COMMUTATIVE ALGEBRA

35

7.2. The Picard group of a ring.

Definition. A line bundle over a ring A is a finitely generated projective A-module
such that the rank function Spec A N is constant with value 1. We call A itself
the trivial line bundle.
We want to put an abelian group structure on the set of isomorphism classes of
line bundles over A. The product will be given by tensor product and the inverse by
a kind of duality. We know that is associative and commutative and the trivial
line bundle is an identity up to isomophism in each case. But to show that the
group operation is well-defined and we have inverses we need to prove a few results.
Definition. If X is a f.g. projective module, we define its dual by
X := HomA (X, A).
Lemma. Let X and Y be f.g. projective A-modules
(1) X A Y is a f.g. projective A-module and rkXA Y = rkX . rkY pointwise;
(2) X is a f.g. projective A-module and rkA = rkA ;
(3) If X is a line bundle then X A X
= A.
Proof. Since X is projective there exists an A-module Z such that X Z
= An is
n
n

free. Then (X A Y ) (Z A Y )
A

Y
Y
.
Since
Y
is
f.g.
projective,
Yn
=
=
A
is f.g. projective and so X A Y is a direct summand of a f.g projetive and so is a
f.g. projective.
n
Next we compute the ranks: if P in Spec(A), then XP
= Am
P and YP = AP say,
nm
then (X A Y )P
= AP and (1) follows.
= XP AP Yp
Similarly HomA (X, A) HomA (Z, A)
= HomA (An , A)
= An and so X is f.g.
projective.
Also as X is finitely presented, (X )P
= AnP and (2) follows.
= (XP )

Finally if m = 1, let : X X A; (f, x) = f (x). Locally we have

P : HomA (AP , AP )
= HomA (X, A)P AP
P

is an isomorphism and so is an isomorphism.

Definition. If A is a ring we may define the Picard group of A, Pic(A) to be the

set of isomorphism classes of line bundles over A, with multiplication given by
and inverses given by X 7 X .
Corollary. (Pic(A), ) is an abelian group.
Examples.
(1) If A is a local ring then all line bundles are trivial, and so Pic(A)
= 0.
(2) If O is the ring of integers of a number field then (as we will see) Pic(O) is
just the ideal class group of O
Exercise. Pic is a functor from rings to abelian groups; more specifically if f : B
C is a ring homomorphism then there is a group homomorphism Pic(f ) : Pic(A)
Pic(B) given by X 7 C B X and if also g : A B is a ring homomorphism then
Pic(f g) = Pic(f ) Pic(g).
As hinted in Example 2, when A is an integral domain, the Picard group of A
is something that arises classically in a familiar way: Since Spec(A) is connected
in this case a f.g. projective A-module is a line bundle if and only if Q(A) A X
=

36

COMMUTATIVE ALGEBRA

Q(A); that is rkX ((0)) = 1. Indeed since projective modules are flat (Ex. Sheet 2
Q2) the exact sequence 0 A Q(A) induces an injection
X
= A A X Q(A) A X Q(A).
Thus X is isomorphic to a finitely generated submodule of Q(A).
Definition. A fractional ideal I of A is a non-zero A-submodule of Q(A) that is
contained in a cyclic A-submodule of Q(A).
In particular every finitely generated A-submodule of Q(A) is a fractional ideal.
It is easy to see that if I and J are both fractional ideals of A then IJ is also
fractional ideal. Thus the set Frac(A) of fractional ideals in A is a commutative
monoid with identity given by A.
Definitions. We call a fractional ideal invertible or a Cartier divisor if it has an
inverse in Frac(A). We call the set of Cartier divisors Cart(A) an abelian group.
If f Q(A) , then the fractional ideal Af is a Cartier divisor with inverse Af 1 .
We call these divisors principal divisors of A.
Now if I, J Cart(A)
with IJ = A then there are x1 , . . . , xn in I and y1 , . . . , yn
P
in J such that
xi yi = 1. So we can define maps fP
: I An by f (a) =
n
(ay1 , . . . , ayn ) and g : A I by g(ei ) = xi and gf (a) = a xi yi = a. Thus I is a
summand of An and so is a f.g. projective A-module. Morover Q(A) A I
= Q(A)
and so I is a line bundle. Thus all Cartier divisors are line bundles.
Now suppose I and J are any Cartier divisors in A, we want to show that
IJ
= I A J and so the map Cart(A) Pic(A) sending a Cartier divisor to its
isomorphism class is a group homomorphism. Since I is projective it is flat (Q2 of
example sheet 2), so I A J I A Q(A)
= Q(A) is an injection and sends xi yi
to xi yi . Thus its image is the set IJ and IJ
= I A J as claimed.
Weve seen that every line bundle is isomorphic to a fractional ideal. In fact
the fractional ideal must be invertible: suppose X is a line bundle isomorphic to
the fractional ideal I. Then X = HomA (X, A)
= J for some fractional ideal J.
Then A
= IJ so IJ = Af some f Q(A) is a principal divisor. Thus
= X X
I(Jf 1 ) = A and I is in Cart(A).
Suppose now that I ker(Cart(A) Pic(A)). Then I
= A and so I is a
principal divisor.
Thus we have proven the following proposition.
Proposition. If A is a integral domain, then every Cartier divisor is a line bundle,
moreover there is an exact sequence of abelian groups
1 A Q(A) Cart(A) Pic(A) 0.
In particular the first non-trivial map is inclusion, the second maps f to the Cartier
divisor f A, and the third just realises a Cartier divisor as a line bundle. Thus
Pic(A)
= Cart(A)/(principal divisors of A).
Definition. We say a ring is a Dedekind domain if every fractional ideal is invertible.
Well see other ways to characterise Dedekind domains later.

