Commutative Algebra III

1
Craig Huneke

April 16, 2013

1
Notes taken by Arindam Banerjee, Alessandro De Stefani, and Branden Stone in
the Fall of 2011 at the University of Kansas
Contents

1 Hilbert Functions and Multiplicities 1

1 Hilbert Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 The Hilbert-Samuel Polynomial . . . . . . . . . . . . . . . 4
2 Multiplicities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3 Superficial Elements . . . . . . . . . . . . . . . . . . . . . . . . . 14
4 Integral Closure of Ideals . . . . . . . . . . . . . . . . . . . . . . 21
5 Associated Graded Ring and Rees Algebra . . . . . . . . . . . . . 26
5.1 Equations defining Rees Algebras . . . . . . . . . . . . . . 26
6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2 Grothendieck Groups 32
1 Basic Lemmas and Remarks . . . . . . . . . . . . . . . . . . . . . 32
1.1 Structure of One Dimensional Local Complete Domains . 42
2 Class Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3 Divisors attached to Modules . . . . . . . . . . . . . . . . . . . . 48
4 Construction and Properties of K0 (R) . . . . . . . . . . . . . . . 56
5 Construction and Properties of K1 (R) . . . . . . . . . . . . . . . 62
6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

3 The Module of Differentials 65

1 First construction of the Module of Differentials . . . . . . . . . . 66
2 Second construction of the Module of Differentials . . . . . . . . 70
3 The Jacobi-Zariski Sequence . . . . . . . . . . . . . . . . . . . . . 73
4 Quasi-unramified maps . . . . . . . . . . . . . . . . . . . . . . . . 77
5 Quasi-smooth maps . . . . . . . . . . . . . . . . . . . . . . . . . 84

4 Basic element theory 89

1 Basic sets and basic elements . . . . . . . . . . . . . . . . . . . . 89
2 Basic elements and Projective modules . . . . . . . . . . . . . . . 96

Bibliography 104
CONTENTS iii

This course will cover a selection of basic topics in commutative algebra.

I will be assuming knowledge of a first course in commutative algebra, as in
the book of Atiyah-MacDonald [1]. I will also assume knowledge of Tor and
Ext. Some of topics which will covered may include Cohen-Macaulay rings,
Gorenstein rings, regular rings, Gröbner bases, the module of differentials, class
groups, Hilbert functions, Grothendieck groups, projective modules, tight clo-
sure, and basic element theory (see [4]). Eisenbuds book [3], the book of Bruns
and Herzog [2], and Matsumuras book [5] are all good reference books for the
course, but there is no book required for the course.
iv CONTENTS
Chapter 1

Hilbert Functions and

Multiplicities

Through out
L these notes, we will need the concept of a graded object. A graded
ring R = i>0 Ri is a commutative ring with identity, decomposed as a direct
sum of abelian groups with
Ri · Rj ⊆ Ri+j .
In particular, each Ri is an R0 -module L
and R0 is a commutative ring itself
(1 ∈ R0 ). Likewise, a graded module M = i∈Z Mi is an R-module, decomposed
as a direct sum of abelian groups with

Ri · Mj ⊆ Mi+j .

Each Mn is an R0 -module.
If R is a graded ring and M , N are graded R-modules, then an R-homomorphism
f : M → N is said to be homogeneous of degree k if f (Mn ) ⊆ Mn+k for all n.
Homogeneous maps are very desirable, so we define a convention to transform
graded maps in to homogeneous maps. By twist, denoted M (n), we mean a new
graded module (the same as M with out grading), but

M (n) := Mi+n .

Example 1. Let R be a graded ring. The new ring R(n) is a graded free
module, isomorphic to R, but has a generator in degree −n. So, R(−n) has a
generator in degree n.

1 Hilbert Functions
In the following, we denote the length of an R-module M by λR (M ). For more
information on length, see [1].
2 Hilbert Functions and Multiplicities

Definition. Let R is a graded ring and M a graded R-module with finite length.
The Hilbert series of M is
X
HM (t) := λR0 (Mi )ti .
i∈Z

1.1 Examples
Before we begin an in-depth study of the Hilbert series, we consider a few
examples.
Example 2. Let R = k[x1 , . . . , xd ] be a polynomial ring over a field k with
deg(xi ) = 1 for all i = 1 . . . d. Here we have that
Rn = khmonomials of degree ni
X
= khxa1 1 . . . xadd | ai = ni.
It is a standard fact that the length of a vector space over

a field is the same as
the vector space dimension. Thus, dimk (Rn ) = n+d−1 d−1 and the Hilbert series
is given by
∞  
X n+d−1 i
HR (t) = t.
i=0
d−1
Example 3. Let k[x, y] be a polynomial ring over a field k and define
R = k[xl , xl−1 y, . . . , xy l−1 , y l ] ⊆ k[x, y].
Assume that each element of {xl , xl−1 y, . . . , xy l−1 , y l } has degree one. Notice
that  
il + 2 − 1
dimk (Ri ) = = il + 1,
1
which gives us a Hilbert series of
∞
X
HR (t) = (il + 1)ti .
i=0

Further, we are able to write HR (t) as a rational function by using differentia-

tion:
X∞ ∞
X
i
HR (t) = l(i + 1)t − (l − 1) ti
i=0 i=0
X 0 X
=l ti − (l − 1) ti
 0
1 l−1
=l −
1−t 1−t
l l−1
= −
(1 − t)2 1−t
(l − 1)t + 1
=
(1 − t)2
1 Hilbert Functions 3

Example 4. Let R be a hypersurface of degree m. That is, let f ∈ Sm where

S = k[x1 , . . . , xd ] and deg xi = 1. Now Set R = S/(f ). To find HR (t), consider
the short exact sequence
·f
0 / S(−m) /S / S/(f ) / 0.

Notice that multiplication by f is a degree zero homomorphism. Since length

is additive, the Hilbert series is

HR (t) = HS (t) − HS(−m) (t)

1 tm
= d
− .
(1 − t) (1 − t)d
To see the Hilbert series of S(−m), notice that
X
HS(−m) (t) = dimk S(−m)i ti
i∈Z
X
= dimk Si−m ti
i∈Z
X
= dimk Sj tj+m
j=i−m∈Z

= tm HS (t)

Remark 1. Suppose that S = k[x1 , . . . , xn ] with the usual grading. Let f1 , . . . , fr

be a regular sequence with deg(fi ) = di . It is natural to guess that

(1 − tdi )
Q
HS/(f1 ,...,fr ) (t) = .
(1 − t)n
Notice that if r = n, then S/(f1 , . . . , fn ) has finite length. Then HS/(f1 ,...,fn ) (t)
is a polynomial! (Total length being the product of the degrees of fi .) Trying
the guess we find that
n
(1 − tdi ) Y
Q
HR (t) = = (1 + t + · · · + tdi−1 ).
(1 − t)n i=1

So, HR (1) = d1 d2 · · · dn .
Example 5. Let R = k[x, y] be a polynomial ring over a field k in two variables.
Let deg(x) = 2 and deg(y) = 3. Calculating the length of each graded piece
gives us the following:
i= 0 1 2 3 4 5 6 7 8 9 10 11 12 13
dimk (Ri ) = 1 0 1 1 1 1 2 1 2 2 2 2 3 2
Given the short exact sequence

0 / R(−2) ·x /R / k[y] / 0,
4 Hilbert Functions and Multiplicities

we have that HR (t) = t2 HR (t) + Hk[y] (t). Since Hk[y] = 1/(1 − t3 ), we have
1
HR (t) = .
(1 − t2 )(1 − t3 )
Remark 2. Consider R = k[x, y]/J where J is the set of all differences of poly-
nomials of the same degree. Notice that dimk (Ri ) = 1 for all i > 0. It can be
shown that
k[x, y]
R= 3 ' k[t2 , t3 ].
(x − y 2 )
Example 6. Another type of grading is multi-degree. For an example of this,
let R = k[x, y] be a polynomial ring in two variables. We let the variables have
the following degrees in N2 :
deg(x) = (1, 0)
deg(y) = (0, 1).
Here we have the R(i,j) = khxi y j i. In this case,
HR (t, s) = 1 + t + s + t2 + st + s2 + · · ·

1.2 The Hilbert-Samuel Polynomial

Before we define the Hilbert-Samuel polynomial, we need some propositions.
Proposition 1. Let R be a graded ring. The following are equivalent:
(1) R is noetherian;
(2) R0 is noetherian and R+ = ⊕i>1 Ri is a finitely generated ideal;
(3) R0 is noetherian and R ' R0 [x1 , . . . , xn ]/I with deg(xi ) = ki where I is
a homogeneous ideal.
Proof. (3) ⇒ (1): This follows from the Hilbert basis theorem.
(1) ⇒ (2): The object R+ is an ideal and hence is finitely generated as R is
noetherian. We thus have that R0 ' R/R+ is noetherian as well.
(2) ⇒ (3): Choose z1 , . . . , zn ∈ R+ with deg(zi ) = ki such that R+ =
Rz1 + · · · + Rzn . We claim that R = R0 [z1 , . . . , zn ]. As we naturally have that
R ⊇ R0 [z1 , . . . , zn ], it is enough to show Ri ⊆ R0 [z1 , . . . , zn ] for all i. To do this
we induct on i. The i = 0 is clear. Let i > 0 and suppose the claim is true up
to i − 1. Let f ∈ R+ and write it as
n
X
f= sj zJ
j=1

where sj ∈ R. If we restrict to the degree i part,

n
X
f= s0j zJ
j=1

where s0j ∈ Ri−kj . By induction we are done because s0j ∈ R0 [z1 , . . . , zn ].

1 Hilbert Functions 5

Remark 3. If M is an R-module, then M is naturally an R/ann(M )-module.

Proposition 2. Let R be a noetherian graded ring and M a finitely generated
graded R-module. Then for all n ∈ Z, Mn is a finitely generated R0 -module. In
particular, if R0 is artinian, then the length of Mn is finite.
Proof. Consider the submodule of M defined by M>n := ⊕i>n Mi . Since M is
noetherian, M>n is a finitely generated graded R-module. Therefore we have
that Mn := M>n /M>n+1 is also a finitely generated graded R-module. By
Remark 3, we have that Mn is a finitely generated graded R/ann(Mn )-module
as well. But R+ = ann(Mn ), hence Mn is a finitely generated R0 = R/R+ -
module.
Theorem 3. Let R be a graded noetherian ring with R0 artinian and let M be a
finitely generated graded R-module. Write R = R0 [z1 , . . . , zs ] with deg(zi ) = ki .
Then
fM (t)
HM (t) = Qs ki
i=1 (1 − t )
where fM (t) ∈ Z[t, t−1 ]. If M is non-negatively graded, then we have that
fM (t) ∈ Z[t].
Proof. By Proposition 2, the length of Mn is finite for all n. Induct on s. For
the s = 0 case, let R = R0 and assume that M is finitely generated over R0 .
Thus M = ⊕ri=−p Mi with λ(Mi ) < ∞ and
X
HM (t) = λ(Mi )ti ∈ Z[t, t−1 ].

Assume that s > 0 and consider the following exact sequence:

0 /K / M (−ks ) ·zs
/M /C /0

where K and C are the kernel and cokernel of the map defined by multiplication
by zs . Both K and C are finitely generated R-modules. Note that zs K = zs C =
0. Therefore K and C are modules over R0 [z10 , . . . , zs−1
0
] = R/zs R. We have
that
HM (t) + HK (t) = HM (−ks ) (t) + HC (t)
= tks HM (t) + HC (t).
Hence, we can solve for HM (t) to get
HC (t) − HK (t)
HM (t) =
(1 − tks )
fc (t) − fk (t)
= Qs ki
.
i=1 (1 − t )

If M is non-negatively graded, then C is non-negatively graded as well. Thus

fC (t) is in Z[t]. Also, M (−ks ) is non-negatively graded since ks > 0. Therefore
K is also non-negatively graded, thus fK (t) is also in Z[t].
6 Hilbert Functions and Multiplicities

Corollary 4. Suppose that in Theorem 3 we have ki = 1 for i = 1, . . . , s and

that M is non-negatively graded. Then there exists a polynomial, PM (x) in Q[x]
such that λ(Mn ) = PM (n) for all n >> 0. Moreover, deg(PM ) 6 s − 1.
Definition. The polynomial PM (x) is called the Hilbert-Samuel polynomial.
Proof. Note that we have for some f (t) in Z[t],
X f (t)
λ(Mi )ti = (1.1)
(1 − t)s
i>0
X s + i − 1
= f (t) ti (1.2)
s−1
i>0

If deg(f ) = N , we can write

f (t) = aN tN + · · · + ao

where ai ∈ Z. The coefficient of tn is a polynomial in n of degree s − 1 with

coefficients in Q. In particular,
N  
X s+n−j−1
λ(Mn ) = .
j=0
s−1

Here we have that

s−1
  z }| {
s+n−j−1 (s + n − j − 1)(s + n − j − 2) · · · (n − j + 1)
=
s−1 (s − 1)!
s−1
n
= + lower terms.
(s − 1)!
We now set
N  
X s+x−j−1
PM (x) = aj .
j=0
s−1

Example 7. Let R = k[x, y] and assume that deg(x) = 2 and deg(y) = 3. Here
we have that
1
HM (t) = = (1 + t2 + t4 + · · · )(1 + t3 + t6 + · · · )
(1 − t2 )(1 − t3 )
Thus the length is given by the coefficients of tn , that is,

λ(Rn ) = |{(a, b) | 2a + 3b = n}|

and this is not a polynomial but a quasi-polynomial.

1 Hilbert Functions 7

Fact. Recall that

 
x+i (x + i)(x + i − 1) · · · (x + 1)
= .
i i!
x+i
}∞


Then { i i=0 are a Q-basis of Q[x].
Example 8. Notice that
   
x x
x2 = 2 + .
2 1
We can use this to find the sum of the first n squares. That is,
n
X
1 + 2 2 + 3 2 + 4 2 + · · · + n2 = x2
x=1
n 
 X n  
x X x
=2 +
x=1
2 x=1
1
   
n+1 n+1
=2 +
3 1
2n(n − 1)(n − 2) 3n(n − 1)
= +
6 6
(2n + 1)(n(n + 1))
= .
6
Pn
Remark 4. We have j=0 j+k = n+k+1



k k+1 .
Remark 5. Notice that x+i


i : i ≥ 0 is a Q-basis of Q[x].
Remark 6. If f (x) ∈ Q[x] and we write:
n  
X x+j
f (x) = bj
j=0
j

and and we assume bn 6= 0, then f (x) is a polynomial of degree n with leading

coefficient bn!n . Moreover, if we set:
s
X
g(s) := f (i),
i=0
bn
then g(s) is a polynomial of degree n + 1 with leading coefficient (n+1)! .

Proof. Clearly deg f = n and the leading coefficient is bn!n . Consider now
s s X n   X n s   X n  
X X i+j X i+j s+j+1
g(s) = f (i) = bj = bj = bj ,
i=0 i=0 j=0
j j=0 i=0
j j=0
j+1
bn
which is now a polynomial of degree n + 1 and leading coefficient (n+1)! .

Remark 7.
Let f (x) ∈ Q[x] be a polynomial of degree n. Write f (x) =

P n x+j
j=0 bj j . If f (m) ∈ N for all m >> 0, then bj ∈ Z for all j and bn > 0.
8 Hilbert Functions and Multiplicities

2 Multiplicities
Throughout this section let (R, m, k) be a noetherian local ring, where m denotes
the
√ unique maximal ideal and k = R/m. Let I ⊆ R be an m-primary ideal (i.e.
I = m) and let M ∈ Modfg (R). Define
M In
grI R =: G = ,
I n+1
n≥0

where I 0 = R, which is called the associated graded ring of R with respect to I.

It is a ring with the product
I n /I n+1 × I m /I m+1 → I n+m /I n+m+1
(r∗ , s∗ ) 7→ r∗ s∗

on the graded components.

Remark 8. G = G0 [G1 ] is a noetherian graded ring. This is because G0 = R/I is
artinian, hence noetherian. Also R is noetherian, therefore I is finitely generated
and so the ideal G+ = ⊕n≥1 I n /I n+1 = G · G1 is finitely generated too. We
conclude by Proposition 1.
Definition. Let I ⊆ R and M be as above. Define
M I nM
M(I) := .
I n+1 M
n≥0

This is a graded G-module generated in degree zero, so it is finitely generated:

M(I) = G (M(I)0 ) .


Corollary 5. λ I n M/I n+1 M = Q(n) is a polynomial in n (for n >> 0)
of degree at most µ(I) − 1, where µ(I) = λ(I/mI) is the minimal number of
generators of the ideal I.
Proof. Since G is generated in degree one, by Corollary 4 the degree of the
polynomial is at most µ(G1 ) − 1. But

µ(G1 ) = µ(I/I 2 ) = µ(I),

where the last equality follows by NAK (Nakayama’s Lemma).

Corollary 6. With the same assumptions as above we have that λ(M/I n M )
is a polynomial in n, for n >> 0, having positive leading coefficient and degree
bounded by µ(I).
Proof. Notice that
n−1
IjM
X  
n
λ(M/I M ) = λ .
j=0
I j+1 M

Then we conclude by Remark 7 and Corollary 5.

2 Multiplicities 9

Definition. We set PI,M (n) to be the polynomial such that

 
M
PI,M (n) = λ for n >> 0
I nM

and we call it the Hilbert polynomial of M with respect to I.

Theorem 7. deg PI,M = dim M .

Before proving it we need some more results.

Lemma 8. Let the notation be as above and assume

0 /N /M /L /0

is a short exact sequence of finitely generated R-modules. Then

(1) deg PI,M = max {deg PI,N , deg PI,L }.

(2) deg (PI,M − PI,N − PI,L ) < deg PI,M .

Proof. (1) Tensor the short exact sequence with R/I n , then we get:

N / M / L /0
I nN I nM I nL

and hence      
M N L
λ ≤λ +λ .
I nM I nN I nL
This implies (for n >> 0):
       
M N L
deg PI,M = deg λ ≤ deg λ + λ =
I nM I nN I nL
      
N L
= max deg λ , deg λ = max {deg PI,N , deg PI,L } .
I nN I nL

Conversely there exists a short exact sequence:

/ N / M / L /0
0
I nM ∩ N I nM I nL

and also, by Artin-Rees Theorem, there exists k ∈ N such that for all n ≥ k

I n M ∩ N = I n−k (I k M ∩ N ) ⊆ I n−k N.

This implies    
N N
λ ≥λ
I nM ∩ N I n−k N
10 Hilbert Functions and Multiplicities

for n >> 0. Therefore, for such n:

         
M L N L N
λ =λ +λ ≥λ +λ .
I nM I nL I nM ∩ N I nL I n−k N
But PI,N (n − k) and PI,N (n) have same degree and leading coefficient (the first
is just a translation of the second one). Hence
       
M N L
deg PI,M = deg λ ≥ deg λ +λ =
I nM I n−k N I nL
      
N L
= max deg λ , deg λ = max {deg PI,N , deg PI,L } .
I n−k N I nL
Therefore (1) follows. For (2) we have
         
L N M L N
λ +λ ≤λ ≤λ +λ ,
I nL I n−k N I nM I nL I nN
therefore the leading coefficients have to cancel, i.e.

LC(PI,M ) = LC(PI,N + PI,L ),

where LC denotes the leading coefficient of a polynomial. This is of course

equivalent to (2).
We are now ready to prove Theorem 7.
Proof of Theorem 7. Take a prime filtration of M and apply Lemma 8. Then,
without loss of generality we can assume M = R/p, where p is a prime ideal.
Furthermore, passing from R to R/annM = R/p we can directly assume that
M = R is a domain. Set d = dim R and choose a sop (system of parameters)
J = (x1 , . . . , xd ) ⊆ I. Then there exists l ∈ N such that I l ⊆ J, and so

I ln ⊆ J n ⊆ I n for all n ∈ N.

This implies      
R R R
λ ≤λ ≤λ
In Jn I ln
and so
PI,R (n) ≤ PJ,R (n) ≤ PI,R (ln).
But deg PI,R (n) = deg PI,R (ln) since l ∈ N is just a constant, therefore deg PJ,R =
deg PI,R . But we know that

deg PI,R = deg PJ,R ≤ µ(J) = d.

Conversely, we induct on d. If d = 0 then clearly deg PI,R ≥ d. Let d > 0 and

pick x ∈ m, x 6= 0. Remember that R is a domain, hence we have a short exact
sequence:
0 /R x /R / R := R/xR / 0.
2 Multiplicities 11

Notice that deg(PI,R − PI,R − PI,R ) < deg PI,R by Lemma 8 (2), therefore

d − 1 = dim R = deg PI,R < deg PI,R ,

that is d ≤ deg PI,R .

√
Definition. Let (R, m) be a d-dimensional noetherian local ring, let I = m
be an m-primary ideal and let M ∈ Modfg (R). Then the multiplicity of M (with
respect to I) is
d! λ (M/I n M )
e(I; M ) := lim .
n→∞ nd
If M = R we will denote e(I) := e(I; R).
Notation. Given a function f we say that a function g is O(f ) if there exists
a constant C such that g(n) ≤ Cf (n) for n >> 0.
Remark 9. e(I; M ) ∈ N and e(I; M ) = 0 if and only if dim M < dim R.
Proof. Set s := dim M and d := dim R. In general s ≤ d. Also
 
M bs
λ n
= ns + O(ns−1 )
I M s!
with bs > 0, and therefore
d! λ(M/I n M ) d! bs d−s
= n + O(nd−s−1 ).
nd s!
Taking the limit shows that

0 s<d
e(I; M ) =
bs s=d

Remark 10. Let t ∈ N, t ≥ 1. Then e(I t ; M ) = e(I; M )td .

