Escolar Documentos
Profissional Documentos
Cultura Documentos
E. L. Lady
August 5, 1998
The assumption in this book is that the reader is either a student of abelian group
theory who knows the bare minimum of commutative ring theory or a commutative
ring theorist whose knowledge of abelian group theory is essentially nil. In this chapter
we valiantly attempt to present a quick and dirty summary of the commutative ring
theory which will be relevant for what follows and which might not be included in the
normal basic graduate algebra course, as well as all the abelian group theory (in disguised
form) which is needed to study finite rank torsion free groups and finite rank torsion free
modules over dedekind domains. Although most of the results in this chapter are well
known, your author has found it hard to track some of them down in standard references,
so it seems worthwhile to gather them all together here.
By inference, then, it might seem plausible that this book is accessible to a reader who
knows no abelian group theory and almost no commutative ring theory, provided that he
knows at least a little non-commutative ring theory, at least enough to be comfortable
when encountering the terms left and right. However it is doubtful that this book
would be a good choice of reading matter for such a reader.
proposition 0.1. (1) A vector space over an infinite field cannot be the union of a finite
number of proper subspaces.
(2) If p1S
, . . . , pn are prime ideals in a commutative ring and a is an ideal such that
n
a 1 pi , then a pi for some i.
proof: (1) Suppose that V = V1 Vn with Vi V . We may assume wlog that V
is not a union of any proper subset of {V1 , . . . , Vn } . Thus we may choose v1 V1 with
v1
/ V2 Vn and v2 V2 with v2
/ V1 . Then for any k 6= 0 K , kv1 + v2
/ V1 V2
so kx + y Vi for i 3 . Since K is infinite and there are only finitely many Vi , it
follows that there exist distinct k, k 0 K such that kv1 + v2 , k 0 v1 + v2 Vi for some
i 3 . But it then follows that v1 Vi , a contradiction.
(2) We may assume without loss of generality that a is not contained in the union of
any proper subset of {p1 , . . . , pn } . Thus there exist ai a with ai pi and ai
/ pj for
j 6= i. Then a2 an pi for i > 1 and a2 S an
/ p1 since p1 is prime. If n > 1 ,
n
/ 1 pi , a contradiction. Thus n = 1 and
let x = a1 + a2 an . Then x a but x
a p1 . X
s
s
s
s
r m
s s0
s rm r sm r m
=
= 0 .
s
ss0
s
ss0
s
s
sm rn
rm sn
rm
rn
=
=
n.
s
s
s
s
s
s
Since localization preserves exactness, applying Hom( , N ) and localizing with respect to
S then yields a commutative diagram
0 HomR (M, N )
S 1 HomR (Rt , N )
S 1 HomR (Rs , N )
.
a
M
ai M
i
1
T
proof: ForTeach k there
L is an obvious map M/ ai M M/ak M and therefore there is
a map M/ ai M
M/ai M . Since ai + aj = R for k 6= j , a prime
Tnideal p can contain
at most one ai , and (ai M )p = Mp if ai 6 p. Thus if ak p then ( 1 ai M )p = ak Mp
and Mp /ai Mp = 0 for i 6= k . Therefore for all p,
!
n
M
M
M
M
M
= Tn
=
ak M p
ai M
ak M p
1 ai M p
1
T
proposition 0.7. A finitely generated projective module M over a local ring R is free.
In fact, if m is the maximal ideal in R and m1 , . . . , mt M are such that the cosets
m
1, . . . , m
t are a basis for M/mM as a vector space over R/m, then m1 , . . . , mt are a
basis for M .
proof: Let m be the unique maximal ideal in R . Choose m1 , . . . , mt M so that
the cosets m
1, . . . , m
t are a basis for the vector
P space M/mM over the field R/m. Let
t
ri mi . It follows easily from Nakayamas
: R M be defined by (r1 , . . . , rt ) =
Lemma that is surjective. Since M is a projective module, splits, so Rt = K L
with K = Ker and L M . Then K is finitely generated. Since induces an
isomorphism from Rt /mRt to M/mM , it follows that K/mK L/mL M/mM . These
are finite dimensional vector spaces over the field R/m and comparing dimensions yields
K/mK = 0 . Thus K = 0 by Nakayamas Lemma. Thus is monic and hence an
isomorphism. X
Projective modules over a local ring are free even if not finitely generated, but the
proof is much more difficult.
