and 3 of Lagranges Four Squares Theorem: That every natural
number can be written in the form
c
2
+/
2
+c
2
+d
2
for integers c. /. c and d.
v
Contents
Chapter 1 Dedekind Domains and Fractional Ideals 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Dedekind Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 The Factorisation of Ideals . . . . . . . . . . . . . . . . . . . . . . . 7
Chapter 2 Dedekind Domains and Orders 13
2.1 Orders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Maximal Orders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 The Factorisation of Two-Sided Ideals . . . . . . . . . . . . . . . . 20
Chapter 3 Maximal Orders and Groupoids 26
3.1 One-Sidedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 Generalising a Familiar Sight . . . . . . . . . . . . . . . . . . . . . 30
3.3 The Factorisation of One-Sided Ideals . . . . . . . . . . . . . . . . . 33
Chapter 4 Factorisation...of Elements 39
4.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Factorisation, the Class Group and The Class Number . . . . . . . 40
4.3 Finiteness of The Class Number of the Ring of Integers . . . . . . . 42
4.4 Factorisation in an Order and the Jordan-Zassenhaus Theorem . . . 44
4.5 Quaternions and a Sum of Four Squares . . . . . . . . . . . . . . . 47
References 51
vi
Chapter 1
Dedekind Domains and Fractional Ideals
To start with, we will look at a standard view of the theory, with the end of the
rainbow not being home to an Irishman with a pot of gold, but rather a uniqueness
result about Dedekind domains. Chapter 1 is mainly to do with the commutative
case, however a more general case will be done after this. So to keep a standard,
unless dened or stated otherwise commutativity is not assumed and every ring is
unital. Firstly though, some preliminaries must be carried out, but most of these
should be familiar to the reader. If 1 is a domain, it is meant that 1 is a ring with
no zero divisors. A ring is then an integral domain if it is a commutative domain.
Assume hereon for the whole document that 1 is a ring with quotient ring 1 where
1 = 1 and note that if 1 is an integral domain then 1 is a eld. For clarity,
the notation will denote a strict subset whereas indicates the possibility of
equality.
1.1 Background
For a ring 1, not necessarily commutative, a left 1-module ` is an additive abelian
group with a map 1` ` : (:. :) :: such that for :. :. 1 1 and :. : `
(: +:): = ::+::
:(:+:) = ::+::
(::): = :(::)
1: = :.
Similarly, a right and two-sided module can be dened. The left 1-module ` is
called nitely generated if it can be written as ` = 1r
1
+ 1r
2
+ + 1r
a
for a
nite number, :, of elements r
i
`. An 1-algebra is a ring such that there
exists a homomorphism o : 1 2() where 2() = {r rc = cr for every
c } is the centre of . In this document, an algebra is always assumed to be
associative. To see an as an 1-module, we dene : r = o(:)r for : 1 and
r . The dimension of a 1-algebra over the eld 1 is its dimension as a
vector space over 1.
Given an integral domain 1, let be a nite dimensional 1-algebra. An element
c is integral over 1 if there is some monic polynomial )(A) 1[A] with )(c) =
0 and similarly a subring 1 is integral over 1 if every element of 1 is integral
over 1. Hence the set of all elements of integral over 1 is called the integral closure
of 1 in and is denoted 1

. 1 is then called integrally closed if the integral closure

of 1 in 1 is 1 itself. For an integral domain 1 and nite dimensional 1-algebra ,
the following two results will be useful at times throughout this document. They
are standard results so will not be proved here, but proofs can be found in Reiner
[1, p.3-6], Jacobson [6, p.408-409] and Janusz [7, p.5-7].
1
Proposition 1.1.1 The following are equivalent for an element c .
a) c is integral over 1.
b) 1[c] is a nitely generated 1-module.
c) There is nitely generated 1-module ` that is a subring of containing c.
Proposition 1.1.2 If 1 is integrally closed, then c is integral over 1 if and
only if the minimal polynomial of c over 1, j(A), is an element of 1[A].
Some concepts of factorisation theory are required in a later chapter, but are
also needed conceptually at times throughout the document, so a basic introduction
will be given here. For the reader not familiar with this material, it can be found
in Jacobson [5, p.140-149], Stewart & Tall [8, p.81-99] and Sivaramakrishnan [10,
2,3]. For an integral domain 1, an element n 1 is called a unit if there is some
1 such that n = n = 1
1
. The set of units in 1, n(1), is then an abelian
subgroup of 1 under multiplication. Then if two elements c. / 1 satisfy c = n/
for some n n(1) we call c and / associates. If c = /c for c. /. c 1, we say that
/ (or equivalently c) divides c and denote this by / c. An element j 1 is called
prime when for two elements c. / 1, if j c/ then either j c or j /. A non-zero
non-unit element : 1 is irreducible if for every factorisation : = c/ for c. / 1,
either c n(1) or / n(1). See that both c and / can not be in n(1) since this
forces : n(1). Also note that irreducible elements are prime, but the converse is
not necessarily true.
Example 1.1.3 Every irreducible element j in an integral domain 1 is prime since
a factorisation j = c/ with c. / 1 has, without loss of generalisation, / n(1).
So j and c are associates and j c.
Now we can dene certain types of rings. Here they are commutative, but the
denitions can be generalised to a non-commutative case. The integral domain 1
is called an Unique Factorisation Domain (or UFD) when for every element : 1,
there exists a factorisation of : into irreducible elements of 1 and if there are two
factorisations : = c
1
c
a
= /
1
/
n
with c
i
. /
)
1 irreducible for all i and ,,
then : = : and c
i
= n
i
/
(i)
for some permutation perm{1. . :} and unit
n
i
n(1) for every i. That is, we can swap around the /
i
s and multiply them by
units to get c
1
c
a
.
Certain rings have a nice property when looking at their ideals, so we will dene
those rings here. Given a subset o 1, the ideal generated by o is

-S
:1 and
is denoted o. An ideal 1 is then called a principal ideal if 1 = : = :1 for some
: 1. That is, it is generated by one element of 1. An integral domain 1 is thus
called a Principal Ideal Domain (or PID) if every ideal of 1 is a principal ideal. It
turns out that the following results hold, but note importantly that the converse of
Theorem 1.1.5 does not hold.
Proposition 1.1.4 Every prime element in a UFD is irreducible.
Proof. Let 1 be a UFD and j 1 be prime. If j = c/ for two elements c. / 1,
then j c/ and without loss of generality say that j c. Then c = jc for some
c 1. But then j = c/ = jc/ and hence c/ = 1 since 1 is an integral domain,
implying / n(1) and j is irreducible.
2
Theorem 1.1.5 Every PID is a UFD.
Proof. Let 1 be a PID and rst suppose that there is some : 1 that is not
expressible as a product of irreducibles and so can not be irreducible itself. Then
: = c
1
/
1
for some non-units c
1
and /
1
where at least one is not expressible as a
product of irreducibles as well. Suppose without loss of generalisation this is c
1
.
Then we can write c
1
= c
2
/
2
for some non-units c
2
and /
2
where c
2
is not expressible
as a product of irreducibles. Continuing this gets an innite chain of proper ideals
: c
1
c
2
. Using Proposition 1.1.7 from ahead gets that this chain
eventually stabilises, so one of the c
i
s must be ireeducible, contradictiong that c
i
is not expressible as a product of irreducibles.
Secondly, write : 1 as : = j
1
j
2
j
a
=
1

2

n
for irreducible j
i
and

)
. Since they are also prime we have j
1

1

2

n
, so j
1

I
for some / and
since j
1
and
I
are irreducible, j
1
= n
1

I
for some unit n
1
n(1). Now j
2
j
a
=

1

I1

I+1

n
, so doing this repeatedly gets : = : and each j
i
an associate of
some
)
with 1 = n
1
n
a
. Thus : is uniquely expressible as a product of irreducible
elements of 1 up to rearrangement and associates, so 1 is a UFD.
Example 1.1.6 The integers have ideals of the form :, so they are a PID and
thus a UFD. The units of are n() = {1. 1} and the irreducible integers are the
prime integers.
Finally, a ring 1 is called left noetherian if its left ideals satisfy the ascending
chain condition (that is, a chain of left ideals 1
0
1
1
1
2
of 1 eventually has
1
a
= 1
.
for every ` :.) and left artinian if its left ideals satisfy the descending
chain condition (similarly to before but with a decreasing chain instead). This
similarly holds for right noetherian and artinian, and a ring is just called noetherian
or artianian if the respective chain conditions hold for both right and left ideals.
Similarly, an 1-module ` is called noetherian or artinian if its submodules satisfy
those same conditions. A prime ideal of an integral domain 1 is a non-zero proper
ideal 1 such that 1,1 has no zero divisors. This denition is equivalent to saying
that given ideals . 1 and 1, if 1 1 implies 1 or 1 1, then 1 is prime
(which is analogous to the usual denition of an element being prime, as given
above). The following result about a PID will be useful. In fact, we have already
used it.
Proposition 1.1.7 Any PID is noetherian.
Proof. Let 1 be a PID, 1
0
1
1
1
2
be an innite chain of ideals in 1 and
1 =
i0
1
i
. Let c. / 1, so c 1
i
and / 1
)
for some i and , and without loss of
generality suppose that i ,. Then since 1
i
1
)
. c +/ 1
)
1 since 1
)
is a group
and for : 1. :/ 1
)
1 also, showing 1 is in fact an ideal of 1. Since 1 is a PID,
every ideal is principal, so 1 = j for some irreducible j 1. Now j 1, so j 1
I
for some / and 1
0
1
1
1
I
= 1
I+1
= , showing 1 is noetherian.
Example 1.1.8 A maximal ideal 1 of an integral domain 1 is a prime ideal since
1,1 is a eld.
3
1.2 Dedekind Domains
Now using these denitions we can dene a certain sort of ring, called a Dedekind
domain or Dedekind ring, which is the assumption taken upon many rings in this
document. As will be found by the end of the document, these types of rings all
have a property that can help when trying to establish when elements of 1 can be
expressed uniquely in terms of its irreducible elements, that is, when 1 is a unique
factorisation domain. But their main property, as we will nd, is their ability to
factorise of ideals.
The theory of Dedekind domains comes from the studies of the factorisation
properties of the ring of algebraic integers
1
in an algebraic number eld 1.
Dedekind (1831-1916) in the latter half of his life, knowing that
1
was not always
a UFD, looked at fractional ideals of
1
rather than elements and saw that there
was always a unique factorisation of them. This lead to the use of the word ideal
in its common usage and the use of his name for the ring we are about to dene.
References for the rest of this chapter are Reiner [1, p.44-50], Curtis & Reiner [4,
18], Jacobson [6, 10], Janusz [7, p.8-18], Samuel [9, p.47-52] and Sivaramakrishnan
[10, p.394-410]. There are other denitions of a Dedekind ring involving projectivity
and localizations and discrete valuation rings, but they will not be looked at. Our
denition will be as follows.
Denition 1.2.1 An integral domain 1 is called a Dedekind domain if it is noethe-
rian, integrally closed and such that every non-zero prime ideal is maximal.
This denition is equivalent to a remarkable fact about the factorisation of
ideals in the integral domain 1 into irreducibles. You might guess that these
irreducibles are the maximal ideals of 1 and in that case, you would be right.
Except when 1 is a Dedekind domain prime ideals coincide with the maximal ideals,
or irreducible ideals, which is reminiscent of the property in unique factorisation
domains that prime elements are irreducible and every element can be written as a
unique product of prime elements. The reader may also guess that principal ideal
domains may be important here, and again they would be right as the following
examples give an indication of.
Example 1.2.2 As with many things in number theory, some concepts are vast
generalisations of properties held by the integers. The PID is a noetherian inte-
grally closed integral domain with quotient eld . Since its prime ideals are of the
form j for j a non-zero prime integer, the prime ideals coincide with the maximal
ideals of and hence it is a Dedekind domain.
Proposition 1.2.3 Every PID is a Dedekind domain.
Proof. Proposition 1.1.7 shows us every PID is noetherian, so we only need to show
the other two requirements. Let 1 be a PID and j be a prime ideal in 1. Then
for two ideals c and / of 1 that have c / = c/ j, it follows that j c/.
But then c j or / j since j is prime, so we have either j c or j /,
showing j is prime as an element of 1. Theorem 1.1.4 then implies j is irreducible,
so j is actually a maximal ideal.
For the proof of being integrally closed, the usual proof will do. Let r 1 and
suppose it is a root of a polynomial A
a
+c
a1
A
a1
+ +c
1
A +c
0
1[A]. Then
4
since 1 is a UFD, r =
j
q
where j and are coprime. Then substituting
j
q
into the
polynomial and multiplying by
a
gets
j
a
+c
a1
j
a1
+ +c
1

a1
j +c
0

a
= 0.
Thus j must follow and = 1, so r 1 and 1 is integrally closed, proving
the proposition.
Example 1.2.4 The ring [

5] =

5 is a Dedekind domain. Firstly,

it is noetherian since is noetherian. Secondly, it is integrally closed since if
=
o
o
+
c
o

5 [

5] with (c. /) and (c. d) pairwise coprime is a root of a monic

polynomial A
a
+c
a1
A
a1
+ +c
1
A +c
0
[

5][A], then substituting A =

and multiplying by (/d)
a
gets
(/d)
a
+c
a1
(/d)(/d)
a1
+ +c
1
(/d)
a1
(/d) +c
0
(/d)
a
= 0.
a ploynomaial in [

5][/d] But /d divides 0, so it must divide the left hand side,

so must divide (/d)
a
. But /d = cd + c/

5, so /d must divide both cd and c/,

implying both / and d are either 1 or 1, so [

5].
Thirdly, let 1 be a prime ideal of [

5]. If c + /

5 1, then both of
5c/ + 5/
2

5. c
2

5 5c/ 1, so adding gets c

2
+ 5/
2
1 also. But an integer
: 1 if and only if :

5 1 too. Thus 1 is of the form j+

5j for some j
. If : = c/ for some c. / [

5], then 1 = (c+c

5)(/+/

5) = 1,
so 1 1 (Note the multiplication of ideals is given by (1.2) ahead, if required).
But 1 and 1 1, so 1 can not be prime. Thus j must be an irreducible
in [

5]. Let 1 be an ideal of [

5] such that 1 1. Then there is some

c 1 such that j c. But by the Euclidean algorithm there exists r.
such that rj +c = 1, so 1 = 1 and 1 is maximal.
This example was given for a particular reason. Since 6 = (1+

5)(1

5) =
2.3, it is not a UFD and thus not a PID. This shows the converse of Proposition
1.2.3 does not necessarily hold.
Example 1.2.5 For another example, given a nite eld extension 1, consider
the integral closure of in 1 and let
1
=
1
. Then
1
is a ring called the
algebraic integers of 1 (or the ring of integers of 1) and is in fact a Dedekind
domain. This example will be reintroduced later in Chapter 4 when talking about
the class number.
However from here the relationship between unique factorisation of elements and
of ideals will be left until Chapter 4. What we need to focus on is what we are trying
to look at. If we want to factorise ideals lets make sure that we know exactly what
we mean when we say an ideal. Before we do this, we again need to build up some
standard theory. Tensor products will be used here, but not in their full power, so
only a small notion of them is required. For readers unfamiliar with tensor product
see Curtis and Reiner [4, 12] and Jacobson [6, p.125-133]. For a basic view of a
construction of a tensor product of two modules, let 1 be an integral domain and
5
consider a right 1-module M, left 1-module N, and the module 1 generated by
their product ` `
1 = {nite sums

i
:
i
(:
i
. :
i
) :
i
`. :
i
`. :
i
1}.
Now look at the subgroup \ of 1 generated by the following subset of 1, for all
:. :

`, :. :

` and : 1.