COMMUTATIVE ALGEBRA

37

Examples.
(1) Every principal ideal domain is a Dedekind domain. Clearly, Pic(A) = 0
in this case. In fact a Dedekind domain is a principal ideal domain if and
only if Pic(A) = 0.
(2) If K is an algebraic number field and O is its ring of algebraic integers,
then O is a Dedekind domain and Pic(O) is known as the ideal class group.
It is known that this group is always finite. Understanding this group is
very important in algebraic number theory.
(3) If X is a smooth affine curve over C, then the coordinate ring C[X] of X
is a Dedekind domain. In particular if X is obtained by removing a single
point from a smooth projective curve (Riemann surface) X, then Pic(C[X])
is known as the Jacobian variety of X and is known to be isomorphic to
(R/Z)2g where g is the genus of X.
(4) (Claborn 1966) Incredibly every abelian group arises as the Picard group
of some Dedekind domain.

8. Integral extensions
8.1. Integral dependence. Suppose that A and B are rings, with A a subring of
B. We say an element of B is integral over A if x is a root of a monic polynomial
with coefficients in A.
Examples. If K is an algebraic extension of Q then the set of integral elements of
K over Z is by definition OK , the algebraic integers of K. In particular an element
of Q is integral over Z if and only if it is an integer. (Exercise if you havent seen
this: cf Numbers and Sets sheet 3 2006).
Proposition. Suppose A is a subring of B and x is an element of B. Write A[x]
for the subring of B generated by x and A. The following are equivalent:
(i)
(ii)
(iii)
(iv)

x is integral over A;
A[x] is a finitely generated A-module;
A[x] is contained in a subring of B that is a finitely generated A-module;
There is an A[x]-module M with AnnA[x] (M ) = 0 which is finitely generated
as an A-module.

Proof. (i) = (ii): if x is a root of monic polynomial f of degree n, then also of

xr .f for each r 0. Thus xn+r is in the A-module generated by 1, x, . . . , xn+r1
for each r 0. Inductively we see that A[x] is generated by 1, . . . , xn1 .
(ii) implies (iii): is clear: A[x] is a subring of B with the required properties.
(iii) implies (iv): Let C be the subring of B given by (iii). We may consider C
is a finitely generated A-module by definition. Moreover if x AnnA[x] (C) then
x.1 = 0 so x = 0.
(iv) implies (i): Let m1 , . . . , mn be a generating set
Pnfor M as an A-module. Then
certainly we can find aij in A such that x.mi =
. Thus the matrix
j=1 aij mjP
B = (bij ) with A[x] coefficients given by xI (aij ) satisfies nj=1 bij mj = 0, it
follows (by Cramers rule) that det B acts as 0 on each mj , and so on the whole
of M . But in A[x] only 0 can act as 0 on M so det B = 0. But det B is a monic
polynomial in x with coefficients in A, and so x is integral over A as required. 

38

COMMUTATIVE ALGEBRA

Lemma. If A B C is a chain of subrings such that B is a finitely generated

A-module and C is a finitely generated B-module, then C is a finitely generated
A-module.
Proof. If b1 , . . . , bm is a generating set for B as an A-module and c1 , . . . , cn is
a generating set for C as a B-module then it is easy to check that {bi cj } is a
generating set for C as an A-module.