proposition 0.8. A finitely generated module M over a noetherian ring is projective if
and only if Mp is a free Rp -module for all prime ideals p.
proof: ( ): Using the criterion that projective modules are just the direct summands
of free modules, it is easy to see that the localization of a projective R-module at p is a
projective module over Rp . It then follows from Proposition 0.7 that this localization is a
free Rp -module.
( ): Suppose now that M is finitely generated and for all p, Mp is a free
Rp -module. To show that M is projective one must show that for every surjection
: X Y , the induced map : HomR (M, X) HomR (M, Y ) is surjective. By
Proposition 0.5, it suffices to prove that for all maximal ideals p, the localized map
(HomR (M, X))p (HomR (M, Y ))p is surjective. But since M is finitely generated, by
Proposition 0.4 and Proposition 0.3 there are natural isomorphisms yielding the following
commutative diagram:
(HomR (M, X))p (HomR (M, Y ))p
y
y
HomRp (Mp , Xp) HomRp (Mp , Yp ) 0
where the bottom map is surjective since Mp is a projective Rp -module . Thus
(HomR (M, X))p
(HomR (M, Y ))p is a surjection, proving the result. X
remark. The hypothesis that M be finitely generated is essential here. In later
chapters we will see many examples of non-finitely generated non-projective modules M
such that Mp is a free Rp -module for all prime ideals p. Such modules will be called
locally free. (See Chapter 9 in particular.)
COMAXIMAL IDEALS.
definition 0.9. Local Property.
proposition 0.10. Let a, b, c R .
(1) If a + b = R then a b = ab .
(2) If a + b = R and a + c = R , then a + (b c) = R .
(3) If a + b = R then for any n 1 , an + bn = R .
proof: In each case the hypothesis and conclusion are local properties, i. e. it suffices
to prove in (1) that ap bp = ap bp for all prime ideals p. Thus wlog we may assume
that R is a local ring.
(1) & (3) Since a + b = R , if R is a local ring it follows that a = R or b = R , If, say,
b = R then a b = a R = a = aR = ab. Furthermore an + bn = an + R = R .
(2) Assuming again that R is a local ring, either a = R or b = c = R . Both cases are
trivial. X
proposition 0.11. If a + b = R then a b R ab.
proof: By Lemma 0.10 the usual short exact sequence
0ababa+b 0
becomes
0 ab a b R 0,
which splits because R is free. Thus a b R ab.
ASSOCIATED PRIMES. One of the most basic concepts in abelian group theory
is that of the order of an element. And for any abelian group G, one is particularly
interested in the set of prime numbers p such that G contains an element of order p.
For modules over a commutative ring, the analogous concepts are the annihilator of an
element and the set of associated primes of a module.
If M is an R-module and m M we will write ann M for the annihilator of M and
ann m for the annihilator of m. I. e.
ann m = {r R | rm = 0}
ann M = {r R | (m M ) rm = 0} =
ann m.
mM
.
lemma 0.12. Let m M and let p be a prime ideal. Then m/1 6= 0 Mp if and only if
ann m p.
proof: In fact, m/1 = 0 if and only if sm = 0 for some s
/ p, i. e. if and only if
ann m 6 p. X
definition 0.13. We say that a prime ideal p is an associated prime for M if there
exists m M such that p = ann m.
We write Ass M (sometimes called the assassinator of M ) for the set of associated
primes for M .