Proof. Assume dim M = dim R = d. We have seen that
 
M e(I; M ) d
λ = n + O(nd−1 ),
I nM d!
hence
e(I; M )td d
 
M e(I; M )
λ = (tn)d + O(nd−1 ) = n + O(nd−1 ).
(I t )n M d! d!
This means
e(I t ; M ) = e(I; M )td .
If dim M < dim R, then the equality still holds since e(I t ; M ) = e(I; M ) =
0.
12 Hilbert Functions and Multiplicities

Remark 11. e(I; M ) is additive on short exact sequences because of Lemma 8

(2).
√
ˆ where Iˆ = I R̂, since I = m and so R/I n ' R̂/Iˆn .
Remark 12. e(I) = e(I),
Theorem 9. If x1 , . . . , xd is a regular sequence, then
R
gr(x) R ' [T1 , . . . , Td ].
(x)
n+d−1


Equivalently, (x)n /(x)n+1 is a free R/(x)-module of rank d−1 .
For the proof see Corollary 22.
Remark 13. If x1 , . . . , xd is an sop for R as above and it is a regular sequence
(this is equivalent to say that R is Cohen-Macaulay), then:
 
R
e((x)) = λ .
(x)
Moreover, if x1 , . . . , xd is just an sop, then
 
R
e((x)) ≤ λ .
(x)
Proof. By Theorem 9 we have
n−1  n−1
(x)i
   X  R  i + d − 1
R X
λ = λ = λ =
(x)n i=0
(x)i+1 i=0
(x) d−1
     d 
R n+d R n d−1
=λ =λ + O(n ) .
(x) d (x) d!
Hence    
d! R R
e((x)) = lim λ =λ .
n→∞ nd (x) (x)
For the general inequality notice that in any case there exists a surjective ring
homomorphism

B = R/(x)[T1 , . . . , Td ]  gr(x) R = G
Ti 7→ xi + (x)2

and therefore
 
R λ(Bn )(d − 1)! λ(Gn )(d − 1)!
λ = lim ≥ lim .
(x) n→∞ nd−1 n→∞ nd−1

In this case we multiplied by (d − 1)! because we know that if λ(B/Bn ) is

eventually a polynomial of degree d, then λ(Bn ) is eventually a polynomial of
degree d − 1, and the leading coefficient, that gives the multiplicity,√doesn’t
change. Notice now that dim R = dim G. In fact, more generally for I = m,
2 Multiplicities 13

λ(I n /I n+1 ) is eventually the Hilbert Polynomial of the associated graded ring
Pn−1 i i+1
G = grI R of degree dim G − 1. But also, for n >> 0, i=0 λ(I /I ) =
n
λ(R/I ) = PI,R is a polynomial of degree both dim G by Remark 6 and dim R
by Theorem 7. Therefore, back to our case:
 
R λ(Gn )(d − 1)!
λ ≥ lim = e((x)).
(x) n→∞ nd−1
Pn−1
The last inequality is again because λ(R/(x)n ) = i=0 λ((x)i /(x)i+1 ) and so
λ((x)n /(x)n+1 ) = λ(Gn ) is eventually a polynomial of degree d − 1 that gives
the multiplicity of (x).
Remark 14. Let (R, m, k) be a RLR (regular local ring), then e(m) = 1.
Proof. By Remark 13 e(m) = λ(R/m) = dimk k = 1.
Definition. e(m) =: e(R) is often called the multiplicity of the ring R.
Remark 15. Let (R, m, k) be artinian, then e(R) = λ(R).
Proof. It follow also by Remark 13, but there is also an easy direct proof. Since
R is artinian we have mn = 0 for n >> 0, hence

e(R) = lim 0! λ(R)n0 = λ(R).

n→∞

Example 9. Let R = k[x, y]/(x2 , xy), then dim R = 1. Also λ(Rn ) = 1 for all
n ∈ N, therefore
0! λ(Rn )
e(R) = lim =1
n→∞ n0
but R is clearly not regular (it is not even Cohen-Macaulay). We will see that
the converse to Remark 14, that is e(R) = 1 ⇒ R is a RLR, holds if R is
unmixed, and it is a theorem of Nagata.
Theorem
√ 10 (Associativity formula). Let (R, m) be a local noetherian ring, let
I = m be an m-primary ideal and let M ∈ Modfg (R). Then
X
e(I; M ) = e(I; R/p)λRp (Mp ).
p ∈ SpecR
dim R/p = dim R

Proof. Take a prime filtration of M :

M0 ⊆ M1 ⊆ . . . ⊆ Ms = M,

with Mi+1 /Mi ' R/pi for some pi ∈ SpecR. Multiplicity is additive on short
exact sequences, hence
s−1
X
e(I; M ) = e(I; R/pi ).
i=0
14 Hilbert Functions and Multiplicities

We just have to count how many times each pi appears in the prime filtration.
First notice that e(I; R/pi ) = 0 unless dim R/pi = dim R, therefore
X
e(I; M ) = e(I; R/pi ).
{i: dim R/pi =dim R}

Fix a p ∈ SpecR such that dim R/p = dim R. Localizing at p we have

  
R 0 pi 6= p
=
pi p (R/p)p = k(p) pi = p

since p is minimal (dim R/p = dim R) and we cannot have pi ⊆ p. Here k(p) =
Rp /pRp = (R/p)p is the residue field of the localzation Rp . Now, localizing at
p the filtration:
(M0 )p ⊆ (M1 )p ⊆ . . . ⊆ Mp
gives us a composition series of the Rp -module Mp , and its length is both
λRp (Mp ) and the times p appears in the original filtration. Hence the asso-
ciativity formula follows.

3 Superficial Elements
Let (R, m, k) be a Noetherian local ring. Let f ∈ m be a non zero element. We
want to understand how do e(R) and e(R/(f )) relate to each other. To do this
we need several tools. the relation between G = grm (R), Ḡ = grm̄ (R̄) where
R̄ = R
f.

Definition. Let I ⊂ R be an ideal. The Rees ring of I is defined to be

M
R(I) := R ⊕ I ⊕ I 2 ⊕ I 3 . . . = I n.
n>0

Equivalently R(I) = R[It] as a subring of R[t]. Also, Proj(R(I)) is the blow up

of V (I) in Spec(R).
Remark 16. One can easily check that

R(I)
grI (R) ' R(I) ⊗R R/I ' .
IR(I)

Definition. Let (R, m, k) be a Noetherian local ring and let I ⊆ R be an ideal.

Let f ∈ R be a non zero element. Since R is local and Noetherian we have
n n
/ I n+1 .
T
n>0 I = (0), therefore there exists n ∈ N such that f ∈ I , but f ∈
Then the leading form of f in grI R is defined to be

f ∗ := [f ] ∈ I n /I n+1 ,

where [f ] denotes the equivalence class of f inside I n /I n+1 .

3 Superficial Elements 15

Example 10. Consider m = (x, y) ⊆ R = k[[x, y]], and f = x2 − y 3 ∈ R. Then,

grm (R) ' k[[x, y]] and f ∗ = (x2 − y 3 )∗ = x2 ∈ m2 /m3 . Notice that
R k[[x, y]]
= 2 ' k[[t2 , t3 ]].
(f ) x − y3
Set R̄ := R/(f ) and m̄ := m/(f ). One can prove that
k[[x, y]] grm (R)
grm̄ (R̄) ' ' .
(x2 ) (f ∗ )
Definition. Let (R, m, K) be a local ring. Set R(t) := R[t]mR[t] and m(t) :=
mR[t]mR[t] . Notice that
R(t)
' k[t]k[t]r{0} =: k(t),
m(t)
therefore (R(t), m(t), k(t)) has infinite residue field. Also R → R(t) is a faithfully
flat extension.
Definition. Let (R, m, k) be a Noetherian local ring, and let I ⊆ R be an ideal.
Then an element x ∈ I r I 2 is said to be a superficial element for I (of degree 1)
if there exists c ∈ N such that
(I n+1 : x) ∩ I c = I n for all n > c.
Proposition 11. Let (R, m, k) be a Noetherian local ring. Assume that R
contains k, and also that k is infinite. Let I ⊆ R be an ideal. Then
(i) Superficial elements exists
(ii) If x is superficial for I and furthermore x is a nonzero divisor in R, then
I n+1 : x = I n for all n >> 0.

Proof. (i) Set G := grI (R). Consider a partial primary decomposition of (0) in
G:
(0) = q1 ∩ q2 ∩ . . . ql ∩ J,
where qi are pi -primary ideals such that G+√6⊆ pi and J is the intersection of all
components containing a power of G+ (i.e. J = G+ ). Note that pi ∩ G1 6= G1 .
Therefore we can choose x ∈ I \ I 2 such that x∗ ∈ Γ/ 1 = is not in pi for all
i = 1, . . . , l.
Claim. x is superficial for I.
Proof of the Claim. Fix c ∈ N such that (G+ )c ⊆ J, and notice that
q1 ∩ . . . ql ∩ (G+ )c ⊆ q1 ∩ . . . ql ∩ J = (0).
Now we induct on n > c, and we want to prove that (I n+1 : x) ∩ I c = I n for
n > c. Notice that we always have the inclusion I n ⊆ (I n+1 : x) ∩ I c for all
n > c. For the other inclusion:
16 Hilbert Functions and Multiplicities

• If n = c we clearly have the other inclusion as well.

• If n > c, let y ∈ (I n+1 : x) ∪ I c . Since (I n+1 : x) ⊆ (I n : x), we have

y ∈ (I n : x) ∩ I c = I n−1 ,

where the last equality holds by induction. By way of contradiction assume

/ I n . Then looking at the initial forms of x and y we get
y∈

x∗ ∈ I/I 2 = G1 and y ∗ ∈ I n−1 /I n = Gn−1 ⊂ (G1 )c .

Observe that x∗ y ∗ = 0, since xy ∈ I n+1 . So

y ∗ ∈ (0 : x∗ ) = (q1 : x∗ ) ∩ . . . ∩ (ql : x∗ ) ∩ (J : x∗ ) ⊆ q1 ∩ . . . ∩ ql .

Therefore
y ∗ ∈ q1 ∩ . . . ∩ ql ∩ (G1 )c = (0),
which is a contradiction. So y ∈ I n and this proves the Claim.

(ii) Since x is a superficial element there exists c ∈ N such that

(I n+1 : x) ∩ I c = I n .

In particular
I n = (I n+1 : x) ∩ I c ⊆ (I n+1 : x).
Conversely, let y ∈ (I n+1 : x). Then

xy ∈ I n+1 ∩ (x) = ((x) ∩ I r )I n+1−r for n >> 0,

where the last equality follows from the Artin-Rees Lemma. Therefore xy = xzi
for some i ∈ I n+1−r and xz ∈ (x) ∩ I r . Since x is a nonzero divisor in R, we get

y = zi ∈ I n+1−r ⊂ I c for n >> 0.

Hence y ∈ (I n+1 : x) ∩ I c = I n .

Corollary 12. Let (R, m, k) be a Noetherian local ring and let I ⊆ R be an

ideal. let x ∈ R be superficial for I. Set G = grI (R) and Ḡ = grI/(x) (R/(x)).
Then there exits a natural map

G
→ Ḡ
(x∗ )

of degree 0, which is an isomorphism in all large degrees.

3 Superficial Elements 17

Proof. Since x is superficial there exists c ∈ N such that (I n : x) ∩ I c = I n−1

for all n > c. Hence for all n > 1 we have

xI n−1 ⊆ x(I n : x) = (x) ∩ I n .

Notice that clearly (G/(x∗ ))0 = R/I ' (Ḡ)0 . For all n > 1 we have a surjection

I n /I n+1
 
G
= '
(x∗ ) n (xI n−1 )/I n+1
In // In
' '
xI n−1 + I n+1 ((x) ∩ I n ) + I n+1
I n + (x)
'( ) = (Ḡ)n .
I n+1 + (x)

To get an isomorphism we need xI n−1 ⊇ (x) ∩ I n , since we have seen that the
other inclusion always holds. By Artin-Rees Lemma, there exists r ∈ N such
that for n > r + c
(x) ∩ I n = ((x) ∩ I r )I n−r ⊆ xI c .
Let y ∈ (x) ∩ I n , then y = xa for some a ∈ I c . But also (x) ∩ I n = x(I n : x),
hence y = xb for some b ∈ (I n : x). Therefore a − b ∈ (0 : x) ⊆ (I n : x), and
hence
a ∈ I c ∩ (I n : x) = I n−1
by superficiality (since n > r + c > c). So, for n >> 0 (more precisely for
n > r + c), the above map is an isomorphism.

Example 11. Let (R, m, k) be a regular local ring of dimension dim R = n.

Let G := grm (R) ' k[x1 , ..., xn ] and let f ∈ m be an non zero element. Define

ord(f ) := max{n ∈ N : f ∈ mn }.

Notice that f ∗ ∈ mn /mn+1 . Then

e(R/(f )) = ord(f ).

Proof. Notice that f ∗ is a nonzero divisor in G because it is a domain. Set

d := odd(f ). It is easy to prove that, being f ∗ a nonzero divisor, we have
mN : f = mN −d for all N > d. For N > d consider the exact sequence

N −d
0 / m / R f
/ R / R / 0,
mN mN mN mN + (f )

mN −d
       
R R N +n−1
so that λ =λ . But λ = , where
mN + (f ) mN mN n
18 Hilbert Functions and Multiplicities

n = dim R. Hence
     
R N +n−1 N −d+n−1
λ = − =
mN + (f ) n n

(N + n − 1) · . . . · N (N − d + n − 1) · . . . · (N − d)
= − =
n! n!
(
n−1 "  n−1 #)
n
 n

Nn 2 N Nn 2 − dn N
= + − + + O(N n−2 ) =
n! n! n! n!

dnN n−1 dN n−1

+ O(N n−2 ) = + O(N n−2 ).
n! (n − 1)!

Therefore the multiplicity is

   
R (n − 1)! R
e = lim λ = d.
(f ) N →∞ N n−1 mN + (f )

Definition. Let (R, m, k) be a Noetherian local ring of dimension dim R = d.

Let I ⊆ R be an m-primary ideal. Then we can write
     
R n+d n+d−1
PI (n) = λ = e0 (I) − e1 (I) + . . . + (−1)d ed (I).
mn d d

The integers ej (I), for j = 0, . . . , d are called the Hilbert coefficients of I. In

particular e0 (I) = e(I) is the multiplicity.
Proposition 13. Let (R, m, k) be a Noetherian local ring, and let I ⊆ R be
an m-primary ideal. Let x ∈ R be a superficial for I, which is also a nonzero
divisor on R. Set R̄ = R/(x) and I¯ = I/(x). Then , ej R (I) = ej R̄ (I)
¯ for all
0 6 j 6 d − 1.
Proof. Set G := grI (R), and Ḡ := grI¯(R̄). By Corollary 12, for n >> 0, there
is a short exact sequence:

/ Gn (−1) x∗ / Gn / Ḡn / 0.
0

Notice that
n−1 n−1
Ij
   
X X R
λ(Gj ) = λ =λ .
j=0 j=0
I j+1 In

Therefore, using the short exact sequence above, for n >> 0 we get:
    n−1 n−2 n−1  
R R X X X R̄
λ −λ = λ(Gj ) − λ(Gj ) = λ(Ḡj ) + C = λ + C,
In I n−1 j=0 j=0 j=0
I¯n
3 Superficial Elements 19

where C is a constant that depends on the fact that the above sequence is exact
only for n >> 0. Hence we have
    d   X d  
R R X n−j+d n−1−j+d
λ −λ = ej (I) − ej (I) =
In I n−1 j=0
d j=0
d

d−1  
¯ n − 1 − j + d + C.
X
= ej (I)
j=0
d−1

Notice that
     
n−j+d n−1−j+d n−1−j+d
− = ,
d d d−1

therefore we get the following equality of polynomials

d   Xd−1  
X n−1−j+d ¯ n−1−j+d
ej (I) = ej (I) +C
j=0
d−1 j=0
d−1

which implies that all the coefficients have to be the same. In particular ej (I) =
¯ for all j = 0, . . . , d − 1.
ej (I)

Proposition 14. Let R ⊆ S be an extension of local Noetherian rings (or

graded rings with R0 = k = S0 ) and mR (resp. mS ) be maximal ideals. (For
the graded case, let mR = R+ , and mS = S+ .) Further, let I be an mR -primary
ideal in R (homogeneous in the graded case). Assume that R is a domain and
that S is module finite over R. Let k = R/mR , L = S/mS , and F be the quotient
field of R. Set
r = rankR (S) = dimk S ⊗R F.
Then,
e(I; R) · r
e(IS; S) = .
[L : k]
Lemma 15. Given the notation in Proposition 14, if M is an S-module of
finite length, then λR (M ) = λS (M ) · [L : k].

Proof. Since length is additive and λR (L) = [L : k], we have that

λR (M ) = λS (M ) · λR (L) = λS (M ) · [L : k].

Proof of Proposition 14. We compute

eS (IS; S)nd
λS (S/I n S) = + O(nd−1 )
d!
20 Hilbert Functions and Multiplicities

where d = dim(S) = dim(R). On the other hand,

λR (S/I n S) = λS (S/I n S) · [L : k].
and
eR (I; S)nd
λR (S/I n S) = + O(nd−1 ).
d!
Comparing terms gives us that
eR (I; S) = eS (IS; S)[L : k].
It remains to prove that eR (I; S) = r · eR (I; R).
To prove this, chose an F -basis of S ⊗R F , say,
s1 sr
,..., .
1 1
Consider the R submodule of S, T = Rs1 + . . . + Rsr . Let φ be the surjection
of Rr onto T , sending the ith basis element to si for i = 1, . . . , r. Because
S ⊗R F = F r , φ ⊗ 1 is an isomorphism. Since ker(φ) and coker(φ) have smaller
dimension than d, we know that
eR (I; S) = eR (I; Rr )
= r · eR (I; R).

Theorem 16 (Nagata). Let (R, m, k) be a Noetherian, local, formally unmixed

ring of dimension dim R = d. Then e(R) = 1 if and only if R is regular.
Proof. If R is regular, then clearly e(R) = 1. To prove the converse let us
assume that R contains a field (even if the theorem holds in the general case).
Also, we can assume that the residue field is infinite passing to R(t) = R[t]mR[t] ,
and complete R, since the multiplicity doesn’t change, and if R(t) d is regular,
then R is regular as well. So without loss of generality R is a complete unmixed
local ring with infinite residue field. Using Associativity formula we have
X
1 = e(R) = e(R/p)λRp (Rp ).
p ∈ SpecR
dim R/p = dim R

Hence there exists a unique prime p such that dim R/p = dim R, and since R is
unmixed this is the only associated prime of R. Also λ(Rp ) = 1 implies that Rp
is a field, and hence R is a domain. Since |k| = ∞, choose a minimal reduction
(x1 , . . . , xd ) of m. Because R is complete, using Cohen structure theorem, we
have a finite extension
S = k[[x1 , . . . , xd ]] ⊆ R
and since both are domains, by Proposition 14, we have
1 = e(R) = e(S) · rankS (R) = rankS (R),
since S is regular, and so e(S) = 1. Therefore rankS (R) = 1, and hence R = S
is regular.
4 Integral Closure of Ideals 21

4 Integral Closure of Ideals

Definition. Let R be a ring and I an ideal of R. An element x is said to be
integral over I if x satisfies a monic equation

xn + i1 xn−1 + · · · + in = 0

such that ij ∈ I j . The set of all integral elements is called the integral closure
of I and is denoted I

Proposition 17. The integral closure of an ideal is an ideal

Proof. This is a corollary to Exercise 16

Example 12. Let R = k[x, y] be the polynomial ring in two variables over a
field k and I = (x2 , y 2 ). Here we have that xy ∈ I. To see this, notice that xy
satisfies the polynomial T 2 − x2 y 2 ∈ R[T ]. It is worth noting that x2 y 2 ∈ I 2 .

Example 13. Let R = CJx1 , . . . , xn K be the power series in n variables over

C and let f ∈ R such that f (0) = 0. Then f is always integral over its partial
derivatives  
∂f ∂f
,..., .
∂x1 ∂xn
Example 14 (Dedekind-Mertens). Let R be a commutative ring and f , g two
elements of R[t], i.e.

f (t) = an tn + · · · + a0 ;
g(t) = bm tm + · · · + b0 ,

If I is the content of the product f g, then ai bj is integral over I for all i, j. An

example of this is the following: let both f and g have degree one. We have that
I is the ideal generated by the coefficients of the product (a1 t + a0 )(b1 t + b0 ).
That is,
I = (a1 b1 , a0 b0 , a1 b0 + a0 b1 ).
Notice that
(a1 b0 )2 = (a1 b0 + a0 b1 )(a1 b0 ) − (a1 b1 )(a0 b0 ).
As (a1 b0 + a0 b1 ) ∈ I and (a1 b1 )(a0 b0 ) ∈ I 2 , we have that a1 b0 satisfies a degree
two monic polynomial in R[T ].
 
∂f ∂f
Open Question. In the context Example 13, is f ∈ m ∂x 1
, . . . , ∂xn ?

To see an example of this question, let f = x3 + y 4 in the ring R = CJx, yK.

Then
∂f ∂f
= 3x2 and = 4y 3 .
∂x ∂y
We thus see that x3 + y 4 ∈ m(x2 , y 3 ) ⊆ m(x2 , y 3 ).
22 Hilbert Functions and Multiplicities

Proposition 18. Let R be a noetherian ring and J ⊆ I ideals in R. The

following are equivalent:

(1) I ⊆ J;

(2) R(I) is module-finite over R(J);

(3) I n = JI n−1 over all n >> 0;

(4) there exists a k such that I n ⊆ J n−k for all n > k.

Further, if R is also local, the above are also equivalent to

(5) Let
R(I)
FI := = R/m ⊕ I/mI ⊕ I 2 /mI 2 ⊕ · · ·
mR(I)
and define A to be the subring of FI generated by

J + mI
⊆ I/mI
mI
over R/m. Then FI is module-finite over A.

Proof. (1) ⇔ (2): Use Exercise 16 and the fact that R(I) is a finitely generated
as a ring over R(J) since I is finitely generated.
(3) ⇒ (4): Fix k such that I n = JI n−1 for all n > k. By induction,

I k+l = JI k+l−1 = J 2 I k+l−2 = · · · = J l+1 I k−1 ⊂ J l+1 .

(4) ⇒ (2): Part (4) gives us that

1
R(J) ⊆ R(I) ⊆ R(J) · .
tk
since R(j) is noetherian, we have that R(I) is finitely generated as a R(J)-
module.
(2) ⇒ (3): As R(I) is module-finite over R(J), consider

R(I) = R ⊕ It ⊕ I 2 t2 ⊕ · · · ⊕ I N tN ⊕ · · ·
R(J) = R ⊕ Jt ⊕ J 2 t2 ⊕ · · · ⊕ J N tN ⊕ · · · .

Say the homogeneous generators u1 , . . . , us are up to degree N in t, that is,

deg(ui ) = di where di 6 N and

R(I) = R(J)u1 + · · · + R(J)us .

So, we have that

I L tL ⊆ (J L−d1 tL−d1 )u1 + · · · + (J L−ds tL−ds )us .

4 Integral Closure of Ideals 23

Since u1 ı ∈ Iid ti for all i = 1, . . . , s, we see that

I L ⊆ J L−d1 I d1 + · · · + J L−ds I ds .
But J ⊆ I, so for L larger than max{dk },
I L ⊆ JI L−1 ⊆ I L .
(2) ⇒ (18): Assume that (R, m) is local. As R(J) ⊆ R(I) is module-finite,
we have that
R(J) R(I)
A= ⊆ = FI
mR(I) ∩ R(J) mR(I)
is module-finite as well.
(18) ⇒ (3): Assume that (R, m) is local. Say FI is generated over A by a
finite number of elements up to degree k. Just as in the proof of (2) ⇒ (3), this
means for n > k,
In JI n−1 + mI n
= .
mI n mI n
Thus, we have that I n = JI n−1 + mI n and hence by Nakayama’s lemma we see
that I n = JI n−1 .
Definition. A local noetherian ring (R, m, k) is formally equidimensional if for
all minimal primes p of R,
b dim(R/p)
b = dim(R).
b
Theorem 19 (Rees). Let (R, m, k) be a formally equidimensional local noethe-
rian ring and J ⊂ I m-primary ideals. Then e(J) = e(I) if and only if I ⊂ J.
Remark 17. The fact that I ⊆ J implies equality of multiplicities does not
require formally equidimesional.
Remark 18. We will prove Theorem 19 only when R contains an infinite field.
The theorem is true otherwise, but is omitted from these notes.
Before we can prove Theorem 19, we will need the following lemma.
Lemma 20. Let (R, m) be a local noetherian ring of dimension d and suppose
that J ⊆ I. Then, I ⊆ J if and only if for all minimal primes p in R,
I +p J +p
⊆ .
p p
Proof. If we assume that I ⊆ J, we can use the same integral equation to obtain
the desired result.
Conversely, fix i in I and consider the multiplicatively closed subset of R,
W = {f (i) | f (t) = tm + j1 tm−1 + · · · + jm }.
Let Rad(R) denote the nilradical of R, that is, the intersection of all prime
ideals in R. If W ∩ Rad(R) 6= ∅, then let f (i) be an element of Rad(R). Thus
there exists an N such that f (i)N = 0 and hence i is integral over J.
If W ∩ Rad(R) = ∅, then there exists a prime q such that q ∩ W = ∅.
Therefore there exists a minimal prime p in R such that p ∩ W = ∅. Thus, i is
not an element of J + p/p, a contradiction.
24 Hilbert Functions and Multiplicities

Proof of Theorem 19. First assume that I ⊆ J. Thus there exists an l such that
for all n > l,
J n ⊆ I n ⊆ J n−l ⊆ I n−l .
Therefore
λ(R/J n−k ) ⊆ λ(R/I n ) ⊆ λ(R/J n )
and hence we have that e(I) = e(J).
Now assume that e(I) = e(J) and that R contains an infinite filed. Not that
e(I R)
b = e(J R). b ⊂ JR
b Further, by Proposition 18, I R b if and only if there exists
nb n−k b
an l such that I R ⊆ J R for all n > l. Thus,

I n = I nR
b ∩ R ⊆ J n−l R
b ∩ R = J n−l .