(:+:

. :) (:. :) (:

. :)
(:. : +:

) (:. :) (:. :

)
(::. :) :(:. :)
(:. ::) :(:. :)
Then we dene the tensor product of ` and ` over 1 to be `
1
` = 1,\ . Here
the tensor product `
1
` is actually an 1-module. This is because for :. : 1
and : `. : `,
:(:(:. :)) = :(::. :) = ((::):. :) = (:(::). :) = (::)(:. :).
So since 1 is an integral domain, :: = :: allows `
1
` to be an 1-module. If
the tensor product is over the quotient eld 1, then `
1
` is more specically
a vector space. In the case we deal with, ` = 1, the quotient eld of 1, and `
is commutative. So it is actually just the vector space 1
1
` and for ease ignore
the
1
and just write
1
1
` = 1 ` = {nite sums

i
/
i
:
i
/
i
1. :
i
`}. (1.1)
That is all that is required of tensor products here (actually (1.1) is the the
main point), so we will move onto some more denitons. For an integral domain 1
and 1-module `, the annihilator of ` is dened by c::
1
` = {: 1 :` = 0}
and similarly the annihilator of an element : ` is dened by c::
1
: = {:
1 :: = 0}. The element : ` is called an 1-torsion element if c::
1
: = 0
and ` is called 1-torsionfree if the only 1-torsion element in ` is 0.
The point here of torsion elements is that we want to visualize ` as 1
1
`
inside 1
1
`. If there is a 1-homomorphism o : ` 1
1
` : : (1. :),
then /c:{o} is the submodule of these torsion elements in `, and ` being 1-
torsionfree sets /c:{o} = 0, giving us what we want. See Reiner [1, p.32-34, 44]
for more details on this. Now we can dene the following for an integral domain 1.
An 1-lattice is a nitely generated 1-torsionfree 1-module. Each 1-lattice 1 is an
1-submodule of the nite dimensional vector space \ = 1
1
1 = 1 1. Hence 1
is called a full 1-lattice in \ to emphasize that 1 has a 1-basis of \ . This leads us
to the following, as descibed in Reiner [1, p.47-48] and Swan & Evans [2, p.83-84],
which turns out to be the same as the other references listed earlier.
Denition 1.2.6 For a Dedekind domain 1, a non-zero full 1-lattice in 1 is called
a fractional 1-ideal.
6
Note here that since the vector space \ of a fractional 1-ideal 1 is actually
the eld 1, it trivially follows that a nitely generated 1-submodule of 1 is 1-
torsionfree (since 1 is an integral domain) and satises 1 1 = \ = 1 (since
1 is a eld). So a fractional 1-ideal can be dened more simply to be just a
nitely generated 1-submodule in 1. Some important things to note are that
for a fractional 1-ideal 1 there is an : 1 satisfying :1 1 since 1 is nitely
generated in 1. Secondly, every ideal 1 of 1 is itself a fractional 1-ideal since it is
a 1-submodule of 1 1 which is noetherian. Thirdly, the multiplication of two
fractional ideals J and 1 is dened by
J1 = {nite sums

i
,
i
|
i
,
i
J. |
i
1}. (1.2)
Since J and 1 are nitely generated, J1 must be nitely generated and is thus a
fractional ideal too.
These fractional ideals are the key part of the theory for factorisation of ideals
into prime ideals in a Dedekind domain. But to show this requires a bit more work.
It turns out to be of immeasurable use to look at the set of elements of 1 that send
a fractional ideal inside 1. The reason being that the product of two fractional
ideals is another fractional ideal and 1 is istelf a fractional 1-ideal, so this set of
elements gives the possible notion of a group with identity 1. As we will soon see
in the next section, this indeed gives us what we want. So to nish this section,
given a Dedekind domain 1 dene for a fractional 1-ideal 1
1
1
= {r 1 r1 1}. (1.3)
1
1
is clearly an 1-module in 1 and by the note above there is an : 1 such that
:1 1, so 1
1
= 0. Now a non-zero r 1 satises 1
1
r 1, so r 1 for every
1
1
. So 1r
1
and 1
1
is an 1-submodule of the principal 1-module 1r
1
and is thus nitely generated itself. So 1
1
is also a fractional 1-ideal.
Example 1.2.7 As before, take the Dedekind domain . Then (j,) is a fractional
ideal for non-zero j. and ((j,))
1
is given by (,j). Actually, all fractional
ideals of are of that form. More generally, for any Dedekind domain 1, r1 is a
fractional ideal for a non-zero r 1 and (r1)
1
is given by r
1
1. But all ideals
of 1 are of that form only if 1 is a PID.
1.3 The Factorisation of Ideals
From here we can start to think about if we can factorise ideals of a Dedekind
domain 1 into maximal (or prime) ideals and even better if we can factorise all
fractional ideals of 1 into prime ideals, which will be the aim of the rest of this
chapter. The order of the results here is sometimes in the opposite direction of the
order given in sources containing this material, but the content is often the same.
References for this section were mentioned at the denition of a Dedekind domain
and are Reiner [1, p.44-50], Curtis & Reiner [4, 18], Jacobson [6, 10], Janusz [7,
p.8-18], Samuel [9, p.49-52] and Sivaramakrishnan [10, p.394-410]. From here, only
a few results are needed to get the group structure that was mentioned before, but
they also are our stepping stones towards our aim of factorising ideals.
7
Proposition 1.3.1 Every proper ideal 1 of a Dedekind domain 1 contains a product
of prime ideals of 1 and has 1 1
1
.
Proof. For completeness, note the ideal 1 = 1 contains any product of prime ideals,
but 1
1
= 1
1
= 1. Let be the set of proper ideals of 1 that do not contain
a product of prime ideals. Since 1 is noetherian there must exist an ideal 1 in
that is contained in no other ideal in , that is, a maximal element of . Now
1 cannot be prime and any ideal properly containing 1 must contain a product of
prime ideals. Since 1 is not prime, there are elements :. : 11 such that :: 1
and hence 1 + :1 and 1 + :1 are ideals properly containing 1, so each contains a
product of prime ideals. But
(1 +:1)(1 +:1) 11 +:1 +:1 +::1 1
and so 1 must contain a product of prime ideals, contradicting the existence of a
maximal element of the set , and is empty.
For the second part of the proposition let 1 be a proper ideal of 1. Then clearly
1
1
1, so we must show 1
1
1 is not empty. Let i 1 be non-zero. Then by the
previous part there are prime ideals 1
1
. . . . . 1
a
such that 1
1
. . . 1
a
i1 1, and
select 1
1
. . . . . 1
a
such that : is minimal. As 1 is noetherian and 1 1 there is some
maximal ideal 1 with 1
1
. . . 1
a
1. If 1
I
= 1 for every /, then there are j
I
1
I
where j
I
, 1 and j
1
. . . j
a
1, so j
I
1 1 and j
1
1. . . j
a
1 = j
1
. . . j
a
1 1,
contradicting 1 being prime (since maximal ideals are prime ideals). Thus 1 = 1
I
for some / and
1
1
. . . 1
a
= 1
1
. . . 1
I1
1
I+1
. . . 1
a
1 i1 1 1.
Now since : is minimal, 1
1
. . . 1
I1
1
I+1
. . . 1
a
i1, so select some element j
1
1
. . . 1
I1
1
I+1
. . . 1
a
i1. Thus i
1
j , 1 and
i
1
j1 i
1
j1 i
1
1
1
. . . 1
I1
1
I+1
. . . 1
a
1 i
1
(i1) = 1
and hence i
1
j 1
1
1.
Proposition 1.3.2 For a Dedekind domain 1 and fractional 1-ideal 1, the set
{r 1 r1 1} = 1.
Proof. Let o = {r 1 r1 1} and note that 1 o since 1 is an 1-module, so
we only need to show the reverse inclusion. Let : o, then for :
:
a
1 = :
a1
(:1) :
a1
1 . . . :1 1.
so 1[:] is an 1-submodule of 1 and is thus nitely generated. If 1 1, then since
1 is integrally closed it follow by Proposition 1.1.1 that : 1. If 1 1 then 1[:]
is a nitely generated submodule of 1[:] and similarly : 1.
Corollary 1.3.3 For a fractional ideal 1 of a Dedekind domain 1, 1
1
1 = 11
1
=
1.
8
Proof. By denition 1
1
1 1 and since 1
1
1 is itself a fractional ideal, it is an
ideal in 1 and so (1
1
1)
1
(1
1
1) is also an ideal in 1. By Proposition 1.3.1, if
1
1
1 is proper in 1 then (1
1
1)
1
1. Now since (1
1
1)
1
(1
1
1) 1 we
have (1
1
1)
1
1
1
1
1
and thus (1
1
1)
1
1 by Proposition 1.3.2. So 1
1
1
can not be proper and we must have 1
1
1 = 1. Also, 11
1
= 1 since 1 is
commutative.
Note also that for a fractional 1-ideal 1, this imples that (1
1
)
1
= 1. Now
do we have a group structure? Firstly, the product of two fractional ideals is a
fractional ideal (and is commutative since 1 is). Secondly the product of fractional
ideals is clearly associative since 1 is. Thirdly and nally; 1 is the unity element
and there exist inverses for every fractional ideal that satisfy Corollary 1.3.3. Thus
we have just proven the following theorem.
Theorem 1.3.4 For a Dedekind domain 1, the set of fractional 1-ideals
1
is a
multiplicative abelian group with identity 1 and the inverse of 1
1
given by
1
1
.
But what about our aim? If we want to be able to factorise every ideal of a
Dedekind domain 1, or even better, every fractional ideal of 1 in terms of prime
ideals of 1, then it makes sense that the set of fractional 1-ideals
1
would be
generated by the prime ideals of 1. Indeed this is the case and the following results
lead to its proof. But rst, say for two fractional ideals 1 and J that 1 divides J
when there is an ideal 1 1 such that J = 11 and denote this by 1 J. Then
there is the following equivalence.
Proposition 1.3.5 Let J and 1 be fractional ideals of the Dedekind domain 1.
Then J 1 if and only if there is a proper ideal 1 1 such that J = 11.
Proof. One way is trivial since if J = 11 then 1 = 11 11 = J. So let J 1.
Then 1 = 1
1
1 1
1
J and 1
1
J is a proper ideal in 1. Letting 1 = 1
1
J gets
11 = 11
1
J = J as required.
Lemma 1.3.6 The ideals in a Dedekind domain can be written uniquely in terms
of its prime ideals up to rearrangement.
Proof. First show an ideal in the Dedekind domain 1 can be written as a product
of prime ideals and afterwards show this expression is unique up to rearrangement.
We dene the ideal 1 of 1 to be an empty multiplication of primes, so exclude
this case. Let o be the set of proper ideals in 1 that can not be expressed as a
product of prime ideals and suppose it is not empty. Since 1 is noetherian there is a
maximal element 1 o that can not be a maximal ideal in 1 (since maximal ideals
are prime). Let J be a maximal ideal of 1 strictly containing 1. By Proposition
1.3.5 there is a proper ideal 1 in 1 such that 1 = J1. Now 1 = 11 J1 = 1, so
we have 1 J. 1 1 and by the maximality of 1 both J and 1 are products of
prime ideals. But then so is J1 = 1 and thus o must be empty.
Now suppose that two products of prime ideals are equal, say 1
1
1
2
. . . 1
a
=
Q
1
Q
2
. . . Q
n
for prime ideals 1
i
. Q
)
and positive integers :. :. As in Proposition
1.3.1, let 1 be a maximal ideal in 1 such that 1
1
1
2
. . . 1
a
1. Then if 1 = 1
i
for
9
every i there are j
i
1
i
where j
i
, 1 such that j
1
j
2
. . . j
a
1. So j
i
1 1
i
for
every i and j
1
1j
2
1. . . j
a
1 = j
1
j
2
. . . j
a
1 1, contradicting 1 being prime. So
let 1 = 1
I
for some /. Similarly Q
1
Q
2
. . . Q
n
1 and thus 1 = Q
|
for some |. So
1
I
= Q
|
and we get
1
1
1
2
. . . 1
I
. . . 1
a
= Q
1
Q
2
. . . Q
|
. . . Q
n
1
1
1
I
1
1
1
2
. . . 1
I1
1
I+1
. . . 1
a
= 1
1
Q
|
Q
1
Q
2
. . . Q
|1
Q
|+1
. . . Q
n
1
1
1
2
. . . 1
I1
1
I+1
. . . 1
a
= Q
1
Q
2
. . . Q
|1
Q
|+1
. . . Q
n
.
By doing this procedure another : 2 times it is clear that : = : and each 1
i
coincides with some Q
)
. Thus the factorisation is unique up to rearrangement.
We now know every proper ideal 1 of a Dedekind domain 1 can be written in
the form 1 = Q
1
Q
2
. . . Q
n
for some positive integer : and prime ideals Q
)
, so we
can rewrite this as 1 = 1
c
1
1
1
c
2
2
. . . 1
c
u
a
for some positive integer : : where all the
1
i
are distinct and c
i
0. This remarkable (and only slightly set up) property as
mentioned earlier is reminiscent of the integers and principal ideal domains and is
where the this theory, the theory of Dedekind domains, was born. Also, it gives us
a nice generalisation of the greatest common divisor and lowest common multiple.
Corollary 1.3.7 The greatest common divisor of two ideals 1. J of 1, denoted
gcd(1. J), and lowest common multiple, denoted lcm(1. J), both exist and are a
product of prime ideals of 1.
Proof. Write 1 = 1
c
1
1
. . . 1
c
u
a
and J = 1
)
1
1
. . . 1
)
u
a
where c
i
and )
)
are natrural
numbers as in the previous lemma. Then gcd(1. J) = 1
min{c
1
,)
1
}
1
. . . 1
min{c
u
,)
u
}
a
and
lcm(1. J) = 1
max{c
1
,)
1
}
1
. . . 1
max{c
u
,)
u
}
a
.
This last Lemma also gives us a nice fact. If 1 = 1
c
1
1
1
c
2
2
. . . 1
c
u
a
is a proper ideal
in 1, then can we use this to write 1
1
in a similar form? Consider 1
c
1
1
1
c
2
2
. . . 1
c
u
a
where 1
c
.
i
= 1
1
i
1
1
i

c
.
times
. Then
11
c
1
1
1
c
2
2
. . . 1
c
u
a
= 1
c
1
1
1
c
2
2
. . . 1
c
u
a
1
c
1
1
1
c
2
2
. . . 1
c
u
a
= 1
and it follows that 1
1
= 1
c
1
1
1
c
2
2
. . . 1
c
u
a
. This then allows us to extend the
preceding Lemma to get one of the main theorems of this chapter.
Theorem 1.3.8 The abelian group of fractional ideals
1
of a Dedekind domain
1 is generated by the prime ideals of 1 and their inverses.
Proof. This will be proved by showing every fractional ideal can be written as a
product of primes and prime inverses. Let 1 be any fractional ideal of 1. If 1 1,
then this is immediate by Lemma 1.3.6 so let 1 1. As mentioned earlier, since 1
is nitely generated there is a non-zero : 1 such that :1 1, so :1 is an ideal in
1. By Lemma 1.3.6 we can write this as :1 = 1
c
1
1
1
c
2
2
. . . 1
c
u
a
for some non-negative
integer :, positive integers c
i
and prime ideals 1
i
. Similarly :1 = Q
)
1
1
Q
)
2
2
. . . Q
)
r
n
for some non-negative integer :, positive integers )
)
and prime ideals Q
)
. So :1 =
10
(:1)1 and thus 1 = (:1)
1
1
c
1
1
1
c
2
2
. . . 1
c
u
a
= Q
)
1
1
Q
)
2
2
. . . Q
)
r
n
1
c
1
1
1
c
2
2
. . . 1
c
u
a
.
That is, 1 can be written as the product of prime ideals and inverses of prime
ideals and this factorisation is unique (up to rearrangement) since the factoriza-
tions of :1 and :1 are themselves. Thus the set of fractional ideals
1
is generated
by the prime ideals of 1 and their inverses.
Example 1.3.9 As in Example 1.2.7, fractional ideals of are of the form
j
q
for
j
q
. So let j and be coprime and write j and as their prime factorisations
j = r
i
1
1
r
i
u
a
and =
)
1
1