The French call Ass M lassassin de M . A literal translation of this would be
the murderer of M , or the killer of M . Since this sounds much more violent in
English than it does in French, English speaking mathematicians have invented the word
assassinator to translate the French.
lemma 0.14. A prime ideal p is an associated prime for M if and only if M contains a
submodule isomorphic to R/p.
proof: For m M , if p = ann(m) then R/p is isomorphic to the cyclic
submodule Rm. X
proposition 0.15. If p is a prime ideal then Ass R/p = {p} .
proof: Clearly p Ass R/p. If q Ass R/p then R/p contains a submodule X
isomorphic to R/q . But then X is cyclic, i. e. is a principal ideal in the integral domain
R/p. Hence X R/p and q = ann X = ann R/p = p. X
For p to be an associated prime of M it is important that we have p = ann m for
some m, not merely p ann m. It is also important that p be a prime ideal. It is not
obvious, however, that there always exist ideals satisfying both of these requirements. In
other words, it is not a priori obvious that if M 6= 0 then Ass M will be non-empty. For
noetherian rings, however, this does turn out to be the case. (See Proposition ** below.)
example 0.16. Let R = Z.
(1) If p is a non-zero prime number, then (p) Ass M if and only if M contains an
element of order p.
(2) 0 Ass M if and only if M contains a torsion free element, i. e. an element m
with rm 6= 0 for all r 6= 0 Z.
We say that a module M over an integral domain R is torsion if for every m M
there exists r 6= 0 R such that rm = 0 . We say that M is torsion free if no
non-trivial submodule of M is torsion, or, equivalently,
(m M ) (r R) m 6= 0 & r 6= 0 rm 6= 0 .
proposition 0.17. Let M be a non-trivial module over a noetherian integral domain R .
Then
(1) M is torsion if and only if 0
/ Ass M .
(2) M is torsion free if and only if Ass M = {0} .
proof: (1) M is torsion if and only if there does not exist an element m M whose
annihilator is the zero ideal.
(2) M is torsion free if and only if the annihilator of every non-trivial element is the
zero ideal. X
If ann M = 0 then we say that M is faithful. In abelian group theory, a group G
which is not faithful as a Z-module is called bounded.
Clearly a module which is not faithful is torsion, however the converse is far from true.
For instance, with R = Z, the module Q/Z is torsion but also faithful.
A classical theorem in abelian group theory implies that every module over a dedekind
domain which is not faithful is a direct sum of cyclic modules. Therefore it is only the
faithful torsion modules which have any real interest in abelian group theory.
The following theorem is known as the Krull Intersection Theorem. In abelian group
theory terms, it says that a finitely generated torsion free module has no (non-trivial)
elements of infinite p-height , for any prime ideal p.
proposition 0.18. If M is a finitely generated
free module over a noetherian
T torsion
k
integral domain and p is any prime ideal then 1 p M = 0 .
note: Counter-examples abound when M is not finitely generated. For
if
T instance,
k
R = Z, M = Q, and p is any non-trivial prime ideal then pM = M so p M = M 6= 0 .
lemma 0.19. Suppose that M is a module over a commutative noetherian ring R and
let m 6= 0 M . Let S be a multiplicative set in R such that S ann m = . Then there
exists p Ass M such that ann m p and S p = . In fact, if q is any prime ideal
such that ann m q then there exists p Ass M with ann m p q .
proof: Since ann m 6= R then by Zorns Lemma there exist ideals q maximal with
respect to the properties ann m q and q S = . We claim that any such ideal must
be prime. In fact, if r, s
/ q then S (r) + q 6= and S (s) + q 6= . Thus there exist
rx + q1 , sy + q2 S , with x, y R , q1 , q2 q . Then rsxy + rxq2 + syq1 + q1 q2 S and
since S q = it follows that rs
/ q.
Let q be such a prime ideal. Since R is noetherian, among the ideals p such that
p q and p = ann rm for r R and rm 6= 0 , there exist ones maximal with
this property. Let p = ann rm be such an ideal. Clearly ann m ann rm = p.
