COMMUTATIVE ALGEBRA

00AO
Contents
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.

Introduction
Conventions
Basic notions
Snake lemma
Finite modules and finitely presented modules
Ring maps of finite type and of finite presentation
Finite ring maps
Colimits
Localization
Internal Hom
Tensor products
Tensor algebra
Base change
Miscellany
Cayley-Hamilton
The spectrum of a ring
Local rings
The Jacobson radical of a ring
Nakayamas lemma
Open and closed subsets of spectra
Connected components of spectra
Glueing functions
More glueing results
Zerodivisors and total rings of fractions
Irreducible components of spectra
Examples of spectra of rings
A meta-observation about prime ideals
Images of ring maps of finite presentation
More on images
Noetherian rings
Locally nilpotent ideals
Curiosity
Hilbert Nullstellensatz
Jacobson rings
Finite and integral ring extensions
Normal rings
Going down for integral over normal
Flat modules and flat ring maps

This is a chapter of the Stacks Project, version d418c10, compiled on Sep 09, 2015.
1

4
4
4
7
7
9
10
11
15
19
20
25
26
28
30
31
36
37
37
38
40
40
44
46
47
49
52
54
57
59
61
62
63
64
72
77
80
82

COMMUTATIVE ALGEBRA

39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
65.
66.
67.
68.
69.
70.
71.
72.
73.
74.
75.
76.
77.
78.
79.
80.
81.
82.
83.
84.
85.
86.

Supports and annihilators

Going up and going down
Separable extensions
Geometrically reduced algebras
Separable extensions, continued
Perfect fields
Universal homeomorphisms
Geometrically irreducible algebras
Geometrically connected algebras
Geometrically integral algebras
Valuation rings
More Noetherian rings
Length
Artinian rings
Homomorphisms essentially of finite type
K-groups
Graded rings
Proj of a graded ring
Noetherian graded rings
Noetherian local rings
Dimension
Applications of dimension theory
Support and dimension of modules
Associated primes
Symbolic powers
Relative assassin
Weakly associated primes
Embedded primes
Regular sequences
Quasi-regular sequences
Blow up algebras
Ext groups
Depth
Functorialities for Ext
An application of Ext groups
Tor groups and flatness
Functorialities for Tor
Projective modules
Finite projective modules
Open loci defined by module maps
Faithfully flat descent for projectivity of modules
Characterizing flatness
Universally injective module maps
Descent for finite projective modules
Transfinite devissage of modules
Projective modules over a local ring
Mittag-Leffler systems
Inverse systems

88
90
93
95
97
98
99
104
106
108
108
112
114
117
118
119
122
123
127
129
132
135
137
138
141
142
145
148
149
151
154
156
159
161
162
163
168
168
170
173
174
175
177
183
184
186
187
189

COMMUTATIVE ALGEBRA

87.
88.
89.
90.
91.
92.
93.
94.
95.
96.
97.
98.
99.
100.
101.
102.
103.
104.
105.
106.
107.
108.
109.
110.
111.
112.
113.
114.
115.
116.
117.
118.
119.
120.
121.
122.
123.
124.
125.
126.
127.
128.
129.
130.
131.
132.
133.
134.

Mittag-Leffler modules
Interchanging direct products with tensor
Coherent rings
Examples and non-examples of Mittag-Leffler modules
Countably generated Mittag-Leffler modules
Characterizing projective modules
Ascending properties of modules
Descending properties of modules
Completion
Completion for Noetherian rings
Taking limits of modules
Criteria for flatness
Base change and flatness
Flatness criteria over Artinian rings
What makes a complex exact?
Cohen-Macaulay modules
Cohen-Macaulay rings
Catenary rings
Regular local rings
Epimorphisms of rings
Pure ideals
Rings of finite global dimension
Regular rings and global dimension
Auslander-Buchsbaum
Homomorphisms and dimension
The dimension formula
Dimension of finite type algebras over fields
Noether normalization
Dimension of finite type algebras over fields, reprise
Dimension of graded algebras over a field
Generic flatness
Around Krull-Akizuki
Factorization
Orders of vanishing
Quasi-finite maps
Zariskis Main Theorem
Applications of Zariskis Main Theorem
Dimension of fibres
Algebras and modules of finite presentation
Colimits and maps of finite presentation
More flatness criteria
Openness of the flat locus
Openness of Cohen-Macaulay loci
Differentials
Finite order differential operators
The naive cotangent complex
Local complete intersections
Syntomic morphisms

189
194
198
200
202
204
205
205
207
211
214
214
221
221
224
227
230
232
233
235
238
240
243
246
247
248
249
252
254
256
256
261
265
269
274
277
281
283
285
288
295
300
302
304
309
311
318
324

COMMUTATIVE ALGEBRA

135. Smooth ring maps

136. Formally smooth maps
137. Smoothness and differentials
138. Smooth algebras over fields
139. Smooth ring maps in the Noetherian case
140. Overview of results on smooth ring maps

141. Etale
ring maps
142. Local homomorphisms
143. Integral closure and smooth base change
144. Formally unramified maps
145. Conormal modules and universal thickenings
146. Formally etale maps
147. Unramified ring maps
148. Henselian local rings
149. Filtered colimits of etale ring maps
150. Henselization and strict henselization
151. Serres criterion for normality
152. Formal smoothness of fields
153. Constructing flat ring maps
154. The Cohen structure theorem
155. Japanese rings
156. Nagata rings
157. Ascending properties
158. Descending properties
159. Geometrically normal algebras
160. Geometrically regular algebras
161. Geometrically Cohen-Macaulay algebras
162. Colimits and maps of finite presentation, II
163. Other chapters
References

330
336
342
343
347
350
351
363
363
365
366
369
370
376
383
385
391
393
397
399
402
407
413
416
418
420
422
422
426
427

1. Introduction
00AP

Basic commutative algebra will be explained in this document. A reference is

[Mat70].
2. Conventions

00AQ

A ring is commutative with 1. The zero ring is a ring. In fact it is the only ring
that does not have a prime ideal. The Kronecker symbol ij will be used. If R S
is a ring map and q a prime of S, then we use the notation p = R q to indicate
the prime which is the inverse image of q under R S even if R is not a subring
of S and even if R S is not injective.
3. Basic notions

00AR

The following is a list of basic notions in commutative algebra. Some of these

notions are discussed in more detail in the text that follows and some are defind
in the list, but others are considered basic and will not be defined. If you are

COMMUTATIVE ALGEBRA

not familiar with most of the italicized concepts, then we suggest looking at an
introductory text on algebra before continuing.
00AS
00AT
00AU
00AV
00AW
00AX
00AY
00AZ

(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)

00B0

(9)

00B1
00B2
00B3
00B4
00B5
00B6
00B7
00B8
00B9
00BA
00BB
00BC
00BD
00BE
00BF

(10)
(11)
(12)
(13)
(14)
(15)
(16)
(17)
(18)
(19)
(20)
(21)
(22)
(23)
(24)

00BG
00BH
00BI
00BJ
0543
00BK
00BL

(25)
(26)
(27)
(28)
(29)
(30)
(31)

00BM
00BN
00BO

(32)
(33)
(34)

00BP
00BQ
00BR
00BS

(35)
(36)
(37)
(38)

R is a ring,
x R is nilpotent,
x R is a zerodivisor,
x R is a unit,
e R is an idempotent,
an idempotent e R is called trivial if e = 1 or e = 0,
: R1 R2 is a ring homomorphism,
: R1 R2 is of finite presentation, or R2 is a finitely presented R1 algebra, see Definition 6.1,
: R1 R2 is of finite type, or R2 is a finite type R1 -algebra, see Definition
6.1,
: R1 R2 is finite, or R2 is a finite R1 -algebra,
R is a (integral) domain,
R is reduced,
R is Noetherian,
R is a principal ideal domain or a PID,
R is a Euclidean domain,
R is a unique factorization domain or a UFD,
R is a discrete valuation ring or a dvr,
K is a field,
K L is a field extension,
K L is an algebraic field extension,
{ti }iI is a transcendence basis for L over K,
the transcendence degree trdeg(L/K) of L over K,
the field k is algebraically closed,
if K L is algebraic, and K k an extension with k algebraically closed,
then there exists a map ring map L k extending the map on K,
I R is an ideal,
I R is radical,

if I is an ideal then we have its radical I,

I R is nilpotent means that I n = 0 for some n N,
I R is locally nilpotent means that every element of I is nilpotent,
p R is a prime ideal,
if p R is prime and if I, J R are ideal, and if IJ p, then I p or
J p.
m R is a maximal ideal,
any nonzero ring has a maximal ideal,
T
the Jacobson radical of R is rad(R) = mR m the intersection of all the
maximal ideals of R,
the ideal (T ) generated by a subset T R,
the quotient ring R/I,
an ideal I in the ring R is prime if and only if R/I is a domain,
an ideal I in the ring R is maximal if and only if the ring R/I is a field,

COMMUTATIVE ALGEBRA

00BT
00BU

00BV
00BW
055Y
00BX
00BY
00BZ
00C0
00C1
00C2
0516
00C3
00C4
00C5
00C6
00C7

00C8

00C9

(39) if : R1 R2 is a ring homomorphism, and if I R2 is an ideal, then

1 (I) is an ideal of R1 ,
(40) if : R1 R2 is a ring homomorphism, and if I R1 is an ideal, then
(I) R2 (sometimes denoted I R2 , or IR2 ) is the ideal of R2 generated
by (I),
(41) if : R1 R2 is a ring homomorphism, and if p R2 is a prime ideal,
then 1 (p) is a prime ideal of R1 ,
(42) M is an R-module,
(43) for m M the annihilator I = {f R | f m = 0} of m in R,
(44) N M is an R-submodule,
(45) M is an Noetherian R-module,
(46) M is a finite R-module,
(47) M is a finitely generated R-module,
(48) M is a finitely presented R-module,
(49) M is a free R-module,
(50) if 0 K L M 0 is a short exact sequence of R-modules and K,
M are free, then L is free,
(51) if N M L are R-modules, then L/M = (L/N )/(M/N ),
(52) S is a multiplicative subset of R,
(53) the localization R S 1 R of R,
(54) if R is a ring and S is a multiplicative subset of R then S 1 R is the zero
ring if and only if S contains 0,
(55) if R is a ring and if the multiplicative subset S consists completely of
nonzerodivisors, then R S 1 R is injective,
(56) if : R1 R2 is a ring homomorphism, and S is a multiplicative subsets
of R1 , then (S) is a multiplicative subset of R2 ,
(57) if S, S 0 are multiplicative subsets of R, and if SS 0 denotes the set of products
SS 0 = {r R | s S, s0 S 0 , r = ss0 } then SS 0 is a multiplicative subset
of R,
(58) if S, S 0 are multiplicative subsets of R, and if S denotes the image of S in
1

00CA
00CB

(59)
(60)

00CC

(61)

00CD

(62)

(S 0 )1 R, then (SS 0 )1 R = S ((S 0 )1 R),

the localization S 1 M of the R-module M ,
the functor M 7 S 1 M preserves injective maps, surjective maps, and
exactness,
if S, S 0 are multiplicative subsets of R, and if M is an R-module, then
(SS 0 )1 M = S 1 ((S 0 )1 M ),
if R is a ring, I and ideal of R and S a multiplicative subset of R, then
1

00CE

00CF

00CG

S 1 I is an ideal of S 1 R, and we have S 1 R/S 1 I = S (R/I), where S

is the image of S in R/I,
(63) if R is a ring, and S a multiplicative subset of R, then any ideal I 0 of S 1 R
is of the form S 1 I, where one can take I to be the inverse image of I 0 in
R,
(64) if R is a ring, M an R-module, and S a multiplicative subset of R, then any
submodule N 0 of S 1 M is of the form S 1 N for some submodule N M ,
where one can take N to be the inverse image of N 0 in M ,
(65) if S = {1, f, f 2 , . . .} then Rf = S 1 R and Mf = S 1 M ,

COMMUTATIVE ALGEBRA

00CH
00CI
03C0
00CJ
00CK
00CL

(66) if S = R \ p = {x R | x 6 p} for some prime ideal p, then it is customary

to denote Rp = S 1 R and Mp = S 1 M ,
(67) a local ring is a ring with exactly one maximal ideal,
(68) a semi-local ring is a ring with finitely many maximal ideals,
(69) if p is a prime in R, then Rp is a local ring with maximal ideal pRp ,
(70) the residue field, denoted (p), of the prime p in the ring R is the quotient
Rp /pRp = (R \ p)1 R/p,
(71) given R and M1 , M2 the tensor product M1 R M2 ,
(72) etc.
4. Snake lemma

07JV

The snake lemma and its variants are discussed in the setting of abelian categories
in Homology, Section 5.

07JW

Lemma 4.1. Suppose given a commutative diagram

X


/U

/Y

/V

/Z

/0


/W

of abelian groups with exact rows, then there is a canonical exact sequence
Ker() Ker() Ker() Coker() Coker() Coker()
Moreover, if X Y is injective, then the first map is injective, and if V W is
surjective, then the last map is surjective.
Proof. The map : Ker() Coker() is defined as follows. Take z Ker().
Choose y Y mapping to z. Then (y) V maps to zero in W . Hence (y) is
the image of some u U . Set z = u the class of u in the cokernel of . Proof of
exactness is omitted.

5. Finite modules and finitely presented modules
0517

Just some basic notation and lemmas.

0518

Definition 5.1. Let R be a ring. Let M be an R-module.

(1) We say M is a finite R-module, or a finitely generated R-module if there
exist n N and x1 , . . . , xn M such that every element of M is a R-linear
combination of the xi . Equivalently, this means there exists a surjection
Rn M for some n N.
(2) We say M is a finitely presented R-module or an R-module of finite presentation if there exist integers n, m N and an exact sequence
Rm Rn M 0
Informally, M is a finitely presented R-module if and only if it is finitely generated
and the module of relations among these generators is finitely generated as well. A
choice of an exact sequence as in the definition is called a presentation of M .

07JX

Lemma 5.2. Let R be a ring. Let : Rn M and : N M be module

maps. If Im() Im(), then there exists an R-module map : Rn N such
that = .

[CE56, III, Lemma

3.3]

COMMUTATIVE ALGEBRA

Proof. Let ei = (0, . . . , 0, 1, 0, . . . , 0) be the ith basis vector of Rn . Let xi N

be
P an element with (ei ) = (xi ) which exists by assumption. Set (a1 , . . . , an ) =
ai xi . By construction = .

0519

Lemma 5.3. Let R be a ring. Let

0 M1 M2 M3 0
be a short exact sequence of R-modules.
(1) If M1 and M3 are finite R-modules, then M2 is a finite R-module.
(2) If M1 and M3 are finitely presented R-modules, then M2 is a finitely presented R-module.
(3) If M2 is a finite R-module, then M3 is a finite R-module.
(4) If M2 is a finitely presented R-module and M1 is a finite R-module, then
M3 is a finitely presented R-module.
(5) If M3 is a finitely presented R-module and M2 is a finite R-module, then
M1 is a finite R-module.
Proof. Proof of (1). If x1 , . . . , xn are generators of M1 and y1 , . . . , ym M2 are
elements whose images in M3 are generators of M3 , then x1 , . . . , xn , y1 , . . . , ym
generate M2 .
Part (3) is immediate from the definition.
Proof of (5). Assume M3 is finitely presented and M2 finite. Choose a presentation
Rm Rn M3 0
By Lemma 5.2 there exists a map Rn M2 such that the solid diagram

Rm

/ Rn

/ M3


/ M1


/ M2


/ M3

/0

id

/0

commutes. This produces the dotted arrow. By the snake lemma (Lemma 4.1) we
see that we get an isomorphism
Coker(Rm M1 )
= Coker(Rn M2 )
In particular we conclude that Coker(Rm M1 ) is a finite R-module. Since
Im(Rm M1 ) is finite by (3), we see that M1 is finite by part (1).
Proof of (4). Assume M2 is finitely presented and M1 is finite. Choose a presentation Rm Rn M2 0. Choose a surjection Rk M1 . By
Lemma 5.2 there exists a factorization Rk Rn M2 of the composition
Rk M1 M2 . Then Rk+m Rn M3 0 is a presentation.
Proof of (2). Assume that M1 and M3 are finitely presented. The argument in the
proof of part (1) produces a commutative diagram
0

/ Rn

/ Rn+m

/ Rm

/0


/ M1


/ M2


/ M3

/0

COMMUTATIVE ALGEBRA

with surjective vertical arrows. By the snake lemma we obtain a short exact sequence
0 Ker(Rn M1 ) Ker(Rn+m M2 ) Ker(Rm M3 ) 0
By part (5) we see that the outer two modules are finite. Hence the middle one is
finite too. By (4) we see that M2 is of finite presentation.

00KZ

Lemma 5.4. Let R be a ring, and let M be a finite R-module. There exists a
filtration by R-submodules
0 = M0 M1 . . . Mn = M
such that each quotient Mi /Mi1 is isomorphic to R/Ii for some ideal Ii of R.
Proof. By induction on the number of generators of M . Let x1 , . . . , xr M be
a minimal number of generators. Let M 0 = Rx1 M . Then M/M 0 has r 1
generators and the induction hypothesis applies. And clearly M 0
= R/I1 with
I1 = {f R | f x1 = 0}.


0560

Lemma 5.5. Let R S be a ring map. Let M be an S-module. If M is finite as

an R-module, then M is finite as an S-module.
Proof. In fact, any R-generating set of M is also an S-generating set of M , since
the R-module structure is induced by the image of R in S.

6. Ring maps of finite type and of finite presentation

00F2
00F3

Definition 6.1. Let R S be a ring map.

(1) We say R S is of finite type, or that S is a finite type R-algebra if there
exists an n N and an surjection of R-algebras R[x1 , . . . , xn ] S.
(2) We say R S is of finite presentation if there exist integers n, m N and
polynomials f1 , . . . , fm R[x1 , . . . , xn ] and an isomorphism of R-algebras
R[x1 , . . . , xn ]/(f1 , . . . , fm )
= S.
Informally, R S is of finite presentation if and only if S is finitely generated as
an R-algebra and the ideal of relations among the generators is finitely generated.
A choice of a surjection R[x1 , . . . , xn ] S as in the definition is sometimes called
a presentation of S.

00F4

Lemma 6.2. The notions finite type and finite presentation have the following
permanence properties.
(1) A composition of ring maps of finite type is of finite type.
(2) A composition of ring maps of finite presentation is of finite presentation.
(3) Given R S 0 S with R S of finite type, then S 0 S is of finite type.
(4) Given R S 0 S, with R S of finite presentation, and R S 0 of
finite type, then S 0 S is of finite presentation.
Proof. We only prove the last assertion. Write S = R[x1 , . . . , xn ]/(f1 , . . . , fm ) and
S 0 = R[y1 , . . . , ya ]/I. Say that the class yi of yi maps to hi mod (f1 , . . . , fm ) in S.
Then it is clear that S = S 0 [x1 , . . . , xn ]/(f1 , . . . , fm , h1 y1 , . . . , ha ya ).


00R2

Lemma 6.3. Let R S be a ring map of finite presentation. For any surjection
: R[x1 , . . . , xn ] S the kernel of is a finitely generated ideal in R[x1 , . . . , xn ].

COMMUTATIVE ALGEBRA

10

Proof. Write S = R[y1 , . . . , ym ]/(f1 , . . . , fk ). Choose gi R[y1 , . . . , ym ] which

are lifts of (xi ). Then we see that S = R[xi , yj ]/(fj , xi gi ). Choose hj
R[x1 , . . . , xn ] such that (hj ) corresponds to yj mod (f1 , . . . , fk ). Consider the
map : R[xi , yj ] R[xi ], xi 7 xi , yj 7 hj . Then the kernel of is the image of
(fj , xi gi ) under and we win.

0561

Lemma 6.4. Let R S be a ring map. Let M be an S-module. Assume R S

is of finite type and M is finitely presented as an R-module. Then M is finitely
presented as an S-module.
Proof. This is similar to the proof of part (4) of Lemma 6.2. We may assume S =
R[x1 , . . . , xn ]/J. P
Choose y1 , . . . , ym M which generate M as an R-module and
choose relations
aij yj = 0, i = 1, . . . , t which generate the kernel of Rm M .
For any i = 1, . . . , n and j = 1, . . . , m write
X
xi yj =
aijk yk
for some aijk P
R. Consider the S-module N generated
P by y1 , . . . , ym subject to
the relations
aij yj = 0, i = 1, . . . , t and xi yj =
aijk yk , i = 1, . . . , n and
j = 1, . . . , m. Then N has a presentation
S nm+t S m N 0
By construction there is a surjective
map : N M . To finish the proof we show
P
is injective. Suppose z =
bj yj N for some bj S. We may think of bj
as a polynomial P
in x1 , . . . , xn with coefficients in R. By applying the relations of
the form xi yj =
aijkP
yk we can inductively lower the degree of the polynomials.
Hence we see that z =
cj yj for some cj R. Hence if (z) = 0 then the vector
(c1 , . . . , cm ) is an R-linear combination of the vectors (ai1 , . . . , aim ) and we conclude
that z = 0 as desired.

7. Finite ring maps

0562
0563

Definition 7.1. Let : R S be a ring map. We say : R S is finite if S is

finite as an R-module.

00GJ

Lemma 7.2. Let R S be a finite ring map. Let M be an S-module. Then M

is finite as an R-module if and only if M is finite as an S-module.
Proof. One of the implications follows from Lemma 5.5. To see the other assume
that M is finite as an S-module. Pick x1 , . . . , xn S which generate S as an
R-module. Pick y1 , . . . , ym M which generate M as an S-module. Then xi yj
generate M as an R-module.


00GL

Lemma 7.3. Suppose that R S and S T are finite ring maps. Then R T
is finite.
Proof. If ti generate T as an S-module and sj generate S as an R-module, then
ti sj generate T as an R-module. (Also follows from Lemma 7.2.)

For more information on finite ring maps, please see Section 35.

COMMUTATIVE ALGEBRA

11

8. Colimits
07N7

Some of the material in this section overlaps with the general discussion on colimits
in Categories, Sections 14 21.

00D3

Definition 8.1. A partially ordered set is a set I together with a relation which
is transitive (if i j and j k then i k) and reflexive (i i for all i I). A
directed set (I, ) is a partially ordered set (I, ) such that I is not empty and such
that i, j I, there exists k I with i k, j k.
It is customary to drop the from the notation when talking about a partially
ordered set (that is, one speaks of the partially ordered set I rather than of the
partially ordered set (I, )). This is the same as the notion defined in Categories,
Section 21.
The notion partially ordered set is commonly abbreviated as poset.

00D4

Definition 8.2. Let (I, ) be a partially ordered set. A system (Mi , ij ) of Rmodules over I consists of a family of R-modules {Mi }iI indexed by I and a family
of R-module maps {ij : Mi Mj }ij such that for all i j k
ii = idMi

ik = jk ij

We say (Mi , ij ) is a directed system if I is a directed set.

This is the same as the notion defined in Categories, Definition 21.1 and Section
21. We refer to Categories, Definition 14.2 for the definition of a colimit of a
diagram/system in any category.
00D5

Lemma 8.3. Let (Mi , ij ) be a system of R-modules over the partially

L ordered
set I. The colimit of the system (Mi , ij ) is the quotient R-module ( iI Mi )/Q
where Q is the R-submodule generated by all elements
i (xi ) j (ij (xi ))
L

where i : Mi
i is the natural inclusion. We denote the colimit M =
iI ML
colimi Mi . We denote : iI Mi M the projection map and i = i : Mi
M.
Proof. This lemma is a special case of Categories, Lemma 14.11 but we will also
prove it directly in this case. Namely, note that i = j ij in the above construction. To show the pair (M, i ) is the colimit we have to show it satisfies the
universal property: for any other such pair (Y, i ) with i : Mi Y , i = j ij ,
there is a unique R-module homomorphism g : M Y such that the following
diagram commutes:
ij
/ Mj
Mi
j

M
g

}
j

  
Y
And this is clear because
we
can
define
g by taking the map i on the summand
L
Mi in the direct sum
Mi .


COMMUTATIVE ALGEBRA

00D6

12

Lemma 8.4. Let (Mi , ij ) be a system of R-modules over the partially ordered
set I. Assume that I is directed. The colimit of the system (Mi , ij ) is canonically
isomorphic to the module M defined as follows:
(1) as a set let
M=

a
iI


Mi /

where for m Mi and m0 Mi0 we have

m m0 ij (m) = i0 j (m0 ) for some j i, i0
(2) as an abelian group for m Mi and m0 Mi0 we define the sum of the
classes of m and m0 in M to be the class of ij (m) + i0 j (m0 ) where j I
is any index with i j and i0 j, and
(3) as an R-module define for m Mi and x R the product of x and the
class of m in M to be the class of xm in M .
The canonical maps i : Mi M are induced by the canonical maps Mi
`
iI Mi .
Proof. Omitted. Compare with Categories, Section 19.
00D7

Lemma 8.5. Let (Mi , ij ) be a directed system. Let M = colim Mi with i :

Mi M . Then, i (xi ) = 0 for xi Mi if and only if there exists j i such that
ij (xi ) = 0.
Proof. This is clear from the description of the directed colimit in Lemma 8.4.

00D8

Example 8.6. Consider the partially ordered set I = {a, b, c} with a < b and a < c
and no other strict inequalities. A system (Ma , Mb , Mc , ab , ac ) over I consists of
three R-modules Ma , Mb , Mc and two R-module homomorphisms ab : Ma Mb
and ac : Ma Mc . The colimit of the system is just
M := colimiI Mi = Coker(Ma Mb Mc )
where the map is ab ac . Thus the kernel of the canonical map Ma M is
Ker(ab ) + Ker(ac ). And the kernel of the canonical map Mb M is the image
of Ker(ac ) under the map ab . Hence clearly the result of Lemma 8.5 is false for
general systems.

00D9

Definition 8.7. Let (Mi , ij ), (Ni , ij ) be systems of R-modules over the same
partially ordered set I. A homomorphism of systems from (Mi , ij ) to (Ni , ij )
is by definition a family of R-module homomorphisms i : Mi Ni such that
j ij = ij i for all i j.
This is the same notion as a transformation of functors between the associated
diagrams M : I ModR and N : I ModR , in the language of categories. The
following lemma is a special case of Categories, Lemma 14.7.

00DA

Lemma 8.8. Let (Mi , ij ), (Ni , ij ) be systems of R-modules over the same partially ordered set. A morphism of systems = (i ) from (Mi , ij ) to (Ni , ij )
induces a unique homomorphism
colim i : colim Mi colim Ni

COMMUTATIVE ALGEBRA

such that

13

/ colim Mi

Mi
i

colim i


/ colim Ni


Ni

commutes for all i I.

Proof. Write M = colim Mi and N = colim Ni and = colim i (as yet to be
constructed). We will use the explicit description of M and N in Lemma 8.3
without further mention. The condition of the lemma is equivalent to the condition
that
L
/M
iI Mi
L

L 

iI


/N

Ni

commutes. Hence it is clear that if exists, then it is unique. To see that exists,
L
it suffices to show that the kernel of the upper horizontal arrow is mapped by
i
to the kernel of the lower horizontal arrow. To see this, let j k and xj Mj .
Then
M
(
i )(xj jk (xj )) = j (xj ) k (jk (xj )) = j (xj ) jk (j (xj ))
which is in the kernel of the lower horizontal arrow as required.
00DB

Lemma 8.9. Let I be a directed partially ordered set. Let (Li , ij ), (Mi , ij ), and
(Ni , ij ) be systems of R-modules over I. Let i : Li Mi and i : Mi Ni be
morphisms of systems over I. Assume that for all i I the sequence of R-modules
Li

/ Mi

/ Ni

is a complex with homology Hi . Then the R-modules Hi form a system over I, the
sequence of R-modules
colimi Li

/ colimi Mi

/ colimi Ni

is a complex as well, and denoting H its homology we have

H = colimi Hi .

/ colimi Mi / colimi Ni is a complex.

Proof. It is clear that colimi Li
For each i I, there is a canonical R-module morphism Hi H (sending each
[m] Hi = Ker(i )/Im(i ) to the residue class in H = Ker()/Im() of the image
of m in colimi Mi ). These give rise to a morphism colimi Hi H. It remains to
show that this morphism is surjective and injective.

We are going to repeatedly use the description of colimits over I as in Lemma 8.4
without further mention. Let h H. Since H = Ker()/Im() we see that h is
the class mod Im() of an element [m] in Ker() colimi Mi . Choose an i such
that [m] comes from an element m Mi . Choose a j i such that ij (i (m)) = 0
which is possible since [m] Ker(). After replacing i by j and m by ij (m) we
see that we may assume m Ker(i ). This shows that the map colimi Hi H is
surjective.

COMMUTATIVE ALGEBRA

14

Suppose that hi Hi has image zero on H. Since Hi = Ker(i )/Im(i ) we may

represent hi by an element m Ker(i ) Mi . The assumption on the vanishing of
hi in H means that the class of m in colimi Mi lies in the image of . Hence there
exists a j i and an l Lj such that j (l) = ij (m). Clearly this shows that the
image of hi in Hj is zero. This proves the injectivity of colimi Hi H.

00DC

Example 8.10. Taking colimits is not exact in general. Consider the partially
ordered set I = {a, b, c} with a < b and a < c and no other strict inequalities, as in
Example 8.6. Consider the map of systems (0, Z, Z, 0, 0) (Z, Z, Z, 1, 1). From the
description of the colimit in Example 8.6 we see that the associated map of colimits
is not injective, even though the map of systems is injective on each object. Hence
the result of Lemma 8.9 is false for general systems.

04B0

Lemma 8.11. Let I be an index category satisfying the assumptions of Categories,

Lemma 19.7. Then taking colimits of diagrams of abelian groups over I is exact
(i.e., the analogue of Lemma 8.9 holds in this situation).
`
Proof. By Categories, Lemma 19.7 we may write I = jJ Ij with each Ij a
filtered category, and J possibly empty. By Categories, Lemma 21.3 taking colimits
over the index categories Ij is the same as taking the colimit over some directed
partially ordered set. Hence Lemma 8.9 applies to these colimits. This reduces the
problem to showing that coproducts in the category of R-modules over the set J
are exact. In other words, exact sequences Lj Mj Nj of R modules we have
to show that
M
M
M
Lj
Mj
Nj
jJ

jJ

jJ

is exact. This can be verified by hand, and holds even if J is empty.

For purposes of reference, we define what it means to have a relation between

elements of a module.
07N8

Definition 8.12. Let R be a ring. Let M be an R-module. Let n 0 and xi M

for i = 1, . . . , n. A relation
P between x1 , . . . , xn in M is a sequence of elements
f1 , . . . , fn R such that i=1,...,n fi xi = 0.

00HA

Lemma 8.13. Let R be a ring and let M be an R-module. Then M is the colimit of
a directed system (Mi , ij ) of R-modules with all Mi finitely presented R-modules.
Proof. Consider any finite subset S M and any finite collection of relations E
among the elements of S. So each s S corresponds
P to xs M and each e E
consists of a vector of elements fe,s R such that
fe,s xs = 0. Let MS,E be the
cokernel of the map
X
R#E R#S , (ge )eE 7 (
ge fe,s )sS .
There are canonical maps MS,E M . If S S 0 and if the elements of E correspond, via this map, to relations in E 0 , then there is an obvious map MS,E MS 0 ,E 0
commuting with the maps to M . Let I be the set of pairs (S, E) with ordering by
inclusion as above. It is clear that the colimit of this directed system is M .


COMMUTATIVE ALGEBRA

15

9. Localization
00CM
00CN

Definition 9.1. Let R be a ring, S a subset of R. We say S is a multiplicative

subset of R is 1 S and S is closed under multiplication, i.e., s, s0 S ss0 S.
Given a ring A and a multiplicative subset S, we define a relation on A S as
follows:
(x, s) (y, t) u S such that (xt ys)u = 0
It is easily checked that this is an equivalence relation. Let x/s (or xs ) be the
equivalence class of (x, s) and S 1 A be the set of all equivalence classes. Define
addition and multiplication in S 1 A as follows:
x/s + y/t = (xt + ys)/st,

x/s y/t = xy/st

One can check that S 1 A becomes a ring under these operations.

00CO

Definition 9.2. This ring is called the localization of A with respect to S.

We have a natural ring map from A to its localization S 1 A,
A S 1 A,

x 7 x/1

which is sometimes called the localization map. In general the localization map is
not injective, unless S contains no zerodivisors. For, if x/1 = 0, then there is a
u S such that xu = 0 in A and hence x = 0 since there are no zerodivisors in S.
The localization of a ring has the following universal property.
00CP

Proposition 9.3. Let f : A B be a ring map that sends every element in S to

a unit of B. Then there is a unique homomorphism g : S 1 A B such that the
following diagram commutes.
A

/B
<

"
S 1 A

Proof. Existence. We define a map g as follows. For x/s S 1 A, let g(x/s) =

f (x)f (s)1 B. It is easily checked from the definition that this is a well-defined
ring map. And it is also clear that this makes the diagram commutative.
Uniqueness. We now show that if g 0 : S 1 A B satisfies g 0 (x/1) = f (x), then
g = g 0 . Hence f (s) = g 0 (s/1) for s S by the commutativity of the diagram.
But then g 0 (1/s)f (s) = 1 in B, which implies that g 0 (1/s) = f (s)1 and hence
g 0 (x/s) = g 0 (x/1)g 0 (1/s) = f (x)f (s)1 = g(x/s).

00CQ

Lemma 9.4. The localization S 1 A is the zero ring if and only if 0 S.

Proof. If 0 S, any pair (a, s) (0, 1) by definition. If 0 6 S, then clearly
1/1 6= 0/1 in S 1 A.


07JY

Lemma 9.5. Let R be a ring. Let S R be a multiplicative subset. The category

of S 1 R-modules is equivalent to the category of R-modules N with the property
that every s S acts as an automorphism on N .

COMMUTATIVE ALGEBRA

16

Proof. The functor which defines the equivalence associates to an S 1 R-module

M the same module but now viewed as an R-module via the localization map
R S 1 R. Conversely, if N is an R-module, such that every s S acts via an
automorphism sN , then we can think of N as an S 1 R-module by letting x/s act
via xN s1
N . We omit the verification that these two functors are quasi-inverse to
each other.

The notion of localization of a ring can be generalized to the localization of a
module. Let A be a ring, S a multiplicative subset of A and M an A-module. We
define a relation on M S as follows
(m, s) (n, t) u S such that (mt ns)u = 0
This is clearly an equivalence relation. Denote by m/s (or m
s ) be the equivalence
class of (m, s) and S 1 M be the set of all equivalence classes. Define the addition
and scalar multiplication as follows
m/s + n/t = (mt + ns)/st,
It is clear that this makes S
07JZ

M an S

m/s n/t = mn/st

A module.

Definition 9.6. The S 1 A-module S 1 M is called the localization of M at S.

Note that there is an A-module map M S 1 M , m 7 m/1 which is sometimes
called the localization map. It satisfies the following universal property.

07K0

Lemma 9.7. Let R be a ring. Let S R a multiplicative subset. Let M , N be

R-modules. Assume all the elements of S act as automorphisms on N . Then the
canonical map
HomR (S 1 M, N ) HomR (M, N )
induced by the localization map, is an isomorphism.
Proof. It is clear that the map is well-defined and R-linear. Injectivity: Let
HomR (S 1 M, N ) and take an arbitrary element m/s S 1 M . Then, since s
(m/s) = (m/1), we have (m/s) = s1 ((m/1)), so is completely determined
by what it does on the image of M in S 1 M . Surjectivity: Let : M N be a
given R-linear map. We need to show that it can be extended to S 1 M . Define
a map of sets
M S N, (m, s) 7 s1 (m)
Clearly, this map respects the equivalence relation from above, so it descends to a
well-defined map : S 1 M N . It remains to show that this map is R-linear, so
take r, r0 R as well as s, s0 S and m, m0 M . Then
(r m/s + r0 m0 /s0 ) = ((r s0 m + r0 s m0 )/(ss0 ))
= (ss0 )1 ((r s0 m + r0 s m0 )
= (ss0 )1 (r s0 (m) + r0 s(m0 )
= r(m/s) + r0 (m0 /s0 )
and we win.

02C5

Example 9.8. Let A be a ring and let M be an A-module. Here are some
important examples of localizations.

COMMUTATIVE ALGEBRA

17

(1) Given p a prime ideal of A consider S = A\p. It is immediately checked that

S is a multiplicative set. In this case we denote Ap and Mp the localization
of A and M with respect to S respectively. These are called the localization
of A, resp. M at p.
(2) Let f A. Consider S = {1, f, f 2 , . . .}. This is clearly a multiplicative
subset of A. In this case we denote Af (resp. Mf ) the localization S 1 A
(resp. S 1 M ). This is called the localization of A, resp. M with respect to
f . Note that Af = 0 if and only if f is nilpotent in A.
(3) Let S = {f A | f is not a zerodivisor in A}. This is a multiplicative
subset of A. In this case the ring Q(A) = S 1 A is called either the total
quotient ring, or the total ring of fractions of A.
00CR

Lemma 9.9. Let R be a ring. Let S R be a multiplicative subset. Let M be an

R-module. Then
S 1 M = colimf S Mf
where the partial ordering on S is given by f f 0 f = f 0 f 00 for some f 00 R in
which case the map Mf 0 Mf is given by m/(f 0 )e 7 m(f 00 )e /f e .
Proof. Omitted. Hint: Use the universal property of Lemma 9.7.

In the following paragraph, let A denote a ring, and M, N denote modules over A.
If S and S 0 are multiplicative sets of A, then it is clear that
SS 0 = {ss0 : s S, s0 S 0 }
is also a multiplicative set of A. Then the following holds.
02C6

Proposition 9.10. Let S be the image of S in S 01 A, then (SS 0 )1 A is isomorphic

to S

(S 01 A).

Proof. The map sending x A to x/1 (SS 01 )A induces a map sending x/s
S 01 A to x/s (SS 01 )A, by universal property. The image of the elements in S are
1
invertible in (SS 01 )A. By the universal property we get a map f : S (S 01 A)
01
0
0
0
0 1
(SS )A which maps (x/t )/(s/s ) to (x/t ) (s/s ) .
1

On the other hand, the map from A to S (S 01 A) sending x A to (x/1)/(1/1)

1
also induces a map g : (SS 01 )A S (S 01 A) which sends x/ss0 to (x/s0 )/(s/1),
by the universal property again. It is immediately checked that f and g are inverse
to each other, hence they are both isomorphisms.

For the module M we have
02C7

Proposition 9.11. View S 01 M as an A-module, then S 1 (S 01 M ) is isomorphic

to (SS 0 )1 M .
Proof. Note that given a A-module M, we have not proved any universal property
for S 1 M . Hence we cannot reason as in the preceding proof; we have to construct
the isomorphism explicitly.
We define the maps as follows
f : S 1 (S 01 M ) (SS 0 )1 M,
g : (SS 0 )1 M S 1 (S 01 M ),

x/s0
7 x/ss0
s
x/s0
x/t 7
for some s S, s0 S 0 , and t = ss0
s

COMMUTATIVE ALGEBRA

18

We have to check that these homomorphisms are well-defined, that is, independent
the choice of the fraction. This is easily checked and it is also straightforward to
show that they are inverse to each other.

If u : M N is an A homomorphism, then the localization indeed induces a
well-defined S 1 A homomorphism S 1 u : S 1 M S 1 N which sends x/s to
u(x)/s. It is immediately checked that this construction is functorial, so that S 1
is actually a functor from the category of A-modules to the category of S 1 Amodules. Moreover this functor is exact, as we show in the following proposition.
00CS

Proposition 9.12. Let L

M
N is an exact sequence of R-modules. Then
S 1 L S 1 M S 1 N is also exact.
Proof. First it is clear that S 1 L S 1 M S 1 N is a complex since localization is a functor. Next suppose that x/s maps to zero in S 1 N for some
x/s S 1 M . Then by definition there is a t S such that v(xt) = v(x)t = 0 in
M , which means xt Ker(v). By the exactness of L M N we have xt = u(y)
for some y in L. Then x/s is the image of y/st. This proves the exactness.


02C8

Lemma 9.13. Localization respects quotients, i.e. if N is a submodule of M , then

S 1 (M/N ) ' (S 1 M )/(S 1 N ).
Proof. From the exact sequence
0 N M M/N 0
we have
0 S 1 N S 1 M S 1 (M/N ) 0
The corollary then follows.

If, in the preceding Corollary, we take N = I and M = A for an ideal I of A, we

see that S 1 A/S 1 I ' S 1 (A/I) as A-modules. The next proposition shows that
they are isomorphic as rings.
00CT

Proposition 9.14. Let I be an ideal of A, S a multiplicative set of A. Then S 1 I

is an ideal of S 1 A and S
image of S in A/I.

(A/I) is isomorphic to S 1 A/S 1 I, where S is the

Proof. The fact that S 1 I is an ideal is clear since I itself is an ideal. Define
f : S 1 A S

(A/I),

x/s 7 x/s

where x and s are the images of x and s in A/I. We shall keep similar notations in
this proof. This map is well-defined by the universal property of S 1 A, and S 1 I
is contained in the kernel of it, therefore it induces a map
f : S 1 A/S 1 I S

(A/I),

x/s 7 x/s

On the other hand, the map A S 1 A/S 1 I sending x to x/1 induces a map
A/I S 1 A/S 1 I sending x to x/1. The image of S is invertible in S 1 A/S 1 I,
thus induces a map
x
1
7 x/s
g : S (A/I) S 1 A/S 1 I,
s
by the universal property. It is then clear that f and g are inverse to each other,
hence are both isomorphisms.


COMMUTATIVE ALGEBRA

19

We now consider how submodules behave in localization.

00CU

Lemma 9.15. Any submodule N 0 of S 1 M is of the form S 1 N for some N M .

Indeed one can take N to be the inverse image of N 0 in M .
Proof. Let N be the inverse image of N 0 in M . Then one can see that S 1 N N 0 .
To show they are equal, take x/s in S 1 N , where s S and x N . This yields
that x/1 N 0 . Since N 0 is an S 1 R-submodule we have x/s = x/1 1/s N 0 .
This finishes the proof.

Taking M = A and N = I an ideal of A, we have the following corollary, which can
be viewed as a converse of the first part of Proposition 9.14.

02C9

Lemma 9.16. Each ideal I 0 of S 1 A takes the form S 1 I, where one can take I
to be the inverse image of I 0 in A.
Proof. Immediate from Lemma 9.15.

10. Internal Hom

0581

If R is a ring, and M , N are R-modules, then

HomR (M, N ) = { : M N }
is the set of R-linear maps from M to N . This set comes with the structure of an
abelian group by setting (+)(m) = (m)+(m), as usual. In fact, HomR (M, N )
is also an R-module via the rule (x)(m) = x(m) = (xm).
Given maps a : M M 0 and b : N N 0 of R-modules, we can pre-compose and
post-compose homomorphisms by a and b. This leads to the following commutative
diagram
/ HomR (M 0 , N 0 )
HomR (M 0 , N )
b


HomR (M, N )


/ HomR (M, N 0 )

In fact, the maps in this diagram are R-module maps. Thus HomR defines an
additive functor
Modopp
R ModR ModR ,
0582

Lemma 10.1. Exactness and HomR . Let R be a ring.

(1) Let M1 M2 M3 0 be a complex of R-modules. Then M1 M2
M3 0 is exact if and only if 0 HomR (M3 , N ) HomR (M2 , N )
HomR (M1 , N ) is exact for all R-modules N .
(2) Let 0 M1 M2 M3 be a complex of R-modules. Then 0 M1
M2 M3 is exact if and only if 0 HomR (N, M1 ) HomR (N, M2 )
HomR (N, M3 ) is exact for all R-modules N .
Proof. Omitted.

0583

(M, N ) 7 HomR (M, N )

Lemma 10.2. Let R be a ring. Let M be a finitely presented R-module. Let N

be an R-module.
(1) For f R we have HomR (M, N )f = HomRf (Mf , Nf ) = HomR (Mf , Nf ),

COMMUTATIVE ALGEBRA

20

(2) for a multiplicative subset S of R we have

S 1 HomR (M, N ) = HomS 1 R (S 1 M, S 1 N ) = HomR (S 1 M, S 1 N ).
Proof. Part (1) is a special case of part (2). The second equality in (2) follows
from Lemma 9.7. Choose a presentation
M
M
R
R M 0.
j=1,...,m

i=1,...,n

By Lemma 10.1 this gives an exact sequence

M
0 HomR (M, N )

i=1,...,n

M
j=1,...,m

Inverting S and using Proposition 9.12 we get an exact sequence

M
M
0 S 1 HomR (M, N )
S 1 N
i=1,...,n

N.

j=1,...,m

S 1 N

and the result follows since S 1 M sits in an exact sequence

M
M
S 1 R
S 1 R S 1 M 0
j=1,...,m

i=1,...,n

which induces (by Lemma 10.1) the exact sequence

M
M
0 HomS 1 R (S 1 M, S 1 N )
S 1 N
i=1,...,n

which is the same as the one above.

j=1,...,m

S 1 N


11. Tensor products

00CV
00CW

Definition 11.1. Let R be a ring, M, N, P be three R-modules. A mapping

f : M N P (where M N is viewed only as Cartesian product of two Rmodules) is said to be R-bilinear if for each x M the mapping y 7 f (x, y) of N
into P is R-linear, and for each y N the mapping x 7 f (x, y) is also R-linear.

00CX

Lemma 11.2. Let M, N be R-modules. Then there exists a pair (T, g) where T
is an R-module, and g : M N T an R-bilinear mapping, with the following
universal property: For any R-module P and any R-bilinear mapping f : M N
P , there exists a unique R-linear mapping f : T P such that f = f g. In other
words, the following diagram commutes:
/P
?

M N
g

f0

Moreover, if (T, g) and (T 0 , g 0 ) are two pairs with this property, then there exists a
unique isomorphism j : T T 0 such that j g = g 0 .
The R-module T which satisfies the above universal property is called the tensor
product of R-modules M and N , denoted as M R N .

COMMUTATIVE ALGEBRA

21

Proof. We first prove the existence of such R-module T . Let M, N be R-modules.

Let T be the quotient module P/Q, where P is the free R-module R(M N ) and Q
is the R-module generated by all elements of the following types: (x M, y N )
(x + x0 , y) (x, y) (x0 , y),
(x, y + y 0 ) (x, y) (x, y 0 ),
(ax, y) a(x, y),
(x, ay) a(x, y)
Let : M N T denote the natural map. This map is R-bilinear, as implied
by the above relations when we check the bilinearity conditions. Denote the image
(x, y) = x y, then these elements generate T . Now let f : M N P be
an R-bilinear map, then we can define f 0 : T P by extending the mapping
f 0 (x y) = f (x, y). Clearly f = f 0 . Moreover, f 0 is uniquely determined by the
value on the generating sets {x y : x M, y N }. Suppose there is another pair
(T 0 , g 0 ) satisfying the same properties. Then there is a unique j : T T 0 and also
j 0 : T 0 T such that g 0 = j g, g = j 0 g 0 . But then both the maps (j j 0 ) g
and g satisfies the universal properties, so by uniqueness they are equal, and hence
j 0 j is identity on T . Similarly (j 0 j) g 0 = g 0 and j j 0 is identity on T 0 . So j is
an isomorphism.

00CY

Lemma 11.3. Let M, N, P be R-modules, then the bilinear maps

(x, y) 7 y x
(x + y, z) 7 x z + y z
(r, x) 7 rx
induce unique isomorphisms
M R N N R M,
(M N ) R P (M R P ) (N R P ),
R R M M
Proof. Omitted.

We may generalize the tensor product of two R-modules to finitely many R-modules,
and set up a correspondence between the multi-tensor product with multilinear
mappings. Using almost the same construction one can prove that:
00CZ

Lemma 11.4. Let M1 , . . . , Mr be R-modules. Then there exists a pair (T, g)

consisting of an R-module T and an R-multilinear mapping g : M1 . . . Mr T
with the universal property: For any R-multilinear mapping f : M1 . . . Mr P
there exists a unique R-module homomorphism f 0 : T P such that f 0 g = f .
Such a module T is unique up to unique isomorphism. We denote it M1 R . . .R Mr
and we denote the universal multilinear map (m1 , . . . , mr ) 7 m1 . . . mr .
Proof. Omitted.

00D0

Lemma 11.5. The homomorphisms

(M R N ) R P M R N R P M R (N R P )
such that f ((xy)z) = xyz and g(xyz) = x(yz), x M, y N, z P
are well-defined and are isomorphisms.

COMMUTATIVE ALGEBRA

22

Proof. We shall prove f is well-defined and is an isomorphism, and this proof

carries analogously to g. Fix any z P , then the mapping (x, y) 7 x y
z, x M, y N , is R-bilinear in x and y, and hence induces homomorphism
fz : M N M N P which sends fz (x y) = x y z. Then consider
(M N ) P M N P given by (w, z) 7 fz (w). The map is R-bilinear and
thus induces f : (M R N ) R P M R N R P and f ((x y) z) = x y z.
To construct the inverse, we note that the map : M N P (M N ) P is
R-trilinear. Therefore, it induces an R-linear map h : M N P (M N ) P
which agrees with the universal property. Here we see that h(xy z) = (xy)z.
From the explicit expression of f and h, f h and hf are identity maps of M N P
and (M N ) P respectively, hence f is our desired isomorphism.

Doing induction we see that this extends to multi-tensor products. Combined with
Lemma 11.3 we see that the tensor product operation on the category of R-modules
is associative, commutative and distributive.
00D1

Definition 11.6. An abelian group N is called an (A, B)-bimodule if it is both an

A-module and a B-module, and the actions A End(M ) and B End(M ) are
compatible in the sense that (ax)b = a(xb) for all a A, b B, x N . Usually we
denote it as A NB .

00D2

Lemma 11.7. For A-module M , B-module P and (A, B)-bimodule N , the modules
(M A N ) B P and M A (N B P ) can both be given (A, B)-bimodule structure,
and moreover
(M A N ) B P
= M A (N B P ).
Proof. A priori M A N is an A-module, but we can give it a B-module structure
by letting
(x y)b = x yb, x M, y N, b B
Thus M A N becomes an (A, B)-bimodule. Similarly for N B P , and thus for
(M A N ) B P and M A (N B P ). By Lemma 11.5, these two modules are
isomorphic as both as A-module and B-module via the same mapping.


00DE

Lemma 11.8. For any three R-modules M, N, P ,

HomR (M R N, P )
= HomR (M, HomR (N, P ))
Proof. An R-linear map f HomR (M R N, P ) corresponds to an R-bilinear map
f : M N P . For each x M the mapping y 7 f (x, y) is R-linear by the
universal property. Thus f corresponds to a map f : M HomR (N, P ). This
map is R-linear since
f (ax + y)(z) = f (ax + y, z) = af (x, z) + f (y, z) = (af (x) + f (y))(z),
for all a R, x M , y M and z N . Conversely, any f HomR (M, HomR (N, P ))
defines an R-bilinear map M N P , namely (x, y) 7 f (x)(y). So this is a natural one-to-one correspondence between the two modules HomR (M R N, P ) and
HomR (M, HomR (N, P )).


00DD

Lemma 11.9 (Tensor products commute with colimits). Let (Mi , ij ) be a system
over the partially ordered set I. Let N be an R-module. Then
colim(Mi N )
= (colim Mi ) N.

COMMUTATIVE ALGEBRA

23

Moreover, the isomorphism is induced by the homomorphisms i 1 : Mi N

M N where M = colimi Mi with natural maps i : Mi M .
Proof. First proof. The functor M 0
7 M 0 R N is left adjoint to the functor
N 0 7 HomR (N, N 0 ) by Lemma 11.8. Thus M 0 7 M 0 R N commutes with all
colimits, see Categories, Lemma 24.4.
Second direct proof. Let P = colim(Mi N ), M = colim Mi . Then for all i j,
the following diagram commutes:
/ M N
Mi N
i 1

ij 1


Mj N

id

j 1


/ M N

By Lemma 8.8, these maps induce a unique homomorphism : P M N , with

i : Mi N P given by i = (i 1).
To construct the inverse map, for each i I, there is the canonical R-bilinear
mapping gi : Mi N Mi N . This induces a unique mapping b : M N P
b
such that (
i 1) = i gi . It is R-bilinear. Thus it induces an R-linear mapping
: M N P . From the commutative diagram below:
Mi N

gi

/ Mi N

i id


M N

id

i id


/P

/ Mi N


/ M N

/P

we see that b = g, the canonical R-bilinear mapping g : M N M N . So

is identity on M N . From the right-hand square and triangle, is also
identity on P .

00DF

Lemma 11.10. Let

f

M1
M2
M3 0
be an exact sequence of R-modules and homomorphisms, and let N be any Rmodule. Then the sequence
00DG

(11.10.1)

f 1

g1

M1 N M2 N M3 N 0

is exact. In other words, the functor R N is right exact, in the sense that
tensoring each term in the original right exact sequence preserves the exactness.
Proof. We apply the functor Hom(, Hom(N, P )) to the first exact sequence. We
obtain
0 Hom(M3 , Hom(N, P )) Hom(M2 , Hom(N, P )) Hom(M1 , Hom(N, P ))
By Lemma 11.8, we have
0 Hom(M3 N, P ) Hom(M2 N, P ) Hom(M1 N, P )
Using the pullback property again, we arrive at the desired exact sequence.
00DH

Remark 11.11. However, tensor product does NOT preserve exact sequences in
general. In other words, if M1 M2 M3 is exact, then it is not necessarily true
that M1 N M2 N M3 N is exact for arbitrary R-module N .

COMMUTATIVE ALGEBRA

00DI

24

Example 11.12. Consider the injective map 2 : Z Z viewed as a map of

Z-modules. Let N = Z/2. Then the induced map Z Z/2 Z Z/2 is NOT
injective. This is because for x y Z Z/2,
(2 1)(x y) = 2x y = x 2y = x 0 = 0
Therefore the induced map is the zero map while Z N 6= 0.

00DJ

Remark 11.13. For R-modules N , if the functor R N is exact, i.e. tensoring

with N preserves all exact sequences, then N is said to be flat R-module. We will
discuss this later in Section 38.

05BS

Lemma 11.14. Let R be a ring. Let M and N be R-modules.

(1) If N and M are finite, then so is M R N .
(2) If N and M are finitely presented, then so is M R N .
Proof. Suppose M is finite. Then choose a presentation 0 K Rn M 0.
This gives an exact sequence K R N N n M R N 0 by Lemma 11.10.
We conclude that if N is finite too then M R N is a quotient of a finite module,
hence finite, see Lemma 5.3. Similarly, if both N and M are finitely presented,
then we see that K is finite and that M R N is a quotient of the finitely presented
module N n by a finite module, namely K R N , and hence finitely presented, see
Lemma 5.3.


00DK

Lemma 11.15. Let M be an R-module. Then the S 1 R-modules S 1 M and

S 1 RR M are canonically isomorphic, and the canonical isomorphism f : S 1 RR
M S 1 M is given by
f ((a/s) m) = am/s, a R, m M, s S
Proof. Obviously, the map f 0 : S 1 R M S 1 M given by f ((a/s, m)) = am/s
is bilinear, and thus by the universal property, this map induces a unique S 1 Rmodule homomorphism f : S 1 R R M S 1 M as in the statement of the
lemma. Actually every element in S 1 M is of the form m/s, m M, s S and
every element in S 1 R R M is of the form 1/s m. To see the latter fact, write
an element in S 1 R R M as
X ak
X ak tk
1 X
1
mk =
mk =
ak tk mk = m
sk
s
s
s
k

Where m = k ak tk mk . Then it is obvious that f is surjective, and if f ( 1s m) =

m/s = 0 then there exists t0 S with tm = 0 in M . Then we have
P

1
1
1
m=
tm =
0=0
s
st
st
Therefore f is injective.
00DL

Lemma 11.16. Let M, N be R-modules, then there is a canonical S 1 R-module

isomorphism f : S 1 M S 1 R S 1 N S 1 (M R N ), given by
f ((m/s) (n/t)) = (m n)/st

COMMUTATIVE ALGEBRA

25

Proof. We may use Lemma 11.7 and Lemma 11.15 repeatedly to see that these
two S 1 R-modules are isomorphic, noting that S 1 R is an (R, S 1 R)-bimodule:
S 1 (M R N )
= S 1 R R (M R N )

= S 1 M R N

= (S 1 M S 1 R S 1 R) R N

= S 1 M S 1 R (S 1 R R N )

= S 1 M S 1 R S 1 N
This isomorphism is easily seen to be the one stated in the lemma.

12. Tensor algebra

00DM

Let R be a ring. Let M be an R-module. We define the tensor algebra of M over

R to be the noncommutative R-algebra
M
T(M ) = TR (M ) =
Tn (M )
n0

with T (M ) = R, T (M ) = M , T (M ) = M R M , T3 (M ) = M R M R M ,
and so on. Multiplication is defined by the rule that on pure tensors we have
(x1 x2 . . . xn ) (y1 y2 . . . ym ) = x1 x2 . . . xn y1 y2 . . . ym
and we extend this by linearity.
We define the exterior algebra (M ) of M over R to be the quotient of T(M ) by
the two sided ideal generated by the elements x x T2 (M ). The image of a pure
tensor x1 . . . xn in n (M ) is denoted x1 . . . xn . These elements generate
n (M ), they are R-linear in each xi and they are zero when two of the xi are equal
(i.e., they are alternating as functions of x1 , x2 , . . . , xn ). The multiplication on
(M ) is graded commutative, i.e., every x M and y M satisfy x y = y x.
An example of this is when M = Rx1 . . . Rxn is a finite free module. In this
case (M ) is free over R with basis the elements
x i1 . . . x ir
with 0 r n and 1 i1 < i2 < . . . < ir n.
We define the symmetric algebra Sym(M ) of M over R to be the quotient of T(M )
by the two sided ideal generated by the elements x y y x T2 (M ). The
image of a pure tensor x1 . . . xn in Symn (M ) is denoted just x1 . . . xn . These
elements generate Symn (M ), these are R-linear in each xi and x1 . . . xn = x01 . . . x0n
if the sequence of elements x1 , . . . , xn is a permutation of the sequence x01 , . . . , x0n .
Thus we see that Sym(M ) is commutative.
An example of this is when M = Rx1 . . . Rxn is a finite free module. In this
case Sym(M ) = R[x1 , . . . , xn ] is a polynomial algebra.
00DN

Lemma 12.1. Let R be a ring. Let M be an R-module. If M is a free R-module,

so is each symmetric and exterior power.
Proof. Omitted, but see above for the finite free case.

COMMUTATIVE ALGEBRA

00DO

26

Lemma 12.2. Let R be a ring. Let M2 M1 M 0 be an exact sequence of

R-modules. There are exact sequences
M2 R Symn1 (M1 ) Symn (M1 ) Symn (M ) 0
and similarly
M2 R n1 (M1 ) n (M1 ) n (M ) 0
Proof. Omitted.

00DP

Lemma 12.3. Let R be a ring. Let M be an R-module. Let xi , i I be a given

system of generators of M as an R-module. Let n 2. There exists a canonical
exact sequence
M
M
M
M
Tn2 (M )
Tn2 (M ) Tn (M ) n (M ) 0
1j1 <j2 n i1 ,i2 I

1j1 <j2 n iI

where the pure tensor m1 . . . mn2 in the first summand maps to

m1 . . . xi1 . . . xi2 . . . mn2
{z
}
|

with xi1 and xi2 occupying slots j1 and j2 in the tensor

m1 . . . xi2 . . . xi1 . . . mn2

{z
}
|

with xi2 and xi1 occupying slots j1 and j2 in the tensor

and m1 . . . mn2 in the second summand maps to

m1 . . . xi . . . xi . . . mn2
|
{z
}

with xi and xi occupying slots j1 and j2 in the tensor

There is also a canonical exact sequence

M
M
Tn2 (M ) Tn (M ) Symn (M ) 0
1j1 <j2 n i1 ,i2 I

where the pure tensor m1 . . . mn2 maps to

m1 . . . xi1 . . . xi2 . . . mn2
|
{z
}

with xi1 and xi2 occupying slots j1 and j2 in the tensor

m1 . . . xi2 . . . xi1 . . . mn2

{z
}
|

with xi2 and xi1 occupying slots j1 and j2 in the tensor

Proof. Omitted.
00DQ

Lemma 12.4. Let R be a ring. Let Mi be a directed system of R-modules. Then

colimi T(M ) = T(colimi Mi ) and similarly for the symmetric and exterior algebras.
Proof. Omitted.


13. Base change

05G3

We formally introduce base change in algebra as follows.

05G4

Definition 13.1. Let : R S be a ring map. Let M be an S-module. Let

R R0 be any ring map. The base change of by R R0 is the ring map
R0 S R R0 . In this situation we often write S 0 = S R R0 . The base change of
the S-module M is the S 0 -module M R R0 .

COMMUTATIVE ALGEBRA

27

If S = R[xi ]/(fj ) for some collection of variables xi , i I and some collection of

polynomials fj R[xi ], j J, then S R R0 = R0 [xi ]/(fj0 ), where fj0 R0 [xi ] is the
image of fj under the map R[xi ] R0 [xi ] induced by R R0 . This simple remark
is the key to understanding base change.
05G5

Lemma 13.2. Let R S be a ring map. Let M be an S-module. Let R R0 be

a ring map and let S 0 = S R R0 and M 0 = M R R0 be the base changes.
(1) If M is a finite S-module, then the base change M 0 is a finite S 0 -module.
(2) If M is an S-module finite presentation, then the base change M 0 is an
S 0 -module of finite presentation.
(3) If R S is of finite type, then the base change R0 S 0 is of finite type.
(4) If R S is of finite presentation, then the base change R0 S 0 is of finite
presentation.
Proof. Proof of (1). Take a surjective, R-linear map Rn M 0. By Lemma
n
11.3 and 11.10 the result after tensoring with R0 is a surjection R0
M 0 0,
0
0
so M is a finitely generated R -module. Proof of (2). Take a presentation Rm
Rn M 0. By Lemma 11.3 and 11.10 the result after tensoring with R0 gives
m
n
a finite presentation R0
R0 M 0 0, of the R0 -module M 0 . Proof of (3).
This follows by the remark preceding the lemma as we can take I to be finite by
assumption. Proof of (4). This follows by the remark preceding the lemma as we
can take I and J to be finite by assumption.

Let : R S be a ring map. Given a S-module N we obtain an R-module NR by
the rule r n = (r)n. This is sometimes called the restriction of N to R.

05DQ

Lemma 13.3. Let R S be a ring map. The functors ModS ModR , N 7 NR

(restriction) and ModR ModS , M 7 M R S (base change) are adjoint functors.
In a formula
HomR (M, NR ) = HomS (M R S, N )
Proof. If : M NR is an R-module map, then we define 0 : M R S N by
the rule 0 (m s) = s(m). If : M R S N is an S-module map, we define
0 : M NR by the rule 0 (m) = (m 1). We omit the verification that these
constructions are mutually inverse.

The lemma above tells us that restriction has a left adjoint, namely base change.
It also has a right adjoint.

08YP

Lemma 13.4. Let R S be a ring map. The functors ModS ModR , N 7 NR

(restriction) and ModR ModS , M 7 HomR (S, M ) are adjoint functors. In a
formula
HomR (NR , M ) = HomS (N, HomR (S, M ))
Proof. If : NR M is an R-module map, then we define 0 : N HomR (S, M )
by the rule 0 (n) = (s 7 (sn)). If : N HomR (S, M ) is an S-module map,
we define 0 : NR M by the rule 0 (n) = (n)(1). We omit the verification that
these constructions are mutually inverse.


08YQ

Lemma 13.5. Let R S be a ring map. Given S-modules M, N and an R-module

P we have
HomR (M S N, P ) = HomS (M, HomR (N, P ))

COMMUTATIVE ALGEBRA

28

Proof. This can be proved directly, but it is also a consequence of Lemmas 13.4
and 11.8. Namely, we have
HomR (M S N, P ) = HomS (M S N, HomR (S, P ))
= HomS (M, HomS (N, HomR (S, P )))
= HomS (M, HomR (N, P ))
as desired.


14. Miscellany

00DR

The proofs in this section should not refer to any results except those from the
section on basic notions, Section 3.

07K1

Lemma 14.1. Let R be a ring, I and J two ideals and p a prime ideal containing
the product IJ. Then p contains I or J.
Proof. Assume the contrary and take x I \ p and y J \ p. Their product is an
element of IJ p, which contradicts the assumption that p was prime.


00DS

Lemma 14.2 (Prime avoidance). Let R be a ring. Let Ii R, i = 1, . . . , r, and

J R be ideals. Assume
(1) J 6 Ii for i = 1, . . . , r, and
(2) all but two of Ii are prime ideals.
Then there exists an x J, x 6 Ii for all i.
Proof. The result is true for r = 1. If r = 2, then let x, y J with x 6 I1 and
y 6 I2 . We are done unless x I2 and y I1 . Then the element x + y cannot be
in I1 (since that would mean x + y y I1 ) and it also cannot be in I2 .
For r 3, assume the result holds for r 1. After renumbering we may assume
that Ir is prime. We may also assume there are no inclusions among the Ii . Pick
x J, x 6 Ii for all i = 1, . . . , r 1. If x 6 Ir we are done. So assume x Ir .
If JI1 . . . Ir1 Ir then J Ir (by Lemma 14.1) a contradiction. Pick y
JI1 . . . Ir1 , y 6 Ir . Then x + y works.


00DT

Lemma 14.3 (Chinese remainder). Let R be a ring.

(1) If I1 , . . . , Ir are ideals such that Ia + Ib = R when a 6= b, then I1 . . . Ir =
I1 I2 . . . Ir and R/(I1 I2 . . . Ir )
= R/I1 . . . R/Ir .
(2) If m1 , . . . , mr are pairwise distinct maximal ideals then ma + mb = R for
a 6= b and the above applies.
Proof. Let us first prove I1 . . .Ir = I1 . . . Ir as this will also imply the injectivity
of the induced ring homomorphism R/(I1 . . . Ir ) R/I1 . . .R/Ir . The inclusion
I1 . . . Ir I1 . . . Ir is always fulfilled since ideals are closed under multiplication
with arbitrary ring elements. To prove the other inclusion, we claim that the ideals
I1 . . . Ii . . . Ir , i = 1, . . . , r
generate the ring R. We prove this by induction on r. It holds when r = 2. If
r > 2, then we see that R is the sum of the ideals I1 . . . Ii . . . Ir1 , i = 1, . . . , r 1.
Hence Ir is the sum of the ideals I1 . . . Ii . . . Ir , i = 1, . . . , r 1. Applying the
same argument with the reverse ordering on the ideals we see that I1 is the sum
of the ideals I1 . . . Ii . . . Ir , i = 2, . . . , r. Since R = I1 + Ir by assumption we see

COMMUTATIVE ALGEBRA

29

that R is the sum of the ideals displayed above. Therefore we can find elements
ai I1 . . . Ii . . . Ir such that their sum is one. Multiplying this equation by an
element of I1 . . . Ir gives the other inclusion. It remains to show that the
canonical map R/(I1 . . . Ir ) P
R/I1 . . . R/Ir is surjective. For this, consider
r
its action on the equation 1 = i=1 ai we derived above. On the one hand, a ring
morphism sends 1 to 1 and on the other hand, the image of any ai is zero in R/Ij
for j 6= i. Therefore, the image of ai in R/Ii is
identity. So given any element
Pthe
r
(b1 , . . . , br ) R/I1 . . . R/Ir , the element i=1 ai bi is an inverse image in R.
To see (2), by the very definition of being distinct maximal ideals, we have ma +mb =
R for a 6= b and so the above applies.

07DQ

Lemma 14.4. Let R be a ring. Let n m. Let A be an n m matrix with

coefficients in R. Let J R be the ideal generated by the m m minors of A.
(1) For any f J there exists a m n matrix B such that BA = f 1mm .
(2) If f R and BA = f 1mm for some m m matrix B, then f m J.
Proof. For I {1, . . . , n} with |I| = m, we denote by EI the m n matrix of the
projection
M
M
Rn =
R
R
i{1,...,n}

iI

and set AI = EI A, i.e., AI is the m m matrix whose rows are the rows of A
with indices in I. Let BI be the adjugate (transpose of cofactor) matrix to AI ,
i.e., such that AI BI = BI AI = det(AI )1mm . The m m minors of A are the
determinants
P det AI for all the I {1, . . . , n} withP|I| = m. If f J then we can
write f =
cI det(AI ) for some cI R. Set B = cI BI EI to see that (1) holds.
P
If f 1mm = BA then by the Cauchy-Binet formula we have f m =
bI det(AI )
where bI is the determinant of the m m matrix whose columns are the columns
of B with indices in I.

080R

Lemma 14.5. Let R be a ring. Let n m. Let A = (aij ) be an n m matrix

with coefficients in R, written in block form as
 
A1
A=
A2
where A1 has size m m. Let B be the adjugate (transpose of cofactor) matrix to
A1 . Then


f 1mm
AB =
C
where f = det(A1 ) and cij is (up to sign) the determinant of the m m minor of
A corresponding to the rows 1, . . . , j, . . . , m, i.
Proof. Since the adjugate has the property A1 B = BA1 = f the first block of the
expression for AB is correct. Note that
X
X
cij =
aik bkj =
(1)j+k aik det(Ajk
1 )
k

Aij
1

where
means A1 with the jth row and kth column removed. This last expression
is the row expansion of the determinant of the matrix in the statement of the
lemma.


COMMUTATIVE ALGEBRA

30

15. Cayley-Hamilton
05G6
00DX

Lemma 15.1. Let R be a ring. Let A = (aij ) be an nn matrix with coefficients in

R. Let P (x) R[x] be the characteristic polynomial of A (defined as det(xidnn
A)). Then P (A) = 0 in Mat(n n, R).
Proof. We reduce the question to the well-known Cayley-Hamilton theorem from
linear algebra in several steps:
(1) If : S R is a ring morphism and bij are inverse images of the aij under
this map, then it suffices to show the statement for S and (bij ) since is a
ring morphism.
(2) If : R , S is an injective ring morphism, it clearly suffices to show the
result for S and the aij considered as elements of S.
(3) Thus we may first reduce to the case R = Z[Xij ], aij = Xij of a polynomial
ring and then further to the case R = Q(Xij ) where we may finally apply
Cayley-Hamilton.


05BT

Lemma 15.2. Let R be a ring. Let M be a finite R-module. Let : M M

be an endomorphism. Then there exists a monic polynomial P R[T ] such that
P () = 0 as an endomorphism of M .
Proof.
Choose a surjective R-module map Rn M , given by (a1 , . . . , an ) 7
P
n
P ai xi for some generators xi M . Choose (ai1 , . . . , ain ) R such that (xi ) =
aij xj . In other words the diagram
Rn
A

Rn

/M


/M

is commutative where A = (aij ). By Lemma 15.1 there exists a monic polynomial

P such that P (A) = 0. Then it follows that P () = 0.

05G7

Lemma 15.3. Let R be a ring. Let I R be an ideal. Let M be a finite R-module.

Let : M M be an endomorphism such that (M ) IM . Then there exists a
monic polynomial P = tn +a1 tn1 +. . .+an R[T ] such that aj I j and P () = 0
as an endomorphism of M .
Proof. Choose a surjective R-module map Rn M , given by (a1 , . . . , an ) 7
P
n
such that (xi ) =
P ai xi for some generators xi M . Choose (ai1 , . . . , ain ) I
aij xj . In other words the diagram
Rn
A

I n

/M


/M

is commutative where A = (aij ). By Lemma 15.1 the polynomial P (t) = det(tidnn

A) has all the desired properties.

As a fun example application we prove the following surprising lemma.

COMMUTATIVE ALGEBRA

05G8

31

Lemma 15.4. Let R be a ring. Let M be a finite R-module. Let : M M be

a surjective R-module map. Then is an isomorphism.
First proof. Write R0 = R[x] and think of M as a finite R0 -module with x acting
via . Set I = (x) R0 . By our assumption that is surjective we have IM = M .
Hence we may apply Lemma 15.3 to M as an R0 -module, the ideal I and the
endomorphism idM . We conclude that (1 + a1 + . . . + an )idM = 0 with aj I.
Write aj P
= bj (x)x for some bj (x) R[x]. Translating back into we see that
idM = ( j=1,...,n bj ()), and hence is invertible.

Second proof. We perform induction on the number of generators of M over R.
If M is generated by one element, then M
= R/I for some ideal I R. In this
case we may replace R by R/I so that M = R. In this case : R R is given by
multiplication on M by an element r R. The surjectivity of forces r invertible,
since must hit 1, which implies that is invertible.
Now assume that we have proven the lemma in the case of modules generated by
n 1 elements, and are examining a module M generated by n elements. Let
A mean the ring R[t], and regard the module M as an A-module by letting t
act via ; since M is finite over R, it is finite over R[t] as well, and since were
trying to prove injective, a set-theoretic property, we might as well prove the
endomorphism t : M M over A injective. We have reduced our problem to the
case our endomorphism is multiplication by an element of the ground ring. Let
M 0 M denote the sub-A-module generated by the first n 1 of the generators of
M , and consider the diagram
0

/ M0
|M 0


/ M0

/M


/M

/ M/M 0

/0

mod M 0


/ M/M 0

/ 0,

where the restriction of to M 0 and the map induced by on the quotient M/M 0
are well-defined since is multiplication by an element in the base, and M 0 and
M/M 0 are A-modules in their own right. By the case n = 1 the map M/M 0
M/M 0 is an isomorphism. A diagram chase implies that |M 0 is surjective hence
by induction |M 0 is an isomorphism. This forces the middle column to be an
isomorphism by the snake lemma.

16. The spectrum of a ring
00DY

We arbitrarily decide that the spectrum of a ring as a topological space is part of

the algebra chapter, whereas an affine scheme is part of the chapter on schemes.

00DZ

Definition 16.1. Let R be a ring.

(1) The spectrum of R is the set of prime ideals of R. It is usually denoted
Spec(R).
(2) Given a subset T R we let V (T ) Spec(R) be the set of primes containing T , i.e., V (T ) = {p Spec(R) | f T, f p}.
(3) Given an element f R we let D(f ) Spec(R) be the set of primes not
containing f .

00E0

Lemma 16.2. Let R be a ring.

COMMUTATIVE ALGEBRA

(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
(13)
(14)
(15)
(16)
(17)

32

The spectrum of a ring R is empty if and only if R is the zero ring.

Every nonzero ring has a maximal ideal.
Every nonzero ring has a minimal prime ideal.
Given an ideal I R and a prime ideal I p there exists a prime I q p
such that q is minimal over I.
If T R, and if (T )
is the ideal generated by T in R, then V ((T )) = V (T ).
If Iis an ideal and I is its radical, see basic notion (27), then V (I) =
V ( I).

T
Given an ideal I of R we have I = Ip p.
If I is an ideal then V (I) = if and only if I is the unit ideal.
If I, J are ideals of R then V (I) V (J) = V (I J).
If (Ia )aA is a set of ideals of R then aA V (Ia ) = V (aA Ia ).
If f R, then D(f ) q V (f ) = Spec(R).
If f R then D(f ) = if and only if f is nilpotent.
If f = uf 0 for some unit u R, then D(f ) = D(f 0 ).
If I R is an ideal, and p is a prime of R with p 6 V (I), then there exists
an f R such that p D(f ), and D(f ) V (I) = .
If f, g R, then D(f g) = D(f
S ) D(g).
If fi R for i I, then iI D(fi ) is the complement of V ({fi }iI ) in
Spec(R).
If f R and D(f ) = Spec(R), then f is a unit.

Proof. We address each part in the corresponding item below.

(1) This is a direct consequence of (2) or (3).
(2) Let A be the set of all proper ideals of R. This set is ordered by inclusion
and is non-empty, since
S (0) A is a proper ideal. Let A be a totally
ordered subset of R. IA I is in fact an ideal. Since 1S
/ I for all I A,
the union does not contain 1 and thus is proper. Hence IA I is in A and
is an upper bound for the set A. Thus by Zorns lemma A has a maximal
element, which is the sought-after maximal ideal.
(3) Since R is nonzero, it contains a maximal ideal which is a prime ideal. Thus
the set A of all prime ideals of R is nonempty. A is ordered by reverseinclusion.
Let A be a totally ordered subset of A. Its pretty clear that
T
J = IA I is in fact an ideal. Not so clear, however, is that it is prime.
Let xyT J. Then xy I for all I A. Now let B = {I A|y I}. Let
K = IB I. Since A is totally ordered, either K = J (and were done,
since then y J) or K J and for all I A such that I is properly
contained in K, we have y
/ I. But that means that for all those I, x I,
since they are prime. Hence x J. In either case, J is prime as desired.
Hence by Zorns lemma we get a maximal element which in this case is a
minimal prime ideal.
(4) This is the same exact argument as (3) except you only consider prime
ideals contained in p and containing I.
(5) (T ) is the smallest ideal containing T . Hence if T I, some ideal, then
(T ) I as well. Hence if I V (T ), then I V ((T )) as well. The other
inclusion is
obvious.

(6) Since I I, V ( I) V (I). Now let p V (I). Let x I. Then

xn I for some n. Hence xn p. Butsince p is prime, a
boring induction
argument gets you that x p. Hence I p and p V ( I).

COMMUTATIVE ALGEBRA

33

(7) Let f R \ I. Then f n

/ I for all n. Hence S = {1, f, f 2 , . . .} is
S 1 R
a multiplicative subset, not containing 0. Take a prime ideal p
1
is a prime
containing S I. Then the pull-back p in R ofTp
ideal containing

I that does not intersect S. This shows that Ip p I. Now if a I,

then an I for some n. Hence if I p, then an p. But since p is prime,
we have a p. Thus the equality is shown.
(8) I is not the unit ideal if and only if I is contained in some maximal ideal
(to see this, apply (2) to the ring R/I) which is therefore prime.
(9) If p V (I) V (J), then I p or J p which means that I J p. Now
if I
TJ p, then IJ p and hence either I of J is in p, since p is prime.
(10) p aA V (Ia ) Ia p, a A p V (aA Ia )
(11) If p is a prime ideal and f R, then either f p or f
/ p (strictly) which
is what the disjoint union says.
(12) If a R is nilpotent, then an = 0 for some n. Hence an p for any prime
ideal. Thus a p as can be shown by induction and D(f ) = . Now, as
shown in (7), if a R is not nilpotent, then there is a prime ideal that does
not contain it.
(13) f p uf p, since u is invertible.
(14) If p
/ V (I), then f I \ p. Then f
/ p so p D(f ). Also if q D(f ),
then f
/ q and thus I is not contained in q. Thus D(f ) V (I) = .
(15) If f g p, then f p or g p. Hence if f
/ p and g
/ p, then f g
/ p. Since
p is S
an ideal, if f g
/ p, then f
/ p and g
/ p.
(16) p iI D(fi ) i I, fi
/ p p Spec(R) \ V ({fi }iI )
(17) If D(f ) = Spec(R), then V (f ) = and hence f R = R, so f is a unit.

The lemma implies that the subsets V (T ) from Definition 16.1 form the closed
subsets of a topology on Spec(R). And it also shows that the sets D(f ) are open
and form a basis for this topology.
00E1

Definition 16.3. Let R be a ring. The topology on Spec(R) whose closed sets
are the sets V (T ) is called the Zariski topology. The open subsets D(f ) are called
the standard opens of Spec(R).
It should be clear from context whether we consider Spec(R) just as a set or as a
topological space.

00E2

Lemma 16.4. Suppose that : R R0 is a ring homomorphism. The induced

map
Spec() : Spec(R0 ) Spec(R), p0 7 1 (p0 )
is continuous for the Zariski topologies. In fact, for any element f R we have
Spec()1 (D(f )) = D((f )).
Proof. It is basic notion (41) that p := 1 (p0 ) is indeed a prime ideal of R. The
last assertion of the lemma follows directly from the definitions, and implies the
first.

If 0 : R0 R00 is a second ring homomorphism then the composition
Spec(R00 ) Spec(R0 ) Spec(R)
equals Spec(0 ). In other words, Spec is a contravariant functor from the
category of rings to the category of topological spaces.

COMMUTATIVE ALGEBRA

00E3

34

Lemma 16.5. Let R be a ring. Let S R be a multiplicative subset. The map

R S 1 R induces via the functoriality of Spec a homeomorphism
Spec(S 1 R) {p Spec(R) | S p = }
where the topology on the right hand side is that induced from the Zariski topology
on Spec(R). The inverse map is given by p 7 S 1 p.
Proof. Denote the right hand side of the arrow of the lemma by D. Choose a
prime p0 S 1 R and let p the inverse image of p0 in R. Since p0 does not contain
1 we see that p does not contain any element of S. Hence p D and we see that
the image is contained in D. Let p D. By assumption the image S does not
1
contain 0. By basic notion (54) S (R/p) is not the zero ring. By basic notion
1
(62) we see S 1 R/S 1 p = S (R/p) is a domain, and hence S 1 p is a prime.
The equality of rings also shows that the inverse image of S 1 p in R is equal to
1
p, because R/p S (R/p) is injective by basic notion (55). This proves that
1
the map Spec(S R) Spec(R) is bijective onto D with inverse as given. It is
continuous by Lemma 16.4. Finally, let D(g) Spec(S 1 R) be a standard open.
Write g = h/s for some h R and s S. Since g and h/1 differ by a unit we have
D(g) = D(h/1) in Spec(S 1 R). Hence by Lemma 16.4 and the bijectivity above
the image of D(g) = D(h/1) is D D(h). This proves the map is open as well. 

00E4

Lemma 16.6. Let R be a ring. Let f R. The map R Rf induces via the
functoriality of Spec a homeomorphism
Spec(Rf ) D(f ) Spec(R).
The inverse is given by p 7 p Rf .
Proof. This is a special case of Lemma 16.5.

It is not the case that every affine open of a spectrum is a standard open. See
Example 26.4.
00E5

Lemma 16.7. Let R be a ring. Let I R be an ideal. The map R R/I induces
via the functoriality of Spec a homeomorphism
Spec(R/I) V (I) Spec(R).
The inverse is given by p 7 p/I.
Proof. It is immediate that the image is contained in V (I). On the other hand, if
p V (I) then p I and we may consider the ideal p/I R/I. Using basic notion
(51) we see that (R/I)/(p/I) = R/p is a domain and hence p/I is a prime ideal.
From this it is immediately clear that the image of D(f + I) is D(f ) V (I), and
hence the map is a homeomorphism.


COMMUTATIVE ALGEBRA

00E6

35

Remark 16.8. A fundamental commutative diagram associated to : R S,

q S and p = 1 (q) is the following
(q) = Sq /qSq o
O

Sq o
O

SO

/ S/q
O

/ (q) = f.f.(S/q)
O

(p) R S = Sp /pSp o
O

Sp o
O

SO

/ S/pS
O

/ (R \ p)1 S/pS
O

(p) = Rp /pRp o

Rp o

/ R/p

/ (p) = f.f.(R/p)

In this diagram the arrows in the outer left and outer right columns are identical.
The horizontal maps induce on the associated spectra always a homeomorphism
onto the image. The lower two rows of the diagram make sense without assuming q
exists. The lower squares induce fibre squares of topological spaces. This diagram
shows that p is in the image of the map on Spec if and only if S R (p) is not the
zero ring.
00E7

Lemma 16.9. Let : R S be a ring map. Let p be a prime of R. The following

are equivalent
(1) p is in the image of Spec(S) Spec(R),
(2) S R (p) 6= 0,
(3) Sp /pSp 6= 0,
(4) (S/pS)p 6= 0, and
(5) p = 1 (pS).
Proof. We have already seen the equivalence of the first two in Remark 16.8. The
others are just reformulations of this.


00E8

Lemma 16.10. Let R be a ring. The space Spec(R) is quasi-compact.

Proof. It suffices to prove that any covering of Spec(R) by standard opens can
be refined by a finite covering. Thus suppose that Spec(R) = D(fi ) for a set of
elements {fi }iI of R. This means that V (fi ) = . According to Lemma 16.2
this means that V ({fi }) = . According to the same lemma this means that the
ideal generated by the fiPis the unit ideal of R. This means that we can write
1 as a finite sum: 1 =
iJ ri fi with J I finite. And then it follows that
Spec(R) = iJ D(fi ).


04PM

Lemma 16.11. Let R be a ring. The topology on X = Spec(R) has the following
properties:
(1) X is quasi-compact,
(2) X has a basis for the topology consisting of quasi-compact opens, and
(3) the intersection of any two quasi-compact opens is quasi-compact.
Proof. The spectrum of a ring is quasi-compact, see Lemma 16.10. It has a basis
for the topology consisting of the standard opens D(f ) = Spec(Rf ) (Lemma 16.6)
which are quasi-compact by the first remark. The intersection of two standard
opens is quasi-compact as D(f ) D(g) = D(f g). Given any two quasi-compact
opens U, V XSwe may write U = D(f1 ). . .D(fn ) and V = D(g1 ). . .D(gm ).
Then U V = D(fi gj ) which is quasi-compact.


COMMUTATIVE ALGEBRA

36

17. Local rings

07BH
07BI

Local rings are the bread and butter of algebraic geometry.

Definition 17.1. A local ring is a ring with exactly one maximal ideal. The
maximal ideal is often denoted mR in this case. We often say let (R, m, ) be a
local ring to indicate that R is local, m is its unique maximal ideal and = R/m
is its residue field. A local homomorphism of local rings is a ring map : R S
such that R and S are local rings and such that (mR ) mS . If it is given that R
and S are local rings, then the phrase local ring map : R S means that is
a local homomorphism of local rings.
A field is a local ring. Any ring map between fields is a local homomorphism of
local rings.

00E9

Lemma 17.2. Let R be a ring. The following are equivalent:

(1)
(2)
(3)
(4)

R is a local ring,
Spec(R) has exactly one closed point,
R has a maximal ideal m and every element of R \ m is a unit, and
R is not the zero ring and for every x R either x or 1 x is invertible
or both.

Proof. Let R be a ring, and m a maximal ideal. If x R \ m, and x is not a unit

then there is a maximal ideal m0 containing x. Hence R has at least two maximal
ideals. Conversely, if m0 is another maximal ideal, then choose x m0 , x 6 m.
Clearly x is not a unit. This proves the equivalence of (1) and (3). The equivalence
(1) and (2) is tautological. If R is local then (4) holds since x is either in m or not.
If (4) holds, and m, m0 are distinct maximal ideals then we may choose x R such
that x mod m0 = 0 and x mod m = 1 by the Chinese remainder theorem (Lemma
14.3). This element x is not invertible and neither is 1 x which is a contradiction.
Thus (4) and (1) are equivalent.

The localization Rp of a ring R at a prime p is a local ring with maximal ideal pRp .
Namely, the quotient Rp /pRp is the fraction field of the domain R/p and every
element of Rp which is not contained in pRp is invertible.
07BJ

Lemma 17.3. Let : R S be a ring map. Assume R and S are local rings.
The following are equivalent:
(1)
(2)
(3)
(4)

is a local ring map,

(mR ) mS , and
1 (mS ) = mR .
For any x R, if (x) is invertible in S, then x is invertible in R.

Proof. Conditions (1) and (2) are equivalent by definition. If (3) holds then (2)
holds. Conversely, if (2) holds, then 1 (mS ) is a prime ideal containing the maximal ideal mR , hence 1 (mS ) = mR . Finally, (4) is the contrapositive of (2) by
Lemma 17.2.

Let : R S be a ring map. Let q S be a prime and set p = 1 (q). Then the
induced ring map Rp Sq is a local ring map.

COMMUTATIVE ALGEBRA

37

18. The Jacobson radical of a ring

0AMD

We recall that the Jacobson radical rad(R) of a ring R is the intersection of all
maximal ideals of R. If R is local then rad(R) is the maximal ideal of R.

0AME

Lemma 18.1. Let R be a ring and let I R be an ideal. The following are
equivalent
(1) I rad(R), and
(2) every element of 1 + I is a unit in R.
In this case every element of R which maps to a unit of R/I is a unit.
Proof. If f rad(R), then f m for all maximal ideals m of R. Hence 1 + f 6 m
for all maximal ideals m of R. Thus the closed subset V (1 + f ) of Spec(R) is empty.
This implies that 1 + f is a unit, see Lemma 16.2.
Conversely, assume that 1 + f is a unit for all f I. If m is a maximal ideal and
I 6 m, then I + m = R. Hence 1 = f + g for some g m and f I. Then
g = 1 + (f ) is not a unit, contradiction.
For the final statement let f R map to a unit in R/I. Then we can find g R
mapping to the multiplicative inverse of f mod I. Then f g = 1 mod I. Hence f g
is a unit of R by (2) which implies that f is a unit.


0B7C

Lemma 18.2. Let : R S be a ring map such that the induced map Spec(S)
Spec(R) is surjective. Then an element x R is a unit if and only if (x) S is
a unit.
Proof. If x is a unit, then so is (x). Conversely, if (x) is a unit, then (x) 6 q
for all q Spec(S). Hence x 6 1 (q) = Spec()(q) for all q Spec(S). Since
Spec() is surjective we conclude that x is a unit by part (17) of Lemma 16.2. 
19. Nakayamas lemma

07RC

We quote from [Mat70]: This simple but important lemma is due to T. Nakayama,
G. Azumaya and W. Krull. Priority is obscure, and although it is usually called
the Lemma of Nakayama, late Prof. Nakayama did not like the name.

00DV

Lemma 19.1 (Nakayamas lemma). Let R be a ring, let M be an R-module, and

let I R be an ideal.
(1) If IM = M and M is finite, then there exists a f 1+I such that f M = 0.
(2) If IM = M , M is finite, and I rad(R), then M = 0.
(3) If N, N 0 M , M = N + IN 0 , and N 0 is finite, then there exists a f 1 + I
such that Mf = Nf .
(4) If N, N 0 M , M = N + IN 0 , N 0 is finite, and I rad(R), then M = N .
(5) If N M is a module map, N/IN M/IM is surjective, and M is
finite, then there exists a f 1 + I such that Nf Mf is surjective.
(6) If N M is a module map, N/IN M/IM is surjective, M is finite,
and I rad(R), then N M is surjective.
(7) If x1 , . . . , xn M generate M/IM and M is finite, then there exists an
f 1 + I such that x1 , . . . , xn generate Mf over Rf .
(8) If x1 , . . . , xn M generate M/IM , M is finite, and I rad(R), then M
is generated by x1 , . . . , xn .
(9) If IM = M , I is nilpotent, then M = 0.

00DW

[Mat70, 1.M Lemma

(NAK) page 11]

COMMUTATIVE ALGEBRA

38

(10) If N, N 0 M , M = N + IN 0 , and I is nilpotent then M = N .

(11) If N M is a module map, I is nilpotent, and N/IN M/IM is
surjective, then N M is surjective.
(12) If {x }A is a set of elements of M which generate M/IM and I is
nilpotent, then M is generated by the x .
Proof. Proof ofP(1). Choose generators y1 , . . . , ym of M over R. For each
P i we
can write yi =
zij yj with zij I (since M = IM ). In other words j (ij
zij )yj = 0. Let f be the determinant of the m m matrix A = (ij zij ). Note
that f 1 + I (since the matrix A is entrywise congruent to the m m identity
matrix modulo I). By Lemma 14.4 (1), there
P exists an m m matrix B such that
BA
=
f
1
.
Writing
out
we
see
that
mm
i bhi aij = f hj for all h and j; hence,
P
P
b
a
y
=
f

y
=
f
y
for
every
h.
In other words, 0 = f yh for every h
hi
ij
j
hj
j
h
i,j
j
P
(since each i satisfies j aij yj = 0). This implies that f annihilates M .
By Lemma 16.2 an element of 1 + rad(R) is invertible element of R. Hence we see
that (1) implies (2). We obtain (3) by applying (1) to M/N which is finite as N 0
is finite. We obtain (4) by applying (2) to M/N which is finite as N 0 is finite. We
obtain (5) by applying (3) to M and the submodules Im(N M ) and M . We
obtain (6) by applying (4) to M and the submodules Im(N M ) and M . We
obtain (7) by applying (5) to the map Rn M , (a1 , . . . , an ) 7 a1 x1 + . . . + an xn .
We obtain (8) by applying (6) to the map Rn M , (a1 , . . . , an ) 7 a1 x1 + . . . +
an xn .
Part (9) holds because if M = IM then M = I n M for all n 0 and I being
nilpotent means I n = 0 for some n  0. Parts (10), (11), and (12) follow from (9)
by the arguments used above.

20. Open and closed subsets of spectra
04PN

It turns out that open and closed subsets of a spectrum correspond to idempotents
of the ring.

00EC

Lemma 20.1. Let R be a ring. Let e R be an idempotent. In this case

Spec(R) = D(e) q D(1 e).
Proof. Note that an idempotent e of a domain is either 1 or 0. Hence we see that
D(e)

= {p Spec(R) | e 6 p}
= {p Spec(R) | e 6= 0 in (p)}
= {p Spec(R) | e = 1 in (p)}

Similarly we have
D(1 e)

{p Spec(R) | 1 e 6 p}

= {p Spec(R) | e 6= 1 in (p)}
= {p Spec(R) | e = 0 in (p)}
Since the image of e in any residue field is either 1 or 0 we deduce that D(e) and
D(1 e) cover all of Spec(R).


COMMUTATIVE ALGEBRA

00ED

39

Lemma 20.2. Let R1 and R2 be rings. Let R = R1 R2 . The maps R R1 ,

(x, y) 7 x and R R2 , (x, y) 7 y induce continuous maps Spec(R1 ) Spec(R)
and Spec(R2 ) Spec(R). The induced map
Spec(R1 ) q Spec(R2 ) Spec(R)
is a homeomorphism. In other words, the spectrum of R = R1 R2 is the disjoint
union of the spectrum of R1 and the spectrum of R2 .
Proof. Write 1 = e1 + e2 with e1 = (1, 0) and e2 = (0, 1). Note that e1 and
e2 = 1 e1 are idempotents. We leave it to the reader to show that R1 = Re1 is the
localization of R at e1 . Similarly for e2 . Thus the statement of the lemma follows
from Lemma 20.1 combined with Lemma 16.6.

We reprove the following lemma later after introducing a glueing lemma for functions. See Section 22.

00EE

Lemma 20.3. Let R be a ring. For each U Spec(R) which is open and closed
there exists a unique idempotent e R such that U = D(e). This induces a 11 correspondence between open and closed subsets U Spec(R) and idempotents
e R.
First proof of Lemma 20.3. Let U Spec(R) be open and closed. Since U is
closedSit is quasi-compact by Lemma 16.10, and similarly for its complement. Write
n
U
Sm= i=1 D(fi ) as a finite union of standard opens. Similarly, write Spec(R) \ U =
j=1 D(gj ) as a finite union of standard opens. Since = D(fi ) D(gj ) = D(fi gj )
we see that fi gj is nilpotent by Lemma 16.2. Let I = (f1 , . . . , fn ) R and let
J = (g1 , . . . , gm ) R. Note that V (J) equals U , that V (I) equals the complement
of U , so Spec(R) = V (I) q V (J). By the remark on nilpotency
above,
S
S we see
that (IJ)N = (0) for some sufficiently large integer N . Since D(fi ) D(gj ) =
Spec(R) we see that I + J = R, see Lemma 16.2. By raising this equation to the
2N th power we conclude that I N + J N = R. Write 1 = x + y with x I N and
y J N . Then 1 = (x + y)2 = x2 + y 2 because I N J N = (0). Then z = x x2
I N J N . Thus zx = 0 and z 2 = 0. Hence (x z) (x z)2 = x x2 z = 0. In
other words, e = x z is an idempotent contained in I N I, and the idempotent
e0 = 1 e = y + z is contained in J N J. This shows that the idempotent e maps
to 1 in every residue field (p) for p V (J) and that e maps to 0 in (p) for every
p V (I).
To see uniqueness suppose that e1 , e2 are distinct idempotents in R. We have to
show there exists a prime p such that e1 p and e2 6 p, or conversely. Write
e0i = 1 ei . If e1 6= e2 , then 0 6= e1 e2 = e1 (e2 + e02 ) (e1 + e01 )e2 = e1 e02 e01 e2 .
Hence either the idempotent e1 e02 6= 0 or e01 e2 6= 0. An idempotent is not nilpotent,
and hence we find a prime p such that either e1 e02 6 p or e01 e2 6 p, by Lemma 16.2.
It is easy to see this gives the desired prime.


00EF

Lemma 20.4. Let R be a nonzero ring. Then Spec(R) is connected if and only if
R has no nontrivial idempotents.
Proof. Obvious from Lemma 20.3.

00EH

Lemma 20.5. Let R be a ring. Let I be a finitely generated ideal. Assume

that I = I 2 . Then V (I) is open and closed in Spec(R), and R/I
= Re for some
idempotent e R.

COMMUTATIVE ALGEBRA

40

Proof. By Nakayamas Lemma 19.1 there exists an element f = 1 + i, i I in

R such that f I = 0. It follows that V (I) = D(f ) by a simple argument. Also,
0 = f i = i + i2 , and hence f 2 = 1 + i + i + i2 = 1 + i = f , so f is an idempotent.
Consider the canonical map R Rf . It is surjective since x/f n = x/f = xf /f 2 =
xf /f = x/1 in Rf . Any element of I is in the kernel since f I = 0. If x 7 0 in Rf ,
then f n x = 0 for some n > 0 and hence (1 + i)x = 0 hence x I.

21. Connected components of spectra
00EB

Connected components of spectra are not as easy to understand as one may think
at first. This is because we are used to the topology of locally connected spaces,
but the spectrum of a ring is in general not locally connected.

04PP

Lemma 21.1. Let R be a ring. Let T Spec(R) be a subset of the spectrum. The
following are equivalent
(1) T is closed and is a union of connected components of Spec(R),
(2) T is an intersection of open and closed subsets of Spec(R), and
(3) T = V (I) where I R is an ideal generated by idempotents.
Moreover, the ideal in (3) if it exists is unique.
Proof. By Lemma 16.11 and Topology,TLemma 11.12 we see that (1) and (2) are
equivalent. Assume (2) and write T = U with U Spec(R) open and closed.
Then U = D(e ) for some idempotent e R by Lemma 20.3. Then setting
I = (1 e ) we see that T = V (I), i.e., (3) holds. Finally, assume (3). Write
T =V
= (e ) for some collection of idempotents e . Then it is clear that
T(I) and I T
T = V (e ) = D(1 e ).
Suppose that I is an ideal generated by idempotents. Let e R be an idempotent
such that V (I) V (e). Then by Lemma 16.2 we see that en I for some n 1.
As e is an idempotent this means that e I. Hence we see that I is generated
by exactly those idempotents e such that T V (e). In other words, the ideal I is
completely determined by the closed subset T which proves uniqueness.


00EG

Lemma 21.2. Let R be a ring. A connected component of Spec(R) is of the form

V (I), where I is an ideal generated by idempotents such that every idempotent of
R either maps to 0 or 1 in R/I.
Proof. Let p be a prime of R. By Lemma 16.11 we have see that the hypotheses
of Topology, Lemma 11.10 are satisfied for the topological space Spec(R). Hence
the connected component of p in Spec(R) is the intersection of open and closed
subsets of Spec(R) containing p. Hence it equals V (I) where I is generated by the
idempotents e R such that e maps to 0 in (p), see Lemma 20.3. Any idempotent
e which is not in this collection clearly maps to 1 in R/I.

22. Glueing functions

00EI

In this section we show that given an open covering

[n
Spec(R) =
D(fi )
i=1

by standard opens, and given an element hi Rfi for each i such that hi = hj as
elements of Rfi fj then there exists a unique h R such that the image of h in Rfi
is hi . This result can be interpreted in two ways:

COMMUTATIVE ALGEBRA

41

(1) The rule D(f ) 7 Rf is a sheaf of rings on the standard opens, see Sheaves,
Section 30.
(2) If we think of elements of Rf as the algebraic or regular functions on
D(f ), then these glue as would continuous, resp. differentiable functions on
a topological, resp. differentiable manifold.
At the end of this section we use this result to reprove the lemma describing open
and closed subsets in terms of idempotents.
00EJ

Lemma 22.1. Let R be a ring, and let f1 , f2 , . . . fn R generate the unit ideal in
R. Then the following sequence is exact:
M
M
0 R
Rfi
Rfi fj
i
i,j
L
L
L
where the maps : R i Rfi and : i Rfi i,j Rfi fj are defined as
!


x
x
x1
xn
xi
xj
,...,
and
, . . . , rn =
rj in Rfi fj .
(x) =
1
1
f1r1
fn
firi
fj
Proof. We first show that is injective, and then that the image of equals the
kernel of . Assume there exists x R such that (x) = (0, . . . , 0). Then x1 = 0 in
Rfi for all i. This means, for all i, there exists a number ni such that
fini x = 0
Since the fi generate R, we can pick ai so
Xn
1=
ai fi
i=1
P
Then for all M
ni , we have

X
M X 
M
ai fi
=
1M =
au1 au2 aunn f1u1 f2u2 fnun
u1 , . . . , un 1 2
where each term has a factor of at least fini for some i. Therefore,
X
M
ai fi
x = 1x = 1M x =
x = 0.
Thus, if (x) = 0, x = 0 and is injective. We check that the image of equals
the kernel of . First, note that for x R,
x
x
x x
,...,
((x)) =
= ( in Rfi fj ) = 0.
1
1
1
1
Therefore, the image of is in the kernel of , and it remains only to verify that if


xn
x1
,
.
.
.
,

= 0,
f1r1
fnrn
then there exists x R so that for all i,
x
xi
= ri
1
fi
Assume we have x1 , . . . , xn such that


x1
xn

,
.
.
.
,
= 0.
f1r1
fnrn
Then, for all pairs i, j, there exists an nij such that
nij

fi

fj ij (fj j xi firi xj ) = 0

COMMUTATIVE ALGEBRA

42

Choosing N so N nij for all i, j, we see that

r

fiN fjN (fj j xi firi xj ) = 0

Define elements xei and fei of R as follows:
fei = fiN +ri , xei = fiN xi .
Notice that

xei
xi
= ri .
fi
fei

Also, we can use this to rewrite the above equation fiN fjN (fj j xi firi xj ) = 0 to
get the following equality, for all i, j,
fej xei = fei xej .
Since f1 , . . . , fn generate R, we clearly have that fe1 , . . . , ff
n also generate R. Therefore, there exist a1 , . . . , an in R so that
Xn
1=
ai fei
i=1

Therefore, we finally conclude that for all i,

Pn
Xn aj fej xei
Xn aj fei xej
ej
xi
xei
j=1 aj x
=
.
=
=
=
ri
j=1
j=1
fi
1
fei
fei
fei
Thus, we have

 Xn
 x
xn
1

aj xej =
, . . . , rn ,
j=1
f1r1
fn
as required. There the sequence is exact.
00EK

Lemma 22.2. Let R be a ring. Let f1 , . . . , fn be elements of R generating the

unit ideal. Let M be an R-module. The sequence
Mn
Mn

0M

Mfi

Mfi fj
i=1

is exact, where (m) = (m/1, . . . , m/1) and

e
mj /fj j )(i,j) .

i,j=1
(m1 /f1e1 , . . . , mn /fnen )

= (mi /fiei

Proof. The same as the proof of Lemma 22.1.

Second proof of Lemma 20.3. Having assured ourselves (Lemma 22.1) that for
generators f1 , . . . , fn for the unit ideal of a ring R the sequence
Mn
M
0R
Rfi
Rfi fj
i=1

00EL

i,j

is exact, we now provide an alternate proof of the surjectivity of the map from
idempotents e of R to open and closed subsets of Spec(R) presented in Lemma 20.3.
Let U Spec(R) be open and closed, and W be S
its complement. We S
can write U
n
m
and V as unions of standard opens such that U = i=1 D(fi ) and W = j=1 D(gj ).
S
S
Since Spec(R) = D(fi ) D(gj ), we observe that the collection {fi ; gj } must
generate the unit ideal in R by Lemma 16.2. So the following sequence is exact.
(22.2.1)
Mn
Mm
M
M
M

0R
Rfi
Rgj
Rfi1 fi2
Rfi gj
Rgj1 gj2
i=1

j=1

i1 ,i2

i,j

j1 ,j2

However, notice that for any pair i, j, D(fi ) D(gj ) = since D(fi ) U and
D(gj ) W ). From part (15) of Lemma 16.2 we recall that D(fi gj ) = D(fi )

COMMUTATIVE ALGEBRA

43

D(gj ) = . Therefore by Lemma 16.5 Spec(Rfi gj ) = D(fi gj ) = , implying that

Rfi gj is the zero ring for each pair i, j by part (3) of Lemma 16.2. Consider the
Ln
Lm
element (1, . . . , 1, 0, . . . , 0) i=1 Rfi j=1 Rgj whose coordinates are 1 in each
Rfi and 0 in each Rgj . This is sent to 0 under the map
M
Mn
Mm
M
Rgj1 gj2
Rfi1 fi2
:
Rfi
Rgj
i=1

j=1

i1 ,i2

j1 ,j2

so by the exactness of the sequence (22.2.1), there must be some element of R

whose image under is (1, . . . , 1, 0, . . . , 0). Call it e. We see that (e2 ) = (e)2 =
(1, . . . , 1, 0, . . . , 0) = (e). Since is injective, e = e2 in R and e is an idempotent
of R. We claim that U = D(e). Notice that for arbitrary j, the map R Rgj maps
k
e to 0. Therefore there must be some positive integer kj such that gj j (e 0) = 0 in
kj
R. Multiplying by e as necessary, we see that (gj e) = 0, so
Sgj e is nilpotent in R.
By Lemma 16.2 D(gj ) D(e) = D(gj e) = . So since V = D(gj ), D(e) V =
and D(e) U . Furthermore, for arbitrary i, the map R Rfi maps e to 1, so
there must be some li such that fili (e 1) = 0 in R. Hence fili e = fili . Suppose
p Spec(R) contains e, then p contains fili e = fili , and since S
p is prime, fi p. So
V (e) V (fi ), implying that D(fi ) D(e). Therefore U = D(fi ) D(e), and
U = D(e). Therefore any open and closed subset of Spec(R) is the standard open
of an idempotent as desired.

The following we have already seen above, but we state it explicitly here for convenience.
00EM

Lemma 22.3. Let R be a ring. If Spec(R) = U q V with both U and V open then
R
= R1 R2 with U
= Spec(R1 ) and V
= Spec(R2 ) via the maps in Lemma 20.2.
Moreover, both R1 and R2 are localizations as well as quotients of the ring R.
Proof. By Lemma 20.3 we have U = D(e) and V = D(1 e) for some idempotent
e. By Lemma 22.1 we see that R
= Re R1e (since clearly Re(1e) = 0 so the
glueing condition is trivial; of course it is trivial to prove the product decomposition
directly in this case). The lemma follows.


0565

Lemma
L 22.4. Let R be a ring. Let f1 , . . . , fn R. Let M be an R-module. Then
M
Mfi is injective if and only if
M
M
M, m 7 (f1 m, . . . , fn m)
i=1,...,n

is injective.
L
Proof. The map M
Mfi is injective if and only if for all m M and
e1 , . . . , en 1 such that fiei m = 0, i = 1, . . . , n we have m = 0. This clearly
implies the displayed map is injective. Conversely, suppose the displayed map is
injective and m M and e1 , . . . , en 1 are such that fiei m = 0, i = 1, . . . , n. If
ei = 1 for all i, then we immediately conclude that m = 0 from the injectivity of
the displayed
map. Next, we prove this holds for any such data by induction on
P
e=
ei . The base case is e = n, and we have just dealt with this. If some ei > 1,
then set m0 = fi m. By induction we see that m0 = 0. Hence we see that fi m = 0,
i.e., we may take ei = 1 which decreases e and we win.


COMMUTATIVE ALGEBRA

44

23. More glueing results

00EN

In this section we put a number of standard results of the form: if something is

true for all members of a standard open covering then it is true. In fact, it often
suffices to check things on the level of local rings as in the following lemma.

00HN

Lemma 23.1. Let R be a ring.

(1) For an element x of an R-module M the following are equivalent
(a) x = 0,
(b) x maps to zero in Mp for all p Spec(R),
(c) x maps to zero in Mm for Q
all maximal ideals m of R.
In other words, the map M m Mm is injective.
(2) Given an R-module M the following are equivalent
(a) M is zero,
(b) Mp is zero for all p Spec(R),
(c) Mm is zero for all maximal ideals m of R.
(3) Given a complex M1 M2 M3 of R-modules the following are equivalent
(a) M1 M2 M3 is exact,
(b) for every prime p of R the localization M1,p M2,p M3,p is exact,
(c) for every maximal ideal m of R the localization M1,m M2,m M3,m
is exact.
(4) Given a map f : M M 0 of R-modules the following are equivalent
(a) f is injective,
(b) fp : Mp Mp0 is injective for all primes p of R,
0
(c) fm : Mm Mm
is injective for all maximal ideals m of R.
(5) Given a map f : M M 0 of R-modules the following are equivalent
(a) f is surjective,
(b) fp : Mp Mp0 is surjective for all primes p of R,
0
(c) fm : Mm Mm
is surjective for all maximal ideals m of R.
(6) Given a map f : M M 0 of R-modules the following are equivalent
(a) f is bijective,
(b) fp : Mp Mp0 is bijective for all primes p of R,
0
(c) fm : Mm Mm
is bijective for all maximal ideals m of R.
Proof. Let x M as in (1). Let I = {f R | f x = 0}. It is easy to see that I
is an ideal (it is the annihilator of x). Condition (1)(c) means that for all maximal
ideals m there exists an f R \ m such that f x = 0. In other words, V (I) does not
contain a closed point. By Lemma 16.2 we see I is the unit ideal. Hence x is zero,
i.e., (1)(a) holds. This proves (1).
Part (2) follows by applying (1) to all elements of M simultaneously.
Proof of (3). Let H be the homology of the sequence, i.e., H = Ker(M2
M3 )/Im(M1 M2 ). By Proposition 9.12 we have that Hp is the homology of
the sequence M1,p M2,p M3,p . Hence (3) is a consequence of (2).
Parts (4) and (5) are special cases of (3). Part (6) follows formally on combining
(4) and (5).


00EO

Lemma 23.2. Let R be a ring. Let M be an R-module. Let S

S be an R-algebra.
Suppose that f1 , . . . , fn is a finite list of elements of R such that D(fi ) = Spec(R)
in other words (f1 , . . . , fn ) = R.

COMMUTATIVE ALGEBRA

45

(1) If each Mfi = 0 then M = 0.

(2) If each Mfi is a finite Rfi -module, then M is a finite R-module.
(3) If each Mfi is a finitely presented Rfi -module, then M is a finitely presented
R-module.
(4) Let M N be a map of R-modules. If Mfi Nfi is an isomorphism for
each i then M N is an isomorphism.
(5) Let 0 M 00 M M 0 0 be a complex of R-module. If 0 Mf00i
Mfi Mf0 i 0 is exact for each i, then 0 M 00 M M 0 0 is
exact.
(6) If each Rfi is Noetherian, then R is Noetherian.
(7) If each Sfi is a finite type R-algebra, so is S.
(8) If each Sfi is of finite presentation over R, so is S.
Proof. We prove each of the parts in turn.
(1) By Proposition 9.10 this implies Mp = 0 for all p Spec(R), so we conclude
by Lemma 23.1.
(2) For each i take a finite generating set Xi of Mfi . Without loss of generality,
we may assume that the elements of Xi are in the image of the localization
map M Mfi , so we take a finite set Yi of preimages of the elements of Xi
in M . Let Y be the union of these sets. This is still a finite set. Consider
the obvious R-linear map RY M sending the basis element ey to y.
By assumption this map is surjective after localizing at an arbitrary prime
ideal p of R, so it surjective by Lemma 23.1 and M is finitely generated.
(3) By (2) we have a short exact sequence
0 K Rn M 0

(4)

(5)

(6)

(7)

(8)

Since localization is an exact functor and Mfi is finitely presented we see

that Kfi is finitely generated for all 1 i n by Lemma 5.3. By (2) this
implies that K is a finite R-module and therefore M is finitely presented.
By Proposition 9.10 the assumption implies that the induced morphism
on localizations at all prime ideals is an isomorphism, so we conclude by
Lemma 23.1.
By Proposition 9.10 the assumption implies that the induced sequence of
localizations at all prime ideals is short exact, so we conclude by Lemma
23.1.
We will show that every ideal of R has a finite generating set: For this, let
I R be an arbitrary ideal. By Proposition 9.12 each Ifi Rfi is an ideal.
These are all finitely generated by assumption, so we conclude by (2).
For each i take a finite generating set Xi of Sfi . Without loss of generality,
we may assume that the elements of Xi are in the image of the localization
map S Sfi , so we take a finite set Yi of preimages of the elements of Xi
in S. Let Y be the union of these sets. This is still a finite set. Consider
the algebra homomorphism R[Xy ]yY S induced by Y . Since it is an
algebra homomorphism, the image T is an R-submodule of the R-module
S, so we can consider the quotient module S/T . By assumption, this is zero
if we localize at the fi , so it is zero by (1) and therefore S is an R-algebra
of finite type.
By the previous item, there exists a surjective R-algebra homomorphism
R[X1 , ..., Xn ] S. Let K be the kernel of this map. This is an ideal in

COMMUTATIVE ALGEBRA

46

R[X1 , ..Xn ], finitely generated in each localization at fi . Since the fi generate the unit ideal in R, they also generate the unit ideal in R[X1 , ..., Xn ],
so an application of (2) finishes the proof.

00EP

Lemma 23.3. Let R SSbe a ring map. Suppose that g1 , . . . , gm is a finite list
of elements of S such that D(gj ) = Spec(S) in other words (g1 , . . . , gm ) = S.
(1) If each Sgi is of finite type over R, then S is of finite type over R.
(2) If each Sgi is of finite presentation over R, then S is of finite presentation
over R.
Proof. Omitted.

The following lemma is better stated and proved in the more general context of flat
descent. However, it makes sense to state it here since it fits well with the above.
00EQ

Lemma 23.4. Let R be a ring. Let f1 , . . . , fn R be elements which generate the

unit ideal in R. Suppose we are given the following data:
(1) For each i an Rfi -module Mi .
(2) For each pair i, j an Rfi fj -module isomorphism ij : (Mi )fj (Mj )fi .
which satisfy the cocycle condition that all the diagrams
/ (Mk )fi fj
9

ik

(Mi )fj fk
ij

%
(Mj )fi fk

jk

commute (for all triples i, j, k). Given this data define

M
M
M = Ker
Mi
1in

1i,jn

(Mi )fi

where (m1 , . . . , mn ) maps to the element whose (i, j)th entry is mi /1 ji (mj /1).
Then the natural map M Mi identifies Mi with Mfi . Moreover ij (m/1) = m/1
for all m M (with obvious notation).
Proof. Omitted.

24. Zerodivisors and total rings of fractions

02LV

The local ring at a minimal prime has the following properties.

00EU

Lemma 24.1. Let p be a minimal prime of a ring R. Every element of the

maximal ideal of Rp is nilpotent. If R is reduced then Rp is a field.
Proof. If some element x of pRp is not nilpotent, then D(x) 6= , see Lemma 16.2.
This contradicts the minimality of p. If R is reduced, then pRp = 0 and hence it is
a field.


00EW

Lemma 24.2. Let R be a reduced ring. Then

(1) R is aQsubring of a product of fields,
(2) R p minimal Rp is an embedding into a product of fields,
S
(3) p minimal p is the set of zerodivisors of R.

COMMUTATIVE ALGEBRA

47

Proof. By Lemma 24.1 each of the rings Rp is a field.

T In particular, the kernel of
the ring map R Rp is p. By Lemma 16.2 we have p p = (0). Hence (2) and (1)
are true. If xy = 0 and y 6= 0, then y 6 p forSsome minimal prime p. Hence x p.
Thus every zerodivisor of R is contained in p minimal p. Conversely, suppose that
x p for some minimal prime p. Then x maps to zero in Rp , hence there exists
y R, y 6 p such that xy = 0. In other words, x is a zerodivisor. This finishes the
proof of (3) and the lemma.

The total ring of fractions Q(R) of a ring R was introduced in Example 9.8.
02LW

Lemma 24.3. Let R be a ring. Let S R be a multiplicative subset consisting of

nonzerodivisors. Then Q(R)
= Q(S 1 R). In particular Q(R)
= Q(Q(R)).
Proof. If x S 1 R is a nonzerodivisor, and x = r/f for some r R, f S, then
r is a nonzerodivisor in R. Whence the lemma.

We can apply glueing results to prove something about total rings of fractions Q(R)
which we introduced in Example 9.8.

02LX

Lemma 24.4. Let R be a ring. Assume that R has finitely many minimal primes
q1 , . . . , qt , and that q1 . . . qt is the set of zerodivisors of R. Then the total ring
of fractions Q(R) is equal to Rq1 . . . Rqt .
Proof. There are natural maps Q(R) Rqi since any nonzerodivisor is contained
in R \ qi . Hence a natural map Q(R) Rq1 . . . Rqt . For any nonminimal prime
p R we see that p 6 q1 . . .qt by Lemma 14.2. Hence Spec(Q(R)) = {q1 , . . . , qt }
(as subsets of Spec(R), see Lemma 16.5). Therefore Spec(Q(R)) is a finite discrete
set and it follows that Q(R) = A1 . . . At with Spec(Ai ) = {qi }, see Lemma
22.3. Moreover Ai is a local ring, which is a localization of R. Hence Ai
= Rqi . 
25. Irreducible components of spectra

00ER

We show that irreducible components of the spectrum of a ring correspond to the

minimal primes in the ring.

00ES

Lemma 25.1. Let R be a ring.

(1) For a prime p R the closure of {p} in the Zariski topology is V (p). In a
formula {p} = V (p).
(2) The irreducible closed subsets of Spec(R) are exactly the subsets V (p), with
p R a prime.
(3) The irreducible components (see Topology, Definition 7.1) of Spec(R) are
exactly the subsets V (p), with p R a minimal prime.
Proof. Note that if p V (I), then I p. Hence, clearly {p} = V (p). In particular
V (p) is the closure of a singleton and hence irreducible. The second assertion implies
the third. To show the second, let V (I) Spec(R) with I a radical ideal. If I is not
prime, then choose a, b R, a, b 6 I with ab I. In this case V (I, a) V (I, b) =
V (I), but neither V (I, b) = V (I) nor V (I, a) = V (I), by Lemma 16.2. Hence V (I)
is not irreducible.

In other words, this lemma shows that every irreducible closed subset of Spec(R) is
of the form V (p) for some prime p. Since V (p) = {p} we see that each irreducible
closed subset has a unique generic point, see Topology, Definition 7.4. In particular,
Spec(R) is a sober topological space. We record this fact in the following lemma.

COMMUTATIVE ALGEBRA

090M

48

Lemma 25.2. The spectrum of a ring is a spectral space, see Topology, Definition
22.1.
Proof. Formally this follows from Lemma 25.1 and Lemma 16.11. See also discussion above.


00ET

Lemma 25.3. Let R be a ring. Let p R be a prime.

(1) the set of irreducible closed subsets of Spec(R) passing through p is in oneto-one correspondence with primes q Rp .
(2) The set of irreducible components of Spec(R) passing through p is in oneto-one correspondence with minimal primes q Rp .
Proof. Follows from Lemma 25.1 and the description of Spec(Rp ) in Lemma 16.5
which shows that Spec(Rp ) corresponds to primes q in R with q p.


00EV

Lemma 25.4. Let R be a ring. Let p be a minimal prime of R. Let W Spec(R)

be a quasi-compact open not containing the point p. Then there exists an f R,
f 6 p such that D(f ) W = .
Proof. Since W is quasi-compact we may write it as a finite union of standard
affine opens D(gi ), i = 1, . . . , n. Since p 6 W we have gi p for all i. By Lemma
24.1 each gi is nilpotent in Rp . Hence we can find an f R, f 6 p such that for all
i we have f gini = 0 for some ni > 0. Then D(f ) works.


04MG

Lemma 25.5. Let R be a ring. Let X = Spec(R) as a topological space. The

following are equivalent
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)

X is profinite,
X is Hausdorff,
X is totally disconnected.
every quasi-compact open of X is closed,
there are no nontrivial inclusions between its prime ideals,
every prime ideal is a maximal ideal,
every prime ideal is minimal,
every standard open D(f ) X is closed, and
add more here.

Proof. First proof. It is clear that (5), (6), and (7) are equivalent. It is clear that
(4) and (8) are equivalent as every quasi-compact open is a finite union of standard
opens. The implication (7) (4) follows from Lemma 25.4. Assume (4) holds. Let
p, p0 be distinct primes of R. Choose an f p0 , f 6 p (if needed switch p with p0 ).
Then p0 6 D(f ) and p D(f ). By (4) the open D(f ) is also closed. Hence p and
p0 are in disjoint open neighbourhoods whose union is X. Thus X is Hausdorff and
totally disconnected. Thus (4) (2) and (3). If (3) holds then there cannot be
any specializations between points of Spec(R) and we see that (5) holds. If X is
Hausdorff then every point is closed, so (2) implies (6). Thus (2), (3), (4), (5), (6),
(7) and (8) are equivalent. Any profinite space is Hausdorff, so (1) implies (2). If
X satisfies (2) and (3), then X (being quasi-compact by Lemma 16.10) is profinite
by Topology, Lemma 21.2.
Second proof. Besides the equivalence of (4) and (8) this follows from Lemma 25.2
and purely topological facts, see Topology, Lemma 22.7.


COMMUTATIVE ALGEBRA

49

26. Examples of spectra of rings

00EX

In this section we put some examples of spectra.

00EY

Example 26.1. In this example we describe X = Spec(Z[x]/(x2 4)). Let p be

an arbitrary prime in X. Let : Z Z[x]/(x2 4) be the natural ring map. Then,
1 (p) is a prime in Z. If 1 (p) = (2), then since p contains 2, it corresponds to
a prime ideal in Z[x]/(x2 4, 2)
= (Z/2Z)[x]/(x2 ) via the map Z[x]/(x2 4)
2
Z[x]/(x 4, 2). Any prime in (Z/2Z)[x]/(x2 ) corresponds to a prime in (Z/2Z)[x]
containing (x2 ). Such primes will then contain x. Since (Z/2Z)
= (Z/2Z)[x]/(x)
is a field, (x) is a maximal ideal. Since any prime contains (x) and (x) is maximal,
the ring contains only one prime (x). Thus, in this case, p = (2, x). Now, if
1 (p) = (q) for q > 2, then since p contains q, it corresponds to a prime ideal in
Z[x]/(x2 4, q)
= (Z/qZ)[x]/(x2 4) via the map Z[x]/(x2 4) Z[x]/(x2 4, q).
Any prime in (Z/qZ)[x]/(x2 4) corresponds to a prime in (Z/qZ)[x] containing
(x2 4) = (x 2)(x + 2). Hence, these primes must contain either x 2 or x + 2.
Since (Z/qZ)[x] is a PID, all nonzero primes are maximal, and so there are precisely
2 primes in (Z/qZ)[x] containing (x 2)(x + 2), namely (x 2) and (x + 2). In
conclusion, there exist two primes (q, x 2) and (q, x + 2) since 2 6= 2 Z/(q).
Finally, we treat the case where 1 (p) = (0). Notice that p corresponds to a
prime ideal in Z[x] that contains (x2 4) = (x 2)(x + 2). Hence, p contains
either (x 2) or (x + 2). Hence, p corresponds to a prime in Z[x]/(x 2) or one
in Z[x]/(x + 2) that intersects Z only at 0, by assumption. Since Z[x]/(x 2)
=Z
and Z[x]/(x + 2)
= Z, this means that p must correspond to 0 in one of these rings.
Thus, p = (x 2) or p = (x + 2) in the original ring.

00EZ

Example 26.2. In this example we describe X = Spec(Z[x]). Fix p X. Let

: Z Z[x] and notice that 1 (p) Spec(Z). If 1 (p) = (q) for q a prime
number q > 0, then p corresponds to a prime in (Z/(q))[x], which must be generated
by a polynomial that is irreducible in (Z/(q))[x]. If we choose a representative
of this polynomial with minimal degree, then it will also be irreducible in Z[x].
Hence, in this case p = (q, fq ) where fq is an irreducible polynomial in Z[x] that
is irreducible when viewed in (Z/(q)[x]). Now, assume that 1 (p) = (0). In this
case, p must be generated by nonconstant polynomials which, since p is prime, may
be assumed to be irreducible in Z[x]. By Gauss lemma, these polynomials are also
irreducible in Q[x]. Since Q[x] is a Euclidean domain, if there are at least two
distinct irreducibles f, g generating p, then 1 = af + bg for a, b Q[x]. Multiplying
through by a common denominator, we see that m = a
f + bg for a
, b Z[x] and
nonzero m Z. This is a contradiction. Hence, p is generated by one irreducible
polynomial in Z[x].

00F0

Example 26.3. In this example we describe X = Spec(k[x, y]) when k is an arbitrary field. Clearly (0) is prime, and any principal ideal generated by an irreducible
polynomial will also be a prime since k[x, y] is a unique factorization domain. Now
assume p is an element of X that is not principal. Since k[x, y] is a Noetherian UFD,
the prime ideal p can be generated by a finite number of irreducible polynomials
(f1 , . . . , fn ). Now, I claim that if f, g are irreducible polynomials in k[x, y] that are
not associates, then (f, g) k[x] 6= 0. To do this, it is enough to show that f and
g are relatively prime when viewed in k(x)[y]. In this case, k(x)[y] is a Euclidean
domain, so by applying the Euclidean algorithm and clearing denominators, we
obtain p = af + bg for p, a, b k[x]. Thus, assume this is not the case, that is,

COMMUTATIVE ALGEBRA

50

that some nonunit h k(x)[y] divides both f and g. Then, by Gausss lemma, for
some a, b k(x) we have ah|f and bh|g for ah, bh k[x] since f.f.(k[x]) = k(x).
By irreducibility, ah = f and bh = g (since h
/ k(x)). So, back in k(x)[y], f, g are
associates, as ab g = f . Since k(x) = f.f.(k[x]), we can write g = rs f for elements
r, s k[x] sharing no common factors. This implies that sg = rf in k[x, y] and so
s must divide f since k[x, y] is a UFD. Hence, s = 1 or s = f . If s = f , then r = g,
implying f, g k[x] and thus must be units in k(x) and relatively prime in k(x)[y],
contradicting our hypothesis. If s = 1, then g = rf , another contradiction. Thus,
we must have f, g relatively prime in k(x)[y], a Euclidean domain. Thus, we have
reduced to the case p contains some irreducible polynomial p k[x] k[x, y]. By
the above, p corresponds to a prime in the ring k[x, y]/(p) = k()[y], where is an
element algebraic over k with minimum polynomial p. This is a PID, and so any
prime ideal corresponds to (0) or an irreducible polynomial in k()[y]. Thus, p is
of the form (p) or (p, f ) where f is a polynomial in k[x, y] that is irreducible in the
quotient k[x, y]/(p).
00F1

Example 26.4. Consider the ring

R = {f Q[z] with f (0) = f (1)}.
Consider the map
: Q[A, B] R
2

defined by (A) = z z and (B) = z 3 z 2 . It is easily checked that (A3 B 2 +

AB) Ker() and that A3 B 2 + AB is irreducible. Assume that is surjective;
then since R is an integral domain (it is a subring of an integral domain), Ker()
must be a prime ideal of Q[A, B]. The prime ideals which contain (A3 B 2 + AB)
are (A3 B 2 +AB) itself and any maximal ideal (f, g) with f, g Q[A, B] such that
f is irreducible mod g. But R is not a field, so the kernel must be (A3 B 2 + AB);
hence gives an isomorphism R Q[A, B]/(A3 B 2 + AB).
To see that is surjective, we must express any f R as a Q-coefficient polynomial
in A(z) = z 2 z and B(z) = z 3 z 2 . Note the relation zA(z) = B(z). Let
a = f (0) = f (1). Then z(z 1) must divide f (z) a, so we can write f (z) =
z(z 1)g(z) + a = A(z)g(z) + a. If deg(g) < 2, then h(z) = c1 z + c0 and f (z) =
A(z)(c1 z + c0 ) + a = c1 B(z) + c0 A(z) + a, so we are done. If deg(g) 2, then
by the polynomial division algorithm, we can write g(z) = A(z)h(z) + b1 z + b0
(deg(h) deg(g) 2), so f (z) = A(z)2 h(z) + b1 B(z) + b0 A(z). Applying division
to h(z) and iterating, we obtain an expression for f (z) as a polynomial in A(z) and
B(z); hence is surjective.
Now let a Q, a 6= 0, 21 , 1 and consider
1
] with f (0) = f (1)}.
za
This is a finitely generated Q-algebra as well: it is easy to check that the functions
2
a
z 2 z, z 3 z, and aza
+ z generate Ra as an Q-algebra. We have the following
inclusions:
1
1
R Ra Q[z,
], R Q[z] Q[z,
].
za
za
Recall (Lemma 16.5) that for a ring T and a multiplicative subset S T , the
ring map T S 1 T induces a map on spectra Spec(S 1 T ) Spec(T ) which is a
Ra = {f Q[z,

COMMUTATIVE ALGEBRA

51

homeomorphism onto the subset

{p Spec(T ) | S p = } Spec(T ).
2

When S = {1, f, f , . . .} for some f T , this is the open set D(f ) T . We now
verify a corresponding property for the ring map R Ra : we will show that the
map : Spec(Ra ) Spec(R) induced by inclusion R Ra is a homeomorphism
onto an open subset of Spec(R) by verifying that is an injective local homeomorphism. We do so with respect to an open cover of Spec(Ra ) by two distinguished
opens, as we now describe. For any r Q, let evr : R Q be the homomorphism
given by evaluation at r. Note that for r = 0 and r = 1 a, this can be extended to
1
is well-defined at z = 1 a,
a homomorphism ev0r : Ra Q (the latter because za
1
since a 6= 2 ). However, eva does not extend to Ra . Write mr = Ker(evr ). We have
m0 = (z 2 z, z 3 z),
ma = ((z 1 + a)(z a), (z 2 1 + a)(z a)), and
m1a = ((z 1 + a)(z a), (z 1 + a)(z 2 a)).
To verify this, note that the right-hand sides are clearly contained in the left-hand
sides. Then check that the right-hand sides are maximal ideals by writing the
generators in terms of A and B, and viewing R as Q[A, B]/(A3 B 2 + AB). Note
that ma is not in the image of : we have
a2 a
+ z) = (z 2 z)2 (a2 a) + (z 2 z)2 (z a)z
za
The left hand side is in ma Ra because (z 2 z)(z a) is in ma and because (z 2
2
a
z)( aza
+ z) is in Ra . Similarly the element (z 2 z)2 (z a)z is in ma Ra because
(z 2 z) is in Ra and (z 2 z)(z a) is in ma . As a 6 {0, 1} we conclude that
(z 2 z)2 ma Ra . Hence no ideal I of Ra can satisfy I R = ma , as such an I
would have to contain (z 2 z)2 , which is in R but not in ma . The distinguished
open set D((z 1 + a)(z a)) Spec(R) is equal to the complement of the closed
set {ma , m1a }. Then check that R(z1+a)(za) = (Ra )(z1+a)(za) ; calling this
localized ring R0 , then, it follows that the map R R0 factors as R Ra R0 .
By Lemma 16.5, then, these maps express Spec(R0 ) Spec(Ra ) and Spec(R0 )
Spec(R) as open subsets; hence : Spec(Ra ) Spec(R), when restricted to D((z
1 + a)(z a)), is a homeomorphism onto an open subset. Similarly, restricted to
D((z 2 + z + 2a 2)(z a)) Spec(Ra ) is a homeomorphism onto the open subset
D((z 2 + z + 2a 2)(z a)) Spec(R). Depending on whether z 2 + z + 2a 2 is
irreducible or not over Q, this former distinguished open set has complement equal
to one or two closed points along with the closed point ma . Furthermore, the ideal
in Ra generated by the elements (z 2 + z + 2a a)(z a) and (z 1 + a)(z a)
is all of Ra , so these two distinguished open sets cover Spec(Ra ). Hence in order
to show that is a homeomorphism onto Spec(R) {ma }, it suffices to show that
these one or two points can never equal m1a . And this is indeed the case, since
1 a is a root of z 2 + z + 2a 2 if and only of a = 0 or a = 1, both of which do
not occur.
(z 2 z)2 (z a)(

Despite this homeomorphism which mimics the behavior of a localization at an

1
element of R, while Q[z, za
] is the localization of Q[z] at the maximal ideal (z a),
the ring Ra is not a localization of R: Any localization S 1 R results in more units
than the original ring R. The units of R are Q , the units of Q. In fact, it is easy

COMMUTATIVE ALGEBRA

52

1
to see that the units of Ra are Q . Namely, the units of Q[z, za
] are c(z a)n for

c Q and n Z and it is clear that these are in Ra only if n = 0. Hence Ra has

no more units than R does, and thus cannot be a localization of R.
1
makes sense at z = 0, 1. We
We used the fact that a 6= 0, 1 to ensure that za
used the fact that a 6= 1/2 in a few places: (1) In order to be able to talk about
the kernel of ev1a on Ra , which ensures that m1a is a point of Ra (i.e., that Ra
is missing just one point of R). (2) At the end in order to conclude that (z a)k+`
can only be in R for k = ` = 0; indeed, if a = 1/2, then this is in R as long as k + `
is even. Hence there would indeed be more units in Ra than in R, and Ra could
possibly be a localization of R.

27. A meta-observation about prime ideals

05K7

This section is taken from the CRing project. Let R be a ring and let S R
be a multiplicative subset. A consequence of Lemma 16.5 is that an ideal I R
maximal with respect to the property of not intersecting S is prime. The reason
is that I = R m for some maximal ideal m of the ring S 1 R. It turns out that
for many properties of ideals, the maximal ones are prime. A general method of
seeing this was developed in [LR08]. In this section, we digress to explain this
phenomenon.
Let R be a ring. If I is an ideal of R and a R, we define
(I : a) = {x R | xa I} .
More generally, if J R is an ideal, we define
(I : J) = {x R | xJ I} .

05K8

Lemma 27.1. Let R be a ring. For a principal ideal J R, and for any ideal
I J we have I = J(I : J).
Proof. Say J = (a). Then (I : J) = (I : a). Since I J we see that any y I
is of the form y = xa for some x (I : a). Hence I J(I : J). Conversely, if
x (I : a), then xJ = (xa) I, which proves the other inclusion.

Let F be a collection of ideals of R. We are interested in conditions that will
guarantee that the maximal elements in the complement of F are prime.

05K9

Definition 27.2. Let R be a ring. Let F be a set of ideals of R. We say F is

an Oka family if R F and whenever I R is an ideal and (I : a), (I, a) F for
some a R, then I F.
Let us give some examples of Oka families. The first example is the basic example
discussed in the introduction to this section.