)
r
n
. Then
j

=
r
i
1
1
r
i
u
a

)
1
1

)
r
n
= (r
i
1
1
) (r
i
u
a
)(
)
1
1
) (
)
r
n
)
= (r
1
)
i
1
(r
a
)
i
u
(
1
)
)
1
(
n
)
)
r
.
Also, the abelian group of fractional ideals

is just the group generated by

{} {2. 3. 5. 7. 11. } {
1
2
.
1
3
.
1
5
.
1
7
.
1
11
}.
So in a similar fashion to the previous Lemma, we can now write any fractional
ideal 1
1
uniquely in the form 1 = 1
c
1
1
1
c
2
2
. . . 1
c
u
a
for some non-negative integer
: with each 1
i
distinct. But this time each c
i
is a non-zero integer, not necessarily
positive. This theorem eectively gives us a characterisation of Dedekind domains
and to conclude the chapter, it turns out that if we went backwards, then everything
works out nicely. But as the next chapters will show us, the uniqueness result about
Dedekind domains that we have found can be generalised. So if we are at the end
of the rainbow now, then maybe the pot of gold would of been better?
Theorem 1.3.10 The following are equivalent for an integral domain 1.
a) The ring 1 is a Dedekind domain.
b) The ring 1 is noetherian, integrally closed and every non-zero prime ideal in
1 is maximal.
c) Every proper ideal 1 in 1 is expressible as a product of prime ideals of 1
uniquely up to rearrangement as 1
c
1
1
. . . 1
c
u
a
with each 1
i
distinct and c
i
0.
d) The set of fractional ideals
1
of 1 is generated by the prime ideals of 1 and
their inverses.
Proof. By denition, a) and b) are equivalent and Theorem 1.3.8 c) and d) must
be equivalent. Also, we have just shown that b) implies c), so we only need to show
that c) or d) implies b) to get the theorem. Assume d) and consider an innite
chain of ideals 1
1
1
2
1
3
. Then write 1 = 1
1
1
a
for prime ideals
1
i
not necessarily distinct. Then the chain can have at most : distinct proper
ideals, so eventually stabilizes and thus 1 is noetherian. Secondly, let c 1
be a root of the polynomial A
a
+ c
a1
A
a1
+ c
1
A + c
0
1[A]. Then c
a
=
c
a1
c
a1
c
1
c c
0
, so letting 1 = c
a1
. . c. 1 we see that c
a
1 and
thus c1 1. But 1 = 1
1
1
I
is a product of prime ideals 1
i
of 1, so multiplying
11
both sided by 1
1
1
1
1
I
gets c1 1, so c 1. Finally, clearly the prime ideals
coincide with the maximal ideals. So 1 is a Dedekind domain, showing d) implies
b) and proving the theorem.
12
Chapter 2
Dedekind Domains and Orders
In Chapter 1, we went through what a Dedekind domain was and how ideals in
Dedekind domains could be written as a product of prime ideals. But what about
doing this more generally? What do we mean here by more generally? We mean
for much larger rings than Dedekind domains, say for rings containing Dedekind
domains. Dedekind domains are of course commutative, so to do this more generally,
we generalise the ideas in Chapter 1 to a non-commutative setting. So as mentioned
at the beginning of Chapter 1, commutativity is not assumed. To start, we extend
our reach and consider algebras over a Dedekind domains quotient eld, then take a
subring of this algebra and see if this subring (which is not necessarily commutative
or a domain) has a similar type of factorisation into prime ideals.
Some preliminaries are required to move forward, so we will say these here.
Recalling a denition from Chapter 1, For a commutative ring o, an o-algebra is a
ring such that there exists a homomorphism : o 2() where 2() = {r
rc = cr for every c } is the centre of . So to see as an o-module, we
dene : r = (:)r for : o and r . An o-algebra is then called simple if
it has no non-trivial two-sided ideals and semisimple if is the direct product of
simple subalgebras.
The use of elds in this chapter require some properties, so again let us dene
them here. Let 1 be a eld and 1,1 be a eld extension. Then 1 is called separable
over 1 if the minimal polynomial of every c 1 over 1 factors into distinct linear
factors over 1. The dimension of an 1-algebra over 1 is its dimension as a
vector space over 1. For the nite dimensional 1-algebra , if is semisimple and
the centre of each simple simple summand of is a separable eld extension of 1,
then is called separable over 1.
Also recall from Chapter 1 that for a Dedekind domain 1 with quotient eld
1 and vector space \ over 1, a full 1-lattice in \ is a nitely generated 1-
torsionfree 1-module ` in \ that satises 1
1
` = 1 ` = \ by the equality
(1.1). This allows us to dene a major part of the theory from hereon. So for the
rest of this chapter let 1 be a Dedekind domain with quotient eld 1, be a
nite dimensional separable semisimple 1-algebra and ` be a full 1-lattice in .
Note that is always associative, but not necessarily commutative and that 1 is
a subring of 2().
2.1 Orders
In Chapter 1 we dened an integral domain to be a Dedekind domain if
I It was noetherian.
II It was integrally closed.
13
III Every prime ideal was maximal.
If we use these three properties as a hint to help nd subrings of able to factorise
ideals as we can in Dedekind domains, then these subrings might not be as compli-
cated or hard to nd as rst thought. Actually, as references Reiner [1, p.108-110],
Swan & Evans [2, p.83-84], Bass [3, p.152-156] and Curtis & Reiner [4, p.515-517]
point out, these types of subrings are starring us in the face; as the next seemingly
abstract denition of an 1-order points out.
Denition 2.1.1 A unital subring of that is a full 1-lattice in is called an
1-order in .
As with Dedekind domains there are also other denitions of an 1-order that
involve Krull rings and primary ideals, but they will not be looked at. So for this
reason Reiner [1] and Swan & Evans [2] are the main references for this chapter,
though Bass [3] and Curtis & Reiner [4] still provide some input, but not as much.
(Note that conceptually orders can be thought of as follows: Starting with a nite
number of points in space, we can extend these by adding an subtracting points
from each other to make a lattice in that space, and if that lattice turns out to be
a subring of the space without needing innitely many points to generate it, then
it is an order).
So does this denition satisfy the three properties above? Since 1 is noetherian
and the 1-order is nitely generated, must also be noetherian for both left
and right ideals. So we will just call noetherian to indicate both left and right
noetherian. Thus property 1 above is satised for and we are o to a good
start. After a few examples of some orders we will focus our aims at the next two
properties.
Example 2.1.2 If we let be the : : matrices over 1, `
a
(1), then the homo-
morphism 1 : r r1
a
shows is a 1-algebra. Let be the subring and
1-module `
a
(1). See that is generated by {(1)
i)
0 i. , :} where (1)
i)
is the matrix with a 1 in the i,
|
position and zeros elsewhere and so is nitely
generated by :
2
elements. Since 1 = and contains the identity 1
a
with 1s
in the main diagonal and zeros elsewhere, it is a unital subring and a full 1-lattice
in , so an 1-order in .
Example 2.1.3 Let 1 = and = i , / be the rational
quaternions, where = i, / is an extension of called the quaternions
where i
2
= ,
2
= /
2
= i,/ = 1. This is a -algebra since is embedded in .
If we let be the subring i , /, then is also a nitely generated
-module containing the identity 1 with = , so it is a -order in .
Example 2.1.4 It is important to note that both the algebras in the previous two
examples were separable and semisimple. In fact, both were simple algebras. But if
orders are to generalise the notion of a Dedekind domain, then letting = 1 and
= 1 should work and as expected, it does.
In Chapter 1 we dened the inverse of a fractional 1-ideal 1 in (1.3) by elements
r 1 that sent r1 inside 1, and then showed in Proposition 1.3.2 that {r 1
r1 1} = 1. So what if we consider something similar for a full 1-lattice ` in
? Take the set o = {r r` `}. Then o is a unital subring of and
clearly an 1-module. But what else can we say about it?
14
Since ` is a full 1-lattice in , r` is an 1-lattice in for every r (not
necessarily a full 1-lattice). Now we can write r = /
1
:
1
+ /
a
:
a
for some /
i
1
and :
i
` since 1 ` = . Since /
i
1, there is an :
i
1 such that :
i
/
i
1.
Thus there is an : 1 such that :r = :/
1
:
1
+ +:/
a
:
a
` and thus :r o.
So r = :
1
:r 1 o and we have 1 o. Hence = 1 o. If we now take
r = 1, then there is an : 1 with : `. So o: ` and o :
1
`. Since
:
1
` is an 1-lattice by above thus so is o. Hence o is actually an 1-order in .
Similarly, the set {r `r `} is an 1-order and we can dene the following.
Denition 2.1.5 The left and right orders of ` are dened respectively as
C
|
(`) = {r r` `}.
C

(`) = {r `r `}.
Now returning to before to see if the other two properties hold, we rst note
that they do not quite do what we want them to at the moment. Looking at the
second property, integrally closed is dened for a ring with respect to its quotient
eld, but we only want to look at an 1-order of , not the whole of or of the
integral closure of . So we just want an 1-order to be integral over . This query
is then solved by the following theorem.
Theorem 2.1.6 The 1-order is integral over 1 and thus the minimal polynomial
of every c over 1, j
c
(A), is in 1[A].
Proof. Let r . Since is an 1-lattice and 1[r] , then 1[r] is a nitely
generated 1-module. By Proposition 1.1.1 it follows that r is integral over 1.
Since r was arbitrary, we have integral over 1. The second statement is then a
result of Proposition 1.1.2.
Example 2.1.7 As in Example 2.1.4, letting = 1 and = 1 gets 1 being
integral over 1 as expected.
We now have properties 1 and 11 and look at the third. From here we let be
an 1-order in and ` be a full 1-lattice in . We dened a prime ideal of an
integral domain in Chapter 1 and here we dene one in an order, referencing Reiner
[1, p.190] and Swan & Evans [2, p.89]. Since 1 is commutative, ideals in 1 are both
left and right ideals. In however, left ideals are not necessarily right ideals. So
at rst we will stick only to two-sided ideals (unless stated otherwise) and look at
one-sided ideals later in the chapter. When prime ideals are spoken about it will be
clear what kind of prime is being referred to, to avoid confusion. We assume that
all ideals of are full 1-lattices in .
Denition 2.1.8 A prime ideal of is a proper two-sided ideal 1 that is a
full 1-lattice in such that for two-sided ideals 1. J in , if 1J 1 then 1 1
or J 1.
15
The multiplication of ideals here is the same as in (1.2) and note that the last
part of the denition is equivalent to saying for two-sided ideals 1. J in ,1,
if 1J = 0 then 1 = 0 or J = 0. (2.1)
reminiscent of an (possibly integral) domain.
Example 2.1.9 Every maximal two-sided ideal ` of is a prime ideal of . By
maximal, it is meant that ` is a proper ideal of that is not contained in any
other proper ideal of .
Proof. This example requires some proof, so we sill do it here. Let ` be a two-sided
maximal ideal of . Two two-sided ideals 1. J of that have 1J `, fall into the
following cases:
1. If one of the ideals contains `, say 1 `, then either 1 = ` or 1 = . The
rst shows 1 ` and the second shows J = J = 1J `, so ` is prime.
2. Otherwise let 1 ` and J `. Since 1J `, (1 + `)J `. But
1 + ` ` and so 1 + ` falls into the rst case. So if 1 + ` = `, then
1 ` and if 1 +` = , then J ` and again ` is prime.
This example can also be proven a dierent way by using (2.1) and the following
Lemma, which will be proven for ideals in general, not just two-sided ideals.
Lemma 2.1.10
a) For an ideal ` of , `,` is an ideal of ,` if and only if ` is an ideal of
containing `.
b) ,` is simple if and only if ` is a maximal two-sided ideal in .
Proof. Consider left ideals and let ` be one such of . To Prove part a, rst let
`,` be an ideal of ,`. Note that ` ` since otherwise `,` is not well
dened. Let : ` and r , so : + ` `,` and r + ` ,`. Then
(r + `)(: + `) = r: + r` + `: + `` `,`. But `: + r` + `` `,
so r: ` must follow. Thus ` is an ideal of containing `. Now to prove
the converse, let ` be an ideal of containing ` and show `,` is an ideal in
,`. Again let : ` and r , so : + ` `,` and r + ` ,`. Then
(r + `)(: + `) = r: + r` + `: + `` `,` with `: + r` + `` `
and r: `. Thus (r +`)(: +`) `,` and `,` is an ideal in ,`, proving
part a.
The proof of part b follows from part a since if ,` is simple then there are
no proper two-sided ideals of strictly containing `, so ` is a maximal two-sided
ideal. Conversely, if ` is a maximal two-sided ideal, then the only two-sided ideals
of ,` are `,` and ,`, that is, the trivial ones. So ,` has no non-trivial
two-sided ideals and is thus simple, proving part b.
This proves the Example 2.1.9 trivially since simple algebras have no non-trivial
ideals, but it also gives us an important application of the next Theorem, which can
be found in Reiner [1, p.190-191] and also in Swan & Evans [2, p.95-96]. The latter,
however, uses a dierent proof to the one given here, instead using the Chinese
16
Remainder Theorem. But rst we must state a standard theorem from the theory
of modules, which can be found in Curtis & Reiner [4, 25].
Theorem 2.1.11 The ring 1 is semisimple as a left 1-module if and only if every
left 1-module is semisimple. Every semisimple ring 1 can be decomposed into a
sum of minimal left ideals 1c
i
of 1 where c
i
are orthogonal idempotents. That is
c
2
i
= c
i
and for i = ,. c
i
c
)
= 0.
Theorem 2.1.12 If 1 is a prime ideal of then
a) 1 = 11 is a prime ideal of 1. Moreover 1 1 gives a surjective relation
between the prime ideals of and the prime ideals of 1.
b) ,1 is a nite dimensional simple algebra over the eld 1,1.
Proof. Let 1 be a prime ideal of , 1 = 1 1 and = ,1. We prove part a
rst as we need 1,1 to be a eld. Since 1 is a full 1-lattice in , by the argument
before Denition 2.1.5 there is an : 1 such that : 1, so 1 is not empty. Also,
since 1 . 1 , 1 and hence 1 , 1, implying 1 1. Now 1 is an ideal of 1
since 11 = 1(1 1) = (11) 1 = 1 1 = 1 since 1 is a full 1-lattice in .
Now if c. 1 satisfy c 1, then (c)() 1. So c 1 or 1,
implying that c 1 or 1 and 1 is a prime ideal of 1.
Now let 1 be a prime ideal of 1. Since is an 1-lattice and 1 1, 1 is
a proper two-sided ideal of and so is contained in some maximal ideal 1 in .
Now 1 1 1 1 and since both 1 and 1 1 are prime in 1 it follows that
1 = 1 1 proving the surjectivity of the relation 1 1 = 1 1 and thus part
a.
For the proof of part b, 1,1 is a eld so is trivially artinian as a ring. Now is a
nitely generated 1-module, so ,1 must be a nitely generated 1,1-module and
is thus artinian, as well as a nite dimensional 1,1-algebra since 1,1 2(,1).
Suppose rst that ,1 is semisimple. Now two separate simple components
i
and

)
of have
i

)

i

)
= 0, so by (2.1) one of them must be zero. Thus
there can only be one simple component and ,1 is actually simple. So it suces
to show that ,1 is semisimple.
Suppose that there is a two-sided ideal A 1 such that A,1 0 is minimal.
Such an A exists since ,1 is artinitan. Now (A,1)
2
A,1, so since A,1 is
minimal either (A,1)
2
= 0 or (A,1)
2
= A,1. The rst implies A,1 = 0 via (2.1)
so it must be the latter. But in that case, it must be also that A
2
= A in . Let
A = 1r
1
+ 1r
a
for some r
i
, :
+
and no r
i
equal to a sum

)=i
:
)
r
)
or product

)=i
:
)
r
)

I=i
:
I
r
I
involving the other terms of A since if any r
i
is,
removing that r
i
will not aect the ideal A (so we have a minimal amount of r
i
).
Then A
2
= 1r
2
1
+ 1r
a
r
1
+1r
1
r
2
+ 1r
a
r
2
+ +1r
1
r
a
+ 1r
2
a
and thus
since each r
i
cannot be expressed in terms of the other elements of A, they must
all be idempotent (that is r
2
i
= r
i
). But as A
2
= A then r
i
r
)
= 0 (and so they are
orthogonal idempotents) must also be the case for all i = ,. By Theorem 2.1.2.1
this means A is actually semisimple. Doing similarly for another minimal two-sided
ideal 1 A it follows that 1 is a direct product of ideals generated by idempotents
of that cancel each other so is semisimple as well. Thus every two-sided ideal in
,1 is semisimple, so by Theorem 2.1.2.1,1 is semisimple itself.
17
This result is quite remarkable and provides a correspondence between the orig-
inal denition of a prime ideal in Chapter 1 and the abstracted denition here in
Chapter 2. Just as importantly, the succeeding corollary of Theorem 2.1.12 allows
us to get the last property we require of , property 111.
Corollary 2.1.13 The prime ideals of are maximal two-sided ideals in .
Proof. For a prime ideal 1 of , Theorem 2.1.12 says that ,1 is simple and
Lemma 2.1.10 then implies 1 is maximal.
This gives us the equivalence of maximal ideals and prime ideals of , so the
order satises the three properties held by a Dedekind domain and we are o to
a good start. Though this is very nice, it still does not mean too much to us about
the factorisation of ideals in this order without actually looking at the ideals of .
So next we will discuss the ideals of , with references for the next section Reiner
[1, p.108-111,192,204-205], Swan & Evans [2, p.83-84,88] and Bass [3, p.152-156].
2.2 Maximal Orders
In the last section, we went briey into the ideals of to show property 111 of
a Dedekind domain held for . Left ideals ` in all have ` = ` and are
assumed to be full 1-lattices in , but what if we consider such ` not contained
in . We did this in Chapter 1 with the denition of a fractional ideal and saw
that each fractional ideal could be factorised uniquely as a product of prime ideals
and inverses of prime ideals and also that the set of these fractional ideals was an
abelian group generated by the prime ideals and their inverses. So we search for
the Irishman the the end of the rainbow and again we do it here with a similar
denition with the aim of seeing if a similar factorisation occurs in an 1-order .
Denition 2.2.1 For the 1-order , a non-zero full 1-lattice in that is a nitely
generated -module is called a fractional -ideal of .
Example 2.2.2 As in Example 2.1.7, letting = 1 and = 1 gets the fractional
-ideals here and the fractional 1-ideals in Chapter 1 coinciding.
We will denote fractional -ideals as we did in Chapter 1 by using 1. J and so
on and as usual the product of two fractional -ideals is another fractional -ideal
given by (1.2). We do however drop the word fractional for ease and just refer to
them as ideals of , assuming that they are in fact fractional -ideals of . If a full
1-lattice is a two-sided ideal of , it is then just an ideal of that is a two-sided
-module. We will soon only look at two-sided ideals and then look at a more
general case later, but for now this two-sidedness is not assumed. First note the
useful fact via the argument given before Denition 2.1.5, that for every ideal 1 of
, the intersection 1 1 = 0.
The ideals considered from Denition 2.1.8 until this point are all fractional -
ideals since they are full 1-lattices in and are contained in , an 1-lattice itself,
implying they are nitely generated -modules. To emphasize when an ideal 1 of
is contained in we call 1 an integral ideal of . As usual, 1 is a maximal integral
ideal of when it is contained properly in no other integral ideal of .
So now we have generalised some of the denitions of Chapter 1, in particular
Dedekind domains and the corresponding fractional ideals. But is this all that is
18
required to generalise the results for Dedekind domains from Chapter 1? At the
moment there are no restrictions on the order we choose. In particular, as we will
soon nd to be important, we have not established any sort of maximality for orders,
which leads to the following.
Denition 2.2.3 An 1-order that is not properly contained in any other 1-order
is called a maximal 1-order.
Example 2.2.4 As in Example 2.2.2 (and as mentioned just before the previous
denition), if = 1 and = 1 then is a maximal 1-order in since adding
any element 11 to 1 gets an innitely generated 1-module, so 1+ cannot
be an 1-order.
Example 2.2.5 As in Example 2.1.2, letting = `
a
(1) and = `
a
(1), then
was an 1-order in . Actually, ` is a maximal 1-order in since if you add
another matrix ` to with, say, an element 1 1 in the |/
|

position. Since is generated by the matrices {(1)

i)
0 i. , :} where (1)
i)
is
the matrix with a 1 in the i,
|
position and zeros elsewhere, the element can be
placed in any position. Thus any multiple of can be placed in any position and
similarly to Example 2.2.4 +` is not an 1-order.
Example 2.2.6 As in Example 2.1.3 with = i , / and =
i , /, then was an order in . But is not a maximal order. Let
c = (1+i+, +/),2 and