/ p then r2 rm 6= 0 and
We claim that p is prime. In fact, if r1 r2 p and r2
p p + (r1 ) ann(r2 rm) q (why?). By the maximality of p we conclude that
p + (r1 ) = p, i. e. r1 p. This shows that p is prime and since p = ann rm, thus
p Ass M . X
& p S = }
10
Proposition 0.15 shows that for any prime ideal p, R/p is p-primary . The following
proposition shows how to generalize in case p is maximal.
proposition 0.31. If p is a maximal ideal then for all n > 0 the cyclic module R/pn is
p-primary .
proof: If p0 is any prime then by Proposition 0.23, p0 Supp R/pn if and only if pn p0
i. e. if and only if p p0 . Since p is maximal it then follows that Supp R/pn = {p} . Since
6= Ass R/pn Supp R/pn it follows that Ass R/pn = {p} . X
If p is not maximal then R/pn need not be p-primary . For dedekind domains,
however, which are the rings of interest in this book, all non-trivial prime ideals are
maximal.
proposition 0.32. If p is minimal in Ass M then Mp is p-primary .
proof: By Proposition 0.28, Ass Mp = {p0 Ass M | p0 p} = {p} .
In abelian group theory, we know that an abelian group G is p-primary for some
prime number p if and only if the order of every element of G is a power of p. The
analogous characterization for modules over a commutative ring is the following:
proposition 0.33. If R is noetherian and M is an R-module and p is a prime ideal,
the following conditions are equivalent:
(1) M is p-primary.
(2) The natural map : M Mp is monic and
(m M ) (k 1) pk m = 0.
proof: (1) (2): Suppose that M is p-primary , i. e. Ass M = {p} . Then by
Lemma 0.12, if m 6= 0 M , then m/1 6= 0 Mp . Therefore : M Mp is monic. Now
/ Ass S 1 M {p} .
let r 6= 0 p and let S = {r k | k 1} . By Proposition 0.28, p
Therefore Ass S 1 M = and so S 1 M = 0 . This means that for each m M there
exists rk S such that r k m = 0 . Since this is true for all r p and p is finitely
0
generated (because R is noetherian) it then follows easily that pk m = 0 for some k 0 .
(2) (1): Now suppose the stated conditions hold. Since M Mp is monic, by
Proposition 0.29, Ass M Ass Mp , so it suffices to prove that Ass Mp = {p} . Therefore
we may assume wlog that M = Mp . Now let q Ass M . Then q p and q = ann m
for some m. By assumption, for some k , pk ann m = q Then p q since q is prime.
Therefore q = p, showing that M is p-primary . X
11
S
proposition 0.35. If R is noetherian then {p | p Ass M } is the set of elements in R
which are zero divisors on M .
proof: By definition, r R is a zero divisor on M if and only if r ann m for some
m 6= 0 M . If this is the case then by Lemma 0.19, r p for some p Ass M .
Conversely, if p Ass M then pm = 0 for some m 6= 0 M , so p consists of zero divisors
on M . X
corollary 0.36. If R is noetherian then
in R .
MODULES WITH FINITE LENGTH. Henceforth we will assume that the ring R
is noetherian.
In abelian group theory, an important role is played by the finite groups. A number of
important definitions and theorems in the theory of finite rank torsion free groups have to
do with the condition that a certain quotient G/H is finite. (For instance, if G is a finite
rank torsion free group and H G, then H is quasi-equal to G if and only if G/H is
finite.)
For many commutative rings R , there do not exist non-trivial R-modules which are
finite as sets. Instead, the appropriate condition is that a module have finite length.
definition 0.37. A module M over a ring R is said to have finite length if and only if
it has a composition series
0 = M0
M1
...
M` = M
where each quotient Mi /Mi1 is a simple module, and in this case we define length M to
be the length ` of this composition series. The Jordan-H
older Theorem asserts that
length M is independent of the particular composition series.