Applying the proposition once again yields the fact that I ⊆ J. So, without
losing any generality, R = R.
b
Using the associativity formula (Theorem 10),
X
e(I) = e(I; R/p)λ(Rp );
X
e(J) = e(J; R/p)λ(Rp ),

where p is a minimal prime in R such that dim(R/p) = dim(R). The fact

that J ⊆ I shows us e(J; R/p) > e(I; R/p). However, since R is formally
equidimensional, we have equality for all minimal primes p ∈ R. By Lemma 20,
if
I +p J +p
⊆
p p
for all minimal primes p in R, then we have that I ⊆ J. Hence, we may assume
R is a complete local domain.
We now turn to a reduction of the ideals I and J. We can replace J by
a system of parameters (x1 , . . . , xd ) ⊆ J. It is enough to show there exists a
system of parameters (x1 , . . . , xd ) ⊆ J such that J ⊆ (x1 , . . . , xd ). If so, then
we know by the easy direction that

e(x1 , . . . , xd ) = e(I).

But then, e(x1 , . . . , xd ) = e(I). If we prove that I ⊆ (x1 , . . . , xd ) then we are

able to deduce that I ⊆ J.
To do this, use Noether normalization on the fiber ring

R J J2
FJ = ⊕ ⊕ ··· .
m mJ mJ 2
Keep in mind that FJ is a finitely generated k algebra. By Noether’s normal-
ization,
k[x∗1 , . . . , x∗l ] ⊆ FJ ,
4 Integral Closure of Ideals 25

where x∗i are elements of J/mJ. But lifting x∗i to xi ∈ J, we may apply Propo-
sition 18 to see that J ⊆ (x1 ,p . . . , xl ).
Note that (x1 , .√. . , xl ) ⊆ (x1 , . . . , xl ) and hence (x1 , . . . , xl ) is m-primary
(this follows since J = m). Therefore, lgsd by Krull’s height theorem. But,
l = dim(FJ ) 6 dim(grJ R) = d,
and hence l = d. (To see this last fact, notice that grJ R/mgrJ R ' FJ .)
Thus, without losing any generality, we may assume that J is generated by
the system of parameters (x1 , . . . , xd ). Next we make I as simple as possible.
Choose any y ∈ I. It is enough to show that y ∈ J. Thus, we can replace I by
(x1 , . . . , xd , y). Note that in this case,
e(x1 , . . . , xd ) = e(x1 , . . . , x, y).
We are now able to make a further reduction on the rings. Since k ⊆ R, by
Cohen’s structure theorem, we can consider the extenstion
R ⊇ B = kJx1 , . . . , xd , yK
| |
kJx1 , . . . , xd K = A
where R is finite over kJx1 , . . . , xd K. We may assume that R = B. To see this,
let the maximal ideal of B be η = (x1 , . . . , xd , y)B and let r = rankB (R). We
know that
eR (I) = eR (ηR) = eB (η) · r
eR (J) = eR ((x1 , . . . , xd )R) = eB (x1 , . . . , xd ) · r
and hence we have that eR (x1 , . . . , xd ) = eB (η). So, if we prove the theorem
is true for R = B, we get that y ∈ (x1 , . . . , xd )B which implies that y ∈
(x1 , . . . , xd )R.
Now, R = kJx1 , . . . , xd , yK is a complete domain of dimension d, x1 , . . . , xd is
an system of parameters, J = (x1 , . . . , xd ), and I is the maximal ideal m. Note
that
R ' kJx1 , . . . , xd , T K/p
where p is a height one prime. But kJx1 , . . . , xd , T K is a unique factorization
domain and hence
p = (T l + a1 T l−1 + · · · + al ) =: (f ).
Now, e(m) = ord(f ) and e(x1 , . . . , xd ) = λ(R/(x1 , . . . , xd )). As
R kJT K
= ,
(x1 , . . . , xd ) Tl
we have that e(x1 , . . . , xd ) = l (here we are assuming that l = ord(f )). This
implies that ai ∈ (x1 , . . . , xd )i , otherwise ord(f ) < l. Since f (y) = 0, this shows
that y ∈ (x1 , . . . , xd ).
26 Hilbert Functions and Multiplicities

5 Associated Graded Ring and Rees Algebra

Throughout this section let R be a noetherian ring. Let I ⊆ R be an ideal, then
we have already defined
M
G := grI R = I n /I n+1
n≥0

the associated graded ring of R with respect to I. We have also defined

M
R(I) ' R[It] = I n tn ⊆ R[t]
n≥0

the Rees ring of I. Notice that

R(I)
G'
IR(I)

and often is more convenient to study R(I) instead of grI R.

5.1 Equations defining Rees Algebras

Let I = (x1 , . . . , xn ) ⊆ R. Then there exists a graded surjective map

ϕ : R[T1 , . . . , Tn ]  R(I)
Ti 7→ xi t

Hence ker ϕ ⊆ R[T1 , . . . , Tn ] is a homogeneous ideal, and if R itself is graded,

then ker ϕ is bigraded.
Remark 19. Set a := ker ϕ. Then a is homogeneous and it is generated by
homogeneous polynomials F (T1 , . . . , Tn ) (say deg F = d) such that

0 = ϕ(F (T1 , . . . , Tn )) = F (x1 t, . . . , xn t) = td F (x1 , . . . , xn ),

that is a = (F ∈ R[T1 , . . . , Tn ] : F (x1 , . . . , xn ) = 0).

Example 15. Let R = k[x, y] and let I = (x, y)2 = (x2 , xy, y 2 ). Then

a = (yT1 − xT2 , yT2 − xT3 , T22 − T1 T3 ).

Definition. Let ai = ”the ideal generated by all homogeneous polynomial of

degree at most i”.

In the previous example a1 = (yT1 − xT2 , yT2 − xT3 ) and a2 = a.

Remark 20. Since R is noetherian, there exists N >> 0 such that aN = a, since
a is finitely generated.

Definition. I is said to be linear type if a = a1 .

5 Associated Graded Ring and Rees Algebra 27

Example 16. Present I = (x1 , . . . , xn ) ⊆ R:

Rm
A / Rn /I /0

ei / xi
P
and let A = (aij ), so that xi aij = 0. Then it is easy to prove that
i

n
!
X
a1 = Ti aij : j = 1, . . . , n .
i=1

Definition. A sequence of elements x1 , . . . , xn is said to be a d-sequence if for

all i ≥ 0
((x1 , . . . , xi ) : xi+1 ) ∩ (x1 , . . . , xn ) = (x1 , . . . , xi ),
where we set x0 = 0.

Example 17. Any regular sequence is a d-sequence since for all i ≥ 0

(x1 , . . . , xi ) : xi+1 = (x1 , . . . , xi ).

Remark 21. If n = 1, then x1 = x is a d-sequence if and only if

0 : x = 0 : x2 .

In fact let x be a d-sequence, then (0 : x) ∩ (x) = 0. Clearly 0 : x ⊆ 0 : x2 .

Let ax2 = 0, then ax ∈ (0 : x) ∩ (x) = 0, that is a ∈ 0 : x. Conversely let
0 : x = 0 : x2 , and let bx ∈ (0 : x)∩(x). Hence bx2 = 0, that is b ∈ 0 : x2 = 0 : x.
So bx = 0 and x is a d-sequence.

Example 18. Let R = k[x, y]/(x2 , xy). Then y ∈ R is not a regular sequence,
but it is a d-sequence since 0 : y = 0 : y 2 = xR.

Definition. Let F ∈ R[T1 , . . . , Tn ] be a homogeneous polynomial. Then we say

that F has weight j if

F ∈ (T1 , . . . , Tj ) r (T1 , . . . , Tj−1 ).

We denote j = wt(F ).

Example 19. Let F = T12 + T1 T2 + T 4 + T4 T3 + T2 T6 ∈ R[T1 , . . . , T6 ]. Then

wt(F ) = 4 since
F ∈ (T1 , . . . , T4 ) r (T1 , . . . , T3 ).

Theorem 21. Let x1 , . . . , xn be a d-sequence and let I = (x1 , . . . , xn ). Then I

is linear type.

Proof. By induction on both degree and weight we will prove the following
28 Hilbert Functions and Multiplicities

Claim. If F (T1 , . . . , Tn ) is homogeneous of degree d ≥ 1 is such that F (x1 , . . . , xn ) ∈

(x1 , . . . , xj )R, then there exists G(T1 , . . . , Tn ) a form of degree d and weight at
most j such that F − G ∈ a1 .

Proof of the Claim. Assume d = 1. By assumption there exists ri ∈ R, for

1 ≤ i ≤ j, such that
j
X
F (x1 , . . . , xn ) = ri xi .
i=1

But deg F = 1, so set

j
X
G := ri Ti ,
i=1

so that (F − G)(x1 , . . . , xn ) = 0, that is F − G ∈ a1 and clearly wt(G) ≤ j.

Assume now d > 1 and induct on the weight of F . If wt(F ) ≤ j, then set
G = F , so that clearly F − G ∈ a1 . Then suppose wt(F ) = k > j and write

F = Tk F1 + F2 ,

where deg F2 = deg F1 + 1 = d and wt(F2 ) < k. Then

F (x1 , . . . , xn ) = xk F1 (x1 , . . . , xn )+F2 (x1 , . . . , xn ) ∈ (x1 , . . . , xj )R ⊆ (x1 , . . . , xk−1 )R.

Since F2 (x1 , . . . , xn ) ∈ (x1 , . . . , xk−1 ) we have that

F1 (x1 , . . . , xn ) ∈ ((x1 , . . . , xk−1 ) : xk ) ∩ (x1 , . . . , xn ) = (x1 , . . . , xk−1 )

since x1 , . . . , xn is a d-sequence. By induction on the degree there exists G1 of

degree d − 1 with wt(G1 ) ≤ k − 1 and such that F1 − G1 ∈ a1 . Set

F 0 := Tk G1 + F2 .

Then F − F 0 = Tk (F − G1 ) ∈ a1 . Note that F (x1 , . . . , xn ) = F 0 (x1 , . . . , xn ) ∈

(x1 , . . . , xj ) and wt(F 0 ) ≤ k − 1 since wt(G1 ), wt(F2 ) ≤ k − 1. By induction
on weight there exists G homogeneous of degree d, with wt(G) ≤ j, such that
F 0 − G ∈ a1 . Hence

F − G = (F − F 0 ) + (F 0 − G) ∈ a1 .

Finally, the Claim implies the theorem with j = 0: let F be a non zero
homogeneous polynomial of degree d ≥ 1 such that F ∈ a, i.e. F (x1 , . . . , xn ) =
0. Then there exists G homogeneous of degree d, with wt(G) ≤ 0, such that
F = G ∈ a1 . But wt(G) = 0, deg G = d ≥ 1 implies that G = 0. Therefore
F ∈ a1 .
5 Associated Graded Ring and Rees Algebra 29

Corollary 22. Let (R, m) be a noetherian local ring and let x1 , . . . , xn be a

regular sequence. Then
R
gr(x) R ' [T1 , . . . , Tn ].
(x)

Proof. If x1 , . . . , xn is a regular sequence, then it is a d-sequence. Also a1 is

generated by the Koszul relations xi Tj − xj Ti by Example 16. Therefore

R[T1 , . . . , Tn ]
R((x)) =  
x 1 . . . xn
2×2
T1 . . . Tn

Hence
R((x)) R[T1 , . . . , Tn ] R
gr(x) R ' '  ' [T1 , . . . , Tn ].
(x)R((x)) x 1 . . . xn (x)
(x) + 2 × 2
T1 . . . Tn
30 Hilbert Functions and Multiplicities

6 Exercises
(1) Assume R is a noetherian and that M , N are finitely generated graded
R-modules. Prove that Hom
(2) Suppose that S = k[x1 , . . . , xn ] with the usual grading. Let f1 , . . . , fr be
a regular sequence with deg(fi ) = di . Find HS/(f1 ,...,fr ) (t).
(3) If R = k[x1 , . . . , xn ] is a polynomial ring in n variables with deg(xi ) = di ,
then
1
HR (t) = Q .
(1 − tdi )
(4) Let R = ⊕i>0 Ri be a graded noetherian ring. Prove there exists and N
such that RN K = (RN )K for all K > 0. That is, the subring

R0 ⊕ RN ⊕ R2N ⊕ · · ·

is generated in degree N .
(5) Let R be as in Exercise 4. Then R is integral and finite over ⊕i>0 RN i .
(6) Let R = k[x1 , . . . , xn ] with deg(xi ) = ki . What is the least N that satisfies
Exercise 4?
(7) Assume G is a noetherian graded ring, (G0 , m0 ) is artinian local and G =
G0 [G1 ]. Let m := m0 ⊕ G1 ⊕ . . . as G-module. Then clearly G/m ' G0 /m0
is a field, and so m is maximal in G. Let R := Gm . Prove that

grm R = G.

(n.b. the fact that G is generated in degree one is crucial).

(8) Let (R, m) be a noetherian local ring and let I ⊆ R be an m-primary ideal.
Then
grI R ' grI R̂ R̂.

(9) Let (R, m, k) be a noetherian local ring, I ⊆ R an m-primary ideal. Prove

that:
• grI R is a domain ⇒ R is a domain.
• grI R is integrally closed ⇒ R is integrally closed.
• grI R is CM (Cohen-Macaulay) ⇒ R is CM.
• grI R is Gorenstein ⇒ R is Gorenstein.
Is it true that if grI R is a UFD then R is a UFD?
(10) Find an example (with prove) of a local domain such that the completion
(with respect to the maximal ideal) is not a domain. (Hint: By Exercises
8 and 9, if grm R ' grm̂ R̂ is a domain, then R̂ is a domain, so a possible
example has to be such that grm R is not a domain.)
6 Exercises 31

(11) Which Artinian local rings A with residue field C, can be embedded in
C[|t|]
(tn ) .
√
(12) If I = m, then e(I, R) = e(IR(t), R(t)).

(13) Complete the proof of the fact that superficial element exits by proving
there exists x ∈ I \ I 2 such that x∗ is not in pi for all i = 1, 2, ..., l.
(14) I ⊂ R, f ∈ R. Suppose f ∈ I d \ I d+1 .TFAE:
(a) For all n ≥ d, I n : f = I n−d .
(b) f ∗ is a NZD in grI (R).

(15) (R, m, k) local f ∈ md \ md+1 . Assume f ∗ is NZD in grm (R). Then

e(R/(f )) = de(R).
(16) Let J ⊆ I be ideal in a commutative ring R. Then I ⊆ J if and only if
R(I) is integral over R(J).
Chapter 2

Grothendieck Groups

Throughout this chapter R will always be a noetherian ring. Define H(R) to

be the free abelian group on the isomorphism classes of finitely generated R-
modules. Given M ∈ Modfg (R) denote < M > its class inside H(R), that is the
generator in H(R) corresponding to the isomorphism class of M . Define also
L(R) to be the subgroup of H(R) generated by elements

{< M > − < M1 > − < M2 >}

for which there exists a short exact sequence 0 → M1 → M → M2 → 0. Finally

we define the Groethendieck Group of R to be
H(R)
G0 (R) := .
L(R)

Give M ∈ Modfg (R) write [M ] for its class inside G0 (R).

Remark 22. If we have two short exact sequences

0 / M1 /M / M2 /0

and
0 / M1 /N / M2 / 0,

then we have [M ] = [N ] in G0 (R).

Remark 23. If we restrict to projective module and we do the same costruction,
we get the K-group K0 (R).
Question. Given M, N ∈ Modfg (R) when is [M ] = [N ]?

1 Basic Lemmas and Remarks

Lemma 23 (Filtration Lemma). Let M ∈ Modfg (R) and suppose

0 = M0 ⊆ M1 ⊆ . . . ⊆ M n = M
1 Basic Lemmas and Remarks 33

is a filtration. Then
n−1
X 
Mi+1
[M ] = .
i=0
Mi

Proof. By induction on n. The step n = 1 is trivial. Let n > 1 and consider

0 / M1 /M / M / 0,
M1

so that [M ] = [M1 ] + [M/M1 ]. Then we have a filtration

M1 M2 Mn M
0= ⊆ ⊆ ... ⊆ =
M1 M1 M1 M1
of length n − 1. By induction we get
n−1
X 
Mi+1
[M ] = [M1 ] + [M/M1 ] = .
i=0
Mi

Corollary 24. G0 (R) is generated by [R/p], for p ∈ SpecR.

Proof. Take a prime filtration of M and apply the Filtration Lemma.
Lemma 25 (Long Exact Sequence Lemma). Given an exact sequence

0 / Mn / Mn−1 / ... / M0 / 0,

then
n
X
(−1)n [Mi ] = 0.
i=0

Proof. By induction again.

Lemma 26 (Additive Map Lemma). Suppose ε : Modfg (R) → Z is an additive
function on short exact sequences. Then there exists an induced homomorphism
εe : G0 (R) → Z
Proof. Given M ∈ Modfg (R), the homomorphism εe is defined to be

εe(< M >) = ε(M )

on elements of the basis of H(R). Since ε is additive on short exact sequences

we have εe(L(R)) = 0, hence there is an induced map εe : G0 (R) → Z.
Example 20. Let R be a PID. Then

G0 (R) ' Z

generated by [R].
34 Grothendieck Groups

Proof. By Corollary 24 we can consider just [R/p], with p prime in R. If p = 0,

then [R/p] = [R]. If p 6= 0, then p = (x) for some x 6= 0, hence the following
sequence is exact

0 /R ·x /R / R/p / 0,

so that in G0 (R) we have [R/p] = 0. Therefore G0 (R) ' Z[R], but we still
have to prove that [R] is not a torsion element. Consider the rank function. By
Additive Map Lemma there exists a homomorphism

G0 (R) /Z

[M ]  / rank([M
g ])

which is surjective since [R] 7→ 1. So G0 (R) cannot be torsion and hence

G0 (R) = Z[R] ' Z.

Example 21. Let (R, m, k) be a regular local ring. Then G0 (R) = Z[R] ' Z.

Proof. Given M ∈ Modfg (R) there exists a free resolution

0 / Fn / Fn−1 / ... / F1 / F0 /M /0

and hence, by Long Exact Sequence Lemma,

n n
!
X X
i i
[M ] = (−1) [Fi ] = (−1) rankFi [R].
i=0 i=0

Therefore G0 (R) = Z[R] and, considering the rank function as in the previous
example, we get G0 (R) ' Z.

Example 22. Let (R, m, k) be an artinian local ring. Then G0 (R) = Z[k] ' Z.

Proof. By Corollary 24 we clearly get G0 (R) = Z[k]. Using the length function,
by Additive Map Lemma we get

G0 (R) /Z

[M ]  / λ([M
e ])

which is surjective since λ(k) = 1. Hence G0 (R) ' Z and moreover [R] =
λ(R)[k].

Question (H. Dao). Let (R, m, k) be a normal local ring, and assume k is an
algebraically closed field of characteristic zero, k ⊆ R. Also, assume G0 (R)⊗Z Q
is a finite dimensional Q-vector space. Does R have rational singularities?
1 Basic Lemmas and Remarks 35

Theorem 27. Let R be a noetherian ring and let I ⊆ R be nilpotent. Then the
map
j : G0 (R/I) / G0 (R)
 / [M ]
[M ]
is an isomorphism.
Proof. First

suppose we 2have shown the theorem when I 2 = 0.
Then G0 (R/I)

'
2 2
G0 R/I since in R/I clearly I = 0. Similarly G0 R/I 2 ' G0 R/I 4 and
so on. For some n ∈ N we have I n = 0, so that R/I n = R and hence we get the
following chain of isomorphisms:
' / G0 R/I 2 ' / G0 R/I 4 ' / ...... ' / G0 (R) .



G0 (R/I)

So suppose I 2 = 0. Notice that under this assumption, given M ∈ Modfg (R),

both IM and M/IM are R/I-modules. Consider the map i : G0 (R) →
G0 (R/I) given by
i([M ]) = [IM ] + [M/IM ].
Claim. i : G0 (R) → G0 (R/I) is well defined.
Proof of the Claim. On a basis {< M >: M ∈ Modfg (R)} of H0 (R) define

i0 : H0 (R) / G0 (R/I)

<M > / [IM ] + [M/IM ]

and extend it to a group homomorphism. We need to prove that i0 (L(R)) = 0.

Let
0 / M1 /M / M2 /0

be a short exact sequence of R-modules. Consider the following short exact

sequences of R/I-modules:

0 / M1 ∩ IM / IM / IM2 /0 (2.1)

0 / M1 +IM / M / M2 /0 (2.2)
IM IM IM

0 / M1 ∩IM / M1 / M1 /0 (2.3)
IM1 IM1 M1 ∩IM

0 / IM1 / M1 ∩ IM / M1 ∩IM /0 (2.4)

IM1

Then by (2.1) and (2.2) we get

 h i
M1 +IM M2
i0 (< M >) = [IM ] + M
  
IM = ([M1 ∩ M ] + [IM2 ]) + IM + IM2 =
M
1 +IM
= i0 (< M2 >) + [M1 ∩ IM ] +

IM .
36 Grothendieck Groups

Notice that (M1 + IM )/IM ' M1 /(M1 ∩ IM ). Finally, by (2.3) and (2.4) we
have
 
M1
i0 (< M >) = i0 (< M2 >) + [IM1 ] + = i0 (< M1 >) + i0 (< M2 >).
IM1

So i : G0 (R) → G0 (R/I) is well defined.

Now let [M ] ∈ G0 (R/I) and consider the composition (i ◦ j)([M ]):

(i ◦ j)([M ]) = i([M ]) = [IM ] + [M/IM ] = [M ]

since M ∈ Modfg (R/I) and hence IM = 0. Similarly, for [M ] ∈ G0 (R):

    
M M
(j ◦ i)([M ]) = j [IM ] + = [IM ] + = [M ]
IM IM

simply using the short exact sequence 0 → IM → M → M/IM → 0. So j is

an isomorphism.

Lemma 28 (Localization Lemma). Let R be a noetherian ring and W a mul-

tiplicatively closed set. Then there exists an exact sequence

α / G0 (R) β
/ G0 (RW ) /0
M
G0 (R/p)
W ∩p6=∅

where α and β are defined as follows:

 α / P[M (p)]
⊕[M (p)]

[M ]  / [MW ].
β

Proof. The first step is to show that

Im(α) = h[M ] | MW = 0i. (2.5)

If M is a R/p module and p∩W 6= ∅, then clearly MW = 0 since p ⊆ ann(M ).

Therefore
Im(α) ⊆ h[M ] | MW = 0i.
To see the other direction, suppose that MW = 0. Since M is finitely
generated, we know that

Supp(M ) = V (ann(M )).

In a prime filtration of M , say,

0 = M0 ⊂ M1 ⊂ · · · ⊂ Mn = M
1 Basic Lemmas and Remarks 37

where Mi+1 /Mi ' R/pi , we have that pi ∈ Supp(M ). Thus the annihilator of
M is contained in each pi . But MW = 0 implies there exists w ∈ W such that
wM = 0. In other words, w ∈ pi for all i and hence pi ∩ W 6= ∅ for all i as well.
This forces [R/pi ] ∈ Im(α) and by the Filtration lemma (Lemma 23) we have
that [M ] ∈ Im(α). This proves (2.5).
Next we show that β is surjective. This is clear since every finitely generated
RW -module is of the form MW for some finitely generated R-module M . (Note
that β is well-defined as localization is flat.)
Observe that Im(α) ⊆ ker(β) since MW = 0, that is, β([M ]) = 0. Abusing
notation, we now have the induced surjection

G0 (R) β
/ / G0 (RW ) .
Im(α)

We want to show there exists a splitting γ, that is, an inverse to β. Define

γ : H(RW ) / G0 (R)
Im(α)

by the following construction.

