The theory of maximal orders originates in the work of Dedekind, who studied
the factorization properties of ideals of R, where R is a ring of algebraic
integers in an algebraic number field K. As shown by Dedekind, the factoriza
tion theory is especially simple in the extreme case where R is the ring of all
algebraic integers in K. This ring is in fact the unique maximal Zorder in K.
In this book we shall consider the generalization of Dedekind's ideal theory
to the case of maximal Rorders in separable Kalgebras. The entire theory
reduces almost immediately to the case of a maximal Rorder A in a central
simple Kalgebra A. It turns out that there are two ideal theories, one con
cerned with twosided ideals, the other with onesided ideals. The twosided
theory is easier, since the group of twosided fractional Aideals in A is the free
abelian group generated by the prime ideals of A. The difficulty in the one
sided theory is that a left Aideal in A is also a right AI ideal, where AI is some
other maximal order in A. It is therefore necessary to consider the set of
normal ideals in A, namely all onesided ideals relative to the various possible
maximal orders in A. These normal ideals need not form a group, and instead
constitute the Brandt groupoid of A.
The deeper aspects of the theory of maximal orders depend on properties of
central simple algebras over local and global fields. Many of these deeper
results, such as splitting theorems, the theory of the different and discriminant,
reduced norms, and so on, become almost trivial in the commutative case.
The theory of maximal orders is of interest in its own right, and is essentially
the study of "noncommutative arithmetic". The beauty of the subject stems
from the fascinating interplay between the arithmetical properties of orders,
and the algebraic properties of the algebras containing them. Apart from aes
thetic considerations, however, this theory provides an excellent introduction
to the general theory of orders (maximal or not). Questions involving integral
representations of groups, and those concerned with matrices with integer
entries, often reduce to the study of nonmaximal orders. Many problems can
be handled by embedding such orders in maximal orders, and then using
known facts on maximal orders.
One aim of this book is to present, in as selfcontained a fashion as prac
ticable, most of the basic algebraic techniques needed for the study of orders,
ix
Vlll Preface Preface
maximal or not. The subj ect matter is arranged in the form of a textbook, also mention the proof of the KrullSchmidt Theorem in
section 6 for
on the level of a secondyear graduate course. The exercises at the end of algebras over local rings.
simple
each section are an integral part of the book. In many cases, the results of the The final third of the book covers the deeper theory of central
exercises will be needed later in the book, and occasionally are needed within algebras over global fields, and maximal orders in such algebras.
Chapter 7
The results of
the section itself. For this reason, detailed hints are given for many treats Brauer groups, crossedproduct and cyclic algebras.
Grunwald
exercises. Chapter 7 are combined with the Hasse Norm Theorem and
Wang Theorem in Chapter 8, to derive some of the major theorems
on simple
The reader is expected to be familiar with basic facts from module theory
is proved in
and algebraic number theory, such as those given in the introductory chapters algebras over global fields. In particular, Eichler's Theorem
order.
of CurtisReiner [1]. Vie have included, for the convenience of the reader, Chapter 8, and is used to calculate the ideal class group of a maximal
of Picard
a lengthy chapter on algebraic preliminaries. This chapter provides brief Chapter 8 also contains an introduction to Frohlich's theory
of nonmax imal
surveys of topics needed later in the book, and may be skimmed quickly in groups of orders, and a discussion of locally free class groups
are in a sense
a first reading. Proofs are often omitted in sections 15, especially when long orders. The last chapter deals with hereditary orders, which
or computational. In the rest of the book, from section 6 on, proofs are not much harder to handle than maximal orders. Chapter 9 also includes
given in detail. The only exceptions are the Hasse Norm Theorem and the some miscellaneous facts about group rings.
y of
GrunwaldWang Theorem; these are stated without proof, since otherwise This book is based on class notes for courses given at the Universit
and Melody
several additional chapters would become necessary. Illinois in 1969 and 1973. I would like to thank Janet Largent
My thanks
This book is intended as an introduction to maximal orders, and no Armstrong for their excellent work in typing these class notes.
with the proofrea ding, and
attempt has been made to compile an encyclopedic treatise, or to provide also go to the members of the classes who helped
Robert L. Long, who read
the historical background of each result. Our approach draws heavily from corrected errors in the notes. I especially thank
attention ; he deserves credit for
that in Deuring [1], and benefits from the numerous simplifications in the entire manuscript with great care and
and in the mathema tical content.
SwanEvans [1]. We have not covered any of the analytic theory, which is catching innumerable mistakes in printing
I am glad to
readily available in the excellent book by Weil [1]. We have also omitted His suggestions have helped clarify the presentations. Finally
on the contents and style of the
the theory of Asano orders, which would require a book of its own. In many thank my wife Irma, not only for her advice
for her constant encourag ement
ways, the theory of orders merges into the vast topic of algebras over com book, and her help in its preparation, but also
mutative rings. Among the many references on this subj ect, we may mention and support.
support
the fine books by Bass [1], DeMeyerIngraham [1], Kaplansky [1], and It is also a pleasure to acknowledge with thanks the financial
Foundat ion and the Science Research
Matsumura [1]. I received from the National Science
Sections are numbered consecutively throughout the book. A list of perma Council, during part of the time when the book was being written.
nent notation precedes Chapter 1. Boldface "Theorem" indicates that the
November, 1974 Irving Reiner
theorem is one of the maj or results proved in the book. We have distributed
such honors lavishlythere are about 50 such results in the book!
This book divides naturally into three parts. The first part consists of the
preliminary material in Chapter 1, which may be skimmed in a first reading,
together with some generalities on orders in Chapter 2. The second part,
Chapters 36, deals mainly with the ideal theory of maximal orders. In
Chapter 3 we consider such orders in skewfields, in the complete local case.
The Morita correspondence, explained in Chapter 4, is used in Chapter 5
to study maximal orders in central simple algebras in the local case. The local
results are then applied in Chapter 6 to obtain the global theory. Many of
the techniques developed in this book are useful for the theory of nonmaximal
orders. Thus, for example, Chapter 6 contains a proof of the Jordan
Zassenhaus Theorem and a discussion of genus for arbitrary orders. We may
Acknow ledgements Contents
2 Orders
8 Definitions and examples 108
9 Reduced norms and traces 112
10 Existence of maximal orders; discriminants 125
11 Localization of orders 131
4 Morita equivalence
15 Pro generators 154
16 Morita correspondence 161
19 Ideals 181
20 Different, discriminant 184
38 Nonmaximal orders 340 diag(a l'  . , an) diagonal matrix with main diagonal entries at . . . , an
I. In identity matrix (in proper context)
9 Hereditary orders aT' al (aR' ar) left (right) multiplication by the element a
39 Local theory of hereditary orders 351 Z = rational integers. Q = rational field
40 Global theory of hereditary orders 367 R = real field, C = complex field
41 Group rings 379 R+ = bER : � � O}
H = quaternion skewfield over R
K* = K {O}.
387
. {I,
References  where K is a field
Index 391 i =j
(Kronecker delta)
(jij = 0, i =1= j
1. Algebraic Preliminaries
This chapter reviews some basic algebraic techniques. Proofs are often
omitted, especially when they are readily available in standard references,
or when they are too long or too detailed to warrant their inclusion in this
preliminary chapter. Many readers may wish to glance briefly at the contents
of this chapter, referring back to the relevant sections when they are needed
later. Occasionally, results are stated somewhat more generally than neces
sary for later use.
Each ring considered below will be assumed to have a unity element, and
each ring homomorphism preserves unity elements. Every module M over a
ring A is assumed unital, that is, the unity element of A acts as identity operator
on M. Let K be a commutative ring; a ring A is a Kalgebra if there is a ring
homomorphism of K into the center of A. Such a homomorphism permits
us to view A, and all Amodules, as Kmodules. If A, Bare Kalgebras, we
call them isomorphic as K algebras if there is a ring isomorphism cp: A � B
such that cp(C(x) = C(cp(x), C( E K, x E A.
1. INTEGRAL CLOSURE
Throughout this section let K be a field, and A a finite dimensional
Kalgebra. Denote its dimension by (A : K) or dimK A.
(1. 1 ) V � I· K[XJ/(hi(X)),
i =l
Each hi (X) is nonzero in K[XJ since (V : K) is finite. As is well known, the
2 (1 . 8 ) MINIMUM AND CHAR.POL.; NORM, TRACE 3
INTEGRAL CLOSURE (1 .2)
h(a) = O. Further, f(X) I g(X), and the polynomials f(X), g(X) have the same
f
characteristic polynomial of cp acting on V is given by
t
irreducible factors in K[ X], apart from multiplicities.
I
(1 .2) char. pol.K cp = IT hi (X).
i= 1 A remark about notation seems called for. If B is a Kalgebra containing A,
On the other hand, let f(X) E K[ X]; then f( <p) acts on V just as f(X) acts then each a E A also lies in B. Viewing a as element of B does not affect the
on �h� right �an� expression in (l.l). Hence f(<p) = 0 if and only if f(X) calculation of min. pol.K a, and so we have suppressed the subscript A in
anm�Ilates thIS nght hand expression. There is thus a unique monic poly this notation. On the other hand, char. POl.A/K a certainly depends on the
nomIalf(X) E K[X], of least degree, such thatf(<p) = 0; call this polynomial choice of A, since it must be computed by letting a act on a Kbasis for A;
the minimum polynomial of cpo It is given by indeed, char. pol.A/K a i= char. POl.B/K a if B i= A, since char. POl.A/K a has
degree (A : K).
(l. 3 ) min. pol.K cp = L.C.M. {hi (X): 1 � i � t}.
From (1 .2) and (1 .3) we have at once
Suppose now that A =
i=fo1 Kui, and let a E A. We may write
m
a · uj = aipi' 1 � j � m.
(1 .4) THEOREM. Let cp be a K linear tran�formation on the finite dimensional iL
=1
Kspace V, and let
Then
f(X) = min. pol. K cp, g(X) = char. pol.K cpo
(l.8) char. pol. K a = det (bi jX  ai )
Then f(X) divides g(X) in K[X] ; furthermore, f(X) and g(X) have the same xm  (TA/K a)Xm  1 + ... + (It NA/K a.
=
irreducible factors in K[X], apart from multiplicities. Finally, if h(X)E K[X],
then h(cp) = 0 if and only if f(X) divides h(X). . We call TA/K the trace map, and NA/K the norm map. Clearly
TA/K a = trace of aL , NA/K a det av
Now let A be a finite dimensional Kalgebra. Each a E A determines a
=
Klinear transformation aL on A, where aL is the left multiplication x � ax, so from (1 .5) we deduce
x E A. Clearly
(1 .5) (1 .9)
{ T(ra + sf3) = rT(a) + sT(f3)
N(af3) = N(a) · N(f3), N(ra) = rmN(a),
fo� all r, s E K, a, f3 E A. Further, aL = 0 if and only if a = 0, since A has a for r, S E K, a, f3 E A. Here, m = (A:K).
umty element. Therefore the map a � aL , a E A, is a Kalgebra isomorphism
of A into the ring HomK(A, A) of Klinear transformations on A. Now define
(l.6) min. po1.K a = min. pol.K aL , char. pol.A/K a = char. pol.K aL . Ib Integral elements
For h(X) E K[ X] we have h(aL) = h(a)L ' and thus h(a) = 0 if and only if Now let R be an integral domain with quotient field K, and let A be a
h(aL) = O. This shows that min. pol.K a is the unique monic polynomial ·finite dimensional Kalgebra. An element a E A is integral over R if f(a) = 0
f(X) E K[ X], of least degree, such that f(a) = O. For later use, we record for some monic polynomialf(X) E R[ Xl
explicitly the following consequence of (1 .4) :
( 1 . 10) THEOREM. For an element a E A, the following three conditions are
(1 . 7) THEOREM. Let A be a finite dimensional K algebra. For a E A, let equivalent :
(i) a is integral over R.
f(X) = min. pol. K a, g(X) = char. pol.A/K a. (ii) R[ a] is a finitely generated Rmodule.
Then f(X) is the unique monic polynomial in K[X], of least degree, such that (iii) a is contained in some subring B of A, such that B is a finitely generated
f(a) = 0; it has the property that f(X) I h(X) for each h(X) E K[X] such that Rmodule.
4 INTEGRAL CLOSURE
(1.11) (1 . 1 2) INTEGRAL CLOSURE 5
Proof If (i) holds, then there is a relation
domain (see section 2a). Keep the above notation, and let x E R[ex, P]' Then
exn a 1 exn  1 + . . . + a n '
= R[x ] is an Rsubmodule of the finitely generated Rmodule R[ex,
R[x] is also finitely generated as Rmodule, and thus x is integral over R
It is then obvious that
R[ex] = R + Rex . + ... + Rexn  \
by ( 1 . 10).
(ii) If ex and P are integral elements which do not commute, it may well occur
that ex + P is not integral. For example, let R Z, K Q, A M2 (Q) (the
= = =
a finit�ly generated Rmodule. Thus (i) implies (ii). Clearly (ii) implies (iii),
algebra of all 2 x 2 matrices with entries in Q). Let
( l)�,
by
choosmg B = R[ex
To prove that (iii) implies (i), let us write 0
ex =
n 0
B = i L R Pi' 2
=l Both ex and P satisfy the equation X = 0, and so are integral over Z. How
Since B is a ring, and ex ever, 2
min. pol.Q ( ex + P) = X  ± ¢ Z[X] .
n
exPi = L aijPj, Therefore ex
j= 1
We may rewrite these equations as Ie Integral closure
n The integral closure of R in A is the set of all elements of A which are
L (ex<> 1 � i � n. integral over R. If A is a commutative Kalgebra, then by (1.11) the integral
j= 1 closure of R in A is a subring of A. If A is not commutative, the integral closure
Let d det (exbij  ai) E B. By Cramer 's Formula, it follows that d . Pj = °
= need not be a ring.
.
for 1 � } � n. The�efore d· B = 0, and hence d = 0, since B is a ring with
.
umty element. ThIS shows that ex DEFINITION. The integral domain R is integrally closed if the integral
( 1 . 1 2)
det (X<>ij  a) E R[X], and thus ex is integral over R. We have now shown closure of R in its quotient field K coincides with R. Thus, R is integrally
that (iii) implies (i), and the theorem is proved. closed if for ex EK,
f(ex)
(1.11) COROLLARY. Let ex, P E A be such that exp = pex. If both ex and p are
integral over R, then so are ex ( 1 .13) GAUSS' LEMMA. Let R be an integrally closed domain with quotient field
K. Let f(X) E R[X] be a monic polynomial, and suppose that
Proof Since ex and P are integral over R, by (1.10) we may write
f(X) = g(X) · h(X),
k
R[ex] where g(X) and h(X) are monic polynomials in K[Xj. Then both g(X) and
i =l j=l J
h(X) lie in R[X].
Since exp
and therefore Proof. Let K' be a finite extension ofK which is a splitting field for f(X), = P
R [ex, P] = L RuivJ.. and write
. .
I,
J
f(X) = IT (X  Xi ),
Thus R [ex, P]
follows from ( 1 . 1 0) that every element of R[ex, Let R' denote the integral closure of R inK'. Since f(X) E R[X] is monic, and
the corollary is established . f(xi) ° for each i, it follows that each Xi E R'.
=
InK'[X] we have
Remarks. (i) A slightly simpler proof can be given when R is a noetherian g(X) . h(X) = IT (X  X i ),
6 INTEGRAL CLOSURE (1.14) I
I
EXERCISES 7
whence both g(X) and h(X) are expressible as partial products II' (X  Xi)' EXERCISES
But R' is a ring by (1.11), so both g(X) and h(X) have coefficients in R'. The 1. Keep the notation and hypotheses of ( 1.14). Show that an element (J. E A is integral
coefficients of g(X) and h(X) are thus in R' n K. However, R' n K = R, over R if and only if char. pol.A/K (J. E R[X].[Hint: Use (1.7) and (1. 14).]
because R is integrally closed. Thus both g(X) and h(X) lie in R[X], and 2. Let A be a finite dimensional Kalgebra, and let L be any extension field of K.
�he lemma is proved. Each elementa E A gives rise to an element 1 ®a of the Lalgebra L ®K A. Prove that
In order to determine whether an element rx E A is integral over R, we char. pol. L® KA,L 1 ®a = char. pol. AIKa.
must in theory consider all monic polynomials f(X) E K[X] such that In other words, char. pol.AIKa is not affected by the change of ground field from K
f(rx) = 0, and then we must see whether any such f(X) lies in R[X]. In to L.
practice, however, this task is considerably simplified by use of the following 3. Let E ::::J K be fields, and let (E: K) = n. Each <X E E corresponds to a matrix
result. aE Mn (K), obtained by letting <X act as left multiplication on a Kbasis of E. Show
by induction on r that for each matrix (<Xij) E Mr
(1.14) THEOREM. Let R be an integrally closed domain with quotient field K, .
and let A be a finite dimensional K algebra. An element rx E A is integral over det (aij ) = NEIK {det(<Xij)}'
R if and only if 4. Let L be a finite galois extension field of K, with galois group G. Show that for
min. pol. K rx E R[Xl aEL,
char. pol. LIKa = n (X  aO'),
Proof. If min. pol. K rx E R[X], it is clear that rx is integral over R. Conversely,
let g(X) min. pol. K (X, and assume that rx is integral over R. Then f( rx) = 0
=
O'EG
and hence
for some monic poJynomial f(X) E R[X]. Therefore f(X) = g(X)' h(X) for
some monic polynomial h(X) E K[X]. It then follows from Gauss' Lemma NL/Ka = n aO'.
O'EG
that g(X) E R[X], as desired.
[Hint: Let
,./
n
We shall conclude this section by listing some examples of integrally f(X)= min. poLKa = n (X ai), ai E L.
i= 1
closed domains. Proofs are available in standard references such as Bourbaki
[4, Ch. 5, § I] and ZariskiSamuel [I, Ch. V, § 3]. Then n (X aO' ) E K[ X], hence is a power of f ( X). Then compare degrees of
O'EG
char. pol.LIKa and n (X a°').]
O'EG
(1. 1 5) Examples of integrally closed domains.
5. Keep the notation of Exercise 3, and let A be a finite dimensional Ealgebra.
Prove that fora E A,
(a) Every principal ideal domain is integrally closed.
(b) Every unique factorization domain is integrally closed. NAIK a = NEIK ( NAlEa).
(c) Every Dedekind domain is integrally closed (see § 4a). [Hint: Let
(d) Let R be a domain with quotient field K, and let K' be any extension
'A = 'L° EXi aXj
' = 'L <XijXi' <Xij E E.
field of K. Then the integral closure of R in K' is always integrally
closed. Relative to a suitable Kbasis of A, the Klinear transformationaL on A corresponds
(e) Let S be any multiplicative subset of R (see § 3a). If R is integrally to the matrix (aij ). Hence
closed, so is the ring of quotients Sl R. TAlKa = trace of «(iu) = 'L trace of an
(f) Let K[Xl' ... , X n] be the polynomial domain in n indeterminates over i
a field K (or more generally, over a unique factorization domain K).
Then K [xl' ..., xnJ is also a unique factorization domain, and thus is
integrally closed by (b). A similar result holds for norms, by Exercise 3.]
8 HOMOLOGICAL ALGEBRA (2. 1 ) (2.3) MODULES 9
cok <p the cokernel of <p, that is, Note that HomA (X, L), HomA (X, M), etc., a re additive g roups �but not
usually A modules), and the maps f , f* a re additive homomo rph Isms.
cok <p = M/<p(L). * .
The map f goes "in the same di rection " as f, namely from L to M. Smce
When the re is no danger of confusion, we shall w rite Hom instead of Hom A . �
f a rises fro a change in the second va riable occu rring in Hom, we say that
A sequence of Amodules and Ahomom o rphisms Hom is covar iant in the second va riable.
Simila rly, f* goes "in the opposite di rection " from f. Since f* �rises fr?m
(2. 1) X1 II j�
X 2 X·3 �X a change in the fi rst va riable in Hom, we call H om contravariant m the fIrst
n

va riable.
is exact at Xi if ke r 1; = im 1;1' The sequence is exact if it is exact at each
If f :L � M and g:M � N a re Ahomomo rphisms, then
place, that is,
ke r 1; = im 1; 1 ' 2 � i � n 1 . (2.4) (gf ) g .f ,
= (gf)* = f* . g*.

* * *
(Cont rava riance reve rses o rder of composition �f mappings !)
An exact sequence of Amodules and Ahomomo rphisms will sometimes be
The concepts of cova riance and cont ravanance a re ext remely useful
called a Aexact s equence.
in wo rk ing with modules . Thus , for example, let � be another ring, an � let
A s hor t exact s equence is an exact sequence of the fo rm
M = AM� be a (A, �)bimodule, that is, M is a left Amodule and fIght
(2.2) O  LLM�N _ O. �module, with the actions of A and � on M commutmg: .
Exactness of (2.2) is equivalent to the following th ree conditions : f is monic, {}. m)b
ke r g = im !, g is epic. In this case, g induces an isomo rphism M/f(L) � N. If L AL is a left Amo dule, then HomA (M, L) can be giv �n the st �uctu r� of
The short exact sequence (2.2) is s plit if f(L) is a di rect summand of M.
=
go. f*
o  HomA(N, Y)  HomA(M, Y)  HomA(L, Y)
is exact.
there is a unique cp: Y' � Y such that g; = g iCP , i = 1, 2.
It is sometimes convenient to use the concepts of pushouts and pullbacks
Next, we recall some facts about tensor products (see CurtisReiner
of pairs of maps. Given a pair of Ahomomorphisms;; : A � M., i 1 2 [1, §12], for example). Given a right Amodule L and a left Amodule M
(notation: LA ' A M), we may form their tensor product L ®A M. This is an
=
we define a A,.module X, called the pushout of the pair (11 : 12 ), by the form � l �
additive group, but not usually a Amodule. The tensor product L ®A M
is covariant in each variable: given Ahomomorphisms I : Ll � L2 ,
X = (M1 +M2 )!{(/1 a, /2 a): a E A}.
g: M1 � M2 , there is an additive homomorphism
Thus X is 0bta�ned fro� the external direct sum M 1 + M2 by identifying
. I ® g : L 1 ®A M I � L 2 ®A M2 .
t�e Image of A ill M1 WIth the image of A in M2 . There is a commutative
dIagram (called a pushout diagram or fibre s um) Consequently, if .,1 is another ring with unity element, and if t.,LA is a bi
module, then for each AM we can make L ® A M into a left ,1module in a
A�M I
natural way. The action of an element (j E.,1 on L ®A M is given by
J�lM2�X1 gl
where gi is defined by composition: Mi � Ml + M2 � X, i 1 , 2. The = In particular, if A is commutative and L, M are viewed as twosided
pushout X is characterized up to isomorphism by a "universal mapping Amodules, then L ® A M is also a Amodule.
property", as follows: given any commutative diagram of Amodules and The effect of applying X ® A ' or . ® A Y to exact sequences is described
Ahomomorphisms in the following result, stated without proof:
12 HOMOLOGICAL ALGEBRA (2.6) (2.9) MODULES 13
A (8) A Y B 0 A Y C 0A Y
� � � 0.
noetherian if the submodules of M
Finally w e give some results on chain conditions. A Amodule M
satisfy the ascending chain condition
is
Next we list a number of natural isomorphisms involving Hom and (8). (A.C.C.), that is, if every ascending chain of submodules terminates The
In most cases there is no difficulty in writing down explicitly the maps which ring A is left noetherian if the left ideals of A satisfy the A.C.C. The following
yield these isomorphisms, and we shall leave this task to the reader. It will result is used repeatedly (see CurtisReiner [1, §l 1 J) :
also be clear from the context whether the modules which occur are left
Amodules or right Amodules, and we shall suppress such information in (2.9) THEOREM. For a ring A, the following are equivalent:
some of the statements below. (i) A is left noetherian.
(ii) Every finitely generated left Amodule is noetherian.
AM, there is a left Aisomorphism
(2.7) THEOREM. (i) Given (iii) Every submodule of a finitely generated left Amodule is finitely
HomA (A, M) M, generated.
� (iv) Every nonempty collection of submodules of a finitely generated left
obtained by mapping each f EHomA (A, M) onto f( I ). The left Astructure of Amodule contains a maximal element.
Hom (A, M) arises from the AAbimodule structure of A.
A (2.10) COROLLARY. Let R be a commutative noetherian ring, and let A be an
(ii) For each index set I,
Ralgebra which is finitely generated as Rmodule. Then A is left and right
HomA(iIoEI Mi' N) iITEI HomA (Mi' N),
�
noetherian as a ring.
Proof. Every left ideal of A is an Rsubmodule of the finitely generated
HomA (M, IT Ni) IT HomA (M, Ni).
Rmodule A. Since R is noetherian, it follows from (2.9 (ii)) that the Rsub
iEI ieI �
modules of A satisfy the A.C.C. Hence the left ideals of A satisfy the A.C.C.,
so A is left noetherian. Likewise A is right noetherian.
Choosing I to be the finite set {I, . . , k}, it follows as special cases of these
.
formulas that Hom commutes with finite direct sums in either variable : Next, a Amodule M
is artinian if its submodules satisfy the descending
chain condition (D.C.C.), that is, if every descending chain of submQdules
HomA
Ct M" N ) it � HOffiA(Mi, N), terminates. The ring A is left artinian if its left ideals satisfy the D.C.C. We
have (see CurtisReiner [ 1 , § 1 1 J)
HomA (M, it N.) .t: � HomA (M, Ni)· (2.11) T HEOREM. The following statements are equivalent:
(i) A is left artinian.
(2.8) THEOREM. (i) GivenAM, there is a left Aisomorphism (ii) Every finitely generated left Amodule is artinian.
(iii) Every nonempty collection of submodules of a finitely generated
A0AM M, � Amodule contains a minimal element .
obtained by mapping I Ai0 mi EA 0 A M onto I Aimi EM. The left Astructure Just as in the preceding discussion, we obtain
of A 0 A M arises from the AAbimodule structure of A.
(ii) Tensor product commutes with direct sums: (2. 1 2) COROLLARY. Let R be a commutative artinian ring, and let A be an
1'''
in such a way as to preserve identity maps and compositions. We assume chosen free with a finite basis.
e. For ex�m�le,
always that our functors are additive, that is, F(rxi + rx2) = Frxi + Frx2 . A A projective Amodule is a direct summand of a free modul
J I E9 J2 where each Ji is a left ideal of �, then J I nd J 2 are prOjectIve
contravariant functor F: d + flA carries AEd onto FAEflA, and carries if A = '
. �
each rx E Hom1 s (A, left Amodules. Obviously, every free modul e IS projec tIve. Let {MJ be any
A') onto .FaE Hom�(FA', FA
clear from the definiti on of project ivity that I �i
composition of maps. family of Amodules. It is . tIve
each Mi is projec tive. As we sh � ll see, prOjec
A covariant functor F is right exact if it caqies each exact sequence is projective if and only if
and m other ar�as of
Al + A2 + A3 + 0 in d onto an exact sequence FAl + FA2 + FA3 + 0 modules play a basic role in homological algebra, .
at the begmmng
in flA. For example, if d6 denotes the category of abelian groups, and X algebra as well. We state without proof (see references listed
is some fixed right A'module, there is a functor F: A.A + sit, given b � of § 2) :
M + X ®A M, and rx + 1 ® rx. We often write F = X ®A'. Theorem 2.6
is just the assertion that F is a right exact functor. 2.14) THEOREM. Let P be any Amodule. The following statements are
On the other hand, let AX be a fixed left Amodule, and define F: AJIt + si6 equivalent:
by M + HomA(X, M), ME A.A . We must also describe the action of F on (i) P is projective.
split.
maps; if rx E HomA (M, M/), we set Frx = rx* (see (2.3)). Then F is covariant, (ii) Every short exact sequence 0 + X + Y + P + 0 must
and (2.5) asserts that F is a left exact functor. We often write F = HomA(X, . ). (iii) Given any diagram with exact bottom row
>/If
We shall not give the definition here, but remark only that the functor
F:A.A + d6, given by F = HomA(', X), is a contravariant functor which
is left exact (by (2.5)).
A functor F: d + flA is exact if F is both left and right exact. We state XF Y + 0,
without proof:
there exists a Ahomomorphism h such that f = gh.
(iv) Given any epimorphism g:X + Y, the induced map
(2.13) THEOREM. Let F: d + flA be a covariant functor. The following condi
tions are equivalent: g*:HomA(P, X) + HomA(P, Y)
(i) F is an exact functor.
(ii) F preserves exactness of short exact sequences, that is, if is also an epimorphism.
16 HOMOLOGICAL ALGEBRA (2.15) (2. 1S) PROJECTIVE AND FLAT MODULES 17
(v) Bom A (P,') is an exact functor. similar situations. In order to prove certain types of theorems about pro
jective modules, one first treats the case of a free module on one generator,
Remark. There is a concept of injective modules, dual to that of pro then the case of an arbitrary free module, and finally the case of projective
jective modules: a Amodule L is injective if every short exact sequence modules. Of course, one needs to know that the functors which occur com
o� L� X� Y � 0 is necessarily split. The analogue of (2.14) holds true. mute with the operation of forming direct sums. In many arguments, it
We shall seldom need to use injective modules in this book. suffices to restrict attention to finite direct sums.
A right Amodule X is flat if X ®A' preserves exactness, that is, if for (2.1S) Remark. We shall show in § 3c that every ring of quotients R' of a
f
every exact sequence of left Amodules commutative ring R is a flat Rmodule. From (2.15) it then follows that for
9 each inclusion of Rmodules L c M, there is an inclusion of R'modules
L�M�N,
R' ®R L c R' ®R M. Further, there is an R' isomorphism
the sequence of additive groups
�
®1, X®AM ®g) X®AN
R' ®R (M/ L) (R' ®RM)/ (R' ®R L),
X®AL 1 1
a result to be used repeatedly.
is also exact. We may rephrase this definition as follows: the module X A Next we note
determines a covariant functor F : A.A � dlJ. as in § 2(b), by setting
F( M) = X ®AM, M EAAt. Then X is flat if and only if F is an exact functor. (2.19) THEOREM. A right Amodule X is flat if every finitely generated sub
We have already remarked in (2.6) that F is always right exact, and the only module of X is flat.
question to be' decided is whether F is also left exact. Hence we have
Proof. See Rotman [1, Corollary 3.3 1 ]' The converse of this theorem is false.
(2.15) THEOREM. A module X A is flat if and only if X® A' preserves mono
Using this, we prove (see also Exercise 5.3):
morphisms, that is, for each monomorphi�m f : L � M of left Amodules, the
additive homomorphism
(2.20) THEOREM. Let R be a principal ideal domain. Then every torsionfreet
R module is flat.
is also monic.
Proof Submodules of torsionfree modules are also torsionfree. Hence by
(2.16) COROLLARY. Every projective module is flat. (2.19), it suffices to prove that every finitely generated torsionfree R module
X is flat. But by the Structure Theorem for modules over principal ideal
Proof The right Amodule AA is flat, because of the isomorphism given in domains (see CurtisReiner [1, §16J), each such X is Rfree. Hence X is
(2.S(i)). Since tensor product commutes with direct sum (see (2.S)), it follows Rflat, by (2.16), and the theorem is proved.
that every free right Amodule is flat. Now let X A be projective; then X is a
direct summand of some free right Amodule F, and we may write F = X EB X' As a matter of fact, the converse of (2.20) also holds: if X is a flat Rmodule,
for some module X'. Let f : L � M be a monomorphism of left Amodules. then X must be torsionfree (see Rotman [1, Exercise 3.11J).
Then there is a commutative diagram
X'®AL
A flat right A module X is faithfully flat if for each left A module L, the
equality X ® A L = 0 implies that L = O. To test for faithful flatness, we may
llx.®f use the result :
X' ®A L. (2.21) THEOREM. Let XA be flat. The following statements are equivalent :
Since F is flat, IF ® f is monic. Hence also Ix ® f is monic, and therefore X (i) X is faithfully flat.
is flat, as claimed. (ii) A sequence of left A modules is exact if and only if it becomes exact
after applying X ® A' to it.
(2.17) Remark. The technique of the preceding proof can be applied to many tAn Rmodule X is torsionfree if for each nonzero x E X and each nonzero r E R, also rx :f:. O.