A module has finite length if and only if it is both noetherian and artinian.
theorem 0.38. If R is a noetherian ring and M is an R-module such that all the
associated primes of M are maximal, then Ass M = Supp M and for each p Ass M , the
canonical map M Mp is surjective and Mp is isomorphic to the p-primary component
of M , i. e. to {m M | (n) pn m = 0} . Furthermore M is the direct sum of its
p-primary components.
proof: By Theorem 0.5, to show that M Mp is surjective, it suffices to prove
that for every maximal ideal m, the induced map Mm (Mp )m is surjective. But
since p is maximal, if m 6= p then m
/ Ass Mp = Supp Mp so that (Mp )m = 0 so
Mm (Mp )m is surely surjective. And if m = p then Mm (Mp )m is essentially
the identity
this shows that for each maximal ideal m, the obvious map
L map. Likewise
L
Mm
is an isomorphism, showing that M
pAss M Mp
Ass M Mp is an
m
isomorphism.
12
13
pk1 M
...
pM
14
15
16
definition 0.49. An integral domain satisfying the conditions of Proposition 0.49 is called
integrally closed (in its quotient field).
Using the terminology we will introduce in Chapter 3, condition (3) in Proposition 0.48
says that W is not quasi-equal to any subring of Q properly containing it.
17
18
theorem 0.59. Let W be an integral domain with quotient field Q . The following
conditions are equivalent:
(1)
(2)
(3)
(4)
proof: (1) (2): By Proposition 0.46 ideals are projective if and only if they are
invertible.
Note now that by Proposition 0.46, projective ideals are finitely generated. Hence (1)
implies that W is noetherian.
(1) (4): The ideals in Wp have the form ap , where a is an ideal in W . By
hypothesis a is projective, so by Proposition 0.8 ap is a free Wp -module. Thus all ideals
of Wp are free, so that Wp is a local principal ideal domain, i. e. a discrete valuation ring.
19
(4) (3): To see that W is integrally closed, it suffices by Lemma 0.51 to see that
Wp is integrally closed for all primes p. But this is true if Wp is a discrete valuation ring,
since principal ideal domains are integrally closed. Now let p be a prime ideal in W . The
prime ideals contained in p correspond to the prime ideals of Wp . Since Wp is a discrete
valuation ring, its only prime ideals are pWp and 0. Thus there are no non-trivial prime
ideals strictly contained in p, so p has height one. It follows that all prime ideals of W
are maximal.
(3) (1): Let a be an ideal in W . Since W is noetherian, a is finitely generated.
Hence by Proposition 0.8, to prove a projective it suffices to prove that ap is a free
Wp -module for all primes p. But since Wp is a principal ideal domain, ap is in fact
free. X
20
To prove that S 1 W = W 0 it suffices to prove that S 1 Wp = Wp0 for all prime ideals p
of W . If p
/ X then S 1 Wp = Q = Wp0 . But if p X then Wp S 1 Wp Wp0 Q .
Since Wp is a discrete valuation ring, by Proposition 0.54 it is a maximal subring of Q , so
S 1 Wp = Wp0 . It follows that S 1 W = W 0 . X
corollary 0.64. If W 0 is a subring of Q containing W and M, N are W 0 -modules ,
then HomW 0 (M, N ) = HomW (M, N ) and M W 0 N = M W N .
proof: By Proposition 0.63, W 0 = S 1 W for some multiplicative set S . Thus the result
reduces to Proposition 0.3. X
proposition 0.65. If G is a torsion module over a dedekind domain, then for every prime
p the p-primary component
of G can be canonically identified with the localization Gp .
L
Furthermore G =
Gp .
proof: Immediate from 0.38 since all primes in Ass G are necessarily maximal.
proposition 0.67. A finitely generated torsion module over a dedekind domain W has
finite length and is a finite direct sum of cyclic modules of the form W/pk , for various
prime ideals p.
proof: If G is torsion then 0
/ Ass G. Since all non-trivial primes in a dedekind
domain are maximal, it follows from Theorem 0.41 that G has finite length. Furthermore,
since L
all the associated primes of G are necessarily maximal, by Proposition 0.38,
G Ass G Gp . It thus suffices to prove that each Gp is a direct sum of cyclic modules
W/pk . But we may think of Gp as a Wp -module, and by Proposition 0.59 Wp is a
principal ideal domain, so this is a standard result. X
We will prove in Chapter 1 that every finitely generated torsion free module over
a dedekind domain is a direct sum of ideals, hence in particular is projective. It then
follows immediately that the torsion submodule of any W -module is a direct summand of
that module.