05KA

Example 27.3. Let R be a ring and let S be a multiplicative subset of R. We

claim that F = {I R | I S 6= } is an Oka family. Namely, suppose that
(I : a), (I, a) F for some a R. Then pick s (I, a) S and s0 (I : a) S.
Then ss0 I S and hence I F. Thus F is an Oka family.

05KB

Example 27.4. Let R be a ring, I R an ideal, and a R. If (I : a) is generated

by a1 , . . . , an and (I, a) is generated by a, b1 , . . . , bm with b1 , . . . , bm I, then I is
generated by aa1 , . . . , aan , b1 , . . . , bm . To see this, note that if x I, then x (I, a)
is a linear combination of a, b1 , . . . , bm , but the coefficient of a must lie in (I : a).
As a result, we deduce that the family of finitely generated ideals is an Oka family.

COMMUTATIVE ALGEBRA

53

05KC

Example 27.5. Let us show that the family of principal ideals of a ring R is an
Oka family. Indeed, suppose I R is an ideal, a R, and (I, a) and (I : a) are
principal. Note that (I : a) = (I : (I, a)). Setting J = (I, a), we find that J is
principal and (I : J) is too. By Lemma 27.1 we have I = J(I : J). Thus we find in
our situation that since J = (I, a) and (I : J) are principal, I is principal.

05KD

Example 27.6. Let R be a ring. Let be an infinite cardinal. The family of ideals
which can be generated by at most elements is an Oka family. The argument is
analogous to the argument in Example 27.4 and is omitted.

05KE

Proposition 27.7. If F is an Oka family of ideals, then any maximal element of

the complement of F is prime.
Proof. Suppose I 6 F is maximal with respect to not being in F but I is not prime.
Note that I 6= R because R F. Since I is not prime we can find a, b R I
with ab I. It follows that (I, a) 6= I and (I : a) contains b 6 I so also (I : a) 6= I.
Thus (I : a), (I, a) both strictly contain I, so they must belong to F. By the Oka
condition, we have I F, a contradiction.

At this point we are able to turn most of the examples above into a lemma about
prime ideals in a ring.

05KF

Lemma 27.8. Let R be a ring. Let S be a multiplicative subset of R. An ideal

I R which is maximal with respect to the property that I S = is prime.
Proof. This is the example discussed in the introduction to this section. For an
alternative proof, combine Example 27.3 with Proposition 27.7.


05KG

Lemma 27.9. Let R be a ring.

(1) An ideal I R maximal with respect to not being finitely generated is prime.
(2) If every prime ideal of R is finitely generated, then every ideal of R is
finitely generated1.
Proof. The first assertion is an immediate consequence of Example 27.4 and Proposition 27.7. For the second, suppose that there exists an ideal I R which is not
finitely generated. The union of a totally ordered chain {IS} of ideals that are not
finitely generated is not finitely generated; indeed, if I = I were generated by
a1 , . . . , an , then all the generators would belong to some I and would consequently
generate it. By Zorns lemma, there is an ideal maximal with respect to being not
finitely generated. By the first part this ideal is prime.


05KH

Lemma 27.10. Let R be a ring.

(1) An ideal I R maximal with respect to not being principal is prime.
(2) If every prime ideal of R is principal, then every ideal of R is principal.
Proof. The first part follows from Example 27.5 and Proposition 27.7. For the
second, suppose that there exists an ideal I R which is not principal. The union
of a totally
S ordered chain {I } of ideals that not principal is not principal; indeed,
if I = I were generated by a, then a would belong to some I and a would
generate it. By Zorns lemma, there is an ideal maximal with respect to not being
principal. This ideal is necessarily prime by the first part.

1Later we will say that R is Noetherian.

COMMUTATIVE ALGEBRA

05KI

54

Lemma 27.11. Let R be a ring.

(1) An ideal maximal among the ideals which do not contain a nonzerodivisor
is prime.
(2) If every nonzero prime ideal in R contains a nonzerodivisor, then R is a
domain.
Proof. Consider the set S of nonzerodivisors. It is a multiplicative subset of R.
Hence any ideal maximal with respect to not intersecting S is prime, see Lemma
27.8. Thus, if every nonzero prime ideal contains a nonzerodivisor, then (0) is
prime, i.e., R is a domain.


05KJ

Remark 27.12. Let R be a ring. Let be an infinite cardinal. By applying

Example 27.6 and Proposition 27.7 we see that any ideal maximal with respect to
the property of not being generated by elements is prime. This result is not so
useful because there exists a ring for which every prime ideal of R can be generated
by 0 elements, but some ideal cannot. Namely, let k be a field, let T be a set
whose cardinality is greater than 0 and let
2
R = k[{xn }n1 , {zt,n }tT,n0 ]/(x2n , zt,n
, xn zt,n zt,n1 )

This is a local ring with unique prime ideal m = (xn ). But the ideal (zt,n ) cannot
be generated by countably many elements.
28. Images of ring maps of finite presentation
00F5

In this section we prove some results on the topology of maps Spec(S) Spec(R)
induced by ring maps R S, mainly Chevalleys Theorem. In order to do this
we will use the notions of constructible sets, quasi-compact sets, retrocompact sets,
and so on which are defined in Topology, Section 11.

00F6

Lemma 28.1. Let U Spec(R) be open. The following are equivalent:

(1) U is retrocompact in Spec(R),
(2) U is quasi-compact,
(3) U is a finite union of standard opens, and
(4) there exists a finitely generated ideal I R such that X \ V (I) = U .
Proof. We have (1) (2) because Spec(R) is quasi-compact, see Lemma 16.10.
We have (2) (3) because
standard opens form a basis for the topology. Proof
S
of (3) (1). Let U = i=1...n D(fi ). To show that U is retrocompact in Spec(R)
it suffices to show that
S U V is quasi-compact for any quasi-compact open V of
Spec(R). Write V = j=1...m D(gj ) which is possible by (2) (3). Each standard
open is homeomorphic to the spectrum of aSring and hence quasi-compact,
see
S
Lemmas 16.6 and 16.10. Thus U V = ( i=1...n D(fi )) ( j=1...m D(gj )) =
S
i,j D(fi gj ) is a finite union of quasi-compact opens hence quasi-compact. To
finish the proof note that (4) is equivalent to (3) by Lemma 16.2.


00F7

Lemma 28.2. Let : R S be a ring map. The induced continuous map

f : Spec(S) Spec(R) is quasi-compact. For any constructible set E Spec(R)
the inverse image f 1 (E) is constructible in Spec(S).
Proof. We first show that the inverse image of any quasi-compact open U
Spec(R) is quasi-compact. By Lemma 28.1 we may write U as a finite open of

COMMUTATIVE ALGEBRA

55

standard opens. Thus by Lemma 16.4 we see that f 1 (U ) is a finite union of standard opens. Hence f 1 (U ) is quasi-compact by Lemma 28.1 again. The second
assertion now follows from Topology, Lemma 14.3.

00F8

Lemma 28.3. Let R be a ring and let T Spec(R) be constructible. Then there
exists a ring map R S of finite presentation such that T is the image of Spec(S)
in Spec(R).
Proof. Let T Spec(R) be constructible. The spectrum of a finite product of
rings is the disjoint union of the spectra, see Lemma 20.2. Hence if T = T1 T2 and
the result holds for T1 and T2 , then the result holds for T . In particular we may
assume that T = U V c , where
U, V S
Spec(R) are retrocompact
open. By Lemma


S
S
28.1 we may write T = ( D(fi )) ( D(gj ))c =
D(fi ) V (g1 , . . . , gm ) . In
fact we may assume that T = D(f ) V (g1 , . . . , gm ) (by the argument on unions
above). In this case T is the image of the map R (R/(g1 , . . . , gm ))f , see Lemmas
16.6 and 16.7.


00F9

Lemma 28.4. Let R be a ring. Let f be an element of R. Let S = Rf . Then the

image of a constructible subset of Spec(S) is constructible in Spec(R).
Proof. We repeatedly use Lemma 28.1 without mention. Let U, V be quasicompact open in Spec(S). We will show that the image of U V c is constructible.
Under the identification Spec(S) = D(f ) of Lemma 16.6 the sets U, V correspond
to quasi-compact opens U 0 , V 0 of Spec(R). Hence it suffices to show that U 0 (V 0 )c
is constructible in Spec(R) which is clear.


00FA

Lemma 28.5. Let R be a ring. Let I be a finitely generated ideal of R. Let

S = R/I. Then the image of a constructible of Spec(S) is constructible in Spec(R).
S
Proof. If I = (f1 , . . . , fm ), then we see that V (I) is the complement of D(fi ), see
Lemma 16.2. Hence it is constructible, by Lemma 28.1. Denote the map R S
by f 7 f . We have to show that if U , V are retrocompact opens of Spec(S),
c
then the image
S of U V in Spec(R)
S is constructible. By Lemma 28.1 we may
write U = D(gi ). Setting U = D(gi ) we see U has image U V (I) which
is constructible in Spec(R). Similarly the image of V equals V V (I) for some
c
retrocompact open V of Spec(R). Hence the image of U V equals U V (I) V c
as desired.


00FB

Lemma 28.6. Let R be a ring. The map Spec(R[x]) Spec(R) is open, and the
image of any standard open is a quasi-compact open.
Proof. It suffices to show that the image of a standard open D(f ), f R[x] is
quasi-compact open. The image of D(f ) is the image of Spec(R[x]f ) Spec(R).
Let p R be a prime ideal. Let f be the image of f in (p)[x]. Recall, see Lemma
16.9, that p is in the image if and only if R[x]f R (p) = (p)[x]f is not the zero
ring. This is exactly the condition that f does not map to zero in (p)[x], in other
words, that some coefficient of f is not in p. Hence we see: if f = ad xd + . . . a0 ,
then the image of D(f ) is D(ad ) . . . D(a0 ).

We prove a property of characteristic polynomials which will be used below.

00FC

Lemma 28.7. Let R A be a ring homomorphism. Assume A

= Rn as an Rmodule. Let f A. The multiplication map mf : A A is R-linear and hence has

COMMUTATIVE ALGEBRA

56

a characteristic polynomial P (T ) = T n +rn1 T n1 +. . .+r0 R[T ]. For any prime

p Spec(R), f acts nilpotently on A R (p) if and only if p V (r0 , . . . , rn1 ).
Proof. This follows quite easily once we prove that the characteristic polynomial
P (T ) (p)[T ] of the multiplication map mf : A R (p) A R (p) which
multiplies elements of AR (p) by f, the image of f viewed in (p), is just the image
of P (T ) in (p)[T ]. Let (aij ) be the matrix of the map mf with entries in R, using
a basis e1 , . . . , en of A as an R-module. Then, A R (p)
= (R R (p))n = (p)n ,
which is an n-dimensional vector space over (p) with basis e1 1, . . . , en 1. The
image f = f 1, and so the multiplication map mf has matrix (aij 1). Thus,
the characteristic polynomial is precisely the image of P (T ).
From linear algebra, we know that a linear transformation acts nilpotently on an ndimensional vector space if and only if the characteristic polynomial is T n (since the
characteristic polynomial divides some power of the minimal polynomial). Hence,
f acts nilpotently on A R (p) if and only if P (T ) = T n . This occurs if and only
if ri p for all 0 i n 1, that is when p V (r0 , . . . , rn1 ).

00FD

Lemma 28.8. Let R be a ring. Let f, g R[x] be polynomials. Assume the leading
coefficient of g is a unit of R. There exists
S elements ri R, i = 1 . . . , n such that
the image of D(f ) V (g) in Spec(R) is i=1,...,n D(ri ).
Proof. Write g = uxd + ad1 xd1 + . . . + a0 , where d is the degree of g, and
hence u R . Consider the ring A = R[x]/(g). It is, as an R-module, finite
free with basis the images of 1, x, . . . , xd1 . Consider multiplication by (the image
of) f on A. This is an R-module map. Hence we can let P (T ) R[T ] be the
characteristic polynomial of this map. Write P (T ) = T d + rd1 T d1 + . . . + r0 . We
claim that r0 , . . . , rd1 have the desired property. We will use below the property
of characteristic polynomials that
p V (r0 , . . . , rd1 ) multiplication by f is nilpotent on A R (p).
This was proved in Lemma 28.7.
Suppose q D(f ) V (g), and let p = q R. Then there is a nonzero map
A R (p) (q) which is compatible with multiplication by f . And f acts as a
unit on (q). Thus we conclude p 6 V (r0 , . . . , rd1 ).
On the other hand, suppose that ri 6 p for some prime p of R and some 0 i d1.
Then multiplication by f is not nilpotent on the algebra A R (p). Hence there
exists a maximal ideal q A R (p) not containing the image of f . The inverse
image of q in R[x] is an element of D(f ) V (g) mapping to p.


00FE

Theorem 28.9 (Chevalleys Theorem). Suppose that R S is of finite presentation. The image of a constructible subset of Spec(S) in Spec(R) is constructible.
Proof. Write S = R[x1 , . . . , xn ]/(f1 , . . . , fm ). We may factor R S as R
R[x1 ] R[x1 , x2 ] . . . R[x1 , . . . , xn1 ] S. Hence we may assume that
S = R[x]/(f1 , . . . , fm ). In this case we factor the map as R R[x] S, and by
Lemma S
28.5 we reduce to the case S = R[x]. By Lemma 28.1 suffices to show that
if T = ( i=1...n D(fi )) V (g1 , . . . , gm ) for fi , gj R[x] then the image in Spec(R)
is constructible. Since finite unions of constructible sets are constructible, it suffices
to deal with the case n = 1, i.e., when T = D(f ) V (g1 , . . . , gm ).

COMMUTATIVE ALGEBRA

57

Note that if c R, then we have

Spec(R) = V (c) q D(c) = Spec(R/(c)) q Spec(Rc )),
and correspondingly Spec(R[x]) = V (c) q D(c) = Spec(R/(c)[x]) q Spec(Rc [x])).
The intersection of T = D(f ) V (g1 , . . . , gm ) with each part still has the same
shape, with f , gi replaced by their images in R/(c)[x], respectively Rc [x]. Note
that the image of T in Spec(R) is the union of the image of T V (c) and T D(c).
Using Lemmas 28.4 and 28.5 it suffices to prove the images of both parts are
constructible in Spec(R/(c)), respectively Spec(Rc ).
Let us assume we have T = D(f ) V (g1 , . . . , gm ) as above, with deg(g1 )
deg(g2 ) . . . deg(gm ). We are going to use descending induction on m, and
on the degrees of the gi . Let d = deg(g1 ), i.e., g1 = cxd1 + l.o.t with c R not
zero. Cutting R up into the pieces R/(c) and Rc we either lower the degree of g1
(and this is covered by induction) or we reduce to the case where c is invertible.
If c is invertible, and m > 1, then write g2 = c0 xd2 + l.o.t. In this case consider
g20 = g2 (c0 /c)xd2 d1 g1 . Since the ideals (g1 , g2 , . . . , gm ) and (g1 , g20 , g3 , . . . , gm )
are equal we see that T = D(f ) V (g1 , g20 , g3 . . . , gm ). But here the degree of g20 is
strictly less than the degree of g2 and hence this case is covered by induction.
The bases case for the induction above are the cases (a) T = D(f ) V (g) where
the leading coefficient of g is invertible, and (b) T = D(f ). These two cases are
dealt with in Lemmas 28.8 and 28.6.

29. More on images
00FF

In this section we collect a few additional lemmas concerning the image on Spec
for ring maps. See also Section 40 for example.

00FG

Lemma 29.1. Let R S be an inclusion of domains. Assume that R S is of

finite type. There exists a nonzero f R, and a nonzero g S such that Rf Sf g
is of finite presentation.
Proof. By induction on the number of generators of S over R.
Suppose that S is generated by a single element over R. Then S = R[x]/q for
some prime ideal q R[x]. If q = (0) there is nothing to prove. If q 6= (0), then
let g q be an element with minimal degree in x. Since K[x] = f.f.(R)[x] is
a PID we see that g is irreducible over K and that f.f.(S) = K[x]/(g). Write
g = ad xd + . . . + a0 with ai R and ad 6= 0. After inverting ad in R we may assume
that g is monic. Hence we see that R R[x]/(g) S with the last map surjective.
But R[x]/(g) = R Rx . . . Rxd1 maps injectively into f.f.(S) = K[x]/(g) =
K Kx . . . Kxd1 . Thus S
= R[x]/(g) is finitely presented.
Suppose that S is generated by n > 1 elements over R. Say x1 , . . . , xn S generate
S. Denote S 0 S the subring generated by x1 , . . . , xn1 . By induction hypothesis
we see that there exist f R and g S 0 nonzero such that Rf Sf0 g is of finite
presentation. Next we apply the induction hypothesis to Sf0 g Sf g to see that
there exist f 0 Sf0 g and g 0 Sf g such that Sf0 gf 0 Sf gf 0 g0 is of finite presentation.
We leave it to the reader to conclude.


00FH

Lemma 29.2. Let R S be a finite type ring map. Denote X = Spec(R) and
Y = Spec(S). Write f : Y X the induced map of spectra. Let E Y = Spec(S)

COMMUTATIVE ALGEBRA

58

be a constructible set. If a point X is in f (E), then {} f (E) contains an

open dense subset of {}.
Proof. Let X be a point of f (E). Choose a point E mapping to . Let
p R be the prime corresponding to and let q S be the prime corresponding
to . Consider the diagram
_


/ EY0

/ Y 0 = Spec(S/q)

/Y


/ f (E) X 0


/ X 0 = Spec(R/p)


/X

By Lemma 28.2 the set E Y 0 is constructible in Y 0 . It follows that we may replace

X by X 0 and Y by Y 0 . Hence we may assume that R S is an inclusion of
domains, is the generic point of X, and is the generic point of Y . By Lemma
29.1 combined with Chevalleys theorem (Theorem 28.9) we see that there exist
dense opens U X, V Y such that f (V ) U and such that f : V U maps
constructible sets to constructible sets. Note that E V is constructible in V , see
Topology, Lemma 14.4. Hence f (E V ) is constructible in U and contains . By
Topology, Lemma 14.14 we see that f (E V ) contains a dense open U 0 U . 
At the end of this section we present a few more results on images of maps on
Spectra that have nothing to do with constructible sets.
00FI

Lemma 29.3. Let : R S be a ring map. The following are equivalent:

(1) The map Spec(S) Spec(R) is surjective.
(2) For any radical ideal I R the inverse image of IS in R is equal to I.
(3) For every prime p of R the inverse image of pS in R is p.
In this case the same is true after any base change: Given a ring map R R0 the
ring map R0 R0 R S has the equivalent properties (1), (2), (3) also.
Proof. The implication (2) (3)Tis immediate. If I R is a radical ideal, then
Lemma 16.2 guarantees that I = Ip p. Hence (3) (2). By Lemma 16.9 we
have p = 1 (pS) if and only if p is in the image. Hence (1) (3). Thus (1), (2),
and (3) are equivalent.
Assume (1) holds. Let R R0 be a ring map. Let p0 R0 be a prime ideal lying
over the prime p of R. To see that p0 is in the image of Spec(R0 R S) Spec(R0 )
we have to show that (R0 R S) R0 (p0 ) is not zero, see Lemma 16.9. But we have
(R0 R S) R0 (p0 ) = S R (p) (p) (p0 )
which is not zero as S R (p) is not zero by assumption and (p) (p0 ) is an
extension of fields.


00FJ

Lemma 29.4. Let R be a domain. Let : R S be a ring map. The following

are equivalent:
(1) The ring map R S is injective.
(2) The image Spec(S) Spec(R) contains a dense set of points.
(3) There exists a prime ideal q S whose inverse image in R is (0).

COMMUTATIVE ALGEBRA

59

Proof. Let K be the field of fractions of the domain R. Assume that R S is

injective. Since localization is exact we see that K S R K is injective. Hence
there is a prime mapping to (0) by Lemma 16.9.
Note that (0) is dense in Spec(R), so that the last condition implies the second.
Suppose the second condition holds. Let f R, f 6= 0. As R is a domain we see
that V (f ) is a proper closed subset of R. By assumption there exists a prime q of
S such that (f ) 6 q. Hence (f ) 6= 0. Hence R S is injective.

00FK

Lemma 29.5. Let R S be an injective ring map. Then Spec(S) Spec(R)

hits all the minimal primes of Spec(R).
Proof. Let p R be a minimal prime. In this case Rp has a unique prime ideal.
Hence it suffices to show that Sp is not zero. And this follows from the fact that
localization is exact, see Proposition 9.12.


00FL

Lemma 29.6. Let R S be a ring map. The following are equivalent:

(1) The kernel of R S consists of nilpotent elements.
(2) The minimal primes of R are in the image of Spec(S) Spec(R).
(3) The image of Spec(S) Spec(R) is dense in Spec(R).
p
T
Proof. Let I = Ker(R S). Note that (0) = qS q, see Lemma 16.2. Hence

T
I = qS R q. Thus V (I) = V ( I) is the closure of the image of Spec(S)
Spec(R). This shows that (1) is equivalent to (3). It is clear that (2) implies (3).
Finally, assume (1). We may replace R by R/I and S by S/IS without affecting
the topology of the spectra and the map. Hence the implication (1) (2) follows
from Lemma 29.5.

30. Noetherian rings

00FM

A ring R is Noetherian if any ideal of R is finitely generated. This is clearly

equivalent to the ascending chain condition for ideals of R. By Lemma 27.9 it
suffices to check that every prime ideal of R is finitely generated.

00FN

Lemma 30.1. Any finitely generated ring over a Noetherian ring is Noetherian.
Any localization of a Noetherian ring is Noetherian.
Proof. The statement on localizations follows from the fact that any ideal J
S 1 R is of the form IS 1 R. Any quotient R/I of a Noetherian ring R is Noetherian
because any ideal J R/I is of the form J/I for some ideal I J R. Thus
it suffices to show that if R is Noetherian so is R[X]. Suppose J1 J2 . . . is
an ascending chain of ideals in R[X]. Consider the ideals Ii,d defined as the ideal
of elements of R which occur as leading coefficients of degree d polynomials in Ji .
Clearly Ii,d Ii0 ,d0 whenever i i0 and d d0 . By the ascending chain condition in
R there are at most finitely many distinct ideals among all of the Ii,d . (Hint: Any
infinite set of elements of N N contains an increasing infinite sequence.) Take i0
so large that Ii,d = Ii0 ,d for all i i0 and all d. Suppose f Ji for some i i0 .
By induction on the degree d = deg(f ) we show that f Ji0 . Namely, there exists
a g Ji0 whose degree is d and which has the same leading coefficient as f . By
induction f g Ji0 and we win.


0306

Lemma 30.2. If R is a Noetherian ring, then so is the formal power series ring
R[[x1 , . . . , xn ]].

COMMUTATIVE ALGEBRA

60

Proof. Since R[[x1 , . . . , xn+1 ]]

= R[[x1 , . . . , xn ]][[xn+1 ]] it suffices to prove the
statement that R[[x]] is Noetherian if R is Noetherian. Let I R[[x]] be a ideal.
We have to show that I is a finitely generated ideal. For each integer d denote
Id = {a R | axd + h.o.t. I}. Then we see that I0 I1 . . . stabilizes as R
is Noetherian. Choose d0 such that Id0 = Id0 +1 = . . .. For each d d0 choose
elements fd,j I (xd ), j = 1, . . . , nd such that if we write fd,j = ad,j xd + h.o.t
then Id = (ad,j ). Denote I 0 = ({fd,j }d=0,...,d0 ,j=1,...,nd ). Then it is clear that I 0 I.
Pick f I. First we may choose cd,i R such that
X
f
cd,i fd,i (xd0 +1 ) I.
Next, we can choose ci,1 R, i = 1, . . . , nd0 such that
X
X
f
cd,i fd,i
ci,1 xfd0 ,i (xd0 +2 ) I.
Next, we can choose ci,2 R, i = 1, . . . , nd0 such that
X
X
X
f
cd,i fd,i
ci,1 xfd0 ,i
ci,2 x2 fd0 ,i (xd0 +3 ) I.
And so on. In the end we see that
X
X X
f=
cd,i fd,i +
(
ci,e xe )fd0 ,i
i

is contained in I 0 as desired.

The following lemma, although easy, is useful because finite type Z-algebras come
up quite often in a technique called absolute Noetherian reduction.
00FO

Lemma 30.3. Any finite type algebra over a field is Noetherian. Any finite type
algebra over Z is Noetherian.
Proof. This is immediate from Lemma 30.1 and the fact that fields are Noetherian
rings and that Z is Noetherian ring (because it is a principal ideal domain).


00FP

Lemma 30.4. Let R be a Noetherian ring.

(1) Any finite R-module is of finite presentation.
(2) Any finite type R-algebra is of finite presentation over R.
Proof. Let M be a finite R-module. By Lemma 5.4 we can find a finite filtration of
M whose successive quotients are of the form R/I. Since any ideal is finitely generated, each of the quotients R/I is finitely presented. Hence M is finitely presented
by Lemma 5.3. This proves (1). To see (2) note that any ideal of R[x1 , . . . , xn ] is
finitely generated by Lemma 30.1.


00FQ

Lemma 30.5. If R is a Noetherian ring then Spec(R) is a Noetherian topological

space, see Topology, Definition 8.1.
Proof. This is because any closed subset of Spec(R) is uniquely of the form V (I)
with I a radical ideal, see Lemma 16.2. And this correspondence is inclusion reversing. Thus the result follows from the definitions.


00FR

Lemma 30.6. If R is a Noetherian ring then Spec(R) has finitely many irreducible
components. In other words R has finitely many minimal primes.
Proof. By Lemma 30.5 and Topology, Lemma 8.2 we see there are finitely many
irreducible components. By Lemma 25.1 these correspond to minimal primes of
R.


COMMUTATIVE ALGEBRA

045I

61

Lemma 30.7. Let k be a field and let R be a Noetherian k-algebra. If k K is

a finitely generated field extension then K k R is Noetherian.
Proof. Since K/k is a finitely generated field extension, there exists a finitely
generated k-algebra B K such that K is the fraction field of B. In other words,
K = S 1 B with S = B \ {0}. Then K k R = S 1 (B k R). Since B k R is a
finite type R-algebra (Lemma 13.2) it follows that S 1 (B k R) is Noetherian by
Lemma 30.1.

Here is fun lemma that is sometimes useful.

06RN

Lemma 30.8. Any surjective endomorphism of a Noetherian ring is an isomorphism.

Proof. If f : R R were such an endomorphism but not injective, then
Ker(f ) Ker(f f ) Ker(f f f ) . . .
would be a strictly increasing chain of ideals.

31. Locally nilpotent ideals

0AMF

Here is the definition.

00IL

Definition 31.1. Let R be a ring. Let I R be an ideal. We say I is locally

nilpotent if for every x I there exists an n N such that xn = 0. We say I is
nilpotent if there exists an n N such that I n = 0.

0544

Lemma 31.2. Let R R0 be a ring map and let I R be a locally nilpotent

ideal. Then IR0 is a locally nilpotent ideal of R0 .
Proof. This follows from the fact that if x, y R0 are nilpotent, then x + y is
nilpotent too. Namely, if xn = 0 and y m = 0, then (x + y)n+m1 = 0.


0AMG

Lemma 31.3. Let R be a ring and let I R be a locally nilpotent ideal. An

element x of R is a unit if and only if the image of x in R/I is a unit.
Proof. If x is a unit in R, then its image is clearly a unit in R/I. It remains to
prove the converse. Assume the image of y R in R/I is the inverse of the image
of x. Then xy = 1 z for some z I. Then every k 1 satisfies
(1 z)(1 + z)(1 + z 2 )(1 + z 4 ) . . . (1 + z 2

k1

) = 1 z2

(as follows by induction over k). But the right hand side is is equal to 1 for
sufficiently large k (since z lies in the locally nilpotent ideal I). Thus 1 z is
invertible in R, and therefore so is x (as xy = 1 z).

00IM

Lemma 31.4. Let R be a Noetherian ring. Let I, J be ideals of R. Suppose

J I. Then J n I for some n. In particular, in a Noetherian ring the notions

of locally nilpotent ideal and nilpotent ideal coincide.
Proof. Say J = (f1 , . . . , fs ). By assumption fidi I. Take n = d1 + d2 + . . . +
ds + 1.


00J9

Lemma 31.5. Let R be a ring. Let I R be a locally nilpotent ideal. Then

R R/I induces a bijection on idempotents.

COMMUTATIVE ALGEBRA

62

First proof of Lemma 31.5. As I is locally nilpotent it is contained in every

prime ideal. Hence Spec(R/I) = V (I) = Spec(R). Hence the lemma follows from
Lemma 20.3.

Second proof of Lemma 31.5. Suppose e R/I is an idempotent. We have to
lift e to an idempotent of R.
First, choose any lift f R of e, and set x = f 2 f . Then, x I, so x is nilpotent
(since I is locally nilpotent). Let now J be the ideal of R generated by x. Then, J
is nilpotent (not just locally nilpotent), since it is generated by the nilpotent x.
Now, assume that we have found a lift e R of e such that e2 e J k for some
k 1. Let e0 = e (2e 1)(e2 e) = 3e2 2e3 , which is another lift of e (since
the idempotency of e yields e2 e I). Then
(e0 )2 e0 = (4e2 4e 3)(e2 e)2 J 2k
by a simple computation.
We thus have started with a lift e of e such that e2 e J k , and obtained a
lift e0 of e such that (e0 )2 e0 J 2k . This way we can successively improve the
approximation (starting with e = f , which fits the bill for k = 1). Eventually,
we reach a stage where J k = 0, and at that stage we have a lift e of e such that
e2 e J k = 0, that is, this e is idempotent.
We thus have seen that if e R/I is any idempotent, then there exists a lift of e
which is an idempotent of R. It remains to prove that this lift is unique. Indeed,
let e1 and e2 be two such lifts. We need to show that e1 = e2 .
By definition of e1 and e2 , we have e1 e2 mod I, and both e1 and e2 are idempotent. From e1 e2 mod I, we see that e1 e2 I, so that e1 e2 is nilpotent
(since I is locally nilpotent). A straightforward computation (using the idempotency of e1 and e2 ) reveals that (e1 e2 )3 = e1 e2 . Using this and induction,
we obtain (e1 e2 )k = e1 e2 for any positive integer k. Since all high enough k
satisfy (e1 e2 )k = 0 (since e1 e2 is nilpotent), this shows e1 e2 = 0, so that
e1 = e2 , which completes our proof.

05BU

Lemma 31.6. Let A be a possibly noncommutative algebra. Let e A be an

2
element such that
Px = ei j e is nilpotent. Then there exists an idempotent of the
0
form e = e + x( ai,j e x ) A with ai,j Z.
Proof. Consider the ring Rn = Z[e]/((e2 e)n ). It is clear that if we can prove the
result for each Rn then the lemma follows. In Rn consider the ideal I = (e2 e)
and apply Lemma 31.5.

32. Curiosity

02JG

Lemma 22.3 explains what happens if V (I) is open for some ideal I R. But what
if Spec(S 1 R) is closed in Spec(R)? The next two lemmas give a partial answer.
For more information see Section 107.

02JH

Lemma 32.1. Let R be a ring. Let S R be a multiplicative subset. Assume the

image of the map Spec(S 1 R) Spec(R) is closed. Then S 1 R
= R/I for some
ideal I R.

COMMUTATIVE ALGEBRA

63

Proof. Let I = Ker(R S 1 R) so that V (I) contains the image. Say the image
is the closed subset V (I 0 ) Spec(R) for some ideal I 0 R. So V (I 0 ) V (I). For
f I 0 we see that f /1 S 1 R is contained in every prime ideal. Hence f n maps
to zero in S 1 R for some n 1 (Lemma 16.2). Hence V (I 0 ) = V (I). Then this
implies every g S is invertible mod I. Hence we get ring maps R/I S 1 R and
S 1 R R/I. The first map is injective by choice of I. The second is the map
S 1 R S 1 (R/I) = R/I which has kernel S 1 I because localization is exact.
Since S 1 I = 0 we see also the second map is injective. Hence S 1 R

= R/I.
02JI

Lemma 32.2. Let R be a ring. Let S R be a multiplicative subset. Assume

the image of the map Spec(S 1 R) Spec(R) is closed. If R is Noetherian, or
Spec(R) is a Noetherian topological space, or S is finitely generated as a monoid,
then R
= S 1 R R0 for some ring R0 .
Proof. By Lemma 32.1 we have S 1 R
= R/I for some ideal I R. By Lemma
22.3 it suffices to show that V (I) is open. If R is Noetherian then Spec(R) is a
Noetherian topological space, see Lemma 30.5. If Spec(R) is a Noetherian topological space, then the complement Spec(R) \ V (I) is quasi-compact, see Topology, Lemma 11.13. Hence there exist finitely many f1 , . . . , fn I such that
V (I) = V (f1 , . . . , fn ). Since each fi maps to zero in S 1 R there exists a g S such
that gfi = 0 for i = 1, . . . , n. Hence D(g) = V (I) as desired. In case S is finitely
generated as a monoid, say S is generated by g1 , . . . , gm , then S 1 R
= Rg1 ...gm and
we conclude that V (I) = D(g1 . . . gm ).

33. Hilbert Nullstellensatz

00FS
00FV

Theorem 33.1 (Hilbert Nullstellensatz). Let k be a field.

00FW

(1) For any maximal ideal m k[x1 , . . . , xn ] the field extension k (m) is
finite.
(2) Any radical ideal I k[x1 , . . . , xn ] is the intersection of maximal ideals
containing it.
The same is true in any finite type k-algebra.

00FX

Proof. It is enough to prove part (1) of the theorem for the case of a polynomial
algebra k[x1 , . . . , xn ], because any finitely generated k-algebra is a quotient of such
a polynomial algebra. We prove this by induction on n. The case n = 0 is clear.
Suppose that m is a maximal ideal in k[x1 , . . . , xn ]. Let p k[xn ] be the intersection
of m with k[xn ].
If p 6= (0), then p is maximal and generated by an irreducible monic polynomial P
(because of the Euclidean algorithm in k[xn ]). Then k 0 = k[xn ]/p is a finite field
extension of k and contained in (m). In this case we get a surjection
k 0 [x1 , . . . , xn1 ] k 0 [x1 , . . . , xn ] = k 0 k k[x1 , . . . , xn ] (m)
and hence we see that (m) is a finite extension of k 0 by induction hypothesis. Thus
(m) is finite over k as well.
If p = (0) we consider the ring extension k[xn ] k[x1 , . . . , xn ]/m. This is a finitely
generated ring extension, hence of finite presentation by Lemmas 30.3 and 30.4.
Thus the image of Spec(k[x1 , . . . , xn ]/m) in Spec(k[xn ]) is constructible by Theorem 28.9. Since the image contains (0) we conclude that it contains a standard

COMMUTATIVE ALGEBRA

64

open D(f ) for some f k[xn ] nonzero. Since clearly D(f ) is infinite we get a
contradiction with the assumption that k[x1 , . . . , xn ]/m is a field (and hence has a
spectrum consisting of one point).
To prove part (2) let I R be radical, with R of finite type over k. Let f R,
f 6 I. Pick a maximal ideal m0 in the nonzero ring Rf /IRf = (R/I)f . Let m R
be the inverse image of m0 in R. We see that I m and f 6 m. If we show that m
is a maximal ideal of R, then we are done. We clearly have
k R/m (m0 ).
By part (1) the field extension k (m0 ) is finite. Hence R/m is a field by Fields,
Lemma 8.10. Thus m is maximal and the proof is complete.

00FY

Lemma 33.2. Let R be a ring. Let K be a field. If R K and K is of finite type

over R, then there exists a f R such that Rf is a field, and Rf K is a finite
field extension.
Proof. By Lemma 29.2 there exist a nonempty open U Spec(R) contained in
the image {(0)} of Spec(K) Spec(R). Choose f R, f 6= 0 such that D(f ) U ,
i.e., D(f ) = {(0)}. Then Rf is a domain whose spectrum has exactly one point and
Rf is a field. Then K is a finitely generated algebra over the field Rf and hence a
finite field extension of Rf by the Hilbert Nullstellensatz (Theorem 33.1).

34. Jacobson rings

00FZ

Let R be a ring. The closed points of Spec(R) are the maximal ideals of R. Often
rings which occur naturally in algebraic geometry have lots of maximal ideals. For
example finite type algebras over a field or over Z. We will show that these are
examples of Jacobson rings.

00G0

Definition 34.1. Let R be a ring. We say that R is a Jacobson ring if every

radical ideal I is the intersection of the maximal ideals containing it.

00G1

Lemma 34.2. Any algebra of finite type over a field is Jacobson.

Proof. This follows from Theorem 33.1 and Definition 34.1.

00G2

Lemma 34.3. Let R be a ring. If every prime ideal of R is the intersection of the
maximal ideals containing it, then R is Jacobson.
Proof. This is immediately clear from the fact that every radical ideal I R is
the intersection of the primes containing it. See Lemma 16.2.


00G3

Lemma 34.4. A ring R is Jacobson if and only if Spec(R) is Jacobson, see

Topology, Definition 17.1.
Proof. Suppose R is Jacobson. Let Z Spec(R) be a closed subset. We have
to show that the set of closed points in Z is dense in Z. Let U Spec(R) be an
open such that U Z is nonempty. We have to show Z U contains a closed point
of Spec(R). We may assume U = D(f ) as standard opens form a basis for the
topology on Spec(R). According to Lemma 16.2 we may assume that Z = V (I),
where I is a radical ideal. We see also that f 6 I. By assumption, there exists a
maximal ideal m R such that I m but f 6 m. Hence m D(f ) V (I) = U Z
as desired.

COMMUTATIVE ALGEBRA

65

Conversely, suppose that Spec(R) is Jacobson. Let I R be a radical ideal. Let

J = Im m be the intersection of the maximal ideals containing I. Clearly J is
radical, V (J) V (I), and V (J) is the smallest closed subset of V (I) containing all
the closed points of V (I). By assumption we see that V (J) = V (I). But Lemma
16.2 shows there is a bijection between Zariski closed sets and radical ideals, hence
I = J as desired.

034J

Lemma 34.5. Let R be a ring. If R is not Jacobson there exist a prime p R,

an element f R such that the following hold
(1) p is not a maximal ideal,
(2) f 6 p,
(3) V (p) D(f ) = {p}, and
(4) (R/p)f is a field.
On the other hand, if R is Jacobson, then for any pair (p, f ) such that (1) and (2)
hold the set V (p) D(f ) is infinite.
Proof. Assume R is not Jacobson. By Lemma 34.4 this means there exists an
closed subset T Spec(R) whose set T0 T of closed points is not dense in T .
Choose an f R such that T0 V (f ) but T 6 V (f ). Note that T D(f ) is
homeomorphic to Spec((R/I)f ) if T = V (I), see Lemmas 16.7 and 16.6. As any
ring has a maximal ideal (Lemma 16.2) we can choose a closed point t of space
T D(f ). Then t corresponds to a prime ideal p R which is not maximal (as
t 6 T0 ). Thus (1) holds. By construction f 6 p, hence (2). As t is a closed point of
T D(f ) we see that V (p) D(f ) = {p}, i.e., (3) holds. Hence we conclude that
(R/p)f is a domain whose spectrum has one point, hence (4) holds (for example
combine Lemmas 17.2 and 24.1).
Conversely, suppose that R is Jacobson and (p, f ) satisfy (1) and (2). If V (p)
V (f ) = {p, q1 , . . . , qt } then p 6= qi implies there exists an element g R such that
g 6 p but g qi for all i. Hence V (p) D(f g) = {p} which is impossible since each
locally closed subset of Spec(R) contains at least one closed point as Spec(R) is a
Jacobson topological space.


00G4

Lemma 34.6. The ring Z is a Jacobson ring. More generally, let R be a ring
such that
(1) R is a domain,
(2) R is Noetherian,
(3) any nonzero prime ideal is a maximal ideal, and
(4) R has infinitely many maximal ideals.
Then R is a Jacobson ring.
Proof.
Let R satisfy (1), (2), (3) and (4). The statement means that (0) =
T
m.
Since R has infinitely many maximal ideals it suffices to show that any
mR
nonzero x R is contained in at most finitely many maximal ideals, in other
words that V (x) is finite. By Lemma 16.7 we see that V (x) is homeomorphic
to Spec(R/xR). By assumption (3) every prime of R/xR is minimal and hence
corresponds to an irreducible component of Spec(R) (Lemma 25.1). As R/xR is
Noetherian, the topological space Spec(R/xR) is Noetherian (Lemma 30.5) and has
finitely many irreducible components (Topology, Lemma 8.2). Thus V (x) is finite
as desired.


COMMUTATIVE ALGEBRA

66

02CC

Example 34.7.Q Let A be an infinite set. For each A, let k be a field. We

claim that R = A k is Jacobson. First, note that any element f R has the
form f = ue, with u R a unit and e R an idempotent (left to the reader). Hence
D(f ) = D(e), and Rf = Re = R/(1 e) is a quotient of R. Actually, any ring with
this property is Jacobson. Namely, say p R is a prime ideal and f R, f 6 p.
We have to find a maximal ideal m of R such that p m and f 6 m. Because Rf
is a quotient of R we see that any maximal ideal of Rf corresponds to a maximal
ideal of R not containing f . Hence the result follows by choosing a maximal ideal
of Rf containing pRf .

00G5

Example 34.8. A domain R with finitely many maximal ideals mi , i = 1, . . . , n

is not a Jacobson ring, except when it is a field. Namely, in this case (0) is not the
intersection of the maximal ideals (0) 6= m1 m2 . . . mn m1 m2 . . . mn 6= 0.
In particular a discrete valuation ring, or any local ring with at least two prime
ideals is not a Jacobson ring.

00GA

Lemma 34.9. Let R S be a ring map. Let m R be a maximal ideal. Let

q S be a prime ideal lying over m such that (m) (q) is an algebraic field
extension. Then q is a maximal ideal of S.
Proof. Consider the diagram
SO

/ S/q
O

/ R/m

/ (q)

We see that (m) S/q (q). Because the field extension (m) (q) is
algebraic, any ring between (m) and (q) is a field (Fields, Lemma 8.10). Thus
S/q is a field, and a posteriori equal to (q).

00FT

Lemma 34.10. Suppose that k is a field and suppose that V is a nonzero vector
space over k. Assume the dimension of V (which is a cardinal number) is smaller
than the cardinality of k. Then for any linear operator T : V V there exists
some monic polynomial P (t) k[t] such that P (T ) is not invertible.
Proof. If not then V inherits the structure of a vector space over the field k(t).
But the dimension of k(t) over k is at least the cardinality of k for example due to
1
the fact that the elements t
are k-linearly independent.

Here is another version of Hilberts Nullstellensatz.

00FU

Theorem 34.11. Let k be a field. Let S be a k-algebra generated over k by the

elements {xi }iI . Assume the cardinality of I is smaller than the cardinality of k.
Then
(1) for all maximal ideals m S the field extension k (m) is algebraic, and
(2) S is a Jacobson ring.
Proof. If I is finite then the result follows from the Hilbert Nullstellensatz, Theorem 33.1. In the rest of the proof we assume I is infinite. It suffices to prove the
result for m k[{xi }iI ] maximal in the polynomial ring on variables xi , since S
is a quotient of this. As I is infinite the set of monomials xei11 . . . xeirr , i1 , . . . , ir I
and e1 , . . . , er 0 has cardinality at most equal to the cardinality of I. Because the

COMMUTATIVE ALGEBRA

67

S
cardinality of I . . . I is the cardinality of I, and also the cardinality of n0 I n
has the same cardinality. (If I is finite, then this is not true and in that case this
proof only works if k is uncountable.)
To arrive at a contradiction pick T (m) transcendental over k. Note that the
k-linear map T : (m) (m) given by multiplication by T has the property that
P (T ) is invertible for all monic polynomials P (t) k[t]. Also, (m) has dimension
at most the cardinality of I over k since it is a quotient of the vector space k[{xi }iI ]
over k (whose dimension is #I as we saw above). This is impossible by Lemma
34.10.
To show that S is Jacobson we argue as follows. If not then there exists a prime
q S and an element f S, f 6 q such that q is not maximal and (S/q)f is a field,
see Lemma 34.5. But note that (S/q)f is generated by at most #I + 1 elements.
Hence the field extension k (R/q)f is algebraic (by the first part of the proof).
This implies that (q) is an algebraic extension of k hence q is maximal by Lemma
34.9. This contradiction finishes the proof.

046V

Lemma 34.12. Let k be a field. Let S be a k-algebra. For any field extension
k K whose cardinality is larger than the cardinality of S we have
(1) for every maximal ideal m of SK the field (m) is algebraic over K, and
(2) SK is a Jacobson ring.
Proof. Choose k K such that the cardinality of K is greater than the cardinality
of S. Since the elements of S generate the K-algebra SK we see that Theorem 34.11
applies.


02CB

Example 34.13. The trick in the proof of Theorem 34.11 really does not work
if k is a countable field and I is countable too. Let k be a countable field. Let
x be a variable, and let k(x) be the field of rational functions in x. Consider the
polynomial algebra R = k[x, {xf }f k[x]{0} ]. Let I = ({f xf 1}f k[x]{0} ). Note
that I is a proper ideal in R. Choose a maximal ideal I m. Then k R/m is
isomorphic to k(x), and is not algebraic over k.

00G6

Lemma 34.14. Let R be a Jacobson ring. Let f R. The ring Rf is Jacobson

and maximal ideals of Rf correspond to maximal ideals of R not containing f .
Proof. By Topology, Lemma 17.5 we see that D(f ) = Spec(Rf ) is Jacobson and
that closed points of D(f ) correspond to closed points in Spec(R) which happen to
lie in D(f ). Thus Rf is Jacobson by Lemma 34.4.


00G7

Example 34.15. Here is a simple example that shows Lemma 34.14 to be false
if R is not Jacobson. Consider the ring R = Z(2) , i.e., the localization of Z at the
prime (2). The localization of R at the element 2 is isomorphic to Q, in a formula:
R2
= Q. Clearly the map R R2 maps the closed point of Spec(Q) to the generic
point of Spec(R).

00G8

Example 34.16. Here is a simple example that shows Lemma 34.14 is false if R
is Jacobson but we localize at infinitely many elements. Namely, let R = Z and
consider the localization (R \ {0})1 R
= Q of R at the set of all nonzero elements.
Clearly the map Z Q maps the closed point of Spec(Q) to the generic point of
Spec(Z).

COMMUTATIVE ALGEBRA

00G9

Lemma 34.17. Let R be a Jacobson ring. Let I R be an ideal. The ring R/I is
Jacobson and maximal ideals of R/I correspond to maximal ideals of R containing
I.
Proof. The proof is the same as the proof of Lemma 34.14.

00GB

68

Proposition 34.18. Let R be a Jacobson ring. Let R S be a ring map of finite

type. Then
(1) The ring S is Jacobson.
(2) The map Spec(S) Spec(R) transforms closed points to closed points.
(3) For m0 S maximal lying over m R the field extension (m) (m0 ) is
finite.
Proof. Let A B C be finite type ring maps. Suppose Spec(C) Spec(B)
and Spec(B) Spec(A) map closed points to closed points, and induce finite
residue field extensions on residue fields at closed points. Then so does Spec(C)
Spec(A). Thus it is clear that if we factor R S as R S 0 S for some finite
type R-algebra S 0 , then it suffices to prove the lemma for R S 0 and then S 0 S.
Writing S = R[x1 , . . . , xn ]/I we see that it suffices to prove the lemma in the cases
S = R[x] and S = R/I. The case S = R/I is Lemma 34.17.
The case S = R[x]. Take an irreducible closed subset Z Spec(R[x]). In other
words Z = V (q) for some prime q R[x]. Set p = q R. Let U Spec(R[x]) be
open such that U Z 6= . We have to find a closed point in U Z. In fact, we
will find
() a closed point y of U Z which maps to a closed point x of Spec(R) such
that additionally (x) (y) is finite.
To do this we may assume U = D(f ) for some f R[x]. In this case U V (q) 6=
means f 6 q. Consider the diagram
R[x]
O

/ R/p[x]
O

/ R/p

It suffices to solve the problem on the right hand side of this diagram. Thus we see
we may assume R is Jacobson, a domain and p = (0).
In case q = (0), write f = ad xd + . . . + a0 . We see that not all ai are zero. Take
any maximal ideal m of R such that ai 6 m for some i (here we use R is Jacobson).
Next, choose a maximal ideal m0 (R/m)[x] not containing the image of f (possible
because (m)[x] is Jacobson). Then the inverse image m0 R[x] defines a closed
point of U Z and maps to m. Also, by construction (m) (m0 ) is finite. Thus
we have shown () in this case.
In case q 6= (0), let K be the fraction field of R. Write qK[x] = (g) for some
irreducible g K[x]. Clearing denominators, we may assume that g R[x], and
hence in q. Write g = be xe + . . . + b0 , bi R with be 6= 0. The maps R Rbe
and R[x] R[x]be satisfies the conclusion of the lemma, by Lemma 34.14 and
moreover induce isomorphisms on residue fields. Hence, in order to prove (), we
may replace R by Rbe and assume that g is monic. In this case we see that R[x]/q
is a quotient of the finite free R-module R[x]/(g) = R Rx . . . Rxe1 . But

COMMUTATIVE ALGEBRA

69

on the other hand we have R[x]/(g) K[x]/(g) = K[x]/qK[x]. Hence q = (g),

and Z = V (q) = V (g). At this point, by Lemma 28.8 the image of D(f ) V (g) in
Spec(R) is D(r1 ) . . . D(rd ) for some ri R (of course it is nonempty). Take
any maximal ideal m R in this image (possible because R is Jacobson) and take
any prime m0 R[x] corresponding to a point of D(f ) V (g) lying over m. Note
that the residue field extension (m) (m0 ) is finite (because g m0 ). By Lemma
34.9 we see that m0 is a closed point. This proves () in this case.
At this point we are done. Namely, () implies that Spec(R[x]) is Jacobson (via
Lemma 34.4). Also, if Z is a singleton closed set, then () implies that Z = {m0 }
with m0 lying over a maximal ideal m R such that (m) (m0 ) is finite.

00GC

Lemma 34.19. Any finite type algebra over Z is Jacobson.

Proof. Combine Lemma 34.6 and Proposition 34.18.

00GD

Lemma 34.20. Let R S be a finite type ring map of Jacobson rings. Denote
X = Spec(R) and Y = Spec(S). Write f : Y X the induced map of spectra. Let
E Y = Spec(S) be a constructible set. Denote with a subscript 0 the set of closed
points of a topological space.
(1) We have f (E)0 = f (E0 ) = X0 f (E).
(2) A point X is in f (E) if and only if {} f (E0 ) is dense in {}.
Proof. We have a commutative diagram of continuous maps
E

/Y


f (E)


/X

Suppose x f (E) is closed in f (E). Then f 1 ({x}) E is nonempty and closed

in E. Applying Topology, Lemma 17.5 to both inclusions
f 1 ({x}) E E Y
we find there exists a point y f 1 ({x}) E which is closed in Y . In other words,
there exists y Y0 and y E0 mapping to x. Hence x f (E0 ). This proves that
f (E)0 f (E0 ). Proposition 34.18 implies that f (E0 ) X0 f (E). The inclusion
X0 f (E) f (E)0 is trivial. This proves the first assertion.
Suppose that f (E). According to Lemma 29.2 the set f (E) {} contains a
dense open subset of {}. Since X is Jacobson we conclude that f (E){} contains
a dense set of closed points, see Topology, Lemma 17.5. We conclude by part (1)
of the lemma.
On the other hand, suppose that {} f (E0 ) is dense in {}. By Lemma 28.3
there exists a ring map S S 0 of finite presentation such that E is the image of
Y 0 := Spec(S 0 ) Y . Then E0 is the image of Y00 by the first part of the lemma
applied to the ring map S S 0 . Thus we may assume that E = Y by replacing S

COMMUTATIVE ALGEBRA

70

by S 0 . Suppose corresponds to p R. Consider the diagram

SO

/ S/pS
O

/ R/p

This diagram and the density of f (Y0 ) V (p) in V (p) shows that the morphism
R/p S/pS satisfies condition (2) of Lemma 29.4. Hence we conclude there exists
a prime q S/pS mapping to (0). In other words the inverse image q of q in S
maps to p as desired.

The conclusion of the lemma above is that we can read off the image of f from the
set of closed points of the image. This is a little nicer in case the map is of finite
presentation because then we know that images of a constructible is constructible.
Before we state it we introduce some notation. Denote Constr(X) the set of constructible Let R S be a ring map. Denote X = Spec(R) and Y = Spec(S).
Write f : Y X the induced map of spectra. Denote with a subscript 0 the set of
closed points of a topological space.
00GE

Lemma 34.21. With notation as above. Assume that R is a Noetherian Jacobson

ring. Further assume R S is of finite type. There is a commutative diagram
Constr(Y )

E7E0

E7f (E)


Constr(X)

/ Constr(Y0 )
E7f (E)


/ Constr(X0 )

E7E0

where the horizontal arrows are the bijections from Topology, Lemma 17.8.
Proof. Since R S is of finite type, it is of finite presentation, see Lemma 30.4.
Thus the image of a constructible set in X is constructible in Y by Chevalleys
theorem (Theorem 28.9). Combined with Lemma 34.20 the lemma follows.

To illustrate the use of Jacobson rings, we give the following two examples.
00GF

Example 34.22. Let k be a field. The space Spec(k[x, y]/(xy)) has two irreducible
components: namely the x-axis and the y-axis. As a generalization, let
R = k[x11 , x12 , x21 , x22 , y11 , y12 , y21 , y22 ]/a,
where a is the ideal in k[x11 , x12 , x21 , x22 , y11 , y12 , y21 , y22 ] generated by the entries
of the 2 2 product matrix



x11 x12
y11 y12
.
x21 x22
y21 y22
In this example we will describe Spec(R).
To prove the statement about Spec(k[x, y]/(xy)) we argue as follows. If p k[x, y]
is any ideal containing xy, then either x or y would be contained in p. Hence the
minimal such prime ideals are just (x) and (y). In case k is algebraically closed,
the max-Spec of these components can then be visualized as the point sets of yand x-axis.

COMMUTATIVE ALGEBRA

71

For the generalization, note that we may identify the closed points of the spectrum
of k[x11 , x12 , x21 , x22 , y11 , y12 , y21 , y22 ]) with the space of matrices





x11 x12
y11 y12
(X, Y ) Mat(2, k) Mat(2, k) | X =
,Y =
x21 x22
y21 y22
at least if k is algebraically closed. Now define a group action of GL(2, k)
GL(2, k) GL(2, k) on the space of matrices {(X, Y )} by
(g1 , g2 , g3 ) (X, Y ) 7 ((g1 Xg21 , g2 Y g31 )).
Here, also observe that the algebraic set
GL(2, k) GL(2, k) GL(2, k) Mat(2, k) Mat(2, k) Mat(2, k)
is irreducible since it is the max spectrum of the domain
k[x11 , x12 , . . . , z21 , z22 , (x11 x22 x12 x21 )1 , (y11 y22 y12 y21 )1 , (z11 z22 z12 z21 )1 ].
Since the image of irreducible an algebraic set is still irreducible, it suffices to
classify the orbits of the set {(X, Y ) Mat(2, k) Mat(2, k)|XY = 0} and take
their closures. From standard linear algebra, we are reduced to the following three
cases:
(1) (g1 , g2 ) such that g1 Xg21 = I22 . Then Y is necessarily 0, which as an
algebraic set is invariant under the group action. It follows that this orbit is contained in the irreducible algebraic set defined by the prime ideal
(y11 , y12 , y21 , y22 ). Taking the closure, we see that (y11 , y12 , y21 , y22 ) is actually a component.
(2) (g1 , g2 ) such that


1 0
g1 Xg21 =
.
0 0
This case occurs if and only if X is a rank 1 matrix, and furthermore, Y is
killed by such an X if and only if
x11 y11 + x12 y21 = 0;

x11 y12 + x12 y22 = 0;

x21 y11 + x22 y21 = 0;

x21 y12 + x22 y22 = 0.

Fix a rank 1 X, such non zero Y s satisfying the above equations form an
irreducible algebraic set for the following reason(Y = 0 is contained the
previous case): 0 = g1 Xg21 g2 Y implies that


0
0
g2 Y =
.
0
0
y22
y21
With a further GL(2, k)-action on the right by g3 , g2 Y can be brought into


0 0
g2 Y g31 =
,
0 1
and thus such Y s form an irreducible algebraic set isomorphic to the image of GL(2, k) under this action. Finally, notice that the rank 1 condition for Xs forms an open dense subset of the irreducible algebraic set
det X = x11 x22 x12 x21 = 0. It now follows that all the five equations
define an irreducible component (x11 y11 + x12 y21 , x11 y12 + x12 y22 , x21 y11 +
x22 y21 , x21 y12 + x22 y22 , x11 x22 x12 x21 ) in the open subset of the space
of pairs of nonzero matrices. It can be shown that the pair of equations

COMMUTATIVE ALGEBRA

72

det X = 0, det Y = 0 cuts Spec(R) in an irreducible component with the

above locus an open dense subset.
(3) (g1 , g2 ) such that g1 Xg21 = 0, or equivalently, X = 0. Then Y can be
arbitrary and this component is thus defined by (x11 , x12 , x21 , x22 ).
00GG

Example 34.23. For another example, consider R = k[{tij }ni,j=1 ]/a, where a is
the ideal generated by the entries of the product matrix T 2 T , T = (tij ). From
linear algebra, we know that under the GL(n, k)-action defined by g, T 7 gT g 1 , T
is classified by the its rank and each T is conjugate to some diag(1, . . . , 1, 0, . . . , 0),
which has r 1s and nr 0s. Thus each orbit of such a diag(1, . . . , 1, 0, . . . , 0) under
the group action forms an irreducible component and every idempotent matrix
is contained in one such orbit. Next we will show that any two different orbits
are necessarily disjoint. For this purpose we only need to cook up polynomial
functions that take different values on different orbits. In characteristic
0 cases,
Pn
such a function can be taken to be f (tij ) = trace(T ) =
t
.
In
positive
i=1 ii
characteristic cases, things are slightly more tricky since we might have trace(T ) = 0
even if T 6= 0. For instance, char = 3

1
1 =3=0
trace
1
Anyway, these components can be separated using other functions. For instance, in
the characteristic 3 case, tr(3 T ) takes value 1 on the components corresponding
to diag(1, 1, 1) and 0 on other components.
35. Finite and integral ring extensions

00GH

Trivial lemmas concerning finite and integral ring maps. We recall the definition.

00GI

Definition 35.1. Let : R S be a ring map.

(1) An element s S is integral over R if there exists a monic polynomial
P (x) R[x] such that P (s) = 0, where P (x) S[x] is the image of P
under : R[x] S[x].
(2) The ring map is integral if every s S is integral over R.

052I

Lemma 35.2. Let : R S be a ring map. Let y S. If there exists a finite

R-submodule M of S such that 1 M and yM M , then y is integral over R.
Proof. Let x1 = 1 M and xi M ,P
i = 2, . . . , n be a finite set of elements generating M as an R-module. Write yxi =
(aij )xj for some aij R. Let P (T ) R[T ]
be the characteristic polynomial of the n n matrix A = (aij ). By Lemma
P 15.1 we
see P (A) = 0. By construction the map : Rn M , (a1 , . . . , an ) 7
(ai )xi
commutes with A : Rn Rn and multiplication by y. In a formula (Av) = y(v).
Thus P (y) = P (y) 1 = P (y) x1 = P (y) ((1, 0, . . . , 0)) = (P (A)(1, 0, . . . , 0)) =
0.


00GK

Lemma 35.3. A finite ring extension is integral.

Proof. Let R S be finite. Let y S. Apply Lemma 35.2 to M = S to see that
y is integral over R.


00GM

Lemma 35.4. Let : R S be a ring map. Let s1 , . . . , sn be a finite set of

elements of S. In this case si is integral over R for all i = 1, . . . , n if and only if
there exists an R-subalgebra S 0 S finite over R containing all of the si .

COMMUTATIVE ALGEBRA

73

Proof. If each si is integral, then the subalgebra generated by (R) and the si
is finite over R. Namely, if si satisfies a monic equation of degree di over R,
then this subalgebra is generated as an R-module by the elements se11 . . . senn with
0 ei di 1. Conversely, suppose given a finite R-subalgebra S 0 containing all
the si . Then all of the si are integral by Lemma 35.3.

02JJ

Lemma 35.5. Let R S be a ring map. The following are equivalent

(1) R S is finite,
(2) R S is integral and of finite type, and
(3) there exist x1 , . . . , xn S which generate S as an algebra over R such that
each xi is integral over R.
Proof. Clear from Lemma 35.4.

00GN

Lemma 35.6. Suppose that R S and S T are integral ring maps. Then
R T is integral.
Proof. Let t T . Let P (x) S[x] be a monic polynomial such that P (t) = 0.
Apply Lemma 35.4 to the finite set of coefficients of P . Hence t is integral over some
subalgebra S 0 S finite over R. Apply Lemma 35.4 again to find a subalgebra
T 0 T finite over S 0 and containing t. Lemma 7.3 applied to R S 0 T 0
shows that T 0 is finite over R. The integrality of t over R now follows from Lemma
35.3.


00GO

Lemma 35.7. Let R S be a ring homomorphism. The set

S 0 = {s S | s is integral over R}
is an R-subalgebra of S.
Proof. This is clear from Lemmas 35.4 and 35.3.

00GP


0

Definition 35.8. Let R S be a ring map. The ring S S of elements integral

over R, see Lemma 35.7, is called the integral closure of R in S. If R S we say
that R is integrally closed in S if R = S 0 .
In particular, we see that R S is integral if and only if the integral closure of R
in S is all of S.

0307

Lemma 35.9. Integral closure commutes with localization: If A B is a ring

map, and S A is a multiplicative subset, then the integral closure of S 1 A in
S 1 B is S 1 B 0 , where B 0 B is the integral closure of A in B.
Proof. Since localization
see that S 1 B 0 S 1 B. Suppose x B 0
P is exact we
d
di
and f S. Then x + i=1,...,d ai x
= 0 in B for some ai A. Hence also
X
(x/f )d +
ai /f i (x/f )di = 0
i=1,...,d

in S B. In this way we see that S 1 B 0 is contained in the integral closure of

S 1 A in S 1 B. Conversely, suppose that x/f S 1 B is integral over S 1 A. Then
we have
X
(x/f )d +
(ai /fi )(x/f )di = 0
i=1,...,d

in S 1 B for some ai A and fi S. This means that

X
(f 0 f1 . . . fd x)d +
f i (f 0 )i f1i . . . fii1 . . . fdi ai (f 0 f1 . . . fd x)di = 0
i=1,...,d

COMMUTATIVE ALGEBRA

74

for a suitable f 0 S. Hence f 0 f1 . . . fd x B 0 and thus x/f S 1 B 0 as desired. 

034K

Lemma 35.10. Let : R S be a ring map. Let x S. The following are

equivalent:
(1) x is integral over R, and
(2) for every prime ideal p R the element x Sp is integral over Rp .
Proof. It is clear that (1) implies (2). Assume (2). Consider the R-algebra S 0 S
generated
P by (R) and x. Let p be a prime ideal of R. Then we know that
xd + i=1,...,d (ai )xdi = 0 in Sp for some ai Rp . Hence we see, by looking
at which
P denominators occur, that for some f R, f 6 p we have ai Rf and
xd + i=1,...,d (ai )xdi = 0 in Sf . This implies that Sf0 is finite over Rf . Since
p was arbitrary and Spec(R) is quasi-compact (Lemma 16.10) we can find finitely
many elements f1 , . . . , fn R which generate the unit ideal of R such that Sf0 i is
finite over Rfi . Hence we conclude from Lemma 23.2 that S 0 is finite over R. Hence
x is integral over R by Lemma 35.4.


02JK

Lemma 35.11. Let R S and R R0 be ring maps. Set S 0 = R0 R S.

(1) If R S is integral so is R0 S 0 .
(2) If R S is finite so is R0 S 0 .
Proof. We prove (1). Let si S be generators for S over R. Each of these satisfies
a monic polynomial equation Pi over R. Hence the elements 1 si S 0 generate S 0
over R0 and satisfy the corresponding polynomial Pi0 over R0 . Since these elements
generate S 0 over R0 we see that S 0 is integral over R0 . Proof of (2) omitted.


02JL

Lemma 35.12. Let R S be a ring map. Let f1 , . . . , fn R generate the unit

ideal.
(1) If each Rfi Sfi is integral, so is R S.
(2) If each Rfi Sfi is finite, so is R S.
Proof. Proof of (1). Let s S. Consider the ideal I R[x] of polynomials P such
that P (s) = 0. Let J R denote the ideal (!) of leading coefficients of elements
of I. By assumption and clearing denominators we see that fini J for all i and
certain ni 0. Hence J contains 1 and we see s is integral over R. Proof of (2)
omitted.


02JM

Lemma 35.13. Let A B C be ring maps.

(1) If A C is integral so is B C.
(2) If A C is finite so is B C.
Proof. Omitted.

0308

Lemma 35.14. Let A B C be ring maps. Let B 0 be the integral closure of

A in B, let C 0 be the integral closure of B 0 in C. Then C 0 is the integral closure of
A in C.
Proof. Omitted.

00GQ

Lemma 35.15. Suppose that R S is an integral ring extension with R S.

Then : Spec(S) Spec(R) is surjective.

COMMUTATIVE ALGEBRA

75

Proof. Let p R be a prime ideal. We have to show pSp 6= Sp , see Lemma

16.9. The localization Rp Sp is injective (as localization is exact) and integral
by Lemma 35.9 or 35.11. Hence we may replace R, S by Rp , Sp and we may
assume R is local
P with maximal ideal m and it suffices to show that mS 6= S.
Suppose 1 =
fi si with fi m and si S in order to get a contradiction. Let
R S0 S P
be such that R S 0 is finite and si S 0 , see Lemma 35.4. The
equation 1 =
fi si implies that the finite R-module S 0 satisfies S 0 = mS 0 . Hence
by Nakayamas Lemma 19.1 we see S 0 = 0. Contradiction.

00GR

Lemma 35.16. Let R be a ring. Let K be a field. If R K and K is integral

over R, then R is a field and K is an algebraic extension. If R K and K is finite
over R, then R is a field and K is a finite algebraic extension.
Proof. Assume that R K is integral. By Lemma 35.15 we see that Spec(R) has
1 point. Since clearly R is a domain we see that R = R(0) is a field (Lemma 24.1).
The other assertions are immediate from this.


00GS

Lemma 35.17. Let k be a field. Let S be a k-algebra over k.

(1) If S is a domain and finite dimensional over k, then S is a field.
(2) If S is integral over k and a domain, then S is a field.
(3) If S is integral over k then every prime of S is a maximal ideal (see Lemma
25.5 for more consequences).
Proof. The statement on primes follows from the statement integral + domain
field. Let S integral over k and assume S is a domain, Take s S. By Lemma
35.4 we may find a finite dimensional k-subalgebra k S 0 S containing s. Hence
S is a field if we can prove the first statement. Assume S finite dimensional over k
and a domain. Pick s S. Since S is a domain the multiplication map s : S S
is surjective by dimension reasons. Hence there exists an element s1 S such that
ss1 = 1. So S is a field.


00GT

Lemma 35.18. Suppose R S is integral. Let q, q0 Spec(S) be distinct primes

having the same image in Spec(R). Then neither q q0 nor q0 q.
Proof. Let p R be the image. By Remark 16.8 the primes q, q0 correspond to
ideals in S R (p). Thus the lemma follows from Lemma 35.17.


05DR

Lemma 35.19. Suppose R S is finite. Then the fibres of Spec(S) Spec(R)

are finite.
Proof. By the discussion in Remark 16.8 the fibres are the spectra of the rings
S R (p). As R S is finite, these fibre rings are finite over (p) hence Noetherian
by Lemma 30.1. By Lemma 35.18 every prime of S R (p) is a minimal prime.
Hence by Lemma 30.6 there are at most finitely many.


00GU

Lemma 35.20. Let R S be a ring map such that S is integral over R. Let
p p0 R be primes. Let q be a prime of S mapping to p. Then there exists a
prime q0 with q q0 mapping to p0 .
Proof. We may replace R by R/p and S by S/q. This reduces us to the situation
of having an integral extension of domains R S and a prime p0 R. By Lemma
35.15 we win.


COMMUTATIVE ALGEBRA

76

The property expressed in the lemma above is called the going up property for
the ring map R S, see Definition 40.1.
0564

Lemma 35.21. Let R S be a finite and finitely presented ring map. Let M
be an S-module. Then M is finitely presented as an R-module if and only if M is
finitely presented as an S-module.
Proof. One of the implications follows from Lemma 6.4. To see the other assume
that M is finitely presented as an S-module. Pick a presentation
S m S n M 0
As S is finite as an R-module, the kernel of S n M is a finite R-module. Thus
from Lemma 5.3 we see that it suffices to prove that S is finitely presented as an
R-module.
Pick y1 , . . . , yn S such that y1 , . . . , yn generate S as an R-module. By Lemma
35.2 each yi is integral over R. Choose mononic polynomials Pi (x) R[x] with
Pi (yi ) = 0. Consider the ring
S 0 = R[x1 , . . . , xn ]/(P1 (x1 ), . . . , Pn (xn ))
Then we see that S is of finite presentation as an S 0 -algebra by Lemma 6.2. Since
S 0 S is surjective we see that S is of finite presentation as an S 0 -module (use
Lemma 6.3). Hence, arguing as in the first paragraph, it suffices to show that
S 0 is of finite presentation as an R-module. To see this we write R S 0 as the
composition
R R[x1 ]/(P1 (x1 )) R[x1 , x2 ]/(P1 (x1 ), P2 (x2 )) . . . S 0
of ring maps of the form R0 R0 [x]/(xd + a1 xd1 + . . . + ad ). Again arguing
as in the first paragraph of the proof it is enough to show that the ith ring in
this sequence is of finite presentation as a module over the (i 1)st one. This is
true because R0 [x]/(xd + a1 xd1 + . . . + ad ) is free as a module over R0 with basis
1, x, . . . , xd1 .


052J

Lemma 35.22. Let R be a ring. Let x, y R be nonzerodivisors. Let R[x/y]

Rxy be the R-subalgebra generated by x/y, and similarly for the subalgebras R[y/x]
and R[x/y, y/x]. If R is integrally closed in Rx or Ry , then the sequence
(1,1)

(1,1)

0 R R[x/y] R[y/x] R[x/y, y/x] 0

is a short exact sequence of R-modules.
Proof. Since x/y y/x = 1 it is clear that the map R[x/y] R[y/x] R[x/y, y/x]
is surjective. Let R[x/y] R[y/x]. To show exactness in the middle we have
to prove that R. By assumption we may write
= a0 + a1 x/y + . . . + an (x/y)n = b0 + b1 y/x + . . . + bm (y/x)m
for some n, m 0 and ai , bj R. Pick some N > max(n, m). Consider the finite
R-submodule M of Rxy generated by the elements
(x/y)N , (x/y)N 1 , . . . , x/y, 1, y/x, . . . , (y/x)N 1 , (y/x)N
We claim that M M . Namely, it is clear that (x/y)i (b0 + b1 y/x + . . . +
bm (y/x)m ) M for 0 i N and that (y/x)i (a0 + a1 x/y + . . . + an (x/y)n ) M
for 0 i N . Hence is integral over R by Lemma 35.2. Note that Rx , so if
R is integrally closed in Rx then R as desired.


COMMUTATIVE ALGEBRA

77

36. Normal rings

037B

We first introduce the notion of a normal domain, and then we introduce the (very
general) notion of a normal ring.

0309

Definition 36.1. A domain R is called normal if it is integrally closed in its field

of fractions.

034L

Lemma 36.2. Let R S be a ring map. If S is a normal domain, then the

integral closure of R in S is a normal domain.
Proof. Omitted.

The following notion is occasionally useful when studying normality.

00GW

Definition 36.3. Let R be a domain.

(1) An element g of the fraction field of R is called almost integral over R if
there exists an element r R, r 6= 0 such that rg n R for all n 0.
(2) The domain R is called completely normal if every almost integral element
of the fraction field of R is contained in R.
The following lemma shows that a Noetherian domain is normal if and only if it is
completely normal.

00GX

Lemma 36.4. Let R be a domain with fraction field K. If u, v K are almost

integral over R, then so are u + v and uv. Any element g K which is integral over
R is almost integral over R. If R is Noetherian then the converse holds as well.
Proof. If run R for all n 0 and v n r0 R for all n 0, then (uv)n rr0 and
(u + v)n rr0 are in R for all n 0. Hence the first assertion. Suppose g K
is integral over R. In this case there exists an d > 0 such that the ring R[g] is
generated by 1, g, . . . , g d as an R-module. Let r R be a common denominator of
the elements 1, g, . . . , g d K. It is follows that rR[g] R, and hence g is almost
integral over R.
Suppose R is Noetherian and g K is almost integral over R. Let r R, r 6= 0 be
as in the definition. Then R[g] 1r R as an R-module. Since R is Noetherian this
implies that R[g] is finite over R. Hence g is integral over R, see Lemma 35.3. 

00GY

Lemma 36.5. Any localization of a normal domain is normal.

Proof. Let R be a normal domain, and let S R be a multiplicative subset.
Suppose g is an
is integral over S 1 R.
Pelementj of the fraction field of R which
d
1
Let P = x + j<d aj x be a polynomial with ai S R such that P (g) = 0.
Choose
P s S such that sai R for all i. Then sg satisfies the monic polynomial
xd + j<d sdj aj xj which has coefficients sdj aj in R. Hence sg R because R

is normal. Hence g S 1 R.

00GZ

Lemma 36.6. A principal ideal domain is normal.

Proof. Let R be a principal ideal domain. Let g = a/b be an element of the
fraction field of R integral over R. Because R is a principal ideal domain we may
divide out a common factor of a and b and assume (a, b) = R. In this case, any
equation (a/b)n + rn1 (a/b)n1 + . . . + r0 = 0 with ri R would imply an (b).
This contradicts (a, b) = R unless b is a unit in R.


COMMUTATIVE ALGEBRA

00H0

Lemma 36.7. Let R be a domain with fraction field K. Suppose f =

an element of K[x].

78

i xi is

(1) If f is integral over R[x] then all i are integral over R, and
(2) If f is almost integral over R[x] then all i are almost integral over R.
Proof. We first prove the second statement. Write f = 0 + 1 x + . . . + r xr with
r 6= 0. By assumption there exists h = b0 + b1 x + . . . + bs xs R[x], bs 6= 0 such
that f n h R[x] for all n 0. This implies that bs rn R for all n 0. Hence r
is almost integral over R. Since the set of almost integral elements form a subring
(Lemma 36.4) we deduce that f r xr = 0 + 1 x + . . . + r1 xr1 is almost
integral over R[x]. By induction on r we win.
In order to prove the first statement we will usePabsolute Noetherian reduction.
Namely, write i = ai /bi and let P (t) = td + j<d fj tj be a polynomial with
P
coefficients fj R[x] such that P (f ) = 0. Let fj =
fji xi . Consider the subring
R0 R generated by the finite list of elements ai , bi , fji of R. It is a domain; let
K0 be its field of fractions. Since R0 is a finite type Z-algebra it is Noetherian, see
Lemma 30.3. It is still the case that f K0 [x] is integral over R0 [x], because all
the identities in R among the elements ai , bi , fji also hold in R0 . By Lemma 36.4
the element f is almost integral over R0 [x]. By the second statement of the lemma,
the elements i are almost integral over R0 . And since R0 is Noetherian, they are
integral over R0 , see Lemma 36.4. Of course, then they are integral over R.

030A

Lemma 36.8. Let R be a normal domain. Then R[x] is a normal domain.

Proof. The result is true if R is a field K because K[x] is a euclidean domain
and hence a principal ideal domain and hence normal by Lemma 36.6. Let g be
an element of the fraction field of R[x] which is integral over R[x]. Because g is
integral over K[x] where K is the fraction field of R we may write g = d xd +
d1 xd1 + . . . + 0 with i K. By Lemma 36.7 the elements i are integral over
R and hence are in R.


0BI0

Lemma 36.9. Let R be a Noetherian normal domain. Then R[[x]] is a Noetherian

normal domain.
Proof. The power series ring is Noetherian by Lemma 30.2. Let f, g R[[x]] be
nonzero elements such that w = f /g is integral over R[[x]]. Let K be the fraction
field of R. Since the ring of Laurent series K((x)) = K[[x]][1/x] is a field, we can
write w = an xn + an+1 xn+1 + . . .) for some n Z, ai K, and an 6= 0. By Lemma
36.4 we see there exists a nonzero element h = bm xm + bm+1 xm+1 + . . . in R[[x]]
with bm 6= 0 such that we h R[[x]] for all e 1. We conclude that n 0 and that
bm aen R for all e 1. Since R is Noetherian this implies that an R by the same
lemma. Now, if an , an+1 , . . . , aN 1 R, then we can apply the same argument to
w an xn . . . aN 1 xN 1 = aN xN + . . .. In this way we see that all ai R and
the lemma is proved.


030B

Lemma 36.10. Let R be a domain. The following are equivalent:

(1) The domain R is a normal domain,
(2) for every prime p R the local ring Rp is a normal domain, and
(3) for every maximal ideal m the ring Rm is a normal domain.

COMMUTATIVE ALGEBRA

79

Proof. This follows easily from the fact that for any domain R we have
\
R=
Rm
m

inside the fraction field of R. Namely, if g is an element of the right hand side then
the ideal I = {x R | xg R} is not contained in any maximal ideal m, whence
I = R.

Lemma 36.10 shows that the following definition is compatible with Definition 36.1.
(It is the definition from EGA see [DG67, IV, 5.13.5 and 0, 4.1.4].)
00GV

Definition 36.11. A ring R is called normal if for every prime p R the localization Rp is a normal domain (see Definition 36.1).
Note that a normal ring is a reduced ring, as R is a subring of the product of its
localizations at all primes (see for example Lemma 23.1).

034M

Lemma 36.12. A normal ring is integrally closed in its total ring of fractions.
Proof. Let R be a normal ring. Let x Q(R) be an element of the total ring of
fractions of R integral over R. Set I = {f R, f x R}. Let p R be a prime.
As R Rp is flat we see that Rp Q(R) Rp . As Rp is a normal domain we
see that x 1 is an element of Rp . Hence we can find a, f R, f 6 p such that
x 1 = a 1/f . This means that f x a maps to zero in Q(R) R Rp = Q(R)p ,
which in turn means that there exists an f 0 R, f 0 6 p such that f 0 f x = f 0 a in
R. In other words, f f 0 I. Thus I is an ideal which isnt contained in any of the
prime ideals of R, i.e., I = R and x R.


037C

Lemma 36.13. A localization of a normal ring is a normal ring.

Proof. Omitted.

00H1

Lemma 36.14. Let R be a normal ring. Then R[x] is a normal ring.

Proof. Let q be a prime of R[x]. Set p = R q. Then we see that Rp [x] is a normal
domain by Lemma 36.8. Hence (R[x])q is a normal domain by Lemma 36.5.


030C

Lemma 36.15. Let R be a ring. Assume R is reduced and has finitely many
minimal primes. Then the following are equivalent:
(1) R is a normal ring,
(2) R is integrally closed in its total ring of fractions, and
(3) R is a finite product of normal domains.
Proof. Let q1 , . . . , qt be the minimal primes of R. By Lemmas 24.2 and 24.4 we
have Q(R) = Rq1 . . . Rqt , and by Lemma 24.1 each factor is a field. Denote
ei = (0, . . . , 0, 1, 0, . . . , 0) the ith idempotent of Q(R).
If R is integrally closed in Q(R), then it contains in particular the idempotents ei ,
and we see that R is a product of t domains (see Sections 21 and 22). Hence it is
clear that R is a finite product of normal domains.
If R is normal, then it is clear that ei Rp for every prime ideal p of R. Hence
we see that R contains the elements ei (see proof of Lemma 36.10). We conclude
that R is a product of t domains as before. Each of these t domains is normal by
Lemma 36.10 and the assumption that R is a normal ring. Hence it follows that R
is a finite product of normal domains.
We omit the verification that (3) implies (1) and (2).

COMMUTATIVE ALGEBRA

037D

80

Lemma 36.16. Let (Ri , ii0 ) be a directed system (Categories, Definition 8.2) of
rings. If each Ri is a normal ring so is R = colimi Ri .
Proof. Let p R be a prime ideal. Set pi = Ri p (usual abuse of notation). Then
we see that Rp = colimi (Ri )pi . Since each (Ri )pi is a normal domain we reduce
to proving the statement of the lemma for normal domains. If a, b R and a/b
satisfies a monic polynomial P (T ) R[T ], then we can find a (sufficiently large)
i I such that a, b, P all come from objects ai , bi , Pi over Ri . Since Ri is normal
we see ai /bi Ri and hence also a/b R.

37. Going down for integral over normal

037E

We first play around a little bit with the notion of elements integral over an ideal,
and then we prove the theorem referred to in the section title.

00H2

Definition 37.1. Let : R S be a ring map. Let I R be an ideal.

We say anPelement g S is integral over I if there exists a monic polynomial
P = xd + j<d aj xj with coefficients aj I dj such that P (g) = 0 in S.
This is mostly used when = idR : R R. In this case the set I 0 of elements
integral over I is called the integral closure of I. We will see that I 0 is an ideal of
R (and of course I I 0 ).

00H3

Lemma
P 37.2. Let : R S be a ring map. Let I R be an ideal. Let
A = I n tn R[t] be the subring of the polynomial ring generated by RIt R[t].
An element s S is integral over I if and only if the element st S[t] is integral
over A.
P
Proof. Suppose st is integral over A. Let P = xd + j<d aj xj be a monic polynomial with coefficients in A such that P (st) = 0. Let a0j A be the degree d j
part of ai , in other words a0j = a00j tdj with a00j I dj . For degree reasons we still
P
have (st)d + j<d (a00j )tdj (st)j = 0. Hence we see that s is integral over I.
P
Suppose that s is integral over I. Say P = xd + j<d aj xj with aj I dj . The
P
we immediately find a polynomial Q = xd + j<d (aj tdj )xj with coefficients in A
which proves that st is integral over A.


00H4

Lemma 37.3. Let : R S be a ring map. Let I R be an ideal. The set of

elements of S which are integral over I form a R-submodule of S. Furthermore, if
s S is integral over R, and s0 is integral over I, then ss0 is integral over I.
Proof. Closure under addition is clear from the characterization of Lemma 37.2.
Any element s S which is integral over R corresponds to the degree 0 element s
of S[x] which is integral over A (because R A). Hence we see that multiplication
by s on S[x] preserves the property of being integral over A, by Lemma 35.7. 

00H5

Lemma 37.4. Suppose : R S is integral. Suppose I R is an ideal. Then

every element of IS is integral over I.
Proof. Immediate from Lemma 37.3.

00H6

Lemma 37.5. Let R be a domain with field of fractions K. Let n, m N and

a0 , . . . , an1 , b0 , . . . , bm1 R. If the polynomial xn + an1 xn1 + . . . + a0 divides
the polynomial xm + bm1 xm1 + . . . + b0 in K[x] then

COMMUTATIVE ALGEBRA

81

(1) a0 , . . . , an1 are integral over the subring of R generated by b0 , . . . , bm1 ,

and
p
(2) each ai lies in (b0 , . . . , bm ).
Proof. Let K R be the fraction field of R. Q
Let L K be a field extension such
m
that we can write xm + bm1 xm1 + . . . + b0 = i=1 (x i ) with i L. Each i is
integral over the subring generated by b0 , . . . , bm1 . Since each ai is a homogeneous
polynomial in 1 , . . . , m we deduce the same for the ai .
Choose c0 , . . . , cmn1 K such that
xm + bm1 xm1 + . . . + b0 =
(xn + an1 xn1 + . . . + a0 )(xmn + cmn1 xmn1 + . . . + c0 ).
By the first part we see that the elements ci are integral over R. Let R0 be the
sub R-algebrapof K generated by
we
pc0 , . . . , cmn1 . By Lemmas 35.15 and 29.3
0
0
bm ). Thus we may replace R by R and
see that R (b0 , . . . , bm )R = (b0 , . . . ,p
assume ci R. Dividing out the radical (b0 , . . . , bm ) we get a reduced ring R.
We have to show that the images ai R are zero. And in R[x] we have the relation
xm = xm + bm1 xm1 + . . . + b0 =
(x + an1 x
+ . . . + a0 )(xmn + cmn1 xmn1 + . . . + c0 ).
n

n1

It is easy to see that this implies ai = 0 for all i. For example one can see this by
localizing at all the minimal primes, see Lemma 24.2.

00H7

Lemma 37.6. Let R S be an inclusion of domains. Assume R is normal. Let

g S be integral over R. Then the minimal polynomial of g has coefficients in R.
Proof. Let P = xm + bm1 xm1 + . . . + b0 be a polynomial with coefficients in R
such that P (g) = 0. Let Q = xn + an1 xn1 + . . . + a0 be the minimal polynomial
for g over the fraction field K of R. Then Q divides P in K[x]. By Lemma 37.5 we
see the ai are integral over R. Since R is normal this means they are in R.


00H8

Proposition 37.7. Let R S be an inclusion of domains. Assume R is normal

and S integral over R. Let p p0 R be primes. Let q0 be a prime of S with
p0 = R q0 . Then there exists a prime q with q q0 such that p = R q. In other
words: the going down property holds for R S, see Definition 40.1.
Proof. Let p, p0 and q0 be as in the statement. We have to show there is a prime
q, q q0 such that R q = p. This is the same as finding a prime of Sq0 mapping to
p. According to Lemma 16.9 we have to show that pSq0 R = p. Pick z pSq0 R.
We may write z = y/g with y pS and g S, g 6 q0 . Written differently we have
zg = y.
By Lemma 37.4 there exists a monic polynomial P = xm + bm1 xm1 + . . . + b0
with bi p such that P (y) = 0.
By Lemma 37.6 the minimal polynomial of g over K has coefficients in R. Write
it as Q = xn + an1 xn1 + . .P
. + a0 . Note that not all ai , i = n 1, . . . , 0 are in p
since that would imply g n = j<n aj g j pS p0 S q0 which is a contradiction.
Since y = zg we see immediately from the above that Q0 = xn + zan1 xn1 +
. . . + z n a0 is the minimal polynomial
for y. Hence Q0 divides P and by Lemma
p
37.5 we see that z j anj (b0 , . . . , bm1 ) p, j = 1, . . . , n. Because not all ai ,
i = n 1, . . . , 0 are in p we conclude z p as desired.


COMMUTATIVE ALGEBRA

82

38. Flat modules and flat ring maps

00H9

One often used result is that if M = colimiI Mi is a colimit of R-modules and if

N is an R-module then
M N = colimiI Mi R N,
see Lemma 11.9. This property is usually expressed by saying that commutes
with colimits. Another often used result is that if 0 N1 N2 N3 0 is an
exact sequence and if M is any R-module, then
M R N1 M R N2 M R N3 0
is still exact, see Lemma 11.10. Both of these properties tell us that the functor
N 7 M R N is right exact. See Categories, Section 23 and Homology, Section 7.
An R-module M is flat if N 7 N R M is also left exact, i.e., if it is exact. Here
is the precise definition.

00HB

Definition 38.1. Let R be a ring.

(1) An R-module M is called flat if whenever N1 N2 N3 is an exact
sequence of R-modules the sequence M R N1 M R N2 M R N3 is
exact as well.
(2) An R-module M is called faithfully flat if the complex of R-modules N1
N2 N3 is exact if and only if the sequence M R N1 M R N2
M R N3 is exact.
(3) A ring map R S is called flat if S is flat as an R-module.
(4) A ring map R S is called faithfully flat if S is faithfully flat as an Rmodule.
Here is an example of how you can use the flatness condition.

0BBY

Lemma 38.2. Let R be a ring. Let I, J R be ideals. Let M be a flat R-module.

Then IM JM = (I J)M .
Proof. Consider the exact sequence 0 I J R R/I R/J. Tensoring with
the flat module M we obtain an exact sequence
0 (I J) R M M M/IM M/JM
Since the kernel of M M/IM M/JM is equal to IM JM we conclude.

05UT

Lemma 38.3. Let R be a ring. Let {Mi , ii0 } be a directed system of flat Rmodules. Then colimi Mi is a flat R-module.
Proof. This follows as commutes with colimits and because directed colimits
are exact, see Lemma 8.9.


00HC

Lemma 38.4. A composition of (faithfully) flat ring maps is (faithfully) flat. If

R R0 is (faithfully) flat, and M 0 is a (faithfully) flat R0 -module, then M 0 is a
(faithfully) flat R-module.
Proof. The first statement of the lemma is a particular case of the second, so it is
clearly enough to prove the latter. Let R R0 be a flat ring map, and M 0 a flat R0 module. We need to prove that M 0 is a flat R-module. Let N1 N2 N3 be an
exact complex of R-modules. Then, the complex R0 R N1 R0 R N2 R0 R N3
is exact (since R0 is flat as an R-module), and so the complex M 0 R0 (R0 R N1 )
M 0 R0 (R0 R N2 ) M 0 R0 (R0 R N3 ) is exact (since M 0 is a flat R0 -module).

COMMUTATIVE ALGEBRA

83

Since M 0 R0 (R0 R N )
= (M 0 R0 R0 ) R N
= M 0 R N for any R-module N
functorially (by Lemmas 11.7 and 11.3), this complex is isomorphic to the complex
M 0 R N1 M 0 R N2 M 0 R N3 , which is therefore also exact. This shows
that M 0 is a flat R-module. Tracing this argument backwards, we can show that if
R R0 is faithfully flat, and if M 0 is faithfully flat as an R0 -module, then M 0 is
faithfully flat as an R-module.

00HD
00HE
00HF
00HG
00HH

Lemma 38.5. Let M be an R-module. The following are equivalent:

(1) M is flat over R.
(2) for every injection of R-modules N N 0 the map N R M N 0 R M is
injective.
(3) for every ideal I R the map I R M R R M = M is injective.
(4) for every finitely generated ideal I R the map I R M R R M = M
is injective.
Proof. The implications (1) implies (2) implies (3) implies (4) are all trivial. Thus
we prove (4) implies (1). Suppose that N1 N2 N3 is exact. Let K =
Ker(N2 N3 ) and Q = Im(N2 N3 ). Then we get maps
N1 R M K R M N2 R M Q R M N3 R M
Observe that the first and third arrows are surjective. Thus if we show that the
second and fourth arrows are injective, then we are done2. Hence it suffices to show
that R M transforms injective R-module maps into injective R-module maps.
Assume K N is an injective R-module map and let x Ker(KR M N R M ).
We have to show that x is zero. The R-module K is the union of its finite Rsubmodules; hence, K R M is the colimit of R-modules of the form Ki R M where
Ki runs over all finite R-submodules of K (because tensor product commutes with
colimits). Thus, for some i our x comes from an element xi Ki R M . Thus we
may assume that K is a finite R-module. Assume this. We regard the injection
K N as an inclusion, so that K N .
The R-module N is the union of its finite R-submodules that contain K. Hence,
N R M is the colimit of R-modules of the form Ni R M where Ni runs over all
finite R-submodules of N that contain K (again since tensor product commutes
with colimits). Notice that this is a colimit over a directed system (since the sum
of two finite submodules of N is again finite). Hence, (by Lemma 8.5) the element
x K R M maps to zero in at least one of these R-modules Ni R M (since x
maps to zero in N R M ). Thus we may assume N is a finite R-module.
Assume N is a finite R-module. Write N = Rn /L and K = L0 /L for some
L L0 Rn . For any R-submodule G Rn , we have a canonical map G R
M M n obtained by composing G R M Rn R M = M n . It suffices to
2Here is the argument in more detail: Assume that we know that the second and fourth arrows
are injective. Lemma 11.10 (applied to the exact sequence K N2 Q 0) yields that the
sequence K R M N2 R M Q R M 0 is exact. Hence, Ker (N2 R M Q R M ) =
Im (K R M N2 R M ). Since Im (K R M N2 R M ) = Im (N1 R M N2 R M )
(due to the surjectivity of N1 R M K R M ) and Ker (N2 R M Q R M ) =
Ker (N2 R M N3 R M ) (due to the injectivity of Q R M N3 R M ), this becomes Ker (N2 R M N3 R M ) = Im (N1 R M N2 R M ), which shows that the functor
R M is exact, whence M is flat.

COMMUTATIVE ALGEBRA

84

prove that L R M M n and L0 R M M n are injective. Namely, if so,

then we see that K R M = L0 R M/L R M M n /L R M is injective too3.
Thus it suffices to show that L R M M n is injective when L Rn is an
R-submodule. We do this by induction on n. The base case n = 1 we handle below.
For the induction step assume n > 1 and set L0 = L R 0n1 . Then L00 = L/L0
is a submodule of Rn1 . We obtain a diagram

L0 R M

/ L R M

/ L00 R M

/0


/M


/ M n


/ M n1

/0

By induction hypothesis and the base case the left and right vertical arrows are
injective. The rows are exact. It follows that the middle vertical arrow is injective
too.
The base case of the induction above is when L R is an ideal. In other words,
we have to show that I R M M is injective for anySideal I of R. We know
this is true when I is finitely generated. However, I = I is the union of the
finitely generated ideals I contained in it. In other words, I = colim I . Since
commutes with colimits we see that I R M = colim I R M and since all
the morphisms I R M M are injective by assumption, the same is true for
I R M M .

05UU

Lemma 38.6. Let {Ri , ii0 } be a system of rings over the directed partially ordered
set I. Let R = colimi Ri . Let M be an R-module such that M is flat as an Ri module for all i. Then M is flat as an R-module.
Proof. Let a R be a finitely generated ideal. By Lemma 38.5 it suffices to show
that a R M M is injective. We can find an i I and a finitely generated ideal
a0 Ri such that a = a0 R. Then a = colimi0 i a0 Ri0 . Hence the map a R M M
is the colimit of the maps
a0 Ri0 Ri0 M M
which are all injective by assumption. Since commutes with colimits and since
colimits over I are exact by Lemma 8.9 we win.


00HI

Lemma 38.7. Suppose that M is flat over R, and that R R0 is a ring map.
Then M R R0 is flat over R0 .
Proof. For any R0 -module N we have a canonical isomorphism N R0 (R0 R M ) =
N R M . Hence the exactness of R0 (R0 R M ) follows from the exactness of
R M .


00HJ

Lemma 38.8. Let R R0 be a faithfully flat ring map. Let M be a module over
R, and set M 0 = R0 R M . Then M is flat over R if and only if M 0 is flat over
R0 .
3This becomes obvious if we identify L0 M and L M with submodules of M n (which
R
R
is legitimate since the maps L R M M n and L0 R M M n are injective and commute
with the obvious map L0 R M L R M ).

COMMUTATIVE ALGEBRA

85

Proof. By Lemma 38.7 we see that if M is flat then M 0 is flat. For the converse,
suppose that M 0 is flat. Let N1 N2 N3 be an exact sequence of R-modules.
We want to show that N1 R M N2 R M N3 R M is exact. We know that
N1 R R0 N2 R R0 N3 R R0 is exact, because R R0 is flat. Flatness of M 0
implies that N1 R R0 R0 M 0 N2 R R0 R0 M 0 N3 R R0 R0 M 0 is exact. We
may write this as N1 R M R R0 N2 R M R R0 N3 R M R R0 . Finally,
faithful flatness implies that N1 R M N2 R M N3 R M is exact.

0584

Lemma 38.9. Let R be a ring. Let S S 0 be a faithfully flat map of R-algebras.

Let M be a module over S, and set M 0 = S 0 S M . Then M is flat over R if and
only if M 0 is flat over R.
Proof. Let N N 0 be an injection of R-modules. By the faithful flatness of
S S 0 we have
Ker(N R M N 0 R M ) S S 0 = Ker(N R M 0 N 0 R M 0 )
Hence the equivalence of the lemma follows from the second characterization of
flatness in Lemma 38.5.


039V

Lemma 38.10. Let R S be a ring map. Let M be an S-module. If M is flat

as an R-module and faithfully flat as an S-module, then R S is flat.
Proof. Let N1 N2 N3 be an exact sequence of R-modules. By assumption
N1 R M N2 R M N3 R M is exact. We may write this as
N1 R S S M N2 R S S M N3 R S S M.
By faithful flatness of M over S we conclude that N1 R S N2 R S N3 R S
is exact. Hence R S is flat.

P
Let R be a ring.P
Let M be an R-module. Let
fi xi = 0 be a relation in M . We
say the relation
fi xi is trivial if there exist an integer m 0, elements yj M ,
j = 1, . . . , m, and elements aij R, i = 1, . . . , n, j = 1, . . . , m such that
X
X
xi =
aij yj , i, and 0 =
fi aij , j.
j

00HK

Lemma 38.11 (Equational criterion of flatness). A module M over R is flat if

and only if every relation in M is trivial.
P
Proof. Assume M is flat and let
fi xi = 0 be a relation
in M . Let I =
P
(f1 , . . . , fn ), and let K = Ker(Rn I, (a1 , . . . , an ) 7P i ai fi ). So we have the
short exact sequence 0 K Rn I 0. Then
fi P
xi is an element of
I R M which maps to zero in R R M = M . By flatnessP fi xi is zero in
n
I R M . Thus there exists
R M .
P an element of K R M mapping to ei xi n R P
Write this element as
kj yj and then write the image of kj in R as
aij ei
to get the result.
Assume
every relation is trivial, let I be a finitely generated ideal, and let x =
P
fi xi be an element
of I R M mapping to zero in R R M = M . This just
P
means exactly that
fi xi is a relation in M . And the fact that it is trivial implies
easily that x is zero, because
X
 X X

X
X
x=
fi xi =
fi
aij yj =
fi aij yj = 0


COMMUTATIVE ALGEBRA

00HL

86

Lemma 38.12. Suppose that R is a ring, 0 M 00 M 0 M 0 a short exact

sequence, and N an R-module. If M is flat then N R M 00 N R M 0 is injective,
i.e., the sequence
0 N R M 00 N R M 0 N R M 0
is a short exact sequence.
Proof. Let R(I) N be a surjection from a free module onto N with kernel K.
The result follows from the snake lemma applied to the following diagram

M 00 R N

(M 00 )(I)

M 00 R K

M 0 R N

(M 0 )(I)

M 0 R K

M R N

M (I)

M R K

0
0

with exact rows and columns. The middle row is exact because tensoring with the
free module R(I) is exact.

00HM

Lemma 38.13. Suppose that 0 M 0 M M 00 0 is a short exact sequence

of R-modules. If M 0 and M 00 are flat so is M . If M and M 00 are flat so is M 0 .
Proof. We will use the criterion that a module N is flat if for every ideal I R the
map N R I N is injective, see Lemma 38.5. Consider an ideal I R. Consider
the diagram
0

M0

M 0 R I

M R I

M 00

M 00 R I

with exact rows. This immediately proves the first assertion. The second follows
because if M 00 is flat then the lower left horizontal arrow is injective by Lemma
38.12.

00HO

Lemma 38.14. Let R be a ring. Let M be an R-module. The following are

equivalent
(1) M is faithfully flat, and
(2) M is flat and for all R-module homomorphisms : N N 0 we have = 0
if and only if idM = 0.
Proof. If M is faithfully flat, then 0 Ker() N 0 is exact if and only
if the same holds after tensoring with M . This proves (1) implies (2). For the
other, assume (2). Let N1 N2 N3 be a complex, and assume the complex
N1 R M N2 R M N3 R M is exact. Take x Ker(N2 N3 ), and consider
the map : R N2 /Im(N1 ), r 7 rx + Im(N1 ). By the exactness of the complex
R M we see that idM is zero. By assumption we get that is zero. Hence
x is in the image of N1 N2 .


00HP

Lemma 38.15. Let M be a flat R-module. The following are equivalent:

(1) M is faithfully flat,

COMMUTATIVE ALGEBRA

87

(2) for all p Spec(R) the tensor product M R (p) is nonzero, and
(3) for all maximal ideals m of R the tensor product M R (m) = M/mM is
nonzero.
Proof. Assume M faithfully flat. Since R (p) is not zero we deduce that
M M R (p) is not zero, see Lemma 38.14.
Conversely assume that M is flat and that M/mM is never zero. Suppose that
N1 N2 N3 is a complex and suppose that N1 R M N2 R M N3 R M is
exact. Let H be the cohomology of the complex, so H = Ker(N2 N3 )/Im(N1
N2 ). By flatness we see that H R M = 0. Take x H and let I = {f R | f x = 0}
be its annihilator. Since R/I H we get M/IM H R M = 0 by flatness of
M . If I 6= R we may choose a maximal ideal I m R. This immediately gives
a contradiction.

00HQ

Lemma 38.16. Let R S be a flat ring map. The following are equivalent:
(1) R S is faithfully flat,
(2) the induced map on Spec is surjective, and
(3) any closed point x Spec(R) is in the image of the map Spec(S)
Spec(R).
Proof. This follows quickly from Lemma 38.15, because we saw in Remark 16.8
that p is in the image if and only if the ring S R (p) is nonzero.


00HR

Lemma 38.17. A flat local ring homomorphism of local rings is faithfully flat.
Proof. Immediate from Lemma 38.16.

00HS

Lemma 38.18. Let R S be flat. Let p p0 be primes of R. Let q0 S be a

prime of S mapping to p0 . Then there exists a prime q q0 mapping to p.
Proof. Namely, consider the flat local ring map Rp0 Sq0 . By Lemma 38.17 this
is faithfully flat. By Lemma 38.16 there is a prime mapping to pRp0 . The inverse
image of this prime in S does the job.

The property of R S described in the lemma is called the going down property.
See Definition 40.1. We finish with some remarks on flatness and localization.

00HT

Lemma 38.19. Let R be a ring. Let S R be a multiplicative subset.

(1) The localization S 1 R is a flat R-algebra.
(2) If M is a S 1 R-module, then M is a flat R-module if and only if M is a
flat S 1 R-module.
(3) Suppose M is an R-module. Then M is a flat R-module if and only if Mp
is a flat Rp -module for all primes p of R.
(4) Suppose M is an R-module. Then M is a flat R-module if and only if Mm
is a flat Rm -module for all maximal ideals m of R.
(5) Suppose R A is a ring map, M is an A-module, and g1 , . . . , gm A are
elements generating the unit ideal of A. Then M is flat over R if and only
if each localization Mgi is flat over R.
(6) Suppose R A is a ring map, and M is an A-module. Then M is a flat
R-module if and only if the localization Mq is a flat Rp -module (with p the
prime of R lying under q) for all primes q of A.

COMMUTATIVE ALGEBRA

88

(7) Suppose R A is a ring map, and M is an A-module. Then M is a

flat R-module if and only if the localization Mm is a flat Rp -module (with
p = R m) for all maximal ideals m of A.
Proof. Let us prove the last statement of the lemma. In the proof we will use
repeatedly that localization is exact and commutes with tensor product, see Sections
9 and 11.
Suppose R A is a ring map, and M is an A-module. Assume that Mm is a flat
Rp -module for all maximal ideals m of A (with p = R m). Let I R be an ideal.
We have to show the map I R M M is injective. We can think of this as a
map of A-modules. By assumption the localization (I R M )m Mm is injective
because (I R M )m = Ip Rp Mm . Hence the kernel of I R M M is zero by
Lemma 23.1. Hence M is flat over R.
Conversely, assume M is flat over R. Pick a prime q of A lying over the prime p
of R. Suppose that I Rp is an ideal. We have to show that I Rp Mq Mq is
injective. We can write I = Jp for some ideal J R. Then the map I Rp Mq Mq
is just the localization (at q) of the map J R M M which is injective. Since
localization is exact we see that Mq is a flat Rp -module.
This proves (7) and (6). The other statements follow in a straightforward way from
the last statement (proofs omitted).

090N

Lemma 38.20. Let R be a ring. Let {Si , ii0 } be a directed system of faithfully
flat R-algebras. Then S = colimi Si is a faithfully flat R-algebra.
Proof. By Lemma 38.3 we see that S is flat. Let m R be a maximal ideal. By
Lemma 38.16 none of the rings Si /mSi is zero. Hence S/mS = colim Si /mSi is
nonzero either as 1 is not equal to zero. Thus the image of Spec(S) Spec(R)
contains m and we see that R S is faithfully flat by Lemma 38.16.

39. Supports and annihilators

080S

Some very basic definitions and lemmas.

00L1

Definition 39.1. Let R be a ring and let M be an R-module. The support of M

is the set
Supp(M ) = {p Spec(R) | Mp 6= 0}

0585

Lemma 39.2. Let R be a ring. Let M be an R-module. Then

M = (0) Supp(M ) = .
Proof. Actually, Lemma 23.1 even shows that Supp(M ) always contains a maximal
ideal if M is not zero.


07T7

Definition 39.3. Let R be a ring. Let M be an R-module.

(1) Given an element m M the annihilator of m is the ideal
AnnR (m) = Ann(m) = {f R | f m = 0}.
(2) The annihilator of M is the ideal
AnnR (M ) = Ann(M ) = {f R | f m = 0 m M }.

COMMUTATIVE ALGEBRA

07T8

89

Lemma 39.4. Let R S be a flat ring map. Let M be an R-module and

m M . Then AnnR (m)S = AnnS (m 1). If M is a finite R-module, then
AnnR (M )S = AnnS (M R S).
Proof. Set I = AnnR (m). By definition there is an exact sequence 0 I
R M where the map R M sends f to f m. Using flatness we obtain an
exact sequence 0 I R S S M R S which proves
T the first assertion. If
m1 , . . . , mn is a setTof generators of M then AnnR (M ) = AnnR (mi ). Similarly
AnnST
(M R S) = Ann
T S (mi 1). Set Ii = AnnR (mi ). Then it suffices to show
that i=1,...,n (Ii S) = ( i=1,...,n Ii )S. This is Lemma 38.2.


00L2

Lemma 39.5. Let R be a ring and let M be an R-module. If M is finite, then

Supp(M ) is closed. More precisely, if I = Ann(M ) is the annihilator of M , then
V (I) = Supp(M ).
Proof. We will show that V (I) = Supp(M ).
Suppose p Supp(M ). Then Mp 6= 0. Hence by Nakayamas Lemma 19.1 we have
M R (p) 6= 0. Hence I p.
Conversely, suppose that p 6 Supp(M ). Then Mp = 0. Let x1 , . . . , xr M be
generators. By Lemma 9.9 there exists an f R, f 6 p such that xi /1 = 0 in Mf .
Hence f ni xi = 0 for some ni 1. Hence f n M = 0 for n = max{ni } as desired. 

07Z5

Lemma 39.6. Let R be a ring, let M be an R-module, and let x M . Then

p V (Ann(m)) if and only if x does not map to zero in Mp .
Proof. We may replace M by Rm M . Then (1) Ann(m) = Ann(M ) and (2) x
does not map to zero in Mp if and only if p Supp(M ). The result now follows
from Lemma 39.5.


051B

Lemma 39.7. Let R be a ring and let M be an R-module. If M is a finitely

presented R-module, then Supp(M ) is a closed subset of Spec(R) whose complement
is quasi-compact.
Proof. Choose a presentation
Rm Rn M 0
Let A Mat(n m, R) be the matrix of the first map. By Nakayamas Lemma
19.1 we see that
Mp 6= 0 M (p) 6= 0 rank(A mod p) < n.
Hence, if I is the ideal of R generated by the n n minors of A, then Supp(M ) =
V (I). Since I is finitely generated, say I = (f1 , . . . , ft ), we see that Spec(R) \ V (I)
is a finite union of the standard opens D(fi ), hence quasi-compact.


00L3

Lemma 39.8. Let R be a ring and let M be an R-module.

(1)
(2)
(3)
(4)

If M is finite then the support of M/IM is Supp(M ) V (I).

If N M , then Supp(N ) Supp(M ).
If Q is a quotient module of M then Supp(Q) Supp(M ).
If 0 N M Q 0 is a short exact sequence then Supp(M ) =
Supp(Q) Supp(N ).

COMMUTATIVE ALGEBRA

90

Proof. The functors M 7 Mp are exact. This immediately implies all but the first
assertion. For the first assertion we need to show that Mp 6= 0 and I p implies
(M/IM )p = Mp /IMp 6= 0. This follows from Nakayamas Lemma 19.1.

40. Going up and going down
00HU

Suppose p, p0 are primes of the ring R. Let X = Spec(R) with the Zariski topology.
Denote x X the point corresponding to p and x0 X the point corresponding to
p0 . Then we have:
x0
x p0 p.
In words: x is a specialization of x0 if and only if p0 p. See Topology, Section 18
for terminology and notation.

00HV

Definition 40.1. Let : R S be a ring map.

(1) We say a : R S satisfies going up if given primes p p0 in R and a
prime q in S lying over p there exists a prime q0 of S such that (a) q q0 ,
and (b) q0 lies over p0 .
(2) We say a : R S satisfies going down if given primes p p0 in R and a
prime q0 in S lying over p0 there exists a prime q of S such that (a) q q0 ,
and (b) q lies over p.
So far we have see the following cases of this:
(1) An integral ring map satisfies going up, see Lemma 35.20.
(2) As a special case finite ring maps satisfy going up.
(3) As a special case quotient maps R R/I satisfy going up.
(4) A flat ring map satisfies going down, see Lemma 38.18
(5) As a special case any localization satisfies going down.
(6) An extension R S of domains, with R normal and S integral over R
satisfies going down, see Proposition 37.7.
Here is another case where going down holds.

0407

Lemma 40.2. Let R S be a ring map. If the induced map : Spec(S)

Spec(R) is open, then R S satisfies going down.
Proof. Suppose that p p0 R and q0 S lies over p0 . As is
g S, g 6 q0 we see that p is in the image of D(g) Spec(S).
Sg R (p) is not zero. Since Sq0 is the directed colimit of these
that Sq0 R (p) is not zero, see Lemmas 9.9 and 11.9. Hence p is
Spec(Sq0 ) Spec(R) as desired.

00HW

open, for every

In other words
Sg this implies
in the image of


Lemma 40.3. Let R S be a ring map.

(1) R S satisfies going down if and only if generalizations lift along the map
Spec(S) Spec(R), see Topology, Definition 18.3.
(2) R S satisfies going up if and only if specializations lift along the map
Spec(S) Spec(R), see Topology, Definition 18.3.
Proof. Omitted.

00HX

Lemma 40.4. Suppose R S and S T are ring maps satisfying going down.
Then so does R T . Similarly for going up.
Proof. According to Lemma 40.3 this follows from Topology, Lemma 18.4

COMMUTATIVE ALGEBRA

00HY

91

Lemma 40.5. Let R S be a ring map. Let T Spec(R) be the image of

Spec(S). If T is stable under specialization, then T is closed.
Proof. We give two proofs.
First proof. Let p R be a prime ideal such that the corresponding point of
Spec(R) is in the closure of T . This means that for ever f R, f 6 p we have
D(f ) T 6= . Note that D(f ) T is the image of Spec(Sf ) in Spec(R). Hence
we conclude that Sf 6= 0. In other words, 1 6= 0 in the ring Sf . Since Sp is the
directed limit of the rings Sf we conclude that 1 6= 0 in Sp . In other words, Sp 6= 0
and considering the image of Spec(Sp ) Spec(S) Spec(R) we see there exists a
p0 T with p0 p. As we assumed T closed under specialization we conclude p is
a point of T as desired.
Second proof. Let I = Ker(R S). We may replace R by R/I. In this case the
ring map R S is injective. By Lemma 29.5 all the minimal primes of R are
contained in the image T . Hence if T is stable under specialization then it contains
all primes.


00HZ

Lemma 40.6. Let R S be a ring map. The following are equivalent:

(1) Going up holds for R S, and
(2) the map Spec(S) Spec(R) is closed.
Proof. It is a general fact that specializations lift along a closed map of topological
spaces, see Topology, Lemma 18.6. Hence the second condition implies the first.
Assume that going up holds for R S. Let V (I) Spec(S) be a closed set.
We want to show that the image of V (I) in Spec(R) is closed. The ring map
S S/I obviously satisfies going up. Hence R S S/I satisfies going up, by
Lemma 40.4. Replacing S by S/I it suffices to show the image T of Spec(S) in
Spec(R) is closed. By Topology, Lemmas 18.2 and 18.5 this image is stable under
specialization. Thus the result follows from Lemma 40.5.


00I0

Lemma 40.7. Let R be a ring. Let E Spec(R) be a constructible subset.

(1) If E is stable under specialization, then E is closed.
(2) If E is stable under generalization, then E is open.
Proof. First proof. The first assertion follows from Lemma 40.5 combined with
Lemma 28.3. The second follows because the complement of a constructible set is
constructible (see Topology, Lemma 14.2), the first part of the lemma and Topology,
Lemma 18.2.
Second proof. Since Spec(R) is a spectral space by Lemma 25.2 this is a special
case of Topology, Lemma 22.5.


00I1

Proposition 40.8. Let R S be flat and of finite presentation. Then Spec(S)

Spec(R) is open. More generally this holds for any ring map R S of finite
presentation which satisfies going down.
Proof. Assume that R S has finite presentation and satisfies going down. It
suffices to prove that the image of a standard open D(f ) is open. Since S Sf
satisfies going down as well, we see that R Sf satisfies going down. Thus after
replacing S by Sf we see it suffices to prove the image is open. By Chevalleys
theorem (Theorem 28.9) the image is a constructible set E. And E is stable under

COMMUTATIVE ALGEBRA

92

generalization because R S satisfies going down, see Topology, Lemmas 18.2 and
18.5. Hence E is open by Lemma 40.7.

037F

Lemma 40.9. Let k be a field, and let R, S be k-algebras. Let S 0 S be a sub

k-algebra, and let f S 0 k R. In the commutative diagram
/ Spec((S 0 k R)f )

Spec((S k R)f )
'
w
Spec(R)
the images of the diagonal arrows are the same.

Proof. Let p R be in the image of the south-west arrow. This means (Lemma
16.9) that
(S 0 k R)f R (p) = (S 0 k (p))f
is not the zero ring, i.e., S 0 k (p) is not the zero ring and the image of f in it
is not nilpotent. The ring map S 0 k (p) S k (p) is injective. Hence also
S k (p) is not the zero ring and the image of f in it is not nilpotent. Hence
(S k R)f R (p) is not the zero ring. Thus (Lemma 16.9) we see that p is in the
image of the south-east arrow as desired.

037G

Lemma 40.10. Let k be a field. Let R and S be k-algebras. The map Spec(S k
R) Spec(R) is open.
Proof. Let f R k S. It suffices to prove that the image of the standard open
D(f ) is open. Let S 0 S be a finite type k-subalgebra such that f S 0 k R.
The map R S 0 k R is flat and of finite presentation, hence the image U of
Spec((S 0 k R)f ) Spec(R) is open by Proposition 40.8. By Lemma 40.9 this is
also the image of D(f ) and we win.

Here is a tricky lemma that is sometimes useful.

00EA

Lemma 40.11. Let R S be a ring map. Let p R be a prime. Assume that

(1) there exists a unique prime q S lying over p, and
(2) either
(a) going up holds for R S, or
(b) going down holds for R S and there is at most one prime of S above
every prime of R.
Then Sp = Sq .
Proof. Consider any prime q0 S which corresponds to a point of Spec(Sp ). This
means that p0 = R q0 is contained in p. Here is a picture
q0

p0

Assume (1) and (2)(a). By going up there exists a prime q00 S with q0 q00 and
q00 lying over p. By the uniqueness of q we conclude that q00 = q. In other words q0
defines a point of Spec(Sq ).

COMMUTATIVE ALGEBRA

93

Assume (1) and (2)(b). By going down there exists a prime q00 q lying over p0 .
By the uniqueness of primes lying over p0 we see that q0 = q00 . In other words q0
defines a point of Spec(Sq ).
In both cases we conclude that the map Spec(Sq ) Spec(Sp ) is bijective. Clearly
this means all the elements of S q are all invertible in Sp , in other words Sp =
Sq .

The following lemma is a generalization of going down for flat ring maps.
080T

Lemma 40.12. Let R S be a ring map. Let N be a finite S-module flat over
R. Endow Supp(N ) Spec(S) with the induced topology. Then generalizations lift
along Supp(N ) Spec(R).
Proof. The meaning of the statement is as follows. Let p p0 R be primes. Let
q0 S be a prime q0 Supp(N ) Then there exists a prime q q0 , q Supp(N )
lying over p. As N is flat over R we see that Nq0 is flat over Rp0 , see Lemma 38.19.
As Nq0 is finite over Sq0 and not zero since q0 Supp(N ) we see that Nq0 Sq0 (q0 )
is nonzero by Nakayamas Lemma 19.1. Thus Nq0 Rp0 (p0 ) is also not zero. We
conclude from Lemma 38.15 that Nq0 Rp0 (p) is nonzero. Let J Sq0 Rp0 (p)
be the annihilator of the finite nonzero module Nq0 Rp0 (p). Since J is a proper
ideal we can choose a prime q S which corresponds to a prime of Sq0 Rp0 (p)/J.
This prime is in the support of N , lies over p, and is contained in q0 as desired. 
41. Separable extensions

030I

In this section we talk about separability for nonalgebraic field extensions. This
is closely related to the concept of geometrically reduced algebras, see Definition
42.1.

030O

Definition 41.1. Let k K be a field extension.

(1) We say K is separably generated over k if there exists a transcendence basis
{xi ; i I} of K/k such that the extension k(xi ; i I) K is a separable
algebraic extension.
(2) We say K is separable over k if for every subextension k K 0 K with
K 0 finitely generated over k, the extension k K 0 is separably generated.
With this awkward definition it is not clear that a separably generated field extension is itself separable. It will turn out that this is the case, see Lemma 43.2.

030P

Lemma 41.2. Let k K be a separable field extension. For any subextension

k K 0 K the field extension k K 0 is separable.
Proof. This is direct from the definition.

030Q

Lemma 41.3. Let k K be a separably generated, and finitely generated field

extension. Set r = trdegk (K). Then there exist elements x1 , . . . , xr+1 of K such
that
(1) x1 , . . . , xr is a transcendence basis of K over k,
(2) K = k(x1 , . . . , xr+1 ), and
(3) xr+1 is separable over k(x1 , . . . , xr ).
Proof. Combine the definition with Fields, Lemma 18.1.

COMMUTATIVE ALGEBRA

04KM

94

Lemma 41.4. Let k K be a finitely generated field extension. There exists a

diagram
/ K0
KO
O
k

/ k0

where k k 0 , K K 0 are finite purely inseparable field extensions such that k 0 K 0

is a separably generated field extension.
Proof. This lemma is only interesting when the characteristic of k is p > 0. Choose
x1 , . . . , xr a transcendence basis of K over k. As K is finitely generated over k
the extension k(x1 , . . . , xr ) K is finite. Let k(x1 , . . . , xr ) Ksep K be the
subextension found in Fields, Lemma 13.6. If K = Ksep then we are done. We will
use induction on d = [K : Ksep ].
Assume that d > 1. Choose a K with = p Ksep and 6 Ksep . Let
P = T d + a1 T d1 + . . . + ad be the minimal polynomial of over k(x1 , . . . , xr ). Let
k k 0 be a finite purely inseparable extension obtained by adjoining pth roots such
1/p
1/p
that each ai is a pth power in k 0 (x1 , . . . , xr ). Such an extension exists; details
omitted. Let L be a field fitting into the diagram
KO

/L
O

k(x1 , . . . , xr )

/ k 0 (x1/p , . . . , x1/p
r )
1
1/p

1/p

We may and do assume L is the compositum of K and k 0 (x1 , . . . , xr ). Let

1/p
1/p
k 0 (x1 , . . . , xr ) Lsep L be the subextension found in Fields, Lemma 13.6.
1/p
1/p
Then Lsep is the compositum of Ksep and k 0 (x1 , . . . , xr ). The element
Lsep is a zero of the polynomial P all of whose coefficients are pth powers in
1/p
1/p
k 0 (x1 , . . . , xr ) and whose roots are pairwise distinct. By Fields, Lemma 27.2
we see that = (0 )p for some 0 Lsep . Clearly, this means that maps to
0 Lsep . In other words, we get the tower of fields
KO

Ksep ()
O

/L
O
/ Lsep

Ksep
O

/ Lsep
O

k(x1 , . . . , xr )
O

/ k 0 (x1/p , . . . , x1/p
r )
1
O

/ k0

COMMUTATIVE ALGEBRA

95

Thus this construction leads to a new situation with [L : Lsep ] < [K : Ksep ]. By
induction we can find k 0 k 00 and L L0 as in the lemma for the extension k 0 L.
Then the extensions k k 00 and K L0 work for the extension k K. This proves
the lemma.

42. Geometrically reduced algebras
05DS

The main result on geometrically reduced algebras is Lemma 43.3. We suggest the
reader skip to the lemma after reading the definition.

030S

Definition 42.1. Let k be a field. Let S be a k-algebra. We say S is geometrically

reduced over k if for every field extension k K the K-algebra K k S is reduced.
Let k be a field and let S be a reduced k algebra. To check that S is geometrically
reduced it will suffice to check that k k S is reduced (where k denotes the algebraic
closure of k). In fact it is enough to check this for finite purely inseparable field
extensions k k 0 . See Lemma 43.3.

030T

Lemma 42.2. Elementary properties of geometrically reduced algebras. Let k be

a field. Let S be a k-algebra.
(1) If S is geometrically reduced over k so is every k-subalgebra.
(2) If all finitely generated k-subalgebras of S are geometrically reduced, then
S is geometrically reduced.
(3) A directed colimit of geometrically reduced k-algebras is geometrically reduced.
(4) If S is geometrically reduced over k, then any localization of S is geometrically reduced over k.
Proof. Omitted. The second and third property follow from the fact that tensor
product commutes with colimits.


04KN

Lemma 42.3. Let k be a field. If R is geometrically reduced over k, and S R

is a multiplicative subset, then the localization S 1 R is geometrically reduced over
k. If R is geometrically reduced over k, then R[x] is geometrically reduced over k.
Proof. Omitted. Hints: A localization of a reduced ring is reduced, and localization commutes with tensor products.

In the proofs of the following lemmas we will repeatedly use the following observation: Suppose that R0 R and S 0 S are inclusions of k-algebras. Then the map
R0 k S 0 R k S is injective.

00I3

Lemma 42.4. Let k be a field. Let R, S be k-algebras.

(1) If R k S is nonreduced, then there exist finitely generated subalgebras R0
R, S 0 S such that R0 k S 0 is not reduced.
(2) If R k S contains a nonzero zerodivisor, then there exist finitely generated subalgebras R0 R, S 0 S such that R0 k S 0 contains a nonzero
zerodivisor.
(3) If R k S contains a nontrivial idempotent, then there exist finitely generated subalgebras R0 R, S 0 S such that R0 k S 0 contains a nontrivial
idempotent.

COMMUTATIVE ALGEBRA

96

P
Proof. Suppose z R k S is nilpotent. We may write z = i=1,...,n xi yi .
Thus we may take R0 the k-subalgebra generated by the xi and S 0 the k-subalgebra
generated by the yi . The second and third statements are proved in the same
way.

034N

Lemma 42.5. Let k be a field. Let S be a geometrically reduced k-algebra. Let R

be any reduced k-algebra. Then R k S is reduced.
Proof. By Lemma 42.4 we may assume that R is of finite type over k. Then R,
as a reduced Noetherian ring, embeds into a finite product of fields (see Lemmas
24.4, 30.6, and 24.1). Hence we may assume R is a finite product of fields. In this
case it follows from Definition 42.1 that R k S is reduced.


030U

Lemma 42.6. Let k be a field. Let S be a reduced k-algebra. Let k K be either

a separable field extension, or a separably generated field extension. Then K k S
is reduced.
Proof. Assume k K is separable. By Lemma 42.4 we may assume that S is of
finite type over k and K is finitely generated over k. Then S embeds into a finite
product of fields, namely its total ring of fractions (see Lemmas 24.1 and 24.4).
Hence we may actually assume that S is a domain. We choose x1 , . . . , xr+1 K
as in Lemma 41.3. Let P k(x1 , . . . , xr )[T ] be the minimal polynomial of xr+1 . It
is a separable polynomial. It is easy to see that k[x1 , . . . , xr ] k S = S[x1 , . . . , xr ]
is a domain. This implies k(x1 , . . . , xr ) k S is a domain as it is a localization of
S[x1 , . . . , xr ]. The ring extension k(x1 , . . . , xr ) k S K k S is generated by a
single element xr+1 with a single equation, namely P . Hence K k S embeds into
f.f.(k(x1 , . . . , xn ) k S)[T ]/(P ). Since P is separable this is a finite product of
fields and we win.
At this point we do not yet know that a separably generated field extension is
separable, so we have to prove the lemma in this case also. To do this suppose
that {xi }iI is a separating transcendence basis for K over k. For any finite set
of elements j K there exists a finite subset T I such that k({xi }iT )
k({xi }iT {j }) is finite separable. Hence we see that K is a directed colimit of
finitely generated and separably generated extensions of k. Thus the argument of
the preceding paragraph applies to this case as well.


07K2

Lemma 42.7. Let k be a field and let S be a k-algebra. Assume that S is reduced
and that Sp is geometrically reduced for every minimal prime p of S. Then S is
geometrically reduced.
Q
Proof. Since S is reduced the map S p minimal Sp is injective, see Lemma 24.2.
If k K is a field extension, then the maps
Y
Y
S k K ( Sp ) k K
Sp k K
are injective: the first as k K is flat and the second by inspection because K is
a free k-module. As Sp is geometrically reduced the ring on the right is reduced.
Thus we see that S k K is reduced as a subring of a reduced ring.


COMMUTATIVE ALGEBRA

97

43. Separable extensions, continued

05DT

In this section we continue the discussion started in Section 41. Let p be a prime
number and let k be a field of characteristic p. In this case we write k 1/p for the
extension of k gotten by adjoining pth roots of all the elements of k to k. (In other
words it is the subfield of an algebraic closure of k generated by the pth roots of
elements of k.)

030W

Lemma 43.1. Let k be a field of characteristic p > 0. Let k K be a field

extension. The following are equivalent:
(1) K is separable over k,
(2) the ring K k k 1/p is reduced, and
(3) K is geometrically reduced over k.
Proof. The implication (1) (3) follows from Lemma 42.6. The implication (3)
(2) is immediate.
Assume (2). Let k L K be a subextension such that L is a finitely generated
field extension of k. We have to show that we can find a separating transcendence
basis of L. The assumption implies that L k k 1/p is reduced. Let x1 , . . . , xr be a
transcendence basis of L over k such that the degree of inseparability of the finite
extension k(x1 , . . . , xr ) L is minimal. If L is separable over k(x1 , . . . , xr ) then we
win. Assume this is not the case to get a contradiction. Then there exists an element
L which is not separable over k(x1 , . . . , xr ). Let P (T ) k(x1 , . . . , xr )[T ] be
the minimal polynomial of over k(x1 , . . . , xr ). After replacing by f for some
nonzero f k[x1 , . . . , xr ] we may and do assume that P lies in k[x1 , . . . , xr , T ].
Because is not separable P is a polynomial in T p , see Fields, Lemma 12.1. Let
dp be the degree of P as a polynomial in T . Since P is the minimal polynomial of
the monomials
xe11 . . . xerr e
for e < dp are linearly independent over k in L. We claim that the element P/xi
k[x1 , . . . , xr , T ] is not zero for at least one i. Namely, if this was not the case, then
P is actually a polynomial in xp1 , . . . , xpr , T p . In that case we can consider P 1/p
k 1/p [x1 , . . . , xr , T ]. This would map to P 1/p (x1 , . . . , xr , ) which is a nilpotent
element of k 1/p k L and hence zero. On the other hand, P 1/p (x1 , . . . , xr , ) is a
k 1/p -linear combination the monomials listed above, hence nonzero in k 1/p k L.
This is a contradiction which proves our claim.
Thus, after renumbering, we may assume that P/x1 is not zero. As P is an
irreducible polynomial in T over k(x1 , . . . , xr ) it is irreducible as a polynomial in
x1 , . . . , xr , T , hence by Gausss lemma it is irreducible as a polynomial in x1 over
k(x2 , . . . , xr , T ). Since the transcendence degree of L is r we see that x2 , . . . , xr ,
are algebraically independent. Hence P (X, x2 , . . . , xr , ) k(x2 , . . . , xr , )[X] is
irreducible. It follows that x1 is separably algebraic over k(x2 , . . . , xr , ). This
means that the degree of inseparability of the finite extension k(x2 , . . . , xr , ) L
is less than the degree of inseparability of the finite extension k(x1 , . . . , xr ) L,
which is a contradiction.


030X

Lemma 43.2. A separably generated field extension is separable.

Proof. Combine Lemma 42.6 with Lemma 43.1.

COMMUTATIVE ALGEBRA

98

In the following lemma we will use the notion of the perfect closure which is defined
in Definition 44.5.
030V

Lemma 43.3. Let k be a field. Let S be a k-algebra. The following are equivalent:
(1) k 0 k S is reduced for every finite purely inseparable extension k 0 of k,
(2) k 1/p k S is reduced,
(3) k perf k S is reduced, where k perf is the perfect closure of k,
(4) k k S is reduced, where k is the algebraic closure of k, and
(5) S is geometrically reduced over k.
Proof. Note that any finite purely inseparable extension k k 0 embeds in k perf .
Moreover, k 1/p embeds into k perf which embeds into k. Thus it is clear that (5)
(4) (3) (2) and that (3) (1).
We prove that (1) (5). Assume k 0 k S is reduced for every finite purely inseparable extension k 0 of k. Let k K be an extension of fields. We have to show that
K k S is reduced. By Lemma 42.4 we reduce to the case where k K is a finitely
generated field extension. Choose a diagram
KO

/ K0
O

/ k0

as in Lemma 41.4. By assumption k 0 k S is reduced. By Lemma 42.6 it follows

that K 0 k S is reduced. Hence we conclude that K k S is reduced as desired.
Finally we prove that (2) (5). Assume k 1/p k S is reduced. Then S is reduced.
Moreover, for each localization Sp at a minimal prime p, the ring k 1/p k Sp is a
localization of k 1/p k S hence is reduced. But Sp is a field by Lemma 24.1, hence
Sp is geometrically reduced by Lemma 43.1. It follows from Lemma 42.7 that S is
geometrically reduced.

44. Perfect fields
05DU

Here is the definition.

030Y

Definition 44.1. Let k be a field. We say k is perfect if every field extension of

k is separable over k.

030Z

Lemma 44.2. A field k is perfect if and only if it is a field of characteristic 0 or

a field of characteristic p > 0 such that every element has a pth root.
Proof. The characteristic zero case is clear. Assume the characteristic of k is p > 0.
If k is perfect, then all the field extensions where we adjoin a pth root of an element
of k have to be trivial, hence every element of k has a pth root. Conversely if every
element has a pth root, then k = k 1/p and every field extension of k is separable by
Lemma 43.1.


030R

Lemma 44.3. Let k K be a finitely generated field extension. There exists a

diagram
/ K0
KO
O
k

/ k0

COMMUTATIVE ALGEBRA

99

where k k 0 , K K 0 are finite purely inseparable field extensions such that k 0 K 0

is a separable field extension. In this situation we can assume that K 0 = k 0 K is the
compositum, and also that K 0 = (k 0 k K)red .
Proof. By Lemma 41.4 we can find such a diagram with k 0 K 0 separably generated. By Lemma 43.2 this implies that K 0 is separable over k 0 . The compositum
k 0 K is a subextension of k 0 K 0 and hence k 0 k 0 K is separable by Lemma 41.2.
n
The ring (k 0 k K)red is a domain as for some n  0 the map x 7 xp maps it
into K. Hence it is a field by Lemma 35.17. Thus (k 0 k K)red K 0 maps it
isomorphically onto k 0 K.

046W

Lemma 44.4. For every field k there exists a purely inseparable extension k k 0
such that k 0 is perfect. The field extension k k 0 is unique up to unique isomorphism.
Proof. If the characteristic of k is zero, then k 0 = k is the unique choice. Assume
the characteristic of k is p > 0. For every n > 0 there exists a unique algebraic
n
n
extension k k 1/p such that (a) every element k has a pn th root in k 1/p
n
n
and (b) for every element k 1/p we have p k. Namely, consider the ring
n
n
map k k 1/p = k, x 7 xp . This is injective and satisfies (a) and (b). It is clear
n
n+1
that S
k 1/p k 1/p
as extensions of k via the map y 7 y p . Then we can take
0
1/pn
k = k
. Some details omitted.


046X

Definition 44.5. Let k be a field. The field extension k k 0 of Lemma 44.4 is

called the perfect closure of k. Notation k k perf .
Note that if k k 0 is any algebraic purely inseparable extension, then k 0 k perf .
Namely, (k 0 )perf is isomorphic to k perf by the uniqueness of Lemma 44.4.

00I4

Lemma 44.6. Let k be a perfect field. Any reduced k algebra is geometrically

reduced over k. Let R, S be k-algebras. Assume both R and S are reduced. Then
the k-algebra R k S is reduced.
Proof. The first statement follows from Lemma 43.3. For the second statement
use the first statement and Lemma 42.5.

45. Universal homeomorphisms

0BR5

Let k k 0 be an algebraic purely inseparable field extension. Then for any kalgebra R the ring map R k 0 k R induces a homeomorphism of spectra. The
reason for this is the slightly more general Lemma 45.6 below.

0BR6

Lemma 45.1. Let : R S be a surjective map with locally nilpotent kernel.

Then induces a homeomorphism of spectra and isomorphisms on residue fields.
For any ring map R R0 the ring map R0 R0 R S is surjective with locally
nilpotent kernel.
Proof. By Lemma 16.7 the map Spec(S) Spec(R) is a homeomorphism onto
the closed subset V (Ker()). Of course V (Ker()) = Spec(R) because every prime
ideal of R contains every nilpotent element of R. This also implies the statement
on residue fields. By right exactness of tensor product we see that Ker()R0 is the
kernel of the surjective map R0 R0 R S. Hence the final statement by Lemma
31.2.