= i, c. Then

is a subring (since it can be

written as

= {(c+/i+c,+d/),2 c / c d mod 2) that contains (since

/ = 2c1 i ,) and is generated by 1. i. ,. c, so is an -order of . Further, it is
maximal. This follows since

is a non-commutative PID, which will be looked at

in Chapter 4. So suppose there is an order

, then is a -ideal also. There

turns out to be an : 1 such that :

(which will be an element of

1
). So
: =

r for some r . But then = C

|
() = C
|
(:) = C
|
(

r) = C
|
(

) =

and

is maximal.
The reason we need maximal orders will soon become apparent as we look into
the ideals of . But rst we must make one more extrapolation from Chapter 1.
As mentioned earlier, in (1.3) we dened for a fractional 1-ideal 1 a set of elements
r 1 that sent r1 inside 1, which happened to be the inverse of 1 in
1
. This
becomes a slight problem here due to the issue of non-commutativity, since for a
fractional -ideal 1 and element r , r1 is not necessarily the same as 1r. So
let o
|
= {r r1 } and o

= {r 1r }. For arguments sake, let 1

be a two-sided ideal, so for r o
|
we have r1 , so 1r1 1 and similarly for
any r o

. 1r1 1 also. So let o = {r 1r1 1}. We have just seen that

o
|
o

o, but are they equal? In general the answer is no.

The most we can say is that, since 1o1 1, we have 1o C
|
(1) and o1
C

(1). But if is a maximal order the the answer is better than yes. In this case
C
|
(1) , so C
|
(1) = and similarly C

(1) = . So we have o = {r 1 r
C
|
(1)} = {r 1 r } = o

and also o = {r r 1 C

(1)} = {r
r 1 } = o
|
, exactly what we want since it is the counterpart to (1.3) in a
maximal order. That is, when is maximal we get o
|
= o

= o. So we will use this

19
idea as motivation and dene in general for an ideal 1 (not necessarily two-sided)
of
1
1
= {r 1 r 1 1}. (2.2)
or equivalently
1
1
= {r 1 r C
|
(1)} = {r r 1 C

(1)}. (2.3)
So we then see that when is maximal and 1 is strictly two-sided
1
1
= {r 1 r } = {r r 1 }. (2.4)
We see that 1
1
is clearly an 1-module, but is it an ideal in ? Since there
is an : 1 1, then :1
1
C
|
(1). Thus 1
1
:
1
C
|
(1), which shows 1
1
is
nitely generated and 1-torsionfree and thus an 1-lattice. If = C
|
(1), then as
in the discussion before Dention 2.1.5 there is an : 11 such that :1 . So
1 :
1
and : = (:
1
) = {r :
1
r } {r 1r } = 1
1
.
So 1
1
is full and indeed an ideal of itself. Similarly, 1
1
is an -ideal also when
= C

(1).
We know now that 1
1
is an ideal, but what else do we know? Using (2.3) we
have that 11
1
C
|
(1), so 1(1
1
C
|
(1)) C
|
(1) and thus 1
1
C
|
(1) 1
1
which
shows 1
1
is a right C
|
(1)-module. Similarly, 1
1
1 C

(1) gets (C

(1)1
1
)1
C

(1) and C

(1)1
1
1
1
which shows that 1
1
is a left C

(1)-module. These
also give us two nice containments,
C

(1
1
) C
|
(1) and C
|
(1
1
) C

(1). (2.5)
This allows us to think of the ideals of an order as a set that contains inverses,
but we want to see if it is a group. Before we see if this is the case, we use the
preceding results to aquire the following useful facts. For the full 1-lattice 1, if
C
|
(1) is maximal we have 11
1
C
|
(1) by denition. Firstly, since 1
1
is an ideal
and the product of two ideals is another ideal, it follows that 11
1
is an integral
ideal in C
|
(1). Now by (2.5) C

(1
1
) = C
|
(1) and thus 11
1
is actually a two-
sided integral ideal of C
|
(1). Secondly, if J 1, then J1
1
11
1
C
|
(1) and
so 1
1
J
1
. Reiterating this instead with C

(1) maximal gets 1

1
1 a two-sided
integral ideal of C

(1) and for J 1, yet again 1

1
J
1
. Repeating the second
fact here, we get
if J 1 then 1
1
J
1
. (2.6)
2.3 The Factorisation of Two-Sided Ideals
So now we have the necessary utensils to see if the set of these fractional ideals is a
group. We do however have to revert to the case of two-sided ideals (two-sidedness
will be stated explicitly), like in the previous section. But this turns out to be
necessary here for our new aim, to see if the two-sided ideals of can be factorised
in a similar fashion to the ideals in a Dedekind domain (which may not be the
case for one-sided ideals). So we continue on with this, trying to establish a group
20
structure along the way. The method we will use to see if these cases hold may seem
familiar to the reader and can be found in Reiner [1, p.204-207], Swan & Evans [2,
p.88-92] and Bass [3, p.155-159]. In fact most of the results (and proofs) in this
section naturally follow from their counterparts in Chapter 1. For example, the
next three results resemble Proposition 1.3.1 and Corollary 1.3.3 and are again our
rst stepping stones.
Lemma 2.3.1 Every two-sided ideal of contains a product of prime ideals of .
Proof. Note rst that ideal here means two-sided ideal and any ideal 1 contains
any product of prime ideals of , so only consider integral ideals from hereon. Let
o be the set of proper ideals of that do not contain a product of prime ideals.
Since is noetherian there must exist a maximal ideal ` in o. Now ` cannot be
prime and any ideal properly containing ` must contain a product of prime ideals.
Since ` is not prime, there are two two-sided ideals 1. J such that 1. J `
and 1J `. But then (1 +`)(J +`) ` also and both 1 +`. J +` `.
Both 1 + ` and J + ` are two-sided ideals and by the maximality of ` they
contain a product of prime ideals. But then so does ` since (1+`)(J +`) `,
contradicting the existence of a maximal element of the set o, and o is empty.
Lemma 2.3.2 Let 1 be a two-sided ideal of the maximal order . If 1 , then
1
1
.
Proof. Let 1 be a two-sided ideal of . Clearly 1
1
, so suppose 1
1
= .
Then there is a maximal ideal 1 such that 1 1 and let : 11. Then by
Lemma 2.3.1, : contains a product of prime ideals 1
i
1 : 1
1
1
2
1
a
and choose these prime ideals such that : is minimal. If 1 = 1
i
for every i, then
there are r
i
1
i
where r
i
, 1 for every i but r
1
r
2
r
a
1. So r
1
r
2
` r
a

1 but r
i
1 for every i. So since 1 is prime and r
1
1, then r
2
r
a
1
must follow. But then r
2
1, so again r
3
r
a
1 follows. Doing this
another : 3 times gets either r
a1
1 or r
a
1, which is a contradiction.
Thus 1 = 1
)
for some , and
1 : 1
1
1
)1
11
)+1
1
a
.
Let 1 = 1
1
1
)1
and C = 1
)+1
1
a
and see that 1 and C are both two-
sided ideals of . So 11C : gets :
1
11C and 1:
1
1C1 1. Thus
:
1
1C1 C

(1) = and so by (2.6) we get :

1
C1 1
1
1
1
= . But this
gets C1 :, so : contains a product of : 1 prime ideals of , contradicting
the minimality of :. Thus 1
1
= .
Theorem 2.3.3 Let 1 be a full 1-lattice of . If C
|
(1) is a maximal order then
1 1
1
= C
|
(1). Similarly, if C

(1) is a maximal order then 1

1
1 = C

(1).
21
Proof. Suppose C
|
(1) is maximal. We know that 11
1
is a two-sided ideal con-
tained in = C
|
(1). Then we also get (11
1
)(11
1
)
1
, implying 1
1
(11
1
)
1

1
1
and thus (11
1
)
1
C

(1
1
). But by (2.5) C

(1
1
) , a maximal order, so
C

(1
1
) = and we get (11
1
)
1
. By Lemma 2.3.2 it follows that 11
1
= .
The second part of the theorem works similarly, but will be shown to indicate the
importance of this result. Again, 1
1
1 is a two-sided ideal contained in

= C

(1),
so (1
1
1)
1
(1
1
1)

gets (1
1
1)
1
1
1
1
1
and hence (1
1
1)
1
C
|
(1
1
).
By (2.5) C
|
(1
1
)

, a maximal order, so C
|
(1
1
) =

and (11
1
)
1

,
where Lemma 2.3.2 again nishes o the proof.
The preceding theorem (which was proven for a full 1-lattice, not a two-sided
ideal) gives us exactly what we want when is a maximal order since if is
maximal, for every two-sided ideal 1 of we have C
|
(1) = C

(1) = . Also, see

that 11
1
= (1
1
)
1
1
1
. But does (1
1
)
1
= 1? In fact it does, not only for
two-sided ideals, but also for one-sided ideals as the following result shows.
Proposition 2.3.4 Let 1 be a full 1-lattice in and = C
|
(1) or C

(1). If
is a maximal order then (1
1
)
1
= 1. Also, any ideal of that contains is the
inverse of an integral ideal of .
Proof. Let = C
|
(1) be maximal. Since 1 is a full 1-lattice, then so is 1
1
. So by
Theorem 2.3.3 and (2.5) we have (1
1
)
1
1
1
= C

(1
1
) = . So (1
1
)
1
1
1
1 =
(1
1
)
1
C

(1) = 1 and thus (1

1
)
1
1 since C

(1) is a unital subring. To show

the reverse inclusion, by (2.5) we get
(1
1
)
1
= C
|
((1
1
)
1
)(1
1
)
1
C

(1
1
)(1
1
)
1
= (1
1
)
1
.
This then gets
(1
1
)
1
11
1
(1
1
)
1
= 1C
|
(1
1
) 1C

(1) = 1.
and the reverse inclusion holds, implying (1
1
)
1
= 1. The proof is done similarly
for = C

(1).
For the second part, again let = C
|
(1) be maximal but also 1. Then
1
1
11
1
= , so 1
1
C

() = and 1
1
is integral. Letting J = 1
1
, we
get J
1
= (1
1
)
1
= 1, so 1 is the inverse of an integral ideal of . Again, the
proof is done similarly for = C

(1).
So now we only need one more lemma to get what we want, and then our
objective of ideal factorisation into prime ideals is done. Unlike the Dedekind
domain case, we have the answer to our aim before before we certify that the set
of fractional ideals is a group. By what we have so far, products of ideals of are
also ideals and this is associative since is, acts as an identity and there is an
inverse for every ideal of . So we have a group structure, but we want to know if
it is abelian since at the moment there is no reason to see that it is. Referencing
Reiner [1, p.206-207], Swan & Evans [2, p.91-92] and Bass [3, p.158], this query is
also answered by the following, which achieves our aim.
22
Lemma 2.3.5 Let be a maximal order. Then multiplication of prime ideals of
is commutative and every integral two-sided ideal of can be written uniquely as
a product of prime ideals of up to rearrangement.
Proof. Let 1 an C be distinct prime ideals of . Then see that 1
1
C1 1
1
1 =
by Theorem 2.3.3. Now 1(1
1
C1) = C1 C, but since C is prime either
1 C or 1
1
C1 C. The rst implies 1 = C, which cannot be true, so we
must have 1
1
C1 C and thus C1 1C. Doing exactly the same procedure
with 1 and C swapped gets the reverse inclusion and thus 1C = C1, showing
multiplication of prime ideals is commutative.
Now for the second part of the proof, suppose that there are two-sided -ideals
that cannot be written as a product of prime ideals and let the set of these be o.
Then there is a maximal two-sided ideal ` o which obviously is not prime. So
there is a prime ideal 1 such that ` 1 and hence 1
1
`
1
and
multiplying by ` gets ` `1
1
. If ` = `1
1
then 1
1
C

(`) = ,
contradicting Lemma 2.3.2. So ` `1
1
and thus `1
1
is a two-sided ideal
strictly containing `, so by the maximality of ` `1
1
is expressible as a product
of prime ideals, say `1
1
= 1
1
1
2
1
a
for some :. But then ` = 1
1
1
a
1, a
product of prime ideals, contradicting the existence of a maximal element of o and
thus o is empty.
Finally we must show that if two expressions of prime ideals are equal then
they are the same prime ideals, but rearranged. So for prime ideals 1
i
and Q
)
, let
1
1
1
2
1
I
= Q
1
Q
2
Q
|
. since both expressions are a product of prime ideals,
they are contained in some prime ideal, say 1. So 1 1
1
1
2
1
I
and by the
proof of Lemma 2.3.2 it follows that 1 = 1

for some :. Similarly 1 Q

1
Q
2
Q
|
and so 1 = Q
-
for some :, so we get
1
1
1
1
11
+1
1
I
= Q
1
Q
-1
1Q: + 1 Q
|
1
1
11
1
1
1
1
+1
1
I
= 1
1
1Q
1
Q
-1
1: + 1 Q
|
1
1
1
1
1
+1
1
I
= Q
1
Q
-1
Q: + 1 Q
|
.
Doing this another / 2 times gets / = | and each 1
i
coinciding with some Q
)
.
Thus the factorisation is unique up to rearrangement.
So as in Chapter 1 we can write two-sided ideals 1 of in the form 1 =
1
q
1
1
1
q
2
2
1
q
u
a
for distinct prime ideals 1
i
of and positive integers
i
. But how
do we write 1
1
? If we look only when is maximal so 11
1
= , then similarly
to Chapter 1 yet again, we see that due to the commutativity of the prime ideals,
1
1
= 1
q
1
1
q
2
1
q
u
. This observation then leads to the major theorem of
this section and our aim, the generalisation of Theorem 1.3.8. But rst, we state a
corollary, which yet again generalises the notion of a greatest common divisor and
lowest commom multiple from Chapter 1.
Corollary 2.3.6 The greatest common divisor of two ideals 1. J of , denoted
gcd(1. J), and lowest common multiple, denoted lcm(1. J), both exist and are a
product of prime ideals of .
23
Proof. Write 1 = 1
c
1
1
. . . 1
c
u
a
and J = 1
)
1
1
. . . 1
)
u
a
where c
i
and )
)
are natrural
numbers as in the previous lemma. Then gcd(1. J) = 1
min{c
1
,)
1
}
1
. . . 1
min{c
u
,)
u
}
a
and
lcm(1. J) = 1
max{c
1
,)
1
}
1
. . . 1
max{c
u
,)
u
}
a
.
Theorem 2.3.7 Let be a maximal order. Then the set of two-sided ideals of
.