Corollary. The set C of elements of B that are integral over A is a subring of B.
Definitions.
We call the ring C the integral closure of A in B.
We say A is integrally closed in B if A = C.
We say B is integral over A if C = B.
Proof of Corollary. It suffices to show that if x, y are in C then xy and xy are in C.
Suppose that x, y are in C and consider the chain of rings A A[x] A[x][y] B.
We have A[x] is a finitely generated A-module. Moreover as y is integral over A
then also over A[x], and so A[x][y] is finitely generated as an A[x]-module. It is
now follows from the lemma that A[x][y] is a finitely generated A-module. Thus
every element of A[x][y] is integral over A by the lemma. In particular x y and
xy are integral.

Definition. We say that an integral domain A is integrally closed or normal if it
is integrally closed in its field of fractions, Q(A).
In particular the ring of integers OK of a number field is integrally closed.
Geometrically, an integral domain A is normal means that the singularities of
Spec(A) all lie in codimension 2. In particular if Spec(A) is an algebraic curve then
A is normal means that the curve is non-singular.
Exercise. Suppose A is a subring of B and x1 , . . . , xn in B are each integral over
A, then A[x1 , . . . , xn ], the subring of B generated by A and each xi is finitely as an
A-module. [Hint the result for n = 2 is the content of the proof of the last result].
Corollary. If A B C is a chain of subrings and C is integral over B and B
is integral over A then C is integral over A.
Proof. Suppose x C then as x is integral over B we may find an equation xn +
b1 xn1 + + bn = 0 with all the bi in B. If we let D be the subring of C generated
by A, b1 , . . . , bn we see that D is a finitely generated A-module by the exercise.
Moreover, D[x] is a finitely generated D-module since x is integral over D by the
equation. So by the lemma above D[x] is a finitely generated A-module. Thus x is
integral over A as required.

Corollary. If A is a subring of B then the integral closure of A in B is integrally
closed in B.
Proof. Let C be the integral closure of A in B and suppose x in B is integral over
C. Then C[x] is integral over A by the above and so x is in C.

Exercise. Suppose that A is a subring of B, and B is integral over A. Then
(1) If I is an ideal in B then B/I is integral over A/(I A).
(2) If S is a m.c. subset of A then BS is integral over AS .

COMMUTATIVE ALGEBRA

39

In fact we can say more than this:

Proposition. If A is a subring of B, S is a m.c. subset of A and C is the integral
closure of A in B then CS is the integral closure of AS in BS .
Proof. By the exercise we have CS is integral over AS .
Suppose that an element x in BS is integral over AS so we have an equation
xn + r1 xn1 + + rn = 0 BS
with ri AS .
Multiplying through by the product s of the denominators of x and the ri we
see that (y) = sx satisfies a monic polynomial with coefficients in (A). So there
is a t S such that ty satisfies a monic polynomial with coefficients in A. Thus ty
is integral over A and so in C and x = (st)1 (y) CS as required.

Now, we can show that to be integrally closed is a local property:
Theorem. If A is an integral domain then the following are equivalent:
(i) A is integrally closed;
(ii) AP is integrally closed for every prime ideal P in A;
(iii) Am is integrally closed for every maximal ideal m in A.
Proof. Notice that Q(A) = Q(AP ) for every prime ideal P , so (i) implies (ii) follows
from the above proposition.
(ii) implies (iii) is trivial as usual.
For (iii) implies (i), let C be the integral closure of A in Q(A) and let : A C
be the inclusion. For each maximal ideal m in A, Cm is the integral closure of Am in
Q(A) so by the above proposition again, we have m : Am Cm is an isomorphism
by assumption. Since isomorphism is a local property of module maps we can
deduce that is also an isomorphism and A = C as required.

8.2. Going-up and Going-down Theorems.
Proposition. Suppose that A B are rings with B integral over A. Let p
Spec(B) and P = p A. Then P is maximal if and only if p is maximal. That is
the natural map Spec(B) Spec(A) sends closed points to closed points and the
fibres of a closed point are all closed.
Proof. By a previous exercise we know that B/p is integral over A/P . So we may
assume that A and B are integral domains, p = 0 and need to show that A is a
field if and only if B is a field.
Suppose that A is a field and b B is non-zero. We want to show that b is a
unit in B. We know that b satisfies a polynomial
bn + a1 bn1 + + an = 0
with ai A. Since B is an integral domain by dividing by b if necessary, we may
assume that an 6= 0. But then (bn1 + a1 bn2 + an1 )b = an is a unit in A
thus b is a unit in B.
Conversely, suppose that B is a field, and x A is non-zero. Now x is invertible
in B so we may write
xm + a1 xm+1 + + am = 0
with ai A. Thus x1 = (a1 + a2 x + am xm1 ) A as required.