Let hN i ∈ H(RW ) and then choose an R-module M such that MW ' N .
Now define
γ(hN i) = [M ] + Im(α).
To show this is well-defined, suppose LW ' N . We need to prove that [M ]−[N ]
is an element of Im(α). Since R is noetherian and all modules are finitely
generated, we have that

(HomR (M, L))W ' HomRW (MW , LW ). (2.6)

h
As MW ' LW , choose some fixed RW -isomorphism g. We can write g = w
where h ∈ HomR (M, L). Replacing g by wg we get that
h
∈ HomRW (MW , LW )
1
is an isomorphism. Consider the following exact sequences:

0 / ker(h) /M / Im(h) /0

0 / Im(h) /L / coker(h) / 0.

In G0 (R), we have that

[M ] − [L] = [ker(h)] + [Im(h)] − [Im(h)] − [coker(h)] = [ker(h)] − [coker(h)].

But, (ker(h))W and (coker(h))W are zero since h1 is an isomorphism (use 2.5).
It follows that [M ] − [L] is in the image of α and thus γ is a well-defined map.
38 Grothendieck Groups

Next, we would like to show that L(RW ) ⊆ ker(γ). Let

f0 g0
0 / N1 /N / N2 /0 (2.7)

be a short exact sequence of RW -modules. We claim that

γ(hN i) = γ(hN1 i) + γ(hN2 i).

To see this, let M and M2 be finitely generated R-modules such that MW = N1

and (M2 )W = N2 . Using (2.6), choose g : M → M2 such that wg = g 0 where
w ∈ W . Let M1 := ker(g) and consider the short exact sequence

0 / M1 /M / Im(g) / 0. (2.8)

Since (Im(g))W = N2 , we have that

γ(hN i) = [M ] + Im(α);
γ(hN i2 ) = [Im(g)] + Im(α).

Thus, γ(hN i) = [M1 ] + Im(α) since the localization of (2.8) at W yields the
short exact sequence (2.7). Therefore, γ(hN i) − γ(hN1 i) − γ(hN2 i) is a coset
of [M ] − [Im(g)] − [M1 ] in G0 (R) /Im(α) and hence is 0 as we have an exact
sequence. This shows that L(RW ) ⊆ ker(γ).
Finally, γ induces a homomorphism γ : G0 (RW ) → G0 (R) /Im(α). Let M
be a finitely generated R-module. We have that γ · β is the identity, that is,

γ · β([M ] + Im(α)) = γ([MW ])

= [M ] + Im(α).

Likewise, we have that β · γ is the identity. Let N be a finitely generated RW -

module and choose M finitely generated as an R-module such that MW = N .
Then,

β · γ([N ]) = β([M ] + Im(α))

= [MW ]
= [N ]

Example 23. Let R = C[[x, y, z]]/(x2 + y 3 + z 5 ). One can prove that R is a

2-dimensional UFD. Then G0 (R) = Z[R] ' Z.
Proof. Notice that (x) ⊆ R is prime. Consider the multiplicatively closed system
W = {xn : n ≥ 1}, and notice that RW is a 1-dimensional regular local ring, so
that G0 (RW ) ' Z. By Localization Lemma we have an exact sequence

α / G0 (R) β
/Z /0
M
G0 (R/p)
W ∩p6=∅
1 Basic Lemmas and Remarks 39

Notice that {p ∈ SpecR : W ∩ p 6= ∅} = {(x, y, z)R, (x)R}. Set S := R/xR '

C[[y, z]]/(y 3 + z 5 ) so that the exact sequence becomes

G0 (C) ⊕ G0 (S)
α / G0 (R) β
/Z /0

By Corollary 24 we have G0 (S) = Z{[S], [C]}, since the only primes in S are
the zero and the maximal ideal. Consider the following short exact sequence:

0 /R ·x /R /S / 0.

Then in the Grothendieck group G0 (R) we have [S] = 0. Also notice that
x, y ∈ R form a regular sequence, therefore we get a long exact sequence

0 /R / R2 /R / R/(x, y)R / 0.

So, inside G0 (R):    

R C[[z]]
= = 5[C] = 0
(x, y)R (z 5 )
But this just means that [C] is torsion in G0 (R). However, y, z also form a
regular sequence in R, therefore

0 /R / R2 /R / R/(y, z)R / 0.

is also exact. This means

   
R C[[x]]
= = 2[C] = 0
(y, z)R (x2 )
in G0 (R), and hence [C] = 0. Since [C] = [S] = 0 in G0 (R), we have that α is
the zero map, and therefore
G0 (R) ' Z.

Theorem 29. Let R be a noetherian ring. Then the map

α : G0 (R) / G0 (R[x])

[M ]  / [M ⊗R R[x]]

is an isomorphism.
Proof. Let M [x] denote M ⊗R R[x]. Since R[x] is free, in particular flat, the
map α is well-defined. We will define a map β : G0 (R[x]) → G0 (R) with the
following construction.
Let N be a finitely generated R[x]-module and consider the exact sequence

0 /K /N x−1
/N /C /0
40 Grothendieck Groups

where K and C are the kernel and cokernel respectively. Notice in G0 (R[x])
that [K] = [C] and that

(x − 1)K = (x − 1)C = 0.

But R ' R[x]/(x − 1) and so K, C are finitely generated R-modules as well.

Define
β([N ]) = [C] − [K] ∈ G0 (R) .
This is a well-defined map. To see this, define β(hN i) = [C] − [K] and consider
the following commutative diagram of R[x]-modules:

0 / N1 /N / N2 /0
x−1 x−1 x−1
  
0 / N1 /N / N2 /0

By the snake lemma, we have the long exact sequence

0 / K1 /K / K2 / C1 /C / C2 /0

where Ki and Ci are the respective kernels and cokernels. Therefore we have
that
[C] − [K] = [C1 ] − [K1 ] + [C2 ] − [K2 ],
giving the desired result of

β(hN i) = β(hN1 i) + β(N2 ).

Hence, β is well-defined.
Observe that the composition β ◦ α is the identity. To see this, not that x − 1
is a non-zero divisor on M [x] and thus the sequence

0 / M [x] x−1
/ M [x] /M /0 (2.9)

is exact (this should be verified by the reader). Hence,

β([M [x]]) = [M ] − [0] = [M ].

To finish the proof, it is enough to show that α is onto. To do this, we use

Noetherian induction:
Assume not and choose and ideal I in R maximal such that

G0 (R/I)
α / G0 (R/I[x])

is not onto. We change notation by letting R/I = R.

Claim 1. The ring R is a domain.
1 Basic Lemmas and Remarks 41

To see this, suppose R is not a domain and let min(R) = {p1 , . . . , ps }. Note
that
min(R[x]) = {p1 R[x], . . . , ps R[x]}.
We thus have the following commutative diagram:

/ G0 (R) /0
M
G0 (R/pi )
α
 
/ G0 (R[x]) /0
M
G0 (R/pi [x])

By the corollary of the Filtration Lemma (page 32), the two horizontal maps
are onto. Further, we have that the far left map is onto by induction. This
forces α to be onto; a contradiction. Therefore, R is a domain.
Now consider the following sequence where W = R \ {0}:
     
M R /
M R[x] /
M R[x] / G0 (R[x]) .
G0 G0 G0
p pR[x] Q 8
p6=0 Q∩W 6=∅
p∈Spec(R) Q∈Spec(R[x])
Q⊇pR[x]

The first map is onto by induction and the second is just a natural map and
hence is also onto. The third map is defined by the Localization Lemma on
page 36. As labeled above, let γ be the composition map.
Let k be the quotient field of R. We now have the following commutative
diagram:
 
M R / G0 (R) / G0 (k) /0
G0
p
γ α αk

  
0 /K / G0 (R[x]) / G0 (k[x]) /0

where K is the respective kernel defined from the Localization Lemma. First
notice that αk is surjective since

G0 (k) = Z = G0 (k[x])

and αk is defined by mapping [k] to [k[x]]. Further, using the induction with
the Filtration Lemma, we see that γ is onto the kernel K. Therefore, α is also
onto by the snake lemma.
A helpful tool in the next theorem is the concept of the conductor.
Definition. Let R be a local domain and S be the integral closure of R. The
conductor, denoted C, is the largest common ideal of both R and S.
42 Grothendieck Groups

Example 24. If R = kJt6 , t10 , t15 K then we have that S = kJtK is the integral
closure. It is not difficult to see that C = t30 R.
Theorem 30 (Conjecture of Herzog). If (R, m, k) is a one dimensional complete
local domain with algebraically closed residue field, then G0 (R) ' Z.
Proof. By the Filtration Lemma, G0 (R) is generated by [R] and [k]. Further,
we have that the rank function G0 (R) → Z is onto and sends [R] to 1. We need
to prove that [k] = 0.
If x ∈ R is non-zero, then [R/xR] = 0 and R/xR has a filtration of copies
of k of λR (R/xR). Therefore
λR (R/xR) · [k] = 0
for any non-zero x in R. Let S = {λ(R/xR) | 0 6= x ∈ m}. It is enough to show
that the gcd of S is one.
Let V be the integral closure of R. Note that V is a one-dimensional (Cohen-
Seidenberg), local (true sense R is complete; exercise) and V is integrally closed.
Thus we have that V is a DVR with mV = (t).
As V is a finitely generated R-module, there exists a conductor C ⊆ R such
that CV ⊆ R. Pick any non-zero x in C. As tx is an element of C, we have that
λV (V /xV ) + 1 = λV (V /txV ).
However, as the residue fields of R and V are the same,
λV (V /xV ) = λR (V /xV )
= rank(V ) · λR (R/xR)
= λR (R/xR).
Therefore we have that λR (R/xR) = λR (R/txR) − 1. Thus we have that the
gcd of S is one.

1.1 Structure of One Dimensional Local Complete Do-

mains
Assume that R contains its residue field k and that k is algebraically closed.
Then R = kJtK since R is a DVR as above and R/mR = k; k ⊆ R. Therefore,
by Cohen’s structure theorem, R = kJtK. Let
S = min{λ(R/xR) | x ∈ m}.
Then there exists an element
ts + αs+1 ts+1 + · · · ∈ R
and an element u = 1 + αs+1 ts + · · · such that ts u is the above element. Note
that ts u ∈ R, but u may not be in R.
If the characteristic of k is zero, then by Hensel’s Lemma there exists v ∈ R
such that v s = u. Let z = tv. Then kJtK = kJzK. Now z s ∈ R and
R = kJz s , higher order powersK ⊆ kJzK.
2 Class Groups 43

2 Class Groups
All along this section R will be an integrally closed noetherian domain.
Definition. Set X 1 (R) := {p ∈ SpecR : ht p = 1}. Also set X(R) = the free
abelian group on generators p ∈ X 1 (R). More explicitly, if D ∈ X(R) we can
write X
D= np p,
p∈x1 (R)

where all but finitely many among the np ’s are zero. Elements in X(R) are
called divisors.
Notice that, since R is integrally closed, for all p ∈ X 1 (R) Rp is a DVR (it is
a 1-dimensional integrally closed local domain). If pRp = (tp ), then every ideal
is a power of the maximal ideal, i.e. I = (tnp ) for some n ∈ N. By definition

(tnp )Rp ∩ R = p(n)

is the n-th symbolic power of p, and it is the p-primary component of pn .
Example 25. Lel k be a field, with Chark 6= 2. Let
k[x, y, z]
R := .
(x2 − yz)
R is an integrally closed domain: R is Cohen-Macaulay, and hence it satisfies
Serre’s condition (S2 ). Also the Jacobian ideal
J = (2x, y, z)
has height two, which means that Rp is regular for all p ∈ X 1 (R) and hence R
satisfies Serre’s condition (R1 ). Hence R is integrally closed.
Consider p = (x, z)R a prime ideal in R, and notice that p ∈ X 1 (R). Then
pRp = xRp , since in Rp y is invertible and hence z = −y −1 x2 . Then p2 =
(x2 , xz, z 2 )R = (xz, z 2 , yz)R and therefore
p(2) = (xz, z 2 , yz)Rp ∩ R = (z)Rp ∩ R = (z) ( p2 .
since y is again invertible in Rp .
Given a non-zero ideal I ⊆ R, set
X
div (I) := vp (I)p,
p∈X 1 (R)

where  
IRp = tvp (I) Rp for p ∈ X 1 (R).
Since I 6= (0) there are only finitely many minimal primes containing I, and
hence only finitely many height one primes containing I. For the others, if
p ∈ X 1 (R) and I 6⊆ p, then IRp = Rp , and hence vp (I) = 0 for such p. So
div (I) is well defined, since we have just proved that the sum is finite.
44 Grothendieck Groups

Remark 24. If ht I > 2, then vp (I) = 0 for all p ∈ X 1 (R) and therefore div (I) =
0.

Definition. Given f = a/b ∈ K r {0}, where K is the fraction field of R, define

div (f ) = div (a) − div (b) .

Define the set of principal divisors

P (R) := {div (f ) : f ∈ K r {0}} ⊆ X(R).

It is a subgroup of X(R) since

div f −1 = −div (f ) .


div (f g) = div (f ) + div (g) and

Finally define the Class Group of R as

Cl (R) := X(R)/P (R).

The class of p ∈ X(R) inside Cl (R) is denoted [p].

Definition. Two divisors D1 , D2 ∈ X(R) are said to be linearly equivalent if

D1 − D2 ∈ P (R) (i.e. [D1 ] = [D2 ]). A divisor D ∈ X(R) is said to be effective
if X
D= np p.
p∈x1 (R)

and np ≥ 0 for all p ∈ X 1 (R).

Lemma 31. Let D ∈ X(R). Then D is linearly equivalent to an effective

divisor.

Proof. Write D = D+ − D− , where D+ and D− are effective. Write

k
X
D− = ni pi
i=1

for some pi ∈ X 1 (R), ni > 0. Choose a non-zero x such that

(n1 ) (nk )
x ∈ p1 ∩ . . . ∩ pk 6= (0),

then E := D+ − D− + div (x) is effective since vpi (x) ≥ ni for all i = 1, . . . , k

and clearly
D − E = div (x) ∈ P (R).

Lemma 32. Let D ∈ X(R) be an effective divisor. Then [D] = 0 if and only if
D = div (x) for some x ∈ R.
2 Class Groups 45

Proof. If D = div (f ) ∈ P (R), then by definition [D] = 0. Conversely assume

D = div (a/b) = div (a) − div (b) for some a, b ∈ R, b 6= 0. Write
X X X
D= vp (a)p − vp (b)p = (vp (a) − vp (b))p,
p∈X 1 (R) p∈X 1 (R) p∈X 1 (R)

where vp (a) − vp (b) > 0 since D is effective. By uniqueness of the minimal part
of the primary decomposition we get
\ \
(a) = pvp (a) ⊆ pvp (b) = (b),
p∈X 1 (R) p∈X 1 (R)

and therefore a/b = x ∈ R, so that D = div (a/b) = div (x).

Theorem 33. R is UFD if and only if Cl (R) = 0.
Proof. R is UFD if and only if every height one prime is principal, if and only
if p = div (x) for some x ∈ R. By the previous Lemma this is true if and only
if [p] = 0 for all p ∈ X(R), if and only if Cl (R) = 0.
Theorem 34 (Localization Lemma). Let W ⊆ R be a multiplicatively closed
set. Then there exists a short exact sequence

0 /H / Cl (R) / Cl (RW ) / 0,

where is the subgroup H =< [p] : p ∈ X 1 (R), p ∩ W 6= ∅ >⊆ Cl (R).

Proof. First we define the map θe : X(R) −→ X(RW ) as follows:

(
p if ∩ W 6= (0)
θ(p) =
e
pW if p ∩ W = ∅

Notice that such a map is well defined since X(R) is free and we can always
define a map on a basis. Notice also that pW are height one primes in RW .
e (R)) ⊆ P (RW ).
Claim. θ(P
Proof of the Claim. By Lemma 31 it is enough to show it for effective divisors.
Let a ∈ R, then
X X
div (a) = vp (a)p + vq (a)q.
p∩W 6=∅ q∩W =∅

Then X a
θ(div
e (a)) = vq (a)qW = div .
1
q∩W =∅

This is because \ \
(a) = pvp (a) ∩ qvq (a)
p∩W 6=∅ q∩W =∅
46 Grothendieck Groups

and localizing a \

= qvq (a) .
1
q∩W =∅

Also, θe is surjective by the correspondence between primes in R and primes

in RW that don’t intersect W . Therefore θe induces a surjective map

Cl (R) / Cl (RW ) θ / 0.

Clearly H ⊆ ker θ. Conversely, let D ∈ ker θ, we can assume that D is effective,

so that X X
D= dp [p] + dq [q],
p∩W 6=∅ q∩W =∅
P
with dp , dq > 0, and dp [p] ∈ H ⊆ ker θ. Also
p∩W 6=∅
 
X X
θ dq [q] = dq [qw ] = 0
q∩W =∅ q∩W =∅

inside Cl (RW ). This means that there exists wa ∈ RW such that

X
div (a/w) = dq [qw ].
q∩W =∅

Since w is a unit in RW we have div (a/w) = div (a/1), so we can assume

X
div (a/1) = dq [qw ].
q∩W =∅

Taking a primary decomposition of (a), we have that the part with {p : p ∩ W =

∅} has to be the same as for D, so that
X X
div (a) = ep p + dq [q].
p∩W 6=∅ q∩W =∅

Clearly D ∼
= div (a) (mod H), so that [D] ∼
= 0 (mod H) in Cl (R) and therefore
[D] ∈ H.
Example 26. Let k be a field with Chark 6= 2. Let
k[x, y, z]
R=
(x2 − yz)

and consider the multiplicatively closed set W = {z n }. Then

k[x, y, z, z −1 ] k[x, y, z, z −1 ]
R[z −1 ] ' ' ' k[x, z, z −1 ],
2
(x − yz) (x2 z −1 − y)
2 Class Groups 47

which is a UFD, and hence Cl (R)W = 0. Using Localization Lemma for Class
Groups we have that

Cl (R) '< [p] : p ∩ {z n } =

6 ∅, p ∈ X 1 (R) > .
√
Notice that zR = (x, z) =: p and ht p = 1, so that it is the only height one
prime intersecting W . Hence Cl (R) ' Z[p]. We have already noticed that

p(2) = (z)

hence div (z) = 2p, which means 2[p] = 0 in Cl (R). We have two cases:
(
Z/2Z
Cl (R) '
0

But R is not a UFD because p is height one but not principal, therefore Cl (R) 6=
0, that is
Z
Cl (R) '
2Z
Theorem 35. Let R be an integrally closed noetherian domain. Then
(1) R[T1 , . . . , Tn ] is an integrally closed noetherian domain.
(2) Cl (R) ' Cl (R[T1 , . . . , Tn ]).
Proof. We prove only (2). By induction is enough to show the case n = 1.
Recall that if p ∈ X 1 (R), then p[T ] = p ⊗R R[T ] ∈ X 1 (R[T ]). Also, if a ∈ R
and
(n ) (n )
(a) = p1 1 ∩ . . . ∩ pk k
is a primary decomposition, then

aR[T ] = p1 [T ](n1 ) ∩ . . . ∩ pk [T ](nk ) . (2.10)

Let W = R r {0}, then (R[T ])W = RW [T ] = K[T ], where K = Q(R) is the

quotient field of R. This is UFD since it is a PID, hence Cl (() R[T ])W ) = 0.
Using Localization Lemma for Class Groups we have

Cl (R[T ]) ' H =< [Q] : Q ∩ W 6= ∅, Q ∈ X 1 (R[T ]) > .

Note that, since Q ∩ R 6= (0), then q = Q ∩ R is a height one prime in R.

Therefore Q ⊇ q[T ]. But ht Q = 1, hence Q = q[T ]. Hence

X(R) /H /0

q / [q[T ]]

By (2.10) the kernel is exactly P (R), hence H ' Cl (R). This is because prin-
cipal divisors in R correspond to principal divisors in R[T ].
48 Grothendieck Groups

Remark 25. In general it is not true that Cl (R) ' Cl (RJT K). If this is the case
R is said to have discrete divisor class group. Notice that we have always
Cl (R) ,→ Cl (RJT K)
Theorem 36 (Danilov). If R satisfies Serre’s conditions (S3 ) and (R2 ) then
Cl (R) ' Cl (RJT K).
 
Remark 26. Similarly we always have Cl (R) ,→ Cl R
b .

Theorem 37 (Flenner).
L Let R be an integrally closed standard
L graded domain,
say R = k[R1 ] = i>0 Ri where k is a field. Set m = i>1 Ri and assume R
satisfies Serre’s condition (R2 ). Then
 
Cl (R) ' Cl (Rm ) ' Cl R bm .

Example 27. Let  

C[x, y, z]
R= .
(x2 + y 3 + z 7 ) (x,y,z)
 
b 6= 0.
Then R is UFD, therefore Cl (R) = 0, but Cl (RJT K) 6= 0 and Cl R
There is a relation between the Class Group and the Grothendieck Group.
First we need the following definition.
Definition. Let R be a noetherian domain. The reduced Grothendieck Group
G
f0 (R) is the subgroup of G0 (R) which is the kernel of the map

G0 (R) / G0 (K) ,

where K = Q(R) is the fraction field of R. Notice that, by Localization Lemma,

f0 (R) is generated by [R/p], where p 6= 0 is prime.
G
Theorem 38. Let R be a integrally closed noetherian domain. Let H be the
subgroup of G0 (R) generated by [R/p], with ht p = 2. Then there is a short
exact sequence

0 /H /G
f0 (R) / Cl (R) / 0.

The proof of the theorem is postponed.

3 Divisors attached to Modules

Throughout this section let R be an integrally closed noetherian domain. The
goal is to construct a map
c : G0 (R) / Cl (R)

[M ]  / c([M ])

We will start the construction restricting to torsion modules.

3 Divisors attached to Modules 49

Definition. Let R be a domain. An R-module T is said to be torsion if there

exists x ∈ R, x 6= 0 such that xT = 0.

Let T be a torsion R-module. Define

X
c(T ) = λ(Tp )[p] ∈ Cl (R) .
p∈X 1 (R)

Since T is torsion we have ht annT ≥ 1. Also, since SuppT = V (annT ), there

exist only finitely many p ∈ X 1 (R) such that inSuppT , namely

SuppT ∩ X 1 (R) = X 1 (R) ∩ Min(annT ).

p
For each of these we get λ(Tp ) < ∞ since (annT )p = pRp , which means that
SuppTp = pRp is just the maximal ideal, hence Tp has finite length.
Remark 27. If we have a short exact sequence of torsion modules

0 / T1 /T / T2 / 0,

then c(T ) = c(T1 ) + c(T2 ) since localization is flat and length is additive.
Let us now go back to the general case. Let M ∈ Modfg (R), and recall that
rank(M ) = dimK M ⊗R K, where K = R(0) is the fraction field of R.
Remark 28. T is torsion of and only if rank(T ) = 0.
Suppose rank(M ) = r, then there is a K-vector space isomorphism

e : R r ⊗R K
α / M ⊗R K.

Since Rr is finitely presented, we have

r
HomR(0) (R(0) , M(0) ) ' (HomR (Rr , M ))(0) ,

which means that there exists α : Rr → M such that α(0) = α

e. We have an
exact sequence

0 / ker α / Rr /M / cokerα / 0,

and also kerα and cokerα are torsion modules because α(0) = α e, which is an
isomorphism. Hence ker α ⊗ 1 = 0 = cokerα ⊗ 1. But ker α ⊆ Rr , and a
submodule of a free module cannot be torsion unless it is zero. Therefore we
have the following short exact sequence

0 /F /M /T / 0,

where F is free and T := cokerα is torsion. Define c(M ) := c(T ).

Claim. c is well defined.
50 Grothendieck Groups

Proof of the Claim. Suppose we have

0 /F /M /T /0

and
0 /G /M / T0 / 0.

We need to prove that c(T ) = c(T 0 ). First of all we can reduce to the case
F ⊆ G. In fact, notice that

F ⊗R K ' G ⊗R K ' M ⊗R K ' K r .