, is an abelian group under multiplication generated by the prime ideals of

with identity and inverse of two-sided ideal 1

given by 1
1
.
Proof. This will be proved by showing every two-sided ideal of can be written
as a product of prime ideals of and prime inverses. Let 1 be an ideal of .
If 1 , then this follows straight away from Lemma 2.3.5 so let 1 . Let
: 1 1
1
so :1 is integral and by Lemma 2.3.5 can be written as a
product :1 = 1
1
1
a
of prime 1
i
ideals of . But :1 = (:)1, so since is
an 1-module we can also write : = Q
1
Q
n
for prime ideals Q
)
of . Thus
1 = (:)
1
1
1
1
a
= Q
1
1
Q
1
n
1
1
1
a
is written as a product of prime ideals
of and prime inverses. This proves the theorem.
So just like the fractional ideals of a Dedekind domain, Theorem 2.3.7 shows
every two-sided fractional ideal 1 of a maximal order can be written as 1 =
1
c
1
1
1
c
2
2
1
c
u
a
for distinct prime ideals 1
i
of and each integer c
i
= 0. That is,
there is unique factorisation of two-sided fractional -ideals up to rearrangement.
Example 2.3.8 We look at factorising some two-sided ideals in a maximal order in
the rational quaternions given in Example 2.2.6. So we let = i , /
and = i , c where c = (1 +i +, +/),2. Then we saw that was a
maximal order (note the change in notation from that example) that could also be
written as = {(c +/i +c, +d/),2 c / c d mod 2}.
So what are the maximal two-sided ideals in ? It turns out that is actually
a non-commutative version of a PID (see Chapter 4), so every left ideal in can
be written as r for some r . As it is a PID, there are irreducible elements
in also. It turns out (which will be proven again in Chapter 4) that if r is
irreducible, then writing r = c +/i +c, +d/ gets the norm of r to be
r = rr = (c +/i +c, +d/)(c /i c, d/) = c
2
+/
2
+c
2
+d
2
.
Furthermore, it is a positive prime integer. So from this we see that all maximal
ideals and maximal two-sided ideals are either of the form r or r for some
irreducible r . Visually,
{Maximal Left -ideals} {Maximal Two-sided -ideals}

{-ideals of form r or r Irreducible r }.

So when is an an ideal two-sided? Clearly the ones of the form r are two-
sided, but are there any irreducible r such that r is a two-sided ideal? Straight
away we can see that (1+i) is actually two-sided since (1+i)(u+ri +, +./) =
(u+ri ., +/)(1 +i). So are there any more? If we let r = c +/i +c, +d/
24
be irreducible, if r = 2, then two of c. /. c and d are 1 and the other two 0, so it
is identical to the case 1 + i, so lets look when r 2.
If r is two-sided, then for every . there must be a ` such that .r = r`,
so ` =
a:a
a
. That is,
aa
a
= . When substituting . = 1 +, and rearranging, we get
that c
2
+/
2
+c
2
+d
2
/ccd. When . = i we get that c
2
+/
2
+c
2
+d
2
cd+/c. cc/d.
Thus it must be that c
2
+ /
2
+ c
2
+ d
2
/c (since c
2
+ /
2
+ c
2
+ d
2
2), implying
that one of / and c is 0. But then c
2
+/
2
+c
2
+d
2
cd +/c = cd implying that one
of c and d is 0. So without loss of generality r = c +/i. By substituting . = 1 +,
again yields c
2
+/
2
2c/ so one of c and / must be zero, implying a prime number
is an integer squared which is obviously nonsense.
So by this long-winded argument every ideal of the form r for r irreducible
with r 2 a maximal left ideal. Thus r are the maximal two-sided -ideals,
that is, the prime ideals of in addition to (1 + i). But what about 2? Lets
apply some of the theory. Since 2 = (1 +i)(1 i) we have that
2 = (1 +i)(1 i) = ((1 +i))
2
.
Now for any positive rational :, write : = 2
o
j
c
1
1
j
c
l
l
q
]
1
1
q
]
o
o
where each j
|
and
|
is a
positive odd prime and c
|
. )
|
0 (assuming no common factors on numerator and
denominator), but c a non-zero integer. Then we have
: = (2
o
)(j
c
1
1
) (j
c
l
o
)(
)
1
1
) (
)
o
o
)
= (2)
o
(j
1
)
c
1
(j
o
)
c
l
(
1
)
)
1
(
o
)
)
o
= ((1 +i))
2o
(j
1
)
c
1
(j
o
)
c
l
(
1
)
)
1
(
o
)
)
o
.
Which seems almost identical to the factorisation of the ideals in the Dedekind
domain given in Example 1.3.9. Note if : is negative, then there is no dierence
since : = (:). Finally the set of two-sided ideals in ,

, is just the set

{. (1 +i).
1 +i
2
} {3. 5. 7. 11. } {
1
3
.
1
5
.
1
7
.
1
11
. }.
25
Chapter 3
Maximal Orders and Groupoids
The nal result of the last chapter is indeed a nice consequence for two-sided ideals
in a maximal 1-order over a nite dimensional separable semisimple 1-algebra,
but if that is the only gold at the end of the rainbow, then the Irishman has run
away with our gold and left only a few pieces behind. So we chase after him by
now considering one-sided ideals to see yet again if they can be factorised uniquely
in terms of irreducible ideals and not only this, but we will again see if the set of
all these one-sided ideals forms a group. So now we focus at both of these aims.
Strictly one-sided ideals, however, cannot be a product of prime ideals since prime
ideals here are by dention two-sided. But there is still a relation between the prime
ideals and maximal one-sided ideals, and this lends a hand later in the piece. So
from hereon we let all ideals be one-sided unless stated otherwise.
3.1 One-Sidedness
To start ourselves o, we will actually give a conclusion to the previous chapter
that shall give us some preliminary argument towards our new aim. In the previous
chapter we saw that the set of two-sided ideals,

, of the maximal 1-order in the

nite dimensional separable semisimple 1-algebra was actually an abelian group
generated by the prime ideals of . But if we look at two dierent maximal orders,
do we get two dierent groups? This query will be answered shortly, but rst we
need some requisites. The reference for this section is Reiner [1, p.197-199,204-205].
So similarly to last section, let 1 be a Dedekind domain with quotient eld 1,
a nite dimensional separable semisimple 1-algebra and an order of .
If we take a full 1-lattice ` in , then C
|
(`) is not necessarily the same as
C

(`). In fact, if

= C
|
(`) = C

(`) then ` is a two-sided fractional

-ideal,
so if ` is strictly a one-sided -ideal for some order , then C
|
(`) = C

(`). This
then introduces some problems. Firstly, what is an integral ideal now in the one-
sided case? Secondly, given two orders and

, is there an -ideal (also

-ideal)
1 with C
|
(1) = and C

(1) =

and does the multiplication of ideals now change

since C
|
(1) = C

(1)?
Upon considering the denition of an integral ideal: a -ideal 1 is an integral
ideal of if 1 ; It is seen that this is independent of left or right multiplication,
so still holds in the one-sided case. Similarly, so does the denition for a maximal
integral ideal. However, it would be nice to know when integrality holds for an
order on each side. We get some insight into this if the left ideal 1 has = C
|
(1),
allowing us the following result, taken for a full 1-lattice in .
26
Proposition 3.1.1 For any full 1-lattice ` in , ` C
|
(`) if and only if
` C

(`). That is, ` is an integral ideal of C

|
(`) if and only if ` is an
integral ideal of C

(`).
Proof. Let ` be a full 1-lattice in and let ` C
|
(`). Then `` C
|
(`)` =
` and it follows by deniton that ` C

(`). Similarly, if ` C

(`), then
`` `C

(`) = ` and ` C
|
(`), completing the proof.
Now looking at the second problem, we see that the multiplication of ideals is as
usual dened by (1.2) since we are in the algebra . But for the full 1-lattices `
and `. `` = (`C

(`))` = `(C

(`)`). So it would again seem optimal that

C

(`) = C
|
(`). Thus for full 1-lattices `
1
. `
2
. . `
a
the product `
1
`
2
`
a
is called proper if C

(`
i
) = C
|
(`
i+1
) for every i {1. 2. . :1}. It is then seen
that
C
|
(`
1
`
2
`
a
) C
|
(`
1
) and C

(`
1
`
2
`
a
) C

(`
a
). (3.1)
Now for two orders and

, is there a full 1-lattice ` with C

|
(`) = and
C

(`) =

? The answer in general is not clear, but this is not needed. If and

are maximal, then the answer is yes. Let ` =

, then C
|
(`) , so
C
|
(`) = and similarly C

(`) =

, so ` shows this (Note that the fact ` was

not constructed via a proper product is not important, since all we wanted was for
` to be a left -ideal and right

-ideal).
Using this, we create some notation by writing a full 1-lattice ` as `
i)
where
we denote
i
= C
|
(`) and
)
= C

(`). Thus a proper product of full 1-lattices

can be written as `
12
`
23
`
a a+1
. So for a full 1 lattice ` = `
i)
with both

i
and
)
maximal, we still have the inverse dened by (2.2) or (2.3). But can
we write `
1
in the same way? By (2.5), we have C
|
(`
1
) C

(`) =
)
, so
C
|
(`
1
) =
)
and similarly C

(`
1
) =
i
. Hence inverting ` swaps the orders
over and we write `
1
= (`
i)
)
1
= `
1
i)
seeing in fact that C
|
(`
1
i)
) =
)
and
C

(`
1
i)
) =
i
.
One nal thing to note for a proper product is given in the following result.
Proposition 3.1.2 A proper product of integral ideals is an integral ideal.
Proof. Let `
12
`
23
`
+1
be a proper product of integral ideals. Then
`
12
`
23
`
+1

2
`
23
`
+1
= `
23
`
+1

3
`
34
`
+1
= `
34
`
+1
.
.
.

`
+1
= `
+1

+1
.
Thus `
12
`
23
`
+1
is an integral ideal, proving the proposition.
Now returning to see if

.
=

for two maximal orders

i
and
)
, the
following result gives us the answer to our query.
27
Theorem 3.1.3 Let
i
and
)
be maximal orders in . For an ideal `
i)
with
C
|
(`
i)
) =
i
and C

(`
i)
) =
)
, the map
o
i)
:

: 1 `
1
i)
1`
i)
is an isomorphism. Furthermore, o
i)
is independent of the choice of `
i)
.
Proof. Let o = o
i)
and the ideal ` = `
i)
have C
|
(`
i)
) =
i
and C

(`
i)
) =

)
. Then for 1

.
, by (3.1) we have C
|
(`
1
1`) = C
|
(`
1
) =
)
and
C

(`
1
1`) = C

(`) =
)
, so `
1
1` is a two-sided
)
ideal and `
1
1`

.
Also see that o(
i
) = `
1

i
` = `
1
` =
)
, so the identity in

.
is sent to
the identity in

. Now to show o is a homomorphism, let 1. 1

.
. Then
o(11

) = `
1
11

` = `
1
1
i
1

` = `
1
1``
1
1

` = o(1)o(1

) and o is a
homomorphism.
Considering similarly the homomorphism o

= o
)i
sending J

to `J`
1

.
, then it is seen that o

o(1) = ``
1
1``
1
= 1 and oo

(J) = `
1
`J`
1
` =
J. So o
)i
is the inverse of o
i)
and they are both isomorphisms. Now for another
ideal ` = `
i)
, see that ``
1
and ``
1
are two-sided
i
-ideals since they both
have left and right orders equal to
i
, so they are elements of

.
. Thus since

.
is abelian, for 1

.
we have
1 = 1
i
= 1``
1
= 1`
)
`
1
= 1``
1
``
1
= ``
1
1``
1
.
So using this with o gets
`
1
1` = o(1) = o(``
1
1``
1
) = `
1
``
1
1``
1
` = `
1
1`.
So o is independent of the choice of ` = `
i)
, proving the theorem.
So we see that

.
=

does not quite hold, but

always does, show-

ing us that the group generated by the prime ideals of a maximal order essentially
does not depend on the maximal order. This is a wonderful result and it will guide
us in the right direction via two of its corollaries. However, we cannot go into these
at the moment without some further theory, so they will be left to the next section.
Instead, we start with some results about one-sided ideals, some of which generalise
further the results of Chapter 2. As before, the reference for the remaineder of this
section is Reiner [1, p.195-196,204]. This will focus on left ideals, but the right case
follows equivalently.
Getting back to one of our usual and aforementioned aims of seeing if the set
of the one-sided ideals is a group, we see that ideals are actually -ideals for some
order , so they are dependent on . Since a full 1-lattice ` in does not
necessarily have C
|
(`) = C

(`), we see that for another full 1-lattice ` such that

C
|
(`) = C
|
(`) and C

(`) = C

(`), ` and ` are left C

|
(`)-ideals, but one is
a right C

(`)-ideal, and the other a C

(`)-ideal, so considering the set of ideals

is not quite what we want. Now we want C
|
(`) or C

(`) to be maximal so that

Theorem 2.3.3 gets one of the products ` `
1
= C
|
(`) or `
1
` = C

(`)
equal to a maximal order, allowing an order to be the identity of the potential
28
group. Thus we must consider a more general case than an ideal over a maximal
order and dene the following.
Denition 3.1.4 A full 1-lattice `
i)
in is called a normal ideal of if
i
is a
maximal order in .
Example 3.1.5 For a maximal order , left ideals 1 of are normal ideals
since C
|
(1) = . So this denition is extending the reach of all the ideals we are
considering to all arbitrary maximal orders.
It will be seen more clearly why this denition is needed later. The denition,
which says the left order is maximal, however says nothing about the maximality
of
)
, so though we want the identity to be a maximal order, this is currently
only plausible for left multiplication. We soon see that for a normal ideal `
i)
that
)
must also be maximal, but we rst need to state the following theorem,
which actually needs another standard theorem, which can be found in Jacobson
[6, p.200-201].
Theorem 3.1.6 1 is a simple artinian ring if and only if 1 is artinian and has a
simple module ` such that ann
1
` = 0.
Now the following theorem is a remarkably important result and will be used
throughout the following sections. So for 1 and 1 as in the theorem below, we say
that 1 belongs to 1.
Theorem 3.1.7 Let 1 be a maximal left integral ideal of maximal order . Then
a) There is a unique prime ideal 1 of such that 1 1 and 1 = c::

,1.
Moreover 1 1 gives a surjective relation between the maximal integral
ideals of and the prime ideals of .
b) The simple ring ,1 has simple left module ,1.
c) Each prime ideal determines a maximal left integral ideal that belongs to it.
Proof. Since 1 is a maximal left integral -ideal we see that 1 = ann

,1 = {r
r 1} and it is a two-sided ideal. Choose an c 1 1, then : = : 1.
But : is a two-sided ideal, so write it as Q
c
1
1
Q
c
I
I
for disctinct prime ideals Q
i
and non-zero c
i
. Then ,: ,1 has only a nite number of two-sided ideals
and is thus artinian, so ,1 is artinian also. Now
ann
1
,1 = {r ,1 r(,1) = (,1)r = 0} = {r ,1 r. r 1} = 0
(note this is 0 in ,1), so since ,1 is a simple module (since 1 is maximal) by
Theorem 3.1.6 ,1 is simple, proving part b and implying 1 is actually a prime
ideal of . This is unique since if Q is another prime ideal contained in 1, then
Q ann

,1, implying Q = 1.
Now given any prime ideal 1 of there is a maximal left integral ideal 1
such that 1 1. Since ,(ann

,1) is simple, then 1 ann

,1 and thus
1 = ann

,1, proving part c and the surjectivity of the relation 1 1, proving

part a.
29
3.2 Generalising a Familiar Sight
So now we want to look at our aims of factorising ideals of maximal orders into
irreducible ideals and looking to see if the set of them is a group. But as we needed
to do in Chapter 2 when generalising Chapter 1, we need to extend the results of
Chapter 2 to one-sided ideals here, where the main reference is Reiner [1, p.201,207-
209]. Also, since we introduced normal ideals and intend on using them, we had
better look at them to see their properties. As we have suggested, this section is
mainly for the creation of utensils to use for looking at our aims in the next section.
Also, as suggested by the name of this section, Theorem 3.1.7 will have its rst use
in the following very familiar Lemma, which is courtesy of Reiner [1, p.207-208].
Lemma 3.2.1 Let 1 be a proper integral ideal of . Then 1
1
.
Proof. If 1 is not maximal then 1 J for some maximal integral ideal J. But then
J
1
1
1
. So it suces to prove the lemma for 1 a maximal integral ideal. Let
1 = ann

(,1) be the prime ideal to which 1 belongs. Then ,1 is a simple left

module over the simple ring = ,1. Hence 1 is the inverse image, under the map
, of some maximal left ideal of . Therefore we may write 1 = (1 c) +1,
where c is such that its image c is a primitive idempotent in . Set ` = c+1,
a right ideal in . Since (1 c)c 1, it follows that 1` 1. On the other hand,
1` is a two-sided ideal of containing 1
2
. It follows that 1` is either 1 or 1
2
.
If 1` = 1
2
, then 1c 1
2
; multiplying by 1
1
, we deduce that c 1, which is
impossible. This shows 1` = 1.
Now choose a non-zero c 1 1 and write c as a product of prime ideals
of . Since c 1, one of the factors must equal 1, and so we may write
c = 1Q = 1`Q, where Q is some two-sided ideal of . If 1
1
= , then
c
1
`Q 1
1
= implies `Q c, which in turn implies ` cQ
1
= 1,
so ` 1. But ` = c + 1, so the last inclusion is impossible. This shows that
1
1
and completes the proof of the lemma.
This then allows the following fundamental result, which was alluded to earlier
and allows the notion of a proper product to be more familiar and not a rare novelty.
Theorem 3.2.2 For a full 1-lattice `
i)
in ,
i
is a maximal order if and only if