18 HOMO LOGICAL ALGEBRAS (2.22) (2.24) EXTENSIONS OF MODULES 19
(iv) For each maximal left ideal m of A, we have X =1= Xm. (2.24)
is called an extension of N by L ; sometimes we refer to the module X itself
Proof See Bourbaki [2, §3, no. 1]. as such an extension. Another extension 0 � L � X' � N � 0 is equivalent
to that in (2.24) if there exists a Aisomorphism e making the following
(2.22) COROLLARY t. Let R be a principal ideal domain, P a prime ideal of R, diagram commutative :
11.1 e) )
and R* the Padic completion of R. Then R* is aflat Rmodule. If R is a dis O�L�X�N�O
crete valuation ring, then R* is faithfully flat as Rmodule. I
N
For each left Rmodule M let M* R* ® R M, a left R*module. Then R* is This gives a Aexact sequence
'"
=
u) ) )
is also exact, and O�K�P�N�O
(ii) L* = 0 if and only if L = o. I
N
is called a deleted projective resolution of N ; it is an exact sequence except (v) If N is a projective module, then 0 � N � N � 0 gives a projective
possibly at Po ' Now let L be a Amodule, and form the sequence of additive resolution of N. It follows at once from Definition (2.27) that Extn(N, L) = 0
groups for all L and all n � 1 . Conversely, it may be shown that if Ext 1 (N, L) 0 =
Hom(P, L) : 0 � Hom(P 0' L) � Hom(P l ' L) ce.4 Hom(P 2 ' L) � " ' , for all L, then N is projective.
where (as usual) Hom is an abbreviation for HomA " By (2.4) we have One of the most important properties of the groups Exe is as follows (see
(<P i + 1 )* . (<pJ* = (<P i ' <P i + 1 )* = 0, i � 1, references) :
and therefore im <pi c ker <pi + l ' Now define (2.29) THEOREM. (Long exact sequence for Ext). Let
(2.27) Ext�(N, L) ker <p! + l /im <P:,
= n � 1. o�ALB� C�O
This yields a sequence of additive groups Ext�(N, L), Ext�(N, L), . . " the be a short exact sequence of Amodules, and let L be any Amodule. Then
first of which coincides with that given in (2.25). there is a long exact sequence of extension groups
22 (2. 3 1 ) EXTENSIONS O F M ODULES 23
HOM OLOGICAL ALGEBRA (2.30)
0 + Hom(C, L) f4 Hom(B, L) L; Hom(A, L) be reduced to that of Extn. This procedure is called dimension shifting. A
more general version is as follows : Given a A exact sequence
� Ext 1 (C, L) � Ext 1 (B, L) .c Ext 1 (A, L)
(2. 3 1 )
� Ext2(C, L) � Ext2(B, L) C Ext2(A, L) � . . . .
where Pi is projective, 0 � i � n, then for each Amodule L and for each
(2.30) � emarks. (i) We h �ve written Hom, Ext instead of HomA' ExtA , for m � 1, there is an isomorphism of additive groups
convemence. The subscnpts on the various maps g*, f* have also been Extm + n + l (N, L) � Extm(K , L).
(2.32) n
omitted ; in accordance with the notation of (2.28(iv)), we should have denoted
by g: the map Extn(C, L) + Extn(B, L) induced by g. (v) Let N be a Amodule, and let
(ii) It is instructive to compare Theorem 2.29 with (2.5(ii)); we now see that o + A L B � C + O
(2.5(ii)) gives the start of the long exact sequence of groups occurring in (2.29). be a Aexact sequence. An analogue of Theorem 2.29 states that there is a
�urthe� ore, we can describe the homomorphism 80 explicitly, provided we long exact sequence of extension groups
VIew Ext (C, L) as the set of equivalence classes of extensions of C by L.
To be precise, given any h E Hom(A, L), we define 80(h) to be the equivalence o + Hom(N, A) � Hom(N, B) � Hom(N, C) � Ext 1 (N, A)
class of the extension occurring in the bottom row of the commutative S Ext 1 (N, B) � Ext 1 (N, C) � Ext2(N, A) + . . . .
diagram
o + A L B .f4 C + O We shall not attempt to describe the maps 8 occurring here.
hi t
lel
't'
Another basic property of extension groups is as follows (see references):
o + L .. X  .. C + O.

(2.35)
Even if A is not necessarily noetherian, it may happen that the module M O � H om
a 1
A (M' , N') + 1
H omA'a l(A , (r)" N')
occurs in some sequence of this type, with r, s finite. In such case, we say that
M is a finitely presented Amodule. We emphasize that when A is left noether in which the vertical maps (x, (X l are precisely those given by (2.36). But (X l is
an isomorphism, since (2.7) gives
ian, every finitely generated left Amodule is necessarily finitely presented.
Now let A and r be Ralgebras and let M, N, . . . be left Amodules. Set r Q9 R HomA(A� >' N) � r Q9 R N(r) � (N')(r ) � HomN (A, (r), N').
A' r Q9 R A,
= M' r Q9 R M, . . . ,
=
It follows at once that (X is monic, and the theorem is proved.
so M' is a left module over the Ralgebra A'. There exists an Rhomomorphism A variant of the preceding result is as follows :
(2.36)
(2.38) Theorem. Let A and r be R algebras, M and N left Amodules. Suppose
given by that r is Rflat, and that M is a finitely presented Amodule. Then the map (X
'y E r, f E HomA (M, N), in (2.36) is a twosided risomorphism.
26 HOMOLOGICAL A LGEBRA (2.39) (2.42) HEREDITARY RINGS 27
tary rings, and some of the important facts about maximal orders are actually a submodule of a free module on k generators is isomorphic to an external
special cases of assertions about hereditary rings. direct sum of at most k left ideals of A.
(iii) There are examples of rings which are left hereditary but not right
(2.44) THEOREM. If A is a left hereditary ring, then every sub module of a free hereditary. However, if A is left and right noetherian, then A is left hereditary
left Amodule M is isomorphic to an external direct sum of left ideals of A, and if and only if A is right hereditary (see Rotman [1, Corollary 9.20J ; the result
is therefore projective. is due to M. Auslander).
(iv) The proof of (2.44), for the case where M is ' not finitely generated,
Proof.t Let {m ' . . . , mk} be a free Abasis of M, and let N be any submodule proceeds in much the same manner as the above proof. The main difference
l
of M. We shall use induction on k to show that N is isomorphic to an external is that k must be allowed to be transfinite, and the proof uses transfinite
direct sum of left ideals of A. The result is obvious for k = 1 , so now let k > 1 , induction.
and assume that the theorem holds for submodules of M', where M ' =
Amz + . . . + Amk • EXERCISES
Each n E N is expressible in the form
1. Let
n = rim 1 + . . + rkmk ,
.
with uniquely determined coefficients {rJ. Let J be the set of all first coeffi
cients r 1 which occur as n ranges over all elements of N. Then J is a left ideal
of A, and there is an epimorphism cp : N � J given by cp (n) = r 1 , n E N .
Clearly ker q> = N n M', so there is a Aexact sequence be a pullback diagram of Amodul,e s. Prove
(i) 92 induces an isomorphism ker 9 1 � ker .f2 ·
o � N n M'  N � J  O. (ii) If f2 is epic, then so is 9 1 .
2. Let
Since A is left hereditary, the left ideal J is Aprojective, and so the above
sequence splits by (2. 1 4). Therefore
N � J + N n M'.
But N n M' is a submodule of M', and thus N n M' is isomorphic to an be a pushout diagram of Amodules. Prove
external direct sum of left ideals of A, by the induction hypothesis. Hence (i) 9 1 induces an isomorphism cok 11 � cok 92 ·
N is also isomorphic to such a sum. Finally, each summand is projective, (ii) If 11 is monic, then so is 92 .
whence so is N.
3. Let r and 11 be rings, and suppose we have modules Mr ' LA ' and a bimodule
AN r . Show that there is a welldefined additive homomorphism
(2.45) Rf>marks. (i) If the ring A has the property that submodules of free left
a: L ® A Homr(M, N) 4 Homr(M, L ® A N),
modules are projective, then in particular all left ideals of A are projective.
By (2.44), we may conclude that A is left hereditary if and only if submodules defined by 1 ® I 4 (I, f), where
of free modules are projective.
(I, f)m l ® jm, t E L, IE Homr(M, N), m E M.
(ii) The importance of (2.44) is that it gives us a structure theorem for =
submodules of free Amodules when A is hereditary. This structure theorem Prove that a is an isomorphism whenever L is l1flat and M is a finitely presented
generalizes the one valid for modules over a principal ideal domain, since rmodule. [Hint : Imitate the proof of (2.3 8), using ® A in place of ® R . J
clearly every such domain is hereditary. The above proof shows further that 4 . Let 11 b e a ring, and let L A b e flat, and M A finitely presented. Use Exercise 3 to
t We give the proof only for the case where M is finitely generated; it is this special case which
prove that
arises most frequently in practice. For the general case, see CartanEilenberg [ 1 ] .
30 LOCALIZATION
(3. 1) RINGS O F QUOTIENTS 31
5. Let .1 be a ring, L� any module, and let M� be a finitely generated projective
module. Prove that
R x S by setting (a, s) (a', S') if and only if t(sa'  s'a) 0 for some t E S.
r.; =
Let a/s (or S  1 a) denote the equivalence class of the pair (a, s). Now define
a b a b ab
ta + sb
[Hint : Use (2. 1 7). ]
�±t= �
6. Let qJ : R � S be a homomorphism of commutative rings, and let
s t ;t '
J be an ideal
in R. Prove that verifying that these operations are indeed well defined. The set of symbols
{a/s} then forms a commutative ring, denoted by S  1 R, with unity element
1/1 and zero element 0/1 . For s E S, we have
as Smodules.
7. Prove Schanuel's Lemma : Given two exact sequences of left Amodules (s/l) . (l/s) 1/1, =
8. Prove the Snake Lemma : Given a commutative diagram of modules, with exact elements of R. The map i permits us to view S 1 R as Rmodule.
rows : If R is an integral domain, the map i may be regarded as an embedding
of R into S 1 R ; in particular, if S R {O}, then S  1 R is precisely the
= 
quotient field of R. For an arbitrary ring R, the map i is monic if and only
if R is Storsionfree (that is, R contains no nonzero Storsion elements, or
equivalently, S contains no divisors of zero).
there is an exact sequence
(3. 1) THEOREM. Let S be a multiplicative subset of R. Then any ring homo
ker a � ker p � ker y � cok a � cok fj � cok y. morphism cp : R � R' which maps each element of S onto a unit of R', extends
uniquely to a ring homomorphism t/J : S 1 R � R' such that t/J i cp.
If qJ is monic, so is qJ * If t/I is epic, so is t/I * . The lemma remains true for a diagram
0 =
of groups and group homomorphisms, if each of the groups A', B' and C' is commuta
.
3a Rings of quotients
so t/J = t/J'.
Throughout this section, R denotes a commutative ring. A multiplicative
subset of R is a subset S closed under mUltiplication, and such that 0 ¢ S, (3.2) COROLLARY. Let cp : R � R' be a ring homomorphism mapping a multi
1 E S. We introduce an equivalence relation on the Cartesian product plicative subset S of R into a multiplicative subset S' of R'. Then there is a
32 LOCALIZATION (3.3) (3 .4) MODULES OF Q UOTIENTS 33
unique ring homomorphism t/J : S  1 R � S'  1 R' which extends cp , that is, an Rsubmodule of S  1 R.
t/J(a/l ) = cp (a)/l , a E R. We shall now define a homomorphism
f : R 1 ®R M � M/M'
3b Modules of quotients
by setting
Let S be a multiplicative subset of the commutative ring R, and M any f(x ® m) = txm + M',
Rmodule. We view S  1 R as Rmodule by means of the canonical map
This map f is welldefined, since if x = aft bit in R ' then a  b is an
i : R � S 1R, and we form the S  1Rmodule
=
1
in R ® R M. Therefore f(1 ® mo ) = 0 in M/M', that is, tmo E M' . This
We shall call S  1 M a module of quotients. Each X E S  1 M is then expressible 1
implies that mo also lies in M', which completes the proof of the theorem.
as a finite sum
(3.4) COROLLARY. The module of quotients S  1 M coincides with the module
Set s = IT Si E S; then whose elements are equivalence classes of ordered pairs (m, s), with (m, s) ,....,
(m', s') if and only if s'm  sm' is an Storsion element of M. The class of
x = L s  1 · a.! s s!� 1 ® m.! = � a !.ss1� 1 m .
S  1 ® L.. 1
= /)(1 m,
S  1 ICJ (m, s) may be identified with the element m/s = S 1 ® m in S  l M.
say. Denote S 1 ® m by m/s; we have thus shown that every element of
The S  1 Rmodule S  lM may be viewed as an Rmodule by use of the
S  1 M is of the form m/s for some m E M, s E S. It is easily checked that
homomorphism i : R � S  1 R; in other words, an element a E R acts on
m/s ± m'/s' = (s'm ± sm')/ss'. S 1 M as left multiplication by the element a/I of S  1 R. We shall say that the
We shall call an element m of an Rmodule M an Storsion element if element a E R acts invertibly on an Rmodule X if the mapping x � ax,
sm = 0 for some s E S. The collection of all Storsion elements of M is called
x E X, gives a onetoone map of X onto itself. It is then clear that every
the Storsion submodule of M, and is an Rsubmodule of M . element of S acts invertibly on the module of quotients S  1 M.
On the other hand, given any Rmodule X on which every element of
(3.3) THEOREM. Let M' be the Storsion submodule of M. Then m/s = 0 in S acts invertibly, we can make X into an S 1 Rmodule by defining
S 1 M if and only if m E M'. (3.5) (a/s) · x = s  l ' ax, a E R, S E S, X E X.
Proof. If m E M', then tm = 0 for some t E S, whence This is the unique way in '.vhich X can be made into an S 1 Rmodule so as
to preserve the action of R on X.
m/s = S 1 ® m = (st)  1 ® tm = O.
Conversely, suppose that S  1 ® mo = 0 in S 1 M. Then I ® mo = (3.6) THEOREM. Let X be an Rmodule on which every element of S acts invertibly,
S(S 1 ® mo) = 0, and therefore I ® mo is expressible as a finite sum of elements and make X into an S  1 Rmodule by means of (3.5). Then there is an S 1 R
of S 1 M of the following three types : isomorphism
S  1R ®R X � X.
(u + u') ® m  u ® m  u' ® m, u ® (m + m')  u ® m  u ® m',
ua ® m  u ® am, Further, for each Rmodule Y, every Rhomomorphism f : Y � X extends
canonically to an S  1 Rhomomorphism f' : S  1 Y � X, given by the formula
with u, u' E S  1 R, m, m' E M, a E R. Choose t E S to be a common denomi
nator for all of the u's which occur in these terms, so that tu E R for each such f'(u ® y) = u · f(y), U E S  1 R, y E Y.
u (or more precisely, tu E i(R )) Let .
t . g(m) 0, that is, g(tm) = 0. But ker 9 im f, and so we may write tm f(l)
= = =
This ring, hereafter denoted by Rp , is called the localization of R at P. Since questions by first settling the local case, and then app�ying this local i�for
every element of R  P is invertible in Rp , it is easily verified that Rp has a mation to the global case. The next few theorems provIde some connectIOns
unique maximal ideal, namely p . Rp . We should remark that the ring between local and global information .
homomorphism i : R 4 Rp , defined in § 3a, enables us to view Rp (and all For each Rmodule M, and each prime ideal P of R, there is an Rhomo
Rpmodules) as Rmodules. Thus P . Rp is the same as i(P) . Rp . morphism
If R happens to be an integral domain, then the mapping i : R _ Rp is
an embedding. In particular, when P = {O} then Rp is precisely the quotient
ip : M  Mp ,
field of the domain R. given by ip(m) = 1 ® m E Rp ® R M, m E M. Conseque�tly there is an Rhomo
.
Returning to the general case, let M be any Rmodule. We define M p = morphism M _ I1 M P
' where P ranges over all maxImal Ideals of R.
be made into an Rpmodule. For each s E S we have Rs + P an ideal in R. If J < R, then t J is contained in some maximal ideal P of
0, so by (3.3) there exists an s E R  P such t� at s . m = O .
R, since P
=
1. =
Taking rth powers, we obtain Hence s E J, s ¢ P, which contradicts the fact that J c P. ThIs shows th �t
J R, and thus m 0 as claimed. The second assertion in the theorem IS
= =
f3s + nr 1 =
then obvious.
for some f3 E R. Therefore s is a unit in the ring R/pr, and hence acts invertibly
on the (R/pr)module M. From (3.6) we deduce that M can be viewed as (3. 1 6) COROLLARY. Let f E HomR (L, M) induce fp : Lp  Mp . Then
Rpmodule, and that there is an Rpisomorphism (i) There are Rpisomorphisms
(3. 1 4)
for each maximal ideal P of R, and each Rmodule M. is exact for each P.
We often refer to problems concerning R modules and Rhomomor
phisms as global problems, whereas those involving Rpmodules are called Proof. Given any f E HomR (L, M), there is an Rexact sequence
local problems. A fundamental technique in algebraic number theory, and t This statement is a consequence of Zorn's Lemma (see CurtisReiner [lJ); it does not
in its generalizations considered in this book, is the method of solving global require the hypothesis that R be noetherian.
38 LOCALIZATION (3 . 1 7) (3.20) LOCALIZATION AT PRIME IDEALS 39
o � ker f � L L M. We may emphasize that when R is noetherian, and A is finitely generated
Since Rp is Rflat, it follows that the sequence as Rmodule, then A is left and right noetherian, and every finitely generated
Amodule is automatically finitely presented.
o � (ker f)p � Lp f4 l\1p Now let A be any Ralgebra; we shall show that the question, as to whether
is Rpexact. This gives an Rpisomorphism (ker f)p � ker fp . Similar a given Aexact sequence is Asplit, is a "local" question, that is, the answer
arguments yield the other assertions in (i). is determined by the answers for the corresponding Apsequences, where
If f is monic, then ker f 0, whence ker fp 0 for each P, by (i). Con
= =
P ranges over the maximal ideals of R.
versely, if each fp is monic, then by (i) and (3. 1 5), also f is monic. This proves
(ii), and the remaining assertions in the theorem follow in a similar manner. (3.20) THEOREM. Let A be an algebra over the commutative ring R, and let
(3. 1 7) COROLLARY. Let R be an integral domain with quotient field K, and (3.21) o � L .4 M 54 N�0
regard R and all of its localizations Rp as embedded in K. Then be a Aexact sequence, where N is a finitely presented Amodule. Then the
R ( j Rp ,
=
sequence is Asplit if and only if for each maximal ideal P of R, the Apexact
p sequence
where P ranges over all maximal ideals of R.
(3.22)
Proof. Obviously, R is contained in the intersection. Conversely, let x E K
be such that x E Rp for all P. Writing x alb, a, b E R, it follows that a E bRp
=
for all P. Let f : R � R be defined by f(r) rb, r E R. Then the image of Proof. The epimorphism 9 induces a map
=
a in cok fp is zero for all P. But by (3. 1 5) and (3. 1 6), the map
9
cok f � TI cok fp * : HomA (N, M) � HomA (N, N).
p
We claim that the sequence (3.21) is split if and only if g* is epic. Indeed, if
is monic. Hence a has zero image in cok f, that is, a E bR. This shows that h : N � M is such that gh I N ' the identity map on N, then for each
x E R, so R contains n Rp , and the proof is complete.
=
Let us next record the following special cases of (2.38) and the "Change *
proof of the claim.
of rings" Theorem 2.43.
N ow let us define
(3. l 8) THEOREM. Let A be an Ralgebra, where R is a commutative ring, and let (gp) * : HomA p {NP ' Mp) � HomA p (Np , Np ),
M be any finitely presented left Amodule. Then for every Amodule N, and where 9 induces gp : Mp � Np . By virtue of (3. l 8), we may identify (gp)
for each prime ideal P of R, there is an R pisomorphism with (g )p . We have seen above that (3.21) is Asplit if and only if g* is epic,*
*
while analogously (3.22) is Apsplit if and only if (g )p is epic. The theorem then
Rp ® R HomA (M, N) � HomA (Mp , Np ).
p
follow immediately from (3. 1 6). *
(3. 1 9) THEOREM. Let A be a left noetherian Ralgebra, where R is a commutative
ring, and let P be any prime ideal of R. Then there is an Rpisomorphism (3.23) COROLLARY. Let A be an Ralgebra, N afinitely presented left Amodule.
Then N is Aprojective if and only if Np is Approjective for each P.
Rp ® R Ext� (M, N) � Ext�p(Mp , Np ), n � 0,
for each pair of left Amodules M, N such that M is finitely generated over A . Proof. Choose a free Amodule F mapping onto N, so there is a Aexact
40 LOCALIZATION (3.24) 41
(3.28) LOCALIZATION AT PRIME IDEALS
sequence ° � L � F � N � 0. If N is projective, then N I F, whence Np i F p
and Np is Approjective. Conversely, if each Np is projective, then each (iii) For every exact sequence (3.26) in which X 0 ' . . . , Xn  1 are projective,
localization of the above sequence is split. Hence the original sequence splits, also Kn is projective.
by (3.22), and so N is projective. We note that hdAM 0 if and only if M is projective. The following
=
p
for every maximal ideal P of R. where P ranges over all maximal ideals of R.
Proof. Suppose that Ap is left hereditary for each P, and let L be any left
ideal of A. Then L is finitely presented as Amodule, since A is left noetherian.
Proof. Denote the sup by s, and let n hdA M. We show first that n � s.
The result is clear if n 00, so assume n finite. Then there is an exact sequence
=
Further, Lp is a left ideal of Ap , and hence Lp is projective. Therefore L is (3.25) with each Xi projective. Localizing at P, we obtain a projective reso
projective, by (3.23). lution of M p' and therefore hdAp Mp � n. This holds for each P, whence
Conversely, let A be left hereditary, and let P be any maximal ideal of R.
s � n, as claimed.
By (3. 1 0), every left ideal of Ap is of the form Lp , for some left ideal L of Conversely, we prove that n � s, and we may assume that s is finite. From
A. Since L is projective as Amodule, it follows as in (3.23) that Lp is Appro
(3. 1 9) we obtain
Rp ® R Ext�+ 1 (M, N) 0
jective, as desired. This complete the proof of the corollary.
=
In order to generalize (3.24), we introduce the concept of homological for all P and for all Amodules N. It now follows from (3. 1 5) that
1
dimension, which is sometimes referred to as "projective dimension". The Ext�+ (M, ) 0, and so hdA M � s. This completes the proof of the theorem.
. =
(3.27) THEOREM. Let M be a left Amodule. The following statements are M be any left Amodule. There is a Aexact sequence
equivalent : O � K � F � M � O,
(i) hdA M � n . in which F is Afree. Since A is hereditary, the submodule K of the free
(ii) Ext� (M, . ) = ° for r � n + 1. module F must be projective, by (2.44). Therefore hdA M � 1 for each M,
w hence also dim A � 1 .
42 LOCALIZAnON (3.29) EXERCISES 43
We are now ready to give a generalization of (3.24). Since Rp is faithfully flat as Rpmodule, it follows that Ext s + 1 (M P ' . ) = 0,
and therefore hd Mp � s. The remaining assertion in (3.30) follows from the
(3.29) THEOREM. Let A be a left noetherian Ralgebra, where R is a commu formula of Auslander used in the proof of (3.29).
tative ring . Then
dim A = sup {dim Ap},
p EXERCISES
where P ranges over all maximal ideals of R. 1 . Let R be a domain with quotient field K, and let M be a direct summand of a
free Rmodule F of rank n. Show that M and all of its localizations M p , where P
Proof. We shall use a formula due to M . Auslander (see Rotman [1, Th. 9. 14J) : ranges over the maximal ideals of R, may be embedded in the ndimensional vector
space K ® R F over K. Then use (3. 1 7) to prove the formula
dim A = sup {hdi\ AI J: J = left ideal of A}.
M = (l Mp ,
But AIJ is a finitely generated Amodule, so by (3.28)
the intersection being taken over all maximal ideals P of R .
hdi\ AIJ = sup {hdi\ ApIJp }. . 2. Let R be a domain with quotient field K , and let X, Y be submodules of an
p P
Rmodule M. Show that if X p = Yp in M p for each P, then X = Y. [Hint : For each
Further, by (3. 1 0), as J ranges over all left ideals of A, Jp ranges over all left P, {(X + Y)/ Y}p = 0.]
ideals of Ap . Therefore we have 3. Let R be a commutative ring, and let S, T be multiplicative subsets of R such
dim A = sup {hdi\ AplJp} =
that S e T. Prove that T  1 R is isomorphic to the ring of quotients of S  1 R relative
sup {dim Ap},
J, P P P to the multiplicative set which is the image of T in S  1 R.
4. Let P ' . . . , P be a set of prime ideals of the commutative ring R, no one of which
as claimed. 1 n
To conclude this section, let us show that homological and global dimen S =R  (P 1 U ... U Pn )'
sions are unchanged by passage to Padic completions. The preceding (i) Show that S is a multiplicative subset of R, and that the distinct maximal ideals
theorems reduce the calculations of hdi\M and dim A to the case where the of S  1 R are Pi s  1 R, I � i � n.
ground ring is Rp , a local ring. If we make the additional hypothesis that (ii) Show that R p, is canonically isomorphic to the localization of S  1 R at the
'
R is noetherian, then each Rp is also noetherian. The Padic completion maximal ideal Pi . S  1 R.
Rp of the ring Rp is then a faithfully flat Rpmodule (see remark following (iii) Show that
(2.23), or Exercise 5.3). Using this fact, we prove
i '" 1
(3.30) THEOREM. Let Rp be the Padic completion of the local noetherian ring if R is an integral domain, the intersection being formed within the quotient field of R.
Rp , and let Ap be a left noetherian Rpalgebra. Set (Reference : Bourbaki [2, Ch. II, § 3, no. 5].)
A.P = Rp ® R Ap , Mp = Rp ®R Mp , 5. Let S be a multiplicative subset of the commutative ring R, and let R' = S  1R .
Let A be an Ralgebra, and set A' = R' ®R A, an R'algebra. Show that every finitely
P l'
where Mp is any finitely generated left Apmodule. Then generated left A'module X contains a finitely generated left Amodule M such that
[Hint : The ring homomorphism A + A' makes X into a left Amodule. Let
for each Mp . Furthermore
dim Ap = dim A.p . X = L A' Xi ' and set M L A X i ' so A'M = X .
n n
i= 1 i= 1
=
Proof. As in the proof of (3.28), we find at once that hdMp � hdMp ' On
The inclusion M c X yields an inclusion A' ®i\ M e A' ® i\ X � X, since R' is
the other hand, if s hd M P ' then for each Apmodule N p we have
Rflat. Hence the map A' ®i\ M + A'M is monic, and thus is an isomorphism. For
=
case of (3 . 1 0),
type of domain beyond principal ideal domains, and share many of their only if N is an Ssaturatedt Rsubmodule of A1. As a special
arithmetical properties. Dedekind domains arise naturally, as follows : let we thus obtain
R be a principal ideal domain with quotient field K, let L be a finite extension
of K, and let S be the integral closure of R in L. Then S is a Dedekind domain (4.0) THEOREM. Let M be an RIattice. There is a onetoone inclusionpre
with quotient field L. As a matter of fact (see (4.4)), this same conclusion holds serving correspondence N ++ W between the set of Rpure sub lattices N of M,
under the weaker hypothesis that R itself is a Dedekind domain. and the set of Ksubspaces W of KM, given by
We may also remark that Dedekind domains are a special kind of maximal
W = KN, N = W n M.
order. One of the main topics of this book is the study of arithmetical proper
ties of maximal orders analogous to those of Dedekind domains. Many of
the results stated in this section are special cases of the theorems to be de 4a Ideal theory, modules, order ideals
veloped in Chapter VI. call R a Dedekind
Let us fix some notation, to be used throughout this section. The annihilator Let R be an integral domain with quotient field K. We
ing three equiva lent conditions :
of an Rmodule M is defined by domain if it satisfies any one of the follow
a projective Rmodule
(4. 1 ) R is a hereditary ring, that is, every ideal of R is
annR M = {a E R : a . M = O} . (see §2t).
every nonzero prime
Likewise we set (4.2) R is a noetherian integrally closed domain such that
annR m = {a E R : a ' m = O}, ideal of R is a maximal ideal.
sible as a product of
for m E M. Call m E M an Rtorsion element if annR m =1= O. The set of all (4.3) Every proper nonzero ideal of R is uniquely expres
order of occurr ence of the factors.
Rtorsion elements of M is the Rtorsion submodule of M, and is the kernel of nonzero prime ideals of R, apart from
A.C.C. (see (2.9)). The
the Rhomomorphism M � K ® R M given by m � 1 ® m, m E M (see §3b). We recall that R is noetherian if its ideals satisfy the
The R rank of M is defined as ring R is integrally closed if each element of K inte.gral
�ver R nec essarily lies
.
ed m R (notatIOn : J < R)
rankRM dimK (K ® R M). in R. A prime ideal of R is an ideal J properly contam
proof that these three
such that RIJ has no zero divisors (see (3 . 1 2)). For the
=
(iv) Let K be an algebraic number field, that is, a finite extension of the
rational field Q . By (4.4), the integral closure of Z in K is a Dedekind domain, J . J' { L Xi Yi : Xi E J, Y i E J'},
=
finite
hereafter denoted by alg. into {K}. Its elements are called algebraic integers.
(v) In many texts, Theorem 4.4 is proved only when L is separable over K. J 1 = {x E K : x J c R} .
For the general version, see ZariskiSamuel [ 1 , Chapter Y, §8, Theorem 1 9] ' Then J + J', J n J', J . J ' and J  1 are fractional Rideals as well. Relative
to the multiplication J . J' defined above, the set of fractional Rideals forms
Let L be a finite field extension of K, and let TL/K be the trace map from a multiplicative group with identity element R, and with J  1 the inverse of J.
L to K, defined as in §la. From T we can construct a bilinear trace form Property (4.3 ) of Dedekind domains extends readily to the factorization of
r : L x L � K by setting fractional ideals, as follows :
(4.5) r(x, y) = TL/ K(xy), x, y E L.
(4.8) THEOREM. The set of fractional Rideals in K is afree abelian group, with
Then r is a symmetric K bilinear form. The form r is called nondegenerate free generators the nonzero prime ideals of R. In other words, every fractional
if for nonzero X E L, the map r(x, ') : L � K is not the zero map. We may ideal is uniquely expressible in the form
represent r by a symmetric matrix 't' as follows : let L = f1' KXi ' where J =� e l . . . Pr et , e i E Z,
i=
n = (L :K), and put where the {PJ are distinct nonzero prime ideals of R, and where no ei is O.
(We agree to write R itself as an "empty product", with t 0) .
= (r(xi, x))1 � i , j� n = ( TL/K( xi x))1 � i, j� n '
=
't'
Then r is nondegenerate if and only if the matrix 1: is nonsingular. A standard Consider next the factorization J = I1 P/i, J' = I1 p/ i , of a pair of
result in field theory is as follows : fractional ideals. If ai � b i for each i, we say that J divides J' (notation : J I J'),
and in this case it is clear that J :=> J'. Conversely, if J :=> J' then J  1J' c R,
(4.6) THEOREM. Let L be a finite field extension of K. The following state whence by (4.8) ai � bi for each i.
ments are equivalent : Given any pair of fractional ideals J, J', we call J + J' their greatest
(i) L is separable over K. common divisor, and J n J' their least common multiple. If the factorizations
(ii) The bilinear trace form from L x L to K is nondegenerate. of J and J' are as above, it follows at once from the preceding paragraph that
(iii) There exists an x E L such that TL/K(x) = 1 . (4.9) J + J' I1 Ptin (ai , b i l , J n J' = I1 p /naX (ai , bi !.
=
(iv) TL/K : L � K is an epimorphism. Two ideals J, J' of R are relatively prime if J + J' = R. By (4.9 ), if J and J'
(We remind the reader that if char K = 0, or more generally if char K are nonzero, they are relatively prime if and only if they have no prime ideal
does not divide the degree (L : K), then L is necessarily separable over K. factor in common. A set of nonzero integral ideals {J l ' . . . , Jn} is pairwise
The same is true whenever K is a finite field.) relatively prime if Ji + Jk = R for 1 � i < k � n.
We give an addendum to (4.4) for the case of separable extensions.
(4. 1 0) THEOREM. (Chinese Remainder Theorem). Let {J l ' . . . , In} be a set of
(4.7) THEOREM . Let R be a Dedekind domain with quotient field K, and let S be pairwise relatively prime integral ideals of R. Then there is a ring isomorphism
48 (4. 1 3) IDEAL THEORY, M ODULES, ORDER IDEALS 49
DEDEKIND DOMAINS (4. 1 1)
R/J1 " · In � (R/J1 ) + . . . + (R/JJ (4. 1 3) THEOREM. Let R be a Dedekind domain. Each R lattice M is R projective,
and is isomorphic to an external direct sum
An easy consequence of (4. 1 0) is the extremely important
M � J1 + · · · + Jn '
(4. 1 1 ) THEOREM. (Strong Approximation Theorem). Let P P be distinct 1" • • n where the {JJ are fractional Rideals, and n rankRM. Further, two such
=
1' n sums L:" Ji and L:" J� are Risomorphic if and only if m n, and the products
m
n
nonzero prime ideals of the Dedekind domain R, and let the elements a . . . , a E K
1" i= 1 i= 1
=
E (PJi . Rp , 1 � i � n, There is a generalization of the invariant factor theorem for modules over
i
P l ' . . , Pn
principal ideal rings, as follows :
b E Rp for all P =1= . '
where P denotes a variable nonzero prime ideal of R. (4. 1 4) THEOREM (Invariant Factor Theorem). Let R be a Dedekind domain,
We shall also record
and let M, N be RIattices such that N c K . M. Then there exist elements
{mJ in M, and fractional Rideals {J J and {EJ, such that
(4. 1 2) THEOREM. LetJ be an integral ideal of the Dedekind domain R, and A
any fractional Rideal. Then there is an Risomorphism
where r rankR M, s rankR N, and where E1 ::;, E2 ::;, . . , ::;, Es . lf N e Al,
R/J � A/JA .
= =
then R ::;, E 1 .