We can think of F and G inside K r (they are not K-vector subspaces). Then
there exists x ∈ R such that xF ⊆ G. However, consider

0 0 ker θ

  
0 / xF /M / T 00 /0

i idM θ
  
0 /F /M /T /0


F  
0 cokerθ
xF

By the Snake Lemma we get ker θ ' F/xF , and hence we have a short exact
sequence

0 / F / T 00 /T /0
xF

These are all torsion modules, and we have already proved that for torsion
modules c is additive. Hence

c(T 00 ) = c(T ) + c(F/xF ) = c(T ) + rc(R/xR).

But
  !
R X R
c( )= λ [p] = [div (x)] = 0 in Cl (R) .
xR xR p
p∈X 1 (R)

Therefore c(T 00 ) = c(T ), and hence without loss of generality we can assume
F ⊆ G.
3 Divisors attached to Modules 51

Now consider

0 /0 /0 / ker δ /0

  
0 /F /M /T /0
idM
i
 
δ

0 /G /M / T0 /0


 
0 / G /0 / cokerδ /0
F
so that, again by Snake Lemma, we get a short exact sequence

0 / G /T / T0 / 0,
F

and therefore
c(T ) = c(T 0 ) + c(G/F ).
We want to prove that c(G/F ) = 0. We have

0 / F ' Rr A / G ' Rr / G / 0,
F
and we need to prove that there exists y ∈ R such that
  !
X G
λ p = div (y) .
1
F p
p∈X (R)

Take y = det A. Let p ∈ X 1 (R), then localizing we get

0 / Rpr Ap
/ Rpr / Gp / 0.
Fp

Since p is a height one prime, Rp is a DVR, and in particular a PID. Then

we can use the fundamental theorem for PID, for which we can diagonalize Ap
changing basis. So assume
 
d1 0 . . . 0
 0 d2 . . . 0 
 
Ap = . . . . . . . . . . . .

. . . . . . . . . . . .
0 0 . . . dr

is diagonal, with di 6= 0 for all i since Ap is an injection. Then

     
Gp R R R
' ⊕ ⊕ ... ⊕
Fp (d1 ) p (d2 ) p (dr ) p
52 Grothendieck Groups

and therefore
  r   !
Gp X R
λ = λ .
Fp i=1
(di ) p

Finally
r   !     !
X R Rp R
λ =λ =λ ,
i=1
(di ) p (d1 · . . . · dr )p (det A) p

so that  
G
c = [div (det a)] = 0 in Cl (R) .
F

We want to prove now that c induces a map on G0 (R). Suppose we have a

short exact sequence

0 / M1 α /M β
/ M2 / 0,

we want to show that c(M ) = C(M1 ) + c(M2 ). Suppose rank(M1 ) = r1 and

rank(M2 ) = r2 and choose free modules Rr1 ⊆ M1 and Rr2 ⊆ M2 with bases
e1 , . . . , er1 and f1 , . . . , fr2 respectively. Set

ui = α(ej ) ∈ M for i = 1, . . . , r1 and vj = β −1 (fj ) ∈ M for j = 1, . . . , r2 .

Claim. R{u1 , . . . , ur1 , v1 , . . . , vr2 } is a free R-module of rank r1 + r2 .

Proof of the Claim. Assume not, so that there exist s1 , . . . , sr2 , t1 , . . . , tr2 ∈ R
not all zero such that

s1 u1 + . . . + sr1 ur1 + t1 v1 + . . . + tr2 vr2 = 0.

Apply β, using the fact that β(ui ) = β(α(ei )) = 0 for all i = 1, . . . , r1 :

r2
X r2
X
0= tj β(vj ) = tj fj ,
j=1 j=1

which implies t1 = . . . = tr2 = 0 since {f1 , . . . , fr2 is a basis of Rr2 . Then we

have
Xr1 r1
X
si ui = α(ei ) = 0,
i=1 i=1

which implies u1 = . . . = ur1 = 0 since {e1 , . . . , er1 } is a basis of Rr1 and α is

injective. This proves the Claim.
3 Divisors attached to Modules 53

Finally we have an exact diagram

0 / Rr 1 / Rr1 +r
_ 2 / Rr 2 /0
_ _

  
0 / M1 /M / M2 /0

  
0 / T1 /T / T2 /0

  
0 0 0
which commutes by construction. The short exact sequence
0 / T1 /T / T2 /0

exists by Snake Lemma, and T1 , T and T2 are all torsion modules, again by
construction. But for torsion modules we know that c is additive, hence
c(M ) = c(T ) = c(T1 ) + c(T2 ) = c(M1 ) + c(M2 ).
Summarizing we have the following theorem.
Theorem 39. There is a surjective group homomorphism

c:G
f0 (R) / Cl (R) /0

Proof. We already proved that there is a map

c : H(R) / Cl (R)

that preserves short exact sequences, i.e. c(L(R)) = 0. This induces a group
homomorphism
c : G0 (R) / Cl (R) .

To prove that it is surjective let [p] ∈ Cl (R) and consider R/p, which is a torsion
R-module. Then
    !
R X R
c = λ [q] = [p].
p 1
p q
q∈X (R)

Finally we can consider the restriction

c:G
f0 (R) / Cl (R) ,

which is still surjective since for all p ∈ X 1 (R) we have [R/p] ∈ G

f0 (R).
54 Grothendieck Groups

f0 (R) → Cl (R) be as above. Then

Proposition 40. Let c : G

ker c = H =< [R/p] : ht p ≥ 2 >,

so that we have a short exact sequence

0 /H /G
f0 (R) / Cl (R) / 0.

Proof. First notice that H ⊆ ker c since if [R/p] ∈ H, then R/p is torsion and
hence X  
c([R/p]) = λ (R/p)q [q] = 0
q∈X 1 (R)

since, being ht p ≥ 2, p cannot be contained in any height one prime. To prove

the converse we want to find a left inverse to c. First let use define a map on
the free abelian group X(R) as follows:

β : X(R) /G
f0 (R) /H

p / [R/p] + H

Notice that β is onto since by the localization lemma we have

f0 (R) =< [R/q] : ht q ≥ 1 >
G

and H already involves all primes of height at least two, while β(X(R)) involves
all primes of height one. We now want to show that P (R) ⊆ ker β. Let a ∈ R,
a 6= 0, and write \
(a) = p(np ) ,
p∈X 1 (R)

so that X
div (a) = np [p] ∈ X(R).
p∈X 1 (R)

Applying β we get
X
β(div (a)) = np [R/p] + H ∈ G
f0 (R) .
p∈X 1 (R)

Consider the following short exact sequence

/ R / R /T / 0,
M
0 (2.11)
(a) p(np )
p∈X 1 (R)

where T is the cokernel of the first map. Also, notice that we have the following
short exact sequence of R-modules

0 /R ·a /R / R/(a) / 0,
3 Divisors attached to Modules 55

so that [R/(a)] = 0 in G0 (R). Therefore

 
 

M R  = R + [T ] = [T ].
1
pnp (a)
p∈X (R)

Claim. ht(annT ) ≥ 2.

Proof of the Claim. Let q ∈ X 1 (R). If a ∈ q, localizing the short exact sequence
(2.11) at q we get
 
 
/ R R /
M R  R / Tq / 0,
0 = = (n )
(a) q q(nq ) p(np ) q q
p∈X 1 (R)
q

L 
so that Tq = 0. Also, if a ∈
/ q, then both (R/(a))q and p∈X 1 (R) R/p(np )
q
are zero, so that Tq = 0 again. This proves the Claim.

By the Claim we have that

 
M R 
 = [T ] ∈ H.
p(np )
p∈X 1 (R)

Finally, one can prove that

 
 
M R  X R
 − np ∈ H,
p(np ) p
p∈X 1 (R) p∈X 1 (R)

so that P (R) ⊆ ker β, and hence we get an induced map

β : Cl (R) /G
f0 (R) /H.

To finish the proof we need to show that β is a left inverse for c, and it is enough
to check it on the generators of Gf0 (R) /H, i.e. [R/p] + H with p ∈ X 1 (R).
Notice that
     !  
R X R R
c +H = λ [q] = ,
p 1
p q p
q∈X (R)

so that       
R R R
β◦c +H =β = + H,
p p p
that is β ◦ c = idG
f0 (R)/H , and c is injective. This implies ker c = H.
56 Grothendieck Groups

Corollary 41. Let R be a integrally closed noetherian domain such that every
p ∈ X 1 (R) has finite free resolution. Then R is a UFD.
Proof. It is enough to show that Cl (R) = 0. Let [p] ∈ Cl (R), then by assump-
tion there is a finite free resolution

0 / Fn ... / F1 / F0 / R /0
p

with the Fi ’s finitely generated free R-modules. In G

f0 (R) we get
  X n
R
= (−1)i [Fi ],
p i=0

and applying c we get

  n
R X
c = [p] = (−1)i c([Fi ]) = 0
p i=0

since c is zero on free modules. Therefore Cl (R) = 0 and R is a UFD.

We now state, without proving it, Eagon’s Theorem.
Theorem 42 (Eagon). Let R be a noetherian ring and let F be the free abelian
group on {hRpi : p ∈ SpecR}. Let W ⊆ F be the submodule generated by all
X
np hR/pi,
p∈Λ

where for q ∈ SpecR and x ∈ / q there exists a prime filtration of R/(q, x) that
has exactly np copies of R/p, for p ∈ Λ. Then

G0 (R) ' F/W.

Remark 29. For q ∈ specR and x ∈

/ q we always have a short exact sequence

0 / R ·x / R / R / 0,
q q (q,x)

therefore [R/(q, x)] = 0 in G0 (R) and we always have a subjective map

F/W → G0 (R) → 0.

4 Construction and Properties of K0 (R)

Definition. An R-module P is said to be projective if there exist an R-module
Q and a free R-module F such that

P ⊕ Q ' F.
4 Construction and Properties of K0 (R) 57

Remark 30. By symmetry, the module Q in the above definition is also projec-
tive.
Remark 31. Free R-modules are projective.
Remark 32. If (R, m) is a local noetherian ring, then every projective R-module
P is free. With the assumption that P is finitely generated this is an easy exer-
cise, using Nakayama’s Lemma. The result for non-finitely generated modules
was proved by Kaplansky.
Remark 33. If R is a noetherian ring and P is Projective, then for all q ∈ SpecR
Pq is a free Rq -module. This is because localization commutes with direct sums,
and because projective modules over a local ring are free by Remark 31. If P is
finitely generated, then the converse holds, i.e. if Pq is free for all q ∈ SpecR,
then P is projective.
Remark 34. Let W ⊆ R be a multiplicatively closed set. Because localization
commutes with direct sums we have that if P is a projective R-module, then
PW is a projective RW -module.
Remark 35. Let P be a R-module. Then P is projective if and only if whenever
we have a short exact sequence
0 /K /M /P /0

then it splits, i.e. M ' P ⊕ K.

Definition. Let R be a noetherian ring. In analogy with G0 (R) we define
Z{finitely generated projective R-modules}
K0 (R) :=
Z{< P ⊕ Q > − < P > − < Q >}
since by Remark 35 every short exact sequence of Projective modules splits.
Remark 36. The case in which (R, m) is local is not interesting since every
Projective module is free by Remark 31.
Remark 37. Clearly there is a map

K0 (R) / G0 (R) rank,e,λ

/Z

 / [P ]
[P ]

so that we can apply the same results concerning the existence of functions

rank, e, λ : K0 (R) /Z

as we did for G0 (R).

Theorem 43. Let R be a noetherian ring and let P, Q be finitely generated
projective R-modules. Then [P ] = [Q] in K0 (R) if and only if there exists a
finitely generated free R-module F such that
P ⊕ F ' Q ⊕ F.
58 Grothendieck Groups

Proof. Clearly if P ⊕ F ' Q ⊕ F for some finitely generated free R-module F ,

then [P ] = [Q] in K0 (R). Conversely assume [P ] = [Q]. Then there exist short
exact sequences
0 / Pi / Pi ⊕ Qi / Qi /0

such that
X X
hP i − hQi = ai (hPi ⊕ Qi i − hPi i − hQi i) − bj (hPj ⊕ Qj i − hPj i − hQj i),
ai >0 bj >0

where we rearrange the sum in order to have positive coefficients. then we can
rewrite
X X X X
hP i+ ai hPi ⊕Qi i+ bj (hPj i+hQj i) = hQi+ ai (hPi i+hQi i)+ bj hPj ⊕Qj i
ai >0 bj >0 ai >0 bj >0

and therefore
h i
b b
P ⊕ Qai i ⊕ Piai ⊕ (Pj ⊕ Qj )bj ' Q ⊕ (Pi ⊕ Qi )ai ⊕ Pj j ⊕ Qjj .
 

b b
Hence P ⊕ F ' Q ⊕ L with L = Qai i ⊕ Piai ⊕ Pj j ⊕ Qjj a projective R-module.
Since L is projective there exist a R-module N and a free module F such that
L ⊕ N ' F . Therefore
P ⊕ F ' P ⊕ L ⊕ N ' Q ⊕ L ⊕ N ' Q ⊕ F,
which completes the proof.
Definition. Let Q be a noetherian R-module. Suppose there exist free R-
modules F and G such that Q ⊕ G ' F , then Q is said to be stably free. Clearly
projective modules are stably free.
Corollary 44. Let R be a noetherian domain. If K0 (R) ' Z, then all finitely
generated projective R-modules are stably free.
Proof. Use the rank function
rank : K0 (R) /Z

[R]  /1

Since by assumption K0 (R) ' Z and since rank is surjective, it must be also
injective and hence an isomorphism. Let Q be a finitely generated projective
R-module, say rank(Q) = r, then
rank([Q]) = r = rank([Rr ]),
and hence [Q] = [Rr ] since the function rank is an isomorphism. By Theorem
43 there exists a free R-module G such that
Q ⊕ G ' Rr ⊕ G ' F a free module.
Therefore Q is stably free.
4 Construction and Properties of K0 (R) 59

Lemma 45 (Schanuel’s Lemma). Let R be a ring, and let

0 / N1 i / P1 α /M /0

0 / N2 j
/ P2 β
/M /0

be short exact sequences of R-modules, with P1 and P2 projective. Then N1 ⊕

P2 ' N2 ⊕ P1 .
Proof. Since P1 is projective there exists a map π : P2 → P1 such that βπ = α.
By diagram chasing we can also get a map g : N1 → N2 , so that the following
diagram commutes:

0 / N1 i / P1 α /M /0

g π

 
0 / N2 j
/ P2 β
/M /0

Then the following is exact:

(g,i) (j,−π)
0 / N1 / N2 ⊕ P1 / P2 /0

If so the sequence splits as P2 is projective, and hence the lemma follows. So we

just have to prove that the above sequence is exact. Clearly (g, i) is injective,
because j is injective. Also (j, −π) is subjective: let z ∈ P2 , since α is surjective
there exists y ∈ P1 such that α(y) = β(z). But β(π(y) − z) = α(y) − β(z) = 0,
therefore there exists u ∈ N2 with j(u) = π(y) − z. Finally z = π(y) − j(u) =
(j, π)(−u, −y). Now clearly Im(g, i) ⊆ ker(j, −π) because π ◦ i = j ◦ g. Let
u ∈ N2 , y ∈ P1 such that j(u) = π(y). Then α(y) = 0 because β(π) ≡ 0, and
hence y = i(v) for some v ∈ N1 . But j(g(v)) = π(i(v))π(y) = j(u), and since j
is injective we have u = g(v). Hence (u, y) = (g, i)(v).
Lemma 46 (Generalized Schanuel’s Lemma). Let R be a ring and suppose we
have long exact sequences

0 / N1 / Pn / Pn−1 / ...... / P0 / M1 /0

'


0 / N2 / Qn / Qn−1 / ...... / Q0 / M2 /0

where M1 ' M2 and all Pi ’s and Qj ’s are projective R-modules. Then

N1 ⊕ Qn ⊕ Pn−1 ⊕ . . . ' N2 ⊕ Pn ⊕ Qn−1 ⊕ . . .

60 Grothendieck Groups

Proof. Start with

0 / K0 / P0 / M1 /0

'


0 / L0 / Q0 / M2 /0

to get, by Schanuel’s Lemma, that P0 ⊕ L0 ' Q0 ⊕ K0 . Now use induction on

0 / N1 / Pn / Pn−1 / ...... / P2 / P1 ⊕ Q0 / K0 ⊕ Q0 /0

'


0 / N2 / Qn / Qn−1 / ...... / Q2 / Q1 ⊕ P0 / L0 ⊕ P0 /0

to get the result, again by Schanuel’s Lemma.

Theorem 47. Let R be a regular ring, dim R = d. Then
G0 (R) ' K0 (R) .
Proof. There exists an obvious map

i : K0 (R) / G0 (R)

[P ]  / [P ]

Let M ∈ Modfg (R), and look at its projective resolution

0 /K / Pd−1 / Pd−2 / ...... / P1 / P0 M / 0.

If q ∈ SpecR, then localizing we get the exact sequence

0 / Kq / (Pd−1 )q / ...... / (P1 )q / (P0 )q Mq / 0.

But globdim(Rq ) 6 d, therefore there exists a free Rq -resolution

0 / (Fd )q / (Fd−1 )q / ...... / (F1 )q / (F0 )q Mq / 0.

By Generalized Schanuel’s Lemma we get

Kq ⊕ (projective) ' (Fd )q ⊕ (projective),
and hence Kq is projective (and hence free) for all q ∈ SpecR. Since K is a
finitely generated locally free R-module it follows that it is projective. Using
the above notation (set Pd := K) define a map

j : H(R)
e / K0 (R)

hM i  / [P0 ] − [P1 ] + . . . + (−1)d [Pd ].

4 Construction and Properties of K0 (R) 61

Notice that e
j is well defined since if we have another projective resolution
0 / Qd / Qd−1 / ...... / Q1 / Q0 M / 0,

then by Generalized Schanuel’s Lemma we have Qeven ⊕ Podd ' Peven ⊕ Qodd ,
that is
[P0 ] − [P1 ] + . . . + (−1)d [Pd ] = [Q0 ] − [Q1 ] + . . . + (−1)d [Qd ]
in K0 (R). Now let 0 → M1 → M → M2 → 0 be a short exact sequence of
R-modules and consider projective resolutions
P· / M1 /0

Q· / M2 /0

Then there exists a projective resolution P· ⊕ Q· → M → 0 of M , so that

j(hM i) = [Peven ⊕ Qeven ] − [Podd ⊕ Qodd ] =
e

= [Peven ] − [Podd ] + [Qeven ] − [Qodd ] = e

j(hM1 i) + e
j(hM2 i).

j induces a homomorphism j : G0 (R) → K0 (R).

Therefore e
Claim. i ◦ j = idG0 (R) and j ◦ i = idK0 (R) .
Proof of the Claim. Let P be a finitely generated projective R-module, then
i([P ]) = [P ] ∈ K0 (R). We have shown that the definition of j does not depend
on the chosen resolution of P , so in particular we can consider
0 / P0 = P /P / 0,

and hence j([P ]) = [P0 ] = [P ]. Let now M ∈ Modfg (R), and consider a projec-
tive resolution
0 / Pd / Pd−1 / Pd−2 / ...... / P1 / P0 M / 0.

Then j([M ]) = [Peven ] − [Podd ] ∈ K0 (R), so that

ij([M ]) = [Peven ] − [Podd ] = [M ] ∈ G0 (R) .

Corollary 48. Let k be a field and let R be the polynomial ring k[x1 , . . . , xn ].
Then every finitely generated projective R-module is stably free.
Proof. We know that R is regular, therefore by Theorem 47 we have
K0 (R) ' G0 (R) ' G0 (k) ' Z.
Since K0 (R) ' Z, by Corollary 44, we have that every finitely generated pro-
jective R-module is stably free.
62 Grothendieck Groups

5 Construction and Properties of K1 (R)

Let R be a commutative ring with 1R . Let us define two categories:

• Objects: Pairs (P, f ) [respectively (F, f )], where P is a finitely generated

projective R-module and f : P → P is an isomorphism [respectively F is
a finitely generated free R-module and f : F → F is an isomorphism].

• Morphisms: Let us define morphisms just for the first category, for the
second the definition is analogous. A morphism h : (P, f ) → (Q, g) con-
sists of an R-module homomorphism h : P → Q such that the following
diagram commutes:
P
f
/P

h h
 
Q /Q
g

In particular, if h is invertible this means that g = hf h−1 . A sequence

0 / (P1 , f1 ) h / (P2 , f2 ) g
/ (P3 , f3 ) /0 (#)

is exact if and only if the following sequence of R-modules is exact

0 / P1 h / P2 g
/ P3 /0

i.e. P2 ' P1 ⊕ P3 since they are projective.

Definition. We define
free abelian group on (P, f ) up to isomorphism
K1 (R) := ,
H
where H is the subgroup generated by the following relations:

(1) Given a short exact sequence as in (#), then we introduce a relation

(P2 , f2 ) − (P1 , f1 ) − (P3 , f3 ).

(2) If (P, f ) and (P, g) are objects, then we introduce the relation

(P, f g) − (P, f ) − (P, g).

By [P, f ] or [(P, f )] we mean the image of (P, f ) in K1 (R). Also, we let K1f (R)
be the same construction, but in the second category of finitely generated free
modules.

Remark 38. We will prove that K1 (R) ' K1f (R).

5 Construction and Properties of K1 (R) 63

Remark 39. For any finitely generated projective module P we have [P, 1P ] = 0.
In fact [P, 1P ] = [P, 1P ◦ 1P ] = [P, 1P ] + [P, 1P ], and hence the remark follows.
Proposition 49. Let L be a field. Then

K1 (L) ' L∗ = L r {0}.

Proof.
64 Grothendieck Groups

6 Exercises
(1) What is the kernel of α?
(2) Is the following exact?

G0 (R/(x)) / G0 (R) / G0 (Rx ) /0

(3) Show that sequence (2.9) on page 40 is exact.

(4) So that G0 (() R[x, y]/(x2 + y 2 ) is Z ⊕ Z/2Z.
Chapter 3

The Module of Differentials

Throughout this chapter, k, R, S will be commutative rings, and k → R → S

will denote ring homomorphisms, so that R and S are k-algebras and S is a
R-algebra.

Definition. Let k → R be a k-algebra, and let M be a R-module. A k-derivation

D : R → M is a map such that

(1) D(r + s) = D(r) + D(s) for all r, s ∈ R.

(2) D(αr) = αD(r) for all r ∈ R and for all α ∈ k.

(3) D(rs) = rD(s) + sD(r) for all r, s ∈ R.

We will denote the set of all k-derivations on M by Derk (R, M ).

Remark 40. The definition of derivation makes sense even if the ring is not
commutative, but in that case we have to respect the order of the multiplication.
For instance in (3) we have to require D(rs) = rD(s)+D(r)s instead of D(rs) =
rD(s) + sD(r).
Remark 41. D(α) = 0 for all α ∈ k.

Proof. Notice that

D(1) = D(1 · 1) = 1 · D(1) + 1 · D(1),

and hence D(1) = 0. It follows that

D(α) = αD(1) = 0

for all α ∈ k.

Remark 42. The set l := D−1 (0) = {r ∈ R : D(r) = 0} is a subring of R, and

also D is a l-linear map. In fact l is the biggest subring S ⊆ R such that D is
S-linear.
66 The Module of Differentials

Proof. If r, s ∈ l, then D(r + s) = D(r) + D(s) = 0. Also D(rs) = rD(s) +

sD(r) = 0 and D(1) = 0, so that l ⊆ R is a subring. Notice that by the previous
remark we have that k ⊆ l. If β ∈ l and r ∈ R, then

D(βr) = βD(r) + rD(β) = βD(r),

so that D is automatically a l-derivation.

Remark 43. Let D be a k-derivation and let r ∈ R. Then for n > 1 we have

D(rn ) = nrn−1 D(r).

Proof. Clearly the claim holds when n = 1. By induction assume it’s true for
n > 1. Then

D(rn+1 ) = rD(rn ) + rn D(r) = rnrn−1 D(r) + rn D(r) = (n + 1)rn D(r).