21
22
proposition 0.72. If a and b are non-trivial ideals in a dedekind domain W then there
exists an ideal a0 with a0 a such that a0 + b = W .
proof: Choose a 6= 0 a . Let c = aa1 W . Now by Proposition 0.71, c is generated
by bc together with an additional element x . Now multiplying the equation
c = xW + bc
by c1 = a/a yields
W = c1 c = xc1 + b =
But clearly ax/a a .
xa
+ b.
a
23
24
M/p
M
such
that
(k)
m
m
(mod pk M ) .
(mk )
k+1
k
1
1
The p-adic completion will be important in this book because of its algebraic
properties. The topology will not be of any real relevance. However, for the record, we
will give a few topological properties.
Q
If we give each module M/pk M the discrete topology, then 1 M/pk M has an
inherits is called the inverse
induced product topology. The relative topology that M
converges to an element x M
with respect
limit topology. A sequence {xt } in M
to the inverse limit topology if and only if the sequence converges in each separate
coordinate. Since the topology on the modules M/pk M is descrete, this means that
if where xt = {mtk }
1 and x = {mk } , then limt xt = x if and only if for each k ,
mtk = mk for sufficiently large t .
(This is related to the finite topology on EndR M as defined by Jacobson if we think
of EndR M as a submodule of M M .)
is the same as the p-adic topology.
proposition 0.82. The inverse limit topology on M
is a complete topological module in this topology. If : M M
is the canonical map,
M
.
then (M ) is topologically dense in M
25
proof: The neighborhood system at 0 in the inverse limit topology has a basis consisting
of those submodules Un consisting of elements whose first n coordinates are zero. Now
Un . On
since the first n coordinates live in M/pk M for k n , it follows that pn M
satisfy the condition mn+k mk (mod pn ) ,
the other hand, since the sequences in M
for all r . Thus Un pk M
, so the
it follows that if mr = 0 for r n then mr pn M
inverse limit topology and the p-adic topology are the same.
It seems clear that the p-adic completion is a contrived device to simplify the proofs
of certain theorems rather than a concept of intrinsic interest in its own right. There is
no natural geometric way of visualizing the p-adic numbers intuitively, and there are no
natural real-world applications which can be naturally represented by p-adic numbers.
However in terms of ring theory, there is one way in which the p-adic completion arises
in a fairly intrinsic way.
proposition 0.83. If p is a (non-zero) prime in a dedekind domain, then
p.
EndW W (p ) W
proof: *****
As an important application of this, we can construct the famous Pontraygin
module.
For convenience, suppose that W is a discrete valuation ring and let p be it unique
non-trivial prime. We will construct a submodule M of Q Q containing W W
such that M (Q 0) = W 0 and M (0 Q) = 0 W , and i (M ) = Q . By
Theorem 0.47, such a submodule will be given by an isomorphism from Q/W to itself.
By Proposition 0.77, Q/W W (p ) , and therefore by Proposition 0.83, corresponds
p.
to an invertible element W
More specifically, M will be the inverse image in Q Q of the module
{(x, (x)) | x Q/W } = {(x, x) | x Q/W } Q/W Q/W .
w
Now x Q/W can be written as a coset k + W for some w W and k 0 . Using
p
this, we will describe M in the following way:
(w, w)
M ={
| w W, k 0 } .
pk
p and so w
/ W.
We need to make sense of this, since W
p ) . It follows that
For given k , there exists ak W such that ak (mod pk W
pk
ak
and k have the same image in Q/W . Now if is the automorphism of Q/W given
p
w
ak w
w
by (x) = x , then ( k + W ) = k + W = k + W . Thus we can describe the
p
p
p
(w, ak w)
| w W, k 0 } .
pk
26
k
of 1 M/p M if the modules M/pn M are considered discrete spaces and
Q
k
topology.