We may partition the set of fractional ideals of R into ideal classes, putting 1
The ideals E ' . . . , Es occurring above are called the invariant factors of
J and J' into the same class whenever J � J' as Rmodules. Note that J � J' the pair M, N; they are uniquely determined by the inclusion map N c K . M.
if and only if J' Jx for some x E K. The ideal classes form a multiplicative
= If N c M and K · N K · M, then M/N is an Rtorsion module, and there
=
V = f: Kvp and set N L RVi ' a free full Rlattice in V. Since M and N are
=
Let us define the order ideal ord X of a finitely generated Rmodule X
i= l
full lattices in the same space V, there exists a nonzero r E R such that r ' M e N. as follows :
But M :;?; r ' M, and thus we conclude that M may be embedded in a free (i) If X 0, ord X
= R. =
Rlattice on n generators. This argument holds for any domain R. 1f however (ii) If X is not an Rtorsion module, ord X O. =
M � J1 + · · · + Jn ' the number of factors equals the number of composition factors of X. The
J
where the { J are fractional Rideals. There must be n summands because ideal p.I can be recovered from the composition factor R/Pi , namely
rankR M n. To complete the discussion of the structure of lattices over a
=
Pi = annR R/Pi ·
Dedekind domain, it is necessary to know under what conditions two such
Since the set of composition factors is uniquely determined by X, according
external direct sums of fractional ideals are isomorphic. The complete answer
is given by a theorem of Steinitz : to the JordanHOlder Theorem, it follows that ord X is welldefined.
50 DEDEKIND DOMAINS (4. 1 7) (4. 1 9) LOCALIZATIONS, VALUATIONS 51
We observe that if X is an Rtorsion module, then ord X = R if and domains. I n the second part, w e shall consider the more general situation in
only if X O. Further, we prove
=
which R is a noetherian integrally closed domain. This more general material
may be skipped in a first reading, since it will occur only peripherally in later
(4. 1 7) THEOREM. (i) For each exact sequence of finitely generated Rmodules chapters.
o � X' � X � X" � 0, Let us introduce some concepts from valuation theory (we shall consider
only rank one valuations I). Let R be the real field, R + the set of nonnegative
we have real numbers.
ord X = (ord X') (ord X").
(4. 1 9) Definition. A valuation of K is a mapping cp : K � R + such that for
(ii) For each nonzero ideal J of R, ord (RjJ) = J. a, b E K,
(i) cp(a) = 0 if and only if a = O.
Proof. (i) is clear from the fact that the composition factors of X are those of (ii) cp(ab) = cp(a)cp(b).
X' together with those of X". For (ii), we can use the factorization J IT Pfi
=
(iii) cp(a + b) � cp(a) + cp(b).
in (4.8) to write an explicit composition series for RjJ, from which it is evident
If the valuation also satisfies the stronger condition
that the composition factors of RjJ are the {RjPi} with multiplicities {eJ
(iv) cp(a + b) � max(cp{a), cp(b)),
From (4. 1 4) we obtain a structure theorem for finitely generated R we call cp nonarchimedean. It is easily verified that every nonarchimedean
modules: valuation satisfies.
(v) cp(a + b) = max(cp(a), cp(b)) whenever cp(a) =1= cp(b).
(4. 1 8) THEOREM. Every finitely generated Rmodule is isomorphic to a finite The trivial valuation is dermed by the formulas cp(O) = 0, cp(a) = 1 for
external direct sum of ideals of R and cyclict modules RjJ, with J an integral a E K, a # O. We shall always exclude the trivial valuation in our discussion
ideal of R. in §§45, so hereafter the term " valuation" always means "nontrivial
valuation".
Proof Given a finitely generated Rmodule X, we can find an Rexact The value group of a valuation cp is the multiplicative group
sequence {cp(a) : a E K, a =1= a}. If this value group is an infinite cyclic group, cp is a
O � N � M�X � O, discrete valuation, and is necessarily nonarchimedean.
Two valuations cp, t/J are equivalent if for a E K,
with M Rfree on r generators, r finite. We then have (in the notation of (4. 1 4))
cp(a) � 1 if and only if t/J(a) � 1.
s . r •
X � MjN � I J i + I CRJEi) , Each valuation cp on K gives rise to a topology on K, by taking as basis for
i=r+ l i= l
the neighborhoods of a point a E K the sets
as claimed. Note that the expression "L'Ji does not occur if X is an Rtorsion
module {x E K : cp(x  a) < B },
where B ranges over all positive real numbers. Equivalent valuations give the
4b Localizations, valuations same topology on K.
Given any nonarchimedean valuation cp on K, let us put
Throughout let R be an integral domain with quotient field K, R # K.
We shall describe some properties of the ring R, and of Rmodules, with R = {a E K : cp{a) � I } .
respect to localization at prime ideals of R. In cases where proofs are omitted, Then R i s a ring, and is called the valuation ring o f cp. The set
the reader may consult the references listed at the beginning of § 4. For the
first part of this subsection, we shall restrict our attention to Dedekind P = {a E K : <p(a) < I}
t An Rmodule is cyclic if it can be generated by one element.
is the unique maximal ideal of R. If cp is a discrete valuation, then P is a
52 DEDEKIND D OMAINS (4.l9) (4.19a) LOCALIZATIONS, VALUATIONS 53
principal ideal, namely P Rn where n is any element of P such that We refer to vp as the exponential valuation associated with P. The proper
cp(n) < I and cp(n) generates the value group of cpo In this case R is a discrete ties of CPP listed in (4. 19) are consequences of the following properties of vp :
=
valuation ring, by which we shall mean a principal ideal domain having a (4.l9a) For any elements a, b E K, we have
(i) vp(a) 00 if and only if a = O.
unique maximal ideal P, and such that P "# O.
One way of obtaining archimedean valuations is as follows : the ordinary =
absolute value I I on the complex field C is an archimedian valuation, whose (ii) vp(ab) vp(a) + vp(b).
=
restriction to any subfield of C is an archimedean valuation on that subfield. (iii) vp(a + b) � min (vp(a), vp(b)), with equality whenever vp(a) "# vp(b).
Now let K be a field which can be embedded in C, and let f.l : K C be an �
elements {(XJ in C such that f (X) TI (X (XJ, and we may define em
= 
=
If P does not occur, set vp(a) = O. Also, we put vp(O) + 00. Now fix some (iii) For any finitely generated Rmodule M, we have Mp 0 if and only if
=
P + annR M R. =
a E K, a "# 0, Proof. Statement (i) is well known (see references), and can easily be deduced
and cpp(O) = O. Then CP P is a discrete nonarchimedean valuation on K, from the formula
whose value group is the cyclic group generated by K. (If instead of K we used vp(J) = min{vp(a) : a E J} .
another real number K', with K' > 1, the valuation cp� thus obtained would be As a consequence of (i), we have R p = Jp for each integral ideal J of R
equivalent to the abovedefined valuation CPP ) .
relatively prime to P.
54 DEDEKIND D OMAINS (4.21 ) (4.22) LOCALIZATIONS, VALUATIONS 55
Statement (ii) follows from the preceding remarks. First o f all, unless X (4.22) Theorem. Let R be a Dedekind domain, M an Rlattice, and let V = KM.
is an Rtorsion module, both ord X and ord Xp are 0. Next, if X is a nonzero For each maximal ideal P of R, let there be given a full Rplattice X(P) in V,
Rtorsion module, it has a composition series such that X(P) = Mp a.e.t Define
X => X' => X" => . . . => X(k ) => 0, N = O X(P),
p
whose composition factors are (say) R/P , R/P2 , • • • , R/Pk , with the {PJ the intersection being formed within V. Then N is a full Rlattice in V, and
l
maximal ideals of R. Then Np = X(P) for all P.
Xp => (X/ )p => . .. => (X ( k»)p => °
Proof Let V =
i= 1
to
Kvp and set L =
i= 1
fo
Rvi • Then L is a full RIattice in V,
is a descending chain of Rmodules, possibly with repetitions, and with
factor modules { (R/P)p } . But and so Lp = Mp a . e.t (See Exercise 4.6). Therefore Lp X(P) a.e., and so
=
P
IT (P)p = (PiIT= P P)p = (ord X)p,
i=P so N is an RIattice in V. Likewise L c bN, so N is a full RIattice in V.
as claimed. (See also Exercise 4.4). It remains for us to show that Np = X(P) for each P. Since N c X(P), it
Finally we prove (iii), remarking in advance that the proof will be valid for follows that
the more general case where R is any commutative ring, and P a maximal Np c Rp ' X(P) X(P). =
Mp S l M; thus M p ° if and only if M is an Storsion module. Since M is where of course s � r. Then each x E X(P) is expressible as a sum x iI ai ni ,
= =
=
finitely generated as Rmodule, we see that M is an Storsion module if and =1
only if there exists an s E S such that s . M = 0. But this occurs if and only if ai E K. By the Strong Approximation Theorem (4. 1 1), we may find elements
ann R M q: P, that is, if and only if P + ann R M = R. This completes the bl , . . . , bs E K such that
proof. bi  ai E Rp , bi E Rp (p ' =1= P), 1 � i � s,
'
So far we have considered only localizations at a single maximal ideal P where P' ranges over all maximal ideals of R distinct from P. Let us put
of R. In order to study relationships between "local" and "global" questions, y = I bini ' Then for P' =1= P, we have Y E R p . N C X(PI) ; on the other hand,
'
one often needs results of the following type, in which the entire set of local y  x I (bi  a)ni E Rp ' N c Np ' so also Y E X(P). Therefore Y E N, and
=
the intersection being formed within KM, and where P ranges over all maximal (4.23) M* = Hom R (M, R), M** = HomR (M*, R).
ideals of R. The evaluation map cp : M � M** is given by
Proof By (4. 1 3), the RIattice M is Rprojective. The desired assertion is now {cp(m)} f = f (m), fE M*.
a consequence of Exercise 3.1 .
Clearly cp = ° ifand only if M* = 0. We call M reflexive ifcp is an isomorphism.
In the same vein, we prove t "a.e." means "almost everywhere", that is, for all but a finite number of P's.
1!1!1
56 DEDE KIND DOMAINS (4.24) (4.26) RAMIFICATION INDEX ; RESIDUE CLASS DEGREE 57
(4.24) THEOREM. Let R be a Dedekind domain. Then every Rlattice is reflexive, converse is false : Krull rings are always integrally closed, but need not be
that is, M � M**. noetherian. For example Z[X l ' X2 , ] is a nonnoetherian Krull ring.
• . •
The evaluation map gives a Kisomorphism V � V**. Since R is noetherian, (4.27) p·S = n P/ i ,
i= 1
M* and M* * are also RIattices, and there are embeddings M* c V*,
M** c V**. Explicitly we have where the {PJ are distinct maximal ideals of S, and the {eJ are positive
M** = {v E V: f(v) E R for all f E M*}. integers. Call ei the ramification index of Pi for the extension LjK, and write
(4.25) THEOREM. Let R be a noetherian integrally closed domain, M an Rlattice, (4.28) 1 � i � g.
and let P range over the minimal primes of R. Then
(i) For each P, Rp is a discrete valuation ring (that is, a principal ideal We say that Pi is unramified in LjK if ei = 1 and SjPi is a separable extension
domain with unique maximal ideal P . R p). of RjP. Likewise, we say that P is unramified in the extension LjK if each Pi
(ii) R = 0 Rp. is unramified in LjK.
(iii) For e� ch nonzero x E R, x is a unit in Rp a.e. Turning to residue class fields, we observe at once that for 1 � i � g,
(iv) M** = O Mp  the field SjPi is an extension field of RjP. We set
p
Remarks. The proof of (i) may be found in Serre [1, p. 19, Prop. 3]' As general (4.29) J; = f(Pp LjK) = (SjP i : RjP),
reference, see Bourbaki [5], Fossum [4J .
Conditions (iHiii) are the postulates for a Krull ring. Thus the theorem and call f the residue class degree of P relative to the extension LjK. The
i
asserts that every noetherian integrally closed domain is a Krull ring. The finiteness of the { J; } is a consequence of the following basic result :
58 DEDEKIND D OMAINS (4.30) (4.32) RAMIFICATION INDEX ; RESIDUE CLASS DEGREE 59
(4.30) THEOREM. Keeping the above notation, we have and so Pi :::J p . S, whence Pj occurs in the factorization of P . S into maximal
9
ideals of S.
iI
=l
eJ'j = (L : K). (iii) We may define a relative norm map NL/K which associates to each
fractional Sideal J of L, a fractional Rideal NL/K (J) of K. For each maximal
Proof. Let n (L : K), so rank RS n. Let Sp Rp ' S ; then Sp is a torsion
ideal Pi of S, we set
NL/K(P) PIi, where P Pi n R, f; f(Pi, LjK).
= = =
Sp � R p(n) . Therefore Since these {PJ form a free basis for the abelian group of fractional Sideals
SjPS � SpjP ' Sp � (RpjP ' Rp)(n ) � (RjP)(n) in L, we may then define NL/K on this group by requiring it to be multiplica
by virtue of (3. 14). This shows that the dimension of SjPS as vector space tive :
over the field RjP is precisely n. NL/K(JJ') NL/K(J) ' NL/K( J').
=
We may also compute this dimension as follows: by (4.10) we have We could also have defined the norm map NL/K as follows : for each non
zero ideal J of S, put
NL/K(J) ordR (SjJ),
=
dimR/iSjP i ei) ei . dimR/iSjP) eJ; , and by (4. 12) the term JjJl' is isomorphic to SjJ'. From (4.17) we then con
= =
clude
for each i, 1 � i � g. This gives
ordR(SjJJ') ordR (SjJ) ' ordR (SjJ'), that is, N(JJ') N(J) ' N(J').
= =
If we now put
and the theorem is proved. NL/K(J1 . J2 1 ) NL/K(J1 ) ' NL/K(J2 )  1
=
for J 1 ' J2 nonzero ideals of S, it follows that NL/K is well defined on the group
(4. 31) Comments. (i) It is easily verified that Sp is the integral closure of Rp in of fractional Sideals in L. We now show that NL/K coincides with NL/K '
L, and that { Pi ' Sp : 1 � i � g} �re the distinct maximal ideals of Spo Further It suffices to verify that
more, (4.33)
Sp S PI n · · · n SPg '
=
using the notation of (4.32). But SjPi is a field extension of RjP of degree 1; ,
where Sp. is the localization of S at Pi ' and where S p Rp ' S. Note that Rp
= so as Rmodules we have SjPi � (RjP)(fi) , and then (4.33) is obvious.
is a discr'ete valuation ring, but Sp is not (unless g 1). In connection with
= (iv) In the special case where K Q, R Z, one defines a counting norm
= =
The latter alternative cannot occur, since 1 1= Pi ' where p is the positive rational prime such that pZ Pi Z. Thus %L/Q(P)
= n
On the other hand, starting with any maximal ideal Pi of S, we may set is the number of elements in the finite field SjPj . As usual, %L/Q is defined on
P Pi n R, a prime ideal of R. Surely P # 0, since for each nonzero a E Pi '
= all fractional Sideals in L, so as to be multiplicative . Then for each nonzero
the constant term in min.pol. K (a) is a nonzero element in P. Thus Pj :::) P, ideal J in S,
60 DEDEKIND DOMAINS (4.33) (4.34) DIFFERENT, DISCRIMINANT 61
k'L/ Q(1) = card (Sj1), (4.34) THEOREM Let J be a fractional Sideal in L. Then
where card (Sj1) denotes the number of elements in the ring Sj J. J = J 1 . '!)(SjR)  1 , J = J.
(v) If L is a galois extension of K, with galois group G, then for each
a E G we have SG = S, where SG = {xG: X E S}. It is well known (see references) Proof Let us abbreviate TL/K as T, and ,!)(SjR) as '!). We have
that G acts transitively on the set of prime ideals {Pi} occurring in (4.27), that
T(J · J  1 ,!)  1 ) = T(S · '!)  1 ) c ' R,
is, there exist elements {aJ E G such that Pi = (P 1 )G; , 1 � i � g. Further�
more, in this case we have so J ":::J J 1 '!)  1 . On the other hand,
e 1 = . . . = eg ' f = . . . = fg . T(JJ) = T(JJ · S),
R :::>
1
whence JJ c '!) 1, so J c J 1 ,!)  1 . This proves that j = J  1 . ,!)  1 . The
4d Different, discriminant
formula for J is then obvious (see also Exercise 4. 1 2).
As in §4c, let L be a finite separable field extension of K, let R be a Dedekind
domain with quotient field K, and let S be the integral closure of R in L. The discriminant of S with respect to R is defined as
Let 7: : L x L � K be the non degenerate bilinear trace form defined in (4 . 5), d(SjR) = NL/K('!) (SjR)) ,
that is,
the norm of the different '!)(SjR). Thus d(SjR) is a nonzero ideal of R, often
7:(X, y) = TL/K(xy), x, y E L. called the discriminant ideal of SjR.
For any set J c L, put
(4.35) THEOREM. (i) For each maximal ideal P of R,
7:(x, 1) = TL/K(x1) = {7:(x, y) : y E J} .
{d(SjR)}p = d(SJRp).
We prove at once that TL1K(S) c.; R. Indeed, each a E S is integral over R,
so char. pol. L/K a E R[X] by Exercise 1.1. Therefore TL/K a E R, as desired. (ii) If
n
Given any fQll RIattice M in L, we can define a complementary lattice S = I O Rxi , n = (L : K),
i= 1
M = {x E L : 7:(x, M) c R}.
then
The map M � M is inclusionreversing. Let us verify that M is a full RIattice
in L. First observe that M is contained in some full free RIattice
n
N = IO RXi ' where n = (L : K). There exist elements {Yj } E L such that Proof For P a maximal ideal of R, and J any Sideal in L, we put Jp = Rp · J,
i=1 an S pideal of L. As remarked earlier, Sp is the integral closure of Rp in L,
7:(xi , Y) = bij , 1 � i, j � n. Then
n
so in the passage from the pair R, S to the pair Rp, Sp' we have introduced a
M ":::J N = jI O Ry . . new pair of Dedekind domains, with quotient fields K, L, respectively. The
=1 J advantage of this procedure is that the new domain Rp is a principal ideal
Similarly, M is contained in some full RIattice in L . Therefore M is itself a domain, and thus Sp is R pfree on n generators, where n = (L : K).
full RIattice in L, as claimed Since the trace map TL/K is Klinear, it follows easily that
In particular, if J is a fractional Sideal in L, then J is an Ssubmodule of
L which is an RIattice, and thus J is also a fractional Sideal of L ; we call j (4.36) '!)( S pjRp) = { ,!) (SjR) }p .
the complementary ideal of J. For the case where J is S itself, we note that Furthermore, for each Sideal J in L, we have
5 ":::J S, so that '8  1 is an integral ideal of S. Let us define the different of S with
respect to R as NL/K(Jp) = { NL/K(1)} P '
,!) (SjR) = 51, that is, forming norms commutes with localization. Applying NL/K to (4.36)
1
and then 5 = '!)(SjR)  may be referred to as the inverse different. yields assertion (i) of the theorem.
62 DEDEKIND DOMAINS (4.37) (4. 38) GLOBAL FIELDS 63
Now suppose that S happens to be Rfree, say S =
i= 1
f: Rx • There exist
i For a E S, we have
elements {Yj } E L with TL/K(X iY) = bij , 1 � i,j � n. Then
S \ where S = to Ry . . Then char . pol. (a) is the characteristic polynomial of the matrix ((Xi)' while
1)(SjR) }
j= 1 char. pol. (a) is that of (ii ii )' which establishes (4.38) at once.
=
d(SjR ) NL/K(1)(SjR)) ordR(Sj1)(SjR)) and where J is an integral ideal of S relatively prime to PI . For each a E PI J
we have ae ° in S, and therefore char. po1. S/ R a = x n. Hence from (4.38)
= =
= ordR(SjS l) = ordR(SjS) (by (4. 12)). we conclude that TL/K(P 1 J) c P. Writing 1) instead of 1)(SjR) for convenience,
Write we then obtain PI J C P1)  1 , and consequently 1) c PI e  1 . Thus PI e  1 1 1),
n and therefore P 1 1 1) whenever e > 1 .
x.! j = a!). . y}. ,
I
1 T o complete the proof, we now assume that e 1 , and we must show that
=
Then (Exercise 4.2) P 1 1 1) if and only if SjP 1 is inseparable over R. Since e = 1, we have
We may remark that this theorem enables us to compute the discriminant Then for a E J,
d(SjR) by calculating each "local" component d(SjR)p , and that by (ii), each char. po1.s/Ii a = Xk · char. pol. S ' iii d .
local component can be obtained once an Rpbasis of Sp is known.
The different 1)(SjR) and the discriminant d(SjR) are of great importance This implies at once that
because they give information about ramification of prime ideals in the TSIR a TS ' /if a', a E J.
=
�roof Let R = RjP, S SjPS, n (L : K); then S is an algebra over the field
= =
A global field K is either an algebraic number field (that is, a finite extension
R, of dimension n. For f (X) E R[X], let /(X) denote its image in R[X] . of the rational field Q), or else a function field (that is, a finite extension of a
We claim that for a E S, field k(X) of rational functions in an indeterminate X over a finite field k).
A prime of K is an equivalence class of valuations of K. We exclude once and
(4.38) char . pol.L/ K a = char. pol.s/I? a. for all the "trivial" valuation q>, defined by q>(0) 0, q>(a) 1 for a E K,
= =
To pr �ve t�is, first replace R by R p ' and S by Sp . Such a change does not a =1= 0. If K is an algebraic number field, there are the archimedean or infinite
affect R or S. Then we may write primes, arising from embeddings of K in the complex field C, and the non
n archimedean or finite primes of K, arising from discrete Padic valuations
n
Sp = S of K, with P ranging over the distinct maximal ideals in the ring alg.int. { K}
i Rp Xi '
I"
=1
= i IO Rxi •
=1 of all algebraic integers in K.
64 DEDE KIND DOMAINS (4.39) EXERCISES 65
If K is a function field, there are no archimedean primes, and the non EXERCISES
archimedean or finite primes arise from discrete valuation rings in K. In this
case, as well as in the number field case, the residue class field associated with Throughout, R is an integral domain with quotient field K. Unless otherwise stated,
a finite prime P is a finite field ; let A1P) denote the number of elements in R is assumed to be a Dedekind domain.
this residue class field (see (4.31 (iv)). 1. Let J be a nonzero ideal of R. Using (4.9), find all ideals of R which contain J.
For each finite prime P of a global field K, we normalize the Padic valua Prove that RjJ is an artinian and noetherian ring. Deduce from this that every finitely
tion q> p by setting generated Rtorsion Rmodule possesses an Rcomposition series.
2. Let R be a principal ideal domain, and let N c M be RIattices with KN KM.
a E K, a i= 0,
=
Suppose that
q>p(O) ° (see §4b). If P is a real prime of K, that is, an infinite prime arising
=
N
r r
a E K.
Show that
If P is a complex prime of K, that is, an infinite prime arising from an em
bedding J.l: K + C such that J.l(K) cj:. R, let ordR MIN = R . det((Xjj)'
[Hint : After change of Rbases of M and N, we may assume that the matrix (aii) is
q>p(a) I J.l(a) 1 2 , a E K.
=
diagonal. Such basis changes have the effect of multiplying det((Xi} by a unit factor from
Thus we have a normalized valuation q>p for each prime P of K. R, and thus do not change R . det((Xjj)' Finally, if ((Xjj) is diagonal, it follows from (4. 1 7)
that ordR MIN is the principal ideal generated by the product of the diagonal entries.]
(4.39) THEOREM (Product Formula). Let K be a global field. Then for each 3. Let X be a finitely generated Rtorsion module. Show that ordRX and annRX
= 1 a.e., and
nonzero a E K, q>p(a) have the same prime ideal factors, apart from multiplicities. Explain the connection
between this result and Theorem 1 .7.
np q>p(a) 1,
4. Let R e S be an inclusion of Dedekind domains, and let X be any finitely gener
=
(4.40) THEOREM (Very Strong Approximation Theorem). Let P i ' . . . , Pn be ideal of R. The result is obvious for X J, while for X
= RjJ it follows from Exercise
=
2.6.]
distinct primes of K (finite or infinite), and let Po be a fixed prime distinct
from these. Given any elements ai , . . " an E K and any e > 0, there exists an 5. Let f E HomR(M, M), where M is an RIattice. Prove that
(4.41) COROLLARY. Let {P i P) be distinct primes of the algebraic number 6. Let M, N be full R lattices in a Kspace V. Prove that Mp Np a.e. [Hint : There
exist nonzero a, 13 E R such that aM c N c pM.]
=
" ' "
field K, and suppose that there exists an infinite prime Po of K distinct from
PI . . . , Pn • Let a I ' . . . , an E K be such that ai E R Pi if Pi is a finite prime,
' 7. Let R be any domain, M a finitely generated Rmodule, and let qJ : M  M** be
where R alg. int. {K}. Let ° < e < 1. Then there exists an a E R such that the evaluation map (see (4.23)). Show that if M is Rtorsionfree, then qJ is monic. Show
0 if and only if M is an Rtorsion module.
=
that M* =
12. Let 1' : V x V + K be a nondegenerate symmetric Kbilinear form on a finite R El3 Ra El3 . . . El3 R an  1 c S,
dimensional Kspace V, and let M be a full RIattice in V. Define and hence prove that the discriminant d(SjR) divides R . disc a.
Nt {x E V: 1'(x, M) c R} . 1 6. Let co be a primitive mth root of l over K, where char K t m. Show that the only
prime ideals of R dividing disc co are those which divide m. Hence if P t Rm, then P is
=
1 7. Give another proof of (4.24) by showing that every finitely generated free R
f�ll RIattice in V. To prove that M M in general, show that the process of forming module is reflexive, and then proving that every finitely generated projective Rmodule
M from M commutes with localization, and thus the problem can be reduced to the
=
is reflexive.
case of RpIattices.]
1 3. Keeping the notation of Exercise 12, let n = rankR M. Define the discriminant
ideal of M with respect to l' to be the Rideal d(M) in K generated by
5, COMPLETIONS AND VALUATIONS
{det ("[(xj , X)) l � i , j � n : X l ' . . . , Xn E M } .
n
Prove that d(Mp) = d( M)p for each maximal ideal P of R. Show that if M
5a Completions
= L' RXi '
i= 1
then We shall review, without proof for the most part, some elementary facts
about completions. The reader is referred to the standard references, listed
d(M) = R . det (1'(x j ' X))l � i , j �n ' at the start of § 4, for details and proofs.
If N c M, where N is another full RIattice in V, prove that Let K be a field with a valuation qJ, topologized as in § 4b. Let K denote
the completion of K relative to this topology; then K is a field whose elements
(4.42) deN) = (ordR MjN)Z ' d(M), are equivalence classes of Cauchy sequences of elements of K, two sequences
and that deN) c d(M). Deduce from this that d(M) = deN) if and only if M N. = being equivalent if their difference is a null sequence. The field K is embedded
[Hint: To prove (4.42), first localize, and then write R in place of Rp . Use the notation in K, and the valuation qJ extends to a valuation {p on K. The field K is
of Exercise 2. Then complete relative to the topology induced by {p, that is, every Cauchy sequence
from K has a limit in K.
If qJ is an archimedean valuation, then so is {p, and K is a complete field
1 4. Let L = K(a) be a finite separable extension of K, and let with respect to an archimedean valuation. The only possibilities for K are
(5.2) COMPLETIONS 69
68 COMPLETIONS AND VALUATIONS (5. 1)
R (the real field) or C (the complex field), and in each case ijJ is equivalent to start with a discrete valuation ring Rp , and pass to its completion Rp . We
the usual absolute value. shall analyze this step in some detail below.
If ({J is nonarchimedean, so also is ijJ ; the two valuations have the same
value group, and the same residue class field (up to isomorphism). In particu (5.2) THEOREM . Let R be a discrete valuation ring with prime element n, quotient
lar, let R be a Dedekind domain with quotient field K, where K i= R, and field K, and let R, K denote completions.
let P be a maximal ideal of R. The completion of K with respect to the Padic (i) For each finitely generated Rmodule M, the map M � R ® R M is an
valuation ({Jp on K (see § 4b) will be denoted by K p (or just K, if there is no inclusion. Denote R ® R M by RM hereafter.
danger of confusion). Call Kp a Padic field, and its elements Padic numbers. (ii) Let V be a finite dimensional K space, and set 11 K ® K V. Theformulas
=
denote a full set of residue class representatives in R of the residue class field kR .
R = R/P, with ° E /7. Each x E Rp is uniquely expressible as of the form R/nkR, k � 0, so it suffices to prove the result for M = R/n
But then M � R ® R M = R/nkR is surely monic (obvious when k = 0, and
x CXo + cx 1 n + cx 2 n 2 + " ' ,
= cxi E /7 , true for k > ° by (5. 1 ) ).
and each y E Kp is uniquely of the form y = nk . x, k E Z, where uniqueness is To prove (ii), use bases ! Suppose first that A1 is a full RIattice in V. We
guaranteed by the condition that CXo i= 0. (If y = 0, take k =  (0). may write
We note that K is dense in K P ' that is, given any cx E K p and any E > 0, n n
n
i
I O RXi '
i= 1 i=1 =1
=
V = I KYi
T= I · RXi ' V = I Kxi ,
i= l i= l
°
i= l
°
Thus T n V is a full RIattice in V such that R ( T n V) = T. This completes each a E K lies in some field L with K c L c K , (L : K) finite (for example,
the proof of the theorem. L= K(a) will do). Set
LK
(5.5) iP(a) = {cp(NL/K a)} l / ( : ) .
We may now prove analogues of (4.21 ) and (4.22) :
Then we find that the value iP(a) is independent of the choice of L, and that
(5.3) Theorem. Let R be a Dedekind domain with quotient field K, let M be an every finite extension of K contained in K is complete with respect to the
Rlattice, and let V = KM. Let P range over all maximal ideals of R. valuation iP .
(i) We have When cp is archimedean, there are only two possibilities :
M = KM n { n M p } .
(i) K = C = K, cp = iP
p (ii) K = R, K = C, iP extends cp,
(ii) F o� each P, let there be given a full R plattice Yep) in K p V, such that where cp and iP are the usual absolute values on R or C.
Yep) = M p a.e. Dt.fme If cp is nonarchimedean, so is iP. However, (p need not be a discrete
N = V n { n Y(P) }. valuation, even if cp is discrete.
p We shall consider discrete valuations in more detail below, and for this
Then N is a full Rlattice in V, and N p = Yep) for all P. purpose let us adopt a more systematic notation. If cp is a discrete valuation
on K, denote by OK its valuation ring, and by PK the maximal ideal of OK '
Proof. By (5.2), M p = KM n M p for each P. Assertion (i) now follows at Let OK = 0K/PK be the residue class field, and let PK = nK ' OK ' so nK is a
once from (4.21). To prove (ii), set X(P) = V n Y(P) for each P. By (5.2), prime element of OK ' Let vK be the exponential valuation on K, defined by
X(P) is a full R pIattice in V. Clearly X(P) = V n M p = M p a.e., and setting
N = n X(P). By (4.22) N is a full RIattice in V, and N p = X(P) for all P. aR = p�K(a) , a E K, a "# 0,
p
Therefore Np = Rp . X(P) = Y(P) for all P, by (5.2), and the theorem is and vK(O) = + 00 .
proved. Any finite extension L of K can be embedded in K, and the restriction of
We indicate the generalization of (5.3) to the case where R is a noetherian iP to L gives a discrete valuation tjJ which extends cp. We define the ramifica
integrally closed domain. Recall that for each minimal prime P of such a tion index e = e(L/K) and residue class degree f = f(L/K) by the formulas
domain R, the localization Rp is a discrete valuation ring, and so we may vL(nK) = e, (OL : OK ) = f.
fonn its completion R p just as above.
(5.6) THEOREM. Let L be a finite extension of the complete field K, and keep the
(5.4) THEOREM. Let M be a full Rlattice in a finite dimensional Kspace V, above notation. Then
where R is a noetherian integrally closed domain with quotient field K. Suppose (i) 0L is the integral closure of OK in L.
that for each mini'!!al prime P of R, we are given an R plattice Y(P) in Vp , (ii) Both e(L/K) and f(L/K) are finite, and
such that Yep) = M p a.e. Set
N = V n { n Y(P) } . e(L/K) ' f(L/K) = (L : K).
p (iii) For each a E L,
Then N is a full Rlattice in V, N is reflexive, and N p = Y(P) for all P. f(L/K)  l . vK(NL/K a).
vL(a) =
Proof. The theorem follows readily from (4.26) and (5.2). (iv) Let a 1 , , af E 0L be such that 0L = LO 0Kai ' and let no , n 1 ,
• . . . . · ,
f(L/K). Then the e f elements {ainj} form a free OKbasis for 0L '
5b Extensions of complete fields
For future use, we state
Throughout this section, let K be a field complete with respect to a valua
tion cp, either archimedean or not, and let K be an algebraic closure of K. (5.7) THEOREM. (Hensel's Lemma). Let f(X) E 0K[X] and suppose that in
Then (see references) we may extend cp to a valuation iP on K, as follows : 0K[ X ] there is a factorization
72 COMPLETIONS AND VALUATIONS (5 . 8) (5. 1 1 ) EXTENSIONS O F COMPLETE FIELDS 73
J(X) = u(X) v(X), u(X) monic, unique unramified extension W of K such that ( W : K) = (ow : O K ) f, =
i= 1
(5.12) T�EOREM. (i) Let L be a completely ramified extension of K of degree n.
Then mm. pol. K nL is an Eisenstein polynomial over OK ' and L K(nL ). The {J;} are necessarily distinct, and we may choose elements { � J E n such
that J;(8J = O. Any Kisomorphism W L � 0 must carry A into a Kconjugate
=
(ii) Let f(X) be an nth degree Eisenstein polynomial over OK ' and let
L = K(Ct), where min. pol. K Ct = f(X). Then L is completely ramified over of some 8j , and since we are interested in embeddings only up to equivalence,
K, (L : K) n, and Ct is a prime element of 0L ' Furthermore; 0L 0K [CtJ . we may indeed choose Pi so that Il j(8) 8j . In this way we obtain r inequiva
=
lent embeddings 11 1 , . . . ,
r
The proof is straightforward (see references), and we omit it. We call L/K
an Eisenstein extension if it is of the type described in (5. 12).