Remark 44. If Char(R) = p > 0, then for all k-derivations we have D(rp ) = 0.
Remark 45. If D : R → M is a k-derivation and f : M → N is a R-
homomorphism, then f ◦ D is a k-derivation.
We want now to construct a R-module ΩR/k and a universal derivation
d : R → ΩR/k such that for any other k-derivation D : R → M there exists a
unique R-module homomorphism

R
d / ΩR/k

D
 ~ f
M

Definition. ΩR/k is called the universal module of differentials (or universal

module of derivations). It is also called Kähler module of differentials.

1 First construction of the Module of Differen-

tials
We present now a first way to contstruct (ΩR/k , d). We will see a second con-
struction, easier to deal with, later in this chapter. Take a free module on
symbols {dr : r ∈ R}, i.e. M
F := Rdr.
r∈R

We want to construct a derivation, therefore let us define H ⊆ F to be the

submodule generated by:
• d(r + s) − dr − ds.
1 First construction of the Module of Differentials 67

• d(rs) − rds − sdr.

• d(αr) − αdr.
where r, s ∈ R, α ∈ k. Then set

ΩR/k := F/H d:R / ΩR/k

r / dr

Notice that d is clearly a derivation. Let now D : R → M be a k-derivation, i.e.

an element of Derk (R, M ). Consider the following diagram
 / dr
r

r R
d / ΩR/k

D
 | f
M

D(r)

Notice that f (dr) = D(r) is forced to make the diagram commute, and also
{dr : r ∈ R} generates ΩR/k . Hence if such f : ΩR/k → M exists it has to
be unique. Notice also that on F we can define freely f on the basis elements
dr. Also f (H) = 0 because D is a derivation, therefore f : F → M induces
a map f : ΩR/k → M . Finally (ΩR/k , d) is unique (up to isomorphism) by
usual universal property arguments. If we consider d0 : R → Ω0 , then we have a
commutative diagram
d0 / Ω0
R G
g

D
  f
ΩR/k
and one can verify that f ◦ g = idΩ0 and g ◦ f = idΩR/k .
Remark 46. By uniqueness and by Remark 45 we have
Derk (R, M ) ' HomR (ΩR/k , M ).
Proposition 50. Let k be a ring and let R = k[xλ ]λ∈Λ be a polynomial ring.
Then M
ΩR/k = Rdxλ
λ∈Λ

a free R-module. Also, for f ∈ R, we define

X ∂f
df = dxλ ∈ ΩR/k ,
i
∂xi

where the sum involves only finitely many i’s.

68 The Module of Differentials

L
Proof. Let F = λ∈Λ Rdxλ . Consider the diagram

k[xλ ]
d /F
9dxλ

D
" ~
M
|
D(xλ )

where the choice dxλ 7→ D(xλ ) is forces by the commutativity of the diagram.
F is free, hence the map is well defined. The key point of this proof is that
d : k[xλ ] → F is in fact a derivation. For f, g ∈ k[xλ ] and α ∈ k we have:
• d(f + g) = i ∂(f∂x+g)
P P ∂f P ∂g
i
dxi = i ∂x i
dxi + i ∂x i
dxi = df + dg.

• d(f g) = i ∂(f g)
P P ∂g P ∂f
∂xi dxi = f i ∂xi dxi + g i ∂xi dxi = f dg + gdf .
P ∂(αf ) P ∂f
• d(αf ) = i ∂xi dxi = α i ∂xi dxi = αdf .

Proposition 51. Let R be a k-algebra and let I ⊆ R be an ideal. Then

ΩR/k
Ω(R/I)/k ' ,
Rhdi : i ∈ Ii
with d(r + I) = dr + Rhdi : i ∈ Ii.
Proof. We will prove that (Ω(R/I)/k , d) satisfies the universal property. Let
D : R/I → M be a k-derivation. Then

R
d / ΩR/k


R/I f ∈HomR (ΩR/k ,M )
D
D

e
)
M

Notice that f (di) = D(i)

e = 0 for all i ∈ I, therefore we get a unique induced
map d, which is clearly a derivation:

R
d / ΩR/k

 {
R/I
d / ΩR/k
Rhdi:i∈Ii

 |
M
1 First construction of the Module of Differentials 69

Example 28. Let R = k[x]/(f ). Find ΩR/k . By the previous proposition we

have
Ωk[x]/k
ΩR/k ' .
k[x]hdi : i ∈ (f )i
Notice that if i ∈ (f ), then i = f g, and hence

di = f dg + gdf = f dg + gf 0 dx,

therefore Rhdi : i ∈ (f )i ⊆ f k[x]dx + f 0 k[x]dx. Conversely

f 0 dx = df ∈ Rhdi : i ∈ (f )i,

and also d(xf ) = xdf + f dx, so that

f dx = d(xf ) − xdf ∈ Rhdi : i ∈ (f )i.

Hence Rhdi : i ∈ (f )i = f k[x]dx + f 0 k[x]dx. This means

k[x]dx k[x] R
ΩR/k ' ' ' 0 .
f k[x]dx + f 0 k[x]dx (f, f 0 ) (f )

Definition. Let R → S be a map of algebras, and assume that S is finitely

generated as an R-algebra. For every R-algebra T and for every ideal J ⊆ T
such that J 2 = 0 define the natural map

θT,J : Homalg alg

R (S, T ) → HomR (S, T /J),

where Homalg
R (·, ·) denotes the module of R-algebra homomorphisms. Then

(A) S is said to be smooth over R if θT,J is surjective for all T, J as above.

(B) S is said to be unramified over R if θT,J is injective for all T, J as above.

(C) S is said to be étale over R if it is both smooth and unramified, i.e. if θT,J
is an isomorphism for all T, J as above.

We will prove the following theorem:

Theorem 52. Let R → S be as in the above definition. Then

(A’) S is smooth over R if and only if S is flat over R and ΩS/R is a projective
S-module.

(B’) S is unramified over R if and only if ΩS/R = 0.

(C’) S is étale over R if and only if S is flat over R and ΩS/R = 0.

70 The Module of Differentials

2 Second construction of the Module of Differ-

entials
Let k be a ring, and let R be a k-algebra. Map

R ⊗k R
µ
/R
P  /
P
i ri ⊗ si i ri si

and let I := ker µ. We will prove that the following is an isomorphism.

I /I 2
' / ΩR/k

(r ⊗ 1 − 1 ⊗ r) + I 2  / dr

Remark 47. R ⊗k R has both left and right module structures, in fact we can
consider
X X X X
r( si ⊗ ti ) = (rsi ) ⊗ ti and ( si ⊗ ti )r = si ⊗ (rti ).
i i i i

Remark 48. With respect to either left or right R-module structure we have

I = Rhr ⊗ 1 − 1 ⊗ r : r ∈ Ri

Proof. We just prove the case with the left module structure. The other case
I i ∈ I , so that
P
is
P similar. Clearly Rhr ⊗ 1 − 1 ⊗ r : r ∈ Ri ⊆ . Let i ri ⊗ s
i ri si = 0. Consider
X X X
− ri (si ⊗ 1 − 1 ⊗ si ) = (−ri si ) ⊗ 1 + ri ⊗ si .
i i i

⊗ si ∈ I and hence the remark follows.

P
Hence i ri

Remark 49. The two R-module structures on I /I 2 are the same.

Proof. It is enough to show it on the generators. Let r, s ∈ R, then

s(r⊗1−1⊗r)−(r⊗1−1⊗r)s = rs⊗1−s⊗r−r⊗s+1⊗rs = (r⊗1−1⊗r)(s⊗1−1⊗s) ∈ I 2 ,

that is s(r ⊗ 1 − 1 ⊗ r) = (r ⊗ 1 − 1 ⊗ r)s in I /I 2 .

Remark 50. The map

d:R / I /I 2
 / r⊗1−1⊗r
r
is a k-derivation.
2 Second construction of the Module of Differentials 71

Proof. Let r, s ∈ R, then

d(r + s) = (r + s) ⊗ 1 − 1 ⊗ (r + s) = (r ⊗ 1 − 1 ⊗ r) + (s ⊗ 1 − 1 ⊗ s) = d(r) + d(s).

Also

d(rs) = (rs) ⊗ 1 − 1 ⊗ (rs) = s(r ⊗ 1 − 1 ⊗ r) + (s ⊗ 1 − 1 ⊗ s)r.

Since in I /I 2 left and right action are the same we get (s ⊗ 1 − 1 ⊗ s)r =
r(s ⊗ 1 − 1 ⊗ s), and therefore

d(rs) = rd(s) + sd(r).

Finally, if α ∈ k and r ∈ R, we get

d(αr) = (αr) ⊗ 1 − 1 ⊗ (αr) = α(r ⊗ 1 − 1 ⊗ r) = αd(r).

Before proving that (I /I 2 , d) is in fact the module of differentials we want

to introduce the notion of idealization. Let R be a ring and let M be a R-module.
Define
S := R n M := {(r, m) : r ∈ R, m ∈ M },
adding componentwise and multiplying using the following law

(r, m) · (s, n) = (rs, rn + sm).

Another way to see this multiplication is ”using distributive property with M 2 =

0”:

(r, m)·(s, n) “ = ” (r+m)·(s+n) = rs+rn+sm+mn = rs+rn+sm “ = ” (rs, rn+sm).

With this choices R n M is a commutative ring with identity (1, 0). Another
way to see this is
  
r m
RnM = : r ∈ R, m ∈ M
0 r

with the usual ring operations.

Remark 51. If R is a Cohen-Macaulay ring and ωR is its canonical module, then
2
S = R n ωR is Gorenstein. Notice also that S/ωR ' R, and ωR = 0 in S, hence
up to radical every Cohen-Macaulay module is Gorenstein.

Question. Let R be a domain and let M be a torsion free R-module. When

•
does there exists an idealization S = R n M such that S is also a domain?

We are now ready to prove the theorem.

72 The Module of Differentials

Theorem 53. With the above notation we have

I /I 2
' / ΩR/k

r⊗1−1⊗r o / dr

Proof. We want to show that (I /I 2 , d) satisfies the universal property, i.e. if

D ∈ Derk (R, M ) for some M R-module, then there exists f : I /I 2 → M such
that
R
d / I /I 2

D
 } f
M
As usual, since I /I = Rhr ⊗ 1 − 1 ⊗ ri, if f exists it is forced to be unique,
2

since
f (r ⊗ 1 − 1 ⊗ r) = f (dr) = D(r).
Let S = R n M . There exists a k-algebra homomorphism

R ⊗k R
h /S

r⊗s  / (rs, sD(r)) = (r, D(r)) · (s, 0)

and this follows from the fact that the map

R ⊗k R
h /S

 / (rs, sD(r)) = (r, D(r)) · (s, 0)

(r, s)

is k-bilinear, and from the universal properties for tensor products. Notice that

h(r ⊗ 1 − 1 ⊗ r) = h(r ⊗ 1) − h(1 ⊗ r) = (r, D(r)) − (r, 0) = (0, D(r)),

hence there exists a map

fe : I /M

 / D(r)
r⊗1−1⊗r

Also fe(I 2 ) = 0 since fe is a map of rings and hence

fe(I 2 ) = fe(I ) · fe(I ) ⊆ M 2 = 0.

Therefore fe induces the desired map f : I /I 2 → M that makes the above

diagram commute.
3 The Jacobi-Zariski Sequence 73

3 The Jacobi-Zariski Sequence

ϕ
Remark 52. Consider k → R → S. Then there exists a natural S-module
homomorphism
S ⊗R ΩR/k / ΩS/k

Proof. There exists a R-linear map

ΩR/k / ΩS/k
 / dϕ(r)
dr

This gives a R-bilinear map

S × ΩR/k / ΩS/k

and hence we get a S-linear homomorphism

S ⊗R ΩR/k / ΩS/k

Question. When is the map an isomorphism? When is there a left inverse?

Definition. We say that a R-module homomorphism α : M → N is left split if

there exists β : N → M a map such that β ◦ α = idM .

Remark 53. α : M → N is left split if and only if for all R-modules K the
induced map HomR (α, K) : HomR (N, K) → HomR (M, K) is onto.

Proof. Assume α is left split, so that N ' M ⊕ L via α. Then just extend any
map f : M → K just by defining it to be zero on L. In other words

HomR (N, K) ' HomR (M, K) ⊕ HomR (L, K),

so that HomR (N, K) → HomR (M, K) is onto.

Conversely assume that the map HomR (α, K) : HomR (N, K) ' HomR (M, K)⊕
HomR (L, K) is onto for all R-modules K. Choose M = K, then we can
lift the identity map idM : M → M to a map β : N → M , and clearly
HomR (α, K)(β) = β ◦ α = idM , so that α is left split.

Remark 54. If M is finitely generated it is enough to check just for all K finitely
generated.
ϕ
Proposition 54. Let k → R → S. Then

(1) The natural map S ⊗R ΩR/k → ΩS/k is left split if and only if for all
S-modules N and for all D ∈ Derk (R, N ), D can be extended to a k-
e : S → N.
derivation D
74 The Module of Differentials

(2) The natural map S ⊗R ΩR/k → ΩS/k is an isomorphism if and only if the
extension is unique.
Proof. (1) Rephrasing the statement we have to prove that S ⊗R ΩR/k → ΩS/k
is left split if and only if Derk (S, N ) → Derk (R, N ) is onto. But we know
that
Derk (S, N ) = HomS (ΩS/k , N ) → HomR (ΩR/k , N ) = Derk (R, N ).
By Hom-tensor adjointness we get
HomS (ΩS/k , N ) → HomR (ΩR/k , HomS (S, N )) ' HomS (ΩR/k ⊗R S, N ),
which is onto if and only if the map S ⊗R ΩR/k → ΩS/k is left split by
Remark 53.
(2) With the same argument used in (1) we have that the extension is unique
if and only if Derk (S, N ) ' Derk (R, N ), if and only if HomS (ΩS/k , N ) '
HomS (ΩR/k ⊗R S, N ), if and only if the map S ⊗R ΩR/k → ΩS/k is an
isomorphism, since the statement is true for all S-modules N .

Corollary 55. Let W ⊆ R be a multiplicatively closed set, where R is a k-

algebra. Then

ΩR/k ⊗R RW ' ΩR/k W ' ΩRW /k .
Proof. Consider k → R → RW . By the above proposition the corollary holds if
and only if
Derk (R, N ) ' Derk (RW , N ) for all N.
Let D : R → N be a k-derivation. Uniqueness is forced, in fact for w ∈ W :
 r r r
D(r) = D w = wD + D(w).
w w w
Hence  r  D(r) − r D(w)
w
D =
w w
is forced. Finally note that using this equality as a definition, we get a k-
derivation D : RW → N , so that Derk (R, N ) ' Derk (RW , N ) and the corollary
is proved.
Remark 55. Suppose k → k 0 is a ring homomorphism and suppose R is a k-
algebra. Write R0 := R ⊗k k 0 . Then
k 0 ⊗k ΩR/k ' ΩR0 /k0 .
Theorem 56. Consider k → S → S/I = R for some I ⊆ S ideal. Then there
exists an exact sequence
I d / ΩS/k / ΩR/k /0
I2 IΩS/k

where d(i + I 2 ) = di + IΩS/k .

3 The Jacobi-Zariski Sequence 75

Proof. The last homomorphism is the natural map

ϕ : R ⊗S ΩS/k / ΩR/k ,

but here R = S/I, hence

ΩS/k
R ⊗S ΩS/k ' .
IΩS/k

Clearly the map is onto, since for all r ∈ R there exists s ∈ S lifting r, and
hence if dr ∈ ΩR/k , then ds 7→ dr. By Proposition 51 we also have

ΩS/k
ΩR/k ' .
Shdi : i ∈ Ii
Notice that IΩS/k ⊆ Shdi : i ∈ Ii, so that

Shdi : i ∈ Ii
ker ϕ = .
IΩS/k

Clearly I does surject onto Shdi : i ∈ Ii via i 7→ di, but also for all i, i0 ∈ I we
have d(ii0 ) = idi0 + i0 di ∈ IΩS/k , so that we get

I d / ΩS/k ϕ
/ ΩR/k /0
I2 7 IΩS/k
%
Shdi:i∈Ii
6 IΩS/k

0 )0

Example 29. Let k be a field of characteristic not equal to 2 or 3. Let also

S = k[x, y] and let R = S/(f ) = k[x, y]/(x2 + y 3 − 1). We know that ΩS/k '
Sdx ⊕ Sdy ' S 2 . Also f ∈ S is a regular sequence, so that
f
' S/(f ) = R.
f2
Also ΩS/k ⊗S R ' Rdx ⊕ Rdy ' R2 . Therefore we have an exact sequence

(2x,3y)
R / R2 / ΩR/k /0

because here the map d is given by the partial derivatives of f . Since 2, 3 are
units in R we have that (2x, 3y)R = R, that is (2x, 3y) is unimodular. Therefore
we get
ΩR/k ⊕ R ' R2
and in particular ΩR/k is projective.
76 The Module of Differentials

Example 30. Describe ΩR/k where R = k[x, y, z]/(x2 , y 3 , z 4 ), assuming 6 is a

unit in k. Then we have an exact sequence

R3 !/ R3 / ΩR/k / 0,
2x 0 0
0 3y 2 0
0 0 4z 3

so that
R R R
ΩR/k ' ⊕ 2 ⊕ 3 .
(x) (y ) (z )

Definition. Let k → S → S/I = R be as above, so that we have an exact

sequence
I d / ΩS/k / ΩR/k / 0.
I2 IΩS/k

If S = k[xλ ] is a polynomial ring, then we denote ker d =: ΓR/k . One can prove
that the definition does not depend on the choice of the presentation S.

Theorem 57 (Jacobi-Zariski sequence). Let k → R → S be ring homomor-

phisms. Then we have an exact sequence of S-modules

ΓS/k / ΓS/R / S ⊗R ΩR/k / ΩS/k / ΩS/R / 0.

Furthermore, if ΩR/k is flat over R, then we can add S ⊗R ΓR/k on the left, i.e.
the following sequence is exact

S ⊗R ΓR/k / ΓS/k / ΓS/R / S ⊗R ΩR/k / ΩS/k / ΩS/R / 0.

Proof. To define Γ·/· we can choose any presentation for R. Let us choose

k[xλ ] A k[xλ , yν ] B
R= =: and S = =: ,
I I L L
where IB ⊆ L. In this way

B R[yν ]
S= = ,
L L

where L = L/IB. Notice that there is a short exact sequence

0 / IB
' L2 +IB / L / L
= L /0
6
L2 ∩IB L2 L2 2 L2 +IB
6 L

I IB
I2 ⊗R S ' LIB
4 Quasi-unramified maps 77

therefore we get a commutative diagram

I / L / L /0
⊗R S 2
I2 L2 L
d⊗1 d d
  
/ (L Rdxλ ) ⊗R S /L /L /0
0 λ λ,ν (Sdxλ ⊕ Sdyν ) ν Sdν
  
ΩR/k ⊗R S ΩR/k ΩS/R
  
0 0 0
where !
I /
M
/ ΩR/k ⊗R S /0
⊗R S Rdxλ ⊗R S
I2
λ

is exact because tensor product is right exact. Hence, by the Snake Lemma we
get a long exact sequence

ker(d ⊗ 1) / ΓS/k / ΓS/R / S ⊗R ΩR/k / ΩS/k / ΩS/R / 0.

For the last statement assume that ΩR/k is flat over R. By definition we have
an exact sequence

/ ΓR/k / I d / / ΩR/k /0
M
0 Rdxλ
I2
; λ

;C
'
0 0
L
and since ΩR/k is flat and λ Rdxλ is free we have that C is flat over R. Tensor
the first half of the sequence to get

TorR
1 (C, S) = 0
/ ΓR/k ⊗R S / I ⊗R S / C ⊗R S /0
I2
d⊗1 L( 
( λ Rdxλ ) ⊗R S

Hence ker(d ⊗ 1) ' ΓR/k ⊗R S.

4 Quasi-unramified maps
Definition. Let k → R be a ring homomorphism. R is quasi-unramified over k
if for all k-algebras T and for all ideals J ⊆ T such that J 2 = 0 the map
Homalg alg
k (R, T ) → Homk (R, T /J)
78 The Module of Differentials

is injective.
Remark 56. The definition is the same as the one of unramified, except for the
fact that we are not assuming that R is a finitely generated k-algebra.
Theorem 58. Let k → R be a ring homomorphism. Then R is quasi-unramified
if and only if ΩR/k = 0.
Proof. Recall that ΩR/k ' I /I 2 , where I is given by

0 /I / R ⊗k R /R /0
r⊗s / rs

Suppose R is quasi-unramified. Set T = R ⊗k R/I 2 and J = I /I 2 ⊆ T , so

that T /J ' R ⊗k R/I ' R. Consider the following diagram
( ,
r r⊗1
R 3+ T
r 2 1⊗r
id

$ x
T /J
{
0
Then we have to liftings of the identity, hence they have to coincide. Hence
r ⊗ 1 + I 2 = 1 ⊗ r + I 2 , i.e.

r ⊗ 1 − 1 ⊗ r ∈ I 2.

These elements generate I , therefore I = I 2 and

I
ΩR/k ' = 0.
I2
Conversely assume ΩR/k = 0 and suppose we have a diagram
α
R ,2 T
β

# {{
T /J

We want to prove that α = β. Consider the ring homomorphism

R ⊗k R
ψ
/T

r⊗s  / α(r)β(s)
4 Quasi-unramified maps 79

Notice that ψ(r ⊗ 1) − ψ(1 ⊗ r) = α(r) − β(r) ∈ J because α = β mod J.

Hence ψ(I ) ⊆ J and therefore ψ(I 2 ) = ψ(I ) · ψ(I ) ⊆ J 2 = 0. But I /I 2 '
ΩR/k = 0, and hence ψ(I ) = ψ(I 2 ) = 0. Therefore, for all r ∈ R

α(r) − β(r) = ψ(r ⊗ 1 − 1 ⊗ r) = 0,

that is α = β.
Corollary 59. Let k → R → S be ring homomorphisms. Then
(1) If S/k (i.e. S is unramified over k) is unramified, then so is S/R.
(2) (Transitivity) If R/k and S/R are unramified, then so is S/k.
Proof. Use Jacobi-Zariski sequence and the previous theorem.
Example 31. Let k be a field and let k ⊆ ` be a finite separable field extension.
Then there exists a primitive element, say
k[x]
` = k(α) ' ,
(f (x))

with f 0 (α) 6= 0. We have an exact sequence

f (x) / `dx / Ω`/k /0

f 2 (x)
' '

 
` /` / Ω`/k /0
f 0 (α)

Then Ω`/k = 0 and therefore `/k is unramified.

Example 32. Let W ⊆ R be a multiplicatively closed set. We have already
seen that ΩRW /R = 0, hence RW /R is unramified.
Example 33. Let k → k[x]. We know that

Ωk[x]/k = k[x]dx ' k[x],

therefore k[x] is not unramified over k.

Example 34. Let R be a ring and let I ⊆ R be an ideal. One can easily prove
that Ω(R/I)/R = 0, hence the map R → R/I is unramified.
Example 35 (Base Change). Let k → R be quasi-unramified, and let k → k 0
be another ring homomorphism. Then

ΩR/k ⊗k k 0 ' ΩR⊗k k0 /k0 ,

therefore k 0 → R is quasi-unramified too.

80 The Module of Differentials

Definition (Separable algebraic). Let k be a field, and let k be its algebraic

closure. Then α ∈ k is said to be separable over k if (f, f 0 ) = 1, where f (x) =
min(α, k) the minimal polynomial of α over k. A filed extension k ⊆ l ⊆ k is
separable over k if every α ∈ l is separable over k.
Remark 57. (f, f 0 ) = 1 if and only if f (x) has distinct roots in k[x].
Remark 58. Let k ⊆ k 0 ⊆ l ⊆ k be field extensions, and assume l is separable
over k. Then l is separable over k 0 , because the minimal polynomial of α over
k 0 divides the minimal polynomial of α over k.
Remark 59. Let k ⊆ l = k(α) be a simple algebraic field extension. Then l is
separable over k if and only if k → l is unramified.
Theorem 60. Let k ⊆ ` ⊆ k be field extensions, with k ⊆ ` finite. Then the
following facts are equivalent:
(1) ` is separable over k.
(2) Ω`/k = 0.
(3) Γ`/k = 0.
Proof. (2) ⇐⇒ (3) We can write

k[x1 , . . . , xn ]
`= ,
m
where m is a maximal ideal, and we know that
m
dim` = n,
m2
since k[x1 , . . . , xn ] is regular. There is an exact sequence of `-vector spaces
n
/ Γ`/k / m d / / Ω`/k / 0.
M
0 `dxi
m2 i=1

Looking at the dimensions we get

dim` Ωl/k = dim` Γ`/k ,

and this implies the equivalence between (2) and (3).