COMMUTATIVE ALGEBRA

0BR7

100

Lemma 45.2. Let k k 0 be a field extension. The following are equivalent

0

(1) for each x k there exists an n > 0 such that x k, and

(2) k 0 = k, or k 0 /k is a purely inseparable extension of fields, or k and k 0 have
characteristic p > 0 and are algebraic extensions of Fp .
Proof. Observe that each of the possibilities listed in (2) satisfies (1). Thus we
assume k 0 /k satisfies (1) and we prove that we are in one of the cases of (2).
Discarding the case k = k 0 we may assume k 0 6= k. It is clear that k 0 /k is algebraic.
0
Hence we may assume that k 0 /k is a nontrivial finite extension. Let k ksep
k0
be the separable subextension found in Fields, Lemma 13.6. We have to show that
0
k = ksep
or that k is an algebraic over Fp . Thus we may assume that k 0 /k is a
nontrivial finite separable extension and we have to show k is algebraic over Fp .
Pick x k 0 , x 6 k. Pick n, m > 0 such that xn k and (x + 1)m k. Let k be an
algebraic closure of k. We can choose embeddings , : k 0 k with (x) 6= (x).
This follows from the discussion in Fields, Section 12 (more precisely, after replacing
k 0 by the k-extension generated by x it follows from Fields, Lemma 12.8). Then
we see that (x) = (x) for some nth root of unity in k. Similarly, we see that
(x + 1) = 0 (x + 1) for some mth root of unity 0 k. Since (x + 1) 6= (x + 1)
we see 0 6= 1. Then
0 ( (x) + 1) = 0 (x + 1) = (x + 1) = (x) + 1 = (x) + 1
implies that
(x)( 0 ) = 1 0
hence 0 6= and
(x) = (1 0 )/( 0 )
Hence every element of k 0 which is not in k is algebraic over the prime subfield.
Since k 0 is generated over the prime subfield by the elements of k 0 which are not in
k, we conclude that k 0 (and hence k) is algebraic over the prime subfield.
Finally, if the characteristic of k is 0, the above leads to a contradiction as follows
(we encourage the reader to find their own proof). For every rational number y we
similarly get a root of unity y such that (x + y) = y (x + y). Then we find
(x) + y = y ( (x) + y)
and by our formula for (x) above we conclude y Q(, 0 ). Since the number
field Q(, 0 ) contains only a finite number of roots of unity we find two distinct
rational numbers y, y 0 with y = y0 . Then we conclude that
y y 0 = (x + y) (x + y 0 ) = y ( (x + y)) y0 (x + y 0 ) = y (y y 0 )
which implies y = 1 a contradiction.
0BR8

Lemma 45.3. Let : R S be a ring map. If

(1) for any x S there exists n > 0 such that xn is in the image of , and
(2) Ker() is locally nilpotent,
then induces a homeomorphism on spectra and induces residue field extensions
satisfying the equivalent conditions of Lemma 45.2.

[Alp14, Lemma
3.1.6]

COMMUTATIVE ALGEBRA

101

Proof. Assume (1) and (2). Let q, q0 be primes of S lying over the same prime
ideal p of R. Suppose x S with x q, x 6 q0 . Then xn q and xn 6 q0 for all
n > 0. If xn = (y) with y R for some n > 0 then
xn q y p xn q0
which is a contradiction. Hence there does not exist an x as above and we conclude
that q = q0 , i.e., the map on spectra is injective. By assumption (2) the kernel
I = Ker() is contained in every prime, hence Spec(R) = Spec(R/I) as topological
spaces. As the induced map R/I S is integral by assumption (1) Lemma 35.15
shows that Spec(S) Spec(R/I) is surjective. Combining the above we see that
Spec(S) Spec(R) is bijective. If x S is arbitrary, and we pick y R such that
(y) = xn for some n > 0, then we see that the open D(x) Spec(S) corresponds
to the open D(y) Spec(R) via the bijection above. Hence we see that the map
Spec(S) Spec(R) is a homeomorphism.
To see the statement on residue fields, let q S be a prime lying over a prime
ideal p R. Let x (q). If we think of (q) as the residue field of the local ring
Sq , then we see that x is the image of some y/z Sq with y S, z S, z 6 q.
Choose n, m > 0 such that y n , z m are in the image of . Then xnm is the residue
of (y/z)nm = (y n )m /(z m )n which is in the image of Rp Sq . Hence xnm is in the
image of (p) (q).

0545

Lemma 45.4. Let p be a prime number. Let n, m > 0 be two integers. There
a
n
m
exists an integer a such that (x + y)p , pa (x + y) Z[xp , pn x, y p , pm y].
Proof. This is clear for pa (x + y) as soon as a n, m. In fact, pick a  n, m.
Write
 a
X
p
pa
(x + y) =
xi y j
i,j0,i+j=pa i, j
For every i, j 0 with i + j = pa write i = qpn + r with r {0, . . . , pn 1} and j =
a
n
m
q 0 pm + r0 with r0 {0, . . . , pm 1}. The condition (x + y)p Z[xp , pn x, y p , pm y]
holds if
 a
0
p
pnr+mr divides
i, j
If r = r0 = 0 then the divisibility holds. If r 6= 0, then we write


 a
p
pa pa 1
=
i i 1, j
i, j
Since r 6= 0 the rational number pa /i has p-adic valuation at least a(n1) (because


pa
i is not divisible by pn ). Thus i,j
is divisible by pan+1 in this case. Similarly, we
a

p
see that if r0 6= 0, then i,j
is divisible by pam+1 . Picking a = npn + mpm + n + m
will work.


0BR9

Lemma 45.5. Let k k 0 be a field extension. Let p be a prime number. The

following are equivalent
(1) k 0 is generated as a field extension of k by elements x such that there exists
n
an n > 0 with xp k and pn x k, and
0
(2) k = k or the characteristic of k and k 0 is p and k 0 /k is purely inseparable.

COMMUTATIVE ALGEBRA

102

Proof. Let x k 0 . If there exists an n > 0 with xp k and pn x k and if the

n
characteristic is not p, then x k. If the characteristic is p, then we find xp k
and hence x is purely inseparable over k.

0BRA

Lemma 45.6. Let : R S be a ring map. Let p be a prime number. Assume

(a) S is generated as an R-algebra by elements x such that there exists an n > 0
n
with xp (R) and pn x (R), and
(b) Ker() is locally nilpotent,
Then induces a homeomorphism of spectra and induces residue field extensions
satisfying the equivalent conditions of Lemma 45.5. For any ring map R R0 the
ring map R0 R0 R S also satisfies (a) and (b).
Proof. Assume (a) and (b). Note that (b) is equivalent to condition (2) of Lemma
45.3. Let T S be the set of elements x S such that there exists an integer n > 0
n
such that xp , pn x (R). We claim that T = S. This will prove that condition
(1) of Lemma 45.3 holds and hence induces a homeomorphism on spectra. By
assumption (a) it suffices to show that T S is an R-sub algebra. If x T and
y R, then it is clear that yx T . Suppose x, y T and n, m > 0 such that
m
n+m
n
xp , y p , pn x, pm y (R). Then (xy)p
, pn+m xy (R) hence xy T . We
have x + y T by Lemma 45.4 and the claim is proved.
Since induces a homeomorphism on spectra, it is in particular surjective on
spectra which is a property preserved under any base change, see Lemma 29.3.
Therefore for any R R0 the kernel of the ring map R0 R0 R S consists of
nilpotent elements, see Lemma 29.6, in other words (b) holds for R0 R0 R S. It
is clear that (a) is preserved under base change. Finally, the condition on residue
fields follows from (a) as generators for S as an R-algebra map to generators for
the residue field extensions.


0BRB

Lemma 45.7. Let : R S be a ring map. Assume

(1) induces an injective map of spectra,
(2) induces purely inseparable residue field extensions.
Then for any ring map R R0 properties (1) and (2) are true for R0 R0 R S.
Proof. Set S 0 = R0 R S so that we have a commutative diagram of continuous
maps of spectra of rings
Spec(S 0 )

/ Spec(S)


Spec(R0 )


/ Spec(R)

Let p0 R0 be a prime ideal lying over p R. If there is no prime ideal of S

lying over p, then there is no prime ideal of S 0 lying over p0 . Otherwise, by Remark
16.8 there is a unique prime ideal r of F = S R (p) whose residue field is purely
inseparable over (p). Consider the ring maps
(p) F (r)
By Lemma 24.1 the ideal r F is locally nilpotent, hence we may apply Lemma
45.1 to the ring map F (r). We may apply Lemma 45.6 to the ring map

COMMUTATIVE ALGEBRA

103

(p) (r). Hence the composition and the second arrow in the maps
(p0 ) (p0 ) (p) F (p0 ) (p) (r)
induces bijections on spectra and purely inseparable residue field extensions. This
implies the same thing for the first map. Since
(p0 ) (p) F = (p0 ) (p) (p) R S = (p0 ) R S = (p0 ) R0 R0 R S
we conclude by the discussion in Remark 16.8.
0BRC

Lemma 45.8. Let : R S be a ring map. Assume

(1) is integral,
(2) induces an injective map of spectra,
(3) induces purely inseparable residue field extensions.
Then induces a homeomorphism from Spec(S) onto a closed subset of Spec(R)
and for any ring map R R0 properties (1), (2), (3) are true for R0 R0 R S.
Proof. The map on spectra is closed by Lemmas 40.6 and 35.20. The properties
are preserved under base change by Lemmas 45.7 and 35.11.


0BRD

Lemma 45.9. Let : R S be a ring map. Assume

(1) is integral,
(2) induces an bijective map of spectra,
(3) induces purely inseparable residue field extensions.
Then induces a homeomorphism on spectra and for any ring map R R0 properties (1), (2), (3) are true for R0 R0 R S.
Proof. Follows from Lemmas 45.8 and 29.3.

09EF

Lemma 45.10. Let : R S be a ring map such that

(1) the kernel of is locally nilpotent, and
(2) S is generated as an R-algebra by elements x such that there exist n > 0
and a polynomial P (T ) R[T ] whose image in S[T ] is (T s)n .
Then Spec(S) Spec(R) is a homeomorphism and R S induces purely inseparable extensions of residue fields. Moreover, conditions (1) and (2) remain true on
arbitrary base change.
Proof. We may replace R by R/Ker(), see Lemma 45.1. Assumption (2) implies
S is generated over R by elements which are integral over R. Hence R S is
integral (Lemma 35.7). In particular Spec(S) Spec(R) is surjective and closed
(Lemmas 35.15, 40.6, and 35.20).
Let x S be one of the generators in (2), i.e., there exists an n > 0 be such that
(T x)n R[T ]. Let p R be a prime. The (p) R S ring is nonzero by the
above and Lemma 16.9. If the characteristic of (p) is zero then we see that nx R
implies 1 x is in the image of (p) (p) R S. Hence (p) (p) R S is an
isomorphism. If the characteristic of (p) is p > 0, then write n = pk m with m
prime to p. In (p) R S[T ] we have
k

(T 1 x)n = ((T 1 x)p )m = (T p 1 xp )m

COMMUTATIVE ALGEBRA
k

104
k

and we see that mxp R. This implies that 1 xp is in the image of (p)
(p) R S. Hence Lemma 45.6 applies to (p) (p) R S. In both cases we conclude that (p) R S has a unique prime ideal with residue field purely inseparable
over (p). By Remark 16.8 we conclude that is bijective on spectra.
The statement on base change is immediate.

46. Geometrically irreducible algebras

00I2

An algebra S over a field k is geometrically irreducible if the algebra S k k 0 has a

unique minimal prime for every field extension k 0 /k. In this section we develop a
bit of theory relevant to this notion.

00I6

Lemma 46.1. Let R S be a ring map. Assume

(a) Spec(R) is irreducible,
(b) R S is flat,
(c) R S is of finite presentation,
(d) the fibre rings S R (p) have irreducible spectra for a dense collection of
primes p of R.
Then Spec(S) is irreducible. This is true more generally with (b) + (c) replaced by
the map Spec(S) Spec(R) is open.
Proof. The assumptions (b) and (c) imply that the map on spectra is open, see
Proposition 40.8. Hence the lemma follows from Topology, Lemma 7.12.


00I7

Lemma 46.2. Let k be a separably algebraically closed field. Let R, S be kalgebras. If R, S have a unique minimal prime, so does R k S.
Proof. Let k k be a perfect closure, see Definition 44.5. By assumption k is
algebraically closed. The ring maps R R k k and S S k k and R k S
(R k S) k k = (R k k) k (S k k) satisfy the assumptions of Lemma 45.6. Hence
we may assume k is algebraically closed.
We may replace R and S by their reductions. Hence we may assume that R and S
are domains. By Lemma 44.6 we see that R k S is reduced. Hence its spectrum is
reducible if and only if it contains a nonzero zerodivisor. By Lemma 42.4 we reduce
to the case where R and S are domains of finite type over k algebraically closed.
Note that the ring map R R k S is of finite presentation and flat. Moreover,
for every maximal ideal m of R we have (R k S) R R/m
= S because k
= R/m
by the Hilbert Nullstellensatz Theorem 33.1. Moreover, the set of maximal ideals
is dense in the spectrum of R since Spec(R) is Jacobson, see Lemma 34.2. Hence
we see that Lemma 46.1 applies to the ring map R R k S and we conclude that
the spectrum of R k S is irreducible as desired.


037K

Lemma 46.3. Let k be a field. Let R be a k-algebra. The following are equivalent
(1) for every field extension k k 0 the spectrum of R k k 0 is irreducible, and
(2) for every finite separable field extension k k 0 the spectrum of R k k 0 is
irreducible.
Proof. Let k k perf be a perfect closure of k, see Definition 44.5. By Lemma 45.6
we may replace R by (R k k perf )reduction and k by k perf (some details omitted).
Hence we may assume that R is geometrically reduced over k.

COMMUTATIVE ALGEBRA

105

Assume R is geometrically reduced over k. For any extension of fields k k 0 we see

irreducibility of the spectrum of R k k 0 is equivalent to R k k 0 being a domain.
Assume (2). Let k k be a separable algebraic closure of k. Using Lemma 42.4
we see that (2) is equivalent to R k k being a domain. For any field extension
0
0
k k 0 , there exists a field extension k k with k 0 k . By Lemma 46.2 we see
0
0
that R k k is a domain. If R k k 0 is not a domain, then also R k k is not a
domain, contradiction.

037L

Definition 46.4. Let k be a field. Let S be a k-algebra. We say S is geometrically

irreducible over k if for every field extension k k 0 the spectrum of S k k 0 is
irreducible4.
By Lemma 46.3 it suffices to check this for finite separable field extensions k k 0 .

037M

Lemma 46.5. Let k be a field. Let R be a k-algebra. If k is separably algebraically

closed then R is geometrically irreducible over k if and only if the spectrum of R is
irreducible.
Proof. Immediate from the remark following Definition 46.4.

037N

Lemma 46.6. Let k be a field. Let S be a k-algebra.

(1) If S is geometrically irreducible over k so is every k-subalgebra.
(2) If all finitely generated k-subalgebras of S are geometrically irreducible, then
S is geometrically irreducible.
(3) A directed colimit of geometrically irreducible k-algebras is geometrically
irreducible.
Proof. Let S 0 S be a subalgebra. Then for any extension k k 0 the ring map
S 0 k k 0 S k k 0 is injective also. Hence (1) follows from Lemma 29.5 (and the
fact that the image of an irreducible space under a continuous map is irreducible).
The second and third property follow from the fact that tensor product commutes
with colimits.


037O

Lemma 46.7. Let k be a field. Let S be a geometrically irreducible k-algebra. Let

R be any k-algebra. The map
Spec(R k S) Spec(R)
induces a bijection on irreducible components.
Proof. Recall that irreducible components correspond to minimal primes (Lemma
25.1). As R Rk S is flat we see by going down (Lemma 38.18) that any minimal
prime of R k S lies over a minimal prime of R. Conversely, if p R is a (minimal)
prime then
R k S/p(R k S) = (R/p) k S f.f.(R/p) k S
by flatness of R R k S. The ring f.f.(R/p) k S has irreducible spectrum by
assumption. It follows that R k S/p(R k S) has a single minimal prime (Lemma
29.5). In other words, the inverse image of the irreducible set V (p) is irreducible.
Hence the lemma follows.

Let us make some remarks on the notion of geometrically irreducible field extensions.
4An irreducible space is nonempty.

COMMUTATIVE ALGEBRA

037P

106

Lemma 46.8. Let k K be a field extension. If k is algebraically closed in K,

then K is geometrically irreducible over k.
Proof. Let k k 0 be a finite separable extension, say generated by k 0 over
k (see Fields, Lemma 18.1). Let P = T d + a1 T d1 + . . . + ad k[T ] be the
minimal polynomial of . Then K k k 0
= K[T ]/(P ). The only way the spectrum
of K[T ]/(P ) can be reducible is if P is reducible in K[T ]. Say P = P1 P2 is a
nontrivial factorization of P into monic polynomials. Let b1 , . . . , bt K be the
coefficients of P1 . Then we see that bi is algebraic over k by Lemma 37.5. Hence
the lemma follows.


037Q

Lemma 46.9. Let k K be a field extension. Consider the subextension k k 0

K such that k k 0 is separable algebraic and k 0 K maximal with this property.
Then K is geometrically irreducible over k 0 . If K/k is a finitely generated field
extension, then [k 0 : k] < .
Proof. Let k 00 K be the algebraic closure of k in K. By Lemma 46.8 we see
that K is geometrically irreducible over k 00 . Since k 0 k 00 is purely inseparable
(Fields, Lemma 13.6) we see from Lemma 45.6 that the extension k 0 K is also
geometrically irreducible. If k K is finitely generated, then k 0 is finite over k by
Fields, Lemma 25.10.


04KP

Lemma 46.10. Let k K be an extension of fields. Let k k be a separable

algebraic closure. Then Gal(k/k) acts transitively on the primes of k k K.
Proof. Let k k 0 K be the subextension found in Lemma 46.9. Note that as
k k is integral all the prime ideals of k k K and k k k 0 are maximal, see Lemma
35.18. By Lemma 46.7 the map
Spec(k k K) Spec(k k k 0 )
is bijective because (1) all primes are minimal primes, (1) k k K = (k k k 0 ) k0 K,
and (3) K is geometrically irreducible over k 0 . Hence it suffices to prove the lemma
for the action of Gal(k/k) on the primes of k k k 0 .
As every prime of k k k 0 is maximal, the residue fields are isomorphic to k. Hence
the prime ideals of k k k 0 correspond one to one to elements of Homk (k 0 , k) with
Homk (k 0 , k) corresponding to the kernel p of 1 : k k k 0 k. In particular
Gal(k/k) acts transitively on this set as desired.

47. Geometrically connected algebras

05DV
037R

Lemma 47.1. Let k be a separably algebraically closed field. Let R, S be kalgebras. If Spec(R), and Spec(S) are connected, then so is Spec(R k S).
Proof. Recall that Spec(R) is connected if and only if R has no nontrivial idempotents, see Lemma 20.4. Hence, by Lemma 42.4 we may assume R and S are of
finite type over k. In this case R and S are Noetherian, and have finitely many
minimal primes, see Lemma 30.6. Thus we may argue by induction on n + m where
n, resp. m is the number of irreducible components of Spec(R), resp. Spec(S). Of
course the case where either n or m is zero is trivial. If n = m = 1, i.e., Spec(R)
and Spec(S) both have one irreducible component, then the result holds by Lemma

COMMUTATIVE ALGEBRA

107

46.2. Suppose that n > 1. Let p R be a minimal prime corresponding to the irreducible closed subset T Spec(R). Let I R be such that T 0 = V (I) Spec(R)
is the closure of the complement of T . Note that this means that T 0 = Spec(R/I)
(Lemma 16.7) has n 1 irreducible components. Then T T 0 = Spec(R), and
T T 0 = V (p + I) = Spec(R/(p + I)) is not empty as Spec(R) is assumed connected. The inverse image of T in Spec(R k S) is Spec(R/p k S), and the inverse
of T 0 in Spec(R k S) is Spec(R/I k S). By induction these are both connected.
The inverse image of T T 0 is Spec(R/(p + I) k S) which is nonempty. Hence
Spec(R k S) is connected.

037S

Lemma 47.2. Let k be a field. Let R be a k-algebra. The following are equivalent
(1) for every field extension k k 0 the spectrum of R k k 0 is connected, and
(2) for every finite separable field extension k k 0 the spectrum of R k k 0 is
connected.
Proof. For any extension of fields k k 0 the connectivity of the spectrum of Rk k 0
is equivalent to R k k 0 having no nontrivial idempotents, see Lemma 20.4. Assume
(2). Let k k be a separable algebraic closure of k. Using Lemma 42.4 we see
that (2) is equivalent to R k k having no nontrivial idempotents. For any field
0
0
extension k k 0 , there exists a field extension k k with k 0 k . By Lemma
0
47.1 we see that R k k has no nontrivial idempotents. If R k k 0 has a nontrivial
0
idempotent, then also R k k , contradiction.


037T

Definition 47.3. Let k be a field. Let S be a k-algebra. We say S is geometrically

connected over k if for every field extension k k 0 the spectrum of S k k 0 is
connected.
By Lemma 47.2 it suffices to check this for finite separable field extensions k k 0 .

037U

Lemma 47.4. Let k be a field. Let R be a k-algebra. If k is separably algebraically

closed then R is geometrically connected over k if and only if the spectrum of R is
connected.
Proof. Immediate from the remark following Definition 47.3.

037V

Lemma 47.5. Let k be a field. Let S be a k-algebra.

(1) If S is geometrically connected over k so is every k-subalgebra.
(2) If all finitely generated k-subalgebras of S are geometrically connected, then
S is geometrically connected.
(3) A directed colimit of geometrically connected k-algebras is geometrically
connected.
Proof. This follows from the characterization of connectedness in terms of the
nonexistence of nontrivial idempotents. The second and third property follow from
the fact that tensor product commutes with colimits.

The following lemma will be superseded by the more general Varieties, Lemma 5.4.

037W

Lemma 47.6. Let k be a field. Let S be a geometrically connected k-algebra. Let

R be any k-algebra. The map
R R k S
induces a bijection on idempotents, and the map
Spec(R k S) Spec(R)

COMMUTATIVE ALGEBRA

108

induces a bijection on connected components.

Proof. The second assertion follows from the first combined with Lemma 21.2. By
Lemmas 47.5 and 42.4 we may assume that R and S are of finite type over k. Then
we see that also Rk S is of finite type over k. Note that in this case all the rings are
Noetherian and hence their spectra have finitely many connected components (since
they have finitely many irreducible components, see Lemma 30.6). In particular,
all connected components in question are open! Hence via Lemma 22.3 we see that
the first statement of the lemma in this case is equivalent to the second. Lets
prove this. As the algebra S is geometrically connected and nonzero we see that all
fibres of X = Spec(R k S) Spec(R) = Y are connected and nonempty. Also,
as R R k S is flat of finite presentation the map X Y is open (Proposition
40.8). Topology, Lemma 6.5 shows that X Y induces bijection on connected
components.

48. Geometrically integral algebras
05DW
05DX

Definition 48.1. Let k be a field. Let S be a k-algebra. We say S is geometrically

integral over k if for every field extension k k 0 the ring of S k k 0 is a domain.
Any question about geometrically integral algebras can be translated in a question
about geometrically reduced and irreducible algebras.

05DY

Lemma 48.2. Let k be a field. Let S be a k-algebra. In this case S is geometrically

integral over k if and only if S is geometrically irreducible as well as geometrically
reduced over k.
Proof. Omitted.

09P9

Lemma 48.3. Let k be a field. Let S be a geometrically integral k-algebra. Let R

be a k-algebra and an integral domain. Then R k S is an integral domain.
Proof. By Lemma 42.5 the ring R k S is reduced and by Lemma 46.7 the ring
R k S is irreducible (the spectrum has just one irreducible component), so R k S
is an integral domain.

49. Valuation rings

00I8

Here are some definitions.

00I9

Definition 49.1. Valuation rings.

(1) Let K be a field. Let A, B be local rings contained in K. We say that B
dominates A if A B and mA = A mB .
(2) Let A be a ring. We say A is a valuation ring if A is a local domain and if
A is maximal for the relation of domination among local rings contained in
the fraction field of A.
(3) Let A be a valuation ring with fraction field K. If R K is a subring of
K, then we say A is centered on R if R A.
With this definition a field is a valuation ring.

00IA

Lemma 49.2. Let K be a field. Let A K be a local subring. Then there exists
a valuation ring with fraction field K dominating A.

COMMUTATIVE ALGEBRA

109

Proof. We consider the collection of local subrings of K as a partially ordered

set using the relation of domination. Suppose
S that {Ai }iI is a totally ordered
collection of local subrings of K. Then B = Ai is a local subring which dominates
all of the Ai . Hence by Zorns Lemma, it suffices to show that if A K is a local
ring whose fraction field is not K, then there exists a local ring B K, B 6= A
dominating A.
Pick t K which is not in the fraction field of A. If t is transcendental over A,
then A[t] K and hence A[t](t,m) K is a local ring distinct from A dominating
A. Suppose t is algebraic over A. Then for some a A the element at is integral
over A. In this case the subring A0 K generated by A and ta is finite over A. By
Lemma 35.15 there exists a prime ideal m0 A0 lying over m. Then A0m0 dominates
A. If A = A0m0 , then t is in the fraction field of A which we assumed not to be the
case. Thus A 6= A0m0 as desired.

00IB

Lemma 49.3. Let A be a valuation ring with maximal ideal m and fraction field
K. Let x K. Then either x A or x1 A or both.
Proof. Assume that x is not in A. Let A0 denote the subring of K generated
by A and x. Since A is a valuation ring we see that there is no prime of A0
Pd
lying over m. Hence
we can write 1 = i=0 ti xi with ti m. This implies that
P
(1 t0 )(x1 )d ti (x1 )di = 0. In particular we see that x1 is integral over A.
Thus the subring A00 of K generated by A and x1 is finite over A and we see there
exists a prime ideal m00 A00 lying over m by Lemma 35.15. Since A is a valuation
ring we conclude that A = (A00 )m00 and hence x1 A.


052K

Lemma 49.4. Let A K be a subring of a field K such that for all x K either
x A or x1 A or both. Then A is a valuation ring with fraction field K.
Proof. If A is not K, then A is not a field and there is a nonzero maximal ideal m.
If m0 is a second maximal ideal, then choose x, y A with x m, y 6 m, x 6 m0 ,
and y m0 (see Lemma 14.2). Then neither x/y A nor y/x A contradicting
the assumption of the lemma. Thus we see that A is a local ring. Suppose that A0
is a local ring contained in K which dominates A. Let x A0 . We have to show
that x A. If not, then x1 A, and of course x1 mA . But then x1 mA0
which contradicts x A0 .


0AS4

Lemma 49.5. Let I be a directed partially ordered set. Let (Ai , ij ) be a system
of valuation rings over I whose transition maps ij are local. Then A = colim Ai
is a valuation ring.
Proof. It is clear that A is a domain. Let a, b A. Lemma 49.4 tells us we have
to show that either a|b or b|a in A. Choose i so large that there exist ai , bi Ai
mapping to a, b. Then Lemma 49.3 applied to ai , bi in Ai implies the result for a, b
in A.


052L

Lemma 49.6. Let K L be an extension of fields. If B L is a valuation ring,

then A = K B is a valuation ring.
Proof. We can replace L by f.f.(B) and K by K f.f.(B). Then the lemma
follows from a combination of Lemmas 49.3 and 49.4.


COMMUTATIVE ALGEBRA

0AAV

110

Lemma 49.7. Let K L be an algebraic extension of fields. If B L is a

valuation ring with fraction field L and not a field, then A = K B is a valuation
ring and not a field.
Proof. By Lemma 49.6 the ring A is a valuation ring. If A is a field, then A = K.
Then A = K B is an integral extension, hence there are no proper inclusions
among the primes of B (Lemma 35.18). This contradicts the assumption that B is
a local domain and not a field.


088Y

Lemma 49.8. Let A be a valuation ring. For any prime ideal p A the quotient
A/p is a valuation ring. The same is true for the localization Ap and in fact any
localization of A.
Proof. Use the characterization of valuation rings given in Lemma 49.4.

088Z

Lemma 49.9. Let A be a valuation ring with residue field K. Let A be a valuation
ring with fraction field K. Then C = { A0 | mod mA0 A} is a valuation
ring.
Proof. Note that mA0 C and C/mA0 = A. In particular, the fraction field of
C is equal to the fraction field of A0 . We will use the criterion of Lemma 49.4 to
prove the lemma. Let x be an element of the fraction field of C. By the lemma we
may assume x A0 . If x mA0 , then we see x C. If not, then x is a unit of A0
and we also have x1 A0 . Hence either x or x1 maps to an element of A by the
lemma again.


00IC

Lemma 49.10. Let A be a valuation ring. Then A is a normal domain.

Proof.
Suppose x is in the field of fractions of A and integral over A, say xd +
P
i
either x A (and were done) or x1 A. In the
i<d ai x = 0. By Lemma 49.4
P
second case we see that x = ai xid A as well.


090P

Lemma 49.11. Let A be a normal domain with fraction field K. For every
x K, x 6 A there exists a valuation ring A V K with fraction field K
such that x 6 V . In other words, A is the intersection of all valuation rings in K
containing A.
Proof. Suppose x K, x 6 A. Consider B = A[x1 ]. Then x 6 B. Namely, if
x = a0 + a1 x1 + . . . + ad xd then xd+1 a0 xd . . . ad = 0 and x is integral over
A in contradiction with the fact that A is normal. Thus x1 is not a unit in B.
Thus V (x1 ) Spec(B) is not empty (Lemma 16.2), and we can choose a prime
p B with x1 p. Choose a valuation ring V K dominating Bp (Lemma
49.2). Then x 6 V as x1 mV .

An totally ordered abelian group is a pair (, ) consisting of an abelian group
endowed with a total ordering such that 0 + 00 0 + 00 for all
, 0 , 00 .

00ID

Lemma 49.12. Let A be a valuation ring with field of fractions K. Set = K /A

(with group law written additively). For , 0 define 0 if and only if 0
is in the image of A {0} . Then (, ) is a totally ordered abelian group.
Proof. Omitted, but follows easily from Lemma 49.3. Note that in case A = K
we obtain the zero group = {0} endowed with its unique total ordering.


COMMUTATIVE ALGEBRA

00IE

111

Definition 49.13. Let A be a valuation ring.

(1) The totally ordered abelian group (, ) of Lemma 49.12 is called the value
group of the valuation ring A.
(2) The map v : A {0} and also v : K is called the valuation
associated to A.
(3) The valuation ring A is called a discrete valuation ring if
= Z.
Note that if
= Z then there is a unique such isomorphism such that 1 0. If the
isomorphism is chosen in this way, then the ordering becomes the usual ordering of
the integers.

00IF

Lemma 49.14. Let A be a valuation ring. The valuation v : A {0} 0 has

the following properties:
(1) v(a) = 0 a A ,
(2) v(ab) = v(a) + v(b),
(3) v(a + b) min(v(a), v(b)).
Proof. Omitted.

090Q

Lemma 49.15. Let A be a ring. The following are equivalent

(1) A is a valuation ring,
(2) A is a local domain and every finitely generated ideal of A is principal.
Proof. Assume A is a valuation ring and let f1 , . . . , fn A. Choose i such that
v(fi ) is minimal among v(fj ). Then (fi ) = (f1 , . . . , fn ). Conversely, assume A is a
local domain and every finitely generated ideal of A is principal. Pick f, g A and
write (f, g) = (h). Then f = ah and g = bh and h = cf + dg for some a, b, c, d A.
Thus ac + bd = 1 and we see that either a or b is a unit, i.e., either g/f or f /g is
an element of A. This shows A is a valuation ring by Lemma 49.4.


00IG

Lemma 49.16. Let (, ) be a totally ordered abelian group. Let K be a field.

Let v : K be a homomorphism of abelian groups such that v(a + b)
min(v(a), v(b)) for a, b K with a, b, a + b not zero. Then
A = {x K | x = 0 or v(x) 0}
is a valuation ring with value group Im(v) , with maximal ideal
m = {x K | x = 0 or v(x) > 0}
and with group of units
A = {x K | v(x) = 0}.
Proof. Omitted.

Let (, ) be a totally ordered abelian group. An ideal of is a subset I such

that all elements of I are 0 and I, 0 implies 0 I. We say that such
an ideal is prime if + 0 I, , 0 0 I or 0 I.
00IH

Lemma 49.17. Let A be a valuation ring. Ideals in A correspond 1 1 with ideals

of . This bijection is inclusion preserving, and maps prime ideals to prime ideals.
Proof. Omitted.

00II

Lemma 49.18. A valuation ring is Noetherian if and only if it is a discrete

valuation ring or a field.

COMMUTATIVE ALGEBRA

112

Proof. Suppose A is a discrete valuation ring with valuation v : A \ {0} Z

normalized so that Im(v) Z0 . By Lemma 49.17 the ideals of A are the subsets
In = {0} v 1 (Zn ). It is clear that any element x A with v(x) = n generates
In . Hence A is a PID so certainly Noetherian.
Suppose A is a Noetherian valuation ring with value group . By Lemma 49.17 we
see the ascending chain condition holds for ideals in . We may assume A is not a
field, i.e., there is a with > 0. Applying the ascending chain condition to
the subsets + 0 with > 0 we see there exists a smallest element 0 which is
bigger than 0. Let be an element > 0. Consider the sequence of elements
, 0 , 20 , etc. By the ascending chain condition these cannot all be > 0.
Let n0 be the last one 0. By minimality of 0 we see that 0 = n0 .
Hence is a cyclic group as desired.

50. More Noetherian rings
00IJ
00IK

Lemma 50.1. Let R be a Noetherian ring. Any finite R-module is of finite

presentation. Any submodule of a finite R-module is finite. The ascending chain
condition holds for R-submodules of a finite R-module.
Proof. We first show that any submodule N of a finite R-module M is finite. We
do this by induction on the number of generators of M . If this number is 1, then
N = J/I M = R/I for some ideals I J R. Thus the definition of Noetherian
implies the result. If the number of generators of M is greater than 1, then we can
find a short exact sequence 0 M 0 M M 00 0 where M 0 and M 00 have
fewer generators. Note that setting N 0 = M 0 N and N 00 = Im(N M 00 ) gives
a similar short exact sequence for N . Hence the result follows from the induction
hypothesis since the number of generators of N is at most the number of generators
of N 0 plus the number of generators of N 00 .
To show that M is finitely presented just apply the previous result to the kernel of
a presentation Rn M .
It is well known and easy to prove that the ascending chain condition for Rsubmodules of M is equivalent to the condition that every submodule of M is
a finite R-module. We omit the proof.


00IN

Lemma 50.2 (Artin-Rees). Suppose that R is Noetherian, I R an ideal. Let

N M be finite R-modules. There exists a constant c > 0 such that I n M N =
I nc (I c M N ) for all n c.
L
n
0
Proof. Consider the ring S = R I I 2 . . . =
n0 I . Convention: I =
R. Multiplication maps I n I m into I n+m by multiplication in R. Note that if
I = (f1 , . . . , ft ) then S is a quotient of the Noetherian ring R[X1 , . . . , Xt ]. The
map just sends the monomial X1e1 . . . Xtet to f1e1 . . . ftet . L
Thus S is Noetherian.
n
Similarly, consider the module M IM I 2 M . . . =
n0 I M . This is a
finitely generated
L S-module. Namely, if x1 , . . . , xr generate M over
L R, then they
also generate n0 I n M over S. Next, consider the submodule n0 I n M N .
This is an S-submodule, as is easily
By Lemma 50.1 it is finitely generated
L verified.
n
as an S-module, say by j
n0 I M N , j = 1, . . . , s. We may assume by
decomposing each j into its homogeneous pieces that each j I dj M N for some

COMMUTATIVE ALGEBRA

113

dj . Set
for all n c every element in I n M N is of the
P c = max{dj }. Then
ndj
form
hj j with hj I
. The lemma now follows from this and the trivial
observation that I ndj (I dj M N ) I nc (I c M N ).

00IO

Lemma 50.3. Suppose that 0 K M

N is an exact sequence of finitely
generated modules over a Noetherian ring R. Let I R be an ideal. Then there
exists a c such that
f 1 (I n N ) = K + I nc f 1 (I c N )

and

f (M ) I n N f (I nc M )

for all n c.
Proof. Apply Lemma 50.2 to Im(f ) N and note that f : I nc M I nc f (M )
is surjective.

00IP

Lemma 50.4 (Krulls intersection theorem). Let R be a Noetherian

local ring.
T
Let I R be a proper ideal. Let M be a finite R-module. Then n0 I n M = 0.
T
Proof. Let N = n0 I n M . Then N = I n M N for all n 0. By the ArtinRees Lemma 50.2 we see that N = I n M N IN for some suitably large n. By
Nakayamas Lemma 19.1 we see that N = 0.


00IQ

Lemma 50.5. Let R be T

a Noetherian ring. Let I R be an ideal. Let M be a
finite R-module. Let N = n I n M .
(1) For every prime p, I p there exists a f R, f 6 p such that Nf = 0.
(2) If I rad(R) is contained in the Jacobson radical of R, then N = 0.
Proof. Proof of (1). Let x1 , . . . , xn be generators for the module N , see Lemma
50.1. For every prime p, I p we see that the image of N in the localization Mp is
zero, by Lemma 50.4. Hence we can find gi R, gi 6 p such that xi maps to zero
in Ngi . Thus Ng1 g2 ...gn = 0.
Part (2) follows from (1) and Lemma 23.1.

00IR

Remark 50.6. Lemma 50.4 in particular implies that n I n = (0) when I R is

a non-unit ideal in a Noetherian local ring R.TMore generally, let R be a Noetherian
ring and I R an ideal. Suppose that f nN I n . Then Lemma 50.5 says that
for every prime ideal I p there exists a g R, g 6 p such that f maps to zero
in Rg . In algebraic geometry we express this by saying that f is zero in an open
neighbourhood of the closed set V (I) of Spec(R).

00IS

Lemma 50.7 (Artin-Tate). Let R be a Noetherian ring. Let S be a finitely

generated R-algebra. If T S is an R-subalgebra such that S is finitely generated
as a T -module, then T is a finite type over R.

Proof. Choose elements x1 , . . . , xn S which generate S as an R-algebra. Choose

y1 , . . .P
, ym in S which generate S as a T -module. Thus there
that
Pexist aij T such
xi =
aij yj . There also exist bijk T such that yi yj =
bijk yk . Let T 0 T be
the sub R-algebra generated by aij and bijk . This is a finitely generated R-algebra,
hence Noetherian. Consider the algebra
X
S 0 = T 0 [Y1 , . . . , Ym ]/(Yi Yj
bijk Yk ).
Note that S 0 is finite over T 0 , namely as a T 0 -module it is generated by the classes
of 1, Y1 , . . . , Ym . Consider the T 0 -algebra homomorphism S 0 S which maps Yi

COMMUTATIVE ALGEBRA

114

to yi . Because aij T 0 we see that xj is in the image of this map. Thus S 0 S

is surjective. Therefore S is finite over T 0 as well. Since T 0 is Noetherian and we
conclude that T S is finite over T 0 and we win.

51. Length
00IU
02LY

Definition 51.1. Let R be a ring. For any R-module M we define the length of
M over R by the formula
lengthR (M ) = sup{n | 0 = M0 M1 . . . Mn = M, Mi 6= Mi+1 }.
In other words it is the supremum of the lengths of chains of submodules. There
is an obvious notion of when a chain of submodules is a refinement of another.
This gives a partial ordering on the collection of all chains of submodules, with the
smallest chain having the shape 0 = M0 M1 = M if M is not zero. We note the
obvious fact that if the length of M is finite, then every chain can be refined to
a maximal chain. But it is not as obvious that all maximal chains have the same
length (as we will see later).

02LZ

Lemma 51.2. Let R be a ring. Let M be an R-module. If lengthR (M ) < then

M is a finite R-module.
Proof. Omitted.

00IV

Lemma 51.3. If 0 M 0 M M 00 0 is a short exact sequence of modules

over R then the length of M is the sum of the lengths of M 0 and M 00 .
Proof. Given filtrations of M 0 and M 00 of lengths n0 , n00 it is easy to make a
corresponding filtration of M of length n0 + n00 . Thus we see that lengthR M
lengthR M 0 + lengthR M 00 . Conversely, given a filtration M0 M1 . . . Mn
of M consider the induced filtrations Mi0 = Mi M 0 and Mi00 = Im(Mi M 00 ).
Let n0 (resp. n00 ) be the number of steps in the filtration {Mi0 } (resp. {Mi00 }). If
0
00
Mi0 = Mi+1
and Mi00 = Mi+1
then Mi = Mi+1 . Hence we conclude that n0 +n00 n.
Combined with the earlier result we win.


00IW

Lemma 51.4. Let R be a local ring with maximal ideal m. Let M be an R-module.
(1) If M is a finite module and mn M 6= 0 for all n 0, then lengthR (M ) = .
(2) If M has finite length then mn M = 0 for some n.
Proof. Assume mn M 6= 0 for all n 0. Choose x M and f1 , . . . , fn m such
that f1 f2 . . . fn x 6= 0. By Nakayamas Lemma 19.1 the first n steps in the filtration
0 Rf1 . . . fn x Rf1 . . . fn1 x . . . Rx M
are distinct. This can also be seen directly. For example, if Rf1 x = Rf1 f2 x , then
f1 x = gf1 f2 x for some g, hence (1 gf2 )f1 x = 0 hence f1 x = 0 as 1 gf2 is a unit
which is a contradiction with the choice of x and f1 , . . . , fn . Hence the length is
infinite, i.e., (1) holds. Combine (1) and Lemma 51.2 to see (2).


00IX

Lemma 51.5. Let R S be a ring map. Let M be an S-module. We always

have lengthR (M ) lengthS (M ). If R S is surjective then equality holds.
Proof. A filtration of M by S-submodules gives rise a filtration of M by Rsubmodules. This proves the inequality. And if R S is surjective, then any
R-submodule of M is automatically a S-submodule. Hence equality in this case. 

COMMUTATIVE ALGEBRA

00IY

115

Lemma 51.6. Let R be a ring with maximal ideal m. Suppose that M is an

R-module with mM = 0. Then the length of M as an R-module agrees with the
dimension of M as a R/m vector space. The length is finite if and only if M is a
finite R-module.
Proof. The first part is a special case of Lemma 51.5. Thus the length is finite if
and only if M has a finite basis as a R/m-vector space if and only if M has a finite
set of generators as an R-module.


00IZ

Lemma 51.7. Let R be a ring. Let M be an R-module. Let S R be a multiplicative subset. Then lengthR (M ) lengthS 1 R (S 1 M ).
Proof. Any submodule N 0 S 1 M is of the form S 1 N for some R-submodule
N M , by Lemma 9.15. The lemma follows.


00J0

Lemma 51.8. Let R be a ring with finitely generated maximal ideal m. (For
example R Noetherian.) Suppose that M is a finite R-module with mn M = 0 for
some n. Then lengthR (M ) < .
Proof. Consider the filtration 0 = mn M mn1 M . . . mM M . All of the
subquotients are finitely generated R-modules to which Lemma 51.6 applies. We
conclude by additivity, see Lemma 51.3.


00J1

Definition 51.9. Let R be a ring. Let M be an R-module. We say M is simple

if M 6= 0 and every submodule of M is either equal to M or to 0.

00J2

Lemma 51.10.
equivalent:

Let R be a ring. Let M be an R-module. The following are

(1) M is simple,
(2) lengthR (M ) = 1, and
(3) M
= R/m for some maximal ideal m R.
Proof. Let m be a maximal ideal of R. By Lemma 51.6 the module R/m has length
1. The equivalence of the first two assertions is tautological. Suppose that M is
simple. Choose x M , x 6= 0. As M is simple we have M = Rx. Let I R be the
annihilator of x, i.e., I = {f R | f x = 0}. The map R/I M , f mod I 7 f x is
an isomorphism, hence R/I is a simple R-module. Since R/I 6= 0 we see I 6= R. Let
I m be a maximal ideal containing I. If I 6= m, then m/I R/I is a nontrivial
submodule contradicting the simplicity of R/I. Hence we see I = m as desired. 
00J3

Lemma 51.11. Let R be a ring. Let M be a finite length R-module. Let ` =

lengthR (M ). Choose any maximal chain of submodules
0 = M0 M1 M2 . . . Mn = M
with Mi 6= Mi1 , i = 1, . . . , n. Then
(1)
(2)
(3)
(4)

n = `,
each Mi /Mi1 is simple,
each Mi /Mi1 is of the form R/mi for some maximal ideal mi ,
given a maximal ideal m R we have
#{i | mi = m} = lengthRm (Mm ).

COMMUTATIVE ALGEBRA

116

Proof. If Mi /Mi1 is not simple then we can refine the filtration and the filtration
is not maximal. Thus we see that Mi /Mi1 is simple. By Lemma 51.10 the modules
Mi /Mi1 have length 1 and are of the form R/mi for some maximal ideals mi . By
additivity of length, Lemma 51.3, we see n = `. Since localization is exact, we see
that
0 = (M0 )m (M1 )m (M2 )m . . . (Mn )m = Mm
is a filtration of Mm with successive quotients (Mi /Mi1 )m . Thus the last statement
follows directly from the fact that given maximal ideals m, m0 of R we have

0
if m 6= m0 ,
0

(R/m )m =
Rm /mRm if m = m0
This we leave to the reader.
02M0

Lemma 51.12. Let A be a local ring with maximal ideal m. Let B be a semi-local
ring with maximal ideals mi , i = 1, . . . , n. Suppose that A B is a homomorphism
such that each mi lies over m and such that
[(mi ) : (m)] < .
Let M be a B-module of finite length. Then
X
lengthA (M ) =
[(mi ) : (m)]lengthBm (Mmi ),
i=1,...,n

in particular lengthA (M ) < .

Proof. Choose a maximal chain
0 = M0 M1 M2 . . . Mn = M
by B-submodules as in Lemma 51.11. Then each quotient Mi /Mi1 is isomorphic
to (mj(i) ) for some j(i) {1, . . . , n}. Moreover lengthA ((mi )) = [(mi ) : (m)]
by Lemma 51.6. The lemma follows by additivity of lengths (Lemma 51.3).

02M1

Lemma 51.13. Let A B be a flat local homomorphism of local rings. Then for
any A-module M we have
lengthA (M )lengthB (B/mA B) = lengthB (M A B).
In particular, if lengthB (B/mA B) < then M has finite length if and only if
M A B has finite length.
Proof. The ring map A B is faithfully flat by Lemma 38.17. Hence if 0 = M0
M1 . . . Mn = M is a chain of length n in M , then the corresponding chain
0 = M0 A B M1 A B . . . Mn A B = M A B has length n also. This proves
lengthA (M ) = lengthB (M A B) = . Next, assume lengthA (M ) < . In
this case we see that M has a filtration of length ` = lengthA (M ) whose quotients
are A/mA . Arguing as above we see that M A B has a filtration of length ` whose
quotients are isomorphic to B A A/mA = B/mA B. Thus the lemma follows. 

02M2

Lemma 51.14. Let A B C be flat local homomorphisms of local rings. Then

lengthB (B/mA B)lengthC (C/mB C) = lengthC (C/mA C)
Proof. Follows from Lemma 51.13 applied to the ring map B C and the Bmodule M = B/mA B


COMMUTATIVE ALGEBRA

117

52. Artinian rings

00J4

Artinian rings, and especially local Artinian rings, play an important role in algebraic geometry, for example in deformation theory.

00J5

Definition 52.1. A ring R is Artinian if it satisfies the descending chain condition

for ideals.

00J6

Lemma 52.2. Suppose R is a finite dimensional algebra over a field. Then R is

Artinian.
Proof. The descending chain condition for ideals obviously holds.

00J7

Lemma 52.3. If R is Artinian then R has only finitely many maximal ideals.
Proof. Suppose that mi , i = 1, 2, 3, . . . are maximal ideals. Then m1 m1 m2
m1 m2 m3 . . . is an infinite descending sequence (because by the Chinese
remainder theorem all the maps R ni=1 R/mi are surjective).


00J8

Lemma 52.4. Let R be Artinian. The radical rad(R) of R is a nilpotent ideal.

Proof. Denote the radical I. Note that I I 2 I 3 . . . is a descending sequence.
Thus I n = I n+1 for some n. Set J = {x R | xI n = 0}. We have to show J = R.
If not, choose an ideal J 0 6= J, J J 0 minimal (possible by the Artinian property).
Then J 0 = J + Rx for some x R. By NAK, Lemma 19.1, we have IJ 0 J.
Hence xI n+1 xI I n J I n = 0. Since I n+1 = I n we conclude x J.
Contradiction.


00JA

Lemma 52.5. Any ring with finitely many maximal ideals and locally nilpotent
radical is the product of its localizations at its maximal ideals. Also, all primes are
maximal.
Proof.
Let R be a ring with finitely many maximal ideals m1 , . . . , mn . Let I =
Tn
m
be the radical of R. Assume I is locally nilpotent. Let p be a prime ideal of
i
i=1
R. Since every prime contains every nilpotent element of R we see p m1 . . .mn .
Since m1 . . . mn m1 . . . mn we conclude p m1 . . . mn . Hence p mi for some
i, and so
Lp = mi . By the Chinese remainder theorem (Lemma 14.3) we have
R/mi which is a product of fields. Hence by Lemma
R/I
=
Q 31.5 there are
idempotents ei , i = 1, . . . , n with ei mod mj = ij . Hence R =
Rei , and each
Rei is a ring with exactly one maximal ideal.


00JB

Lemma 52.6. A ring R is Artinian if and only if it has finite length as a module
over itself. Any such ring R is both Artinian and Noetherian, any prime ideal of R
is a maximal ideal, and R is equal to the (finite) product of its localizations at its
maximal ideals.
Proof. If R has finite length over itself then it satisfies both the ascending chain
condition and the descending chain condition for ideals. Hence it is both Noetherian
and Artinian. Any Artinian ring is equal to product of its localizations at maximal
ideals by Lemmas 52.3, 52.4, and 52.5.
Suppose that R is Artinian. We will show R has finite length over itself. It suffices
to exhibit a chain of submodules whose successive quotients have finite length.
By what we said above we may assume that R is local, with maximal ideal m.
By Lemma 52.4 we have mn = 0 for some n. Consider the sequence 0 = mn

COMMUTATIVE ALGEBRA

118

mn1 . . . m R. By Lemma 51.6 the length of each subquotient mj /mj+1

is the dimension of this as a vector space over (m). This has to be finite since
otherwise we would have an infinite descending chain of sub vector spaces which
would correspond to an infinite descending chain of ideals in R.

53. Homomorphisms essentially of finite type
07DR

Some simple remarks on localizations of finite type ring maps.

00QM

Definition 53.1. Let R S be a ring map.

(1) We say that R S is essentially of finite type if S is the localization of an
R-algebra of finite type.
(2) We say that R S is essentially of finite presentation if S is the localization
of an R-algebra of finite presentation.

07DS

Lemma 53.2. The class of ring maps which are essentially of finite type is preserved under composition. Similarly for essentially of finite presentation.
Proof. Omitted.

0AUF

Lemma 53.3. The class of ring maps which are essentially of finite type is preserved by base change. Similarly for essentially of finite presentation.
Proof. Omitted.

07DT

Lemma 53.4. Let R S be a ring map. Assume S is an Artinian local ring with
maximal ideal m. Then
(1) R S is finite if and only if R S/m is finite,
(2) R S is of finite type if and only if R S/m is of finite type.
(3) R S is essentially of finite type if and only if the composition R S/m
is essentially of finite type.
Proof. If R S is finite, then R S/m is finite by Lemma 7.3. Conversely,
assume R S/m is finite. As S has finite length over itself (Lemma 52.6) we can
choose a filtration
0 I1 . . . In = S
by ideals such that Ii /Ii1
= S/m as S-modules. Thus S has a filtration by Rsubmodules Ii such that each successive quotient is a finite R-module. Thus S is a
finite R-module by Lemma 5.3.
If R S is of finite type, then R S/m is of finite type by Lemma 6.2. Conversely,
assume that R S/m is of finite type. Choose f1 , . . . , fn S which map to
generators of S/m. Then A = R[x1 , . . . , xn ] S, xi 7 fi is a ring map such that
A S/m is surjective (in particular finite). Hence A S is finite by part (1) and
we see that R S is of finite type by Lemma 6.2.
If R S is essentially of finite type, then R S/m is essentially of finite type
by Lemma 53.2. Conversely, assume that R S/m is essentially of finite type.
Suppose S/m is the localization of R[x1 , . . . , xn ]/I. Choose f1 , . . . , fn S whose
congruence classes modulo m correspond to the congruence classes of x1 , . . . , xn
modulo I. Consider the map R[x1 , . . . , xn ] S, xi 7 fi with kernel J. Set
A = R[x1 , . . . , xn ]/J S and p = A m. Note that A/p S/m is equal to the
image of R[x1 , . . . , xn ]/I in S/m. Hence (p) = S/m. Thus Ap S is finite by
part (1). We conclude that S is essentially of finite type by Lemma 53.2.


COMMUTATIVE ALGEBRA

119

The following lemma can be proven using properness of projective space instead of
the algebraic argument we give here.
0AUG

Lemma 53.5. Let : R S be essentially of finite type with R and S local (but not necessarily local). Then there exists an n and a maximal ideal
m R[x1 , . . . , xn ] lying over mR such that S is a localization of a quotient of
R[x1 , . . . , xn ]m .
Proof. We can write S as a localization of a quotient of R[x1 , . . . , xn ]. Hence
it suffices to prove the lemma in case S = R[x1 , . . . , xn ]q for some prime q
R[x1 , . . . , xn ]. If q + mR R[x1 , . . . , xn ] 6= R[x1 , . . . , xn ] then we can find a maximal
ideal m as in the statement of the lemma with q m and the result is clear.
Choose a valuation ring A (q) which dominates the image of R (q) (Lemma
49.2). If the image i (q) of xi is contained in A, then q is contained in
the inverse image of mA via R[x1 , . . . , xn ] A which means we are back in the
preceding case. Hence there exists an i such that 1
A and such that j /i A
i
for all j = 1, . . . , n (because the value group of A is totally ordered, see Lemma
49.12). Then we consider the map
R[y0 , y1 , . . . yn ] R[x1 , . . . , xn ]q ,

y0 7 1/xi ,

yj 7 xj /xi

Let q0 R[y0 , . . . , yn ] be the inverse image of q. Since y0 6 q0 it is easy to see

that the displayed arrow defines an isomorphism on localizations. On the other
hand, the result of the first paragraph applies to R[y0 , . . . , yn ] because yj maps to
an element of A. This finishes the proof.

54. K-groups
00JC

Let R be a ring. We will introduce two abelian groups associated to R. The first
of the two is denoted K00 (R) and has the following properties:
(1) For every finite R-module M there is given an element [M ] in K00 (R),
(2) for every short exact sequence 0 M 0 M M 00 0 we have the
relation [M ] = [M 0 ] + [M 00 ],
(3) the group K00 (R) is generated by the elements [M ], and
(4) all relations in K00 (R) are Z-linear combinations of the relations coming
from exact sequences as above.
The actual construction is a bit more annoying since one has to take care that
the collection of all finitely generated R-modules is a proper class. However, this
problem can be overcome by taking as set of generators of the group K00 (R) the
elements [Rn /K] where n ranges over all integers and K ranges over all submodules
K Rn . The generators for the subgroup of relations imposed on these elements
will be the relations coming from short exact sequences whose terms are of the form
Rn /K. The element [M ] is defined by choosing n and K such that M
= Rn /K
n
and putting [M ] = [R /K]. Details left to the reader.

00JD

Lemma 54.1. If R is an Artinian local ring then the length function defines a
natural abelian group homomorphism lengthR : K00 (R) Z.
Proof. The length of any finite R-module is finite, because it is the quotient of Rn
which has finite length by Lemma 52.6. And the length function is additive, see
Lemma 51.3.


COMMUTATIVE ALGEBRA

120

The second of the two is denoted K0 (R) and has the following properties:
(1) For every finite projective R-module M there is given an element [M ] in
K0 (R),
(2) for every short exact sequence 0 M 0 M M 00 0 of finite projective
R-modules we have the relation [M ] = [M 0 ] + [M 00 ],
(3) the group K0 (R) is generated by the elements [M ], and
(4) all relations in K0 (R) are Z-linear combinations of the relations coming
from exact sequences as above.
The construction of this group is done as above.
We note that there is an obvious map K0 (R) K00 (R) which is not an isomorphism
in general.
00JE

Example 54.2. Note that if R = k is a field then we clearly have K0 (k) =

K00 (k)
= Z with the isomorphism given by the dimension function (which is also
the length function).

00JF

Example 54.3. Let k be a field. Then K0 (k[x]) = K00 (k[x]) = Z.

Since R = k[x] is a principal ideal domain, any finite projective R-module is free.
In a short exact sequence of modules
0 M 0 M M 00 0
we have rank(M ) = rank(M 0 ) + rank(M 00 ), which gives K0 (k[x]) = Z.
As for K00 , the structure theorem for modules of a PID says that any finitely generated R-module is of the form M = Rr R/(d1 ) . . . R/(dk ). Consider the short
exact sequence
0 (di ) R R/(di ) 0
Since the ideal (di ) is isomorphic to R as a module (it is free with generator di ),
in K00 (R) we have [(di )] = [R]. Then [R/(di )] = [(di )] [R] = 0. From this it
follows that any torsion part disappears in K00 . Again the rank of the free part
determines that K00 (k[x]) = Z, and the canonical homomorphism from K0 to K00 is
an isomorphism.

00JG

Example 54.4. Let k be a field. Let R = {f k[x] | f (0) = f (1)}, compare

Example 26.4. In this case K0 (R)
= k Z, but K00 (R) = Z.

00JH

Lemma 54.5. Let R = R1 R2 . Then K0 (R) = K0 (R1 ) K0 (R2 ) and K00 (R) =
K00 (R1 ) K00 (R2 )
Proof. Omitted.

00JI

Lemma 54.6. Let R be an Artinian local ring. The map lengthR : K00 (R) Z of
Lemma 54.1 is an isomorphism.
Proof. Omitted.

00JJ

Lemma 54.7. Let R be a local ring. Every finite projective R-module is finite
free. The map rankR : K0 (R) Z defined by [M ] rankR (M ) is well defined and
an isomorphism.

COMMUTATIVE ALGEBRA

121

Proof. Let P be a finite projective R-module. The n generators of P give a

surjection Rn P , and since P is projective it follows that Rn
= P Q for some
projective module Q.
If m R is the maximal ideal, then P/m and Q/m are R/m-vector spaces, with
P/m Q/m
= (R/m)n . Say that dim P = p, dim Q = q, so p + q = n.
Choose elements a1 , . . . , ap in P and b1 , . . . , bq in Q lying above bases for P/m and
Q/m. The homomorphism Rn P Q
= Rn given by (r1 , . . . , rn ) 7 r1 a1 + . . . +
rp ap + rp+1 b1 + . . . + rn bq is a matrix A which is invertible over R/m. Let B be
a matrix over R lying over the inverse of A in R/m. AB = I + M , where M is
a matrix whose entries all lie in m. Thus det AB = 1 + x, for x m, so AB is
invertible, so A is invertible.
The homomorphism Rp P given by (r1 , . . . , rp ) 7 r1 a1 + . . . + rp ap inherits
injectivity and surjectivity from A. Hence, P
= Rp .
Next we show that the rank of a finite projective module over R is well defined: if
P
= R , then = . This is immediate in the vector space case, and so it is
= R
true in the general module case as well, by dividing out the maximal ideal on both
sides. If 0 R R R 0 is exact, the sequence splits, so R
= R R ,
so = + .
So far we have seen that the map rankR : K0 (R) Z is a well-defined homomorphism. It is surjective because rankR [R] = 1. It is injective because the element of
K0 (R) with rank is uniquely [R ].

00JK

Lemma 54.8. Let R be a local Artinian ring. There is a commutative diagram

/ K00 (R)

K0 (R)

lengthR

rankR


Z

lengthR (R)


/Z

where the vertical maps are isomorphisms by Lemmas 54.6 and 54.7.
Proof. By induction on the rank of M . Suppose [M ] K0 (R). Then M is
a finite projective R-module over a local ring, so M is free; M
= Rn for some
n. The claim is that rank(M )lengthR (R) = lengthR (M ), or equivalently that
nlengthR (R) = lengthR (Rn ) for all n 1. When n = 1, this is clearly true.
Suppose that (n 1)lengthR (R) = lengthR (Rn1 ). Then since there is a split
short exact sequence
0 R Rn Rn1 0
by Lemma 51.3 we have
lengthR (Rn )

as desired.

lengthR (R) + lengthR (Rn1 )

lengthR (R) + (n 1)lengthR (R)

nlengthR (R)


COMMUTATIVE ALGEBRA

122

55. Graded rings

00JL

A graded
L ring will be for us a ring S endowed with a direct sum decomposition
S =
nonzero
d0 Sd such that Sd Se Sd+e . Note that we do not allow
L
elements in negative degrees. The irrelevant ideal is the ideal S+ = d>0 Sd . A
gradedL
module will be an S-module M endowed with a direct sum decomposition
M =
nZ Mn such that Sd Me Md+e . Note that for modules we do allow
nonzero elements in negative degrees. We think of S as a graded S-module by
setting Sk = (0) for k > 0. An element x (resp. f ) of M (resp. S) is called
homogeneous if x Md (resp. f Sd ) for some d. A map of graded S-modules is a
map of S-modules : M M 0 such that (Md ) Md0 . We do not allow maps to
shift degrees. Let us denote GrHom0 (M, N ) the S0 -module of homomorphisms of
graded modules from M to N .
At this point there are the notions of graded ideal, graded quotient ring, graded
submodule, graded quotient module, graded tensor product, etc. We leave it to the
reader to find the relevant definitions, and lemmas. For example: A short exact
sequence of graded modules is short exact in every degree.
Given a graded ring S, a graded S-module M and n Z we denote M (n) the
graded S-module with M (n)d = Mn+d . This is called the twist of M by n. In
particular we get modules S(n), n Z which will play an important role in the
study of projective schemes. There are some obvious functorial isomorphisms such
as (M N )(n) = M (n) N (n), (M S N )(n) = M S N (n) = M (n) S N . In
addition we can define a graded S-module structure on the S0 -module
M
GrHom(M, N ) =
GrHomn (M, N ), GrHomn (M, N ) = GrHom0 (M, N (n)).
nZ

We omit the definition of the multiplication.

L
Let S be a graded ring. Let d 1 be an integer. We set S (d) =
n0 Snd .
We think of S (d) as a graded ring with degree n summand L
(S (d) )n = Snd . Given
a graded S-module M we can similarly consider M (d) =
nZ Mnd which is a
(d)
graded S -module.
077G

Lemma 55.1. Let R S be a homomorphism of graded rings. Let S 0 S be the

integral closure of R in S. Then
M
S0 =
S 0 Sd ,
d0

i.e., S is a graded R-subalgebra of S.

Proof. We have to show the following: If s = sn + sn+1 + . . . + sm S 0 , then
each homogeneous part sj S 0 . We will prove this by induction on m n over all
homomorphisms R S of graded rings. First note that it is immediate that s0
is integral over R0 (hence over R) as there is a ring map S S0 compatible with
the ring map R R0 . Thus, after replacing s by s s0 , we may assume n > 0.
Consider the extension of graded rings R[t, t1 ] S[t, t1 ] where t has degree 0.
There is a commutative diagram
S[t, t1 ]
O
R[t, t1 ]

s7tdeg(s) s

r7tdeg(r) r

/ S[t, t1 ]
O
/ R[t, t1 ]

COMMUTATIVE ALGEBRA

123

where the horizontal maps are ring automorphisms. Hence the integral closure C
of S[t, t1 ] over R[t, t1 ] maps into itself. Thus we see that
tm (sn + sn+1 + . . . + sm ) (tn sn + tn+1 sn+1 + . . . + tm sm ) C
which implies by induction hypothesis that each (tm ti )si C for i = n, . . . , m1.
Note that for any ring A and m > i n > 0 we have A[t, t1 ]/(tm ti 1)
=
A[t]/(tm ti 1) A because t(tm1 ti1 ) = 1 in A[t]/(tm ti 1). Since
tm ti maps to 1 we see the image of si in the ring S[t]/(tm ti 1) is integral
over R[t]/(tm ti 1) for i = n, . . . , m 1. Since R R[t]/(tm ti 1) is finite
we see that si is integralPover R by transitivity, see Lemma 35.6. Finally, we also
conclude that sm = s i=n,...,m1 si is integral over R.

56. Proj of a graded ring
00JM

Let S be a graded ring. A homogeneous ideal is simply an ideal I S which is also

a graded submodule of S. Equivalently, it is an ideal generated by homogeneous
elements. Equivalently, if f I and
f = f0 + f1 + . . . + fn
is the decomposition of f into homogeneous parts in S then fi I for each i. To
check that a homogeneous ideal p is prime it suffices to check that if ab p with
a, b homogeneous then either a p or b p.

00JN

Definition 56.1. Let S be a graded ring. We define Proj(S) to be the set of

homogeneous, prime ideals p of S such that S+ 6 p. As Proj(S) is a subset of
Spec(S) and we endow it with the induced topology. The topological space Proj(S)
is called the homogeneous spectrum of the graded ring S.
Note that by construction there is a continuous map
Proj(S) Spec(S0 )
Let S = d0 Sd be a graded ring. Let f Sd and assume that d 1. We
define S(f ) to be the subring of Sf consisting of elements of the form r/f n with
r homogeneous and deg(r) = nd. If M is a graded S-module, then we define the
S(f ) -module M(f ) as the sub module of Mf consisting of elements of the form x/f n
with x homogeneous of degree nd.

00JO

Lemma 56.2. Let S be a Z-graded ring containing a homogeneous invertible

element of positive degree. Then the set G Spec(S) of Z-graded primes of S
(with induced topology) maps homeomorphically to Spec(S0 ).
Proof. First we show that the map is a bijection by constructing an inverse. Let
f Sd , d > 0 be invertible in S. If p0 is a prime of S0 , then p0 S is a Z-graded
ideal of S such that p0 S S0 = p0 . And if ab p0 S with a, b homogeneous, then
ad bd /f deg(a)+deg(b) p0 . Thus either ad /f deg(a) p0 or bd /f deg(b) p0 , in other
words either ad p0 S or bd p0 S. It follows that p0 S is a Z-graded prime ideal
of S whose intersection with S0 is p0 .
To show that the
P map is a homeomorphism we show that the image of G D(g)
is
open.
If
g
=
gi with gi Si , then by the above G D(g) maps onto the set
S
D(gid /f i ) which is open.


COMMUTATIVE ALGEBRA

124

For f S homogeneous of degree > 0 we define

D+ (f ) = {p Proj(S) | f 6 p}.
Finally, for a homogeneous ideal I S we define
V+ (I) = {p Proj(S) | I p}.
We will use more generally the notation V+ (E) for any set E of homogeneous
elements E S.
00JP

Lemma 56.3 (Topology on Proj). Let S = d0 Sd be a graded ring.

(1) The sets D+ (f ) are open in Proj(S).
(2) We have D+ (f f 0 ) = D+ (f ) D+ (f 0 ).
(3) Let g = g0 + . . . + gm be an element of S with gi Si . Then
[
D(g) Proj(S) = (D(g0 ) Proj(S))
D+ (gi ).
i1

(4) Let g0 S0 be a homogeneous element of degree 0. Then

[
D(g0 ) Proj(S) =
D+ (g0 f ).
f Sd , d1

(5) The open sets D+ (f ) form a basis for the topology of Proj(S).
(6) Let f S be homogeneous of positive degree. The ring Sf has a natural
Z-grading. The ring maps S Sf S(f ) induce homeomorphisms
D+ (f ) {Z-graded primes of Sf } Spec(S(f ) ).
(7) There exists an S such that Proj(S) is not quasi-compact.
(8) The sets V+ (I) are closed.
(9) Any closed subset T Proj(S) is of the form V+ (I) for some homogeneous
ideal I S.

(10) For any graded ideal I S we have V+ (I) = if and only if S+ I.

Proof. Since D+ (f ) = Proj(S) D(f ), these sets are open. Similarly the sets
V+ (I) = Proj(S) V (E) are closed.
Suppose that T Proj(S) is closed. Then we can write T = Proj(S) V (J) for
some ideal J S. By definition of a homogeneous ideal if g J, g = g0 + . . . + gm
with gd Sd then gd p for all p T . Thus, letting I S be the ideal generated
by the homogeneous parts of the elements of J we have T = V+ (I).
The formula for Proj(S) D(g), with g S is direct from the definitions. Consider
the formula for Proj(S) D(g0 ). The inclusion of the right hand side in the left
hand side is obvious. For the other inclusion, suppose g0 6 p with p Proj(S). If
all g0 f p for all homogeneous f of positive degree, then we see that S+ p which
is a contradiction. This gives the other inclusion.
The collection of opens D(g) Proj(S) forms a basis for the topology since the
standard opens D(g) Spec(S) form a basis for the topology on Spec(S). By the
formulas above we can express D(g) proj(S) as a union of opens D+ (f ). Hence
the collection of opens D+ (f ) forms a basis for the topology also.
First we note that D+ (f ) may be identified with a subset (with induced topology)
of D(f ) = Spec(Sf ) via Lemma 16.6. Note that the ring Sf has a Z-grading. The
homogeneous elements are of the form r/f n with r S homogeneous and have
degree deg(r/f n ) = deg(r) n deg(f ). The subset D+ (f ) corresponds exactly to

COMMUTATIVE ALGEBRA

125

those prime ideals p Sf which are Z-graded ideals (i.e., generated by homogeneous
elements). Hence we have to show that the set of Z-graded prime ideals of Sf maps
homeomorphically to Spec(S(f ) ). This follows from Lemma 56.2.
Let S = Z[X1 , X2 , X3 , . . .] with grading such that each Xi has degree 1. Then it is
easy to see that
[
Proj(S) =
D+ (Xi )
i=1

does not have a finite refinement.

Let I S be a graded ideal. If I S+ then V+ (I) = since every prime

p Proj(S) does not contain S+ by definition. Conversely, suppose that S+ 6 I.

Then we can find an element f S+ such that f is not nilpotent modulo I. Clearly
this means that one of the homogeneous parts of f is not nilpotent modulo I, in
other words we may (and do) assume that f is homogeneous. This implies that
ISf 6= 0, in other words that (S/I)f is not zero. Hence (S/I)(f ) 6= 0 since it is a
ring which maps into (S/I)f . Pick a prime q (S/I)(f ) . This corresponds to a
graded prime of S/I, not containing the irrelevant ideal (S/I)+ . And this in turn
corresponds to a graded prime ideal p of S, containing I but not containing S+ as
desired.

00JQ

Example 56.4. Let R be a ring. If S = R[X] with deg(X) = 1, then the natural
map Proj(S) Spec(R) is a bijection and in fact a homeomorphism. Namely,
suppose p Proj(S). Since S+ 6 p we see that X 6 p. Thus if aX n p with a R
and n > 0, then a p. It follows that p = p0 S with p0 = p R.
If p Proj(S), then we define S(p) to be the ring whose elements are fractions
r/f where r, f S are homogeneous elements of the same degree such that f 6 p.
As usual we say r/f = r0 /f 0 if and only if there exists some f 00 S homogeneous,
f 00 6 p such that f 00 (rf 0 r0 f ) = 0. Given a graded S-module M we let M(p) be the
S(p) -module whose elements are fractions x/f with x M and f S homogeneous
of the same degree such that f 6 p. We say x/f = x0 /f 0 if and only if there exists
some f 00 S homogeneous, f 00 6 p such that f 00 (xf 0 x0 f ) = 0.

00JR

Lemma 56.5. Let S be a graded ring. Let M be a graded S-module. Let p be an

element of Proj(S). Let f S be a homogeneous element of positive degree such that
f 6 p, i.e., p D+ (f ). Let p0 S(f ) be the element of Spec(S(f ) ) corresponding to
p as in Lemma 56.3. Then S(p) = (S(f ) )p0 and compatibly M(p) = (M(f ) )p0 .
Proof. We define a map : M(p) (M(f ) )p0 . Let x/g M(p) . We set
(x/g) = (xg deg(f )1 /f deg(x) )/(g deg(f ) /f deg(g) ).
This makes sense since deg(x) = deg(g) and since g deg(f ) /f deg(g) 6 p0 . We omit
the verification that is well defined, a module map and an isomorphism. Hint:
the inverse sends (x/f n )/(g/f m ) to (xf m )/(gf n ).

Here is a graded variant of Lemma 14.2.

00JS

Lemma 56.6. Suppose S is a graded ring, pi , i = 1, . . . , r homogeneous prime

ideals and I S+ a graded ideal. Assume I 6 pi for all i. Then there exists a
homogeneous element x I of positive degree such that x 6 pi for all i.

COMMUTATIVE ALGEBRA

126

Proof. We may assume there are no inclusions among the pi . The result is true
for r = 1. Suppose the result holds for r 1. Pick x I homogeneous of positive
degree such that x 6 pi for all i = 1, . . . , r 1. If x 6 pr we are done. So assume
x pr . If Ip1 . . . pr1 pr then I pr a contradiction. Pick y Ip1 . . . pr1
homogeneous and y 6 pr . Then xdeg(y) + y deg(x) works.

00JT

Lemma 56.7. Let S be a graded ring. Let p S be a prime. Let q be the

homogeneous ideal of S generated by the homogeneous elements of p. Then q is a
prime ideal of S.
Proof. Suppose f, g S are such that f g q. Let fd (resp. ge ) be the homogeneous
part of f (resp. g) of degree d (resp. e). Assume P
d, e are maxima such that fd 6= 0
and ge 6= 0. By assumption we canPwrite f g =
ai fi with fi p homogeneous.
Say deg(fi ) = di . Then fd ge =
a0i fi with a0i to homogeneous par of degree
d + e di of ai (or 0 if d + e di < 0). Hence fd p or ge p. Hence fd q or
ge q. In the first case replace f by f fd , in the second case replace g by g ge .
Then still f g q but the discrete invariant d + e has been decreased. Thus we may
continue in this fashion until either f or g is zero. This clearly shows that f g q
implies either f q or g q as desired.


00JU

Lemma 56.8. Let S be a graded ring.

(1) Any minimal prime of S is a homogeneous ideal of S.
(2) Given a homogeneous ideal I S any minimal prime over I is homogeneous.
Proof. The first assertion holds because the prime q constructed in Lemma 56.7
satisfies q p. The second because we may consider S/I and apply the first
part.


07Z2

Lemma 56.9. Let R be a ring. Let S be a graded R-algebra. Assume that S is

of finite type over R. Then for every homogeneous f S+ the ring S(f ) is of finite
type over R.
Proof. Choose f1 , . . . , fn S which generate S as an R-algebra. We may assume that each fi is homogeneous (by decomposing each fi into its homogeneous
components). An element of S(f ) is a sum of the form
X
e1 ...en f1e1 . . . fnen /f e
P
e deg(f )=

ei deg(fi )

with e1 ...en R. Thus S(f ) isP

generated as an R-algebra by the f1e1 . . . fnen /f e with
the property that e deg(f ) = ei deg(fi ). If ei deg(f ) then we can write this as
deg(f )

f1e1 . . . fnen /f e = fi

e deg(f )

/f deg(fi ) f1e1 . . . fi i

. . . fnen /f edeg(fi )

deg(f )

Thus we only need

/f deg(fi ) as well as the elements f1e1 . . . fnen /f e
Pthe elements fi
with e deg(f ) = ei deg(fi ) and ei deg(f ). This is a finite list and we win. 
052N

Lemma 56.10. Let R be a ring. Let R0 be a finite type R-algebra, and let M be
a finite R0 -module. There exists a graded R-algebra S, a graded S-module N and
an element f S homogeneous of degree 1 such that
(1) R0
= S(f ) and M
= N(f ) (as modules),
(2) S0 = R and S is generated by finitely many elements of degree 1 over R,
and

COMMUTATIVE ALGEBRA

127

(3) N is a finite S-module.

Proof. We may write R0 = R[x1 , . . . , xn ]/I for some ideal I. For an element
g R[x1 , . . . , xn ] denote g R[x0 , . . . , xn ] the element homogeneous of minimal
degree such that g = g(1, x1 , . . . , xn ). Let I R[X0 , . . . , Xn ] generated by all
elements g, g I. Set S = R[X0 , . . . , Xn ]/I and denote f the image of X0 in S.
By construction we have an isomorphism
S(f ) R0 ,

Xi /X0 7 xi .

To do the same thing with the module M we choose a presentation

X
M = (R0 )r /
R0 kj
jJ

with kj = (k1j , . . . , krj ). Let dij = deg(kij ). Set dj = max{dij }. Set Kij =
d d
X0 j ij kij which is homogeneous of degree dj . With this notation we set
M

(Kij )
S(dj ) S r
N = Coker
jJ

which works. Some details omitted.

57. Noetherian graded rings

00JV

A bit of theory on Noetherian graded rings including some material on Hilbert

polynomials.

07Z4

Lemma 57.1. Let S be a graded ring. A set of homogeneous elements fi S+

generates S as an algebra over S0 if and only if they generate S+ as an ideal of S.
Proof. If the fi generate S as an algebra over S0 then every element in S+ is a
polynomial without constant term in the fP
i and hence S+ is generated by the fi as
an ideal. Conversely, suppose that S+ =
Sfi . We will prove that any element
f of S can be written as a polynomial in the fi with coefficients in S0 . It suffices
to do this for homogeneous elements. Say f has degree d. Then we may perform
induction on d.PThe case d = 0 is immediate. If d > 0 then f S+ hence we
can write f =
gi fi for some gi S. As S is graded we can replace gi by its
homogeneous component of degree d deg(fi ). By induction we see that each gi is
a polynomial in the fi and we win.


00JW

Lemma 57.2. A graded ring S is Noetherian if and only if S0 is Noetherian and

S+ is finitely generated as an ideal of S.
Proof. It is clear that if S is Noetherian then S0 = S/S+ is Noetherian and S+ is
finitely generated. Conversely, assume S0 is Noetherian and S+ finitely generated
as an ideal of S. Pick generators S+ = (f1 , . . . , fn ). By decomposing the fi into
homogeneous pieces we may assume each fi is homogeneous. By Lemma 57.1 we
see that S0 [X1 , . . . Xn ] S sending Xi to fi is surjective. Thus S is Noetherian
by Lemma 30.1.


00JX

Definition 57.3. Let A be an abelian group. We say that a function f : n 7

f (n) A defined for all sufficient large integers n is a numerical polynomial if there
exists r 0, elements a0 , . . . , ar A such that
Xr  n 
f (n) =
ai
i=0
i

COMMUTATIVE ALGEBRA

128

for all n  0.
The reason for using the binomial coefficients is the elementary fact that any polynomial P Q[T ] all
of whose values at integer points are integers, is equal to a

P T
+1
sum P (T ) =
ai i with ai Z. Note that in particular the expressions Ti+1
are of this form.
00JY

Lemma 57.4. If A A0 is a homomorphism of abelian groups and if f : n 7

f (n) A is a numerical polynomial, then so is the composition.
Proof. This is immediate from the definitions.

00JZ

Lemma 57.5. Suppose that f : n 7 f (n) A is defined for all n sufficiently

large and suppose that n 7 f (n) f (n 1) is a numerical polynomial. Then f is
a numerical polynomial.


Pr
n
Proof. Let f (n) f (n 1) =
i=0 i ai for all n  0. Set g(n) = f (n)


Pr
n+1
i=0 i+1 ai . Then g(n) g(n 1) = 0 for all n  0. Hence g is eventually con

Pr
stant, say equal to a1 . We leave it to the reader to show that a1 + i=0 n+1
i+1 ai
has the required shape (see remark above the lemma).


00K0

Lemma 57.6. If M is a finitely generated graded S-module, and if S is finitely

generated over S0 , then each Mn is a finite S0 -module.
Proof. Suppose the generators of M are mi and the generators of S are fi . By
taking homogeneous components we may assume that the mi and the fi are homogeneous and we may assume fi SQ
+ . In this case it is clear that each Mn is
generated over S0 by the monomials fiei mj whose degree is n.


00K1

Proposition 57.7. Suppose that S is a Noetherian graded ring and M a finite

graded S-module. Consider the function
Z K00 (S0 ),

n 7 [Mn ]

see Lemma 57.6. If S+ is generated by elements of degree 1, then this function is a

numerical polynomial.
Proof. We prove this by induction on the minimal number of generators of S1 . If
this number is 0, then Mn = 0 for all n  0 and the result holds. To prove the
induction step, let x S1 be one of a minimal set of generators, such that the
induction hypothesis applies to the graded ring S/(x).
First we show the result holds if x is nilpotent on M . This we do by induction on
the minimal integer r such that xr M = 0. If r = 1, then M is a module over S/xS
and the result holds (by the other induction hypothesis). If r > 1, then we can find
a short exact sequence 0 M 0 M M 00 0 such that the integers r0 , r00 are
strictly smaller than r. Thus we know the result for M 00 and M 0 . Hence we get the
result for M because of the relation [Md ] = [Md0 ] + [Md00 ] in K00 (S0 ).
If x is not nilpotent on M , let M 0 M be the largest submodule on which x
is nilpotent. Consider the exact sequence 0 M 0 M M/M 0 0 we see
again it suffices to prove the result for M/M 0 . In other words we may assume that
multiplication by x is injective.

COMMUTATIVE ALGEBRA

129

Let M = M/xM . Note that the map x : M M is not a map of graded S-modules,
since it does not map Md into Md . Namely, for each d we have the following short
exact sequence
x
0 Md
Md+1 M d+1 0
This proves that [Md+1 ] [Md ] = [M d+1 ]. Hence we win by Lemma 57.5.

02CD

Remark 57.8. If S is still Noetherian but S is not generated in degree 1, then

the function associated to a graded S-module is a periodic polynomial (i.e., it is a
numerical polynomial on the congruence classes of integers modulo n for some n).

00K2

Example 57.9. Suppose that S = k[X1 , . . . , Xd ]. By Example 54.2 we may

identify K0 (k) = K00 (k) = Z. Hence any finitely generated graded k[X1 , . . . , Xd ]module gives rise to a numerical polynomial n 7 dimk (Mn ).

00K3

Lemma 57.10. Let k be a field. Suppose that I k[X1 , . . . , Xd ] is a nonzero

graded ideal. Let M = k[X1 , . . . , Xd ]/I. Then the numerical polynomial n 7
dimk (Mn ) (see Example 57.9) has degree < d 1 (or is zero if d = 1).
Proof. The numerical
polynomial associated to the graded module k[X1 , . . . , Xn ]


is n 7 n1+d
.
For
any
nonzero homogeneous f I of degree e and any degree
d1


n >> e we have In f k[X1 , . . . , Xd ]ne and hence dimk (In ) ne1+d
. Hence
d1




ne1+d
dimk (Mn ) n1+d

.
We
win
because
the
last
expression
has
degree
d1
d1
< d 1 (or is zero if d = 1).

58. Noetherian local rings

00K4

In all of this section (R, m, ) is a Noetherian local ring. We develop some theory
on Hilbert functions of modules in this section. Let M be a finite R-module. We
define the Hilbert function of M to be the function
M : n 7 lengthR (mn M/mn+1 M )
defined for all integers n 0. Another important invariant is the function
M : n 7 lengthR (M/mn+1 M )
defined for all integers n 0. Note that we have by Lemma 51.3 that
Xn
M (n) =
M (i).
i=0

There is a variant of this construction which uses an ideal of definition.

07DU

Definition 58.1. Let (R, m) be a local Noetherian ring. An ideal I R such that

I = m is called an ideal of definition of R.

Let I R be an ideal of definition. Because R is Noetherian this means that
mr I for some r, see Lemma 31.4. Hence any finite R-module annihilated by a
power of I has a finite length, see Lemma 51.8. Thus it makes sense to define
I,M (n) = lengthR (I n M/I n+1 M )

and I,M (n) = lengthR (M/I n+1 M )

for all n 0. Again we have that

I,M (n) =

Xn
i=0

I,M (i).

COMMUTATIVE ALGEBRA

00K5

130

Lemma 58.2. Suppose that M 0 M are finite R-modules with finite length
quotient. Then there exists a constants c1 , c2 such that for all n c2 we have
c1 + I,M 0 (n c2 ) I,M (n) c1 + I,M 0 (n)
Proof. Since M/M 0 has finite length there is a c2 0 such that I c2 M M 0 . Let
c1 = lengthR (M/M 0 ). For n c2 we have
I,M (n)

lengthR (M/I n+1 M )

= c1 + lengthR (M 0 /I n+1 M )
c1 + lengthR (M 0 /I n+1 M 0 )
= c1 + I,M 0 (n)
On the other hand, since I c2 M M 0 , we have I n M I nc2 M 0 for n c2 . Thus
for n c2 we get
I,M (n)

lengthR (M/I n+1 M )

c1 + lengthR (M 0 /I n+1 M )

c1 + lengthR (M 0 /I n+1c2 M 0 )

c1 + I,M 0 (n c2 )

which finishes the proof.

00K6

Lemma 58.3. Suppose that 0 M 0 M M 00 0 is a short exact sequence

of finite R-modules. Then there exists a submodule N M 0 with finite colength l
and c 0 such that
I,M (n) = I,M 00 (n) + I,N (n c) + l
and
I,M (n) = I,M 00 (n) + I,N (n c)
for all n c.
Proof. Note that M/I n M M 00 /I n M 00 is surjective with kernel M 0 /M 0 I n M .
By the Artin-Rees Lemma 50.2 there exists a constant c such that M 0 I n M =
I nc (M 0 I c M ). Denote N = M 0 I c M . Note that I c M 0 N M 0 . Hence
lengthR (M 0 /M 0 I n M ) = lengthR (M 0 /N ) + lengthR (N/I nc N ) for n c. From
the short exact sequence
0 M 0 /M 0 I n M M/I n M M 00 /I n M 00 0
and additivity of lengths (Lemma 51.3) we obtain the equality
I,M (n 1) = I,M 00 (n 1) + I,N (n c 1) + lengthR (M 0 /N )
for n c. We have I,M (n) = I,M (n) I,M (n 1) and similarly for the modules
M 00 and N . Hence we get I,M (n) = I,M 00 (n) + I,N (n c) for n c.


00K7

Lemma 58.4. Suppose that I, I 0 are two ideals of definition for the Noetherian
local ring R. Let M be a finite R-module. There exists a constant a such that
I,M (n) I 0 ,M (an) for n 1.
Proof. There exists an integer c such that (I 0 )c I. Hence we get a surjection
M/(I 0 )c(n+1) M M/I n+1 M . Whence the result with a = c + 1.


COMMUTATIVE ALGEBRA

131

00K8

Proposition 58.5. Let R be a Noetherian local ring. Let M be a finite R-module.

Let I R be an ideal of definition. The Hilbert function I,M and the function
I,M are numerical polynomials.
L
Proof. Consider the graded ring S = R/I I/I 2 I 2 /I 3 . . . = d0 I d /I d+1 .
L
Consider the graded S-module N = M/IM IM/I 2 M . . . = d0 I d M/I d+1 M .
This pair (S, N ) satisfies the hypotheses of Proposition 57.7. Hence the result for
I,M follows from that proposition and Lemma 54.1. The result for I,M follows
from this and Lemma 57.5.


09CA

Definition 58.6. Let R be a Noetherian local ring. Let M be a finite R-module.

The Hilbert polynomial of M over R is the element P (t) Q[t] such that P (n) =
M (n) for n  0.
By Proposition 58.5 we see that the Hilbert polynomial exists.

00K9

Lemma 58.7. Let R be a Noetherian local ring. Let M be a finite R-module.

(1) The degree of the numerical polynomial I,M is independent of the ideal of
definition I.
(2) The degree of the numerical polynomial I,M is independent of the ideal of
definition I.
Proof. Part (2) follows immediately from Lemma 58.4. Part (1) follows from (2)
because I,M (n) = I,M (n) I,M (n 1) for n 1.


00KA

Definition 58.8. Let R be a local Noetherian ring and M a finite R-module. We

denote d(M ) the element of {, 0, 1, 2, . . .} defined as follows:
(1) If M = 0 we set d(M ) = ,
(2) if M 6= 0 then d(M ) is the degree of the numerical polynomial M .
If mn M 6= 0 for all n, then we see that d(M ) is the degree +1 of the Hilbert
polynomial of M .

00KB

Lemma 58.9. Let R be a Noetherian local ring. Let I R be an ideal of definition.

Let M be a finite R-module which does not have finite length. If M 0 M is
a submodule with finite colength, then I,M I,M 0 is a polynomial of degree <
degree of either polynomial.
Proof. Follows from Lemma 58.2 by elementary calculus.

00KC

Lemma 58.10. Let R be a Noetherian local ring. Let I R be an ideal of

definition. Let 0 M 0 M M 00 0 be a short exact sequence of finite
R-modules. Then
(1) if M 0 does not have finite length, then I,M I,M 00 I,M 0 is a numerical
polynomial of degree < the degree of I,M 0 ,
(2) max{deg(I,M 0 ), deg(I,M 00 )} = deg(I,M ), and
(3) max{d(M 0 ), d(M 00 )} = d(M ),
Proof. We first prove (1). Let N M 0 be as in Lemma 58.3. By Lemma 58.9 the
numerical polynomial I,M 0 I,N has degree < the common degree of I,M 0 and
I,N . By Lemma 58.3 the difference
I,M (n) I,M 00 (n) I,N (n c)

COMMUTATIVE ALGEBRA

132

is constant for n  0. By elementary calculus the difference I,N (n) I,N (n c)

has degree < the degree of I,N which is bigger than zero (see above). Putting
everything together we obtain (1).
Note that the leading coefficients of I,M 0 and I,M 00 are nonnegative. Thus the
degree of I,M 0 + I,M 00 is equal to the maximum of the degrees. Thus if M 0 does
not have finite length, then (2) follows from (1). If M 0 does have finite length, then
I n M I n M 00 is an isomorphism for all n  0 by Artin-Rees (Lemma 50.2). Thus
M/I n M M 00 /I n M 00 is a surjection with kernel M 0 for n  0 and we see that
I,M (n) I,M 00 (n) = length(M 0 ) for all n  0. Thus (2) holds in this case also.
Proof of (3). This follows from (2) except if one of M , M 0 , or M 00 is zero. We omit
the proof in these special cases.

59. Dimension
00KD
00KE

Definition 59.1. The Krull dimension of the ring R is the Krull dimension of the
topological space Spec(R), see Topology, Definition 9.1. In other words it is the
supremum of the integers n 0 such that there exists a chain of prime ideals of
length n:
p0 p1 . . . pn , pi 6= pi+1 .

00KF

Definition 59.2. The height of a prime ideal p of a ring R is the dimension of the
local ring Rp .

00KG

Lemma 59.3. The Krull dimension of R is the supremum of the heights of its
(maximal) primes.
Proof. This is so because we can always add a maximal ideal at the end of a chain
of prime ideals.


00KH

Lemma 59.4. A Noetherian ring of dimension 0 is Artinian. Conversely, any

Artinian ring is Noetherian of dimension zero.
Proof. By Lemma 30.5 the space Spec(R) is Noetherian. By Topology, Lemma
8.2 we see that Spec(R) has finitely many irreducible components, say Spec(R) =
Z1 . . . Zr . According to Lemma 25.1, each Zi = V (pi ) with pi a minimal ideal.
Since the dimension is 0 these pi are also maximal. Thus Spec(R) is the discrete
topological space with elements pi . All elements f of the radical I = pi are
nilpotent since otherwise Rf would not be the zero ring and we would have another
prime. Since I is finitely generated we conclude that I is nilpotent, Lemma 31.4.
By Lemma 52.5 R is the product of its local rings. By Lemma 51.8 each of these
has finite length over R. Hence we conclude that R is Artinian by Lemma 52.6.
If R is Artinian then by Lemma 52.6 it is Noetherian. All of its primes are maximal
by a combination of Lemmas 52.3, 52.4 and 52.5.

In the following we will use the invariant d() defined in Definition 58.8. Here is a
warm up lemma.

00KI

Lemma 59.5. Let R be a Noetherian local ring. Then dim(R) = 0 d(R) = 0.

Proof. This is because d(R) = 0 if and only if R has finite length as an R-module.
See Lemma 52.6.


COMMUTATIVE ALGEBRA

00KJ

Proposition 59.6. Let R be a ring. The following are equivalent:

(1) R is Artinian,
(2) R is Noetherian and dim(R) = 0,
(3) R has finite length as a module over itself,
(4) R is a finite product of Artinian local rings,
(5) R is Noetherian and Spec(R) is a finite discrete topological space,
(6) R is a finite product of Noetherian local rings of dimension 0,
(7) R is a finite product of Noetherian local rings Ri with d(Ri ) = 0,
(8) R is a finite product of Noetherian local rings Ri whose maximal ideals are
nilpotent,
(9) R is Noetherian, has finitely many maximal ideals and its radical ideal is
nilpotent, and
(10) R is Noetherian and there are no strict inclusions among its primes.
Proof. This is a combination of Lemmas 52.5, 52.6, 59.4, and 59.5.

00KK
00KL
00KM
00KN
00KO
00KP

133

Lemma 59.7. Let R be a local Noetherian ring. The following are equivalent:
(1)
(2)
(3)
(4)
(5)

dim(R) = 1,
d(R) = 1,
there exists an x m, x not nilpotent such that V (x) = {m},
p
there exists an x m, x not nilpotent such that m = (x), and
there exists an ideal of definition generated by 1 element, and no ideal of
definition is generated by 0 elements.

Proof. First, assume that dim(R) = 1. Let pi be the minimal primes of R. Because
the dimension is 1 the only other prime of R is m. According to Lemma 30.6 there
are finitely many. Hence we can find x m, x 6 pi , see Lemma 14.2. Thus the only
prime containing x is m and hence (3).
p
If (3) then m = (x) by Lemma 16.2, and hence (4). The converse is clear as well.
The equivalence of (4) and (5) follows from directly the definitions.
Assume (5). Let I = (x) be an ideal of definition. Note that I n /I n+1 is a quotient
of R/I via multiplication by xn and hence lengthR (I n /I n+1 ) is bounded. Thus
d(R) = 0 or d(R) = 1, but d(R) = 0 is excluded by the assumption that 0 is not an
ideal of definition.
Assume (2). To get a contradiction, assume there exist primes p q m, with
both inclusions strict. Pick some ideal of definition I R. We will repeatedly use
Lemma 58.10. First of all it implies, via the exact sequence 0 p R R/p 0,
that d(R/p) 1. But it clearly cannot be zero. Pick x q, x 6 p. Consider the
short exact sequence
0 R/p R/p R/(xR + p) 0.
This implies that I,R/p I,R/p I,R/(xR+p) = I,R/(xR+p) has degree < 1. In
other words, d(R/(xR + p) = 0, and hence dim(R/(xR + p)) = 0, by Lemma 59.5.
But R/(xR + p) has the distinct primes q/(xR + p) and m/(xR + p) which gives
the desired contradiction.

00KQ
00KR

Proposition 59.8. Let R be a local Noetherian ring. Let d 0 be an integer.

The following are equivalent:
(1) dim(R) = d,

COMMUTATIVE ALGEBRA

00KS
00KT

134

(2) d(R) = d,
(3) there exists an ideal of definition generated by d elements, and no ideal of
definition is generated by fewer than d elements.
Proof. This proof is really just the same as the proof of Lemma 59.7. We will
prove the proposition by induction on d. By Lemmas 59.5 and 59.7 we may assume
that d > 1. Denote the minimal number of generators for an ideal of definition of
R by d0 (R). We will prove that the inequalities dim(R) d0 (R) d(R) dim(R),
and hence they are all equal.
First, assume that dim(R) = d. Let pi be the minimal primes of R. According
to Lemma 30.6 there are finitely many. Hence we can find x m, x 6 pi , see
Lemma 14.2. Note that every maximal chain of primes starts with some pi , hence
the dimension of R/xR is at most d 1. By induction there are x2 , . . . , xd which
generate an ideal of definition in R/xR. Hence R has an ideal of definition generated
by (at most) d elements.
Assume d0 (R) = d. Let I = (x1 , . . . , xd ) be
an ideal of definition. Note that
copies R/I via multiplication by
I n /I n+1 is a quotient of a direct sum of d+n1
d1
all degree n monomials in x1 , . . . , xn . Hence lengthR (I n /I n+1 ) is bounded by a
polynomial of degree d 1. Thus d(R) d.
Assume d(R) = d. Consider a chain of primes p q q2 . . . pe = m,
with all inclusions strict, and e 2. Pick some ideal of definition I R. We
will repeatedly use Lemma 58.10. First of all it implies, via the exact sequence
0 p R R/p 0, that d(R/p) d. But it clearly cannot be zero. Pick
x q, x 6 p. Consider the short exact sequence
0 R/p R/p R/(xR + p) 0.
This implies that I,R/p I,R/p I,R/(xR+p) = I,R/(xR+p) has degree < d.
In other words, d(R/(xR + p)) d 1, and hence dim(R/(xR + p)) d 1, by
induction. Now R/(xR+p) has the chain of prime ideals q/(xR+p) q2 /(xR+p)
. . . qe /(xR + p) which gives e 1 d 1. Since we started with an arbitrary
chain of primes this proves that dim(R) d(R).
Reading back the reader will see we proved the circular inequalities as desired.

Let (R, m) be a Noetherian local ring. From the above it is clear that m cannot be
generated by fewer than dim(R) variables. By Nakayamas Lemma 19.1 the minimal
number of generators of m equals dim(m) m/m2 . Hence we have the following
fundamental inequality
dim(R) dim(m) m/m2 .
It turns out that the rings where equality holds have a lot of good properties. They
are called regular local rings.
00KU

Definition 59.9. Let (R, m) be a Noetherian local ring of dimension d.

(1) A system of parameters of R is a sequence of elements x1 , . . . , xd m which
generates an ideal of definition of R,
(2) if there exist x1 , . . . , xd m such that m = (x1 , . . . , xd ) then we call R a
regular local ring and x1 , . . . , xd a regular system of parameters.
The following lemmas are clear from the proofs of the lemmas and proposition
above, but we spell them out so we have convenient references.

COMMUTATIVE ALGEBRA

00KV

135

Lemma 59.10. Let R be a Noetherian ring. Let x R.

(1) If p is minimal over (x) then the height of p is 0 or 1.
(2) If p, q Spec(R) and q is minimal over (p, x), then there is no prime
strictly between p and q.
Proof. Proof of (1). If p is minimal over x, then the only prime ideal of Rp
containing x is the maximal ideal pRp . This is true because the primes of Rp
correspond 1-to-1 with the primes of R contained in p, see Lemma 16.5. Hence
Lemma 59.7 shows dim(Rp ) = 1 if x is not nilpotent in Rp . Of course, if x is
nilpotent in Rp the argument gives that pRp is the only prime ideal and we see
that the height is 0.
Proof of (2). By part (1) we see that p/q is a prime of height 1 or 0 in R/q. This
immediately implies there cannot be a prime strictly between p and q.


0BBZ

Lemma 59.11. Let R be a Noetherian ring. Let f1 , . . . , fr R.

(1) If p is minimal over (f1 , . . . , fr ) then the height of p is r.
(2) If p, q Spec(R) and q is minimal over (p, f1 , . . . , fr ), then every chain of
primes between p and q has length at most r.
Proof. Proof of (1). If p is minimal over f1 , . . . , fr , then the only prime ideal of Rp
containing f1 , . . . , fr is the maximal ideal pRp . This is true because the primes of
Rp correspond 1-to-1 with the primes of R contained in p, see Lemma 16.5. Hence
Proposition 59.8 shows dim(Rp ) r.
Proof of (2). By part (1) we see that p/q is a prime of height r. This immediately
implies the statement about chains of primes between p and q.


00KW

Lemma 59.12. Suppose that R is a Noetherian local ring and x m an element

of its maximal ideal. Then dim R dim R/xR + 1. If x is not contained in any of
the minimal primes of R then equality holds. (For example if x is a nonzerodivisor.)
Proof. If x1 , . . . , xdim R/xR R map to elements of R/xR which generate an ideal
of definition for R/xR, then x, x1 , . . . , xdim R/xR generate an ideal of definition for
R. Hence the inequality by Proposition 59.8. On the other hand, if x is not
contained in any minimal prime of R, then the chains of primes in R/xR all give
rise to chains in R which are at least one step away from being maximal.


02IE

Lemma 59.13. Let (R, m) be a Noetherian local ring. Suppose x1 , . . . , xd m

generate an ideal of definition and d = dim(R). Then dim(R/(x1 , . . . , xi )) = d i
for all i = 1, . . . , d.
Proof. Follows either from the proof of Proposition 59.8, or by using induction on
d and Lemma 59.12.

60. Applications of dimension theory

02IF

We can use the results on dimension to prove certain rings have infinite spectra and
to produce more Jacobson rings.

02IG

Lemma 60.1. Let R be a Noetherian local domain of dimension 2. A nonempty

open subset U Spec(R) is infinite.

COMMUTATIVE ALGEBRA

136

Proof. To get a contradiction, assume that U Spec(R) is finite. In this case

(0) U and {(0)} is an open subset of U (because the complement of {(0)} is
the union of the closures of the other points). Thus we may assume U = {(0)}.
Let m R be the maximal ideal. We can find an x m, x 6= 0 such that
V (x) U = Spec(R). In other words we see that D(x) = {(0)}. In particular we
see that dim(R/xR) = dim(R) 1 1, see Lemma 59.12. Let y 2 , . . . , y dim(R)
R/xR generate an ideal of definition of R/xR, see Proposition 59.8. Choose lifts
y2 , . . . , ydim(R) R, so that x, y2 , . . . , ydim(R) generate an ideal of definition in R.
This implies that dim(R/(y2 )) = dim(R) 1 and dim(R/(y2 , x)) = dim(R) 2,
see Lemma 59.13. Hence there exists a prime p containing y2 but not x. This
contradicts the fact that D(x) = {(0)}.

The rings k[[t]] where k is a field, or the ring of p-adic numbers are Noetherian
rings of dimension 1 with finitely many primes. This is the maximum dimension
for which this can happen.
0ALV

Lemma 60.2. A Noetherian ring with finitely many primes has dimension 1.
Proof. Let R be a Noetherian ring with finitely many primes. If R is a local
domain, then the lemma follows from Lemma 60.1. If R is a domain, then Rm
has dimension 1 for all maximal ideals m by the local case. Hence dim(R) 1
by Lemma 59.3. If R is general, then dim(R/q) 1 for every minimal prime
q of R. Since every prime contains a minimal prime (Lemma 16.2), this implies
dim(R) 1.


0ALW

Lemma 60.3. Let S be a nonzero finite type algebra over a field k. Then dim(S) =
0 if and only if S has finitely many primes.
Proof. Recall that Spec(S) is sober, Noetherian, and Jacobson, see Lemmas 25.2,
30.5, 34.2, and 34.4. If it has dimension 0, then every point defines an irreducible
component and there are only a finite number of irreducible components (Topology,
Lemma 8.2). Conversely, if Spec(S) is finite, then it is discrete by Topology, Lemma
17.6 and hence the dimension is 0.


00KX

Lemma 60.4. Noetherian Jacobson rings.

(1) Any Noetherian domain R of dimension 1 with infinitely many primes is
Jacobson.
(2) Any Noetherian ring such that every prime p is either maximal or contained
in infinitely many prime ideals is Jacobson.
Proof. Part (1) is a reformulation of Lemma 34.6.
Let R be a Noetherian ring such that every non-maximal prime p is contained in infinitely many prime ideals. Assume Spec(R) is not Jacobson to get a contradiction.
By Lemmas 25.1 and 30.5 we see that Spec(R) is a sober, Noetherian topological
space. By Topology, Lemma 17.3 we see that there exists a non-maximal ideal
p R such that {p} is a locally closed subset of Spec(R). In other words, p is not
maximal and {p} is an open subset of V (p). Consider a prime q R with p q.
Recall that the topology on the spectrum of (R/p)q = Rq /pRq is induced from that
of Spec(R), see Lemmas 16.5 and 16.7. Hence we see that {(0)} is a locally closed
subset of Spec((R/p)q ). By Lemma 60.1 we conclude that dim((R/p)q ) = 1. Since
this holds for every q p we conclude that dim(R/p) = 1. At this point we use

COMMUTATIVE ALGEBRA

137

the assumption that p is contained in infinitely many primes to see that Spec(R/p)
is infinite. Hence by part (1) of the lemma we see that V (p)
= Spec(R/p) is the
closure of its closed points. This is the desired contradiction since it means that
{p} V (p) cannot be open.

61. Support and dimension of modules
00KY
00L0

Lemma 61.1. Let R be a Noetherian ring, and let M be a finite R-module. There
exists a filtration by R-submodules
0 = M0 M1 . . . Mn = M
such that each quotient Mi /Mi1 is isomorphic to R/pi for some prime ideal pi of
R.

00L4

Proof. By Lemma 5.4 it suffices to do the case M = R/I for some ideal I. Consider
the set S of ideals J such that the lemma does not hold for the module R/J, and
order it by inclusion. To arrive at a contradiction, assume that S is not empty.
Because R is Noetherian, S has a maximal element J. By definition of S, the
ideal J cannot be prime. Pick a, b R such that ab J, but neither a J nor
b J. Consider the filtration 0 aR/(J aR) R/J. Note that aR/(J aR) is
a quotient of R/(J + bR) and the second quotient equals R/(aR + J). Hence by
maximality of J, each of these has a filtration as above and hence so does R/J.
Contradiction.

S
Lemma 61.2. Let R, M , Mi , pi as in Lemma 61.1. Then Supp(M ) = V (pi )
and in particular pi Supp(M ).
Proof. This follows from Lemmas 39.5 and 39.8.

00L5

Lemma 61.3. Suppose that R is a Noetherian local ring with maximal ideal m.
Let M be a nonzero finite R-module. Then Supp(M ) = {m} if and only if M has
finite length over R.
Proof. Assume that Supp(M ) = {m}. It suffices to show that all the primes pi in
the filtration of Lemma 61.1 are the maximal ideal. This is clear by Lemma 61.2.
Suppose that M has finite length over R. Then mn M = 0 by Lemma 51.4. Since
some element of m maps to a unit in Rp for any prime p 6= m in R we see Mp = 0. 

00L6

Lemma 61.4. Let R be a Noetherian ring. Let I R be an ideal. Let M be a

finite R-module. Then I n M = 0 for some n 0 if and only if Supp(M ) V (I).
Proof. It is clear that I n M = 0 for some n 0 implies Supp(M ) V (I). Suppose
that Supp(M ) V (I). Choose a filtration 0 = M0 M1 . . . Mn = M as in
Lemma 61.1. Each of the primes pi is contained in V (I) by Lemma 61.2. Hence
I pi and I annihilates Mi /Mi1 . Hence I n annihilates M .


00L7

Lemma 61.5. Let R, M , Mi , pi as in Lemma 61.1. The minimal elements of the

set {pi } are the minimal elements of Supp(M ). The number of times a minimal
prime p occurs is
#{i | pi = p} = lengthRp Mp .

COMMUTATIVE ALGEBRA

138

S
Proof. The first statement follows because Supp(M ) = V (pi ), see Lemma 61.2.
Let p Supp(M ) be minimal. The support of Mp is the set consisting of the
maximal ideal pRp . Hence by Lemma 61.3 the length of Mp is finite and > 0. Next
we note that Mp has a filtration with subquotients (R/pi )p = Rp /pi Rp These are
zero if pi 6 p and equal to (p) if pi p because by minimality of p we have pi = p
in this case. The result follows since (p) has length 1.

00L8

Lemma 61.6. Let R be a Noetherian local ring. Let M be a finite R-module.

Then d(M ) = dim(Supp(M )).
Proof. Let Mi , pi be as in Lemma 61.1. By Lemma 58.10 we obtain the equality
d(M ) = max{d(R/pi )}. By Proposition 59.8 we have d(R/pi ) = dim(R/pi ). Trivially dim(R/pi ) = dim V (pi ). Since all minimal primes of Supp(M ) occur among
the pi (Lemma 61.5) we win.


0B51

Lemma 61.7. Let R be a Noetherian ring. Let 0 M 0 M M 00 0 be a

short exact sequence of finite R-modules. Then max{dim(Supp(M 0 )), dim(Supp(M 00 ))} =
dim(Supp(M )).
Proof. If R is local, this follows immediately from Lemmas 61.6 and 58.10. A
more elementary argument, which works also if R is not local, is to use that
Supp(M 0 ), Supp(M 00 ), and Supp(M ) are closed (Lemma 39.5) and that Supp(M ) =
Supp(M 0 ) Supp(M 00 ) (Lemma 39.8).

62. Associated primes

00L9

Here is the standard definition. For non-Noetherian rings and non-finite modules
it may be more appropriate to use the definition in Section 65.

00LA

Definition 62.1. Let R be a ring. Let M be an R-module. A prime p of R is

associated to M if there exists an element m M whose annihilator is p. The set
of all such primes is denoted AssR (M ) or Ass(M ).

0586

Lemma 62.2. Let R be a ring. Let M be an R-module. Then Ass(M ) Supp(M ).

Proof. If m M has annihilator p, then in particular no element of R\p annihilates
m. Hence m is a nonzero element of Mp , i.e., p Supp(M ).


02M3

Lemma 62.3. Let R be a ring. Let 0 M 0 M M 00 0 be a short

exact sequence of R-modules. Then Ass(M 0 ) Ass(M ) and Ass(M ) Ass(M 0 )
Ass(M 00 ).
Proof. Omitted.

00LB

Lemma 62.4. Let R be a ring, and M an R-module. Suppose there exists a

filtration by R-submodules
0 = M0 M1 . . . Mn = M
such that each quotient Mi /Mi1 is isomorphic to R/pi for some prime ideal pi of
R. Then Ass(M ) {p1 , . . . , pn }.
Proof. By induction on the length n of the filtration {Mi }. Pick m M whose
annihilator is a prime p. If m Mn1 we are done by induction. If not, then m
maps to a nonzero element of M/Mn1
= R/pn . Hence we have p pn . If equality
does not hold, then we can find f pn , f 6 p. In this case the annihilator of f m
is still p and f m Mn1 . Thus we win by induction.


COMMUTATIVE ALGEBRA

00LC

Lemma 62.5. Let R be a Noetherian ring. Let M be a finite R-module. Then

Ass(M ) is finite.
Proof. Immediate from Lemma 62.4 and Lemma 61.1.

02CE

139

Proposition 62.6. Let R be a Noetherian ring. Let M be a finite R-module. The

following sets of primes are the same:
(1) The minimal primes in the support of M .
(2) The minimal primes in Ass(M ).
(3) For any filtration 0 = M0 M1 . . . Mn1 Mn = M with
Mi /Mi1
= R/pi the minimal primes of the set {pi }.
Proof. Choose a filtration as in (3). In Lemma 61.5 we have seen that the sets in
(1) and (3) are equal.
Let p be a minimal element of the set {pi }. Let i be minimal such that p = pi .
Pick m Mi , m 6 Mi1 . The annihilator of m is contained in pi = p and contains
p1 p2 . . . pi . By our choice of i and p we have pj 6 p for j < i and hence we have
p1 p2 . . . pi1 6 pi . Pick f p1 p2 . . . pi1 , f 6 p. Then f m has annihilator p.
In this way we see that p is an associated prime of M . By Lemma 62.2 we have
Ass(M ) Supp(M ) and hence p is minimal in Ass(M ). Thus the set of primes in
(1) is contained in the set of primes of (2).
Let p be a minimal element of Ass(M ). Since Ass(M ) Supp(M ) there is a
minimal element q of Supp(M ) with q p. We have just shown that q Ass(M ).
Hence q = p by minimality of p. Thus the set of primes in (2) is contained in the
set of primes of (1).


0587

Lemma 62.7. Let R be a Noetherian ring. Let M be an R-module. Then

M = (0) Ass(M ) = .
Proof. If M = (0), then Ass(M ) = by definition. If M 6= 0, pick any nonzero
finitely generated submodule M 0 M , for example a submodule generated by
a single nonzero element. By Lemma 39.2 we see that Supp(M 0 ) is nonempty.
By Proposition 62.6 this implies that Ass(M 0 ) is nonempty. By Lemma 62.3 this
implies Ass(M ) 6= .


05BV

Lemma 62.8. Let R be a Noetherian ring. Let M be an R-module. Any p

Supp(M ) which is minimal among the elements of Supp(M ) is an element of
Ass(M ).
Proof. If M is a finiteS
R-module, then this is a consequence of Proposition 62.6.
In general write
M
=
M as the union
S
S of its finite submodules, and use that
Supp(M ) = Supp(M ) and Ass(M ) = Ass(M ).


00LD

Lemma
62.9. Let R be a Noetherian ring. Let M be an R-module. The union
S
qAss(M ) q is the set of elements of R which are zerodivisors on M .
Proof. Any element in any associated prime clearly is a zerodivisor on M . Conversely, suppose x R is a zerodivisor on M . Consider the submodule N = {m
M | xm = 0}. Since N is not zero it has an associated prime q by Lemma 62.7.
Then x q and q is an associated prime of M by Lemma 62.3.


COMMUTATIVE ALGEBRA

0B52

140

Lemma 62.10. Let R is a Noetherian local ring, M a finite R-module, and f m

an element of the maximal ideal of R. Then
dim(Supp(M/f M )) dim(Supp(M )) dim(Supp(M/f M )) + 1
If f is not in any of the minimal primes of the support of M (for example if f is
a nonzerodivisor on M ), then equality holds for the right inequality.
Proof. (The parenthetical statement follows from Lemma 62.9.) The first inequality follows from Supp(M/f M ) Supp(M ), see Lemma 39.8. For the second inequality, note that Supp(M/f M ) = Supp(M ) V (f ), see Lemma 39.8. It follows,
for example by Lemma 61.2 and elementary properties of dimension, that it suffices
to show dim V (p) dim(V (p) V (f )) + 1 for primes p of R. This is a consequence
of Lemma 59.12. Finally, if f is not contained in any minimal prime of the support
of M , then the chains of primes in Supp(M/f M ) all give rise to chains in Supp(M )
which are at least one step away from being maximal.


05BW

Lemma 62.11. Let : R S be a ring map. Let M be an S-module. Then

Spec()(AssS (M )) AssR (M ).
Proof. If q AssS (M ), then there exists an m in M such that the annihilator of
m in S is q. Then the annihilator of m in R is q R.


05BX

Remark 62.12. Let : R S be a ring map. Let M be an S-module. Then it

is not always the case that Spec()(AssS (M )) AssR (M ). For example, consider
the ring map R = k S = k[x1 , x2 , x3 , . . .]/(x2i ) and M = S. Then AssR (M ) is
not empty, but AssS (S) is empty.

05DZ

Lemma 62.13. Let : R S be a ring map. Let M be an S-module. If S is

Noetherian, then Spec()(AssS (M )) = AssR (M ).
Proof. We have already seen in Lemma 62.11 that Spec()(AssS (M )) AssR (M ).
For the converse, choose a prime p AssR (M ). Let m M be an element such
that the annihilator of m in R is p. Let I = {g S | gm = 0} be the annihilator
of m in S. Then R/p S/I is injective, hence there exists a prime q S lying
over p, see Lemma 29.5. By Proposition 62.6 we see that q is an associated prime
of S/I, hence an associated prime of M by Lemma 62.3 and we win.


05BY

Lemma 62.14. Let R be a ring. Let I be an ideal. Let M be an R/I-module. Via

the canonical injection Spec(R/I) Spec(R) we have AssR/I (M ) = AssR (M ).
Proof. Omitted.

0310

Lemma 62.15. Let R be a ring. Let M be an R-module. Let p R be a prime.

(1) If p Ass(M ) then pRp Ass(Mp ).
(2) If p is finitely generated then the converse holds as well.
Proof. If p Ass(M ) there exists an element m M whose annihilator is p. As
localization is exact (Proposition 9.12) we see that the annihilator of m/1 in Mp
is pRp hence (1) holds. Assume pRp Ass(Mp ) and p = (f1 , . . . , fn ). Let m/g be
an element of Mp whose annihilator is pRp . This implies that the annihilator of m
is contained in p. As fi m/g = 0 in Mp we see there exists a gi R, gi 6 p such
that gi fi m = 0 in M . Combined we see the annihilator of g1 . . . gn m is p. Hence
p Ass(M ).


COMMUTATIVE ALGEBRA

05BZ

141

Lemma 62.16. Let R be a ring. Let M be an R-module. Let S R be a

multiplicative subset. Via the canonical injection Spec(S 1 R) Spec(R) we have
(1) AssR (S 1 M ) = AssS 1 R (S 1 M ),
(2) AssR (M ) Spec(S 1 R) AssR (S 1 M ), and
(3) if R is Noetherian this inclusion is an equality.
Proof. The first equality follows, since if m S 1 M , then the annihilator of m
in R is the intersection of the annihilator of m in S 1 R with R. The displayed
inclusion and equality in the Noetherian case follows from Lemma 62.15 since for
p R, S p = we have Mp = (S 1 M )S 1 p .


05C0

Lemma 62.17. Let R be a ring. Let M be an R-module. Let S R be a

multiplicative subset. Assume that every s S is a nonzerodivisor on M . Then
AssR (M ) = AssR (S 1 M ).
Proof. As M S 1 M by assumption we get the inclusion Ass(M ) = Ass(S 1 M )
from Lemma 62.3. Conversely, suppose that n/s S 1 M is an element whose
annihilator is a prime ideal p. Then the annihilator of n M is also p.


00LL

Lemma 62.18. Let R be a Noetherian local ring with maximal ideal m. Let I m
be an ideal. Let M be a finite R-module. The following are equivalent:
(1) There exists an x I which is not a zerodivisor on M .
(2) We have I 6 q for all q Ass(M ).
Proof. If there exists a nonzerodivisor x in I, then x clearly cannot be in any
associated prime of M . Conversely, suppose I 6 q for all q Ass(M ). In this
case we can choose x I, x 6 q for all q Ass(M ) by Lemmas 62.5 and 14.2. By
Lemma 62.9 the element x is not a zerodivisor on M .


0311

Lemma 62.19. Let R be a ring. Let M be an R-module. If R is Noetherian the

map
Y
M
Mp
pAss(M )

is injective.
Proof. Let x M be an element of the kernel of the map. Then if p is an
associated prime of Rx M we see on the one hand that p Ass(M ) (Lemma
62.3) and on the other hand that (Rx)p Mp is not zero. This contradiction shows
that Ass(Rx) = . Hence Rx = 0 by Lemma 62.7.

63. Symbolic powers
05G9

Here is the definition.

0313

Definition 63.1. Let R be a ring. Let p be a prime ideal. For n 0 the nth
symbolic power of p is the ideal p(n) = Ker(R Rp /pn Rp ).
Note that pn p(n) but equality does not always hold.

0314

Lemma 63.2. Let R be a Noetherian ring. Let p be a prime ideal. Let n > 0.
Then Ass(R/p(n) ) = {p}.
Proof. If q is an associated prime of R/p(n) then clearly p q. On the other hand,
any element x R, x 6 p is a nonzerodivisor on R/p(n) . Namely, if y R and
xy p(n) = R pn Rp then y pn Rp , hence y p(n) . Hence the lemma follows. 

COMMUTATIVE ALGEBRA

0BC0

142

Lemma 63.3. Let R S be flat ring map. Let p R be a prime such that
q = pS is a prime of S. Then p(n) S = q(n) .
Proof. Since p(n) = Ker(R Rp /pn Rp ) we see using flatness that p(n) S is the
kernel of the map S Sp /pn Sp . On the other hand q(n) is the kernel of the map
S Sq /qn Sq = Sq /pn Sq . Hence it suffices to show that
Sp /pn Sp Sq /pn Sq
is injective. Observe that the right hand module is the localization of the left
hand module by elements f S, f 6 q. Thus it suffices to show these elements are
nonzerodivisors on Sp /pn Sp . By flatness, the module Sp /pn Sp has a finite filtration
whose subquotients are
pi Sp /pi+1 Sp
= pi Rp /pi+1 Rp R Sp
= V (p) (S/q)p
p

where V is a (p) vector space. Thus f acts invertibly as desired.

64. Relative assassin

05GA

Discussion of relative assassins. Let R S be a ring map. Let N be an S-module.

In this situation we can introduce the following sets of primes q of S:
A with p = R q we have that q AssS (N R (p)),
A0 with p = R q we have that q is in the image of AssS(p) (N R (p))
under the canonical map Spec(S R (p)) Spec(S),
Af in with p = R q we have that q AssS (N/pN ),
A0f in for some prime p0 R we have q AssS (N/p0 N ),
B for some R-module M we have q AssS (N R M ), and
Bf in for some finite R-module M we have q AssS (N R M ).
Let us determine some of the relations between theses sets.

05GB

Lemma 64.1. Let R S be a ring map. Let N be an S-module. Let A, A0 ,

Af in , B, and Bf in be the subsets of Spec(S) introduced above.
(1) We always have A = A0 .
(2) We always have Af in A, Bf in B, Af in A0f in Bf in and A B.
(3) If S is Noetherian, then A = Af in and B = Bf in .
(4) If N is flat over R, then A = Af in = A0f in and B = Bf in .
(5) If R is Noetherian and N is flat over R, then all of the sets are equal, i.e.,
A = A0 = Af in = A0f in = B = Bf in .
Proof. Some of the arguments in the proof will be repeated in the proofs of later
lemmas which are more precise than this one (because they deal with a given module
M or a given prime p and not with the collection of all of them).
Proof of (1). Let p be a prime of R. Then we have
AssS (N R (p)) = AssS/pS (N R (p)) = AssSR (p) (N R (p))
the first equality by Lemma 62.14 and the second by Lemma 62.16 part (1). This
prove that A = A0 . The inclusion Af in A0f in is clear.
Proof of (2). Each of the inclusions is immediate from the definitions except perhaps
Af in A which follows from Lemma 62.16 and the fact that we require p = R q
in the formulation of Af in .

COMMUTATIVE ALGEBRA

143

Proof of (3). The equality A = Af in follows from Lemma 62.16 part (3) if S is
Noetherian. Let q = (g1 , . . . , gm ) be a finitely generated prime ideal of S. Say
z N R M is an element whose annihilator is q. We may pick a finite submodule
M 0 M such that z is the image of z 0 N R M 0 . Then AnnS (z 0 ) q = AnnS (z).
Since N R commutes with colimits and since M is the directed colimit of finite
R-modules we can find M 0 M 00 M such that the image z 00 N R M 00 is
annihilated by g1 , . . . , gm . Hence AnnS (z 00 ) = q. This proves that B = Bf in if S is
Noetherian.
Proof of (4). If N is flat, then the functor N R is exact. In particular, if M 0 M ,
then N R M 0 N R M . Hence if z N R M is an element whose annihilator
q = AnnS (z) is a prime, then we can pick any finite R-submodule M 0 M such
that z N R M 0 and we see that the annihilator of z as an element of N R M 0 is
equal to q. Hence B = Bf in . Let p0 be a prime of R and let q be a prime of S which
is an associated prime of N/p0 N . This implies that p0 S q. As N is flat over R we
see that N/p0 N is flat over the integral domain R/p0 . Hence every nonzero element
of R/p0 is a nonzerodivisor on N/p0 . Hence none of these elements can map to an
element of q and we conclude that p0 = R q. Hence Af in = A0f in . Finally, by
Lemma 62.17 we see that AssS (N/p0 N ) = AssS (N R (p0 )), i.e., A0f in = A.
Proof of (5). We only need to prove A0f in = Bf in as the other equalities have been
proved in (4). To see this let M be a finite R-module. By Lemma 61.1 there exists
a filtration by R-submodules
0 = M0 M1 . . . Mn = M
such that each quotient Mi /Mi1 is isomorphic to R/pi for some prime ideal pi of
R. Since N is flat we obtain a filtration by S-submodules
0 = N R M0 N R M1 . . . N R Mn = N R M
such that each subquotient
S is isomorphic to N/pi N . By Lemma 62.3 we conclude
that AssS (N R M ) AssS (N/pi N ). Hence we see that Bf in A0f in . Since the
other inclusion is part of (2) we win.

We define the relative assassin of N over S/R to be the set A = A0 above. As a
motivation we point out that it depends only on the fibre modules N R (p) over
the fibre rings. As in the case of the assassin of a module we warn the reader that
this notion makes most sense when the fibre rings S R (p) are Noetherian, for
example if R S is of finite type.
05GC

Definition 64.2. Let R S be a ring map. Let N be an S-module. The relative

assassin of N over S/R is the set
AssS/R (N ) = {q S | q AssS (N R (p)) with p = R q}.
This is the set named A in Lemma 64.1.
The spirit of the next few results is that they are about the relative assassin, even
though this may not be apparent.

0312

Lemma 64.3. Let R S be a ring map. Let M be an R-module, and let N be

an S-module. If N is flat as R-module, then
[
AssS (M R N )
AssS (N/pN )
pAssR (M )

and if R is Noetherian then we have equality.

COMMUTATIVE ALGEBRA

144

Proof. If p AssR (M ) then there exists an injection R/p M . As N is flat over

R we obtain an injection R/p R N M R N . Since R/p R N = N/pN we
conclude that AssS (N/pN ) AssS (M R N ), see Lemma 62.3. Hence the right
hand side is contained in the left hand side.
S
Write M = S M as the union of its finitely generated R-submodules. Then also
N R M = N R MS(as N is R-flat). By definition of associated
primes we see
S
that AssS (N R M ) = AssS (N R M ) and AssR (M ) = Ass(M ). Hence we
may assume M is finitely generated.
Let q AssS (M R N ), and assume R is Noetherian and M is a finite R-module.
To finish the proof we have to show that q is an element of the right hand side.
First we observe that qSq AssSq ((M R N )q ), see Lemma 62.15. Let p be the
corresponding prime of R. Note that
(M R N )q = M R Nq = Mp Rp Nq
If pRp 6 AssRp (Mp ) then there exists an element x pRp which is a nonzerodivisor
in Mp (see Lemma 62.18). Since Nq is flat over Rp we see that the image of x in
qSq is a nonzerodivisor on (M R N )q . This is a contradiction with the assumption
that qSq AssS ((M R N )q ). Hence we conclude that p is one of the associated
primes of M .
Continuing the argument we choose a filtration
0 = M0 M1 . . . M n = M
such that each quotient Mi /Mi1 is isomorphic to R/pi for some prime ideal pi of
R, see Lemma 61.1. (By Lemma 62.4 we have pi = p for at least one i.) This gives
a filtration
0 = M0 R N M1 R N . . . Mn R N = M R N
with subquotients isomorphic to N/pi N . If pi 6= p then q cannot be associated to the
module N/pi N by the result of the preceding paragraph (as AssR (R/pi ) = {pi }).
Hence we conclude that q is associated to N/pN as desired.

05C1

Lemma 64.4. Let R S be a ring map. Let N be an S-module. Assume N is

flat as an R-module and R is a domain with fraction field K. Then
AssS (N ) = AssS (N R K) = AssSR K (N R K)
via the canonical inclusion Spec(S R K) Spec(S).
Proof. Note that S R K = (R \ {0})1 S and N R K = (R \ {0})1 N . For any
nonzero x R multiplication by x on N is injective as N is flat over R. Hence the
lemma follows from Lemma 62.17 combined with Lemma 62.16 part (1).


05C2

Lemma 64.5. Let R S be a ring map. Let M be an R-module, and let N be

an S-module. Assume N is flat as R-module. Then
[
AssS (M R N )
AssSR (p) (N R (p))
pAssR (M )

where we use Remark 16.8 to think of the spectra of fibre rings as subsets of Spec(S).
If R is Noetherian then this inclusion is an equality.
Proof. This is equivalent to Lemma 64.3 by Lemmas 62.14, 38.7, and 64.4.

COMMUTATIVE ALGEBRA

05E0

145

Remark 64.6. Let R S be a ring map. Let N be an S-module. Let p be a

prime of R. Then
AssS (N R (p)) = AssS/pS (N R (p)) = AssSR (p) (N R (p)).
The first equality by Lemma 62.14 and the second by Lemma 62.16 part (1).
65. Weakly associated primes

0546

This is a variant on the notion of an associated prime that is useful for nonNoetherian ring and non-finite modules.

0547

Definition 65.1. Let R be a ring. Let M be an R-module. A prime p of R is

weakly associated to M if there exists an element m M such that p is minimal
among the prime ideals containing the annihilator Ann(m) = {f R | f m = 0}.
The set of all such primes is denoted WeakAssR (M ) or WeakAss(M ).
Thus an associated prime is a weakly associated prime. Here is a characterization
in terms of the localization at the prime.

0566

Lemma 65.2. Let R be a ring. Let M be an R-module. Let p be a prime of R.

The following are equivalent:
(1) p is weakly associated to M ,
(2) pRp is weakly associated to Mp , and
(3) Mp contains an element whose annihilator has radical equal to pRp .
Proof. Assume (1). Then there exists an element m M such that p is minimal
among the primes containing the annihilator I = {x R | xm = 0} of m. As
localization is exact, the annihilator of m in Mp is Ip . Hence pRp is a minimal
prime of Rp containing thepannihilator Ip of m in Mp . This implies (2) holds, and
also (3) as it implies that Ip = pRp .
Applying the implication (1) (3) to Mp over Rp we see that (2) (3).
Finally, assume (3). This means there exists an element m/f Mp whose annihilator has radicalp
equal to pRp . Then the annihilator I = {x R | xm = 0} of m in
M is such that Ip = pRp . Clearly this means that p contains I and is minimal
among the primes containing I, i.e., (1) holds.


0548

Lemma 65.3. Let R be a ring. Let 0 M 0 M M 00 0 be a short exact

sequence of R-modules. Then WeakAss(M 0 ) WeakAss(M ) and WeakAss(M )
WeakAss(M 0 ) WeakAss(M 00 ).
Proof. We will use the characterization of weakly associated primes of Lemma
65.2. Let p be a prime of R. As localization is exact we obtain the short exact
sequence 0 Mp0 Mp Mp00 0. Suppose that m Mp is an element
whose annihilator has radical pRp . Then either the image m of m in Mp00 is zero
and m Mp0 , or the annihilator of m is pRp . This proves that WeakAss(M )
WeakAss(M 0 ) WeakAss(M 00 ). The inclusion WeakAss(M 0 ) WeakAss(M ) is
immediate from the definitions.


0588

Lemma 65.4. Let R be a ring. Let M be an R-module. Then

M = (0) WeakAss(M ) =

COMMUTATIVE ALGEBRA

146

Proof. If M = (0) then WeakAss(M ) = by definition. Conversely, suppose

that M 6= 0. Pick a nonzero element m M . Write I = {x R | xm = 0}
the annihilator of m. Then R/I M . Hence WeakAss(R/I) WeakAss(M ) by
Lemma 65.3. But as I 6= R we have V (I) = Spec(R/I) contains a minimal prime,
see Lemmas 16.2 and 16.7, and we win.

0589

Lemma 65.5. Let R be a ring. Let M be an R-module. Then

Ass(M ) WeakAss(M ) Supp(M ).

05C3

Proof. The first inclusion is immediate from the definitions. If p WeakAss(M ),

then by Lemma 65.2 we have Mp 6= 0, hence p Supp(M ).

S
Lemma 65.6. Let R be a ring. Let M be an R-module. The union qWeakAss(M ) q
is the set elements of R which are zerodivisors on M .
Proof. Suppose f q WeakAss(M ). Then there exists an element m M
such that q is minimal over I = {x R | xm = 0}. Hence there exists a g R,
g 6 q and n > 0 such that f n gm = 0. Note that gm 6= 0 as g 6 I. If we take
n minimal as above, then f (f n1 gm) = 0 and f n1 gm 6= 0, so f is a zerodivisor
on M . Conversely, suppose f R is a zerodivisor on M . Consider the submodule
N = {m M | f m = 0}. Since N is not zero it has a weakly associated prime q
by Lemma 65.4. Clearly f q and by Lemma 65.3 q is a weakly associated prime
of M .


05C4

Lemma 65.7. Let R be a ring. Let M be an R-module. Any p Supp(M ) which

is minimal among the elements of Supp(M ) is an element of WeakAss(M ).
Proof. Note that Supp(Mp ) = {pRp } in Spec(Rp ). In particular Mp is nonzero,
and hence WeakAss(Mp ) 6= by Lemma 65.4. Since WeakAss(Mp ) Supp(Mp ) by
Lemma 65.5 we conclude that WeakAss(Mp ) = {pRp }, whence p WeakAss(M )
by Lemma 65.2.


058A

Lemma 65.8. Let R be a ring. Let M be an R-module. Let p be a prime ideal of

R which is finitely generated. Then
p Ass(M ) p WeakAss(M ).
In particular, if R is Noetherian, then Ass(M ) = WeakAss(M ).
Proof. Write p = (g1 , . . . , gn ) for some gi R. It is enough the prove the implication as the other implication holds in general, see Lemma 65.5. Assume
p WeakAss(M ). By Lemma 65.2 there exists an element m Mp such that
I = {x Rp | xm = 0} has radical pRp . Hence for each i there exists a smallest
ei > 0 such that giei m = 0 in Mp . If ei > 1 for some i, then we can replace m
P
by giei 1 m 6= 0 and decrease
ei . Hence we may assume that the annihilator of
m Mp is (g1 , . . . , gn )Rp = pRp . By Lemma 62.15 we see that p Ass(M ).


05C5

Remark 65.9. Let : R S be a ring map. Let M be an S-module. Then it

is not always the case that Spec()(WeakAssS (M )) WeakAssR (M ) contrary to
the case of associated primes (see Lemma 62.11). An example is to consider the
ring map
R = k[x1 , x2 , x3 , . . .] S = k[x1 , x2 , x3 , . . . , y1 , y2 , y3 , . . .]/(x1 y1 , x2 y2 , x3 y3 , . . .)

COMMUTATIVE ALGEBRA

147

P
and M = S. In this case q =
xi S is a minimal prime of S, hence a weakly
associated prime of M = S (see Lemma 65.7). But on the other hand, for any
nonzero element of S the annihilator in R is finitely generated, and hence does not
have radical equal to R q = (x1 , x2 , x3 , . . .) (details omitted).
05C6

Lemma 65.10. Let : R S be a ring map. Let M be an S-module. Then we

have Spec()(WeakAssS (M )) WeakAssR (M ).
Proof. Let p be an element of WeakAssR (M ). Then there exists an m Mp
whose annihilator I = {x Rp | xm = 0} has radical pRp . Consider the radical
J = {x Sp | xm = 0} of m in Sp . As ISp J we see that any minimal prime
q Sp over J lies over p. Moreover such a q corresponds to a weakly associated
prime of M for example by Lemma 65.2.


05C7

Remark 65.11. Let : R S be a ring map. Let M be an S-module. Denote

f : Spec(S) Spec(R) the associated map on spectra. Then we have
f (AssS (M )) AssR (M ) WeakAssR (M ) f (WeakAssS (M ))
see Lemmas 62.11, 65.10, and 65.5. In general all of the inclusions may be strict, see
Remarks 62.12 and 65.9. If S is Noetherian, then all the inclusions are equalities
as the outer two are equal by Lemma 65.8.

05E1

Lemma 65.12. Let : R S be a ring map. Let M be an S-module. Denote

f : Spec(S) Spec(R) the associated map on spectra. If is a finite ring map,
then
WeakAssR (M ) = f (WeakAssS (M )).
Proof. One of the inclusions has already been proved, see Remark 65.11. To prove
the other assume q WeakAssS (M ) and let p be the corresponding prime of R. Let
m M be an element such that q is a minimal prime over J = {g S | gm = 0}.
Thus the radical of JSq is qSq . As R S is finite there are finitely many primes
q = q1 , q2 , . . . , ql over p, see Lemma 35.19. Pick x q with x 6 qi for i > 1, see
Lemma 14.2. By the above there exists an element y S, y 6 q and an integer
t > 0 such that yxt m = 0. Thus the element ym M is annihilated by xt , hence
ym maps to zero in Mqi , i = 2, . . . , l. To be sure, ym does not map to zero in Sq .
The ring Sp is semi-local with maximal ideals qi Sp by going up for finite ring maps,
see Lemma 35.20. If f pRp then some power of f ends up in JSq hence for some
n > 0 we see that f t ym maps to zero in Mq . As ym vanishes at the other maximal
ideals of Sp we conclude that f t ym is zero in Mp , see Lemma 23.1. In this way we
see that p is a minimal prime over the annihilator of ym in R and we win.