Now let E be any finite extension of K, and suppose that the residue class Now
field OK is perfect, that is, OK has no inseparable extensions. This certainly L = K(8) � K[X]/( f(X)),
is the case when OK is a finite field. If W is the inertia field of the extension and thus
E/K, then we have
011' 0E ' � I " K [X]/(J;(X) ) � I " Li •
r r
followed by a completely ramified step from W to E. We have seen that E (5. 14) char. pol. L/K a11 char. POl.£; / K Il j(a),
is an Eisenstein extension of W, and if OK is finite, then W is a cyclotomic
=
j= 1
extension of E. since a acts on K ®K L just as I pj(a) acts on I" L j • In particular, we obtain
from (5. 1 4) the important formulas
5c Extensions of valuations TL/K a I Tid K Il la),
= NL/ K a 11 N Li t' k. Il i(a). =
j
Throughout this section, K denotes a field with a valuation <p, archimedean
j
� not, and (p the extension of <p to the algebraic closure n of the completion Briefly, global trace is the sum of local traces, and global norm the product
K, as in § 5b. Given a finite separable extension L over K, we wish to determine of local norms.
all extensions of the valuation <p from K to L. Each such extension determines Suppose now that R is a Dedekind domain with quotient field K, and S
an embedding of L in n which preserves the embedding of K in K. Two em
the integral closure of R in L. For each maximal ideal P of R, let
beddings j1, j1' of L in n are called equivalent if there exists a K isomorphism 
(J : p(L) � p' (L) such that (J1l 11' · Let PI , . . . , Ilr be a full set of inequivalent
=
j=1
isomorphisms of L into n which preserve the embedding of K in K. Let
Lj K ' p/L), the composite of K and pj(L) in n (that is, the smallest subfield
= be the factorization of P ' S into a product of powers of distinct maximal
ideals {PJ of S. Then there are precisely r inequivalent valuations t/J l " ' " t/Jr
on L which extend the Padic valuation <p p on K, obtained by choosing
of n containing both), and set nj (Lj : K). Then (see references) there are
precisely r inequivalent valuations t/J l ' . . . , t/Jr of L which extend <p, and
=
these are given by the formula t/Jj to be the P jadic valuation on L. The fields Lj defined above are precisely
the P j adic completions of L, and we have
(5. 1 3) I � i � r.
nj = (L j : K p ) = ejJ; , ej
e(Pi , L/K) = e(L/K p ),
=
<{Jp on Kp . Thus Padic completion. For M a finitely generated Rmodule, we set !VI R ® Jl M. =
Prove
(i) M/nkM � !VI/nk!VI, k � 1 .
(ii) M n nk !VI = 7r!' M, k � 0, where we view M as a submodule of M as in (5.2).
[Hint : We can express M as a direct sum of cyclic modules of the form R/P", n � 0,
Hence for a E L,
r
(This also holds if P is archimedean, with the standard normalizations of Identifying M with M, it is then obvious that (i) and (ii) hold true.]
<{J p , <(J Pi as in § 4e).
6. RADICALS OF RINGS
EXERCISES Throughout this section, A is a ring (with unity element), not necessarily
1. Let P be a maximal ideal of the Dedekind domain R, and let commutative. All modules are unital. For an Amodule M, define its annihila
+ + '" +
tor as
f(X) = X" a 1 X"  1 a n E R[ X],
annA AI = {a E A : aM = O},
2
�here � ach ai E P, and Q".¢ p . (Call f an Eisenstein polynomial). Show that f(X) is and omit the subscript A when there is no danger of confusion. We shall
rreduc ble m . K[X]. [Hmt: By Gauss' Lemma, it suffices to show that f(X) is
� � .
lrredu�lble In R[X] . Let bars denote passage to R/P R. If f gh in R[X], then also write Hom instead of HomA . By an epimorphism A + B of the ring A
onto the ring B, we shall mean always a ring homomorphism which is sur
= =
2. �e� R be a complete discrete valuation ring, R its residue class field, and suppose
that R IS a timte field with q elements. Using Hensel's Lemma (5.7), show that the theory.) As general reference for the material in this section, we cite Jacobson
.
polynomial xq  1  1 splits into q  1 distinct linear factors in R[Xl [1, 2] '
3. L �t R ?e the Padic completion of the discrete valuation ring R. Prove directlyt 6a Jacobson radical
that R IS faIthfully flat as Rmodule. [Hint : We show that R is Rflat that is for
each inclusion i: L c M of Rmodules, the map 1 ® i : R ® L . R ® it is als� an
An Amodule M is irreducible (or simple) if M =I= 0 and M contains no sub
inclusion, where ® is over R. If (1 ® i)u 0, this equation in R ® M holds as a
=
modules except 0 and M. If M is simple, then for each nonzero m E M we
have M = Am. The module epimorphism A + M defined by a + am,
�onse � uence of finitely many defining relations of the tensor product R ® M. Hence
It suffIces to prove that i' : R' ® L . R' ® M is monic, where R' is some finitely a E A, has kernel a left ideal L in A, and thus M � AIL. Since M is simple,
generated Rsubmodule of R. But then R' is Rfree, and so i' is clearly monic. L must be a maximal left ideal of A. Conversely, for each such L, the Amodule
To show that � is fa2 t�fully flat: we show that R ® M =1= 0 if M =1= O. Choose any AIL is simple.
nonzero m E M. Smce R Is 'Rflat, R ® R m is a submodule of R ® M. But R m � R/J, The Jacobson radical of A, denoted by rad A, is defined as
J annR m < R, and then
rad A = n ann M n ann AIL,
R ® (R/J) � R/JR =1= 0.]
=
(6. 1) =
L
be
M
4. Let Rp the Padic completion of a Dedekind domain R. Prove directlyt that where M ranges over all simple left Amodules, and L over all maximal left
t That is, without using (2.23). ideals of A. Since each ann M is a twosided ideal of A, so is rad A.
78 RADICALS OF RINGS 79
(6.2) ( 6 . 4) JACOBSON RADICAL
"semisimple in the sense of Jacobson".) that f(x) = O. From the hypothesis that f(M) M, it follows that M =fn(M),
=
n+
and so we may write x = fn(y) for some y E M. Then 0 = f(x) f 1 (y), =
(6.2) THEOREM. A/rad A is a semisimple ring. whence y E Kn + 1 But then y E Kn , and thus x fn(y) = O. This completes
• =
M, we may view M as a simple Amodule. Now let x E A be such that As an immediate consequence of the above, we have
x + rad A E rad A. Then (x + rad A) M 0, whence xM O. Hence = =
F or the reverse inclusion, let M be any simple A module. Then (1  ba)b = 0, and therefore 1 ba = O. This completes the proof.

An element u E A is a unit if there exists an element v E A such that A(1  (1  t)) = A, that is At = A . Hence there exists u E A such that
uv vu 1 . We shall denote by u(A) the group of units of A. In most of the
= = ut 1, and thus
=
cases arising in this' book, each element of A having a onesided inverse is u = ut(l  y) = I y. 
�ecessarily a unit of A. To prove this, we begin with a result of independent This proves that t is a twosided inverse of I  y, so 1  Y E u(A), as desired.
mterest : (ii) Let x E A be such that 1  axb E u(A) for all a, b E A . To show that
x E rad A, we need prove that xM = 0 for every simple left Amodule M.
(6.3a) 1HEOREM. Let M be a noetherian left Amodule, and let f E HomA(M, M) Let m E M, m i= 0; if xm i= 0, then M A · xm, so m = axm for some
=
if Xk = 0 for some k. An element e E A is idempotent if e =1= 0 and e 2 = e. Proof. If L is a nilpotent left ideaL and X E L, then for all a, b E A also axb is
Obviously, a nilpotent ideal cannot contain any idempotent elements. nilpotent ; namely,
k 1
(axb) = ax ' (baxt  . b E r · b = O.
(6.7) THEOREM. Let N be a nilpotent left ideal of A, and let X E A be a non But then I  axb E u(A), since
nilpotent element such that x 2  x E N. Then the left ideal Ax contains an
1 + axb + (axb) 2 + ...
idempotent y such that y  x E N.
is a twosided inverse of 1  axb. This proves that L c rad A, by (6.5).
Proof· Let Nk = 0, and set nl = x2  X E N. If nl = 0, choose y = x, and On the other hand, rad A must itself be nilpotent ; for, if not, then by (6.8)
we are done. If n1 =1= 0, let rad A contains an idempotent e. Hence 1  e E u(A) by (6.5), which is im
Xl = X + nl  2xnl E Ax. possible since (1  e)e = 0 but e =1= O.
Then x, xl and nl commute with each other, and hence if X l is nilpotent, We may remark that if A is left artinian, then rad A is also the largest
.
so also IS x Xl  nl + 2xnl , a contradiction. Thus Xl is a nonnilpotent
= nilpotent right ideal. For if J is any nilpotent right ideal of A, then AJ is a
element of Ax, and direct calculation gives nilpotent twosided ideal, whence J c AJ c rad A.
X 2l  X l = 4 nl3  3 n 2l . Now we prove some easy properties of radicals.
The element n2 = xi  X l is nilpotent, contains ni as factor, and commutes (6. 1 0) THEOREM. If f : A � B is an epimorphism of rings, then f(rad A) c rad B.
with � l ' Continuing in this manner, we may construct a sequence {xJ of Hence f induces an epimorphism A/rad A � B/rad B.
nonmlpotent elements of Ax, such that nf occurs as a factor in x�  x . . If
we choose i so that 2i � k, then we have x;  Xi = O. Furth�r, x. # 0 Proof. Let x E rad A, y = f(x), and let b1 , b2 E B. Then there exist a1 , a2 E A
�ince Xi is � onnilpotent. Thus we may take y = Xi ' the desired idemp�tent such that bi = f(ai), i = 1 , 2. But 1  a1 x a2 E u(A), whence I  b1yb 2 E u(B).
III Ax such that y  X E N . Hence y E rad B, by (6.5).
(6.8) COROLLARY. Let L be a nonnilpotent left ideal in a left artinian ring A. (6.1 1 ) THEOREM (Nakayama' s Lemma). Let M be a finitely generated left
Then L contains an idempotent element. Amodule such that (rad A)M = M. Then M = O.
Proof. Let Ll be a minimal member of the (nonempty) set of nonnilpotent Proof. If M =1= 0, let m 1 ' . . . , mk be a minimal set of generators of the left
left ideals contained in L; then L/ c L1' whence L/ Ll. Now let I be
=
Amodule M. Since m1 E M = (rad A)M, we may write
minimal in the set of left ideals contained in Ll such that Ll . I "# 0, and
m1 = r 1 m1 + . . , + rk mk , ri E rad A.
choose a E I so that Ll . a "# O. Then Ll . Ll a "# 0 and Lla c I, whence
Ll a = I. Hence a = xa for some X E L I ' and thus a = xka for all k � O. But then I  r 1 E u(A), so we can solve for m1 in terms of m 2 , . . . , mk ' thereby
Therefore I contains the nonnilpotent element x. reducing the number of generators of M.
If we now set
(6. 12) COROLLARY. Let M be a finitely generated left Amodule, and N a
submodule such that
82 RADICALS OF RINGS (6. 1 3 ) (6 . 1 5 ) LOCAL RINGS 83
N + (rad A)M = M. P rad R, and residue class field R R/P. Let A be an Ralgebra, finitely
= =
Then N = M. generated as R module, and let qJ :A ...... A = A/PA be the natural epimorphism
of A onto the finite dimensional Ralgebra A. Then we have
Proof· The hypothesis implies that M/N is finitely generated, and 1
(rad A) (M/N) M/N. The result then follows from (6. 1 1).
=
PA c qJ  (rad A) = rad A,
and (rad A)t c PA for some t. Further, qJ induces a ring isomorphism
(6 13 ) COROLLARY. Every maximal twosided ideal J of A contains rad A.
.
A/rad A � A/rad A.
Proof· �f J � rad A, then J + rad A is a twosided ideal of A properly
contaInIng J. Hence J + rad A A, whence J = A by applying (6. 12) with
=
Proof. We show first that rad A :::) PA. Let M be any simple left Amodule ;
N J, M A. then M = Am for each nonzero m E M, whence M is finitely generated as
Rmodule. Now PM is an Asubmodule of M, hence is ° or M. But PM ::f:. M,
= =
multiplicative group).
Next, rad A is a nilpotent ideal of the left (and right) artinian ring A, by
(6.9), and hence (rad A)! ° for some n. Thus qJ{ rad A )t} = 0, whence
Proof. Suppose first that A is local, with J its unique maximal left. ideal, so
=
z E A with zy = 1, and hence z = x. Thus x E u(A), a contradiction. This is a free Rmodule, and let P denote the maximal ideal of R. We define a
proves that S J, and shows also that each element of A J is a unit in Padic topology on X, by taking as basis for the neighborhoods of a point
X E X the sets {x + pm X : m 0, 1, 2, . . . }. Given a sequence {Yn } from X,
= 
Next, suppose that S is a left ideal of A. Since every proper onesided ideal we may write
of A lies in S, it is clear that S is the unique maximal left ideal (and also Y = r t l )x + . . . + r(nk)xk '
n n 1
rn(i) E R.
right ideal), so A is local. Then {Yn } is a Cauchy sequence in X if and only if, for each i, {r� i )} is a Cauchy
Finally, suppose that A/rad A is a skewfield. Clearly rad A c S, and we sequence in R. Further, {y,,} converges if and only if each {r�:)} converges,
need only prove the reverse inclusion. If x E A  rad A, there exists Y E A and we have
�uch that yx == 1 + n, for some n E rad A. Hence yx E u(A), and x has a left
Inverse in A. Analogously, x has a right inverse, whence x E u(A). Thus
n t, oo n + <Xl
A  rad A c u(A), so rad A c S, as desired. This completes the proof.
Thus if R is complete in the Padic topology, then so is X. When R is not
(6. 1 5) THEOREM. Let R be a commutative local ring, with maximal ideal complete, we may form its Padic completion it Then R ® R X is a free
84 RADICALS OF RINGS (6. 1 6) (6 . 1 8) LIFTING IDEMPOTENTS 85
Rmodule, and can be topologized as above. If X denotes the Padic com idempotents in A. Further, each ej is primitive, that is, ej cannot be expressed
pletion of X, the above discussion shows that there is an Risomorphism as a sum e' + e" of orthogonal idempotents; for if ej = e' + e", then Lj =
Ae' E8 Ae" gives a decomposition of Lj
X � R ® R X,
•
reader to Bourbaki [2, Ch. 3J or Matsumura [1 J for proofs : N ow let A = A /N, where N is a twosided ideal of A such that N rad A,
and let a E A have image a E A. If e E A is idempotent, then e is an idempotent
c
(6.1 6) THEoREM.j et R be a commutative noetherian local ring, with maximal in A; indeed, e 2 = e implies that e2 = e, while if e 0 then e E rad A, an
impossibility (see proof of (6.9)). Thus each decomposition 1 = I ej into
=
and let A be an Ralgebra which is finitely generated as Rmodule. Then A is Cauchy sequence (relative to this topology) converges to an element of A.
a complete Hausdorff space relative to the Padic topology on A . If N rad A, there are two obvious cases where A is necessarily complete
c
We shall discuss here the relationship between idempotents in a ring A, for some t. Hence any Cauchy sequence (from A) relative to the N adic
and idempotents in a factor ring A = A/N, where N is a twosided ideal of topology is also a Cauchy sequence in the Padic topology and thus converges
A contained in rad A. To begin with, suppose that there is a direct sum by (6 . 1 7).
decomposition For the remainder of this section, we assume that A is complete in the
A = L1 E8 . . . E8 Lk , N adic topology.
where each Li is an indecomposable left ideal of A. We may write (6. 1 8) THEOREM. Each idempotent C E A can be lifted to an idempotent e E A,
that is, e = c . If e l ' e2 are idempotents of A, then Ae 1 � Ae2 as left Amodules
if and only if Ae1 � Ae2 as left Amodules.
For each x E Lj ' we obtain x = I xei , whence xei = x and xej = 0 for
j =1= i. This gives Proof. We examine the proof of (6.7). Given an idempotent C E ..4, choose any
e; = ej , ejej = 0 U =1= i), Li = Aei . X l E A with X l = c. Then nl = xi  Xl E N . Once Xi and nj are chosen, let
Thus each ei is idempotent, and we call {e l , . . . , ek } a full set of orthogonal
86 RADICALS OF RINGS (6. 1 9) (6.20) LIFTING IDEMPOTENTS 87
2 i
Then ni E N , and so {xJ is a Cauchy sequence in the N adic topology of
en e  en  1 ; then 1 e1 + ... + en is the desired orthogonal decompo
A . Let e = lim x . E A. Then
= =
1     
e1 <   
)
._     
   '    
>
j e,
idempotents of A. The proof of (6. 1 9) shows that there exist orthogonal
idempotents e 1 , e 2 E A such that e 1 = 81 , e 2 = 8 2 , and e = e 1 + e 2 • This
completes the proof.
J (6.21 ) THEOREM. Let A be a left artinian ring, and let A = A/rad A. Let
1 = e 1 + . . . + en be a decomposition of 1 E A into orthogonal primitive
g
idempotents in A. Then
the vertical arrows are epic. Since A = Aej EB A(l  eJ, each Aej is A A = Ae1 EB . . . EB Aen
projective. Hence we can find Ahomomorphisms f, g lifting j, g, respectively.
is a decomposition of A into indecomposable left ideals of A, and
We claim that f is an isomorphism, and need only show that fJ = g . f is
an automorphism of Ae1 (for then, by symmetry, f · g is an automorphism A = Ae1 EB · · · EB Aen
of Ae 2 ). Clearly fJ lifts g . j, and hence (fJ  l)Ae 1 c Ne 1 Let � = 1  fJ ;
is a decomposition of A into minimal left ideals. Further, Aej � Ae ). if and only
•
of Ae1 ' and is a twosided inverse of 1  �. Since 1  � = fJ, this shows that
fJ is an Aautomorphism of Ael ' as claimed. This completes the proof., n
Proof. Since A is left artinian, it can be expressed as a finite direct sum L:' Lj
1
(6. 19) THEOREM. Let i = 81 + . . . + 8n be a decomposition of I into orthogonal
idempotents in A. Then there exist orthogonal idempotents e 1 , • • . , en E A of indecomposable left ideals of A. Writing I = L: ej , ej E Lj ' we obtain
such that a decomposition of I into orthogonal primitive idempotents. Then I = L: ej
is such a decomposition in A, and so each Aej is an indecomposable left ideal
of A. But A is a semisimple artinian ring, so (see (7. 1 )) every indecomposable
Proof. We proceed by induction on n, the result being trivial when n = 1 . left ideal of A is a minimal left ideal. The last statement in the theorem has
Assume that n > 1 and that the result holds at n  1. Set b = 8 n  1 + 8n ' already been established in (6. 18), so the proof is completed.
so now 1 = 8 1 + . . . + 8n  2 + b is an orthogonal decomposition. By the
_
induction hypothesis there exists an orthogonal decomposition (6.22) COROLLARY. Let A be an Ralgeqra, finitely generated as Rmodule,
where R is a complete noetherian commutative local ring. Then the assertions
1 = e1 + ... + en  2 + e, e = b, in the preceding theorem remain valid for this case.
Choose a E A such that a = 8n 1 ' and set b = eae. Then b b . 8n  1 . b = =
corollary. The first is true, since in the present case A is a left noetherian
ring. The second holds true, by virtue of (6.1 5). Thus the proof of (6.21) applies
equally well in this case, and the corollary is established. V 1 is monic and cp  1 II I . V 1 M l ' Hence there is a
N 1 � M l ' Thus both v1 and II I
is the identity on splitting
N 1 � M 1 EEl ker cp lll l ' and therefore are iso
morphisms.
EXERCISES
Next verify that the sum
1. Let J be a twosided ideal of a ring A, such that AjJ is semisimple. Prove that
(J + rad A)jJ c rad (AjJ)
M' = N l EEl M 2 EEl · · · EEl M[
J :=J rad A. [Hint : By (6. 1 0), O.J
x E A map onto X E A/rad A. Show that x E u(A) if and only if x E u(A/rad A).
=
cp(J) is
rad A, so J
a nilpotent ideal in the and thus M1 M'. Prove next that the map
1 =
Therefore p induces
and let A be an Ralgebra which is finitely generated as R module. We in clude here
the special case where R is a field, and {O} is its maximal ideal.
defines an Aisomorphism of M onto M', carrying M1 onto Nl '
an Aisomorphism
1 is the only idempotent in A. [Hint : Let
#
4. Prove that A is a local ring if and only if
A be local, and let eEA be idempotent, e 1. Since e( l  e) 0, both e and 1
= e  M 2 EB · · · EEl Mt � M/N l � N 2 EB · · · EB Nu '
are nonunits, whence so is their sum ; this is a contradiction. Conversely, if A is not
local, then A/rad A is not a skewfield. If P = rad R, then by (6. 1 5) we know that An induction argument completes the proof.J
§ 7, 7.
contains an idempotent other than 1 . Therefore A contains an idempotent different
A/rad A is a finite dimensional semisimple (R/P)algebra. Hence by A/rad A Let L, M, N be finitely generated left A modu les such that
L + M � L + N.
from 1, by ( 6 . 1 8).J
5. indecomposable left Amodule is a nonzero left Amodule which is not expres
An Prove that M � N. [Hint : Express L, M, N as direct sums of indecomposable modules,
sible as a direct sum of nontrivial submodules. Let M be a nonzero finitely generated and then use the KrullSchmidt Theorem proved above.]
left Amodule, and set E HomA (M, M). Show that E is an Ralgebra which is
=
8. Let {M1 ' . . . , Mt} be finitely generated indecomposable left Amodules, and let L be
finitely generated as Rmodule. Using Exercise 6.4, deduce that M is indecomposable
a left A module which is isomorphic to a direct summand of M1 + . . . + Mt• S how that
M M 1 EEl M 2 into Asubmodules, let 11:1 : M + M 1 be the corresponding projection
as Amodule if and only if E is a local ring. [Hint : Given a nontrivial decomposition
j
e we obtain eJ = e J2 = eJ · J. Therefore eJ 0 by Nakayama's = Lemma. But
M J Hence some /1 1 Vj must be a unit In this ring, that is, must be an
l' J :=J PA by (6. 1 5), and so p . eA O. This gives eA 0, so e = 0 and 0, a
automorphism of M 1 . If (say) II I VI is an automorphism
HomA (M = = 8 =
X = dl ® 1 +
l
By virtue of (7.5), we may speak: unambiguously of artinian simple or Now let b E D', b =P 0; then y = (1 ® b) · X · ( 1 ® b  ) E X, and
semisimple rings, without having to distinguish between "left" and "right" .
A quick proof of (7.4) is sketched in Exercise 7.4. See also Exercises 1 6.9 and
:L di ® bbib  l .
m
y = dl ® 1 +
1 6. 10. 2
Since x  y E X and x  y is shorter than x, this gives y = x. Therefore each
7b Splitting fields
b. commutes with each nonzero b E D', whence each bi E K. Thus x E D,
Let K, L, E denote fields, and D, D' skewfields. The center of an algebra A a�d so x is a unit in D ® D', whence X = D ® D'. This completes the proof.
will be denoted by AC. A central simple Kalgebra is a simple Kalgebra A
for which AC K and (A : K) is finite. Call D a skewfield over K if De ::> K
= (7.8) COROLLARY. Let A be a central simple Kalgebra, L ::> K. Then L ® K A
and (D : K) is finite. By Wedderburn's Theorem, every central simple Kalgebra is a central simple Lalgebra.
A is of the form Mn(D) for some skewfield D over K. Further,
Let A be any ring, M a left Amodule. Each a E A determines a left multi
AC = {aIn : a E De } � De,
plication· aL : m � am, m E M. The map a � aL is a ring homomorp�ism of A
2
so D has center K, and (A : K) = n (D : K). onto a subring AL of Homz (M, M). Call M a faithful Amodule if A � AL
is monic, that is, if a . M = 0 implies that a = o.
(7.6) THEOREM . Let A be a central simple Kalgebra, B an artinian simple K We shall say that the pair (A, M) has the double centralizer property if
algebra, not necessarily finite dimensional over K. Then B ® K A is an artinian AL HomD (M, M), where D = Hom A(M, M). It is trivially verified that
=
simple algebra with center BC. (A, AA) has the double centralizer property, since in this case D = AR , the
set of right multiplications by elements of A.
Proof. We shall first compute (B ® AY, where we write ® instead of ® K for
brevity. Let B = I· Kei , so B ® A :L. ei ® A. If u = :L ei ® ai E (B ® A)e,
(7.9) THEOREM. (i) For each Amodule N, the pair (A, AA + N) has the double
=
(ii) Let M be an Amodule, k a positive integer. If (A, M(k») has the double
for some bo lE B. But now u commutes with b ® 1 for all b E B, whence
centralizer property, then so does (A, M).
bo E B'. Identifying B with B ® 1 in B ® A, we have thus shown that
(B ® AY c Be. The reverse inclusion is obvious, and therefore (B ® AY Be. =
Proof. (i) Let D HomA(M, M), where M AA i N. Each d E D is repre
Since B is artinian and (A : K) is finite, it is clear that B ® A is also artinian.
= =
G �)
=
2 =
(0, ano )' Then the centralizer of B in A. Then B' is a simple artinian ring, and B is its centralizer
in A.
while
Proof. Let V be a simple left Amodule, and let D = Homi V, V), viewed as
dn ({J(a, n) dno (({J 1 1 a, ({J 2 2 n) = (0, (({J I I a)no)'
= left operator domain on V. Then A = Hom D(V, V), and D is a skew field with
o
center K. For a E A, d E D, let aL and dL denote left multiplications on V.
Thus (({J I 1 a)no = ({J 2 (anO) for all (a, no) E M, whence (taking a = 1 ) Then bL ' dL = dL . bL for all b E B, d E D, and therefore we may make V
( )
2
a
(({J l 1 (1))no ({J22(nO)' But ({J commutes with R O E D for each a E A ' into a left D ® K Bmodule by setting
° °
=
(d ® b)v = dL b L v, V E V.
whence ({J u commutes with each aR , and so ({Ju = (aO)L for some ao E A.
Thence ({J22(nO) = aono ' and so ({J is (aO) L on .M. This completes the proof of (i). By (7.6), D ® K B is a simple artinian algebra of finite Kdimension. Both
To prove (ii), let D = Hom A(M, M), and let V = M(k) be the set of 1 x k DL and BL are Ksubalgebras of Hom K(V, V), and the elements of DL com
row vectors with entries in M. If we think of D as a domain of right opera mute with those of BL. Hence the map D ® K B + DL . BL is a Kalgebra
tors on M, then M may be viewed as a bimodule AMD' In that case, isomorphism (the kernel must be a twosided ideal of D ® K B), and so letting
.
E = HomA(V, V) � M (D), and we may view V as a bimodule A VE . Now S denote DL ' BL , it follows that S is a simple algebra. Further, V is a finitely
let f E HomD(M, M); we k must prove that f aL for some a E A Define
=
generated faithful Smodule, so the pair (S, V) has the double centralizer
f* : V + V by property, by (7. 1 0).
Let ({J E Homs(V, V) ; then ({J E HomDL (V, V) = AL , so ({J = aL for some
a E A which centralizes BL . Since both A and B act faithfully on V, this shows
It is easily checked that f* E HomE (V, V). But (A, V) has the double centralizer that a E B', and therefore
property, whence f* aL for some a E A. Then also f = aL (on M), as
=
=
by (7.9(i)) the pair (A, M(k») has the double centralizer property, whence so For later use, we give two important consequences of (7. 1 2) :
does (A, M) by (7.9(ii) .
(7. 1 3) COROLLARY. Keeping the notation of (7. 1 1 ), let V be a simple left
We are now ready to apply this corollary to the study of simple subalgebras Amodule, and set D HomA(V, V). Then
=
(7.1 1 ) Theorem. Let K c B c A, where B is a simple subring of the central Proof. The first assertion is a restatement of (7. 1 2). To prove the second, let
simple Kalgebra A. Let W be a simple left B'module. Since V is a left B' module, we have V� W(k)
96 SEMISIMPLE RINGS AND SIMPLE ALGEBRAS (7. 14) (7. 1 5) SPLITTING FIELDS 97
where k dim V/dim }if!, and dim denotes dim K . Let do dim Do , where
=
=
(dim B) (dim B') = (k 2 do/d) ' dom2 = k 2 d�m2 /d over K, and hence lies in K (see Exercise 3 1 .6).
The important problem for us is to show the existence of splitting fields
= d 2 n 2/d = dim A, which are finite separable extensions of K. This fact is contained in the follow
which completes the proof. ing basic result :
(7. 1 4) COROLLARY. In the notation of (7.1 1 ), we have (7. 1 5) Theorem. Let D be a skewfield with center K, and let (D : K) be finite.
(i) Every maximal subfield E of D contains K, and is a splitting field for D.
A ® K IJO � Mr(B'), where r = (B : K). Further, �f m (E : K), then
=
(D : K ) m 2 , E ® K D � Mm(E).
Further, B ® K B' � A if B has center K.
=
B' � Mm(D�PP), we need only show that dim B kn/m. But this is clear, since
= equality must hold, since for each x E B', E(x) is a subfield of D containing E.
Thus B' = B E, and (7. 1 3) becomes
=
D ® E � Hom i V, V) � Mr(E),
K
=
For the second part of the corollary, we observe that B is a central simple
Kalgebra, and thus B ® B' is simple. Therefore the algebra homomorphism where r = (V: E). But
B ® K B' � B ' B' c A r2 (D ® K E : E) (D : K) = (E : K) 2 ,
= =
and the corollary is established. trivial when n 1. It clearly suffices to handle the case where char K p =I: O.
= =
min. pol.K X = f (Xpe) suffices to prove the result when A is a central simple Ealgebra. We may
write L � K[ X] /(f(X)� where f(X) is an irreducible separable polynomial
for some separable polynomial f (X) E K[X]. In this case, we need only in K [X]. Let f (X) IIJ;(X) be the factorization of f(X) into irreducible
=
choose E = K(xpe). So now suppose that each X E D  K is purely inseparable polynomials in E[X]. Since f(X) is separable, the {h (Xn are distinct, and
over K. Then for each x E D, min. pol. KX is of the form Xpe  a, a E K, therefore
e � O. We note that p i n, since n is a multiple of (K(x) : K) for each X E D  K. "
Choose F to be any maximal subfield of D, so by (i) there is an isomorphism L@ K E � E[XJ/(f(X)) � I E[X] /( h(X)) .
of F alge bras Setting Fi = E[X] /(h (X)), we see that each Fi is a field containing E.
fl : F @K D � Mr(F) for some r. Therefore
(7. 1 8) THEOREM. Let A be a finite dimensional K algebra. The following state (a) rad A =1= 0 , or
ments are equivalent : (b) A is semisimple, but sorp.e Ki is inseparable over K, (using the notation
(i) A is a separable Kalgebra. of the first part of the pro()f).
In case (a), (iii) is false Gust pick F = K !). In case (b), Ki contains an element
(ii) There exists a finite separable field extension E of K such that E ® K A
is a direct sum of full matrix algebras over E. y such that
1
(iii) For every field F ::) K, F ® K A is semisimple . min. pol.K y = (XP)" + cc 1 (XP)"  + . . . + ccn E K[ X],
l
where char K p =1= O. Choose F to be the field K(cc//P , , ccn /p), and
=
• • • let
Proof. Let A be a separable Kalgebra with simple components B 1 , . . . , Bt ,
1
and let Ki be the center of Bi ' so Ki is separable over K, 1 � i � t. For each z =
1 ® yn + CC 1 1 /P ® y n  1 + . . . + ccn / P ® 1 E F ®K Ki .
i, by (7. 1 5) we may choose a separable finite extension field E.I of K I. which .
splits Bi ' say Then z =1= 0, since { l , y, . . . , yn} are Klinearly independent elements of Ki
Further, n
Ei ® K i Bi � Mri (E i)· zP = 1 ® ypn + cc1 ® yp (  1 ) + . . . + ccn ® 1 =
O.
Let us set Ki � K[ XJj(J;(X)), where J;(X) is a separable polynomial over K, Thus F ® K Ki contains a nonzero nilpotent element z. Since
and le� E; be the splitting field of J;(X) over K. Then E; is a finite separable F ® K Bi � (F ®K Ki) ® K i Bi '
extensIOn of K for each i. By Exercise 7.6, we can find a finite separable
exte �sion E of K �hich contains the fields E 1 , . " , Ep E� , . . . , E� (or, more it follows that z ® 1 is a nonzero nilpotent element in the center of F ® K Bi '
.
precIsely, E contams KIsomorphic copies of the {EJ and {EJ). Therefore and thus generates a nonzero nilpotent ideal in F ® K Bi . Therefore neither
F ® K Bi nor F ® K A can be semisimple. This completes the proof of the
E ®K Ki � E[X]/(J;(X)) � L 'E, theorem.
where the number of summands on the right equals the degree of f(X).
I
We then have E ® K A (7. 1 9) COROLLARY. If A and B are separable Kalgebras, so is A ® K B.