(2) + (3) ⇒ (1) Let α ∈ `, and consider k ⊆ k(α) ⊆ `. Using the Jacobi-Zariski
sequence we get

Γ`/k(α) / Ωk(α)/k ⊗k(α) ` Ω`/k / Ω`/k(α) / 0.

By assumption Ω`/k = 0, and hence Ω`/k(α) = 0. But since we have already

shown that (2) ⇒ (3) we have also that Γ`/k(α) = 0, and therefore

Ωk(α)/k ⊗k(α) ` = 0.
4 Quasi-unramified maps 81

Since k(α) → ` is faithfully flat (it is just a field extension) we have Ωk(α)/k = 0,
and hence α is separable over k.
(1) ⇒ (2) Use induction on m = [` : k]. The case m = 1 is clear. Now assume
m > 1, and choose α ∈ ` r k. Then ` is separable over k(α) by transitivity. By
induction Ω`/k(α) = 0, and also Ωk(α)/k = 0 since α is separable over k. Then,
using the Jacobi-Zariski sequence we get

Ωk(α)/k ⊗k(α) ` = 0 / Ω`/k / Ω`/k(α) = 0,

and therefore Ω`/k = 0.

Theorem 61. Let k be a field and let R be a finitely generated k-algebra. Then
the following are equivalent:
(1) k → R is unramified.
(2) R ' k1 × . . . × kr , where each ki is field which is finite and separable over
k.
Proof. (2) ⇒ (1) One can prove that if R ' k1 × . . . × kr , then
r
M
ΩR/k ' Ωki /k .
i=1

Since each ki is finite and separable over k, by Theorem 60 we have Ωki /k = 0

for all i = 1, . . . , r, and hence ΩR/k = 0.
(1) ⇒ (2) Let m ∈ MaxSpec(R). Then R → R/m2 is unramified, and by
transitivity k → m/m2 is also unramified. Call A = R/m2 and call mA = m/m2
its maximal ideal. Since R is a finitely generated k-algebra, by Nullstellensatz
l = A/mA is a finite algebraic extension of k. In particular dimk A < ∞, because
[l : k] < ∞ and A is artinian, hence diml A < ∞. Consider

0 /I / A ⊗k A /A / 0,

so that
I /I 2 ' ΩA/k = 0.
Then I = I 2 and it is finitely generated, hence I = (e), where e2 = e is an
idempotent. Assume now k = k is algebraically closed, so that l = k. Notice
that
A ⊗k A
' (A ⊗k A) ⊗A l = (A ⊗k A) ⊗A k = k ⊗k k = k,
mA ⊗k A + A ⊗k mA
is a field, therefore mA ⊗k A+A⊗k mA is a maximal ideal in A⊗k A, but it is also
nilpotent since m2A = 0. This means that A ⊗k A has only one maximal ideal,
and therefore it is local. But a local ring has no trivial idempotents, and hence
e = 0, 1. Clearly e 6= 1, otherwise A = 0. So I = 0 and therefore A ⊗k A ' A.
But then
dimk A = (dimk A)2 ,
82 The Module of Differentials

which implies dimk A = 1, and then mA = 0, which is m = m2 . This again

implies that m is generated by an idempotent. Since R is noetherian, and m
was an arbitrary maximal ideal in R, R must have only finitely many maximal
ideals (a noetherian ring cannot have infinitely many idempotents) and also

R ' k × . . . × k,

proving the theorem in the case k = k. Back to the general case, by base change
we have
Ω(R⊗k k)/k ' ΩR/k ⊗k k = 0,

so that R ⊗k k ' k × . . . × k (r copies) by what we have already proved.

Therefore
dimk R = dimk R ⊗k k = r.
Moreover k ,→ k is of course a flat extension, therefore

R = R ⊗k k ,→ R ⊗k k.

This implies that R is reduced, and hence is a direct product of fields:

R ' k1 × . . . × ks ,
Ps
Ls ki ⊆ k is finite over k and i=1 [ki : k] = r. Finally, since 0 =
where each
ΩR/k = i=1 Ωki /k we have that each Ωki /k = 0 for all i = 1, . . . , s and hence
each ki is separable over k by Theorem 60.
Proposition 62. Let ϕ : R → S be a homomorphism of noetherian rings,
and assume S is a finitely generated R-algebra. Then the following facts are
equivalent:
(1) R → S is unramified.
(2) For all Q ∈ Spec(S), set q = ϕ−1 (Q), then k(q) → S⊗R k(q) is unramified.
Proof. (1) ⇒ (2) Let Q ∈ Spec(S) and let q be as above. By base change we
get
Ω(S⊗R k(q))/k(q) ' ΩS/R ⊗R k(q) = 0
since R → S is unramified. Therefore S ⊗R k(q) is unramified for all Q ∈
Spec(S).
(2) ⇒ (1) Since S is a finitely generated R-algebra we have that ΩS/R is a
finitely generated S-module. Then to show ΩS/R = 0 it is enough to show that
(ΩS/R )Q = 0 for all Q ∈ Spec(S), and hence it is enough to show (ΩS/R )q = 0
for all q ∈ Spec(R) since

(ΩS/R )q = (ΩS/R )ϕ(Rrq)

and then

(ΩS/R )Q = (ΩS/R )ϕ(Rrq) Q
4 Quasi-unramified maps 83

is just a further localization. By base change

(ΩS/R )q = ΩS/R ⊗R Rq ' ΩSq /Rq .

ΩSq /Rq is finitely generated, and hence by NAK it is enough to show that

ΩSq /Rq
= 0.
qΩSq /Rq

Finally, by base change

ΩSq /Rq
Rq
' ΩSq /Rq ⊗Rq ' Ω(S⊗R k(q))/k(q) = 0.
qΩSq /Rq qRq

Theorem 63. Let R → S be a homomorphism of noetherian rings, and assume

S is a finitely generated R-algebra. Then the following facts are equivalent:

(1) R → S is unramified.

(2) For all q ∈ Spec(R), S ⊗R k(q) ' k1 × . . . × krq , where each k1 is finite
and separable over k(q).

Proof. Follows immediately from Theorem 61 and Proposition 62.

Theorem 64 (Local Structure Theorem). Let (R, mR ) → (S, mS ) be essentially

of finite type (i.e. S is the localization of a finitely generated R-algebra). Assume
R → S is quasi-unramified. Then there exists a R-algebra
 
R[x]
T =
(f (x)) Q

such that

(1) f is monic and f 0 (x) ∈ Q.

(2) There exists a surjective map R → T  S such that

T S
' .
mR T mS

Example 36. Consider C → C[x] and let q = (0). Then

C[x] ⊗C k(q) = C[x] ⊗C C = C[x],

which is not finite over C. Then C → C[x] is not unramified.

84 The Module of Differentials

2 3
Example 37. Consider C[x] → (xC[x,y]
2 −y 3 ) ' C[t , t ]. Prime ideals in C[x] are (0)

and (x − α), with α ∈ C. First let us check q = (0):

 
C[x, y] C(x)[y]
C(X) → 2 3
= 2 .
(x − y ) q (x − y 3 )

The polynomial x2 − y 3 is irreducible in C(x)[y], hence C(X) → (x C(x)[y]

2 −y 3 ) is

unramified. Let now q = (x − α), so that we get

 
C[x, y] C[y]
C = C ⊗ k((x − α)) −→ ⊗ k((x − α)) ' 2 .
(x2 − y 3 ) (α − y 3 )

If α = 0, then this extension is not unramified, so that C[x] → (xC[x,y]

2 −y 3 ) itself is

not unramified. Notice that to make it unramified it is enough to invert y or

y 2 , i.e. to consider  
C[x, y]
C[x] →
(x2 − y 3 ) y
or similarly  
C[x, y]
C[x] → .
(x2 − y 3 ) y2

5 Quasi-smooth maps
Definition. Let k → R be a ring homomorphism. R is quasi-smooth over k if
for all k-algebras T and for all ideals J ⊆ T such that J 2 = 0 the map

Homalg alg
k (R, T ) → Homk (R, T /J)

is surjective. R is quasi-étale over k if k → R is quasi-unramified and quasi-

smooth.
Remark 60. As for quasi-unramified and unramified, these definitions the same
as the ones of smooth and étale, except for the fact that we are not assuming
that R is a finitely generated k-algebra.
Theorem 65. Let k → R be ring homomorphism. Then it is quasi-smooth if
and only if

ΓR/k = 0 and ΩR/k is a projective R − module. (3.1)

Proof. We want to find a condition equivalent to (3.1). Write R = k[xλ ]/I, then
we have to following defining sequence:

/ ΓR/k / I d / / ΩR/k / 0,
M
0 Rdxi (3.2)
I2

Rdxi → I/I 2 splitting map

L
Hence (3.1) holds if and only if there exists ψ :
for d.
5 Quasi-smooth maps 85

Claim. (3.2) holds if and only if

∃δλ ∈ I such that f (xλ − δλ ) ∈ I 2 ∀f ∈ I. (3.3)

Proof
L of the Claim.2 Assume (3.2) holds. This is true if and only if there exists
ψ: Rdxi → I/I such that for f ∈ I we get
X 
∂f
ψ ) dxλ = f + I 2 ,
∂xλ

that is if and only if there exists δλ = ψ(dxλ ) such that

X ∂f
δλ = f (xλ ) + I 2 .
∂xλ

Recall now that

X ∂f
f (x1 , . . . , xn ) = f (α1 , . . . , αn ) + (xj − αj ) mod (x1 − α1 , . . . , xn − αn )2 .
∂xj

Set αλ = xλ − δλ , so that
X ∂f
f (x1 , . . . , xn ) ≡ f (xλ − δλ ) + δλ mod (δλ )2 .
∂xλ
∂f
Since δλ = ψ(dxλ ) ∈ I we have that (δλ )2 ⊆ I 2 . But also f (xλ ) =
P
∂xλ δλ
mod I 2 . Hence we get
f (xλ − δλ ) ∈ I 2 .

Conversely assume f (xλ − δλ ) ∈ I 2 , then we can define ψ : dxλ 7→ δλ + I 2 .

So it is enough to prove the equivalence quasi-smooth if and only if (3.3).

Assume (3.3) holds, then consider a diagram

k[xλ ] T

π


/ T
R ϕ J

with J 2 = 0. Let rλ = π(xλ ) = xλ + I, and ϕ(rλ ) = tλ + J. k[xλ ] is just a

polynomial ring, therefore we can define a map

k[xλ ]
φ
/T

xλ  / tλ
86 The Module of Differentials

We need to find ελ ∈ J such that for all f ∈ I, f (tλ − ελ ) = 0. If we can find

such ελ then we can send rλ 7→ tλ − ελ and get the required lifting of ϕ. Set
ελ = φ(δλ ) ∈ J, because φ(I) ⊆ J by commutativity of the following diagram

k[xλ ]
φ
/T

π


R / T
ϕ J

Finally, since by assumption we have f (xλ − δλ ) ∈ I 2 , we get

f (tλ − ελ ) = φ(f (xλ − δλ )) ∈ J 2 = 0.

Conversely assume k → R is quasi-smooth, then construct δλ as follows. If

R = k[xλ ]/I set T = k[xλ ]/I 2 , and if J = I/I 2 then T /J ' R. By assumption
we get a diagram
9T
ϕ

/ T
R =R
idR J
Therefore ϕ(xλ + I) = xλ − δλ for some δλ ∈ I. This implies, for f ∈ I:

0 = ϕ(f ) = f (xλ − δλ ) ∈ T,

so that f (xλ − δλ ) ∈ I 2 .

Exercise 1. Let W ⊆ S be a multiplicatively closed set, R → S a ring homo-

morphism. Then

ΓS/R W ' ΓSW /R

Corollary 66 (of Theorem 65). A ring homomorphism R → S is quasi-étale if

and only if ΓS/R = ΩS/R = 0.

Proof. Quasi-étale if and only if quasi-smooth and quasi-unramified, if and only

if

 ΓS/R = 0 and ΩS/R is projective (quasi-smooth)

ΩS/R = 0 (quasi-unramified)


which is clearly equivalent to the condition ΓS/R = ΩS/R = 0.

Corollary 67 (Transitivity). Let k → R → S be ring homomorphisms. If

k → R and R → S are (quasi-)smooth, then so is k → S.
5 Quasi-smooth maps 87

Proof. Use the Jacobi-Zariski sequence:

ΓR/k ⊗R S / ΓS/k / ΓS/R / ΩR/k ⊗R S / ΩS/k / ΩS/R / 0,

where the first map comes from the fact that k → R is quasi-smooth, and hence
ΩR/k is projective, and in particular flat. Also, since R → S is quasi-smooth,
we have ΓS/R = 0 and ΩS/R is projective. Therefore the final part of sequence
above splits: M

ΩS/k ' ΩS/R ΩR/k ⊗R S .
Moreover, ΩR/k is a projective R-module, and as a consequence ΩR/k ⊗R S is a
projective S-module. This shows that ΩS/k is a projective S-module. Finally,
ΓS/R = ΓR/k = 0 since k → R and R → S are quasi-smooth, therefore ΓS/k = 0
by the sequence above. By Theorem 65 we get that k → S is quasi-smooth.
Theorem 68 (Cartier-Mac Lane). Let k be a field and let k ⊆ l be a finitely
generated field extension. Then

dim` Ωl/k = dim` Γl/k + tr.degk `.

Proof. We have discussed the case in which tr.degk ` < ∞, i.e. when ` is al-
gebraic over k. Now assume n = tr.degk ` > 0, and fix a transcendence basis
xi , . . . , xn of ` over k. Set E := k(x1 , . . . , xn ), and consider the inclusions
k ⊆ E ⊆ `, where E ⊆ ` is now algebraic. Then we have

ΓE/k ⊗E ` / Γ`/k / Γ`/E / ΩE/k ⊗E ` / Ω`/k / Ω`/E / 0,

where the first map comes from the fact that ΩE/k is clearly flat, being E a
field. Consider k → E, which is a purely transcendental extension. Notice that,
if we set R := k[x1 , . . . , xn ] and W = R r {0}, then we have that E = RW .
Since R is a polynomial ring over k we have that
n
M
ΩR/k ' Rdxi ,
i=1

and in particulare it is a free R-module. Localizing at W we get that

n
M
ΩE/k ' Edxi .
i=1

Also, ΓR/k = 0 since we can just choose I = 0 in the presentation of R as an

algebra k[x1 , . . . , xn ]/I over k. Putting things together, and going back to the
Jacobi-Zariski sequence we get

0 / Γ`/k / Γ`/E / `n / Ω`/k / Ω`/E / 0,

because !
n
M n
M
ΩE/k ⊗E ` ' Edxi ⊗E ` ' `dxi ' `n .
i=1 i=1
88 The Module of Differentials

Every module is finitely generated in the sequence, therefore dimension over l

are finite. From the sequence we get

dim` Ω`/E + n + dim` Γl/k = dim` Ω`/k + dim` Γ`/E .

From the algebraic case, since E ⊆ ` is algebraic, we know that dim` Ω`/E =
dim` Γ`/E , therefore

tr.degk ` + dim` Γ`/k = n + dim` Γ`/k = dim` Ω`/k .

Chapter 4

Basic element theory

1 Basic sets and basic elements

Definition. Let R be a commutative ring with 1R . A subset X ⊆ SpecR is
said to be basic if
(i) X is Noetherian (i.e. it has DCC on closed sets)
\ \
(ii) If pα ∈ X for α ∈ Λ, and pα ∈ SpecR, then pα ∈ X.
α∈Λ α∈Λ

Remark 61. Finite intersection of primes which are not nested are never prime.
Therefore the interesting cases in (ii) of the above definition happen when the
intersection is infinite.
Remark 62. If R is Noetherian, then SpecR is Noetherian. But the converse
does not hold in general.
Example 38. (1) When R is Noetherian, X = SpecR clearly is basic.
(2) When R is Noetherian,
 

 

\
X = j − SpecR = p ∈ SpecR | p = m

 p ⊆ m


m maximal

is basic.
(3) When R is Noetherian,

X i := {p ∈ SpecR | htp 6 i}

is basic.
(4) If X is basic and F ⊆ SpecR is closed, then X ∩ F is basic.
90 Basic element theory

Discussion. Closed sets in the Zariski topology on R are of the form

√
V (I) = {p ∈ SpecR | I ⊆ p} = V ( I) ←→ SpecR/I.
It’s the weakest topology that makes ring homomorphism continuous, that is if
ϕ : R → S is a ring homomorphism, then there is an induced map ϕ∗ : SpecS →
SpecR on spectra, and it has continuous in the Zariski topology. A basis of open
sets is given by {D(f )}f ∈R , where
D(f ) = {p ∈ SpecR | f ∈
/ p} ←→ SpecRf .
Definition. Let X be basic, and let p ∈ X. Define
dimX p := sup{n | ∃p = p0 ( p1 ( . . . ( pn , with pi ∈ X∀i}.
If X = SpecR then notice that dimX p = dim(R/p), the Krull dimension.
Proposition 69. Let X be basic. Then
(i) Every closed set F ⊆ X is a finite union of irreducible closed sets in X.
Recall that a set is irreducible if it cannot be written as a proper union of
smaller subsets.
(ii) If F ⊆ X is not empty, closed and irreducible, then F has a generic point
in X, i.e. F = V (p) ∩ X for some p ∈ X.
Proof. (i) It follows from the Noetherian property: given F ⊆ X closed, if it
is irreducible we are done. If it is reducible we can write it as a union of
two smaller closed sets. Then repeat of the smaller sets, and the process
ends by DCC.
(ii) Write F = V (I) ∩ X for some ideal I ⊆ R. Set
\
p0 = p.
p∈F

Notice that p0 ∈ X since X is basic, and also F = V (p0 ) ∩ X. In fact

F ⊆ V (p0 ) ∩ X, because if p ∈ F , then p ⊇ p0 . On the other hand
V (p0 ) ∩ X ⊆ V (I) ∩ X = F since I ⊆ p for all p ∈ V (p0 ) by definition
of p0 . Hence, without loss of generality we may assume I = p0 .We want
to prove that p0 is prime. If not there exist ab ∈ p0 such that a ∈ / p0 and
b∈/ p0 . Set F1 := V (p0 , a) ∩ X and F2 := V (p0 , b) ∩ X. Then F = F1 ∪ F2 ,
in fact it is clear that F1 ∪ F2 ⊆ V (p0 ) ∩ X = F . On the other hand, if
p ∈ F , then p0 ⊆ p and therefore ab ∈ p. But p is prime, therefore a ∈ p
or b ∈ p. So F = F1 ∪ F2 . By irreducibility, we have F = F1 or F = F2 ,
say F = F1 . But then (p0 , a) ⊆ p0 , that is a ∈ p0 and hence p0 is prime.

Notation. Let R be a ring, M ∈ Modfg (R) and let p ∈ SpecR. Define

µp (M ) := dimk(p) Mp /pMp = µRp (Mp ),
where the last equality follows by Nakayama’s lemma.
1 Basic sets and basic elements 91

Lemma 70. Let M ∈ Modfg (R). The set

Ft = {q ∈ SpecR | µq (M ) > t} ⊆ SpecR

is closed.

Proof. For t ∈ N∗ consider the ideal

X  
It = (mi1 , . . . , mit−1 ) :R M ,
mi1 ,...,mit−1 ∈M

i.e. the sum of colons of M into any submodule of M generated by t−1 elements.
Suppose p ∈ SpecR is such that It ⊆ p. Then µp (M ) > t, in fact if not there
exist t − 1 elements m1 , . . . , mt−1 such that (m1 . . . , mt−1 )p = Mp , and since
M is finitely generated this implies (m1 , . . . , mt−1 ) :R M 6⊆ p (basically clearing
denominators). On the other hand, if It 6⊆ p, then there exist m1 , . . . , mt−1 such
that (m1 , . . . , mt−1 ) :R M 6⊆ p, that is Mp = (m1 , . . . , mt−1 )p , and therefore
µp (M ) 6 t − 1. This shows that It defines Ft . i.e. Ft = V (It ) is closed.

Crucial Lemma 71. Let M ∈ Modfg (R) and let X ⊆ SpecR be basic. Then
there exists a finite set of primes Λ ⊆ X such that if p ∈ X r Λ, there exists
q ( p such that
µq (M ) = µp (M ) .

Proof. By Lemma 70, for all t ∈ N∗ Ft is a closed subset of SpecR. By Propo-

sition 69 Ft ∩ X is a finite union of irreducible closed sets, that is
[
Ft ∩ X = V (pi,t ) ∩ X,
i∈Λt

for prime ideals pi,t ∈ Ft . Define

[
Λ := {pi,t | t = 0, . . . , µ(M ), i ∈ Λt } ⊆ X,

which is a finite set. If p ∈ X r Λ, set t = µp (M ). Then p ∈ Ft ∩ X, i.e.

p ∈ V (pi,t ) for some i ∈ Λt . Then pi,t ( p, and the containment is strict
because p ∈/ Λ, while pi,t ∈ Λ. But by definition pi,t has at least t generators,
therefore
t = µp (M ) > µpi,t (M ) > t,
where the second inequality follows from the fact that pi,t ⊆ p, and therefore
the minimal number of generators potentially decreases when further localizing
at pi,t . But then we have

µp (M ) = µpi,t (M ).
92 Basic element theory

Definition. Let R be a ring and let M be an R-module. Then x ∈ M is a

p-basic element if
µp (M ) > µp (M/Rx) ,
which means that x is part of a minimal generating set of Mp . An element
x ∈ M is called X-basic if it is p-basic for all p ∈ X.

Theorem 72 (Eisenbud-Evans). Let R be a commutative ring with 1R (not

necessarily Noetherian). Let M ∈ Modfg (R) and let X be a basic set. Assume
that

(1) (a, y) ∈ R ⊕ M is X-basic.

(2) For all p ∈ X we have µp (M ) > 1 + dimX p.

Then there exists z ∈ M such that y + az is X-basic.

To prove the theorem we need a few more auxiliary definitions and results.

Notation. In the assumptions of the theorem, for p ∈ X and for S ⊆ M set

δp (S) := µp (M ) − µp (M/RS) ,

where RS denotes the R-submodule of M generated by the elements in S.

Remark 63. There is a short exact sequence

0 / RS /M / M/RS / 0.

Localizing at p and then tensoring with k(p) gives

Rp S α / Mp / Mp / 0.
pRp S pMp Rp S + pMp

These are k(p)-vector spaces and, if we denote by k(p)S = Im(α), we have

δp (S) = dimk(p) k(p)S.

Remark 64. Let X ⊆ SpecR be basic and let S ⊆ M be as above. Let Λ ⊆ X

be as in the Crucial Lemma 71. Then for all p ∈ X r Λ there exists q ∈ X such
that q ( p and δq (S) 6 δp (S).

Proof. Let M 0 = M/RS. By the Crucial Lemma 71 we know that except for
finitely many primes p ∈ X (the ones not in Λ) there exists q ( p such that
µp (M 0 ) = µq (M 0 ). Also, µp (M ) > µq (M ) always holds since it is a further
localization. But then

δq (S) = µq (M ) − µq (M 0 ) 6 µp (M ) − µp (M 0 ) = δp (S) .
1 Basic sets and basic elements 93

Definition. Let X be basic and let p ∈ X. A subset S ⊆ M , where S =

{x1 , . . . , xn } is called p-basic if

δp (S) > min{n, 1 + dimX p}.