)
is a maximal order.
Proof. Let ` = `
i)
be a normal ideal. We rst prove the theorem for integral
ideals, but see that if ` =
i
, then
)
= C

(`) =
i
is trivially maximal. So
assume from hereon that ` is proper integral. For a contradiction, suppose that
` is a maximal counterexample to the theorem. That is,
)
is not maximal and if
1
iI
`
i)
is also a left
i
-ideal, then
I
is maximal. In fact, let 1 = 1
iI
be one such
that 1,` is simple, that is, 1 is a minimal left ideal in
i
,`. Let ` = 1
1
`, then
` 1
1
1 =
I
since if ` = 1
1
` = 1
1
1 =
I
then ` = 1, a contradiction.
Also, since 1,` is simple it follows that ` is a maximal integral ideal in
I
. Now
C
|
(`) C

(1) =
I
, so C
|
(`) =
I
and C

(`) C

(`) =
)
. But if r C

(`),
then `r = 1(`r) 1(`) = `, so r C

(`) and C

(`) = C

(`) =
)
and
` = `
I)
.
30
Since
)
is not maximal, let

)

)
be maximal and consider the product
`

)
`
1
. It is a full 1-lattice containing the unity since `

)
`
1
``
1
=
I
,
and
I
is a unital full 1-lattice. It is also a subring of since `

)
`
1
`

)
`
1

)
`
1
`

)
`
1
and hence `

)
`
1
is actually an order. In fact since
`

)
`
1

I
it follows that `

)
`
1
=
I
is a maximal order.
Now from Lemma 3.2.1 we get the following chain of inclusions, ` `

)

`

)
`
1
=
I
and since ` is a maximal left
I
-ideal it follows that either `

)
=
` or `

)
=
I
. If the rst was true, then

)
C

(`) =
)
and so
)
is
maximal, which is a contradiction and so `

)
=
I
must hold. But in this case,

I
= `

)
`
1
=
I
`
1
and thus `
1
C

(
I
) =
I
, contradicting Lemma
3.2.1. Therefore the maximal normal ideal `
i)
with
)
not maximal is ctitious.
By doing the converse similarly to above, the theorem then follows for all integral
ideals.
Finally if `
i)
is not integral, then `
i
and by Proposition 2.3.4 ` = J
1
for some integral
i
-ideal J with
)
= C

(J
1
) = C
|
(J) and
i
= C
|
(J
1
) = C

(J).
So it suces to show that J (which has maximal right order) has a maximal left
order, which is covered by above. Again this similarly holds for a non-integral `
i)
with
)
maximal and proves the theorem.
From this we see that a normal ideal `
i)
has both
i
and
)
maximal, so the
ideals that have their left order maximal actually coincide with the ideals that have
their right order maximal. At this point we mention that these normal ideals now
replace the fractional ideals in a maximal order in our train of thought and see
that, as mentioned earlier, that all fractional -ideals are actually normal ideals. So
we are actually looking to see if the set of normal ideals is a group, and the previous
theorem simplies remarkably our considerations of the set of normal ideals of
since the right orders of each element of the set are also maximal, allowing us to
have a maximal order as both a left and right identity element. However, there
seems to be the problem that there may be several identities, since left and right
multiplication by a maximal order may not acquire the same answer. Thus we might
need to consider a larger set than the previous sections, and the set may actually be
a groupoid rather than a group, where we dene a groupoid as following, referencing
Reiner [1, p.201]. First, a partial function GG G, is an operation that is not
dened on the whole of GG. Then dene the following.
Denition 3.2.3 A set G with partial function GG G : (`
i)
. `
I|
) `
i)
`
I|
is
called a groupoid if it satises the following axioms. For elements `
i)
. `
I|
. `
na
G,
There are unique c
i
. c
)
G, called the left and right units of `
i)
respectively,
such that the following are dened and hold true: c
i
`
i)
= `
i)
= `
i)
c
)
. c
i
c
i
=
c
i
and c
)
c
)
= c
)
.
`
i)
`
I|
is dened if and only if , = /, that is, when c
)
= c
I
.
For any two units c
i
. c
)
G, there is an element `
i)
G with left unit c
i
and
right unit c
)
.
If `
i)
`
I|
and `
I|
`
na
are both dened, then so are both (`
i)
`
I|
)`
na
and
`
i)
(`
I|
`
na
) and they are equal.
There is another element `
1
i)
G with left unit c
)
, right unit c
i
and such
that `
i)
`
1
i)
= c
i
and `
1
i)
`
i)
= c
)
both hold.
31
So a groupoid is just a generalisation of a group, with possibly multiple identity
elements and a partial function replacing the binary operation. Thus a group is
a specic case of a groupoid and we see that the set of fractional ideals over a
Dedekind domain 1,
1
, and the set of two-sided ideals of a maximal order ,

,
are actually groupoids themselves. Now leaving the group (or groupoid) structure
on the backburner, for the remainder of the section we look at some properties of
normal ideals, referencing again Reiner [1, p.197-198,202,209]. We rstly prove an
uniqueness property for normal ideals and then show a generalisation for divisors
of an ideal given in Proposition 1.3.5.
Proposition 3.2.4 For normal ideals ` and `, the product `` is proper if and
only if `` ``

for every normal ideal `

strictly containing `, or equivalently

`` `

` for every normal ideal `

strictly containing `.
Proof. Let ` = `
12
. ` = `
34
and `

= `

56
. Suppose `` ``

for all normal

ideals ` `

. Then `` = (`
2
)` = `(
2
`) implies that ` =
2
` so
3
=

2
and `` is proper. Now let `` be proper and suppose that ` `

. Clearly
`` ``

, but if `` = ``

then `

2
`

= `
1
``

= `
1
`` =

2
` = `. This is a contradiction, so `` ``

. The proof for `` `

`
for every normal ideal `

containing ` follows similarly.

Proposition 3.2.5 Let ` and ` be normal ideals.
a) If ` = `
12
and ` = `
34
, then ` ` if and only if there are integral ideals
A
13
and 1
42
such that `
12
= A
13
`
34
1
42
.
b) If ` = `
12
and ` = `
14
, then ` ` if and only if there is an integral
ideal 1
42
such that `
12
= `
14
1
42
.
c) If ` = `
12
and ` = `
32
, then ` ` if and only if there is an integral
ideal A
13
such that `
12
= A
13
`
32
.
Proof. Let `
12
and `
34
be normal ideals. If `
12
= A
13
`
34
1
42
for some integral
ideals A
13
and 1
42
, then `
12

3
`
34

4
= `
34
as required. Now suppose that
` = `
12
`
34
= `. We of course want `
12
= A
13
`
34
1
42
for some integral ideals
A = A
13
and 1 = 1
42
, so we let A = `
12
(
3
`
12
)
1
and 1 = `
1
34
`
12
. Then
it is seen that A and 1 are both normal ideals and have left and right orders as
indicated by their subscripts. Since `
3
`, we have (
3
`)
1
`
1
and so
A ``
1
=
1
, so A is an integral ideal. Also, 1 `
1
` =
4
so 1 is an
integral ideal. Now we get
A`1 = `(
3
`)
1
``
1
` = `(
3
`)
1

3
` = `
2
= `
and part a) of the the proposition is proven.
To prove b), see that when
3
=
1
in a), then A = `
12
(
1
`
12
)
1
=
1
getting `
12
= `
14
1
42
. Similarly, if we prove a) instead by letting A = `
12
`
1
34
and 1 = (`
12

4
)
1
`
12
, then when
4
=
2
we get 1 = (`
12

2
)
1
`
12
=
2
and so
`
12
= A
13
`
32
, proving c).
32
These results then get us the following theorem, which is essential if we are to
look at factorising ideals of a maximal order into irreducible ideals of . By
irreducible we of course mean maximal integral ideals in , so the next theorem
shows that if an ideal is maximal as a left ideal, then it must be maximal as a right
ideal also and vice versa.
Theorem 3.2.6 The normal ideal `
12
is a maximal left integral ideal in
1
if and
only if it is a maximal right integral ideal in
2
.
Proof. Let the normal ideal ` = `
12
be a maximal left integral ideal in
1
. By
Proposition 3.1.1, `
2
. If ` =
2
, then
1
= C
|
(`) =
2
= `, contradicting
` being a maximal ideal, so `
2
. Suppose there is a normal ideal ` = `
32
such that ` `
2
. If ` =
3
, then again
2
= C

(`) =
3
= `, so
`
3
. Now by Proposition 3.2.5 we can write ` as `
12
= A
13
`
32
for some
integral ideal A = A
13
. But then
` = A
13
`
32
A
13

3
= A
13

1
and so A = ` or A =
1
. If A = `, then
3
=
2
and ` = `
1
`` = `
1
` =

2
, which is a contradiction. So A =
1
must hold. But then
3
= C

(A
13
) =
1
and thus ` = `. This is a contradiction and proves that `
12
is also a maximal
right integral ideal in
2
. Similarly, the reverse statement holds and the theorem is
proven.
So a normal ideal `
i)
that is either a maximal left or right integral ideal can just
be called a maximal integral ideal, indicating by the theorem above that it is indeed
both a left and right maximal integral in its left and right orders respectively. With
this theorem we now have the required tools to look at our aims of factorising ideals
and seeing if the set of these ideals is a group or groupoid. So in the next section
this is what we do.
3.3 The Factorisation of One-Sided Ideals
Now we use the technical results of the previous section to see if a the factorisation
of ideals in Dedekind domains and for two-sided ideals in maximal orders holds
similarly for one-sided ideals in maximal orders, but without the use of prime ideals,
replacing them with maximal integral ideals. The main reference for this section is
again Reiner [1, p.196-201]. As mentioned in the rst section of this chapter, there
are two important corollaries to Theorem 3.1.3. These are given here.
Corollary 3.3.1 Let the normal ideal `
12
be a maximal integral ideal which belongs
to the prime ideals 1
1
of
1
and 1
2
of
2
. Then 1
2
= o
12
(1
1
).
Proof. First let ` = `
12
and see that o
12
(1
1
) = `
1
1
1
`. Let J. 1

2
such
that J1 o
12
(1
1
) = `
1
1
1
`. Then `J
2
1`
1
= `J`
1
`1`
1
1
1
.
Since `J`
1
. `1`
1

1
and 1
1
is prime, without loss of generality we have
`J`
1
1
1
. But then J `
1
1
1
` and thus o
12
(1
1
) is prime in
2
. Now
since 1
1
`, we have `
1
1
1
1
. Thus `
1
1
1
` 1
1
1
1
1
` = ` and so
`
1
1
1
1
` is a prime ideal in
2
containing `. By the uniqueness of 1
2
it follows
that o
12
(1
1
) = `
1
1
1
1
` = 1
2
and the corollary is proven.
33
This corollary gives us a notion of similarity between ideals of diering orders.
So with o
12
as in Corollary 3.3.1 and for an ideal 1 in
1
we say that 1 and o
12
(1)
are similar. So this corollary shows that each maximal integral ideal identies a
pair of similar prime ideals. However, it is not always the case that the converse
holds. That is, if there are two prime ideals that are similar, there is not always
a maximal integral ideal belonging to both of them. The second corollary is given
here, and will be used later.
Corollary 3.3.2 Let ` = `
23
be a normal maximal integral ideal belonging to the
prime ideal 1
2
of
2
. Then for each normal ideal ` = `
12
, `,`` is a simple
left
1
-module and c::

1
`,`` is similar to 1
2
.
Proof. First we see that ann

1
`,`` = {r
1
r` ``}. Now if there is
an
1
-module 1 such that `` 1 `, then ` `
1
1
2
. Thus `
1
1
is either ` or
2
, implying 1 = `` or 1 = ` and `,`` is a simple left

1
-module.
Now similarly to the proof of Theorem 3.1.7 we have
ann

1
A.
(`,``) = {r
1
,`` r(`,``) = (`,``)r = 0}
= {r
1
,`` r`. `r ``} = 0
(note this is 0 in
1
,``) so ann

1
`,`` is a prime ideal of
1
, say 1
1
. But
considering o
21
(1
2
) = `1
2
`
1
, we see that o
21
(1
2
)` = `1
2
``, so
o
21
(1
2
) 1
1
and thus o
21
(1
2
) = 1
1
= ann

1
`,``, proving the corollary.
Finally the Irishman is in sight and we can look at trying to factorise one-sided
ideals. Unfortunately due to the non-commutativity of and order , if there is a
factorisation then it will probably not be unique. This will be remedied somewhat
in the next few results, but rst we show that there is a factorisation of integral
one-sided ideals into maximal integral ideals via the proper product and that the
number of factors stays the same.
Lemma 3.3.3 Let `
i)
be a normal integral ideal, and be either
i
or
)
. If the
composition length of ,`
i)
is :, then `
i)
can be written as a proper product of :
normal maximal integral ideals `
1
`
2
`
-
where C
|
(`) = C
|
(`
1
) and C

(`) =
C

(`
-
). Also, : is uniquely determined by `
i)
and is independent of the choice of

i
or
)
.
Proof. For the rst statement, let =
i
and see that if ` is itself maximal then
the result trivially holds true, so taking this as the : = 1 case we prove the theorem
by induction on :. Suppose the result is true for : 1 and let ` be such that

i
,`
i)
has composition length : +1. Then there is a normal maximal integral ideal
` = `
iI
such that ` `
i
and `,` has composition length :. Consider
34
`
1
`, a left ideal of
I
, and suppose
I
,`
1
` has composition length t. Then
there is an unrenable chain of t left
I
-modules
`
1
` = 1
1
1
2
1
|

I
.
But then by Proposition 3.2.4
` = ``
1
` = `1
1
`1
2
`1
|
`
I
= `.
and t : since `,` has composition length :. Similarly there is an unrenable
chain of :
i
-modules from ` to
I
, so multiplying by `
1
on the left gets : t,
showing : = t and so
I
,`
1
` has composition length : also. By induction
there are : normal maximal integral ideals `
|
such that `
1
` = `
2
`
|+1
and
C
|
(`
1
`) = C
|
(`
2
) and C

(`
1
`) = C

(`
-+1
). But then ` = ``
2
`
-+1
is a product of : +1 normal maximal integral ideals. Finally, C
|
(`) =
i
= C
|
(`)
and C

(`
-+1
) =
)
= C

(`). The case where =

)
holds similarly to above.
For the second statement, suppose that ` can be written as a proper product
of normal maximal integral ideals ` = `
1
`
2
`
-
with C
|
(`) = C
|
(`
1
) and
C

(`) = C

(`
-
). Then by Proposition 3.2.4
` = `
1
`
-
`
1
`
-1
`
1
`
2
`
1

i
(3.2)
and
` = `
1
`
-
`
2
`
-
`
-1
`
-
`
-

)
(3.3)
are unrenable chains of : left
i
and
)
-modules. So the composition length of
,`
i)
for either =
i
or
)
is : and by the Jordan-Holder Theorem all decom-
positions are isomorphic to either (3.2) or (3.3), thus : is uniquely determined by
`
i)
, proving the lemma.
So this shows that every normal integral ideal can be expressed as a product of
maximal normal integral ideals. But what about ideals that are not integral? First
consider what the inverse of an integral ideal would look like. If we write an normal
integral ideal 1
i)
out as a proper product as in Lemma 3.3.3 1 = 1
i)
= `
1
`
-
.
Then if we let 1
1
= `
1
-
`
1
1
, which is still a proper product, then everything
works out as we want it to. For instance,
11
1
= `
1
`
-
`
1
-
`
1
1
= `
1
`
-1
C
|
(`
-
)`
1
-1
`
1
1
.
.
.
= `
1
C
|
(`
2
)`
1
1
= C
|
(`
1
) =
i
and similarly, 1
1
1 =
)
. We can now use this new found property to achieve our
aim, catch the Irishman and get a factorisation of every normal ideal in .
35
Theorem 3.3.4 Every normal ideal in can be written as a proper product of
maximal normal integral ideals and inverses of maximal normal integral ideals in
.
Proof. Let ` = `
i)
be a normal ideal in . If `
i
, then the result follows
immediately from Lemma 3.3.3, so let `
i
. Since there is an : 1`
1
, then
:`
i
and by Lemma 3.3.3 write :` = `
1
`
-
for maximal normal integral
ideals `
|
. But since :` = (:
i
)`, then we can write ` = (:
i
)
1
:`. Since
:
i

i
and it is two-sided, we can write :
i
= 1
1
1
I
for maximal normal
integral ideals 1
|
with C
|
(1
1
) = C