40

COMMUTATIVE ALGEBRA

Corollary. If A B are rings with B integral over A then the natural map
: Spec(B) Spec(A) preserves strict inclusions and is surjective.
Proof. Let P be in Spec(A). Then BP = AP A B is integral over AP and AP has
a unique maximal ideal PP . Thus by the proposition above, P (pP ) = PP if and
only if pP is a maximal ideal of BP . Thus for p Spec(B), (p) = P if and only
if pP is a maximal ideal in BP .
Both parts follow easily since Spec(BP ) is in natural order-preserving 1-1 correspodance with points of Spec(B) that meet A\P trivially: for the first, if p q
are in Spec(B) and both map to P in Spec(A) then pP and qP are both maximal
and so equal. Hence p = q. Finally, for each P Spec(A), BP has a maximal ideal
which lifts to a prime ideal p Spec(B) and then (p) = P by construction.

Suppose that we have a ring homomorphism f : A B.
Definition. The going-up theorem is said to hold for f if for every pair of primes
P Q in Spec(A) and every p Spec(B) such that f (p) = P there is q Spec(B)
such that p q and f (q) = Q.
Definition. Similarly the going-down theorem is said to hold for f if for every pair
of primes P Q in Spec(A) and every q Spec(B) such that f (q) = Q there is
p Spec(B) such that p q and f (p) = P .
Theorem. If A B are rings and B is integral over A then the going-up theorem
holds for the natural inclusion map.
Proof. Let P Q in Spec(A) and p Spec(B) with (p) = P . By the exercise
above B/p is integral over A/P . Thus by the corollary above, there is q/p
Spec(B/p) such that (Q/P ) = q/p. The result follows easily.

The proof of the going-down theorem is a little more subtle, and depends on the
following fact that we will not prove:
Theorem. Suppose that A is an integrally closed integral domain, F is a normal
field extension of Q(A) (in the sense of Galois theory) and B is the integral closure
of A in F . Then Aut(F/Q(A)) acts transitively on {p Spec(B)| (p) = P } for
each P Spec(A).
Proof. Omitted

Theorem. If A B are rings, A is integrally closed and B is integral over A then

the going-down theorem holds for the natural inclusion map.
Proof. Let P Q in Spec(A) and q Spec(B) with (q) = Q.
Let F be the normal closure of Q(B) over Q(A) and let C be the integral closure
of A in F . Since C is integral over A, : Spec(C) Spec(A) is surjective, so there
is an X Spec(C) such that X A = (X) = P . Since also B is integral over
A, there is a Y Spec(C) such that Y B = q. By the going-up theorem for the
exentension A C we may also find Z Spec(C) such that Z X and Z A = Q.
Now Z A = Y A = Q so by the unproved theorem the is a Aut(F/Q(A))
such that (Z) = Y . Now (Z) (X) and so q = (Z) B (X) B. Thus if
we take p = (X) B we have p A = (X) A = X A = P and we are done. 

COMMUTATIVE ALGEBRA

41

8.3. Dimension. We are now ready to discuss the (Krull) dimension of a ring.
Definition. Given a ring A and P Spec(A) we define the height of P to be
ht(P ) = sup{n|there is a chain in Spec(A) P0 ( P1 ( ( Pn = P }
Examples.
(1) A minimal prime ideal has height zero.
(2) If A is an integral domain then a minimal non-zero prime has height one.
Definition. We define Krull dimension of A by
K-dim(A) := sup{ht(P )|P Spec(A)}.
Remark. Notice that we always have ht(P ) = K-dim(AP ).
Well see later that if A is Noetherian then every prime ideal has finite height
but that it is not true that every Noetherian ring has finite Krull dimension.
Geometrically, the height of a prime ideal P corresponds to the maximal length
of a chain of irreducible subvarities of Spec(A) containing V (P ). Thus we think
of ht(P ) as being the codimension of P in the largest irreducible component of
Spec(A) containing P .
Notice that Example Sheet 2 Q11 says a Noetherian ring has Krull dimension 0
if and only if it is Artinian.
Lemma. If A B are rings and B is integral over A then K-dim(B) = K-dim(A).
Proof. Weve already seen that : Spec(B) Spec(A) is strictly order-preserving
when B is an integral extension of A and so K-dim(A) K-dim B. Conversely, it
follows from the Going-up theorem and the surjectivity of for integral extensions
that every chain of prime ideals in A can be lifted to a chain of prime ideals in B;
so K-dim(B) K-dim(A).