05C8

Lemma 65.13. Let R be a ring. Let I be an ideal. Let M be an R/I-module.

Via the canonical injection Spec(R/I) Spec(R) we have WeakAssR/I (M ) =
WeakAssR (M ).
Proof. Omitted.

05C9

Lemma 65.14. Let R be a ring. Let M be an R-module. Let S R be a

multiplicative subset. Via the canonical injection Spec(S 1 R) Spec(R) we have
WeakAssR (S 1 M ) = WeakAssS 1 R (S 1 M ) and
WeakAss(M ) Spec(S 1 R) = WeakAss(S 1 M ).

COMMUTATIVE ALGEBRA

148

Proof. Suppose that m S 1 M . Let I = {x R | xm = 0} and I 0 = {x0

S 1 R | x0 m = 0}. Then I 0 = S 1 I and I S = unless I = R (verifications
omitted). Thus primes in S 1 R minimal over I 0 correspond bijectively to primes in
R minimal over I and avoiding S. This proves the equality WeakAssR (S 1 M ) =
WeakAssS 1 R (S 1 M ). The second equality follows from Lemma 62.15 since for
p R, S p = we have Mp = (S 1 M )S 1 p .

05CA

Lemma 65.15. Let R be a ring. Let M be an R-module. Let S R be a

multiplicative subset. Assume that every s S is a nonzerodivisor on M . Then
WeakAss(M ) = WeakAss(S 1 M ).
Proof. As M S 1 M by assumption we obtain WeakAss(M ) WeakAss(S 1 M )
from Lemma 65.3. Conversely, suppose that n/s S 1 M is an element with
annihilator I and p a prime which is minimal over I. Then the annihilator of
n M is I and p is a prime minimal over I.


05CB

Lemma 65.16. Let R be a ring. Let M be an R-module. The map

Y
M
Mp
pWeakAss(M )

is injective.
Proof. Let x M be an element of the kernel of the map. Set N = Rx M . If
p is a weakly associated prime of N we see on the one hand that p WeakAss(M )
(Lemma 65.3) and on the other hand that Np Mp is not zero. This contradiction
shows that WeakAss(N ) = . Hence N = 0, i.e., x = 0 by Lemma 65.4.

05CC

Lemma 65.17. Let R S be a ring map. Let N be an S-module. Assume N is

flat as an R-module and R is a domain with fraction field K. Then
WeakAssS (N ) = WeakAssSR K (N R K)
via the canonical inclusion Spec(S R K) Spec(S).
Proof. Note that S R K = (R \ {0})1 S and N R K = (R \ {0})1 N . For any
nonzero x R multiplication by x on N is injective as N is flat over R. Hence the
lemma follows from Lemma 65.15.

66. Embedded primes

02M4

Here is the definition.

02M5

Definition 66.1. Let R be a ring. Let M be an R-module.

(1) The associated primes of M which are not minimal among the associated
primes of M are called the embedded associated primes of M .
(2) The embedded primes of R are the embedded associated primes of R as an
R-module.
Here is a way to get rid of these.

02M6

Lemma 66.2. Let R be a Noetherian ring. Let M be a finite R-module. Consider

the set of R-submodules
{K M | Supp(K) nowhere dense in Supp(M )}.
This set has a maximal element K and the quotient M 0 = M/K has the following
properties

COMMUTATIVE ALGEBRA

149

(1) Supp(M ) = Supp(M 0 ),

(2) M 0 has no embedded associated primes,
(3) for any f R which is contained in all embedded associated primes of M
we have Mf
= Mf0 .
Proof. Let q1 , . . . , qt denote the minimal primes in the support of M . Let p1 , . . . , ps
denote the embedded associated primes of M . Then
Q Ass(M ) = {qj , pi }. There are
finitely many of these, see Lemma 62.5. Set I = i=1,...,s pi . Then I 6 qj for any
j. Hence by Lemma 14.2 we can find an f I such that f 6 qj for all j = 1, . . . , t.
Set M 0 = Im(M Mf ). This implies that Mf
= Mf0 . Since M 0 Mf we see that
0
0
Ass(M ) Ass(Mf ) = {qj }. Thus M has no embedded associated primes.
Moreover, the support of K = Ker(M M 0 ) is contained in V (p1 ) . . . V (ps ),
because Ass(K) Ass(M ) (see Lemma 62.3) and Ass(K) contains none of the qi
by construction. Clearly, K is in fact the largest submodule of M whose support
is contained in V (p1 ) . . . V (pt ). This implies that K is the maximal element of
the set displayed in the lemma.

02M7

Lemma 66.3. Let R be a Noetherian ring. Let M be a finite R-module. For any
f R we have (M 0 )f = (Mf )0 where M M 0 and Mf (Mf )0 are the quotients
constructed in Lemma 66.2.
Proof. Omitted.

02M8

Lemma 66.4. Let R be a Noetherian ring. Let M be a finite R-module without

embedded associated primes. Let I = {x R | xM = 0}. Then the ring R/I has
no embedded primes.
Proof. We may replace R by R/I. Hence we may assume every nonzero element
of R acts nontrivially on M . By Lemma 39.5 this implies that Spec(R) equals
the support of M . Suppose that p is an embedded prime of R. Let x R be
an element whose annihilator is p. Consider the nonzero module N = xM M .
It is annihilated by p. Hence any associated prime q of N contains p and is also
an associated prime of M . Then q would be an embedded associated prime of M
which contradicts the assumption of the lemma.

67. Regular sequences

0AUH
00LF

In this section we develop some basic properties of regular sequences.

Definition 67.1. Let R be a ring. Let M be an R-module. A sequence of elements
f1 , . . . , fr of R is called an M -regular sequence if the following conditions hold:
(1) fi is a nonzerodivisor on M/(f1 , . . . , fi1 )M for each i = 1, . . . , r, and
(2) the module M/(f1 , . . . , fr )M is not zero.
If I is an ideal of R and f1 , . . . , fr I then we call f1 , . . . , fr a M -regular sequence
in I. If M = R, we call f1 , . . . , fr simply a regular sequence (in I).
Please pay attention to the fact that the definition depends on the order of the
elements f1 , . . . , fr (see examples below). Some papers/books drop the requirement
that the module M/(f1 , . . . , fr )M is nonzero. This has the advantage that being a
regular sequence is preserved under localization. However, we will use this definition
mainly to define the depth of a module in case R is local; in that case the fi are
required to be in the maximal ideal a condition which is not preserved under
going from R to a localization Rp .

COMMUTATIVE ALGEBRA

150

00LG

Example 67.2. Let k be a field. In the ring k[x, y, z] the sequence x, y(1x), z(1
x) is regular but the sequence y(1 x), z(1 x), x is not.

00LH

Example 67.3. Let k be a field. Consider the ring k[x, y, w0 , w1 , w2 , . . .]/I where
I is generated by ywi , i = 0, 1, 2, . . . and wi xwi+1 , i = 0, 1, 2, . . .. The sequence
x, y is regular, but y is a zerodivisor. Moreover you can localize at the maximal
ideal (x, y, wi ) and still get an example.

00LJ

Lemma 67.4. Let R be a local Noetherian ring. Let M be a finite R-module. Let
x1 , . . . , xc be an M -regular sequence. Then any permutation of the xi is a regular
sequence as well.
Proof. First we do the case c = 2. Consider K M the kernel of x2 : M M . For
any z K we know that z = x1 z 0 for some z 0 M because x2 is a nonzerodivisor on
M/x1 M . Because x1 is a nonzerodivisor on M we see that x2 z 0 = 0 as well. Hence
x1 : K K is surjective. Thus K = 0 by Nakayamas Lemma 19.1. Next, consider
multiplication by x1 on M/x2 M . If z M maps to an element z M/x2 M in
the kernel of this map, then x1 z = x2 y for some y M . But then since x1 , x2 is
a regular sequence we see that y = x1 y 0 for some y 0 M . Hence x1 (z x2 y 0 ) = 0
and hence z = x2 y 0 and hence z = 0 as desired.
For the general case, observe that any permutation is a composition of transpositions of adjacent indices. Hence it suffices to prove that
x1 , . . . , xi2 , xi , xi1 , xi+1 , . . . , xc
is an M -regular sequence. This follows from the case we just did applied to the
module M/(x1 , . . . , xi2 ) and the length 2 regular sequence xi1 , xi .


00LM

Lemma 67.5. Let R, S be local rings. Let R S be a flat local ring homomorphism. Let x1 , . . . , xr be a sequence in R. Let M be an R-module. The following
are equivalent
(1) x1 , . . . , xr is an M -regular sequence in R, and
(2) the images of x1 , . . . , xr in S form a M R S-regular sequence.
Proof. This is so because R S is faithfully flat by Lemma 38.17.

061L

Lemma 67.6. Let R be a Noetherian ring. Let M be a finite R-module. Let p be

a prime. Let x1 , . . . , xr be a sequence in R whose image in Rp forms an Mp -regular
sequence. Then there exists a g R, g 6 p such that the image of x1 , . . . , xr in Rg
forms an Mg -regular sequence.
Proof. Set
Ki = Ker (xi : M/(x1 , . . . , xi1 )M M/(x1 , . . . , xi1 )M ) .
This is a finite R-module whose localization at p is zero by assumption. Hence there
exists a g R, g 6 p such that (Ki )g = 0 for all i = 1, . . . , r. This g works.


065K

Lemma 67.7. Let A be a ring. Let I be an ideal generated by a regular sequence

f1 , . . . , fn in A. Let g1 , . . . , gm A be elements whose images g 1 , . . . , g m form a
regular sequence in A/I. Then f1 , . . . , fn , g1 , . . . , gm is a regular sequence in A.
Proof. This follows immediately from the definitions.

07DV

Lemma 67.8. Let R be a ring. Let M be an R-module. Let f1 , . . . , fr R be

M -regular. Then for e1 , . . . , er > 0 the sequence f1e1 , . . . , frer is M -regular too.

COMMUTATIVE ALGEBRA

151

Proof. We will prove this by induction on r. If r = 1 this follows from the fact
that a power of an M -regular element is an M -regular element. If r > 1, then by
induction applied to M/f1 M we have that f1 , f2e2 , . . . , frer is an M -regular sequence.
Thus it suffices to show that f1e , f2 , . . . , fr is an M -regular sequence if f1 , . . . , fr is
an M -regular sequence. We will prove this by induction on e. The case e = 1 is
trivial. Since f1 is a nonzerodivisor we have a short exact sequence
f e1

1
0 M/f1 M
M/f1e M M/f1e1 M 0

By induction the elements f2 , . . . , fr are M/f1 M and M/f1e1 M -regular sequences.

It follows from the snake lemma that they are also M/f1e M -regular sequences. 
07DW

Lemma 67.9. Let R be a ring. Let f1 , . . . , fr R which do not generate the unit
ideal. The following are equivalent:
(1) any permutation of f1 , . . . , fr is a regular sequence,
(2) any subsequence of f1 , . . . , fr (in the given order) is a regular sequence, and
(3) f1 x1 , . . . , fr xr is a regular sequence in the polynomial ring R[x1 , . . . , xr ].
Proof. It is clear that (1) implies (2). We prove (2) implies (1) by induction on r.
The case r = 1 is trivial. The case r = 2 says that if a, b R are a regular sequence
and b is a nonzerodivisor, then b, a is a regular sequence. This is clear because the
kernel of a : R/(b) R/(b) is isomorphic to the kernel of b : R/(a) R/(a) if both
a and b are nonzerodivisors. The case r > 2. Assume (2) holds and say we want
to prove f(1) , . . . , f(r) is a regular sequence for some permutation . We already
know that f(1) , . . . , f(r1) is a regular sequence by induction. Hence it suffices to
show that fs where s = (r) is a nonzerodivisor modulo f1 , . . . , fs , . . . , fr . If s = r
we are done. If s < r, then note that fs and fr are both nonzerodivisors in the ring
R/(f1 , . . . , fs , . . . , fr1 ) (by induction hypothesis again). Since we know fs , fr is a
regular sequence in that ring we conclude by the case of sequence of length 2 that
fr , fs is too.
Note that R[x1 , . . . , xr ]/(f1 x1 , . . . , fi xi ) as an R-module is a direct sum of the
modules
R/IE xe11 . . . xerr
indexed by multi-indices E = (e1 , . . . , er ) where IE is the ideal generated by fj for
1 j i with ej > 0. Hence fi+1 xi is a nonzerodivisor on this if and only if fi+1
is a nonzerodivisor on R/IE for all E. Taking E with all positive entries, we see
that fi+1 is a nonzerodivisor on R/(f1 , . . . , fi ). Thus (3) implies (2). Conversely, if
(2) holds, then any subsequence of f1 , . . . , fi , fi+1 is a regular sequence by Lemma
67.8, i.e., hence fi+1 is a nonzerodivisor on all R/IE . In this way we see that (2)
implies (3).

68. Quasi-regular sequences

061M

061N

There is a notion of regular sequence which is slightly weaker than that of a regular
sequence and easier to use. Let R be a ring and let f1 , . . . , fc R. Set J =
(f1 , . . . , fc ). Let M be an R-module. Then there is a canonical map
M
(68.0.1)
M/JM R/J R/J[X1 , . . . , Xc ]
J n M/J n+1 M
n0

of graded R/J[X1 , . . . , Xc ]-modules defined by the rule

m X1e1 . . . Xcec 7 f1e1 . . . fcec m mod J e1 +...+ec +1 M.

COMMUTATIVE ALGEBRA

152

Note that (68.0.1) is always surjective.

061P

Definition 68.1. Let R be a ring. Let M be an R-module. A sequence of elements

f1 , . . . , fc of R is called M -quasi-regular if (68.0.1) is an isomorphism. If M = R,
we call f1 , . . . , fc simply a quasi-regular sequence.
So if f1 , . . . , fc is a quasi-regular sequence, then
M
R/J[X1 , . . . , Xc ] =

n0

J n /J n+1

where J = (f1 , . . . , fc ). It is clear that being a quasi-regular sequence is independent

of the order of f1 , . . . , fc .
00LN

Lemma 68.2. Let R be a ring.

(1) A regular sequence f1 , . . . , fc of R is a quasi-regular sequence.
(2) Suppose that M is an R-module and that f1 , . . . , fc is an M -regular sequence. Then f1 , . . . , fc is an M -quasi-regular sequence.
Proof. Set J = (f1 , . . . , fc ). We prove
P the first assertion by induction on c. We
have to show that given any relation |I|=n aI f I J n+1 with aI R we actually
n+1
have
is of the form
P aI IJ for all multi-indices I. Since any element of J
b
f
with
b

J
we
may
assume,
after
replacing
a
by
a

b
I
I
I
I , the relation
|I|=n I
P
I
reads |I|=n aI f = 0. We can rewrite this as

Xn X
I0
0 ,e f
a
fce = 0
I
0
e=0

|I |=ne

Here and below the primed multi-indices I 0 are required to be of the form I 0 =
(i1 , . . . , id1 , 0). We will show by descending induction on l {0, . . . , n} that if we
have a relation

Xl X
I0
0
a
f
fce = 0
I ,e
e=0
|I 0 |=ne
P
0
then aI 0 ,e J for all I 0 , e. Namely, set J 0 = (f1 , . . . , fc1 ). Observe that |I 0 |=nl aI 0 ,l f I
is mapped into (J 0 )nl+1 by fcl . By induction hypothesis (for the induction on c)
we see that fcl aI 0 ,l J 0 . Because fc is not a zerodivisor on R/J 0 (as f1 , . . . , fc is a
regular sequence) we conclude that aI 0 ,l J 0 . This allows us to rewrite the term
P
P
0
0
( |I 0 |=nl aI 0 ,l f I )fcl in the form ( |I 0 |=nl+1 fc bI 0 ,l1 f I )fcl1 . This gives a new
relation of the form

X
Xl2 X
I0
l1
I0
(aI 0 ,l1 + fc bI 0 ,l1 )f )fc +
aI 0 ,e f
fce = 0
0
0
|I |=nl+1

e=0

|I |=ne

Now by the induction hypothesis (on l this time) we see that all aI 0 ,l1 + fc bI 0 ,l1
J and all aI 0 ,e J for e l 2. This, combined with aI 0 ,l J 0 J seen above,
finishes the proof of the induction step.
P
The second assertion means that given any formal expression F = |I|=n mI X I ,
P
mI M with
mI f I J n+1 M , then all the coefficients mI are in J. This is
proved in exactly the same way as we prove the corresponding result for the first
assertion above.

065L

Lemma 68.3. Let R R0 be a flat ring map. Let M be an R-module. Suppose that f1 , . . . , fr R form an M -quasi-regular sequence. Then the images of
f1 , . . . , fr in R0 form a M R R0 -quasi-regular sequence.

COMMUTATIVE ALGEBRA

153

Proof. Set J = (f1 , . . . , fr ), J 0 = JR0 and M 0 = M R L

R0 . We have to show
0
0
0
0
0
the canonical map : R /J [X1 , . . . Xn ] R0 /J 0 M /J M (J 0 )n M 0 /(J 0 )n+1 M 0
is an isomorphism. Because R R0 is flat the sequences 0 J n M M and
0 J n+1 M J n M J n M/J n+1 M 0 remain exact on tensoring with R0 .
This first implies that J n M R R0 = (J 0 )n M 0 and then that (J 0 )n M 0 /(J 0 )n+1 M 0 =
J n M/J n+1 M R R0 . Thus is the tensor product of (68.0.1), which is an isomorphism by assumption, with idR0 and we conclude.

061Q

Lemma 68.4. Let R be a Noetherian ring. Let M be a finite R-module. Let p be

a prime. Let x1 , . . . , xc be a sequence in R whose image in Rp forms an Mp -quasiregular sequence. Then there exists a g R, g 6 p such that the image of x1 , . . . , xc
in Rg forms an Mg -quasi-regular sequence.
Proof. Consider the kernel K of the map (68.0.1). As M/JM R/J R/J[X1 , . . . , Xc ]
is a finite R/J[X1 , . . . , Xc ]-module and as R/J[X1 , . . . , Xc ] is Noetherian, we see
that K is also a finite R/J[X1 , . . . , Xc ]-module. Pick homogeneous generators
k1 , . . . , kt K. By assumption for each i = 1, . . . , t there exists a gi R, gi 6 p
such that gi ki = 0. Hence g = g1 . . . gt works.


061R

Lemma 68.5. Let R be a ring. Let M be an R-module. Let f1 , . . . , fc R

be an M -quasi-regular sequence. For any i the sequence f i+1 , . . . , f c of R =
R/(f1 , . . . , fi ) is an M = M/(f1 , . . . , fi )M -quasi-regular sequence.
Proof. It suffices to prove this for i = 1. Set J = (f 2 , . . . , f c ) R. Then
n

J M /J

n+1

M = (J n M + f1 M )/(J n+1 M + f1 M )
= J n M/(J n+1 M + J n M f1 M ).

Thus, in order to prove the lemma it suffices to show that J n+1 M + J n M f1 M =

L
n
n+1
J n+1 M +f1 J n1 M because that will show that n0 J M /J
M is the quotient
L
of n0 J n M/J n+1
= M/JM [X1 , . . . , Xc ] by X1 . Actually, we have J n M f1 M =
L n
f1 J n1 M . Namely, if m 6 J n1 M , then f1 m 6 J n M because
J M/J n+1 M is
the polynomial algebra M/J[X1 , . . . , Xc ] by assumption.

061S

Lemma 68.6. Let (R, m) be a local Noetherian ring. Let M be a nonzero finite
R-module. Let f1 , . . . , fc m be an M -quasi-regular sequence. Then f1 , . . . , fc is
an M -regular sequence.
Proof. Set J = (f1 , . . . , fc ). Let us show that f1 is a nonzerodivisor on M . Suppose
x M is not zero. By the Artin-Rees lemma there exists an integer r such that
x J r M but x 6 J r+1 M , see Lemma 50.4. Then f1 x J r+1 M is L
an element whose
class in J r+1 M/J r+2 M is nonzero by the assumed structure of
J n M/J n+1 M .
Whence f1 x 6= 0.
Now we can finish the proof by induction on c using Lemma 68.5.

061T

Remark 68.7 (Koszul regular sequences). In the paper [Kab71] the author introduces two more regularity conditions for sequences x1 , . . . , xr of elements of a
ring R. Namely, we say the sequence is Koszul-regular if Hi (K (R, x )) = 0 for
i 1 where K (R, x ) is the Koszul complex. The sequence is called H1 -regular if
H1 (K (R, x )) = 0. If R is a local ring (possibly nonnoetherian) and the sequence
consists of elements of the maximal ideal, then one has the implications regular
Koszul-regular H1 -regular quasi-regular. By examples the author shows that

COMMUTATIVE ALGEBRA

154

these implications cannot be reversed in general. We introduce these notions in

more detail in More on Algebra, Section 23.
065M

Remark 68.8. Let k be a field. Consider the ring

A = k[x, y, w, z0 , z1 , z2 , . . .]/(y 2 z0 wx, z0 yz1 , z1 yz2 , . . .)
In this ring x is a nonzerodivisor and the image of y in A/xA gives a quasi-regular
sequence. But it is not true that x, y is a quasi-regular sequence in A because
(x, y)/(x, y)2 isnt free of rank two over A/(x, y) due to the fact that wx = 0 in
(x, y)/(x, y)2 but w isnt zero in A/(x, y). Hence the analogue of Lemma 67.7 does
not hold for quasi-regular sequences.

065N

Lemma 68.9. Let R be a ring. Let J = (f1 , . . . , fr ) be an ideal of R. Let M

T
T
be an R-module. Set R = R/ n0 J n , M = M/ n0 J n M , and denote f i the
image of fi in R. Then f1 , . . . , fr is M -quasi-regular if and only if f 1 , . . . , f r is
M -quasi-regular.
n
n+1
M.
Proof. This is true because J n M/J n+1 M
= J M /J

69. Blow up algebras

052P

In this section we make some elementary observations about blowing up.

052Q

Definition 69.1. Let R be a ring. Let I R be an ideal.

(1) The blowup algebra, or the Rees algebra, associated to the pair (R, I) is the
graded R-algebra
M
BlI (R) =
In = R I I2 . . .
n0

where the summand I n is placed in degree n.

(2) Let a I be an element. Denote a(1) the element a seen as an element of
degree 1 in the Rees algebra. Then the affine blowup algebra R[ aI ] is the
algebra (BlI (R))(a(1) ) constructed in Section 56.
In other words, an element of R[ aI ] is represented by an expression of the form x/an
with x I n . Two representatives x/an and y/am define the same element if and
only if ak (am x an y) = 0 for some k 0.
07Z3

Lemma 69.2. Let R be a ring, I R an ideal, and a I. The image of a in the

blowup algebra R0 = R[ aI ] is a nonzerodivisor and IR0 = aR0 .
Proof. Immediate from the description of R[ aI ] above.

0BIP

Lemma 69.3. Let R S be a ring map. Let I R be an ideal and a I. Set

J = IS and let b J be the image of a. Then S[ Jb ] is the quotient of S R R[ aI ] by
the ideal of elements annihilated by some power of b.
Proof. Let S 0 be the quotient of S R R[ aI ] by its b-power torsion elements. The
ring map
S R R[ aI ] S[ Jb ]
is surjective and annihilates a-power torsion as b is a nonzerodivisor in S[ Jb ]. Hence
we obtain a surjective map S 0 S[ Jb ]. To see that the kernel is trivial, we construct
an inverse map. Namely, let z = y/bn be an element of S[ Jb ], i.e., y J n . Write

COMMUTATIVE ALGEBRA

155

P
P
y=
xi si with xi I n and si S. We map z to the class of
si xi /an in S 0 .
This is well defined because an element of the kernel of the map S R I n J n is
annihilated by an , hence maps to zero in S 0 .

080U

Lemma 69.4. Let R be a ring, I R an ideal, and a I. Set R0 = R[ aI ].

If f R is such that V (f ) = V (I), then f maps to a nonzerodivisor in R0 and
Rf0 = Ra0 = Rf .
Proof. We will use the results of Lemma 69.2 without further mention. The assumption V (f ) = V (I) implies V (f R0 ) = V (IR0 ) = V (aR0 ). Hence an = f b and
f m = ac for some b, c R0 . The lemma follows.


0BBI

Lemma 69.5. Let R be a ring, I R an ideal, a I, and f R. Set R0 = R[ aI ]

and R00 = R[ ff aI ]. Then there is a surjective R-algebra map R0 R00 whose kernel
is the set of f -power torsion elements of R0 .
Proof. The map is given by sending x/an for x I n to f n x/(f a)n . It is straightforward to check this map is well defined and surjective. Since af is a nonzero
divisor in R00 (Lemma 69.2) we see that the set of f -power torsion elements are
mapped to zero. Conversely, if x R0 and f n x 6= 0 for all n > 0, then (af )n x 6= 0
for all n as a is a nonzero divisor in R0 . It follows that the image of x in R00 is not
zero by the description of R00 following Definition 69.1.


052S

Lemma 69.6. If R is reduced then every (affine) blowup algebra of R is reduced.

Proof. Let I R be an ideal and a I. Suppose x/an with x I n is a nilpotent
element of R[ aI ]. Then (x/an )m = 0. Hence aN xm = 0 in R for some N 0.
After increasing N if necessary we may assume N = me for some e 0. Then
(ae x)m = 0 and since A is reduced we find ae x = 0. This means that x/an = 0 in

R[ aI ].

052R

Lemma 69.7.
domain.

If R is a domain then every (affine) blowup algebra of R is a

Proof. Let I R be an ideal and a I nonzero. Suppose x/an , y/am with x I n ,

y I m are elements of R[ aI ] whose product is zero. Then aN xy = 0 in R. Since R
is a domain we conclude that either x = 0 or y = 0.

052T

Lemma 69.8. Let R be a ring. Let I R be an ideal. Let a I. If a is

not contained in any minimal prime of R, then Spec(R[ aI ]) Spec(R) has dense
image.
Proof. If ak x = 0 for x R, then x is contained in all the minimal primes of R
and hence nilpotent, see Lemma 16.2. Thus the kernel of R R[ aI ] consists of
nilpotent elements. Hence the result follows from Lemma 29.6.


0BIQ

Lemma 69.9. Let R be a Noetherian ring. Let a, a2 , . . . , ar be a regular sequence

in R. With I = (a, a2 , . . . , ar ) the blowup algebra R0 = R[ aI ] is isomorphic to
R00 = R[y2 , . . . , yr ]/(ayi ai ).
Proof. There is a canonical map A00 P
A0 sending yi to the class ofPai /a. Since
every element x of I can be written as ra+ ri ai we see that x/a = r+P ri ai /a is in
the image of the map. Hence our map is surjective. Suppose that z =
rE y E A00
0
maps to zero in A . Here we use the multi-index notation E = (e2 , . . . , er ) and

COMMUTATIVE ALGEBRA

156

P
ei of the multiy E = y2e2 . . . yrer . Let d be the maximum of the degrees |E| =
indices which occur with a nonzero coefficient rE in z. Then we see that
X
ad z =
rE ad|E| ae22 . . . aerr
is zero in R; here we use that a is a nonzerodivisor on R. Since a regular sequence
is quasi-regular by Lemma 68.2 we conclude that rE I for all E. This means
that z is divisible by a in A00 . Say z = az 0 . Then z 0 is in the kernel of A00 A0
0
and
T nwe00 see that z00 is divisible by a and so on. In other words, z is an element of
a R . Since R is Noetherian by Krulls intersection theorem z maps to zero in
Rp00 for every prime ideal p containing aR00 , see Remark 50.6. On the other hand,
if p R00 does not contain a, then Ra00
= Ra
= Ra0 and we find that z maps to zero
00
in Rp as well. We conclude that z is zero by Lemma 23.1.

052M

Lemma 69.10. Let (R, m) be a local domain with fraction field K. Let R A K
be a valuation ring which dominates R. Then
A = colim R[ aI ]
is a directed colimit of affine blowups R R[ aI ] with the following properties
(1) a I m,
(2) I is finitely generated, and
(3) the fibre ring of R R[ aI ] at m is not zero.
Proof. Consider a finite subset E A. Say E = {e1 , . . . , en }. Choose a nonzero
a R such that we can write ei = fi /a for all i = 1, . . . , n. Set I = (f1 , . . . , fn , a).
We claim that R[ aI ] A. This is clear as an element of R[ aI ] can be represented
as a polynomial in the elements ei . The lemma follows immediately from this
observation.

70. Ext groups

00LO

In this section we do a tiny bit of homological algebra, in order to establish some

fundamental properties of depth over Noetherian local rings.

00LP

Lemma 70.1. Let R be a ring. Let M be an R-module.

(1) There exists an exact complex
. . . F2 F1 F0 M 0.
with Fi free R-modules.
(2) If R is Noetherian and M finite over R, then we can choose the complex
such that Fi is finite free. In other words, we can find an exact complex
. . . Rn2 Rn1 Rn0 M 0.
Proof. Let us explain only the Noetherian case. As a first step choose a surjection
Rn0 M . Then having constructed an exact complex of length e we simply
choose a surjection Rne+1 Ker(Rne Rne1 ) which is possible because R is
Noetherian.


00LQ

Definition 70.2. Let R be a ring. Let M be an R-module.

(1) A (left) resolution F M of M is an exact complex
. . . F2 F1 F0 M 0
of R-modules.

COMMUTATIVE ALGEBRA

157

(2) A resolution of M by free R-modules is a resolution F M where each

Fi is a free R-module.
(3) A resolution of M by finite free R-modules is a resolution F M where
each Fi is a finite free R-module.
We often use the notation F to denote a complex of R-modules
. . . Fi Fi1 . . .
In this case we often use di or dF,i to denote the map Fi Fi1 . In this section we
are always going to assume that F0 is the last nonzero term in the complex. The ith
homology group of the complex F is the group Hi = Ker(dF,i )/Im(dF,i+1 ). A map
of complexes : F G is given by maps i : Fi Gi such that i1 dF,i =
dG,i1 i . Such a map induces a map on homology Hi () : Hi (F ) Hi (G ). If
, : F G are maps of complexes, then a homotopy between and is given
by a collection of maps hi : Fi Gi+1 such that i i = dG,i+1 hi + hi1 dF,i .
We will use a very similar notation regarding complexes of the form F which look
like
di
. . . F i F i+1 . . .
There are maps of complexes, homotopies, etc. In this case we set H i (F ) =
Ker(di )/Im(di1 ) and we call it the ith cohomology group.
00LR

Lemma 70.3. Any two homotopic maps of complexes induce the same maps on
(co)homology groups.
Proof. Omitted.

00LS

Lemma 70.4. Let R be a ring. Let M N be a map of R-modules. Let F M

be a resolution by free R-modules and let N N be an arbitrary resolution. Then
(1) there exists a map of complexes F N inducing the given map
M = Coker(F1 F0 ) Coker(N1 N0 ) = N
(2) two maps , : F N inducing the same map M N are homotopic.
Proof. Proof of (1). Because F0 is free we can find a map F0 N0 lifting the
map F0 M N . We obtain an induced map F1 F0 N0 which ends up in
the image of N1 N0 . Since F1 is free we may lift this to a map F1 N1 . This in
turn induces a map F2 F1 N1 which maps to zero into N0 . Since N is exact
we see that the image of this map is contained in the image of N2 N1 . Hence
we may lift to get a map F2 N2 . Repeat.
Proof of (2). To show that , are homotopic it suffices to show the difference
= is homotopic to zero. Note that the image of 0 : F0 N0 is contained
in the image of N1 N0 . Hence we may lift 0 to a map h0 : F0 N1 . Consider
the map 10 = 1 h0 dF,1 . By our choice of h0 we see that the image of 10 is
contained in the kernel of N1 N0 . Since N is exact we may lift 10 to a map
h1 : F1 N2 . At this point we have 1 = h0 dF,1 + dN,2 h1 . Repeat.

At this point we are ready to define the groups ExtiR (M, N ). Namely, choose a resolution F of M by free R-modules, see Lemma 70.1. Consider the (cohomological)
complex
HomR (F , N ) : HomR (F0 , N ) HomR (F1 , N ) HomR (F2 , N ) . . .

COMMUTATIVE ALGEBRA

158

We define ExtiR (M, N ) to be the ith cohomology group of this complex.5 The
following lemma explains in what sense this is well defined.
00LT

Lemma 70.5. Let R be a ring. Let M1 , M2 , N be R-modules. Suppose that F

is a free resolution of the module M1 , and G is a free resolution of the module
M2 . Let : M1 M2 be a module map. Let : F G be a map of complexes
inducing on M1 = Coker(dF,1 ) M2 = Coker(dG,1 ), see Lemma 70.4. Then the
induced maps
H i () : H i (HomR (F , N )) H i (HomR (G , N ))
are independent of the choice of . If is an isomorphism, so are all the maps
H i (). If M1 = M2 , F = G , and is the identity, so are all the maps Hi ().
Proof. Another map : F G inducing is homotopic to by Lemma 70.4.
Hence the maps HomR (F , N ) HomR (G , N ) are homotopic. Hence the independence result follows from Lemma 70.3.
Suppose that is an isomorphism. Let : M2 M1 be an inverse. Choose
: G F be a map inducing : M2 = Coker(dG,1 ) M1 = Coker(dF,1 ), see
Lemma 70.4. OK, and now consider the map H i () H i () = H i ( ). By the
above the map H i ( ) is the same as the map H i (idG ) = id. Similarly for the
composition H i () H i (). Hence H i () and H i () are inverses of each other. 

00LU

Lemma 70.6. Let R be a ring. Let M be an R-module. Let 0 N 0 N

N 00 0 be a short exact sequence. Then we get a long exact sequence
0 HomR (M, N 0 ) HomR (M, N ) HomR (M, N 00 )
Ext1R (M, N 0 ) Ext1R (M, N ) Ext1R (M, N 00 ) . . .
Proof. Pick a free resolution F M . Since each of the Fi are free we see that
we get a short exact sequence of complexes
0 HomR (F , N 0 ) HomR (F , N ) HomR (F , N 00 ) 0
Thus we get the long exact sequence from the snake lemma applied to this.

065P

Lemma 70.7. Let R be a ring. Let N be an R-module. Let 0 M M

M 00 0 be a short exact sequence. Then we get a long exact sequence
0 HomR (M 00 , N ) HomR (M, N ) HomR (M 0 , N )
Ext1R (M 00 , N ) Ext1R (M, N ) Ext1R (M 0 , N ) . . .
Proof. Pick sets of generators {m0i0 }i0 I 0 and {m00i00 }i00 I 00 of M 0 and M 00 .L
For each
00
i00 IL
choose a lift m
00i00 M of the element m00i00 M 00 . Set F 0 =
i0 I 0 R,
F 00 = i00 I 00 R and F = F 0 F 00 . Mapping the generators of these free modules
to the corresponding chosen generators gives surjective R-module maps F 0 M 0 ,
F 00 M 00 , and F M . We obtain a map of short exact sequences
0
0

M0

F0

M 00

F 00

0
0

By the snake lemma we see that the sequence of kernels 0 K 0 K K 00 0 is

short exact sequence of R-modules. Hence we can continue this process indefinitely.
5At this point it would perhaps be more appropriate to say an in stead of the Ext-group.

COMMUTATIVE ALGEBRA

159

In other words we obtain a short exact sequence of resolutions fitting into the
diagram
0 M 0 M M 00 0

0 F0 F F00 0
Because each of the sequences 0 Fn0 Fn Fn00 0 is split exact (by construction) we obtain a short exact sequence of complexes
0 HomR (F00 , N ) HomR (F , N ) HomR (F0 , N ) 0
by applying the HomR (, N ) functor. Thus we get the long exact sequence from
the snake lemma applied to this.

00LV

Lemma 70.8. Let R be a ring. Let M , N be R-modules. Any x R such that

either xN = 0, or xM = 0 annihilates each of the modules ExtiR (M, N ).
Proof. Pick a free resolution F of M . Since ExtiR (M, N ) is defined as the cohomology of the complex HomR (F , N ) the lemma is clear when xN = 0. If xM = 0, then
we see that multiplication by x on F lifts the zero map on M . Hence by Lemma
70.5 we see that it induces the same map on Ext groups as the zero map.


08YR

Lemma 70.9. Let R be a Noetherian ring. Let M , N be finite R-modules. Then

ExtiR (M, N ) is a finite R-module for all i.
Proof. This holds because ExtiR (M, N ) is computed as the cohomology groups of
a complex HomR (F , N ) with each Fn a finite free R-module, see Lemma 70.1. 
71. Depth

00LE
00LI

Here is our definition.

Definition 71.1. Let R be a ring, and I R an ideal. Let M be a finite R-module.
The I-depth of M , denoted depthI (M ), is defined as follows:
(1) if IM 6= M , then depthI (M ) is the supremum in {0, 1, 2, . . . , } of the
lengths of M -regular sequences in I,
(2) if IM = M we set depthI (M ) = .
If (R, m) is local we call depthm (M ) simply the depth of M .
Explanation. By Definition 67.1 the empty sequence is not a regular sequence on
the zero module, but for practical purposes it turns out to be convenient to set the
depth of the 0 module equal to +. Note that if I = R, then depthI (M ) = for
all finite R-modules M . If I is contained in the radical ideal of R (e.g., if R is local
and I mR ), then M 6= 0 IM 6= M by Nakayamas lemma. A module M has
I-depth 0 if and only if M is nonzero and I does not contain an M -regular element.
Example 67.2 shows depth does not behave well even if the ring is Noetherian,
and Example 67.3 shows that it does not behave well if the ring is local but nonNoetherian. We will see depth behaves well if the ring is local Noetherian.

0AUI

Lemma 71.2. Let R be a ring, I R an ideal, and M a finite R-module. Then

depthI (M ) is equal to the supremum of the lengths of sequences f1 , . . . , fr I such
that fi is a nonzerodivisor on M/(f1 , . . . , fi1 )M .

COMMUTATIVE ALGEBRA

160

Proof. Suppose that IM = M . Then Lemma 19.1 shows there exists an f I such
that f : M M is idM . Hence f, 0, 0, 0, . . . is an infinite sequence of successive
nonzerodivisors and we see agreement holds in this case. If IM 6= M , then we see
that a sequence as in the lemma is an M -regular sequence and we conclude that
agreement holds as well.

00LK

Lemma 71.3. Let R be a Noetherian local ring. Let M be a nonzero finite Rmodule. Then dim(Supp(M )) depth(M ).
Proof. If f is an element of the maximal ideal of R and a nonzerodivisor on M , then
dim(Supp(M/f M )) dim(Supp(M ))1, by Lemma 62.10. The result follows. 

0AUJ

Lemma 71.4. Let R be a Noetherian ring, I R an ideal, and M a finite nonzero

R-module such that IM 6= M . Then depthI (M ) < .
Proof. Let I p be a prime ideal. As localization is flat we see that M -regular sequences in I maps to Mp -regular sequences in Ip . Hence depthI (M ) depthIp (Mp ).
The latter is depth(Mp ) which is < by Lemma 71.3.


00LW

Lemma 71.5. Let R be a Noetherian local ring with maximal ideal m. Let M be
a nonzero finite R-module. Then depth(M ) is equal to the smallest integer i such
that ExtiR (R/m, M ) is nonzero.
Proof. Let (M ) denote the depth of M and let i(M ) denote the smallest integer
i such that ExtiR (R/m, M ) is nonzero. We will see in a moment that i(M ) < .
By Lemma 62.18 we have (M ) = 0 if and only if i(M ) = 0, because m Ass(M )
exactly means that i(M ) = 0. Hence if (M ) or i(M ) is > 0, then we may choose
x m such that (a) x is a nonzerodivisor on M , and (b) depth(M/xM ) = (M )
1. Consider the long exact sequence of Ext-groups associated to the short exact
sequence 0 M M M/xM 0 by Lemma 70.6:
0 HomR (, M ) HomR (, M ) HomR (, M/xM )
Ext1R (, M ) Ext1R (, M ) Ext1R (, M/xM ) . . .
Since x m all the maps ExtiR (, M ) ExtiR (, M ) are zero, see Lemma 70.8.
Thus it is clear that i(M/xM ) = i(M )1. Induction on (M ) finishes the proof. 

00LX

Lemma 71.6. Let R be a local Noetherian ring. Let 0 N 0 N N 00 0 be

a short exact sequence of finite R-modules.
(1) depth(N 00 ) min{depth(N ), depth(N 0 ) 1}
(2) depth(N 0 ) min{depth(N ), depth(N 00 ) + 1}
Proof. Use the characterization of depth using the Ext groups Exti (, N ), see
Lemma 71.5, and use the long exact cohomology sequence
0 HomR (, N 0 ) HomR (, N ) HomR (, N 00 )
Ext1R (, N 0 ) Ext1R (, N ) Ext1R (, N 00 ) . . .
from Lemma 70.6.

090R

Lemma 71.7. Let R be a local Noetherian ring and M a nonzero finite R-module.
(1) If x m is a nonzerodivisor on M , then depth(M/xM ) = depth(M ) 1.
(2) Any M -regular sequence x1 , . . . , xr can be extended to an M -regular sequence of length depth(M ).

COMMUTATIVE ALGEBRA

161

Proof. Part (2) is a formal consequence of part (1). Let x R be as in (1). By

the short exact sequence 0 M M M/xM 0 and Lemma 71.6 we see that
the depth drops by at most 1. On the other hand, if x1 , . . . , x m is a regular
sequence for M/xM , then x, x1 , . . . , xr is a regular sequence for M . Hence we see
that the depth drops by at least 1.

0BK4

Lemma 71.8. Let (R, m) be a local Noetherian ring and M a finite R-module.
For p Ass(M ) we have dim(R/p) depth(M ).
Proof. If m Ass(M ) then depth(M ) = 0 and there is nothing to prove. This
proves the lemma if depth(M ) = 1. We will prove the lemma in general by induction
on depth(M ) which we assume to be > 1. Pick x m with x 6 p for p Ass(M ).
The existence of such an x follows from Lemmas 14.2 and 62.5. Then x is a
nonzerodivisor on M by Lemma 62.9. Hence depth(M/xn M ) = depth(M ) 1 by
Lemma 71.7. For p Ass(M ) pick an embedding R/p M . The Artin-Rees lemma
(Lemma 50.2) shows that (R/p) xn M is contained in x(R/p) for n large enough.
Thus the image N M/xn M of R/p is a submodule whose support is equal to
V (p + (x)). By Lemma 59.12 the dimension of this is dim(R/p) 1. It follows that
M/xn M has an associated prime of dimension dim(R/p) 1, see Lemmas 62.8 and
62.3. By induction we find dim(R/p) 1 depth(M ) 1 and we win.


0AUK

Lemma 71.9. Let (R, m) be a Noetherian local ring. Let R S be a finite ring
map. Let m1 , . . . , mn be the maximal ideals of S. Let N be a finite S-module. Then
mini=1,...,n depth(Nmi ) = depth(N )
Proof. By Lemmas 35.18, 35.20, and Lemma 35.19 the maximal ideals of S are
exactly the primes of S lying over m and there are finitely many of them. Hence
the statement of the lemma makes sense. We will prove the lemma by induction on
k = mini=1,...,n depth(Nmi ). If k = 0, then depth(Nmi ) = 0 for some i. By Lemma
71.5 this means mi Smi is an associated prime of Nmi and hence mi is an associated
prime of N (Lemma 62.16). By Lemma 62.13 we see that m is an associated prime
of N as an R-module. Whence depth(N ) = 0. This proves the base case. If k > 0,
then we see that mi 6 AssS (N ). Hence m 6 AssR (N ), again by Lemma 62.13.
Thus we can find f m which is not a zerodivisor on N , see Lemma 62.18. By
Lemma 71.7 all the depths drop exactly by 1 when passing from N to N/f N and
the induction hypothesis does the rest.

72. Functorialities for Ext

087M

In this section we briefly discuss the functoriality of Ext with respect to change of
ring, etc. Here is a list of items to work out.
(1) Given R R0 , an R-module M and an R0 -module N 0 the R-module
ExtiR (M, N 0 ) has a natural R0 -module structure. Moreover, there is a
canonical R0 -linear map ExtiR0 (M R R0 , N 0 ) ExtiR (M, N 0 ).
(2) Given R R0 and R-modules M , N there is a natural R-module map
ExtiR (M, N ) ExtiR (M, N R R0 ).

087N

Lemma 72.1. Given a flat ring map R R0 , an R-module M , and an R0 -module

N 0 the natural map
ExtiR0 (M R R0 , N 0 ) ExtiR (M, N 0 )
is an isomorphism for i 0.

COMMUTATIVE ALGEBRA

162

Proof. Choose a free resolution F of M . Since R R0 is flat we see that F R R0

is a free resolution of M R R0 over R0 . The statement is that the map
HomR0 (F R R0 , N 0 ) HomR (F , N 0 )
induces an isomorphism on homology groups, which is true because it is an isomorphism of complexes by Lemma 13.3.

73. An application of Ext groups
02HN

Here it is.

02HO

Lemma 73.1. Let R be a Noetherian local ring with maximal ideal m. Let N M
be a homomorphism of finite R-modules. Suppose that there exists arbitrarily large
n such that N/mn N M/mn M is a split injection. Then N M is a split
injection.
Proof. Assume : N M satisfies the assumptions of the lemma. Note that this
implies that Ker() mn N for arbitrarily large n. Hence by Lemma 50.4 we see
that is injection. Let Q = M/N so that we have a short exact sequence
0 N M Q 0.
Let
d

2
1
F2
F1
F0 Q 0

be a finite free resolution of Q. We can choose a map : F0 M lifting the map

F0 Q. This induces a map : F1 N such that d2 = 0. The extension
above is split if and only if there exists a map : F0 N such that = d1 . In
other words, the class of in Ext1R (Q, N ) is the obstruction to splitting the short
exact sequence above.
Suppose n is a large integer such that N/mn N M/mn M is a split injection. This
implies
0 N/mn N M/mn M Q/mn Q 0.
is still short exact. Also, the sequence
d

1
F1 /mn F1
F0 /mn F0 Q/mn Q 0

is still exact. Arguing as above we see that the map : F1 /mn F1 N/mn N
induced by is equal to n d1 for some map n : F0 /mn F0 N/mn . Since F0 is
free we can lift n to a map n : F0 N and then we see that n d1 is a map
from F1 into mn N . In other words we conclude that


Im HomR (F0 , N ) HomR (F1 , N ) + mn HomR (F1 , N ).
for this n.
Since we have this property for arbitrarily large n by assumption we conclude
(by Lemma 50.2) that is actually in the image of the map HomR (F0 , N )
HomR (F1 , N ) as desired.


COMMUTATIVE ALGEBRA

163

74. Tor groups and flatness

00LY

In this section we use some of the homological algebra developed in the previous
section to explain what Tor groups are. Namely, suppose that R is a ring and that
M , N are two R-modules. Choose a resolution F of M by free R-modules. See
Lemma 70.1. Consider the homological complex
F R N : . . . F2 R N F1 R N F0 R N
We define TorR
i (M, N ) to be the ith homology group of this complex. The following
lemma explains in what sense this is well defined.

00LZ

Lemma 74.1. Let R be a ring. Let M1 , M2 , N be R-modules. Suppose that F is

a free resolution of the module M1 and that G is a free resolution of the module
M2 . Let : M1 M2 be a module map. Let : F G be a map of complexes
inducing on M1 = Coker(dF,1 ) M2 = Coker(dG,1 ), see Lemma 70.4. Then the
induced maps
Hi () : Hi (F R N ) Hi (G R N )
are independent of the choice of . If is an isomorphism, so are all the maps
Hi (). If M1 = M2 , F = G , and is the identity, so are all the maps Hi ().
Proof. The proof of this lemma is identical to the proof of Lemma 70.5.

Not only does this lemma imply that the Tor modules are well defined, but it also
provides for the functoriality of the constructions (M, N ) 7 TorR
i (M, N ) in the
first variable. Of course the functoriality in the second variable is evident. We
leave it to the reader to see that each of the TorR
i is in fact a functor
ModR ModR ModR .
Here ModR denotes the category of R-modules, and for the definition of the product
category see Categories, Definition 2.20. Namely, given morphisms of R-modules
M1 M2 and N1 N2 we get a commutative diagram

00M0

TorR
i (M1 , N1 )

/ TorR
i (M1 , N2 )


TorR
(M
2 , N1 )
i


/ TorR
(M
2 , N2 )
i

Lemma 74.2. Let R be a ring and let M be an R-module. Suppose that 0

N 0 N N 00 0 is a short exact sequence of R-modules. There exists a long
exact sequence
M R N
0
TorR
1 (M, N )

M R N M R N 00 0
R
00
TorR
1 (M, N ) Tor1 (M, N )

Proof. The proof of this is the same as the proof of Lemma 70.6.

COMMUTATIVE ALGEBRA

164

Consider a homological double complex of R-modules

...

/ A2,0
O

...

/ A0,0
O

/ A2,1
O

...

/ A1,0
O
/ A1,1
O

/ A0,1
O

/ A2,2
O

/ A1,2
O

/ A0,2
O

...

...

...

This means that di,j : Ai,j Ai1,j and i,j : Ai,j Ai,j1 have the following
properties
(1) Any composition of two di,j is zero. In other words the rows of the double
complex are complexes.
(2) Any composition of two i,j is zero. In other words the columns of the
double complex are complexes.
(3) For any pair (i, j) we have i1,j di,j = di,j1 i,j . In other words, all
the squares commute.
The correct thing to do is to associate a spectral sequence to any such double
complex. However, for the moment we can get away with doing something slightly
easier.
Namely, for the purposes of this section only, given a double complex (A, , d, )
set R(A)j = Coker(A1,j A0,j ) and U (A)i = Coker(Ai,1 Ai,0 ). (The letters R
and U are meant to suggest Right and Up.) We endow R(A) with the structure
of a complex using the maps . Similarly we endow U (A) with the structure
of a complex using the maps d. In other words we obtain the following huge
commutative diagram
...

/ U (A)2
O

/ U (A)1
O

/ U (A)0
O

...

/ A2,0
O

/ A1,0
O

/ A0,0
O

...

/ A2,1
O

...

/ A2,2
O

/ A1,1
O

/ A1,2
O

/ A0,1
O

/ A0,2
O

/ R(A)0
O

/ R(A)1
O

/ R(A)2
O

...

...

...

...

(This is no longer a double complex of course.) It is clear what a morphism :

(A, , d, ) (B, , d, ) of double complexes is, and it is clear that this induces
morphisms of complexes R() : R(A) R(B) and U () : U (A) U (B) .

COMMUTATIVE ALGEBRA

00M1

165

Lemma 74.3. Let (A, , d, ) be a double complex such that

(1) Each row A,j is a resolution of R(A)j .
(2) Each column Ai, is a resolution of U (A)i .
Then there are canonical isomorphisms
Hi (R(A) )
= Hi (U (A) ).
The isomorphisms are functorial with respect to morphisms of double complexes
with the properties above.
Proof. We will show that Hi (R(A) )) and Hi (U (A) ) are canonically isomorphic
to a third group. Namely
Hi (A) :=

{(ai,0 , ai1,1 , . . . , a0,i ) | d(ai,0 ) = (ai1,1 ), . . . , d(a1,i1 ) = (a0,i )}

{d(ai+1,0 ) (ai,1 ), d(ai,1 ) (ai1,2 ), . . . , d(a1,i ) (a0,i+1 )}

Here we use the notational convention that ai,j denotes an element of Ai,j . In other
words, an element of Hi is represented by a zig-zag, represented as follows for i = 2

/ d(a2,0 ) = (a1,1 )
a2,0
O
_
a1,1 

/ d(a1,1 ) = (a0,2 )
O
_
a0,2

Naturally, we divide out by trivial zig-zags, namely the submodule generated by

elements of the form (0, . . . , 0, (at+1,ti ), d(at+1,ti ), 0, . . . , 0). Note that there
are canonical homomorphisms
Hi (A) Hi (R(A) ),

(ai,0 , ai1,1 , . . . , a0,i ) 7 class of image of a0,i

Hi (A) Hi (U (A) ),

(ai,0 , ai1,1 , . . . , a0,i ) 7 class of image of ai,0

and

First we show that these maps are surjective. Suppose that r Hi (R(A) ). Let
r R(A)i be a cocycle representing the class of r. Let a0,i A0,i be an element
which maps to r. Because (r) = 0, we see that (a0,i ) is in the image of d. Hence
there exists an element a1,i1 A1,i1 such that d(a1,i1 ) = (a0,i ). This in turn
implies that (a1,i1 ) is in the kernel of d (because d((a1,i1 )) = (d(a1,i1 )) =
((a0,i )) = 0. By exactness of the rows we find an element a2,i2 such that
d(a2,i2 ) = (a1,i1 ). And so on until a full zig-zag is found. Of course surjectivity
of Hi Hi (U (A)) is shown similarly.
To prove injectivity we argue in exactly the same way. Namely, suppose we are
given a zig-zag (ai,0 , ai1,1 , . . . , a0,i ) which maps to zero in Hi (R(A) ). This means
that a0,i maps to an element of Coker(Ai,1 Ai,0 ) which is in the image of :
Coker(Ai+1,1 Ai+1,0 ) Coker(Ai,1 Ai,0 ). In other words, a0,i is in the image
of d : A0,i+1 A1,i A0,i . From the definition of trivial zig-zags we see that we
may modify our zig-zag by a trivial one and assume that a0,i = 0. This immediately
implies that d(a1,i1 ) = 0. As the rows are exact this implies that a1,i1 is in the
image of d : A2,i1 A1,i1 . Thus we may modify our zig-zag once again by a

COMMUTATIVE ALGEBRA

166

trivial zig-zag and assume that our zig-zag looks like (ai,0 , ai1,1 , . . . , a2,i2 , 0, 0).
Continuing like this we obtain the desired injectivity.
If : (A, , d, ) (B, , d, ) is a morphism of double complexes both of which
satisfy the conditions of the lemma, then we clearly obtain a commutative diagram
Hi (U (A) ) o

Hi (A)

/ Hi (R(A) )


Hi (U (B) ) o


Hi (B)


/ Hi (R(B) )

This proves the functoriality.

00M2

Remark 74.4. The isomorphism constructed above is the correct one only up
to signs. A good part of homological algebra is concerned with choosing signs for
various maps and showing commutativity of diagrams with intervention of suitable
signs. For the moment we will simply use the isomorphism as given in the proof
above, and worry about signs later.

00M3

Lemma 74.5. Let R be a ring. For any i 0 the functors ModR ModR ModR ,
R
(M, N ) 7 TorR
i (M, N ) and (M, N ) 7 Tori (N, M ) are canonically isomorphic.
Proof. Let F be a free resolution of the module M and let G be a free resolution
of the module N . Consider the double complex (Ai,j , d, ) defined as follows:
(1) set Ai,j = Fi R Gj ,
(2) set di,j : Fi R Gj Fi1 Gj equal to dF,i id, and
(3) set i,j : Fi R Gj Fi Gj1 equal to id dG,j .
This double complex is usually simply denoted F R G .
Since each Gj is free, and hence flat we see that each row of the double complex is
exact except in homological degree 0. Since each Fi is free and hence flat we see
that each column of the double complex is exact except in homological degree 0.
Hence the double complex satisfies the conditions of Lemma 74.3.
To see what the lemma says we compute R(A) and U (A) . Namely,
R(A)i

Coker(A1,i A0,i )

Coker(F1 R Gi F0 R Gi )

Coker(F1 F0 ) R Gi

M R Gi

In fact these isomorphisms are compatible with the differentials and we see that
R(A) = M R G as homological complexes. In exactly the same way we see that
U (A) = F R N . We get
TorR
i (M, N )

= Hi (F R N )
= Hi (U (A) )
= Hi (R(A) )
= Hi (M R G )
= Hi (G R M )
=

TorR
i (N, M )

COMMUTATIVE ALGEBRA

167

Here the third equality is Lemma 74.3, and the fifth equality uses the isomorphism
V W = W V of the tensor product.
Functoriality. Suppose that we have R-modules M , N , = 1, 2. Let : M1
M2 and : N1 N2 be morphisms of R-modules. Suppose that we have free
resolutions F, for M and free resolutions G, for N . By Lemma 70.4 we may
choose maps of complexes : F1, F2, and : G1, G2, compatible with
and . We claim that the pair (, ) induces a morphism of double complexes
: F1, R G1, F2, R G2,
This is really a very straightforward check using the rule that F1,i R G1,j F2,i R
G2,j is given by i j where i , resp. j is the degree i, resp. j component of ,
resp. . The reader also readily verifies that the induced maps R(F1, R G1, )
R(F2, R G2, ) agrees with the map M1 R G1, M2 R G2, induced by .
Similarly for the map induced on the U () complexes. Thus the statement on
functoriality follows from the statement on functoriality in Lemma 74.3.

00M4

Remark 74.6. An interesting case occurs when M = N in the above. In this case
R
we get a canonical map TorR
i (M, M ) Tori (M, M ). Note that this map is not the
identity, because even when i = 0 this map is not the identity! For example, if V is
a vector space of dimension n over a field, then the switch map V k V V k V
has (n2 + n)/2 eigenvalues +1 and (n2 n)/2 eigenvalues 1. In characteristic 2
it is not even diagonalizable. Note that even changing the sign of the map will not
get rid of this.

0AZ4

Lemma 74.7. Let R be a Noetherian ring. Let M , N be finite R-modules. Then

TorR
p (M, N ) is a finite R-module for all p.
Proof. This holds because TorR
p (M, N ) is computed as the cohomology groups of
a complex F R N with each Fn a finite free R-module, see Lemma 70.1.


00M5

Lemma 74.8. Let R be a ring. Let M be an R-module. The following are

equivalent:
(1) The module M is flat over R.
(2) For all i > 0 the functor TorR
i (M, ) is zero.
(3) The functor TorR
1 (M, ) is zero.
(4) For all ideals I R we have TorR
1 (M, R/I) = 0.
(5) For all finitely generated ideals I R we have TorR
1 (M, R/I) = 0.
Proof. Suppose M is flat. Let N be an R-module. Let F be a free resolution of
N . Then F R M is a resolution of N R M , by flatness of M . Hence all higher
Tor groups vanish.
It now suffices to show that the last condition implies that M is flat. Let I R
be an ideal. Consider the short exact sequence 0 I R R/I 0. Apply
Lemma 74.2. We get an exact sequence
TorR
1 (M, R/I) M R I M R R M R R/I 0
Since obviously M R R = M we conclude that the last hypothesis implies that
M R I M is injective for every finitely generated ideal I. Thus M is flat by
Lemma 38.5.


COMMUTATIVE ALGEBRA

00M6

168

Remark 74.9. The proof of Lemma 74.8 actually shows that

TorR
1 (M, R/I) = Ker(I R M M ).
75. Functorialities for Tor

00M7

In this section we briefly discuss the functoriality of Tor with respect to change of
ring, etc. Here is a list of items to work out.
(1) Given a ring map R R0 , an R-module M and an R0 -module N 0 the
0
0
R-modules TorR
i (M, N ) have a natural R -module structure.
0
(2) Given a ring map R R and R-modules M , N there is a natural R-module
R0
0
0
map TorR
i (M, N ) Tori (M R R , N R R ).
0
(3) Given a ring map R R an R-module M and an R0 -module N 0 there
R0
0
0
0
exists a natural R0 -module map TorR
i (M, N ) Tori (M R R , N ).

00M8

Lemma 75.1. Given a flat ring map R R0 and R-modules M , N the natural
R0
0
0
0
R-module map TorR
i (M, N ) R R Tori (M R R , N R R ) is an isomorphism
for all i.
Proof. Omitted. This is true because a free resolution F of M over R stays exact
when tensoring with R0 over R and hence (F R N )R R0 computes the Tor groups
over R0 .

The following lemma does not seem to fit anywhere else.

0BNF

Lemma 75.2. Let R be a ring. Let M = colim Mi be a filtered colimit of RR

modules. Let N be an R-module. Then TorR
n (M, N ) = colim Torn (Mi , N ) for all
n.
Proof. Choose a free resolution F of N . Then F R M = colim F R Mi as
complexes by Lemma 11.9. Thus the result by Lemma 8.9.

76. Projective modules

05CD

Some lemmas on projective modules.

05CE

Definition 76.1. Let R be a ring. An R-module P is projective if and only if the

functor HomR (P, ) : ModR ModR is an exact functor.
The functor HomR (M, ) is left exact for any R-module M , see Lemma 10.1.
Hence the condition for P to be projective really signifies that given a surjection of
R-modules N N 0 the map HomR (P, N ) HomR (P, N 0 ) is surjective.

05CF

Lemma 76.2. Let R be a ring. Let P be an R-module. The following are equivalent
(1) P is projective,
(2) P is a direct summand of a free R-module, and
(3) Ext1R (P, M ) = 0 for every R-module M .
Proof. Assume P is projective. Choose a surjection : F P where F is a free
R-module. As P is projective there exists a i HomR (P, F ) such that i = idP .
In other words F
= Ker() i(P ) and we see that P is a direct summand of F .
Conversely,
assume that P Q = F is a freeQ
R-module. Note that the free Q
module
L
F = iI R is projective as HomR (F, M ) = iI M and the functor M 7 iI M

COMMUTATIVE ALGEBRA

169

is exact. Then HomR (F, ) = HomR (P, ) HomR (Q, ) as functors, hence both
P and Q are projective.
Assume P Q = F is a free R-module. Then we have a free resolution F of the
form
a
b
...F
F
F P 0
where the maps a, b alternate and are equal to the projector onto P and Q. Hence
the complex HomR (F , M ) is split exact in degrees 1, whence we see the vanishing
in (3).
Assume Ext1R (P, M ) = 0 for every R-module M . Pick a free resolution F P .
Set M = Im(F1 F0 ) = Ker(F0 P ). Consider the element Ext1R (P, M )
given by the class of the quotient map : F1 M . Since is zero there exists a
map s : F0 M such that = s (F1 F0 ). Clearly, this means that
F0 = Ker(s) Ker(F0 P ) = P Ker(F0 P )
and we win.
065Q

Lemma 76.3. A direct sum of projective modules is projective.

Proof. This is true by the characterization of projectives as direct summands of
free modules in Lemma 76.2.


07LV

Lemma 76.4. Let R be a ring. Let I R be a nilpotent ideal. Let P be a projective

R/I-module. Then there exists a projective R-module P such that P/IP
= P.
L
Proof. We can choose a set A and a direct sum decomposition A R/I = P K
L
for some R/I-module K. Write F = A R for the free R-module on A. Choose a
L
lift p : F F of the projector p associated to the direct summand P of A R/I.
Note that p2 p EndR (F ) is a nilpotent endomorphism of F (as I is nilpotent and
the matrix entries of p2 p are in I; more precisely, if I n = 0, then (p2 p)n = 0).
Hence by Lemma 31.6 we can modify our choice of p and assume that p is a
projector. Set P = Im(p).


05CG

Lemma 76.5. Let R be a ring. Let I R be an ideal. Let M be an R-module.

Assume
(1) I is nilpotent,
(2) M/IM is a projective R/I-module,
(3) M is a flat R-module.
Then M is a projective R-module.
Proof. By Lemma 76.4 we can find a projective R-module P and an isomorphism
P/IP M/IM . We are going to show that M is isomorphic to P which will finish
the proof. Because P is projective we can lift the map P P/IP M/IM to
an R-module map P M which is an isomorphism modulo I. By Nakayamas
Lemma 19.1 the map P M is surjective. It remains to show that P M is
injective. Since I n = 0 for some n, we can use the filtrations
0 = I n M I n1 M . . . IM M
0 = I n P I n1 P . . . IP P

COMMUTATIVE ALGEBRA

170

to see that it suffices to show that the induced maps I a P/I a+1 P I a M/I a+1 M
are injective. Since both P and M are flat R-modules we can identify this with the
map
I a /I a+1 R/I P/IP I a /I a+1 R/I M/IM
induced by P M . Since we chose P M such that the induced map P/IP
M/IM is an isomorphism, we win.

77. Finite projective modules
00NV
00NW

Definition 77.1. Let R be a ring and M an R-module.

(1) We say that M is locally free if we can cover Spec(R) by standard opens
D(fi ), i I such that Mfi is a free Rfi -module for all i I.
(2) We say that M is finite locally free if we can choose the covering such that
each Mfi is finite free.
(3) We say that M is finite locally free of rank r if we can choose the covering
such that each Mfi is isomorhic to Rfr
.
i
Note that a finite locally free R-module is automatically finitely presented by
Lemma 23.2.

00NX

Lemma 77.2. Let R be a ring and let M be an R-module. The following are
equivalent
(1) M is finitely presented and R-flat,
(2) M is finite projective,
(3) M is a direct summand of a finite free R-module,
(4) M is finitely presented and for all p Spec(R) the localization Mp is free,
(5) M is finitely presented and for all maximal ideals m R the localization
Mm is free,
(6) M is finite and locally free,
(7) M is finite locally free, and
(8) M is finite, for every prime p the module Mp is free, and the function
M : Spec(R) Z,

p 7 dim(p) M R (p)

is locally constant in the Zariski topology.

Proof. First suppose M is finite projective, i.e., (2) holds. Take a surjection Rn
M and let K be the kernel. Since M is projective, 0 K Rn M 0 splits.
Hence (2) (3). The implication (3) (2) follows from the fact that a direct
summand of a projective is projective, see Lemma 76.2.
Assume (3), so we can write K M
= Rn . So K is a direct summand of Rn and
thus finitely generated. This shows M = Rn /K is finitely presented. In other
words, (3) (1).
Assume M is finitely presented and flat, i.e., (1) holds. We will prove that (7)
holds. Pick any prime p and x1 , . . . , xr M which map to a basis of M R (p).
By Nakayamas Lemma 19.1 these elements generate Mg for some g R, g 6 p.
The corresponding surjection : Rgr Mg has the following two properties: (a)
Ker() is a finite Rg -module (see Lemma 5.3) and (b) Ker() (p) = 0 by flatness

COMMUTATIVE ALGEBRA

171

of Mg over Rg (see Lemma 38.12). Hence by Nakayamas lemma again there exists
a g 0 Rg such that Ker()g0 = 0. In other words, Mgg0 is free.
A finite locally free module is a finite module, see Lemma 23.2, hence (7) (6).
It is clear that (6) (7) and that (7) (8).
A finite locally free module is a finitely presented module, see Lemma 23.2, hence
(7) (4). Of course (4) implies (5). Since we may check flatness locally (see
Lemma 38.19) we conclude that (5) implies (1). At this point we have
(2) ks

+3 (3)

+3 (1)
KS
(5) ks

+3 (7) ks

(4)

+3 (6)
$
(8)

Suppose that M satisfies (1), (4), (5), (6), and (7). We will prove that (3) holds. It
suffices to show that M is projective. We have to show that HomR (M, ) is exact.
Let 0 N 00 N N 0 0 be a short exact sequence of R-module. We have to
0
show that 0 HomR (M, N 00 ) HomR (M, N ) HomR (M,
S N ) 0 is exact. As
M is finite locally free there exist a covering Spec(R) = D(fi ) such that Mfi is
finite free. By Lemma 10.2 we see that
0 HomR (M, N 00 )fi HomR (M, N )fi HomR (M, N 0 )fi 0
is equal to 0 HomRfi (Mfi , Nf00i ) HomRfi (Mfi , Nfi ) HomRfi (Mfi , Nf0 i ) 0
which is exact as Mfi is free and as the localization 0 Nf00i Nfi Nf0 i 0
is exact (as localization is exact). Whence we see that 0 HomR (M, N 00 )
HomR (M, N ) HomR (M, N 0 ) 0 is exact by Lemma 23.2.
Finally, assume that (8) holds. Pick a maximal ideal m R. Pick x1 , . . . , xr M
which map to a (m)-basis of M R (m) = M/mM . In particular M (m) = r. By
Nakayamas Lemma 19.1 there exists an f R, f 6 m such that x1 , . . . , xr generate
Mf over Rf . By the assumption that M is locally constant there exists a g R,
g 6 m such that M is constant equal to r on D(g). We claim that
X
: Rfr
(a1 , . . . , ar ) 7
ai xi
g Mf g ,
is an isomorphism. This claim will show that M is finite locally free, i.e., that (7)
holds. To see the claim it suffices to show that the induced map on localizations
p : Rpr Mp is an isomorphism for all p D(f g), see Lemma 23.1. By our
(p)
choice of f the map p is surjective. By assumption (8) we have Mp
= Rp M
and by our choice of g we have M (p) = r. Hence p determines a surjection
Rpr Mp
= Rpr whence is an isomorphism by Lemma 15.4. (Of course this last
fact follows from a simple matrix argument also.)

00NY

Remark 77.3. It is not true that a finite R-module which is R-flat is automatically projective. A counter example is where R = C (R) is the ring of infinitely
differentiable functions on R, and M = Rm = R/I where m = {f R | f (0) = 0}
and I = {f R | ,  > 0 : f (x) = 0 x, |x| < }.

00NZ

Lemma 77.4. (Warning: see Remark 77.3.) Suppose R is a local ring, and M is
a finite flat R-module. Then M is finite free.

COMMUTATIVE ALGEBRA

172

Proof. Follows from the equational criterion of flatness, see Lemma 38.11. Namely,
suppose that x1 , . . . , xr M map to a basis of M/mM . By Nakayamas Lemma
19.1 these elements generate M . We want to show there is no relation among the
xi . Instead, we will show by induction on n that if x1 , . . . , xn M are linearly
independent in the vector space M/mM then they are independent over R.
The base case of the induction is where we have x M , x 6 mM and a relation
f x =P0. By the equational criterion there exist yj M and aj R such that
x=
aj yj and f aj = 0 for all j. Since x 6 mM we see that at least one aj is a
unit and hence f = 0.
P
Suppose that
fi xi is a relation among x1 , . . . , xn . By our choice of xi we have
fi m. According to thePequational criterion
of flatness there exist aij R and
P
yj M such that xi =
aij yj and
fi aij = 0. Since xn 6 mM we see that
P
Pn1
anj 6 m for at least one j. Since
fi aij = 0 we get fn = i=1 (aij /anj )fi . The
P
Pn1
relation fi xi = 0 now can be rewritten as i=1 fi (xi + (aij /anj )xn ) = 0. Note
that the elements xi + (aij /anj )xn map to n 1 linearly independent elements of
M/mM . By induction assumption we get that all the fi , i n 1 have to be zero,
Pn1
and also fn = i=1 (aij /anj )fi . This proves the induction step.

00O1

Lemma 77.5. Let R S be a flat local homomorphism of local rings. Let M

be a finite R-module. Then M is finite projective over R if and only if M R S is
finite projective over S.
Proof. Suppose that M R S is finite projective over S. By Lemma 77.2 it is
finite free. Pick x1 , . . . , xr M whose residue classes generate M/mR M . Clearly
we see that x1 1, .P
. . , xr 1 are a basis for M R S. This implies that the map
Rr M, (ai ) 7
ai xi becomes an isomorphism after tensoring with S. By
faithful flatness of R S, see Lemma 38.17 we see that it is an isomorphism. 

02M9

Lemma 77.6. Let R be a semi-local ring. Let M be a finite locally free module. If
M has constant rank, then M is free. In particular, if R has connected spectrum,
then M is free.
Proof. Omitted. Hints: First show that M/mi M has the same dimension d for all
maximal ideal m1 , . . . , mn of R using the spectrum is connected. Next, show that
there exist elements x1 , . . . , xd M which form a basis for each M/mi M by the
Chinese remainder theorem. Finally show that x1 , . . . , xd is a basis for M .

Here is a technical lemma that is used in the chapter on groupoids.

03C1

Lemma 77.7. Let R be a local ring with maximal ideal m and infinite residue
field. Let R S be a ring map. Let M be an S-module and let N M be an
R-submodule. Assume
(1) S is semi-local and mS is contained in the radical of S,
(2) M is a finite free S-module, and
(3) N generates M as an S-module.
Then N contains an S-basis of M .
Proof. Assume M is free of rank n. Let I = rad(S). By Nakayamas Lemma 19.1
a sequence of elements m1 , . . . , mn is a basis for M if and only if mi M/IM
generate M/IM . Hence we may replace M by M/IM , N by N/(N IM ), R
by R/m, and S by S/IS. In this case we see that S is a finite product of fields

COMMUTATIVE ALGEBRA

173

S = k1 . . . kr and M = k1n . . . krn . The fact that N M generates

M
linear combination
P as an S-module means that there exist xj N such that an
aj xj with aj S has a nonzero component in each factor ki . Because
R=k
P
is an infinite field, this means that also some linear combination y =
cj xj with
cj k has a nonzero component in each factor. Hence y N generates a free
direct summand Sy M . By induction on n the result holds for M/Sy and the
submodule N = N/(N Sy). In other words there exist y 2 , . . . , y n in N which
(freely) generate M/Sy. Then y, y2 , . . . , yn (freely) generate M and we win.

78. Open loci defined by module maps
05GD

The set of primes where a given module map is surjective, or an isomorphism is

sometimes open. In the case of finite projective modules we can look at the rank
of the map.

05GE

Lemma 78.1. Let R be a ring. Let : M N be a map of R-modules with N a

finite R-module. Then we have the equality
U = {p R | p : Mp Np is surjective}
= {p R | (p) : M (p) N (p) is surjective}
and U is an open subset of Spec(R). Moreover, for any f R such that D(f ) U
the map Mf Nf is surjective.
Proof. The equality in the displayed formula follows from Nakayamas lemma.
Nakayamas lemma also implies that U is open. See Lemma 19.1 especially part
(3). If D(f ) U , then Mf Nf is surjective on all localizations at primes of Rf ,
and hence it is surjective by Lemma 23.1.


05GF

Lemma 78.2. Let R be a ring. Let : M N be a map of finitely presented

R-modules. Then
U = {p R | p : Mp Np is an isomorphism}
is an open subset of Spec(R).
P
Proof. Let p U . Pick a presentation N = Rn / j=1,...,m Rkj . Denote ei the
image in N of the ith basis vector of Rn . For each i {1, . . . , n} choose an
element mi Mp such that (mi ) = fi ei for some fi R, fi 6 p. This is possible
as p is an isomorphism. Set f = f1 . . . fn and let : Rfn M be the map which
maps the ith basis vector to mi /fi . Note that f is the localization at f of the
given map Rn N . As p is an isomorphism we see that (kj ) is an element of
M which maps to zero in Mp . Hence we see that there exist gj R, gj 6 p such
that gj (kj ) = 0. Setting g = g1 . . . gm , we see that g factors through Nf g to give
a map Nf g Mf g . By construction this map is inverse to f g . Hence f g is an
isomorphism, which implies that D(f g) U as desired.


00O0

Lemma 78.3. Let R be a ring. Let : P1 P2 be a map of finite projective

modules. Then
(1) The set U of primes p Spec(R) such that (p) is injective is open
and for any f R such that D(f ) U we have
(a) P1,f P2,f is injective, and
(b) the module Coker()f is finite projective over Rf .

COMMUTATIVE ALGEBRA

174

(2) The set W of primes p Spec(R) such that (p) is surjective is open
and for any f R such that D(f ) W we have
(a) P1,f P2,f is surjective, and
(b) the module Ker()f is finite projective over Rf .
(3) The set V of primes p Spec(R) such that (p) is an isomorphism is
open and for any f R such that D(f ) V the map : P1,f P2,f is an
isomorphism of modules over Rf .
Proof. To prove the set U is open we may work locally on Spec(R). Thus we may
replace R by a suitable localization and assume that P1 = Rn1 and P2 = Rn2 , see
Lemma 77.2. In this case injectivity of (p) is equivalent to n1 n2 and some
n1 n1 minor f of the matrix of being invertible in (p). Thus D(f ) U . This
argument also shows that P1,p P2,p is injective for p U .
Now suppose D(f ) U . By the remark in the previous paragraph and Lemma
23.1 we see that P1,f P2,f is injective, i.e., (1)(a) holds. By Lemma 77.2 to prove
(1)(b) it suffices to prove that Coker() is finite projective locally on D(f ). Thus,
as we saw above, we may assume that P1 = Rn1 and P2 = Rn2 and that some
minor of the matrix of is invertible in R. If the minor in question corresponds to
the first n1 basis vectors of Rn2 , then using the last n2 n1 basis vectors we get a
map Rn2 n1 Rn2 Coker() which is easily seen to be an isomorphism.
Openness of W and (2)(a) for d(f ) W follow from Lemma 78.1. Since P2,f is
projective over Rf we see that f : P1,f P2,f has a section and it follows that
Ker()f is a direct summand of P2,f . Therefore Ker()f is finite projective. Thus
(2)(b) holds as well.
It is clear that V = U W is open and the other statement in (3) follows from
(1)(a) and (2)(a).

79. Faithfully flat descent for projectivity of modules
058B

In the next few sections we prove, following Raynaud and Gruson [GR71], that the
projectivity of modules descends along faithfully flat ring maps. The idea of the
proof is to use devissage `
a la Kaplansky [Kap58] to reduce to the case of countably
generated modules. Given a well-behaved filtration of a module M , devissage allows
us to express M as a direct sum of successive quotients of the filtering submodules
(see Section 83). Using this technique, we prove that a projective module is a
direct sum of countably generated modules (Theorem 83.5). To prove descent
of projectivity for countably generated modules, we introduce a Mittag-Leffler
condition on modules, prove that a countably generated module is projective if and
only if it is flat and Mittag-Leffler (Theorem 92.3), and then show that the property
of being a Mittag-Leffler module descends (Lemma 94.1). Finally, given an arbitrary
module M whose base change by a faithfully flat ring map is projective, we filter
M by submodules whose successive quotients are countably generated projective
modules, and then by devissage conclude M is a direct sum of projectives, hence
projective itself (Theorem 94.5).
We note that there is an error in the proof of faithfully flat descent of projectivity
in [GR71]. There, descent of projectivity along faithfully flat ring maps is deduced
from descent of projectivity along a more general type of ring map ([GR71, Example
3.1.4(1) of Part II]). However, the proof of descent along this more general type

COMMUTATIVE ALGEBRA

175

of map is incorrect. In [Gru73], Gruson explains what went wrong, although he

does not provide a fix for the case of interest. Patching this hole in the proof of
faithfully flat descent of projectivity comes down to proving that the property of
being a Mittag-Leffler module descends along faithfully flat ring maps. We do this
in Lemma 94.1.
80. Characterizing flatness
058C

In this section we discuss criteria for flatness. The main result in this section is
Lazards theorem (Theorem 80.4 below), which says that a flat module is the colimit
of a directed system of free finite modules. We remind the reader of the equational
criterion for flatness, see Lemma 38.11. It turns out that this can be massaged
into a seemingly much stronger property.

058D

Lemma 80.1. Let M be an R-module. The following are equivalent:

(1) M is flat.
(2) If f : Rn M is a module map and x Ker(f ), then there are module
maps h : Rn Rm and g : Rm M such that f = g h and x Ker(h).
(3) Suppose f : Rn M is a module map, N Ker(f ) any submodule, and
h : Rn Rm a map such that N Ker(h) and f factors through h.
0
Then given any x Ker(f ) we can find a map h0 : Rn Rm such that
N + Rx Ker(h0 ) and f factors through h0 .
(4) If f : Rn M is a module map and N Ker(f ) is a finitely generated
submodule, then there are module maps h : Rn Rm and g : Rm M
such that f = g h and N Ker(h).
Proof. That (1) is equivalent to (2) is just a reformulation of the equational criterion for flatness6. To show (2) implies (3), let g : Rm M be the map such that
0
f factors as f = g h. By (2) find h00 : Rm Rm such that h00 kills h(x) and
g : Rm M factors through h00 . Then taking h0 = h00 h works. (3) implies (4) by
induction on the number of generators of N Ker(f ) in (4). Clearly (4) implies
(2).


058E

Lemma 80.2. Let M be an R-module. Then M is flat if and only if the following
condition holds: if P is a finitely presented R-module and f : P M a module
map, then there is a free finite R-module F and module maps h : P F and
g : F M such that f = g h.
Proof. This is just a reformulation of condition (4) from Lemma 80.1.

058F

Lemma 80.3. Let M be an R-module. Then M is flat if and only if the following
condition holds: for every finitely presented R-module P , if N M is a surjective
R-module map, then the induced map HomR (P, N ) HomR (P, M ) is surjective.
6In fact, a module map f : Rn M corresponds to a choice of elements x , x , . . . , x of M
n
1
2
(namely, the images of the standard basis elements e1 , e2 , . . . , en ); furthermore, an P
element x
Ker(f ) corresponds to a relation between these x1 , x2 , . . . , xn (namely, the relation i fi xi = 0,
where the fi are the coordinates of x). The module map h (represented as an m n-matrix)
corresponds to the matrix (aij ) from Lemma 38.11, and the yj of Lemma 38.11 are the images of
the standard basis vectors of Rm under g.

COMMUTATIVE ALGEBRA

176

Proof. First suppose M is flat. We must show that if P is finitely presented, then
given a map f : P M , it factors through the map N M . By Lemma 80.2 the
map f factors through a map F M where F is free and finite. Since F is free,
this map factors through N M . Thus f factors through N M .
Conversely, suppose the condition of the lemma holds. Let f : P M be a
map from a finitely presented module P . Choose a free module N with a surjection
N M onto M . Then f factors through N M , and since P is finitely generated,
f factors through a free finite submodule of N . Thus M satisfies the condition of
Lemma 80.2, hence is flat.

058G

Theorem 80.4 (Lazards theorem). Let M be an R-module. Then M is flat if

and only if it is the colimit of a directed system of free finite R-modules.
Proof. A colimit of a directed system of flat modules is flat, as taking directed
colimits is exact and commutes with tensor product. Hence if M is the colimit of
a directed system of free finite modules then M is flat.
For the converse, first recall that any module M can be written as the colimit of
a directed system of finitely presented modules, in the following way. Choose a
surjection f : RI M for some set I, and let K be the kernel. Let E be the set
of ordered pairs (J, N ) where J is a finite subset of I and N is a finitely generated
submodule of RJ K. Then E is made into a directed partially ordered set by
defining (J, N ) (J 0 , N 0 ) if and only if J J 0 and N N 0 . Define Me = RJ /N
for e = (J, N ), and define fee0 : Me Me0 to be the natural map for e e0 .
Then (Me , fee0 ) is a directed system and the natural maps fe : Me M induce an

=
isomorphism colimeE Me
M.
Now suppose M is flat. Let I = M Z, write (xi ) for the canonical basis of RI , and
take in the above discussion f : RI M to be the map sending xi to the projection
of i onto M . To prove the theorem it suffices to show that the e E such that Me
is free form a cofinal subset of E. So let e = (J, N ) E be arbitrary. By Lemma
80.2 there is a free finite module F and maps h : RJ /N F and g : F M such
h

that the natural map fe : RJ /N M factors as RJ /N

F
M . We are going
to realize F as Me0 for some e0 e.
Let {b1 , . . . , bn } be a finite basis of F . Choose n distinct elements i1 , . . . , in I
such that i`
/ J for all `, and such that the image of xi` under f : RI M equals
the image of b` under g : F M . This is possible since every element of M can
be written as f (xi ) for infinitely many distinct i I (by our choice of I). Now let
0
J 0 = J {i1 , . . . , in }, and define RJ F by xi 7 h(xi ) for i J and xi` 7 b` for
0
` = 1, . . . , n. Let N 0 = Ker(RJ F ). Observe:
(1) The square
0
/F
RJ _

RI
0


/M

is commutative, hence N K = Ker(f );

0
(2) RJ F is a surjection onto a free finite module, hence it splits and so N 0
is finitely generated;

COMMUTATIVE ALGEBRA

177

(3) J J 0 and N N 0 .
By (1) and (2) e0 = (J 0 , N 0 ) is in E, by (3) e0 e, and by construction Me0 =
0
RJ /N 0

= F is free.
81. Universally injective module maps
058H

Next we discuss universally injective module maps, which are in a sense complementary to flat modules (see Lemma 81.5). We follow Lazards thesis [Laz69]; also
see [Lam99].

058I

Definition 81.1. Let f : M N be a map of R-modules. Then f is called

universally injective if for every R-module Q, the map f R idQ : M R Q
N R Q is injective. A sequence 0 M1 M2 M3 0 of R-modules is called
universally exact if it is exact and M1 M2 is universally injective.

058J

Example 81.2. Examples of universally exact sequences.

(1) A split short exact sequence is universally exact since tensoring commutes
with taking direct sums.
(2) The colimit of a directed system of universally exact sequences is universally
exact. This follows from the fact that taking directed colimits is exact and
that tensoring commutes with taking colimits. In particular the colimit of
a directed system of split exact sequences is universally exact. We will see
below that, conversely, any universally exact sequence arises in this way.
Next we give a list of criteria for a short exact sequence to be universally exact. They
are analogues of criteria for flatness given above. Parts (3)-(6) below correspond,
respectively, to the criteria for flatness given in Lemmas 38.11, 80.1, 80.3, and
Theorem 80.4.

058K

Theorem 81.3. Let

f1

f2

0 M1 M2 M3 0
be an exact sequence of R-modules. The following are equivalent:
(1) The sequence 0 M1 M2 M3 0 is universally exact.
(2) For every finitely presented R-module Q, the sequence
0 M 1 R Q M 2 R Q M 3 R Q 0
is exact.
(3) Given elements xi M1 (i = 1, . . . , n), yj M2 (j = 1, . . . , m), and
aij R (i = 1, . . . , n, j = 1, . . . , m) such that for all i
X
f1 (xi ) =
aij yj ,
j

there exists zj M1 (j = 1, . . . , m) such that for all i,

X
xi =
aij zj .
j

(4) Given a commutative diagram of R-module maps

/ Rm

Rn

M1

f1


/ M2

COMMUTATIVE ALGEBRA

178

where m and n are integers, there exists a map Rm M1 making the top
triangle commute.
(5) For every finitely presented R-module P , the R-module map HomR (P, M2 )
HomR (P, M3 ) is surjective.
(6) The sequence 0 M1 M2 M3 0 is the colimit of a directed system
of split exact sequences of the form
0 M1 M2,i M3,i 0
where the M3,i are finitely presented.
Proof. Obviously (1) implies (2).
P
Next we show (2) implies (3). Let f1 (xi ) = j aij yj be relations as in (3). Let
m
(d
Rn , and Rm Rn the map given by dj 7
Pj ) be a basis for R , (ei ) a basis for
m
a
e
.
Let
Q
be
the
cokernel
of
R

Rn . Then tensoring Rm Rn Q 0
i ij i
by the map f1 : M1 M2 , we get a commutative diagram
M1m

/ M n

/ M 1 R Q

/0


M2m


/ M n


/ M 2 R Q

/0

where M1m M1n is given by

X
X
(z1 , . . . , zm ) 7 (
a1j zj , . . . ,
anj zj ),
j

M2m

M2n

and

is given similarly. We want to show x = (x1 , . . . , xn ) M1n

is in the image of M1m M1n . By (2) the map M1 Q M2 Q is injective,
hence by exactness of the top row it is enough to show x maps to 0 in M2 Q, and
so by exactness of the bottom row it is enough to show the image of x in M2n is
in the image of M2m M2n . This is true by assumption.
Condition (4) is just a translation of (3) into diagram form.
Next we show (4) implies (5). Let : P M3 be a map from a finitely presented
R-module P . We must show that lifts to a map P M2 . Choose a presentation
of P ,
g1

g2

Rn Rm P 0.
Using freeness of Rn and Rm , we can construct h2 : Rm M2 and then h1 : Rn
M1 such that the following diagram commutes
Rn

g1

/ Rm

f1


/ M2

h1


/ M1

g2

/P

/0

f2


/ M3

h2

/ 0.

By (4) there is a map k1 : Rm M1 such that k1 g1 = h1 . Now define h02 : Rm

M2 by h02 = h2 f1 k1 . Then
h02 g1 = h2 g1 f1 k1 g1 = h2 g1 f1 h1 = 0.

COMMUTATIVE ALGEBRA

179

Hence by passing to the quotient h02 defines a map 0 : P M2 such that 0 g2 =

h02 . In a diagram, we have
g2
/P
Rm
h02

 |
M2

0
f2


/ M3 .

where the top triangle commutes. We claim that 0 is the desired lift, i.e. that
f2 0 = . From the definitions we have
f2 0 g2 = f2 h02 = f2 h2 f2 f1 k1 = f2 h2 = g2 .
Since g2 is surjective, this finishes the proof.
Now we show (5) implies (6). Write M3 as the colimit of a directed system of
finitely presented modules M3,i , see Lemma 8.13. Let M2,i be the fiber product of
M3,i and M2 over M3 by definition this is the submodule of M2 M3,i consisting
of elements whose two projections onto M3 are equal. Let M1,i be the kernel of the
projection M2,i M3,i . Then we have a directed system of exact sequences
0 M1,i M2,i M3,i 0,
and for each i a map of exact sequences
/ M2,i
/ M1,i
0

/ M1


/ M2

/ M3,i

/0


/ M3

/0

compatible with the directed system. From the definition of the fiber product M2,i ,
it follows that the map M1,i M1 is an isomorphism. By (5) there is a map
M3,i M2 lifting M3,i M3 , and by the universal property of the fiber product
this gives rise to a section of M2,i M3,i . Hence the sequences
0 M1,i M2,i M3,i 0
split. Passing to the colimit, we have a commutative diagram
/ colim M1,i
/ colim M3,i
/ colim M2,i
0


/ M1


/ M2

/0


/ M3

/0

with exact rows and outer vertical maps isomorphisms. Hence colim M2,i M2 is
also an isomorphism and (6) holds.
Condition (6) implies (1) by Example 81.2 (2).

The previous theorem shows that a universally exact sequence is always a colimit of
split short exact sequences. If the cokernel of a universally injective map is finitely
presented, then in fact the map itself splits:
058L

Lemma 81.4. Let

0 M1 M2 M3 0
be an exact sequence of R-modules. Suppose M3 is of finite presentation. Then
0 M1 M2 M3 0

COMMUTATIVE ALGEBRA

180

is universally exact if and only if it is split.

Proof. A split short exact sequence is always universally exact, see Example 81.2.
Conversely, if the sequence is universally exact, then by Theorem 81.3 (5) applied
to P = M3 , the map M2 M3 admits a section.

The following lemma shows how universally injective maps are complementary to
flat modules.
058M

Lemma 81.5. Let M be an R-module. Then M is flat if and only if any exact
sequence of R-modules
0 M 1 M2 M 0
is universally exact.
Proof. This follows from Lemma 80.3 and Theorem 81.3 (5).

058N

Example 81.6. Non-split and non-flat universally exact sequences.

(1) In spite of Lemma 81.4, it is possible to have a short exact sequence of
R-modules
0 M1 M2 M3 0
that is L
universally exact butQnon-split. For instance, take R = Z, let

M1 =
n=1 Z, let M2 =
n=1 Z, and let M3 be the cokernel of the
inclusion M1 M2 . Then M1 , M2 , M3 are all flat since they are torsionfree (More on Algebra, Lemma 16.11), so by Lemma 81.5,
0 M1 M2 M3 0
is universally exact. However there can be no section s : M3 M2 . In
fact, if x is the image of (2, 22 , 23 , . . .) M2 in M3 , then any module map
s : M3 M2 must kill x. This is because x 2n M3 for any n 1, hence
s(x) is divisible by 2n for all n 1 and so must be 0.
(2) In spite of Lemma 81.5, it is possible to have a short exact sequence of
R-modules
0 M1 M2 M3 0
that is universally exact but with M1 , M2 , M3 all non-flat. In fact if M is
any non-flat module, just take the split exact sequence
0 M M M M 0.
For instance over R = Z, take M to be any torsion module.
(3) Taking the direct sum of an exact sequence as in (1) with one as in (2), we
get a short exact sequence of R-modules
0 M1 M2 M3 0
that is universally exact, non-split, and such that M1 , M2 , M3 are all nonflat.

058P

Lemma 81.7. Let 0 M1 M2 M3 0 be a universally exact sequence of

R-modules, and suppose M2 is flat. Then M1 and M3 are flat.

COMMUTATIVE ALGEBRA

181

Proof. Let 0 N N 0 N 00 0 be a short exact sequence of R-modules.

Consider the commutative diagram
/ M2 R N
/ M 3 R N
M 1 R N

M1 R N 0


/ M 2 R N 0


/ M3 R N 0


M1 R N 00


/ M2 R N 00


/ M3 R N 00

(we have dropped the 0s on the boundary). By assumption the rows give short
exact sequences and the arrow M2 N M2 N 0 is injective. Clearly this implies
that M1 N M1 N 0 is injective and we see that M1 is flat. In particular the left
and middle columns give rise to short exact sequences. It follows from a diagram
chase that the arrow M3 N M3 N 0 is injective. Hence M3 is flat.

05CH

Lemma 81.8. Let R be a ring. Let M M 0 be a universally injective R-module

map. Then for any R-module N the map M R N M 0 R N is universally
injective.
Proof. Omitted.

05CI

Lemma 81.9. Let R be a ring. A composition of universally injective R-module

maps is universally injective.
Proof. Omitted.

05CJ

Lemma 81.10. Let R be a ring. Let M M 0 and M 0 M 00 be R-module maps.

If their composition M M 00 is universally injective, then M M 0 is universally
injective.
Proof. Omitted.

05CK

Lemma 81.11. Let R S be a faithfully flat ring map. Then R S is universally injective as a map of R-modules. In particular R IS = I for any ideal
I R.
Proof. Let N be an R-module. We have to show that N N R S is injective.
As S is faithfully flat as an R-module, it suffices to prove this after tensoring with
S. Hence it suffices to show that N R S N R S R S, n s 7 n 1 s is
injective. This is true because there is a section, namely, n s s0 7 n ss0 . 

05CL

Lemma 81.12. Let R S be a ring map. Let M M 0 be a map of S-modules.

The following are equivalent
(1) M M 0 is universally injective as a map of R-modules,
(2) for each prime q of S the map Mq Mq0 is universally injective as a map
of R-modules,
0
(3) for each maximal ideal m of S the map Mm Mm
is universally injective
as a map of R-modules,
(4) for each prime q of S the map Mq Mq0 is universally injective as a map
of Rp -modules, where p is the inverse image of q in R, and
0
(5) for each maximal ideal m of S the map Mm Mm
is universally injective
as a map of Rp -modules, where p is the inverse image of m in R.

COMMUTATIVE ALGEBRA

182

Proof. Let N be an R-module. Let q be a prime of S lying over the prime p of R.

Then we have
(M R N )q = Mq R N = Mq Rp Np .
Moreover, the same thing holds for M 0 and localization is exact. Also, if N is
an Rp -module, then Np = N . Using this the equivalences can be proved in a
straightforward manner.
For example, suppose that (5) holds. Let K = Ker(M R N M 0 R N ). By
the remarks above we see that Km = 0 for each maximal ideal m of S. Hence
K = 0 by Lemma 23.1. Thus (1) holds. Conversely, suppose that (1) holds. Take
any q S lying over p R. Take any module N over Rp . Then by assumption
Ker(M R N M 0 R N ) = 0. Hence by the formulae above and the fact that
N = Np we see that Ker(Mq Rp N Mq0 Rp N ) = 0. In other words (4) holds.
Of course (4) (5) is immediate. Hence (1), (4) and (5) are all equivalent. We
omit the proof of the other equivalences.

05CM

Lemma 81.13. Let : A B be a ring map. Let S A and S 0 B be

multiplicative subsets such that (S) S 0 . Let M M 0 be a map of B-modules.
(1) If M M 0 is universally injective as a map of A-modules, then (S 0 )1 M
(S 0 )1 M 0 is universally injective as a map of A-modules and as a map of
S 1 A-modules.
(2) If M and M 0 are (S 0 )1 B-modules, then M M 0 is universally injective
as a map of A-modules if and only if it is universally injective as a map of
S 1 A-modules.
Proof. You can prove this using Lemma 81.12 but you can also prove it directly as
follows. Assume M M 0 is A-universally injective. Let Q be an A-module. Then
QA M QA M 0 is injective. Since localization is exact we see that (S 0 )1 (QA
M ) (S 0 )1 (Q A M 0 ) is injective. As (S 0 )1 (Q A M ) = Q A (S 0 )1 M and
similarly for M 0 we see that Q A (S 0 )1 M Q A (S 0 )1 M 0 is injective, hence
(S 0 )1 M (S 0 )1 M 0 is universally injective as a map of A-modules. This proves
the first part of (1). To see (2) we can use the following two facts: (a) if Q is an
S 1 A-module, then QA S 1 A = Q, i.e., tensoring with Q over A is the same thing
as tensoring with Q over S 1 A, (b) if M is any A-module on which the elements
of S are invertible, then M A Q = M S 1 A S 1 Q. Part (2) follows from this
immediately.


0AS5

Lemma 81.14. Let R be a ring and let M M 0 be a map of R-modules. If M 0

is flat, then M M 0 is universally injective if and only if M/IM M 0 /IM 0 is
injective for every finitely generated ideal I of R.
Proof. It suffices to show that M R Q M 0 R Q is injective for every finite
R-module Q, see Theorem 81.3. Then Q has a finite filtration 0 = Q0 Q1
. . . Qn = Q by submodules whose subquotients are isomorphic to cyclic modules
R/Ii , see Lemma 5.4. Since M 0 is flat, we obtain a filtration
M Q1

/ M Q2

/ ...

/ M Q



M 0 Q1 


/ M 0 Q2  

/ . . . 


/ M0 Q

COMMUTATIVE ALGEBRA

183

of M 0 R Q by submodules M 0 R Qi whose successive quotients are M 0 R R/Ii =

M 0 /Ii M 0 . A simple induction argument shows that it suffices to check M/Ii M
M 0 /Ii M 0 is injective. Note that the collection of finitely generated ideals Ii0 Ii
is a directed set. Thus M/Ii M = colim M/Ii0 M is a filtered colimit, similarly for
M 0 , the maps M/Ii0 M M 0 /Ii0 M 0 are injective by assumption, and since filtered
colimits are exact (Lemma 8.9) we conclude.

82. Descent for finite projective modules
058Q

In this section we give an elementary proof of the fact that the property of being a
finite projective module descends along faithfully flat ring maps. The proof does not
apply when we drop the finiteness condition. However, the method is indicative of
the one we shall use to prove descent for the property of being a countably generated
projective modulesee the comments at the end of this section.

058R

Lemma 82.1. Let M be an R-module. Then M is finite projective if and only if

M is finitely presented and flat.
Proof. This is part of Lemma 77.2. However, at this point we can give a more
elegant proof of the implication (1) (2) of that lemma as follows. If M is finitely
presented and flat, then take a surjection Rn M . By Lemma 80.3 applied to
P = M , the map Rn M admits a section. So M is a direct summand of a free
module and hence projective.

Here are some properties of modules that descend.

03C4

Lemma 82.2. Let R S be a faithfully flat ring map. Let M be an R-module.

Then
(1) if the S-module M R S is of finite type, then M is of finite type,
(2) if the S-module M R S is of finite presentation, then M is of finite presentation,
(3) if the S-module M R S is flat, then M is flat, and
(4) add more here as needed.
Proof. Assume M P
R S is of finite type. Let y1 , . . . , ym be generators of M R S
over S. Write yj =
xi fi for some x1 , . . . , xn M . Then we see that the map
: Rn M has the property that idS : S n M R S is surjective. Since
R S is faithfully flat we see that is surjective, and M is finitely generated.
Assume M R S is of finite presentation. By (1) we see that M is of finite type.
Choose a surjection Rn M and denote K the kernel. As R S is flat we see
that K R S is the kernel of the base change S n M R S. As M R S is of
finite presentation we conclude that K R S is of finite type. Hence by (1) we see
that K is of finite type and hence M is of finite presentation.
Part (3) is Lemma 38.8.

058S

Proposition 82.3. Let R S be a faithfully flat ring map. Let M be an Rmodule. If the S-module M R S is finite projective, then M is finite projective.
Proof. Follows from Lemmas 82.1 and 82.2.

The next few sections are about removing the finiteness assumption by using
devissage to reduce to the countably generated case. In the countably generated
case, the strategy is to find a characterization of countably generated projective

COMMUTATIVE ALGEBRA

184

modules analogous to Lemma 82.1, and then to prove directly that this characterization descends. We do this by introducing the notion of a Mittag-Leffer module
and proving that if a module M is countably generated, then it is projective if and
only if it is flat and Mittag-Leffler (Theorem 92.3). When M is finitely generated,
this statement reduces to Lemma 82.1 (since, according to Example 90.1 (1), a
finitely generated module is Mittag-Leffler if and only if it is finitely presented).
83. Transfinite d
evissage of modules
058T

In this section we introduce a devissage technique for decomposing a module into a

direct sum. The main result is that a projective module is a direct sum of countably
generated modules (Theorem 83.5 below). We follow [Kap58].

058U

Definition 83.1. Let M be an R-module. A direct sum devissage of M is a family

of submodules (M )S , indexed by an ordinal S and increasing (with respect to
inclusion), such that:
(0) M0 =S0;
(1) M = M ;
S
(2) if S is a limit ordinal, then M = < M ;
(3) if + 1 S, then M is a direct summand of M+1 .
If moreover
(4) M+1 /M is countably generated for + 1 S,
then (M )S is called a Kaplansky devissage of M .
The terminology is justified by the following lemma.

058V

Lemma 83.2.LLet M be an R-module. If (M )S is a direct sum devissage of

M , then M
= +1S M+1 /M .
Proof. By property (3) of a direct sum devissage, there is an inclusion M+1 /M
M for each S. Consider the map
M
f:
M+1 /M M
+1S

given by the sum of these inclusions. Transfinite induction on S shows that the
image contains M for every S: for = 0 this is true by (0); if + 1 is a
successor ordinal then it is clearly true; and if is a limit ordinal and it is true for
< , then it is true for by (2). Hence f is surjective by (1).
Transfinite induction on S also shows that for every S the restriction
M
f :
M+1 /M M
+1

of f is injective: For = 0 it is true. If it is true for all 0 < , then let x be in the
kernel and write x = (x+1 )+1 in terms of its components x+1 M+1 /M .
By property (3) both (x+1 )+1< and x+1 map to 0. Hence x+1 = 0 and, by
the assumption that the restriction f 0 is injective for all 0 < , also x+1 = 0 for
every + 1 < . So x = 0 and f is injective, which finishes the induction. We
conclude that f is injective since f is for each S.

058W

Lemma 83.3. Let M be an R-module. Then M is a direct sum of countably

generated R-modules if and only if it admits a Kaplansky devissage.

COMMUTATIVE ALGEBRA

185

Proof. The lemma takes care of the if direction. Conversely, suppose M =

L
Well-order I so that
iI Ni where each Ni is a countably generated R-module.
L
we can think of it as an ordinal. Then setting Mi = j<i Nj gives a Kaplansky
devissage (Mi )iI of M .

058X

Theorem 83.4. Suppose M is a direct sum of countably generated R-modules. If

P is a direct summand of M , then P is also a direct sum of countably generated
R-modules.
Proof. Write M = P Q. We are going to construct a Kaplansky devissage
(M )S of M which, in addition to the defining properties (0)-(4), satisfies:
(5) Each M is a direct summand of M ;
(6) M = P Q , where P = P M and Q = Q M .
(Note: if properties (0)-(2) hold, then in fact property (3) is equivalent to property
(5).)
To see how this implies the theorem, it is enough to show that (P )S forms a
Kaplansky devissage of P . Properties (0), (1), and (2) are clear. By (5) and (6)
for (M ), each P is a direct summand of M . Since P P+1 , this implies P is
a direct summand of P+1 ; hence (3) holds for (P ). For (4), note that
M+1 /M
= P+1 /P Q+1 /Q ,
so P+1 /P is countably generated because this is true of M+1 /M .
L
It remains to construct the M . Write M = iI Ni where each Ni is a countably
generated R-module. Choose a well-ordering of I. By transfinite induction we are
going to define an increasing family of submodules M of M , one for each ordinal
, such that M is a direct sum of some subset of the Ni .
For = 0 let M0 =S0. If is a limit ordinal and M has been defined for all < ,
then define M = < M . Since each M for < is a direct sum of a subset
of the Ni , the same will be true of M . If + 1 is a successor ordinal and M
has been defined, then define M+1 as follows. If M = M , then let M+1 = M .
If not, choose the smallest j I such that Nj is not contained in M . We will
construct an infinite matrix (xmn ), m, n = 1, 2, 3, . . . such that:
(1) Nj is contained in the submodule of M generated by the entries xmn ;
(2) if we write any entry xk` in terms of its P - and Q-components, xk` =
yk` + zk` , then the matrix (xmn ) contains a set of generators for each Ni
for which yk` or zk` has nonzero component.
Then we define M+1 to be the submodule of M generated by M and all xmn ;
by property (2) of the matrix (xmn ), M+1 will be a direct sum of some subset of
the Ni . To construct the matrix (xmn ), let x11 , x12 , x13 , . . . be a countable set of
generators for Nj . Then if x11 = y11 + z11 is the decomposition into P - and Qcomponents, let x21 , x22 , x23 , . . . be a countable set of generators for the sum of the
Ni for which y11 or z11 have nonzero component. Repeat this process on x12 to get
elements x31 , x32 , . . ., the third row of our matrix. Repeat on x21 to get the fourth
row, on x13 to get the fifth, and so on, going down along successive anti-diagonals

COMMUTATIVE ALGEBRA

186

as indicated below:

x11

x21

x31

x41

x12
x22
x32

x13
x23

x14
...

...

...

...

...
Transfinite induction on I (using the fact that we constructed M+1 to contain Nj
for the smallest j such that Nj is not contained in M ) shows that for each i I,
Ni is contained in some M . Thus, there is some large enough ordinal S satisfying:
for each i I there is S such that Ni is contained in M . This means (M )S
satisfies property (1) of a Kaplansky devissage of M . The family (M )S moreover
satisfies the other defining properties, and also (5) and (6) above: properties (0),
(2), (4), and (6) are clear by construction; property (5) is true because each M
is by construction a direct sum of some Ni ; and (3) is implied by (5) and the fact
that M M+1 .

As a corollary we get the result for projective modules stated at the beginning of
the section.
058Y

Theorem 83.5. If P is a projective R-module, then P is a direct sum of countably

generated projective R-modules.
Proof. A module is projective if and only if it is a direct summand of a free module,
so this follows from Theorem 83.4.

84. Projective modules over a local ring

058Z

In this section we prove a very cute result: a projective module M over a local ring
is free (Theorem 84.4 below). Note that with the additional assumption that M is
finite, this result is Lemma 77.4. In general we have:

0590

Lemma 84.1. Let R be a ring. Then every projective R-module is free if and only
if every countably generated projective R-module is free.
Proof. Follows immediately from Theorem 83.5.

Here is a criterion for a countably generated module to be free.

0591

Lemma 84.2. Let M be a countably generated R-module. Suppose any direct

summand N of M satisfies: any element of N is contained in a free direct summand
of N . Then M is free.
Proof. Let x1 , x2 , . . . be a countable set of generators for M . By the assumption
on M , we can construct
Ln by induction free R-modules F1 , F2 , . . . such that for every
positive integer
n,
i=1 Fi is a direct summand of M and contains x1 , . . . , xn .
L
Then M = i=1 Fi .


0592

Lemma 84.3. Let P be a projective module over a local ring R. Then any element
of P is contained in a free direct summand of P .

COMMUTATIVE ALGEBRA

187

Proof. Since P is projective it is a direct summand of some free R-module F , say

F = P Q. Let x P be the element that we wish to show is contained in a free
direct summand of P . Let B be a basis of F such that
Pnthe number of basis elements
needed in the expression of x is minimal, say x = i=1 ai ei for some ei B and
ai R.
P Then no aj can be expressed as a linear combination of the other ai ; for if
aj = i6=j ai bi for some bi R, then replacing ei by ei + bi ej for i 6= j and leaving
unchanged the other elements of B, we get a new basis for F in terms of which x
has a shorter expression.
Let ei = yi + zi , yi P,
Pzni Q be the decomposition of ei into its P - and Qcomponents. Write yi = j=1 bij ej +ti , where ti is a linear combination of elements
in B other than e1 , . . . , en . To finish the proof it suffices to show that the matrix
(bij ) is invertible. For then the map F F sending ei 7 yi for i = 1, . . . , n
and fixing B \ {e1 , . . . , en } is an isomorphism, so that y1 , . . . , yn together with
B \ {e1 , . . . , en } form a basis for F . Then the submodule N spanned by y1 , . . . , yn
is a free submodule of P ; N is a direct summand of P since N P and
Pn both N
and
P
are
direct
summands
of
F
;
and
x

N
since
x

P
implies
x
=
i=1 ai ei =
Pn
i=1 ai yi .
Pn
Pn
Now we prove that (bij ) is invertible. Plugging yi = j=1 bij ej +ti into i=1 ai ei =
Pn
Pn
i=1 ai yi and equating the coefficients of ej gives aj =
i=1 ai bij . But as noted
above, our choice of B guarantees that no aj can be written as a linear combination
of the other ai . Thus bij is a non-unit for i 6= j, and 1 bii is a non-unitso in
particular bii is a unitfor all i. But a matrix over a local ring having units along
the diagonal and non-units elsewhere is invertible, as its determinant is a unit. 
0593

Theorem 84.4. If P is a projective module over a local ring R, then P is free.

Proof. Follows from Lemmas 84.1, 84.2, and 84.3.

85. Mittag-Leffler systems

0594

The purpose of this section is to define Mittag-Leffler systems and why it is a useful
property.
In the following, I will be a directed partially ordered set, see Categories, Definition
21.2. Let (Ai , ji : Aj Ai ) be an inverse system of sets or of modules indexed by
I, see Categories, Definition 21.2. This is a directed inverse system as we assumed
I directed. For each
T i I, the images ji (Aj ) Ai for j i form a decreasing
family. Let A0i = ji ji (Aj ). Then ji (A0j ) A0i for j i, hence by restricting
we get a directed inverse system (A0i , ji |A0j ). From the construction of the limit of
an inverse system in the category of sets or modules, we have lim Ai = lim A0i . The
Mittag-Leffler condition on (Ai , ji ) is that A0i equals ji (Aj ) for some j i (and
hence equals ki (Ak ) for all k j):

0595

Definition 85.1. Let (Ai , ji ) be a directed inverse system of sets over I. Then
we say (Ai , ji ) is Mittag-Leffler inverse system if for each i I, the decreasing
family ji (Aj ) Ai for j i stabilizes. Explicitly, this means that for each i I,
there exists j i such that for k j we have ki (Ak ) = ji (Aj ). If (Ai , ji ) is a
directed inverse system of modules over a ring R, we say that it is Mittag-Leffler if
the underlying inverse system of sets is Mittag-Leffler.

COMMUTATIVE ALGEBRA

188

0596

Example 85.2. If (Ai , ji ) is a directed inverse system of sets or of modules and

the maps ji are surjective, then clearly the system is Mittag-Leffler. Conversely,
suppose (Ai , ji ) is Mittag-Leffler. Let A0i Ai be the stable image of ji (Aj ) for
j i. Then ji |A0j : A0j A0i is surjective for j i and lim Ai = lim A0i . Hence
the limit of the Mittag-Leffler system (Ai , ji ) can also be written as the limit of a
directed inverse system over I with surjective maps.

0597

Lemma 85.3. Let (Ai , ji ) be a directed inverse system over I. Suppose I is

countable. If (Ai , ji ) is Mittag-Leffler and the Ai are nonempty, then lim Ai is
nonempty.
Proof. Let i1 , i2 , i3 , . . . be an enumeration of the elements of I. Define inductively
a sequence of elements jn I for n = 1, 2, 3, . . . by the conditions: j1 = i1 , and
jn in and jn jm for m < n. Then the sequence jn is increasing and forms a
cofinal subset of I. Hence we may assume I = {1, 2, 3, . . .}. So by Example 85.2 we
are reduced to showing that the limit of an inverse system of nonempty sets with
surjective maps indexed by the positive integers is nonempty. This is obvious. 
The Mittag-Leffler condition will be important for us because of the following exactness property.

0598

Lemma 85.4. Let

fi

gi

0 Ai Bi Ci 0
be an exact sequence of directed inverse systems of abelian groups over I. Suppose
I is countable. If (Ai ) is Mittag-Leffler, then
0 lim Ai lim Bi lim Ci 0
is exact.
Proof. Taking limits of directed inverse systems is left exact, hence we only need
to prove surjectivity of lim Bi lim Ci . So let (ci ) lim Ci . For each i I, let
Ei = gi1 (ci ), which is nonempty since gi : Bi Ci is surjective. The system of
maps ji : Bj Bi for (Bi ) restrict to maps Ej Ei which make (Ei ) into an
inverse system of nonempty sets. It is enough to show that (Ei ) is Mittag-Leffler.
For then Lemma 85.3 would show lim Ei is nonempty, and taking any element of
lim Ei would give an element of lim Bi mapping to (ci ).
By the injection fi : Ai Bi we will regard Ai as a subset of Bi . Since (Ai ) is
Mittag-Leffler, if i I then there exists j i such that ki (Ak ) = ji (Aj ) for
k j. We claim that also ki (Ek ) = ji (Ej ) for k j. Always ki (Ek ) ji (Ej )
for k j. For the reverse inclusion let ej Ej , and we need to find xk Ek such
that ki (xk ) = ji (ej ). Let e0k Ek be any element, and set e0j = kj (e0k ). Then
gj (ej e0j ) = cj cj = 0, hence ej e0j = aj Aj . Since ki (Ak ) = ji (Aj ), there
exists ak Ak such that ki (ak ) = ji (aj ). Hence
ki (e0k + ak ) = ji (e0j ) + ji (aj ) = ji (ej ),
so we can take xk = e0k + ak .

COMMUTATIVE ALGEBRA

189

86. Inverse systems

03C9

In many papers (and in this section) the term inverse system is used to indicate
an inverse system over the partially ordered set (N, ). We briefly discuss such
systems in this section. This material will be discussed more broadly in Homology,
Section 27. Suppose we are given a ring R and a sequence of R-modules
2

M1 M2 M3 . . .
with maps as indicated. By composing successive maps we obtain maps ii0 : Mi
Mi0 whenever i i0 such that moreover ii00 = i0 i00 ii0 whenever i i0 i00 .
Conversely, given the system of maps ii0 we can set i = i(i1) and recover the
maps displayed above. In this case
Y
lim Mi = {(xi )
Mi | i (xi ) = xi1 , i = 2, 3, . . .}
compare with Categories, Section 15. As explained in Homology, Section 27 this is
actually a limit in the category of R-modules, as defined in Categories, Section 14.
03CA

Lemma 86.1. Let R be a ring. Let 0 Ki Li Mi 0 be short exact

sequences of R-modules, i 1 which fit into maps of short exact sequences
0

/ Ki
O

/ Li
O

/ Mi
O

/0

/ Ki+1

/ Li+1

/ Mi+1

/0

If for every i there exists a c = c(i) i such that Im(Kc Ki ) = Im(Kj Ki )

for all j c, then the sequence
0 lim Ki lim Li lim Mi 0
is exact.
Proof. This is a special case of the more general Lemma 85.4.

87. Mittag-Leffler modules

0599

A Mittag-Leffler module is (very roughly) a module which can be written as a

directed limit whose dual is a Mittag-Leffler system. To be able to give a precise
definition we need to do a bit of work.

059A

Definition 87.1. Let (Mi , fij ) be a directed system of R-modules. We say that
(Mi , fij ) is a Mittag-Leffler directed system of modules if each Mi is an R-module
of finite presentation and if for every R-module N , the inverse system
(HomR (Mi , N ), HomR (fij , N ))
is Mittag-Leffler.
We are going to characterize those R-modules that are colimits of Mittag-Leffler
directed systems of modules.

059B

Definition 87.2. Let f : M N and g : M M 0 be maps of R-modules. Then

we say g dominates f if for any R-module Q, we have Ker(f R idQ ) Ker(gR idQ ).
It is enough to check this condition for finitely presented modules.

COMMUTATIVE ALGEBRA

059C

190

Lemma 87.3. Let f : M N and g : M M 0 be maps of R-modules. Then g

dominates f if and only if for any finitely presented R-module Q, we have Ker(f R
idQ ) Ker(g R idQ ).
Proof. Suppose Ker(f R idQ ) Ker(g R idQ ) for all finitely presented modules
Q. If Q is an arbitrary module, write Q = colimiI Qi as a colimit of a directed
system of finitely presented modules Qi . Then Ker(f R idQi ) Ker(g R idQi )
for all i. Since taking directed colimits is exact and commutes with tensor product,
it follows that Ker(f R idQ ) Ker(g R idQ ).


0AUM

Lemma 87.4. Let f : M N and g : M M 0 be maps of R-modules. Consider

the pushout of f and g,
/N
M
f

g0



f0
/ N0
M0
Then g dominates f if and only if f 0 is universally injective.
Proof. Recall that N 0 is M 0 N modulo the submodule consisting of elements
(g(x), f (x)) for x M . From the construction of N 0 we have a short exact
sequence
0 Ker(f ) Ker(g) Ker(f ) Ker(f 0 ) 0.
Since tensoring commutes with taking pushouts, we have such a short exact sequence
0 Ker(f idQ ) Ker(g idQ ) Ker(f idQ ) Ker(f 0 idQ ) 0
for every R-module Q. So f 0 is universally injective if and only if Ker(f idQ )
Ker(g idQ ) for every Q, if and only if g dominates f .

The above definition of domination is sometimes related to the usual notion of
domination of maps as the following lemma shows.
059D

Lemma 87.5. Let f : M N and g : M M 0 be maps of R-modules. Suppose

Coker(f ) is of finite presentation. Then g dominates f if and only if g factors
through f , i.e. there exists a module map h : N M 0 such that g = h f .
Proof. Consider the pushout of f and g as in the statement of Lemma 87.4. From
the construction of the pushout it follows that Coker(f 0 ) = Coker(f ), so Coker(f 0 )
is of finite presentation. Then by Lemma 81.4, f 0 is universally injective if and only
if
f0
0 M 0 N 0 Coker(f 0 ) 0
splits. This is the case if and only if there is a map h0 : N 0 M 0 such that
h0 f 0 = idM 0 . From the universal property of the pushout, the existence of such
an h0 is equivalent to g factoring through f .


059E

Proposition 87.6. Let M be an R-module. Let (Mi , fij ) be a directed system of

finitely presented R-modules, indexed by I, such that M = colim Mi . Let fi : Mi
M be the canonical map. The following are equivalent:
(1) For every finitely presented R-module P and module map f : P M , there
exists a finitely presented R-module Q and a module map g : P Q such
that g and f dominate each other, i.e., Ker(f R idN ) = Ker(g R idN ) for
every R-module N .

COMMUTATIVE ALGEBRA

191

(2) For each i I, there exists j i such that fij : Mi Mj dominates

fi : Mi M .
(3) For each i I, there exists j i such that fij : Mi Mj factors through
fik : Mi Mk for all k i.
(4) For every R-module N , the inverse system (HomR (Mi , N ), HomR (fij , N ))
is Mittag-Leffler.
Q
(5) For N = sI Ms , the inverse system (HomR (Mi , N ), HomR (fij , N )) is
Mittag-Leffler.
Proof. First we prove the equivalence of (1) and (2). Suppose (1) holds and let
i I. Corresponding to the map fi : Mi M , we can choose g : Mi Q as in
(1). Since Mi and Q are of finite presentation, so is Coker(g). Then by Lemma
87.5, fi : Mi M factors through g : Mi Q, say fi = h g for some h : Q M .
Then since Q is finitely presented, h factors through Mj M for some j i, say
h = fj h0 for some h0 : Q Mj . In total we have a commutative diagram
>M a
fj

fi
fij

Mi
g

/ Mj
>
h0

Q
Thus fij dominates g. But g dominates fi , so fij dominates fi .
Conversely, suppose (2) holds. Let P be of finite presentation and f : P M a
module map. Then f factors through fi : Mi M for some i I, say f = fi g 0
for some g 0 : P Mi . Choose by (2) a j i such that fij dominates fi . We have
a commutative diagram
f

P
g0


Mi

fi

fij

/M
= O
fj

/ Mj

From the diagram and the fact that fij dominates fi , we find that f and fij g 0
dominate each other. Hence taking g = fij g 0 : P Mj works.
Next we prove (2) is equivalent to (3). Let i I. It is always true that fi dominates
fik for k i, since fi factors through fik . If (2) holds, choose j i such that fij
dominates fi . Then since domination is a transitive relation, fij dominates fik for
k i. All Mi are of finite presentation, so Coker(fik ) is of finite presentation for
k i. By Lemma 87.5, fij factors through fik for all k i. Thus (2) implies
(3). On the other hand, if (3) holds then for any R-module N , fij R idN factors
through fik R idN for k i. So Ker(fik R idN ) Ker(fij R idN ) for k i. But
Ker(fi R idN : Mi R N M R N ) is the union of Ker(fik R idN ) for k i.
Thus Ker(fi R idN ) Ker(fij R idN ) for any R-module N , which by definition
means fij dominates fi .

COMMUTATIVE ALGEBRA

192

It
Q is trivial that (3) implies (4) implies (5). We show (5) implies (3). Let N =
sI Ms . If (5) holds, then given i I choose j i such that
Im(Hom(Mj , N ) Hom(Mi , N )) = Im(Hom(Mk , N ) Hom(Mi , N ))
for all k j. Passing the product over s I outside of the Homs and looking at
the maps on each component of the product, this says
Im(Hom(Mj , Ms ) Hom(Mi , Ms )) = Im(Hom(Mk , Ms ) Hom(Mi , Ms ))
for all k j and s I. Taking s = j we have
Im(Hom(Mj , Mj ) Hom(Mi , Mj )) = Im(Hom(Mk , Mj ) Hom(Mi , Mj ))
for all k j. Since fij is the image of id Hom(Mj , Mj ) under Hom(Mj , Mj )
Hom(Mi , Mj ), this shows that for any k j there is h Hom(Mk , Mj ) such that
fij = h fik . If j k then we can take h = fkj . Hence (3) holds.

059F

Definition 87.7. Let M be an R-module. We say that M is Mittag-Leffler if the

equivalent conditions of Proposition 87.6 hold.
In particular a finitely presented module is Mittag-Leffler.

059G

Remark 87.8. Let M be a flat R-module. By Lazards theorem (Theorem 80.4)

we can write M = colim Mi as the colimit of a directed system (Mi , fij ) where
the Mi are free finite R-modules. For M to be Mittag-Leffler, it is enough for the
inverse system of duals (HomR (Mi , R), HomR (fij , R)) to be Mittag-Leffler. This
follows from criterion (4) of Proposition 87.6 and the fact that for a free finite Rmodule F , there is a functorial isomorphism HomR (F, R) R N
= HomR (F, N ) for
any R-module N .

05CN

Lemma 87.9. If R is a ring and M , N are Mittag-Leffler modules over R, then

M R N is a Mittag-Leffler module.
Proof. Write M = colimiI Mi and N = colimjJ Nj as directed colimits of
finitely presented R-modules. Denote fii0 : Mi Mi0 and gjj 0 : Nj Nj 0 the
transition maps. Then Mi R Nj is a finitely presented R-module (see Lemma
11.14), and M R N = colim(i,j)IJ Mi R Mj . Pick (i, j) I J. By the definition of a Mittag-Leffler module we have Proposition 87.6 (3) for both systems. In
other words there exist i0 i and j 0 j such that for every choice of i00 i and
j 00 j there exist maps a : Mi00 Mi0 and b : Mj 00 Mj 0 such that fii0 = a fii00
and gjj 0 = b gjj 00 . Then it is clear that a b : Mi00 R Nj 00 Mi0 R Nj 0 serves the
same purpose for the system (Mi R Nj , fii0 gjj 0 ). Thus by the characterization
Proposition 87.6 (3) we conclude that M R N is Mittag-Leffler.


05CP

Lemma 87.10. Let R be a ring and M an R-module. Then M is Mittag-Leffler

if and only if for every finite free R-module F and module map f : F M , there
exists a finitely presented R-module Q and a module map g : F Q such that g and
f dominate each other, i.e., Ker(f R idN ) = Ker(g R idN ) for every R-module
N.
Proof. Since the condition is clear weaker than condition (1) of Proposition 87.6
we see that a Mittag-Leffler module satisfies the condition. Conversely, suppose
that M satisfies the condition and that f : P M is an R-module map from
a finitely presented R-module P into M . Choose a surjection F P where F
is a finite free R-module. By assumption we can find a map F Q where Q

COMMUTATIVE ALGEBRA

193

is a finitely presented R-module such that F Q and F M dominate each

other. In particular, the kernel of F Q contains the kernel of F P , hence we
obtain an R-module map g : P Q such that F Q is equal to the composition
F P Q. Let N be any R-module and consider the commutative diagram
F R N

/ Q R N
8


P R N

/ M R N

By assumption the kernels of F R N QR N and F R N M R N are equal.

Hence, as F R N P R N is surjective, also the kernels of P R N Q R N
and P R N M R N are equal.

05CQ

Lemma 87.11. Let R S be a finite and finitely presented ring map. Let M be
an S-module. If M is a Mittag-Leffler module over S then M is a Mittag-Leffler
module over R.
Proof. Assume M is a Mittag-Leffler module over S. Write M = colim Mi as a
directed colimit of finitely presented S-modules Mi . As M is Mittag-Leffler over S
there exists for each i an index j i such that for all k j there is a factorization
fij = h fik (where h depends on i, the choice of j and k). Note that by Lemma
35.21 the modules Mi are also finitely presented as R-modules. Moreover, all the
maps fij , fik , h are maps of R-modules. Thus we see that the system (Mi , fij )
satisfies the same condition when viewed as a system of R-modules. Thus M is
Mittag-Leffler as an R-module.


05CR

Lemma 87.12. Let R be a ring. Let S = R/I for some finitely generated ideal I.
Let M be an S-module. Then M is a Mittag-Leffler module over R if and only if
M is a Mittag-Leffler module over S.
Proof. One implication follows from Lemma 87.11. To prove the other, assume
M is Mittag-Leffler as an R-module. Write M = colim Mi as a directed colimit
of finitely presented S-modules. As I is finitely generated, the ring S is finite and
finitely presented as an R-algebra, hence the modules Mi are finitely presented
as R-modules, see Lemma 35.21. Next, let N be any S-module. Note that for
each i we have HomR (Mi , N ) = HomS (Mi , N ) as R S is surjective. Hence the
condition that the inverse system (HomR (Mi , N ))i satisfies Mittag-Leffler, implies
that the system (HomS (Mi , N ))i satisfies Mittag-Leffler. Thus M is Mittag-Leffler
over S by definition.


05CS

Remark 87.13. Let R S be a finite and finitely presented ring map. Let M
be an S-module which is Mittag-Leffler as an R-module. Then it is in general not
the case that if M is Mittag-Leffler as an S-module. For example suppose that S is
the ring of dual numbers over R, i.e., S = R R with 2 = 0. Then an S-module
consists of an R-module M endowed with a square zero R-linear endomorphism
 : M M . Now suppose that M0 is an R-module which is not Mittag-Leffler.
u
Choose a presentation F1
F0 M0 0 with F1 and F0 free R-modules. Set
M = F1 F0 with


0 0
=
: M M.
u 0

COMMUTATIVE ALGEBRA

194

Then M/M
= F1 M0 is not Mittag-Leffler over R = S/S, hence not MittagLeffler over S (see Lemma 87.12). On the other hand, M/M = M S S/S which
would be Mittag-Leffler over S if M was, see Lemma 87.9.
88. Interchanging direct products with tensor
059H

Let M be an R-module and

be a family of R-modules. Then there
Qlet (Q
)A Q
is a canonical map M R
Q

A
A (M R Q ) given on pure tensors
by x (q ) 7 (x q ). This map is not necessarily injective or surjective, as the
following example shows.

059I

Example 88.1.
Q Take R = Z, M = Q, and consider the family Qn = Z/n
Q for
n 1. Then n (M Qn ) = 0. HoweverQthere is an injection Q

(
n Qn )
Q
obtained by Q
tensoring the
injection
Z

Q
by
M
,
so
M

(
Q
)
is
nonzero.
n n
n n
Q
Thus M ( n Qn ) n (M Qn ) is not injective.
On the other hand,
= Q, and let Qn = Z for n 1. The
Q take again
Q R = Z, M Q
image of M ( n Qn ) n (M Qn ) = n M consists precisely of sequences of
the form (an /m)n1 with an Z and m some nonzero integer. Hence the map is
not surjective.
We
Q determine
Q below the precise conditions needed on M for the map M R
( Q ) (M R Q ) to be surjective, bijective, or injective for all choices of
(Q )A . This is relevant because the modules for which it is injective turn out to
be exactly Mittag-Leffler modules (Proposition 88.5).QIn what follows, if M is an
R-module and A a set, we write M A for the product A M .

059J

Proposition 88.2. Let M be an R-module. The following are equivalent:

(1) M is finitely generated.
Q
(2) For
Q every family (Q )A of R-modules, the canonical map M R ( Q )
(M R Q ) is surjective.
(3) For every R-module Q and every set A, the canonical map M R QA
(M R Q)A is surjective.
(4) For every set A, the canonical map M R RA M A is surjective.
Proof. First we prove (1) implies (2). Choose a surjection Rn M and consider
the commutative diagram

Q
= / Q
n
Rn R ( Q )
(R R Q )

M R (


Q

Q )


/ Q (M R Q ).

The top arrow is an isomorphism and the vertical arrows are surjections. We
conclude that the bottom arrow is a surjection.
Obviously (2) implies (3) implies (4), so it remains to prove (4) implies (1). In fact
for (1) to hold it suffices that the element d = (x)
of M M is in the image of
PxM
n
M
M
the map f : M R R M . In this case d = i=1 f (xi ai ) for some xi M
and ai RM . If for x M we write px : M M M for the projection onto the
x-th factor, then
Xn
Xn
x = px (d) =
px (f (xi ai )) =
px (ai )xi .
i=1

i=1

COMMUTATIVE ALGEBRA

195

Thus x1 , . . . , xn generate M .
059K

Proposition 88.3. Let M be an R-module. The following are equivalent:

(1) M is finitely presented.
Q
(2) For
Q every family (Q )A of R-modules, the canonical map M R ( Q )
(M R Q ) is bijective.
(3) For every R-module Q and every set A, the canonical map M R QA
(M R Q)A is bijective.
(4) For every set A, the canonical map M R RA M A is bijective.
Proof. First we prove (1) implies (2). Choose a presentation Rm Rn M and
consider the commutative diagram
Q
/ Rm R (Q Q )
/ M R (Q Q )
/0
Rm R ( Q )

/ Q (Rn R Q )

m
(R R Q )


/ Q (M R Q )

/ 0.

The first two vertical arrows

are isomorphisms
and the rows are exact. This implies
Q
Q
that the map M R ( Q ) (M R Q ) is surjective and, by a diagram
chase, also injective. Hence (2) holds.
Obviously (2) implies (3) implies (4), so it remains to prove (4) implies (1). From
Proposition 88.2, if (4) holds we already know that M is finitely generated. So we
can choose a surjection F M where F is free and finite. Let K be the kernel. We
must show K is finitely generated. For any set A, we have a commutative diagram
K R RA
f3


/ KA

/ F R R A

/ M R R A

f2
=

f1
=

/ FA


/ MA

/0

/ 0.

The map f1 is an isomorphism by assumption, the map f2 is a isomorphism since

F is free and finite, and the rows are exact. A diagram chase shows that f3 is
surjective, hence by Proposition 88.2 we get that K is finitely generated.

We need the following lemma for the next proposition.
059L

Lemma 88.4. Let M be an R-module, P a finitely presented R-module, and

f : P M a map. Let Q be an R-module and suppose x Ker(P Q M Q).
Then there exists a finitely presented R-module P 0 and a map f 0 : P P 0 such
that f factors through f 0 and x Ker(P Q P 0 Q).
Proof. Write M as a colimit M = colimiI Mi of a directed system of finitely
presented modules Mi . Since P is finitely presented, the map f : P M factors
through Mj M for some j I. Upon tensoring by Q we have a commutative
diagram
Mj Q
9
P Q

%
/ M Q.

COMMUTATIVE ALGEBRA

196

The image y of x in Mj Q is in the kernel of Mj Q M Q. Since M Q =

colimiI (Mi Q), this means y maps to 0 in Mj 0 Q for some j 0 j. Thus we
may take P 0 = Mj 0 and f 0 to be the composite P Mj Mj 0 .

059M

Proposition 88.5. Let M be an R-module. The following are equivalent:

(1) M is Mittag-Leffler.
Q
(2) For
Q every family (Q )A of R-modules, the canonical map M R ( Q )
(M R Q ) is injective.
Proof. First we prove (1) implies
(2). Q
Suppose M is Mittag-Leffler and let x
Q
be in the kernel of M R ( Q ) (M R Q ). Write M as a colimit
M = colim
of finitely presented modules Mi . Then
Q iI Mi of a directed system Q
M R ( QQ
) is the colimit of Mi R ( Q ). So x is the image
Q of an element
xi Mi R ( Q ). We must show that xi maps to 0 in Mj R ( Q ) for some
j i. Since M is Mittag-Leffler, we may choose j i such that Mi Mj and
Mi M dominate each other. Then consider the commutative diagram
Q
Q
/
Q )
M R ( Q )
(M
O
O R
M i R (

Q )

Q
Mj R ( Q )


/ Q (Mj R Q )

(Mi

R Q )

whose bottom two horizontal

to Proposition
Q maps are isomorphisms, according
Q
88.3. Since xi mapsQto 0 in (M RQ
Q ), its image in (Mi R Q ) is in the
kernel of Q
the map (Mi R
QQ ) (M R Q ). But this kernel equals the
kernel of (M

Q
)

(Mj R Q ) according to
Qi R
Qthe choice of j. Thus xi
maps to 0 in (Mj R Q ) and hence to 0 in Mj R ( Q ).
Now suppose (2) holds. We prove M satisfies formulation (1) of being MittagLeffler from Proposition 87.6. Let f : P M be a map from a finitely presented
module P to M . Choose a set B of representatives of the isomorphism classes of
finitely presented R-modules. Let A be the set of pairs (Q, x) where Q B and
x Ker(P Q M Q). For = (Q, x) A, we write Q for Q and x for x.
Consider the commutative diagram
Q
Q
/
Q )
M R ( Q )
(M
O
O R
P R (

Q )

/ Q (P R Q )

The top arrow is an injection by assumption,

Q and the bottom arrow is an isomorphism by Q
Proposition 88.3. Let x P R ( Q ) be the element corresponding
Q
to (x ) Q
(P R Q ) under this isomorphism. Then x Ker(P R ( Q )
M R ( Q )) since the top arrow in the diagram is injective. By Lemma 88.4,
we get a finitely presented module P 0 andQa map f 0 : P P 0 Q
such that f : P M
factors through f 0 and x Ker(P R ( Q ) P 0 R ( Q )). We have a

COMMUTATIVE ALGEBRA

197

commutative diagram
Q
P 0 R ( Q )
O
P R (

Q )

(P

R Q )
O

/ Q (P R Q )

where both the top and bottom arrows are isomorphisms by Proposition 88.3. Thus
since x is in the kernel of the left vertical map, (x ) is in the kernel of the right
vertical map. This means x Ker(P R Q P 0 R Q ) for every A. By
the definition of A this means Ker(P R Q P 0 R Q) Ker(P R Q M R Q)
for all finitely presented Q and, since f : P M factors through f 0 : P P 0 ,
actually equality holds. By Lemma 87.3, f and f 0 dominate each other.