(1
I
) =
i
. But then (:
i
)
1
= 1
1
I
1
1
1
, so
we can write ` = 1
1
I
1
1
1
`
1
`
-
. That is, ` is a proper product of maximal
normal integral ideals and inverses of maximal normal integral ideals.
So we have caught the Irishman, taken his gold and found that every normal
ideal in can be factorised, even though it is not unique. We have thus achieved
one of our aims, but what of the second? What can we say about the set of all these
normal ideals in ? The answer is given in the following theorem, which follows
from the preceding theorem.
Theorem 3.3.5 The set of all normal ideals in is a groupoid when endowed
with the proper product. It is called the Brandt Groupoid assosciated with and is
generated by the maximal normal integral ideals of and the inverses of maximal
normal integral ideals of , where the proper product is dened.
As expected, the units of the Brandt Groupoid associated with are the max-
imal orders in and the inverse of the normal ideal 1
i)
in with left unit
i
and
right unit
)
is 1
1
i)
. But returning to the corollaries stated earleier, how do they
help us? Though the factorisation of a normal ideal of is not unique, the factors
do however belong to the same prime ideals, up to similarity.
Corollary 3.3.6 Let `
i)
be a normal integral ideal written as a proper product
` = `
1
`
2
`
-
of normal maximal integral ideals `
|
, where : is the composi-
tion length of
i
,`
i)
. Then the prime ideals that each `
|
belongs to is uniquely
determined by `
i)
up to similarity and order of occurrence.
Proof. Relabeling the product as ` = `
1 -+1
= `
12
`
23
`
- -+1
, consider the
series of composition factors of
1
,`

1
,`
12
. `
12
,`
12
`
23
. . `
12
`
-1 -
,`
12
`
- -+1
.
If we let 1
i
be a prime ideal of
i
such that `
i i+1
belongs to it, then by Corollary
3.3.2 we get that
o
11
(1
1
) = 1
1
= ann

1
(
1
,`
12
).
o
21
(1
2
) = ann

1
(`
12
,`
12
`
23
).
.
.
.
o
-1
(1
-
) = ann

1
(`
12
`
-1 -
,`
12
`
- -+1
).
36
So the prime ideals 1
i
are determined up to similarity and order of occurence by
1, proving the corollary.
But what else can we say about the factorisation, or at least a factorisation? It
turns out that we can specify the composition factors of
1
,1
1-
in advance, as the
following theorem states. For the proof of this theorem, see Reiner [1, p.200-201].
Theorem 3.3.7 Let 1 = 1
i)
be a normal ideal with the composition lenngth
of
i
,1 equal to : and let {o
1
. o
2
. . o
-
} be the composition factors of the
i
-
module
i
,1, arranged in any preassigned order. Then there is a factorisation
1 = `
1
`
-
of 1 into a product of maximal normal integral ideals `
I
such that
the factor modules

i
,`
1
. `
1
,`
1
`
2
. . `
1
`
-1
,1
are precisely o
1
. o
2
. . o
I
, in that order.
To conclude our look on the factorisation of ideals we will look at, yet again,
the rational quaternions and state a theorem, which is included more for aesthetic
reasons rather than for usefulness.
Theorem 3.3.8 Let be a maximal order in . There are surjective relations
from the maximal integral ideals of to the prime ideals of and from the prime
ideals of to the prime ideals of 1,
{Maximal Integral Ideals of } {Prime Ideals of } {Prime Ideals of 1}.
Proof. These are given by Theorems 2.1.12 and 3.1.7.
Example 3.3.9 We look at factorising some one-sided ideals in a maximal order in
the rational quaternions. So let = i , / and = i , c
where c = (1+i+,+/),2. Then we saw that was a maximal order that could also
be written as = {(c +/i +c, +d/),2 c / c d mod 2}. By Example 2.3.8,
we also saw that all the maximal left ideals of are of the form r for irreducible
r . This was because was a non-commutative version of a PID.
Since is a PID, it follows that every element can be expressed as a product of
irreducible elements (for more on this see Chapter 4). So let . be a non-zero
non-unit element. Then we can write . = r
1
r
2
r
a
for some : irreducible elements
r
i
. Thus we can write . = r
1
r
a
. But r is a maximal left ideal, so
relabeling =
1
, suppose
1
r
1
has right order
2
. Then
2
must also be a
37
maximal order and r
1

2
must be a maximal right ideal since
1
r
1
is a maximal
left ideal. So r
1
is irreducible in
2
and hence we can write
. = (
1
r
1

2
)r
2
r
a
= (
1
r
1
)(
2
r
2
)r
3
r
a
.
Similarly we can say
2
r
2
has right order
3
and r
2
is irreducible in
3
so
. = (
1
r
1
)(
2
r
2
)(
3
r
3
)r
4
r
a
.
Continuing this we then express the left ideal . as a product of normal maximal
integral ideals
. = (
1
r
1
)(
2
r
2
) (
a
r
a
).
The Brandt Groupoid associated with is however hard to nd, as all maximal
orders in must be considered. So it will just be left as a thought.
38
Chapter 4
Factorisation...of Elements
So far we have just considered factorising ideals of specic rings into a product of
irreducibles, or maximal ideals, in that ring. But what about factorising elements
in the ring? More specically, we pose the question: Do the results we have found
about Dedekind domains and orders allow us to uniquely factorise elements? Before
we answer this question, we recall some denitions at the beginning of Chapter 1
as they will be used here.
4.1 Background
Let 1 be an integral domain and n(1) be the multiplicative subgroup of units of
1. An element j 1 is called prime if for two elements c. / 1, if j c/ then
either j c or j /. A non-zero non-unit element : 1 is irreducible if for every
factorisation : = c/ for c. / 1, either c n(1) or / n(1) (see that both c and
/ can not be in n(1) since this forces : n(1)). Call 1 an Unique Factorisation
Domain (or UFD) when for every element : 1, if there are two factorisations
: = c
1
c
a
= /
1
/
n
with c
i
. /
)
1 for all i and ,, then : = : and c
i
= n
i
/
(i)
for some permutation {1. . :} and unit n
i
n(1) for every i. That is, we
can swap around the /
i
s and multiply them by units to get c
1
c
a
. Call 1 a
Principal Ideal Domain (or PID) if every ideal of 1 is a principal ideal.
If for the integral domain 1 there is a function i : 1 0 such that the
following are satised
For c 1 and / 1 0 there are . : 1 such that c = / + : and
i(:) < i(/),
For c. / 1 0, i(c) i(c/),
then 1 is called an Euclidean domain. The denition shows us that every Euclidean
domain has an Euclidean algorithm and this can be used to nd the greatest
common divisor of two elements of 1. From this it we can establish the following
result.
Theorem 4.1.1 Every Euclidan domain 1 with function i is a PID and all proper
ideals 1 in 1 are generated by an element r 1 with i(r) = min
i1
{i(i)}. That is,
i(r) is minimal.
Proof. Let 1 be a proper ideal of 1. Since i : 1 , then there is a minimal value
of {i(r) r 1}, and let . 1 be one such that i(.) is minimal. Then 1 ..
But for any r 1 there are . : 1 such that r = . + : and i(:) < i(.). The
minimality of i(.) implies i(:) = 0 and so : = 0 since 1 is an integral domain. So
r . and thus 1 = .. This also proves that 1 is a PID.
39
There are the following examples of Euclidean domains.
Example 4.1.2 All elds 1 are Euclidean domains via the function i(r) = 1 for
all r 1 0. But not all Euclidean domains are elds, as the Euclidean domain
with function i(r) = r indicates.
Now returning to the question: Do the results we have found about Dedekind
domains and orders allow us to uniquely factorise elements in that Dedekind do-
main? The answer is actually very easy to get, as we have already seen in Example
1.2.4 and as repeated below.
Example 4.1.3 The ring 1 = [

5] is a Dedekind domain, but 6 = (1+

5)(1

5) = 2.3, so 1 is not a UFD.

So the answer is no. But it turns out that it is pretty close. The number 6
above was expressed twice above and it turns out that
6 = (1 +

5)(1

5) = 2.3
are the only expressions that equal it except for rearrangement or multiplication by
units in [

5]. So even though [

5] is not a UFD, it seems to be within arms

reach of being one.
Of course when we say factorise elements, we are being very informal when
this is in regard to orders since they are not necessarily commutative. So First we
consider if it is the case in Dedekind domains and then move onto orders later,
redening some factorisation theory notions with non-commutativity allowed. Note
though that this chapter is more a summary of some nice results rather than an in
depth account, so many theorems will not have proofs provided, but references to
proofs will be given.
4.2 Factorisation, the Class Group and The Class Number
Back in Chapter 1 we showed every PID was a Dedekind domain via Proposition
1.2.3. So if we look at the following chain of inclusions,
{Fields} {Integral Domains}

{Euclidean Domains} {Unique Factorisation Domains}

{Principal Ideal Domains}.
do {Dedekind domains} t in somewhere? As we see by Example 4.1.3, if it does
then it is in between UFDs and integral domains. However, the next proposition
shows us that it does not t into the chain since not all UFDs are Dedekind
domains, but if we replace UFDs with Dedekind domaains in the chain above then
40
the resulting chain of inclusions below still holds.
{Fields} {Integral Domains}

{Euclidean Domains} {Dedekind Domains}

{Principal Ideal Domains}.
Theorem 4.2.1 A Dedekind domain is a UFD if and only if it is a PID.
Proof. See Sivaramakrishnan [10, p.406].
But how do we actually know when a Dedekind domain 1 is a PID? To do this
we must consider classes of set of Fractional ideals
1
of 1, a Dedekind domain
from hereon. These classes are dened as following. Two fractional ideals 1. J
1
are elements of the same ideal class if 1 = rJ for some r 1 and this ideal class
can then be denoted [1] (or equivalently [J]). See here that 1

= J as modules via
1 J : c rc with inverse rc r
1
rc = c, so these ideal classes can be thought
of as isomorphism classes. Ideal classes can be multiplied together to get another
ideal class by [1][J] = [1J]. This works since if r1 [1] and J [J] for some
r. 1, then r1J = (r)1J [1J]. Thus we have proven the following.
Denition 4.2.2 The ideal classes of
1
form an abelian multiplicative group
with identity element [1] called the ideal class group of 1. This is denoted by
1
.
Now if a fractional ideal of 1 is generated by one element it is, as expected,
called a principal fractional ideal. So by Theorem 1.3.8 if 1 is a PID, then the
ideal class group of 1 is actually just {[1]}, an one element set. Conversely, if the
ideal class group of 1 is just {[1]} then 1 is obviously a PID, so 1 is a PID if and
only if its ideal class group is just the class [1]. But what if it is not just [1], that
is, if 1 is not a PID? If this is the case, we can use
1
as a measure of how far
1 is from being one, if it happens to be nite. So when
1
is indeed nite, we
let (1) =
1
(and say (1) is innite otherwise) and call this number the class
number of 1. From the preceding discussion we see that 1 is a PID if and only if
(1) = 1 and thus we get the following corollary.
Corollary 4.2.3 A Dedekind domain is a UFD if and only if the class number of
1 is 1.
Proof. By above a Dedekind domain is a PID if and only if (1) = 1. Theorem
4.2.1 then proves the corollary.
This corollary says that in the case of a Dedekind domain 1 with class number
1, there is actually more than just a unique factorisation of ideals, there is a unique
factorisation of elements. So when looking at 1, one can use the ideals of 1 instead
of the elements of 1, and vice versa. Class group are discussed more in delpth see
Reiner [1, p.48], Jacobson [6, 643-649], Janusz [7, p.16-18], Stewart & Tall [8, 9]
and Sivaramakrishnan [10, p.436-441].
41
4.3 Finiteness of The Class Number of the Ring of Integers
Algebraic number theory is, obviously, a large part of number theory and uses vast
amounts of algebraic techniques. Here we take a certain type of Dedekind domain
that is important to algebraic number theory and see that its class number is nite.
More detailed accounts of what will be discussed here can be found in Janusz [7,
p.52-65], Stewart & Tall [8, 9], Samuel [9, p.57-59] and Sivaramakrishnan [10,
p.436-441]. This certain type of ring is the ring of integers of a number eld and
was a motivating example for the denition of a Dedekind domain. For this section
we ignore the use of 1 and in the rest of the document as the quotient eld of
1 and 1-algebra, since they will soon be redened.
To start with, we must take a specic case of integrality over a ring 1. A
complex number c is an algebraic number if there is a polynomial j(A) [A]
such that j(c) = 0. The ring could actually be replaced by since multiplying
the polynomial by all the denominators of the coecients gets a polynomial with
integer coecients. The set of all algebraic numbers is denoted by and is actually
a subeld of .
Corollary 4.3.1 A complex number c is an algebraic number if and only if [(c) :
] is nite.
Proof. If c , then c is a root of a polynomial j(A) [A] of say degree :.
Making this polynomial monic in [A] and multiplying by c we get cj(c) = 0,
so c
a+1
is expressible as a polynomial of c of degree :, thus [(c) : ] is nite.
Conversely if [(c) : ] is nite of say degree n, then 1. c. . c
a
are linearly
dependent over , so c .
Theorem 4.3.2 The set of all algebraic numbers is a subeld of .
Proof. For c.
[(c. ) : ] = [(c. ) : (c)][(c) : ].
which is nite. Since c . c. c, (c. ), then each is an algebraic number
by Corollary 4.3.1.
It turns out that [ : ] is not nite, so to make life easier we dene the
following. A subeld 1 is an algebraic number eld if [1 : ] is nite. Now
obviously 1 and since [1 : ] is nite we may write 1 = (c
1
. . c
a
) for
some algebraic numbers c
i
, but can we say anything else? Since 1 is a eld it
is an Euclidean domain and there is a gcd algorithm applicable in 1. So letting
c = gcd(c
1
. . c
a
) we see that 1 = (c), showing us every algebraic number
eld is just a nite extension of generated by one algebraic number.
Given the algebraic numbers , lets look at a subset of them as follows. For
an c , it was mentioned that there is a polynomial in [A] with c as a root.
But this polynomial may not be monic. We then call an algebraic integer if
there is a monic polynomial j(A) [A] such that j() = 0 and let the set of all
algebraic integers be denoted by 1. This subset actually turns out to be a subring,
as the following shows.
42
Theorem 4.3.3 The set of algebraic integers, 1, is a subring of the algebraic
numbers .
Proof. An algebraic integer is just an element r that is integral over . So by
Proposition 1.1.1 r is an algebraic integer if and only if [r] is a nitely generated
-module. Now for two algebraic integers r. , [r] and [] are nitely
generated -modules. Thus [r. ] must be a nitely generated -module also.
But since r + . r [r. ], then they are also algebraic integers and thus 1 is a
subring of .
This set is useful to us in the following way. Let 1 be an algebraic number eld
and let
1
= 11. Then
1
is called the ring of integers of 1. As the denition
suggests,
1
is a subring of 1 since both 1 and 1 are rings. But
1
turns out
to be more than just a ring, it is a Dedekind domain. If this sounds familiar, it
may be because it was mentioned back in Example 1.2.5 where it was dened as the
integral closure of in a nite extension 1,,
1
=
1
. Both of these denitions
create the same ring, which the following theorem shows.
Theorem 4.3.4 For an algebraic number eld 1, the ring of integers
1
of 1 is
a Dedekind domain.
Proof. See Sivaramakrishnan [10, p.409-410].
Now we look at the ideas discussed in the previous section and consider the
ideal class group of
1
. Is the ideal class group of
1
nite? The answer to this is
given in the following theorem, which will not be proved here. One proof, though
not straightforward, uses fairly straightforward ideas from geometry, lattices and
Minkowskis theorem to get bounds on the norm of an ideal using the discriminant
of a matrix formed by a basis of 1 over and monomorphisms 1 (this proof
can be found in the references listed at the beginning of this section). But the
previous seemingly complicated sentence is remedied by the next seemingly simple
one.
Theorem 4.3.5 The ideal class group of
1
is nite.
Theorem 4.3.5 shows that the class group of 1 is nite, but how small is it?
There is one nice property associated with
1
that can be found without too much
work. It is not always the case that
1
is a PID, so an ideal is not always generated
by one element. However, if it is not generated by one element then it is generated
by two.
Theorem 4.3.6 Every ideal in
1
is generated by at most two elements.
Proof. See Stewart & Tall [8, p.131-132] and Sivaramakrishnan [10, p.426-427].
As we mentioned before, one proof of Theorem 4.3.5 uses geometric ideas. But
there is another way using the denitions in Chapters 2 and 3 and algebraic tech-
niques. A benet of this more complicated theorem is that it is a generalised version
of the niteness theorem above.
43
4.4 Factorisation in an Order and the Jordan-Zassenhaus Theorem
Theorem 4.3.5 shows us that the ideal class number is nite for algebraic number
elds. But the proof that was mentioned uses ideas dierent from Chapters 2 and
3. If similar ideas to Chapters 2 and 3 are to be used, it turns out that a more
general theorem can be established, named the Jordan-Zassenhaus Theorem. So to
start, as usual we let be a separable semisimple 1-algebra and an 1-order in
.
As shown earlier, the set of fractional ideals of a Dedekind domain 1 can be
partitioned into ideal classes, with the set of all these classes called the ideal class
group. This was done by saying two 1-ideals 1 and J are in the same class when
1 = rJ for some r 1. But this says that 1

= J as 1-modules. So we use
the latter to classify ideals of for 1-orders in . Two fractional left -ideals
1 and J are elements of the same ideal class if 1

= J as -modules. But each
isomorphism between 1 and J can be extended to an automorphism 11