L 'E ® K Bp and
=
This shows that (i) implies (ii). and then choose a field E containing both F and F' (Exercise 7.6). Then
To prove that (ii) implies (iii), suppose that
E ®K A � L ' Mri (E),
and so by (7.7)
and let F ::) K. By Exercise 7.6, we may find a field L containing both E and Mrs (E).
E ®K (A ® K B) � (E ® K A) ® E (E ®K B) �
F, and then
L ®K A � L ® E (E ®K A) � L' Mr i (L). Thus A ® K B is separable over K, as claimed.
On the other hand,
Let AO denote the opposite ring of A, that is, AO = {xo : x E A}, with
L ®K A � L ® F (F ®K A). x, Y E A.
If F ® K A were not semisimple, it would contain a nonzero nilpotent ideal
�, and then L �F .N. would. be a nonzero nilpotent ideal in the semisimple Then A and AO have the same center. We set
nng L ® K A. ThIS IS ImpossIble, and so F ® K A is semisimple, as claimed. Ae = A ®K Ao,
Finally, we show that (iii) implies (i), by proving that if (i) is false' then so a finite dimensional Kalgebra called the enveloping algebra of A. We may
is (iii). If (i) is false, then either
102 SEMI SIMPLE RINGS AND SIMPLE ALGEBRAS (7.20) (7.21) SKOLEMNOETHER THEOREM 103
view A as a left Aemodule, by means of the formula Let n: M + a K projection of M onto N, so n2 = n. We set
N be
x, a E A, n' v'(l) · n L xinYi
= = .
(Indeed, every twosided AAbimodule can be viewed as a left Aemodu1e.) Since n = 1 on N, for each n E N we have
There is a left A eepimorphism
n'( n) L x in(yin) = L x iyin = n,
=
fl : Ae A, +
whence (n')2 n'. Finally, for b E B we have
=
obvious that such a map v exists if and only if for each Kbasis { X l ' . . . xn }
'
= v'((l ® bO) . l)n = (1 ® bO )n' n'b. =
fl · V = 1, while (ii) asserts that v is an Aehomomorphism. Conversely, given X axa  1 , X E A, and this automorphism fixes each element of K. The
+
a Kbasis {x J of A, and given elements {y J E A satisfying (i) and (ii), we SkolemNoether Theorem, to be proved below, asserts the converse : every
need only set v(a) a · L Xi ® y�, a E A.
=
algebra automorphism of A fixing each element of K must be an inner
automorphism. We shall prove a slightly more general version of this result,
(7.20) THEOREM. Let A be a finite dimensional Kalgebra. Then A is a separable namely :
Kalgebra if and only if A is Aeprojective.
(7. 21) Theorem. (SkolemNoether). Let K B A, where B is a simple
subring of the central simple Kalgebra A. Then every Kisomorphism qJ of B
c c
We have
Be = B ® F Bo (F ® K A) ® F (F ® AO)
=
Proof. Let ® stand for ® K throughout this proof, and let us use the notation
� F ® K (A ® K AO) F ® K Ae. = introduced in the first paragraph of the proof of (7. 1 1). We have seen there
Let fl' = 1 ® fl : F ® K Ae F ® K A, and likewise let v' 1 ® v . Then v' that a simple left Amodule V is a left D ® Bmodu1e, under the action
E V.
+ =
and have the same Kdimension. It follows at once that there is a left
2. Let V(k) be the direct sum of copies of the left Amodule V, and let E
D ® Bisomorphism e : V � V. In other words, there is an isomorphism =
3. Let A be a ring direct sum L: " B i , where each Bi is a ring having no twosided
If we put d 1, and remember that A acts faithfully on V, we may conclude i= 1
ideals except 0 and B i . Show that every twosided ideal J of A is a sub sum L:" B iv '
=
that
[Hint : J . B i = 0 or B i for each i, and J L:" JB i .]
a·b q>(b) ' a,
=
= b E B,
4. Let L be a minimal left ideal of the simple left artinian ring A, and let D
which proves the theorem.
=
HomA (L, L). Show that A = HomD (L, L). Using "Exercises 1 and 2, prove that D is a
skewfield, and that A � Mn(DO), where n is the dimension of L as left Dspace. [Hint :
(7.23) COROLLARY. (i) Every Kautomorphism of A is inner. This is the harder part of the Wedderburn Structure Theorem (7.4). The following
(ii) Any two Kisomorphic subfields L and E of A are conjugate, that is, quick proof is due to Milnor [ 1 ] . Since L · A is a twosided ideal of A, we have
E = aLa 1 for some invertible a E A. L · A = A . Let
so in (7.25) it suffices to let a range over the left coset representatives of L* in rad (E ® K A ). For each L, N n (L ® K A) is a nilpotent ideal of L ® K A, hence is zero.
D*. Hence there are (D* : L*) sets {aL*a 1 } occurring in (7.25), each of cardi But each x E N lies in some L ® K A.]
nality card L* ; these sets are not disjoint, since each contains 1: Thus D* 6. Let L 1 '
• . ., Ln be a set of field extensions of K. Show that there exists a field 0
cannot be their union, unless there is only one such set, that is, D* = L* . containing K, for which there exist Kisomorphisms Ili : Li  0, 1 � i � n. Further,
Thus D L K, as desired.
= = if each (Li : K) is finite, the field 0 may be chosen so that (0 : K) is finite. If in addition
106 SEMI SIMPLE RINGS AND SIMPLE ALGEBRAS EXERCISES 1 07
each Lj is separable over K, then 0 may be chosen finite separable over K. [Hint : U E A, by Exercise 1 1 . But every maximal left ideal of A is of the form A(1  e) for
Let A L 1 ®K ' " ®K Ln , a commutative Kalgebra. By Zorn's Lemma, A has a some primitive idempotent e.]
maximal ideal M. Set 0
=
group algebra KG is a separable Kalgebra ifand only if char K t n. [Hint : If char K i n, F ®K A L' (F ®K K j ) ®Kj A j •
I
L gj is a nilpotent element in the center of KG, so KG is not semisimple. If
11
then
(7. 1 6), hence
1 Since K.IK is finite separable, the Fa1gebra F ®K K j is semisimple by
char K t n, set A KG, and define v : A ®K A o
A Ae by is expre�sible as a direct sum of fields F ' Each F i} contains Kisomor phic copies of
i}
= + =
v(l ) =
1
n  L 9j ® gj l, F ® 1 and 1 ® Kj .]
j
v(a) = a ' v( I ), a E A. Show that v is an Aehomomorphism such that JlV = 1, where
J,l : Ae + A is as in § 7c.]
9. Let A be an Ralgebra, where R is a commutative ring. Call A Rseparable if
A is Aeprojective, where Ae A ® R A Show that if A is Rseparable, then every
= o.
A = HomD ( V, V), where V is a right vector space over the skewfield D. Since
V = e V EB (I e) V is a Ddecomposition of V, then relative to a suitable Dbasis

of V, e is represented by a diagonal matrix diag (1, . . . , 1 , 0, . . . , 0). The same holds for
e', with the same number of I 's, since Ae � A e . If U E A gives a suitable change of
'
left ideal of A. Show that xA is a maximal right ideal. Prove that if Ax and Ay are
any pair of maximal left ideals in A, then U · Ax ' u  1 Ay for some unit u E A. =
is a homomorphic image of xA, and x A =1= A, whence xA is also a maximal right ideal.
Further, if e and e' are primitive idempotents of A, then e' ueu  1 for some unit =
(8. 1 ) DEFINITIONS AND EXAMPLES 109
Then RG = { 2:: axx : ax E R} is an Rorder in A.
xeG
From example (ii) it is evident that the study of orders includes as a special
case the classical subject of algebraic number theory. On the other hand, in
dealing with representations of a finite group G by means of matrices with
entries in a domain R, one of the first steps is to pass from matrix theory to
2. Orders the study of RGmodules (see CurtisReiner [1] ), so as to be able to take
advantage of the ring structure of RG. Thus the study of orders also includes
Throughout this chapter let R denote a noetherian integral domain with the theory of integral representations of finite groups.
quotient field K, and let A be a finite dimensional Kalgebra. Our aim here Let us show at once that every Kalgebra A contains Rorders. Let M be
is to define orders, and to develop some of their basic properties. In later any full RIattice in A; such M 's exist, and indeed if A = 2:: . KXi ' then
chapters we shall concentrate on maximal orders. Usually R will be a Dede 2:: . RXi is a full RIattice in A. We define the left order of M as
kind domain, or at the very least, an integrally closed domain (see (1.12)), but
occasionally theorems will be stated for the general situation in which R is (8 .1) 0 z (M) = { x E A : xM e M} .
only assumed to be noetherian. Clearly O/(M) is a subring of A, and is an Rmodule. Let us verify that
0z(M) is a full RIattice in A. For each Y E A, yM is an RIattice in A, and so
8. DEFINITIONS AND EXAMPLES by §4 there exists a nonzero r E R such that r yM c M. Thus ry E 0z(M),
.
An Rorder in the Kalgebra A is a subring A of A, having the same unity (8.2) 0 r (M) = {x E A : Mx c M}
element as A, and such that A is a full RIattice in A. Note that A is both left is an Rorder in A.
and right noetherian, since A is finitely generated over the noetherian For a full RIattice M in A, and an Rorder r in A such that r · M c M,
domain R. let us compute Homr(M, M). Each q; E Homr(M, M) extends uniquely to
Let us give some examples of orders : an element of HomA(KM, KM� hence is given by a right multiplication by
(i) If A = Mn(K), �he algebra of all n x n matrices over K, then A Mn(R). = an element of 0r(M). Thus we have an identification
is an Rorder in A.
(ii) Let R be a Dedekind domain, and let L be a finite separable extension (8.3)
of K. Denote by S the integral closure of R in L. Then S is an Rorder in L where M is a left r module. Note that Homr (M, M) does not depend on the
(see (4.7)). In particular, taking R Z, we see that alg. int. {L} is a Zorder
=
choice of the Rorder r in A.
in L. Now let R' be an integral domain containing R, with quotient field K'.
(iii) Let a E A be integral over R, that is, a is a zero of a monic polynomia l Then R' ® R M is a full R'lattice in the K'algebra K' ® K A. Since M is
over R. Then the ring R[a] is an Rorder in the Kalgebra K[aJ . finitely generated over the noetherian ring r, we have by (2.39)
(iv) Let G be a finite group, and let A = KG be its group algebra over K,
that is, KG consists of all formal sums 2:: ax ' x, rJ..;c E K, with R' ® R Homr(M, M) � Hom R'@ R r (R' ® R M, R' ® R M),
xeG provided that R' is Rflat. It then follows from (8 .3) that there is a ring iso
( 2:: ax ' x) + (2:: f3x ' x) = 2:: (ax + f3 J ' x, morphism
(2:: ax . x) ( 2:: f3 y . y)
xI
= axf3 y ' (x y). (8.4)
,y
108
1 10 ORDERS (8.5) (8.8) DEFINITIONS AND EXAMPLES 111
This holds in particular if R' is any ring of quotients of R. Hence it follows that A' is a ring containing R. Further, the map A + E). , A E A',
embeds A' in the RIattice r', and hence A' is also an RIattice.
(8.5) The above shows that A' is an Rorder in A. Therefore A' A, since A is
=
for each prime ideal P of R. Here, the subscript P indicates localization at maximal. Hence every entry of every matrix in r' lies in A, whence r' c r.
P (see § 3d). This shows that r r', and proves that r is a maximal order. If R is integrally
=
The following result is fundamental : closed, then R is a maximal Rorder in K, so the second assertion in the
theorem is the special case of the first assertion, in which A = K, A R. =
case where A is commutative, so that Rei is a ring by (1.1 1), it may happen (ii) Every R'order A' (in a finite dimensional Kalgebra A) contains an
that Rei is not an RIattice. Thus, in example (ii) of maximal orders given Rorder A in A, such that A' R'A.
=
above, if we drop the hypothesis of separability of L over K, then S need not (iii) Let A' R' A, as in (ii). Then every left A'lattice M' contains a left
=
be finitely generated as Rmodule (see Artin [1]); in this case, there are no Alattice M such that M' A'M R'M.
= =
maximal Rorders in L.
By way of reassurance, as well as for later use, we now prove
Proof. It is clear that K is the quotient field of R'. The domain R' is also a
Dedekind domain (see references listed in §4), though we shall not require
(8.7) THEOREM. If A is a maximal Rorder in A, then for each n, Mn (A) is a
this fact here. To prove (i), we note that any R'lattice M' can be written as a
sum
maximal Rorder in Mn( A). If R is integrally closed, then Mn (R) is a maximal s
i= 1
Set M I Rmi ; then M is an R lattice in M' such that R'M M', as desired.
As usual, we identify an RIattice M with its image 1 ® M in K ® R M,
Proof. Let B Mn (A), r Mn (A), where A is a maximal Rorder in A.
= =
= =
{
ORDERS (9. 1 ) REDUCED NORMS AND TRACES
In order to apply the "Change of Rings" Theorem (2.37), we shall need to T(ab) = T(ba), T(a + b) = T(a) + T(b), T(ra) = rT(a) ,
know that R' is Rflat. If X is any finitely generated Rsubmodule of R', then (9. 1)
N(ab) = N(a) N(b), N(ra) = rm N(a),
X is an RIattice. Therefore X is Rprojective, and hence X is Rflat by (2. 1 6).
This shows that every finitely generated Rsubmodule of R' is Rflat, whence for a, b E A, r E K.
also R' is Rflat by (2. 19). The above definitions make no use of the assumption that A is a central
Now let !\' be an R' order in A. By (i), we may choose an RIattice L in simple algebra. For central simple algebras, it is possible to introduce more
A' such that A' = R'L. We set useful concepts, namely those of reduced characteristic polynomial, reduced
norm and reduced trace. We proceed to show how this may be done.
A = 0r(L) = {x E A : Lx c L}, Since A is a simple algebra with center K, there exists by § 7b an extension
Then A and r are Rorders in A, and there is an identification A =
field E of K which splits A. This means that there is an isomorphism of
Homr(L, L), where L is viewed as left rmodule. By (2.37) we obtain Ealgebras
R'M = R'(AM) = A'M. This completes the proof of the theorem. nomial. This show that char. pol. h(u) does not depend on the choice of
the Eisomorphism h in (9.2).
For a E A, define its reduced characteristic polynomial as
EXERCISE
red. char. pol.AIK a = char. pol. h(l ® a).
Let R be a noetherian integrally closed domain with quotient field K, and let
1.
A be a finite dimensional Kalgebra. Let r be a subring of A containing R, and We shall omit the subscript AjK when there is no danger of confusion.
suppose that r is a finitely generated Rmodule. Show that r is contained in an
Rorder in A. [Hint : if M is any full RIattice in A, then so is M · r, and r c 0 r(M . n.] (9.3) THEOREM. For each a E A, red. char. pol.A /K a lies in K[X], and is indepen
dent of the choice of the splitting field E of A used to define red. char. pol.
9. REDUCED NORMS AND TRACES Proof. Let us show first that if F => K is another splitting field for A, then
red. char. pol. obtained using F is the same as that obtained by using E.
9a Central simple algebras By Exercise 7.6, there exists Kembeddings E � a, F � a, where Q is some
Throughout this subsection, let A denote a central simple Kalgebra, field containing K. It suffices to prove that red. char. pol. via E is the same
that is, a simple Kalgebra with center K, such that (A : K) is finite. We have as red. char. pol. via Q. We have an aalgebra isomorphism
seen in § la how to assign to each a E K a characteristic polynomial, norm and
I ® h : Q ® E (E ® K A ) � a ® E Mn (E).
trace, as follows : let aL E HomK (A, A) be left multiplication by a on A, and
set After canonical identifications, we may rewrite the above aisomorphism
char. pol.AIK a char. pol. of a L
=
as (say)
l h' : Q ®K A � Mn(a),
=
xm  (TAlK a) xm + . . . + (  It NAIK a,
where m (A : K). Here T TAlK is the trace map, and N = NAIK the norm
= = so Q is also a splitting field for A. Any element u E E ® K A may also be viewed
map. For any X, Y E M (K) we know that trace (X Y) = trace ( YX). Therefore as element of a ® K A. When this is done, h'(u) is precisely the matrix h(u)
m
1 14 ORDERS (9.4) (9.5) REDUCED NORMS AND TRACES 1 15
viewed as element of Mn(O). Consequently Clearly A has center Q , and E Q(i) is a maximal subfield of A, hence is a
=
char. pol. h(u) = char. pol. h'(u), splitting field for A. Each a E A is uniquely expressible as
f® 1
E ®K A E' �K A X � a /3
= det
i 
I
h h' i . f3 X  &.
Y
f* = X2  (a + &')X + (a&. /3,8) E Q [X].
Mn(E) Mn(E') . +
Here, h is the Eisomorphism in (9.2), and f* takes a matrix in M n(E) onto Hence (see (9.6a) below
a matrix in MneE') by applying f to each of its entries. The map hi is then tr A /Q a = a + &.,
defined by requiring the diagram to be commutative. Then h' is an isomor
phism of E'algebras, since h is an Eisomorphism. Hence for a E A, Returning to the general case, we have
f*{ h( l ® a)} = { h '(f ® In (1 ® a) = h'(1 ® a). (9.5) THEOREM. If (A : K) = n2, then
Therefore char. pol. A / K a = {red. char. pol. A/ K a } n, a E A.
red. char. pol. a (via E') = char. pol. h' ( l ® a)
Proof. Let E K split A, so E ® K A � Mn(E). By Exercise 1 .2,
:::>
=
=
(9.8)
{ tr(ab)
nr(ab)
=
=
tr(ba), tr(a + b) tr(a) + tr (b),
nr(a) ' nr(b),
=
f(X) =
i
L ai X E L[X],
define i
for a, b E A, s E K. I(X) = L Iii X E Mn (K [X]).
(9.9) THEOREM. The reduced trace map gives rise to a symmetric nondegenerate We now define a norm map NL/K : L[X] � K [X ] by setting
K bilinear form T : A x A � K, by setting
NL/K f(X) = det l (X), f(X) E L[X].
T(a, b) = tr(ab), a, b E A.
Let us establish some basic properties of this norm :
Further, T is associative, that is,
T(ab, c) = T(a, bc) for a, b, c E A. (9.10) THEOREM. Let K c L be fields, with (L : K) = n. Then
(i) N L/K is multiplicative.
Proof. It is clear that T is symmetric and associative, and we need only prove (ii) For a I ' . . . , as E L, we have
that T is nondegenerate. Let E ::J K split A, so E (8) K A � Mn(E). We may
extend T to an Ebilinear map (9. 1 1 ) NL/ K (XS + a 1 xs  1 + . . . + a)
n n 1
= x s + (TL/K ( ) X s  + . . . + NL/K as '
T' : (E (8) A) x (E (8) A) � E, 1
and indeed from the definition of reduced trace, we have (iii) Let V be an sdimensional Lspace, and let cp be an Lendomorphism
of V. We may view cp as a K endomorphism of V. Then
T'(U, v) = trace (uv),
(9. 12) char. pol. Y/K cp = N L/K (char. po1 . Y/ L cp).
where trace (uv) is the ordinary trace of the matrix uv. Furthermore, r is
nondegenerate if and only if T' is non degenerate. Proof. Write N in place of N L/K ' and let f, g E L[ X]. The map a � Ii, a E L,
Let is a ring homomorphism. Therefore jg = jg, and so
B = {u E Mn(E) : trace (uv) = 0 for all v E Mn(E)}. N(fg) = det (jg) = (det l) (det g) = (Nf) (Ng).
1 18 ORDERS (9. 1 2 ) (9. 1 3) REDUCED NORMS AND TRACES 1 19
This establishes (i). Now perform elementary row operations on this determinant, from the
Next we have bottom up : add X times the sth row of blocks to the (s  l )th row, then X
N(XS + a 1 X s  1 + . as ) det ( l XS + a 1 XS  1 + ' " times the (s  l)th row of blocks to the (s  2)th row, and so on. This gives
. . + = + as )
0" 5 5 l e X)
= x ns + (tr a 1 )Xns  1 + . . . + det as '
1 5 ... 6 *
with the second equality following from an obvious calculation. This yields
char. pol. Y/K cP =
assertion (ii), since by § I we have
6 6 ... 0 *
tr a 1 = TL/K a 1 ' det as = NL/K as ' 6 6 ' " 1 *
To prove (iii), we first write V = '
I � where each � is an Lsubspace of = det l e x) = N L/K f(x) .
V such that cp( �) c � . Then
char. pol. Y/K cP = IT char. pol. Y; /K cP, This completes the proof of (iii), and establishes the theorem.
i
and likewise with K replaced by L. In view of assertion (i), it therefore suffices Any semisimple K algebra A may be written as a direct sum I ' Ai of simple
to establish (9. 1 2) for each � . Changing notation, we may hereafter assume algebras Ai ' whose centers are finite extensions of K. In order to define
that V is an indecomposable K[ cpJmodule. Thus we may write red. char. POLA/K a for each a E A, we shall first treat the case in which A is
a simple Kalgebra ; the extension to the semisimple case will be straight
V = L [ X] /( f (X» for some f(X) E L[X] , forward.
with cp acting on V as left multiplication by X. Then
(9. 1 3) Definition. Let B denote a central simple Lalgebra with (B : L) m2 , =
t
char. pol. B/K b = NL/K (char. pol. B/L b) by (9. 12) (9.20) red. char. pol. A / K a = TI red. char. pol. A;/K Q i
i= 1
= N L/K {(red. char. pol.B/L bt } by (9.5) t
= { NL/K (red. char. pol. B/L b)} m by (9. 1 0)
= TI NK;/K (red. char. pol.A dKi a J
i= 1
= {red. char. pol. B/K b} m. As usual, we define tr A/K and nrA/K by the equation
This proves (9. 16) ; since (9. 1 7) is an obvious consequence of (9. 1 6), the (9.21) red. char. pol. A/K a = xr  (trA/K a)X r  l + . . . + (  l y nrA/K a
theorem is established. for each a E A, where r = L mi (K i : K). For future use, we record the following
obvious consequences of these definitions :
(9. 1 8) COROLLARY. The analogues of formulas (9.8) hold for trB/K and nr B/K ' t t
(9.22) trA/K a L trA ;/ K a i = L TK ;/K (trA ;/K i ai)'
=
Proof. The assertion is clear from (9.8) and (9. 1 5). We remark merely that i= 1 i= 1
t t
a E K, b E B. (9.23) nrA /K a = TI nr A ;/K a i TINK ;/K (nrA ;/K i aJ
=
i= 1 i =1
We are now ready to generalize Theorem 9.9, as follows : From (9. 1 6), we also obtain
t t
char. pol. AIK a = TI char. pol. A dK ai = TI {red. char. Pol. Ai/K aJ m i,
(9. 1 9) THEOR EM . Let B be a simple Kalgebra whose center L is separable over
(9.24)
i= 1 i=1
K. Then the map tr B/K , gives rise to a symmetric associative nondegenerate t t
bilinear form from B x B to K. (9.25) TA lK a = L m i trA d K a i ' NAI K a = TI {nr A ; /K aJ m i.
i= 1 i= 1
Proof. The form is given by r(a, b) = trB/K (ab), a, b E B, and we need only Of course, the analogues of formulas (9.8) remain true for trA IK and nrAI K '
check that r is nondegenerate. If r(a, b) = 0 for all b E B, then From (9. 1 9) we have at once
TL/K (a . tr B/L (ab)) = TL/K (tr B/ L (a ' ab)) = 0 for all a E L, b E B. (9.26) THEOR EM . Let A be a separable Kalgebra. Then the map trA 1 K gives
rise to a symmetric associative nondegenerate bilinear trace form from
But TL/K gives a nondegenerate trace form, since L is separable over K, and A x A to K.
1 22 ORDERS (9.27) (9.3 1) REDUCED NORMS AND TRACES 1 23
Finally, we show that reduced characteristic polynomials, norms and (9.3 1) THEOREM. Let A be a separable Kalgebra. Then
traces, are not affected by change of ground field. We prove char. pol. K aL = char. po1. K aR
for each a E A.
.
(9.27) THEOREM. Let A be a separable Kalgebra, and let F be any field contain
ing K (not necessarily finite dimensional over K). View A as embedded in
the separable F algebra F ® A, where ® means ® K ' Then for all a E A, n
Proof. Let A = I KXi ' and let {y 1 ' . . . , Y } be a dual basis of A relative
n
=1 i
(9.28) red. char. POI . F ® A/F a = red. char. pol. A/K a,
.
to the reduced trace form, so
(9.29)
n
A = I Kyj , 1 � i, j � n,
Proof. By Exercise 7. 1 3, F ® A is a separable Falgebra. We have j= 1
(9.30) char. po1.F ® A/F a = char. pol. A/K a where tr is an abbreviation for trA/K . We shall compute char. pol.K aL by
by Exercise 1 .2, a fact which we shall use presently. It suffices to prove the letting aL act on the Kbasis { Xi }' and we shall compute char. pol. K aR by
theorem for the case where A is a central simple Lalgebra, with L a finite letting aR act on the Kbasis {yj }. For 1 � j � n, let
separable extension of K. By Exercise 7. 14, we may write aL(x) = aXj = I !X ij Xi ' aR(y) = YP = I Pij Yi '
i i
d
. tr YPXi .
a b 1 + . . . + bd , bj E Bj . Then (9.24) YIelds
= Cl ij = tr y iaxj , Pij =
d
Thus the matrices (!X ij ) and (Pij ) are transposes of one another, and hence
char. pol.F ® A/F a = Tl {red. char. pol. Bj/F bj} m have the same characteristic polynomial. This completes the proof of the
j= l
theorem.
= {red. char. pol.F ®A/F a}m.
no inseparable extensions). Using (9.5), prove directly that red. char. pol.A/ K a E K[ X] a = ( exi) E Mr(D). Show that
for each a E A. [Hint : Show that if E is an extension field of K, and f(X) E E[X] is r
�
=
j= 1
fl : D + E ®K D Ms(E).
Let nj : E ®K A + Bj . Prove that for every a E A, Then fl' : A + E ®K A � Mrs(E) is given by
d ' (fl (exi) ) 1
fl (a) =
(9.33) red. char. pol. A /K a =
j
I1 char. pol. q>jn/l ® a). � ".j � r '
=1 and
r
[Hint : Let f(X) denote the expression on the right hand side of (9.33). As in the proof trA/K a = trace of fl'(a) = � trace of fl( exu )
i= 1
of (9.3), the isomorphism q>j is unique up to an inner automorphism of Mn/E), and
hence f (X) does not depend on the choice of the maps { q>j }. The proof of (9.3) carries
r
=
L trD/K (exJ ' ]
over to this case, and shows that f(X) does not depend on the choice of E, and that i= 1
f(X) E K[X].
To prove (9.33), it suffices to treat the case where A is a central simple Lalgebra,
with LjK separable. Choose E => L so that 10. EXISTENCE OF MAXIMAL ORDERS ; DISCRIMINANTS
. Throughout this section, A is a separable Kalgebra, where K is the quotient
field of a noetherian integrally closed domain R. Let A be an Rorder in A.
and so that
In dealing with reduced characteristic polynomials and reduced norms and
�
n
traces, we shall omit the subscripts AjK when there is no danger of confusion.
E ®K L L ' E i ' where (L : K) n,
i=1
=
A = I · Kuj ,
i= I
In brief, (10. 1 ) asserts that integral elements have integral reduced norms
and reduced traces. We shall use this fact repeatedly. Let us define the Then a i= 0 since A is separable over K. We claim that
discriminant of A to be the ideal d(A) of R generated by the set of elements
{det (tr x I.xJ. ) I "" I, J�
� . .
m
}, XI' E A, i= I
where m = (A : K). Each x.x
•
I J. E A, so the m x m matrix (tr x.x
I J. ) has entries Let X E r, and write X =
I I , r.I E K. Then
.i..J r.u.
"
m R, and thus d(A) c R. Furthermore, if X l , . . . , X are chosen so as to be m
linearly independent over K, then since the bil�ar reduced trace form tr xUj = I rj ' tr uiuj ' 1 � j � m.
A x A + K is nondegenerate, it follows at once that det(tr x.x i= I
I J.) i= O. Thus
d(A) is a nonzero ideal of R. Since each xuJ. E r, we have tr xuj E R. If we use Cramer's Rule to solve the
above system of equations for the {rJ, we thus obtain
( 1 0.2) THEOREM. Let A have a free Rbasis U I ' • . . , um ' Then the discriminant
d(A) is the principal ideal R . det (tr uiuj ). ri = (element of R)/ a, 1 � i � m.
Proof. This principal ideal surely lies in d(A). On the other hand, let X ' This proves that r c a  I. I Rui , and hence that r is an RIattice.
I
. . . , Xm E A . We may write Xi I rik uk , rik E R, and then
=
(10.4) Corollary . Every Rorder in A is contained in a maximal Rorder
tr xixj = I r ik rj1 ' tr ukul ' 1 � i, j � m. in A. There exists at least one maxi m al Rorder in A.
k, l
Therefore Proof. Let A be an R order in A, and let C be the collection of Rorders in
det(tr XiX) = det ( rik ) det (tr UkU1 ) · det (rjl )',
•
A containing A. Then C is nonempty. If {Ao J is a chain of orders containing
where ' denotes transpose. But det ( rik ) E R, and thus we deduce that A, let
d(A) c R det (tr UjUj), which completes the proof.
'
ex ex
If P is any prime ideal of R, and Rp is the localization of R at P, then A. Each X E A' lies
Then A' is a subring of A containing R, and K · A' =
and suppose that each a E r is integral over R. Then r is an Rorder in A. We remark that the crucial step in the proof of (10.3) is the fact that (tr uiu)
Conversely, every Rorder in A has these properties. is a nonsingular matrix, or equivalently that trAIK gives rise to a nondegenerate
bilinear form A x A + K. If char K 0, it is clear from (9.25) that the
=
Proof. We already know that every Rorder has the indicated properties. Now ordinary trace TAlK already gives a nondegenerate form, and so the proof
let r be a subring satisfying the given conditions ; we need only check that of (10.3) could have been given using ordinary traces instead of reduced
1 28 ORDERS (1 0 .5) (l 0.6) EXISTENCE OF MAXIMAL ORDERS ; DISCRIMINANTS 129
traces. Nevertheless, even when char K 0, the reduced trace and reduced
=
Proof. Suppose that A is a separable K algebra until further notice, and let
norm will play a basic role in our later discussions. A be a maximal Rorder in � . Then
(Without the hypothesis that A be a separable Kalgebra, it may happen A = A(e 1 + . . . + et ) c Ae1 EB ' " EB Aet ,
that there are no maximal Rorders in A, as already pointed out in §8. The and each Ae. is an Rorder in A . . If Ae. e r e A I. , where r. is a maximal
following example, due to Faddeev [ 1 , & 2 5J (see Roggenkamp and Huber Rotder, the� I " � is an Rorde ; in A � ontai ning A. Theref�re A = I " � ,
Dyson [1, pp. 201202]) shows how to construct an infinite strictly increasing and so Ae. = r. for each i.
chain of orders, in case A is not semisimple. Conver;ely, l�t Ai c Ai be a maximaI Rorder, I � i � t, and let A = I · Ai '
Let A be any Rorder in the Kalgebra A, and assume that rad A =1= O. Set If A e r e A with r maximal, then A . e re. , whence A. = r eIo for each i.
Lk = A n (rad A)\ k � L Therefore r = I· rei = A, proving th;t A is � maximal o�der.
We claim next that R.I is an RIattice, and is also a noetherian integrally
closed domain. Indeed, Rj is an RIattice by (10.3), since Kj is a separable
Then each Lk is an Rpure Asubmodule of A (see (4.0» , and is a full RIattice
in (rad At Since rad A is nilpotent, there exists an integer n � 1 such that
(rad A)n =1= 0, (rad At + I = O. Let r be a nonunit in R, and set Kalgebra. Thus Rj is noetherian. Further, if a E Kj is integral over Rp then
Ak = A + r k L I + r 2 k L 2 + . . + r nk Ln , k � 1 .
R.[a] is a finitely generated Rjmodule, hence is also finitely generated over
  . 
R: But then it follows by ( 1 . 1 0) that every element of RJa] is integral over R.
Then each A is an Rorder in A , and A = A o C A l C A 2 C . . . . We show Thus a is integral over R, and so a E Rj . This completes the proof that Ri is
k integrally closed.
k that Ak = A k + 1 for some k.
that the chain is strictly increasing. For suppose
Multiplying by ,n(k + 1 ), it follows that Ln c r A. Therefore Now let Ai be a maximal Rorder in Ai ' Since the elements of Rj commute
with those of A.I , and both R. and A.I are RIattices, also R . . A. is an RIattice.
Ln c �A n (rad A)n = �Ln ' Thus R . · A. is an Rorder in A . , whence A. = R . . A. , whi ch shows that
I
the equality following from the fact that Ln is Rpure in A. This gives A. is ne�ess �rily an R.order in A : . But any R.I � orde; in A. is also an Rorder
Ln = �L a contradiction since L is an RIattice and r is not a unit of R. id A . , since R. is finitely generated over R. Hence A.l � ust be a maximal
In parti�ular, the above shows that A is always strictly contained in the R .o ;der in A l� . Conversely, the same remarks show that every maximal
larger Rorder Al ' and thus A has no maximal Rorders whatsoever.) R I�order in A.I is also a maximal Rorder in A . . This completes the proof of
I
(i)(iii).