1 M/p M is given the product
Q
is a closed subspace of
(5) M
M/pk M .
1 Q
(6) The diagonal map from M to 1 M/pk M embeds M/p M as a pure dense
.
submodule of M
Q
proof: Q
(1) The factors M/pk M are all Rp -modules , hence so is 1 M/pk M . And if
27
proof: The crucial fact here is that if M is p-primary then for every m M there
exists n (dependent on m, of course) such that pn m = 0 . Now consider a sequence
Q
R/pk R .; We will define rm = rn m. This is actually independent of n ,
r = {rk } R
provided that pn m = 0 because of the condition that rn+k rn (mod pn ) . From this, it
is easy to see that all the conditions are satisfied for M to be an R-module.
If W is the ring of integers and M has finite rank, then all M/pn M are finite and so
is compact by Tychonofs Theorem. In general, if p is a maximal ideal we can only
M
say that the M/pn M have finite length. But it still seems intuitively plausible that M
would have all the algebraic properties that one would expect of a compact module. In
belongs to the class of algebraically
fact, abelian group theorists will recognize that M
compact modules, which are characterized by the following Proposition 0.* below. The
p is the inverse limit of the family of
proof is a simple consequence of the fact that W
modules W/pk W . What this means is given by the following proposition:
p W/pk W and
proposition 0.88. Let n : W
n : W/pn+1 W W/pk W be the
canonical maps. Let M be a W -module and for each n 1 let n : M R/pk W .
p such
Suppose that for every n ,
n n+1 = n . Then there exists a unique : M W
that n = n for all n .
proof: ****
p ) Hom(M, W/pW ) given by 7 1 is
corollary 0.89. The map Hom(M, W
surjective.
proof: ****
is a pure submodule of any (not
proposition 0.90. If p is a maximal ideal and M
is a summand of X .
necessarily torsion free) W -module X , then M
proof: ****
is a finite rank W -module, except in the trivial case
One should not expect that M
p -module .
torsion free Wp -module . Furthermore, if p is maximal then M is a free W
proof: ****
ADDITIVE FUNCTORS. It is an absolute prerequisite for much of this book that
the reader be comfortable with the ideas of categories and natural transformations. All
that your author can do to help with this is to present here a few topics which may seem,
from the prospective of a standard graduate algebra course, a bit specialized.
28
The categories of modules (whether right or left) over a commutative ring, along
with almost all of the other categories used in this book, are what are called additive
categories. This means that if , Hom(X, Y ) then + and are defined,
and these operators make Hom(X, Y ) into an abelian group. Furthermore, it is required
that composition of functors be bilinear, i. e. in the same notation if Hom(W, X) and
Hom(Y, Z) then ( + ) = + and ( + ) = + . Furthermore, any two
objects X and Y in these categories have a direct sum X Y , which is simultaneously
a product and a coproduct. Finally, an additive category always has a zero object 0 ,
analogous to the trivial module in the category of modules over a ring and characterized
by the property that Hom(0, X) = Hom(X, 0) = 0 for all objects X .
Essentially, the point of an additive category is that one may use all of the instincts
and habits one has learned from working with modules over a ring without getting
into trouble. The one exception is that for some of the categories used in this book
statements that involve kernels or cokernels, or the word epimorphism, may not work
exactly the way one thinks they should. This does not cause problems if one simply uses
common sense. However for technical reasons it has seemed better to avoid the term
epimorphism in favor of the safer term surjection.
The following examples may be skipped for now by anyone who is confused rather than
enlightened by them.
examples 0.92.
(1) For an integral domain R , we can consider the category of
torsion free R-modules, i. e. those modules M such that
(r R)(m M ) r 6= 0 & m 6= 0 rm 6= 0 . This is an additive category
and it causes no special problems unless one has been trained so thoroughly in
category theory that one no longer has common sense. For instance if K is the
quotient field of R and : R K is the inclusion map, then within the category
in question (assuming K 6= R ) it would not make sense to talk about the cokernel
of , since the cokernel in the usual sense, i. e. K/R , is not torsion free and hence
does not exist within the category. Now it turns out that within the category
of torsion free modules over R does indeed have a cokernel, namely the zero
module, and that is an epimorphism. However only a bloody-minded fool
would think in those terms. Which is why in this book we will avoid the word
epimorphism.