= 1J
from to . Hence the ideal class containing 1, denoted [1] as previously, is
actually [1] = {1r r n()}. When = 1 this coincides with the ideal classes
looked at previously, but in that case the multiplication by a unit can be on either
side of the ideal. In an arbitrary , however, the multiplication must be on the
right to keep 1r a left -ideal. The set of these ideal classes, denoted

, is not
always a group due to this right multiplcation by an element in n(), but as we did
with Dedekind domains, we only want to see if it is nite. So if it is nite we let
() =

(and say () innite otherwise) and call this the ideal class number
of . See Reiner [1, p.224] for more details on this.
In a Dedekind domain 1, when we had (1) = 1 then we knew 1 was a PID. So
can we get something similar for an order in an algebra? First we need to clear up
some concepts, for instance, a PID is by denition commutative so it will need to
be generalised (see Sivaramakrishnan [10, p.244-246]). The concept of an integral
domain being a PID, however, can be easily dened for a non-commutative ring,
so until we revert back to orders, let be any arbitrary non-commutative ring and
not necessarily an order (or a domain). The ring is a left principal ideal ring (or
left PIR) if every left integral ideal in is generated by one element. That is, every
left integral ideal 1 of can be written as 1 = r for some r . Similarly the
denition holds for a ring being a right PIR and if the ring is both a left and right
PIR, then it is just called a PIR.
A unit in the ring is an element n such that there exists a where
n = 1. Note this also implies that n = 1 since n = , so (n 1) = 0 and
multiplying by n on the left gets n = 1. The set of units of . n(), is then a
subgroup of under multiplication. The element r is then called irreducible if
every factorisation r = c/ for c. / , has either c n() or / n(). This gets
the following result on irreducibility.
Proposition 4.4.1 For the ring , r is irreducible if and only if r is a
maximal left integral ideal.
Proof. Suppose that r is not irreducible, then r = c/ for some non-zero non-
units c. / . Then r = c/ /, showing one implication. If r is not
44
maximal, then r / for some non-zero non-unit / . So r = c/ for some
non-zero non-unit c and r is not irreducible, proving the proposition.
Note the result could equally have been proven for r a maximal right ideal, but
we continue on our path. Two elements r. are called left associates if r = n
for some n n() and right associates if r = n. This is equivalent to saying that
r and are left associates when ,r

= , and similarly right associates when

,r

= ,. It turns out that when dening a UFD, or in this case a UFR, this
is the more useful way. When looking at a factorisation of an element r into
irreducibles of , say r = j
1
j
a
with j
i
irreducible for every i, then there is the
following chain of inclusions
r = j
1
. . . j
a
j
2
j
a
j
a
. (4.1)
Each subset, though, must be strict since j
i1
j
a
,j
i
j
a

= ,j
i
and j
i
is a maximal left ideal in . Thus we get the following series of non-zero quotient
rings
,j
1
. ,j
2
. . ,j
a
. (4.2)
This allows us to dene the following. A ring is an unique factorisation ring (or
UFR) if for every non-zero element r , there is a factorisation of r as a product of
irreducible elements of and if there are two factorisations r = j
1
j
a
=
1

n
for irreducibles j
i
.
)
, then : = : and their series of ideals as in (4.2) are
isomorphic in the sense that ,j
i

= ,
(i)
for some perm{1. . :} for
every i. That is, j
i
is a left associate of
(i)
.
Finally, call a set \ well-ordered if there is a partial order (a reexive, anti-
symmetric and transitive binary relation) on \ such that every subset of \ has a
minimal element according to this order. We will call the ring a left Euclidean
ring if there is a function i : \ where \ is a well-ordered set (usually )
such that for r and 0, there exist . : such that r = + : and
i(:) < i(/) (see Brungs [11]). Similarly is a right Euclidean ring if the same
conditions hold but the multiplication by is on the right, so r = + :. These
denitions do not always coincide, so if a ring is both a left and right Eudlcidean
ring, then it is just called an Euclidean ring. Similarly to the commutative case, a
division ring 1 is an Euclidean ring with i(r) = 1 for all r 10 and Theorems
1.1.5 and 4.1.1 hold, allowing a similar chain of inclusions
{Division Rings} {Unique Factorisation Rings}

{Euclidean Rings} {Principal Ideal Rings}
Returning to looking at 1-orders in the algebra , we let again be an 1-
order in . So what can we now say when () = 1? When (1) = 1 for the
Dedekind domain 1 then 1 was a PID, when replacing 1 with , we get the
following Theorem.
45
Theorem 4.4.2 The 1-order in is a left PIR if and only if () = 1 for left
-ideals.
Proof. If has () = 1 for left ideals, then since is a left -ideal itself, the only
class of left -ideals in

must be [] = {n n n()}, showing is a left

PIR. Secondly, if is a left PIR then take an integral ideal of the form r for some
r . By the Brandt Groupoid associated with , there is an inverse to r, say
1, which will be a right -ideal. But r1 = , so there are some and . 1
such that r. = 1, implying that r n(). Similary, there is a n n() such
that nr = 1, implying that r n() and thus r is in the class [], so () = 1
for left ideals.
That is, if the ideal class number for left ideals is 1, then there is a factorisation
of elements of and not just ideals, just like the Dedekind domain scenario, even
though it is non-commutative (see Reiner [1, p.230]). If () = 1, then for an
element r , nd a chain as in (4.1)
r = r
1
r
2
r
a
=
where r = r
1
and r
a
= 1. Then as in Proposition 4.4.1 we can write r
i
= .
i
r
i+1
for
some non-zero non-unit .
i
. Since r
i+1
,r
i

= ,.
i
then .
i
is a maximal
ideal and by Proposition 4.4.1 .
i
is irreducible. But then r = .
1
.
2
.
a1
r
a
=
.
1
.
2
.
a1
, a product of irreducible elements of . Furthermore, we get a series
of quotient rings as in (4.2)
,n
1
. ,n
2
. . ,n
a
.
So when looking at , one can use the ideals of instead of the elements of ,
and vice versa, as in the Dedekind domain case. Moreover, Corollary 3.3.6 gives us
the prime ideals that each .
i
belongs to up to similarity and occurrence. So each
.
i
is associated with some prime ideal of and these prime ideals can actually be
uniquely determined by the element r up to similarity and occurence.
More generally, however, seeing just when () is nite is not always easy. So
when can we say it is nite? We have seen previously that when is an algebraic
number eld 1 and the ring of integers
1
, then (
1
) is nite. This was in
fact Theorem 4.3.5 and it was stated that a standard proof of this uses ideas from
geometry and lattices and is relatively unrelated to the material here. However there
is another proof that uses Chapters 2 and 3 that actually establishes a more general
result. The proof (including Lemma 4.4.3), which is somewhat more complicated
than the standard proof of Theorem 4.3.5, will not be given, but it can be found in
Reiner [1, p.224-229], Swan & Evans [2, p.43-54], Curtis & Reiner [4, 79] and Bass
[3, p.539-548].
To state the theorem we rst need to give some denitions. Recall that a full
1-lattice ` in is an 1-lattice such that 1` = and fractional ideals in the 1-
order are full 1-lattices in that are nitely generated -modules. The 1-rank
of a 1-lattice ` in is then dened as rank
1
` = dim
1
(1`). For the 1-order
, the Jordan-Zassenhaus condition, denoted J2() holds true if for each t
+
,
the number of -isomorphism classes of left 1-lattices of 1-rank at most t is nite.
46
Note that if t = [ : 1] then clearly the -isomorphism classes of rank t are the
left ideal classes of and so if J2() holds then indeed () is nite. In fact, we
get the following lemma, which is proven when is a skeweld over 1, which is
dened as follows. The division ring 1 is called a skeweld over 1 if 1 2(1)
and [1 : 1] is nite, where 2(1) is the centre of 1.
Lemma 4.4.3 Let be an 1-order in a skeweld 1 over 1. Then J2() holds if
and only if () is nite.
This lemma then gives us the generalisation of Theorem 4.3.5 by its use in
the following theorem, the Jordan-Zassenhaus Theorem, which uses the following
denition, allowing it to expand its reach. A function eld is a nite extension
of the eld /(A) of rational functions in an indeterminate A over a nite eld /.
The theorem will just be stated without any discussion, but note that it can also
be proven for certain o-orders where o is a commutative ring, not necessarily a
Dedekind domain, see Reiner [1, p.228-229].
Theorem 4.4.4 Let 1 be a Dedekind domain such that 1 is an algebraic number
eld or function eld. Then for each 1-order in a semisimple 1-algebra , J2()
holds true.
4.5 Quaternions and a Sum of Four Squares
Now we look at a specic example. We have seen already in Examples 3.3.9, 2.3.8
and 2.2.6 that the rational quaternions give good examples of all the theoretical
aspects discussed up to Chapter 3. But certain facts about them have been used so
that they can be such a good example. So here we will show these facts, with the
rational quaternions yet again providing a good example of the theory looked at in
this chapter so far. For this section, refer to Reiner [1, p.229-231]. As a historical
note, Hamilton was the inventor of the quaternions in 1843. He intended to extend
the complex number to three dimensions over for applied purposes, but it took
him several years to nd out that this would not work. He realised that he needed
four dimensions and that commutativity does not hold, which prompted the start
of non-commutative ring theory, a pure topic.
So from hereon let = i, / be the separable semisimple -algebra
of rational quaternions, recalling that it is an extension of the rational complex
numbers i where elements , and / act as roots of 1 and multiplication
follows the rule i
2
= ,
2
= /
2
= i,/ = 1. Let = i , / and
= i , c where c = (1 + i + , + /),2. We have seen that is an
order and is a maximal order containing . Also, it is easy to show that we can
rewrite as
= {(c +/i +c, +d/),2 c. /. c. d such that c / c d mod 2}. (4.3)
Also, it is easy to see that for every r there is a n n() such that nr .
This will actually become a key point in our dicussions.
We will show that the class number of is nite. In fact, we will see it is 1 and
do this by proving is an Euclidean ring. To do this we rst dene the function
47
i :
+
{0} : r r, where acts as a norm like in the complex numbers.
That is, if r = c +/i +c, +d/ , then
i(r) = r = (c +/i +c, +d/)(c +/i +c, +d/)
= (c +/i +c, +d/)(c /i c, d/)
= c
2
+/
2
+c
2
+d
2
.
The function i is also multiplicative since for r.
i(r) = rr = rr = i(r) = i(r) = i(r)i().
But what if we restrict i to ? Let r = (c + /i + c, + d/),2 as in (4.3), then
(r) =
1
4
(c
2
+/
2
+c
2
+d
2
). If the integers c. /. c and d are all congruent to 0 mod 2,
then their squares are congruent to 0 mod 4 and i(r) . Otherwise if they are
congruent to 1 mod 2, then for some integers c

. /

. c

and d

we can write
1
4
(c
2
+/
2
+c
2
+d
2
) =
1
4
((2c

+ 1)
2
+ (2/

+ 1)
2
+ (2c

+ 1)
2
+ (2d

+ 1)
2
)
= c
2
+c

+/
2
+/

+c
2
+c

+d
2
+d

+ 1.
Getting (r) yet again. Thus i restricted to always achieves a non-negative
integer, so we can say i : instead.
This allows us to look at the units of , since if n n(), then there is a
n() such that n = 1. So 1 = i(n) = i(n)i() and since i(n) and i() must
be non-negative integers, they are both 1. Thus the group of units of is exactly
the following 24 elements
n() = {n i(n) = 1} = {1. i. ,. /. (1 i , /),2}.
Now we have all we need to know about i to show that is an Euclidean ring, and
thus () = 1. So we do it here.
Theorem 4.5.1 The maximal order i , c in i , / where
c = (1 +i +, +/),2 is an Euclidean ring
Proof. Let = i , c and = i , / as usual an let r
and 0. Since is a division ring, we look at the quotient r
1
in . Since
is a full -lattice in there is some and :

such that r
1
= + :

where :

= c
1
+c
2
i +c
3
, +c
4
/ with c
i

1
2
for every i. In fact, if c
i
=
1
2
for every
i then :

and we can write r

1
=

+ :

where

= + :

and :

= 0.
Thus at least one of the c
i
has c
i
<
1
2
and (:) <
1
4
+
1
4
+
1
4
+
1
4
= 1. So multiplying
by we get r = + : where : = :

. But both r and are in , so : also.

Finally, since i(:) = i(:

) = i(:

)i() < i() it follows that is a left Euclidean

ring and by considering
1
r it follows that is actually an Euclidean ring.
Furthermore, since is a domain, is actually a non-commutative version of a
Euclidean domain and thus a principal ideal domain. So the following corollary is
now trivial.
48
Corollary 4.5.2 The maximal order i , c in i , / where
c = (1 +i +, +/),2 has ( i , c) = 1.
Proof. Theorem 4.5.1 says that i , c is an Euclidean ring, so it is a
PIR and thus ( i , c) = 1.
But now that we know is a PID, every integral ideal of can be written as
r for some r . This unexpectedly allows us to prove a nice number theoretic
result (with inspiration from Reiner [1, p.215]) that was rst proved in 1770 by
Lagrange (obviously via a dierent proof), the Four Squares Theorem. The theorem
also allows us to relate the theory of orders back to arithmetic, and even non-
commutative arithmetic, even though it seems to be an abstract algebraic being.
This theorem is given here.
Theorem 4.5.3 Every natural number can be expressed as a sum of four squares.
Proof. Let j be prime. Then there exists an irreducible r so that r is a
maximal integral ideal such that j r. Now intersecting these ideal with we
get j = j r. Now if . r, then . and . r, so . = r
for some . But then r = rr = .r and thus =
:
a
r. Thus r r,
so j r. But j is prime, so either r = 1 or r = j. The rst implies that r
is a unit, contradicting that r is irreducible, so it must be that r = j. But now
there is a n n() such that nr , so nr = c+/i +c, +d/ for some c. /. c. d .
So we have
j = r = n r = nr = c
2
+/
2
+c
2
+d
2
.
That is, each prime is a sum of four squares. Finally, if : = j
c
1
1
j
c
I
I
where
each j
i
is prime and c
i
0, then there are irreducible r
i
such that j
i
= r
i
for
every i. So by the multiplicative property of and for some unit n() such
that r
c
1
1
r
c
I
I
= c +/i +c, +d/ , we have
: = j
c
1
1
j
c
I
I
= r
1

c
1
r
I

c
I
= r
c
1
1
r
c
I
I

= r
c
1
1
r
c
I
I
= r
c
1
1
r
c
I
I
= r
c
1
1
r
c
I
I

= c
2
+/
2
+c
2
+d
2
.
proving the theorem.
To conclude we state a similar result to the Four Squares Theorem above, which
can be shown via a similar method. Also, it seems plausible that the changes in the
setup between this next theorem and the previous theorem can be continued for
other values. However Ramanujan [12, 20] and Kloosterman [13] show that there
are at most 54 dierent combinations of u. r. . .
+
such that expressions of the
form uc
2
+ r/
2
+ c
2
+ .d
2
can express every natural number. From this the only
values that make an order, up to rearrangement, are u = 1. r = 3. = 1 and
. = 3.
49
Theorem 4.5.4 Let =

3i ,

3/ and =

3i , c
where c = (c+

3i +, +

3/),2. Then is a non-commutative Euclidean domain

and every natural number can be written in the form c
2
+3/
2
+c
2
+3d
2
for integers
c. /. c and d.
50
References
[1] Reiner, I., Maximal Orders, Academic Press, 1975.
[2] Swan, R. and Evans, E,. K-Theory of Finite Groups and Orders, Springer-
Verlag, 1970.
[3] Bass, H., Algebraic K-Theory, W.A. Benjamin, Inc , 1968.
[4] Curtis, C. and Reiner, I., Representation Theory of Finite Groups and Asso-
ciative Algebras, John Wiley & Sons, Inc, 1962.
[5] Jacobson, N., Basic Algebra I, Dover Publications, Inc, 2009.
[6] Jacobson, N., Basic Algebra II, Dover Publications, Inc, 2009.
[7] Janusz, G., Algebraic Number Fields, Academic Press, 1973.
[8] Stewart, I. and Tall, D., Algebraic Number Theory, Chapman & Hall, 1987.
[9] Samuel, P., Silberger, A., Algebraic Theory of Numbers, Dover Publications,
Inc, 2008.
[10] Sivaramakrishnan, R., Certain Number-Theoretic Episodes in Algebra, Chap-
man & Hall, 2007.
[11] Brungs, H., Left Euclidean rings, Pacic Journal of Mathematics, Volume 45,
No. 1, 1973.
[12] Ramanujan, S., Collected papers of Srinivasa Ramanujan, Cambridge Uni-
versity Press, 1927.
[13] Kloosterman, H., On the Representation of Numbers in the Form cr
2
+/
2
+
c.
2
+dt
2
, London Mathematical Society, 1926.
[14] The MacTutor History of Mathematics archive, http://www-history.mcs.st-
and.ac.uk/, School of Mathematics and Statistics, University of St Andrews,
Scotland, October 2009.
51