Let us now return to the case where A is a separable Kalgebra. We To prove (iv), let us now assume that R is a complete discrete valuation
shall now prove that a decomposition of A into simple components A = ring, and that A is a semisimple Kalgebra. The hypothesis of separability
A l ffi . . ffi At yields a corresponding decomposition of maximal orders and
hereditary orders in A. We may write 1 = e 1 + . . . + et , ej E Aj ; the {ej } are
.
was used to establish that each R.I is an RIattice, and is a noetherian integrally
then central idempotents of A such that ejej ::::: 0, i =1= j, and Ai Aej for each i.
closed domain. But these facts hold even without this hypothesis, by § 5,
=
since we are assuming that R is complete. Indeed, each Rj is also a complete
discrete valuation ring, and is finitely generated over R by Theorem 5.6.
(10 . 5) THEOREM. Let A be a separable Kalgebra with simple components The preceding proofs of (i)(iii) thus hold equally well in the present case,
{A.}, and let R.l be the integral closure of R in the center Ki of Ai ' Then and the proof of the theorem is complete.
(i) For each maximal Rorder 7\ in A, we have A = I " Aep where the {eJ
are the central idempotents oj A such that Aj = Aej . Further, each A ej is a The preceding theorem shows that the study of maximal orders always
maximal R order in A . . .
reduces to the case of simple algebras, and even to the central simple case
(ii) If Aj is a max/mal Rorder in Aj ' 1 � i � t, then I " Ai is a maximal if the ground ring is a Dedekind domain. As we shall see eventually, when
R order in A. R is a Dedekind domain, every maximal Rorder is hereditary. We shall not
(iii) An Rorder Aj in Aj is a maximal Rorder if and only if Aj is a maximal use this result in the discussion below, however.
R(order. Let A be any Rorder in A ; by a left Alattice we mean a left Amodule M
(iv) When R is a complete discrete valuation ring, all the preceding assertions which is an RIattice. In analogy with the results in §4a, we prove
remain true under the weaker hypothesis that A is a semisimple Kalgebra,
not necessarily K separable. (10.6) THEOREM. Let A be an Rorder in a semis imp Ie Kalgebra A. Then
1 30 ORDERS ( 1 0.7) (1 0.9) LOCALIZATION OF ,ORDERS 131
every left Alattice M is Aisomorphic to a submodule of a free module A(k) by the discussion in § 2f, the ring A is left hereditary if and only if A is right
for some k. hereditary. We may remark that hereditary orders enjoy yet another pro
perty of maximal orders ; in (40.7) we shall prove the following result of
Proof. We may embed M in KM ( = K ® R M), and KM is a finitely generated Harada [2] :
left Amodule. Hence there is an Aexact sequence
A (k ) � KM � O (10.9) THEOREM Every hereditary Rorder Aj in Ai is also an Rjorder, where
.
result gives at once a = (1 + i + j + k)/2, and find d(Ao)' Prove that Ao is a maximal Zorder in A, where
A = QA.
(1 0.7) COROLLARY. The Rorder A is left hereditary if and only if every left 3. Let A c A' be a pair of Rorders in A. Show that d(A') l d(A), and that deAf) =
Alattice is projective. d(A) if and only if A' A . [Hint : First localize so as to reduce the problem to the case
where R is a principal ideal domain.]
=
( 1 0.8) THEOREM. IfA is a left hereditary Rorder in the semisimple Kalgebra A, 4. Let A be any Rorder in A. Deduce from Exercise 3 that A is contained in a
with central idempotents {eJ as in ( l 0.5), then A = I' Aei , and each Aei is a maximal Rorder in A.
left hereditary Rorder in Ai ' Conversely, the direct sum of left hereditary 5. Let r be a subring of A containing R, such that r is finitely generated as R
orders is left hereditary. module. Prove that r lies in a maximal Rorder in A. [Hint : Use Exercise 8. 1 .]
6. Let A be an Rorder in a separable Kalgebra A, and let S be a domain containing
Proof. Let A be a left hereditary order. There is an exact sequence of left R, with quotient field L. Then S ® R A is an Sorder in L ® K A. Prove the discriminant
Amodules
o � A' � A � Aei � 0,
formula :
It is easily seen that A' is also left hereditary. Now Aei is a left Alattice, where the last equality is an identification.
hence is Aprojective by (1 0.7), and so the above sequence splits, that is, 7. Let A be an Rorder in A, and let L be a left Alattice in A, M a right Alattice in
A' is a direct summand of A : A, such that K . L = K . M = A. Prove that for each a E A,
Multiplying by ei , we find that J = Aej . An induction argument then proves [Hint : By (4.20) or Exercise 4.4, it suffices to prove the result when R is a discrete
that A = I' Aej . valuation ring. In this case, L and M have free Rbases, which are also Kbases for
Suppose co�versely that Ai is a left hereditary Rorder in Aj' I � ! � t, A. If L �. RXj and we write ax j = � (Xj j xj > (Xi j E R, then ord� L/aL R . det((Xj)
by Exercise 4.2. But det(iXjj ) = NA tK a by § 1 . For the correspondmg statement about
= =
and let A = I Aj . Each left Alattice M then decomposes as M = I e jM, M, use (9.32).] ,
and eiM is a left Arlattice. Thus each e jM is Arprojective, and so M is
A projective.
11. LOCALIZATION OF ORDERS
It should be pointed out that an Rorder A is left and right noetherian, so Here we investigate the relation between an R..,order A in A and its
(1 1 .3) LOCALIZATION OF ORDERS l 33
l 32 ORDERS (1 1 . 1)
localizations Ap and completions Ap , where P ranges over the prime ideals The following result shows that in order to decide whether A is maximal,
of R. We keep the hypotheses of section 10. it suffices to consider the ring structure of the localizations of A at the minimal
primes of R, together with the structure of A as an Rmodule. We recall
( 1 1 . 1) THEOREM. Let S be a multiplicatively closed subset ofR. F or each maximal (see page 55) that A is a reflexive Rmodule if A A * * , where =
Proof. By ( 1 . 1 5e) and (3. 1 1 ), we know that S  1 R i s noetherian and integrally We saw in (4.25) that for any noetherian integrally closed domain R, we have
closed, since R has these properties by hypothesis. Let i: A � S  1 A be the ( 1 1 .3) A**
n Ap ,
homomorphism x � 1 ® x, X E A, defined as in (3. 1 0). p
=
Now S  1 A is an S  1 Rorder in A ; let r be any larger order in A, and choose where P ranges over the minimal left ideals of R. Using this, we prove
a nonzero a E R such that a ' r c: S lA. Then Ll = i  1 (ar) is an Ssaturated
submodule of A such that ar = S  1 Ll. Since i is a twosided Ahomomor ( 1 1.4) THEOREM. (AuslanderGoldman [1] ). A n Rorder A is maximal if and
phism, Ll is a twosided Amodule. Then the right order 0 (Ll) coincides only if A is a reflexive Rmodule such that for each minimal prime P of R,
with A, since 0r(Ll) => A and A is maximal. Hence by (8.4) we h ave Ap is a maximal Rporder.
S 1A = O r(S  l Ll) = 0r(ar) = r, Proof. Obviously A c: A ** since A c: Ap for each P. By (4.26), A ** is a full
which completes the proof. RIattice in A. Furthermore, A ** is a ring by ( 1 1 .3), since it is an intersection
of rings. Thus A ** is an Rorder in A containing A. If A is a maximal Rorder,
( 1 1 .2) CoROLLARY. A
n Rorder A in A is maximal if and only if for each then we have A ** A, and hence A is reflexive. The fact that each Ap is
=
maximal ideal P of R, Ap is a maximal Rporder in A . maximal has already been shown in (1 1 . 1).
Conversely, let A be reflexive, and let each Ap be maximal. Then
Proof. If A is maximal, so is each Ap by (1 1 . 1). Conversely, assume that each A = A** = n Ap , P minimal.
Ap is maximal, and let A c: r, where r is any Rorder in A. Then Ap c: rp , p
so that Ap rp for all P. But then (r/A)p 0 for all P, whence r A by
= = = If A c: r, with r an Rorder in A, then Ap c: rp for each P, whence Ap = rp
(3. 1 5). for each P. Thence
r c: n rp n Ap A,
The following material, up to ( 1 1 . 5), can be skipped by the reader interested
= =
p
only in the case where R is a Dedekind domain. The above result shows that so A is maximal. This completes the proof.
the question of whether an Rorder A is maximal is a "local" one, and can be
decided by knowing whether the localizations Ap are maximal orders, where
P ranges over the maximal ideals of R. Now if pI c: P are a pair of prime In applying ( 1 1 .2) or ( 1 1 .4), we are faced with the problem of deciding
ideals of R, we have an inclusion Ap c: Ap' , and indeed whether the Rporder Ap in A is maximal. The underlying ring Rp is an
integrally closed local noetherian domain, with unique maximal ideal
Ap' � (Rp)p' . Rp ® Rp Ap , p . Rp . For such a domain, we now show that we may pass to the completion,
that is, Ap' is a localization of Ap at the prime ideal pl . Rp of Rp . Hence if and then consider the problem in the complete case.
Ap is a maximal Rporder, then by ( 1 1 . 1 ), Ap' is necessarily a maximal
Rp,order. Thus the hypothesis that Ap be a maximal order for each maximal (1 1 .5) THEOREM. Let R be an integrally closed local noetherian domain, R its
ideal P of R is a rather strong one, since it implies that Ap' is a maximal completion, and K the quotient field of R. Let A be an Rorder in A, and set
Rp, order for every prime ideal pI of R. The weakest hypothesis would be A = R ® R A,
the assumption that Ap' is a maximal order for every minimal prime pI of
R. (Recall that a minimal prime of R is any minimal member in the set of so A is an Rorder in A. Then A is a maximal Rorder in A if and only if A is
nonzero prime ideals of R .) a maximal Rorder in A.
1 34 ORDERS
We are going to show that D contains a unique maximal Rorder Ll. The
argument will depend strongly on the use of Hensel's Lemma (5.7). Let v be
the exponential Padic valuation defined on K (see § 5b, where v was denoted
1
by vK ). Then it is easily seen that for a, b E K we have
( 1 2. 1 )
(i)
(ii)
v(a) = 00 if and only if a = °
v(ab) = v(a) + v(b) = v(ba)
(iii) v(a + b) � min (v(a), v(b))
(iv) The value group of v is infinite cyclic,
where the value group of v is defined to be {v(a) : a =F O}.
Proof. Let t = min {V(ClJ ° � i � n}. If the theorem is false, then t <
min (V(Clo )' V(Cl )). Let r be the largest subscript for which V(ClJ = t. Then
n
r =1= 0, r =1= n, and
C(; 1 . f(X) = poxn + ... + p,xn ' + . . . + Pn , Pi E R,
where f3, = 1 and f3,+ l ' . . . , P E P. However, in R[X] the right hand expres
n
sion is a product xn  ' ( 1 + Pr  1 X + . . . + PoX') of two relatively prime
1 35
1 36 MAXIMAL ORDERS IN SKEWFIELDS (LOCAL CASE) ( 1 2.3) ( 1 2.7) UNIQUENESS OF MAXIMAL O RDERS 137
polynomials, with the first factor monic. It follows by Hensel's Lemma that v satisfies (ii). For a E K, we have w(a) = v(a) by (1 2.4). Next, the value group
a; 1 . f(X) is reducible, whence so is f(X), a contradiction. This completes of w, {w(a) : a E D, a =1= O}, is a nonzero additive subgroup of m  1 Z, and is
the proof. therefore isomorphic to Z.
It remains for us to show that w satisfies (iii), and for this it suffices to con
Now let ND/K and TD/K be the norm and trace maps, defined by sider the case where a = 1 and w(b) � 0, with b E D. But then b is integral
over R by (1 2.5), so by ( 1 . 1 1) also 1 + b is integral over R. Hence
char. pol. D/K a = xm  (TD/ K a)Xm  1 + . . + (  I t ND/ K a,
. a E A,
w(1 + b) � 0 = min (w( I ), w(b)),
where for the rest of this section we put (D : K) = m. We now set
as desired. This completes the proof.
(1 2.3) w(a) = m  1 ' v(ND/K a), a E D.
Let us set
(1 2.4) THEOREM. For a E D, let f(X) = min. pol. K a. Then
l l (1 2.7) 1\ = {a E D : w(a) � O} = {a E D : ND/K a E R} .
w(a) = (K(a) : K) . v(f(O)) = (K(a) : K)  . v(NK(a)/K a).
Then 1\ i s a ring containing R, b y ( 1 2 6) . W e call 1\ the valuation ring o f w.
.
Proof. Since D is a skewfield, f(X) is irreducible, and therefore char. pol. D/K a = We shall show in ( 1 3.3) that 1\ is finitely generated as Rmodule. Taking this
f(x)m/n , where n = degree of f(X) = (K(a) : K). Therefore for granted, we have
ND/K a = (  It · {f(o)}m/n,
whence ( 12.8 ) Theorem. 1\ is the unique maximal Rorder in D, and is the integral
. v( {j{o)}m/n) = n  1 . v(f(O)), closure of R in D.
w(a) = m 1
which completes the proof. Proof Clearly K 1\ = D, and so 1\ is an Rorder in D. But by (8.6), every
.
properties (12. 1 ).
vD(a) = (elm) . v(ND/ Ka)
Proof. Let u : D + be another such extension of v. If u i= w, there
R u { oo} (13.1)
= e ( K (a) : K )  l v(f(O)), where f(X) = min. pol. K a.
exists a nonzero a E D such that u(a) i= w( a ). Replacing a by a  l if need be,
'
f(X) = min. pol. K a = xn + alX n  l + . . . + an E K[X] . vD(nD) = 1 ; call nD a prime element of �. We have at once
=
Then w(a) = v(an)ln by (12.4), while v(a) � (jln) v(an) by (12.9), since f(X)
. is vD(n) e.
irreducible. Therefore The group of units of � is given by
· .
u(aJ. an  J ) = v(aJ.) + (n  j) u(a) >
j n j
n ' v(a ) + 
n
n
' v(an) = v(an), u(�) = {a E � : vD(a) = O} = {a E � : ND/ K a E u(R)}.
( 1 3.2) THEOREM. Let nD be a prime element of �, and set p = nD� ' Then
This gives a contradiction, so the theorem is proved. every nonzero onesided ideal of � is a twosided ideal, and is a power of p. The
residue class ring �/p is a skewfield over the field R, and p n R = P.
The preceding discussion is of course valid for the special case where D
is an extension field of the complete Padic field K. It shows that the Padic Proof. For L any nonzero left ideal in �, let
I
valuation on K can be extended to a discrete valuation on D, by means of
= min {v D(a) : a E L},
formula (1 2.3), and that this extension is unique (up to equivalence).
and choose an element x E L such that vD(x) = I. Then for each y E L we have
yx  1 E �, so Y E �x
c L. This proves that L = �x. Furthermore, �x = x�
13. RAMIFICATION INDEX, INERTIAL DEGREE
since
Keeping the notation of § 1 2, let D be a skewfield of finite dimension over a
vD(x  1 ax) = vD(a), a E D.
complete Padic field K contained in the center of D, and let � be the unique
maximal Rorder in D. We shall define the ramification index e = e(DIK ) Thus L is a twosided ideal of �, and clearly
and inertial degree f = f(DIK), and shall show that the analogue of Theorem
L = nb� = �nb p i , since n;; 1 E �. X '
5.6 holds true in this case. In fact, the discussion below does not make use of
=
any of the earlier results from §5, and so provides a proof of Theorem 5.6, by The above shows that p is a maximal left ideal of �: The ring K = �/p
choosing D to be a field. therefore has no left ideals except 0 and K, and thus K is a skewfield. Finally,
Let v be the exponential valuation on K, and let w be the extension of v to p n R is a prime ideal of R, and is nonzero since p is a full Rlattice in D.
the skewfield D. We have seen that the value group of v is Z, while the value Thus p n R = P, whence P . K 0, so K is a skewfield over R.
=
1 40 MAXIMAL ORDERS IN SKEWFIELDS (LOCAL CASE) 141
(1 3 . 3) ( 1 3.5) RAMIFICATION INDEX, INERTIAL DEGREE
We now define the inertial degree of D over K as Passing t o the residue class ring K, we obtain the relation I Y l iai = �,
f = f(D/K) = (� : R). with Y 1 1 =f. 0, which contradicts the linear independence of the {aJ over R.
We have therefore established that the s · t products {aib ;} are elements of D
Note that � = IJ./p is a vector space over R, since P c p. The fact that f is linearly independent over K, so st � rn. This proves that f is finite, and shows
finite follows from the theorem below, which generalizes Theorem 5.6, and also that ef � m.
indeed provides an independent proof thereof. N ow let at ' . . . , af ' bo ' . . . , b e _ 1 E IJ. be elements satisfying the hypotheses
of the theorem. The e ' f elements { a ib) are linearly independent over K,
(1 3.3) THEOREM. Let D be a skewfield of finite dimension m over a compLete by the first part of the proof, and we need only show that IJ. = � R aibj ' Let
Padic field K contained in the center of D. Let e = e(D/K) be the ramification I, )
index of D over K, and f = f(D/K) the inertiaL degree of D over K. Both e and X E IJ., x =f. 0, and let vD(x) = ke + j, where ° �j � e  1 . Then
f are finite, and x = (n kb)u, u E u(IJ.).
e . f = (D : K) = m. Since the {aJ form an Rbasis for 'X, we may find elements {r J in R such that
Let a l ' . . . , af E IJ. be such that � = I:,0 Rai ' and let bo ' . . . , be  1 E IJ. be such . u = r al + . . . + rfaf + Z, with Z E p. Hence we have
I
that vD(bj ) j, ° � j � e  1 . Then the e ·f products {ai bj } form a free
=
�roof. F�rst, let { a i : 1 � i � s} be elements in IJ. whose images {aJ in If x I 0, we may repeat this procedure with Xl ' and continue if need be with
=f.
� are
lInearly mdependent over R, and let { b .: 1 � j � t} be elements of K such x2 ' X3 ' Note that k is the greatest integer in vD (x)/e. Hence after n ' e
• • . •
that the integers {vD(bj)} are incongr u�nt mod e. We claim that the s · t steps, we obtain an equality
products {a ibj} are linearly independent over K. If not, there is a relation f
et .
( 1 3.5) x I I (rU ) nk + dP nk + I + ' " + r�j) nkt "  l ) aibj + Y" ,
( 1 3.4) Q:: Cl 1 i aJb 1 + . . . + CL Clt iaJbt = 0, Clji E K, =
j=o i= l
where we may assume that for each b). , at least one of the coefficients where vD( Y ) � vD(x) + ne.
.
Clj 1 ,
, Cljs IS nonzero. Write
• . .
For fixed" i, j, the sequence
nj = . min { V(CljJ : 1 � i � s}, Clji = nni . f3ji ' r��
I)
)n\ r�I)� )n k + r��)n
I)
k+ \ . . .
Then each f3ji E R, and for each j there is an index k for which f3jk rI P. We is a Cauchy sequence from R, relative to the Padic valuation of R. Since R
may rewrite ( 1 3 .4) as is assumed complete, this sequence has a limit Sij E R. If we set x' =
r��J
I)
E K, n � 1.
14. FINITE RESIDUE CLASS FIELD CASE
Set
vD(xn  xn  l ) = e ' kn + ln '
=
Throughout this section we assume that R is a finite field with q elements.
2
Let D be a skewfield with center K, and let (D : K) n . We call n the index
=
so also lim kn
n + if
= 00. We have then of D. Denote by v the valuation vK on K, and by vD that on D, as in ( 1 3 . 1 ).
Xn  xn  l nkn . Yn ' Yn E d . Let d be the unique maximal Rorder in D. We shall see here that the struc
tures of D and d can be described explicitly in this case, and depend only on
Write Yn as an Rlinear combination of the {aibj } ; then by ( 1 3.3), it follows the index n and some integer r such that 1 � r � n, (I', n) = 1 .
from the above equation that The discussion to follow depends heavily on the results listed in §5b, and
r��)  r�� l we shall use the notation introduced there. Since R is finite, we know from
 ) E nknR •
=
I) I)
(5. 1 0) and (5. 1 1 ) that for each positive integer f, there is a unique unramified
Hence {r�� ), r�J ), . . . } is a Cauchy sequence from K, with limit p . . , say. Then
I) extension W of K with (W: K) = f, given by W K(w), where w is a primi
obviously tive (qf  l )th root of 1 . If Ow denotes the valuation ring of W, and Ow the
residue class field of ow ' then
2 Ow = R[w], ( W: K) = (ow : R) = f
(13.7) THEOREM. Let K be the center of the skewfield D, and set (D : K) = n ,
e = e(D/K), f = f(D/K). Then Further, TV;'K is a galois extension with cyclic galois group of order f, generated
e l n, n lf by the Frobenius automorphism (J defined by w � wq• Likewise, 0w/R is a
galois extension, with galois group cyclic of order f, generated by the auto
Proof Let w be the extension of v to D defined in ( 12.3). By definition of e, morphism if which maps ill onto illq •
we have w(nD) = l/e. Then K(nD) is a subfield of D, and has ramification Before turning to the study of skewfields, we need one more preliminary
inde� e. Hence by (5.6) or ( 1 3.3), we have e l (K(nD) : K). But K(nD) lies in some result.
maxImal subfield E of D, whence (K(nD) : K) I (E : K). Since (E : K) = n by
(7. 1 5), the result follows. (14.1) THEOREM. Let W be an unramified extension of K of degree f, and let v
be the valuation on K. Given any element r:x E K, the equation
EXERCISES NW K X
'
= r:x,
1. Let Y be any full set of representatives in d of the residue classes in K, where we is solvable for x if and only if f divides v(r:x).
assume that 0 E Y. Prove that every nonzero X E D is uniquely expressible as
Proof The prime element n of R is also a prime element for Ow . Each nonzero
x = 7r�D()C) . (s o + Sl7rD + S27r� + . . . ), Si E Y , So f= O.
2. Prove that the order ideal ordR Li eq uals pI, where f
x E W is thus expressible as x = nkxo , Xo E u(ow ). Then vw(x o ) = 0, whence
f(D/ K).
[Hint : X � RU) as Rmodules.]
Nx o E u(R), where N denotes NW/ K ' But
=
n
Nx n fN(x o )'
3. Prove that v(ND{K 7rD) = f, and that v(nrD{K 7rD) = me.
=
generator if: w � wq, the distinct algebraic conjugates of w over R are the proof of the theorem.
{wq i : 0 � i � f  I} . Now we are ready to consider skewfields. The first important consequence
Each nonzero element of Ow is of the form wt, 1 � t � qf  1 , and (by of the hypothesis that R be finite is as follows :
Exercise 1 .4) we have
( 1 4.3) Theorem. If (D : K) =
n2, then
N(wt ) W t( 1 + q + ' " + qf  1) (jj t(qf  1 )/(q  1 ) .
= =
=
e(D/K) f(D/K) = n.
Thus N(wt)= 1 if and only if (q  1 ) I t. Hence exactly  l)/(q  1 ) (qf
elements of O w have norm 1 , s o there are q  1 distinct nonzero norms. This
Proof. � is a skewfield of finite dimension f over R, hence is a fi�e sk:,wfield.
proves that N is epic.
By Wedderburn's Theorem (7.24), � must be a field. Hence � =
R(a) for
Continuing with the proof, let a E u(R). Since N is epic, we may choose
some a E �, so by (1 3.3) or (5.6) we have
b o E Ow such that a == Nbo mod nR . For i � 1 , suppose that we have found
an element (K (a) : K) � (� : R) = f.
But K(a) is a subfie1d of D, whence (K(a) : K) � n by (7. 1 5). Therefore f � �,
such that = =
so f n by (1 3.7). Finally, ef n2 implies that e n, and the proof IS
=
�. Then we have
Replacing rt by n1a if need be, we may assume that rt E �. Now let vD(rt) = k, (n� )" = {3n{3  1 = n, n�w'n� l = {3 · w tqr . p 1 = (wl)qr,
and put h n/(k, n) . Then
=
8 E u(�), b E Z, so the choice of r does not depend on which root of unity is used. We must
still show that r does not depend on the choice of the prime element nD . Let
and h is the least positive integer for which such an expression for rth is nl be any prime element of �, and suppose that
possible. Then
w qh = rth • W · rt h 8 ' W · 8  1 == w (mod p),
=
nl w . nl  1 = wq S
•
148 MAXIMAL ORDERS IN SKEWFIELDS (LOCAL CASE) (14.6) (14.7 ) FINITE RESIDUE CLASS FIELD CASE 149
d.
for some s between 1 and n. We may write n' = ynD with y E u(8), and then a Ksubalgebra of M ( W). We shall verify that D is the desired skewfiel
D is expressi ble as a Klinea r combin ation of the n2
Each element a E
wqS ynD . w . (ynD )  l = y . wqr . y  l . ble
elements { (W*)i . (n; Y } . Since (W*)i = (wi)*, it follows that a is expressi
=
Passing to � we obtain the equality QjqS = Qjqr. Therefore s = r, since Qj is a in the form
primitive (qn  l )th root of unity. This completes the proof of the theorem . n l
a  I rxj* (nD* )j' rxJ. E W.
The above shows that once the complete field K is given, the skewfield D j=O
[ �'  1
is completely determined by its index n, and by the integer r. Indeed, we first Since
form the field K(w), with w any primitive (qn  l )th root of 1 . Then we pick
any prime n E R, and adjoin to the field W an element nD satisfying the
0
(n*D )1 =
conditions listed in ( 14.5). This determines the skewfield D = K(w, nD ) up n1J.
to Kisomorphism. We call the fraction r/n the Hasse invariantt of D. we obtain
We are still left with the problem of deciding whether each fraction r/n
rxo rx l rx 2 rx n  l
arises from some skewfield.
nf)(rxn _ l ) e(rx o ) e(rx l ) e(rxn )
2
(14.6) THEOREM. Let 1 � r � n, (r, n) = 1 . Given the complete field K, there nf) 2 ( rxn _ ) nf) 2 ( rx f) 2 ( rxo ) e 2 (rxn _ 3 )
(14.7) a= 2 n_ 1)
exists a skewfield D with center K, index n, and Hasse invariant r/n.
for some u E K, which shows that K is the center of D. Finally, the equation
=
for some integer m, and m � 0 since Li � /1. Call � the inverse different, and n 1
define the different as /1 = IO ow · (n D Y,
:D = :D(/1jR) = n� . /1. j= O
we have
F or a nonzero ideal L n1 /1 of /1, define the norm n 1
n  (n  1 ) /1 = n  1 . nD /1 I (n  1 ow ) . (nD Y ill Ow .
=
= °
1 . Let D be a skewfield whose center K is infinite. Let a E D  K, f(X) red. char. POL D/K (X char. PO I. W/K (x, (x E w.
min. pol. K a. Show that there are infinitely many elements a' E D such that f(a') = O.
=
=
o � i � f  1, wq is conjugate to w in D. tion, with a finite residue class field R (as in §14). Let
i
3. Let K be the 2adic completion of the rational field, R its ring of 2adic integers,
K EB K i E8 Kj E8 Kk (quaternions over K),
nr D = {nrD/K(d) : d E D}.
and VK the valuation on K. Let D
Prove that nr D = K. [Hint : Use the notation and results of the preceding exercise.
=
and set
+ + +
Then
A = R + Ri + Rj + Rk + Ra, a = (1 i j k)j2.
nr D :::> N W/K( W) = {x E K : n l vK(x)},
Show how to extend vK to a valuation on D, and prove that A is its valuation ring.
Find the discriminant of A, and compute e(DjK), f(DjK), a prime element 1lD ' the by (14. 1 ). But nr D also contains ( _ l)n  11l, whence nr D = K.]
residue class field K, an Rbasis for �, and the inertia fields of D over K.
4. Let K be the 3adic completion of Q. Show that there exists an element a E K
with a 2  2. Let <I>4(X) = X 4 + 1, the cyclotomic polynomial of order 4. Show that
=
5. Keeping the notation of (14.5), let W be an inertia field for D. Prove that
n rD/K(1lD) ( _ l)n  1 1l,
= nrD/K(cr:) = Nw/icr:) , cr: E W
[Hint : Since xn  1l is irreducible in K[X] by Exercise 5. 1, it follows that
min. pol. K 1lD = xn  1l. But then
red. char. pol. D/K 1l D = min. pol. K 1l D '
by Exercise 9.1. This gives the desired formula for nr(1lD). Next, let cr: E W and let
f(X) min. pol.K cr:. By Exercise 9.1,
=
K
red. char. PO l. D/K cr: {f(X)} (W : (a)) . =
But also
K
char. pol. W/K cr: = { f(X)} ( W: (a))
(1 5.1) PROGENERATORS 1 55
Let id.rJ be the identity functor on d. Two categories d and f!J are called
equivalent if there exist functors F, G :
such that GF '" idSli and F G idEM' For example, if A denotes the opposite
'" 0
4. Morita Equivalence ring of A (same elements, but multiplication is reversed), then the categories
AJIt and JItAO are equivalent.
We shall need to study the relationship between Amodules and Mn{A) N ow let F: d + r!J be any functor. Call F faithful if the map
modules, where as usual Mn(A) is the ring of all n x n matrices over the ring
(15.1) a E Hom (A, A') + Fa E Hom(FA, FA')
A. The generalization of this relationship leads to the concept of Morita
equivalence, and we sketch here some of the main results of this theory. As is injective, that is, if Fa = 0 implies that a = O. For example, if M is a
references, we cite Bass [1], P . M. Cohn [1], Faith [1]. While the general faithfully flat right Amodule, then the functor M ®A ' : AJIt + dlJ is faithful
theory holds for abelian categories, we shall make the simplifying assumption (see (2.2 1)).
that our categories are in fact categories of modu les. Denote by . 11,\ the On the other hand, the functor F : d + r!J is full if the map in (15 . 1 ) is
category of right Amodules, by /).JltA the category of (�, A)bimodules, and surjective, that is, if each (3 E Hom(FA, FA') equals some Fa . For example,
so on . let i\ =AI I where I is a two sided ideal of A. Let d be the category of pro
jective left A modules, f!J the corresponding category for A, and let F : d + r!J
be given by F(M) = M/IM. Then F is a full functor, since each Ahomomor
15. PROGENERATORS
phism {3 : M/IM + N/IN, where M, N E d, can be lifted to a Ahomo
We shall use the notation of § 2b. Let d be a category with objects A, A', . . ,
. morphism a : M + N, and thus (3 = Fa.
and f!J a category with objects B, B', . . We shall write A E d to indicate that
. . We recall from (2. 14) that an object X E d is projective if the functor
A is an object in d. All functors F : d + f!J considered below are assumed to
Hom(X, ' ) : d + dlJ
be additive. Unless otherwise stated, functors are taken to be covariant.
Let F : d + r!J and G : d + f!J be a pair of functors. A natural transforma
is exact. This functor carries A E d onto the abelian group Hom(X, A), and
tion t : F + G is a family
takes each a E Hom(A, A') onto
a* : Hom(X, A) + Hom(X, A'),
where for each A E d,
where a* f af, f E Hom(X, A).
=
tA E Hom.@ {FA, GA),
Whether or not X is projective, let us consider the functor Hom(X, . ). We
and where for each a E Hom(A, A'), the diagram call X a generator of the category d if Hom(X, ' ) is a faithful functor, that is,
if a* = 0 implies that a = O. In other words, X is a generator if and only if
FA � GA
Fal Gal for each nonzero a: A + A' there exists an f : X + A such that af =1= O.
Dually, X E d is injective if the contravariant functor Hom{ " X) is exact,
FA' � GA' while X is a cogenerator for d if Hom( " X) is faithful.
is commutative. If furthermore each tA is an isomorphism, we call the functors
( 15.2) THEOREM. An object X in the category d is a generator for d if and
F and G naturally equivalent, and write F '" G.
only if every A E d is a quotient of a direct sum of copies of x.
Example. Let d f!J AJIt, F the identity functor, G the functor A ®A
= = • .
Proof. Let X be a generator, and let A E d. For each f E Hom(X, A), let
For each A E d, let tA : A + A ®A A be the left Aisomorphism given in (2.8).
XJ be a copy of X, and set Y = I· XJ ' Each XJ maps into A (by means of f),
Then t is a natural equivalence from F to G. J
1 54
1 57
1 56 MORITA EQUIVALENCE ( 1 5.3) ( 1 5. 6) PROGENERATORS
so there is a map 11 : Y � A, defined by 11 = I"! Thus direct sum IO X, V There are the usual injection maps {i;,} and projection
maps {p;, } , with
11 ( I xj ) I f (xj ) , Xj E Xj " ;, .
X;. � I XV '
=
Clearly p;. V
(15.3) im fl = f(X) c A. We shall say that a functor F : d � 14 preserves direct sums if for each family
j eHom (X, A) { X;. }, there is an isomorphism
Let us show that im fI A ; if not, there exists a nonzero map
= F( I o Xv ) � IO FXv '
cr:: A � A jim fl.
Since X is a generator, there must be an f : X � A such that rtf =1= O. But induced by the maps {Fi;. , Fp;. }. For example, tensor product functors
(af)X rt(f(X)) 0, since f(X) c im 11. This gives a contradiction, and
= =
preserve direct sums. So also does the functor HomA(P, ' ), where P is any
shows that 11 is an epimorphism of Y onto A. Therefore A is a quotient of a finitely generated projective Amodule (see Exercise 15.3).
direct sum of copies of X, as claimed.