If S is p-basic for all p ∈ X, then S is said to be X-basic.

Remark 65. Suppose S = {x} and p ∈ X. Then S is p-basic if and only if x is
p-basic, i.e. the two definitions coincide when S consists of one element.
Proof. If S is p-basic, then δp (S) > min{1, 1 + dimX p} = 1, then

µp (M/Rx) < µp (M ) ,

i.e. x is p-basic. Conversely, if x is p-basic, we have µp (M ) − µp (M/Rx) > 1,

that is δp (S) > 1 = min{1, 1 + dimX p}, so that S is p-basic. Notice that
we always have µp (M ) − µp (M/Rx) 6 1, therefore if S = {x} is p-basic we
necessarily have δp (S) = 1.
Remark 66. Suppose X is basic and M satisfies (2) in Theorem 72, that is
µp (M ) > 1 + dimX p for all p ∈ X. Let S be any (finite) set of generators of
M . Then for all p ∈ X

δp (S) = µp (M ) − µp (0) = µp (M ) > 1 + dimX p > min{|S|, 1 + dimX p},

and therefore S is X-basic.

Main Lemma 73. Let S = {x1 , . . . , xn } ⊆ M be X-basic. Assume that
(a, x1 ) ∈ R ⊕ M is X-basic. Then there exist a1 , . . . , an−1 ∈ R such that

S 0 = {x01 , . . . , x0n−1 } = {x1 + aa1 xn , x2 + a2 xn , . . . , xn−1 + an−1 xn }

is X-basic.
Proof. We claim that for any choice of a1 , . . . , an−1 , S 0 is p-basic for all but
finitely many primes in X. In fact recall Remark 64, and let p ∈ X r Λ. Notice
that R(S 0 ∪ {xn }) = RS, and therefore

δp (S 0 ) = δp (S) > min{n, 1 + dimX p},

because S is X-basics by assumption. But we also have

µp (M/RS) = µp (M/R(S 0 ∪ {xn })) 6 µp (M/RS 0 ) − 1,

therefore δp (S 0 ) > δp (S) − 1. By Remark 64 there exists q ( p, q ∈ X such

that δq (S) 6 δp (S). Putting things together:

δp (S 0 ) > δp (S) − 1 > δq (S) − 1 > min{n, dimX q} − 1 >

> min{n − 1, dimX q} > min{n − 1, 1 + dimX p},

94 Basic element theory

where the last inequality follows from the fact that the containment q ( p is
strict. Now set {p1 , . . . , pr } = Λ, which are the primes for which the claim does
not hold. Choose pr minimal among the pi ’s in Λ. By induction on r we assume
that we can choose a1 , . . . , an−1 such that S 0 is pi -basic for all i = 1, . . . , r − 1.
By minimality we can now choose c ∈ p1 · . . . · pr−1 r pr . Set
 00
 x1 = x01 + acb1 xn
x00 = x02 + cb2 xn



 2


.........

 .........
.........



 00
xn−1 = n0n−1 + cbn−1 xn


for some b1 , . . . , bn−1 to be determined. Set S 00 = {x001 , . . . , x00n }. Fix 1 6 i 6 r−1

and set M (pi ) = M ⊗ k(pi ). Since c ∈ qi we have x00j = x0j in M (pi ), for all
j = 1, . . . , n − 1. Therefore δpi (S 0 ) = δpi (S 00 ), and so S 00 is pi -basic since S 0 is.
Now we need to choose b1 , . . . , bn−1 so that S 00 is also pr -basic. We distinguish
three cases:
(a) x01 , . . . , x0n−1 are linearly independent in M (pr ). Then we have a short exact
sequence

n−1
/ / M (pr ) / P M (pr ) M / 0.
X
0 k(pr )x0i n−1 0
= 0
(pr )
i=1 i=1 k(p )x
r i RS

Since they are linearly independent we get

δpr (S 0 ) = n − 1 > min{n − 1, 1 + dimX pr },

so just take b1 = . . . = bn−1 = 0.

(b) x01 = 0. Since (a, x1 ) is X-basic by assumption, we have a ∈
/ pr , otherwise
(a, x1 ) = 0 in (R ⊕ M ) ⊗ k(pr ), and it cannot be basic. Set b1 = 1, b2 =
. . . = bn−1 = 0. The image of S 00 in M (pr ) is then

k(pr )S = k(pr ){acxn , x02 , . . . , x0n−1 }.

By choice of c and by what we just observed we have that ac is a unit in

k(pr ), hence
k(pr )S 00 = k(pr ){x02 , . . . , x0n−1 , xn }.
But x0i = xi + cbi xn , therefore

k(pr )S 00 = k(pr ){x2 , . . . , xn−1 , xn }.

Finally, by definition of x01 we have 0 = x01 = x1 + λxn for some λ, and

therefore x1 = λxn in k(pr ). Therefore

k(pr )S 00 = k(pr ){x1 , . . . , xn } = k(pr )S,

1 Basic sets and basic elements 95

which gives

δpr (S 00 ) = δpr (S) > min{n, 1 + dimX pr } > min{n − 1, 1 + dimX pr },

that is S 00 is pr basic.
(c) x01 6= 0 and x01 , . . . , x0n−1 are linearly dependent in k(pr ). There exists
1 6 i 6 n − 1 such that x0i is not zero and can be expressed in terms of the
others. Set bi = 1 and bj = 0 for all i 6= j. Then

k(pr )S 00 = k(pr ){x01 , . . . , x0i + cxn , . . . , x0n−1 } =

= k(pr ){x01 , . . . , xbi , . . . , x0n−1 , . . . , cxn }.

because x0i can be expressed in terms of the others. Now, c is a unit in

k(pr ), therefore

k(pr )S 00 = k(pr ){x01 , . . . , xbi , . . . , x0n−1 , xn } = k(pr ){x1 , . . . , xbi , . . . , xn }.

Finally, since

x0i = xi + λxn ∈ k(pr ){x01 , . . . , x0i , . . . , x0n−1 , xn } = k(pr ){x1 , . . . , xbi , . . . , xn }

b

we also get xi ∈ k(pr ){x1 , . . . , xbi , . . . , xn }, that is

k(pr )S 00 = k(pr ){x1 , . . . , xi , . . . , xn } = k(pr )S

and therefore S 00 is pr -basic again.

We are now ready to prove Eisenbud-Evans’ theorem 72.

Proof of Theorem 72. Let S = {y, x2 , . . . , xn } be a generating set of M . Set
x1 := y. By condition (2), that is µp (M ) > 1 + dimX p for all p ∈ X and by
Remark 66 we have that x1 is X-basic. By the Main Lemma 73 we can find
a1 , . . . , an−1 such that

S 0 = {x1 + aa1 xn , x02 , . . . , x0n−1 }

is X-basic. Notice that the cardinality of S 0 is one less than the cardinality of
S. Repeat the process until the cardinality of the set, say S 0 ”, is one. But then
S 00 = {x1 + az} = {y + az} for some z ∈ M , and S 00 is X-basic. But by Remark
65 this exactly means that y + az is X-basic.
Corollary 74 (of Theorem 72). Let R be a commutative ring with 1R , let X
be basic and let M ∈ Modfg (R) be such that µp (M ) > 1 + dimX p for all p ∈ X
(i.e. it satisfies just (2) in Theorem 72). Then there exists z ∈ M which is
X-basic.
96 Basic element theory

Proof. Set a = 1, y = 0. Then (1, 0) is always X-basic in R ⊕ M , and therefore

by Theorem 72 there exists z ∈ M such that z = y + az is X-basic.

Corollary 75 (of Theorem 72). Let R be a Noetherian ring of dimension d and

let I ⊆ R be an ideal. Then there exist d + 1 elements a1 , . . . , ad+1 such that
p
I ⊆ (a1 , . . . , ad+1 ).

Proof. Let M = ⊕d I, and take X = SuppM , which is closed in SpecR (because

M is finitely generated) and hence basic. Note that for p ∈ X

µp (M ) > (d + 1)µp (I) > d + 1 > 1 + dimX p.

The first inequality follows from Nakayama’s Lemma, while the second is be-
cause p ∈ X = SuppM . By Corollary 74 there exists z = (a1 , . . . , ad+1 ) ∈ M
which is X-basic.
p
Claim. I ⊆ (a1 , . . . , ad+1 ).
In fact if not there exists p ⊇ (a1 , . . . , ad+1 ) such that p 6⊇ I. But then

Mp = ⊕d Ip = Rpd+1 ,

i.e. (a1 , . . . , ad+1 ) ∈ pRpd+1 . By Nakayama’s Lemma this contradicts the fact
that z = (a1 , . . . , ad+1 ) is basic.

2 Basic elements and Projective modules

Question. What does it mean for any element z ∈ P to be basic at q ∈ SpecR
if P is a finitely generated projective module?

Discussion. Assume that P = Rt is free, and that z = (z1 , . . . , zt ) ∈ P . Then

z is q-basic if and only if it is a minimal generator for Rpt , i,e, z ∈ / qRqt . But
this is equivalent to say that (z1 , . . . , zt ) 6⊆ q (where here we mean the ideal
generated by z1 , . . . , zt in R). This means that (z1 , . . . , zt ) is a unimodular row
in Rqt , and therefore
Rqt ' zR ⊕ Rqt−1 .
If P is any projective module, locally we have Pq ' Rqt for some t. Therefore
z ∈ P is q-basic if and only if

Pq ' Rqt ' zRq ⊕ Qq

for some Q ∈ Modfg (R).

Question. Given P ∈ Modfg (R) a projective module and given z ∈ P , what is

Uz := {q ∈ SpecR | z is q − basic}?
2 Basic elements and Projective modules 97

Discussion. First suppose P ' Rt is free. Then by the above discussion it is

easy to see that for z = (z1 , . . . , zt ) ∈ Rt we have

Uz = SpecR r V ((z1 , . . . , zt )).

For any M ∈ Modfg (R) define M ∗ (z) := {f (z) | f ∈ M ∗ }. If P is just projective

and z ∈ P , then we get

Uz = SpecR r V (P ∗ (z)),

since we can localize and easily reduce to the free case.

Corollary 76 (Serre’s Theorem). Let R be a Noetherian commutative ring with

1R . Let X = j-SpecR and d = dim(j-SpecR). Let P ∈ f gR be a projective
module and assume rankPm > d for all m ∈ max SpecR. Then P ' R ⊕ Q.

Proof. First notice that it is stated as a Corollary because it will follow from
Corollary 74 of Theorem 72. Notice that for all q ∈ X = j-SpecR we have (for
q ⊆ m ∈ max SpecR)

µq (P ) = rankPq = rankPm > d + 1 > dimX q + 1.

Therefore there exists a X-basic element z ∈ P , and then P ∗ (z) = R, i.e.

P ' Rz ⊕ Q.

Definition. With the same notation as in Chapter 2, we say that n ∈ N defines

a stable range for GL(R) if whenever r > n and (a1 , . . . , ar ) is unimodular, then
there exist b1 , . . . , br−1 ∈ R such that (a1 + ar b1 , a2 + ar b2 , . . . , ar−1 + ar br−1 )
is unimodular.

We need the following theorem. We are going to prove just the first part.

Theorem 77. If n defines a stable range of GL(R), then

GLm (R) GL(R)
(1) → is onto for all m > n.
Em (R) E(R)
(2) Er (R)  GLr (R) for all r.

(3) GLr (R)/Er (R) is abelian for r > 2n.

Lemma 78. Let n define a stable range for GL(R) and let r > n. If (a1 , . . . , ar )
is unimodular, then there exists A ∈ Er (R) such that
t
((a1 , . . . , ar )A) = (1, 0, . . . , 0)t .

Sketch. By adding multiples of the last row to the first r −1 rows we can assume
that a1 , . . . , ar−1 is unimodular. Then, since it is unimodular, one can add
multiples of the first r − 1 rows to get (a1 , . . . , ar−1 , 1). Now, adding multiples
of 1 we can clearly get rid of a1 , . . . , ar−1 , and finally get to (0, . . . , 0, 1).
98 Basic element theory

Remark 67. With r > n as above we have that GLr (R)Er+1 (R) = GLr+1 (R).
Proof. Clearly GLr (R)Er+1 (R) ⊆ GLr+1 (R). If A ∈ GLr+1 (R), then the last
row is unimodular, so there exists E ∈ Er+1 (R) such that
 0 
A f
AE = ,
0 1

for A0 ∈ GLr (R). But notice that

 
I ∗
∈ Er+1 (R),
0 1

and hence
(−A0 )−1 f
  0
(−A0 )−1 f
    0 
0 I A f I A 0
AE = AE = = ,
0 1 0 1 0 1 0 1

with E 0 ∈ Er+1 (R). But this means

 0 
A 0
A= (E 0 )−1 .
0 1

This also proves (1) in Theorem 77.

Theorem 79 (Bass’ Stable Range Theorem). Let R be a commutative Noethe-
rian ring with 1R . Then n = dim(j-SpecR)+1 defines a stable range for GL(R).
Proof. Let r > n = dim(j-SpecR) + 1, and let (a1 , . . . , ar ) be unimodular in Rr .
Let M = Rr−1 be the first r − 1 copies of R, so that y = (a1 , . . . , ar−1 ) ∈ M .
Then (y, a) ∈ M ⊕ R, with a := ar . Since (a, y) is unimodular, it is in particular
basic. Also, for all q ∈ j-SpecR we have Mq ' Rqr−1 , therefore

µq (M ) = r − 1 > dim (j-SpecR) ,

and hence µq (M ) > dim (j-SpecR) + 1 > dimj -SpecR q + 1, i.e. (2) of Theorem
72 holds as well. Then there exists z ∈ M which is j-SpecR basic, i.e. there is
z = (b1 , . . . , br−1 ) ∈ M such that

y + az = (a1 + ar b1 , . . . , ar−1 + ar br−1 )

is j-SpecR basic, that is (a1 + ar b1 , . . . , ar−1 + ar br−1 ) is unimodular.

Theorem 80 (Bourbaki’s Theorem). Let R be an integrally closed Noetherian
domain, and let M ∈ Modfg (R) be a torsion free module of rank r. Then there
exists an ideal I ⊆ R and an exact sequence

0 / Rr−1 /M /I / 0.

Notice that in particular this implies that cl(M ) = cl(I).

2 Basic elements and Projective modules 99

Proof. Let us induct on r = rankM .

• If r = 1 then M ' I is an ideal.
• If r > 1, then µq (M ) > rankMq > 2 for all q ∈ SpecR. Apply Corollary
74 to X = X 1 = {q ∈ SpecR | htq > 1}. Notice that condition (2)
of Theorem 72 is satisfied, therefore there exists x = x1 ∈ M which is
X 1 -basic, with R ' Rx ⊆ M (because M is torsion free).
Claim. N := M/R is torsion free.
In fact let 0 6= z ∈ R, we need to prove that z is a nonzero divisor in N .
Consider the following commutative exact diagram
0 0

 
0 /R /M /N /0
·z ·z µz =·z
  
0 /R /M /N /0
where the first two vertical maps are inceptive because R is a domain, and
M is torsion free. Also, µz is just the map induced by the diagram, and
it is again multiplication by z. By the Snake Lemma we get an inclusion

0 / ker µz / coker(R ·z / R) = R/Rz.

It is enough to show then that Ass(ker µz ) = ∅. Let p ∈ Ass(ker µz ), then

we have inclusions
R/p ,→ ker µz ,→ R/Rz,
and therefore p ∈ Ass(R/Rz). But R is an integrally closed domain, and
then z is a nonzero divisor in R. Therefore htp = 1. But localizing at p
we then have that Rp is a DVR, and Mp is torsion free over Rp . However,
torsion free modules over DVR are free, that is Mp is free. Recall that
x = x1 is X 1 -basic, and therefore it is p-basic. This means that
Mp ' xRp ⊕ Np ,
since x is basic (i.e. unimodular) in a free module. This implies in particu-
lar that Np is torsion free, and therefore (ker µz )p = 0. Since this happens
for all p ∈ Ass(ker µz ) we must have ker µz = 0, that is N is torsion free.
Now apply the inductive hypothesis to N (since rankN = r − 1 < rankM ) to
get an ideal I ⊆ R and a short exact sequence

0 / Rr−2 /N /I / 0.

Since N ' M/R we finally get

0 / Rr−1 /M /I / 0.
100 Basic element theory

Remark 68. Recall that, given X basic and M ∈ Modfg (R), a submodule M 0 ⊆
M , M 0 = R{m1 , . . . , mn }, is called X-basic if for all p ∈ X
0
δp (M ) = µp (M ) − µp (M/M 0 ) > min{n, 1 + dimX p}.

It follows from Theorem 72 that if M 0 ⊆ M is X-basic, then there exists z ∈ M 0

such that z is X-basic in M .

Theorem 81 (Bass’ Cancellation Theorem). Let d = dim X, where X = j-

SpecR. Let P ∈ Modfg (R) be a projective module such that rankPq > d + 1
for all q ∈ X (which is equivalent to the same condition for all q ∈ SpecR).
Let Q ∈ Modfg (R) be another projective module, and let M ∈ Modfg (R) be any
other module such that
Q ⊕ M ' Q ⊕ P.

Then M ' P .

Proof. Notice that M is automatically projective to start with, since we are

assuming that Q ⊕ M ' Q ⊕ P , and both Q and P are projective. Then we can
choose Q0 a finitely generated projective R-module such that Q ⊕ Q0 ' Rn , i.e.
we can assume that
Rn ⊕ M ' Rn ⊕ P.

Also, by induction it is enough to show that M ' P whenever R ⊕ M ' R ⊕ P .

Set α : R ⊕ M → R ⊕ P the isomorphism, and notice that if α((1, 0)) = (1, 0),
then clearly we have a commutative diagram

0 /R / R⊕M /M /0
idR ' ' h '
  
0 /R / R⊕P /P /0

so that M ' P . If this is not the case, let α((1, 0)) = (a, x1 ). The goal is to
show that

R⊕M
α / R⊕P β
/ R⊕P γ
/ R⊕P η
/ R⊕P

(1, 0)  / (a, x1 )  / (1, ∗)  / (1, 0)

are all isomorphism, with β, γ and η to be defined, in order to repeat the argu-
ment above. Notice that, since (1, 0) is basic in R ⊕ M and α is an isomorphism,
α((1, 0)) = (a, x1 ) is basic in R ⊕ P . Write P = R{x1 , . . . , xn }, and notice that
for all p ∈ X we have

rankPp = µp (P ) > 1 + dimX p

2 Basic elements and Projective modules 101

Pn By Theorem 72 there exists z = x1 + ax ∈ P which is X-basic.

by assumption.
Write x = i=1 ri xi . Define

f :R /P

1 /x

Now, the map β : R ⊕ P → R ⊕ P defined by the matrix

 
1R 0
f 1P

is an isomorphism since det β = 1. Also,

   t
1R 0 a
β(α((1, 0))) = β((a, x1 )) = = (a, f (a) + x1 ) = (a, z).
f 1P x1

Recall that, since z is basic, we have P ' Rz ⊕ P 0 , and hence there is a splitting
map ϕ : P → R such that ϕ(z) = 1 − a. Define γ : R ⊕ P → R ⊕ P via the
matrix  
1R ϕ
,
0 1P
and notice that it is again an isomorphism since det γ = 1. Also notice that
   t
1R ϕ a
γ(β(α((0, 1)))) = γ((a, z)) = = (a + ϕ(z), z) = (1, z).
0 1P z

Finally, let
g:R /P
 /z
1
and define η : R ⊕ P → R ⊕ P via the matrix
 
1R 0
,
−g 1P

which has determinant one, and hence it is again an isomorphism. We have

   t
1R 0 1
η(γ(β(α((1, 0))))) = η((1, z)) = = (1, 0).
−g 1P z

This proves the theorem.

Theorem 82 (Forester-Swan). Let R be a Noetherian ring, and let M ∈
Modfg (R). Set X = j-SpecR ∩ SuppM , which is basic since SuppM is closed
in SpecR and j-SpecR is basic. Then

µ(M ) 6 sup {dimX p + µp (M )}.

p∈X
102 Basic element theory

Proof. Set n := µ(M ) and t := supp∈X {dimX p + µp (M )}. We want to show

that n 6 t. Since we can pass to R/ann(M ) we can assume without loss of
generality that X = j-SpecR. By way of contradiction assume that n > t,
which in particular means that t < ∞ since M is finitely generated. Since M is
minimally generated by n elements there exists a short exact sequence

0 / M0 / Rn /M / 0,

and tensoring with k(p) for p ∈ X we get

M 0 ⊗ k(p) / k(p)n / M ⊗ k(p) / 0.

Counting dimensions:
dimk(p) (Im (M 0 ⊗ k(p) → k(p)n )) = n − µp (M ) > t − µp (M ) > dimX p.
Therefore
dimk(p) (Im (M 0 ⊗ k(p) → k(p)n )) > 1 + dimX p
for all p ∈ X, and hence M 0 is X-basic. But then, by Corollary 74 there exists
z ∈ M 0 which is basic in Rn , i.e. it is unimodular (since Rn is free). This means
Rn ' Rz ⊕ P.
But rankP = n − 1 > t, therefore we have P ' Rn−1 by Bass cancellation
Thorem 81. Since z ∈ M 0 , we still have a surjection

Rn /Rz /M / 0,

and by what we have just shown we have Rn /Rz ' P ' Rn−1 , which gives a
surjection
Rn−1 /M / 0,

contradicting the minimality of n = µ(M ).

Corollary 83. In a Dedekind domain D every ideal is minimally generated by
at most two elements.
Proof. Any ideal I ⊆ D is projective, since it is locally principal, i.e. µp (I) = 1
for all p ∈ SpecD. If p is maximal, then dimX p = 0, therefore dimX p+µp (I) =
1 in this case. If p = 0, then dimX 0 + µ(0) (I) 6 1 + 1 = 2. Therefore
µ(I) 6 sup {dimX p + µp (() I)} = 2.
p∈SpecD

Remark 69. Notice that, for example when D is semi-local, (0) ∈

/ X = j-SpecD,
so in this case if 0 6= I ⊆ D is an ideal we have
µ(I) = sup {dimX p + µp (I)} = 0 + 1 = 1,
p∈X

so ideals are principally generated, and D is a PID.

2 Basic elements and Projective modules 103

Example 39. Let R = k[x1 , . . . , xn ] and let M = m be a maximal ideal. By

the Nullstellensatz we know that j-SpecR = SpecR. If p 6= m we get

dimX p + µp (m) = dim (R/p) + µp (R)p = dim (R/p) + 1,

where dim (R/p) is the Krull dimension. If p = m we get

dimX m + µm m = 0 + n,

therefore
sup {dim (R/p) + 1, n} = dim(R/0) + 1 = n + 1.
p∈SpecR

Hence by Theorem 82 we know that µ(m) 6 n + 1. Notice that this is off by

one, since we know that µ(m) = n.
Conjecture 1 (Proved). For R = k[x1 , . . . , xn ] one can exclude the prime
p = 0, so that in the example above we get a tight upper bound.
Bibliography

[1] M. F. Atiyah and I. G. Macdonald. Introduction to commutative algebra.

Addison-Wesley Publishing Co., Reading, Mass.-London-Don Mills, Ont.,
1969.
[2] Winfried Bruns and Jürgen Herzog. Cohen-Macaulay rings, volume 39 of
Cambridge Studies in Advanced Mathematics. Cambridge University Press,
Cambridge, 1993.

[3] David Eisenbud. Commutative algebra, volume 150 of Graduate Texts in

Mathematics. Springer-Verlag, New York, 1995. With a view toward alge-
braic geometry.
[4] David Eisenbud and Graham Evans Jr. Basic elements: Theorems from al-
gebraic k-theory. Bulletin of the American Mathematical Society, 78(4):546–
549, July, 1972.
[5] Hideyuki Matsumura. Commutative ring theory, volume 8 of Cambridge
Studies in Advanced Mathematics. Cambridge University Press, Cambridge,
second edition, 1989. Translated from the Japanese by M. Reid.