0AS6

Lemma 88.6. Let M be a flat Mittag-Leffler module over R. Let F be an Rmodule and let x F R M . Then there exists a smallest submodule F 0 F such
that x F 0 R M .
Proof. Since M is flat we have F 0 R M F R M if F 0 F is a submodule,
hence the statement makes sense. Let I = {F 0 F | x F 0 R M } and for i I
denote Fi F the corresponding submodule. Then x maps to zero under the map
Y
F R M
(F/Fi R M )
whence by Proposition 88.5 x maps to zero under the map
Y

F/Fi R M
F R M
T
Since M is flat the kernel of this arrow is ( Fi ) R M which proves the lemma.

059N

Lemma 88.7. Let 0 M1 M2 M3 0 be a universally exact sequence of

R-modules. Then:
(1) If M2 is Mittag-Leffler, then M1 is Mittag-Leffler.
(2) If M1 and M3 are Mittag-Leffler, then M2 is Mittag-Leffler.
Proof. For any family (Q )A of R-modules we have a commutative diagram
/ M2 R (Q Q )
/ M3 R (Q Q )
/0
/ M1 R (Q Q )
0


/ Q (M1 Q )


/ Q (M2 Q )


/ Q (M3 Q )

with exact rows. Thus (1) and (2) follow from Proposition 88.5.
0AS7

/0


Lemma 88.8. If M = colim Mi is the colimit of a directed system of Mittag-Leffler

R-modules Mi with universally injective transition maps, then M is Mittag-Leffler.
Proof.
LetQ(Q )A be a family of R-modules. We haveQto show Q
that M R
Q
( Q ) M R Q is injective and we know that Mi R ( Q ) Mi R Q
is injective for each i, seeQ
Proposition 88.5.QSince commutes with filtered colimits,
it suffices to show that Mi R Q M R Q is injective. This is clear as
each of the maps Mi R Q M R Q is injective by our assumption that the
transition maps are universally injective.


COMMUTATIVE ALGEBRA

059P

198

L
Lemma 88.9. If M = iI Mi is a direct sum of R-modules, then M is MittagLeffler if and only if each Mi is Mittag-Leffler.
Proof. The only if direction follows from Lemma 88.7 (1) and the fact that a
split short exact sequence is universally exact. The converse follows from Lemma
88.8 but we can also argue it directly as follows. First note that if I is finite
then this follows from Lemma 88.7 (2). For general I, if all Mi are Mittag-Leffler
then we prove the same of M by verifying condition (1) of Proposition 87.6. Let
f : P M be a map from a finitely presented module P . Then f factors as
L
f0 L
0
0
P

i0 I 0 Mi ,
iI Mi for some finite subset I of I. By the finite case
L
0
M
is
Mittag-Leffler
and
hence
there
exists
a
finitely
presented module Q
i
i0 I 0
and a map g : P Q such that g and f 0 dominate each other. Then also g and f
dominate each other.


05CT

Lemma 88.10. Let R S be a ring map. Let M be an S-module. If S is MittagLeffler as an R-module, and M is flat and Mittag-Leffler as an S-module, then M
is Mittag-Leffler as an R-module.
Proof. We deduce this from the characterization of Proposition 88.5. Namely,
suppose that Q is a family of R-modules. Consider the composition
Q
Q
M R Q = M S S R Q

M S

Q 
(S R Q )

 Q
(M

Q
)
= (M R Q )
S
R

The first arrows is injective as M is flat over S and S is Mittag-Leffler over R and the
second arrow is injective as M is Mittag-Leffler over S. Hence M is Mittag-Leffler
over R.


05CU

05CV

89. Coherent rings

Q
We use the discussion on interchanging
and to determine for which rings
products of flat modules are flat. It turns out that these are the so-called coherent
rings. You may be more familiar with the notion of a coherent OX -module on a
ringed space, see Modules, Section 12.
Definition 89.1. Let R be a ring. Let M be an R-module.
(1) We say M is a coherent module if it is finitely generated and every finitely
generated submodule of M is finitely presented over R.
(2) We say R is a coherent ring if it is coherent as a module over itself.
Thus a ring is coherent if and only if every finitely generated ideal is finitely presented as a module. The category of coherent modules is abelian.

05CW

Lemma 89.2. Let R be a ring.

(1) A finite submodule of a coherent module is coherent.
(2) Let : N M be a homomorphism from a finite module to a coherent
module. Then Ker() is finite.

COMMUTATIVE ALGEBRA

199

(3) Let : N M be a homomorphism of coherent modules. Then Ker()

and Coker() are coherent modules.
(4) Given a short exact sequence of R-modules 0 M1 M2 M3 0 if
two out of three are coherent so is the third.
Proof. The first statement is immediate from the definition. During the rest of
the proof we will use the results of Lemma 5.3 without further mention.
Let : N M satisfy the assumptions of (2). Suppose that N is
by
Lgenerated
n
x1 , . . . , xn . By Definition 89.1 the kernel K of the induced map
R

M
,
i=1
ei 7 L
(xi ) is of finite type. Hence Ker() which is the image of the composition
n
K i=1 R N is of finite type. This proves (2).
Let : N M satisfy the assumptions of (3). By (2) the kernel of is of finite
type and hence by (1) it is coherent.
With the same hypotheses let us show that Coker() is coherent. Since M is
xi Coker(). We have to show that the kernel of the
finite so is Coker(). Let L
n
associated morphism : i=1 R Coker() is finite. Choose xi M lifting xi .
Ln
Thus lifts to : i=1 R M . Consider the following diagram
L
/ n R
/0
/ Ker()
/M
0
i=1

/ Ker()

/ Ln

i=1 R


/ Coker()

/0

By the snake lemma we get a short exact sequence 0 Ker() Ker()

Im() 0. Hence we see that Ker() is finite.
Statement (4) follows from (3).
Let 0 M1 M2 M3 0 be a short exact sequence of R-modules. It suffices
to prove that if M1 and M3 are coherent so is M2 . By Lemma 5.3 we see that M2
is finite. Let x1 , . . . , xn be finitely many elements of M2 . We have to show that the
module of relations K between them is finite. Consider the following commutative
diagram
L
L
/0
/ n R
/ n R
/0
0
i=1
i=1


/ M1


/ M2


/ M3

/0

with obvious notation. By the snake lemma we get an exact sequence 0 K

K3 M1 where K3 is the module of relations among the images of the xi in M3 .
Since M3 is coherent we see that K3 is a finite module. Since M1 is coherent we
see that the image I of K3 M1 is coherent. Hence K is the kernel of the map
K3 I between a finite module and a coherent module and hence finite by (2). 
05CX

Lemma 89.3. Let R be a ring. If R is coherent, then a module is coherent if and

only if it is finitely presented.
Proof. It is clear that a coherent module is finitely presented (over any ring).
Conversely, if R is coherent, then Rn is coherent and so is the cokernel of any
map Rm Rn , see Lemma 89.2.


COMMUTATIVE ALGEBRA

05CY

200

Lemma 89.4. A Noetherian ring is a coherent ring.

Proof. By Lemma 30.4 any finite R-module is finitely presented. In particular any
ideal of R is finitely presented.


05CZ

Proposition 89.5. Let R be a ring. The following are equivalent

(1) R is coherent,
(2) any product of flat R-modules is flat, and
(3) for every set A the module RA is flat.
Proof. Assume R coherent,
and
Q
Q let Q , A be a set of flat R-modules. We have
to show that I R Q Q is injective for every finitely generated ideal I
of R, see Lemma
Q 38.5. Since
Q R is coherent I is an R-module of finite presentation.
Hence I R Q =
I R Q by Proposition 88.3. The desired injectivity
follows as I R Q Q is injective by flatness of Q .
The implication (2) (3) is trivial.
Assume that the R-module RA is flat for every set A. Let I be a finitely generated
ideal in R. Then I R RA RA is injective by assumption. By Proposition 88.2
and the finiteness of I the image is equal to I A . Hence I R RA = I A for every set
A and we conclude that I is finitely presented by Proposition 88.3.

90. Examples and non-examples of Mittag-Leffler modules

059Q

We end this section with some examples and non-examples of Mittag-Leffler modules.

059R

Example 90.1. Mittag-Leffler modules.

(1) Any finitely presented module is Mittag-Leffler. This follows, for instance,
from Proposition 87.6 (1). In general, it is true that a finitely generated
module is Mittag-Leffler if and only it is finitely presented. This follows
from Propositions 88.2, 88.3, and 88.5.
(2) A free module is Mittag-Leffler since it satisfies condition (1) of Proposition
87.6.
(3) By the previous example together with Lemma 88.9, projective modules
are Mittag-Leffler.
We also want to add to our list of examples power series rings over a Noetherian
ring R. This will be a consequence the following lemma.

059S

Lemma 90.2. Let M be a flat R-module. The following are equivalent

(1) M is Mittag-Leffler, and
(2) if F is a finite free R-module and x F R M , then there exists a smallest
submodule F 0 of F such that x F 0 R M .
Proof. The implication (1) (2) is a special case of Lemma 88.6. Assume (2).
By Theorem 80.4 we can write M as the colimit M = colimiI Mi of a directed
system (Mi , fij ) of finite free R-modules. By Remark 87.8, it suffices to show that
the inverse system (HomR (Mi , R), HomR (fij , R)) is Mittag-Leffler. In other words,
fix i I and for j i let Qj be the image of HomR (Mj , R) HomR (Mi , R); we
must show that the Qj stabilize.
Since Mi is free and finite, we can make the identification HomR (Mi , Mj ) =
HomR (Mi , R) R Mj for all j. Using the fact that the Mj are free, it follows that

This is [Cha60,
Theorem 2.1].

COMMUTATIVE ALGEBRA

201

for j i, Qj is the smallest submodule of HomR (Mi , R) such that fij Qj R Mj .

Under the identification HomR (Mi , M ) = HomR (Mi , R) R M , the canonical map
fi : Mi M is in HomR (Mi , R) R M . By the assumption on M , there exists a
smallest submodule Q of HomR (Mi , R) such that fi Q R M . We are going to
show that the Qj stabilize to Q.
For j i we have a commutative diagram
Qj R Mj

/ HomR (Mi , R) R Mj


Qj R M


/ HomR (Mi , R) R M.

Since fij Qj R Mj maps to fi HomR (Mi , R) R M , it follows that fi

Qj R M . Hence, by the choice of Q, we have Q Qj for all j i.
Since the Qj are decreasing and Q Qj for all j i, to show that the Qj stabilize
to Q it suffices to find a j i such that Qj Q. As an element of
HomR (Mi , R) R M = colimjJ (HomR (Mi , R) R Mj ),
fi is the colimit of fij for j i, and fi also lies in the submodule
colimjJ (Q R Mj ) colimjJ (HomR (Mi , R) R Mj ).
It follows that for some j i, fij lies in Q R Mj . Since Qj is the smallest
submodule of HomR (Mi , R) with fij Qj R Mj , we conclude Qj Q.

05D0

Lemma 90.3. Let R be a Noetherian ring and A a set. Then M = RA is a flat

and Mittag-Leffler R-module.
Proof. Combining Lemma 89.4 and Proposition 89.5 we see that M is flat over
R. We show that M satisfies the condition of Lemma 90.2. Let F be a free finite
R-module. If F 0 is any submodule of F then it is finitely presented since R is
Noetherian. So by Proposition 88.3 we have a commutative diagram
F 0 R M


/ F R M

(F 0 )A

/ FA

by which we can identify the map F 0 R M F R M with (F 0 )A F A . Hence if

x F R M corresponds to (x ) F A , then the submodule of F 0 of F generated
by the x is the smallest submodule of F such that x F 0 R M .

059T

Lemma 90.4. Let R be a Noetherian ring and n a positive integer. Then the
R-module M = R[[t1 , . . . , tn ]] is flat and Mittag-Leffler.
Proof. As an R-module, we have M = RA for a (countable) set A. Hence this
lemma is a special case of Lemma 90.3.


059U

Example 90.5. Non Mittag-Leffler modules.

(1) By Example 88.1 and Proposition 88.5, Q is not a Mittag-Leffler Z-module.

COMMUTATIVE ALGEBRA

202

(2) We prove below (Theorem 92.3) that for a flat and countably generated
module, projectivity is equivalent to being Mittag-Leffler. Thus any flat,
countably generated, non-projective module M is an example of a nonMittag-Leffler module. For such an example, see Remark
Q 77.3.
(3) Let k be a field. Let R = k[[x]]. The R-module M = nN R/(xn ) is not
Mittag-Leffler. Namely, consider the element = (1 , 2 , 3 , . . .) defined by
m1
2m = x2
and n = 0 else, so
= (0, x, 0, x2 , 0, 0, 0, x4 , 0, 0, 0, 0, 0, 0, 0, x8 , . . .)
m

m1

Then the annihilator of in M/x2 M is generated x2

for m  0. But
if M was Mittag-Leffler, then there would exist a finite R-module Q and
an element 0 Q such that the annihilator of 0 in Q/xl Q agrees with
the annihilator of in M/xl M for all l 1, see Proposition 87.6 (1). Now
you can prove there exists an integer a 0 such that the annihilator of
0 in Q/xl Q is generated by either xa or xla for all l  0 (depending on
whether 0 Q is torsion or not). The combination of the above would give
for all l = 2m >> 0 the equality a = l/2 or l a = l/2 which
L is nonsensical.
n
(4) The same argument shows that (x)-adic completion of
nN R/(x ) is
not Mittag-Leffler over R = k[[x]] (hint: is actually an element of this
completion).
(5) Let R = k[a, b]/(a2 , ab, b2 ). Let S be the finitely presented R-algebra with
presentation S = R[t]/(at b). Then as an R-module S is countably
generated and indecomposable (details omitted). On the other hand, R
is Artinian local, hence complete local, hence a henselian local ring, see
Lemma 148.9. If S was Mittag-Leffler as an R-module, then it would be a
direct sum of finite R-modules by Lemma 148.13. Thus we conclude that
S is not Mittag-Leffler as an R-module.
91. Countably generated Mittag-Leffler modules
05D1

It turns out that countably generated Mittag-Leffler modules have a particularly

simple structure.

059W

Lemma 91.1. Let M be an R-module. Write M = colimiI Mi where (Mi , fij )

is a directed system of finitely presented R-modules. If M is Mittag-Leffler and
countably generated, then there is a directed countable subset I 0 I such that
M
= colimiI 0 Mi .
Proof. Let x1 , x2 , . . . be a countable set of generators for M . For each xn choose
i I such that xn is in the image of the canonical map fi : Mi M ; let I00 I
be the set of all these i. Now since M is Mittag-Leffler, for each i I00 we can
choose j I such that j i and fij : Mi Mj factors through fik : Mi Mk
for all k i (condition (3) of Proposition 87.6); let I10 be the union of I00 with all
of these j. Since I10 is a countable, we can enlarge it to a countable directed set
I20 I. Now we can apply the same procedure to I20 as we did to I00 to get a new
countable set I30 I. Then we enlarge I30 to a countable directed set I40 . Continuing
in this wayadding in a j as in Proposition 87.6 (3) for each i I`0 if ` is odd and
0
enlarging I`0 to a directed
S 0set if ` is evenwe get a sequence of subsets I` I for
0
` 0. The union I = I` satisfies:
(1) I 0 is countable and directed;

COMMUTATIVE ALGEBRA

203

(2) each xn is in the image of fi : Mi M for some i I 0 ;

(3) if i I 0 , then there is j I 0 such that j i and fij : Mi Mj factors
through fik : Mi Mk for all k I with k i. In particular Ker(fik )
Ker(fij ) for k i.
We claim that the canonical map colimiI 0 Mi colimiI Mi = M is an isomorphism. By (2) it is surjective. For injectivity, suppose x colimiI 0 Mi maps to 0
in colimiI Mi . Representing x by an element x
Mi for some i I 0 , this means
that fik (
x) = 0 for some k I, k i. But then by (3) there is j I 0 , j i, such
that fij (
x) = 0. Hence x = 0 in colimiI 0 Mi .

Lemma 91.1 implies that a countably generated Mittag-Leffler module M over R
is the colimit of a system
M1 M2 M3 M 4 . . .
with each Mn a finitely presented R-module. To see this argue as in the proof
of Lemma 85.3 to see that a countable directed partially ordered set has a cofinal
subset isomorphic to (N, ). Suppose R = k[x1 , x2 , x3 , . . .] and M = R/(xi ). Then
M is finitely generated but not finitely presented, hence not Mittag-Leffler (see
Example 90.1 part (1)). But of course you can write M = colimn Mn by taking
Mn = R/(x1 , . . . , xn ), hence the condition that you can write M as such a limit
does not imply that M is Mittag-Leffler.
05D2

Lemma 91.2. Let R be a ring. Let M be an R-module. Assume M is MittagLeffler and countably generated. For any R-module map f : P M with P finitely
generated there exists an endomorphism : M M such that
(1) : M M factors through a finitely presented R-module, and
(2) f = f .
Proof. Write M = colimiI Mi as a directed colimit of finitely presented Rmodules with I countable, see Lemma 91.1. The transition maps are denoted
fij
Q
and we use fi : Mi M to denote the canonical maps into M . Set N = sI Ms .
Denote
Y
Mi = HomR (Mi , N ) =
HomR (Mi , Ms )
sI

so that (Mi ) is an inverse system of R-modules over I. Note that HomR (M, N ) =
lim Mi . As M is Mittag-Leffler, we find for every i I an index k(i) i such that
\

Ei :=
Im(Mi0 Mi ) = Im(Mk(i)
Mi )
0
i i

Choose and fix j I such that Im(P M ) Im(Mj M ). This is possible as

P is finitely generated. Set k = k(j). Let x = (0, . . . , 0, idMk , 0, . . . , 0) Mk and
note that this maps to y = (0, . . . , 0, fjk , 0, . . . , 0) Mj . By our choice of k we see
that y Ej . By Example 85.2 the transition maps Ei Ej are surjective for each
i j and lim Ei = lim Mi = HomR (M, N ). Hence Lemma 85.3 guarantees there
exists an element z HomR (M, N ) which maps to y in Ej Mj . Let zk be the
kth component of z. Then zk : M Mk is a homomorphism such that
MO
fj

Mj

zk

/ Mk
=

fjk

COMMUTATIVE ALGEBRA

204

commutes. Let : M M be the composition fk zk : M Mk M . Then

factors through a finitely presented module by construction and fj = fj . Since
the image of f is contained in the image of fj this also implies that f = f . 
We will see later (see Lemma 148.13) that Lemma 91.2 means that a countably
generated Mittag-Leffler module over a henselian local ring is a direct sum of finitely
presented modules.
92. Characterizing projective modules
059V

The goal of this section is to prove that a module is projective if and only if it
is flat, Mittag-Leffler, and a direct sum of countably generated modules (Theorem
92.3 below).

059X

Lemma 92.1. Let M be an R-module. If M is flat, Mittag-Leffler, and countably

generated, then M is projective.
Proof. By Lazards theorem (Theorem 80.4), we can write M = colimiI Mi for a
directed system of finite free R-modules (Mi , fij ) indexed by a set I. By Lemma
91.1, we may assume I is countable. Now let
0 N1 N2 N3 0
be an exact sequence of R-modules. We must show that applying HomR (M, )
preserves exactness. Since Mi is finite free,
0 HomR (Mi , N1 ) HomR (Mi , N2 ) HomR (Mi , N3 ) 0
is exact for each i. Since M is Mittag-Leffler, (HomR (Mi , N1 )) is a Mittag-Leffler
inverse system. So by Lemma 85.4,
0 limiI HomR (Mi , N1 ) limiI HomR (Mi , N2 ) limiI HomR (Mi , N3 ) 0
is exact. But for any R-module N there is a functorial isomorphism HomR (M, N )
=
limiI HomR (Mi , N ), so
0 HomR (M, N1 ) HomR (M, N2 ) HomR (M, N3 ) 0
is exact.

059Y

Remark 92.2. Lemma 92.1 does not hold without the countable generation assumption. For example, the Z-module M = Z[[x]] is flat and Mittag-Leffler but not
projective. It is Mittag-Leffler by Lemma 90.4. Subgroups of free abelian groups
are free, hence a projective Z-module is in fact free and so are its submodules.
Thus to show M is not projective it suffices to produce a non-free submodule.
P Fixi
a prime p and consider the submodule N consisting of power series f (x) =
ai x
m
such
that
for
every
integer
m

1,
p
divides
a
for
all
but
finitely
many
i.
Then
i
P
ai pi xi is in N for all ai Z, so N is uncountable. Thus if N were free it would
have uncountable rank and the dimension of N/pN over Z/p would be uncountable.
This is not true as the elements xi N/pN for i 0 span N/pN .

059Z

Theorem
satisfies:
(1) M
(2) M
(3) M

92.3.

Let M be an R-module. Then M is projective if and only it

is flat,
is Mittag-Leffler,
is a direct sum of countably generated R-modules.

COMMUTATIVE ALGEBRA

205

Proof. First suppose M is projective. Then M is a direct summand of a free module, so M is flat and Mittag-Leffler since these properties pass to direct summands.
By Kaplanskys theorem (Theorem 83.5), M satisfies (3).
Conversely, suppose M satisfies (1)-(3). Since being flat and Mittag-Leffler passes
to direct summands, M is a direct sum of flat, Mittag-Leffler, countably generated
R-modules. Lemma 92.1 implies M is a direct sum of projective modules. Hence
M is projective.

05A0

Lemma 92.4. Let f : M N be universally injective map of R-modules. Suppose

M is a direct sum of countably generated R-modules, and suppose N is flat and
Mittag-Leffler. Then M is projective.
Proof. By Lemmas 81.7 and 88.7, M is flat and Mittag-Leffler, so the conclusion
follows from Theorem 92.3.


05A1

Lemma 92.5. Let R be a Noetherian ring and let M be a R-module. Suppose M

is a direct sum of countably generated R-modules, and suppose there is a universally
injective map M R[[t1 , . . . , tn ]] for some n. Then M is projective.
Proof. Follows from Lemmas 92.4 and 90.4.

93. Ascending properties of modules

05A2

All of the properties of a module in Theorem 92.3 ascend along arbitrary ring maps:

05A3

Lemma 93.1. Let R S be a ring map. Let M be an R-module. Then:

(1) If M is flat, then the S-module M R S is flat.
(2) If M is Mittag-Leffler, then the S-module M R S is Mittag-Leffler.
(3) If M is a direct sum of countably generated R-modules, then the S-module
M R S is a direct sum of countably generated S-modules.
(4) If M is projective, then the S-module M R S is projective.
Proof. All are obvious except (2). For this, use formulation (3) of being MittagLeffler from Proposition 87.6 and the fact that tensoring commutes with taking
colimits.

94. Descending properties of modules

05A4

We address the faithfully flat descent of the properties from Theorem 92.3 that
characterize projectivity. In the presence of flatness, the property of being a MittagLeffler module descends:

05A5

Lemma 94.1. Let R S be a faithfully flat ring map. Let M be an R-module.

If the S-module M R S is Mittag-Leffler, then M is Mittag-Leffler.
Proof. Write M = colimiI Mi as a directed colimit of finitely presented R-modules
Mi . Using Proposition 87.6, we see that we have to prove that for each i I there
exists i j, j I such that Mi Mj dominates Mi M .
Take N the pushout
Mi

/ Mj


M


/N

Email from Juan

Pablo Acosta Lopez
dated 12/20/14.

COMMUTATIVE ALGEBRA

206

Then the lemma is equivalent to the existence of j such that Mj N is universally

injective, see Lemma 87.4. Observe that the tensorization by S
M i R S

/ M j R S


M R S


/ N R S

Is a pushout diagram. So because M R S = colimiI Mi R S expresses M R S as

a colimit of S-modules of finite presentation, and M R S is Mittag-Leffler, there
exists j i such that Mj R S N R S is universally injective. So using that
R S is faithfully flat we conclude that Mj N is universally injective too. 
At this point the faithfully flat descent of countably generated projective modules
follows easily.
05A6

Lemma 94.2. Let R S be a faithfully flat ring map. Let M be an R-module.

If the S-module M R S is countably generated and projective, then M is countably
generated and projective.
Proof. Follows from Lemma 82.2, Lemma 94.1, the fact that countable generation
descends, and Theorem 92.3.

All that remains is to use devissage to reduce descent of projectivity in the general
case to the countably generated case. First, two simple lemmas.

05A7

Lemma 94.3. Let R S be a ring map, let M be an R-module, and let Q

be a countably generated S-submodule of M R S. Then there exists a countably
generated R-submodule P of M such that Im(P R S M R S) contains Q.
P
Proof. Let y1 , y2 , . . . be generators for Q and write yj = k xjk sjk for some
xjk M and sjk S. Then take P be the submodule of M generated by the
xjk .


05A8

Lemma 94.4.
Let R S be a ring map, and let M be an R-module. Suppose
L
M R S = iI Qi is a direct sum of countably generated S-modules Qi . If N is a
countably generated submodule of M , then there is a countably generated
submodule
L
N 0 of M such that N 0 N and Im(N 0 R S M R S) =
Q
for some
0
i
iI
subset I 0 I.
Proof. Let N00 = N . We construct by induction an increasing sequence of countably generated submodules N`0 M for ` = 0, 1, 2, . . . such that: if I`0 is the set of
i I such that the projection of Im(N`0 R S M R S) onto Qi is nonzero, then
0
0
Im(N`+1
R S M R S) contains Qi for all i I`0 . To construct N`+1
from N`0 ,
0
let Q be the sum of (the countably many) Qi for i I` , choose P as in LemmaS94.3,
0
and then S
let N`+1
= N`0 + P . Having constructed the N`0 , just take N 0 = ` N`0
0
0
and I = ` I` .


05A9

Theorem 94.5. Let R S be a faithfully flat ring map. Let M be an R-module.

If the S-module M R S is projective, then M is projective.
Proof. We are going to construct a Kaplansky devissage of M to show that it
is a direct sum of projective
modules and hence projective. By Theorem 83.5 we
L
can write M R S = iI Qi as a direct sum of countably generated S-modules

COMMUTATIVE ALGEBRA

207

Qi . Choose a well-ordering on M . By transfinite induction we are going to define

an increasing family of submodules M of M , one for each ordinal , such that
M R S is a direct sum of some subset of the Qi .
For = 0 let M0 =S0. If is a limit ordinal and M has been defined for all < ,
then define M = < M . Since each M R S for < is a direct sum of a
subset of the Qi , the same will be true of M R S. If +1 is a successor ordinal and
M has been defined, then define M+1 as follows. If M = M , then let M+1 = M .
Otherwise choose the smallest x M (with respect to the fixed well-ordering) such
that x
/ M . Since S is flat over R, (M/M ) R S = M R S/M R S, so since
M R S is a direct sum of some Qi , the same is true of (M/M ) R S. By Lemma
94.4, we can find a countably generated R-submodule P of M/M containing the
image of x in M/M and such that P R S (which equals Im(P R S
LM R S)
since S is flat over R) is a direct sum of some Qi . Since M R S =
iI Qi is
projective and projectivity passes to direct summands, P R S is also projective.
Thus by Lemma 94.2, P is projective. Finally we define M+1 to be the preimage
of P in M , so that M+1 /M = P is countably generated and projective. In
particular M is a direct summand of M+1 since projectivity of M+1 /M implies
the sequence 0 M M+1 M+1 /M 0 splits.
Transfinite induction on M (using the fact that we constructed M+1 to contain
the smallest x M not contained in M ) shows that each x M is contained in
some M . Thus, there is some large enough ordinal S satisfying: for each x M
there is S such that x M . This means (M )S satisfies property (1)
of a Kaplansky devissage
L of M . The other properties are clear by construction.
We conclude M =
+1S M+1 /M . Since each M+1 /M is projective by
construction, M is projective.

95. Completion
00M9

Suppose that R is a ring and I is an ideal. We define the completion of R with

respect to I to be the limit
R = limn R/I n .
An element of R is given by a sequence of elements fn R/I n such that fn
fn+1 mod I n for all n. We will view R as an R-algebra. Similarly, if M is an
R-module then we define the completion of M with respect to I to be the limit
M = limn M/I n M.
An element of M is given by a sequence of elements mn M/I n M such that
mn mn+1 mod I n M for all n. We will view M as an R -module. From this
description it is clear that there are always canonical maps
M M

and M R R M .

Moreover, given a map : M N of modules we get an induced map : M

N on completions making the diagram
M

/N


M


/ N

COMMUTATIVE ALGEBRA

208

commute. In general completion is not an exact functor, see Examples, Section 8.

Here are some initial positive results.
0315

Lemma 95.1. Let R be a ring. Let I R be an ideal. Let : M N be a map

of R-modules.
(1) If M/IM N/IN is surjective, then M N is surjective.
(2) If M N is surjective, then M N is surjective.
(3) If 0 K M N 0 is a short exact sequence of R-modules and N is
flat, then 0 K M N 0 is a short exact sequence.
(4) The map M R R M is surjective for any finite R-module M .
Proof. Assume M/IM N/IN is surjective. Then the map M/I n M N/I n N
is surjective for each n 1 by Nakayamas lemma. More precisely, apply Lemma
19.1 part (11) to the map M/I n M N/I n N over the ring R/I n and the nilpotent
ideal I/I n to see this. Set Kn = {x M | (x) I n N }. Thus we get short exact
sequences
0 Kn /I n M M/I n M N/I n N 0
n+1
n
We claim that the canonical
P map Kn+1 /I n M Kn /I M is surjective. Namely,
if x Kn writeP(x) = zj nj with zj I , nj N . By assumption we can write
nj = (mj ) + zjk njk with mj M , zjk I and njk N . Hence
X
X
(x
zj mj ) =
zj zjk njk .
P
This means that x0 = x zj mj Kn+1 maps to x which proves the claim. Now
we may apply Lemma 86.1 to the inverse system of short exact sequences above to
see (1). Part (2) is a special case of (1). If the assumptions of (3) hold, then for
each n the sequence
0 K/I n K M/I n M N/I n N 0
is short exact by Lemma 38.12. Hence we can directly apply Lemma 86.1 to conclude (3) is true. To see (4)Pchoose generators xi M , i = 1, . . . , n. Then the map
Rn M , (a1 ,P
. . . , an ) 7
ai xi is surjective. Hence by (2) we see (R )n M ,
(a1 , . . . , an ) 7 ai xi is surjective. Assertion (4) follows from this.


0BNG

Lemma 95.2. Let R be a ring. Let I R be an ideal. Let 0 M N Q 0

be an exact sequence of R-modules such that Q is annihilated by a power of I. Then
completion produces an exact sequence 0 M N Q 0. In particular,
(I n M ) = Ker(M M/I n M ) for all n.
Proof. Say I c Q = 0. It is immediate that the canonical map Q Q is an
isomorphism. By Lemma 95.1 we see that N Q 0 is exact. Denote : M
N the given map.
Let (yn ), yn N represent an element of N which maps to zero in Q. Then
yn = (xn ) for some xn M for n c. We have xn+1 xn 1 (I n N ).
Since I c Q = 0 we see that I n N I nc (M ) = (I nc M ) for n c. Hence
xn+1 xn I nc M . Thus (xn+c ) gives a well defined element of M which maps
to our element (yn ) in N .
Let (xn ), xn M represent an element of M which maps to zero in N . Then
(xn ) I n N . Arguing as above we can find x0n I nc M such that (xn x0n ) = 0
hence xn = x0n . This implies that the element (xn ) is zero in M .


COMMUTATIVE ALGEBRA

0317

209

Definition 95.3. Let R be a ring. Let I R be an ideal. Let M be an R-module.

We say M is I-adically complete if the map
M M = limn M/I n M
is an isomorphism7. We say R is I-adically complete if R is I-adically complete as
an R-module.
It is not true that the completion of an R-module M with respect to I is I-adically
complete. For an example see Examples, Section 6. Here is a lemma from an
unpublished note of Lenstra and de Smit.

0318

Lemma 95.4. Let R be a ring. Let I R be an ideal. Let M be an R-module.

Denote Kn = Ker(M M/I n M ). Then M is I-adically complete if and only
if Kn is equal to I n M for all n 1.
Proof. The module I n M is contained in Kn . Thus for each n 1 there is a
canonical exact sequence
0 Kn /I n M M /I n M M/I n M 0.
As I n M maps onto I n M/I n+1 M we see that Kn+1 + I n M = Kn . Thus the
inverse system {Kn /I n M }n1 has surjective transition maps. By Lemma 86.1 we
see that there is a short exact sequence
0 limn Kn /I n M (M ) M 0
Hence M is complete if and only if Kn /I n M = 0 for all n 1.

05GG

Lemma 95.5. Let R be a ring. Let I be a finitely generated ideal of R. Let M be

an R-module. Then
(1) the completion M is I-adically complete, and
(2) I n M = Ker(M M/I n M ) = (I n M ) for all n 1.
In particular R is I-adically complete, I n R = (I n ) , and R /I n R = R/I n .
Proof. By Lemma 95.2 we have the second equality in (2). Let Kn = Ker(M
M/I n M ). By Lemma 95.4 it suffices to show that Kn = I n M . Write I =
(f1 , . . . , ft ). Let z Kn . Write z = (z m ) with z m M/I m M . Choose zm M
mapping to z m in M/I m M . Then zm+1 = zm mod I m . Write zn+1 = zn + n ,
zn+2 = zn+1 + n+1 , etc. Then m I m M . Thus the infinite sum
z = zn + n + n+1 + n+2 + . . .

converges in M . For m n we have m I m M hence we can write

X
m =
f1j1 . . . ftjt J,m
j1 +...+jt =n

with J,m I mn M . Our assumption z Kn means zn I n M hence we can also

write
X
zn =
f1j1 . . . ftjt J
j1 +...+jt =n

with J M . Then we can set

zJ = J + J,n + J,n+1 + J,n+2 + . . .
P
as an element of M . By construction z = J f1j1 . . . ftjt zJ . Hence z is an element
of I n M as desired.


7This includes the condition that T I n M = (0).

[Mat78, Theorem
15]

COMMUTATIVE ALGEBRA

05GI

210

Lemma 95.6. Let R be a ring, let I R be an ideal, and let R = lim R/I n .
(1) any element of R which maps to a unit of R/I is a unit,
(2) any element of 1 + I maps to an invertible element of R ,
(3) any element of 1 + IR is invertible in R , and
(4) the ideals IR and Ker(R R/I) are contained in the radical of R .
Proof. Let x R map to a unit x1 in R/I. Then x maps to a unit xn in R/I n
n

for every n by Lemma 31.3. Hence y = (x1

n ) lim R/I = R is an inverse to
x. Parts (2) and (3) follow immediately from (1). Part (4) follows from (1) and
Lemma 18.1.


090S

Lemma 95.7. Let A be a ring. Let I = (f1 , . . . , fr ) be a finitely generated ideal.

If M lim M/fin M is surjective for each i, then M lim M/I n M is surjective.
Proof. Note that lim M/I n M = lim M/(f1n , . . . , frn )M as I n (f1n , . . . , frn )
I rn . An element of lim M/(f1n , . . . , frn )M can be symbolically written as
X
X
=
fin xn,i
n0

with
Pxn,i nM .
to
xn,i fi in
090T

If M lim M/fin M is surjective,

then there is an xi M
P
lim M/fin M . Then x = xi maps to in lim M/I n M .

mapping


Lemma 95.8. Let A be a ring. Let I J A be ideals. If M is J-adically

complete and I is finitely generated, then M is I-adically complete.
Proof.
Assume M T
is J-adically complete and I is finitely generated. We have
T n
I M = 0 because J n M = 0. By Lemma 95.7 it suffices to prove the surjectivity
of M lim M/I n M in case I is generated by a single element. Say I = (f ). Let
xn M with xn+1 xn f n M . We have to show there exists an x M such that
xn x f n M for all n. As xn+1 xn J n M and as M is J-adically complete,
there exists an element x M such that xn x J n M . Replacing xn by xn x
we may assume that xn J n M . To finish the proof we will show that this implies
xn I n M . Namely, write xn xn+1 = f n zn . Then
xn = f n (zn + f zn+1 + f 2 zn+2 + . . .)
The sum zn + f zn+1 + f 2 zn+2 + . . . converges in M as f c J c . The sum f n (zn +
f zn+1 +f 2 zn+2 +. . .) converges in M to xn because the partial sums equal xn xn+c
and xn+c J n+c M .


0319

Lemma 95.9. Let R be a ring. Let I, J be ideals of R. Assume there exist integers
c, d > 0 such that I c J and J d I. Then completion with respect to I agrees
with completion with respect to J for any R-module. In particular an R-module M
is I-adically complete if and only if it is J-adically complete.
Proof. Consider the system of maps M/I n M M/J bn/dc M and the system of
maps M/J m M M/I bm/cc M to get mutually inverse maps between the completions.


031A

Lemma 95.10. Let R be a ring. Let I be an ideal of R. Let M be an I-adically

complete R-module, and let K M be an R-submodule. The following are equivalent
T
(1) K = (K + I n M ) and
(2) M/K is I-adically complete.

COMMUTATIVE ALGEBRA

211

Proof. Set N = M/K. By Lemma 95.1 the map M = M N is surjective.

Hence NT N is surjective. It is easy to see that the kernel of N N is the
module (K + I n M )/K.

031B

Lemma 95.11. Let R be a ring. Let I be an ideal of R. Let M be T

an R-module.
If (a) R is I-adically complete, (b) M is a finite R-module, and (c) I n M = (0),
then M is I-adically complete.
Proof. By Lemma 95.1 the
M = M R R = M R R M is surjective.
T map
n
The kernel of this map is I M hence zero by assumption. Hence M
= M and
M is complete.


031D

Lemma 95.12. Let R be a ring. Let I R be an ideal. Let M be an R-module.

Assume
(1) R
T is I-adically complete,
(2) n1 I n M = (0), and
(3) M/IM is a finite R/I-module.
Then M is a finite R-module.
Proof. Let x1 , . . . , xn M be elements whose images in M/IM generate M/IM
as a R/I-module. Denote M 0 M the R-submodule generated
T n by x1 , . . . , xn . By
0

Lemma 95.1 the

map
(M
)

M
is
surjective.
Since
I M = 0 we see in
T
particular that I n M 0 = (0). Hence by Lemma 95.11 we see that M 0 is complete,
and we conclude that MT0 M is surjective. Finally, the kernel of M M is
zero since it is equal to I n M = (0). Hence we conclude that M
= M0
= M is
finitely generated.

96. Completion for Noetherian rings

0BNH

In this section we discuss completion with respect to ideals in Noetherian rings.

00MA

Lemma 96.1. Suppose R is Noetherian.

(1) If N M is an injective map of finite R-modules, then the map on completions N M is injective.
(2) If M is a finite R-module, then M = M R R .
Proof. For the first statement, by the Artin-Rees Lemma 50.2, we have a constant
c such that I n M N equals I nc (I c M N ) I nc N . Thus if (ni ) N maps
to zero in M , then each ni maps to zero in N/I ic N . And hence nic = 0. Thus
N M is injective.
For the second statement let 0 K Rt M 0 be the presentation of M
corresponding to the generators x1 , . . . , xt of M . By Lemma 95.1 (Rt ) M is
surjective, and for any finitely generated R-module the canonical map M R R
M is surjective. Hence to prove the second statement it suffices to prove the kernel
of (Rt ) M is exactly K .
Let (xn ) (Rt ) be in the kernel. Note that each xn is in the image of the map
K/I n K (R/I n )t . Choose c such that (I n )t K I nc K, which is possible by
Artin-Rees (Lemma 50.2). For each n 0 choose yn K/I n+c K mapping to xn+c ,
and set zn = yn mod I n K. The elements zn satisfy zn+1 zn mod I n K = yn+1
yn mod I n K, and yn+1 yn I n+c Rt by construction. Hence zn+1 = zn mod I n K
by the choice of c above. In other words (zn ) K maps to (xn ) as desired.


COMMUTATIVE ALGEBRA

00MB

212

Lemma 96.2. Let R be a Noetherian ring. Let I R be an ideal.

(1) The ring map R R is flat.
(2) The functor M 7 M is exact on the category of finitely generated Rmodules.
Proof. Consider I R R R R R = R . According to Lemma 96.1 this is
identified with I R and I R is injective. Part (1) follows from Lemma
38.5. Part (2) follows from part (1) and Lemma 96.1 part (2).


00MC

Lemma 96.3. Let R be a Noetherian local ring. Let m R be the maximal ideal.
Let I m be an ideal. The ring map R R is faithfully flat. In particular the
completion with respect to m, namely limn R/mn is faithfully flat.
Proof. By Lemma 96.2 it is flat. The composition R R R/m where the last
map is the projection map R R/I combined with R/I R/m shows that m is
in the image of Spec(R ) Spec(R). Hence the map is faithfully flat by Lemma
38.15.


031C

Lemma 96.4. Let R be a Noetherian ring. Let I be an ideal of R. Let M be an

R-module. Then the completion M of M with respect to I is I-adically complete,
I n M = (I n M ) , and M /I n M = M/I n M .
Proof. This is a special case of Lemma 95.5 because I is a finitely generated
ideal.


05GH

Lemma 96.5. Let R be a ring. Let I R be an ideal. Assume

(1) R/I is a Noetherian ring,
(2) I is finitely generated.
Then R is a Noetherian ring complete with respect to IR .
Proof. By Lemma 95.5 we see that R is I-adically complete. Hence it is also IR adically complete. Since R /IR = R/I is Noetherian we see that after replacing
R by R we may in addition to assumptions (1) and (2) assume that also R is
I-adically complete.
Let f1 , . . . , ft be generators of I. Then there is a surjection
R/I[T1 , . . . , Tt ]
L
Lof nrings
I n /I n+1 mapping Ti to the element f i I/I 2 . Hence
I /I n+1 is a Noetherian
ring. Let J R be an ideal. Consider the ideal
M
M
J I n /J I n+1
I n /I n+1 .
Let g 1 , . . . , g m be generators of this ideal. We may choose g j to be a homogeneous
element of degree dj and we may pick gj J I dj mapping to g j J I dj /J I dj +1 .
We claim that g1 , . . . , gm generate J.
P
Let x J I n . There exist aj I max(0,ndj ) such
that x aj gj J I n+1 .
P
The reason is that J I n /J I n+1 is equal to
g j I ndj /I ndj +1 by our choice
of g1 , . . . , gm . Hence starting with x J we can find a sequence of vectors
(a1,n , . . . , am,n )n0 with aj,n I max(0,ndj ) such that
X
X
x=
aj,n gj mod I N +1
n=0,...,N
j=1,...,m
P
P
Setting Aj = n0 aj,n we see that x =
Aj gj as R is complete. Hence J is
finitely generated and we win.


COMMUTATIVE ALGEBRA

0316

213

Lemma 96.6. Let R be a Noetherian ring. Let I be an ideal of R. The completion

R of R with respect to I is Noetherian.
Proof. This is a consequence of Lemma 96.5. It can also be seen directly as follows.
Choose generators f1 , . . . , fn of I. Consider the map
R[[x1 , . . . , xn ]] R ,

xi 7 fi .

This is a well defined and surjective ring map (details omitted). Since R[[x1 , . . . , xn ]]
is Noetherian (see Lemma 30.2) we win.

Suppose R S is a local homomorphism of local rings (R, m) and (S, n). Let S be
the completion of S with respect to n. In general S is not the m-adic completion
of S. If nt mS for some t 1 then we do have S = lim S/mn S by Lemma 95.9.
In some cases this even implies that S is finite over R .
0394

Lemma 96.7. Let R S be a local homomorphism of local rings (R, m) and

(S, n). Let R , resp. S be the completion of R, resp. S with respect to m, resp. n.
If m and n are finitely generated and dim(m) S/mS < , then
(1) S is equal to the m-adic completion of S, and
(2) S is a finite R -module.
Proof. We have mS n because R S is a local ring map. The assumption
dim(m) S/mS < implies that S/mS is an Artinian ring, see Lemma 52.2. Hence

has dimension 0, see Lemma 59.4, hence n = mS. This and the fact that n is
finitely generated implies that nt mS for some t 1. By Lemma 95.9 we see that
S can be identified with the m-adic completion of S. As m is finitely generated
we see from Lemma 95.5 that S and R are m-adically complete. At this point
we may apply Lemma 95.12 to S as an R -module to conclude.


07N9

Lemma 96.8. Let R be a Noetherian ring. Let R S be a finite ring map. Let
p R be a prime and let q1 , . . . , qm be the primes of S lying over p (Lemma 35.19).
Then
Rp R S = Sq1 . . . Sqm
where the local rings Rp and Sqi are completed with respect to their maximal ideals.
Proof. We may replace R by the localization Rp and S by Sp = S R Rp . Hence
we may assume that R is a local Noetherian ring and that p = m is its maximal
ideal. The qi Sqi -adic completion Sqi is equal to the m-adic completion by Lemma
96.7. For every n 1 prime ideals of S/mn S are in 1-to-1 correspondence with
the maximal ideals
Q q1 , . . . , qm of S (by going up for S over R, see Lemma 35.20).
Hence S/mn S = Sqi /mn Sqi by Lemma 52.6 (using for example Proposition 59.6
to
that S/mn S is Artinian). Hence the m-adic completion S of S is equal to
Q see

Sqi . Finally, we have R R S = S by Lemma 96.1.



05D3

Lemma 96.9. Let R be a ring. Let I R be an ideal. Let 0 K P M 0

be a short exact sequence of R-modules. If M is flat over R and M/IM is a
projective R/I-module, then the sequence of I-adic completions
0 K P M 0
is a split exact sequence.

COMMUTATIVE ALGEBRA

214

Proof. As M is flat, each of the sequences

0 K/I n K P/I n P M/I n M 0
is short exact, see Lemma 38.12 and the sequence 0 K P M 0
is a short exact sequence, see Lemma 95.1. It suffices to show that we can find
splittings sn : M/I n M P/I n P such that sn+1 mod I n = sn . We will construct
these sn by induction on n. Pick any splitting s1 , which exists as M/IM is a
projective R/I-module. Assume given sn for some n > 0. Set Pn+1 = {x P |
x mod I n P Im(sn )}. The map : Pn+1 /I n+1 Pn+1 M/I n+1 M is surjective
(details omitted). As M/I n+1 M is projective as a R/I n+1 -module by Lemma 76.5
we may choose a section t : M/I n+1 M Pn+1 /I n+1 Pn+1 of . Setting sn+1
equal to the composition of t with the canonical map Pn+1 /I n+1 Pn+1 P/I n+1 P
works.

97. Taking limits of modules
09B7

In this section we discuss what happens when we take a limit of modules.

09B8

Lemma 97.1. Let A be a ring. Let I A be an ideal. Let (Mn ) be an inverse

system of A-modules. Set M = lim Mn . If Mn = Mn+1 /I n Mn+1 and I is finitely
generated then M/I n M = Mn and M is I-adically complete.
Proof. As Mn+1 Mn is surjective, the map M M1 L
is surjective. Pick xt M ,
t T mapping to generators of M1 . This gives a map tT A M . Note that
the images of xt in Mn generate Mn for all n too. Consider the exact sequences
M
0 Kn
A/I n Mn 0
tT

We claim
Kn+1 Kn is surjective. Namely, if y Kn choose a lift
L the map
n+1
y0
A/I
.
Then y 0 maps to an element of I n Mn+1 by our assumption
tT
n
0
M
n+1 /I Mn+1 . Hence we can modify our choice of y by an element of
Ln = M
n
n+1
0
0
so that y maps to zero in Mn+1 . Then y Kn+1 maps to y.
tT I /I
Hence (Kn ) is a sequence of modules with surjective transition maps and we obtain
an exact sequence
M

0 lim Kn
A M 0
tT

by Lemma 86.1. Fix an integer L

m. As I is finitely
the completion with
L generated,L
m
respect to I is complete and ( tT A) /I m ( tT A) =
(Lemma
tT A/I
95.5). We obtain a short exact sequence
M
(lim Kn )/I m (lim Kn )
A/I m M/I m M 0
tT

Since lim Kn Km is surjective we conclude that M/I m M = Mm . It follows in

particular that M is I-adically complete.

98. Criteria for flatness
00MD

In this section we prove some important technical lemmas in the Noetherian case.
We will (partially) generalize these to the non-Noetherian case in Section 127.

00ME

Lemma 98.1. Suppose that R S is a local homomorphism of Noetherian local

rings. Denote m the maximal ideal of R. Let M be a flat R-module and N a finite
S-module. Let u : N M be a map of R-modules. If u : N/mN M/mM is
injective then u is injective. In this case M/u(N ) is flat over R.

COMMUTATIVE ALGEBRA

215

Proof. First we claim that un : N/mn N M/mn M is injective for all n 1. We

proceed by induction, the base case is that u = u1 is injective. By our assumption
that M is flat over R we have a short exact sequence 0 M R mn /mn+1
M/mn+1 M M/mn M 0. Also, M R mn /mn+1 = M/mM R/m mn /mn+1 .
We have a similar exact sequence N R mn /mn+1 N/mn+1 N N/mn N 0
for N except we do not have the zero on the left. We also have N R mn /mn+1 =
N/mN R/m mn /mn+1 . Thus the map un+1 is injective as both un and the map
u idmn /mn+1 are.
By Krulls intersectionTtheorem (Lemma 50.4) applied to N over the ring S and
the ideal mS we have mn N = 0. Thus the injectivity of un for all n implies u is
injective.
To show that M/u(N ) is flat over R, it suffices to show that I R M/u(N )
M/u(N ) is injective for every ideal I R, see Lemma 38.5. Consider the diagram

N/IN

N R I

M/IM

M R I

M/(IN + u(N ))

M/u(N )

M/u(N ) R I

0
0
0

The arrow M R I M is injective. By the snake lemma (Lemma 4.1) we see

that it suffices to prove that N/IN injects into M/IM . Note that R/I S/IS
is a local homomorphism of Noetherian local rings, N/IN M/IM is a map of
R/I-modules, N/IN is finite over S/IS, and M/IM is flat over R/I and u mod I :
N/IN M/IM is injective modulo m. Thus we may apply the first part of the
proof to u mod I and we conclude.

00MF

Lemma 98.2. Suppose that R S is a flat and local ring homomorphism of

Noetherian local rings. Denote m the maximal ideal of R. Suppose f S is a
nonzerodivisor in S/mS. Then S/f S is flat over R, and f is a nonzerodivisor in
S.
Proof. Follows directly from Lemma 98.1.

00MG

Lemma 98.3. Suppose that R S is a flat and local ring homomorphism of

Noetherian local rings. Denote m the maximal ideal of R. Suppose f1 , . . . , fc is a
sequence of elements of S such that the images f 1 , . . . , f c form a regular sequence
in S/mS. Then f1 , . . . , fc is a regular sequence in S and each of the quotients
S/(f1 , . . . , fi ) is flat over R.
Proof. Induction and Lemma 98.2.

00MH

Lemma 98.4. Let R S be a local homomorphism of Noetherian local rings.

Let m be the maximal ideal of R. Let M be a finite S-modules. Suppose that (a)
M/mM is a free S/mS-module, and (b) M is flat over R. Then M is free and S
is flat over R.
Proof. Let x1 , . . . , xn be a basis for the free module M/mM . Choose x1 , . . . , xn
M with xi mapping to xi . Let u : S n M be the map which maps the ith

COMMUTATIVE ALGEBRA

216

standard basis vector to xi . By Lemma 98.1 we see that u is injective. On the other
hand, by Nakayamas Lemma 19.1 the map is surjective. The lemma follows.

00MI

Lemma 98.5. Let R S be a local homomorphism of local Noetherian rings.

Let m be the maximal ideal of R. Let 0 Fe Fe1 . . . F0 be a finite
complex of finite S-modules. Assume that each Fi is R-flat, and that the complex
0 Fe /mFe Fe1 /mFe1 . . . F0 /mF0 is exact. Then 0 Fe Fe1
. . . F0 is exact, and moreover the module Coker(F1 F0 ) is R-flat.
Proof. By induction on e. If e = 1, then this is exactly Lemma 98.1. If e > 1, we
see by Lemma 98.1 that Fe Fe1 is injective and that C = Coker(Fe Fe1 )
is a finite S-module flat over R. Hence we can apply the induction hypothesis to
the complex 0 C Fe2 . . . F0 . We deduce that C Fe2 is injective
and the exactness of the complex follows, as well as the flatness of the cokernel of
F1 F0 .

In the rest of this section we prove two versions of what is called the local criterion
of flatness. Note also the interesting Lemma 127.1 below.

00MJ

Lemma 98.6. Let R be a local ring with maximal ideal m and residue field =
R/m. Let M be an R-module. If TorR
1 (, M ) = 0, then for every finite length
(N,
M
)
=
0.
R-module N we have TorR
1
Proof. By descending induction on the length of N . If the length of N is 1, then
N
= and we are done. If the length of N is more than 1, then we can fit N
into a short exact sequence 0 N 0 N N 00 0 where N 0 , N 00 are finite
length R-modules of smaller length. The vanishing of TorR
1 (N, M ) follows from the
R
00
0
vanishing of TorR
1 (N , M ) and Tor1 (N , M ) (induction hypothesis) and the long
exact sequence of Tor groups, see Lemma 74.2.


00MK

Lemma 98.7 (Local criterion for flatness). Let R S be a local homomorphism

of local Noetherian rings. Let m be the maximal ideal of R, and let = R/m. Let
M be a finite S-module. If TorR
1 (, M ) = 0, then M is flat over R.
Proof. Let I R be an ideal. By Lemma 38.5 it suffices to show that I R M M
is injective. By Remark 74.9 we see that this kernel is equal to TorR
1 (M, R/I). By
Lemma 98.6 we see that J R M M is injective for all ideals of finite colength.
Choose n >> 0 and consider the following short exact sequence
0 I mn I mn I + mn 0
This is a sub sequence of the short exact sequence 0 R R2 R 0. Thus
we get the diagram
(I mn ) R M

/ I R M mn R M

/ (I + mn ) R M


M


/ M M


/M

Note that I + mn and mn are ideals of finite colength. Thus a diagram chase shows
that Ker((I mn ) R M M ) Ker(I R M M ) is surjective. We conclude in
particular that K = Ker(I R M M ) is contained in the image of (I mn ) R M
in I R M . By Artin-Rees, Lemma 50.2 we see that K is contained in mnc (I R M )
for some c > 0 and all n >> 0. Since I R M is a finite S-module (!) and since S

COMMUTATIVE ALGEBRA

217

is Noetherian, we see that this implies K = 0. Namely, the above implies K maps
to zero in the mS-adic completion of I R M . But the map from S to its mS-adic
completion is faithfully flat by Lemma 96.3. Hence K = 0, as desired.

In the following we often encounter the conditions M/IM is flat over R/I and
TorR
1 (R/I, M ) = 0. The following lemma gives some consequences of these conditions (it is a generalization of Lemma 98.6).
051C

Lemma 98.8. Let R be a ring. Let I R be an ideal. Let M be an R-module. If

M/IM is flat over R/I and TorR
1 (R/I, M ) = 0 then
(1) M/I n M is flat over R/I n for all n 1, and
(2) for any module N which is annihilated by I m for some m 0 we have
TorR
1 (N, M ) = 0.
In particular, if I is nilpotent, then M is flat over R.
Proof. Assume M/IM is flat over R/I and TorR
1 (R/I, M ) = 0. Let N be an
R/I-module. Choose a short exact sequence
M
0K
R/I N 0
iI

By the long exact sequence of Tor and the vanishing of TorR

1 (R/I, M ) we get
M
0 TorR
R/I) R M N R M 0
1 (N, M ) K R M (
iI
L
But since K, iI R/I, and N are all annihilated by I we see that
M
(

iI

K R M = K R/I M/IM,
M
R/I) R M = (
R/I) R/I M/IM,
iI

N R M = N R/I M/IM.
As M/IM is flat over R/I we conclude that
M
0 K R/I M/IM (
R/I) R/I M/IM N R/ M/IM 0
iI

is exact. Combining this with the above we conclude that TorR

1 (N, M ) = 0 for any
R-module N annihilated by I.
In particular, if we apply this to the module I/I 2 , then we conclude that the
sequence
0 I 2 R M I R M I/I 2 R M 0
is short exact. This implies that I 2 R M M is injective and it implies that
I/I 2 R/I M/IM = IM/I 2 M .
Let us prove that M/I 2 M is flat over R/I 2 . Let I 2 J be an ideal. We have to
show that J/I 2 R/I 2 M/I 2 M M/I 2 M is injective, see Lemma 38.5. As M/IM
is flat over R/I we know that the map (I +J)/I R/I M/IM M/IM is injective.
The sequence
(I J)/I 2 R/I 2 M/I 2 M J/I 2 R/I 2 M/I 2 M (I + J)/I R/I M/IM 0
is exact, as you get it by tensoring the exact sequence 0 (I J) J
(I + J)/I 0 by M/I 2 M . Hence suffices to prove the injectivity of the map (I
J)/I 2 R/I M/IM IM/I 2 M . However, the map (I J)/I 2 I/I 2 is injective
and as M/IM is flat over R/I the map (I J)/I 2 R/I M/IM I/I 2 R/I M/IM

COMMUTATIVE ALGEBRA

218

is injective. Since we have previously seen that I/I 2 R/I M/IM = IM/I 2 M we
obtain the desired injectivity.
Hence we have proven that the assumptions imply: (a) TorR
1 (N, M ) = 0 for all N
annihilated by I, (b) I 2 R M M is injective, and (c) M/I 2 M is flat over R/I 2 .
Thus we can continue by induction to get the same results for I n for all n 1. 
0AS8

Lemma 98.9. Let R be a ring. Let I R be an ideal. Let M be an R-module.

(1) If M/IM is flat over R/I and M R I/I 2 IM/I 2 M is injective, then
M/I 2 M is flat over R/I 2 .
(2) If M/IM is flat over R/I and M R I n /I n+1 I n M/I n+1 M is injective
for n = 1, . . . , k, then M/I k+1 M is flat over R/I k+1 .
Proof. The first statement is a consequence of Lemma 98.8 applied with R replaced
by R/I 2 and M replaced by M/I 2 M using that
R/I 2

Tor1

(M/I 2 M, R/I) = Ker(M R I/I 2 IM/I 2 M ),

see Remark 74.9. The second statement follows in the same manner using induction
on n to show that M/I n+1 M is flat over R/I n+1 for n = 1, . . . , k. Here we use that
R/I n+1

Tor1
for every n.
00ML

(M/I n+1 M, R/I) = Ker(M R I n /I n+1 I n M/I n+1 M )



Lemma 98.10 (Variant of the local criterion). Let R S be a local homomorphism of Noetherian local rings. Let I 6= R be an ideal in R. Let M be a finite
S-module. If TorR
1 (M, R/I) = 0 and M/IM is flat over R/I, then M is flat over
R.
Proof. First proof: By Lemma 98.8 we see that TorR
1 (, M ) is zero where is the
residue field of R. Hence we see that M is flat over R by Lemma 98.7.
Second proof: Let m be the maximal ideal of R. We will show
P that m R M M
is injective,
and
then
apply
Lemma
98.7.
Suppose
that
fi xi m R M
P
and that
fi xi = 0 in M . By the equational criterion for flatness Lemma 38.11
applied to M/IM
P over R/I we seePthere exist aij R/I and y j M/IM such that
xi mod IM = j aij y j and 0 = i (fi mod I)aij . Let aij R be a lift of aij and
similarly let yj M be a lift of y j . Then we see that
X
X
X
X
fi xi =
fi xi +
fi aij yj
fi aij yj
X
X
XX
=
fi (xi
aij yj ) +
(
fi aij ) yj
P
P
Since xi aij yj IM and
fi aij IPwe see that there exists an element in
I R M which maps to our given element
fi xi in m R M . But I R M M
is injective by assumption (see Remark 74.9) and we win.

In particular, in the situation of Lemma 98.10, suppose that I = (x) is generated
by a single element x which is a nonzerodivisor in R. Then TorR
1 (M, R/(x)) = (0)
if and only if x is a nonzerodivisor on M .

0523

Lemma 98.11. Let R S be a ring map. Let I R be an ideal. Let M be an

S-module. Assume
(1) R is a Noetherian ring,

COMMUTATIVE ALGEBRA

219

(2) S is a Noetherian ring,

(3) M is a finite S-module, and
(4) for each n 1 the module M/I n M is flat over R/I n .
Then for every q V (IS) the localization Mq is flat over R. In particular, if S is
local and IS is contained in its maximal ideal, then M is flat over R.
Proof. We are going to use Lemma 98.10. By assumption M/IM is flat over
R/I. Hence it suffices to check that TorR
1 (M, R/I) is zero on localization at q. By
Remark 74.9 this Tor group is equal to K = Ker(I R M M ). We know for each
n 1 that the kernel Ker(I/I n R/I n M/I n M M/I n M ) is zero. Since there
is a module map I/I n R/I n M/I n M (I R M )/I n1 (I R M ) we conclude
that K I n1 (I R M ) for each n. By the Artin-Rees lemma, and more precisely
Lemma 50.5 we conclude that Kq = 0, as desired.

00MM

Lemma 98.12. Let R R0 R00 be ring maps. Let M be an R-module.

0
00
Suppose that M R R0 is flat over R0 . Then the natural map TorR
1 (M, R )R0 R
R
00
Tor1 (M, R ) is onto.
Proof. Let F be a free resolution of M over R. The complex F2 R R0 F1 R
0
00
00
R0 F0 R R0 computes TorR
1 (M, R ). The complex F2 R R F1 R R
R
F0 R R00 computes Tor1 (M, R00 ). Note that Fi R R0 R0 R00 = Fi R R00 . Let
K 0 = Ker(F1 R R0 F0 R R0 ) and similarly K 00 = Ker(F1 R R00 F0 R R00 ).
Thus we have an exact sequence
0 K 0 F1 R R0 F0 R R0 M R R0 0.
By the assumption that M R R0 is flat over R0 , the sequence 0 K 0 R0 R00
F1 R R00 F0 R R00 M R R00 0 is still exact. This means that K 00 =
R
0
0
00
K 0 R0 R00 . Since TorR
1 (M, R ) is a quotient of K and Tor1 (M, R ) is a quotient
00
of K we win.


00MN

Lemma 98.13. Let R R0 be a ring map. Let I R be an ideal and I 0 = IR0 .

0
0
Let M be an R-module and set M 0 = M R R0 . The natural map TorR
1 (R /I , M )
R0
0
0
0
Tor1 (R /I , M ) is surjective.
Proof. Let F2 F1 F0 M 0 be a free resolution of M over R. Set
Fi0 = Fi R R0 . The sequence F20 F10 F00 M 0 0 may no longer be exact
at F10 . A free resolution of M 0 over R0 therefore looks like
F20 F200 F10 F00 M 0 0
for a suitable free module F200 over R0 . Next, note that Fi R R0 /I 0 = Fi0 /IFi0 =
0
0
Fi0 /I 0 Fi0 . So the complex F20 /I 0 F20 F10 /I 0 F10 F00 /I 0 F00 computes TorR
1 (M, R /I ).
0
0
0
0
0
0
00
On the other hand Fi R0 R /I = Fi /I Fi and similarly for F2 . Thus the complex
0
0
0
0
F20 /I 0 F20 F200 /I 0 F200 F10 /I 0 F10 F00 /I 0 F00 computes TorR
1 (M , R /I ). Since the
vertical map on complexes
F20 /I 0 F20

/ F10 /I 0 F10

/ F00 /I 0 F00


F20 /I 0 F20 F200 /I 0 F200


/ F10 /I 0 F10


/ F00 /I 0 F00

clearly induces a surjection on cohomology we win.

COMMUTATIVE ALGEBRA

00MO

220

Lemma 98.14. Let

SO

/ S0
O

/ R0
R
be a commutative diagram of local homomorphisms of local Noetherian rings. Let
I R be a proper ideal. Let M be a finite S-module. Denote I 0 = IR0 and
M 0 = M S S 0 . Assume that
(1) S 0 is a localization of the tensor product S R R0 ,
(2) M/IM is flat over R/I,
R0
0
0
0
(3) TorR
1 (M, R/I) Tor1 (M , R /I ) is zero.
Then M 0 is flat over R0 .
Proof. Since S 0 is a localization of S R R0 we see that M 0 is a localization of
M R R0 . Note that by Lemma 38.7 the module M/IM R/I R0 /I 0 = M R
R0 /I 0 (M R R0 ) is flat over R0 /I 0 . Hence also M 0 /I 0 M 0 is flat over R0 /I 0 as the
localization of a flat module is flat. By Lemma 98.10 it suffices to show that
0
0
0
0
0
0
TorR
1 (M , R /I ) is zero. Since M is a localization of M R R , the last assumption
0
R
0
0
0
0
implies that it suffices to show that Tor1 (M, R/I) R R TorR
1 (M R R , R /I )
is surjective.
0

R
0
0
0
0
0
By Lemma 98.13 we see that TorR
1 (M, R /I ) Tor1 (M R R , R /I ) is surjective.
R
R
0
0
0
So now it suffices to show that Tor1 (M, R/I)R R Tor1 (M, R /I ) is surjective.
This follows from Lemma 98.12 by looking at the ring maps R R/I R0 /I 0 and
the module M .


Please compare the lemma below to Lemma 100.8 (the case of a nilpotent ideal)
and Lemma 127.8 (the case of finitely presented algebras).
00MP

Lemma 98.15 (Crit`ere de platitude par fibres; Noetherian case). Let R, S, S 0

be Noetherian local rings and let R S S 0 be local ring homomorphisms. Let
m R be the maximal ideal. Let M be an S 0 -module. Assume
(1) The module M is finite over S 0 .
(2) The module M is not zero.
(3) The module M/mM is a flat S/mS-module.
(4) The module M is a flat R-module.
Then S is flat over R and M is a flat S-module.
Proof. Set I = mS S. Then we see that M/IM is a flat S/I-module because of
(3). Since m R S 0 I S S 0 is surjective we see that also m R M I S M is
surjective. Consider
m R M I S M M.
As M is flat over R the composition is injective and so both arrows are injective.
In particular TorS1 (S/I, M ) = 0 see Remark 74.9. By Lemma 98.10 we conclude
that M is flat over S. Note that since M/mS 0 M is not zero by Nakayamas Lemma
19.1 we see that actually M is faithfully flat over S by Lemma 38.15 (since it forces
M/mS M 6= 0).
Consider the exact sequence 0 m R 0. This gives an exact sequence
0 TorR
1 (, S) m R S I 0. Since M is flat over S this gives an exact

COMMUTATIVE ALGEBRA

221

sequence 0 TorR
1 (, S) S M m R M I S M 0. By the above this
implies that TorR
(,
S) S M = 0. Since M is faithfully flat over S this implies
1
that TorR
(,
S)
=
0
and
we conclude that S is flat over R by Lemma 98.7.

1
99. Base change and flatness
051D
00MQ

Some lemmas which deal with what happens with flatness when doing a base
change.
Lemma 99.1. Let
SO

/ S0
O

/ R0
R
be a commutative diagram of local homomorphisms of local rings. Assume that S 0
is a localization of the tensor product S R R0 . Let M be an S-module and set
M 0 = S 0 S M .
(1) If M is flat over R then M 0 is flat over R0 .
(2) If M 0 is flat over R0 and R R0 is flat then M is flat over R.
In particular we have
(3) If S is flat over R then S 0 is flat over R0 .
(4) If R0 S 0 and R R0 are flat then S is flat over R.
Proof. Proof of (1). If M is flat over R, then M R R0 is flat over R0 by Lemma
38.7. If W S R R0 is the multiplicative subset such that W 1 (S R R0 ) = S 0
then M 0 = W 1 (M R R0 ). Hence M 0 is flat over R0 as the localization of a flat
module, see Lemma 38.19 part (5). This proves (1) and in particular, we see that
(3) holds.
Proof of (2). Suppose that M 0 is flat over R0 and R R0 is flat. By (3) applied to
the diagram reflected in the northwest diagonal we see that S S 0 is flat. Thus
S S 0 is faithfully flat by Lemma 38.17. We are going to use the criterion of
Lemma 38.5 (3) to show that M is flat. Let I R be an ideal. If I R M M
has a kernel, so does (I R M ) S S 0 M S S 0 = M 0 . Note that I R R0 = IR0
as R R0 is flat, and that
(I R M ) S S 0 = (I R R0 ) R0 (M S S 0 ) = IR0 R0 M 0 .
From flatness of M 0 over R0 we conclude that this maps injectively into M 0 . This
concludes the proof of (2), and hence (4) is true as well.

100. Flatness criteria over Artinian rings
051E

We discuss some flatness criteria for modules over Artinian rings. Note that an
Artinian local ring has a nilpotent maximal ideal so that the following two lemmas
apply to Artinian local rings.

051F

Lemma 100.1. Let (R, m) be a local ring with nilpotent maximal ideal m. Let M
be a flat R-module. If A is a set and x M , A is a collection of elements of
M , then the following are equivalent:
(1) {x }A forms a basis for the vector space M/mM over R/m, and
(2) {x }A forms a basis for M over R.

COMMUTATIVE ALGEBRA

222

Proof. The implication (2) (1) is immediate. We will prove the other implication by using induction on n to show that {x }A forms a basis for M/mn M over
R/mn . The case n = 1 holds by assumption (1). Assume the statement holds for
some n 1. By Nakayamas Lemma 19.1 the elements x generate M , in particular
M/mn+1 M . The exact sequence 0 mn /mn+1 R/mn+1 R/mn 0 gives on
tensoring with M the exact sequence
0 mn M/mn+1 M M/mn+1 M M/mn M 0
Here we are using that M is flat. Moreover, we have mn M/mn+1 M = M/mM R/m
P
mn /mn+1 by flatness of M again. Now suppose that
f x = 0 in M/mn+1 M .
n
Then by induction hypothesisPf m for each . By the short exact sequence
f x is zero in mn /mn+1 R/m M/mM . Since
above we then conclude that
x forms a basis we conclude that each of the congruence classes f mn /mn+1 is
zero and we win.

051G

Lemma 100.2. Let R be a local ring with nilpotent maximal ideal. Let M be an
R-module. The following are equivalent
(1) M is flat over R,
(2) M is a free R-module, and
(3) M is a projective R-module.
Proof. Since any projective module is flat (as a direct summand of a free module)
and every free module is projective, it suffices to prove that a flat module is free.
Let M be a flat module. Let A be a set and let x M , A be elements such
that x M/mM forms a basis over the residue field of R. By Lemma 100.1 the
x are a basis for M over R and we win.


051H

Lemma 100.3. Let R be a ring. Let I R be an ideal. Let M be an R-module.

Let A be a set and let x M , A be a collection of elements of M . Assume
(1) I is nilpotent,
(2) {x }A forms a basis for M/IM over R/I, and
(3) TorR
1 (R/I, M ) = 0.
Then M is free on {x }A over R.
Proof. Let R, I, M , {x }A be as in the lemma and satisfy assumptions (1), (2),
and (3). By Nakayamas Lemma 19.1 the elements x generate M over R. The
assumption TorR
1 (R/I, M ) = 0 implies that we have a short exact sequence
0 I R M M M/IM 0.
P

Let
f x = P
0 be a relation in M . By choice of x we see that f I. Hence we
conclude that
f x = 0 in I R M . The map I R M I/I 2 R/I M/IM
and the fact that {x }A forms a basis for M/IM implies that f I 2 ! Hence
we conclude that there are no relations among the images of the x in M/I 2 M . In
other words, we see that M/I 2 M is free with basis the images of the x . Using the
map I R M I/I 3 R/I 2 M/I 2 M we then conclude that f I 3 ! And so on.
Since I n = 0 for some n by assumption (1) we win.

051I

Lemma 100.4. Let : R R0 be a ring map. Let I R be an ideal. Let M be

an R-module. Assume
(1) M/IM is flat over R/I, and

COMMUTATIVE ALGEBRA

223

(2) R0 R M is flat over R0 .

Set I2 = 1 ((I 2 )R0 ). Then M/I2 M is flat over R/I2 .
Proof. We may replace R, M , and R0 by R/I2 , M/I2 M , and R0 /(I)2 R0 . Then
I 2 = 0 and is injective. By Lemma 98.8 and the fact that I 2 = 0 it suffices to
0
0
prove that TorR
1 (R/I, M ) = K = Ker(I R M M ) is zero. Set M = M R R
0
0
0
0
0
and I = IR . By assumption the map I R0 M M is injective. Hence K maps
to zero in
I 0 R0 M 0 = I 0 R M = I 0 R/I M/IM.
Then I I 0 is an injective map of R/I-modules. Since M/IM is flat over R/I the
map
I R/I M/IM I 0 R/I M/IM
is injective. This implies that K is zero in I R M = I R/I M/IM as desired. 
051J

Lemma 100.5. Let : R R0 be a ring map. Let I R be an ideal. Let M be

an R-module. Assume
(1)
(2)
(3)
(4)

I is nilpotent,
R R0 is injective,
M/IM is flat over R/I, and
R0 R M is flat over R0 .

Then M is flat over R.

Proof. Define inductively I1 = I and In+1 = 1 ((In )2 R0 ) for n 1. Note that
by Lemma 100.4 we find that M/In M is flat over R/In for each n 1. It is clear
n
that (In ) (I)2 R0 . Since I is nilpotent we see that (In ) = 0 for some n. As
is injective we conclude that In = 0 for some n and we win.

Here is the local Artinian version of the local criterion for flatness.
051K

Lemma 100.6. Let R be an Artinian local ring. Let M be an R-module. Let

I R be a proper ideal. The following are equivalent
(1) M is flat over R, and
(2) M/IM is flat over R/I and TorR
1 (R/I, M ) = 0.
Proof. The implication (1) (2) follows immediately from the definitions. Assume M/IM is flat over R/I and TorR
1 (R/I, M ) = 0. By Lemma 100.2 this implies
that M/IM is free over R/I. Pick a set A and elements x M such that the
images in M/IM form a basis. By Lemma 100.3 we conclude that M is free and
in particular flat.

It turns out that flatness descends along injective homomorphism whose source is
an Artinian ring.

051L

Lemma 100.7. Let R S be a ring map. Let M be an R-module. Assume

(1) R is Artinian
(2) R S is injective, and
(3) M R S is a flat S-module.
Then M is a flat R-module.

COMMUTATIVE ALGEBRA

224

Proof. First proof: Let I R be the radical of R. Then I is nilpotent and M/IM
is flat over R/I as R/I is a product of fields, see Section 52. Hence M is flat by an
application of Lemma 100.5.
Q
Second proof: By Lemma 52.6 we may write R = Ri as a finite product of local
Artinian rings. This induces similar product decompositions for both R and S.
Hence we reduce to the case where R is local Artinian (details omitted).
Assume that R S, M are as in the lemma satisfying (1), (2), and (3) and in
addition that R is local with maximal ideal m. Let A be a set and x A be
elements such that x forms a basis for M/mM over R/m. By Nakayamas Lemma
19.1 we see that the elements x generate M as an R-module. Set N = S R M
and I = mS. Then {1 x }A is a family of elements of N which form a basis
for N/IN . Moreover, since N is flat over S we have TorS1 (S/I, N ) = 0. Thus we
conclude from Lemma 100.3 that N is free on {1 x }A . The injectivity of
R S then guarantees that there cannot be a nontrivial relation among the x
with coefficients in R.

Please compare the lemma below to Lemma 98.15 (the case of Noetherian local
rings) and Lemma 127.8 (the case of finitely presented algebras).
06A5

Lemma 100.8 (Crit`ere de platitude par fibres: Nilpotent case). Let

/ S0
>

S_
R

be a commutative diagram in the category of rings. Let I R be a nilpotent ideal

and M an S 0 -module. Assume
(1) The module M/IM is a flat S/IS-module.
(2) The module M is a flat R-module.
Then M is a flat S-module and Sq is flat over R for every q S such that M S (q)
is nonzero.
Proof. As M is flat over R tensoring with the short exact sequence 0 I R
R/I 0 gives a short exact sequence
0 I R M M M/IM 0.
Note that I R M IS S M is surjective. Combined with the above this means
both maps in
I R M IS S M M
are injective. Hence TorS1 (IS, M ) = 0 (see Remark 74.9) and we conclude that M
is a flat S-module by Lemma 98.8. To finish we need to show that Sq is flat over
R for any prime q S such that M S (q) is nonzero. This follows from Lemma
38.15 and 38.10.

101. What makes a complex exact?
00MR

Some of this material can be found in the paper [BE73] by Buchsbaum and Eisenbud.

COMMUTATIVE ALGEBRA

00MS

225

Situation 101.1. Here R is a ring, and we have a complex

e

e1

i+1

i1

0 Rne Rne1 . . . Rni Rni1 . . . Rn0

In other words we require i i+1 = 0 for i = 1, . . . , e 1.
00MT

Lemma 101.2. In Situation 101.1. Suppose R is a local ring with maximal ideal
m. Suppose that for some i, e i 1 some matrix coefficient of the map i is
invertible. Then the complex 0 Rne Rne1 . . . Rn0 is isomorphic to the
direct sum of a complex 0 Rne . . . Rni 1 Rni1 1 . . . Rn0 and
the complex 0 0 . . . R R 0 . . . 0 where the map R R is the
identity map.
Proof. The assumption means, after a change of basis of Rni and Rni1 that the
first basis vector of Rni is mapped via i to the first basis vector of Rni1 . Let
ej denote P
the jth basis vector of Rni and fk the kth basis vector of Rni1 . Write
i (ej ) =
ajk fk . So a1k = 0 unless k = 1 and a11 = 1. Change basis on Rni
again by setting e0j = ej aj1 e1 for j > 1. After this change of coordinates we have
aj1 = 0 for j > 1. Note the image of Rni+1 Rni is contained in the subspace
spanned by ej , j > 1. Note also that Rni1 Rni2 has to annihilate f1 since it is
in the image. These conditions and the shape of the matrix (ajk ) for i imply the
lemma.

Let us say that an acyclic complex of the form . . . 0 R R 0 . . . is
trivial. The lemma above clearly says that any finite complex of finite free modules
over a local ring is up to direct sums with trivial complexes the same as a complex
all of whose maps have all matrix coefficients in the maximal ideal.

00MU

Lemma 101.3. In Situation 101.1. Let R be a Artinian local ring. Suppose

that 0 Rne Rne1 . . . Rn0 is an exact complex. Then the complex is
isomorphic to a direct sum of trivial complexes.
P
Proof. By induction on the integer
ni . Clearly Ass(R) = {m}. Pick x R,
x 6= 0, mx = 0. Pick a basis vector ei Rne . Since xei is not mapped to zero
by exactness of the complex we deduce that some matrix coefficient
P of the map
Rne Rne1 is not in m. Lemma 101.2 then allows us to decrease
ni .

Below we define the rank of a map of finite free modules. This is just one possible
definition of rank. It is just the definition that works in this section; there are
others that may be more convenient in other settings.

00MV

Definition 101.4. Let R be a ring. Suppose that : Rm Rn is a map of finite

free modules.
(1) The rank of is the maximal r such that r : r Rm r Rn is nonzero.
(2) We let I() R be the ideal generated by the r r minors of the matrix
of , where r is the rank as defined above.

00MW

Lemma 101.5. In Situation 101.1, suppose the complex is isomorphic to a direct

sum of trivial complexes. Then we have
(1) the maps i have rank ri = ni ni+1 + . . . + (1)ei1 ne1 + (1)ei ne ,
(2) for all i, 1 i e we have rank(i+1 ) + rank(i ) = ni ,
(3) each I(i ) = R.

COMMUTATIVE ALGEBRA

226

Proof. We may assume the complex is the direct sum of trivial complexes. Then
for each i we can split the standard basis elements of Rni into those that map to a
basis element of Rni1 and those that are mapped to zero (and these are mapped
onto by basis elements of Rni+1 ). Using descending induction starting with i = e
it is easy to prove that there are ri+1 -basis elements of Rni which are mapped to
zero and ri which are mapped to basis elements of Rni1 . From this the result
follows.

00MX

Lemma 101.6. Let R be a local Noetherian ring. Suppose that : Rm Rn is

a map of finite free modules. The following are equivalent
(1) is injective.
(2) the rank of is m and either I() = R or it contains a nonzerodivisor.
Proof. If any matrix coefficient of is not in m, then we apply Lemma 101.2 to
write as the sum of 1 : R R and a map 0 : Rm1 Rn1 . It is easy to
see that the lemma for 0 implies the lemma for . Thus we may assume from the
outset that all the matrix coefficients of are in m.
Suppose is injective. We may assume m > 0. Let q Ass(R). Let x R be an
element whose annihilator is q. Note that induces a injective map xRm xRn
which is isomorphic to the map q : (R/q)m (R/q)n induced by . Since R/q is
a domain we deduce immediately by localizing to its fraction field that the rank of
q is m and that I(q ) is not the zero ideal. Hence we conclude by Lemma 62.18.
Conversely, assume that the rank of is m and that I() contains a nonzerodivisor
x. The rank being m implies n m. By Lemma 14.4 we can find a map : Rn
Rm such that = xidRm . Thus is injective.


00MY

Lemma 101.7. In Situation 101.1. Suppose R is a local Noetherian ring with

maximal ideal m. Assume m Ass(R), in other words R has depth 0. Suppose
that the complex is exact. In this case the complex is isomorphic to a direct sum of
trivial complexes.
Proof. The proof is the same as in Lemma 101.3, except using Lemma 101.6 to
guarantee that I(e ) = R, and hence some matrix coefficient of e is not in m. 

00MZ

Lemma 101.8. In Situation 101.1, suppose R is a local Noetherian ring, and

suppose that the complex is exact. Let x be an element of the maximal ideal which
is a nonzerodivisor. The complex 0 (R/xR)ne . . . (R/xR)n1 is still exact.
Proof. Follows easily from the snake lemma.

00N0

Lemma 101.9 (Acyclicity lemma). Let R be a local Noetherian ring. Let 0

Me Me1 . . . M0 be a complex of finite R-modules. Assume depth(Mi ) i.
Let i be the largest index such that the complex is not exact at Mi . If i > 0 then
Ker(Mi Mi1 )/Im(Mi+1 Mi ) has depth 1.
Proof. Let H = Ker(Mi Mi1 )/Im(Mi+1 Mi ) be the cohomology group
in question. We may break the complex into short exact sequences 0 Me
Me1 Ke2 0, 0 Kj Mj Kj1 0, for i + 2 j e 2,
0 Ki+1 Mi+1 Bi 0, 0 Ki Mi Mi1 , and 0 Bi Ki
H 0. We proceed up through these complexes to prove the statements about
depths, repeatedly using Lemma 71.6. First of all, since depth(Me ) e, and
depth(Me1 ) e 1 we deduce that depth(Ke2 ) e 1. At this point the

COMMUTATIVE ALGEBRA

227

sequences 0 Kj Mj Kj1 0 for i + 2 j e 2 imply similarly that

depth(Kj1 ) j for i + 2 j e 2. The sequence 0 Ki+1 Mi+1 Bi 0
then shows that depth(Bi ) i + 1. The sequence 0 Ki Mi Mi1 shows
that depth(Ki ) 1 since Mi has depth i 1 by assumption. The sequence
0 Bi Ki H 0 then implies the result.

00N1

Proposition 101.10. In Situation 101.1, suppose R is a local Noetherian ring.

The complex is exact if and only if for all i, 1 i e the following two conditions
are satisfied:
(1) we have rank(i+1 ) + rank(i ) = ni , and
(2) I(i ) = R, or I(i ) contains a regular sequence of length i.
Proof. This proof is very similar to the proof of Lemma 101.6. As in the proof of
Lemma 101.6 we may assume that all matrix entries of each i are elements of the
maximal ideal. We may also assume that e 1.
Assume the complex is exact. Let q Ass(R). (There is at least one such prime.)
Note that the ring Rq has depth 0. We apply Lemmas 101.7 and 101.5 to the
localized complex over Rq . All of the ideals I(i )q , e i 1 are equal to Rq . Thus
none of the ideals I(i ) is contained in q. This implies that I(e )I(e1 ) . . . I(1 )
is not contained in any of the associated primes of R. By Lemma 14.2 we may
choose x I(e )I(e1 ) . . . I(1 ), x 6 q for all q Ass(R). According to Lemma
101.8 the complex 0 (R/xR)ne . . . (R/xR)n1 is exact. By induction on e
all the ideals I(i )/xR have a regular sequence of length i 1. This proves that
I(i ) contains a regular sequence of length i.
Assume the two conditions on the ranks of i and the ideals I(i ) is satisfied. Note
that I(i ) m for all i because of what was said in the first paragraph of the proof.
Hence the assumption in particular implies that depth(R) e. By induction on
the dimension of R we may assume the complex is exact when localized at any
nonmaximal prime of R. Thus Ker(i )/Im(i+1 ) has support {m} and hence (if
nonzero) depth 0. By Lemma 101.9 we see that the complex is exact.

102. Cohen-Macaulay modules

00N2

Here we show that Cohen-Macaulay modules have good properties. We postpone

using Ext groups to establish the connection with duality and so on.

00N3

Definition 102.1. Let R be a Noetherian local ring. Let M be a finite R-module.

We say M is Cohen-Macaulay if dim(Support(M )) = depth(M ).
We start with an innocuous observation.

0AAD

Lemma 102.2. Let R S be a surjective homomorphism of Noetherian local

rings. Let N be a finite S-module. Then N is Cohen-Macaulay as an S-module if
and only if N is Cohen-Macaulay as an R-module.
Proof. Omitted.

Let R be a local Noetherian ring. Let M be a Cohen-Macaulay module, and let

f1 , . . . , fd be an M -regular sequence with d = dim(Support(M )). We say that
g m is good with respect to (M, f1 , . . . , fd ) if for all i = 0, 1, . . . , d 1 we have
dim(Support(M ) V (g, f1 , . . . , fi )) = d i 1. This is equivalent to the condition
that dim(Support(M/(f1 , . . . , fi )M ) V (g)) = d i 1 for i = 0, 1, . . . , d 1.

COMMUTATIVE ALGEBRA

00N4

228

Lemma 102.3. Notation and assumptions as above. If g is good with respect to

(M, f1 , . . . , fd ), then (a) g is a nonzerodivisor on M , and (b) M/gM is CohenMacaulay with maximal regular sequence f1 , . . . , fd1 .
Proof. We prove the lemma by induction on d. If d = 0, then M is finite and
there is no case to which the lemma applies. If d = 1, then we have to show that
g : M M is injective. The kernel K has support {m} because by assumption
dim Supp(M ) V (g) = 0. Hence K has finite length. Hence f1 : K K injective
implies the length of the image is the length of K, and hence f1 K = K, which
by Nakayamas Lemma 19.1 implies K = 0. Also, dim Supp(M/gM ) = 0 and so
M/gM is Cohen-Macaulay of depth 0.
For d > 1 we essentially argue in the same way. Let K M be the kernel of
multiplication by g. As above f1 : K K cannot be surjective if K 6= 0 Consider
the commutative diagram
f1

M
g

f1

M
g

M/f1 M
g

M/f1 M

This shows that the kernel K1 of g : M/f1 M M/f1 M cannot be zero if K is not
zero. But g is good for (M/f1 M, f2 , . . . , fd ), as is easy seen from the definition.
We conclude that K1 = 0, and so K = 0. From the snake lemma we see that
0 M/gM M/gM M/(f1 , g)M 0 is exact. By induction, we have that
M/(g, f1 )M is Cohen-Macaulay with regular sequence f2 , . . . , fd1 . Thus M/gM
is Cohen-Macaulay with regular sequence f1 , . . . , fd1 .

00N5

Lemma 102.4. Let R be a Noetherian local ring. Let M be a Cohen-Macaulay

module over R. Suppose g m is such that dim(Supp(M )V (g)) = dim(Supp(M ))
1. Then (a) g is a nonzerodivisor on M , and (b) M/gM is Cohen-Macaulay of depth
one less.
Proof. Choose a M -regular sequence f1 , . . . , fd with d = dim(Supp(M )). If g is
is good with respect to (M, f1 , . . . , fd ) we win by Lemma 102.3. In particular the
lemma holds if d = 1. (The case d = 0 does not occur.) Assume d > 1. Choose
an element h R such that (a) h is good with respect to (M, f1 , . . . , fd ), and (b)
dim(Supp(M )V (h, g) = d2. To see h exists, let {qi } be the (finite) set of minimal
primes of the closed sets Supp(M ), Supp(M ) V (f1 , . . . , fi ), i = 1, . . . , d 1, and
Supp(M ) V (g). None of these qi is equal to m and hence we may find h m,
h 6 qi by Lemma 14.2. It is clear that h satisfies (a) and (b). At this point we may
apply Lemma 102.3 to conclude that M/hM is Cohen-Macaulay. By (b) we see
that the pair (M/hM, g) satisfies the induction hypothesis. Hence M/(h, g)M is
Cohen-Macaulay, and g : M/hM M/hM is injective. From this it follows easily
that g : M M is injective, by a snake lemma argument. This in its turn implies
that h : M/gM M/gM is injective. Combined with the fact that M/(g, h)M is
Cohen-Macaulay this finishes the proof.


00N6

Proposition 102.5. Let R be a Noetherian local ring, with maximal ideal m. Let
M be a Cohen-Macaulay module over R whose support has dimension d. Suppose
that g1 , . . . , gc are elements of m such that dim(Supp(M/(g1 , . . . , gc )M )) = d c.
Then g1 , . . . , gc is an M -regular sequence, and can be extended to a maximal M regular sequence.

COMMUTATIVE ALGEBRA

229

Proof. Let Z = Supp(M ) Spec(R). By Lemma 59.12 in the chain Z Z

V (g1 ) . . . Z V (g1 , . . . , gc ) each step decreases the dimension at most by 1.
Hence by assumption each step decreases the dimension by exactly 1 each time.
Thus we may successively apply Lemma 102.4 to the modules M/(g1 , . . . , gi ) and
the element gi+1 .
To extend g1 , . . . , gc by one element if c < d we simply choose an element gc+1 m
which is not in any of the finitely many minimal primes of Z V (g1 , . . . , gc ), using
Lemma 14.2.

00NF

Definition 102.6. Let R be a Noetherian local ring. A finite module M over R

is called a maximal Cohen-Macaulay module if depth(M ) = dim(R).
In other words, a maximal Cohen-Macaulay module over a Noetherian local ring
is a finite module with the largest possible depth over that ring. Equivalently,
a maximal Cohen-Macaulay module over a Noetherian local ring R is a CohenMacaulay module of dimension equal to the dimension of the ring. In particular, if
M is a Cohen-Macaulay R-module with Spec(R) = Supp(M ), then M is maximal
Cohen-Macaulay. Thus the following two lemmas are on maximal Cohen-Macaulay
modules.

0AAE

Lemma 102.7. Let R be a Noetherian local ring. Assume there exists a CohenMacaulay module M with Spec(R) = Supp(M ). Then any maximal chain of ideals
p0 p1 . . . pn has length n = dim(R).
Proof. We will prove this by induction on dim(R). If dim(R) = 0, then the
statement is clear. Assume dim(R) > 0. Then n > 0. Choose an element x p1 ,
with x not in any of the minimal primes of R, and in particular x 6 p0 . (See
Lemma 14.2.) Then dim(R/xR) = dim(R) 1 by Lemma 59.12. The module
M/xM is Cohen-Macaulay over R/xR by Proposition 102.5 and Lemma 102.2.
The support of M/xM is Spec(R/xR) by Lemma 39.8. By induction the chain
p1 /xR . . . pn /xR in R/xR has length dim(R/xR) = dim(R) 1.


0AAF

Lemma 102.8. Suppose R is a Noetherian local ring. Assume there exists a

Cohen-Macaulay module M with Spec(R) = Supp(M ). Then for a prime p R we
have
dim(R) = dim(Rp ) + dim(R/p).
Proof. Follows immediately from Lemma 102.7.

0AAG

Lemma 102.9. Suppose R is a Noetherian local ring. Let M be a Cohen-Macaulay

module over R. For any prime p R the module Mp is Cohen-Macaulay over Rp .
Proof. Choose a maximal chain of primes p = pc pc1 . . . p1 m. If we
prove the result for Mp1 over Rp1 , then the lemma will follow by induction on c.
Thus we may assume that there is no prime strictly between p and m.
If Mp = 0, then the lemma holds. Assume Mp 6= 0. We have dim(Supp(Mp ))
dim(Supp(M )) 1 as a chain of primes in the support of Mp is a chain a primes in
the support of M not including m. Thus it suffices to show that the depth of Mp
is at least the depth of M minus 1. We will prove by induction on the depth of M
that there exists an M -regular sequence f1 , . . . , fdepth(M )1 in p. This will prove
the lemma since localization at p is exact. Since depth(M ) = dim((Supp(M ))

COMMUTATIVE ALGEBRA

230

dim(Supp(Mp )) + 1 1 we see that the base case happens when the depth of M is
1 and this case is trivial. Assume the depth of M is at least 2.
Let I R be the annihilator of M such that Spec(R/I) = V (I) = Supp(M )
(Lemma 39.5). By Lemmas 102.2 and 102.7 every maximal chain of primes in V (I)
has length 2. Hence none of the minimal primes of V (I) are equal to p. Thus we
can use Lemma 14.2 to find a f1 p which is not contained in any of the minimal
primes of V (I). Then f1 is a nonzerodivisor on M and M/f1 M has depth exactly
one less by Lemma 102.4. By induction we can extend to an M -regular sequence
f1 , . . . , fr p with r = depth(M ) 1 as desired.

0AAH

Definition 102.10. Let R be a Noetherian ring. Let M be a finite R-module.

We say M is Cohen-Macaulay if Mp is a Cohen-Macaulay module over Rp for all
primes p of R.
By Lemma 102.9 it suffices to check this in the maximal ideals of R.

0AAI

Lemma 102.11. Let R be a Noetherian ring. Let M be a Cohen-Macaulay module

over R. Then M R R[x1 , . . . , xn ] is a Cohen-Macaulay module over R[x1 , . . . , xn ].
Proof. By induction on the number of variables it suffices to prove this for M [x] =
M R R[x] over R[x]. Let m R[x] be a maximal ideal, and let p = R m.
Let f1 , . . . , fd be a Mp -regular sequence in the maximal ideal of Rp of length d =
dim(Supp(Mp )). Note that since R[x] is flat over R the localization R[x]m is flat over
Rp . Hence, by Lemma 67.5, the sequence f1 , . . . , fd is a M [x]m -regular sequence of
length d in R[x]m . The quotient
Q = M [x]m /(f1 , . . . , fd )M [x]m = Mp /(f1 , . . . , fd )Mp Rp R[x]m
has support equal to the primes lying over p because Rp R[x]m is flat and the
support of Mp /(f1 , . . . , fd )Mp is equal to {p} (details omitted; hint: follows from
Lemmas 39.4 and 39.5). Hence the dimension is 1. To finish the proof it suffices
to find a single Q-regular element f m. Since m is a maximal ideal, the field
extension (p) (m) is finite (Theorem 33.1). Hence we can find f m which
viewed as a polynomial in x has leading coefficient not in p. Such an f acts as a
nonzerodivisor on
M
Mp /(f1 , . . . , fd )Mp R R[x] =
Mp /(f1 , . . . , fd )Mp xn
n0

and hence acts as a nonzerodivisor on Q.

103. Cohen-Macaulay rings

00N7

Most of the results of this section are special cases of the results in Section 102.

00N8

Definition 103.1. A Noetherian local ring R is called Cohen-Macaulay if it is

Cohen-Macaulay as a module over itself.
Note that this is equivalent to requiring the existence of a R-regular sequence
x1 , . . . , xd of the maximal ideal such that R/(x1 , . . . , xd ) has dimension 0. We will
usually just say regular sequence and not R-regular sequence.

02JN

Lemma 103.2. Let R be a Noetherian local Cohen-Macaulay ring with maximal

ideal m. Let x1 , . . . , xc m be elements. Then
x1 , . . . , xc is a regular sequence dim(R/(x1 , . . . , xc )) = dim(R) c

COMMUTATIVE ALGEBRA

231

If so x1 , . . . , xc can be extended to a regular sequence of length dim(R) and each

quotient R/(x1 , . . . , xi ) is a Cohen-Macaulay ring of dimension dim(R) i.
Proof. Special case of Proposition 102.5.
00N9

Lemma 103.3. Let R be Noetherian local. Suppose R is Cohen-Macaulay of

dimension d. Any maximal chain of ideals p0 p1 . . . pn has length n = d.
Proof. Special case of Lemma 102.7.

00NA

Lemma 103.4. Suppose R is a Noetherian local Cohen-Macaulay ring of dimension d. For any prime p R we have
dim(R) = dim(Rp ) + dim(R/p).
Proof. Follows immediately from Lemma 103.3. (Also, this is a special case of
Lemma 102.8.)


00NB

Lemma 103.5. Suppose R is a Cohen-Macaulay local ring. For any prime p R

the ring Rp is Cohen-Macaulay as well.
Proof. Special case of Lemma 102.9.

00NC

Definition 103.6. A Noetherian ring R is called Cohen-Macaulay if all its local

rings are Cohen-Macaulay.

00ND

Lemma 103.7. Suppose R is a Cohen-Macaulay ring. Any polynomial algebra

over R is Cohen-Macaulay.
Proof. Special case of Lemma 102.11.

00NE

Lemma 103.8. Let R be a Noetherian local Cohen-Macaulay ring of dimension

d. Let 0 K Rn M 0 be an exact sequence of R-modules. Then either
depth(K) > depth(M ) or depth(K) = depth(M ) = d.
Proof. If depth(M ) = 0 the lemma is clear. Let x m be a nonzerodivisor on M
and on R. Then x is a nonzerodivisor on M and on K and it follows by an easy
diagram chase that 0 K/xK (R/xR)n M/xM 0 is exact. Thus the
result follows from the result for K/xK over R/xR which has smaller dimension. 

00NG

Lemma 103.9. Let R be a local Noetherian Cohen-Macaulay ring of dimension

d. Let M be a finite R module of depth e. There exists an exact complex
0 K Fde1 . . . F0 M 0
with each Fi finite free and K maximal Cohen-Macaulay.
Proof. Immediate from the definition and Lemma 103.8.

06LC

Lemma 103.10. Let : A B be a map of local rings. Assume that B is

p
Noetherian and Cohen-Macaulay and that mB = (mA )B. Then there exists
a sequence of elements f1 , . . . , fdim(B) in A such that (f1 ), . . . , (fdim(B) ) is a
regular sequence in B.
Proof. By induction on dim(B) it suffices to prove: If dim(B) 1, then we can
find an element f of A which maps to a nonzerodivisor in B. By Lemma 103.2 it
suffices to find f A whose image in B is not contained in any ofp
the finitely many
minimal primes q1 , . . . , qr of B. By the assumption that mB = (mA )B we see
that mA 6 1 (qi ). Hence we can find f by Lemma 14.2.


COMMUTATIVE ALGEBRA

232

104. Catenary rings

00NH
00NI

Definition 104.1. A ring R is said to be catenary if for any pair of prime ideals
p q, all maximal chains of primes p = p0 p1 . . . pe = q have the same
(finite) length.

02IH

Lemma 104.2. A ring R is catenary if and only if the topological space Spec(R)
is catenary (see Topology, Definition 10.4).
Proof. Immediate from the definition and the characterization of irreducible closed
subsets in Lemma 25.1.

In general it is not the case that a finitely generated R-algebra is catenary if R is.
Thus we make the following definition.

00NL

Definition 104.3. A ring R is said to be universally catenary if R is Noetherian

and every R algebra of finite type is catenary.
By Lemma 104.6 this just means that R is Noetherian and that each polynomial
algebra R[x1 , . . . , xn ] is catenary.

00NJ

Lemma 104.4. Any localization of a (universally) catenary ring is (universally)

catenary.
Proof. Let A be a ring and let S A be a multiplicative subset. The description
of Spec(S 1 A) in Lemma 16.5 shows that if A is catenary, then so is S 1 A. If
S 1 A C is of finite type, then C = S 1 B for some finite type ring map A B.
Hence if A is universally catenary, then B is catenary and we see that C is catenary
too. Combined wtih Lemma 30.1 this proves the lemma.


0AUN

Lemma 104.5. Let R be a ring. The following are equivalent

(1) R is catenary,
(2) Rp is catenary for all prime ideals p,
(3) Rm is catenary for all maximal ideals m.
Assume R is Noetherian. The following are equivalent
(1) R is universally catenary,
(2) Rp is universally catenary for all prime ideals p,
(3) Rm is universally catenary for all maximal ideals m.
Proof. The implication (1) (2) follows from Lemma 104.4 in both cases. The
implication (2) (3) is immediate in both cases. Assume Rm is catenary for all
maximal ideals m of R. If p q are primes in R, then choose a maximal ideal
q m. Chains of primes ideals between p and q are in 1-to-1 correspondence with
chains of prime ideals between pRm and qRm hence we see R is catenary. Assume
R is Noetherian and Rm is universally catenary for all maximal ideals m of R. Let
R S be a finite type ring map. Let q be a prime ideal of S lying over the prime
p R. Choose a maximal ideal p m in R. Then Rp is a localization of Rm hence
universally catenary by Lemma 104.4. Then Sp is catenary as a finite type ring
over Rp . Hence Sq is catenary as a localization. Thus S is catenary by the first
case treated above.


00NK

Lemma 104.6.
catenary.

Any quotient of a (universally) catenary ring is (universally)

COMMUTATIVE ALGEBRA

233

Proof. Let A be a ring and let I A be an ideal. The description of Spec(A/I)

in Lemma 16.7 shows that if A is catenary, then so is A/I. If A/I B is of finite
type, then A B is of finite type. Hence if A is universally catenary, then B is
catenary. Combined wtih Lemma 30.1 this proves the lemma.

0AUP

Lemma 104.7. Let R be a Noetherian ring.

(1) R is catenary if and only if R/p is catenary for every minimal prime p.
(2) R is universally catenary if and only if R/p is universally catenary for every
minimal prime p.
Proof. If a b is an inclusion of primes of R, then we can find a minimal prime
p a and the first assertion is clear. We omit the proof of the second.


00NM

Lemma 104.8. A Cohen-Macaulay ring is universally catenary. More generally,

if R is a Noetherian ring and M is a Cohen-Macaulay R-module with Supp(M ) =
Spec(R), then R is universally catenary.
Proof. Since a polynomial algebra over R is Cohen-Macaulay, by Lemma 103.7, it
suffices to show that a Cohen-Macaulay ring is catenary. Let R be Cohen-Macaulay
and p q primes of R. By definition Rq and Rp are Cohen-Macaulay. Take a
maximal chain of primes p = p0 p1 . . . pn = q. Next choose a maximal chain
of primes q0 q1 . . . qm = p. By Lemma 103.3 we have n + m = dim(Rq ).
And we have m = dim(Rp ) by the same lemma. Hence n = dim(Rq ) dim(Rp ) is
independent of choices.
To prove the more general statement, argue exactly as above but using Lemmas
102.11 and 102.7.

105. Regular local rings

00NN

It is not that easy to show that all prime localizations of a regular local ring are
regular. In fact, quite a bit of the material developed so far is geared towards a
proof of this fact. See Proposition 109.5, and trace back the references.

00NO

Lemma 105.1. Let R be a regular local ring with maximal ideal m. The graded
L n n+1
ring
m /m
is isomorphic to the graded polynomial algebra (m)[X1 , . . . , Xd ].
Proof. Let x1 , . . . , xd be a minimal set of generators for the maximal ideal m.
By Definition 59.9 this implies
dim(R) = d. Write = (m). There is a
L that
surjection [X1 , . . . , Xd ]
mn /mn+1 , which maps the class of xi in m/m2 to
Xi . Since d(R) = d by Proposition 59.8 we know that the numerical polynomial
n 7 dim mn /mn+1 has degree
1. By Lemma 57.10 we conclude that the
L n d n+1
surjection [X1 , . . . , Xd ]
m /m
is an isomorphism.


00NP

Lemma 105.2. Any regular local ring is a domain.

T
Proof. We will use that mn = 0 by Lemma 50.4. Let f, g R such that
f g = 0. Suppose that f ma and g mb , with a, b maximal. Since f g = 0
ma+b+1 we see from the result of Lemma 105.1 that either f ma+1 or g mb+1 .
Contradiction.


00NQ

Lemma 105.3. Let R be a regular local ring and let x1 , . . . , xd be a minimal set
of generators for the maximal ideal m. Then x1 , . . . , xd is a regular sequence, and
each R/(x1 , . . . , xc ) is a regular local ring of dimension d c. In particular R is
Cohen-Macaulay.

COMMUTATIVE ALGEBRA

234

Proof. Note that R/x1 R is a Noetherian local ring of dimension d1 by Lemma

59.12 with x2 , . . . , xd generating the maximal ideal. Hence it is a regular local ring
by definition. Since R is a domain by Lemma 105.2 x1 is a nonzerodivisor.

00NR

Lemma 105.4. Let R be a regular local ring. Let I R be an ideal such that
R/I is a regular local ring as well. Then there exists a minimal set of generators
x1 , . . . , xd for the maximal ideal m of R such that I = (x1 , . . . , xc ) for some 0
c d.
Proof. Say dim(R) = d and dim(R/I) = d c. Denote m = m/I the maximal
ideal of R/I. Let = R/m. We have
dim ((I + m2 )/m2 ) = dim (m/m2 ) dim(m/m2 ) = d (d c) = c
by the definition of a regular local ring. Hence we can choose x1 , . . . , xc I whose
images in m/m2 are linearly independent and supplement with xc+1 , . . . , xd to get
a minimal system of generators of m. The induced map R/(x1 , . . . , xc ) R/I is
a surjection between regular local rings of the same dimension (Lemma 105.3). It
follows that the kernel is zero, i.e., I = (x1 , . . . , xc ). Namely, if not then we would
have dim(R/I) < dim(R/(x1 , . . . , xc )) by Lemmas 105.2 and 59.12.


00NS

Lemma 105.5. Let R be a Noetherian local ring. Let x m. Let M be a finite

R-module such that x is a nonzerodivisor on M and M/xM is free over R/xR.
Then M is free over R.
Proof. Let m1 , . . . , mr be elements of M which map to P
a R/xR-basis of M/xM .
By Nakayamas Lemma 19.1 m1 , . . . , mr generate M . If
ai mi = 0 is a relation,
then ai xR for all i. Hence ai = bi x for some bi R. Hence the kernel K of
Rr M satisfies xK = K and hence is zero by Nakayamas lemma.


00NT

Lemma 105.6. Let R be a regular local ring. Any maximal Cohen-Macaulay

module over R is free.
Proof. Let M be a maximal Cohen-Macaulay module over R. Let x m be part of
a regular sequence generating m. Then x is a nonzerodivisor on M by Proposition
102.5, and M/xM is a maximal Cohen-Macaulay module over R/xR. By induction
on dim(R) we see that M/xM is free. We win by Lemma 105.5.


00NU

Lemma 105.7. Suppose R is a Noetherian local ring. Let x m be a nonzerodivisor such that R/xR is a regular local ring. Then R is a regular local ring.
More generally, if x1 , . . . , xr is a regular sequence in R such that R/(x1 , . . . , xr ) is
a regular local ring, then R is a regular local ring.
Proof. This is true because x together with the lifts of a system of minimal generators of the maximal ideal of R/xR will give dim(R) generators of m. Use Lemma
59.12. The last statement follows from the first and induction.


07DX

Lemma 105.8. Let (Ri , ii0 ) be a directed system of local rings whose transition
maps are local ring maps. If each Ri is a regular local ring and R = colim Ri is
Noetherian, then R is a regular local ring.
Proof. Let m R be the maximal ideal; it is the colimit of the maximal ideal
mi Ri . We prove the lemma by induction on d = dim m/m2 . If d = 0, then
R = R/m is a field and R is a regular local ring. If d > 0 pick an x m, x 6 m2 . For
some i we can find an xi mi mapping to x. Note that R/xR = colimi0 i Ri0 /xi Ri0

COMMUTATIVE ALGEBRA

235

is a Noetherian local ring. By Lemma 105.3 we see that Ri0 /xi Ri0 is a regular local
ring. Hence by induction we see that R/xR is a regular local ring. Since each Ri is
a domain (Lemma 105.1) we see that R is a domain. Hence x is a nonzerodivisor
and we conclude that R is a regular local ring by Lemma 105.7.

106. Epimorphisms of rings
04VM

In any category there is a notion of an epimorphism. Some of this material is taken

from [Laz69] and [Maz68].

04VN

Lemma 106.1. Let R S be a ring map. The following are equivalent

(1) R S is an epimorphism,
(2) the two ring maps S S R S are equal,
(3) either of the ring maps S S R S is an isomorphism, and
(4) the ring map S R S S is an isomorphism.
Proof. Omitted.

04VP

Lemma 106.2. The composition of two epimorphisms of rings is an epimorphism.

Proof. Omitted. Hint: This is true in any category.

04VQ

Lemma 106.3. If R S is an epimorphism of rings and R R0 is any ring

map, then R0 R0 R S is an epimorphism.
Proof. Omitted. Hint: True in any category with pushouts.

04VR

Lemma 106.4. If A B C are ring maps and A C is an epimorphism, so

is B C.
Proof. Omitted. Hint: This is true in any category.

This means in particular, that if R S is an epimorphism with image R S, then

R S is an epimorphism. Hence while proving results for epimorphisms we may
often assume the map is injective. The following lemma means in particular that
every localization is an epimorphism.
04VS

Lemma 106.5. Let R S be a ring map. The following are equivalent:

(1) R S is an epimorphism, and
(2) Rp Sp is an epimorphism for each prime p of R.
Proof. Since Sp = Rp R S (see Lemma 11.15) we see that (1) implies (2) by
Lemma 106.3. Conversely, assume that (2) holds. Let a, b : S A be two ring
maps from S to a ring A equalizing the map R S. By assumption we see that
for every prime
Q p of R the induced maps ap , bp : Sp Ap are the same. Hence
a = b as A p Ap , see Lemma 23.1.


04VT

Lemma 106.6. Let R S be a ring map. The following are equivalent

(1) R S is an epimorphism and finite, and
(2) R S is surjective.
Proof. (This lemma seems to have been reproved many times in the literature, and
has many different proofs.) It is clear that a surjective ring map is an epimorphism.
Suppose that R S is a finite ring map such that S R S S is an isomorphism.

COMMUTATIVE ALGEBRA

236

Our goal is to show that R S is surjective. Assume S/R is not zero. The exact
sequence R S S/R 0 leads to an exact sequence
R R S S R S S/R R S 0.
Our assumption implies that the first arrow is an isomorphism, hence we conclude
that S/R R S = 0. Hence also S/R R S/R = 0. By Lemma 5.4 there exists a
surjection of R-modules S/R R/I for some proper ideal I R. Hence there
exists a surjection S/R R S/R R/I R R/I = R/I 6= 0, contradiction.

04VU

Lemma 106.7. A faithfully flat epimorphism is an isomorphism.

Proof. This is clear from Lemma 106.1 part (3) as the map S S R S is the
map R S tensored with S.


04VV

Lemma 106.8. If k S is an epimorphism and k is a field, then S = k or S = 0.

Proof. This is clear from the result of Lemma 106.7 (as any nonzero algebra over
k is faithfully flat), or by arguing directly that R R k R cannot be surjective
unless dimk (R) 1.


04VW

Lemma 106.9. Let R S be an epimorphism of rings. Then

(1) Spec(S) Spec(R) is injective, and
(2) for q S lying over p R we have (p) = (q).
Proof. Let p be a prime of R. The fibre of the map is the spectrum of the fibre
ring S R (p). By Lemma 106.3 the map (p) S R (p) is an epimorphism,
and hence by Lemma 106.8 we have either S R (p) = 0 or S R (p) = (p)
which proves (1) and (2).


04VX

Lemma 106.10. Let R be a ring. Let M , N be R-modules. Let {xi }iI be a set
of generators of M . Let {yj }jJ be a set of generators of N . Let {mj }jJ be a
family of elements of M with mj = 0 for all but finitely many j. Then
X
mj yj = 0 in M R N
jJ

is equivalent to the following: There exist ai,j R with ai,j = 0 for all but finitely
many pairs (i, j) such that
X
mj =
ai,j xi for all j J,
iI
X
0=
ai,j yj for all i I.
jJ
P
Proof. The sufficiency is immediate. Suppose that jJ mj yj = 0. Consider
the short exact sequence
M
0K
RN 0
jJ
L
where the jth basis vector of jJ R maps to yj . Tensor this with M to get the
exact sequence
M
K R M
M N R M 0.
jJ
P
The assumption implies that there exist elements ki K such that
ki xi maps
to the element (mj )jJ of the middle. Writing ki = (ai,j )jJ and we obtain what
we want.


COMMUTATIVE ALGEBRA

04VY

237

Lemma 106.11. Let : R S be a ring map. Let g S. The following are

equivalent:
(1) g 1 = 1 g in S R S, and
(2) there exist n 0 and elements yi , zj S and xi,j R for 1 i, j n such
that
P
(a) g = i,jn xi,j yi zj ,
P
(b) for each j we have P xi,j yi (R), and
(c) for each i we have
xi,j zj (R).
Proof. It is clear that (2) implies (1). Conversely, suppose that g 1 = 1 g.
Choose generators {si }iI of S as an R-module with 0, 1 I and s0 = 1 and s1 = g.
Apply Lemma 106.10 to the relation
g s0 +P
(1) s1 = 0. We see that there
P
existPai,j R such that g = i ai,0 si , 1 =P i ai,1 si , and for j 6= 0, 1 we have
0 = i ai,j si , and moreover for all i we have j ai,j sj = 0. Then we have
X
ai,j si sj = g + a0,0
i,j6=0
P
and for each j 6= 0 we have i6=0 ai,j si R. This proves that g + a0,0 can be
written as in (2). It follows that g can be written as in (2). Details omitted. Hint:
Show that the set of elements of S which have an expression as in (2) form an
R-subalgebra of S.


04VZ

Remark 106.12. Let R S be a ring map. Sometimes the set of elements

g S such that g 1 = 1 g is called the epicenter of S. It is an R-algebra.
By the construction of Lemma 106.11 we get for each g in the epicenter a matrix
factorization
(g) = Y XZ
with X Mat(n n, R), Y Mat(1 n, S), and Z Mat(n 1, S). Namely,
let xi,j , yi , zj be as in part (2) of the lemma. Set X = (xi,j ), let y be the row
vector whose entries are the yi and let z be the column vector whose entries are the
zj . With this notation conditions (b) and (c) of Lemma 106.11 mean exactly that
Y X Mat(1 n, R), XZ Mat(n 1, R). It turns out to be very convenient to
consider the triple of matrices (X, Y X, XZ). Given n N and a triple (P, U, V ) we
say that (P, U, V ) is a n-triple associated to g if there exists a matrix factorization
as above such that P = X, U = Y X and V = XZ.

04W0

Lemma 106.13. Let R S be an epimorphism of rings. Then the cardinality of

S is at most the cardinality of R. In a formula: |S| |R|.
Proof. The condition that R S is an epimorphism means that each g S
satisfies g1 = 1g, see Lemma 106.1. We are going to use the notation introduced
in Remark 106.12. Suppose that g, g 0 S and suppose that (P, U, V ) is an n-triple
which is associated to both g and g 0 . Then we claim that g = g 0 . Namely, write
(P, U, V ) = (X, Y X, XZ) for a matrix factorization (g) = Y XZ of g and write
(P, U, V ) = (X 0 , Y 0 X 0 , X 0 Z 0 ) for a matrix factorization (g 0 ) = Y 0 X 0 Z 0 of g 0 . Then
we see that
(g) = Y XZ = U Z = Y 0 X 0 Z = Y 0 P Z = Y 0 XZ = Y 0 V = Y 0 X 0 Z 0 = (g 0 )
and hence g = g 0 . This implies that the cardinality of S is bounded by the number
of possible triples, which has cardinality at most supnN |R|n . If R is infinite then
this is at most |R|, see [Kun83, Ch. I, 10.13].

COMMUTATIVE ALGEBRA

238

If R is a finite ring then the argument above only proves that S is at worst countable.
In fact in this case R is Artinian and the map R S is surjective. We omit the
proof of this case.

08YS

Lemma 106.14. Let R S be an epimorphism of rings. Let N1 , N2 be Smodules. Then HomS (N1 , N2 ) = HomR (N1 , N2 ). In other words, the restriction
functor ModS ModR is fully faithful.
Proof. Let : N1 N2 be an R-linear map. For any x N1 consider the map
S R S N2 defined by the rule g g 0 7 g(g 0 x). Since both maps S S R S
are isomorphisms (Lemma 106.1), we conclude that g(g 0 x) = gg 0 (x) = (gg 0 x).
Thus is S-linear.

107. Pure ideals

04PQ

The material in this section is discussed in many papers, see for example [Laz67],
[Bko70], and [DM83].

04PR

Definition 107.1. Let R be a ring. We say that I R is pure if the quotient

ring R/I is flat over R.

04PS

Lemma 107.2. Let R be a ring. Let I R be an ideal. The following are

equivalent:
(1) I is pure,
(2) for every ideal J R we have J I = IJ,
(3) for every finitely generated ideal J R we have J I = JI,
(4) for every x R we have (x) I = xI,
(5) for every x I we have x = yx for some y I,
(6) for every x1 , . . . , xn I there exists a y I such that xi = yxi for all
i = 1, . . . , n,
(7) for every prime p of R we have IRp = 0 or IRp = Rp ,
(8) Supp(I) = Spec(R) \ V (I),
(9) I is the kernel of the map R (1 + I)1 R,
(10) R/I
= S 1 R as R-algebras for some multiplicative subset S of R, and
(11) R/I
= (1 + I)1 R as R-algebras.
Proof. For any ideal J of R we have the short exact sequence 0 J R
R/J 0. Tensoring with R/I we get an exact sequence J R R/I R/I
R/I + J 0 and J R R/I = R/JI. Thus the equivalence of (1), (2), and (3)
follows from Lemma 38.5. Moreover, these imply (4).
The implication (4) (5) is trivial. Assume (5) and let x1 , . . . , xn Q
I. Choose yi
I such that xi = yi xi . Let y I be the element such that 1 y = i=1,...,n (1 yi ).
Then xi = yxi for all i = 1, . . . , n. Hence (6) holds, and it follows that (5) (6).
Assume (5). Let x I. Then x = yx for some y I. Hence x(1 y) = 0,
which shows that x maps to zero in (1 + I)1 R. Of course the kernel of the map
R (1+I)1 R is always contained in I. Hence we see that (5) implies (9). Assume
(9). Then for any x I we see that x(1 y) = 0 for some y I. In other words,
x = yx. We conclude that (5) is equivalent to (9).
Assume (5). Let p be a prime of R. If p 6 V (I), then IRp = Rp . If p V (I), in
other words, if I p, then x I implies x(1 y) = 0 for some y I, implies x
maps to zero in Rp , i.e., IRp = 0. Thus we see that (7) holds.

COMMUTATIVE ALGEBRA

239

Assume (7). Then (R/I)p is either 0 or Rp for any prime p of R. Hence by Lemma
38.19 we see that (1) holds. At this point we see that all of (1) (7) and (9) are
equivalent.
As IRp = Ip we see that (7) implies (8). Finally, if (8) holds, then this means
exactly that Ip is the zero module if and only if p V (I), which is clearly saying
that (7) holds. Now (1) (9) are equivalent.
Assume (1) (9) hold. Then R/I (1 + I)1 R by (9) and the map R/I
(1 + I)1 R is also surjective by the description of localizations at primes afforded
by (7). Hence (11) holds.
The implication (11) (10) is trivial. And (10) implies that (1) holds because a
localization of R is flat over R, see Lemma 38.19.

04PT

Lemma 107.3. Let R be a ring. If I, J R are pure ideals, then V (I) = V (J)
implies I = J.
Proof.
For example, by property (7) of Lemma 107.2 we see that I = Ker(R
Q

pV (I) Rp ) can be recovered from the closed subset associated to it.

04PU

Lemma 107.4. Let R be a ring. The rule I 7 V (I) determines a bijection

{I R pure} {Z Spec(R) closed and closed under generalizations}
Proof. Let I be a pure ideal. Then since R R/I is flat, by going up generalizations lift along the map Spec(R/I) Spec(R). Hence V (I) is closed under
generalizations. This shows that the map is well defined. By Lemma 107.3 the map
is injective. Suppose that Z Spec(R) is closed and closed under generalizations.
Let J R be the radical ideal such that Z = V (J). Let I = {x R : x xJ}.
Note that I is an ideal. We claim that I is pure and that V (I) = V (J). If the
claim is true then the map of the lemma is surjective and the lemma holds.
Note that I J, so that V (J) V (I). Let I p be a prime. Consider the
multiplicative subset S = (R \ p)(1 + J). By definition of I and I p we see that
0 6 S. Hence we can find a prime q of R which is disjoint from S, see Lemmas 9.4
and 16.5. Hence q p and q (1 + J) = . This implies that q + J is a proper
ideal of R. Let m be a maximal ideal containing q + J. Then we get m V (J) and
hence q V (J) = Z as Z was assumed to be closed under generalization. This in
turn implies p V (J) as q p. Thus we see that V (I) = V (J).

Finally, since V (I) = V (J) (and J radical) we see that J = I. Pick x I, so

that x = xy for some y J by definition. Then x = xy = xy 2 = . . . = xy n . Since
y n I for some n > 0 we conclude that property (5) of Lemma 107.2 holds and we
see that I is indeed pure.


05KK

Lemma 107.5.
equivalent
(1)
(2)
(3)
(4)

Let R be a ring. Let I R be an ideal. The following are

I is pure and finitely generated,

I is generated by an idempotent,
I is pure and V (I) is open, and
R/I is a projective R-module.

COMMUTATIVE ALGEBRA

240

Proof. If (1) holds, then I = I I = I 2 by Lemma 107.2. Hence I is generated

by an idempotent by Lemma 20.5. Thus (1) (2). If (2) holds, then I = (e) and
R = (1e)(e) as an R-module hence R/I is flat and I is pure and V (I) = D(1e)
is open. Thus (2) (1) + (3). Finally, assume (3). Then V (I) is open and
closed, hence V (I) = D(1 e) for some idempotent e of R, see Lemma 20.3. The
ideal J = (e) is a pure ideal such that V (J) = V (I) hence I = J by Lemma
107.3. In this way we see that (3) (2). By Lemma 77.2 we see that (4) is
equivalent to the assertion that I is pure and R/I finitely presented. Moreover,
R/I is finitely presented if and only if I is finitely generated, see Lemma 5.3.
Hence (4) is equivalent to (1).

We can use the above to characterize those rings for which every finite flat module
is finitely presented.
052U

Lemma 107.6. Let R be a ring. The following are equivalent:

(1) every Z Spec(R) which is closed and closed under generalizations is also
open, and
(2) any finite flat R-module is finite locally free.
Proof. If any finite flat R-module is finite locally free then the support of R/I
where I is a pure ideal is open. Hence the implication (2) (1) follows from
Lemma 107.3.
For the converse assume that R satisfies (1). Let M be a finite flat R-module.
The support Z = Supp(M ) of M is closed, see Lemma 39.5. On the other hand,
if p p0 , then by Lemma 77.4 the module Mp0 is free, and Mp = Mp0 Rp0 Rp
Hence p0 Supp(M ) p Supp(M ), in other words, the support is closed under
generalization. As R satisfies (1) we see that the support of M is open and closed.
Suppose that M is generated by r elements m1 , . . . , mr . The modules i (M ),
i = 1, . . . , r are finite flat R-modules also, because i (M )p = i (Mp ) is free over
Rp . Note that Supp(i+1 (M )) Supp(i (M )). Thus we see that there exists a
decomposition
Spec(R) = U0 q U1 q . . . q Ur
by open and closed subsets such that the support of i (M ) is Ur . . . Ui for all
i = 0, . . . , r. Let p be a prime of R, and say p Ui . Note that i (M ) R (p) =
i (M R (p)). Hence, after possibly renumbering m1 , . . . , mr we may assume that
m1 , . . . , mi generate M R (p). By Nakayamas Lemma 19.1 we get a surjection
X
Rfi Mf , (a1 , . . . , ai ) 7
ai mi
for some f R, f 6 p. We may also assume that D(f ) Ui . This means
that i (Mf ) = i (M )f is a flat Rf module whose support is all of Spec(Rf ).
By the above it is generated by a single element, namely m1 . . . mi . Hence
i (M )f
= Rf /J for some pure ideal J Rf with V (J) = Spec(Rf ). Clearly this
means that J = (0), see Lemma 107.3. Thus m1 . . . mi is a basis for i (Mf )
and it follows that the displayed map is injective as well as surjective. This proves
that M is finite locally free as desired.

108. Rings of finite global dimension

00O2

The following lemma is often used to compare different projective resolutions of a

given module.

COMMUTATIVE ALGEBRA

00O3

241

Lemma 108.1 (Schanuels lemma). Let R be a ring. Let M be an R-module.

Suppose that 0 K P1 M 0 and 0 L P2 M 0 are two short
exact sequences, with Pi projective. Then K P2
= L P1 .
Proof. Consider the module N defined by the short exact sequence 0 N
P1 P2 M 0, where the last map is the sum of the two maps Pi M . It is easy
to see that the projection N P1 is surjective with kernel L, and that N P2 is
surjective with kernel K. Since Pi are projective we have N
= K P2
= LP1 . 

00O4

Definition 108.2. Let R be a ring. Let M be an R-module. We say M has finite

projective dimension if it has a finite length resolution by projective R-modules.
The minimal length of such a resolution is called the projective dimension of M .
It is clear that the projective dimension of M is 0 if and only if M is a projective
module. The following lemma explains to what extent the projective dimension is
independent of the choice of a projective resolution.

00O5

Lemma 108.3. Let R be a ring. Suppose that M is an R-module of projective

dimension d. Suppose that Fe Fe1 . . . F0 M 0 is exact with Fi
projective and e d 1. Then the kernel of Fe Fe1 is projective (or the kernel
of F0 M is projective in case e = 0).
Proof. We prove this by induction on d. If d = 0, then M is projective. In this
case there is a splitting F0 = Ker(F0 M ) M , and hence Ker(F0 M ) is
projective. This finishes the proof if e = 0, and if e > 0, then replacing M by
Ker(F0 M ) we decrease e.
Next assume d > 0. Let 0 Pd Pd1 . . . P0 M 0 be a minimal
length finite resolution with Pi projective. According to Schanuels Lemma 108.1
we have P0 Ker(F0 M )
= F0 Ker(P0 M ). This proves the case d = 1,
e = 0, because then the right hand side is F0 P1 which is projective. Hence now
we may assume e > 0. The module F0 Ker(P0 M ) has the finite projective
resolution 0 Pd F0 Pd1 F0 . . . P1 F0 Ker(P0 M ) F0 0
of length d 1. By induction on d we see that the kernel of Fe P0 Fe1 P0
is projective. This implies the lemma.


065R

Lemma 108.4. Let R be a ring. Let M be an R-module. Let n 0. The following

are equivalent
(1) M has projective dimension n,
(2) ExtiR (M, N ) = 0 for all R-modules N and all i n + 1, and
(3) Extn+1
R (M, N ) = 0 for all R-modules N .
Proof. Assume (1). Choose a free resolution F M of M . Denote de : Fe
Fe1 . By Lemma 108.3 we see that Pe = Ker(de ) is projective for e n 1. This
implies that Fe
= Pe Pe1 for e n where de maps the summand Pe1 isomorphically to Pe1 in Fe1 . Hence, for any R-module N the complex HomR (F , N )
is split exact in degrees n + 1. Whence (2) holds. The implication (2) (3) is
trivial.
Assume (3) holds. If n = 0 then M is projective by Lemma 76.2 and we see that
(1) holds. If n > 0 choose a free R-module F and a surjection F M with kernel
K. By Lemma 70.7 and the vanishing of ExtiR (F, N ) for all i > 0 by part (1) we
see that ExtnR (K, N ) = 0 for all R-modules N . Hence by induction we see that K

COMMUTATIVE ALGEBRA

242

has projective dimension n 1. Then M has projective dimension n as any

finite projective resolution of K gives a projective resolution of length one more for
M by adding F to the front.

065S

Lemma 108.5. Let R be a ring. Let 0 M 0 M M 00 0 be a short exact

sequence of R-modules.
(1) If M has projective dimension n and M 00 has projective dimension
n + 1, then M 0 has projective dimension n.
(2) If M 0 and M 00 have projective dimension n then M has projective dimension n.
(3) If M 0 has projective dimension n and M has projective dimension n+1
then M 00 has projective dimension n + 1.
Proof. Combine the characterization of projective dimension in Lemma 108.4 with
the long exact sequence of ext groups in Lemma 70.7.


00O6

Definition 108.6. Let R be a ring. The ring R is said to have finite global
dimension if there exists an integer n such that every R-module has a resolution
by projective R-modules of length at most n. The minimal such n is then called
the global dimension of R.
The argument in the proof of the following lemma can be found in the paper [Aus55]
by Auslander.

065T

Lemma 108.7. Let R be a ring. The following are equivalent

(1) R has finite global dimension n,
(2) every finite R-module has projective dimension n, and
(3) every cyclic R-module R/I has projective dimension n.
Proof. It is clear that (1) (3). Assume (3). Since every finite R-module has a
finite filtration by cyclic modules, see Lemma 5.4 we see that (2) follows by Lemma
108.5.
Assume (2). Let M be an arbitrary R-module. Choose a set E M of generators
of M . Choose a well ordering on E. For e E denote M
Se the submodule of M
generated by the elements e0 E with e0 e. Then M = eE Me . Note that for
each e E the quotient
[
Me /
M e0
0
e <e

is either zero or generated by one element, hence has projective dimension n. To

finish S
the proof we claim that any time we have
S a well-ordered set E and a module
M = eE Me such that the quotients Me / e0 <e Me0 have projective dimension
n, then M has projective dimension n.
We may prove this statement by induction on n. If n = 0, then we will show, by
transfinite induction
for each e E we may choose
S that M is projective. Namely, S
0 Pe because Pe = Me /
0
a splitting Me = e0 <e
M
e
e0 <e Me is projective. Hence
L
it follows that M = eE Pe and we conclude that M is projective, see Lemma
76.3.
If n > 0, then for e E we denote Fe the free R-module on the set of elements of
Me . Then we have a system of short exact sequences
0 Ke Fe Me 0

COMMUTATIVE ALGEBRA

243

over the well-ordered set E.SNote that the S

transition maps Fe0 Fe and Ke0 Ke
are injective too. Set F = Fe and K = Ke . Then
[
[
[
0 Ke /
Ke0 Fe /
Fe 0 M e /
M e0 0
0
0
e <e
e <e
e0 <e
S
0 is the free R-module
is a short exact sequence of R-modules too and Fe / e0 <e FeS
on the set of elements in Me which are not contained in Se0 <e Me0 . Hence by
Lemma 108.5 we see that the projective dimension of Ke / e0 <e Ke0 is at most
n 1. By induction we conclude that K has projective dimension at most n 1.
Whence M has projective dimension at most n and we win.

00O8

Lemma 108.8. Let R be a ring. Let M be an R-module. Let S R be a

multiplicative subset.
(1) If M has projective dimension n, then S 1 M has projective dimension
n over S 1 R.
(2) If R has finite global dimension n, then S 1 R has finite global dimension
n.
Proof. Let 0 Pn Pn1 . . . P0 M 0 be a projective resolution. As
localization is exact, see Proposition 9.12, and as each S 1 Pi is a projective S 1 Rmodule, see Lemma 93.1, we see that 0 S 1 Pn . . . S 1 P0 S 1 M 0
is a projective resolution of S 1 M . This proves (1). Let M 0 be an S 1 R-module.
Note that M 0 = S 1 M 0 . Hence we see that (2) follows from (1).

109. Regular rings and global dimension

065U

We can use the material on rings of finite global dimension to give another characterization of regular local rings.

00O7

Proposition 109.1. Let R be a regular local ring of dimension d. Every finite

R-module M of depth e has a finite free resolution
0 Fde . . . F0 M 0.
In particular a regular local ring has global dimension d.
Proof. This is clear in view of Lemma 105.6 and Lemma 103.9.

00O9

Lemma 109.2. Let R be a Noetherian ring. Then R has finite global dimension
if and only if there exists an integer n such that for all maximal ideals m of R the
ring Rm has global dimension n.
Proof. We saw, Lemma 108.8 that if R has finite global dimension n, then all
the localizations Rm have finite global dimension at most n. Conversely, suppose
that all the Rm have global dimension n. Let M be a finite R-module. Let 0
Kn Fn1 . . . F0 M 0 be a resolution with Fi finite free. Then
Kn is a finite R-module. According to Lemma 108.3 and the assumption all the
modules Kn R Rm are projective. Hence by Lemma 77.2 the module Kn is finite
projective.


00OA

Lemma 109.3. Suppose that R is a Noetherian local ring with maximal ideal m
and residue field . In this case the projective dimension of is dim m/m2 .

COMMUTATIVE ALGEBRA

244

Proof. Let x1 , . . . xn be elements of m whose images in m/m2 form a basis. Consider the Koszul complex on x1 , . . . , xn . This is the complex
0 n Rn n1 Rn n2 Rn . . . i Rn . . . Rn R
with maps given by
ej1 . . . eji 7

i
X

(1)i+1 xja ej1 . . . eja . . . eji

a=1

It is easy to see that this is a complex K (R, x ). Note that the cokernel of the last
map of K (R, x ) is clearly .
Now, let F by any finite resolution by finite free R-modules. By Lemma 101.2
we may assume all the maps in the complex F have to property that Im(Fi
Fi1 ) mFi1 , because removing a trivial summand from the resolution can at
worst shorten the resolution. By Lemma 70.4 we can find a map of complexes
: K (R, x ) F inducing the identity on . We will prove by induction that
the maps i : i Rn = Ki (R, x ) Fi have the property that i : i n Fi
are injective. This will prove the lemma since it clearly shows that Fn 6= 0.