C ? nversely, suppose that for each A E .91 there is an epimorphism ( 15.6) THEOREM. Let F: .91 � f!4 and G: .91 � 14 be a pair of right exact functors
e : I Xi � A, where each Xi is a copy ofX. We must show that X is a generator which preserve direct sums. Let t: F � G be a natural transformation, and let
for d. Let rt : A � A' be nonzero ; then also tXe =1= 0, since e is epic. If we write X be a generator for d. If tx : FX � GX is an isomorphism in the category
e = LO ep where ei : X i � A, then tXe IO tXei. Therefore tXei =1= 0 for some
=
f!4, then t is a natural e q uivalence of functors.
i. But X:i � X, and so there exists an f : X � A such that tXf =1= O. This proves
that X IS a generator for .91, as desired. Proof. Since t {tA : A E d } is a natural transformation, each
= ct: A � A'
gives rise to a commutative diagram
The construction in (15.3) is of special importance in the following circum
stance. Let X E AvH, and define the trace ideal of X as FA � FA'
(15.4) trace X = f(X) c A.
tAl J
tA
GA � GA'.
that
It is easily verified that trace X is a twosided ideal of A. We are assuming that tx is an isomorphism, and are trying to prove
each tA is an isomorphism.
I X
°
(15.5) COROLLARY. The left Amodule X is a generator for AvH if and only Since X is a generator of .91, each A E .91 is a quotient of a direct sum
if trace X A. =
of copies of X. Hence there is an dexact sequenc e
Proof. If trace X A, then the proof of the preceding theorem shows that U � V � A � 0, where U = IO X, V = IO x.
=
A is a quotient module of IO Xi ' where each Xi � X. But every Amodule But F is right exact, and so the sequence
A is a quotient of a direct sum IO Aj ' with each Aj � A.. Hence A is also a
quotient of a direct sum of copies of X, whence X is a generator for AvH
FU � FV � FA � O
by ( 1 5.2). is 14exact. Hence there is a commutative diagram with exact rows :
Suppose conversely that X is a generator. The proof of (15.2) shows that FU  FV  FA  O
im 11 =A in (15.3), where A is any Amodule. Taking A AA, we see that = tu t tV t tA t
the trace ideal of X equals im 11, and thus coincides with A as claimed. G U  GV + GA  O.
dire �t su�s by
It is clear from (15.2), or directly from the definition of generator, that Now F U � IO FX, G U � I O GX, because F and G preserve
F U U . LIkewIse tv
A is itself a generator for the category AvH. Indeed, every nonzero free hypoth esis. Since tx : FX � GX, it follows that tu : � G
above diagram shows that
module is a generator. is an isomor phism, and the commutativity of the
Given a possibly infinite family {X;,} of objects in .91, we may form their tA is also an isomorphism, as desired.
1 58 MORITA EQUIVALENCE ( 1 5.7) ( 1 5.9) PROGENERATORS 1 59
An object P E .91 is faithfully projective if it satisfies the following three Now let A and A' be arbitrary, and let
conditions :
(i) P is a projective object in .91. U + V + A + O
(ii) P is a generator for .91. be an .91 exact sequence in which U and V are direct sums of copies of P.
(iii) Hom(P, ' ) preserves direct sums. Then the sequence
In particular, considering the category /\At, we call a left A  module P a
progenerator for /\Jt if FU + FV + FA + °
(i) P is a finitely generated projective Amodule, and is !!Aexact, since F is an exact functor. Since Hom is a left exact functor, we
(ii) P is a generator for /\Jt. obtain a commutative diagram with exact rows
Note each such P is faithfully projective, since by Exercise 1 5.3 Hom(P, ' )
necessarily preserves direct sums. ° + Hom (A, A')  Hom (V, A')  Hom (U, A')
t FI t F2 t F3
(15.7) THEOREM. Let P be a faithfully projective object in the category .91, and ° + Hom (FA, FA') + Hom (F V, FA') + Hom (F U, FA').
let
A = Homd (P, P), !!A = /\Jt. By the preceding paragraph, both F2 and F 3 are isomorphisms. Hence so
is F l ' and this completes the proof that F : .91 + !!A is a full and faithful
View P as right Amodule. Then the functor functor. (For another proof see Exercise 1 5.6.)
In order to deduce that F gives an equivalence of categories, it remains
Hom(P, . ) : .91 + !!A
for us to show that every left Amodule M is of the form FA for some A E d.
gives an equivalence of categories. Let us first write a Aexact sequence
x0 ' X 1 Afree.
Proof. It is clear that A is a ring, and that P can be viewed as a right Amodule.
Then for each A E .91, Homd (P, A) is a left Amodule. Thus the functor
F : .91 + !!A, given by F(A) Hom(P, A), is well defined. Clearly FP A.
=
Since A F(P), we may write Xi = F( Ui), with � a direct sum of copies of
=
=
7 . Let Pll. be a progenerator for vIt ll. ' and let G, H : vIt ll. � .916 be the covariant
EXERCISES
functors defined by
1. Let F : .91 � P.I be an equivalence of categories. Show that A is projective if and '
G(L) = Homll.(P, L), H(L ) = L ®A P*,
only if FA is projective.
where P* = Homll.(P, I1) (viewed as left l1module). Show that the functors G, H are
2. Let F : .91 � � be a full faithful functor such that every B E fA is the image FA of
some A E d. Prove that F is a category equivalence. naturally equivalent. [Hint : Prove
(i) P* is a finitely generated projective left l1module.
3. Let P be a finitely generated projective left Amodule. Show that the functor
(ii) The functors G and H are exact, and preserve direct sums.
HomA (P, . ) : A vIt � .916 (iii) There is a natural transformation t: H � G, given by
preserves direct sums. [Hint : First check the case where P = A, then use the fact that
HomA(P + p" . ) '" HomA(P, ' ) + HomA(P', ' ).] where tL is defined by the formula
4. Let AvIt � ll.vIt be an equivalence of categories, where A and 11 are rings. Show {tL(l ® cp) } x = i· cp(x), lE L , CP E P*, X E P.
that A is left hereditary if a�d only if 11 is left hereditary. [Hint : A is left hereditary if
(iv) tll. : H(M � G(M·
and only if subobjects of projective objects in A vIt are projective.]
Then use ( 1 5.6).]
5. Let .916° denote the category which is opposite to .916, that is, .916° has the same
objects as .916, but all arrows are reversed.
(i) Prove that direct products in .916 correspond to direct sums in .916 0 . 16. MORITA CORRESPONDENCE
(ii) Given an object A ' in a category .91 , define a functor G : .91 � .9I6 ° by setting
Let MA be a nonzero right Llmodule, and set
G(A) = { Homd (A, A') f E .9I6 � for each A E d.
(16.1) A = HomA (M, M), M* = HomA(M, Ll).
(There is an obvious definition of the action of the functor G on maps in d.) Prove
that G is a right exact covariant functor which preserves direct sums ! Let A act on the left on M, so there is a bimodule structure j\MA· By definition,
6. Use Theorem 1 5 .6 to prove that the map F in (1 5.8) is an isomorphism. [Hint : M* = HomA(j\MA, ALlA)'
Keep the notation of the proof of ( 1 5 . 7), so
the set of all right
Llhomomorp hisms from M to Ll. The left action of Ll on
F(A') = Homd (P, A') for each A' E d.
the bimodule ALlA enables us to make M* into a left Llmodule. The left
Keeping A' fixed, define functors G, H : .91 � .9IlJo by action of A on the bimodule A MA likewise makes M* a right Amodule
(see §2). It is easily verified that M* is a bimodule AM* A '
G(A) = { Homd (A, A')r, H(A) = { Hom�(F A, FA') t, A E .91,
We define a map
with obvious definitions of G and H on maps. Since P is a fai thfully projective object (16.2) m(fm')
J1 : M ® A M* + A, m ® f + (m, f), where (m, f)m' =
in .91, the functor F is exact and preserves direct sums. Use this to prove that H is a
right exact covariant functor which preserves direct sums. for m, m' E M, f E M*. Likewise, let
The functor F : .91 � � determines homomorphisms
(16.3) T: M* ®A M + Ll, f ® m' + [f, m'] , where [f, m'] = fin' ·
FA, A' : Homd (A, A') � Hom!i (F A, FA'), A, A' E d.
It is easily checked that J1 and T are well defined, J1 is a twosided Ahomo
For each A E d, define tA : HA � GA by tA = {FA ' A ' r, Prove that t : H � G is a morphism, and T a twosided Llhomomorphism. Further, one obtains
natural transformation .
In order to prove that the map F in ( 1 5.8) is an isomorphism, it suffices to establish (16.4) (m, f)m' = m [f, m'] , g(m, f) = [g, m] f,
that each tA is an isomorphism in .9IlJo. By ( 1 5.6), this will follow once we know that
tp is an isomorphism in .9IIJO. But this is true since
for m, m' E M, f, g E M*. Finally, we note that
HP = Hom� (FP, FA') = HomA(A, FA') � FA'.] where "trace" is defined as in (15.4).
162 MORITA EQUIVALENCE (1 6.6) (1 6.9) MORITA CORRESPONDENCE 163
(16.6) Definition. A Morita context consists of a pair of bimodules A Mt. ' generator for .,I(t. . Finally, J1 epic implies J1 monic, by Exercise 16.1, and
t.NA relative to rings �, A, and bimodule maps likewise for T. This completes the proof.
J1 : M ® t. N � A, T : N ® A M � �,
The same argument shows
given symbolically by J1(m ® n) = (m, n), T(n ® m) = [n, m], interrelated by
the formulas (16.8) COROLLARY. For any Morita context (16.6), we have
(m, n)m' m en, m'], n(m, n') En, m]n',
lA E im J1 � J1 epic � J1 monic,
= =
for all m, m' E M, n, n' E N. If both J1 and T are isomorphisms, we call the it. E im T � r epic ====9 r monic.
rings �, A Morita equivalent.
Recall that Mt. is a progenerator of .,1(.1. if M is a finitely generated projective
We have just seen how to construct a Morita context by starting with Mt. ' �module which is a generator of the category .,I(t. . The preceding results
and defining A, M* by (16. 1). We call it the Morita context derived from the give at once
right �module M . We now proceed to investigate this derived context
more fully, keeping the notation of ( 16. 1)(16.5). (1 6.9) Corollary. The module Mt. is a progenerator of At. if and only if both
of the maps J1, T in ( 1 6.7) are isomorphisms. If Mt. is a progenerator, then the
(16.7) THEOREM. Let M be a nonzero right �module, and let A = Homt.(M, M). Morita context derived from M gives a Morita equivalence between the
Let rings � and Homt.(M, M) .
J1 : M ® t. Homt.(M, �) � Homt.(M, M),
Remark. Let Mt. be a progenerator of .,I(t. . Then M is a projective object
T : Homt.(M, �) ® A M � �, in the category .,I(t. ' and by ( 15.9) there is an equivalence of categories
be the maps defined in (16.2) and (16.3). Then J1 is epic if and only if M is a ( 1 6. 1 0) Homt.(M ) : At. � .,I(A'
, ·
A t.; in this case, T is also monic. useful to exhibit this category equivalence in terms of tensor products,
Finally, J1 is epic if and only if lA E im J1, and T is epic if and only if it. E im T. rather than Hom. Define M* as in (16. 1 ) ; by Exercise 2.5, there is an iso
morphism
Proof The last statement in the theorem is obvious, since im J1 is a twosided
ideal in A, and im T a twosided ideal in �. and it is easily verified that this is a right Aisomorphism which is natural in
The identity map 1 M on M is the identity element lA . Hence if J1 is epic, L. In other words, there is an equivalence · of functors
we may write
(16.1 1)
h E Homt. (M, �). (This also follows from Exercise 1 5.7.) Hence the equivalence in ( 1 6. 10) can
i=1
also be given as
fJ
Define right �homomorphisms �(r) � M by
a(a l , · · · , ar ) = L miai , [3(m) = (JI m , . . . , f,.m). We shall show below that every Morita equivalence of rings �, A is
1 obtained in the manner described above, by means of the Morita context
It is easily seen that a[3 = 1 M , whence M I � (r), and thus Mt. is finitely generated derived from a progenerator Mt. of category .,I( t. . As a first step in establish
and projective. The argument can be reversed, and thus J1 is epic if and only ing this fact, let us start with an arbitrary Morita context (16.6), not neces
if Mt. is finitely generated and projective. sarily derived from a �module M, and let us see what conclusions may be
Next, im T trace Mt. . Hence by ( 15.5), T is epic if and only if M is a
=
drawn if we assume that J1 is epic.
1 64 M ORITA EQUIVALENCE ( 1 6. 1 2) ( 1 6. 1 3) MORITA CORRESPONDENCE 1 65
(16. 1 2) THEOREM. Suppose that the map J1 occurring in the Morita context whereas
(1 6.6) is epic. Then both MA and A N are finitely generated projective modules,
and A acts faithfullyt on N. Furthermore, n' · { (J1' (tjJ ® 1) · (m ® n)} = n' · { /( tjJm ® n)} = n' · {tjJm, n}
= < n ' , tjJm ) · n {(tjJm)n' } · n en' , m] · n.
M � RomA (N, L\), = =
Then A ' is a ring, and there are bimodule structures N " ' N* ' and If m E ker tjJ, then [N, m] = 0, whence
bimodule homomorphisms A A A A
m = I (mi , ni )m = I mJni , m] = 0,
J1' : N* ® A N + A ', f ® n + {f, n } , and thus tjJ is monic. To prove tjJ epic, let f E N* and set
!' : N ®A ' N* + L\, n ® f + < n, f ) f(n), m = I mi ' f(nJ .
=
where
We show that tjJ(m) = f by checking that for each n E o N, {tjJ(m)}n = f(n).
{f, n}g f < n, g ) . We have
[n, I m i · f(ni )]
=
where we use ( , ) and [ , ] for the maps J1, ! occurring in (1 6.6). Also define a
ring homomorphism This completes the verification that tjJ is an isomorphism. It follows at once
that tjJ ® 1 is also an isomorphism.
qJ : A + A' , A + AR ,
Now we note that
where AR is right mUltiplication by A on N. Let us verify that the following 1A' qJJ1(z) J1' { (tjJ ® 1)z},
= =
diagram is commutative : whence by (16.8) J1' is an isomorphism. Since (16.13) commutes, it follows
I/I ® l that qJ is an isomorphism, as claimed. This in turn implies that A acts faith
M ® A N +) N* ® A N fully on N.
'
( 16 . 1 3) fl t t fl
0
To prove equality of two elements of A ' , we compute their action on an we may conclude that MA is finitely generated and projective. This completes
arbitrary element n' E N. We have the proof.
n ' {(qJ J1) · (m ® n)} = n ' · (m, n) = en' , m]n,
We called the rings L\, A Morita equivalent if there exists a Morita context
( 1 6.6) in which both J1 and ! are epimorphisms (and hence isomorphisms).
t This means that for A E A, if N A = 0 then A = o.
We saw in (1 6.9) that the Morita context derived from a progenerator MA
166 MORITA EQUIVALENCE (16. l 6) MORITA CORRES PONDENCE 1 67
(1 6. 1 4)
of At� always gives a Morita equivalence, and in that special case we con The first equation in (16. l 5) then asserts that AM has trace ideal A, whence
structed an equivalence of categories At� � At", where A = Hom�(M, M). AM is a generator of AAt by (1 5 . 5). Likewise, the second equation in ( 1 6. l 5)
We now prove that every Morita equivalence of rings is derived from a implies that M� is a generator of At� . Further, since J1 is epic, it follows from
progenerator, and gives an equivalence of categories. ( 1 6. 1 2) that M� is finitely generated and projective ; the same holds for A M
since L is epic. This proves the first part of (i), and the second part follows by
(16. 1 4) T�eorem. Let the rings .1 and A be Morita equivalent relative to a symmetry.
Morita context AM� , �NA as in (16.6). Then The assertions in (ii) and (iii) are immediate consequences of ( 16.12), using
(i) M is a progenerator for both At� and AAt. symmetry arguments. (See also Exercise 16.7). To prove (iv) we observe that
N is a progenerator for both �At and AtA •
for each X E AtA '
(ii) There are bimodule isomorphisms {( . ® N)( · ® M) } X = (X ®A M) ®� N � X ®A (M ® � N)
N � Hom�(M, .1) � HomA(M, A), � X ®A A � X,
(iii) There are ring isomorphisms since M ® � N � A as AAbimodules . All of these isomorphisms are
A � Hom�(M, M) � Hom�(N, N), natural in X, and thus ( . ® N) ( . ® M) is naturally equivalent to the identity
functor on ..#A. Analogous arguments hold for the other functors in (iv),
where Hom�(N, N) acts as a ring of right operators on N, and likewise so we obtain category equivalences, as claimed. The natural equivalences of
HomA(M, M) on M. the type · ®A M HomA (N, · ) follow from (ii) and (16.1 1).
'"
(iv) There are inverse pairs of equivalences of categories Finally S(A) = M, so the right ideal J of A maps onto S(1) = J ® /\ M.
But J ® A M � J M as .1modules, since A M is projective. This implies the
first statement in (v). The second statement follows by symmetry. Finally,
where if 1 is a twosided ideal of .1, then 1 corresponds to the subbimodule M1 of
AM� , and this in turn corresponds (via T) to the twosided ideal Hom�(M, M1).
S = . ®A M '" HomA (N , · ), T = · ® � N "' Hom�(M, · ), This completes the proof.
S' = N ® A · '" HomA (M , · ), T' = M ®� . '" Hom�(N, · ).
Let us briefly consider the special case where M� is free on r generators,
(v) The correspondence J + JM gives an isomorphism of the lattice of and where
right ideals of A and the lattice of submodules of M� , and the twosided ideals
( 16. 16)
correspond to subbimodules of AM � .
The correspondence 1 + M1 gives an isomorphism of the lattice of left Clearly M� is a pro generator for vH.." and there is a category equivalence
ideals of .1 and the lattice of submodules of AM, and the twosided ideals Hom�(M, . ) : At� + AtA.
correspond to subbimodules of AM� .
The correspondence 1 + Hom�(M, M1) gives an isomorphism between the The isomorphism in ( 16. 1 6) can be described explicitly, once we start with a
lattice of twosided ideals of .1 and the lattice of twosided ideals of A, once we free .1basis {m 1 , , mr} of M. Each f E A determines a matrix Tf E Mr(.1),
• • •
(16. 18) COROLLARY. The map I + Mr(l) gives an isomorphism between the left noethe rian Ralgeb ra. and let M be a finitely generat
�
ed left Amod u e . Sh.,?w that
of 71..11, where /\. R Q9 R A
lattice of twosided ideals of .1 and the lattice of twosided ideals of M,(.1). M is a generator of "JIt if and only if R Q9 R M is a genera tor =
Proof By (16.14(v)), the map I + Hom&(M, MI) gives the desired isomorphism, 7. Show that the isomor phisms in ( 1 6. 14(ii)) are given
as follows :
once we identify A with Mr(.1) asin (16. 16). We have N � Hom/j, (M, tl.), n _ En, . ] ; N � Hom,,(M, A), n  ( " n).
. M � H o m . (N, tl.), m  [ ', m J .
M � H om"(N ' A) , m  (m , ' ) , a
a simple left artinian ring, and that tl. Hom,,(V, V)' [Hmt : VIew V � s a
L nj ® gj L (mj , f)nj ® gj L mi Q9 [ f , nj ] gj that A is =
=
.
ideals satIsfy the D.C.C., SInce
j i,j =
bimodu le " VA ' By (16. 1 8), the ring A is simple. Its left
set of tl.subsp aces of V by ( 1 6. 14). Finally, tl. � Hom,,(V, V)
they are in bijection with the
by ( 16. 14). This proves the easier part of Wedderb urn's Theorem (7.4).J
2. Consider an equivalence of categories A" +
AA ' where A, tl. are rings. Show
that this equivalence comes from a Morita context. [Hint : Set M T(tl.), N S(A). = =
Then there are bimodule structures /j,M" , "N/j, which give a Morita equivalence be
tween A and tl.. Further, the functors S and Q9 N are naturally equivalent, as are T
.
and , Q9 M.]
3. Let M be a free right tl.module on r generators, and set A Hom/j,(M, M). =
(1 7.3) Theorem. (i) Let A = M,(fl). Then A is a maximal Rorder in A, and has a
unique maximal twosided ideal nDA. The powers
(nDAr = n';; A, m = 0, 1, 2, . . . ,
give all of the nonzero twosided ideals of A.
5. Maximal Orders Over Discrete Valuation Rings (ii) Every maximal Rorder in A is of the form uAu  1 for some u E u(A),
and each such ring is a maximal order.
.
Throughout this chapter, R denot es a discrete valuation ring with (iii) Every maximal order A' is left and ri� ht heredit� ry, and each of Its n� �
quotie nt sided ideals is principal. The unique maxImal twoSIded Ideal . of uAu _ IS
field K, maximal ideal P nR i= 0, and residu e class field R 1 1
RIP. We . 1 ' unDu  . u Au  .
=
have already seen in (10.5) that the study of maximal Rorders U ' nDA u  , t h at IS,
=
in separable
Kalgebras can be reduced to the case of central simple algebras.
We shall Proof. We know by (8.7) that A is a maxi�al � rder. �eeping the nota
investigate this case in detail here, by using the results from Chapt
maximal orders in skewfields, and then applying the Morit er IlI on tion (17.2), we set M = I' m ifl ; then we may IdentIfy � wIth Hom� (M, M).
a equivalence
between skewfields and full matrix algebras over skewfi elds. Since M� is a progenerator for the ca�egory JIt �, It �ollows from ( � 6.9)
We assum e throughout that A is a centra l simple Kalgebra, and that A is Morita equivalent to fl, relatIve to the Monta context derlved
.
is a minimal left ideal of A. Set D HomA (V, V), and view VD as a that V from M � . By ( 16. 1 8), the twosided ideals of A are given �y M,(l), :Ith I
bimodule.
If t v 1 ' , v,} is any right Dbasis for V, then for each a E AA we
=
ranging over the twosided ideals of fl. But each nonzero I IS some nDfl, by
• • •
may write (13.2). This proves (i), once we observe that
(17.1 ) a(v 1 , . . . , v) = (V 1 ' . . . , V,)(O: i)' (O:ij) E Mr(D), Mr(n�fl) n� ' Mr(fl) (nDA)m.
and the map defined by a * (o:J permits us to identify A with Mr(D)
= =
. as the
(1 7.2) a(m 1 ' . . . , mJ
Now let us prove that every right ideal of A is principal,t that IS, ?
=
) (O:i) E M/fl).
(m 1 ' . . . , m,)(O: i ' form xA for some x E A. By ( 16. 14), every right ideal of A can be wntten as
The map defined by a * (O:i) then gives an identification of Hom (M, L) for some flsubmodule L of M. However, M is flfree, and every
Hom�(M, M) right ideal of fl is principal, and is isomorphic to fl. Hence by (2.44) we may
with Mr(fl), and this identification is consistent with that obtained
from (17. 1).
Finally, we set write L f; l .fl, a free flmodule on s generators, where necessarily s � r.
. J
=
r
Let fls x d�� �te the set of s x r matrices with entries in fl. There is then a
bijection Homd (M, L) � fls X " given by f � U f ' where
17. MAXIMAL ORDERS (COMPLETE LOCAL CASE)
f(m 1 , . . . , m,) = (1 1 , . . , I) U f '
.
onesided ideals of fl, and that these ideals are necessarily twosided ideals. and thus we have
We set � fl/:p, a skewfield of finite dimension over the field R.
= f(m 1 , . . . , m,) = (m 1 ' . . . , mr) . BU f'
Let A Mr(D) be a central simple Kalgebra. Our aim here is to show that
r it follows that Hom � (M, L) rcan be identified
=
Mr(fl), and its conjugates, give all possible maximal Rorders in A. We shall Comparing this with (17.2),
also study the onesided and twosided ideals of such orders. with the right ideal B · fls x of M,(fl). But for each U E fls x , we have
t Caution : A principal ideal xA is not isomorphic to A, except when x E u(A).
1 70
1 72 MAXIMAL ORDERS OVER DISCRETE VALU ATION RINGS 1 73
(1 7.4) ( 1 7.6) M AXIMAL ORDERS (COMPLETE LO CAL C A SE)
B U = [B OJ [�J Proof By (1 7.4) we may choose A = Mr(Ll). We may write rad. A n;� for
some m > O. If m > 1, then A/rad A contains the nonzero mlpotent Ideal
=
where the righthand expression is a product of two r x r matrices. Therefore A/rad A � Mr(A/nDLl) = M/Ll/rad Ll).
This finishes·the proof.
and so we have shown that the given right ideal of A is principal, with gener Now let X be any full RIattice in A, and define its left and right orders as in
ator [B 0] . A corresponding statement holds for left ideals of A, by sym (8. 1 ), (8.2).
metry. (A rather different proof of these facts is given in (1 8.7).)
Turning next to the proof of (ii), let A' be another maximal order in A. ( 1 7.6) THEOREM. O/(X) is a maximal order if and only if o/X) is a maximal order.
Then A'A is a full right Alattice in A, so there exists a nonzero rx E R such that
rxA'A c A. Then rxA'A is a right ideal of A, hence is principal by the above, Proof Let A = o/X) be maximal, and set A' = O/(X). Then X is a right
and so A'A = uA for some u E A. Clearly A uA, whence u E u(A) by (6.4).
=
Al attice in A, so as in (17.3) we may write X = uA for some u E u(A). Then
Therefore uAu  1 . X C X,
uA = A'A A" A'A = A" u A =:;, A'u,
=
whence u Au  1c A'. Since uA u 1 is a maximal order, we obtain uAu  1 = A',

1
there exists a full free Lllattice M in V such that A Hom A (M, M). Con
=
The integers a I ' . . . , as are uniquely determined by x. (Possibly some oj the
versely, each such A is maximal. {ai} are infinite, and we interpret nrc; as 0.)
Proof Let N be any free LlIattice spanning V, and set A' = Hom A (N, N). Proof One can imitate the usual proof for the case where Ll is a (commutative)
By the above theorem, there exists u E u(A) such that A uA'u  1 • But then=
principal ideal domain. Indeed, the argument is even simpler in �his cas e,
choose M uN, and we have
=
. . .
since Ll is a local ring without zero divisors, and the fact that multIplIcatIOn In
Hom A (uN, uN) = U · Hom A (N, N) ' u  1 = A, Ll is noncommutative does not cause any difficulties. We shall not give the
details of the proof. For more general theorems of this type, see Knebusch [1].
as desired. Note that since NA is free, so is MA • The fact that each Hom A (M, M)
is maximal has already been observed in ( 1 7.3(i)). (1 7.8) THEOREM. (i) Set A = A/rad A Mr( �)' There is a onet �one cor�es
�
pondence J � J between maximal right ideals oj A and those of A. Here, J is
Another important consequence of (1 7.3) is the image of J in X, and A/ J � X/J. The module A/J is a simple A module.
(ii) Let Xl diag (nD , 1 , . . . , 1) E M/Ll). Then x l A is a maximal right ideal
=
(1 7.5) COROLLARY. For every maximal Rorder A in A, we hav(! oj A, and its conjugates
rad A = nDA, A/rad A � M/Ll/rad Ll), { u . X A . u  1 : u E u(An
1
where "rad" denotes Jacobson radical, and where Ll/rad Ll is a skewfield. give all of the maximal right ideals of A.
1 74 MAXIM AL ORDERS OVER DISCRE TE VALUAT
1 75
ION RINGS (17.8) (1 8.1) MAXIMAL O RDERS (LOCAL CASE)
(iii) For x E A, the right ideal xA is maximal if and only if the left ideal Ax is
maximal.
be a discrete valuation ring, not necessarily complete. Let R denote the P
adic completion of R, and K the quotient field of R . Given any separable
Kalgebra B, we may form the Kalgebra B = K ® K B, which is a separable
Proof. Each maximal right ideal J of A contains rad A, by (6.3) and (6.6). Kalgebra by Exercise 7.1 3. Of course, when B is a central simple Kalgebra,
Therefore we know from (7.8) that B is a central simple Kalgebra. However, if B is
A/J � (A/rad A)/(J/rad A) = A/J. merely assumed separable over K, then it may well happen that 13 will have
Since J is maxima� A/J is a simple Amodule, whence A/J is a simple A more simple components than B. This may occur even when B is a field (see
mod�e. Therefore J is also maximal. Conver sely for each maximal right ideal Exercise 7.7, for example). Even in the case where B is central simple over K,
Y of A, the k�rnel of the epimorphism A + A/Y is a maximal right ideal J of the full matrix algebras B, 13 may consist of different size matrices. Despite
A such that J Y. This proves (i).
= these precautionary remarks, however, we can obtain results on Rorders in
Next, we have B by using the theorems in §17 on R orders in B.
1 X l = e l l diag (1 , 0, . . . , 0) E A, If r is any Rorder in B, then setting t = R ®R r, we obtain an R order
t in B. If M is any r module, then M R ® R M is a tmodule. We shall use
=


=
onto itself, and carries xA onto uxv . A. Hence we have a = . . = a 0' (i) For each s � 1, there is a ring isomorphism
1 r l
. =
a
r
= 1, and thus A/nsA � A /nsA .
uxv = diag (1, . . . , 1, nJ = Xn (ii) There are ring isomorphisms
say. But xr = tX l t' for some t, t' E u(A), and thus uxv = tX l t' , so A/rad A � A/rad A � A/rad A.
xA = u 1t . X A1 = (u  1 t) . X 1 A . (u  1 t)  l . (iii) Let M, N be finitely generated left Amodules. Then M � N as A
modules if and only if M � N as A modules .

18. MAXIMAL ORDERS (LOCAL CASE) Exercise 5.3), it follows at once by (2.38) that
Keeping the notation introduced at the beginning of this chapter, let R R ® R HomA(M, N) � HOll1A (iW , N).
1 76 MAXIMAL ORDERS OVER DISCRETE VALUATION RINGS (18.3) (1 8.5) MAXIMAL ORDERS ( L O C A L CASE) 1 77
Thus we may view HomA(. M, N) as the Padic completion of Homt\(M, N). Proof Maximal orders have the indicated properties, by ( 1 8.1) and (18.3).
But Hom A(M, N) is dense in its completion (see §5, or (6. 1 6)), and so we can Conversely, let A be hereditary, with unique maximal twosided ideal rad A,
find an f E HomA(M, N) such that and let r be an order properly containing A. Then r is a left Alattice, hence
is a finitely generated projective left Amodule by ( 10.7). It follows then from
f == <p (mod n ' HomA (M, N)).
(1 6.7) that the map
Likewise, we may choose g E HomA (N, M) such that g == l/J (mod n). Then fl : HomA (r, A) ® A r + HomA (r, r)
gfE HomA (M, M) is such that is epic, where
m  (gf)m E nM , m E M. J1(f ® y) =
(f, y), y'(f, y) =
[y',f]y,
However, M n nM nM by Exercise 5.7, and thus m
m E M. This gives
=  (gf)m E nM for all and [y ', f] is f evaluated at y '. Suppose that J1{ L� ® yJ = 1. Then
LY y ) = y for all y E r, that is,
U; ,
M = (gf)M + nM.
L [y, }JYi y, Y E r.
=
(1 8.3) Th,eorem. Let A be a maximal Rorder in a central simple Kalgebra A. ideal of A. If T = A, then r is a progenerator of A.A. This gives (by (16.14))
Then A has a unique maximal twosided ideal �, given by � An rad A . =
A = Homn(r, r), where n = HomA(r, r),
Then rad A � �, and every nonzero twosided ideal of A is a power of �,
Further, rad A is the Padic completion of rad A. and where Arn is viewed as a bimodule. But then n = r, since every Aendo
morphism of Ar is given by right multiplication by an element of r. Therefore
A
�
�roof" Since is .a maximal R order in the central simple J(algebra A, the we obtain A Homn(r, r) = r, a contradiction.
=
nng A/rad A I� sIm ple. Hence also A/rad A is simple, by ( 1 8.2). But every We have now shown that T is a proper ideal of A, and thus T is contained in
. the unique maximal twosided ideal rad A of A. Since T r r, it follows that
maxImal twosIded Ideal of A contains rad A by (6. 1 3), and thus rad A is the =
unique maximal twosided ideal of A. Furthermore, (18.2(ii)) implies at once (rad A)r = r, which is impossible by Nakayama's Lemma. This shows that
that A is a maximal order, and completes the proof of the theorem.
rad A = A n rad A, R ® R rad A rad A. =
Next, since A is simple, every nonzero twosided ideal M of A is such that We are going to show that maximal Rorders in separable Kalgebras are
KM A. Thus M ::> nSA for some s, and hence M is the inverse image of a
=
principal ideal rings, that is, every onesided ideal is principal. Theorems
twosided ideal of A/nsA. But A/ns A � A/nsA, and all twosided ideals of 10.5 and 18.2 will enable us to reduce this question to the case of maximal
A/nsA are images of powers of rad A. It follows at once that M itself is a orders in central simple algebras. The key to the proof is the fact that such an
power of �, as claimed. This completes the proof of the theorem. order A is a primary ring, that is, A/rad A is a simple artinian ring. We begin
with a result on primary algebras.
W,e next give a criterion, due to AuslanderGoldman [1], for an order to be
maxImal. THEOREM. Let B be a primary finite dimensional algebra over a field k,
(1 8.5)
and let M, N be finitely generated projective left Bmodules. Then M � N if
(18.4) THEOREM. Let A be an Rorder in the central simple Kalgebra A. Then and only if (M : k) = (N : k).
A is maximal if and only if A is hereditary, and rad A is its unique maximal
twosided ideal. Proof Let 13 = B/rad B, a simple artinian ring, and let b E B map onto 5 E 13.