(2) In Chapter 3 we will introduce a category which is absolutely fundamental to
the study of torsion free modules. Namely the category of torsion free modules
under quasi-homomorphisms. The objects in this category are simply the
torsion free modules, but instead of using ordinary homomorphisms we define
the morphisms from M to N to be the elements of KHom(M, N ) , where
KHom(M, N ) denotes the localization of Hom(M, N ) at the zero ideal (or, if one
prefers, K R Hom(M, N ) ). Thus morphisms in this category are not (usually)
homomorphisms from M to N in the customary sense. Nonetheless, the set of
morphisms KHom(M, N ) is an abelian group, and this is an additive category,
29
with the direct sum of two modules being simply their ordinary direct sum. We
will see in Chapter 3 that this category presents no particular problems, and that
the way to approach it is simply to forget that theres anything strange about it
and pretend that one is actually in the category from part (1) above.
Unlike epimorphism, the word exact is too important for us to be able to live
without. Fortunately, exactness poses no real conflict here between category theory and
common sense. Exactness in the category of torsion free R-modules (Example 0.92 (1))
means exactly what is does for the category of all R-modules . (So that for what its
worth although as noted above the inclusion map R K is an epimorphism in the
category of torsion free modules, the sequence R K 0 is not exact in the category.)
And we can avoid confusion when working in the category of torsion free modules under
quasi-homomorphisms (Example 0.92 (2)) by the simple expedient of using the term
quasi-exact rather than exact.
As Peter Freyd has pointed out, the whole reason one defines categories in the first
place is in order to define functors. And the reason that we have made a fuss about the
concept of an additive category is not because it itself has any crucial importance in what
follows, but because we want to be able to define the concept of an additive functor.
definition 0.93. A functor F between additive categories C and D is called an additive
functor if the function it induces from HomC (C1 , C2 ) to HomD (F (C1 ), F (C2 )) is a
homomorphism. In other words F ( + ) = F () + F () .
Additive functors are to additive categories what continuous maps are to topological
spaces, or what homomorphisms are to modules. In other words, just about any functor
that anyone ever uses between two additive categories and in particular virtually any
functor used in this book, is additive. The concept is worth noting, however, because of a
couple of useful propositions.
proposition 0.94. An additive functor preserves direct sums.
proof: ****
Actually, the converse is also true: A functor between additive categories is additive if
and only if it preserves direct sums (see [Freyd], for instance). However this converse is
not especially useful, at least not for us.
proposition 0.95. If R is a commutative ring and F is an additive functor from the
category of R-modules into the category of abelian groups, then for every R-module M ,
the abelian group F (M ) has a canonical R-module structure, so that from practical
purposes one may as well think of F as a functor from the category of R-modules into
itself.
proof: ****
30
There is a wonderful bit of black magic which is probably the main reason for
discussing all of this at all.
proposition 0.96. If S and T are additive functors from the category of R-modules
to some other additive category, and if : S T is a natural transformation, and if
R : S(R) T (R) is an isomorphism, then P : S(P ) T (P ) is an isomorphism for
every finitely generated projective module P . Furthermore if S and T preserve arbitrary
(i. e. infinite) coproducts, then P is an isomorphism for all projective modules P .
proof: ****
definition 0.97. If C and D are categories and if F : C D and G : D C
are functors, then we say that F is a left adjoint to G and that G is a right
adjoint to F is HomD (F (C), D) is naturally isomorphic to HomC (C, G(D)) for
pair of objects C C and D D . In other words, there exists an isomorphism
y
y
y
y
G( )
31
X
HomR (M, M R X) HomR (M, Y )
and that 1 sends HomR (X, HomR (M, Y )) to the map given by the composition
R 1M
X R M HomR (M, Y ) R M
Y.