0
The result is clear for i = 0 because the composition R
F0 is nonzero.
Note that F0 must have rank 1 since otherwise the map F1 F0 whose cokernel
is a single copy of cannot have image contained in mF0 . For 1 we use that
x1 , . . . , xn is a minimal set of generators for m. Namely, we saw above that F0 = R
and F1 F0 = R has image m. We have a commutative diagram

Rn

K1 (R, x ) K0 (R, x ) = R

F1

F0
= R

where the rightmost vertical arrow is given by multiplication by a unit. Hence we

see that the image of the composition Rn F1 F0 = R is also equal to m. Thus
the map Rn F1 has to be injective since dim (m/m2 ) = n.
Suppose the injectivity of j has been proved for all j i 1. Consider the
commutative diagram
i Rn

Ki (R, x ) Ki1 (R, x )

Fi

Fi1

= i1 Rn

We know that i1 n Fi1 is injective. This proves that i1 n m/m2

Fi1 m/m2 is injective. Also, by our choice of the complex, Fi maps into mFi1 ,
and similarly for the Koszul complex. Hence we get a commutative diagram
i n

Fi

i1 n m/mn

Fi1 m/m2

At this point it suffices to verify the map i n i1 n m/mn is injective, which

can be done by hand.

00OB

Lemma 109.4. Let R be a Noetherian local ring. Suppose that the residue field
has finite projective dimension n over R. In this case dim(R) n.

COMMUTATIVE ALGEBRA

245

Proof. Let F be a finite resolution of by finite free R-modules. By Lemma 101.2

we may assume all the maps in the complex F have to property that Im(Fi
Fi1 ) mFi1 , because removing a trivial summand from the resolution can at
worst shorten the resolution. Say Fn 6= 0 and Fi = 0 for i > n, so that the projective
dimension of is n. By Proposition 101.10 we see that depthI(n ) (R) n since
I(n ) cannot equal R by our choice of the complex. Thus by Lemma 71.3 also
dim(R) n.

00OC

Proposition 109.5. A Noetherian local ring whose residue field has finite projective dimension is a regular local ring. In particular a Noetherian local ring of finite
global dimension is a regular local ring.
Proof. By Lemmas 109.3 and 109.4 we see that dim(R) dim (m/m2 ). Thus the
result follows immediately from Definition 59.9.


0AFS

Lemma 109.6. A Noetherian local ring R is a regular local ring if and only if it
has finite global dimension. In this case Rp is a regular local ring for all primes p.
Proof. By Propositions 109.5 and 109.1 we see that a Noetherian local ring is a
regular local ring if and only if it has finite global dimension. Furthermore, any
localization Rp has finite global dimension, see Lemma 108.8, and hence is a regular
local ring.

By Lemma 109.6 it makes sense to make the following definition, because it does
not conflict with the earlier definition of a regular local ring.

00OD

Definition 109.7. A Noetherian ring R is said to be regular if all the localizations

Rp at primes are regular local rings.
It is enough to require the local rings at maximal ideals to be regular. Note that
this is not the same as asking R to have finite global dimension, even assuming R
is Noetherian. This is because there is an example of a regular Noetherian ring
which does not have finite global dimension, namely because it does not have finite
dimension.

00OE

Lemma 109.8. Let R be a Noetherian ring. The following are equivalent:

(1) R has finite global dimension n,
(2) there exists an integer n such that all the localizations Rm at maximal ideals
are regular of dimension n with equality for at least one m, and
(3) there exists an integer n such that all the localizations Rp at prime ideals
are regular of dimension n with equality for at least one p.
Proof. This is a reformulation of Lemma 109.2 in view of the discussion surrounding Definition 109.7. See especially Propositions 109.1 and 109.5.


00OF

Lemma 109.9. Let R S be a local homomorphism of local Noetherian rings.

Assume that R S is flat and that S is regular. Then R is regular.
Proof. Let m R be the maximal ideal and let = R/m be the residue field. Let
d = dim S. Choose any resolution F with each Fi a finite free R-module. Set
Kd = Ker(Fd1 Fd2 ). By flatness of R S the complex 0 Kd R S
Fd1 R S . . . F0 R S R S 0 is still exact. Because the global
dimension of S is d, see Proposition 109.1, we see that Kd R S is a finite free
S-module (see also Lemma 108.3). By Lemma 77.5 we see that Kd is a finite free

COMMUTATIVE ALGEBRA

246

R-module. Hence has finite projective dimension and R is regular by Proposition

109.5.

110. Auslander-Buchsbaum
090U

The following result can be found in [AB57].

090V

Proposition 110.1. Let R be a Noetherian local ring. Let M be a nonzero finite

R-module which has finite projective dimension pdR (M ). Then we have
depth(R) = pdR (M ) + depth(M )
Proof. We prove this by induction on depth(M ). The most interesting case is the
case depth(M ) = 0. In this case, let
0 Rne Rne1 . . . Rn0 M 0
be a minimal finite free resolution, so e = pdR (M ). By Lemma 101.2 we may
assume all matrix coefficients of the maps in the complex are contained in the
maximal ideal of R. Then on the one hand, by Proposition 101.10 we see that
depth(R) e. On the other hand, breaking the long exact sequence into short
exact sequences
0 Rne Rne1 Ke2 0,
0 Ke2 Rne2 Ke3 0,
...,
0 K0 R

n0

M 0

we see, using Lemma 71.6, that

depth(Ke2 ) depth(R) 1,
depth(Ke3 ) depth(R) 2,
...,
depth(K0 ) depth(R) (e 1),
depth(M ) depth(R) e
and since depth(M ) = 0 we conclude depth(R) e. This finishes the proof of the
case depth(M ) = 0.
Induction step. If depth(M ) > 0, then we pick x m which is a nonzerodivisor on
both M and R. This is possible, because either pdR (M ) > 0 and depth(R) > 0 by
the aforementioned Proposition 101.10 or pdR (M ) = 0 in which case M is finite
free hence also depth(R) = depth(M ) > 0. Thus depth(R M ) > 0 by Lemma
71.6 (for example) and we can find an x m which is a nonzerodivisor on both R
and M . Let
0 Rne Rne1 . . . Rn0 M 0
be a minimal resolution as above. An application of the snake lemma shows that
0 (R/xR)ne (R/xR)ne1 . . . (R/xR)n0 M/xM 0
is a minimal resolution too. Thus pdR (M ) = pdR/xR (M/xM ). By Lemma 71.7 we
have depth(R/xR) = depth(R) 1 and depth(M/xM ) = depth(M ) 1. Till now
depths have all been depths as R modules, but we observe that depthR (M/xM ) =
depthR/xR (M/xM ) and similarly for R/xR. By induction hypothesis we see that

COMMUTATIVE ALGEBRA

247

the Auslander-Buchsbaum formula holds for M/xM over R/xR. Since the depths
of both R/xR and M/xM have decreased by one and the projective dimension has
not changed we conclude.

111. Homomorphisms and dimension
00OG

This section contains a collection of easy results relating dimensions of rings when
there are maps between them.

00OH

Lemma 111.1. Suppose R S is a ring map satisfying either going up, see Definition 40.1, or going down see Definition 40.1. Assume in addition that Spec(S)
Spec(R) is surjective. Then dim(R) dim(S).
Proof. Assume going up. Take any chain p0 p1 . . . pe of prime ideals in
R. By surjectivity we may choose a prime q0 mapping to p0 . By going up we may
extend this to a chain of length e of primes qi lying over pi . Thus dim(S) dim(R).
The case of going down is exactly the same. See also Topology, Lemma 18.8 for a
purely topological version.


00OI

Lemma 111.2. Suppose that R S is a ring map with the going up property,
see Definition 40.1. If q S is a maximal ideal. Then the inverse image of q in R
is a maximal ideal too.
Proof. Trivial.

00OJ

Lemma 111.3. Suppose that R S is a ring map such that S is integral over R.
Then dim(R) dim(S), and every closed point of Spec(S) maps to a closed point
of Spec(R).
Proof. Immediate from Lemmas 35.18 and 111.2 and the definitions.

00OK

Lemma 111.4. Suppose R S and S integral over R. Then dim(R) = dim(S).

Proof. This is a combination of Lemmas 35.20, 35.15, 111.1, and 111.3.

00OL

Definition 111.5. Suppose that R S is a ring map. Let q S be a prime

lying over the prime p of R. The local ring of the fibre at q is the local ring
Sq /pSq = (S/pS)q = (S R (p))q

00OM

Lemma 111.6. Let R S be a homomorphism of Noetherian rings. Let q S

be a prime lying over the prime p. Then
dim(Sq ) dim(Rp ) + dim(Sq /pSq ).
Proof. We use the characterization of dimension of Proposition 59.8. Let x1 , . . . , xd
be elements of p generating an ideal of definition of Rp with d = dim(Rp ). Let
y1 , . . . , ye be elements of q generating an ideal of definition of Sq /pSq with e =
dim(Sq /pSq ). It is clear that Sq /(x1 , . . . , xd , y1 , . . . , ye ) has a nilpotent maximal
ideal. Hence x1 , . . . , xd , y1 , . . . , ye generate an ideal of definition if Sq .


00ON

Lemma 111.7. Let R S be a homomorphism of Noetherian rings. Let q S be

a prime lying over the prime p. Assume the going down property holds for R S
(for example if R S is flat, see Lemma 38.18). Then
dim(Sq ) = dim(Rp ) + dim(Sq /pSq ).

COMMUTATIVE ALGEBRA

248

Proof. By Lemma 111.6 we have an inequality dim(Sq ) dim(Rp )+dim(Sq /pSq ).

To get equality, choose a chain of primes pS q0 q1 . . . qd = q with d =
dim(Sq /pSq ). On the other hand, choose a chain of primes p0 p1 . . . pe = p
with e = dim(Rp ). By the going down theorem we may choose q1 q0 lying over
pe1 . And then we may choose q2 qe1 lying over pe2 . Inductively we keep
going until we get a chain qe . . . qd of length e + d.

031E

Lemma 111.8. Let R S be a local homomorphism of local Noetherian rings.

Assume
(1) R is regular,
(2) S/mR S is regular, and
(3) R S is flat.
Then S is regular.
Proof. By Lemma 111.7 we have dim(S) = dim(R) + dim(S/mR S). Pick generators x1 , . . . , xd mR with d = dim(R), and pick y1 , . . . , ye mS which generate the maximal ideal of S/mR S with e = dim(S/mR S). Then we see that
x1 , . . . , xd , y1 , . . . , ye are elements which generate the maximal ideal of S and e+d =
dim(S).

The lemma below will later be used to show that rings of finite type over a field
are Cohen-Macaulay if and only if they are quasi-finite flat over a polynomial ring.
It is a partial converse to Lemma 127.1.

00R5

Lemma 111.9. Let R S be a local homomorphism of Noetherian local rings.

Assume R Cohen-Macaulay. If S is finite flat over R, or if S is flat over R and
dim(S) dim(R), then S is Cohen-Macaulay and dim(R) = dim(S).
Proof. Let x1 , . . . , xd mR be a regular sequence of length d = dim(R). By
Lemma 67.5 this maps to a regular sequence in S. Hence S is Cohen-Macaulay if
dim(S) d. This is true if S is finite flat over R by Lemma 111.4. And in the
second case we assumed it.

112. The dimension formula

02II

Recall the definitions of catenary (Definition 104.1) and universally catenary (Definition 104.3).

02IJ

Lemma 112.1. Let R S be a ring map. Let q be a prime of S lying over the
prime p of R. Assume that
(1) R is Noetherian,
(2) R S is of finite type,
(3) R, S are domains, and
(4) R S.
Then we have
height(q) height(p) + trdegR (S) trdeg(p) (q)
with equality if R is universally catenary.
Proof. Suppose that R S 0 S is a finitely generated R-subalgebra of S. In this
case set q0 = S 0 q. The lemma for the ring maps R S 0 and S 0 S implies the
lemma for R S by additivity of transcendence degree in towers of fields (Fields,

COMMUTATIVE ALGEBRA

249

Lemma 25.5). Hence we can use induction on the number of generators of S over
R and reduce to the case where S is generated by one element over R.
Case I: S = R[x] is a polynomial algebra over R. In this case we have trdegR (S) = 1.
Also R S is flat and hence
dim(Sq ) = dim(Rp ) + dim(Sq /pSq )
see Lemma 111.7. Let r = pS. Then trdeg(p) (q) = 1 is equivalent to q = r, and
implies that dim(Sq /pSq ) = 0. In the same vein trdeg(p) (q) = 0 is equivalent to
having a strict inclusion r q, which implies that dim(Sq /pSq ) = 1. Thus we are
done with case I with equality in every instance.
Case II: S = R[x]/n with n 6= 0. In this case we have trdegR (S) = 0. Denote
q0 R[x] the prime corresponding to q. Thus we have
Sq = (R[x])q0 /n(R[x])q0
By the previous case we have dim((R[x])q0 ) = dim(Rp ) + 1 trdeg(p) (q). Since
n 6= 0 we see that the dimension of Sq decreases by at least one, see Lemma 59.12,
which proves the inequality of the lemma. To see the equality in case R is universally
catenary note that n R[x] is a height one prime as it corresponds to a nonzero
prime in f.f.(R)[x]. Hence any maximal chain of primes in R[x]q0 /n corresponds
to a maximal chain of primes with length 1 greater between q0 and (0) in R[x]. If
R is universally catenary these all have the same length equal to the height of q0 .
This proves that dim(R[x]q0 /n) = dim(R[x]q0 ) 1 as desired.

The following lemma says that generically finite maps tend to be quasi-finite in
codimension 1.
02MA

Lemma 112.2. Let A B be a ring map. Assume

(1) A B is an extension of domains.
(2) A is Noetherian,
(3) A B is of finite type, and
(4) the extension f.f.(A) f.f.(B) is finite.
Let p A be a prime of height 1. Then there are at most finitely many primes of
B lying over p and they all have height 1.
Proof. By the dimension formula (Lemma 112.1) for any prime q lying over p we
have
dim(Bq ) dim(Ap ) trdeg(p) (q).
As the domain Bq has at least 2 prime ideals we see that dim(Bq ) 1. We conclude
that dim(Bq ) = 1 and that the extension (p) (q) is algebraic. Hence q defines
a closed point of its fibre Spec(B A (p)), see Lemma 34.9. Since B A (p) is
a Noetherian ring the fibre Spec(B A (p)) is a Noetherian topological space, see
Lemma 30.5. A Noetherian topological space consisting of closed points is finite,
see for example Topology, Lemma 8.2.

113. Dimension of finite type algebras over fields

00OO

In this section we compute the dimension of a polynomial ring over a field. We

also prove that the dimension of a finite type domain over a field is the dimension
of its local rings at maximal ideals. We will establish the connection with the
transcendence degree over the ground field in Section 115.

COMMUTATIVE ALGEBRA

00OP

250

Lemma 113.1. Let m be a maximal ideal in k[x1 , . . . , xn ]. The ideal m is generated

by n elements. The dimension of k[x1 , . . . , xn ]m is n. Hence k[x1 , . . . , xn ]m is a
regular local ring of dimension n.
Proof. By the Hilbert Nullstellensatz (Theorem 33.1) we know the residue field
= (m) is a finite extension of k. Denote i the image of xi . Denote
i = k(1 , . . . , i ) , i = 1, . . . , n and 0 = k. Note that i = k[1 , . . . , i ]
by field theory. Define inductively elements fi m k[x1 , . . . , xi ] as follows: Let
Pi (T ) i1 [T ] be the monic minimal polynomial of i over i1 . Let Qi (T )
k[x1 , . . . , xi1 ][T ] be a monic lift of Pi (T ) (of the same degree). Set fi = Qi (xi ).
Note that if di = degT (Pi ) = degT (Qi ) = degxi (fi ) then d1 d2 . . . di = [i : k] by
Fields, Lemmas 7.6 and 9.2.
We claim that for all i = 0, 1, . . . , n there is an isomorphism
i : k[x1 , . . . , xi ]/(f1 , . . . , fi )
= i .
By construction the composition k[x1 , . . . , xi ] k[x1 , . . . , xn ] is surjective
onto i and f1 , . . . , fi are in the kernel. This gives a surjective homomorphism.
We prove i is injective by induction. It is clear for i = 0. Given the statement for i we prove it for i + 1. The ring extension k[x1 , . . . , xi ]/(f1 , . . . , fi )
k[x1 , . . . , xi+1 ]/(f1 , . . . , fi+1 ) is generated by 1 element over a field and one irreducible equation. By elementary field theory k[x1 , . . . , xi+1 ]/(f1 , . . . , fi+1 ) is a field,
and hence i is injective.
This implies that m = (f1 , . . . , fn ). Moreover, we also conclude that
k[x1 , . . . , xn ]/(f1 , . . . , fi )
= i [xi+1 , . . . , xn ].
Hence (f1 , . . . , fi ) is a prime ideal. Thus
(0) (f1 ) (f1 , f2 ) . . . (f1 , . . . , fn ) = m
is a chain of primes of length n. The lemma follows.

00OQ

Proposition 113.2. A polynomial algebra in n variables over a field is a regular

ring. It has global dimension n. All localizations at maximal ideals are regular local
rings of dimension n.
Proof. By Lemma 113.1 all localizations k[x1 , . . . , xn ]m at maximal ideals are regular local rings of dimension n. Hence we conclude by Lemma 109.8.


00OR

Lemma 113.3. Let k be a field. Let p q k[x1 , . . . , xn ] be a pair of primes.

Any maximal chain of primes between p and q has length height(q) height(p).
Proof. By Proposition 113.2 any local ring of k[x1 , . . . , xn ] is regular. Hence all
local rings are Cohen-Macaulay, see Lemma 105.3. The local rings at maximal
ideals have dimension n hence every maximal chain of primes in k[x1 , . . . , xn ] has
length n, see Lemma 103.3. Hence every maximal chain of primes between (0) and
p has length height(p), see Lemma 103.4 for example. Putting these together leads
to the assertion of the lemma.


00OS

Lemma 113.4. Let k be a field. Let S be a finite type k-algebra which is an

integral domain. Then dim(S) = dim(Sm ) for any maximal ideal m of S. In words:
every maximal chain of primes has length equal to the dimension of S.

COMMUTATIVE ALGEBRA

251

Proof. Write S = k[x1 , . . . , xn ]/p. By Proposition 113.2 and Lemma 113.3 all the
maximal chains of primes in S (which necessarily end with a maximal ideal) have
length n height(p). Thus this number is the dimension of S and of Sm for any
maximal ideal m of S.

Recall that we defined the dimension dimx (X) of a topological space X at a point
x in Topology, Definition 9.1.
00OT

Lemma 113.5. Let k be a field. Let S be a finite type k-algebra. Let X = Spec(S).
Let p S be a prime ideal and let x X be the corresponding point. The following
numbers are equal
(1) dimx (X),
(2) max dim(Z) where the maximum is over those irreducible components Z of
X passing through x, and
(3) min dim(Sm ) where the minimum is over maximal ideals m with p m.
S
Proof. Let X = iI Zi be the decomposition of X into its irreducible components. There are finitely many
of them (see Lemmas 30.3 and 30.5). Let
S
I 0 = {i | x Zi }, and let T = i6I 0 Zi . Then U = X \ T is an open subset of X
containing the point x. The number (2) is maxiI 0 dim(Zi ). For any open W U ,
with x W the irreducible components of W are the irreducible sets Wi = Zi W
for i I 0 . Note that each Wi , i I 0 contains a closed point because X is Jacobson,
see Section 34. Since Wi Zi we have dim(Wi ) dim(Zi ). The existence of a
closed point implies, via Lemma 113.4, that there is a chain of irreducible closed
subsets of length equal to dim(Zi ) in the open Wi . Thus dim(Wi ) = dim(Zi ) for
any i I 0 . Hence dim(W ) is equal to the number (2). This proves that (1) = (2).
Let m p be any maximal ideal containing p. Let x0 X be the corresponding
point. First of all, x0 is contained in all the irreducible components Zi , i I 0 . Let
qi denote the minimal primes of S corresponding to the irreducible components Zi .
For each i such that x0 Zi (which is equivalent to m qi ) we have a surjection
Sm Sm /qi Sm = (S/qi )m
Moreover, the primes qi Sm so obtained exhaust the minimal primes of the Noetherian local ring Sm , see Lemma 25.3. We conclude, using Lemma 113.4, that the
dimension of Sm is the maximum of the dimensions of the Zi passing through x0 .
To finish the proof of the lemma it suffices to show that we can choose x0 such that
x0 Zi i I 0 . Because S is Jacobson (as we saw above) it is enough to show
that V (p) \ T (with T as above) is nonempty. And this is clear since it contains the
point x (i.e. p).


00OU

Lemma 113.6. Let k be a field. Let S be a finite type k-algebra. Let X = Spec(S).
Let m S be a maximal ideal and let x X be the associated closed point. Then
dimx (X) = dim(Sm ).
Proof. This is a special case of Lemma 113.5.

00OV

Lemma 113.7. Let k be a field. Let S `

be a finite type k algebra. Assume that
S is Cohen-Macaulay. Then Spec(S) =
Td is a finite disjoint union of open
and closed subsets Td with Td equidimensional (see Topology, Definition 9.5) of
dimension d. Equivalently, S is a product of rings Sd , d = 0, . . . , dim(S) such that
every maximal ideal m of Sd has height d.

COMMUTATIVE ALGEBRA

252

Proof. The equivalence of the two statements follows from Lemma 22.3. Let m S
be a maximal ideal. Every maximal chain of primes in Sm has the same length equal
to dim(Sm ), see Lemma 103.3. Hence, the dimension of the irreducible components
passing through the point corresponding to m all have dimension equal to dim(Sm ),
see Lemma 113.4. Since Spec(S) is a Jacobson topological space the intersection
of any two irreducible components of it contains a closed point if nonempty, see
Lemmas 34.2 and 34.4. Thus we have shown that any two irreducible components
that meet have the same dimension. The lemma follows easily from this, and the
fact that Spec(S) has a finite number of irreducible components (see Lemmas 30.3
and 30.5).

114. Noether normalization
00OW

In this section we prove variants of the Noether normalization lemma. The key
ingredient we will use is contained in the following two lemmas.

051M

Lemma 114.1. Let n N. Let N be a finite nonempty

set of multi-indices
P
= (1 , . . . , n ). Given e = (e1 , . . . , en ) we set e =
ei i . Then for e1  e2 
. . .  en1  en we have: If , 0 N then
(e = e 0 ) ( = 0 )
Proof. Say N = {j } with j = (j1 , . . . , jn ). Let Ai = maxj ji minj ji . If
for each i we have ei1 > Ai ei + Ai+1 ei+1 + . . . + An en then the lemma holds. For
suppose that e ( 0 ) = 0. Then for n 2,
Xn
e1 (1 10 ) =
ei (i0 i ).
i=2

0. If (1 10 ) > 0, then
Xn
Xn
e1 (1 10 ) e1 > A2 e2 + . . . + An en
ei |i0 i |

We may assume that (1

10 )

i=2

i=2

ei (i0 i ).

This contradiction implies that 10 = 1 . By induction, i0 = i for 2 i n.

051N

Lemma 114.2. Let R be a ring. Let g R[x1 , . . . , xn ] be an element which is

nonconstant, i.e., g 6 R. For e1  e2  . . .  en1  en = 1 the polynomial
g(x1 + xen1 , x2 + xen2 , . . . , xn1 + xnen1 , xn ) = axdn + lower order terms in xn
where d > 0 and a R is one of the nonzero coefficients of g.
P

Proof. Write g =
N a x with a R not zero. Here N is a finite set of
multi-indices as in Lemma 114.1 and x = x11 . . . xnn . Note that the leading term
in
(x1 + xen1 )1 . . . (xn1 + xnen1 )n1 xnn is xne1 1 +...+en1 n1 +n .
Hence the lemma follows from Lemma 114.1 which guarantees that there is exactly
one nonzero term a x of g which gives rise to the leading term of g(x1 + xen1 , x2 +
e
xen2 , . . . , xn1 + xnn1 , xn ), i.e., a = a for the unique N such that e is
maximal.


00OX

Lemma 114.3. Let k be a field. Let S = k[x1 , . . . , xn ]/I for some ideal I.
If I 6= 0, then there exist y1 , . . . , yn1 k[x1 , . . . , xn ] such that S is finite over
k[y1 , . . . , yn1 ]. Moreover we may choose yi to be in the Z-subalgebra of k[x1 , . . . , xn ]
generated by x1 , . . . , xn .

COMMUTATIVE ALGEBRA

253

Proof. Pick f I, f 6= 0. It suffices to show the lemma for k[x1 , . . . , xn ]/(f ) since
S is a quotient of that ring. We will take yi = xi xeni , i = 1, . . . , n 1 for suitable
integers ei . When does this work? It suffices to show that xn k[x1 , . . . , xn ]/(f )
is integral over the ring k[y1 , . . . , yn1 ]. The equation for xn over this ring is
f (y1 + xen1 , . . . , yn1 + xnen1 , xn ) = 0.
Hence we are done if we can show there exists integers ei such that the leading
coefficient with respect to xn of the equation above is a nonzero element of k.
This can be achieved for example by choosing e1  e2  . . .  en1 , see Lemma
114.2.

00OY

Lemma 114.4. Let k be a field. Let S = k[x1 , . . . , xn ]/I for some ideal I. There
exist r 0, and y1 , . . . , yr k[x1 , . . . , xn ] such that (a) the map k[y1 , . . . , yr ] S
is injective, and (b) the map k[y1 , . . . , yr ] S is finite. In this case the integer
r is the dimension of S. Moreover we may choose yi to be in the Z-subalgebra of
k[x1 , . . . , xn ] generated by x1 , . . . , xn .
Proof. By induction on n, with n = 0 being trivial. If I = 0, then take r = n
and yi = xi . If I 6= 0, then choose y1 , . . . , yn1 as in Lemma 114.3. Let S 0 S
be the subring generated by the images of the yi . By induction we can choose r
and z1 , . . . , zr k[y1 , . . . , yn1 ] such that (a), (b) hold for k[z1 , . . . , zr ] S 0 . Since
S 0 S is injective and finite we see (a), (b) hold for k[z1 , . . . , zr ] S. The last
assertion follows from Lemma 111.4.


00OZ

Lemma 114.5. Let k be a field. Let S be a finite type k algebra and denote
X = Spec(S). Let q be a prime of S, and let x X be the corresponding point.
There exists a g S, g 6 q such that dim(Sg ) = dimx (X) =: d and such that there
exists a finite injective map k[y1 , . . . , yd ] Sg .
Proof. Note that by definition dimx (X) is the minimum of the dimensions of Sg
for g S, g 6 q, i.e., the minimum is attained. Thus the lemma follows from
Lemma 114.4.


051P

Lemma 114.6. Let k be a field. Let q k[x1 , . . . , xn ] be a prime ideal. Set

r = trdegk (q). Then there exists a finite ring map : k[y1 , . . . , yn ] k[x1 , . . . , xn ]
such that 1 (q) = (yr+1 , . . . , yn ).
Proof. By induction on n. The case n = 0 is clear. Assume n > 0. If r = n, then
q = (0) and the result is clear. Choose a nonzero f q. Of course f is nonconstant.
After applying an automorphism of the form
k[x1 , . . . , xn ] k[x1 , . . . , xn ],

xn 7 xn ,

xi 7 xi + xeni (i < n)

we may assume that f is monic in xn over k[x1 , . . . , xn ], see Lemma 114.2. Hence
the ring map
k[y1 , . . . , yn ] k[x1 , . . . , xn ],

yn 7 f,

yi 7 xi (i < n)

is finite. Moreover yn q k[y1 , . . . , yn ] by construction. Thus q k[y1 , . . . , yn ] =

pk[y1 , . . . , yn ] + (yn ) where p k[y1 , . . . , yn1 ] is a prime ideal. Note that (p)
(q) is finite, and hence r = trdegk (p). Apply the induction hypothesis to
the pair (k[y1 , . . . , yn1 ], p) and we obtain a finite ring map k[z1 , . . . , zn1 ]
k[y1 , . . . , yn1 ] such that p k[z1 , . . . , zn1 ] = (zr+1 , . . . , zn1 ). We extend the ring

COMMUTATIVE ALGEBRA

254

map k[z1 , . . . , zn1 ] k[y1 , . . . , yn1 ] to a ring map k[z1 , . . . , zn ] k[y1 , . . . , yn ]

by mapping zn to yn . The composition of the ring maps
k[z1 , . . . , zn ] k[y1 , . . . , yn ] k[x1 , . . . , xn ]
solves the problem.
07NA

Lemma 114.7. Let R S be an injective finite type map of domains. Then there
exists an integer d and factorization
R R[y1 , . . . , yd ] S 0 S
by injective maps such that S 0 is finite over R[y1 , . . . , yd ] and such that Sf0
= Sf
for some nonzero f R.
Proof. Pick x1 , . . . , xn S which generate S over R. Let K = f.f.(R) and SK =
S R K. By Lemma 114.4 we can find y1 , . . . , yd S such that K[y1 , . . . , yd ] SK
is a finite injective map. Note that yi S because we may pick the yj in the Zalgebra generated by x1 , . . . , xn . As a finite ring map is integral (see Lemma 35.3)
we can find monic Pi K[y1 , . . . , yd ][T ] such that Pi (xi ) = 0 in SK . Let f R
be a nonzero element such that f Pi R[y1 , . . . , yd ][T ] for all i. Set x0i = f xi and
let S 0 S be the subalgebra generated by y1 , . . . , yd and x01 , . . . , x0n . Note that x0i
is integral over R[y1 , . . . , yd ] as we have Qi (x0i ) = 0 where Qi = f degT (Pi ) Pi (T /f )
which is a monic polynomial in T with coefficients in R[y1 , . . . , yd ] by our choice of
f . Hence R[y1 , . . . , yn ] S 0 is finite by Lemma 35.5. By construction Sf0
= Sf and
we win.

115. Dimension of finite type algebras over fields, reprise

07NB

This section is a continuation of Section 113. In this section we establish the

connection between dimension and transcendence degree over the ground field for
finite type domains over a field.

00P0

Lemma 115.1. Let k be a field. Let S be a finite type k algebra which is an

integral domain. Let K = f.f.(S) be the field of fractions of S. Let r = trdeg(K/k)
be the transcendence degree of K over k. Then dim(S) = r. Moreover, the local
ring of S at every maximal ideal has dimension r.
Proof. We may write S = k[x1 , . . . , xn ]/p. By Lemma 113.3 all local rings of S
at maximal ideals have the same dimension. Apply Lemma 114.4. We get a finite
injective ring map
k[y1 , . . . , yd ] S
with d = dim(S). Clearly, k(y1 , . . . , yd ) K is a finite extension and we win.

06RP

Lemma 115.2. Let k be a field. Let S be a finite type k-algebra. Let q q0 S

be distinct prime ideals. Then trdegk (q0 ) < trdegk (q).
Proof. By Lemma 115.1 we have dim V (q) = trdegk (q) and similarly for q0 .
Hence the result follows as the strict inclusion V (q0 ) V (q) implies a strict inequality of dimensions.

The following lemma generalizes Lemma 113.6.

COMMUTATIVE ALGEBRA

00P1

255

Lemma 115.3. Let k be a field. Let S be a finite type k algebra. Let X = Spec(S).
Let p S be a prime ideal, and let x X be the corresponding point. Then we
have
dimx (X) = dim(Sp ) + trdegk (p).
Proof. By Lemma 115.1 we know that r = trdegk (p) is equal to the dimension
of V (p). Pick any maximal chain of primes p p1 . . . pr starting with p in
S. This has length r by Lemma 113.4. Let qj , j J be the minimal primes of S
which are contained in p. These correspond 1 1 to minimal primes in Sp via the
rule qj 7 qj Sp . By Lemma 113.5 we know that dimx (X) is equal to the maximum
of the dimensions of the rings S/qj . For each j pick a maximal chain of primes
qj p01 . . . p0s(j) = p. Then dim(Sp ) = maxjJ s(j). Now, each chain
qi p01 . . . p0s(j) = p p1 . . . pr
is a maximal chain in S/qj , and by what was said before we have dimx (X) =
maxjJ r + s(j). The lemma follows.

The following lemma says that the codimension of one finite type Spec in another
is the difference of heights.

00P2

Lemma 115.4. Let k be a field. Let S 0 S be a surjection of finite type k

algebras. Let p S be a prime ideal, and let p0 be the corresponding prime ideal of
S 0 . Let X = Spec(S), resp. X 0 = Spec(S 0 ), and let x X, resp. x0 X 0 be the
point corresponding to p, resp. p0 . Then
dimx0 X 0 dimx X = height(p0 ) height(p).
Proof. Immediate from Lemma 115.3.

00P3

Lemma 115.5. Let k be a field. Let S be a finite type k-algebra. Let k K be a

field extension. Then dim(S) = dim(K k S).
Proof. By Lemma 114.4 there exists a finite injective map k[y1 , . . . , yd ] S with
d = dim(S). Since K is flat over k we also get a finite injective map K[y1 , . . . , yd ]
K k S. The result follows from Lemma 111.4.


00P4

Lemma 115.6. Let k be a field. Let S be a finite type k-algebra. Set X = Spec(S).
Let k K be a field extension. Set SK = K k S, and XK = Spec(SK ). Let q S
be a prime corresponding to x X and let qK SK be a prime corresponding to
xK XK lying over q. Then dimx X = dimxK XK .
Proof. Choose a presentation S = k[x1 , . . . , xn ]/I. This gives a presentation K k
S = K[x1 , . . . , xn ]/(K k I). Let q0K K[x1 , . . . , xn ], resp. q0 k[x1 , . . . , xn ] be the
corresponding primes. Consider the following commutative diagram of Noetherian
local rings
/ (K k S)q
K[x1 , . . . , xn ]q0K
K
O
O
k[x1 , . . . , xn ]q0

/ Sq

Both vertical arrows are flat because they are localizations of the flat ring maps
S SK and k[x1 , . . . , xn ] K[x1 , . . . , xn ]. Moreover, the vertical arrows have the
same fibre rings. Hence, we see from Lemma 111.7 that height(q0 ) height(q) =

COMMUTATIVE ALGEBRA

256

0
height(q0K ) height(qK ). Denote x0 X 0 = Spec(k[x1 , . . . , xn ]) and x0K XK
=
0
0
Spec(K[x1 , . . . , xn ]) the points corresponding to q and qK . By Lemma 115.4 and
what we showed above we have

n dimx X

dimx0 X 0 dimx X

height(q0 ) height(q)

height(q0K ) height(qK )

0
dimx0K XK
dimxK XK

n dimxK XK

and the lemma follows.

116. Dimension of graded algebras over a field

00P5

Here is a basic result.

00P6

Lemma 116.1. Let k be a field. Let S be a finitely generated graded algebra over
k. Assume S0 = k. Let P (T ) Q[T ] be the polynomial such that dim(Sd ) = P (d)
for all d  0. See Proposition 57.7. Then
(1) The irrelevant ideal S+ is a maximal ideal m.
(2) Any minimal prime of S is a homogeneous ideal and is contained in S+ = m.
(3) We have dim(S) = deg(P ) + 1 = dimx Spec(S) (with the convention that
deg(0) = 1) where x is the point corresponding to the maximal ideal S+ =
m.
(4) The Hilbert function of the local ring R = Sm is equal to the Hilbert function
of S.
Proof. The first statement is obvious. The second follows from Lemma 56.8. The
equality dim(S) = dimx Spec(S) follows from the fact that every irreducible component passes through x according to (2). Hence we may compute this dimension
as the dimension of the local ring R = Sm with m = S+ by Lemma 113.6. Since
md /md+1
= md R/md+1 R we see that the Hilbert function of the local ring R is
equal to the Hilbert function of S, which is (4). We conclude the last equality of
(3) by Proposition 59.8.

117. Generic flatness

051Q

Basically this says that a finite type algebra over a domain becomes flat after
inverting a single element of the domain. There are several versions of this result
(in increasing order of strength).

051R

Lemma 117.1. Let R S be a ring map. Let M be an S-module. Assume

(1) R is Noetherian,
(2) R is a domain,
(3) R S is of finite type, and
(4) M is a finite type S-module.
Then there exists a nonzero f R such that Mf is a free Rf -module.
Proof. Let K be the fraction field of R. Set SK = K R S. This is an algebra of
finite type over K. We will argue by induction on d = dim(SK ) (which is finite for
example by Noether normalization, see Section 114). Fix d 0. Assume we know
that the lemma holds in all cases where dim(SK ) < d.

COMMUTATIVE ALGEBRA

257

Suppose given R S and M as in the lemma with dim(SK ) = d. By Lemma

61.1 there exists a filtration 0 M1 M2 . . . Mn = M so that Mi /Mi1
is isomorphic to S/q for some prime q of S. Note that dim((S/q)K ) dim(SK ).
Also, note that an extension of free modules is free (see basic notion 50). Thus we
may assume M = S and that S is a domain of finite type over R.
If R S has a nontrivial kernel, then take a nonzero f R in this kernel. In this
case Sf = 0 and the lemma holds. (This is really the case d = 1 and the start
of the induction.) Hence we may assume that R S is a finite type extension of
Noetherian domains.
Apply Lemma 114.7 and replace R by Rf (with f as in the lemma) to get a
factorization
R R[y1 , . . . , yd ] S
where the second extension is finite. Note that f.f.(R[y1 , . . . , yd ]) f.f.(S) is a
finite extension of fields. Choose z1 , . . . , zr S which form a basis for f.f.(S) over
f.f.(R[y1 , . . . , yd ]). This gives a short exact sequence
(z1 ,...,zr )

0 R[y1 , . . . , yd ]r S N 0
By construction N is a finite R[y1 , . . . , yd ]-module whose support does not contain
the generic point (0) of Spec(R[y1 , . . . , yd ]). By Lemma 39.5 there exists a nonzero
g R[y1 , . . . , yd ] such that g annihilates N , so we may view N as a finite module
0
over S 0 = R[y1 , . . . , yd ]/(g). Since dim(SK
) < d by induction there exists a nonzero
f R such that Nf is a free Rf -module. Since (R[y1 , . . . , yd ])f
= Rf [y1 , . . . , yd ]
is free also we conclude by the already mentioned fact that an extension of free
modules is free.

051S

Lemma 117.2. Let R S be a ring map. Let M be an S-module. Assume

(1) R is a domain,
(2) R S is of finite presentation, and
(3) M is an S-module of finite presentation.
Then there exists a nonzero f R such that Mf is a free Rf -module.
Proof. Write S = R[x1 , . . . , xn ]/(g1 , . . . ,P
gm ). For g R[x1 , . . . , xn ] denote g its
image in S. We may write M = S t / Sni for some ni S t . Write ni =
(g i1 , . . . , g it ) for some gij R[x1 , . . . , xn ]. Let R0 R be the subring generated
by all the coefficients of all the elements gi , gij P
R[x1 , . . . , xn ]. Define S0 =
R0 [x1 , . . . , xn ]/(g1 , . . . , gm ). Define M0 = S0t / S0 ni . Then R0 is a domain
of finite type over Z and hence Noetherian (see Lemma 30.1). Moreover via the
injection R0 R we have S
= R R0 S0 and M
= R R0 M0 . Applying Lemma
117.1 we obtain a nonzero f R0 such that (M0 )f is a free (R0 )f -module. Hence
Mf = Rf (R0 )f (M0 )f is a free Rf -module.


051T

Lemma 117.3. Let R S be a ring map. Let M be an S-module. Assume

(1) R is a domain,
(2) R S is of finite type, and
(3) M is a finite type S-module.
Then there exists a nonzero f R such that
(a) Mf and Sf are free as Rf -modules, and

COMMUTATIVE ALGEBRA

258

(b) Sf is a finitely presented Rf -algebra and Mf is a finitely presented Sf module.

Proof. We first prove the lemma for S = R[x1 , . . . , xn ], and then we deduce the
result in general.
Assume S = R[x1 , . . . , xn ]. Choose elements m1 , . . . , mt which generate M . This
gives a short exact sequence
(m1 ,...,mt )

0 N S t M 0.
Denote K the fraction field of R. Denote SK = K R S = K[x1 , . . . , xn ], and
similarly NK = K R N , MK = K R M . As R K is flat the sequence remains
flat after tensoring with K. As SK = K[x1 , . . . , xn ] is a Noetherian ring (see Lemma
30.1) we can find finitely many elements
n01 , . . . , n0s NK which generate it. Choose
P
0
n1 , . . . , nr N such that ni = aij nj for some aij K. Set
X
M 0 = S t /
Sni
i=1,...,r

By construction M is a finitely presented S-module, and there is a surjection

0
M 0 M which induces an isomorphism MK
= MK . We may apply Lemma 117.2
0
to R S and M and we find an f R such that Mf0 is a free Rf -module. Thus
Mf0 Mf is a surjection of modules over the domain Rf where the source is a
free module and which becomes an isomorphism upon tensoring with K. Thus it
0
is injective as Mf0 MK
as it is free over the domain Rf . Hence Mf0 Mf is an
isomorphism and the result is proved.
For the general case, choose a surjection R[x1 , . . . , xn ] S. Think of both S
and M as finite modules over R[x1 , . . . , xn ]. By the special case proved above
there exists a nonzero f R such that both Sf and Mf are free as Rf -modules
and finitely presented as Rf [x1 , . . . , xn ]-modules. Clearly this implies that Sf is a
finitely presented Rf -algebra and that Mf is a finitely presented Sf -module.


051U

Let R S be a ring map. Let M be an S-module. Consider the following condition

on an element f R:

is of finite presentation over Rf

Sf
Mf
is of finite presentation as Sf -module
(117.3.1)

Sf , Mf
are free as Rf -modules
We define

051V

(117.3.2)

U (R S, M ) =

[
f R with (117.3.1)

D(f )

which is an open subset of Spec(R).

051W

Lemma 117.4. Let R S be a ring map. Let 0 M1 M2 M3 0 be a

short exact sequence of S-modules. Then
U (R S, M1 ) U (R S, M3 ) U (R S, M2 ).
Proof. Let u U (R S, M1 ) U (R S, M3 ). Choose f1 , f3 R such that
u D(f1 ), u D(f3 ) and such that (117.3.1) holds for f1 and M1 and for f3 and
M3 . Then set f = f1 f3 . Then u D(f ) and (117.3.1) holds for f and both M1
and M3 . An extension of free modules is free, and an extension of finitely presented
modules is finitely presented (Lemma 5.3). Hence we see that (117.3.1) holds for f
and M2 . Thus u U (R S, M2 ) and we win.


COMMUTATIVE ALGEBRA

051X

259

Lemma 117.5. Let R S be a ring map. Let M be an S-module. Let f R.

Using the identification Spec(Rf ) = D(f ) we have U (Rf Sf , Mf ) = D(f )
U (R S, M ).
Proof. Suppose that u U (Rf Sf , Mf ). Then there exists an element g
Rf such that u D(g) and such that (117.3.1) holds for the pair ((Rf )g
(Sf )g , (Mf )g ). Write g = a/f n for some a R. Set h = af . Then Rh = (Rf )g ,
Sh = (Sf )g , and Mh = (Mf )g . Moreover u D(h). Hence u U (R S, M ).
Conversely, suppose that u D(f ) U (R S, M ). Then there exists an element
g R such that u D(g) and such that (117.3.1) holds for the pair (Rg Sg , Mg ).
Then it is clear that (117.3.1) also holds for the pair (Rf g Sf g , Mf g ) = ((Rf )g
(Sf )g , (Mf )g ). Hence u U (Rf Sf , Mf ) and we win.


051Y

Lemma 117.6. Let R S be a ring map. Let M be an S-module.

Let U
S
Spec(R) be a dense open. Assume there is a covering U = iI D(fi ) of opens
such that U (Rfi Sfi , Mfi ) is dense in D(fi ) for each i I. Then U (R S, M )
is dense in Spec(R).
Proof. In view of Lemma 117.5 this is a purely topological statement. Namely, by
that lemma we see that U (R S, M ) D(fi ) is dense in D(fi ) for each i I. By
Topology, Lemma 20.4 we see that U (R S, M ) U is dense in U . Since U is
dense in Spec(R) we conclude that U (R S, M ) is dense in Spec(R).


051Z

Lemma 117.7. Let R S be a ring map. Let M be an S-module. Assume

(1) R S is of finite type,
(2) M is a finite S-module, and
(3) R is reduced.
Then there exists a subset U Spec(R) such that
(1) U is open and dense in Spec(R),
(2) for every u U there exists an f R such that u D(f ) U and such
that we have
(a) Mf and Sf are free over Rf ,
(b) Sf is a finitely presented Rf -algebra, and
(c) Mf is a finitely presented Sf -module.
Proof. Note that the lemma is equivalent to the statement that the open U (R
S, M ), see Equation (117.3.2), is dense in Spec(R). We first prove the lemma for
S = R[x1 , . . . , xn ], and then we deduce the result in general.
Proof of the case S = R[x1 , . . . , xn ] and M any finite module over S. Note that in
this case Sf = Rf [x1 , . . . , xn ] is free and of finite presentation over Rf , so we do
not have to worry about the conditions regarding S, only those that concern M .
We will use induction on n.
There exists a finite filtration
0 M1 M2 . . . M t = M
such that Mi /Mi1
= S/Ji for some ideal Ji S, see Lemma 5.4. Since a finite
intersection of dense opens is dense open, we see from Lemma 117.4 that it suffices
to prove the lemma for each of the modules R/Ji . Hence we may assume that
M = S/J for some ideal J of S = R[x1 , . . . , xn ].

COMMUTATIVE ALGEBRA

260

Let I R be the ideal generated by the coefficients of elements of J. Let U1 =

Spec(R) \ V (I) and let
U2 = Spec(R) \ U1 .
Then it is clear that U = U1 U2 is dense in Spec(R). Let f R be an element
such that either (a) D(f ) U1 or (b) D(f ) U2 . If for any such f the lemma
holds for the pair (Rf Rf [x1 , . . . , xn ], Mf ) then by Lemma 117.6 we see that
U (R S, M ) is dense in Spec(R). Hence we may assume either (a) I = R, or (b)
V (I) = Spec(R).
In case (b) we actually have I = 0 as R is reduced! Hence J = 0 and M = S and
the lemma holds in this case.
In case (a) we have to do a little bit more work. Note that every element of I
is actually the coefficient of a monomial of an element of J, because the set of
coefficients of elements of J forms an ideal (details omitted). Hence we find an
element
X
g=
a K xK J
KE

where E is a finite set of multi-indices K = (k1 , . . . , kn ) with at least one coefficient

aK0 a unit in R. Actually we can find one which has a coefficient equal to 1 as 1 I
in case (a). Let m = #{K E | aK is not a unit}. Note that 0 m #E 1.
We will argue by induction on m.
The case m = 0. In this case all the coefficients aK , K E of g are units and
E 6= . If E = {K0 } is a singleton and K0 = (0, . . . , 0), then g is a unit and J = S
so the result holds for sure. (This happens in particular when n = 0 and it provides
the base case of the induction on n.) If not E = {(0, . . . , 0)}, then at least one K
is not equal to (0, . . . , 0), i.e., g 6 R. At this point we employ the usual trick of
Noether normalization. Namely, we consider
G(y1 , . . . , yn ) = g(y1 + yne1 , y2 + yne2 , . . . , yn1 + ynen1 , yn )
with 0  en1  en2  . . .  e1 . By Lemma 114.2 it follows that G(y1 , . . . , yn )
as a polynomial in yn looks like
P
kn + i=1,...,n1 ei ki

aK yn

+ lower order terms in yn

As aK is a unit we conclude that M = R[x1 , . . . , xn ]/J is finite over R[y1 , . . . , yn1 ].

Hence U (R R[x1 , . . . , xn ], M ) = U (R R[y1 , . . . , yn1 ], M ) and we win by
induction on n.
The case m > 0. Pick a multi-index K E such that aK is not a unit. As before
set U1 = Spec(RaK ) = Spec(R) \ V (aK ) and set
U2 = Spec(R) \ U1 .
Then it is clear that U = U1 U2 is dense in Spec(R). Let f R be an element
such that either (a) D(f ) U1 or (b) D(f ) U2 . If for any such f the lemma
holds for the pair (Rf Rf [x1 , . . . , xn ], Mf ) then by Lemma 117.6 we see that
U (R S, M ) is dense in Spec(R). Hence we may assume either (a) aK R = R, or
(b) V (aK ) = Spec(R). In case (a) the number m drops, as aK has turned into a
unit. In case (b), since R is reduced, we conclude that aK = 0. Hence the set E
decreases so the number m drops as well. In both cases we win by induction on m.
At this point we have proven the lemma in case S = R[x1 , . . . , xn ]. Assume that
(R S, M ) is an arbitrary pair satisfying the conditions of the lemma. Choose

COMMUTATIVE ALGEBRA

261

a surjection R[x1 , . . . , xn ] S. Observe that, with the notation introduced in

(117.3.2), we have
U (R S, M ) = U (R R[x1 , . . . , xn ], S) U (R R[x1 , . . . , xn ], S)
Hence as weve just finished proving the right two opens are dense also the open
on the left is dense.

118. Around Krull-Akizuki
00P7

One application of Krull-Akizuki is to show that there are plenty of discrete valuation rings. More generally in this section we show how to construct discrete
valuation rings dominating Noetherian local rings.
First we show how to dominate a Noetherian local domain by a 1-dimensional
Noetherian local domain by blowing up the maximal ideal.

00P8

Lemma 118.1. Let R be a local Noetherian domain with fraction field K. Assume
R is not a field. Then there exist R R0 K with
(1) R0 local Noetherian of dimension 1,
(2) R R0 a local ring map, i.e., R0 dominates R, and
(3) R R0 essentially of finite type.
Proof. Choose any valuation ring A K dominating R (which exist by Lemma
49.2). Denote v the corresponding valuation. Let x1 , . . . , xr be a minimal set of
generators of the maximal ideal m of R. We may and do assume that v(xr ) =
min{v(x1 ), . . . , v(xr )}. Consider the ring
S = R[x1 /xr , x2 /xr , . . . , xr1 /xr ] K.
Note that mS = xr S is a principal ideal. Note that S A and that v(xr ) > 0, hence
we see that xr S 6= S. Choose a minimal prime q over xr S. Then height(q) = 1 by
Lemma 59.10 and q lies over m. Hence we see that R0 = Sq is a solution.


0BHZ

Lemma 118.2 (Koll

ar). Let (R, m) be a local Noetherian ring. Then exactly one
of the following holds:
(1) (R, m) is Artinian,
(2) (R, m) is regular of dimension 1,
(3) depth(R) 2, or
(4) there exists a finite ring map R R0 which is not an isomorphism whose
kernel and cokernel are annihilated by a power of m such that m is not an
associated prime of R0 .
Proof. Observe that (R, m) is not Artinian if and only if V (m) Spec(R) is
nowhere dense. See Proposition 59.6. We assume this from now on.
Let J R be the largest ideal killed by a power of m. If J 6= 0 then R R/J
shows that (R, m) is as in (4).
Otherwise J = 0. In particular m is not an associated prime of R and we see that
there is a nonzerodivisor x m by Lemma 62.18. If m is not an associated prime
of R/xR then depth(R) 2 by the same lemma. Thus we are left with the case
when there is an y R, y 6 xR such that ym xR.
If ym xm then we can consider the map : m m, f 7 yf /x (well defined as
x is a nonzerodivisor). By the determinantal trick of Lemma 15.2 there exists a

This is taken from a

forthcoming paper
by J
anos Koll
ar
entitled Variants of
normality for
Noetherian
schemes.

COMMUTATIVE ALGEBRA

262

monic polynomial P with coefficients in R such that P () = 0. We conclude that

P (y/x) = 0 in Rx . Let R0 Rx be the ring generated by R and y/x. Then R R0
and R0 /R is a finite R-module annihilated by a power of x. Thus R is as in (4).
Otherwise there is a t m such that yt = ux for some unit u of R. After replacing t
by u1 t we get yt = x. In particular y is a nonzerodivisor. For any t0 m we have
yt0 = xs for some s R. Thus y(t0 st) = xs xs = 0. Since y is not a zero-divisor
this implies that t0 = ts and so m = (t). Thus (R, m) is regular of dimension 1. 
00P9

Lemma 118.3. Let R be a local ring with maximal ideal m. Assume R is Noetherian, has dimension 1, and that dim(m/m2 ) > 1. Then there exists a ring map
R R0 such that
(1) R R0 is finite,
(2) R R0 is not an isomorphism,
(3) the kernel and cokernel of R R0 are annihilated by a power of m, and
(4) m is not an associated prime of R0 .
Proof. This follows from Lemma 118.2 and the fact that R is not Artinian, not
regular, and does not have depth 2 (the last part because the depth does not
exceed the dimension by Lemma 71.3).


00PA

Example 118.4. Consider the Noetherian local ring

R = k[[x, y]]/(y 2 )
It has dimension 1 and it is Cohen-Macaulay. An example of an extension as in
Lemma 118.3 is the extension
k[[x, y]]/(y 2 ) k[[x, z]]/(z 2 ), y 7 xz
in other words it is gotten by adjoining y/x to R. The effect of repeating the
construction n > 1 times is to adjoin the element y/xn .

00PB

Example 118.5. Let k be a field of characteristic p > 0 such that k has infinite degree over its subfield k p of pth powers. For example k = Fp (t1 , t2 , t3 , . . .).
Consider the ring
nX
o
A=
ai xi k[[x]] such that [k p (a0 , a1 , a2 , . . .) : k p ] <
Then A is a discrete valuation ring and its completion is A = k[[x]]. Note that
the field extension f.f.(A) f.f.(k[[x]]) is infinite purely inseparable. Choose any
f k[[x]], f 6 A. Let R = A[f ] k[[x]]. Then R is a Noetherian local domain of
dimension 1 whose completion R is nonreduced (think!).

00PC

Remark 118.6. Suppose that R is a 1-dimensional semi-local Noetherian domain.

If there is a maximal ideal m R such that Rm is not regular, then we may apply
Lemma 118.3 to (R, m) to get a finite ring extension R R1 . (For example one
can do this so that Spec(R1 ) Spec(R) is the blow up of Spec(R) in the ideal
m.) Of course R1 is a 1-dimensional semi-local Noetherian domain with the same
fraction field as R. If R1 is not a regular semi-local ring, then we may repeat the
construction to get R1 R2 . Thus we get a sequence
R R1 R2 R3 . . .
of finite ring extensions which may stop if Rn is regular for some n. Resolution
of singularities would be the claim that eventually Rn is indeed regular. In reality

COMMUTATIVE ALGEBRA

263

this is not the case. Namely, there exists a characteristic 0 Noetherian local domain
A of dimension 1 whose completion is nonreduced, see [FR70, Proposition 3.1] or
our Examples, Section 15. For an example in characteristic p > 0 see Example
118.5. Since the construction of blowing up commutes with completion it is easy to
see the sequence never stabilizes. See [Ben73] for a discussion (mostly in positive
characteristic). On the other hand, if the completion of R in all of its maximal
ideals is reduced, then the procedure stops (insert future reference here).
00PD

Lemma 118.7. Let A be a ring. The following are equivalent.

(1) The ring A is a discrete valuation ring.
(2) The ring A is a valuation ring and Noetherian.
(3) The ring A is a regular local ring of dimension 1.
(4) The ring A is a Noetherian local domain with maximal ideal m generated
by a single nonzero element.
(5) The ring A is a Noetherian local normal domain of dimension 1.
In this case if is a generator of the maximal ideal of A, then every element of A
can be uniquely written as u n , where u A is a unit.
Proof. The equivalence of (1) and (2) is Lemma 49.18. Moreover, in the proof of
Lemma 49.18 we saw that if A is a discrete valuation ring, then A is a PID, hence
(3). Note that a regular local ring is a domain (see Lemma 105.2). Using this the
equivalence of (3) and (4) follows from dimension theory, see Section 59.
Assume (3) and let be a generator of the maximal ideal m. For all n 0 we
have dimA/m mn /mn+1 = 1 because it is generated by n (and it cannot be zero).
L n n+1
In particular mn = ( n ) and the graded ring
m /m
is isomorphic to the
polynomial ring A/m[T ]. For x A \ {0} define v(x) = max{n | x mn }. In other
words x = u v(x) with u A . By the remarks above we have v(xy) = v(x) + v(y)
for all x, y A \ {0}. We extend this to the field of fractions K of A by setting
v(a/b) = v(a)v(b) (well defined by multiplicativity shown above). Then it is clear
that A is the set of elements of K which have valuation 0. Hence we see that A
is a valuation ring by Lemma 49.16.
A valuation ring is a normal domain by Lemma 49.10. Hence we see that the
equivalent conditions (1) (3) imply (5). Assume (5). Suppose that m cannot be
generated by 1 element to get a contradiction. Then Lemma 118.3 implies there is a
finite ring map A A0 which is an isomorphism after inverting any nonzero element
of m but not an isomorphism. In particular A0 f.f.(A). Since A A0 is finite it
is integral (see Lemma 35.3). Since A is normal we get A = A0 a contradiction. 

09DZ

Definition 118.8. Let A be a discrete valuation ring. A uniformizer is an element

A which generates the maximal ideal of A.
By Lemma 118.7 any two uniformizers of a discrete valuation ring are associates.

00PE

Lemma 118.9. Let R be a domain with fraction field K. Let M be an R-submodule

of K r . Assume R is local Noetherian of dimension 1. For any nonzero x R we
have lengthR (R/xR) < and
lengthR (M/xM ) r lengthR (R/xR).
Proof. If x is a unit then the result is true. Hence we may assume x m the
maximal ideal of R. Since x is not zero and R is a domain we have dim(R/xR) = 0,

COMMUTATIVE ALGEBRA

264

and hence R/xR has finite length. Consider M K r as in the lemma. We may
assume that the elements of M generate K r as a K-vector space after replacing
K r by a smaller subspace if necessary.
Suppose first that M is a finite R-module. In that case we can clear denominators
and assume M Rr . Since M generates K r as a vectors space we see that
Rr /M has finite length. In particular there exists an integer c 0 such that
xc Rr M . Note that M xM x2 M . . . is a sequence of modules with
successive quotients each isomorphic to M/xM . Hence we see that
nlengthR (M/xM ) = lengthR (M/xn M ).
The same argument for M = Rr shows that
nlengthR (Rr /xRr ) = lengthR (Rr /xn Rr ).
By our choice of c above we see that xn M is sandwiched between xn Rr and
xn+c Rr . This easily gives that
r(n + c)lengthR (R/xR) nlengthR (M/xM ) r(n c)lengthR (R/xR)
Hence in the finite case we actually get the result of the lemma with equality.
Suppose now that M is not finite. Suppose that the length of M/xM is k for
some natural number k. Then we can find
0 N0 N1 N2 . . . Nk M/xM
with Ni 6= Ni+1 for i = 0, . . . k 1. Choose an element mi M whose congruence
class mod xM falls into Ni but not into Ni1 for i = 1, . . . , k. Consider the finite
R-module M 0 = Rm1 + . . . + Rmk M . Let Ni0 M 0 /xM 0 be the inverse
0
image of Ni . It is clear that Ni0 6= Ni+1
by our choice of mi . Hence we see that
0
0
lengthR (M /xM ) k. By the finite case we conclude k rlengthR (R/xR) as
desired.

Here is a first application.
031F

Lemma 118.10. Let R S be a homomorphism of domains inducing an injection

of fraction fields K L. If R is Noetherian local of dimension 1 and [L : K] <
then
(1) each prime ideal ni of S lying over the maximal ideal m of R is maximal,
(2) there are finitely many of these, and
(3) [(ni ) : (m)] < for each i.
Proof. Pick x m nonzero. Apply Lemma 118.9 to the submodule S L
= K n
where n = [L : K]. Thus the ring S/xS has finite length over R. It follows that
S/mS has finite length over (m). In other words, dim(m) S/mS is finite (Lemma
51.6). Thus S/mS is Artinian (Lemma 52.2). The structural results on Artinian
rings implies parts (1) and (2), see for example Lemma 52.6. Part (3) is implied by
the finiteness established above.


00PF

Lemma 118.11. Let R be a domain with fraction field K. Let M be an Rsubmodule of K r . Assume R is Noetherian of dimension 1. For any nonzero
x R we have lengthR (M/xM ) < .

COMMUTATIVE ALGEBRA

265

Proof. Since R has dimension 1 we see that x is contained in finitely many primes
mi , i = 1, . . . , n, each maximal. Since R is Noetherian
Qwe see that R/xR is Artinian,
see Proposition 59.6. Hence R/xR is a quotient of R/mei i for certain ei because
that me11 . . . menn (x) for suitably large ei as R/xR
Q is Artinian (see
Q Section 52).
Hence M/xM similarly decomposes as a product (M/xM )mi = M/(mei i , x)M
of its localizations at the mi . By Lemma 118.9 applied to Mmi over Rmi we see
each Mmi /xMmi = (M/xM )mi has finite length over Rmi . It easily follows that
M/xM has finite length over R.

00PG

Lemma 118.12 (Krull-Akizuki). Let R be a domain with fraction field K. Let

K L be a finite extension of fields. Assume R is Noetherian and dim(R) = 1. In
this case any ring A with R A L is Noetherian.
Proof. To begin we may assume that L is the fraction field of A by replacing L by
the fraction field of A if necessary. Let I A be an ideal. Clearly I generates L as
a K-vector space. Hence we see that I R 6= (0). Pick any nonzero x I R. Then
we get I/xA A/xA. By Lemma 118.11 the R-module A/xA has finite length as
an R-module. Hence I/xA has finite length as an R-module. Hence I is finitely
generated as an ideal in A.


00PH

Lemma 118.13. Let R be a Noetherian local domain with fraction field K. Assume that R is not a field. Let K L be a finitely generated field extension. Then
there exists discrete valuation ring A with fraction field L which dominates R.
Proof. If L is not finite over K choose a transcendence basis x1 , . . . , xr of L over
K and replace R by R[x1 , . . . , xr ] localized at the maximal ideal generated by mR
and x1 , . . . , xr . Thus we may assume K L finite.
By Lemma 118.1 we may assume dim(R) = 1.
Let A L be the integral closure of R in L. By Lemma 118.12 this is Noetherian.
By Lemma 35.15 there is a prime ideal q A lying over the maximal ideal of R. By
Lemma 118.7 the ring Aq is a discrete valuation ring dominating R as desired. 
119. Factorization

034O

Here are some notions and relations between them that are typically taught in a
first year course on algebra at the undergraduate level.

034P

Definition 119.1. Let R be a domain.

(1) Elements x, y R are called associates if there exists a unit u R such
that x = uy.
(2) An element x R is called irreducible if it is nonzero, not a unit and
whenever x = yz, y, z R, then y is either a unit or an associate of x.
(3) An element x R is called prime if the ideal generated by x is a prime
ideal.

034Q

Lemma 119.2. Let R be a domain. Let x, y R. Then x, y are associates if and

only if (x) = (y).
Proof. If x = uy for some unit u R, then (x) (y) and y = u1 x so also
(y) (x). Conversely, suppose that (x) = (y). Then x = f y and y = gx for
some f, g A. Then x = f gx and since R is a domain f g = 1. Thus x and y are
associates.


COMMUTATIVE ALGEBRA

034R

266

Lemma 119.3. Let R be a domain. Consider the following conditions:

(1) The ring R satisfies the ascending chain condition for principal ideals.
(2) Every nonzero, nonunit element a R has a factorization a = b1 . . . bk with
each bi an irreducible element of R.
Then (1) implies (2).
Proof. Let x be a nonzero element, not a unit, which does not have a factorization
into irreducibles. Set x1 = x. We can write x = yz where neither y nor z is
irreducible or a unit. Then either y does not have a factorization into irreducibles,
in which case we set x2 = y, or z does not have a factorization into irreducibles, in
which case we set x2 = z. Continuing in this fashion we find a sequence
x1 |x2 |x3 | . . .
of elements of R with xn /xn+1 not a unit. This gives a strictly increasing sequence
of principal ideals (x1 ) (x2 ) (x3 ) . . . thereby finishing the proof.


034S

Definition 119.4. A unique factorization domain, abbreviated UFD, is a domain R such that if x R is a nonzero, nonunit, then x has a factorization into
irreducibles, and if
x = a1 . . . am = b1 . . . bn
are factorizations into irreducibles then n = m and there exists a permutation
: {1, . . . , n} {1, . . . , n} such that ai and b(i) are associates.

034T

Lemma 119.5. Let R be a domain. Assume every nonzero, nonunit factors into
irreducibles. Then R is a UFD if and only if every irreducible element is prime.
Proof. Assume R is a UFD and let x R be an irreducible element. Say ab (x),
i.e., ab = cx. Choose factorizations a = a1 . . . an , b = b1 . . . bm , and c = c1 . . . cr .
By uniqueness of the factorization
a1 . . . an b1 . . . bm = c1 . . . cr x
we find that x is an associate of one of the elements a1 , . . . , bm . In other words,
either a (x) or b (x) and we conclude that x is prime.
Assume every irreducible element is prime. We have to prove that factorization
into irreducibles is unique up to permutation and taking associates. Say a1 . . . am =
b1 . . . bn with ai and bj irreducible. Since a1 is prime, we see that bj (a1 ) for some
j. After renumbering we may assume b1 (a1 ). Then b1 = a1 u and since b1 is
irreducible we see that u is a unit. Hence a1 and b1 are associates and a2 . . . an =
ub2 . . . bm . By induction on n + m we see that n = m and ai associate to b(i) for
i = 2, . . . , n as desired.


0AFT

Lemma 119.6. Let R be a Noetherian domain. Then R is a UFD if and only if

every height 1 prime ideal is principal.
Proof. Assume R is a UFD and let p be a height 1 prime ideal. Take x p nonzero
and let x = a1 . . . an be a factorization into irreducibles. Since p is prime we see
that ai p for some i. By Lemma 119.5 the ideal (ai ) is prime. Since p has height
1 we conclude that (ai ) = p.
Assume every height 1 prime is principal. Since R is Noetherian every nonzero
nonunit element x has a factorization into irreducibles, see Lemma 119.3. It suffices
to prove that an irreducible element x is prime, see Lemma 119.5. Let (x) p be

COMMUTATIVE ALGEBRA

267

a prime minimal over (x). Then p has height 1 by Lemma 59.10. By assumption
p = (y). Hence x = yz and z is a unit as x is irreducible. Thus (x) = (y) and we
see that x is prime.

0AFU

Lemma 119.7 (Nagatas criterion for factoriality). Let A be a domain. Let S A

be a multiplicative subset generated by prime elements. Let x A be irreducible.
Then
(1) the image of x in S 1 A is irreducible or a unit, and
(2) x is prime if and only if the image of x in S 1 A is a unit or a prime
element in S 1 A.
Moreover, then A is a UFD if and only if every element of A has a factorization
into irreducibles and S 1 A is a UFD.
Proof. Say x = for , S 1 A. Then = a/s and = b/s0 for a, b A,
s, s0 S. Thus we get ss0 x = ab. By assumption we can write ss0 = p1 . . . pr for
some prime elements pi . For each i the element pi divides either a or b. Dividing
we find a factorization x = a0 b0 and a = s00 a0 , b = s000 b0 for some s00 , s000 S. As x
is irreducible, either a0 or b0 is a unit. Tracing back we find that either or is a
unit. This proves (1).
Suppose x is prime. Then A/(x) is a domain. Hence S 1 A/xS 1 A = S 1 (A/(x))
is a domain or zero. Thus x maps to a prime element or a unit.
Suppose that the image of x in S 1 A is a unit. Then yx = s for some s S and
y A. By assumption s = p1 . . . pr with pi a prime element. For each i either
pi divides y or pi divides x. In the second case pi and x are associates (as x is
irreducible) and we are done. But if the first case happens for all i = 1, . . . , r, then
x is a unit which is a contradiction.
Suppose that the image of x in S 1 A is a prime element. Assume a, b A and
ab (x). Then sa = xy or sb = xy for some s S and y A. Say the first case
happens. By assumption s = p1 . . . pr with pi a prime element. For each i either
pi divides y or pi divides x. In the second case pi and x are associates (as x is
irreducible) and we are done. If the first case happens for all i = 1, . . . , r, then
a (x) as desired. This completes the proof of (2).
The final statement of the lemma follows from (1) and (2) and Lemma 119.5.

0BC1

Lemma 119.8. A polynomial ring over a UFD is a UFD. In particular, if k is a

field, then k[x1 , . . . , xn ] is a UFD.
Proof. Let R be a UFD. Let S R be the multiplicative subset generated by
prime elements. Since every nonunit of R is a product of prime elements we see
that K = S 1 R is the fraction field of R. Observe that every prime element of R
maps to a prime element of R[x] and that S 1 (R[x]) = S 1 R[x] = K[x] is a UFD
(and even a PID). Thus we may apply Lemma 119.7 to conclude.


0AFV

Lemma 119.9. A unique factorization domain is normal.

Proof. Let R be a UFD. Let x be an element of the fraction field of R which
is integral over R. Say xd a1 xd1 . . . ad = 0 with ai R. We can write
x = upe11 . . . perr with u a unit, ei Z, and p1 , . . . , pr irreducible elements which are

[Nag57, Lemma 2]

COMMUTATIVE ALGEBRA

268

not associates. To prove the lemma we have to show ei 0. If not, say e1 < 0,
then for N  0 we get
X
2 +N
1 N
r +N
ud pde
. . . pde
= pde
p2 . . . pN
ai xdi ) (p1 )
r
r (
2
1
i=1,...,d

which contradicts uniqueness of factorization in R.

034U

Definition 119.10. A principal ideal domain, abbreviated PID, is a domain R

such that every ideal is a principal ideal.

034V

Lemma 119.11. A principal ideal domain is a unique factorization domain.

Proof. As a PID is Noetherian this follows from Lemma 119.6.

034W

Definition 119.12. A Dedekind domain is a domain R such that every nonzero

ideal I R can be written as a product
I = p1 . . . pr
of nonzero prime ideals uniquely up to permutation of the pi .

0AUQ

Lemma 119.13. A PID is a Dedekind domain.

Proof. Let R be a PID. Since every nonzero ideal of R is principal, and R is a
UFD (Lemma 119.11), this follows from the fact that every irreducible element in
R is prime (Lemma 119.5) so that factorizations of elements turn into factorizations
into primes.


09ME

Lemma 119.14. Let A be a ring. Let I and J be nonzero ideals of A such that
IJ = (f ) for some nonzerodivisor f A. Then I and J are finitely generated ideals
and finitely locally free of rank 1 as A-modules.
Proof. It suffices to show that I and J P
are finite locally free A-modules of rank 1,
see Lemma 77.2. To do this, write f = i=1,...,n xi yi with xi I and yi J. We
can also write xi yi = ai f for some ai A. Since f is a nonzerodivisor we see that
P
ai = 1. Thus it suffices to show that each Iai and Jai is free of rank 1 over Aai .
After replacing A by Aai we conclude that f = xy for some x I and y J. Note
that both x and y are nonzerodivisors. We claim that I = (x) and J = (y) which
finishes the proof. Namely, if x0 I, then x0 y = af = axy for some a A. Hence
x0 = ax and we win.


034X

Lemma 119.15. Let R be a ring. The following are equivalent

(1) R is a Dedekind domain,
(2) R is a Noetherian domain, and for every maximal ideal m the local ring Rm
is a discrete valuation ring, and
(3) R is a Noetherian, normal domain, and dim(R) 1.
Proof. Assume (1). The argument is nontrivial because we did not assume that
R was Noetherian in our definition of a Dedekind domain. Let p R be a prime
ideal. Observe that p 6= p2 by uniqueness of the factorizations in the definition.
Pick x p with x 6 p2 . Let y p be a second element (for example y = 0). Write
(x, y) = p1 . . . pr . Since (x, y) p at least one of the primes pi is contained in p.
But as x 6 p2 there is at most one. Thus exactly one of p1 , . . . , pr is contained in
p, say p1 p. We conclude that (x, y)Rp = p1 Rp is prime for every choice of y.
We claim that (x)Rp = pRp . Namely, pick y p. By the above applied with y 2 we

COMMUTATIVE ALGEBRA

269

see that (x, y 2 )Rp is prime. Hence y (x, y 2 )Rp , i.e., y = ax + by 2 in Rp . Thus
(1 by)y = ax (x)Rp , i.e., y (x)Rp as desired.
Writing (x) = p1 . . . pr anew with p1 p we conclude that p1 Rp = pRp , i.e., p1 = p.
Moreover, p1 = p is a finitely generated ideal of R by Lemma 119.14. We conclude
that R is Noetherian by Lemma 27.9. Moreover, it follows that Rm is a discrete
valuation ring for every prime ideal p, see Lemma 118.7.
The equivalence of (2) and (3) follows from Lemmas 36.10 and 118.7. Assume (2)
and (3) are satisfied. Let I R be an ideal. We will construct a factorization of
I. If I is prime, then there is nothing to prove. If not, pick I p with p R
maximal. Let J = {x R | xp I}. We claim Jp = I. It suffices to check
this after localization at the maximal ideals m of A (the formation of J commutes
with localization and we use Lemma 23.1). Then either pRm = Rm and the result
is clear, or pRm = mRm . In the last case pRm = () and the case where p is
principal is immediate. By Noetherian induction the ideal J has a factorization
and we obtain the desired factorization of I. We omit the proof of uniqueness of
the factorization.

The following is a variant of the Krull-Akizuki lemma.
09IG

Lemma 119.16. Let A be a Noetherian domain of dimension 1 with fraction field

K. Let K L be a finite extension. Let B be the integral closure of A in L. Then
B is a Dedekind domain and Spec(B) Spec(A) is surjective, has finite fibres,
and induces finite residue field extensions.
Proof. By Krull-Akizuki (Lemma 118.12) the ring B is Noetherian. By Lemma
111.4 dim(B) = 1. Thus B is a Dedekind domain by Lemma 119.15. Surjectivity
of the map on spectra follows from Lemma 35.15. The last two statements follow
from Lemma 118.10.

120. Orders of vanishing

02MB
02MC

Lemma 120.1. Let R be a semi-local Noetherian ring of dimension 1. If a, b R

are nonzerodivisors then
lengthR (R/(ab)) = lengthR (R/(a)) + lengthR (R/(b))
and these lengths are finite.
Proof. We saw the finiteness in Lemma 118.11. Additivity holds since there is a
short exact sequence 0 R/(a) R/(ab) R/(b) 0 where the first map is
given by multiplication by b. (Use length is additive, see Lemma 51.3.)


02MD

Definition 120.2. Suppose that K is a field, and R K is a local8 Noetherian

subring of dimension 1 with fraction field K. In this case we define the order of
vanishing along R
ordR : K Z
by the rule
ordR (x) = lengthR (R/(x))
if x R and we set ordR (x/y) = ordR (x) ordR (y) for x, y R both nonzero.
8We could also define this when R is only semi-local but this is probably never really what you
want!

COMMUTATIVE ALGEBRA

270

We can use the order of vanishing to compare lattices in a vector space. Here is
the definition.
02ME

Definition 120.3. Let R be a Noetherian local domain of dimension 1 with

fraction field K. Let V be a finite dimensional K-vector space. A lattice in V is a
finite R-submodule M V such that V = K R M .
The condition V = K R M signifies that M contains a basis for the vector space
K. We remark that in many places in the literature the notion of a lattice may
be defined only in case the ring R is a discrete valuation ring. If R is a discrete
valuation ring then any lattice is a free R-module, and this may not be the case in
general.

02MF

Lemma 120.4. Let R be a Noetherian local domain of dimension 1 with fraction

field K. Let V be a finite dimensional K-vector space.
(1) If M is a lattice in V and M M 0 V is an R-submodule of V containing
M then the following are equivalent
(a) M 0 is a lattice,
(b) lengthR (M 0 /M ) is finite, and
(c) M 0 is finitely generated.
(2) If M is a lattice in V and M 0 M is an R-submodule of M then M 0 is a
lattice if and only if lengthR (M/M 0 ) is finite.
(3) If M , M 0 are lattices in V , then so are M M 0 and M + M 0 .
(4) If M M 0 M 00 V are lattices in V then
lengthR (M 00 /M ) = lengthR (M 0 /M ) + lengthR (M 00 /M 0 ).
(5) If M , M 0 , N , N 0 are lattices in V and N M M 0 , M + M 0 N 0 , then
we have
lengthR (M/M M 0 ) lengthR (M 0 /M M 0 )
= lengthR (M/N ) lengthR (M 0 /N )
= lengthR (M + M 0 /M 0 ) lengthR (M + M 0 /M )
= lengthR (N 0 /M 0 ) lengthR (N 0 /M )
Proof. Proof of (1). Assume (1)(a). Say y1 , . . . , ym generate M 0 . Then each yi =
xi /fi for some xi M and nonzero fi R. Hence we see that f1 . . . fm M 0 M .
Since R is Noetherian local of dimension 1 we see that mn (f1 . . . fm ) for some
n (for example combine Lemmas 59.12 and Proposition 59.6 or combine Lemmas
118.9 and 51.4). In other words mn M 0 M for some n Hence length(M 0 /M ) <
by Lemma 51.8, in other words (1)(b) holds. Assume (1)(b). Then M 0 /M is a
finite R-module (see Lemma 51.2). Hence M 0 is a finite R-module as an extension
of finite R-modules. Hence (1)(c). The implication (1)(c) (1)(a) follows from
the remark following Definition 120.3.
Proof of (2). Suppose M is a lattice in V and M 0 M is an R-submodule. We have
seen in (1) that if M 0 is a lattice, then lengthR (M/M 0 ) < . Conversely, assume
that lengthR (M/M 0 ) < . Then M 0 is finitely generated as R is Noetherian and
for some n we have mn M M 0 (Lemma 51.4). Hence it follows that M 0 contains
a basis for V , and M 0 is a lattice.
Proof of (3). Assume M , M 0 are lattices in V . Since R is Noetherian the submodule
M M 0 of M is finite. As M is a lattice we can find x1 , . . . , xn M which form

COMMUTATIVE ALGEBRA

271

a K-basis for V . Because M 0 is a lattice we can write xi = yi /fi with yi M 0

and fi R. Hence fi xi M M 0 . Hence M M 0 is a lattice also. The fact that
M + M 0 is a lattice follows from part (1).
Part (4) follows from additivity of lengths (Lemma 51.3) and the exact sequence
0 M 0 /M M 00 /M M 00 /M 0 0
Part (5) follows from repeatedly applying part (4).
02MG

Definition 120.5. Let R be a Noetherian local domain of dimension 1 with

fraction field K. Let V be a finite dimensional K-vector space. Let M , M 0 be two
lattices in V . The distance between M and M 0 is the integer
d(M, M 0 ) = lengthR (M/M M 0 ) lengthR (M 0 /M M 0 )
of Lemma 120.4 part (5).
In particular, if M 0 M , then d(M, M 0 ) = lengthR (M/M 0 ).

02MH

Lemma 120.6. Let R be a Noetherian local domain of dimension 1 with fraction

field K. Let V be a finite dimensional K-vector space. This distance function has
the property that
d(M, M 00 ) = d(M, M 0 ) + d(M 0 , M 00 )
whenever given three lattices M , M 0 , M 00 of V . In particular we have d(M, M 0 ) =
d(M 0 , M ).
Proof. Omitted.

02MI

Lemma 120.7. Let R be a Noetherian local domain of dimension 1 with fraction

field K. Let V be a finite dimensional K-vector space. Let : V V be a K-linear
isomorphism. For any lattice M V we have
d(M, (M )) = ordR (det())
Proof. We can see that the integer d(M, (M )) does not depend on the lattice M
as follows. Suppose that M 0 is a second such lattice. Then we see that
d(M, (M ))

= d(M, M 0 ) + d(M 0 , (M ))
=

d(M, M 0 ) + d((M 0 ), (M )) + d(M 0 , (M 0 ))

Since is an isomorphism we see that d((M 0 ), (M )) = d(M 0 , M ) = d(M, M 0 ),

and hence d(M, (M )) = d(M 0 , (M 0 )). Moreover, both sides of the equation (of
the lemma) are additive in , i.e.,
ordR (det( )) = ordR (det()) + ordR (det())
and also
d(M, (((M )))

= d(M, (M )) + d((M ), ((M )))

=

d(M, (M )) + d(M, (M ))

by the independence shown above. Hence it suffices to prove the lemma for generators of GL(V ). Choose an isomorphism K n
= V . Then GL(V ) = GLn (K) is
generated by elementary matrices E. The result is clear for E equal to the identity
matrix. If E = Eij () with i 6= j, K, 6= 0, for example

1
...
1 . . .
E12 () = 0
... ... ...

COMMUTATIVE ALGEBRA

272

then with respect to a different basis we get E12 (1). The result is clear for E =
E12 (1) by taking as lattice Rn K n . Finally, if E = Ei (a), with a K for
example

a
0 ...
1 . . .
E1 (a) = 0
... ... ...
then E1 (a)(Rb ) = aR Rn1 and it is clear that d(Rn , aR Rn1 ) = ordR (a)
as desired.

02MJ

Lemma 120.8. Let A B be a ring map. Assume

(1) A is a Noetherian local domain of dimension 1,
(2) A B is a finite extension of domains.
Let K = f.f.(A) and L = f.f.(B) so that L is a finite field extension of K. Let
y L and x = NmL/K (y). In this situation B is semi-local. Let mi , i = 1, . . . , n
be the maximal ideals of B. Then
X
ordA (x) =
[(mi ) : (mA )]ordBmi (y)
i

where ord is defined as in Definition 120.2.

Proof. The ring B is semi-local by Lemma 112.2. Write y = b/b0 for some b, b0 B.
By the additivity of ord and multiplicativity of Nm it suffices to prove the lemma
for y = b or y = b0 . In other words we may assume y B. In this case the left
hand side of the formula is
X
[(mi ) : (mA )]lengthBm ((B/yB)mi )
i

By Lemma 51.12 this is equal to lengthA (B/yB). By Lemma 120.7 we have

y

lengthA (B/yB) = d(B, yB) = ordA (detK (L

L)).
y

Since x = NmL/K (y) = detK (L

L) by definition the lemma is proved.

We can extend some of the results above to reduced 1-dimensional Noetherian local
rings which are not domains by the following lemma.
07K3

Lemma 120.9. Let (R, m) be a reduced Noetherian local ring of dimension 1 and
let x m be a nonzerodivisor. Let q1 , . . . , qr be the minimal primes of R. Then
X
lengthR (R/(x)) =
ordR/qi (x)
i

Proof. Note that Ri = R/qi is a Noetherian 1-dimensional local domain. Denote

Ki = f.f.(Ri ). If x is a unit in R, then
Q both sides are zero. Hence we may assume
x m. Consider the map : R Ri . As R is reduced
this map is injective,
Q
see Lemma 16.2. By Lemma 24.4 we have Q(R) =
Ki . Hence the finite Rmodule Coker() is annihilated by a nonzerodivisor y R, hence has support {m},
is annihilated by some power of x and has finite length over R, see Lemma 61.3.
Consider the short exact sequence
Y
0R
Ri Coker() 0
Applying multiplication by xn to this for n  0 we obtain from the snake lemma
Y
0 Coker() R/xn R
Ri /xn Ri Coker() 0

COMMUTATIVE ALGEBRA

273

Thus we see that

X
Y
lengthR (R/xn R) = lengthR ( Ri /xn Ri ) =
lengthR (Ri /xn Ri )
by Lemma 51.3. By Lemma 51.5 we have lengthR (Ri /xn Ri ) = lengthRi (Ri /xn Ri ).
Now the result follows from the additivity of Lemma 120.1 and the definition of the
order of vanishing along Ri .

02QF

Lemma 120.10. Let R be a Noetherian local ring with maximal ideal m. Let M
be a finite R-module. Let x R. Assume that
(1) dim(Supp(M )) 1, and
(2) dim(Supp(M/xM )) 0.
Write Supp(M ) = {m, q1 , . . . , qt }. Then
X
lengthR (Mx ) lengthR (x M ) =
ordR/qi (x)lengthRq (Mqi ).
i=1,...,t

where Mx = M/xM and x M = Ker(x : M M ).

Proof. We first make some preparatory remarks. The result of the lemma holds
if M has finite length, i.e., if t = 0, because in this case the exact sequence 0
x M M M Mx 0 and additivity of length shows that lengthR (Mx ) =
lengthR (x M ). Also, if we have a short exact sequence 0 M M 0 M 00 0
of modules satisfying (1) and (2), then lemma for 2 out of 3 of these implies the
lemma for the third by the snake lemma.
Denote Mi the image
L of M in Mqi , so Supp(Mi ) = {m, qi }. The kernel and cokernel
of the map M
Mi have support {m} and hence have finite length. By our
preparatory remarks, it follows that it suffices to prove the lemma for each Mi . Thus
we may assume that Supp(M ) = {m, q}. In this case we can filter M by powers
of q. Again additivity shows that it suffices to prove the lemma in the case M is
annihilated by q. In this case we can view M as a R/q-module, i.e., we may assume
that R is a Noetherian local domain of dimension 1 with fraction field K. Dividing
by the torsion submodule, i.e., by the kernel of M M R K = V (the torsion
has finite length hence is handled by our preliminary remarks) we may assume that
M V is a lattice. Then length(x M ) = 0 and lengthR (Mx ) = d(M, xM ). Since
lengthK (V ) = dimK (V ) we see that det(x : V V ) = xdimK (V ) and ordR (det(x :
V V )) = dimK (V )ordR (x). Thus the lemma follows from Lemma 120.7 in this
particular case.

02QG

Lemma 120.11. Let R be a Noetherian local ring with maximal ideal m. Let
I R be an ideal and let x R. Assume x is a nonzerodivisor on R/I and that
dim(R/I) = 1. Then
X
lengthR (R/(x, I)) =
lengthR (R/(x, qi ))lengthRq ((R/I)qi )
i

where q1 , . . . , qn are the minimal primes over I. More generally if M is any finite
Cohen-Macaulay module of dimension 1 over R and dim(Supp(M/xM )) = 0, then
X
lengthR (M/xM ) =
lengthR (R/(x, qi ))lengthRq (Mqi ).
i

where q1 , . . . , qt are the minimal primes of the support of M .

Proof. These are special cases of Lemma 120.10.

COMMUTATIVE ALGEBRA

274

121. Quasi-finite maps

02MK

Consider a ring map R S of finite type. A map Spec(S) Spec(R) is quasifinite at a point if that point is isolated in its fibre. This means that the fibre is
zero dimensional at that point. In this section we study the basic properties of this
important but technical notion. More advanced material can be found in the next
section.

00PJ

Lemma 121.1. Let k be a field. Let S be a finite type k algebra. Let q be a prime
of S. The following are equivalent:
(1)
(2)
(3)
(4)
(5)
(6)

q is an isolated point of Spec(S),

Sq is finite over k,
there exists a g S, g 6 q such that D(g) = {q},
dimq Spec(S) = 0,
q is a closed point of Spec(S) and dim(Sq ) = 0, and
the field extension k (q) is finite and dim(Sq ) = 0.

In this case S = Sq S 0 for some finite type k-algebra S 0 . Also, the element g as
in (3) has the property Sq = Sg .
Proof. Suppose q is an isolated point of Spec(S), i.e., {q} is open in Spec(S).
Because Spec(S) is a Jacobson space (see Lemmas 34.2 and 34.4) we see that q is a
closed point. Hence {q} is open and closed in Spec(S). By Lemmas 20.3 and 22.3
we may write S = S1 S2 with q corresponding to the only point Spec(S1 ). Hence
S1 = Sq is a zero dimensional ring of finite type over k. Hence it is finite over k for
example by Lemma 114.4. We have proved (1) implies (2).
Suppose Sq is finite over k. Then Sq is Artinian local, see Lemma 52.2. So
Spec(Sq ) = {qSq } by Lemma 52.6. Consider the exact sequence 0 K S
Sq Q 0. It is clear that Kq = Qq = 0. Also, K is a finite S-module as S is
Noetherian and Q is a finite S-modules since Sq is finite over k. Hence there exists
g S, g 6 q such that Kg = Qg = 0. Thus Sq = Sg and D(g) = {q}. We have
proved that (2) implies (3).
Suppose D(g) = {q}. Since D(g) is open by construction of the topology on Spec(S)
we see that q is an isolated point of Spec(S). We have proved that (3) implies (1).
In other words (1), (2) and (3) are equivalent.
Assume dimq Spec(S) = 0. This means that there is some open neighbourhood of q
in Spec(S) which has dimension zero. Then there is an open neighbourhood of the
form D(g) which has dimension zero. Since Sg is Noetherian we conclude that Sg
is Artinian and D(g) = Spec(Sg ) is a finite discrete set, see Proposition 59.6. Thus
q is an isolated point of D(g) and, by the equivalence of (1) and (2) above applied
to qSg Sg , we see that Sq = (Sg )qSg is finite over k. Hence (4) implies (2). It is
clear that (1) implies (4). Thus (1) (4) are all equivalent.
Lemma 113.6 gives the implication (5) (4). The implication (4) (6) follows
from Lemma 115.3. The implication (6) (5) follows from Lemma 34.9. At this
point we know (1) (6) are equivalent.
The two statements at the end of the lemma we saw during the course of the proof
of the equivalence of (1), (2) and (3) above.


COMMUTATIVE ALGEBRA

00PK

275

Lemma 121.2. Let R S be a ring map of finite type. Let q S be a prime

lying over p R. Let F = Spec(S R (p)) be the fibre of Spec(S) Spec(R),
see Remark 16.8. Denote q F the point corresponding to q. The following are
equivalent
(1) q is an isolated point of F ,
(2) Sq /pSq is finite over (p),
(3) there exists a g S, g 6 q such that the only prime of D(g) mapping to p
is q,
(4) dimq (F ) = 0,
(5) q is a closed point of F and dim(Sq /pSq ) = 0, and
(6) the field extension (p) (q) is finite and dim(Sq /pSq ) = 0.
Proof. Note that Sq /pSq = (S R (p))q . Moreover S R (p) is of finite type
over (p). The conditions correspond exactly to the conditions of Lemma 121.1 for
the (p)-algebra S R (p) and the prime q, hence they are equivalent.


00PL

Definition 121.3. Let R S be a finite type ring map. Let q S be a prime.

(1) If the equivalent conditions of Lemma 121.2 are satisfied then we say R S
is quasi-finite at q.
(2) We say a ring map A B is quasi-finite if it is of finite type and quasi-finite
at all primes of B.

00PM

Lemma 121.4. Let R S be a finite type ring map. Then R S is quasi-finite

if and only if for all primes p R the fibre S R (p) is finite over (p).
Proof. If the fibres are finite then the map is clearly quasi-finite. For the converse,
note that S R (p) is a (p)-algebra of finite type over k of dimension 0. Hence it
is finite over k for example by Lemma 114.4.


077H

Lemma 121.5. Let R S be a finite type ring map. Let q S be a prime lying
over p R. Let f R, f 6 p and g S, g 6 q. Then R S is quasi-finite at q if
and only if Rf Sf g is quasi-finite at qSf g .
Proof. The fibre of Spec(Sf g ) Spec(Rf ) is homeomorphic to an open subset
of the fibre of Spec(S) Spec(R). Hence the lemma follows from part (1) of the
equivalent conditions of Lemma 121.2.


00PN

Lemma 121.6. Let

SO

/ S0
O

q0

/ R0

p0

be a commutative diagram of rings with primes as indicated. Assume R S of

finite type, and S R R0 S 0 surjective. If R S is quasi-finite at q, then R0 S 0
is quasi-finite at q0 .
Proof. Write S R (p) = S1 S2 with S1 finite over (p) and such that q corresponds to a point of S1 as in Lemma 121.1. Because S R R0 S 0 surjective the
canonical map (S R (p)) (p) (p0 ) S 0 R0 (p0 ) is surjective. Let Si0 be the
image of Si (p) (p0 ) in S 0 R0 (p0 ). Then S 0 R0 (p0 ) = S10 S20 and S10 is finite
over (p0 ). The map S 0 R0 (p0 ) (q0 ) factors through S10 (i.e. it annihilates the

COMMUTATIVE ALGEBRA

276

factor S20 ) because the map S R (p) (q) factors through S1 (i.e. it annihilates
the factor S2 ). Thus q0 corresponds to a point of Spec(S10 ) in the disjoint union
decomposition of the fibre: Spec(S 0 R0 (p0 )) = Spec(S10 ) q Spec(S10 ). (See Lemma
20.2.) Since S10 is finite over a field, it is Artinian ring, and hence Spec(S10 ) is a
finite discrete set. (See Proposition 59.6.) We conclude q0 is isolated in its fibre as
desired.

00PO

Lemma 121.7. A composition of quasi-finite ring maps is quasi-finite.

Proof. Suppose A B and B C are quasi-finite ring maps. By Lemma 6.2
we see that A C is of finite type. Let r C be a prime of C lying over q B
and p A. Since A B and B C are quasi-finite at q and r respectively, then
there exist b B and c C such that q is the only prime of D(b) which maps to p
and similarly r is the only prime of D(c) which maps to q. If c0 C is the image
of b B, then r is the only prime of D(cc0 ) which maps to p. Therefore A C is
quasi-finite at r.


00PP

Lemma 121.8. Let R S be a ring map of finite type. Let R R0 be any ring
map. Set S 0 = R0 R S.
(1) The set {q0 | R0 S 0 quasi-finite at q0 } is the inverse image of the corresponding set of Spec(S) under the canonical map Spec(S 0 ) Spec(S).
(2) If Spec(R0 ) Spec(R) is surjective, then R S is quasi-finite if and only
if R0 S 0 is quasi-finite.
(3) Any base change of a quasi-finite ring map is quasi-finite.
Proof. Let p0 R0 be a prime lying over p R. Then the fibre ring S 0 R0 (p0 ) is
the base change of the fibre ring SR (p) by the field extension (p) (p0 ). Hence
the first assertion follows from the invariance of dimension under field extension
(Lemma 115.6) and Lemma 121.1. The stability of quasi-finite maps under base
change follows from this and the stability of finite type property under base change.
The second assertion follows since the assumption implies that given a prime q S
we can find a prime q0 S 0 lying over it.

The following lemma is not quite about quasi-finite ring maps, but it does not seem
to fit anywhere else so well.

02ML

Lemma 121.9. Let R S be a ring map of finite type. Let p R be a minimal

prime. Assume that there are at most finitely many primes of S lying over p. Then
there exists a g R, g 6 p such that the ring map Rg Sg is finite.
Proof. Let x1 , . . . , xn be generators of S over R. Since p is a minimal prime we
have that pRp is a locally nilpotent ideal, see Lemma 24.1. Hence pSp is a locally
nilpotent ideal, see Lemma 31.2. By assumption the finite type (p)-algebra Sp /pSp
has finitely many primes. Hence (for example by Lemmas 60.3 and 114.4) (p)
Sp /pSp is a finite ring map. Thus we may find monic polynomials Pi Rp [X] such
that Pi (xi ) maps to zero in Sp /pSp . By what we said above there exist ei 1 such
that P (xi )ei = 0 in Sp . Let g1 R, g1 6 p be an element such that Pi R[1/g1 ]
for all i. Next, let g2 R, g2 6 p be an element such that P (xi )ei = 0 in Sg1 g2 .
Setting g = g1 g2 we win.


COMMUTATIVE ALGEBRA

277

122. Zariskis Main Theorem

00PI

In this section our aim is to prove the algebraic version of Zariskis Main theorem.
This theorem will be the basis of many further developments in the theory of
schemes and morphisms of schemes later in the project.
Let R S be a ring map of finite type. Our goal in this section is to show that
the set of points of Spec(S) where the map is quasi-finite is open (Theorem 122.13).
In fact, it will turn out that there exists a finite ring map R S 0 such that in
some sense the quasi-finite locus of S/R is open in Spec(S 0 ) (but we will not prove
this in the algebra chapter since we do not develop the language of schemes here
for the case where R S is quasi-finite see Lemma 122.15). These statements are
somewhat tricky to prove and we do it by a long list of lemmas concerning integral
and finite extensions of rings. This material may be found in [Ray70], and [Pes66].
We also found notes by Thierry Coquand helpful.

00PQ

Lemma 122.1. Let : R S be a ring map. Suppose t S satisfies the relation

(a0 ) + (a1 )t + . . . + (an )tn = 0. Then (an )t is integral over R.
Proof. Namely, multiply the equation (a0 ) + (a1 )t + . . . + (an )tn = 0 with
) + (a1 ann2 )((an )t) + . . . + ((an )t)n = 0. 
(an )n1 and write it as (a0 an1
n
The following lemma is in some sense the key lemma in this section.

00PT

Lemma 122.2. Let R be a ring. Let : R[x] S be a ring map. Let t S.

Assume that (a) t is integral over R[x], and (b) there exists a monic p R[x] such
that t(p) Im(). Then there exists a q R[x] such that t (q) is integral over
R.
Proof. Write t(p) = (r) for some r R[x]. Using euclidean division, write
r = qp + r0 with q, r0 R[x] and deg(r0 ) < deg(p). We may replace t by t (q)
which is still integral over R[x], so that we obtain t(p) = (r0 ). In the ring St we
may write this as (p) (1/t)(r0 ) = 0. This implies that (x) gives an element of
the localization St which is integral over (R)[1/t] St . On the other hand, t is
integral over the subring (R)[(x)] S. Combined we conclude that t is integral
over the subring (R)[1/t] P
St , see Lemma 35.6. In other words there exists an
equation of the form td + i<d ((ri )/tni )ti = 0 in St with ri R. This means
P
that td+N + i<d (ri )ti+N ni = 0 in S for some N large enough. In other words
t is integral over R.


00PU

Lemma 122.3. Let R be a ring and let : R[x] S be a ring map. Let t S.
If t is integral over R[x], then there exists an ` 0 such that for every a R the
element (a)` t is integral over a : R[y] S, defined by y 7 (ax) and r 7 (r)
for r R.
P
Proof. Say td + i<d (fi )ti = 0 with fi R[x]. Let ` be the maximum degree in
P
x of all the fi . Multiply the equation by (a)` to get (a)` td + i<d (a` fi )ti = 0.
Note that each (a` fi ) is in the image of a . The result follows from Lemma
122.1.


00PV

Lemma 122.4. Let R be a ring. Let : R[x] S be a ring map. Let t S.

Assume t is integral over R[x]. Let p R[x], p = a0 + a1 x + . . . + ak xk such that
t(p) Im(). Then there exists a q R[x] and n 0 such that (ak )n t (q) is
integral over R.

COMMUTATIVE ALGEBRA

278

Proof. By Lemma 122.3 there exists an ` 0 such that the element (ak )` t
is integral over the map 0 : R[y] S, 0 (y) = (ak x) and 0 (r) = (r), for
r R. The polynomial p0 = ak1
a0 + ak2
a1 y + . . . + y k is monic and t0 (p0 ) =
k
k
k1
(ak )t(p) Im(). By definition of 0 this implies there exists a n k 1 such
that (ank )t0 (p0 ) Im(0 ). If also n `, then (ak )n t is still integral over R[y].
By Lemma 122.2 we see that (ak )n t 0 (q) is integral over R for some q R[y].
Again by the simple relationship between 0 and this implies the lemma.

00PW

Situation 122.5. Let R be a ring. Let : R[x] S be finite. Let

J = {g S | gS Im()}
be the conductor ideal of . Assume (R) S integrally closed in S.

00PX

Lemma 122.6. In Situation 122.5. Suppose u S, a0 , . . . , ak R, u(a0 + a1 x +

. . . + ak xk ) J. Then there exists an m 0 such that u(ak )m J.
Proof. Assume that S is generated by t1 , . . . , tn as an R[x]-module. In this case
J = {g S | gti Im() for all i}. Note that each element uti is integral over
R[x], see Lemma 35.3. We have (a0 + a1 x + . . . + ak xk )uti Im(). By Lemma
122.4, for each i there exists an integer ni and an element qi R[x] such that
(ank i )uti (qi ) is integral over R. By assumption this element is in (R) and
hence (ank i )uti Im(). It follows that m = max{n1 , . . . , nn } works.


00PY

Lemma 122.7. In Situation 122.5. Suppose u S, a0 , . . . , ak R, u(a0 + a1 x +

. . . + ak xk ) J. Then u(ai ) J for all i.

Proof. Under the assumptions of the lemma we have un (a0 +a1 x+. . .+ak xk )n
J for some n 1. By Lemma 122.6 we deduce un (anm
k ) J for some m 1.
k
and
so
u(a
+
a
x
+
.
.
.
+
a
x
)

u(ak ) = u(a0 + a1 x +
Thus u(ak ) J,
0
1
k

. . . + ak1 xk1 ) J. We win by induction on k.


This lemma suggests the following definition.

00PZ

Definition 122.8. Given an inclusion of rings R S and an element x S we

say that x is strongly transcendental over R if whenever u(a0 +a1 x+. . .+ak xk ) = 0
with u S and ai R, then we have uai = 0 for all i.
Note that if S is a domain then this is the same as saying that x as an element of
the fraction field of S is transcendental over the fraction field of R.

00Q0

Lemma 122.9. Suppose R S is an inclusion of reduced rings and suppose that

x S is strongly transcendental over R. Let q S be a minimal prime and let
p = R q. Then the image of x in S/q is strongly transcendental over the subring
R/p.
Proof. Suppose u(a0 + a1 x + . . . + ak xk ) q. By Lemma 24.1 the local ring Sq
is a field, and hence u(a0 + a1 x + . . . + ak xk ) is zero in Sq . Thus uu0 (a0 + a1 x +
. . . + ak xk ) = 0 for some u0 S, u0 6 q. Since x is strongly transcendental over R
we get uu0 ai = 0 for all i. This in turn implies that uai q.


00Q1

Lemma 122.10. Suppose R S is an inclusion of domains and let x S.

Assume x is (strongly) transcendental over R and that S is finite over R[x]. Then
R S is not quasi-finite at any prime of S.

COMMUTATIVE ALGEBRA

279

Proof. As a first case, assume that R is normal, see Definition 36.11. By Lemma
36.14 we see that R[x] is normal. Take a prime q S, and set p = R q. Assume
that the extension (p) (q) is finite. This would be the case if R S is
quasi-finite at q. Let r = R[x] q. Then since (p) (r) (q) we see that
the extension (p) (r) is finite too. Thus the inclusion r pR[x] is strict. By
going down for R[x] S, see Proposition 37.7, we find a prime q0 q, lying over
the prime pR[x]. Hence the fibre Spec(S R (p)) contains a point not equal to q,
namely q0 , whose closure contains q and hence q is not isolated in its fibre.
If R is not normal, let R R0 K be the integral closure R0 of R in its field
of fractions K. Let S S 0 L be the subring S 0 of the field of fractions L of
S generated by R0 and S. Note that by construction the map S R R0 S 0 is
surjective. This implies that R0 [x] S 0 is finite. Also, the map S S 0 induces
a surjection on Spec, see Lemma 35.15. We conclude by Lemma 121.6 and the
normal case we just discussed.

00Q2

Lemma 122.11. Suppose R S is an inclusion of reduced rings. Assume x S

be strongly transcendental over R, and S finite over R[x]. Then R S is not
quasi-finite at any prime of S.
Proof. Let q S be any prime. Choose a minimal prime q0 q. According to
Lemmas 122.9 and 122.10 the extension R/(R q0 ) S/q0 is not quasi-finite at the
prime corresponding to q. By Lemma 121.6 the extension R S is not quasi-finite
at q.


00Q8

Lemma 122.12. Let R be a ring. Let S = R[x]/I. Let q S be a prime. Assume

R S is quasi-finite at q. Let S 0 S be the integral closure of R in S. Then there
exists an element g S 0 , g 6 q such that Sg0
= Sg .
Proof. Let p be the image of q in Spec(R). The assumption that R S is quasifinite at q implies there exists an f I, f = an xn + . . . + a0 such that some ai 6 p.
In particular there exists a relation bm xm + . . . + b0 = 0 with bj S 0 , j = 0, . . . , m
and bj 6 q S 0 for some j. We prove the lemma by induction on m.
The case bm q. In this case we have bm x S 0 by Lemma 122.1. Set b0m1 =
bm x + bm1 . Then
b0m1 xm1 + bm2 xm2 + . . . + b0 = 0
Since b0m1 is congruent to bm1 modulo S 0 q we see that it is still the case that
one of b0m1 , bm2 , . . . , b0 is not in S 0 q. Thus we win by induction on m.
The case bm 6 q. In this case x is integral over Sb0 m , in fact bm x S 0 by Lemma
122.1. Hence the injective map Sb0 m Sbm is also surjective, i.e., an isomorphism
as desired.


00Q9

Theorem 122.13 (Zariskis Main Theorem). Let R be a ring. Let R S be a

finite type R-algebra. Let S 0 S be the integral closure of R in S. Let q S be a
prime of S. If R S is quasi-finite at q then there exists a g S 0 , g 6 q such that
Sg0
= Sg .
Proof. There exist finitely many elements x1 , . . . , xn S such that S is finite over
the R-sub algebra generated by x1 , . . . , xn . (For example generators of S over R.)
We prove the proposition by induction on the minimal such number n.

COMMUTATIVE ALGEBRA

280

The case n = 0 is trivial, because in this case S 0 = S, see Lemma 35.3.

The case n = 1. We may and do replace R by its integral closure in S, in particular
this means that R S. Consider the map : R[x] S, x 7 x1 . (We will see
that is not injective below.) By assumption is finite. Hence we are in Situation
122.5. Let J S be the conductor ideal defined in Situation 122.5. Consider the
diagram

/ S/ J o
/S
R/(R J)[x]
R[x]
O
`
O
6

/ R/(R

J)

According to Lemma
122.7
the
image
of
x
in
the
quotient
S/
J is strongly

transcen
dental over R/(R J). Hence by Lemma
122.11 the ring map R/(R J) S/ J
is not quasi-finite at any prime of S/ J. By Lemma 121.6 we deduce that q does
not lie in V (J) Spec(S). Thus there exists an element s J, s 6 q. By
definition of J we may write s = (f ) for some polynomial f R[x]. Now let
I = Ker(R[x] S). Since (f ) J we get (R[x]/I)f
= S(f ) . Also s 6 q means
that f 6 1 (q). Thus 1 (q) is a prime of R[x]/I at which R R[x]/I is quasifinite, see Lemma 121.5. Let C R[x]/I be the integral closure of R. By Lemma
122.12 there exists an element h C, h 6 1 (q) such that Ch
= (R[x]/I)h . We
conclude that (R[x]/I)f h = S(f h) is isomorphic to a principal localization Ch0 of
C for some h0 C, h0 6 1 (q). Since (C) S 0 we get g = (h0 ) S 0 , g 6 q and
moreover the injective map Sg0 Sg is also surjective because by our choice of h0
the map Ch0 Sg is surjective.
The case n > 1. Consider the subring R0 S which is the integral closure of
R[x1 , . . . , xn1 ] in S. By Lemma 121.6 the extension S/R0 is quasi-finite at q.
Also, note that S is finite over R0 [xn ]. By the case n = 1 above, there exists a
g 0 R0 , g 0 6 q such that (R0 )g0
= Sg0 . At this point we cannot apply induction to
R R0 since R0 may not be finite type over R. Since S is finitely generated over R
we deduce in particular that (R0 )g0 is finitely generated over R. Say the elements
g 0 , and y1 /(g 0 )n1 , . . . , yN /(g 0 )nN with yi R0 generate (R0 )g0 over R. Let R00 be the
R-sub algebra of R0 generated by x1 , . . . , xn1 , y1 , . . . , yN , g 0 . This has the property
(R00 )g0
= Sg0 . Surjectivity because of how we chose yi , injectivity because R00 R0 ,
and localization is exact. Note that R00 is finite over R[x1 , . . . , xn1 ] because of our
choice of R0 , see Lemma 35.4. Let q00 = R00 q. Since (R00 )q00 = Sq we see that
R R00 is quasi-finite at q00 , see Lemma 121.2. We apply our induction hypothesis
to R R00 , q00 and x1 , . . . , xn1 R00 and we find a subring R000 R00 which is
integral over R and an element g 00 R000 , g 00 6 q00 such that (R000 )g00
= (R00 )g00 . Write
0
00 00
000
00 n
000
000
the image of g in (R )g as g /(g ) for some g R . Set g = g 00 g 000 R000 .
Then it is clear that g 6 q and (R000 )g
= Sg . Since by construction we have R000 S 0
0
we also have Sg = Sg as desired.

00QA

Lemma 122.14. Let R S be a finite type ring map. The set of points q of
Spec(S) at which S/R is quasi-finite is open in Spec(S).
Proof. Let q S be a point at which the ring map is quasi-finite. By Theorem
122.13 there exists an integral ring extension R S 0 , S 0 S and an element
g S 0 , g 6 q such that Sg0
= Sg . Since S and hence Sg are of finite type over R we

COMMUTATIVE ALGEBRA

281

may find finitely many elements y1 , . . . , yN of S 0 such that Sg00

= S where S 00 S 0 is
00
the sub R-algebra generated by g, y1 , . . . , yN . Since S is finite over R (see Lemma
35.4) we see that S 00 is quasi-finite over R (see Lemma 121.4). It is easy to see
that this implies that Sg00 is quasi-finite over R, for example because the property
of being quasi-finite at a prime depends only on the local ring at the prime. Thus
we see that Sg is quasi-finite over R. By the same token this implies that R S
is quasi-finite at every prime of S which lies in D(g).

00QB

Lemma 122.15. Let R S be a finite type ring map. Suppose that S is quasifinite over R. Let S 0 S be the integral closure of R in S. Then
(1) Spec(S) Spec(S 0 ) is a homeomorphism onto an open subset,
(2) if g S 0 and D(g) is contained in the image of the map, then Sg0
= Sg , and
(3) there exists a finite R-algebra S 00 S 0 such that (1) and (2) hold for the
ring map S 00 S.
Proof. Because S/R is quasi-finite we may apply Theorem 122.13 to each point
q of Spec(S). Since Spec(S) is quasi-compact, see Lemma 16.10, we may choose a
finite number of gi S 0 , i = 1, . . . , n such that Sg0 i = Sgi , and such that g1 , . . . , gn
generate the unit ideal in S (in other words the standard opens of Spec(S) associated
to g1 , . . . , gn cover all of Spec(S)).
S
Suppose that D(g) Spec(S 0 ) is contained in the image. Then D(g) D(gi ).
0

In other words, g1 , . . . , gn generate the unit ideal of Sg0 . Note that Sgg
= Sggi by
i
0
our choice of gi . Hence Sg = Sg by Lemma 23.2.
We construct a finite algebra S 00 S 0 as in (3). To do this note that each Sg0 i
= Sgi
is a finite type R-algebra. For each i pick some elements yij S 0 such that each
Sg0 i is generated as R-algebra by 1/gi and the elements yij . Then set S 00 equal to
the sub R-algebra of S 0 generated by all gi and all the yij . Details omitted.

123. Applications of Zariskis Main Theorem

03GB

02MM

Here is an immediate application characterizing the finite maps of 1-dimensional

semi-local rings among the quasi-finite ones as those where equality always holds
in the formula of Lemma 120.8.
Lemma 123.1. Let A B be an extension of domains. Assume
(1) A is a local Noetherian ring of dimension 1,
(2) A B is of finite type, and
(3) the extension K = f.f.(A) L = f.f.(B) is a finite field extension.
Then B is semi-local. Let x mA , x 6= 0. Let mi , i = 1, . . . , n be the maximal
ideals of B. Then
X
[L : K]ordA (x)
[(mi ) : (mA )]ordBmi (x)
i

where ord is defined as in Definition 120.2. We have equality if and only if A B

is finite.
Proof. The ring B is semi-local by Lemma 112.2. Let B 0 be the integral closure
of A in B. By Lemma 122.15 we can find a finite A-subalgebra C B 0 such that
on setting ni = C mi we have Cni
= Bmi and the primes n1 , . . . , nn are pairwise

COMMUTATIVE ALGEBRA

282

distinct. The ring C is semi-local by Lemma 112.2. Let pj , j = 1, . . . , m be the

other maximal ideals of C (the missing points). By Lemma 120.8 we have
X
X
ordA (x[L:K] ) =
[(ni ) : (mA )]ordCni (x) +
[(pj ) : (mA )]ordCpj (x)
i

hence the inequality follows. In case of equality we conclude that m = 0 (no

missing points). Hence C B is an inclusion of semi-local rings inducing a
bijection on maximal ideals and an isomorphism on all localizations at maximal
ideals. So if b B, then I = {x C | xb C} is an ideal of C which is not
contained in any of the maximal ideals of C, and hence I = C, hence b C. Thus
B = C and B is finite over A.

Here is a more standard application of Zariskis main theorem to the structure of
local homomorphisms of local rings.
052V

Lemma 123.2. Let (R, mR ) (S, mS ) be a local homomorphism of local rings.

Assume
(1) R S is essentially of finite type,
(2) (mR ) (mS ) is finite, and
(3) dim(S/mR S) = 0.
Then S is the localization of a finite R-algebra.
Proof. Let S 0 be a finite type R-algebra such that S = Sq0 0 for some prime q0 of
S 0 . By Definition 121.3 we see that R S 0 is quasi-finite at q0 . After replacing
S 0 by Sg0 0 for some g 0 S 0 , g 0 6 q0 we may assume that R S 0 is quasi-finite,
see Lemma 122.14. Then by Lemma 122.15 there exists a finite R-algebra S 00 and
elements g 0 S 0 , g 0 6 q0 and g 00 S 00 such that Sg0 0
= Sg0000 as R-algebras. This
proves the lemma.


07NC

Lemma 123.3. Let R S be a ring map, q a prime of S lying over p in R. If

(1) R is Noetherian,
(2) R S is of finite type, and
(3) R S is quasi-finite at q,
then Rp R S = Sq B for some Rp -algebra B.
Proof. There exists a finite R-algebra S 0 S and an element g S 0 , g 6 q0 = S 0 q
such that Sg0 = Sg and in particular Sq0 0 = Sq , see Lemma 122.15. We have
Rp R S 0 = (Sq0 0 ) B 0
by Lemma 96.8. Note that we have a commutative diagram
Rp R S
O

/ Sq
O

Rp R S 0

/ (S 0 0 )
q

where the right vertical is an isomorphism and the lower horizontal arrow is the
projection map of the product decomposition above. The lemma follows.


COMMUTATIVE ALGEBRA

283

124. Dimension of fibres

00QC

We study the behaviour of dimensions of fibres, using Zariskis main theorem.

Recall that we defined the dimension dimx (X) of a topological space X at a point
x in Topology, Definition 9.1.

00QD

Definition 124.1. Suppose that R S is of finite type, and let q S be a prime

lying over a prime p of R. We define the relative dimension of S/R at q, denoted
dimq (S/R), to be the dimension of Spec(S R (p)) at the point corresponding to
q. We let dim(S/R) be the supremum of dimq (S/R) over all q. This is called the
relative dimension of S/R.
In particular, R S is quasi-finite at q if and only if dimq (S/R) = 0. The following
lemma is more or less a reformulation of Zariskis Main Theorem.

00QE

Lemma 124.2. Let R S be a finite type ring map. Let q S be a prime.

Suppose that dimq (S/R) = n. There exists a g S, g 6 q such that Sg is quasifinite over a polynomial algebra R[t1 , . . . , tn ].
Proof. The ring S = S R (p) is of finite type over (p). Let q be the prime of S
corresponding to q. By definition of the dimension of a topological space at a point
there exists an open U Spec(S) with q U and dim(U ) = n. Since the topology
on Spec(S) is induced from the topology on Spec(S) (see Remark 16.8), we can
find a g S, g 6 q with image g S such that D(g) U . Thus after replacing S
by Sg we see that dim(S) = n.
Next, choose generators x1 , . . . , xN for S as an R-algebra. By Lemma 114.4 there
exist elements y1 , . . . , yn in the Z-subalgebra of S generated by x1 , . . . , xN such
that the map R[t1 , . . . , tn ] S, ti 7 yi has the property that (p)[t1 . . . , tn ] S
is finite. In particular, S is quasi-finite over R[t1 , . . . , tn ] at q. Hence, by Lemma
122.14 we may replace S by Sg for some g S, g 6 q such that R[t1 , . . . , tn ] S
is quasi-finite.


0520

Lemma 124.3. Let R S be a ring map. Let q S be a prime lying over the
prime p of R. Assume
(1) R S is of finite type,
(2) dimq (S/R) = n, and
(3) trdeg(p) (q) = r.
Then there exist f R, f 6 p, g S, g 6 q and a quasi-finite ring map
: Rf [x1 , . . . , xn ] Sg
1

such that

(qSg ) = (p, xr+1 , . . . , xn )Rf [xr+1 , . . . , xn ]

Proof. After replacing S by a principal localization we may assume there exists a

quasi-finite ring map : R[t1 , . . . , tn ] S, see Lemma 124.2. Set q0 = 1 (q). Let
q0 (p)[t1 , . . . , tn ] be the prime corresponding to q0 . By Lemma 114.6 there exists
a finite ring map (p)[x1 , . . . , xn ] (p)[t1 , . . . , tn ] such that the inverse image of
q0 is (xr+1 , . . . , xn ). Let hi (p)[t1 , . . . , tn ] be the image of xi . We can find an
element f R, f 6 p and hi Rf [t1 , . . . , tn ] which map to hi in (p)[t1 , . . . , tn ].
Then the ring map
Rf [x1 , . . . , xn ] Rf [t1 , . . . , tn ]
becomes finite after tensoring with (p). In particular, Rf [t1 , . . . , tn ] is quasifinite over Rf [x1 , . . . , xn ] at the prime q0 Rf [t1 , . . . , tn ]. Hence, by Lemma 122.14

COMMUTATIVE ALGEBRA

284

there exists a g Rf [t1 , . . . , tn ], g 6 q0 Rf [t1 , . . . , tn ] such that Rf [x1 , . . . , xn ]

Rf [t1 , . . . , tn , 1/g] is quasi-finite. Thus we see that the composition
Rf [x1 , . . . , xn ] Rf [t1 , . . . , tn , 1/g] S(g)
is quasi-finite and we win.
00QF

Lemma 124.4. Let R S be a finite type ring map. Let q S be a prime lying
over p R. If R S is quasi-finite at q, then dim(Sq ) dim(Rp ).
Proof. If Rp is Noetherian (and hence Sq Noetherian since it is essentially of finite
type over Rp ) then this follows immediately from Lemma 111.6 and the definitions.
In the general case we can use Zariskis Main Theorem 122.13 to write Sq = Sq0 0 for
some ring S 0 integral over Rp . Thus the result follows from Lemma 111.3.


00QG

Lemma 124.5. Let k be a field. Let S be a finite type k-algebra. Suppose there
is a quasi-finite k-algebra map k[t1 , . . . , tn ] S. Then dim(S) n.
Proof. By Lemma 113.1 the dimension of any local ring of k[t1 , . . . , tn ] is at most
n. Thus the result follows from Lemma 124.4.


00QH

Lemma 124.6. Let R S be a finite type ring map. Let q S be a prime.

Suppose that dimq (S/R) = n. There exists an open neighbourhood V of q in Spec(S)
such that dimq0 (S/R) n for all q0 V .
Proof. By Lemma 124.2 we see that we may assume that S is quasi-finite over
a polynomial algebra R[t1 , . . . , tn ]. Considering the fibres, we reduce to Lemma
124.5.

In other words, the lemma says that the set of points where the fibre has dimension
n is open in Spec(S). The next lemma says that formation of this open commutes
with base change. If the ring map is of finite presentation then this set is quasicompact open (see below).

00QI

Lemma 124.7. Let R S be a finite type ring map. Let R R0 be any ring
map. Set S 0 = R0 R S and denote f : Spec(S 0 ) Spec(S) the associated map on
spectra. Let n 0. The inverse image f 1 ({q Spec(S) | dimq (S/R) n}) is
equal to {q0 Spec(S 0 ) | dimq0 (S 0 /R0 ) n}.
Proof. The condition is formulated in terms of dimensions of fibre rings which are
of finite type over a field. Combined with Lemma 115.6 this yields the lemma. 

00QJ

Lemma 124.8. Let R S be a ring homomorphism of finite presentation. Let

n 0. The set
Vn = {q Spec(S) | dimq (S/R) n}
is a quasi-compact open subset of Spec(S).
Proof. It is open by Lemma 124.6. Let S = R[x1 , . . . , xn ]/(f1 , . . . , fm ) be a presentation of S. Let R0 be the Z-subalgebra of R generated by the coefficients of
the polynomials fi . Let S0 = R0 [x1 , . . . , xn ]/(f1 , . . . , fm ). Then S = R R0 S0 .
By Lemma 124.7 Vn is the inverse image of an open V0,n under the quasi-compact
continuous map Spec(S) Spec(S0 ). Since S0 is Noetherian we see that V0,n is
quasi-compact.


COMMUTATIVE ALGEBRA

00QK

285

Lemma 124.9. Let R be a valuation ring with residue field k and field of fractions
K. Let S be a domain containing R such that S is of finite type over R. If S R k
is not the zero ring then
dim(S R k) = dim(S R K)
In fact, Spec(S R k) is equidimensional.
Proof. It suffices to show that dimq (S/k) is equal to dim(S R K) for every prime
q of S containing mR S. Pick such a prime. By Lemma 124.6 the inequality
dimq (S/k) dim(S R K) holds. Set n = dimq (S/k). By Lemma 124.2 after
replacing S by Sg for some g S, g 6 q there exists a quasi-finite ring map
R[t1 , . . . , tn ] S. If dim(S R K) < n, then K[t1 , . . . , tn ] S R K has a nonzero
P
kernel. Say f =
aI ti11 . . . tinn . After dividing f by a nonzero coefficient of f with
minimal valuation, we may assume f R[t1 , . . . , tn ] and some aI does not map to
zero in k. Hence the ring map k[t1 , . . . , tn ] S R k has a nonzero kernel which
implies that dim(S R k) < n. Contradiction.

125. Algebras and modules of finite presentation

05N4

In this section we discuss some standard results where the key feature is that the
assumption involves a finite type or finite presentation assumption.

00QP

Lemma 125.1. Let R S be a ring map. Let R R0 be a faithfully flat ring

map. Set S 0 = R0 R S. Then R S is of finite type if and only if R0 S 0 is of
finite type.
Proof. It is clear that if R S is of finite type then R0 S 0 is of finite type.
0
0
Assume
Say y1 , . . . , ym generate S 0 over R0 . Write
P that R S is of finite type.
0
yj = i aij xji for some aij R and xji S. Let A S be the R-subalgebra
generated by the xij . By flatness we have A0 := R0 R A S 0 , and by construction
yj A0 . Hence A0 = S 0 . By faithful flatness A = S.


00QQ

Lemma 125.2. Let R S be a ring map. Let R R0 be a faithfully flat ring

map. Set S 0 = R0 R S. Then R S is of finite presentation if and only if R0 S 0
is of finite presentation.
Proof. It is clear that if R S is of finite presentation then R0 S 0 is of finite
presentation. Assume that R0 S 0 is of finite presentation. By Lemma 125.1
we see that R S is of finite type. Write S = R[x1 , . . . , xn ]/I. By flatness
S 0 = R0 [x1 , P
. . . , xn ]/R0 I. Say g1 , . . . , gm generate R0 I over R0 [x1 , . . . , xn ].
Write gj = i aij fji for some aij R0 and fji I. Let J I be the ideal
generated by the fij . By flatness we have R0 R J R0 R I, and both are ideals
over R0 [x1 , . . . , xn ]. By construction gj R0 R J. Hence R0 R J = R0 R I. By
faithful flatness J = I.


05N5

Lemma 125.3. Let R be a ring. Let I R be an ideal. Let S R be a

multiplicative subset. Set R0 = S 1 (R/I) = S 1 R/S 1 I.
(1) For any finite R0 -module M 0 there exists a finite R-module M such that
S 1 (M/IM )
= M 0.
(2) For any finitely presented R0 -module M 0 there exists a finitely presented
R-module M such that S 1 (M/IM )
= M 0.

COMMUTATIVE ALGEBRA

286

Proof. Proof of (1). Choose a short exact sequence 0 K 0 (R0 )n M 0 0.

Let K Rn be the inverse image of K 0 under the map Rn (R0 )n . Then
M = Rn /K works.
Proof of (2). Choose a presentation (R0 )m (R0 )n M 0 0. Suppose that
the first map is given by the matrix A0 = (a0ij ) and the second map is determined
by generators x0i M 0 , i = 1, . . . , n. As R0 = S 1 (R/I) we can choose s S and
a matrix A = (aij ) with coefficients in R such that a0ij = aij /s mod S 1 I. Let
M be the finitely presented R-module with presentation Rm Rn M 0
where the first map is given by the matrix A and the second map is determined
by generators xi M , i = 1, . . . , n. Then the map M M 0 , xi 7 x0i induces an
isomorphism S 1 (M/IM )

= M 0.
05N6

Lemma 125.4. Let R be a ring. Let S R be a multiplicative subset. Let M be

an R-module.
(1) If S 1 M is a finite S 1 R-module then there exists a finite R-module M 0
and a map M 0 M which induces an isomorphism S 1 M 0 S 1 M .
(2) If S 1 M is a finitely presented S 1 R-module then there exists an R-module
M 0 of finite presentation and a map M 0 M which induces an isomorphism S 1 M 0 S 1 M .
Proof. Proof of (1). Let x1 , . . . , xn M be elements which generate S 1 M as an
S 1 R-module. Let M 0 be the R-submodule of M generated by x1 , . . . , xn .
Proof of (2). Let x1 , . . . , xn M be elements which generate S 1 M as an S 1 Rmodule.
Let K = Ker(Rn M ) where the map is given by the rule (a1 , . . . , an ) 7
P
ai xi . By Lemma 5.3 we see that S 1 K is a finite S 1 R-module. By (1) we can
find a finite submodule K 0 K with S 1 K 0 = S 1 K. Take M 0 = Coker(K 0
Rn ).


05GJ

Lemma 125.5. Let R be a ring. Let p R be a prime ideal. Let M be an

R-module.
(1) If Mp is a finite Rp -module then there exists a finite R-module M 0 and a
map M 0 M which induces an isomorphism Mp0 Mp .
(2) If Mp is a finitely presented Rp -module then there exists an R-module M 0
of finite presentation and a map M 0 M which induces an isomorphism
Mp0 Mp .
Proof. This is a special case of Lemma 125.4

00QR

Lemma 125.6. Let : R S be a ring map. Let q S be a prime lying over

p R. Assume
(1) S is of finite presentation over R,
(2) induces an isomorphism Rp
= Sq .
Then there exist f R, f 6 p and an Rf -algebra C such that Sf
= Rf C as
Rf -algebras.
Proof. Write S = R[x1 , . . . , xn ]/(g1 , . . . , gm ). Let ai Rp be an element mapping
to the image of xi in Sq . Write ai = bi /f for some f R, f 6 p. After replacing R
by Rf and xi by xi ai we may assume that S = R[x1 , . . . , xn ]/(g1 , . . . , gm ) such
that xi maps to zero in Sq . Then if cj denotes the constant term of gi we conclude
that ci maps to zero in Rp . After another replacement of R we may assume that

COMMUTATIVE ALGEBRA

287

the constant coefficients cj of the gj are zero. Thus we obtain an R-algebra map
S R, xi 7 0 whose kernel is the ideal (x1 , . . . , xn ).
P
Note that q = pS + (x1 , . . . , xn ). Write gj =
aji xi + h.o.t.. Since Sq = Rp
we have p (p) = q (q). It follows that m n matrix A = (aij ) defines a
surjective map (p)m (p)n . Thus after
P inverting some element of R not in p
we may assume there are bij R such that
bij gj = xi + h.o.t.. We conclude that
(x1 , . . . , xn ) = (x1 , . . . , xn )2 in S. It follows from Lemma 20.5 that (x1 , . . . , xn ) is
generated by an idempotent e. Setting C = eS finishes the proof.

00QS

Lemma 125.7. Let R be a ring. Let S, S 0 be of finite presentation over R. Let

q S and q0 S 0 be primes. If Sq
= Sq0 as R-algebras, then there exist g S,
g 6 q and g 0 S 0 , g 0 6 q0 such that Sg
= Sg0 0 as R-algebras.
Proof. Let : Sq Sq0 be the isomorphism of the hypothesis of the lemma.
Write S = R[x1 , . . . , xn ]/(f1 , . . . , fr ) and S 0 = R[y1 , . . . , ym ]/J. For each i =
1, . . . , n choose a fraction hi /gi with hi , gi R[y1 , . . . , ym ] and gi mod J not in
q0 which represents the image of xi under . After replacing S 0 by Sg0 1 ...gn and
R[y1 , . . . , ym , ym+1 ] (mapping ym+1 to 1/(g1 . . . gn )) we may assume that (xi ) is
the image of some hi R[y1 , . . . , ym ]. Consider the elements fj (h1 , . . . , hn )
R[y1 , . . . , ym ]. Since kills each fj we see that there exists a g R[y1 , . . . , ym ],
g mod J 6 q0 such that gfj (h1 , . . . , hn ) J for each j = 1, . . . , r. After replacing
S 0 by Sg0 and R[y1 , . . . , ym , ym+1 ] as before we may assume that fj (h1 , . . . , hn ) J.
Thus we obtain a ring map S S 0 , xi 7 hi which induces on local rings. By
Lemma 6.2 the map S 0 S is of finite presentation. By Lemma 125.6 we may
assume that S = S 0 C. Thus localizing S at the idempotent corresponding to the
factor C we obtain the result.


07RD

Lemma 125.8. Let R be a ring. Let I R be a locally nilpotent ideal. Let S S 0

be an R-algebra map such that S S 0 /IS 0 is surjective and such that S 0 is of finite
type over R. Then S S 0 is surjective.
0
Proof. Write
P S J= R[x1 , . . . , xm ]/K for some ideal K. By assumption there exist
gj = xj + j,J x R[x1 , . . . , xn ] with j,J I and with gj mod K Im(S S 0 ).
Hence it suffices to show that g1 , . . . , gm generate R[x1 , . . . , xn ]. Let R0 R be
a finitely generated Z-subalgebra of R containing at least the j,J . Then R0 I
is a nilpotent ideal (by Lemma 31.4). It follows that R0 [x1 , . . . , xn ] is generated
by g1 , . . . , gm (because xj 7 gj defines an automorphism of R0 [x1 , . . . , xm ]; details
omitted). Since R is the union of the subrings R0 we win.


087P

Lemma 125.9. Let R be a ring. Let I R be an ideal. Let S S 0 be an

R-algebra map. Let IS q S be a prime ideal. Assume that
(1) S S 0 is surjective,
(2) Sq /ISq Sq0 /ISq0 is an isomorphism,
(3) S is of finite type over R,
(4) S 0 of finite presentation over R, and
(5) Sq0 is flat over R.
Then Sg Sg0 is an isomorphism for some g S, g 6 q.
Proof. Let J = Ker(S S 0 ). By Lemma 6.2 J is a finitely generated ideal.
Since Sq0 is flat over R we see that Jq /IJq Sq /ISq (apply Lemma 38.12 to
0 J S S 0 0). By assumption (2) we see that Jq /IJq is zero. By

COMMUTATIVE ALGEBRA

288

Nakayamas lemma (Lemma 19.1) we see that there exists a g S, g 6 q such that
Jg = 0. Hence Sg

= Sg0 as desired.
07RE

Lemma 125.10. Let R be a ring. Let I R be an ideal. Let S S 0 be an

R-algebra map. Assume that
(1)
(2)
(3)
(4)
(5)

I is locally nilpotent,
S/IS S 0 /IS 0 is an isomorphism,
S is of finite type over R,
S 0 of finite presentation over R, and
S 0 is flat over R.

Then S S 0 is an isomorphism.
Proof. By Lemma 125.8 the map S S 0 is surjective. As I is locally nilpotent, so
are the ideals IS and IS 0 (Lemma 31.2). Hence every prime ideal q of S contains
IS and (trivially) Sq /ISq
= Sq0 /ISq0 . Thus Lemma 125.9 applies and we see that
0
Sq Sq is an isomorphism for every prime q S. It follows that S S 0 is
injective for example by Lemma 23.1.

126. Colimits and maps of finite presentation
00QL

In this section we prove some preliminary lemmas which will eventually help us
prove result using absolute Noetherian reduction. We begin discussing how we will
think about colimits in this section.
Let (, ) a partially ordered set. A system of rings over is given by a ring R
for every , and a morphism R R whenever . These morphisms
have to satisfy the rule that R R R is equal to the map R R for all
. See Categories, Section 21. We will often assume that (I, ) is directed,
which means that is nonempty and given , there exists a with
and . Recall that the colimit colim R is sometimes called a direct
limit in this case (but we will not use this terminology).

00QN

Lemma 126.1. Let R A be a ring map. There exists a directed system A

of R-algebras of finite presentation such that A = colim A . If A is of finite type
over R we may arrange it so that all the transition maps are surjective.
Proof. Compare with the proof of Lemma 8.13. Consider any finite subset S A,
and any finite collection of polynomial relations E among the elements of S. So
each s S corresponds to xs A and each e E consists of a polynomial fe
R[Xs ; s S] such that fe (xs ) = 0. Let AS,E = R[Xs ; s S]/(fe ; e E) which is
a finitely presented R-algebra. There are canonical maps AS,E A. If S S 0
and if the elements of E correspond, via the map R[Xs ; s S] R[Xs ; s S 0 ], to
a subset of E 0 , then there is an obvious map AS,E AS 0 ,E 0 commuting with the
maps to A. Thus, setting equal the set of pairs (S, E) with ordering by inclusion
as above, we get a directed partially ordered set. It is clear that the colimit of this
directed system is A.
For the last statement, suppose A = R[x1 , . . . , xn ]/I. In this case, consider the
subset 0 consisting of those systems (S, E) above with S = {x1 , . . . , xn }. It is
easy to see that still A = colim0 0 A0 . Moreover, the transition maps are clearly
surjective.


COMMUTATIVE ALGEBRA

289

It turns out that we can characterize ring maps of finite presentation as follows.
This in some sense says that the algebras of finite presentation are the compact
objects in the category of R-algebras.
00QO

Lemma 126.2. Let : R S be a ring map. Then is of finite presentation if

and only if for every directed system A of R-algebras we have
colim HomR (S, A ) = HomR (S, colim A )
Proof. Suppose S = R[x1 , . . . , xn ]/(f1 , . . . , fm ). If : S colim A is a map,
then each xi maps to some element in the image of some Ai . We may pick i ,
i = 1, . . . , n and assume (xi ) is the image of yi A for i = 1, . . . , n. Consider
zj = fj (y1 , . . . , yn ) A . Since is a homomorphism the image of zj in colim A
is zero. Hence there exists a j such that zj maps to zero in Aj . Pick j ,
j = 1, . . . , m. Then the images of z1 , . . . , zm are zero in A . This exactly means
that the yi map to elements yi0 A which satisfy the relations fj (y10 , . . . , yn0 ) = 0.
Thus we obtain a ring map S A as desired.
Conversely, suppose the displayed formula holds always. By Lemma 126.1 we may
write S = colim S with S of finite presentation over R. Then the identity map
factors as
S S S
for some . This implies that S is finitely presented over S by Lemma 6.2 part
(4) applied to S S S. Applying part (2) of the same lemma to R S S
we conclude that S is of finite presentation over R.

But more is true. Namely, given R = colim R we see that the category of finitely
presented R-modules is equivalent to the limit of the category of finitely presented
R -modules. Similarly for the categories of finitely presented R-algebras.