1 78 MAXIMAL ORDERS OVER DISCRETE VALUATION RINGS (18.6) (18.7) MAXIMAL ORDERS (LOCAL CASE) 1 79
Suppose that (1 8.7) Theorem. Let A be a maximal Rorder in a central simple Kalgebra A.
B = Be l EB · · · EB Ben (i) Let M and N be left Alattices. Then M � N if and only if M and N have
the same R rank.
is a decomposition of B into a direct sum of indecomposable (ii) Every onesided ideal of A is principal.
left ideals
�h!�e th� i eJ are primitive idem Eotents in B. By (6.21), Bei � Bej if and onl ; (iii) Every maximal R order in A is a conjugate uAu  1 of A, where u E u(A).
I� Bei � B�j : and the � odules {BeJ are minimal left ideals of
E. Since E is (iv) Let A K ® K A � MiE), where E is a skewfield with center K, and
=
sImple artIman, the {BeJ are mutually isomorphic. Therefore Be. let Q be the unique maximal R order in E. Then
� Be 1 £or
a11 1.. But M and N are direct summands of B(s) for some s, whenc ! 
We need one more preliminary result : is Aprojective, since if M I A(r) then M I A(r). Now let M, N be any left Alattices.
Then by ( 1 8.6), M � N if and only if M � N as Amodules. Since A/rad A is a
(18.6) THEOREM. Let A be any Rorder in a finite dimensional Kalgebra simple artinian ring by (1 7.5). it follows from (18.2) that A is a primary ring.
KA,
�d let M, N be projective left Alattices. Set A A/nA, M M/nM, Hence by ( 1 8.5), M � N if and only if (M: R) = (N : R). Finally, we note that
N = NlnN. Then M � N as Amodules if and only if M � it as Amodules. (M : R) = rankR M. This proves that M � N if and only if M and N have the
= =
same Rrank.
Proof� �lear1y � � N implies that M � N. Conversely, let cp : M � Suppose next that M is a left ideal of A. Then KM is a left ideal of the simple
. both M and N are Aproje N
be a A Isomor�hIsm. Sm�e ctive, we may find artinian ring A, whence KM Ae = K . Ae for some idempotent e E A.
cp, cp  , respect ively, making the following
. 1
=
A homomorphIsms a, f3 lIftmg Hence A e � M by (i) ; if this isomorphism maps e onto m, then M = Am . This
.
dIagram commute : completes the proof of (ii). It should be pointed out that this proof is com
pletely independent of that given for the corresponding statement in (17.3(iii)),
and so constitutes another proof of that statement.
The argument in ( 17.3) which proves that every maximal order is conjugate
to A, carries over unchanged to the present case. It remains for us to prove
(P
(iv). Keeping the notation of (iv), we showed in ( 1 7.5) that
M I N.
( (p _ 1
A/rad A � Mt(Q/rad Q).
Thus the desired formula for A/rad A holds by virtue of ( 1 8.2). This completes
For m E M, we have (f3a) m  m E ,,:M. Hence M = f3aM + nM, so M f3aM =
the proof of the theorem.
by �akayama ,s Le�ma. Then f3a IS an automorphism of M by (6.3a). Likewise,
af3 IS an automorphIsm of N, and so a : M � N, as desired. (1 8.8) Remarks. (i) The integer t occurring in ( 1 8.7(iv)) is called the capacity
of rad A. If A � M/D), where D is a skewfield, and if
There is a generalization of the above theorem, in which the hypothesis on
M and � IS . drop e . he general version states : If KA is Kseparable, then fj K ® K D � M/E)
� � �
there �XIStS a pOSItIve mteger q depending on A, such that for any pair of
=
potent e in the semisimple ring A. Thus M � Ae by the first part of this proof,
is not a skew field, then �/rad � is certainly not a skewfield. Indeed, in terms and the theorem is established.
of the notation of the first remark above, we have
�/rad � � M (n/rad n),
s EXERCISES
where n is the maximal R order in E. Unless otherwise stated, R denotes a discrete valuation ring, not necessarily complete.
Generalizing (1 7.8), we have 1. Let � be a maximal Rorder in a skewfield D with center K. Show that every
left �lattice is free. [Hint : Since � is hereditary, each left �lattice M is isomorphic
to a direct sum of ideals of �. Each such ideal is principal, hence is isomorphic to �.]
(18.9) THEOREM. Let A be a maximal Rorder in a central simple Kalgebra A.
(i) There is a onetoone correspondence J � J between the set of maximal 2. Prove ( 1 8.9(ii), (iii)) without using the results for the case where R is comE.let�
right ideals J of A, and the set of maximal right ideals J of A/rad A. [Hint : Let A A/rad A, and let xA, yA be � aximal rig?t ideals of A. T?en xA, �
are maximal right ideals of the simple ring A. By ExerCIse 7. 1 2, there eXIsts u E u(A)
=
(ii) All maximal right ideals of A are mutually conjugate under inner auto
such that U ' xA . u I yA . Choose v E A with v u ; then v E u(A) by Exercise 6.2.
Therefore v . xA . V I yA, since these maximal right ideals have the same image in
morphisms by units of A.
= =
A.
(iii) If xA is a maximal right ideal of A, then Ax is a maximal left ideal of A.
=
that p. == 1 (mod Pi)' Pi == 0 (mod P), j =1= i. Set f I PJ , and show that f:M � N
'
with v == u (mod nA.) ; since v, u have the same image in A/nA, and since
is an e �imorphism, by verifying that fPi : MPi � NPi is epic for 1 � i � n. Deduce from
=
nA c rad A, it follows from Exercise 6.2 that v E u(A). But then yA and vxA
are a pair of maximal right ideals of A, having the same image in A/rad A. this that f is an isomorphism.]
It follows from (i) that yA vxA = V ' xA ' V  I as desired, which completes
=
4. Which of the results of this section remain true when R is assumed to be a semi
the proof. local Dedekind domain, rather than a discrete valuation ring?
1. Show t hat in a proper product of maximal integral ideals, the factors need not
�(A/R) = �(A/R) ' A.
, one another. [Hint : Suppose that M ' M
commute WIth M 2 ' M ' where M l ' M 2
are maXIma I Integra I Ideals. If the algebra A is M n(D), where n > 1,l then M 1 is not a
1 2 =
Proof. Let x (d i) E M/D), and let A = ((3i ) E Mr(A). Then by Exercise 9.5,
' ' '
twosided O,(M l)ideal. Therefore
=
2, Show that in some cases an integral ideal may be expressed in more than one way Hence if each d" E X, then x E A. On the other hand, let A have entry {3
( ) ( )
as a pr� p�r product ;f m aXimal integral ideals, [Hint : Let A M 2 (K), A M 2 (R),
O
= = at position (k, i), and zeros elsewhere, with (3 variable in A. Then
tr A/K (XA) = trDI K(dzk {3), so if x E A then dZk E X. This proves that A = M/X). If
Xl 0 ' X2 0 , where n is a prime element of R. Then X l ' X 2 commute, and
X = (rad A)  m, then
= =
n 1
1
nA = AX 1 ' x; (Ax 2 ) X 1 = A X 2 . X ; l (Ax 1 )X 2
A = Mr((rad A)  m) = { Mr(rad A)} m = { rad A}  m,
gives two different factorizations of nA.]

containing A (by ( 10. 1)). Let � = rad A; then by (19.3) we have A = �  m �(A/R) = �(Q/R) ' A.
for some m � O. Call A the inverse different, and define the different But �(Q/R) = (rad Q)m  l by (1 4.9), which completes the proof.
� = �(A/R) = �m. For any left ideal L of A, define its norm as
(20. 1) THEOREM. If R is the Padic completion of R, then �/ K(L) = ordR A/L.
�(A/R) R ® R �(A/R).
=
(20.4) THEOREM. The discriminant d(A/R) is given by
Proof. By (9.29), tr A/R extends the map trA/K ' Hence for each x E A we have d(A/R) = NAI K (�(A/R)).
1 86 MAXIMAL ORDERS OVER DISCRETE VALUATION RINGS (20.4)
Proof Both the discriminant and the different behave properly under the
passage from R to R, by (20. 1) and Exercise 10.6. So likewise does the norm
and thus it suffices to prove the result for the case where R is complete. But
then A � M/O) as in the proof of (20.3), and !:l(A/R) = n� . A for some k.
Therefore
N(!:l(A/R)) = ord R A/�A = ordR n; kA/A = ordR A/A.
6. Maximal Orders over Dedekind Domains
The rest of the proof is exactly like that of (14.8).
Throughout this chapter, R denotes a Dedekind domain with quotient
field K, R i= K, and P ranges over the maximal ideals of R. Let Rp be the
localization of R at P, and Rp the Padic completion of R. Let Kp be the
quotient field of Rp . For A any Rorder in the separable Kalgebra A, and M
any Amodule, set
.A P = Rp ®R A, M p = Rp ®R M, Ap = Kp ®K A.
Then .Ap is an Rporder in Ap, and Mp is a A.pmodule. It follows from
Exercise 7 . 1 3 that Ap is a separable Kpalgebra.
In this chapter we shall generalize the theorems obtained in Chapters III
and V. There are two ways of accomplishing this. In the first approach,
followed in §§ 2122, we use the local results from Chapter V, together with
the techniques of § Sa, to obtain global results. The second approach, given
in § 23, is independent of the first, and also yields the basic theorems for the
global case. In this second method, we give some straightforward generaliza
tions of some of the familiar proofs in algebraic number theory. Indeed, this
same dichotomy already occurs in algebraic number theory. One can develop
the ideal theory of Dedekind domains either from the standpoint of valuation
theory (first approach), or by manipulations with the ideals themselves
(second approach). Just as in algebraic number theory, however, we cannot
in the long run avoid the use of Padic completions. Many of the deeper
results depend on the interrelations between the global and local theories.
Indeed, we have already seen in (20.3) that when R is a discrete valuation ring,
we may calculate the different !:l(A/R) by passing to completions.
1 87
188 MAXIMAL ORDERS OVER DEDEKIND DOMAINS (21.2) (21.6) BASIC RESULTS 1 89
by (1 1 .6) Ap i � a maximal R porder in Ap , and so by (10.5) there is a de is split. This gives a decomposition of M, unless W 0 or W KM, which
composition Ap IO A i with each Ai a maximal Ri order in the central
= =
= '
completes the proof.
simple Ki algebra Ai . We may then apply the results of Chapter V, on maximal
orders in central simple algebras over complete fields, to the orders At. . We are now ready to determine all maximal Rorders in A. We have seen
Conversely, if the order A p is known to have a decomposition as in (21 . 1) that it suffices to handle the case of central simple algebras . We prove
for each P, and if each Ai is a maximal order in Ai ' then we may conclude
from (10.5) that A is necessarily a maximal order in A. As an illustration of (21 .6) Theorem . Let A = HomD CV, V) be a simple algebra, where V is a right
these remarks, we prove a number of basic results. vector space of dimension r over the skewfield D with center K. Let d be some
fixed maximal Rorder in D, and let M be any full right dlattice in V. �hen
(21 .2) Theorem. Let M be a full RIattice in A. Then 01(M) is a maximal Hom� (M, M) is a maximal Rorder in A. If A' is any maximal Rorder m A,
Rorder in A if and only if 0r(M) is a maximal Rorder in A. then there exists a full right dIattice N in V such that A' Hom� (N, N).
=
Proof. We know that Rp is Rflat, by (2.22) or Exercise 5.4. As in the proof of Proof Letting A = Hom� (M, M), we have
(8.5), it follows that
Ap = Rp ® R A = Hom�p (Mp, Mp )
(21.3)
for each P. B y (18.7) every right ideal o f/). p i s principal, whence Mp i s dpfre� .
and likewise for left orders. Now suppose that A 01 (M) is maximal, and
=
Hence by (8.7) Ap is a maximal Rporder. This holds for each P, and so A IS
l�t r = qr (M). Keeping the notat�on of (21 .1), there is also a decomposition maximal by (1 1 .2). .
Mp I Mi , where Mi is a full Rp lattice in Ai ' and where A.t 01 (M.t ). Now let A' be any maximal Rorder in A . Since A and A' are a pair of full
RIattices in A, we have A� = Ap a.e., say except at P I " ' " Pn• By (18.7),
= =
may take up 1 for P =1= P I ' . . . , Pn. For each PJ set X(P) u pMp ; then we
=
=
such that Np = X(;) for all P. But then A' and Hom� (N, N) are a pair of
(21 .4) Theorem. Maximal orders are left and right hereditary. full RIattices in A, with the same localizations for each P. Therefore
A' = Hom� (N, N) by Exercise 3.2, and the theorem is proved.
Proof Let A be a maximal Rorder in A, and use the notation of (21 . 1). Each
�
Ai is hereditary by (17.3), whence Ap is hereditary for each P. Thus A is itself (21 . 7) COROLLARY. Keeping the above notation, every maximal order in A is
hereditary, by (3.24) and (3.30). Morita equivalent to d.
(21 .5) COROLLARY. Let A be a maximal order. Then every left Alattice M Proof Every maximal order A in A is of the form A �� m� (M, M), wh ere
=
Proof· Since A is hereditary, M is projective by (10.7). If M decomposes, so functor. But for each dmodule L, we known that
does KM. Conversely, let W be any Asubmodule of KM, and set N = M n W
Then N is a Asubmodule of M, and by (4.0) the Amodule MIN is an R { Hom� (M, L) } p � Hom�p (Mp, Lp )
lattice. Hence MIN is Aprojective, and so the Aexact sequence
for each P. Since Mp is dpfree by Exercise 18.1, it follows. from §1 � that
o � N � M � MIN � 0 Hom �p (Mp , . ) is a faithful functor. Therefore Hom� (M, ' ) IS also faithful,
191
(22.4) IDEAL THEORY
190 MAXIMAL O RDERS OVER DEDEKIND DOMAINS (22. 1)
bra over the field RIP.
and so we have proved that MA is a progenerator for AlA ' Hence A and ,1. are of R, and 7\. = A/� is afinite dimensional simple alge
Morita equivalent, by ( 1 6.9), and the result is established. .
Proof Let � be a prime Ideal of A , an
d· set P � n R A A/� Sinc
=
.11 ¢ � .


·
full Rlattice in A, we have P =f
EXERCISES
If rx, p E R, then
M � aA c � or pA c �
1. Let A be a maximal Rorder in a separable Kalgebra A, and let be any left
rxp E P ==> rxA ' PA c ===>
Alattice. Show that the endomorphism ring EndA M is a maximal Rorder in the
separable Kalgebra End A KM. ==> rx E P or P E P.
over the
[ Hint : It suffices to prove the result when A is a central simple Kalgebra, and where 7\ is a finite dimensional algebra
Thus P is a prime ideal of R, and
R is a discrete valuation ring. Set r
left A  module. Then D = EndA V
= EndA M, B = End A K M, and let
is the skewfield part of A. We have KM
V be a simple
� V(t) fiel
���{ �e observe that rad A is nilpotent, since is artin
A ian. Thus rad � 0 =
1) auto 
as claimed. ] 1' be any maxIma" l tWO SI ded 1' deaI o f A . Then (22. :. .
eonversely, l et m . ' sIbI lItIe s for
T contam m 1', sm�e th� on 1 y pos
matically holds true when S and and esta IS
bI" h es
s that � is a pnme Ideal,
22. IDEAL THEORY S and T are � and A. This show
the theorem.
In this section we shall develop the theory of onesided and twosided
prove.
ideals in maximal orders over a Dedekind ring R with quotient field K, Keeping the above notation, we
by using the results of §§ 1 8, 19 for the local case. Let P, Q, . . denote prime
simple K � lgeb.r a A'
.
ideals of R, always assumed to be nonzero. To begin with, let A be any Rorder (22 4) THEOREM Let A be a maximal Rorder in a central
set of pnme 1d ea Is
'
in a separable Kalgebra A.
.
Hence (22. 1) is equivalent to the following condition : for each pair oftwosided Ap are powers of rad Ap . The re f,ore
By ( 1 8.3), all twosided ideals of
ideals J, j in A/�,
�p =
� . Ap = rad Ap ,
J' = O. cative set
(22.2) J · J' = O ==> J = O or torsionfree relative to the multipli
as claimed. Furthermore, A/� is
(22.3) THEOREM. The prime ideals of A coincide with the maximal twosided R P, so by (3. 10) we obtain
� A n � p A n rad Ap .

Conversely, given a prime ideal P of R, we may form the twosided ideal L · L 1 = A. The argument also yields the formula
PA of A. Then PA # A, since A is an RIattice, and so there exists a maximal (Lp)  1 . Lp = y  1 Ap Y = 0r {Lp)
twosided ideal � of A with PA c �. Clearly P R n �, since the inter= for each P, whence L 1 . L = Or{L). Likewise (L 1 )  1 = L, since this holds
section is a prime ideal of R containing P. The preceding discussion shows locally at each P, and the theorem is established.
that � = A n rad Ap, so P uniquely determines �. This completes the proof
of the theorem. (22.8) COROLLARY. Keeping the above notation, we have
Now let L be any full RIattice in the separable Kalgebra A, and define O lL 1 ) = 0r{L), Or{L 1 ) = OlL).
Proof In order to prove that L · L 1 = A, it suffices by Exercise 3.2 to prove Proof Let � 1 ' �2 be distinct prime ideals of A, and let Pi = �i n R, so
{
th at P =I P2• For Q any prime ideal of R, we have
1
Lp ' (Lp) 1  = Ap AQ ' Q # Pi '
(22. 1 1 ) {�i ) Q =
for each P. By ( 1 8. 10) we may write Lp = ApY for some Y E A. Then A = Ay, {�)Q ' Q = Pi '
•
which implies by (6.3a) that y E u{A ). Therefore But then � �2 = �2 � ' since this is true locally at every �.
1
•
1 .
Now let M be a twosided Aideal in A. Then for each Q, MQ IS a tWOSIded
{Lp)  1 = {x E A : Ap Y ' x c Ap } = y  1Ap 
1 94 (22. 1 5) IDEAL THEORY 195
MAXIMAL ORDERS OVER DEDEKIND DOMAINS (22 .12)
A Qideal in A, and so by (19.3) we have ei e(Pi , LIK). Each Pi determines a prime ideal �i of A by (22.4), and we may
=
where n is � h� p rime id �l � f A correspond ing to �. It should be pointed out for some positive integers {mi } .
: .
t� at t �e defInItIOn of nQ gIven ill (19.3) agrees wIth the definition of inverse (i) The factorization of PA into prime ideals of A is given by
gIven m (22.6). Furthermore, m
Q 0 a.e., since M Q A Q a.e. It follows at
d
i= 1
=
M I1 .Q Q ,
III
Q
=
(ii) For each i, mi is the ramification index of the skewfield part of the central
since this result holds everywhere l ocally. Indeed, for any P we have simple Lalgebra A, where denotes P(adic completion .
�
(iii) If the residue class field RIP is finite, then for each i, mi is the index of the
m I1 .Q pm Q, skewfield part of A .
Mp = �p p =
Q
�y (22. l 1 �. Finally, the ideal group is free abelian on the generators {�}, Proof From (4.27) we have P S = TI P t \ whence
smce the Ideal M uniquely determines the exponents in the factorization of PA TI (Pi Ati TI �tm i
=
for some j, � � j � t, �nd some prime ideal �j in Aj " Conversely, each such re
. On the other hand, rad A is the p(adic completion of �i by ( 18.3). Therefo
expreSSIOn IS a pnme Ideal of A. It follows at once that if we express M J. as a equatio n A we
when we take completions on both sides of the Pi �ti =
(22. 1 3 a)
is a powe r product of the corresponding prime ideals of A of the form (22. 1 3). Let us turn next to the theory of onesided ideals in a separable K algebra
. .
Smce � IS the product (for 1 � j � t) of the expressions in (22 . 1 3a), it is clear A. A normal ideal M is called a maximal integral ideal in A if M is a maximal
that M IS a power product of prime ideals of A. It is then also obvious that left ideal in its left order O I(M).
the prime ideals of A generate a free abelian group, and so the result is proved.
(22. l 5) THEOREM. For each maximal left ideal M of a maximal Rorder A in a
�e h �ve at our disposal all of the information necessary to describe how separable K algebra A, there is a unique prime ideal � of A such that
a pnme Ideal P of R behaves in a maximal Rorder in a simple Kalgebra : annA AIM {x E A : xA e M} .
(22. 1 6) � c M e A, � = =
(22.14) THEOREM: Let A be a central simple Lalgebra, and let L be a finite We say that M belongs to � . Then AIM is a simple left module over the simple
separable extenswn of K. Let S denote the integral closure of R in L. Given a ring A/� . Conversely, each � determines a maximal left ideal M of A which
prime ideal P of R, let P I ' . . . , Pd be the distinct primes of S dividing P, and set belongs to � .
1 96 (22. 17) IDEAL THEORY 1 97
MAXIMAL ORDERS OVER DEDEKIND DOMAINS (22. 19)
e that
Proof Given a maximal ideal M of A, we can decompose M and A according Proof Use induction on k, the result being trivial when k 1. Assum a
=
ann R AIM = P, whence by (3.6) we have Ap i Mp � AIM. Thus Mp is a maximal conta in is to divide " (see H. Cohn [1]). In the
number theory dictum "To
proof, it will be conve nient to use the notati on N N ij
left ideal of Ap. If we set A' = 0r{M), it then follows from (19.4) that Mp is = to indicate that the
a maximal right ideal of A� . a prope r produ ct
normal ideal N has left order A,, and right order A)'. Thus . .
might be written as N 1 2N 23 ' where A 1 , A 2 and A 3 need
On the other hand, let Q be any prime ideal of R different from P. Then not be dlstm ct.
�Q MQ = AQ , and so
=
(22. 1 8) T�eorem. Let M be a full left ideal of the maximal Rorder A in the Proof To prove (i), we show by induction on k that
separable Kalgebra A, and suppose that the Amodule AIM has composition M 1 2 M 2 3 · , · Mk , k + 1 c A k + 1 ,
length k. Then M is expressible as a proper product of k maximal integral 1 , so let k > 1 , and
ideals M 1 . . . Mk ' such that if the M's are integral ideals. The result is clear for k =
O/{M) = O /{M 1 )' 0r{M) = 0r{Mk )· assume the result for k  1 factors. Then
198 (22. 22) 1 99
MAXIMAL ORDERS OVER DEDEKIND DOMAINS (22.21) IDEAL THEORY
M 1 2 · · · Mk l , k Mk, k + C Ak Mk, k + = M M 1 2 used to define q> 1 2 ' Finally, q> 1 2 is an isomorphism since it has an inverse
.  l l k, k + l c Ak + l ,
as desIred. q> 2 1 ' given by q> 2 1 ( Y) = M YM  l , Y E 1(A2)· This completes the proof.
We n�xt prov� (iii), since (ii) will then follo'N as a special case. If (22.20)
holds wIth B, C mtegral, then (22.22) COROLLARY. Let M 1 2 be a maximal integral ideal which belongs to the
M 1 2 C A3 N 34 A4 = N 34 ' prime ideal � 1 of A I ' and to the prime ideal � 2 of A2 · Then � 2 = q>1 2( � 1 ) '
as claimed. Conversely, let M l 2 c N 3 4' and set
Proof It is clear from (22.21) that q>1 2( � 1 ) is a prime ideal of A2 , and that
B 1 3 = M 1 2(A3 M 1 2 )  l , C42 = (N34)  1 M 1 2 · q> 1 2( � 1 ) = M  l � l M. We have
1 l 1
It is easily ve rified that B, C are normal ideals with the indicated orders.
. Next, �1 e M => M l c � � � M  � l M c � � � 1 M M. =
the normal Ideals M nd A3M have the same right order, and M Thus q> 1 2(� 1 ) is a prime ideal of A2 contained in M, whence M belongs to
� c A3M,
so by (22.6) we obtam M  l ::J (A3 M)  l . Therefore B e M , M l q> 1 2( � I )' This completes the proof.
 =A
and hence B is integra l. Further, C = N  1 M C N l N = A 4 '
so C is al��
integral. Finally, (22.23) COROLLARY. Let M 23 be a maximal integral ideal belonging to the
B 1 3 N34 C42 M(A3 M)  l . N · N l M = M(A3 M)  l . A3 M prime ideal �2 of A2 • Then for each normal ideal B = B 1 2, the quotient
B 1 2 IB 1 2 M 23 is a simple left A 1 module, and
=
= M · A2 M,
annA l BIBM = {x E A l : xB BM} = q> 2 1 ( � 2) '
=
which completes the proof of (iii). To prove (ii) note that when A = A
we obtain B = M(AI M) l = A l in the above. This establishes the theore�: 1
Proof If BM c X c B is a chain of left AI modules, then M c B  X C A2
1
C?f course, the analogue of (ii) holds for a pair of normal ideals with the same is a chain of left A2modules. Thus B  X is either M or A2 , whence X is
nght order. either BM or B. This shows that BIBM is a simple left A I module. As in the
1
proof of (22. 15), annA l BIBM is a prime ideal of AI ' But q> 2 1 ( �2) = B�2 B 
Let us now investigate the uniqueness properties of the factorization of an is a prime ideal of A l which annihilates BIBM. Therefore annA l BIBM =
integral ideal into a proper product of maximal integral ideals. As we have q> 2 1 ( � 2 ), and the corollary is established.
already seen in Exercises 19.1 and 1 9.2, the factors need not commute' and
there may be more than one possible factorization. To some extent the diffi Let q>1 2 : l(A l ) � 1(A2 ) be the isomorphism described in (22.2 1). For
culty arises from the fact that we must consider not just one maxi�al order XE l(A ), we call the ideals X and q>1 2 (X) similar. Thus, by (22.22), each
but the sequence of maximal orders associated with the successive factors. Le� maxim;l integral ideal determines a pair of similar prime ideals. We may
us prove some results on change of orders. caution that the converse need not hold, that is, given a pair of similar prime
ideals � 1 and �2 ' it is not necessarily true (even in the complete local case)
(22.21) THEOREM . Let AI ' A2 be a pair of maximal Rorders in the separable that there exists a maximal integral ideal M 1 2 belonging to both � 1 and � 2
Kalgebra A. Then there exists a normal ideal M = M 1 2' If l(A . ) denotes the (see Exercise 22.1). . ' .
group of twosided A { ideals in A (j 1 , 2), there is an isomorphism We are now ready to deal with the uniqueness propertIes of factorIZatIOns
into proper products of maximal integral ideals, still assuming that A is any
=
N X N. Therefore q> 1 2 does not depend on the choice of the normal ideal Proof Let L = L l r be an integral ideal, and suppose that
200 M AXIMAL ORDERS OVER DEDEKIND D OMAINS (22.25) (22.28 ) IDEAL THEORY 201
n
(22.25) 'll = ann"T. Then 'll is a twosided ideal of A, and so we may write 'll = TI
1
�t ;
where each M i , i + 1 is a maximal integral ideal, and Mk, k + 1 is interpreted to where the {'l3J are distinct prime ideals of A. Then T is a left (A/'ll)module.
be Mkr . By (22.23), the chain of left A 1 modules But
n
(22.27)
are therefore the composition factors of the left A 1 module A 1 /L . By the � = {x E T : ��i x = O} = {x E T : �� x = o for some m} .
JordanHolder Theorem, these c.omposition factors are uniquely determined
by L, up to A 1 isomorphism and order of occurrence. (Call T. the � .primary component of T.)
There are k such factor modules, so L determines k uniquely. Furthermore, It s�ffices t� show that the left Amodule � has a composition series in
let �i denote the prime ideal of Ai to which M i, i + 1 belongs, 1 :::; i :::; k. which the factors occur in a preassigned order. Each composition factor
Then by (22.23) we have X of � is a simple left Amodule, hence as in (22. 1 5), ann" X is a prime ideal of
A. But then ann" X � i ' since we already know that �ti X = O. Hence X
=
e ).e . = e . . Call ei the left unit of aij , and ej the right unit of aij
Kalgebra A, and let {S 1 " ' " Sk } be the composition factors of the Amodule
•
6i) ai bkl is defined if and only if j = k, that is, if and only if the right unit
/
AIL, arranged in any preassigned order. Then there is afactorization (22.25) of
of aij equals the left unit of bkl •
L into a product of maximal integral ideals, such that the factor modules in
(iii) If ab and bc are defined, so are (ab)c and a(bc), and these are equal.
(22.27) are precisely S 1 , . . . , Sk ' in that order. 1
(iv) For each aij E G, there exists an aD E G with left unit ej , right unit ei ,
such that
Proof Any composition series for the left Amodule AIL lifts to an unrefinable
chain of left ideals of A, and this chain can be expressed in the form (22.26)
'
for some choice of M s (by (22. 19» . Hence it suffices to prove that AIL has a (v) Given any pair of units e, e' E G, there is an element aij E G with left unit
composition series in which the composition factors occur in the preassigned e, right unit e'.
order S 1 , . . . , Sk ' Furthermore, a decomposition of A into simple components The collection of all normal ideals in A, relative to proper multiplication, is
yields corresponding decompositions of A and L, and hence of AIL. It is the Brandt groupoid associated with A. The units are the maximal orders in A ;
1
therefore sufficient to deal with the case where A is a central simple Kalgebra. the inverse ofa normal ideal M is the normal ideaI M  . IfA and A' are maximal
Let us set T AIL, an Rtorsion finitely generated left Amodule, and let
=
.
Qrders, the normal ideal A A' has left unit A, right unit A'.
202 MAXIMAL ORDERS OVER DEDEKIND DOMAINS
EXERCISES 203
EXERCISES 5. Any simple left Amodule is of the form AIM, where M is a maximal left ideal
Unless otherwise stated, A denotes a maximal Rorder in a separable Kalgebra A. of A. Let P R n M. Show that P is a prime ideal of R, and that P
= annR AI M. =
increasing either factor increases the product MN. Then 8. Let L, M be left ideals in A such that
MN = MAz ' N = M ' A zN ==:;;> N = AzN ==> Az = A3 . ann R AIL + ann R AIM = R.
Conversely, let Az = A3 and let M'N = MN, where M' is an RIattice properl y Show that there exists a left ideal N such that
containing M. Then
M' c M'Az =
M" N N  I M ' NN  I = = M, AI N � AIL i AIM.
a contradiction .] [Hin t : Define a left Ah omomorphism
3. Let � l ' . . . , � be distinc t prime ideals of A.
n Let I:A � AIL i AIM
n n
W= n
1
� ti, � = nI �/i, by I(),,) = ( )" + L, )" + M), }" E A For each prime ideal P of R, either (A/L)p = 0 or
(AI M)p = O. Thus Ip is epic for each P, whence also I is epic. Now choose N = ker I.]
Prove
9. Prove (22.9) directly. [Hint : M c O/(M) � M·M c M � M c Or(M).]
(i) 2I => � if and only if ai � hi ' 1 � i � n.
1 0. V�rify that the definition of inverses in (1 9.3) agrees with the definition in (22 .6).
n
(iii) 2I n � =
n �ra,, (a" bi l .
1
l
right multiplications by elements of M  . The formula M · M 
I
Al is then equival =
[Hint : Consider the Ralgebra A/PA, where R R/P, and imitate the proof of (4.30).]
=
For any full RIattice L = Lij in A, define L 1 as in (22.5). Then for x E L 1 ,
Lx c A t. ===> L' xA !. c A.! ===> xA t· c L 1 ,
23. ALTERNATE APPROACH TO GLOBAL IDEAL THEORY whence L 1 is a right A(mo dule. Thus we obtain
In §§2122, we derived results on ideals and orders in the global case by (23.3) Or(L� 1 ) ::J Ai ' Oz(L� 1 ) ::J Aj '
.
makmg use of the local theorems from Chapter V. In this section we shall and we now
give a� alternate approach to some of the global theorems obtained in §§2122. Our next step is to develop the theory of twosided ideals,
The dIrect global approach given below generalizes some of the standard prove
arguments of algebraic number theory. We shall not use any of those results
from the preceding sections, whose proofs were based on localization tech (23.4) LEMMA. Let M be a twosided ideal in the maximal order A. If M < A,
niques. This section may be skipped if desired without affecting the con then M  1 > A.
tinuity of the exposition. sly
As usual, let A be a separable Kalgebra, where K is the quotient field of the Proof It is clear from (23.3) that M  1 is a twosided Aideal in A, and obviou
that 1 A, and obtain a contra diction . Since
Dedekind domain R. In §22, we defined prime ideals of an Rorder A in A, and M  1 ::J A. Let us assume M  =
Let rx be a
showed that these coincide with the maximal twosided ideals of A. We now M < A, we can choos e a prime ideal � of A with M c � c A.
in A). Then by
recall some other definitions from §22. All ideals M considered below are nonzero element of R (\ � (such exist, since � is a full RIattice
assumed to be full RIattices in A. Call M a normal ideal if 0z( M) is a maximal (23.1)rxA contains a product of prime ideals, and so we may write
Rorder. An integral ideal is a norPlal ideal M such that M c ° (M). A maximal � ::J rx A ::J � � 2 ' " �r '
1
integral ideal is an integral ideal M which is a maximal left ide�l in ° ( M). (We �r
where the �i are prime ideals, and where r is minimal. Since
� ::J � • . •
shall eventually remedy this asymmetry in the definitions (see (23.2), (2 3 .10), and 1
and then
and � is a prime ideal, it follow s that � contains some factor
�j '
(23. 1 1).) Let us write M M ij to indicate that M is an ideal with
=
Much more than documents.
Discover everything Scribd has to offer, including books and audiobooks from major publishers.
Cancel anytime.