05LI

Lemma 126.3. Let A be a ring and let M, N be A-modules. Suppose that R =

colimiI Ri is a directed colimit of A-algebras.
(1) If M is a finite A-module, and u, u0 : M N are A-module maps such
that u 1 = u0 1 : M A R N A R then for some i we have u 1 =
u0 1 : M A Ri N A Ri .
(2) If N is a finite A-module and u : M N is an A-module map such
that u 1 : M A R N A R is surjective, then for some i the map
u 1 : M A Ri N A Ri is surjective.
(3) If N is a finitely presented A-module, and v : N A R M A R is an Rmodule map, then there exists an i and an Ri -module map vi : N A Ri
M A Ri such that v = vi 1.
(4) If M is a finite A-module, N is a finitely presented A-module, and u :
M N is an R-module map such that u 1 : M A R N A R is an
isomorphism, then for some i the map u 1 : M A Ri N A Ri is an
isomorphism.
Proof. To prove (1) assume u is as in (1) and let x1 , . . . , xm M be generators.
Since N A R = colimi N A Ri we may pick an i I such that u(xj )1 = u0 (xj )1
in M A Ri , j = 1, . . . , m. For such an i we have u1 = u0 1 : M A Ri N A Ri .
To prove (2) assume u 1 surjective and let y1 , . . . , ym N be generators. Since
N A R = colimi N A Ri we may pick an i I and zj M A Ri , j = 1, . . . , m

COMMUTATIVE ALGEBRA

290

whose images in N A R equal yj 1. For such an i the map u 1 : M A Ri

N A Ri is surjective.
m
To prove (3) let y1 , . . . , ym N be generators.
N ) where
P Let K = Ker(A
the map is given by the rule (a1 , . . . , am ) 7
aj xj . Let k1 , . . . , kt be generators
for K. Say ks = (ks1 , . . . , ksm ). Since M A R = colimi M A Ri we may pick an
i I and zj M A Ri , j = 1, . . . , m whose images in M A R equal v(yj 1). We
want to use the zj to define the map vi : N A Ri M A Ri . Since K
PA Ri
Rim N A Ri 0 is a presentation, it suffices to check that s = j ksj zj is
zero in M A Ri for each s = 1, . . . , t. This may not be the case, but since the
image of s in M A R is zero we see that it will be the case after increasing i a bit.

To prove (4) assume u 1 is an isomorphism, that M is finite, and that N is finitely

presented. Let v : N A R M A R be an inverse to u 1. Apply part (3) to
get a map vi : N A Ri M A Ri for some i. Apply part (1) to see that, after
increasing i we have vi (u 1) = idM R Ri and (u 1) vi = idN R Ri .

05N7

Lemma 126.4. Suppose that R = colimiI Ri is a directed colimit of rings. Then

the category of finitely presented R-modules is the colimit of the categories of finitely
presented R -modules. More precisely
(1) Given a finitely presented R-module M there exists a and a finitely
presented R -module M such that M
= M R R.
(2) Given a , finitely presented R -modules M , N , and an R-module
map : M R R N R R, then there exists a and an R module map : M R R N R R such that = 1R .
(3) Given a , finitely presented R -modules M , N , and R-module maps
, : M N such that 1R = 1R , then 1R = 1R for
some .
Proof. To prove (1) choose a presentation Rm Rn M 0. Suppose that
the first map is given by the matrix A = (aij ). We can choose a and a matrix
A = (a,ij ) with coefficients in R which maps to A in R. Then we simply let
M be the R -module with presentation Rm Rn M 0 where the first
arrow is given by A .
Parts (2) and (3) follow from Lemma 126.3.

05N8

Lemma 126.5. Let A be a ring and let B, C be A-algebras. Suppose that R =

colimiI Ri is a directed colimit of A-algebras.
(1) If B is a finite type A-algebra, and u, u0 : B C are A-algebra maps
such that u 1 = u0 1 : B A R C A R then for some i we have
u 1 = u 0 1 : B A R i C A R i .
(2) If C is a finite type A-algebra and u : B C is an A-algebra map such
that u 1 : B A R C A R is surjective, then for some i the map
u 1 : B A Ri C A Ri is surjective.
(3) If C is of finite presentation over A and v : C A R B A R is an Ralgebra map, then there exists an i and an Ri -algebra map vi : C A Ri
B A Ri such that v = vi 1.
(4) If B is a finite type A-algebra, C is a finitely presented A-algebra, and
u 1 : B A R C A R is an isomorphism, then for some i the map
u 1 : B A Ri C A Ri is an isomorphism.

COMMUTATIVE ALGEBRA

291

Proof. To prove (1) assume u is as in (1) and let x1 , . . . , xm B be generators.

Since B A R = colimi B A Ri we may pick an i I such that u(xj )1 = u0 (xj )1
in B A Ri , j = 1, . . . , m. For such an i we have u1 = u0 1 : B A Ri C A Ri .
To prove (2) assume u 1 surjective and let y1 , . . . , ym C be generators. Since
BA R = colimi BA Ri we may pick an i I and zj BA Ri , j = 1, . . . , m whose
images in C A R equal yj 1. For such an i the map u 1 : B A Ri C A Ri
is surjective.
To prove (3) let c1 , . . . , cm C be generators.
Let K = Ker(A[x1 , . . . , xm ] N )
P
where the map is given by the rule xj 7 cj . Let f1 , . . . , ft be generators for K as
an ideal in A[x1 , . . . , xm ]. We think of fj = fj (x1 , . . . , xm ) as a polynomial. Since
B A R = colimi B A Ri we may pick an i I and zj B A Ri , j = 1, . . . , m
whose images in B A R equal v(cj 1). We want to use the zj to define a map
vi : C A Ri B A Ri . Since K A Ri Ri [x1 , . . . , xm ] C A Ri 0 is
a presentation, it suffices to check that s = fj (z1 , . . . , zm ) is zero in B A Ri for
each s = 1, . . . , t. This may not be the case, but since the image of s in B A R
is zero we see that it will be the case after increasing i a bit.
To prove (4) assume u 1 is an isomorphism, that B is a finite type A-algebra, and
that C is a finitely presented A-algebra. Let v : B A R C A R be an inverse
to u 1. Let vi : C A Ri B A Ri be as in part (3). Apply part (1) to see that,
after increasing i we have vi (u 1) = idBR Ri and (u 1) vi = idCR Ri .

05N9

Lemma 126.6. Suppose that R = colimiI Ri is a directed colimit of rings. Then

the category of finitely presented R-algebras is the colimit of the categories of finitely
presented R -algebras. More precisely
(1) Given a finitely presented R-algebra A there exists a and a finitely
presented R -algebra A such that A
= A R R.
(2) Given a , finitely presented R -algebras A , B , and an R-algebra
map : A R R B R R, then there exists a and an R algebra map : A R R B R R such that = 1R .
(3) Given a , finitely presented R -algebras A , B , and R-algebra maps
, : A B such that 1R = 1R , then 1R = 1R for
some .
Proof. To prove (1) choose a presentation A = R[x1 , . . . , xn ]/(f1 , . . . , fm ). We can
choose a and elements f,j R [x1 , . . . , xn ] mapping to fj R[x1 , . . . , xn ].
Then we simply let A = R [x1 , . . . , xn ]/(f,1 , . . . , f,m ).
Parts (3) and (4) follow from Lemma 126.5.

00QT

Lemma 126.7. Suppose R S is a local homomorphism of local rings. There

exists a directed set (, ), and a system of local homomorphisms R S of local
rings such that
(1) The colimit of the system R S is equal to R S.
(2) Each R is essentially of finite type over Z.
(3) Each S is essentially of finite type over R .
Proof. Denote : R S the ring map. Let m R be the maximal ideal of R
and let n S be the maximal ideal of S. Let
= {(A, B) | A R, B S, #A < , #B < , (A) B}.

COMMUTATIVE ALGEBRA

292

As partial ordering we take the inclusion relation. For each = (A, B) we let
R0 be the sub Z-algebra generated by a A, and we let S0 be the sub Z-algebra
generated by b, b B. Let R be the localization of R0 at the prime ideal R0 m
and let S be the localization of S0 at the prime ideal S0 n. In a picture
BO

/ S0
O

/ S
O

/S .
O

/ R0

/ R

/R

The transition maps are clear. We leave the proofs of the other assertions to the
reader.

00QU

Lemma 126.8. Suppose R S is a local homomorphism of local rings. Assume

that S is essentially of finite type over R. Then there exists a directed set (, ),
and a system of local homomorphisms R S of local rings such that
(1) The colimit of the system R S is equal to R S.
(2) Each R is essentially of finite type over Z.
(3) Each S is essentially of finite type over R .
(4) For each the map S R R S presents S as the localization
of a quotient of S R R .
Proof. Denote : R S the ring map. Let m R be the maximal ideal of R and
let n S be the maximal ideal of S. Let x1 , . . . , xn S be elements such that S is
a localization of the sub R-algebra of S generated by x1 , . . . , xn . In other words, S
is a quotient of a localization of the polynomial ring R[x1 , . . . , xn ].
Let = {A R | #A < } be the set of finite subsets of R. As partial ordering
we take the inclusion relation. For each = A we let R0 be the sub Z-algebra
generated by a A, and we let S0 be the sub Z-algebra generated by (a), a A
and the elements x1 , . . . , xn . Let R be the localization of R0 at the prime ideal
R0 m and let S be the localization of S0 at the prime ideal S0 n. In a picture
(A) q {xi }
O

/ S0
O

/ S
O

/S
O

/ R0

/ R

/R

It is clear that if A B corresponds to in , then there are canonical maps

R R , and S S and we obtain a system over the directed set .
The assertion that R = colim R is clear because all the maps R R are injective
and any element of R eventually is in the image. The same argument works for
S = colim S . Assertions (2), (3) are true by construction. The final assertion
holds because clearly the maps S0 R0 R0 S0 are surjective.

00QV

Lemma 126.9. Suppose R S is a local homomorphism of local rings. Assume

that S is essentially of finite presentation over R. Then there exists a directed set
(, ), and a system of local homomorphism R S of local rings such that
(1) The colimit of the system R S is equal to R S.
(2) Each R is essentially of finite type over Z.
(3) Each S is essentially of finite type over R .

COMMUTATIVE ALGEBRA

293

(4) For each the map S R R S presents S as the localization

of S R R at a prime ideal.
Proof. By assumption we may choose an isomorphism : (R[x1 , . . . , xn ]/I)q S
where I R[x1 , . . . , xn ] is a finitely generated ideal, and q R[x1 , . . . , xn ]/I
is a prime. (Note that R q is equal to the maximal ideal m of R.) We also
choose generators f1 , . . . , fm I for the ideal I. Write R in any way as a colimit
R = colim R over a directed set (, ), with each R local and essentially of
finite type over Z. There exists some 0 such that fj is the image of some
fj,0 R0 [x1 , . . . , xn ]. For all 0 denote fj, R [x1 , . . . , xn ] the image of
fj,0 . Thus we obtain a system of ring maps
R [x1 , . . . , xn ]/(f1, , . . . , fn, ) R[x1 , . . . , xn ]/(f1 , . . . , fn ) S
Set q the inverse image of q. Set S = (R [x1 , . . . , xn ]/(f1, , . . . , fn, ))q . We
leave it to the reader to see that this works.

00QW

Remark 126.10. Suppose that R S is a local homomorphism of local rings,

which is essentially of finite presentation. Take any system (, ), R S with
the properties listed in Lemma 126.8. What may happen is that this is the wrong
system, namely, it may happen that property (4) of Lemma 126.9 is not satisfied.
Here is an example. Let k be a field. Consider the ring
R = k[[z, y1 , y2 , . . .]]/(yi2 zyi+1 ).
Set S = R/zR. As system take = N and Rn = k[[z, y1 , . . . , yn ]]/({yi2
zyi+1 }in1 ) and Sn = Rn /(z, yn2 ). All the maps Sn Rn Rn+1 Sn+1 are not
2
localizations (i.e., isomorphisms in this case) since 1 yn+1
maps to zero. If we
0
0
0
are isomorphisms.
take instead Sn = Rn /zRn then the maps Sn Rn Rn+1 Sn+1
The moral of this remark is that we do have to be a little careful in choosing the
systems.

00QX

Lemma 126.11. Suppose R S is a local homomorphism of local rings. Assume

that S is essentially of finite presentation over R. Let M be a finitely presented
S-module. Then there exists a directed set (, ), and a system of local homomorphisms R S of local rings together with S -modules M , such that
(1) The colimit of the system R S is equal to R S. The colimit of the
system M is M .
(2) Each R is essentially of finite type over Z.
(3) Each S is essentially of finite type over R .
(4) Each M is finite over S .
(5) For each the map S R R S presents S as the localization
of S R R at a prime ideal.
(6) For each the map M S S M is an isomorphism.
Proof. As in the proof of Lemma 126.9 we may first write R = colim R as a
directed colimit of local Z-algebras which are essentially of finite type. Next, we
may assume that for some 1 there exist fj,1 R1 [x1 , . . . , xn ] such that
S = colim1 S , with S = (R [x1 , . . . , xn ]/(f1, , . . . , fm, ))q
Choose a presentation
S s S t M 0

COMMUTATIVE ALGEBRA

294

of M over S. Let A Mat(t s, S) be the matrix of the presentation. For some

2 , 2 1 we can find a matrix A2 Mat(t s, S2 ) which maps to A. For
A

all 2 we let M = Coker(Ss

St ). We leave it to the reader to see that
this works.


00QY

Lemma 126.12. Suppose R S is a ring map. Then there exists a directed set
(, ), and a system of ring maps R S such that
(1) The colimit of the system R S is equal to R S.
(2) Each R is of finite type over Z.
(3) Each S is of finite type over R .
Proof. This is the non-local version of Lemma 126.7. Proof is similar and left to
the reader.


00QZ

Lemma 126.13. Suppose R S is a ring map. Assume that S is of finite type

over R. Then there exists a directed set (, ), and a system of ring maps R S
such that
(1) The colimit of the system R S is equal to R S.
(2) Each R is of finite type over Z.
(3) Each S is of finite type over R .
(4) For each the map S R R S presents S as a quotient of
S R R .
Proof. This is the non-local version of Lemma 126.8. Proof is similar and left to
the reader.


00R0

Lemma 126.14. Suppose R S is a ring map. Assume that S is of finite

presentation over R. Then there exists a directed set (, ), and a system of ring
maps R S such that
(1) The colimit of the system R S is equal to R S.
(2) Each R is of finite type over Z.
(3) Each S is of finite type over R .
(4) For each the map S R R S is an isomorphism.
Proof. This is the non-local version of Lemma 126.9. Proof is similar and left to
the reader.


00R1

Lemma 126.15. Suppose R S is a ring map. Assume that S is of finite

presentation over R. Let M be a finitely presented S-module. Then there exists a
directed set (, ), and a system of ring maps R S together with S -modules
M , such that
(1) The colimit of the system R S is equal to R S. The colimit of the
system M is M .
(2) Each R is of finite type over Z.
(3) Each S is of finite type over R .
(4) Each M is finite over S .
(5) For each the map S R R S is an isomorphism.
(6) For each the map M S S M is an isomorphism.
In particular, for every we have
M = M S S = M R R.

COMMUTATIVE ALGEBRA

295

Proof. This is the non-local version of Lemma 126.11. Proof is similar and left to
the reader.

127. More flatness criteria
00R3

The following lemma is often used in algebraic geometry to show that a finite morphism from a normal surface to a smooth surface is flat. It is a partial converse to
Lemma 111.9 because an injective finite local ring map certainly satisfies condition
(3).

00R4

Lemma 127.1. Let R S be a local homomorphism of Noetherian local rings.

Assume
(1) R is regular,
(2) S Cohen-Macaulay,
(3) dim(S) = dim(R) + dim(S/mR S).
Then R S is flat.
Proof. By induction on dim(R). The case dim(R) = 0 is trivial, because then R
is a field. Assume dim(R) > 0. By (3) this implies that dim(S) > 0. Let q1 , . . . , qr
be the minimal primes of S. Note that qi 6 mR S since
dim(S/qi ) = dim(S) > dim(S/mR S)
the first equality by Lemma 103.3 and the inequality by (3). Thus pi = R qi
is not equal to mR . Pick x m, x 6 m2 , and x 6 pi , see Lemma 14.2. Hence
we see that x is not contained in any of the minimal primes of S. Hence x is a
nonzerodivisor on S by (2), see Lemma 103.2 and S/xS is Cohen-Macaulay with
dim(S/xS) = dim(S) 1. By (1) and Lemma 105.3 the ring R/xR is regular with
dim(R/xR) = dim(R) 1. By induction we see that R/xR S/xS is flat. Hence
we conclude by Lemma 98.10 and the remark following it.


07DY

Lemma 127.2. Let R S be a homomorphism of Noetherian local rings. Assume

that R is a regular local ring and that a regular system of parameters maps to a
regular sequence in S. Then R S is flat.
Proof. Suppose that x1 , . . . , xd are a system of parameters of R which map to
a regular sequence in S. Note that S/(x1 , . . . , xd )S is flat over R/(x1 , . . . , xd )
as the latter is a field. Then xd is a nonzerodivisor in S/(x1 , . . . , xd1 )S hence
S/(x1 , . . . , xd1 )S is flat over R/(x1 , . . . , xd1 ) by the local criterion of flatness
(see Lemma 98.10 and remarks following). Then xd1 is a nonzerodivisor in
S/(x1 , . . . , xd2 )S hence S/(x1 , . . . , xd2 )S is flat over R/(x1 , . . . , xd2 ) by the local criterion of flatness (see Lemma 98.10 and remarks following). Continue till one
reaches the conclusion that S is flat over R.

The following lemma is the key to proving that results for finitely presented modules
over finitely presented rings over a base ring follow from the corresponding results
for finite modules in the Noetherian case.

00R6

Lemma 127.3. Let R S, M , , R S , M be as in Lemma 126.11.

Assume that M is flat over R. Then for some the module M is flat over
R .

COMMUTATIVE ALGEBRA

296

Proof. Pick some and consider

TorR
1 (M , R /m ) = Ker(m R M M ).

See Remark 74.9. The right hand side shows that this is a finitely generated S module (because S is Noetherian and the modules in question are finite). Let
1 , . . . , n be generators. Because M is flat over R we have that 0 = Ker(m R R
M M ). Since commutes with colimits we see there exists a 0 such that
each i maps to zero in m R0 R0 M0 . Hence we see that
R

TorR
1 (M , R /m ) Tor1 (M0 , R0 /m R0 )

is zero. Note that M R R /m is flat over R /m because this last ring is a

field. Hence we may apply Lemma 98.14 to get that M0 is flat over R0 .

Using the lemma above we can start to reprove the results of Section 98 in the
non-Noetherian case.
046Y

Lemma 127.4. Suppose that R S is a local homomorphism of local rings.

Denote m the maximal ideal of R. Let u : M N be a map of S-modules. Assume
(1) S is essentially of finite presentation over R,
(2) M , N are finitely presented over S,
(3) N is flat over R, and
(4) u : M/mM N/mN is injective.
Then u is injective, and N/u(M ) is flat over R.
Proof. By Lemma 126.11 and its proof we can find a system R S of local ring
maps together with maps of S -modules u : M N satisfying the conclusions
(1) (6) for both N and M of that lemma and such that the colimit of the maps
u is u. By Lemma 127.3 we may assume that N is flat over R for all sufficiently
large . Denote m R the maximal ideal and = R /m , resp. = R/m the
residue fields.
Consider the map
: M /m M M/mM.
Since S /m S is essentially of finite type over the field we see that the tensor
product S /m S is essentially of finite type over . Hence it is a Noetherian
ring and we conclude the kernel of is finitely generated. Since M/mM is the
colimit of the system M /m M and is the colimit of the fields there exists a
0 > such that the kernel of is generated by the kernel of
,0 : M /m M 0 M0 /m0 M0 .
By construction there exists a multiplicative subset W S R R0 such that
S0 = W 1 (S R R0 ) and
W 1 (M /m M 0 ) = M0 /m0 M0 .
Now suppose that x is an element of the kernel of
0 : M0 /m0 M0 0 M/mM.
Then for some w W we have wx M /m M . Hence wx Ker( ). Hence
wx is a linear combination of elements in the kernel of ,0 . Hence wx = 0 in
M0 /m0 M0 0 , hence x = 0 because w is invertible in S0 . We conclude that
the kernel of 0 is zero for all sufficiently large 0 !

COMMUTATIVE ALGEBRA

297

By the result of the preceding paragraph we may assume that the kernel of is
zero for all sufficiently large, which implies that the map M /m M M/mM
is injective. Combined with u being injective this formally implies that also u :
M /m M N /m N is injective. By Lemma 98.1 we conclude that (for all
sufficiently large ) the map u is injective and that N /u (M ) is flat over R .
The lemma follows.

046Z

Lemma 127.5. Suppose that R S is a local ring homomorphism of local rings.

Denote m the maximal ideal of R. Suppose
(1) S is essentially of finite presentation over R,
(2) S is flat over R, and
(3) f S is a nonzerodivisor in S/mS.
Then S/f S is flat over R, and f is a nonzerodivisor in S.
Proof. Follows directly from Lemma 127.4.

0470

Lemma 127.6. Suppose that R S is a local ring homomorphism of local rings.

Denote m the maximal ideal of R. Suppose
(1) R S is essentially of finite presentation,
(2) R S is flat, and
(3) f1 , . . . , fc is a sequence of elements of S such that the images f 1 , . . . , f c
form a regular sequence in S/mS.
Then f1 , . . . , fc is a regular sequence in S and each of the quotients S/(f1 , . . . , fi )
is flat over R.
Proof. Induction and Lemma 127.5.

Here is the version of the local criterion of flatness for the case of local ring maps
which are locally of finite presentation.
0471

Lemma 127.7. Let R S be a local homomorphism of local rings. Let I 6= R be

an ideal in R. Let M be an S-module. Assume
(1)
(2)
(3)
(4)

S is essentially of finite presentation over R,

M is of finite presentation over S,
TorR
1 (M, R/I) = 0, and
M/IM is flat over R/I.

Then M is flat over R.

Proof. Let , R S , M be as in Lemma 126.11. Denote I R the inverse
image of I. In this case the system R/I S/IS, M/IM , R S /I S , and
M /I M satisfies the conclusions of Lemma 126.11 as well. Hence by Lemma
127.3 we may assume (after shrinking the index set ) that M /I M is flat for all
. Pick some and consider

TorR
1 (M , R /I ) = Ker(I R M M ).

See Remark 74.9. The right hand side shows that this is a finitely generated
S -module (because S is Noetherian and the modules in question are finite).
Let 1 , . . . , n be generators. Because Tor1R (M, R/I) = 0 and since commutes

COMMUTATIVE ALGEBRA

298

with colimits we see there exists a 0 such that each i maps to zero in
R
Tor1 0 (M0 , R0 /I0 ). The composition of the maps
R0 R Tor1R (M , R /I )


surjective by Lemma 98.12

TorR
1 (M , R0 /I R0 )

surjective up to localization by Lemma 98.13


R
Tor1 0 (M0 , R0 /I R0 )

surjective by Lemma 98.12

Tor1 0 (M0 , R0 /I0 ).

is surjective up to a localization by the reasons indicated. The localization is
necessary since M0 is not equal to M R R0 . Namely, it is equal to M S S0
and S0 is the localization of S R R0 whence the statement up to a localization
(or tensoring with S0 ). Note that Lemma 98.12 applies to the first and third
arrows because M /I M is flat over R /I and because M0 /I M0 is flat over
R0 /I R0 as it is a base change of the flat module M /I M . The composition
maps the generators i to zero as we explained above. We finally conclude that
R
Tor1 0 (M0 , R0 /I0 ) is zero. This implies that M0 is flat over R0 by Lemma
98.10.

Please compare the lemma below to Lemma 98.15 (the case of Noetherian local
rings) and Lemma 100.8 (the case of a nilpotent ideal in the base).
00R7

Lemma 127.8 (Crit`ere de platitude par fibres). Let R, S, S 0 be local rings and
let R S S 0 be local ring homomorphisms. Let M be an S 0 -module. Let m R
be the maximal ideal. Assume
(1) The ring maps R S and R S 0 are essentially of finite presentation.
(2) The module M is of finite presentation over S 0 .
(3) The module M is not zero.
(4) The module M/mM is a flat S/mS-module.
(5) The module M is a flat R-module.
Then S is flat over R and M is a flat S-module.
Proof. As in the proof of Lemma 126.9 we may first write R = colim R as a
directed colimit of local Z-algebras which are essentially of finite type. Denote p
the maximal ideal of R. Next, we may assume that for some 1 there exist
fj,1 R1 [x1 , . . . , xn ] such that
S = colim1 S , with S = (R [x1 , . . . , xn ]/(f1, , . . . , fu, ))q
For some 2 , 2 1 there exist gj,2 R2 [x1 , . . . , xn , y1 , . . . , ym ] with images
g j,2 S2 [y1 , . . . , ym ] such that
S 0 = colim2 S0 , with S0 = (S [y1 , . . . , ym ]/(g 1, , . . . , g v, ))q0
Note that this also implies that
S0 = (R [x1 , . . . , xn , y1 , . . . , ym ]/(g1, , . . . , gv, ))q0

COMMUTATIVE ALGEBRA

299

Choose a presentation
(S 0 )s (S 0 )t M 0
of M over S 0 . Let A Mat(t s, S 0 ) be the matrix of the presentation. For some
3 , 3 2 we can find a matrix A3 Mat(t s, S3 ) which maps to A. For
A

all 3 we let M = Coker((S0 )s

(S0 )t ).

With these choices, we have for each 3 that S R R S is a

localization, S0 S S S0 is a localization, and the map M S0 S M is
an isomorphism. This also implies that S0 R R S0 is a localization. Thus,
since M is flat over R we see by Lemma 127.3 that for all big enough the module
M is flat over R . Moreover, note that m = colim p , S/mS = colim S /p S ,
S 0 /mS 0 = colim S0 /p S0 , and M/mM = colim M /p M . Also, for each 3
we see (from the properties listed above) that
S0 /p S0 S /p S S /p S S0 /p S0
is a localization, and the map
M /p M S0 /p S0 S /p S0 M /p M
is an isomorphism. Hence the system (S /p S S0 /p S0 , M /p M ) is a system
as in Lemma 126.11 as well. We may apply Lemma 127.3 again because M/mM
is assumed flat over S/mS and we see that M /p M is flat over S /p S for all
big enough. Thus for big enough the data R S S0 , M satisfies the
hypotheses of Lemma 98.15. Pick such a . Then S = S R R is flat over R, and
M = M S S0 is flat over S (since the base change of a flat module is flat). 
The following is an easy consequence of the crit`ere de platitude par fibres Lemma
127.8. For more results of this kind see More on Flatness, Section 1.
05UV

Lemma 127.9. Let R, S, S 0 be local rings and let R S S 0 be local ring

homomorphisms. Let M be an S 0 -module. Let m R be the maximal ideal. Assume
(1) R S 0 is essentially of finite presentation,
(2) R S is essentially of finite type,
(3) M is of finite presentation over S 0 ,
(4) M is not zero,
(5) M/mM is a flat S/mS-module, and
(6) M is a flat R-module.
Then S is essentially of finite presentation and flat over R and M is a flat S-module.
Proof. As S is essentially of finite presentation over R we can write S = Cq for
some finite type R-algebra C. Write C = R[x1 , . . . , xn ]/I. Denote q R[x1 , . . . , xn ]
be the prime ideal corresponding to q. Then we see that S = B/J where B =
R[x1 , . . . , xn ]q is essentially of finite presentation over R and J = IB. We can find
f1 , . . . , fk J such that the images f i B/mB generate the image J of J in the
Noetherian ring B/mB. Hence there exist finitely generated ideals J 0 J such
that B/J 0 B/J induces an isomorphism
(B/J 0 ) R R/m B/J R R/m = S/mS.
For any J 0 as above we see that Lemma 127.8 applies to the ring maps
R B/J 0 S 0

COMMUTATIVE ALGEBRA

300

and the module M . Hence we conclude that B/J 0 is flat over R for any choice J 0
as above. Now, if J 0 J 0 J are two finitely generated ideals as above, then we
conclude that B/J 0 B/J 00 is a surjective map between flat R-algebras which are
essentially of finite presentation which is an isomorphism modulo m. Hence Lemma
127.4 implies that B/J 0 = B/J 00 , i.e., J 0 = J 00 . Clearly this means that J is finitely
generated, i.e., S is essentially of finite presentation over R. Thus we may apply
Lemma 127.8 to R S S 0 and we win.

128. Openness of the flat locus
00R8
00R9

Lemma 128.1. Let k be a field. Let S be a finite type k-algebra. Let f1 , . . . , fi be

elements of S. Assume that S is Cohen-Macaulay and equidimensional of dimension
d, and that dim V (f1 , . . . , fi ) d i. Then equality holds and f1 , . . . , fi forms a
regular sequence in Sq for every prime q of V (f1 , . . . , fi ).
Proof. If S is Cohen-Macaulay and equidimensional of dimension d, then we have
dim(Sm ) = d for all maximal ideals m of S, see Lemma 113.7. By Proposition
102.5 we see that for all maximal ideals m V (f1 , . . . , fi ) the sequence is a regular
sequence in Sm and the local ring Sm /(f1 , . . . , fi ) is Cohen-Macaulay of dimension
d i. This actually means that S/(f1 , . . . , fi ) is Cohen-Macaulay and equidimensional of dimension d i.


00RA

Lemma 128.2. Suppose that R S is a ring map which is finite type, flat. Let d
be an integer such that all fibres S R (p) are Cohen-Macaulay and equidimensional
of dimension d. Let f1 , . . . , fi be elements of S. The set
{q V (f1 , . . . , fi ) | f1 , . . . , fi are a regular sequence in Sq /pSq where p = R q}
is open in V (f1 , . . . , fi ).
Proof. Write S = S/(f1 , . . . , fi ). Suppose q is an element of the set defined in
the lemma, and p is the corresponding prime of R. We will use relative dimension
as defined in Definition 124.1. First, note that d = dimq (S/R) = dim(Sq /pSq ) +
trdeg(p) (q) by Lemma 115.3. Since f1 , . . . , fi form a regular sequence in the
Noetherian local ring Sq /pSq general dimension theory tells us that dim(S q /pS q ) =
dim(Sq /pSq ) i. By the same Lemma 115.3 we then conclude that dimq (S/R) =
dim(S q /pS q )+trdeg(p) (q) = di. By Lemma 124.6 we have dimq0 (S/R) di
for all q0 V (f1 , . . . , fi ) = Spec(S) in a neighbourhood of q. Thus after replacing
S by Sg for some g S, g 6 q we may assume that the inequality holds for all q0 .
The result follows from Lemma 128.1.


00RB

Lemma 128.3. Let R S is a ring map. Consider a finite homological complex

of finite free S-modules:
e

e1

i+1

i1

F : 0 S ne S ne1 . . . S ni S ni1 . . . S n0
For every prime q of S consider the complex F ,q = F,q R (p) where p is inverse
image of q in R. Assume there exists an integer d such that R S is finite type,
flat with fibres S R (p) Cohen-Macaulay of dimension d. The set
{q Spec(S) | F ,q is exact}
is open in Spec(S).

COMMUTATIVE ALGEBRA

301

Proof. Let q be an element of the set defined in the lemma. We are going to use
Proposition 101.10 to show there exists a g S, g 6 q such that D(g) is contained
in the set defined in the lemma. In other words, we are going to show that after
replacing S by Sg , the set of the lemma is all of Spec(S). Thus during the proof
we will, finitely often, replace S by such a localization. Recall that Proposition
101.10 characterizes exactness of complexes in terms of ranks of the maps i and
the ideals I(i ), in case the ring is local. We first address the rank condition. Set
ri = ni ni+1 + . . . + (1)ei ne . Note that ri + ri+1 = ni and note that ri is the
expected rank of i (in the exact case).
By Lemma 98.5 we see that if F ,q is exact, then the localization F,q is exact.
In particular the complex F becomes exact after localizing by an element g S,
g 6 q. In this case Proposition 101.10 applied to all localizations of S at prime
ideals implies that all (ri + 1) (ri + 1)-minors of i are zero. Thus we see that
the rank of of i is at most ri .
Let Ii S denote the ideal generated by the ri ri -minors of the matrix of i .
By Proposition 101.10 the complex F ,q is exact if and only if for every 1 i e
we have either (Ii )q = Sq or (Ii )q contains a Sq /pSq -regular sequence of length i.
Namely, by our choice of ri above and by the bound on the ranks of the i this is
the only way the conditions of Proposition 101.10 can be satisfied.
If (Ii )q = Sq , then after localizing S at some element g 6 q we may assume that
Ii = S. Clearly, this is an open condition.
If (Ii )q 6= Sq , then we have a sequence f1 , . . . , fi (Ii )q which form a regular
sequence in Sq /pSq . Note that for any prime q0 S such that (f1 , . . . , fi ) 6 q0 we
have (Ii )q0 = Sq0 . Thus the result follows from Lemma 128.2.

00RC

Theorem 128.4. Let R be a ring. Let R S be a ring map of finite presentation.

Let M be a finitely presented S-module. The set
{q Spec(S) | Mq is flat over R}
is open in Spec(S).
Proof. Let q Spec(S) be a prime. Let p R be the inverse image of q in R.
Note that Mq is flat over R if and only if it is flat over Rp . Let us assume that Mq
is flat over R. We claim that there exists a g S, g 6 q such that Mg is flat over
R.
We first reduce to the case where R and S are of finite type over Z. Choose a
directed partially ordered set and a system (R S , M ) as in Lemma 126.15.
Set p equal to the inverse image of p in R . Set q equal to the inverse image of
q in S . Then the system
((R )p , (S )q , (M )q )
is a system as in Lemma 126.11. Hence by Lemma 127.3 we see that for some
the module M is flat over R at the prime q . Suppose we can prove our claim for
the system (R S , M , q ). In other words, suppose that we can find a g S ,
g 6 q such that (M )g is flat over R . By Lemma 126.15 we have M = M R R
and hence also Mg = (M )g R R. Thus by Lemma 38.7 we deduce the claim for
the system (R S, M, q).

COMMUTATIVE ALGEBRA

302

At this point we may assume that R and S are of finite type over Z. We may write
S as a quotient of a polynomial ring R[x1 , . . . , xn ]. Of course, we may replace S by
R[x1 , . . . , xn ] and assume that S is a polynomial ring over R. In particular we see
that R S is flat and all fibres rings S R (p) have global dimension n.
Choose a resolution F of M over S with each Fi finite free, see Lemma 70.1. Let
Kn = Ker(Fn1 Fn2 ). Note that (Kn )q is flat over R, since each Fi is flat over
R and by assumption on M , see Lemma 38.13. In addition, the sequence
0 Kn /pKn Fn1 /pFn1 . . . F0 /pF0 M/pM 0
R

is exact upon localizing at q, because of vanishing of Tori p ((p), Mq ). Since the

global dimension of Sq /pSq is n we conclude that Kn /pKn localized at q is a finite
free module over Sq /pSq . By Lemma 98.4 (Kn )q is free over Sq . In particular,
there exists a g S, g 6 q such that (Kn )g is finite free over Sg .
By Lemma 128.3 there exists a further localization Sg such that the complex
0 Kn Fn1 . . . F0
is exact on all fibres of R S. By Lemma 98.5 this implies that the cokernel of
F1 F0 is flat. This proves the theorem in the Noetherian case.

129. Openness of Cohen-Macaulay loci
00RD

In this section we characterize the Cohen-Macaulay property of finite type algebras

in terms of flatness. We then use this to prove the set of points where such an
algebra is Cohen-Macaulay is open.

00RE

Lemma 129.1. Let S be a finite type algebra over a field k. Let : k[y1 , . . . , yd ]
S be a finite ring map. As subsets of Spec(S) we have
{q | Sq flat over k[y1 , . . . , yd ]} = {q | Sq CM and dimq (S/k) = d}
For notation see Definition 124.1.
Proof. Let q S be a prime. Denote p = k[y1 , . . . , yd ] q. Note that always
dim(Sq ) dim(k[y1 , . . . , yd ]p ) by Lemma 124.4 for example. Moreover, the field
extension (p) (q) is finite and hence trdegk ((p)) = trdegk ((q)).
Let q be an element of the left hand side. Then Lemma 111.9 applies and we
conclude that Sq is Cohen-Macaulay and dim(Sq ) = dim(k[y1 , . . . , yd ]p ). Combined
with the equality of transcendence degrees above and Lemma 115.3 this implies that
dimq (S/k) = d. Hence q is an element of the right hand side.
Let q be an element of the right hand side. By the equality of transcendence
degrees above, the assumption that dimq (S/k) = d and Lemma 115.3 we conclude
that dim(Sq ) = dim(k[y1 , . . . , yd ]p ). Hence Lemma 127.1 applies and we see that q
is an element of the left hand side.


00RF

Lemma 129.2. Let S be a finite type algebra over a field k. The set of primes q
such that Sq is Cohen-Macaulay is open in S.
This lemma is a special case of Lemma 129.4 below, so you can skip straight to the
proof of that lemma if you like.

COMMUTATIVE ALGEBRA

303

Proof. Let q S be a prime such that Sq is Cohen-Macaulay. We have to show

there exists a g S, g 6 q such that the ring Sg is Cohen-Macaulay. For any g S,
g 6 q we may replace S by Sg and q by qSg . Combining this with Lemmas 114.5 and
115.3 we may assume that there exists a finite injective ring map k[y1 , . . . , yd ] S
with d = dim(Sq ) + trdegk ((q)). Set p = k[y1 , . . . , yd ] q. By construction we
see that q is an element of the right hand side of the displayed equality of Lemma
129.1. Hence it is also an element of the left hand side.
By Theorem 128.4 we see that for some g S, g 6 q the ring Sg is flat over
k[y1 , . . . , yd ]. Hence by the equality of Lemma 129.1 again we conclude that all
local rings of Sg are Cohen-Macaulay as desired.

00RG

Lemma 129.3. Let k be a field. Let S be a finite type k algebra. The set of
Cohen-Macaulay primes forms a dense open U Spec(S).
Proof. The set is open by Lemma 129.2. It contains all minimal primes q S
since the local ring at a minimal prime Sq has dimension zero and hence is CohenMacaulay.


00RH

Lemma 129.4. Let R be a ring. Let R S be of finite presentation and flat.

For any d 0 the set


q Spec(S) such that setting p = R q the fibre ring
Sq /pSq is Cohen-Macaulay and dimq (S/R) = d
is open in Spec(S).
Proof. Let q be an element of the set indicated, with p the corresponding prime
of R. We have to find a g S, g 6 q such that all fibre rings of R Sg are CohenMacaulay. During the course of the proof we may (finitely many times) replace S
by Sg for a g S, g 6 q. Thus by Lemma 124.2 we may assume there is a quasifinite ring map R[t1 , . . . , td ] S with d = dimq (S/R). Let q0 = R[t1 , . . . , td ] q.
By Lemma 129.1 we see that the ring map
R[t1 , . . . , td ]q0 /pR[t1 , . . . , td ]q0 Sq /pSq
is flat. Hence by the crit`ere de platitude par fibres Lemma 127.8 we see that
R[t1 , . . . , td ]q0 Sq is flat. Hence by Theorem 128.4 we see that for some g S,
g 6 q the ring map R[t1 , . . . , td ] Sg is flat. Replacing S by Sg we see that for
every prime r S, setting r0 = R[t1 , . . . , td ] r and p0 = R r the local ring map
R[t1 , . . . , td ]r0 Sr is flat. Hence also the base change
R[t1 , . . . , td ]r0 /p0 R[t1 , . . . , td ]r0 Sr /p0 Sr
is flat. Hence by Lemma 129.1 applied with k = (p0 ) we see r is in the set of the
lemma as desired.


00RI

Lemma 129.5. Let R be a ring. Let R S be flat of finite presentation. The set
of primes q such that the fibre ring Sq R (p), with p = R q is Cohen-Macaulay
is open and dense in every fibre of Spec(S) Spec(R).
Proof. The set, call it W , is open by Lemma 129.4. It is dense in the fibres
because the intersection of W with a fibre is the corresponding set of the fibre to
which Lemma 129.3 applies.


COMMUTATIVE ALGEBRA

00RJ

304

Lemma 129.6. Let k be a field. Let S be a finite type k-algebra. Let k K be a

field extension, and set SK = K k S. Let q S be a prime of S. Let qK SK be
a prime of SK lying over q. Then Sq is Cohen-Macaulay if and only if (SK )qK is
Cohen-Macaulay.
Proof. During the course of the proof we may (finitely many times) replace S
by Sg for any g S, g 6 q. Hence using Lemma 114.5 we may assume that
dim(S) = dimq (S/k) =: d and find a finite injective map k[x1 , . . . , xd ] S. Note
that this also induces a finite injective map K[x1 , . . . , xd ] SK by base change.
By Lemma 115.6 we have dimqK (SK /K) = d. Set p = k[x1 , . . . , xd ] q and pK =
K[x1 , . . . , xd ] qK . Consider the following commutative diagram of Noetherian
local rings
/ (SK )q
Sq
O
O K
k[x1 , . . . , xd ]p

/ K[x1 , . . . , xd ]p
K

By Lemma 129.1 we have to show that the left vertical arrow is flat if and only if
the right vertical arrow is flat. Because the bottom arrow is flat this equivalence
holds by Lemma 99.1.

00RK

Lemma 129.7. Let R be a ring. Let R S be of finite type. Let R R0 be any

ring map. Set S 0 = R0 R S. Denote f : Spec(S 0 ) Spec(S) the map associated
to the ring map S S 0 . Set W equal to the set of primes q such that the fibre
ring Sq R (p), p = R q is Cohen-Macaulay, and let W 0 denote the analogue for
S 0 /R0 . Then W 0 = f 1 (W ).
Proof. Trivial from Lemma 129.6 and the definitions.

00RL

Lemma 129.8. Let R be a ring. Let R S be a ring map which is (a) flat,
(b) of finite presentation, (c) has Cohen-Macaulay fibres. Then we can write S =
S0 . . . Sn as a product of R-algebras Sd such that each Sd satisfies (a), (b), (c)
and has all fibres equidimensional of dimension d.
Proof. For each integer d denote
` Wd Spec(S) the set defined in Lemma 129.4.
Clearly we have Spec(S) =
Wd , and each Wd is open by the lemma we just
quoted. Hence the result follows from Lemma 22.3.

130. Differentials

00RM

In this section we define the module of differentials of a ring map.

00RN

Definition 130.1. Let : R S be a ring map and let M be an S-module. A

derivation, or more precisely an R-derivation into M is a map D : S M which
is additive, annihilates elements of (R), and satisfies the Leibniz rule: D(ab) =
aD(b) + D(a)b.
Note that D(ra) = rD(a) if r R and a S. The set of all R-derivations
forms an S-module: Given two R-derivations D, D0 the sum D + D0 : S M ,
a 7 D(a) + D0 (a) is an R-derivation, and given an R-derivation D and an element
c S the scalar multiple cD : S M , a 7 cD(a) is an R-derivation. We denote
this S-module
DerR (S, M ).

COMMUTATIVE ALGEBRA

305

Also, if : M N is an S-module map, then the composition D is an Rderivation into N . In this way the assignment M 7 DerR (S, M ) is a covariant
functor.
Consider the following map of free S-modules
M
M
M
S[(a, b)]
S[(f, g)]
2
2
(a,b)S

(f,g)S

rR

S[r]

M
aS

S[a]

defined by the rules

[(a, b)] 7 [a + b] [a] [b],

[(f, g)] 7 [f g] f [g] g[f ],

[r] 7 [(r)]

with obvious notation. Let S/R be the cokernel of this map. There is a map
d : S S/R which maps a to the class da of [a] in the cokernel. This is an
R-derivation by the relations imposed on S/R , in other words
d(a + b) = da + db,

d(f g) = f dg + gdf,

dr = 0

where a, b, f, g S and r R.
07BK

Definition 130.2. The pair (S/R , d) is called the module of K

ahler differentials
or the module of differentials of S over R.

00RO

Lemma 130.3. The module of differentials of S over R has the following universal
property. The map
HomS (S/R , M ) DerR (S, M ),

7 d

is an isomorphism of functors.
Proof. By definition an R-derivation is a rule which L
associates to each a S an
element D(a) M . Thus D gives rise to a map [D] :
S[a] M . However, the
conditions of being an R-derivation exactly mean that [D] annihilates the image of
the map in the displayed presentation of S/R above.

031G

Lemma 130.4. Let I be a directed partially ordered set. Let (Ri Si , ii0 ) be a
system of ring maps over I, see Categories, Section 21. Then we have
S/R = colimi Si /Ri .
Proof. This is clear from the presentation of S/R given above.

00RP

Lemma 130.5. Suppose that R S is surjective. Then S/R = 0.

Proof. You can see this either because all R-derivations clearly have to be zero,
or because the map in the presentation of S/R is surjective.

Suppose that
SO

00RQ

(130.5.1)

/ S0
O

/ R0

is a commutative diagram of rings. In this case there is a natural map of modules

of differentials fitting into the commutative diagram
/ S 0 /R0
S/R
O
O
d

/ S0

COMMUTATIVE ALGEBRA

306

To construct the map just use the obvious map between the presentations for S/R
and S 0 /R0 . Namely,
L 0 0 0
L 0 0 0
L 0 0
L 0 0
/
S [(a , b )]
S [(f , g )]
S [r ]
S [a ]
O
O
[(a, b)] 7 [((a), (b))]
[(f, g)] 7 [((f ), (g))]
[r] 7 [(r)]
L

S[(a, b)]

[a]7[(a)]

S[(f, g)]

S[r]

/ L S[a]

The result is simply that f dg S/R is mapped to (f )d(g).

00RR

Lemma 130.6. In diagram (130.5.1), suppose that S S 0 is surjective with

kernel I S. Then S/R S 0 /R0 is surjective with kernel generated as an Smodule by the elements da, where a S is such that (a) (R0 ). (This includes
in particular the elements d(i), i I.)
Proof. Consider the map of presentations above. Clearly the right vertical map
of free modules is surjective. Thus the map is surjective. A diagram chase shows
that the following elements generate the kernel as an S-module for sure: ida, i
I, a S, and da, with a S such that (a) = (r0 ) for some r0 R0 . Note that
(i) = (ia) = 0 = (0), and that d(ia) = ida + adi. Hence ida = d(ia) adi is
an S-linear combination of elements of the second kind.


00RS

Lemma 130.7. Let A B C be ring maps. Then there is a canonical exact

sequence
C B B/A C/A C/B 0
of C-modules.
Proof. We get a diagram (130.5.1) by putting R = A, S = C, R0 = B, and
S 0 = C. By Lemma 130.6 the map C/A C/B is surjective, and the kernel is
generated by the elements d(c), where c C is in the image of B C. The lemma
follows.


00RT

Lemma 130.8. Let : A B be a ring map.

(1) If S A is a multiplicative subset mapping to invertible elements of B,
then B/A = B/S 1 A .
(2) If S B is a multiplicative subset then S 1 B/A = S 1 B/A .
Proof. To show the equality of (1) it is enough to show that any A-derivation
D : B M annihilates the elements (s)1 . This is clear from the Leibniz
rule applied to 1 = (s)(s)1 . To show (2) note that there is an obvious map
S 1 B/A S 1 B/A . To show it is an isomorphism it is enough to show that
there is a A-derivation d0 of S 1 B into S 1 B/A . To define it we simply set
d0 (b/s) = (1/s)db (1/s2 )bds. Details omitted.


00RU

Lemma 130.9. In diagram (130.5.1), suppose that S S 0 is surjective with

kernel I S, and assume that R0 = R. Then there is a canonical exact sequence
of S 0 -modules
I/I 2 S/R S S 0 S 0 /R 0
The leftmost map is characterized by the rule that f I maps to df 1.

COMMUTATIVE ALGEBRA

307

Proof. The middle term is S/R S S/I. For f I denote f the image of f in
I/I 2 . To show that the map f 7 df 1 is well defined we just have to check that
df1 f2 1 = 0 if f1 , f2 I. And this is clear from the Leibniz rule df1 f2 1 =
(f1 df2 + f2 df1 ) 1 = df2 f1 + df1 f2 = 0. A similar computation show this
map is S 0 = S/I-linear.
The map S/R S S 0 S 0 /R is the canonical S 0 -linear map associated to the
S-linear map S/R S 0 /R . It is surjective because S/R S 0 /R is surjective
by Lemma 130.6.
The composite of the two maps is zero because df maps to zero in S 0 /R for f I.
Note that exactness just says that the kernel of S/R S 0 /R is generated as an
S-submodule by the submodule IS/R together with the elements df , with f I.
We know by Lemma 130.6 that this kernel is generated by the elements d(a) where
(a) = (r) for some r R. But then a = (r) + a (r), so d(a) = d(a (r)).
And a (r) I since (a (r)) = (a) ((r)) = (r) (r) = 0. We
conclude the elements df with f I already generate the kernel as an S-module,
as desired.

02HP

Lemma 130.10. In diagram (130.5.1), suppose that S S 0 is surjective with

kernel I S, and assume that R0 = R. Moreover, assume that there exists an
R-algebra map S 0 S which is a right inverse to S S 0 . Then the exact sequence
of S 0 -modules of Lemma 130.9 turns into a short exact sequence
0 I/I 2 S/R S S 0 S 0 /R 0
which is even a split short exact sequence.
Proof. Let : S 0 S be the right inverse to the surjection : S S 0 , so
S = I (S 0 ). Clearly we can use : S 0 /R S/R , to get a right inverse to the
map S/R S S 0 S 0 /R . On the other hand, consider the map
D : S I/I 2 ,

x 7 x ((x))

It is easy to show that D is an R-derivation (omitted). Moreover xD(s) = 0 if x

I, s S. Hence, by the universal property D induces a map : S/R R S 0 I/I 2 .
We omit the verification that it is a left inverse to d : I/I 2 S/R S S 0 . Hence
we win.

02HQ

Lemma 130.11. Let R S be a ring map. Let I S be an ideal. Let n 1 be

an integer. Set S 0 = S/I n+1 . The map S/R S 0 /R induces an isomorphism
S/R S S/I n S 0 /R S 0 S/I n .
Proof. This follows from Lemma 130.9 and the fact that d(I n+1 ) I n S/R by
the Leibniz rule for d.


00RV

Lemma 130.12. Suppose that we have ring maps R R0 and R S. Set

S 0 = S R R0 , so that we obtain a diagram (130.5.1). Then the canonical map
defined above induces an isomorphism S/R R R0 = S 0 /R0 .
P
Proof. Let d0 : S 0 = S R R0 S/R R R0 denote the map d0 ( ai xi ) =
d(ai ) xi . It exists because the map S R0 S/R R R0 , (a, x) 7 da R x is Rbilinear. This is an R0 -derivation, as can be verified by a simple computation. We
will show that (S/R R R0 , d0 ) satisfies the universal property. Let D : S 0 M 0
be an R0 derivation into an S 0 -module. The composition S S 0 M 0 is an

COMMUTATIVE ALGEBRA

308

R-derivation, hence we get an S-linear map D : S/R M 0 . We may tensor this

with R0 and get the map 0D : S/R R R0 M 0 , 0D ( x) = xD (). It is clear
that D = 0D d0 .

The multiplication map S R S S is the R-algebra map which maps a b to ab
in S. It is also an S-algebra map, if we think of S R S as an S-algebra via either
of the maps S S R S.
00RW

Lemma 130.13. Let R S be a ring map. Let J = Ker(S R S S) be the

kernel of the multiplication map. There is a canonical isomorphism of S-modules
S/R J/J 2 , adb 7 a b ab 1.
Proof. First we show that the rule adb 7 a b ab 1 is well defined. In order
to do this we have to show that dr and adb + bda d(ab) map to zero. The first
because r 1 1 r = 0 by definition of the tensor product. The second because
(a b ab 1) + (b a ba 1) (1 ab ab 1) = (a 1 1 a)(1 b b 1)
is in J 2 .
We construct a map in the other direction. We may think of S S R S, a 7 a1
as the base change of R S. Hence we have SR S/S = S/R S (S R S), by
Lemma 130.12. At this point the sequence of Lemma 130.9 gives a map
J/J 2 SR S/S SR S S = (S/R S (S R S)) SR S S = S/R .
We leave it to the reader to see it is the inverse of the map above.

00RX

Lemma 130.14. If S = R[x1 , . . . , xn ], then S/R is a finite free S-module with

basis dx1 , . . . , dxn .
Proof. We first show that dx1 , . . . , dxn generatePS/R as an S-module. To prove
this we show that dg can be expressed as a sum
gi dxi for any g R[x1 , . . . , xn ].
We do this by induction on the (total) degree of g. It is clear if the degree
P of g is 0,
because then dg = 0. If the degree of g is > 0, then we may write P
g as c+ gP
i xi with
c R and deg(gi ) < deg(g). By the Leibniz rule we have dg =
gi dxi + xi dgi ,
and hence we win by induction.
Consider the R-derivation /xi : R[x1 , . . . , xn ] R[x1 , . . . , xn ]. (We leave it to
the reader to define this; the defining property being that /xi (xj ) = ij .) By the
universal property this corresponds to an S-module map li : S/R R[x1 , . . . , xn ]
which maps dxi to 1 and dxj to 0 for j 6= i. Thus it is clear that there are no
S-linear relations among the elements dx1 , . . . , dxn .


00RY

Lemma 130.15. Suppose R S is of finite presentation. Then S/R is a finitely

presented S-module.
Proof. Write S = R[x1 , . . . , xn ]/(f1 , . . . , fm ). Write I = (f1 , . . . , fm ). According
to Lemma 130.9 there is an exact sequence of S-modules
I/I 2 R[x1 ,...,xn ]/R R[x1 ,...,xn ] S S/R 0
The result follows from the fact that I/I 2 is a finite S-module (generated by the
images of the fi ), and that the middle term is finite free by Lemma 130.14.


00RZ

Lemma 130.16. Suppose R S is of finite type. Then S/R is finitely generated

S-module.
Proof. This is very similar to, but easier than the proof of Lemma 130.15.

COMMUTATIVE ALGEBRA

309

131. Finite order differential operators

09CH
09CI

In this section we introduce differential operators of finite order.

Definition 131.1. Let R S be a ring map. Let M , N be S-modules. Let k 0
be an integer. We inductively define a differential operator D : M N of order k
to be an R-linear map such that for all g S the map m 7 D(gm) gD(m) is a
differential operator of order k 1. For the base case k = 0 we define a differential
operator of order 0 to be an S-linear map.
If D : M N is a differential operator of order k, then for all g S the map gD
is a differential operator of order k. The sum of two differential operators of order
k is another. Hence the set of all these
Diffk (M, N ) = DiffkS/R (M, N )
is an S-module. We have
Diff0 (M, N ) Diff1 (M, N ) Diff2 (M, N ) . . .

09CJ

Lemma 131.2. Let R S be a ring map. Let L, M, N be S-modules. If D : L

M and D0 : M N are differential operators of order k and k 0 , then D0 D is a
differential operator of order k + k 0 .
Proof. Let g S. Then the map which sends x L to
D0 (D(gx)) gD0 (D(x)) = D0 (D(gx)) D0 (gD(x)) + D0 (gD(x)) gD0 (D(x))
is a sum of two compositions of differential operators of lower order. Hence the
lemma follows by induction on k + k 0 .


09CK

Lemma 131.3. Let R S be a ring map. Let M be an S-module. Let k 0.

k
There exists an S-module PS/R
(M ) and a canonical isomorphism
k
DiffkS/R (M, N ) = HomS (PS/R
(M ), N )

functorial in the S-module N .

k
Proof. The existence of PS/R
(M ) follows from general category theoretic argumentsL(insert future reference here), but we will also give a construction. Set
F = mM S[m] where [m] is a symbol indicating the basis element in the summand corresponding to m. Given any differential operator D : M N we obtain
an S-linear map LD : F N sending [m] to D(m). If D has order 0, then LD
annihilates the elements

[m + m0 ] [m] [m0 ],

g0 [m] [g0 m]

where g0 S and m, m M . If D has order 1, then LD annihilates the elements

[m + m0 ] [m] [m0 ],

f [m] [f m],

g0 g1 [m] g0 [g1 m] g1 [g0 m] + [g1 g0 m]

where f R, g0 , g1 S, and m M . If D has order k, then LD annihilates the

elements [m + m0 ] [m] [m0 ], f [m] [f m], and the elements
X
g0 g1 . . . gk [m]
g0 . . . gi . . . gk [gi m] + . . . + (1)k+1 [g0 . . . gk m]
Conversely, if L : F N is an S-linear map annihilating all the elements listed in
the previous sentence, then m 7 L([m]) is a differential operator of order k. Thus
k
we see that PS/R
(M ) is the quotient of F by the submodule generated by these
elements.


COMMUTATIVE ALGEBRA

09CL

310

Definition 131.4. Let R S be a ring map. Let M be an S-module. The

k
module PS/R
(M ) constructed in Lemma 131.3 is called the module of principal
parts of order k of M .
Note that the inclusions
Diff0 (M, N ) Diff1 (M, N ) Diff2 (M, N ) . . .
correspond via Yonedas lemma (Categories, Lemma 3.5) to surjections
2
1
0
. . . PS/R
(M ) PS/R
(M ) PS/R
(M ) = M

09CM

Example 131.5. Let R S be a ring map and let N be an S-module. Observe

that Diff1 (S, N ) = DerR (S, N ) N . Namely, if D : S N is a differential operator
of order 1 then D : S N defined by D (g) := D(g) gD(1) is an R-derivation
and D = D + D(1) where x : S N is the linear map sending g to gx. It follows
1
that PS/R
= S/R S by the universal property of S/R .

09CN

Lemma 131.6. Let R S be a ring map. Let M be an S-module. There is a

canonical short exact sequence
1
0 S/R S M PS/R
(M ) M 0

functorial in M called the sequence of principal parts.

1
Proof. The map PS/R
(M ) M is given above. Let N be an S-module and let
D : M N be a differential operator of order 1. For m M the map

g 7 D(gm) gD(m)
is an R-derivation S N by the axioms for differential operators of order 1.
Thus it corresponds to a linear map Dm : S/R N determined by the rule
adb 7 aD(bm) abD(m) (see Lemma 130.3). The map
S/R M N,

(, m) 7 Dm ()

is S-bilinear (details omitted) and hence determines an S-linear map

D : S/R S M N
In this way we obtain a map Diff1 (M, N ) HomS (S/R S M, N ), D 7 D
functorial in N . By the Yoneda lemma this corresponds a map S/R S M
1
PS/R
(M ). It is immediate from the construction that this map is functorial in M .
The sequence
1
(M ) M 0
S/R S M PS/R
is exact because for every module N the sequence
0 HomS (M, N ) Diff1 (M, N ) HomS (S/R S M, N )
is exact by inspection.
1
To see that S/R S M PS/R
(M ) is injective we argue as follows. Choose an
exact sequence
0 M0 F M 0

with F a free S-module. This induces an exact sequence

0 Diff1 (M, N ) Diff1 (F, N ) Diff1 (M 0 , N )

COMMUTATIVE ALGEBRA

311

for all N . This proves that in the commutative diagram

0

/ S/R S M 0

/ P 1 (M 0 )
S/R

/ M0

/0


/ S/R S F


/ P 1 (F )
S/R


/F

/0


/ S/R S M

/ P1


/M

/0

S/R (M )


0


0


0

the middle column is exact. The left column is exact by right exactness of S/R S
. By the snake lemma (see Section 4) it suffices to prove exactness on the left for
1
the free module F . Using that PS/R
() commutes with direct sums we reduce to
the case M = S. This case is a consequence of the discussion in Example 131.5. 
09CP

Remark 131.7. Suppose given a commutative diagram of rings

BO

/ B0
O

/ A0

a B-module M , a B 0 -module M 0 , and a B-linear map M M 0 . Then we get a

compatible system of module maps
...

/ P 2 0 0 (M 0 )
B /A
O

...

/ P 2 (M )
B/A

/ P 1 0 0 (M 0 )
B /A
O
/ P1

B/A (M )

/ P 0 0 0 (M 0 )
B /A
O
/ P 0 (M )
B/A

These maps are compatible with further composition of maps of this type. The
k
easiest way to see this is to use the description of the modules PB/A
(M ) in terms
of generators and relations in the proof of Lemma 131.3 but it can also be seen
directly from the universal property of these modules. Moreover, these maps are
compatible with the short exact sequences of Lemma 131.6.
132. The naive cotangent complex
00S0

Let R S be a ring map. Denote R[S] the polynomial ring whose variables are
the elements s S. Lets denote [s] R[S] the variable corresponding to s S.
Thus R[S] is a free R-module on the basis elements [s1 ] . . . [sn ] where s1 , . . . , sn is
an unordered sequence of elements of S. There is a canonical surjection

07BL

(132.0.1)

R[S] S,

[s] 7 s

whose kernel we denote I R[S]. It is a simple observation that I is generated by

the elements [s + s0 ] [s] [s0 ], [s][s0 ] [ss0 ] and [r] r. According to Lemma 130.9

COMMUTATIVE ALGEBRA

312

there is a canonical map

07BM

(132.0.2)

I/I 2 R[S]/R R[S] S

whose cokernel is canonically isomorphic to S/R . Observe that the S-module

R[S]/R R[S] S is free on the generators d[s].
07BN

Definition 132.1. Let R S be a ring map. The naive cotangent complex NLS/R
is the chain complex (132.0.2)


NLS/R = I/I 2 R[S]/R R[S] S
with I/I 2 placed in (homological) degree 1 and R[S]/R R[S] S placed in degree 0.
We will denote H1 (LS/R ) = H1 (NLS/R )9 the homology in degree 1.
Before we continue let us say a few words about the actual cotangent complex
(Cotangent, Section 3). Given a ring map R S there exists a canonical simplicial
R-algebra P whose terms are polynomial algebras and which comes equipped with
a canonical homotopy equivalence
P S
The cotangent complex LS/R of S over R is defined as the chain complex associated
to the cosimplicial module
P /R P S
The naive cotangent complex as defined above is canonically isomorphic to the
truncation 1 LS/R (see Homology, Section 13 and Cotangent, Section 10). In
particular, it is indeed the case that H1 (NLS/R ) = H1 (LS/R ) so our definition
is compatible with the one using the cotangent complex. Moreover, H0 (LS/R ) =
H0 (NLS/R ) = S/R as weve seen above.
Let R S be a ring map. A presentation of S over R is a surjection : P S of
R-algebras where P is a polynomial algebra (on a set of variables). Often, when S
is of finite type over R we will indicate this by saying: Let R[x1 , . . . , xn ] S be
a presentation of S/R, or Let 0 I R[x1 , . . . , xn ] S 0 be a presentation
of S/R if we want to indicate that I is the kernel of the presentation. Note that
the map R[S] S used to define the naive cotangent complex is an example of a
presentation.
Note that for every presentation we obtain a two term chain complex of S-modules
NL() : I/I 2 P/R P S.
Here the term I/I 2 is placed in degree 1 and the term P/R S is placed in degree
0. The class of f I in I/I 2 is mapped to df 1 in P/R S. The cokernel of this
complex is canonically S/R , see Lemma 130.9. We call the complex NL() the
naive cotangent complex associated to the presentation : P S of S/R. Note
that if P = R[S] with its canonical surjection onto S, then we recover NLS/R . If
L
P = R[x1 , . . . , xn ] then will sometimes use the notation I/I 2 i=1,...,n Sdxi to
denote this complex.
9This module is sometimes denoted
S/R in the literature.

COMMUTATIVE ALGEBRA

313

Suppose we are given a commutative diagram

SO
06RQ

/ S0
O

(132.1.1)
/ R0

of rings. Let : P S be a presentation of S over R and let : P 0 S 0 be

a presentation of S 0 over R0 . A morphism of presentations from : P S to
0 : P 0 S 0 is defined to be an R-algebra map
: P P0
such that = 0 . Note that in this case (I) I 0 , where I = Ker()
and I 0 = Ker(0 ). Thus induces a map of S-modules I/I 2 I 0 /(I 0 )2 and by
functoriality of differentials also an S-module map P/R S P 0 /R0 S 0 . These
maps are compatible with the differentials of NL() and NL(0 ) and we obtain a
map of naive cotangent complexes
NL() NL(0 ).
It is often convenient to consider the induced map NL() S S 0 NL(0 ).
In the special case that P = R[S] and P 0 = R0 [S 0 ] the map : S S 0 induces a
canonical ring map : P P 0 by the rule [s] 7 [(s)]. Hence the construction
above determines canonical(!) maps of chain complexes
NLS/R NLS 0 /R0 ,

and

NLS/R S S 0 NLS 0 /R0

associated to the diagram (132.1.1). Note that this construction is compatible with
composition: given a commutative diagram
SO
R

/ S0
O
/ R0

/ S 00
O
/ R00

we see that the composition of

NLS/R NLS 0 /R0 NLS 00 /R00
is the map NLS/R NLS 00 /R00 given by the outer square.
It turns out that NL() is homotopy equivalent to NLS/R and that the maps constructed above are well defined up to homotopy (homotopies of maps of complexes
are discussed in Homology, Section 12 but we also spell out the exact meaning of
the statements in the lemma below in its proof).
00S1

Lemma 132.2. Suppose given a diagram (132.1.1). Let : P S and 0 : P 0

S 0 be presentations.
(1) There exists a morphism of presentations from to 0 .
(2) Any two morphisms of presentations induce homotopic morphisms of complexes NL() NL(0 ).
(3) The construction is compatible with compositions of morphisms of presentations (see proof for exact statement).

COMMUTATIVE ALGEBRA

314

(4) If R R0 and S S 0 are isomorphisms, then for any map of presentations from to 0 the induced map NL() NL(0 ) is a homotopy
equivalence and a quasi-isomorphism.
In particular, comparing to the canonical presentation (132.0.1) we conclude
there is a quasi-isomorphism NL() NLS/R well defined up to homotopy and
compatible with all functorialities (up to homotopy).
Proof. Since P is a polynomial algebra over R we can write P = R[xa , a A]
for some set A. As 0 is surjective, we can choose for every a A an element
fa P 0 such that 0 (fa ) = ((xa )). Let : P = R[xa , a A] P 0 be the unique
R-algebra map such that (xa ) = fa . This gives the morphism in (1).
Let and 0 morphisms of presentations from to 0 . Let I = Ker() and
I 0 = Ker(0 ). We have to construct the diagonal map h in the diagram

01

/ P/R P S

I/I 2
  x
J/J 2

00

 
/ P 0 /R0 P 0 S 0

where the vertical maps are induced by , 0 such that

1 01 = h d

and 0 00 = d h

Consider the map D = 0 : P P 0 . Since both and are compatible with

and 0 we conclude that 0 : P I 0 . Also 0 is R-linear and
( 0 )(f g) = (f )( 0 )(g) + ( 0 )(f )0 (g)
Hence the induced map D : P I 0 /(I 0 )2 is a R-derivation. Thus we obtain a
canonical map h : P/R P S I 0 /(I 0 )2 such that D = h d. A calculation
(omitted) shows that h is the desired homotopy.
Suppose that we have a commutative diagram
SO
R

/ S0
O
/ R0

/ S 00
O
/ R00

and that
(1) : P S,
(2) 0 : P 0 S 0 , and
(3) 00 : P 00 S 00
are presentations. Suppose that
(1) : P P is a morphism of presentations from to 0 and
(2) 0 : P 0 P 00 is a morphism of presentations from 0 to 00 .
Then it is immediate that 0 : P P 00 is a morphism of presentations from
to 00 and that the induced map NL() NL(00 ) of naive cotangent complexes is
the composition of the maps NL() NL(0 ) and NL() NL(0 ) induced by
and 0 .
In the simple case of complexes with 2 terms a quasi-isomorphism is just a map
that induces an isomorphism on both the cokernel and the kernel of the maps

COMMUTATIVE ALGEBRA

315

between the terms. Note that homotopic maps of 2 term complexes (as explained
above) define the same maps on kernel and cokernel. Hence if is a map from
a presentation of S over R to itself, then the induced map NL() NL()
is a quasi-isomorphism being homotopic to the identity by part (2). To prove
(4) in full generality, consider a morphism 0 from 0 to which exists by (1).
The compositions NL() NL(0 ) NL() and NL(0 ) NL() NL(0 ) are
homotopic to the identity maps by (3), hence these maps are homotopy equivalences
by definition. It follows formally that both maps NL() NL(0 ) and NL(0 )
NL() are quasi-isomorphisms. Some details omitted.

08Q1

Lemma 132.3. Let A B be a polynomial algebra. Then NLB/A is homotopy

equivalent to the chain complex (0 B/A ) with B/A in degree 0.
Proof. Follows from Lemma 132.2 and the fact that idB : B B is a presentation
of B over A with zero kernel.

The following lemma is part of the motivation for introducing the naive cotangent
complex. The cotangent complex extends this to a genuine long exact cohomology
sequence. If B C is a local complete intersection, then one can extend the
sequence with a zero on the left, see More on Algebra, Lemma 25.6.

00S2

Lemma 132.4 (Jacobi-Zariski sequence). Let A B C be ring maps. Choose

a presentation : A[xs , s S] B with kernel I. Choose a presentation :
B[yt , t T ] C with kernel J. Let : A[xs , yt ] C be the induced presentation
of C with kernel K. Then we get a canonical commutative diagram
0

/ A[x ]/A C
s
O

/ A[x ,y ]/A C
s t
O

/ B[y ]/B C
t
O

/0

I/I 2 C

/ K/K 2

/ J/J 2

/0

with exact rows. We get the following exact sequence of homology groups
H1 (NLB/A B C) H1 (LC/A ) H1 (LC/B ) C B B/A C/A C/B 0
of C-modules extending the sequence of Lemma 130.7. If TorB
1 (B/A , C) = 0, then
H1 (NLB/A B C) = H1 (LB/A ) B C.
Proof. The precise definition of the maps is omitted. The exactness of the top row
follows as the dxs , dyt form a basis for the middle module. The map factors
A[xs , yt ] B[yt ] C
with surjective first arrow and second arrow equal to . Thus we see that K J
is surjective. Moreover, the kernel of the first displayed arrow is IA[xs , yt ]. Hence
I/I 2 C surjects onto the kernel of K/K 2 J/J 2 . Finally, we can use Lemma
132.2 to identify the terms as homology groups of the naive cotangent complexes.
The final assertion follows as the degree 0 term of the complex NLB/A is a free
B-module.

07VC

Remark 132.5. Let A B and : B C be ring maps. Then the composition NLB/A NLC/A NLC/B is homotopy equivalent to zero. Namely, this

COMMUTATIVE ALGEBRA

316

composition is the functoriality of the naive cotangent complex for the square
BO

/C
O
/B

Write J = Ker(B[C] C). An explicit homotopy is given by the map A[B]/A A

B J/J 2 which maps the basis element d[b] to the class of [(b)] b in J/J 2 .
07BP

Lemma 132.6. Let A B be a surjective ring map with kernel I. Then NLB/A
is homotopy equivalent to the chain complex (I/I 2 0) with I/I 2 in degree 1. In
particular H1 (LB/A ) = I/I 2 .
Proof. Follows from Lemma 132.2 and the fact that A B is a presentation of
B over A.


065V

Lemma 132.7. Let A B C be ring maps. Assume A C is surjective

(so also B C is). Denote I = Ker(A C) and J = Ker(B C). Then the
sequence
I/I 2 J/J 2 B/A B B/J 0
is exact.
Proof. Follows from Lemma 132.4 and the description of the naive cotangent complexes NLC/B and NLC/A in Lemma 132.6.


00S4

Lemma 132.8 (Flat base change). Let R S be a ring map. Let : P S be

a presentation. Let R R0 be a flat ring map. Let 0 : P 0 R R0 S 0 = S R R0
be the induced presentation. Then NL() R R0 = NL() S S 0 = NL(0 ). In
particular, the canonical map
NLS/R R R0 NLSR R0 /R0
is a homotopy equivalence if R R0 is flat.
Proof. This is true because Ker(0 ) = R0 R Ker() since R R0 is flat.

07BQ

Lemma 132.9. Let Ri Si be a system of ring maps over the directed partially
ordered set I. Set R = colim Ri and S = colim Si . Then NLS/R = colim NLSi /Ri .
L
Proof. Recall that NLS/R is the complex I/I 2 sS Sd[s] where I R[S] is
the kernel of the canonical presentation R[S] S. Now it is clear that R[S] =
colim Ri [Si ] and similarly that I = colim Ii where Ii = Ker(Ri [Si ] Si ). Hence
the lemma is clear.


07BR

Lemma 132.10. If S A is a multiplicative subset of A, then NLS 1 A/A is

homotopy equivalent to the zero complex.
Proof. Since A S 1 A is flat we see that NLS 1 A/A A S 1 A NLS 1 A/S 1 A
is a homotopy equivalence by flat base change (Lemma 132.8). Since the source
of the arrow is isomorphic to NLS 1 A/A and the target of the arrow is zero (by
Lemma 132.6) we win.


07BS

Lemma 132.11. Let S A is a multiplicative subset of A. Let S 1 A B be a

ring map. Then NLB/A NLB/S 1 A is an homotopy equivalence.

COMMUTATIVE ALGEBRA

317

Proof. Choose a presentation : P B of B over A. Then : S 1 P B is a

presentation of B over S 1 A. A direct computation shows that we have NL() =
NL() which proves the lemma as the naive cotangent complex is well defined up
to homotopy by Lemma 132.2.

08JZ

Lemma 132.12. Let A B be a ring map. Let g B. Suppose : P B is a

presentation with kernel I. Then a presentation of Bg over A is the map
: P [x] Bg
extending and sending x to 1/g. The kernel J of is generated by I and the
element f x 1 where f P is an element mapped to g B by . In this situation
we have
(1) J/J 2 = (I/I 2 )g Bg (f x 1),
(2) P [x]/A P [x] Bg = P/A P Bg Bg dx,
g
(3) NL() = NL() B Bg (Bg
Bg )
Hence the canonical map NLB/A B Bg NLBg /A is a homotopy equivalence.
Proof. Since P [x]/(I, f x 1) = B[x]/(gx 1) = Bg we get the statement about I
and f x1 generating J. To prove the other statements one can use the commutative
diagram
0

/ P/A Bg
O

/ P [x]/A Bg
O

/ B[x]/B Bg
O

/0

(I/I 2 )g

/ J/J 2

/ (f x 1)/(f x 1)2

/0

with exact rows of Lemma 132.4. Then the only question left over is: why is
(I/I 2 )g J/J 2 injective? It is enough to show that (I J 2 ) = 0 where : I
If /If2 is the canonical map. So consider the extension of to the P -algebra map
: P [x] Pf /If2 given by (x) = 1/f , and note that (J 2 ) = (J)2 = (I[x])2 =
0, since (f x 1) = 0 implies (J) = (I[x]).

00S7

Lemma 132.13. Let A B be a ring map. Let S B be a multiplicative subset.

The canonical map NLB/A B S 1 B NLS 1 B/A is a quasi-isomorphism.
Proof. We have S 1 B = colimgS Bg where we think of S as a directed partially
ordered set (ordering by divisibility), see Lemma 9.9. By Lemma 132.12 each of the
maps NLB/A B Bg NLBg /A are quasi-isomorphisms. The lemma follows from
Lemma 132.9.


00S3

Lemma 132.14. Let R be a ring. Let A1 A0 , and B1 B0 be two two term

complexes. Suppose that there exist morphisms of complexes : A B and
: B A such that and are homotopic to the identity maps. Then
A1 B0
= B1 A0 as R-modules.
Proof. Choose a map h : A0 B1 such that
idA1 1 1 = h dA and idA0 0 0 = dA h.
Similarly, choose a map h0 : B0 A1 such that
idB1 1 1 = h dB and idB0 0 0 = dB h.

COMMUTATIVE ALGEBRA

A trivial computation shows that


 
idA1 h0 1 + h 0
1
=
0
idB0
dB

h
0

318



1
dA

h0
0

This shows that both matrices on the right hand side are invertible and proves the
lemma.

00S5

Lemma 132.15. Let R S be a ring map of finite type. For any presentations
: R[x1 , . . . , xn ] S, and : R[y1 , . . . , ym ] S we have
I/I 2 S m
= J/J 2 S n
as S-modules where I = Ker() and J = Ker().
Proof. See Lemmas 132.2 and 132.14.

00S6

Lemma 132.16. Let R S be a ring map of finite type. Let g S. For any
presentations : R[x1 , . . . , xn ] S, and : R[y1 , . . . , ym ] Sg we have
(I/I 2 )g Sgm
= J/J 2 Sgn
as Sg -modules where I = Ker() and J = Ker().
Proof. By Lemma 132.15, we see that it suffices to prove this for a single choice
of and . Thus we may take the presentation of Lemma 132.12 and the result
is clear.

133. Local complete intersections

00S8

The property of being a local complete intersection is somehow an intrinsic property

of a Noetherian local ring. However, for the moment we just define this property
for finite type algebras over a field.

00S9

Definition 133.1. Let k be a field. Let S be a finite type k-algebra.

(1) We say that S is a global complete intersection over k if there exists a
presentation S = k[x1 , . . . , xn ]/(f1 , . . . , fc ) such that dim(S) = n c.
(2) We say that S
S is a local complete intersection over k if there exists a covering
Spec(S) = D(gi ) such that each of the rings Sgi is a global complete
intersection over k.
We will also use the convention that the zero ring is a global complete intersection
over k.
Suppose S is a global complete intersection S = k[x1 , . . . , xn ]/(f1 , . . . , fc ) as in
the definition. Recall that dim(S) = n c means that all irreducible components
of Spec(S) have dimension n c. Since all maximal ideals of the polynomial
ring have local rings of dimension n we conclude that all irreducible components
of Spec(S) have dimension n c. See Section 59. In other words, Spec(S) is
equidimensional of dimension n c.

00SA

Lemma 133.2. Let k be a field. Let S be a finite type k-algebra. Let g S.

(1) If S is a global complete intersection so is Sg .
(2) If S is a local complete intersection so is Sg .

COMMUTATIVE ALGEBRA

319

Proof. The second statement follows immediately from the first. For the first,
say that S = k[x1 , . . . , xn ]/(f1 , . . . , fc ) with n c = dim(S). By the remarks
above S is equidimensional of dimension n c, so dim(Sg ) = n c as well (or
it is the zero ring in which case the lemma is true by convention). Let g 0
k[x1 , . . . , xn ] be an element whose residue class corresponds to g. Then Sg =
k[x1 , . . . , xn , xn+1 ]/(f1 , . . . , fc , xn+1 g 0 1) as desired.

00SB

Lemma 133.3. Let k be a field. Let S be a finite type k-algebra. If S is a local

complete intersection, then S is a Cohen-Macaulay ring.
Proof. Choose a maximal prime m of S. We have to show that Sm is CohenMacaulay. By assumption we may assume S = k[x1 , . . . , xn ]/(f1 , . . . , fc ) with
dim(S) = n c. Let m0 k[x1 , . . . , xn ] be the maximal ideal corresponding to
m. According to Proposition 113.2 the local ring k[x1 , . . . , xn ]m0 is regular local of
dimension n. In particular it is Cohen-Macaulay by Lemma 105.3. By dimension
theory (see Section 59) the ring Sm = k[x1 , . . . , xn ]m0 /(f1 , . . . , fc ) has dimension
n c. By assumption dim(Sm ) n c. Thus we get equality. This implies that
f1 , . . . , fc is a regular sequence in k[x1 , . . . , xn ]m0 and that Sm is Cohen-Macaulay,
see Proposition 102.5.

The following is the technical key to the rest of the material in this section. An
important feature of this lemma is that we may choose any presentation for the
ring S, but that condition (1) does not depend on this choice.

00SC

Lemma 133.4. Let k be a field. Let S be a finite type k-algebra. Let q be a

prime of S. Choose any presentation S = k[x1 , . . . , xn ]/I. Let q0 be the prime
of k[x1 , . . . , xn ] corresponding to q. Set c = height(q0 ) height(q), in other words
dimq (S) = n c (see Lemma 115.4). The following are equivalent
(1) There exists a g S, g 6 q such that Sg is a global complete intersection
over k.
(2) The ideal Iq0 k[x1 , . . . , xn ]q0 can be generated by c elements.
(3) The conormal module (I/I 2 )q can be generated by c elements over Sq .
(4) The conormal module (I/I 2 )q is a free Sq -module of rank c.
(5) The ideal Iq0 can be generated by a regular sequence in the regular local ring
k[x1 , . . . , xn ]q0 .
In this case any c elements of Iq0 which generate Iq0 /q0 Iq0 form a regular sequence
in the local ring k[x1 , . . . , xn ]q0 .
Proof. Set R = k[x1 , . . . , xn ]q0 . This is a Cohen-Macaulay local ring of dimension
height(q0 ), see for example Lemma 133.3. Moreover, R = R/IR = R/Iq0 = Sq is
a quotient of dimension height(q). Let f1 , . . . , fc Iq0 be elements which generate
(I/I 2 )q . By Lemma 19.1 we see that f1 , . . . , fc generate Iq0 . Since the dimensions
work out, we conclude by Proposition 102.5 that f1 , . . . , fc is a regular sequence in
R. By Lemma 68.2 we see that (I/I 2 )q is free. These arguments show that (2),
(3), (4) are equivalent and that they imply the last statement of the lemma, and
therefore they imply (5).
If (5) holds, say Iq0 is generated by a regular sequence of length e, then height(q) =
dim(Sq ) = dim(k[x1 , . . . , xn ]q0 )e = height(q0 )e by dimension theory, see Section
59. We conclude that e = c. Thus (5) implies (2).

COMMUTATIVE ALGEBRA

320

We continue with the notation introduced in the first paragraph. For each fi we
may find di k[x1 , . . . , xn ], di 6 q0 such that fi0 = di fi k[x1 , . . . , xn ]. Then
it is still true that Iq0 = (f10 , . . . , fc0 )R. Hence there exists a g 0 k[x1 , . . . , xn ],
g 0 6 q0 such that Ig0 = (f10 , . . . , fc0 ). Moreover, pick g 00 k[x1 , . . . , xn ], g 00 6 q0 such
that dim(Sg00 ) = dimq Spec(S). By Lemma 115.4 this dimension is equal to n c.
Finally, set g equal to the image of g 0 g 00 in S. Then we see that
Sg
= k[x1 , . . . , xn , xn+1 ]/(f10 , . . . , fc0 , xn+1 g 0 g 00 1)
and by our choice of g 00 this ring has dimension n c. Therefore it is a global
complete intersection. Thus each of (2), (3), and (4) implies (1).
Assume (1). Let Sg
= k[y1 , . . . , ym ]/(f1 , . . . , ft ) be a presentation of Sg as a global
complete intersection. Write J = (f1 , . . . , ft ). Let q00 k[y1 , . . . , ym ] be the prime
corresponding to qSg . Note that t = m dim(Sg ) = height(q00 ) height(q),
see Lemma 115.4 for the last equality. As seen in the proof of Lemma 133.3
(and also above) the elements f1 , . . . , ft form a regular sequence in the local ring
k[y1 , . . . , ym ]q00 . By Lemma 68.2 we see that (J/J 2 )q is free of rank t. By Lemma
132.16 we have
J/J 2 Sgn
= (I/I 2 )g Sgm
Thus (I/I 2 )q is free of rank t + n m = m dim(Sg ) + n m = n dim(Sg ) =
height(q0 ) height(q) = c. Thus we obtain (4).

The result of Lemma 133.4 suggests the following definition.
00SD

Definition 133.5. Let k be a field. Let S be a local k-algebra essentially of finite

type over k. We say S is a complete intersection (over k) if there exists a local
k-algebra R and elements f1 , . . . , fc mR such that
(1)
(2)
(3)
(4)

R is essentially of finite type over k,

R is a regular local ring,
f1 , . . . , fc form a regular sequence in R, and
S
= R/(f1 , . . . , fc ) as k-algebras.

By the Cohen structure theorem (see Theorem 154.8) any complete Noetherian local
ring may be written as the quotient of some regular complete local ring. Hence we
may use the definition above to define the notion of a complete intersection ring for
any complete Noetherian local ring. We will discuss this in Divided Power Algebra,
Section 8. In the meantime the following lemma shows that such a definition makes
sense.
00SE

Lemma 133.6. Let A B C be surjective local ring homomorphisms. Assume

A and B are regular local rings. The following are equivalent
(1)
(2)
(3)
(4)

Ker(A C) is generated by a regular sequence,

Ker(A C) is generated by dim(A) dim(C) elements,
Ker(B C) is generated by a regular sequence, and
Ker(B C) is generated by dim(B) dim(C) elements.

Proof. A regular local ring is Cohen-Macaulay, see Lemma 105.3. Hence the equivalences (1) (2) and (3) (4), see Proposition 102.5. By Lemma 105.4 the ideal
Ker(A B) can be generated by dim(A) dim(B) elements. Hence we see that
(4) implies (2).

COMMUTATIVE ALGEBRA

321

It remains to show that (1) implies (4). We do this by induction on dim(A)dim(B).

The case dim(A) dim(B) = 0 is trivial. Assume dim(A) > dim(B). Write
I = Ker(A C) and J = Ker(A B). Note that J I. Our assumption is
that the minimal number of generators of I is dim(A) dim(C). Let m A be the
maximal ideal. Consider the maps
J/mJ I/mI m/m2
By Lemma 105.4 and its proof the composition is injective. Take any element x J
which is not zero in J/mJ. By the above and Nakayamas lemma x is an element
of a minimal set of generators of I. Hence we may replace A by A/xA and I by
I/xA which decreases both dim(A) and the minimal number of generators of I by
1. Thus we win.

00SF

Lemma 133.7. Let k be a field. Let S be a local k-algebra essentially of finite

type over k. The following are equivalent:
(1) S is a complete intersection over k,
(2) for any surjection R S with R a regular local ring essentially of finite
presentation over k the ideal Ker(R S) can be generated by a regular
sequence,
(3) for some surjection R S with R a regular local ring essentially of finite
presentation over k the ideal Ker(R S) can be generated by dim(R)
dim(S) elements,
(4) there exists a global complete intersection A over k and a prime a of A such
that S
= Aa , and
(5) there exists a local complete intersection A over k and a prime a of A such
that S
= Aa .
Proof. It is clear that (2) implies (1) and (1) implies (3). It is also clear that
(4) implies (5). Let us show that (3) implies (4). Thus we assume there exists
a surjection R S with R a regular local ring essentially of finite presentation
over k such that the ideal Ker(R S) can be generated by dim(R) dim(S)
elements. We may write R = (k[x1 , . . . , xn ]/J)q for some J k[x1 , . . . , xn ] and
some prime q k[x1 , . . . , xn ] with J q. Let I k[x1 , . . . , xn ] be the kernel of
the map k[x1 , . . . , xn ] S so that S
= (k[x1 , . . . , xn ]/I)q . By assumption (I/J)q
is generated by dim(R) dim(S) elements. We conclude that Iq can be generated
by dim(k[x1 , . . . , xn ]q ) dim(S) elements by Lemma 133.6. From Lemma 133.4 we
see that for some g k[x1 , . . . , xn ], g 6 q the algebra (k[x1 , . . . , xn ]/I)g is a global
complete intersection and S is isomorphic to a local ring of it.
To finish the proof of the lemma we have to show that (5) implies (2). Assume (5)
and let : R S be a surjection with R a regular local k-algebra essentially of finite
type over k. By assumption we have S = Aa for some local complete intersection A
over k. Choose a presentation R = (k[y1 , . . . , ym ]/J)q with J q k[y1 , . . . , ym ].
We may and do assume that J is the kernel of the map k[y1 , . . . , ym ] R. Let
I k[y1 , . . . , ym ] be the kernel of the map k[y1 , . . . , ym ] S = Aa . Then J I
and (I/J)q is the kernel of the surjection : R S. So S = (k[y1 , . . . , ym ]/I)q .
By Lemma 125.7 we see that there exist g A, g 6 a and g 0 k[y1 , . . . , ym ], g 0 6 q
such that Ag
= (k[y1 , . . . , ym ]/I)g0 . After replacing A by Ag and k[y1 , . . . , ym ] by
k[y1 , . . . , ym+1 ] we may assume that A
= k[y1 , . . . , ym ]/I. Consider the surjective

COMMUTATIVE ALGEBRA

322

maps of local rings

k[y1 , . . . , ym ]q R S.
We have to show that the kernel of R S is generated by a regular sequence. By
Lemma 133.4 we know that k[y1 , . . . , ym ]q Aa = S has this property (as A is a
local complete intersection over k). We win by Lemma 133.6.

00SG

Lemma 133.8. Let k be a field. Let S be a finite type k-algebra. Let q be a prime
of S. The following are equivalent:
(1) The local ring Sq is a complete intersection ring (Definition 133.5).
(2) There exists a g S, g 6 q such that Sg is a local complete intersection
over k.
(3) There exists a g S, g 6 q such that Sg is a global complete intersection
over k.
(4) For any presentation S = k[x1 , . . . , xn ]/I with q0 k[x1 , . . . , xn ] corresponding to q any of the equivalent conditions (1) (5) of Lemma 133.4
hold.
Proof. This is a combination of Lemmas 133.4 and 133.7 and the definitions.

00SH

Lemma 133.9. Let k be a field. Let S be a finite type k-algebra. The following
are equivalent:
(1) The ring S is a local complete intersection over k.
(2) All local rings of S are complete intersection rings over k.
(3) All localizations of S at maximal ideals are complete intersection rings over
k.
Proof. This follows from Lemma 133.8, the fact that Spec(S) is quasi-compact
and the definitions.

The following lemma says that being a complete intersection is preserved under
change of base field (in a strong sense).

00SI

Lemma 133.10. Let k K be a field extension. Let S be a finite type algebra

over k. Let qK be a prime of SK = K k S and let q be the corresponding prime
of S. Then Sq is a complete intersection over k (Definition 133.5) if and only if
(SK )qK is a complete intersection over K.
Proof. Choose a presentation S = k[x1 , . . . , xn ]/I. This gives a presentation
SK = K[x1 , . . . , xn ]/IK where IK = K k I. Let q0K K[x1 , . . . , xn ], resp.
q0 k[x1 , . . . , xn ] be the corresponding prime. We will show that the equivalent conditions of Lemma 133.4 hold for the pair (S = k[x1 , . . . , xn ]/I, q) if and
only if they hold for the pair (SK = K[x1 , . . . , xn ]/IK , qK ). The lemma will follow
from this (see Lemma 133.8).
By Lemma 115.6 we have dimq S = dimqK SK . Hence the integer c occurring
in Lemma 133.4 is the same for the pair (S = k[x1 , . . . , xn ]/I, q) as for the pair
(SK = K[x1 , . . . , xn ]/IK , qK ). On the other hand we have
I k[x1 ,...,xn ] (q0 ) (q0 ) (q0K )

= I k[x1 ,...,xn ] (q0K )

= I k[x1 ,...,xn ] K[x1 , . . . , xn ] K[x1 ,...,xn ] (q0K )
=

(K k I) K[x1 ,...,xn ] (q0K )

= IK K[x1 ,...,xn ] (q0K ).

COMMUTATIVE ALGEBRA

323

Therefore, dim(q0 ) I k[x1 ,...,xn ] (q0 ) = dim(q0K ) IK K[x1 ,...,xn ] (q0K ). Thus it
follows from Nakayamas Lemma 19.1 that the minimal number of generators of
Iq0 is the same as the minimal number of generators of (IK )q0K . Thus the lemma
follows from characterization (2) of Lemma 133.4.

00SJ

Lemma 133.11. Let k K be a field extension. Let S be a finite type k-algebra.

Then S is a local complete intersection over k if and only if S k K is a local
complete intersection over K.
Proof. This follows from a combination of Lemmas 133.9 and 133.10. But we also
give a different proof here (based on the same principles).
Set S 0 = S k K. Let : k[x1 , . . . , xn ] S be a presentation with kernel I. Let
0 : K[x1 , . . . , xn ] S 0 be the induced presentation with kernel I 0 .
Suppose that S is a local complete intersection. Pick a prime q S 0 . Denote q0 the
corresponding prime of K[x1 , . . . , xn ], p the corresponding prime of S, and p0 the
corresponding prime of k[x1 , . . . , xn ]. Consider the following diagram of Noetherian
local rings
K[x1 , . . . , xn ]q0
Sq0 o
O
O
Sp o

k[x1 , . . . , xn ]p0

By Lemma 133.4 we know that Sp is cut out by some regular sequence f1 , . . . , fe

in k[x1 , . . . , xn ]p0 . Since the right vertical arrow is flat we see that the images of
f1 , . . . , fc form a regular sequence in K[x1 , . . . , xn ]q0 . Because tensoring with K over
k is an exact functor we have Sq0 = K[x1 , . . . , xn ]q0 /(f1 , . . . , fe ). Hence by Lemma
133.4 again we see that S 0 is a local complete intersection in a neighbourhood of q.
Since q was arbitrary we see that S 0 is a local complete intersection over K.
Suppose that S 0 is a local complete intersection. Pick a maximal ideal m of S. Let
m0 denote the corresponding maximal ideal of k[x1 , . . . , xn ]. Denote = (m) the
residue field. By Remark 16.8 the primes of S 0 lying over m correspond to primes
in K k . By the Hilbert-Nullstellensatz Theorem 33.1 we have [ : k] < .
Hence K k is finite nonzero over K. Hence K k has a finite number > 0
of primes which are all maximal, each of which has a residue field finite over K
(see Section 52). Hence there are finitely many > 0 prime ideals n S 0 lying over
m, each of which is maximal and has a residue field which is finite over K. Pick
one, say n S 0 , and let n0 K[x1 , . . . , xn ] denote the corresponding prime ideal of
K[x1 , . . . , xn ]. Note that since V (mS 0 ) is finite, we see that n is an isolated closed
point of it, and we deduce that mSn0 is an ideal of definition of Sn0 . This implies
that dim(Sm ) dim(Sn0 ), for example by Lemma 111.6 or by the characterization
of dimension in terms of minimal number of generators of ideal of definition, see
Section 59. (In reality the dimensions are equal but we do not need this.) Consider
the corresponding diagram of Noetherian local rings
Sn0 o
O

K[x1 , . . . , xn ]n0
O

Sm o

k[x1 , . . . , xn ]m0

COMMUTATIVE ALGEBRA

324

According to Lemma 132.8 we have NL() S S 0 = NL(0 ), in particular I 0 /(I 0 )2 =

I/I 2 S S 0 . Thus (I/I 2 )m Sm and (I 0 /(I 0 )2 )n Sn0 (n) have the same dimension.
Since (I 0 /(I 0 )2 )n is free of rank ndim Sn0 we deduce that (I/I 2 )m can be generated
by n dim Sn0 n dim Sm elements. By Lemma 133.4 we see that S is a local
complete intersection in a neighbourhood of m. Since m was any maximal ideal we
conclude that S is a local complete intersection.

We end with a lemma which we will later use to prove that given ring maps T
A B where B is syntomic over T , and B is syntomic over A, then A is syntomic
over T .
02JP

Lemma 133.12. Let

BO o

SO

Ao

be a commutative square of local rings. Assume

(1)
(2)
(3)
(4)

R and S = S/mR S are regular local rings,

A = R/I and B = S/J for some ideals I, J,
J S and J = J/mR J S are generated by regular sequences, and
A B and R S are flat.

Then I is generated by a regular sequence.

Proof. Set B = B/mR B = B/mA B so that B = S/J. Let f1 , . . . , fc J be
elements such that f 1 , . . . , f c J form a regular sequence generating J. Note that
c = dim(S) dim(B), see Lemma 133.6. By Lemma 98.3 the ring S/(f1 , . . . , fc )
is flat over R. Hence S/(f1 , . . . , fc ) + IS is flat over A. The map S/(f1 , . . . , fc ) +
IS B is therefore a surjection of finite S/IS-modules flat over A which is an
isomorphism modulo mA , and hence an isomorphism by Lemma 98.1. In other
words, J = (f1 , . . . , fc ) + IS.
By Lemma 133.6 again the ideal J is generated by a regular sequence of c =
dim(S) dim(B) elements. Hence J/mS J is a vector space of dimension c. By
the description of J above there exist g1 , . . . , gcc I such that J is generated
by f1 , . . . , fc , g1 , . . . , gcc (use Nakayamas Lemma 19.1). Consider the ring A0 =
R/(g1 , . . . , gcc ) and the surjection A0 A. We see from the above that B =
S/(f1 , . . . , fc , g1 , . . . , gcc ) is flat over A0 (as S/(f1 , . . . , fc ) is flat over R). Hence
A0 B is injective (as it is faithfully flat, see Lemma 38.17). Since this map
factors through A we get A0 = A. Note that dim(B) = dim(A) + dim(B), and
dim(S) = dim(R) + dim(S), see Lemma 111.7. Hence c c = dim(R) dim(A)
by elementary algebra. Thus I = (g1 , . . . , gcc ) is generated by a regular sequence
according to Lemma 133.6.

134. Syntomic morphisms
00SK

Syntomic ring maps are flat finitely presented ring maps all of whose fibers are local
complete intersections. We discuss general local complete intersection ring maps in
More on Algebra, Section 25.

COMMUTATIVE ALGEBRA

00SL

325

Definition 134.1. A ring map R S is called syntomic, or we say S is a flat

local complete intersection over R if it is flat, of finite presentation, and if all of its
fibre rings S R (p) are local complete intersections, see Definition 133.1.
Clearly, an algebra over a field is syntomic over the field if and only if it is a local
complete intersection. Here is a pleasing feature of this definition.

00SM

Lemma 134.2. Let R S be a ring map. Let R R0 be a faithfully flat ring

map. Set S 0 = R0 R S. Then R S is syntomic if and only if R0 S 0 is
syntomic.
Proof. By Lemma 125.2 and Lemma 38.8 this holds for the property of being flat
and for the property of being of finite presentation. The map Spec(R0 ) Spec(R)
is surjective, see Lemma 38.16. Thus it suffices to show given primes p0 R0 lying
over p R that S R (p) is a local complete intersection if and only if S 0 R0 (p0 )
is a local complete intersection. Note that S 0 R0 (p0 ) = S R (p) (p) (p0 ).
Thus Lemma 133.11 applies.


00SN

Lemma 134.3. Any base change of a syntomic map is syntomic.

Proof. This is true for being flat, for being of finite presentation, and for having
local complete intersections as fibres by Lemmas 38.7, 6.2 and 133.11.


00SO

Lemma 134.4. Let R S be a ring map. Suppose we have g1 , . . . gm S

which generate the unit ideal such that each R Sgi is syntomic. Then R S is
syntomic.
Proof. This is true for being flat and for being of finite presentation by Lemmas 38.19 and 23.3. The property of having fibre rings which are local complete
intersections is local on S by its very definition, see Definition 133.1.


00SP

Definition 134.5. Let R S be a ring map. We say that R S is a relative

global complete intersection if we are given a presentation S = R[x1 , . . . , xn ]/(f1 , . . . , fc )
such that every nonempty fibre has dimension n c.
The following lemma is occasionally useful to find global presentations.

07CF

Lemma 134.6. Let S be a finitely presented R-algebra which has a presentation

S = R[x1 , . . . , xn ]/I such that I/I 2 is free over S. Then S has a presentation
S = R[y1 , . . . , ym ]/(f1 , . . . , fc ) such that (f1 , . . . , fc )/(f1 , . . . , fc )2 is free with basis
given by the classes of f1 , . . . , fc .
Proof. Note that I is a finitely generated ideal by Lemma 6.3. Let f1 , . . . , fc I
be elements which map to a basis of I/I 2 . By Nakayamas lemma (Lemma 19.1)
there exists a g 1 + I such that
g I (f1 , . . . , fc )
Hence we see that
S
= R[x1 , . . . , xn ]/(f1 , . . . , fc )[1/g]
= R[x1 , . . . , xn , xn+1 ]/(f1 , . . . , fc , gxn+1 1)
as desired. It follows that f1 , . . . , fc , gxn+1 1 form a basis for (f1 , . . . , fc , gxn+1
1)/(f1 , . . . , fc , gxn+1 1)2 for example by applying Lemma 132.12.


COMMUTATIVE ALGEBRA

00SQ

326

Example 134.7. Let n, m 1 be integers. Consider the ring map

R = Z[a1 , . . . , an+m ]

S = Z[b1 , . . . , bn , c1 , . . . , cm ]

a1

b1 + c1

a2

b2 + b1 c1 + c2

...

...

...

an+m

bn cm

In other words, this is the unique ring map of polynomial rings as indicated such
that the polynomial factorization
xn + a1 xn1 + . . . + an+m = (xn + b1 xn1 + . . . + bn )(xm + c1 xm1 + . . . + cm )
holds. Note that S is generated by n + m elements over R (namely, bi , cj ) and that
there are n + m equations (namely ak = ak (bi , cj )). In order to show that S is a
relative global complete intersection over R it suffices to prove that all fibres have
dimension 0.
To prove this, let R k be a ring map into a field k. Say ai maps to i k.
Consider the fibre ring Sk = k R S. Let k K be a field extension. A k-algebra
map of Sk K is the same thing as finding 1 , . . . , n , 1 , . . . , m K such that
xn + 1 xn1 + . . . + n+m = (xn + 1 xn1 + . . . + n )(xm + 1 xm1 + . . . + m ).
Hence we see there are at most finitely many choices of such n+m-tuples in K. This
proves that all fibres have finitely many closed points (use Hilberts Nullstellensatz
to see they all correspond to solutions in k for example) and hence that R S is
a relative global complete intersection.
Another way to argue this is to show Z[a1 , . . . , an+m ] Z[b1 , . . . , bn , c1 , . . . , cm ] is
actually also a finite ring map. Namely, by Lemma 37.5 each of bi , cj is integral
over R, and hence R S is finite by Lemma 35.4.
00SR

Example 134.8. Consider the ring map

R = Z[a1 , . . . , an ]
a1

S = Z[1 , . . . , n ]
1 + . . . + n

...

...

...

an

1 . . . n

In other words this is the unique ring map of polynomial rings as indicated such
that
Yn
xn + a1 xn1 + . . . + an =
(x + i )
i=1

holds in Z[i , x]. Another way to say this is that ai maps to the ith elementary symmetric function in 1 , . . . , n . Note that S is generated by n elements over R subject
to n equations. Hence to show that S is a global relative complete intersection over
R we have to show that the fibre rings S R (p) have dimension 0. This follows as
in Example 134.7 because the ring map Z[a1 , . . . , an ] Z[1 , . . . , n ] is actually
finite since each i S satisfies the monic equation xn a1 xn1 + . . . + (1)n an
over R.
03HS

Lemma 134.9. Suppose that A is a ring, and P (x) = xn +b1 xn1 +. . .+bn A[x]
is a monic polynomial over A. Then there exists a syntomic, finite locally free,

COMMUTATIVE ALGEBRA

faithfully flat ring extension A A0 such that P (x) =

i A0 .

327

i=1,...,n (x i )

for certain

Proof. Take A0 = A R S, where R and S are as in Example 134.8, where R A

maps ai to bi , and let i = 1 i .

00SS

Lemma 134.10. Let S = R[x1 , . . . , xn ]/(f1 , . . . , fc ) be a relative global complete

intersection over R.
(1) For any R R0 the base change R0 R S = R0 [x1 , . . . , xn ]/(f1 , . . . , fc ) is a
relative global complete intersection.
(2) For any g S which is the image of h R[x1 , . . . , xn ] the ring Sg =
R[x1 , . . . , xn , xn+1 ]/(f1 , . . . , fc , hxn+1 1) is a relative global complete intersection.
(3) If R S factors as R Rf S for some f R. Then the ring
S = Rf [x1 , . . . , xn ]/(f1 , . . . , fc ) is a relative global complete intersection
over Rf .
Proof. By Lemma 115.5 the fibres of a base change have the same dimension
as the fibres of the original map. Moreover R0 R R[x1 , . . . , xn ]/(f1 , . . . , fc ) =
R0 [x1 , . . . , xn ]/(f1 , . . . , fc ). Thus (1) follows. The proof of (2) is that the localization at one element can be described as Sg
= S[xn+1 ]/(gxn+1 1). Assertion (3)
follows from (1) since under the assumptions of (3) we have Rf R S

= S.

00ST

Lemma 134.11. Let R be a ring. Let S = R[x1 , . . . , xn ]/(f1 , . . . , fc ). We will

find h R[x1 , . . . , xn ] which maps to g S such that
Sg = R[x1 , . . . , xn , xn+1 ]/(f1 , . . . , fc , hxn+1 1)
is a relative global complete intersection over R in each of the following cases:
(1) Let I R be an ideal. If the fibres of Spec(S/IS) Spec(R/I) have
dimension n c, then we can find (h, g) as above such that g maps to
1 S/IS.
(2) Let p R be a prime. If dim(S R (p)) = n c, then we can find (h, g)
as above such that g maps to a unit of S R (p).
(3) Let q S be a prime lying over p R. If dimq (S/R) = n c, then we can
find (h, g) as above such that g 6 q.
Proof. Ad (1). By Lemma 124.6 there exists an open subset W Spec(S) containing V (IS) such that all fibres of W Spec(R) have dimension n c. Say
W = Spec(S) \ V (J). Then V (J) V (IS) = hence we can find a g J which
maps to 1 S/IS. Let h R[x1 , . . . , xn ] be any preimage of g.
Ad (2). By Lemma 124.6 there exists an open subset W Spec(S) containing
Spec(S R (p)) such that all fibres of W Spec(R) have dimension n c. Say
W = Spec(S) \ V (J). Then V (J S R (p)) = . Hence we can find a g J
which maps to a unit in S R (p) (details omitted). Let h R[x1 , . . . , xn ] be any
preimage of g.
Ad (3). By Lemma 124.6 there exists a g S, g 6 q such that all nonempty fibres
of R Sg have dimension n c. Let h R[x1 , . . . , xn ] be any element that
maps to g.

The following lemma says we can do absolute Noetherian approximation for relative
complete intersections.

COMMUTATIVE ALGEBRA

00SU

328

Lemma 134.12. Let R be a ring. Let S be a relative global complete intersection with presentation S = R[x1 , . . . , xn ]/(f1 , . . . , fc ). There exist a finite type
Z-subalgebra R0 R such that fi R0 [x1 , . . . , xn ] and such that
S0 = R0 [x1 , . . . , xn ]/(f1 , . . . , fc )
is a relative global intersection over R0 .
Proof. Let R0 R be the Z-algebra of R generated by all the coefficients of the
polynomials f1 , . . . , fc . Let S0 = R0 [x1 , . . . , xn ]/(f1 , . . . , fc ). Clearly, S = RR0 S0 .
Pick a prime q S and denote p R, q0 S0 , and p0 R0 the primes it lies
over. Because dim(S R (p)) = n c we also have dim(S0 R0 (p0 )) = n c,
see Lemma 115.5. By Lemma 124.6 there exists a g S0 , g 6 q0 such that all
nonempty fibres of R0 (S0 )g have dimension n c. As q was arbitrary and
Spec(S) quasi-compact, we can find finitely many g1 , . . . , gm S0 such that (a) for
j = 1, . . . , m the nonempty fibres of R0 (S0 )gj have dimension n c and (b)
the image of Spec(S) Spec(S0 ) is contained in D(g1 ) . . . D(gm ). In other
words, the images of g1 , . . . , gm in S = R R0 S0 generate the unit ideal. After
increasing R0 we may assume that g1 , . . . , gm generate the unit ideal in S0 . By (a)
the nonempty fibres of R0 S0 all have dimension n c and we conclude. 

00SV

Lemma 134.13. Let R be a ring. Let S = R[x1 , . . . , xn ]/(f1 , . . . , fc ) be a relative

global complete intersection. For every prime q of S, let q0 denote the corresponding
prime of R[x1 , . . . , xn ]. Then
(1) f1 , . . . , fc is a regular sequence in the local ring R[x1 , . . . , xn ]q0 ,
(2) each of the rings R[x1 , . . . , xn ]q0 /(f1 , . . . , fi ) is flat over R, and
(3) the S-module (f1 , . . . , fc )/(f1 , . . . , fc )2 is free with basis given by the elements fi mod (f1 , . . . , fc )2 .
Proof. First, by Lemma 68.2, part (3) follows from part (1). Parts (1) and (2)
immediately reduce to the Noetherian case by Lemma 134.12 (some minor details
omitted). Assume R is Noetherian. By Lemma 133.4 for example we see that
f1 , . . . , fc form a regular sequence in the local ring R[x1 , . . . , xn ]q0 R (p). Moreover, the local ring R[x1 , . . . , xn ]q0 is flat over Rp . Since R, and hence R[x1 , . . . , xn ]q0
is Noetherian we may apply Lemma 98.3 to conclude.


00SW

Lemma 134.14. A relative global complete intersection is syntomic, i.e., flat.

Proof. Let R S be a relative global complete intersection. The fibres are global
complete intersections, and S is of finite presentation over R. Thus the only thing
to prove is that R S is flat. This is true by (2) of Lemma 134.13.


00SY

Lemma 134.15. Let R S be a ring map. Let q S be a prime lying over the
prime p of R. The following are equivalent:
(1) There exists an element g S, g 6 q such that R Sg is syntomic.
(2) There exists an element g S, g 6 q such that Sg is a relative global
complete intersection over R.
(3) There exists an element g S, g 6 q, such that R Sg is of finite
presentation, the local ring map Rp Sq is flat, and the local ring Sq /pSq
is a complete intersection ring over (p) (see Definition 133.5).
Proof. The implication (1) (3) is Lemma 133.8. The implication (2) (1) is
Lemma 134.14. It remains to show that (3) implies (2).

COMMUTATIVE ALGEBRA

329

Assume (3). After replacing S by Sg for some g S, g 6 q we may assume

S is finitely presented over R. Choose a presentation S = R[x1 , . . . , xn ]/I. Let
c = n dimq (S/R). Let q0 R[x1 , . . . , xn ] be the prime corresponding to q. Write
(p) = k. Note that S R k = k[x1 , . . . , xn ]/I where I k[x1 , . . . , xn ] is the ideal
generated by the image of I. Let q0 k[x1 , . . . , xn ] be the prime ideal generated
by the image of q0 . By Lemma 133.8 we see that Lemma 133.4 holds for I and q0 .
Thus the dimension of I q0 /q0 I q0 over (q0 ) is c. Pick f1 , . . . , fc I mapping to a
basis of this vector space. The images f j I generate I q0 (by Lemma 133.4). Set
S 0 = R[x1 , . . . , xn ]/(f1 , . . . , fc ). Let J be the kernel of the surjection S 0 S. Since
S is of finite presentation J is a finitely generated ideal (Lemma 6.2). Consider the
short exact sequence
0 J S0 S 0
As Sq is flat over R we see that Jq0 R k Sq0 0 R k is injective (Lemma 38.12).
However, by construction Sq0 0 R k maps isomorphically to Sq R k. Hence we
conclude that Jq0 R k = Jq0 /pJq0 = 0. By Nakayamas lemma (Lemma 19.1) we
conclude that there exists a g R[x1 , . . . , xn ], g 6 q0 such that Jg = 0. In other
words Sg0
= Sg . After further localizing we see that S 0 (and hence S) becomes a
relative global complete intersection by Lemma 134.11 as desired.

07BT

Lemma 134.16. Let R be a ring. Let S = R[x1 , . . . , xn ]/I for some finitely
generated ideal I. If g S is such that Sg is syntomic over R, then (I/I 2 )g is a
finite projective Sg -module.
Proof. By Lemma 134.15 there exist finitely many elements g1 , . . . , gm S which
generate the unit ideal in Sg such that each Sggj is a relative global complete
intersection over R. Since it suffices to prove that (I/I 2 )ggj is finite projective, see
Lemma 77.2, we may assume that Sg is a relative global complete intersection. In
this case the result follows from Lemmas 132.16 and 134.13.


00SZ

Lemma 134.17. Let R S, S S 0 be ring maps.

(1) If R S and S S 0 are syntomic, then R S 0 is syntomic.
(2) If R S and S S 0 are relative global complete intersections, then R
S 0 is a relative global complete intersection.
Proof. Assume R S and S S 0 are syntomic. This implies that R S 0 is flat
by Lemma 38.4. It also implies that R S 0 is of finite presentation by Lemma 6.2.
Thus it suffices to show that the fibres of R S 0 are local complete intersections.
Choose a prime p R. We have a factorization
(p) S R (p) S 0 R (p).
By assumption S R (p) is a local complete intersection, and by Lemma 134.3 we
see that S R (p) is syntomic over S R (p). After replacing S by S R (p) and
S 0 by S 0 R (p) we may assume that R is a field. Say R = k.
Choose a prime q0 S 0 lying over the prime q of S. Our goal is to find a g 0 S 0 ,
g 0 6 q0 such that Sg0 0 is a global complete intersection over k. Choose a g S, g 6 q
such that Sg = k[x1 , . . . , xn ]/(f1 , . . . , fc ) is a global complete intersection over k.
Since Sg Sg0 is still syntomic also, and g 6 q0 we may replace S by Sg and S 0 by
Sg0 and assume that S = k[x1 , . . . , xn ]/(f1 , . . . , fc ) is a global complete intersection
over k. Next we choose a g 0 S 0 , g 0 6 q0 such that S 0 = S[y1 , . . . , ym ]/(h1 , . . . , hd )

COMMUTATIVE ALGEBRA

330

is a relative global complete intersection over S. Hence we have reduced to part

(2) of the lemma.
Suppose that R S and S S 0 are relative global complete intersections. Say
S = R[x1 , . . . , xn ]/(f1 , . . . , fc ) and S 0 = S[y1 , . . . , ym ]/(h1 , . . . , hd ). Then
S0
= R[x1 , . . . , xn , y1 , . . . , ym ]/(f1 , . . . , fc , h0 , . . . , h0 )
1

h0j

for some lifts

R[x1 , . . . , xn , y1 , . . . , ym ] of the hj . Hence it suffices to bound
the dimensions of the fibres. Thus we may yet again assume R = k is a field.
In this case we see that we have a ring, namely S, which is of finite type over k
and equidimensional of dimension n c, and a finite type ring map S S 0 all
of whose nonempty fibre rings are equidimensional of dimension m d. Then, by
Lemma 111.6 for example applied to localizations at maximal ideals of S 0 , we see
that dim(S 0 ) n c + m d as desired.

The following lemma will be improved later, see Smoothing Ring Maps, Proposition
4.2.
00T0

Lemma 134.18. Let R be a ring and let I R be an ideal. Let R/I S be

a syntomic map. Then there exists elements g i S which generate the unit ideal
of S such that each S gi
= Si /ISi for some relative global complete intersection Si
over R.
Proof. By Lemma 134.15 we find a collection of elements g i S which generate
the unit ideal of S such that each S gi is a relative global complete intersection
over R/I. Hence we may assume that S is a relative global complete intersection. Write S = (R/I)[x1 , . . . , xn ]/(f 1 , . . . , f c ) as in Definition 134.5. Choose
f1 , . . . , fc R[x1 , . . . , xn ] lifting f 1 , . . . , f c . Set S = R[x1 , . . . , xn ]/(f1 , . . . , fc ).
Note that S/IS
= S. By Lemma 134.11 we can find g S mapping to 1 in S such
that Sg is a relative global complete intersection over R. Since S
= Sg /ISg this
finishes the proof.

135. Smooth ring maps

00T1

Let us motivate the definition of a smooth ring map by an example. Suppose R is

a ring and S = R[x, y]/(f ) for some nonzero f R[x, y]. In this case there is an
exact sequence
S Sdx Sdy S/R 0
f
where the first arrow maps 1 to f
x dx + y dy see Section 132. We conclude that
S/R is locally free of rank 1 if the partial derivatives of f generate the unit ideal
in S. In this case S is smooth of relative dimension 1 over R. But it can happen
that S/R is locally free of rank 2 namely if both partial derivatives of f are zero.
For example if for a prime p we have p = 0 in R and f = xp + y p then this happens.
Here R S is a relative global complete intersection of relative dimension 1 which
is not smooth. Hence, in order to check that a ring map is smooth it is not sufficient
to check whether the module of differentials is free. The correct condition is the
following.

00T2

Definition 135.1. A ring map R S is smooth if it is of finite presentation

and the naive cotangent complex NLS/R is quasi-isomorphic to a finite projective
S-module placed in degree 0.

COMMUTATIVE ALGEBRA

331

In particular, if R S is smooth then the module S/R is a finite projective

S-module. Moreover, by Lemma 135.2 the naive cotangent complex of any presentation has the same structure. Thus, for a surjection : R[x1 , . . . , xn ] S with
kernel I the map
I/I 2 R[x1 ,...,xn ]/R R[x1 ,...,xn ] S
Ln
is a split injection. In other words
Sdxi
= I/I 2 S/R as S-modules. This
i=1

implies that I/I 2 is a finite projective S-module too!

05GK

Lemma 135.2. Let R S be a ring map of finite presentation. If for some presentation of S over R the naive cotangent complex NL() is quasi-isomorphic to
a finite projective S-module placed in degree 0, then this holds for any presentation.
Proof. Immediate from Lemma 132.2.

00T3

Lemma 135.3. Let R S be a smooth ring map. Any localization Sg is smooth

over R. If f R maps to an invertible element of S, then Rf S is smooth.
Proof. By Lemma 132.13 the naive cotangent complex for Sg over R is the base
change of the naive cotangent complex of S over R. The assumption is that the
naive cotangent complex of S/R is S/R and that this is a finite projective Smodule. Hence so is its base change. Thus Sg is smooth over R.
The second assertion follows in the same way from Lemma 132.11.

00T4

Lemma 135.4. Let R S be a smooth ring map. Let R R be any ring map.
Then the base change R0 S 0 = R0 R S is smooth.
Proof. Let : R[x1 , . . . , xn ] S be a presentation with kernel I. Let 0 :
R0 [x1 , . . . , xn ] R0 R S be the induced presentation. Let I 0 = Ker(0 ). Since
0 I R[x1 , . . . , xn ] S 0 is exact, the sequence R0 R I R0 [x1 , . . . , xn ]
R0 R S 0 is exact. Thus R0 R I I 0 is surjective. By Definition 135.1 there
is a short exact sequence
0 I/I 2 R[x1 ,...,xn ]/R R[x1 ,...,xn ] S S/R 0
and the S-module S/R is finite projective. In particular I/I 2 is a direct summand
of R[x1 ,...,xn ]/R R[x1 ,...,xn ] S. Consider the commutative diagram
R0 R (I/I 2 )

/ R0 R (R[x ,...,x ]/R R[x ,...,x ] S)

1
n
1
n


I 0 /(I 0 )2


/ R0 [x ,...,x ]/R0 R0 [x ,...,x ] (R0 R S)
1
n
1
n

Since the right vertical map is an isomorphism we see that the left vertical map is
injective and surjective by what was said above. Thus we conclude that NL(0 ) is
quasi-isomorphic to S 0 /R0
= S 0 S S/R . And this is finite projective since it is
the base change of a finite projective module.

00T5

Lemma 135.5. Let k be a field. Let S be a smooth k-algebra. Then S is a local

complete intersection.
Proof. By Lemmas 135.4 and 133.11 it suffices to prove this when k is algebraically
closed. Choose a presentation : k[x1 , . . . , xn ] S with kernel I. Let m be a maximal ideal of S, and let m0 I be the corresponding maximal ideal of k[x1 , . . . , xn ].

COMMUTATIVE ALGEBRA

332

We will show that condition (5) of Lemma 133.4 holds (with m instead of q). We
may write m0 = (x1 a1 , . . . , xn an ) for some ai k, because k is algebraically
closed, see Theorem
By our assumption that k S is smooth the S-module
L33.1.
n
2
map d : I/I

Sdx
i is a split injection. Hence the corresponding map
i=1
L
I/m0 I
(m0 )dxi is injective. Say dim(m0 ) (I/m0 I) = c and pick f1 , . . . , fc I
which map to a (m0 )-basis of I/m0 I. By Nakayamas Lemma 19.1 we see that
f1 , . . . , fc generate Im0 over k[x1 , . . . , xn ]m0 . Consider the commutative diagram


k[x1 ,...,xn ]/k

/ I/m0 I

/ I/I 2

L 
Sdxi

dxi 7xi ai


/ m0 /(m0 )2

(proof commutativity omitted). The middle vertical map is the one defining the
naive cotangent complex
of . Note that the right lower horizontal arrow induces
L
an isomorphism
(m0 )dxi m0 /(m0 )2 . Hence our generators f1 , . . . , fc of Im0
map to a collection of elements in k[x1 , . . . , xn ]m0 whose classes in m0 /(m0 )2 are
linearly independent over (m0 ). Therefore they form a regular sequence in the
ring k[x1 , . . . , xn ]m0 by Lemma 105.3. This verifies condition (5) of Lemma 133.4
hence Sg is a global complete intersection over k for some g S, g 6 m. As this
works for any maximal ideal of S we conclude that S is a local complete intersection
over k.

00T6

Definition 135.6. Let R be a ring. Given integers n c 0 and f1 , . . . , fc

R[x1 , . . . , xn ] we say
S = R[x1 , . . . , xn ]/(f1 , . . . , fc )
is a standard smooth algebra over R if the polynomial

f1 /x1 f2 /x1 . . . fc /x1

f1 /x2 f2 /x2 . . . fc /x2

g = det
...
...
...
...
f1 /xc f2 /xc . . . fc /xc
maps to an invertible element in S.

00T7

Lemma 135.7. Let S = R[x1 , . . . , xn ]/(f1 , . . . , fc ) = R[x1 , . . . , xn ]/I be a standard smooth algebra. Then
(1) the ring map R S is smooth,
(2) the S-module S/R is free on dxc+1 , . . . , dxn ,
(3) the S-module I/I 2 is free on the classes of f1 , . . . , fc ,
(4) for any g S the ring map R Sg is standard smooth,
(5) for any ring map R R0 the base change R0 R0 R S is standard smooth,
(6) if f R maps to an invertible element in S, then Rf S is standard
smooth, and
(7) the ring S is a relative global complete intersection over R.
Proof. Consider the naive cotangent complex of the given presentation
Mn
(f1 , . . . , fc )/(f1 , . . . , fc )2
Sdxi
i=1

Let us compose this map with the projection onto the first c direct summands
of the direct sum. According to the definition of a standard smooth algebra

COMMUTATIVE ALGEBRA

333

Lc
the classes fi mod (f1 , . . . , fc )2 map to a basis of
i=1 Sdxi . We conclude that
(f1 , . . . , fc )/(f1 , . . . , fc )2 is free of rank c with a basis given by the elements fi mod
(f1 , . . . , fc )2 , and that the homology in degree 0, i.e., S/R , of the naive cotangent
complex is a free S-module with basis the images of dxc+j , j = 1, . . . , n c. In
particular, this proves R S is smooth.
The proofs of (4) and (6) are omitted. But see the example below and the proof of
Lemma 134.10.
Let : R R0 be any ring map. Denote S 0 = R0 [x1 , . . . , xn ]/(f1 , . . . , fc ) where
f is the polynomial obtained from f R[x1 , . . . , xn ] by applying to all the
coefficients. Then S 0
= R0 R S. Moreover, the determinant of Definition 135.6 for
0
0

S /R is equal to g . Its image in S 0 is therefore the image of g via R[x1 , . . . , xn ]

S S 0 and hence invertible. This proves (5).
To prove (7) it suffices to show that S R (p) has dimension n c. By (5) it
suffices to prove that any standard smooth algebra k[x1 , . . . , xn ]/(f1 , . . . , fc ) over
a field k has dimension n c. We already know that k[x1 , . . . , xn ]/(f1 , . . . , fc ) is a
local complete intersection by Lemma 135.5. Hence, since I/I 2 is free of rank c we
see that it dimension n c, by Lemma 133.4 for example.

00T8

Example 135.8. Let R be a ring. Let f1 , . . . , fc R[x1 , . . . , xn ]. Let

f1 /x1 f2 /x1 . . . fc /x1

f1 /x2 f2 /x2 . . . fc /x2
.
h = det
...
...
...
...
f1 /xc f2 /xc . . . fc /xc
Set S = R[x1 , . . . , xn+1 ]/(f1 , . . . , fc , xn+1 h 1). This is an example of a standard
smooth algebra, except that the presentation is wrong and the variables should be
in the following order: x1 , . . . , xc , xn+1 , xc+1 , . . . , xn .

00T9

Lemma 135.9. A composition of standard smooth ring maps is standard smooth.

Proof. Suppose that R S and S S 0 are standard smooth. We choose presentations S = R[x1 , . . . , xn ]/(f1 , . . . , fc ) and S 0 = S[y1 , . . . , ym ]/(g1 , . . . , gd ). Choose
elements gj0 R[x1 , . . . , xn , y1 , . . . , ym ] mapping to the gj . In this way we see
S 0 = R[x1 , . . . , xn , y1 , . . . , ym ]/(f1 , . . . , fc , g10 , . . . , gd0 ). To show that S 0 is standard
smooth it suffices to verify that the determinant

f1 /x1 . . . fc /x1 g1 /x1 . . . gd /x1

...
...
...
...
...
...

f1 /xc . . . fc /xc g1 /xc . . . gd /xc

det
0
...
0
g1 /y1 . . . gd /y1

...
...
...
...
...
...
0
...
0
g1 /yd . . . gd /yd
is invertible in S 0 . This is clear since it is the product of the two determinants
which were assumed to be invertible by hypothesis.


00TA

Lemma 135.10. Let R S be a smooth ring map. There exists an open covering
of Spec(S) by standard opens D(g) such that each Sg is standard smooth over R.
In particular R S is syntomic.

COMMUTATIVE ALGEBRA

334

Proof. Choose a presentation : R[x1 , . . . , xn ] S with kernel I = (f1 , . . . , fm ).

For every subset E {1, . . . , m} consider the open subset UE where the classes
fe , e E freely generate the finite projective S-module I/I 2 , see Lemma 78.3. We
may cover Spec(S) by standard opens D(g) each completely contained in one of
the opens UE . For such a g we look at the presentation
: R[x1 , . . . , xn , xn+1 ] Sg
mapping xn+1 to 1/g. Setting J = Ker() we use Lemma 132.12 to see that
J/J 2
= (I/I 2 )g Sg is free. We may and do replace S by Sg . Then using Lemma
134.6 we may assume we have a presentation : R[x1 , . . . , xn ] S with kernel
I = (f1 , . . . , fc ) such that I/I 2 is free on the classes of f1 , . . . , fc .
Using the presentation obtained at the end of the previous paragraph, we more
or less repeat this argument with the basis elements dx1 , . . . , dxn of R[x1 ,...,xn ]/R .
Namely, for any subset E {1, . . . , n} of cardinality c we may consider the open
subset UE of Spec(S) where the differential of NL() composed with the projection
M
S c
Sdxi
= I/I 2 R[x1 ,...,xn ]/R R[x1 ,...,xn ] S
iE

is an isomorphism. Again we may find a covering of Spec(S) by (finitely many)

standard opens D(g) such that each D(g) is completely contained in one of the opens
UE . By renumbering, we may assume E = {1, . . . , c}. For a g with D(g) UE we
look at the presentation
: R[x1 , . . . , xn , xn+1 ] Sg
mapping xn+1 to 1/g. Setting J = Ker() we conclude from Lemma 132.12 that
J = (f1 , . . . , fc , f xn+1 1) where (f ) = g and that the composition
Mc
J/J 2 R[x1 ,...,xn+1 ]/R R[x1 ,...,xn+1 ] Sg
Sg dxi Sg dxn+1
i=1

is an isomorphism. Reordering the coordinates as x1 , . . . , xc , xn+1 , xc+1 , . . . , xn we

we conclude that Sg is standard smooth over R as desired.
This finishes the proof as standard smooth algebras are syntomic (Lemmas 135.7
and 134.14) and being syntomic over R is local on S (Lemma 134.4).

00TB

Definition 135.11. Let R S be a ring map. Let q be a prime of S. We say

R S is smooth at q if there exists a g S, g 6 q such that R Sg is smooth.
For ring maps of finite presentation we can characterize this as follows.

07BU

Lemma 135.12. Let R S be of finite presentation. Let q be a prime of S. The

following are equivalent
(1) R S is smooth at q,
(2) H1 (LS/R )q = 0 and S/R,q is a projective Sq -module, and
(3) H1 (LS/R )q = 0 and S/R,q is a flat Sq -module.
Proof. We will use without further mention that formation of the naive cotangent
complex commutes with localization, see Section 132, especially Lemma 132.13. It
is clear that (1) implies (2) implies (3). Assume (3) holds. Note that S/R is
a finitely presented S-module, see Lemma 130.15. Hence S/R,q is a finite free
module by Lemma 77.4. Writing Sq as the colimit of principal localizations we see
from Lemma 126.4 that we can find a g S, g 6 q such that (S/R )g is finite free.

COMMUTATIVE ALGEBRA

335

Choose a presentation : R[x1 , . . . , xn ] S with kernel I. We may work with

NL() instead of NLS/R , see Lemma 132.2. The surjection
R[x1 ,...,xn ]/R R S S/R 0
has a right inverse after inverting g because (S/R )g is projective. Hence the image
of d : (I/I 2 )g R[x1 ,...,xn ]/R R Sg is a direct summand and this map has a right
inverse too. We conclude that H1 (LS/R )g is a quotient of (I/I 2 )g . In particular
H1 (LS/R )g is a finite Sg -module. Thus the vanishing of H1 (LS/R )q implies the
vanishing of H1 (LS/R )gg0 for some g 0 S, g 0 6 q. Then R Sgg0 is smooth by
definition.

00TC

Lemma 135.13. Let R S be a ring map. Then R S is smooth if and only

if R S is smooth at every prime q of S.
Proof. The direct implication is trivial. Suppose that R S is smooth at every
prime q of S. Since Spec(S)
S is quasi-compact, see Lemma 16.10, there exists a
finite covering Spec(S) = D(gi ) such that each Sgi is smooth. By Lemma 23.3
this implies that S is of finite presentation over R. According to Lemma 132.13
we see that NLS/R S Sgi is quasi-isomorphic to a finite projective Sgi -module.
By Lemma 77.2 this implies that NLS/R is quasi-isomorphic to a finite projective
S-module.


00TD

Lemma 135.14. A composition of smooth ring maps is smooth.

Proof. This follows from a combination of Lemmas 135.10, 135.9 and 135.13. (You
can also prove this in many different ways; including easier ones.)


00TE

Lemma 135.15. Let R be a ring. Let S = R[x1 , . . . , xn ]/(f1 , . . . , fc ) be a relative

global complete intersection. Let q S be a prime. Then R S is smooth at
q if and only if there exists a subset I {1, . . . , n} of cardinality c such that the
polynomial
gI = det(fj /xi )j=1,...,c, iI .
does not map to an element of q.
Proof. By Lemma 134.13 we see that the naive cotangent complex associated to
the given presentation of S is the complex
Mc
Mn
X fj
dxi .
S fj
S dxi , fj 7
j=1
i=1
xi
The maximal minors of the matrix giving the map are exactly the polynomials gI .
Assume gI maps to g S, with g 6 q. Then the algebra Sg is smooth over
R. Namely, its naive cotangent complex is quasi-isomorphic to the complex above
localized at g, see Lemma 132.13. And by construction it is quasi-isomorphic to a
free rank n c module in degree 0.
Conversely, suppose that all gI end up in q. In this case the complex above tensored
with (q) does not have maximal rank, and hence there is no localization by an
element g S, g 6 q where this map becomes a split injection. By Lemma 132.13
again there is no such localization which is smooth over R.


00TF

Lemma 135.16. Let R S be a ring map. Let q S be a prime lying over the
prime p of R. Assume
(1) there exists a g S, g 6 q such that R Sg is of finite presentation,

COMMUTATIVE ALGEBRA

336

(2) the local ring homomorphism Rp Sq is flat,

(3) the fibre S R (p) is smooth over (p) at the prime corresponding to q.
Then R S is smooth at q.
Proof. By Lemmas 134.15 and 135.5 we see that there exists a g S such
that Sg is a relative global complete intersection. Replacing S by Sg we may
assume S = R[x1 , . . . , xn ]/(f1 , . . . , fc ) is a relative global complete intersection.
For any subset I {1, . . . , n} of cardinality c consider the polynomial gI =
det(fj /xi )j=1,...,c,iI of Lemma 135.15. Note that the image g I of gI in the
polynomial ring (p)[x1 , . . . , xn ] is the determinant of the partial derivatives of the
images f j of the fj in the ring (p)[x1 , . . . , xn ]. Thus the lemma follows by applying
Lemma 135.15 both to R S and to (p) S R (p).

Note that the sets U, V in the following lemma are open by definition.
00TG

Lemma 135.17. Let R S be a ring map of finite presentation. Let R R0

be a flat ring map. Denote S 0 = R0 R S the base change. Let U Spec(S) be
the set of primes at which R S is smooth. Let V Spec(S 0 ) the set of primes
at which R0 S 0 is smooth. Then V is the inverse image of U under the map
f : Spec(S 0 ) Spec(S).
Proof. By Lemma 132.8 we see that NLS/R S S 0 is homotopy equivalent to NLS 0 /R0 .
This already implies that f 1 (U ) V .
Let q0 S 0 be a prime lying over q S. Assume q0 V . We have to show that
q U . Since S S 0 is flat, we see that Sq Sq0 0 is faithfully flat (Lemma 38.17).
Thus the vanishing of H1 (LS 0 /R0 )q0 implies the vanishing of H1 (LS/R )q . By Lemma
77.5 applied to the Sq -module (S/R )q and the map Sq Sq0 0 we see that (S/R )q
is projective. Hence R S is smooth at q by Lemma 135.12.


02UQ

Lemma 135.18. Let k K be a field extension. Let S be a finite type algebra

over k. Let qK be a prime of SK = K k S and let q be the corresponding prime of
S. Then S is smooth over k at q if and only if SK is smooth at qK over K.
Proof. This is a special case of Lemma 135.17.

04B1

Lemma 135.19. Let R be a ring and let I R be an ideal. Let R/I S be a

smooth ring map. Then there exists elements g i S which generate the unit ideal
of S such that each S gi
= Si /ISi for some (standard) smooth ring Si over R.
Proof. By Lemma 135.10 we find a collection of elements g i S which generate the
unit ideal of S such that each S gi is standard smooth over R/I. Hence we may assume that S is standard smooth over R/I. Write S = (R/I)[x1 , . . . , xn ]/(f 1 , . . . , f c )
as in Definition 135.6. Choose f1 , . . . , fc R[x1 , . . . , xn ] lifting f 1 , . . . , f c . Set
f
S = R[x1 , . . . , xn , xn+1 ]/(f1 , . . . , fc , xn+1 1) where = det( xji )i,j=1,...,c as in
Example 135.8. This proves the lemma.

136. Formally smooth maps

00TH

In this section we define formally smooth ring maps. It will turn out that a ring map
of finite presentation is formally smooth if and only if it is smooth, see Proposition
136.13.

COMMUTATIVE ALGEBRA

00TI

337

Definition 136.1. Let R S be a ring map. We say S is formally smooth over

R if for every commutative solid diagram
/ A/I
O

SO

/A

where I A is an ideal of square zero, a dotted arrow exists which makes the
diagram commute.
00TJ

Lemma 136.2. Let R S be a formally smooth ring map. Let R R0 be any

ring map. Then the base change S 0 = R0 R S is formally smooth over R0 .
Proof. Let a solid diagram
SO

/ R 0 R S
O

/ R0

/ A/I
O
$/)

as in Definition 136.1 be given. By assumption the longer dotted arrow exists. By

the universal property of tensor product we obtain the shorter dotted arrow.

031H

Lemma 136.3. A composition of formally smooth ring maps is formally smooth.

Proof. Omitted. (Hint: This is completely formal, and follows from considering a
suitable diagram.)


00TK

Lemma 136.4. A polynomial ring over R is formally smooth over R.

Proof. Suppose we have a diagram as in Definition 136.1 with S = R[xj ; j J].
Then there exists a dotted arrow simply by choosing lifts aj A of the elements in
A/I to which the elements xj map to under the top horizontal arrow.


00TL

Lemma 136.5. Let R S be a ring map. Let P S be a surjective R-algebra

map from a polynomial ring P onto S. Denote J P the kernel. Then R S is
formally smooth if and only if there exists an R-algebra map : S P/J 2 which
is a right inverse to the surjection P/J 2 S.
Proof. Assume R S is formally smooth. Consider the commutative diagram
SO

/ P/J
O

!
/ P/J 2

By assumption the dotted arrow exists. This proves that exists.

Conversely, suppose we have a as in the lemma. Let a solid diagram
SO
R

/ A/I
O
!

/A

COMMUTATIVE ALGEBRA

338

as in Definition 136.1 be given. Because P is formally smooth by Lemma 136.4,

there exists an R-algebra homomorphism : P A which lifts the map P
S A/I. Clearly (J) I and since I 2 = 0 we conclude that (J 2 ) = 0.
Hence factors as : P/J 2 A. The desired dotted arrow is the composition
: S A.

00TM
031I

Remark 136.6.
smooth over R.

Lemma 136.5 holds more generally whenever P is formally

Lemma 136.7. Let R S be a ring map. Let P S be a surjective R-algebra

map from a polynomial ring P onto S. Denote J P the kernel. Then R S is
formally smooth if and only if the sequence
0 J/J 2 P/R R S S/R 0
of Lemma 130.9 is a split exact sequence.
Proof. Assume S is formally smooth over R. By Lemma 136.5 this means there
exists an R-algebra map S P/J 2 which is a left inverse to the canonical map
P/J 2 S. This means that
P/J 2
= S J/J 2
as R-algebras. Note that the middle term of the exact sequence is P/R P S
=
(P/J 2 )/R R S by Lemma 130.11. A direct computation shows that
(SJ/J 2 )/R (SJ/J 2 ) S = S/R J/J 2
as desired.
Assume the exact sequence of the lemma is split exact. Choose a splitting :
S/R P/R R S. For each S choose x P which maps to . Next, for
each S choose f J such that
df = dx (d)
in the middle term of the exact sequence. We claim that s : 7 x f mod J 2
is an R-algebra homomorphism s : S P/J 2 . To prove this we will repeatedly
use that if h J and dh = 0 in P/R R S, then h J 2 . Let , S. Then
(d + d d( + )) = 0. This implies
d(x + x x+ f f + f+ ) = 0
which means that x + x x+ f f + f+ J 2 , which in turn means that
s() + s() = s( + ). Similarly, we have (d + d d) = 0 which implies
that
(dx df ) + (dx df ) dx df = 0
in the middle term of the exact sequence. Moreover we have
d(x x ) = x dx + x dx = dx + dx
in the middle term again. Combined these equations mean that x x x f
f + f J 2 which means that s()s() = s(). If R, then d = 0 and we
see that df = dx , hence x + f J 2 and hence s() = as desired. At this
point we can apply Lemma 136.5 to conclude that S/R is formally smooth.


031J

Proposition 136.8. Let R S be a ring map. Consider a formally smooth

R-algebra P and a surjection P S with kernel J. The following are equivalent

COMMUTATIVE ALGEBRA

339

(1)
(2)
(3)
(4)

S is formally smooth over R,

for some P S as above there exists a section to P/J 2 S,
for all P S as above there exists a section to P/J 2 S,
for some P S as above the sequence 0 J/J 2 P/R S S/R 0
is split exact,
(5) for all P S as above the sequence 0 J/J 2 P/R S S/R 0
is split exact, and
(6) the naive cotangent complex NLS/R is quasi-isomorphic to a projective Smodule placed in degree 0.

Proof. It is clear that (1) implies (3) implies (2), see first part of the proof of
Lemma 136.5. It is also true that (3) implies (5) implies (4) and that (2) implies
(4), see first part of the proof of Lemma 136.7. Finally, Lemma 136.7 applied to
the canonical surjection R[S] S (132.0.1) shows that (1) implies (6).
Assume (4) and lets prove (6). Consider the sequence of Lemma 132.4 associated
to the ring maps R P S. By the implication (1) (6) proved above we
see that NLP/R R S is quasi-isomorphic to P/R P S placed in degree 0. Hence
H1 (NLP/R P S) = 0. Since P S is surjective we see that NLS/P is homotopy
equivalent to J/J 2 placed in degree 1 (Lemma 132.6). Thus we obtain the exact
sequence 0 H1 (LS/R ) J/J 2 P/R P S S/R 0. By assumption we
see that H1 (LS/R ) = 0 and that S/R is a projective S-module. Thus (6) follows.
Finally, lets prove that (6) implies (1). The assumption means that the complex
J/J 2 P/R S where P = R[S] and P S is the canonical surjection (132.0.1).
Hence Lemma 136.7 shows that S is formally smooth over R.

031K

Lemma 136.9. Let A B C be ring maps. Assume B C is formally

smooth. Then the sequence
0 B/A B C C/A C/B 0
of Lemma 130.7 is a split short exact sequence.
Proof. Follows from Proposition 136.8 and Lemma 132.4.

06A6

Lemma 136.10. Let A B C be ring maps with A C formally smooth and

B C surjective with kernel J B. Then the exact sequence
0 J/J 2 B/A B C C/A 0
of Lemma 130.9 is split exact.
Proof. Follows from Proposition 136.8, Lemma 132.4, and Lemma 130.9.

06A7

Lemma 136.11. Let A B C be ring maps. Assume A C is surjective

(so also B C is) and A B formally smooth. Denote I = Ker(A C) and
J = Ker(B C). Then the sequence
0 I/I 2 J/J 2 B/A B B/J 0
of Lemma 132.7 is split exact.
Proof. Since A B is formally smooth there exists a ring map : B A/I 2
whose composition with A B equals the quotient map A A/I 2 . Then
induces a map J/J 2 I/I 2 which is inverse to the map I/I 2 J/J 2 .


031L

Lemma 136.12. Let R S be a ring map. Let I R be an ideal. Assume

COMMUTATIVE ALGEBRA

340

(1) I 2 = 0,
(2) R S is flat, and
(3) R/I S/IS is formally smooth.
Then R S is formally smooth.
Proof. Assume (1), (2) and (3). Let P = R[{xt }tT ] S be a surjection of Ralgebras with kernel J. Thus 0 J P S 0 is a short exact sequence of
flat R-modules. This implies that I R S = IS, I R P = IP and I R J = IJ as
well as J IP = IJ. We will use throughout the proof that
(S/IS)/(R/I) = S/R S (S/IS) = S/R R R/I = S/R /IS/R
and similarly for P (see Lemma 130.12). By Lemma 136.7 the sequence
031M

0 J/(IJ + J 2 ) P/R P S/IS S/R S S/IS 0

L
is split exact. Of course the middle term is
tT S/ISdxt . Choose a splitting
2
: P/R P S/IS J/(IJ + J ). For each t T choose an element ft J which
maps to (dxt ) in J/(IJ + J 2 ). This determines a unique S-module map
M

: P/R R S =
Sdxt J/J 2
(136.12.1)

with the property that

(dxt ) = ft . As is a section to d the difference
= idJ/J 2
d
is a self map J/J 2 J/J 2 whose image is contained in (IJ + J 2 )/J 2 . In particular
((IJ + J 2 )/J 2 ) = 0 because I 2 = 0. This means that factors as

J/J 2 J/(IJ + J 2 ) (IJ + J 2 )/J 2 J/J 2

where is a S/IS-module map. Using again that the sequence (136.12.1) is split,
we can find a S/IS-module map : P/R P S/IS (IJ + J 2 )/J 2 such that d
is equal to . In the same manner as above the map determines an S-module
map : P/R P S J/J 2 . After replacing
by
+ a simple computation
shows that = 0. In other words
is a section of J/J 2 P/R P S. By Lemma
136.7 we conclude that R S is formally smooth.

00TN

Proposition 136.13. Let R S be a ring map. The following are equivalent

(1) R S is of finite presentation and formally smooth,
(2) R S is smooth.
Proof. Follows from Proposition 136.8 and Definition 135.1. (Note that S/R
is a finitely presented S-module if R S is of finite presentation, see Lemma
130.15.)


00TP

Lemma 136.14. Let R S be a smooth ring map. Then there exists a subring
R0 R of finite type over Z and a smooth ring map R0 S0 such that S
=
R R0 S0 .
Proof. We are going to use that smooth is equivalent to finite presentation and
formally smooth, see Proposition 136.13. Write S = R[x1 , . . . , xn ]/(f1 , . . . , fm ) and
denote I = (f1 , . . . , fm ). Choose a right inverse : S R[x1 , . . . , xn ]/I 2 to the
projection to S as in Lemma 136.5. Choose hi R[x1 , . . . , xn ] such that (xi mod

COMMUTATIVE ALGEBRA

341

I) = hi mod I 2 . The fact that is an R-algebra homomorphism R[x1 , . . . , xn ]/I

R[x1 , . . . , xn ]/I 2 is equivalent to the condition that
X
aj1 j2 fj1 fj2
fj (h1 , . . . , hn ) =
j1 j2

for certain akl R[x1 , . . . , xn ]. Let R0 R be the subring generated over Z by all
the coefficients of the polynomials fj , hi , akl . Set S0 = R0 [x1 , . . . , xn ]/(f1 , . . . , fm ),
with I0 = (f1 , . . . , fm ). Let 0 : S0 R0 [x1 , . . . , xn ]/I02 defined by the rule
xi 7 hi mod I02 ; this works since the alk are defined over R0 and satisfy the same
relations. Thus by Lemma 136.5 the ring S0 is formally smooth over R0 .

06CM

Lemma 136.15. Let R S be a ring map. Let R R0 be a faithfully flat ring

map. Set S 0 = S R R0 . Then R S is formally smooth if and only if R0 S 0 is
formally smooth.
Proof. If R S is formally smooth, then R0 S 0 is formally smooth by Lemma
136.2. To prove the converse, assume R0 S 0 is formally smooth. Note that
N R R0 = N S S 0 for any S-module N . In particular S S 0 is faithfully flat
also. Choose a polynomial ring P = R[{xi }iI ] and a surjection of R-algebras
P S with kernel J. Note that P 0 = P R R0 is a polynomial algebra over R0 .
Since R R0 is flat the kernel J 0 of the surjection P 0 S 0 is J R R0 . Hence the
split exact sequence (see Lemma 136.7)
0 J 0 /(J 0 )2 P 0 /R0 P 0 S 0 S 0 /R0 0
is the base change via S S 0 of the corresponding sequence
J/J 2 P/R P S S/R 0
see Lemma 130.9. As S S 0 is faithfully flat we conclude two things: (1) this
sequence (without 0 ) is exact too, and (2) S/R is a projective S-module. Namely,
S 0 /R0 is projective as a direct sum of the free module P 0 /R0 P 0 S 0 and S/R S
S 0 = S 0 /R0 by what we said above. Thus (2) follows by descent of projectivity
through faithfully flat ring maps, see Theorem 94.5. Hence the sequence 0
J/J 2 P/R P S S/R 0 is exact also and we win by applying Lemma
136.7 once more.

It turns out that smooth ring maps satisfy the following strong lifting property.

07K4

Lemma 136.16. Let R S be a smooth ring map. Given a commutative solid

diagram
/ A/I
SO
O
!
/A
R
where I A is a locally nilpotent ideal, a dotted arrow exists which makes the
diagram commute.
Proof. By Lemma 136.14 we can extend the diagram to a commutative diagram
SO 0

/S
O

/ A/I
O

R0

/R

/A

COMMUTATIVE ALGEBRA

342

with R0 S0 smooth, R0 of finite type over Z, and S = S0 R0 R. Let x1 , . . . , xn

S0 be generators of S0 over R0 . Let a1 , . . . , an be elements of A which map to the
same elements in A/I as the elements x1 , . . . , xn . Denote A0 A the subring
generated by the image of R0 and the elements a1 , . . . , an . Set I0 = A0 I. Then
A0 /I0 A/I and S0 A/I maps into A0 /I0 . Thus it suffices to find the dotted
arrow in the diagram
/ A0 /I0
SO 0
O
R0

"

/ A0

The ring A0 is of finite type over Z by construction. Hence A0 is Noetherian,

whence I0 is nilpotent, see Lemma 31.4. Say I0n = 0. By Proposition 136.13 we can
successively lift the R0 -algebra map S0 A0 /I0 to S0 A0 /I02 , S0 A0 /I03 , . . .,
and finally S0 A0 /I0n = A0 .

137. Smoothness and differentials
05D4

Some results on differentials and smooth ring maps.

04B2

Lemma 137.1. Given ring maps A B C with B C smooth, then the

sequence
0 C B B/A C/A C/B 0
of Lemma 130.7 is exact.
Proof. This follows from the more general Lemma 136.9 because a smooth ring
map is formally smooth, see Proposition 136.13. But it also follows directly from
Lemma 132.4 since H1 (LC/B ) = 0 is part of the definition of smoothness of B
C.


06A8

Lemma 137.2. Let A B C be ring maps with A C smooth and B C

surjective with kernel J B. Then the exact sequence
0 J/J 2 B/A B C C/A 0
of Lemma 130.9 is split exact.
Proof. This follows from the more general Lemma 136.10 because a smooth ring
map is formally smooth, see Proposition 136.13.


06A9

Lemma 137.3. Let A B C be ring maps. Assume A C is surjective (so

also B C is) and A B smooth. Denote I = Ker(A C) and J = Ker(B
C). Then the sequence
0 I/I 2 J/J 2 B/A B B/J 0
of Lemma 132.7 is exact.
Proof. This follows from the more general Lemma 136.11 because a smooth ring
map is formally smooth, see Proposition 136.13.


05D5

Lemma 137.4. Let : R S be a smooth ring map. Let : S R be a left

inverse to . Set I = Ker(). Then
(1) I/I 2 is a finite locally free R-module, and

COMMUTATIVE ALGEBRA

343

(2) if I/I 2 is free, then S

= R[[t1 , . . . , td ]] as R-algebras, where S is the
I-adic completion of S.
Proof. By Lemma 130.10 applied to R S R we see that I/I 2 = S/R S, R.
Since by definition of a smooth morphism the module S/R is finite locally free
over S we deduce that (1) holds. If I/I 2 is free, then choose f1 , . . . , fd I whose
images in I/I 2 form an R-basis. Consider the R-algebra map defined by
: R[[x1 , . . . , xd ]] S ,

xi 7 fi .

Denote P = R[[x1 , . . . , xd ]] and J = (x1 , . . . , xd ) P . We write n : P/J n S/I n

for the induced map of quotient rings. Note that S/I 2 = (R) I/I 2 . Thus 2
is an isomorphism. Denote 2 : S/I 2 P/J 2 the inverse of 2 . We will prove by
induction on n that for all n > 2 there exists an inverse n : S/I n P/J n of n .
Namely, as S is formally smooth over R (by Proposition 136.13) we see that in the
solid diagram
/ P/J n
S
n1

" 
P/J n1

of R-algebras we can fill in the dotted arrow by some R-algebra map : S P/J n
making the diagram commute. This induces an R-algebra map : S/I n P/J n
which is equal to n1 modulo J n . By construction the map n is surjective and
now n is an R-algebra endomorphism of P/J n which maps xi to xi + i,n with
i,n J n1 /J n . It follows that n is an isomorphism and hence it has an inverse
n . This proves the lemma.

138. Smooth algebras over fields
00TQ

Warning: The following two lemmas do not hold over nonperfect fields in general.

00TR

Lemma 138.1. Let k be an algebraically closed field. Let S be a finite type kalgebra. Let m S be a maximal ideal. Then
dim(m) S/k S (m) = dim(m) m/m2 .
Proof. Consider the exact sequence
m/m2 S/k S (m) (m)/k 0
of Lemma 130.9. We would like to show that the first map is an isomorphism. Since
k is algebraically closed the composition k (m) is an isomorphism by Theorem
33.1. So the surjection S (m) splits as a map of k-algebras, and Lemma 130.10
shows that the sequence above is exact on the left. Since (m)/k = 0, we win. 

00TS

Lemma 138.2. Let k be an algebraically closed field. Let S be a finite type kalgebra. Let m S be a maximal ideal. The following are equivalent:
(1) The ring Sm is a regular local ring.
(2) We have dim(m) S/k S (m) dim(Sm ).
(3) We have dim(m) S/k S (m) = dim(Sm ).
(4) There exists a g S, g 6 m such that Sg is smooth over k. In other words
S/k is smooth at m.

COMMUTATIVE ALGEBRA

344

Proof. Note that (1), (2) and (3) are equivalent by Lemma 138.1 and Definition
109.7.
Assume that S is smooth at m. By Lemma 135.10 we see that Sg is standard smooth
over k for a suitable g S, g 6 m. Hence by Lemma 135.7 we see that Sg /k is
free of rank dim(Sg ). Hence by Lemma 138.1 we see that dim(Sm ) = dim(m/m2 )
in other words Sm is regular.
Conversely, suppose that Sm is regular. Let d = dim(Sm ) = dim m/m2 . Choose a
presentation S = k[x1 , . . . , xn ]/I such that xi maps to an element of m for all i. In
other words, m00 = (x1 , . . . , xn ) is the corresponding maximal ideal of k[x1 , . . . , xn ].
Note that we have a short exact sequence
I/m00 I m00 /(m00 )2 m/(m)2 0
Pick c = n d elements f1 , . . . , fd I such that their images in m00 /(m00 )2 span
the kernel of the map to m/(m)2 . This is clearly possible. Denote J = (f1 , . . . , fc ).
So J I. Denote S 0 = k[x1 , . . . , xn ]/J so there is a surjection S 0 S. Denote
m0 = m00 S 0 the corresponding maximal ideal of S 0 . Hence we have
k[x1 , . . . , xn ]
O

/ S0
O

/S
O

m00

/ m0

/m

By our choice of J the exact sequence

J/m00 J m00 /(m00 )2 m0 /(m0 )2 0
0
shows that dim(m0 /(m0 )2 ) = d. Since Sm
0 surjects onto Sm we see that dim(Sm0 )
0
d. Hence by the discussion preceding Definition 59.9 we conclude that Sm
0 is regular
0
of dimension d as well. Because S was cut out by c = n d equations we conclude
that there exists a g 0 S 0 , g 0 6 m0 such that Sg0 0 is a global complete intersection
0
over k, see Lemma 133.4. Also the map Sm
0 Sm is a surjection of Noetherian
local domains of the same dimension and hence an isomorphism. By Lemma 125.7
we see that Sg0 0
= Sg for some g S, g 6 m and g 0 S 0 , g 0 6 m0 . All in all we
conclude that after replacing S by a principal localization we may assume that S
is a global complete intersection.

At this point we may write S = k[x1 , . . . , xn ]/(f1 , . . . , fc ) with dim S = n c.

Recall that the naive cotangent complex of this algebra is given by
M
M
S fj
S dxi
see Lemma 134.13. By Lemma 135.15 in order to show that S is smooth at m we
have to show that one of the c c minors gI of the matrix A giving the map
above does not vanish at m. By Lemma 138.1 the matrix A mod m has rank c.
Thus we win.

00TT

Lemma 138.3. Let k be any field. Let S be a finite type k-algebra. Let X =
Spec(S). Let q S be a prime corresponding to x X. The following are equivalent:
(1) The k-algebra S is smooth at q over k.
(2) We have dim(q) S/k S (q) dimx X.
(3) We have dim(q) S/k S (q) = dimx X.

COMMUTATIVE ALGEBRA

345

Moreover, in this case the local ring Sq is regular.

Proof. If S is smooth at q over k, then there exists a g S, g 6 q such that Sg is
standard smooth over k, see Lemma 135.10. A standard smooth algebra over k has
a module of differentials which is free of rank equal to the dimension, see Lemma
135.7 (use that a relative global complete intersection over a field has dimension
equal to the number of variables minus the number of equations). Thus we see that
(1) implies (3). To finish the proof of the lemma it suffices to show that (2) implies
(1) and that it implies that Sq is regular.
Assume (2). By Nakayamas Lemma 19.1 we see that S/k,q can be generated by
dimx X elements. We may replace S by Sg for some g S, g 6 q such that S/k
is generated by at most dimx X elements. Let K k be an algebraically closed
field extension such that there exists a k-algebra map : (q) K. Consider
SK = K k S. Let m SK be the maximal ideal corresponding to the surjection
/ K k (q) idK / K.

SK = K k S

Note that mS = q, in other words m lies over q. By Lemma 115.6 the dimension of
XK = Spec(SK ) at the point corresponding to m is dimx X. By Lemma 113.6 this
is equal to dim((SK )m ). By Lemma 130.12 the module of differentials of SK over K
is the base change of S/k , hence also generated by at most dimx X = dim((SK )m )
elements. By Lemma 138.2 we see that SK is smooth at m over K. By Lemma
135.17 this implies that S is smooth at q over k. This proves (1). Moreover, we
know by Lemma 138.2 that the local ring (SK )m is regular. Since Sq (SK )m is
flat we conclude from Lemma 109.9 that Sq is regular.

The following lemma can be significantly generalized (in several different ways).
00TU

Lemma 138.4. Let k be a field. Let R be a Noetherian local ring containing k.

Assume that the residue field = R/m is a finitely generated separable extension
of k. Then the map
d : m/m2 R/k R (m)
is injective.
Proof. We may replace R by R/m2 . Hence we may assume that m2 = 0. By
assumption we may write = k(x1 , . . . , xr , y) where x1 , . . . , xr is a transcendence
basis of over k and y is separable algebraic
P over k(x1 , . . . , xr ). Say its minimal
equation is P (y) = 0 with P (T ) = T d + i<d ai T i , with ai k(x1 , . . . , xr ) and
P 0 (y) 6= 0. Choose any lifts xi R of the elements xi . This gives a commutative
diagram
/
Re
O

k(x1 , . . . , xr )
of k-algebras. We want to extend the left upwards arrow to a k-algebra map from
to R. To do this choose any y R lifting y. To see that it defines a k-algebra
map defined on
= k(x1 , . . . , xr )[T ]/(P ) all we have to show is that we may choose
y such that P (y) = 0. If not then we compute for m that
P (y + ) = P (y) + P 0 (y)

COMMUTATIVE ALGEBRA

346

because m2 = 0. Since P 0 (y) = P 0 (y) we see that we can adjust our choice as
desired. This shows that R
= m as k-algebras! From a direct computation of
m/k the lemma follows.

00TV

Lemma 138.5. Let k be a field. Let S be a finite type k-algebra. Let q S be a

prime. Assume (q) is separable over k. The following are equivalent:
(1) The algebra S is smooth at q over k.
(2) The ring Sq is regular.
Proof. Denote R = Sq and denote its maximal by m and its residue field . By
Lemma 138.4 and 130.9 we see that there is a short exact sequence
0 m/m2 R/k R /k 0
Note that R/k = S/k,q , see Lemma 130.8. Moreover, since is separable over k
we have dim /k = trdegk (). Hence we get
dim R/k R = dim m/m2 + trdegk () dim R + trdegk () = dimq S
(see Lemma 115.3 for the last equality) with equality if and only if R is regular.
Thus we win by applying Lemma 138.3.


00TW

Lemma 138.6. Let R S be a Q-algebra map. Let f S be such that S/R =

Sdf C for some S-submodule C. Then
(1) f is not nilpotent, and
(2) if S is a Noetherian local ring, then f is a nonzerodivisor in S.
Proof. For a S write d(a) = (a)df + c(a) for some (a) S and c(a) C.
Consider the R-derivation S S, a 7 (a). Note that (f ) = 1.
If f n = 0 with n > 1 minimal, then 0 = (f n ) = nf n1 contradicting the minimality of n. We conclude that f is not nilpotent.
Suppose f a = 0. If f is a unit then a = 0 and we win. Assume f is not a unit. Then
0 = (f a) = f (a) + a by the Leibniz rule and hence a (f ). By induction suppose
we have shown f a = 0 a (f n ). Then writing a = f n b we get 0 = (f n+1 b) =
(n + 1)f n b + f n+1 (b). Hence a =Tf n b = f n+1 (b)/(n + 1) (f n+1 ). Since in the
Noetherian local ring S we have (f n ) = 0, see Lemma 50.4 we win.

The following is probably quite useless in applications.

00TX

Lemma 138.7. Let k be a field of characteristic 0. Let S be a finite type k-algebra.

Let q S be a prime. The following are equivalent:
(1) The algebra S is smooth at q over k.
(2) The Sq -module S/k,q is (finite) free.
(3) The ring Sq is regular.
Proof. In characteristic zero any field extension is separable and hence the equivalence of (1) and (3) follows from Lemma 138.5. Also (1) implies (2) by definition
of smooth algebras. Assume that S/k,q is free over Sq . We are going to use the
notation and observations made in the proof of Lemma 138.5. So R = Sq with
maximal ideal m and residue field . Our goal is to prove R is regular.
If m/m2 = 0, then m = 0 and R
= . Hence R is regular and we win.
If m/m2 6= 0, then choose any f m whose image in m/m2 is not zero. By
Lemma 138.4 we see that df has nonzero image in R/k /mR/k . By assumption

COMMUTATIVE ALGEBRA

347

R/k = S/k,q is finite free and hence by Nakayamas Lemma 19.1 we see that
df generates a direct summand. We apply Lemma 138.6 to deduce that f is a
nonzerodivisor in R. Furthermore, by Lemma 130.9 we get an exact sequence
(f )/(f 2 ) R/k R R/f R (R/f R)/k 0
This implies that (R/f R)/k is finite free as well. Hence by induction we see that
R/f R is a regular local ring. Since f m was a nonzerodivisor we conclude that
R is regular, see Lemma 105.7.

00TY

Example 138.8. Lemma 138.7 does not hold in characteristic p > 0. The standard
examples are the ring maps
Fp Fp [x]/(xp )
whose module of differentials is free but is clearly not smooth, and the ring map
(p > 2)
Fp (t) Fp (t)[x, y]/(xp + y 2 + )
which is not smooth at the prime q = (y, xp ) but is regular.
Using the material above we can characterize smoothness at the generic point in
terms of field extensions.

07ND

Lemma 138.9. Let R S be an injective finite type ring map with R and S
domains. Then R S is smooth at q = (0) if and only if f.f.(R) f.f.(S) is a
separable extension of fields.
Proof. Assume R S is smooth at (0). We may replace S by Sg for some nonzero
g S and assume that R S is smooth. Set K = f.f.(R). Then K S R K
is smooth (Lemma 135.4). Moreover, for any field extension K K 0 the ring map
K 0 S R K 0 is smooth as well. Hence S R K 0 is a regular ring by Lemma 138.3,
in particular reduced. It follows that S R K is a geometrically reduced over K.
Hence f.f.(S) is geometrically reduced over K, see Lemma 42.3. Hence f.f.(S)/K
is separable by Lemma 43.1.
Conversely, assume that f.f.(R) f.f.(S) is separable. We may assume R S is
of finite presentation, see Lemma 29.1. It suffices to prove that K S R K is
smooth at (0), see Lemma 135.17. This follows from Lemma 138.5, the fact that a
field is a regular ring, and the assumption that f.f.(R) f.f.(S) is separable. 
139. Smooth ring maps in the Noetherian case

02HR
02HS

Definition 139.1. Let : B 0 B be a ring map. We say is a small extension

if B 0 and B are local Artinian rings, is surjective and I = Ker() has length 1 as
a B 0 -module.
Clearly this means that I 2 = 0 and that I = (x) for some x B 0 such that m0 x = 0
where m0 B 0 is the maximal ideal.

02HT

Lemma 139.2. Let R S be a ring map. Let q be a prime ideal of S lying

over p R. Assume R is Noetherian and R S of finite type. The following are
equivalent:
(1) R S is smooth at q,

COMMUTATIVE ALGEBRA

348

(2) for every surjection of local R-algebras (B 0 , m0 ) (B, m) with Ker(B 0 B)

having square zero and every solid commutative diagram
SO

/B
O

/ B0

such that q = S m there exists a dotted arrow making the diagram commute,
(3) same as in (2) but with B 0 B ranging over small extensions, and
(4) same as in (2) but with B 0 B ranging over small extensions such that in
addition S B induces an isomorphism (q)
= (m).
Proof. Assume (1). This means there exists a g S, g 6 q such that R Sg is
smooth. By Proposition 136.13 we know that R Sg is formally smooth. Note
that given any diagram as in (2) the map S B factors automatically through Sq
and a fortiori through Sg . The formal smoothness of Sg over R gives us a morphism
Sg B 0 fitting into a similar diagram with Sg at the upper left corner. Composing
with S Sg gives the desired arrow. In other words, we have shown that (1)
implies (2).
Clearly (2) implies (3) and (3) implies (4).
Assume (4). We are going to show that (1) holds, thereby finishing the proof of
the lemma. Choose a presentation S = R[x1 , . . . , xn ]/(f1 , . . . , fm ). This is possible
as S is of finite type over R and therefore of finite presentation (see Lemma 30.4).
Set I = (f1 , . . . , fm ). Consider the naive cotangent complex
Mm
d : I/I 2
Sdxj
j=1

of this presentation (see Section 132). It suffices to show that when we localize this
complex at q then the map becomes a split injection, see Lemma 135.12. Denote
S 0 = R[x1 , . . . , xn ]/I 2 . By Lemma 130.11 we have
Mm
S S 0 S 0 /R = S R[x1 ,...,xn ] R[x1 ,...,xn ]/R =
Sdxj .
j=1

Thus the map

d : I/I 2 S S 0 S 0 /R
is the same as the map in the naive cotangent complex above. In particular the truth
of the assertion we are trying to prove depends only on the three rings R S 0 S.
Let q0 R[x1 , . . . , xn ] be the prime ideal corresponding to q. Since localization
commutes with taking modules of differentials (Lemma 130.8) we see that it suffices
to show that the map
02HU

d : Iq0 /Iq20 Sq Sq0 0 S 0 0 /R

(139.2.1)
coming from R

Sq0 0

Sq is a split injection.

Let N N be an integer. Consider the ring

0
BN
= Sq0 0 /(q0 )N Sq0 0 = (S 0 /(q0 )N S 0 )q0
0
0
0
and its quotient BN = BN
/IBN
. Note that BN
= Sq /qN Sq . Observe that BN
is an Artinian local ring since it is the quotient of a local Noetherian ring by a

COMMUTATIVE ALGEBRA

349

0
power of its maximal ideal. Consider a filtration of the kernel IN of BN
BN by
0
BN -submodules
0 JN,1 JN,2 . . . JN,n(N ) = IN
0
such that each successive quotient JN,i /JN,i1 has length 1. (As BN
is Artinian
such a filtration exists.) This gives a sequence of small extensions
0
0
0
0
0
BN
BN
/JN,1 BN
/JN,2 . . . BN
/JN,n(N ) = BN
/IN = BN = Sq /qN Sq

Applying condition (4) successively to these small extensions starting with the map
S BN we see there exists a commutative diagram
SO

/ BN
O

/ B0
N

0
0
0
Clearly the ring map S BN
factors as S Sq BN
where Sq BN
is a local
0
to the
homomorphism of local rings. Moreover, since the maximal ideal of BN
0
factors through Sq /(q)N Sq = BN .
N th power is zero we conclude that Sq BN
In other words we have shown that for all N N the surjection of R-algebras
0
BN
BN has a splitting.

Consider the presentation

0 B 0 /R B /R 0
IN BN BN
N
N

0
coming from the surjection BN
BN with kernel IN (see Lemma 130.9). By the
0
BN has a right inverse. Hence by Lemma 130.10
above the R-algebra map BN
we see that the sequence above is split exact! Thus for every N the map
0 B 0 /R
IN BN BN
N

is a split injection. The rest of the proof is gotten by unwinding what this means
exactly. Note that
IN = Iq0 /(Iq20 + (q0 )N Iq0 )
By Artin-Rees (Lemma 50.2) we find a c 0 such that
Sq /qN c Sq Sq IN = Sq /qN c Sq Sq Iq0 /Iq20
for all N c (these tensor product are just a fancy way of dividing by qN c ). We
may of course assume c 1. By Lemma 130.11 we see that
0
0 N c 0
0 /R = S 0 /(q )
Sq0 0 /(q0 )N c Sq0 0 Sq0 0 BN
Sq0 Sq0 0 S 0 0 /R
q
q

we can further tensor this by BN = Sq /q

N c

Sq /q

to see that

N c
0 /R = Sq /q
Sq Sq0 0 BN
Sq Sq0 0 S 0 0 /R .
q

Since a split injection remains a split injection after tensoring with anything we see
that
0 B 0 /R )
Sq /qN c Sq Sq (139.2.1) = Sq /qN c Sq Sq (IN BN BN
N

is a split injection for all N c. By Lemma 73.1 we see that (139.2.1) is a split
injection. This finishes the proof.


COMMUTATIVE ALGEBRA

350

140. Overview of results on smooth ring maps

00TZ

Here is a list of results on smooth ring maps that we proved in the preceding
sections. For more precise statements and definitions please consult the references
given.
(1) A ring map R S is smooth if it is of finite presentation and the naive
cotangent complex of S/R is quasi-isomorphic to a finite projective Smodule in degree 0, see Definition 135.1.
(2) If S is smooth over R, then S/R is a finite projective S-module, see discussion following Definition 135.1.
(3) The property of being smooth is local on S, see Lemma 135.13.
(4) The property of being smooth is stable under base change, see Lemma
135.4.
(5) The property of being smooth is stable under composition, see Lemma
135.14.
(6) A smooth ring map is syntomic, in particular flat, see Lemma 135.10.
(7) A finitely presented, flat ring map with smooth fibre rings is smooth, see
Lemma 135.16.
(8) A finitely presented ring map R S is smooth if and only if it is formally
smooth, see Proposition 136.13.
(9) If R S is a finite type ring map with R Noetherian then to check that
R S is smooth it suffices to check the lifting property of formal smoothness along small extensions of Artinian local rings, see Lemma 139.2.
(10) A smooth ring map R S is the base change of a smooth ring map
R0 S0 with R0 of finite type over Z, see Lemma 136.14.
(11) Formation of the set of points where a ring map is smooth commutes with
flat base change, see Lemma 135.17.
(12) If S is of finite type over an algebraically closed field k, and m S a
maximal ideal, then the following are equivalent
(a) S is smooth over k in a neighbourhood of m,
(b) Sm is a regular local ring,
(c) dim(Sm ) = dim(m) S/k S (m).
see Lemma 138.2.
(13) If S is of finite type over a field k, and q S a prime ideal, then the
following are equivalent
(a) S is smooth over k in a neighbourhood of q,
(b) dimq (S/k) = dim(q) S/k S (q).
see Lemma 138.3.
(14) If S is smooth over a field, then all its local rings are regular, see Lemma
138.3.
(15) If S is of finite type over a field k, q S a prime ideal, the field extension
k (q) is separable and Sq is regular, then S is smooth over k at q, see
Lemma 138.5.
(16) If S is of finite type over a field k, if k has characteristic 0, if q S a prime
ideal, and if S/k,q is free, then S is smooth over k at q, see Lemma 138.7.
Some of these results were proved using the notion of a standard smooth ring
map, see Definition 135.6. This is the analogue of what a relative global complete

COMMUTATIVE ALGEBRA

351

intersection map is for the case of syntomic morphisms. It is also the easiest way
to make examples.

141. Etale
ring maps
00U0

An etale ring map is a smooth ring map whose relative dimension is equal to zero.
This is the same as the following slightly more direct definition.

00U1

Definition 141.1. Let R S be a ring map. We say R S is etale if it is of

finite presentation and the naive cotangent complex NLS/R is quasi-isomorphic to
zero. Given a prime q of S we say that R S is etale at q if there exists a g S,
g 6 q such that R Sg is etale.
In particular we see that S/R = 0 if S is etale over R. If R S is smooth, then
R S is etale if and only if S/R = 0. From our results on smooth ring maps
we automatically get a whole host of results for etale maps. We summarize these
in Lemma 141.3 below. But before we do so we prove that any etale ring map is
standard smooth.

00U9

Lemma 141.2. Any etale ring map is standard smooth. More precisely, if R S
is etale, then there exists a presentation S = R[x1 , . . . , xn ]/(f1 , . . . , fn ) such that
the image of det(fj /xi ) is invertible in S.
Proof. Let R S be etale. Choose a presentation S = R[x1 , . . . , xn ]/I. As
R S is etale we know that
M
d : I/I 2
Sdxi
i=1,...,n

is an isomorphism, in particular I/I is a free S-module. Thus by Lemma 134.6 we

may assume (after possibly changing the presentation), that I = (f1 , . . . , fc ) such
that the classes fi mod I 2 form a basis of I/I 2 . It follows immediately from the fact
that the displayed map above is an isomorphism that c = n and that det(fj /xi )
is invertible in S.

00U2

Lemma 141.3. Results on etale ring maps.

(1) The ring map R Rf is etale for any ring R and any f R.
(2) Compositions of etale ring maps are etale.
(3) A base change of an etale ring map is etale.
(4) The property of being etale is local: Given a ring map R S and elements
g1 , . . . , gm S which generate the unit ideal such that R Sgj is etale for
j = 1, . . . , m then R S is etale.
(5) Given R S of finite presentation, and a flat ring map R R0 , set
S 0 = R0 R S. The set of primes where R0 S 0 is etale is the inverse
image via Spec(S 0 ) Spec(S) of the set of primes where R S is etale.
(6) An etale ring map is syntomic, in particular flat.
(7) If S is finite type over a field k, then S is etale over k if and only if
S/k = 0.
(8) Any etale ring map R S is the base change of an etale ring map R0 S0
with R0 of finite type over Z.
(9) Let A = colim Ai be a filtered colimit of rings. Let A B be an etale ring
map. Then there exists an etale ring map Ai Bi for some i such that
B
= A Ai Bi .

COMMUTATIVE ALGEBRA

352

(10) Let A be a ring. Let S be a multiplicative subset of A. Let S 1 A B 0 be

etale. Then there exists an etale ring map A B such that B 0
= S 1 B.
Proof. In each case we use the corresponding result for smooth ring maps with a
small argument added to show that S/R is zero.
Proof of (1). The ring map R Rf is smooth and Rf /R = 0.
Proof of (2). The composition A C of smooth maps A B and B C is
smooth, see Lemma 135.14. By Lemma 130.7 we see that C/A is zero as both
C/B and B/A are zero.
Proof of (3). Let R S be etale and R R0 be arbitrary. Then R0 S 0 = R0 R S
is smooth, see Lemma 135.4. Since S 0 /R0 = S 0 S S/R by Lemma 130.12 we
conclude that S 0 /R0 = 0. Hence R0 S 0 is etale.
Proof of (4). Assume the hypotheses of (4). By Lemma 135.13 we see that R S
is smooth. We are also given that Sgi /R = (S/R )gi = 0 for all i. Then S/R = 0,
see Lemma 23.2.
Proof of (5). The result for smooth maps is Lemma 135.17. In the proof of that
lemma we used that NLS/R S S 0 is homotopy equivalent to NLS 0 /R0 . This reduces
us to showing that if M is a finitely presented S-module the set of primes q0 of S 0
such that (M S S 0 )q0 = 0 is the inverse image of the set of primes q of S such that
Mq = 0. This is true (proof omitted).
Proof of (6). Follows directly from the corresponding result for smooth ring maps
(Lemma 135.10).
Proof of (7). Follows from Lemma 138.3 and the definitions.
Proof of (8). Lemma 136.14 gives the result for smooth ring maps. The resulting
smooth ring map R0 S0 satisfies the hypotheses of Lemma 129.8, and hence we
may replace S0 by the factor of relative dimension 0 over R0 .
Proof of (9). Follows from (8) since R0 A will factor through Ai for some i.
Proof of (10). Follows from (9), (1), and (2) since S 1 A is a filtered colimit of
principal localizations of A.

Next we work out in more detail what it means to be etale over a field.
00U3

Lemma 141.4. Let k be a field. A ring map k S is etale if and only if S is

isomorphic as a k-algebra to a finite product of finite separable extensions of k.
Proof. If k k 0 is a finite separable field extension then we can write k 0 = k()
=
k[x]/(f ). Here f is the minimal polynomial of the element . Since k 0 is separable
over k we have gcd(f, f 0 ) = 1. This implies that d : k 0 f k 0 dx is an isomorphism.
Hence k k 0 is etale.
Conversely, suppose that k S is etale. Let k be an algebraic closure of k. Then
S k k is etale over k. Suppose we have the result over k. Then S k k is reduced
and hence S is reduced. Also, S Q
k k is finite over k and hence S is finite over k.
Hence S is a finite product S =
ki of fields, see Lemma 52.2 and Proposition
59.6. The result over k means S k k is isomorphic to a finite product of copies
of k, which implies that each k ki is finite separable, see for example Lemmas
43.1 and 43.3. Thus we have reduced to the case k = k. In this case Lemma 138.2
(combined with S/k = 0) we see that Sm
= k for all maximal ideals m S. This

COMMUTATIVE ALGEBRA

353

implies the result because S is the product of the localizations at its maximal ideals
by Lemma 52.2 and Proposition 59.6 again.

00U4

Lemma 141.5. Let R S be a ring map. Let q S be a prime lying over p in

R. If S/R is etale at q then
(1) we have pSq = qSq is the maximal ideal of the local ring Sq , and
(2) the field extension (p) (q) is finite separable.
Proof. First we may replace S by Sg for some g S, g 6 q and assume that
R S is etale. Then the lemma follows from Lemma 141.4 by unwinding the fact
that S R (p) is etale over (p).


00U5

Lemma 141.6. An etale ring map is quasi-finite.

Proof. Let R S be an etale ring map. By definition R S is of finite type.
For any prime p R the fibre ring S R (p) is etale over (p) and hence a finite
products of fields finite separable over (p), in particular finite over (p). Thus
R S is quasi-finite by Lemma 121.4.


00U6

Lemma 141.7. Let R S be a ring map. Let q be a prime of S lying over a

prime p of R. If
(1) R S is of finite presentation,
(2) Rp Sq is flat
(3) pSq is the maximal ideal of the local ring Sq , and
(4) the field extension (p) (q) is finite separable,
then R S is etale at q.
Proof. Apply Lemma 121.2 to find a g S, g 6 q such that q is the only prime
of Sg lying over p. We may and do replace S by Sg . Then S R (p) has a
unique prime, hence is a local ring, hence is equal to Sq /pSq
= (q). By Lemma
135.16 there exists a g S, g 6 q such that R Sg is smooth. Replace S by
Sg again we may assume that R S is smooth. By Lemma 135.10 we may even
assume that R S is standard smooth, say S = R[x1 , . . . , xn ]/(f1 , . . . , fc ). Since
S R (p) = (q) has dimension 0 we conclude that n = c, i.e., if R S is etale. 

08WD

Lemma 141.8. A ring map is etale if and only if it is flat, unramified, and of
finite presentation.
Proof. This follows by combining Lemmas 141.3 (flatness of etale maps), 141.5
(etale maps are unramified), and 141.7 (flat and unramified maps of finite presentation are etale).

Here is a completely new phenomenon.

00U7

Lemma 141.9. Let R S and R S 0 be etale. Then any R-algebra map S 0 S

is etale.
Proof. First of all we note that S 0 S is of finite presentation by Lemma 6.2.
Let q S be a prime ideal lying over the primes q0 S 0 and p R. By Lemma
141.5 the ring map Sq /pSq Sq0 0 /pSq0 0 is a map finite separable extensions of (p).
In particular it is flat. Hence by Lemma 127.8 we see that Sq0 0 Sq is flat. Thus
S 0 S is flat. Moreover, the above also shows that q0 Sq is the maximal ideal of
Sq and that the residue field extension of Sq0 0 Sq is finite separable. Hence from

COMMUTATIVE ALGEBRA

354

Lemma 141.7 we conclude that S 0 S is etale at q. Since being etale is local (see
Lemma 141.3) we win.

00U8

Lemma 141.10. Let : R S be a ring map. If R S is surjective, flat and

finitely presented then there exist an idempotent e R such that S = Re .
Proof. Since Spec(S) Spec(R) is a homeomorphism onto a closed subset (see
Lemma 16.7) and is open (see Proposition 40.8) we see that the image is D(e) for
some idempotent e R (see Lemma 20.3). Thus Re S induces a bijection on
spectra. Now this map induces an isomorphism on all local rings for example by
Lemmas 77.4 and 19.1. Then it follows that Re S is also injective, for example
see Lemma 23.1.


04D1

Lemma 141.11. Let R be a ring and let I R be an ideal. Let R/I S be an

etale ring map. Then there exists an etale ring map R S such that S
= S/IS as
R/I-algebras.
Proof. By Lemma 141.2 we can write S = (R/I)[x1 , . . . , xn ]/(f 1 , . . . , f n ) as in
f i
Definition 135.6 with = det( x
)i,j=1,...,n invertible in S. Just take some lifts fi
j
fi
)i,j=1,...,c
and set S = R[x1 , . . . , xn , xn+1 ]/(f1 , . . . , fc , xn+1 1) where = det( x
j
as in Example 135.8. This proves the lemma.


05YT

Lemma 141.12. Consider a commutative diagram

0

/J
O

/ B0
O

/B
O

/0

/I

/ A0

/A

/0

with exact rows where B 0 B and A0 A are surjective ring maps whose kernels
are ideals of square zero. If A B is etale, and J = I A B, then A0 B 0 is
etale.
Proof. By Lemma 141.11 there exists an etale ring map A0 C such that C/IC =
B. Then A0 C is formally smooth (by Proposition 136.13) hence we get an A0 algebra map : C B 0 . Since A0 C is flat we have I A B = I A C/IC = IC.
Hence the assumption that J = I A B implies that induces an isomorphism
IC J and an isomorphism C/IC B 0 /IB 0 , whence is an isomorphism.

00UA

Example 141.13. Let n, m 1 be integers. Consider the ring map

R = Z[a1 , . . . , an+m ]

S = Z[b1 , . . . , bn , c1 , . . . , cm ]

a1

b1 + c1

a2

b2 + b1 c1 + c2

...

...

...

an+m

bn cm

of Example 134.7. Write symbolically

S = R[b1 , . . . , cm ]/({ak (bi , cj ) ak }k=1,...,n+m )

COMMUTATIVE ALGEBRA

where for example a1 (bi , cj ) = b1 + c1 .

1 c1 . . .
0
1 c1

. . . . . . . . .

0 ... 0

1 b1 . . .

0
1 b1

. . . . . . . . .
0 ... 0

355

The matrix of partial derivatives is

cm 0 . . . 0
. . . cm . . . 0

. . . . . . . . . . . .

1
c1 . . . c m

bn
0 ... 0

. . . bn . . . 0

. . . . . . . . . . . .
1
b1 . . . bn

The determinant of this matrix is better known as the resultant of the polynomials
g = xn + b1 xn1 + . . . + bn and h = xm + c1 xm1 + . . . + cm , and the matrix above
is known as the Sylvester matrix associated to g, h. In a formula = Resx (g, h).
The Sylvester matrix is the transpose of the matrix of the linear map
S[x]<m S[x]<n

a b 7

S[x]<n+m
ag + bh

Let q S be any prime. By the above the following are equivalent:

(1) R S is etale at q,
(2) = Resx (g, h) 6 q,
(3) the images g, h (q)[x] of the polynomials g, h are relatively prime in
(q)[x].
The equivalence of (2) and (3) holds because the image of the Sylvester matrix in
Mat(n + m, (q)) has a kernel if and only if the polynomials g, h have a factor in
common. We conclude that the ring map
1
1
R S[ ] = S[
]

Resx (g, h)
is etale.
Lemma 141.2 tells us that it does not really make sense to define a standard etale
morphism to be a standard smooth morphism of relative dimension 0. As a model
for an etale morphism we take the example given by a finite separable extension
k k 0 of fields. Namely, we can always find an element k 0 such that k 0 = k()
and such that the minimal polynomial f (x) k[x] of has derivative f 0 which is
relatively prime to f .
00UB

Definition 141.14. Let R be a ring. Let g, f R[x]. Assume that f is monic

and the derivative f 0 is invertible in the localization R[x]g /(f ). In this case the
ring map R R[x]g /(f ) is said to be standard etale.

00UC

Lemma 141.15. Let R R[x]g /(f ) be standard etale.

(1) The ring map R R[x]g /(f ) is etale.
(2) For any ring map R R0 the base change R0 R0 [x]g /(f ) of the standard
etale ring map R R[x]g /(f ) is standard etale.
(3) Any principal localization of R[x]g /(f ) is standard etale over R.
(4) A composition of standard etale maps is not standard etale in general.
Proof. Omitted. Here is an example for (4). The ring map F2 F22 is standard
etale. The ring map F22 F22 F22 F22 F22 is standard etale. But the ring
map F2 F22 F22 F22 F22 is not standard etale.


COMMUTATIVE ALGEBRA

356

Standard etale morphisms are a convenient way to produce etale maps. Here is an
example.
00UD

Lemma 141.16. Let R be a ring. Let p be a prime of R. Let (p) L be a finite

separable field extension. There exists an etale ring map R R0 together with a
prime p0 lying over p such that the field extension (p) (p0 ) is isomorphic to
(p) L.
Proof. By the theorem of the primitive element we may write L = (p)[]. Let
f (p)[x] denote the minimal polynomial for (in particular this is monic). After
replacing by c for some c R, c 6 p we may assume all the coefficients of f
are in the image of R (p) (verification omitted). Thus we can find a monic
polynomial f R[x] which maps to f in (p)[x]. Since (p) L is separable, we
0
0
see that gcd(f , f ) = 1. Hence there is an element L such that f () = 1.
Thus we get a R-algebra map
R[x, 1/f 0 ]/(f )

x 7

1/f 0

The left hand side is a standard etale algebra R0 over R and the kernel of the ring
map gives the desired prime.

00UE

Proposition 141.17. Let R S be a ring map. Let q S be a prime. If R S

is etale at q, then there exists a g S, g 6 q such that R Sg is standard etale.
Proof. The following proof is a little roundabout and there may be ways to shorten
it.
Step 1. By Definition 141.1 there exists a g S, g 6 q such that R Sg is etale.
Thus we may assume that S is etale over R.
Step 2. By Lemma 141.3 there exists an etale ring map R0 S0 with R0 of finite
type over Z, and a ring map R0 R such that R = R R0 S0 . Denote q0 the prime
of S0 corresponding to q. If we show the result for (R0 S0 , q0 ) then the result
follows for (R S, q) by base change. Hence we may assume that R is Noetherian.
Step 3. Note that R S is quasi-finite by Lemma 141.6. By Lemma 122.15 there
exists a finite ring map R S 0 , an R-algebra map S 0 S, an element g 0 S 0
such that g 0 6 q such that S 0 S induces an isomorphism Sg0 0
= Sg0 . (Note
that of course S 0 is not etale over R in general.) Thus we may assume that (a)
R is Noetherian, (b) R S is finite and (c) R S is etale at q (but no longer
necessarily etale at all primes).
Step 4. Let p R be the prime corresponding to q. Consider the fibre ring
S R (p). This is a finite algebra over (p). Hence it is Artinian (see Lemma 52.2)
and so a finite product of local rings
Yn
S R (p) =
Ai
i=1

see Proposition 59.6. One of the factors, say A1 , is the local ring Sq /pSq which is
isomorphic to (q), see Lemma 141.5. The other factors correspond to the other
primes, say q2 , . . . , qn of S lying over p.

COMMUTATIVE ALGEBRA

357

Step 5. We may choose a nonzero element (q) which generates the finite
separable field extension (p) (q) (so even if the field extension is trivial we do
not allow = 0). Note that for any (p) the element also generates (q)
over (p). Consider the element
Yn
t = (, 0, . . . , 0)
Ai = S R (p).
i=1

After possibly replacing by as above we may assume that t is the image of

t S. Let I R[x] be the kernel of the R-algebra map R[x] S which maps x to
t. Set S 0 = R[x]/I, so S 0 S. Here is a diagram
R[x]
O

6/ S

/ S0
=

R
By construction the primes qj , j 2 of S all lie over the prime (p, x) of R[x],
whereas the prime q lies over a different prime of R[x] because 6= 0.
Step 6. Denote q0 S 0 the prime of S 0 corresponding to q. By the above q is
the only prime of S lying over q0 . Thus we see that Sq = Sq0 , see Lemma 40.11
(we have going up for S 0 S by Lemma 35.20 since S 0 S is finite as R S
is finite). It follows that Sq0 0 Sq is finite and injective as the localization of the
finite injective ring map S 0 S. Consider the maps of local rings
Rp Sq0 0 Sq
The second map is finite and injective. We have Sq /pSq = (q), see Lemma 141.5.
Hence a fortiori Sq /q0 Sq = (q). Since
(p) (q0 ) (q)
and since is in the image of (q0 ) in (q) we conclude that (q0 ) = (q). Hence
by Nakayamas Lemma 19.1 applied to the Sq0 0 -module map Sq0 0 Sq , the map
Sq0 0 Sq is surjective. In other words, Sq0 0
= Sq .
Step 7. By Lemma 125.7 there exist g S, g 6 q and g 0 S 0 , g 0 6 q0 such that
Sg0 0
= Sg . As R is Noetherian the ring S 0 is finite over R because it is an Rsubmodule of the finite R-module S. Hence after replacing S by S 0 we may assume
that (a) R is Noetherian, (b) S finite over R, (c) S is etale over R at q, and (d)
S = R[x]/I.
Step 8. Consider the ring S R (p) = (p)[x]/I where I = I (p)[x] is the ideal
generated by I in (p)[x]. As (p)[x] is a PID we know that I = (h) for some monic
h (p). After replacing h by h for some (p) we may assume that h is the
image of some h R[x]. (The problem is that we do not know if we may choose h
monic.) Also, as in Step 4 we know that S R (p) = A1 . . . An with A1 = (q)
a finite separable extension of (p) and A2 , . . . , An local. This implies that
e2

en

h = h1 h2 . . . hn

for certain pairwise coprime irreducible monic polynomials hi (p)[x] and certain
ei
e2 , . . . , en 1. Here the numbering is chosen so that Ai = (p)[x]/(hi ) as (p)[x]algebras. Note that h1 is the minimal polynomial of (q) and hence is a
separable polynomial (its derivative is prime to itself).

COMMUTATIVE ALGEBRA

358

Step 9. Let m I be a monic element; such an element exists because the ring
extension R R[x]/I is finite hence integral. Denote m the image in (p)[x]. We
may factor
d1 d2

dn

m = kh1 h2 . . . hn

for some d1 1, dj ej , j = 2, . . . , n and k (p)[x] prime to all the hi . Set

f = ml + h where l deg(m) > deg(h), and l 2. Then f is monic as a polynomial
over R. Also, the image f of f in (p)[x] factors as
e2

en

l ld1 ld2

ldn

e2

en

l ld1 1 ld2
h2

f = h1 h2 . . . hn +k h1 h2 . . . hn = h1 (h2 . . . hn +k h1

ldn

. . . hn ) = h1 w

with w a polynomial relatively prime to h1 . Set g = f 0 (the derivative with respect

to x).
Step 10. The ring map R[x] S = R[x]/I has the properties: (1) it maps f to
zero, and (2) it maps g to an element of S \ q. The first assertion is clear since f
is an element of I. For the second assertion we just have to show that g does not
map to zero in (q) = (p)[x]/(h1 ). The image of g in (p)[x] is the derivative of
f . Thus (2) is clear because
g=

df
dh1
dw
=w
+ h1
,
dx
dx
dx

w is prime to h1 and h1 is separable.

Step 11. We conclude that : R[x]/(f ) S is a surjective ring map, R[x]g /(f ) is
etale over R (because it is standard etale, see Lemma 141.15) and (g) 6 q. Pick
an element g 0 R[x]/(f ) such that also (g 0 ) 6 q and S(g0 ) is etale over R (which
exists since S is etale over R at q). Then the ring map R[x]gg0 /(f ) S(g) is a
surjective map of etale algebras over R. Hence it is etale by Lemma 141.9. Hence
it is a localization by Lemma 141.10. Thus a localization of S at an element not in
q is isomorphic to a localization of a standard etale algebra over R which is what
we wanted to show.

The following two lemmas say that the etale topology is coarser than the topology
generated by Zariski coverings and finite flat morphisms. They should be skipped
on a first reading.
00UF

Lemma 141.18. Let R S be a standard etale morphism. There exists a ring

map R S 0 with the following properties
(1) R S 0 is finite, finitely presented, and flat (in other words S 0 is finite
projective as an R-module),
(2) Spec(S 0 ) Spec(R) is surjective,
(3) for every prime q S, lying over p R and every prime q0 S 0 lying over
p there exists a g 0 S 0 , g 0 6 q0 such that the ring map R Sg0 0 factors
through a map : S Sg0 0 with 1 (q0 Sg0 0 ) = q.
Proof. Let S = R[x]g /(f ) be a presentation of S as in Definition 141.14. Write
f = xn + a1 xn1 + . . . + an with ai R. By Lemma 134.9Qthere exists a finite
locally free and faithfully flat ring map R S 0 such that f = (x i ) for certain
i S 0 . Hence R S 0 satisfies conditions (1), (2). Let q R[x]/(f ) be a prime

COMMUTATIVE ALGEBRA

359

ideal with g 6 q (i.e., it corresponds to a prime of S). Let p = R q and let q0 S 0

be a prime lying over p. Note that there are n maps of R-algebras
i : R[x]/(f )

S0

To finish the proof we have to show that for some i we have (a) the image of i (g)
0
0
in (q0 ) is not zero, and (b) 1
i (q ) = q. Because then we can just take g = i (g),
and = i for that i.
Let f denote the image of f in (p)[x]. Note that as a point of Spec((p)[x]/(f ))
the prime q corresponds to an irreducible factor f1 of f . Moreover, g 6 q means
that f1 does not divide the image g of g in (p)[x]. Denote 1 , . . . , n the images
of 1 , . . . , n in (q0 ). Note that the polynomial f splits completely in (q0 )[x],
namely
Y
f=
(x i )
i

Moreover i (g) reduces to g(i ). It follows we may pick i such that f1 (i ) = 0 and
g(i ) 6= 0. For this i properties (a) and (b) hold. Some details omitted.

00UG

Lemma 141.19. Let R S be a ring map. Assume that

(1) R S is etale, and
(2) Spec(S) Spec(R) is surjective.
Then there exists a ring map R S 0 such that
(1) R S 0 is finite, finitely presented, and flat (in other words it is finite
projective as an R-module),
(2) Spec(S 0 ) Spec(R) is surjective,
(3) for every prime q0 S 0 there exists a g 0 S 0 , g 0 6 q0 such that the ring
map R Sg0 0 factors as R S Sg0 0 .
Proof. By Proposition 141.17 and the quasi-compactness of Spec(S) (see Lemma
16.10) we can find g1 , . . . , gn S generating the unit ideal of S such that
Q each R
Sgi is standard etale. If we prove the lemma for the ring map R i=1,...,n Sgi
then Q
the lemma follows for the ring map R S. Hence we may assume that
S = i=1,...,n Si is a finite product of standard etale morphisms.
For each i choose a ring map R Si0 as in Lemma 141.18 adapted to the standard
etale morphism R Si . Set S 0 = S10 R . . . R Sn0 ; we will use the R-algebra
maps Si0 S 0 without further mention below. We claim this works. Properties
(1) and (2) are immediate. For property (3) suppose that q0 S 0 is a prime.
Denote p its image in Spec(R). Choose i {1, . . . , n} such that p is in the image
of Spec(Si ) Spec(R); this is possible by assumption. Set q0i Si0 the image of
q0 in the spectrum of Si0 . By construction of Si0 there exists a gi0 Si0 such that
R (Si0 )gi0 factors as R Si (Si0 )gi0 . Hence also R Sg0 0 factors as
i

R Si (Si0 )gi0 Sg0 i0

as desired.
00UH

Lemma 141.20. Let R be a ring. Let f R[x] be a monic polynomial. Let p be

a prime of R. Let f mod p = gh be a factorization of the image of f in (p)[x]. If
gcd(g, h) = 1, then there exist
(1) an etale ring map R R0 ,

COMMUTATIVE ALGEBRA

360

(2) a prime p0 R0 lying over p, and

(3) a factorization f = gh in R0 [x]
such that
(1) (p) = (p0 ),
(2) g = g mod p0 , h = h mod p0 , and
(3) the polynomials g, h generate the unit ideal in R0 [x].
Proof. Suppose g = b0 xn + b1 xn1 + . . . + bn , and h = c0 xm + c1 xm1 + . . . + cm
with b0 , c0 (p) nonzero. After localizing R at some element of R not contained
in p we may assume b0 is the image of an invertible element b0 R. Replacing
g by g/b0 and h by b0 h we reduce to the case where g, h are monic (verification
omitted). Say g = xn + b1 xn1 + . . . + bn , and h = xm + c1 xm1 + . . . + cm . Write
f = xn+m + a1 xn1 + . . . + an+m . Consider the fibre product
R0 = R Z[a1 ,...,an+m ] Z[b1 , . . . , bn , c1 , . . . , cm ]
where the map Z[ak ] Z[bi , cj ] is as in Examples 134.7 and 141.13. By construction there is an R-algebra map
R0 = R Z[a1 ,...,an+m ] Z[b1 , . . . , bn , c1 , . . . , cm ] (p)
which maps bi to bi and cj to cj . Denote p0 R0 the kernel of this map. Since
by assumption the polynomials g, h are relatively prime we see that the element
= Resx (g, h) Z[bi , cj ] (see Example 141.13) does not map to zero in (p) under
the displayed map. We conclude that R R0 is etale at p0 . In fact a solution to the
problem posed in the lemma is the ring map R R0 [1/] and the prime p0 R0 [1/].
Because Resx (f, g) is invertible in this ring the Sylvester matrix is invertible over
R0 and hence 1 = ag + bh for some a, b R0 [x] see Example 141.13.

The following lemmas say roughly that after an etale extension a quasi-finite ring
map becomes finite. To help interpret the results recall that the locus where a finite
type ring map is quasi-finite is open (see Lemma 122.14) and that formation of this
locus commutes with arbitrary base change (see Lemma 121.8).
00UI

Lemma 141.21. Let R S 0 S be ring maps. Let p R be a prime. Let

g S 0 be an element. Assume
(1) R S 0 is integral,
(2) R S is finite type,
(3) Sg0
= Sg , and
(4) g invertible in S 0 R (p).
Then there exists a f R, f 6 p such that Rf Sf is finite.
Proof. By assumption the image T of V (g) Spec(S 0 ) under the morphism
Spec(S 0 ) Spec(R) does not contain p. By Section 40 especially, Lemma 40.6
we see T is closed. Pick f R, f 6 p such that T V (f ) = . Then we see that
g becomes invertible in Sf0 . Hence Sf0
= Sf . Thus Sf is both of finite type and
integral over Rf , hence finite.


00UJ

Lemma 141.22. Let R S be a ring map. Let q S be a prime lying over the
prime p R. Assume R S finite type and quasi-finite at q. Then there exists
(1) an etale ring map R R0 ,
(2) a prime p0 R0 lying over p,

COMMUTATIVE ALGEBRA

361

(3) a product decomposition

R 0 R S = A B
with the following properties
(1) (p) = (p0 ),
(2) R0 A is finite,
(3) A has exactly one prime r lying over p0 , and
(4) r lies over q.
Proof. Let S 0 S be the integral closure of R in S. Let q0 = S 0 q. By Zariskis
Main Theorem 122.13 there exists a g S 0 , g 6 q0 such that Sg0
= Sg . Consider
0
0
0
the fibre rings F = S R (p) and F = S R (p). Denote q the prime of F 0
corresponding to q0 . Since F 0 is integral over (p) we see that q0 is a closed point
of Spec(F 0 ), see Lemma 35.17. Note that q defines an isolated closed point q of
Spec(F ) (see Definition 121.3). Since Sg0
= Sg we have Fg0
= Fg , so q and q0 have
isomorphic open neighbourhoods in Spec(F ) and Spec(F 0 ). We conclude the set
{q0 } Spec(F 0 ) is open. Combined with q0 being closed (shown above) we conclude
that q0 defines an isolated closed point of Spec(F 0 ) as well.
An additional small remark is that under the map Spec(F ) Spec(F 0 ) the point
q is the only point mapping to q0 . This follows from the discussion above.
By Lemma 22.3 we may write F 0 = F10 F20 with Spec(F10 ) = {q0 }. Since F 0 =
S 0 R (p), there exists an s0 S 0 which maps to the element (r, 0) F10 F20 = F 0
for some r R, r 6 p. In fact, what we will use about s0 is that it is an element of
S 0 , not contained in q0 , and contained in any other prime lying over p.
Let f (x) R[x] be a monic polynomial such that f (s0 ) = 0. Denote f (p)[x]
the image. We can factor it as f = xe h where h(0) 6= 0. By Lemma 141.20 we can
find an etale ring extension R R0 , a prime p0 lying over p, and a factorization
f = hi in R0 [x] such that (p) = (p0 ), xe = h mod p0 , i = i mod p0 , and we can
write ah + bi = 1 in R0 [x] (for suitable a, b).
Consider the elements h(s0 ), i(s0 ) R0 R S 0 . By construction we have h(s0 )i(s0 ) =
f (s0 ) = 0. On the other hand they generate the unit ideal since a(s0 )h(s0 ) +
b(s0 )i(s0 ) = 1. Thus we see that R0 R S 0 is the product of the localizations at these
elements:
R0 R S 0 = (R0 R S 0 )h(s0 ) (R0 R S 0 )i(s0 ) = S10 S20
Moreover this product decomposition is compatible with the product decomposition
we found for the fibre ring F 0 ; this comes from our choice of s0 , h which guarantee
that q0 is the only prime of F 0 which does not contain the image of h(s0 ) in F 0 .
Here we use that the fibre ring of R0 R S 0 over R0 at p0 is the same as F 0 due to the
fact that (p) = (p0 ). It follows that S10 has exactly one prime, say r0 , lying over
p0 and that this prime lies over q. Hence the element g S 0 maps to an element of
S10 not contained in r0 .
The base change R0 R S inherits a similar product decomposition
R0 R S = (R0 R S)h(s0 ) (R0 R S)i(s0 ) = S1 S2
It follows from the above that S1 has exactly one prime, say r, lying over p0 (consider
the fibre ring as above), and that this prime lies over q.

COMMUTATIVE ALGEBRA

362

Now we may apply Lemma 141.21 to the ring maps R0 S10 S1 , the prime p0
and the element g to see that after replacing R0 by a principal localization we can
assume that S1 is finite over R0 as desired.

00UK

Lemma 141.23. Let R S be a ring map. Let p R be a prime. Assume

R S finite type. Then there exists
(1) an etale ring map R R0 ,
(2) a prime p0 R0 lying over p,
(3) a product decomposition
R 0 R S = A1 . . . An B
with the following properties
(1) we have (p) = (p0 ),
(2) each Ai is finite over R0 ,
(3) each Ai has exactly one prime ri lying over p0 , and
(4) R0 B not quasi-finite at any prime lying over p0 .
Proof. Denote F = S R (p) the fibre ring of S/R at the prime p. As F is of
finite type over (p) it is Noetherian and hence Spec(F ) has finitely many isolated
closed points. If there are no isolated closed points, i.e., no primes q of S over p
such that S/R is quasi-finite at q, then the lemma holds. If there exists at least
one such prime q, then we may apply Lemma 141.22. This gives a diagram
SO

/ R 0 R S
O

/ R0

A8 1 B 0

as in said lemma. Since the residue fields at p and p0 are the same, the fibre rings of
S/R and (A B)/R0 are the same. Hence, by induction on the number of isolated
closed points of the fibre we may assume that the lemma holds for R0 B and p0 .
Thus we get an etale ring map R0 R00 , a prime p00 R00 and a decomposition
R00 R0 B 0 = A2 . . . An B
We omit the verification that the ring map R R00 , the prime p00 and the resulting
decomposition
R00 R S = (R00 R0 A1 ) A2 . . . An B
is a solution to the problem posed in the lemma.
00UL

Lemma 141.24. Let R S be a ring map. Let p R be a prime. Assume

R S finite type. Then there exists
(1) an etale ring map R R0 ,
(2) a prime p0 R0 lying over p,
(3) a product decomposition
R 0 R S = A1 . . . An B
with the following properties
(1) each Ai is finite over R0 ,
(2) each Ai has exactly one prime ri lying over p0 ,
(3) the finite field extensions (p0 ) (ri ) are purely inseparable, and

COMMUTATIVE ALGEBRA

363

(4) R0 B not quasi-finite at any prime lying over p0 .

Proof. The strategy of the proof is to make two etale ring extensions: first we
control the residue fields, then we apply Lemma 141.23.
Denote F = S R (p) the fibre ring of S/R at the prime p. As in the proof of
Lemma 141.23 there are finitely may primes, say q1 , . . . , qn of S lying over R at
which the ring map R S is quasi-finite. Let (p) Li (qi ) be the subfield
such that (p) Li is separable, and the field extension Li (qi ) is purely
inseparable. Let (p) L be a finite Galois extension into which Li embeds for
i = 1, . . . , n. By Lemma 141.16 we can find an etale ring extension R R0 together
with a prime p0 lying over p such that the field extension (p) (p0 ) is isomorphic
to (p) L. Thus the fibre ring of R0 R S at p0 is isomorphic to F (p) L. The
primes lying over qi correspond to primes of (qi )(p) L which is a product of fields
purely inseparable over L by our choice of L and elementary field theory. These
are also the only primes over p0 at which R0 R0 R S is quasi-finite, by Lemma
121.8. Hence after replacing R by R0 , p by p0 , and S by R0 R S we may assume
that for all primes q lying over p for which S/R is quasi-finite the field extensions
(p) (q) are purely inseparable.
Next apply Lemma 141.23. The result is what we want since the field extensions
do not change under this etale ring extension.

142. Local homomorphisms
053J
053K

Lemma 142.1. Let (R, mR ) (S, mS ) be a local homomorphism of local rings.

Assume S is the localization of an etale ring extension of R. Then there exists
a finite, finitely presented, faithfully flat ring map R S 0 such that for every
maximal ideal m0 of S 0 there is a factorization
0
R S Sm
0.
0
of the ring map R Sm
0.

Proof. Write S = Tq for some etale R-algebra T . By Proposition 141.17 we may

assume T is standard etale. Apply Lemma 141.18 to the ring map R T to get
R S 0 . Then in particular for every maximal ideal m0 of S 0 we get a factorization
: T Sg0 0 for some g 0 6 m0 such that q = 1 (m0 Sg0 0 ). Thus induces the desired
0
local ring map S Sm

0.
143. Integral closure and smooth base change
03GC
03GD

Lemma 143.1. Let R be a ring. Let f R[x] be a monic polynomial. Let R B

0
be a ring map. If
Ph iB[x]/(f ) is integral over R, then the element f h can be
0
written as f h = i bi x with bi B integral over R.
Proof. Say he + r1 he1 + . . . + re = 0 in the ring B[x]/(f ) with ri R. There
exists a finite free ring extension B B 0 such that f = (x 1 ) . . . (x d ) for
some i B 0 , see Lemma 134.9. Note that each i is integral over R. We may

COMMUTATIVE ALGEBRA

364

represent h = h0 + h1 x + . . . + hd1 xd1 with hi B. Then it is a universal fact

that
X
f 0h
h(i )(x 1 ) . . . (x\
i ) . . . (x d )
i=1,...,d

as elements of B[x]/(f ). You prove this by evaluating both sides at the points i
over the ring Buniv = Z[i , hj ] (some details omitted). By our assumption that h
satisfies he + r1 he1 + . . . + re = 0 in the ring B[x]/(f ) we see that
h(i )e + r1 h(i )e1 + . . . + re = 0
0
0
in
i ) is integral over R. Using the formula above we see that f h
P B . Hence 0 h(
j
0
0
0
0
j=0,...,d1 bj x in B [x]/(f ) with bj B integral over R. However, since f h
d1
0
0
0
B[x]/(f ) and since 1, x, . . . , x
is a B -basis for B [x]/(f ) we see that bj B as
desired.


03GE

Lemma 143.2. Let R S be an etale ring map. Let R B be any ring map.
Let A B be the integral closure of R in B. Let A0 S R B be the integral
closure of S in S R B. Then the canonical map S R A A0 is an isomorphism.
Proof. The map S R A A0 is injective because A B and R S is flat. We
are going to use repeatedly that taking integral closure commutes with localization,
see Lemma 35.9. Hence we may localize on S, by Lemma 23.2 (the criterion for
checking whether an S-module map is an isomorphism). Thus we may assume
that S = R[x]g /(f ) = (R[x]/(f ))g is standard etale over R, see Proposition 141.17.
Applying localization one more time we see that A0 is (A00 )g where A00 is the integral
00
closure of R[x]/(f ) in B[x]/(f ). Suppose that a A
P . Iti suffices to show that a0
0
is in S R A. By Lemma 143.1 we see that f a =
ai x with ai A. Since f
is invertible in B[x]g /(f ) (by definition of a standard etale ring map) we conclude
that a S R A as desired.


03GF

Example 143.3. Let p be a prime number. The ring extension

R = Z[1/p] R0 = Z[1/p][x]/(xp1 + . . . + x + 1)
has the following property: For d < p there exist elements 0 , . . . , d1 R0 such
that
Y
(i j )
0i<j<d

is a unit in R . Namely, take i equal to the class of xi in R0 for i = 0, . . . , p 1.

Then we have
Y
Tp 1 =
(T i )
i=0,...,p1
p1

in R [T ]. Namely, the ring Q[x]/(x

+ . . . + x + 1) is a field because the cyclotomic polynomial xp1 + . . . + x + 1 is irreducible over Q and the i are pairwise
distinct roots of T p 1, whence the equality. Taking derivatives on both sides and
substituting T = i we obtain
pip1 = (i 1 ) . . . (\
i i ) . . . (i 1 )
and we see this is invertible in R0 .
03GG

Lemma 143.4. Let R S be a smooth ring map. Let R B be any ring map.
Let A B be the integral closure of R in B. Let A0 S R B be the integral
closure of S in S R B. Then the canonical map S R A A0 is an isomorphism.

COMMUTATIVE ALGEBRA

365

Proof. Arguing as in the proof of Lemma 143.2 we may localize on S. Hence we

may assume that R S is a standard smooth ring map, see Lemma 135.10. By
definition of a standard smooth ring map we see that S is etale over a polynomial
ring R[x1 , . . . , xn ]. Since we have seen the result in the case of an etale ring extension (Lemma 143.2) this reduces us to the case where S = R[x]. Thus we have to
show
X
f=
bi xi integral over R[x] each bi integral over R.
The implication from right to left holds because the set of elements in B[x] integral
over R[x] is a ring (Lemma 35.7) and contains x.
P
Suppose that f B[x] is integral over R[x], and assume that f = i<d bi xi has
degree < d. Since integral closure and localization commute, it suffices to show
there exist distinct primes p, q such that each bi is integral both over R[1/p] and
over R[1/q]. Hence, we can find a finite free Q
ring extension R R0 such that
R0 contains 1 , . . . , d with the property that i<j (i j ) is a unit in R0 , see
Example 143.3. In this case we have the universal equality
f=

X
i

f (i )

(x 1 ) . . . (x\
i ) . . . (x d )
.
(i 1 ) . . . (\
i i ) . . . (i d )

OK, and the elements f (i ) are integral over R0 since (R0 R B)[x] R0 R B,
h 7 h(i ) is a ring map. Hence we see that the coefficients of f in (R0 R B)[x]
are integral over over R0 . Since R0 is finite over R (hence integral over R) we see
that they are integral over R also, as desired.

144. Formally unramified maps
00UM

It turns out to be logically more efficient to define the notion of a formally unramified map before introducing the notion of a formally etale one.

00UN

Definition 144.1. Let R S be a ring map. We say S is formally unramified

over R if for every commutative solid diagram
SO
R

/ A/I
O
!

/A

where I A is an ideal of square zero, there exists at most one dotted arrow
making the diagram commute.
00UO

Lemma 144.2. Let R S be a ring map. The following are equivalent:

(1) R S is formally unramified,
(2) the module of differentials S/R is zero.
Proof. Let J = Ker(S R S S) be the kernel of the multiplication map. Let
Auniv = S R S/J 2 . Recall that Iuniv = J/J 2 is isomorphic to S/R , see Lemma
130.13. Moreover, the two R-algebra maps 1 , 2 : S Auniv , 1 (s) = s
1 mod J 2 , and 2 (s) = 1 s mod J 2 differ by the universal derivation d : S
S/R = Iuniv .
Assume R S formally unramified. Then we see that 1 = 2 . Hence d(s) = 0
for all s S. Hence S/R = 0.

COMMUTATIVE ALGEBRA

366

Assume that S/R = 0. Let A, I, R A, S A/I be a solid diagram as in

Definition 144.1. Let 1 , 2 : S A be two dotted arrows making the diagram
commute. Consider the R-algebra map Auniv A defined by the rule s1 s2 7
1 (s1 )2 (s2 ). We omit the verification that this is well defined. Since Auniv
= S as
Iuniv = S/R = 0 we conclude that 1 = 2 .

04E8

Lemma 144.3. Let R S be a ring map. The following are equivalent:

(1) R S is formally unramified,
(2) R Sq is formally unramified for all primes q of S, and
(3) Rp Sq is formally unramified for all primes q of S with p = R q.
Proof. We have seen in Lemma 144.2 that (1) is equivalent to S/R = 0. Similarly,
by Lemma 130.8 we see that (2) and (3) are equivalent to (S/R )q = 0 for all q.
Hence the equivalence follows from Lemma 23.1.


04E9

Lemma 144.4. Let A B be a formally unramified ring map.

(1) For S A a multiplicative subset, S 1 A S 1 B is formally unramified.
(2) For S B a multiplicative subset, A S 1 B is formally unramified.
Proof. Follows from Lemma 144.3. (You can also deduce it from Lemma 144.2
combined with Lemma 130.8.)


07QE

Lemma 144.5. Let R be a ring. Let I be a directed partially ordered set. Let
(Si , ii0 ) be a system of R-algebras over I. If each R Si is formally unramified,
then S = colimiI Si is formally unramified over R
Proof. Consider a diagram as in Definition 144.1. By assumption there exists at
most one R-algebra map Si A lifting the compositions Si S A/I. Since
every element of S is in the image of one of the maps Si S we see that there is
at most one map S A fitting into the diagram.

145. Conormal modules and universal thickenings

04EA

It turns out that one can define the first infinitesimal neighbourhood not just for
a closed immersion of schemes, but already for any formally unramified morphism.
This is based on the following algebraic fact.

04EB

Lemma 145.1. Let R S be a formally unramified ring map. There exists a

surjection of R-algebras S 0 S whose kernel is an ideal of square zero with the
following universal property: Given any commutative diagram
SO
R

/ A/I
O

/A

where I A is an ideal of square zero, there is a unique R-algebra map a0 : S 0 A

such that S 0 A A/I is equal to S 0 S A/I.
Proof. Choose a set of generators zi S, i I for S as an R-algebra. Let
P = R[{xi }iI ] denote the polynomial ring on generators xi , i I. Consider the
R-algebra map P S which maps xi to zi . Let J = Ker(P S). Consider the
map
d : J/J 2 P/R P S

COMMUTATIVE ALGEBRA

367

see Lemma 130.9. This is surjective since S/R = 0 by assumption, see Lemma
144.2. Note that P/R is free on dxi , and hence the module P/R P S is free over
S. Thus we may choose a splitting of the surjection above and write
J/J 2 = K P/R P S
Let J 2 J 0 J be the ideal of P such that J 0 /J 2 is the second summand in the
decomposition above. Set S 0 = P/J 0 . We obtain a short exact sequence
0 J/J 0 S 0 S 0
and we see that J/J 0
= K is a square zero ideal in S 0 . Hence
SO

/S
O
/ S0

is a diagram as above. In fact we claim that this is an initial object in the category
of diagrams. Namely, let (I A, a, b) be an arbitrary diagram. We may choose an
R-algebra map : P A such that
SO
R

/S
O
/P

/ A/I
O
/3 A

is commutative. Now it may not be the case that (J 0 ) = 0, in other words it may
not be true that factors through S 0 = P/J 0 . But what is clear is that (J 0 ) I
and since (J) I and I 2 = 0 we have (J 2 ) = 0. Thus the obstruction to
finding a morphism from (J/J 0 S 0 , 1, R S 0 ) to (I A, a, b) is the corresponding
S-linear map : J 0 /J 2 I. The choice in picking lies in the choice of (xi ). A
different choice of , say 0 , is gotten by taking 0 (xi ) = (xi ) + i with i I. In
this case, for g J 0 , we obtain
X
g
0 (g) = (g) +
i
.
i
xi
g
dxi is an isomorphism
Since the map d|J 0 /J 2 : J 0 /J 2 P/R P S given by g 7 x
i
by construction, we see that there is a unique choice of i I such that 0 (g) = 0
for all g J 0 . (Namely, i is (g) where g J 0 /J 2 is the unique element with
g
xj = 1 if i = j and 0 else.) The uniqueness of the solution implies the uniqueness
required in the lemma.


In the situation of Lemma 145.1 the R-algebra map S 0 S is unique up to unique

isomorphism.
04EC

Definition 145.2. Let R S be a formally unramified ring map.

(1) The universal first order thickening of S over R is the surjection of Ralgebras S 0 S of Lemma 145.1.
(2) The conormal module of R S is the kernel I of the universal first order
thickening S 0 S, seen as a S-module.
We often denote the conormal module CS/R in this situation.

COMMUTATIVE ALGEBRA

04ED

Lemma 145.3. Let I R be an ideal of a ring. The universal first order thickening of R/I over R is the surjection R/I 2 R/I. The conormal module of R/I
over R is C(R/I)/R = I/I 2 .
Proof. Omitted.

04EE

368

Lemma 145.4. Let A B be a formally unramified ring map. Let : B 0 B

be the universal first order thickening of B over A.
(1) Let S A be a multiplicative subset. Then S 1 B 0 S 1 B is the universal first order thickening of S 1 B over S 1 A. In particular S 1 CB/A =
CS 1 B/S 1 A .
(2) Let S B be a multiplicative subset. Then S 0 = 1 (S) is a multiplicative
subset in B 0 and (S 0 )1 B 0 S 1 B is the universal first order thickening
of S 1 B over A. In particular S 1 CB/A = CS 1 B/A .
Note that the lemma makes sense by Lemma 144.4.
Proof. With notation and assumptions as in (1). Let (S 1 B)0 S 1 B be the
universal first order thickening of S 1 B over S 1 A. Note that S 1 B 0 S 1 B is
a surjection of S 1 A-algebras whose kernel has square zero. Hence by definition
we obtain a map (S 1 B)0 S 1 B 0 compatible with the maps towards S 1 B.
Consider any commutative diagram
BO

/ S 1 B
O

/ D/I
O

/ S 1 A

/D

where I D is an ideal of square zero. Since B 0 is the universal first order

thickening of B over A we obtain an A-algebra map B 0 D. But it is clear that
the image of S in D is mapped to invertible elements of D, and hence we obtain a
compatible map S 1 B 0 D. Applying this to D = (S 1 B)0 we see that we get a
map S 1 B 0 (S 1 B)0 . We omit the verification that this map is inverse to the
map described above.
With notation and assumptions as in (2). Let (S 1 B)0 S 1 B be the universal
first order thickening of S 1 B over A. Note that (S 0 )1 B 0 S 1 B is a surjection
of A-algebras whose kernel has square zero. Hence by definition we obtain a map
(S 1 B)0 (S 0 )1 B 0 compatible with the maps towards S 1 B. Consider any
commutative diagram
/ S 1 B
/ D/I
BO
O
O
A

/A

/D

where I D is an ideal of square zero. Since B 0 is the universal first order

thickening of B over A we obtain an A-algebra map B 0 D. But it is clear that
the image of S 0 in D is mapped to invertible elements of D, and hence we obtain a
compatible map (S 0 )1 B 0 D. Applying this to D = (S 1 B)0 we see that we get
a map (S 0 )1 B 0 (S 1 B)0 . We omit the verification that this map is inverse to
the map described above.


COMMUTATIVE ALGEBRA

04EF

369

Lemma 145.5. Let R A B be ring maps. Assume A B formally

unramified. Let B 0 B be the universal first order thickening of B over A. Then
B 0 is formally unramified over A, and the canonical map A/R A B B 0 /R B 0 B
is an isomorphism.
Proof. We are going to use the construction of B 0 from the proof of Lemma 145.1
although in principle it should be possible to deduce these results formally from
the definition. Namely, we choose a presentation B = P/J, where P = A[xi ] is a
polynomial ring over A. Next, we choose elements fi J such that dfi = dxi 1
in P/A P B. Having made these choices we have B 0 = P/J 0 with J 0 = (fi ) + J 2 ,
see proof of Lemma 145.1.
Consider the canonical exact sequence
J 0 /(J 0 )2 P/A P B 0 B 0 /A 0
see Lemma 130.9. By construction the classes of the fi J 0 map to elements of the
module P/A P B 0 which generate it modulo J 0 /J 2 by construction. Since J 0 /J 2
is a nilpotent ideal, we see that these elements generate the module altogether
(by Nakayamas Lemma 19.1). This proves that B 0 /A = 0 and hence that B 0 is
formally unramified over A, see Lemma 144.2.
L
Since P is a polynomial ring over A we have P/R = A/R A P
P dxi . We
are going to use this decomposition. Consider the following exact sequence
J 0 /(J 0 )2 P/R P B 0 B 0 /R 0
see Lemma 130.9. We may tensor this with B and obtain the exact sequence
J 0 /(J 0 )2 B 0 B P/R P B B 0 /R B 0 B 0
If we remember that J 0 = (fi ) + J 2 then we see that the first arrow annihilates
the submodule J 2 /(J 0 )2 . In terms of the direct sum decomposition P/R P B =
L
A/R A B
Bdxi given we L
see that the submodule (fi )/(J 0 )2 B 0 B maps
isomorphically onto the summand
Bdxi . Hence what is left of this exact sequence
is an isomorphism A/R A B B 0 /R B 0 B as desired.

146. Formally
etale maps

00UP
00UQ

Definition 146.1. Let R S be a ring map. We say S is formally etale over R

if for every commutative solid diagram
SO
R

/ A/I
O
!

/A

where I A is an ideal of square zero, there exists a unique dotted arrow making
the diagram commute.
Clearly a ring map is formally etale if and only if it is both formally smooth and
formally unramified.
00UR

Lemma 146.2. Let R S be a ring map of finite presentation. The following

are equivalent:

COMMUTATIVE ALGEBRA

370

(1) R S is formally etale,

(2) R S is etale.
Proof. Assume that R S is formally etale. Then R S is smooth by Proposition 136.13. By Lemma 144.2 we have S/R = 0. Hence R S is etale by
definition.
Assume that R S is etale. Then R S is formally smooth by Proposition
136.13. By Lemma 144.2 it is formally unramified. Hence R S is formally
etale.

031N

Lemma 146.3. Let R be a ring. Let I be a directed partially ordered set. Let
(Si , ii0 ) be a system of R-algebras over I. If each R Si is formally etale, then
S = colimiI Si is formally etale over R
Proof. Consider a diagram as in Definition 146.1. By assumption we get unique
R-algebra maps Si A lifting the compositions Si S A/I. Hence these are
compatible with the transition maps ii0 and define a lift S A. This proves
existence. The uniqueness is clear by restricting to each Si .


04EG

Lemma 146.4. Let R be a ring. Let S R be any multiplicative subset. Then

the ring map R S 1 R is formally etale.
Proof. Let I A be an ideal of square zero. What we are saying here is that
given a ring map : R A such that (f ) mod I is invertible for all f S we
have also that (f ) is invertible in A for all f S. This is true because A is the
inverse image of (A/I) under the canonical map A A/I.

147. Unramified ring maps

00US

The definition of a G-unramified ring map is the one from EGA. The definition of
an unramified ring map is the one from [Ray70].

00UT

Definition 147.1. Let R S be a ring map.

(1) We say R S is unramified if R S is of finite type and S/R = 0.
(2) We say R S is G-unramified if R S is of finite presentation and
S/R = 0.
(3) Given a prime q of S we say that S is unramified at q if there exists a g S,
g 6 q such that R Sg is unramified.
(4) Given a prime q of S we say that S is G-unramified at q if there exists a
g S, g 6 q such that R Sg is G-unramified.
Of course a G-unramified map is unramified.

00UU

Lemma 147.2. Let R S be a ring map. The following are equivalent

(1) R S is formally unramified and of finite type, and
(2) R S is unramified.
Moreover, also the following are equivalent
(1) R S is formally unramified and of finite presentation, and
(2) R S is G-unramified.
Proof. Follows from Lemma 144.2 and the definitions.

00UV

Lemma 147.3. Properties of unramified and G-unramified ring maps.

COMMUTATIVE ALGEBRA

371

(1) The base change of an unramified ring map is unramified. The base change
of a G-unramified ring map is G-unramified.
(2) The composition of unramified ring maps is unramified. The composition
of G-unramified ring maps is G-unramified.
(3) Any principal localization R Rf is G-unramified and unramified.
(4) If I R is an ideal, then R R/I is unramified. If I R is a finitely
generated ideal, then R R/I is G-unramified.
(5) An etale ring map is G-unramified and unramified.
(6) If R S is of finite type (resp. finite presentation), q S is a prime and
(S/R )q = 0, then R S is unramified (resp. G-unramified) at q.
(7) If R S is of finite type (resp. finite presentation), q S is a prime and
S/R S (q) = 0, then R S is unramified (resp. G-unramified) at q.
(8) If R S is of finite type (resp. finite presentation), q S is a prime lying
over p R and (SR (p)/(p) )q = 0, then R S is unramified (resp.
G-unramified) at q.
(9) If R S is of finite type (resp. presentation), q S is a prime lying over
p R and (SR (p)/(p) ) SR (p) (q) = 0, then R S is unramified
(resp. G-unramified) at q.
(10) If R S is a ring map, g1 , . . . , gm S generate the unit ideal and R
Sgj is unramified (resp. G-unramified) for j = 1, . . . , m, then R S is
unramified (resp. G-unramified).
(11) If R S is a ring map which is unramified (resp. G-unramified) at every
prime of S, then R S is unramified (resp. G-unramified).
(12) If R S is G-unramified, then there exists a finite type Z-algebra R0 and
a G-unramified ring map R0 S0 and a ring map R0 R such that
S = R R0 S0 .
(13) If R S is unramified, then there exists a finite type Z-algebra R0 and an
unramified ring map R0 S0 and a ring map R0 R such that S is a
quotient of R R0 S0 .
Proof. We prove each point, in order.
Ad (1). Follows from Lemmas 130.12 and 13.2.
Ad (2). Follows from Lemmas 130.7 and 13.2.
Ad (3). Follows by direct computation of Rf /R which we omit.
Ad (4). We have (R/I)/R = 0, see Lemma 130.5, and the ring map R R/I is of
finite type. If I is a finitely generated ideal then R R/I is of finite presentation.
Ad (5). See discussion following Definition 141.1.
Ad (6). In this case S/R is a finite S-module (see Lemma 130.16) and hence there
exists a g S, g 6 q such that (S/R )g = 0. By Lemma 130.8 this means that
Sg /R = 0 and hence R Sg is unramified as desired.
Ad (7). Use Nakayamas lemma (Lemma 19.1) to see that the condition is equivalent
to the condition of (6).
Ad (8) & (9). These are equivalent in the same manner that (6) and (7) are
equivalent. Moreover SR (p)/(p) = S/R S (S R (p)) by Lemma 130.12.
Hence we see that (9) is equivalent to (7) since the (q) vector spaces in both are
canonically isomorphic.

COMMUTATIVE ALGEBRA

372

Ad (10). Follows from from Lemmas 23.2 and 130.8.

Ad (11). Follows from (6) and (7) and the fact that the spectrum of S is quasicompact.
Ad (12). Write S = R[x1 , . . . , xn ]/(g1 , . . . , gm ). As S/R = 0 we can write
X
X
dxi =
hij dgj +
aijk gj dxk
in R[x1 ,...,xn ]/R for some hij , aijk R[x1 , . . . , xn ]. Choose a finitely generated Zsubalgebra R0 R containing all the coefficients of the polynomials gi , hij , aijk .
Set S0 = R0 [x1 , . . . , xn ]/(g1 , . . . , gm ). This works.
Ad (13). Write S = R[x1 , . . . , xn ]/I. As S/R = 0 we can write
X
X
0
dxi =
hij dgij +
gik
dxk
0
in R[x1 ,...,xn ]/R for some hij R[x1 , . . . , xn ] and gij , gik
I. Choose a finitely
generated Z-subalgebra R0 R containing all the coefficients of the polynomials
0
0
gij , hij , gik
. Set S0 = R0 [x1 , . . . , xn ]/(gij , gik
). This works.


02FL

Lemma 147.4. Let R S be a ring map. If R S is unramified, then there

exists an idempotent e S R S such that S R S S is isomorphic to S R S
(S R S)e .
Proof. Let J = Ker(S R S S). By assumption J/J 2 = 0, see Lemma 130.13.
Since S is of finite type over R we see that J is finitely generated, namely by
xi 1 1 xi , where xi generate S over R. We win by Lemma 20.5.


00UW

Lemma 147.5. Let R S be a ring map. Let q S be a prime lying over p in

R. If S/R is unramified at q then
(1) we have pSq = qSq is the maximal ideal of the local ring Sq , and
(2) the field extension (p) (q) is finite separable.
Proof. We may first replace S by Sg for some g S, g 6 q and assume that
R S is unramified. The base change S R (p) is unramified over (p) by
Lemma 147.3. By Lemma 138.3 it is smooth hence etale over (p). Hence we see
that S R (p) = (R \ p)1 S/pS is a product of finite separable field extensions of
(p) by Lemma 141.4. This implies the lemma.


02UR

Lemma 147.6. Let R S be a finite type ring map. Let q be a prime of S.

If R S is unramified at q then R S is quasi-finite at q. In particular, an
unramified ring map is quasi-finite.
Proof. An unramified ring map is of finite type. Thus it is clear that the second
statement follows from the first. To see the first statement apply the characterization of Lemma 121.2 part (2) using Lemma 147.5.


02FM

Lemma 147.7. Let R S be a ring map. Let q be a prime of S lying over a

prime p of R. If
(1) R S is of finite type,
(2) pSq is the maximal ideal of the local ring Sq , and
(3) the field extension (p) (q) is finite separable,
then R S is unramified at q.

COMMUTATIVE ALGEBRA

373

Proof. By Lemma 147.3 (8) it suffices to show that SR (p)/(p) is zero when
localized at q. Hence we may replace S by S R (p) and R by (p). In other
words, we may assume that R = k is a field and S is a finite type k-algebra. In this
case the hypotheses imply that Sq
= (q) and hence S = (q) S 0 (see Lemma
121.1). Hence (S/k )q = (q)/k which is zero as desired.

0395

Proposition 147.8. Let R S be a ring map. Let q S be a prime. If R S

is unramified at q, then there exist
(1) a g S, g 6 q,
(2) a standard etale ring map R S 0 , and
(3) a surjective R-algebra map S 0 Sg .
Proof. This proof is the same as the proof of Proposition 141.17. The proof is
a little roundabout and there may be ways to shorten it.
Step 1. By Definition 147.1 there exists a g S, g 6 q such that R Sg is
unramified. Thus we may assume that S is unramified over R.
Step 2. By Lemma 147.3 there exists an unramified ring map R0 S0 with
R0 of finite type over Z, and a ring map R0 R such that S is a quotient of
R R0 S0 . Denote q0 the prime of S0 corresponding to q. If we show the result for
(R0 S0 , q0 ) then the result follows for (R S, q) by base change. Hence we may
assume that R is Noetherian.
Step 3. Note that R S is quasi-finite by Lemma 147.6. By Lemma 122.15 there
exists a finite ring map R S 0 , an R-algebra map S 0 S, an element g 0 S 0
such that g 0 6 q such that S 0 S induces an isomorphism Sg0 0
= Sg0 . (Note that
S 0 may not unramified over R.) Thus we may assume that (a) R is Noetherian,
(b) R S is finite and (c) R S is unramified at q (but no longer necessarily
unramified at all primes).
Step 4. Let p R be the prime corresponding to q. Consider the fibre ring
S R (p). This is a finite algebra over (p). Hence it is Artinian (see Lemma 52.2)
and so a finite product of local rings
Yn
S R (p) =
Ai
i=1

see Proposition 59.6. One of the factors, say A1 , is the local ring Sq /pSq which is
isomorphic to (q), see Lemma 147.5. The other factors correspond to the other
primes, say q2 , . . . , qn of S lying over p.
Step 5. We may choose a nonzero element (q) which generates the finite
separable field extension (p) (q) (so even if the field extension is trivial we do
not allow = 0). Note that for any (p) the element also generates (q)
over (p). Consider the element
Yn
t = (, 0, . . . , 0)
Ai = S R (p).
i=1

After possibly replacing by as above we may assume that t is the image of

t S. Let I R[x] be the kernel of the R-algebra map R[x] S which maps x to

COMMUTATIVE ALGEBRA

374

t. Set S 0 = R[x]/I, so S 0 S. Here is a diagram

R[x]
O

/ S0
=

6/ S

R
By construction the primes qj , j 2 of S all lie over the prime (p, x) of R[x],
whereas the prime q lies over a different prime of R[x] because 6= 0.
Step 6. Denote q0 S 0 the prime of S 0 corresponding to q. By the above q is
the only prime of S lying over q0 . Thus we see that Sq = Sq0 , see Lemma 40.11
(we have going up for S 0 S by Lemma 35.20 since S 0 S is finite as R S
is finite). It follows that Sq0 0 Sq is finite and injective as the localization of the
finite injective ring map S 0 S. Consider the maps of local rings
Rp Sq0 0 Sq
The second map is finite and injective. We have Sq /pSq = (q), see Lemma 147.5.
Hence a fortiori Sq /q0 Sq = (q). Since
(p) (q0 ) (q)
and since is in the image of (q0 ) in (q) we conclude that (q0 ) = (q). Hence
by Nakayamas Lemma 19.1 applied to the Sq0 0 -module map Sq0 0 Sq , the map
Sq0 0 Sq is surjective. In other words, Sq0 0
= Sq .
Step 7. By Lemma 125.7 there exist g S, g 6 q and g 0 S 0 , g 0 6 q0 such that
Sg0 0
= Sg . As R is Noetherian the ring S 0 is finite over R because it is an Rsubmodule of the finite R-module S. Hence after replacing S by S 0 we may assume
that (a) R is Noetherian, (b) S finite over R, (c) S is unramified over R at q, and
(d) S = R[x]/I.
Step 8. Consider the ring S R (p) = (p)[x]/I where I = I (p)[x] is the ideal
generated by I in (p)[x]. As (p)[x] is a PID we know that I = (h) for some monic
h (p). After replacing h by h for some (p) we may assume that h is the
image of some h R[x]. (The problem is that we do not know if we may choose h
monic.) Also, as in Step 4 we know that S R (p) = A1 . . . An with A1 = (q)
a finite separable extension of (p) and A2 , . . . , An local. This implies that
e2

en

h = h1 h2 . . . hn

for certain pairwise coprime irreducible monic polynomials hi (p)[x] and certain
ei
e2 , . . . , en 1. Here the numbering is chosen so that Ai = (p)[x]/(hi ) as (p)[x]algebras. Note that h1 is the minimal polynomial of (q) and hence is a
separable polynomial (its derivative is prime to itself).
Step 9. Let m I be a monic element; such an element exists because the ring
extension R R[x]/I is finite hence integral. Denote m the image in (p)[x]. We
may factor
d1 d2

dn

m = kh1 h2 . . . hn

for some d1 1, dj ej , j = 2, . . . , n and k (p)[x] prime to all the hi . Set

f = ml + h where l deg(m) > deg(h), and l 2. Then f is monic as a polynomial

COMMUTATIVE ALGEBRA

375

over R. Also, the image f of f in (p)[x] factors as

e2

en

l ld1 ld2

ldn

e2

en

l ld1 1 ld2
h2

f = h1 h2 . . . hn +k h1 h2 . . . hn = h1 (h2 . . . hn +k h1

ldn

. . . hn ) = h1 w

with w a polynomial relatively prime to h1 . Set g = f 0 (the derivative with respect

to x).
Step 10. The ring map R[x] S = R[x]/I has the properties: (1) it maps f to
zero, and (2) it maps g to an element of S \ q. The first assertion is clear since f
is an element of I. For the second assertion we just have to show that g does not
map to zero in (q) = (p)[x]/(h1 ). The image of g in (p)[x] is the derivative of
f . Thus (2) is clear because
df
dh1
dw
=w
+ h1
,
dx
dx
dx
w is prime to h1 and h1 is separable.
g=

Step 11. We conclude that : R[x]/(f ) S is a surjective ring map, R[x]g /(f ) is
etale over R (because it is standard etale, see Lemma 141.15) and (g) 6 q. Thus
the map (R[x]/(f ))g S(g) is the desired surjection.

00UX

Lemma 147.9. Let R S be a ring map. Let q be a prime of S lying over p R.

Assume that R S is of finite type and unramified at q. Then there exist
(1) an etale ring map R R0 ,
(2) a prime p0 R0 lying over p.
(3) a product decomposition
R 0 R S = A B
with the following properties
(1) R0 A is surjective, and
(2) p0 A is a prime of A lying over p0 and over q.
Proof. We may replace (R S, p, q) with any base change (R0 R0 R S, p0 , q0 )
by a etale ring map R R0 with a prime p0 lying over p, and a choice of q0 lying
over both q and p0 . Note also that given R R0 and p0 a suitable q0 can always be
found.
The assumption that R S is of finite type means that we may apply Lemma
141.24. Thus we may assume that S = A1 . . .An B, that each R Ai is finite
with exactly one prime ri lying over p such that (p) (ri ) is purely inseparable
and that R B is not quasi-finite at any prime lying over p. Then clearly q = ri
for some i, since an unramified morphism is quasi-finite (see Lemma 147.6). Say
q = r1 . By Lemma 147.5 we see that (p) (r1 ) is separable hence the trivial field
extension, and that p(A1 )r1 is the maximal ideal. Also, by Lemma 40.11 (which
applies to R A1 because a finite ring map satisfies going up by Lemma 35.20)
we have (A1 )r1 = (A1 )p . It follows from Nakayamas Lemma 19.1 that the map of
local rings Rp (A1 )p = (A1 )r1 is surjective. Since A1 is finite over R we see that
there exists a f R, f 6 p such that Rf (A1 )f is surjective. After replacing R
by Rf we win.


00UY

Lemma 147.10. Let R S be a ring map. Let p be a prime of R. If R S is

unramified then there exist
(1) an etale ring map R R0 ,

COMMUTATIVE ALGEBRA

376

(2) a prime p0 R0 lying over p.

(3) a product decomposition
R 0 R S = A1 . . . An B
with the following properties
(1) R0 Ai is surjective,
(2) p0 Ai is a prime of Ai lying over p0 , and
(3) there is no prime of B lying over p0 .
Proof. We may apply Lemma 141.24. Thus, after an etale base change, we may
assume that S = A1 . . . An B, that each R Ai is finite with exactly one
prime ri lying over p such that (p) (ri ) is purely inseparable, and that R B
is not quasi-finite at any prime lying over p. Since R S is quasi-finite (see Lemma
147.6) we see there is no prime of B lying over p. By Lemma 147.5 we see that
(p) (ri ) is separable hence the trivial field extension, and that p(Ai )ri is the
maximal ideal. Also, by Lemma 40.11 (which applies to R Ai because a finite
ring map satisfies going up by Lemma 35.20) we have (Ai )ri = (Ai )p . It follows
from Nakayamas Lemma 19.1 that the map of local rings Rp (Ai )p = (Ai )ri is
surjective. Since Ai is finite over R we see that there exists a f R, f 6 p such
that Rf (Ai )f is surjective. After replacing R by Rf we win.

148. Henselian local rings
04GE

In this section we discuss a bit the notion of a henselian local ring. Let (R, m, ) be
a local ring. For a R we denote a the image of a in . For a polynomial f R[T ]
we often denote f the image of f in [T ]. Given a polynomial f R[T ] we denote
0
f 0 the derivative of f with respect to T . Note that f = f 0 .

04GF

Definition 148.1. Let (R, m, ) be a local ring.

(1) We say R is henselian if for every monic f R[T ] and every root a0
of f such that f 0 (a0 ) 6= 0 there exists an a R such that f (a) = 0 and
a0 = a.
(2) We say R is strictly henselian if R is henselian and its residue field is
separably algebraically closed.
Note that the condition f 0 (a0 ) 6= 0 is equivalent to the condition that a0 is a simple
root of the polynomial f . In fact, it implies that the lift a R, if it exists, is unique.

06RR

Lemma 148.2. Let (R, m, ) be a local ring. Let f R[T ]. Let a, b R such that
f (a) = f (b) = 0, a = b mod m, and f 0 (a) 6 m. Then a = b.
Proof. Write f (x + y) f (x) = f 0 (x)y + g(x, y)y 2 in R[x, y] (this is possible as one
sees by expanding f (x + y); details omitted). Then we see that 0 = f (b) f (a) =
f (a + (b a)) f (a) = f 0 (a)(b a) + c(b a)2 for some c R. By assumption f 0 (a)
is a unit in R. Hence (b a)(1 + f 0 (a)1 c(b a)) = 0. By assumption b a m,
hence 1 + f 0 (a)1 c(b a) is a unit in R. Hence b a = 0 in R.

Here is the characterization of henselian local rings.

04GG

Lemma 148.3. Let (R, m, ) be a local ring. The following are equivalent
(1) R is henselian,
(2) for every f R[T ] and every root a0 of f such that f 0 (a0 ) 6= 0 there
exists an a R such that f (a) = 0 and a0 = a,

COMMUTATIVE ALGEBRA

377

(3) for any monic f R[T ] and any factorization f = g0 h0 with gcd(g0 , h0 ) = 1
there exists a factorization f = gh in R[T ] such that g0 = g and h0 = h,
(4) for any monic f R[T ] and any factorization f = g0 h0 with gcd(g0 , h0 ) = 1
there exists a factorization f = gh in R[T ] such that g0 = g and h0 = h
and moreover degT (g) = degT (g0 ),
(5) for any f R[T ] and any factorization f = g0 h0 with gcd(g0 , h0 ) = 1 there
exists a factorization f = gh in R[T ] such that g0 = g and h0 = h,
(6) for any f R[T ] and any factorization f = g0 h0 with gcd(g0 , h0 ) = 1 there
exists a factorization f = gh in R[T ] such that g0 = g and h0 = h and
moreover degT (g) = degT (g0 ),
(7) for any etale ring map R S and prime q of S lying over m with = (q)
there exists a section : S R of R S,
(8) for any etale ring map R S and prime q of S lying over m with = (q)
there exists a section : S R of R S with q = 1 (m),
(9) any finite R-algebra is a product of local rings,
(10) any finite R-algebra is a finite product of local rings,
(11) any finite type R-algebra S can be written as A B with R A finite and
R B not quasi-finite at any prime lying over m,
(12) any finite type R-algebra S can be written as A B with R A finite such
that each irreducible component of Spec(B R ) has dimension 1, and
(13) any quasi-finite R-algebra S can be written as S = A B with R A finite
such that B R = 0.
Proof. Here is a list of the easier implications:
21
53
64
43
65
87
109
1112
1113

because in (2) we consider all polynomials and in (1) only monic ones,
because in (5) we consider all polynomials and in (3) only monic ones,
because in (6) we consider all polynomials and in (4) only monic ones,
is obvious,
is obvious,
is obvious,
is obvious,
by definition of being quasi-finite at a prime,
by definition of being quasi-finite,

Proof of 18. Assume (1). Let R S be etale, and let q S be a prime ideal such
that (q)
= . By Proposition 141.17 we can find a g S, g 6 q such that R Sg
is standard etale. After replacing S by Sg we may assume that S = R[t]g /(f ) is
standard etale. Since the prime q has residue field it corresponds to a root a0 of
f which is not a root of g. By definition of a standard etale algebra this also means
that f 0 (a0 ) 6= 0. Since also f is monic by definition of a standard etale algebra
again we may use that R is henselian to conclude that there exists an a R with
a0 = a such that f (a) = 0. This implies that g(a) is a unit of R and we obtain the
desired map : S = R[t]g /(f ) R by the rule t 7 a. By construction 1 (q) = m.
This proves (8) holds.
Proof of 78. (This is really unimportant and should be skipped.) Assume (7)
holds and assume R S is etale. Let q1 , . . . , qr be the other primes of S lying
over m. Then we can find a g S, g 6 q and g qi for i = 1, . . . , r, see Lemma
14.2. Apply (7) to the etale ring map R Sg and the prime qSg . This gives a

COMMUTATIVE ALGEBRA

378

section g : Sg R such that the composition : S Sg R has the property

1 (q) = m. Minor details omitted.
Proof of 811. Assume (8) and let R S be a finite type ring map. Apply Lemma
141.23. We find an etale ring map R R0 and a prime m0 R0 lying over m with
= (m0 ) such that R0 R S = A0 B 0 with A0 finite over R0 and B 0 not quasi-finite
over R0 at any prime lying over m0 . Apply (8) to get a section : R0 R with
m = 1 (m0 ). Then use that
S = (S R R0 ) R0 , R = (A0 B 0 ) R0 , R = (A0 R0 , R) (B 0 R0 , R)
which gives a decomposition as in (11).
Proof of 810. Assume (8) and let R S be a finite ring map. Apply Lemma
141.23. We find an etale ring map R R0 and a prime m0 R0 lying over m with
= (m0 ) such that R0 R S = A01 . . . A0n B 0 with A0i finite over R0 having
exactly one prime over m0 and B 0 not quasi-finite over R0 at any prime lying over
m0 . Apply (8) to get a section : R0 R with m = 1 (m0 ). Then we obtain
S = (S R R0 ) R0 , R
= (A01 . . . A0n B 0 ) R0 , R
= (A01 R0 , R) . . . (A01 R0 , R) (B 0 R0 , R)
= A1 . . . An B
The factor B is finite over R but R B is not quasi-finite at any prime lying over
m. Hence B = 0. The factors Ai are finite R-algebras having exactly one prime
lying over m, hence they are local rings. This proves that S is a finite product of
local rings.
Proof of 910. This holds because if S is finite over the local ring R, then it has at
most finitely many maximal ideals. Namely, by going up for R S the maximal
ideals of S all lie over m, and S/mS is Artinian hence has finitely many primes.
Proof of 101. Assume (10). Let f R[T ] be a monic polynomial and a0
a simple root of f . Then S = R[T ]/(f ) is a finite R-algebra. Applying (10)
we get S = A1 . . . QAr is a finite product of local R-algebras. In particular
we see that S/mS =
Ai /mAi is the decomposition of [T ]/(f ) as a product
of local rings. This means that one of the factors, say A1 /mA1 is the quotient
[T ]/(f ) [T ]/(T a0 ). Since A1 is a summand of the finite free R-module S it
is a finite free R-module itself. As A1 /mA1 is a -vector space of dimension 1 we see
that A1
= R as an R-module. Clearly this means that R A1 is an isomorphism.
Let a R be the image of T under the map R[T ] S A1 R. Then f (a) = 0
and a = a0 as desired.
Proof of 131. Assume (13). Let f R[T ] be a monic polynomial and a0 a
simple root of f . Then S1 = R[T ]/(f ) is a finite R-algebra. Let g R[T ] be any
element such that g = f /(T a0 ). Then S = (S1 )g is a quasi-finite R-algebra such
that S R
= [T ]g /(f )
= [T ]/(T a0 )
= . Applying (13) to S we get S = AB
with A finite over R and B R = 0. In particular we see that
= S/mS = A/mA.
Since A is a summand of the flat R-algebra S we see that it is finite flat, hence
free over R. As A/mA is a -vector space of dimension 1 we see that A
= R as an
R-module. Clearly this means that R A is an isomorphism. Let a R be the

COMMUTATIVE ALGEBRA

379

image of T under the map R[T ] S A R. Then f (a) = 0 and a = a0 as

desired.
Proof of 82. Assume (8). Let f R[T ] be any polynomial and let a0 be a
simple root. Then the algebra S = R[T ]f 0 /(f ) is etale over R. Let q S be the
prime generated by m and T b where b R is any element such that b = a0 .
Apply (8) to S and q to get : S R. Then the image (T ) = a R works in
(2).
At this point we see that (1), (2), (7), (8), (9), (10), (11), (12), (13) are all equivalent. The weakest assertion of (3), (4), (5) and (6) is (3) and the strongest is (6).
Hence we still have to prove that (3) implies (1) and (1) implies (6).
Proof of 31. Assume (3). Let f R[T ] be monic and let a0 be a simple root of
f . This gives a factorization f = (T a0 )h0 with h0 (a0 ) 6= 0, so gcd(T a0 , h0 ) = 1.
Apply (3) to get a factorization f = gh with g = T a0 and h = h0 . Set S =
R[T ]/(f ) which is a finite free R-algebra. We will write g, h also for the images of
g and h in S. Then gS + hS = S by Nakayamas Lemma 19.1 as the equality holds
modulo m. Since gh = f = 0 in S this also implies that gS hS = 0. Hence by
the Chinese Remainder theorem we obtain S = S/(g) S/(h). This implies that
A = S/(g) is a summand of a finite free R-module, hence finite free. Moreover, the
rank of A is 1 as A/mA = [T ]/(T a0 ). Thus the map R A is an isomorphism.
Setting a R equal to the image of T under the maps R[T ] S A R gives
an element of R with f (a) = 0 and a = a0 .
Proof of 16. Assume (1) or equivalently all of (1), (2), (7), (8), (9), (10), (11), (12),
(13). Let f R[T ] be a polynomial. Suppose that f = g0 h0 is a factorization with
gcd(g0 , h0 ) = 1. We may and do assume that g0 is monic. Consider S = R[T ]/(f ).
Because we have the factorization we see that the coefficients of f generate the unit
ideal in R. This implies that S has finite fibres over R, hence is quasi-finite over
R. It also implies that S is flat over R by Lemma 98.2. Combining (13) and (10)
we may write S = A1 . . . An B where each Ai is local and finite over R, and
B R = 0. After reordering the factors A1 , . . . , An we may assume that
[T ]/(g0 ) = A1 /mA1 . . . Ar /mAr , [T ]/(h0 ) = Ar+1 /mAr+1 . . . An /mAn
as quotients of [T ]. The finite flat R-algebra A = A1 . . . Ar is free as an Rmodule, see Lemma 77.4. Its rank is degT (g0 ). Let g R[T ] be the characteristic
polynomial of the R-linear operator T : A A. Then g is a monic polynomial of
degree degT (g) = degT (g0 ) and moreover g = g0 . By Cayley-Hamilton (Lemma
15.1) we see that g(TA ) = 0 where TA indicates the image of T in A. Hence
we obtain a well defined surjective map R[T ]/(g) A which is an isomorphism
by Nakayamas Lemma 19.1. The map R[T ] A factors through R[T ]/(f ) by
construction hence we may write f = gh for some h. This finishes the proof.

04GH

Lemma 148.4. Let (R, m, ) be a henselian local ring.

(1) If R S is a finite ring extension then S is a finite product of henselian
local rings.
(2) If R S is a finite local homomorphism of local rings, then S is a henselian
local ring.
(3) If R S is a finite type ring map, and q is a prime of S lying over m at
which R S is quasi-finite, then Sq is henselian.

COMMUTATIVE ALGEBRA

380

(4) If R S is quasi-finite then Sq is henselian for every prime q lying over

m.
Proof. Part (2) implies part (1) since S as in part (1) is a finite product of its
localizations at the primes lying over m. Part (2) follows from Lemma 148.3 part
(10) since any finite S-algebra is also a finite R-algebra. If R S and q are as in
(3), then Sq is a local ring of a finite R-algebra by Lemma 148.3 part (11). Hence
(3) follows from (1). Part (4) follows from part (3).

04GJ

Lemma 148.5. Let (R, m, ) be a henselian local ring. Any finite type R-algebra
S can be written as S = A1 . . . An B with Ai local and finite over R and
R B not quasi-finite at any prime of B lying over m.
Proof. This is a combination of parts (11) and (10) of Lemma 148.3.

06DD

Lemma 148.6. Let (R, m, ) be a strictly henselian local ring. Any finite type
R-algebra S can be written as S = A1 . . . An B with Ai local and finite over R
and (mAi ) finite purely inseparable and R B not quasi-finite at any prime
of B lying over m.
Proof. First write S = A1 . . . An B as in Lemma 148.5. The field extension
(mAi ) is finite and is separably algebraically closed, hence it is finite purely
inseparable.


04GK

Lemma 148.7. Let (R, m, ) be a henselian local ring. The category of finite etale
ring extensions R S is equivalent to the category of finite etale algebras S
via the functor S 7 S/mS.
Proof. Denote C D the functor of categories of the statement. Suppose that
R S is finite etale. Then we may write
S = A1 . . . An
with Ai local and finite etale over S, use either Lemma 148.5 or Lemma 148.3 part
(10). In particular Ai /mAi is a finite separable field extension of , see Lemma
141.5. Thus we see that every object of C and D decomposes canonically into
irreducible pieces which correspond via the given functor. Next, suppose that S1 ,
S2 are finite etale over R such that 1 = S1 /mS1 and 2 = S2 /mS2 are fields (finite
separable over ). Then S1 R S2 is finite etale over R and we may write
S1 R S2 = A1 . . . An
as before. Then we see that HomR (S1 , S2 ) is identified with the set of indices
i {1, . . . , n} such that S2 Ai is an isomorphism. To see this use that given any
R-algebra map : S1 S2 the map 1 : S1 R S2 S2 is surjective, and hence
is equal to projection onto one of the factors Ai . But in exactly the same way we
see that Hom (1 , 2 ) is identified with the set of indices i {1, . . . , n} such that
2 Ai /mAi is an isomorphism. By the discussion above these sets of indices
match, and we conclude that our functor is fully faithful. Finally, let 0 be a
finite separable field extension. By Lemma 141.16 there exists an etale ring map
R S and a prime q of S lying over m such that (q) is isomorphic to the
given extension. By part (1) we may write S = A1 . . . An B. Since R S
is quasi-finite we see that there exists no prime of B over m. Hence Sq is equal to
Ai for some i. Hence R Ai is finite etale and produces the given residue field
extension. Thus the functor is essentially surjective and we win.


COMMUTATIVE ALGEBRA

04GL

381

Lemma 148.8. Let (R, m, ) be a strictly henselian local ring. Let R S be an

unramified ring map. Then
S = A1 . . . An B
with each R Ai surjective and no prime of B lying over m.
Proof. First write S = A1 . . . An B as in Lemma 148.5. Now we see
that R Ai is finite unramified and Ai local. Hence the maximal ideal of Ai is
mAi and its residue field Ai /mAi is a finite separable extension of , see Lemma
147.5. However, the condition that R is strictly henselian means that is separably
algebraically closed, so = Ai /mAi . By Nakayamas Lemma 19.1 we conclude that
R Ai is surjective as desired.


04GM

Lemma 148.9. Let (R, m, ) be a complete local ring, see Definition 154.1. Then
R is henselian.
Proof. Let f R[T ] be monic. Denote fn R/mn+1 [T ] the image. Denote fn0
the derivative of fn with respect to T . Let a0 be a simple root of f0 . We lift
this to a solution of f over R inductively as follows: Suppose given an R/mn+1
such that an mod m = a0 and fn (an ) = 0. Pick any element b R/mn+2 such that
an = b mod mn+1 . Then fn+1 (b) mn+1 /mn+2 . Set
0
an+1 = b fn+1 (b)/fn+1
(b)
0
(Newtons method). This makes sense as fn+1
(b) R/mn+1 is invertible by the
condition on a0 . Then we compute fn+1 (an+1 ) = fn+1 (b)fn+1 (b) = 0 in R/mn+2 .
Since the system of elements an R/mn+1 so constructed is compatible we get an
element a lim R/mn = R (here we use that R is complete). Moreover, f (a) = 0

since it maps to zero in each R/mn . Finally a = a0 and we win.

06RS

Lemma 148.10. Let (R, m) be a local ring of dimension 0. Then R is henselian.

Proof. Let R S be a finite ring map. By Lemma 148.3 it suffices to show that S
is a product of local rings. By Lemma 35.19 S has finitely many primes m1 , . . . , mr
which all lie over m. There are no inclusions among these primes, see Lemma 35.18,
hence they are all maximal. Every element of m1 . . . mr is nilpotent by Lemma
16.2. It follows S is the product of the localizations of S at the primes mi by Lemma
52.5.

The following lemma will be the key to the uniqueness and functorial properties of
henselization and strict henselization.

08HQ

Lemma 148.11. Let R S be a ring map with S henselian local. Given

(1) an etale ring map R A,
(2) a prime q of A lying over p = R mS ,
(3) a (p)-algebra map : (q) S/mS ,
then there exists a unique homomorphism of R-algebras f : A S such that
q = f 1 (mS ) and f mod q = .
Proof. Consider A R S. This is an etale algebra over S, see Lemma 141.3. Moreover, the kernel
q0 = Ker(A R S (q) (p) (mS ) (mS ))

COMMUTATIVE ALGEBRA

382

of the map using the map given in (3) is a prime ideal lying over mS with residue
field equal to the residue field of S. Hence by Lemma 148.3 there exists a unique
splitting : A R S S with 1 (mS ) = q0 . Set f equal to the composition
A A R S S.

04GX

Lemma 148.12. Let : R S be a local homomorphism of strictly henselian local

rings. Let P1 , . . . , Pn R[x1 , . . . , xn ] be polynomials such that R[x1 , . . . , xn ]/(P1 , . . . , Pn )
is etale over R. Then the map
Rn S n ,

(h1 , . . . , hn ) 7 ((h1 ), . . . , (hn ))

induces a bijection between

{(r1 , . . . , rn ) Rn | Pi (r1 , . . . , rn ) = 0, i = 1, . . . , n}
and
where

Pi0

{(s1 , . . . , sn ) S n | Pi0 (s1 , . . . , sn ) = 0, i = 1, . . . , n}

S[x1 , . . . , xn ] are the images of the Pi under .

Proof. The first solution set is canonically isomorphic to the set

HomR (R[x1 , . . . , xn ]/(P1 , . . . , Pn ), R).
As R is henselian the map R R/mR induces a bijection between this set and
the set of solutions in the residue field R/mR , see Lemma 148.3. The same is true
for S. Now since R[x1 , . . . , xn ]/(P1 , . . . , Pn ) is etale over R and R/mR is separably
algebraically closed we see that R/mR [x1 , . . . , xn ]/(P1 , . . . , Pn ) is a finite product
of copies of R/mR . Hence the tensor product
R/mR [x1 , . . . , xn ]/(P1 , . . . , Pn ) R/mR S/mS = S/mS [x1 , . . . , xn ]/(P10 , . . . , Pn0 )
is also a finite product of copies of S/mS with the same index set. This proves the
lemma.

05D6

Lemma 148.13. Let R be a henselian local ring. Any countably generated MittagLeffler module over R is a direct sum of finitely presented R-modules.
Proof. Let M be a countably generated and Mittag-Leffler R-module. We claim
that for any element x M there exists a direct sum decomposition M = N K
with x N , the module N finitely presented, and K Mittag-Leffler.
Suppose the claim is true. Choose generators x1 , x2 , x3 , . . . of M . By the claim we
can inductively find direct sum decompositions
M = N1 N2 . . . Nn Kn
with Ni finitely presented, x1 , . . . , xn L
N1 . . . Nn , and Kn Mittag-Leffler.
Repeating ad infinitum we see that M =
Ni .
We still have to prove the claim. Let x M . By Lemma 91.2 there exists an
endomorphism : M M such that factors through a finitely presented module,
and (x) = x. Say factors as
M

/P

/M

Set a = i : P P , so i a = 3 . By Lemma 15.2 there exists a monic

polynomial P R[T ] such that P (a) = 0. Note that this implies formally that
2 P () = 0. Hence we may think of M as a module over R[T ]/(T 2 P ). Assume
that x 6= 0. Then (x) = x implies that 0 = 2 P ()x = P (1)x hence P (1) = 0 in

COMMUTATIVE ALGEBRA

383

R/I where I = {r R | rx = 0} is the annihilator of x. As x 6= 0 we see I mR ,

hence 1 is a root of P = P mod mR R/mR [T ]. As R is henselian we can find a
factorization
T 2 P = (T 2 Q1 )Q2
for some Q1 , Q2 R[T ] with Q2 = (T 1)e mod mR R[T ] and Q1 (1) 6= 0 mod mR ,
see Lemma 148.3. Let N = Im(2 Q1 () : M M ) and K = Im(Q2 () : M M ).
As T 2 Q1 and Q2 generate the unit ideal of R[T ] we get a direct sum decomposition
M = N K. Moreover, Q2 acts as zero on N and T 2 Q1 acts as zero on K.
Note that N is a quotient of P hence is finitely generated. Also x N because
2 Q1 ()x = Q1 (1)x and Q1 (1) is a unit in R. By Lemma 88.9 the modules N and
K are Mittag-Leffler. Finally, the finitely generated module N is finitely presented
as a finitely generated Mittag-Leffler module is finitely presented, see Example 90.1
part (1).

149. Filtered colimits of
etale ring maps
0BSG

This section is a precursor to the section on ind-etale ring maps (Pro-etale Cohomology, Section 7). The material will also be useful to prove uniqueness properties
of the henselization and strict henselization of a local ring.

0BSH

Lemma 149.1. Let R A and R R0 be ring maps. If A is a filtered colimit

of etale ring maps, then so is R0 R0 R A.
Proof. This is true because colimits commute with tensor products and etale ring
maps are preserved under base change (Lemma 141.3).


0BSI

Lemma 149.2. Let A B C be ring maps. If A B is a filtered colimit of

etale ring maps and B C is a filtered colimit of etale ring maps, then A C is
a filtered colimit of etale ring maps.
Proof. Write B = colimiI Bi where I is a directed partially ordered set and
where Bi is an etale A-algebra. Write C = colimjJ Cj where J is a directed
partially ordered set and where Cj is an etale B-algebra. By Lemma 141.3 for
each j J we can find an i(j) I and an etale ring map Bi(j) Cj0 such that
Cj = B Bi(j) Cj0 . Then A Cj0 is etale as compositions of etale ring maps are
etale. Then C = colim Cj0 and the proof is complete.


0BSJ

Lemma 149.3. Let R be a ring. Let A = colim Ai be a filtered colimit of Ralgebras such that each Ai is a filtered colimit of etale R-algebras. Then A is a
filtered colimit of etale R-algebras.
Proof. Write Ai = colimjJi Aj where Ji is a directed partially ordered set and
Aj is an etale R-algebra. For each i i0 and j Ji there exists an j 0 Ji0 and an
R-algebra map jj 0 : Aj Aj 0 making the diagram
/ Ai 0
O

AO i
Aj

jj 0

/ Aj 0

commute. This is true because R Aj is of finite`presentation so that Lemma

126.2 applies. Let J be the category with objects iI Ji and morphisms triples

COMMUTATIVE ALGEBRA

384

(j, j 0 , jj 0 ) as above (and obvious composition law). Then J is a filtered category

and A = colimJ Aj . Details omitted.

08HS

Lemma 149.4. Let R be a ring. Let A B be an R-algebra homomorphism. If

A and B are filtered colimits of etale R-algebras, then B is a filtered colimit of etale
A-algebras.
Proof. Write A = colim Ai and B = colim Bj as filtered colimits with Ai and Bj
etale over R. For each i we can find a j such that Ai B factors through Bj ,
see Lemma 126.2. The factorization Ai Bj is etale by Lemma 141.9. Since
A A Ai Bj is etale (Lemma 141.3) it suffices to prove that B = colim A Ai Bj
where the colimit is over pairs (i, j) and factorizations Ai Bj B of Ai B
(this is a directed system; details omitted). This is clear because colimits commute
with tensor products and hence colim A Ai Bj = A A B = B.


08HR

Lemma 149.5. Let R S be a ring map with S henselian local. Given

(1) an R-algebra A which is a filtered colimit of etale R-algebras,
(2) a prime q of A lying over p = R mS ,
(3) a (p)-algebra map : (q) S/mS ,
then there exists a unique homomorphism of R-algebras f : A S such that
q = f 1 (mS ) and f mod q = .
Proof. Write A = colim Ai as a filtered colimit of etale R-algebras. Set qi = Ai q.
We obtain fi : Ai S by applying Lemma 148.11. Set f = colim fi .


08HT

Lemma 149.6. Let R be a ring. Given a commutative diagram of ring maps

SO

/K
O

/ S0

where S, S 0 are henselian local, S, S 0 are filtered colimits of etale R-algebras, K is

a field and the arrows S K and S 0 K identify K with the residue field of both
S and S 0 . Then there exists an unique R-algebra isomorphism S S 0 compatible
with the maps to K.
Proof. Follows immediately from Lemma 149.5.

The following lemma is not strictly speaking about colimits of etale ring maps.
04GI

Lemma 149.7. A filtered colimit of henselian local rings along local homomorphisms is henselian.
Proof. Categories, Lemma 21.3 says that this is really just a question about a
colimit of henselian local rings over a directed partially ordered set. Let (Ri , ii0 )
be such a system with each ii0 local. Then R = colimi Ri is local, and its residue
field is colim i (argument omitted). Suppose that f R[T ] is monic and that
a0 is a simple root of f . Then for some large enough i there exists an fi Ri [T ]
mapping to f and an a0,i i mapping to a0 . Since fi (a0,i ) i , resp. fi0 (a0,i ) i
maps to 0 = f (a0 ) , resp. 0 6= f 0 (a0 ) we conclude that a0,i is a simple root
of fi . As Ri is henselian we can find ai Ri such that fi (ai ) = 0 and a0,i = ai .
Then the image a R of ai is the desired solution. Thus R is henselian.


COMMUTATIVE ALGEBRA

385

150. Henselization and strict henselization

0BSK

In this section we construct the henselization. We encourge the reader to keep in

mind the uniquess already proved in Lemma 149.6 and the functorial behaviour
pointed out in Lemma 149.5 while reading this material.

04GN

Lemma 150.1. Let (R, m, ) be a local ring. There exists a local ring map R Rh
with the following properties
(1) Rh is henselian,
(2) Rh is a filtered colimit of etale R-algebras,
(3) mRh is the maximal ideal of Rh , and
(4) = Rh /mRh .
Proof. Consider the category of pairs (S, q) where R S is an etale ring map, and
q is a prime of S lying over m with = (q). A morphism of pairs (S, q) (S 0 , q0 )
is given by an R-algebra map : S S 0 such that 1 (q0 ) = q. We set
Rh = colim(S,q) S.
Let us show that the category of pairs is filtered, see Categories, Definition 19.1.
The category contains the pair (R, m) and hence is not empty, which proves part
(1) of Categories, Definition 19.1. Note that for any pair (S, q) the prime ideal q
is maximal, for example since S/q (q) are isomorphisms. Suppose that
(S, q) and (S 0 , q0 ) are two objects. Set S 00 = S R S 0 and q00 = qS 00 + q0 S 00 . Then
S 00 /q00 = S/q R S 0 /q0 = by what we said above. Moreover, R S 00 is etale by
Lemma 141.3. This proves part (2) of Categories, Definition 19.1. Next, suppose
that , : (S, q) (S 0 , q0 ) are two morphisms of pairs. Consider
S 00 = (S 0 ,S, S 0 ) S 0 R S 0 S 0
with prime ideal
q00 = (q0 S 0 + S 0 q0 ) S 0 + (S 0 ,S, S 0 ) q0
Arguing as above (base change of etale maps is etale, composition of etale maps is
etale) we see that S 00 is etale over R. Moreover, the canonical map S 0 S 00 (using
the right most factor for example) equalizes and . This proves part (3) of
Categories, Definition 19.1. Hence we conclude that Rh consists of triples (S, q, f )
with f S, and two such triples (S, q, f ), (S 0 , q0 , f 0 ) define the same element of Rh if
and only if there exists a pair (S 00 , q00 ) and morphisms of pairs : (S, q) (S 00 , q00 )
and 0 : (S 0 , q0 ) (S 00 , q00 ) such that (f ) = 0 (f 0 ).
Suppose that x Rh . Represent x by a triple (S, q, f ). Let q1 , . . . , qr be the other
primes of S lying over m. Then we can find a g S, g 6 q and g qi for i = 1, . . . , r,
see Lemma 14.2. Consider the morphism of pairs (S, q) (Sg , qSg ). In this way
we see that we may always assume that
x is given by a triple (S, q, f ) where q is
the only prime of S lying over m, i.e., mS = q. But since R S is etale, we have
mSq = qSq , see Lemma 141.5. Hence we actually get that mS = q.
Suppose that x 6 mRh . Represent x by a triple (S, q, f ) with mS = q. Then
f 6 mS, i.e., f 6 q. Hence (S, q) (Sf , qSf ) is a morphism of pairs such that the
image of f becomes invertible. Hence x is invertible with inverse represented by
the triple (Sf , qSf , 1/f ). We conclude that Rh is a local ring with maximal ideal
mRh . The residue field is since we can define Rh /mRh by mapping a triple
(S, q, f ) to the residue class of f module q.

COMMUTATIVE ALGEBRA

386

We still have to show that Rh is henselian. Namely, suppose that P Rh [T ] is a

monic polynomial and a0 is a simple root of the reduction P [T ]. Then we
can find a pair (S, q) such that P is the image of a monic polynomial Q S[T ]. Since
S Rh induces an isomorphism of residue fields we see that S 0 = S[T ]/(Q) has a
prime ideal q0 = (q, T a0 ) at which S S 0 is standard etale. Moreover, = (q0 ).
Pick g S 0 , g 6 q0 such that S 00 = Sg0 is etale over S. Then (S, q) (S 00 , q0 S 00 ) is
a morphism of pairs. Now that triple (S 00 , q0 S 00 , class of T ) determines an element

a Rh with the properties P (a) = 0, and a = a0 as desired.
04GP

Lemma 150.2. Let (R, m, ) be a local ring. Let sep be a separable algebraic
closure. There exists a commutative diagram
O

/
O

/ sep
O

/ Rh

/ Rsh

with the following properties

(1)
(2)
(3)
(4)
(5)

the map Rh Rsh is local

Rsh is strictly henselian,
Rsh is a filtered colimit of etale R-algebras,
mRsh is the maximal ideal of Rsh , and
sep = Rsh /mRsh .

Proof. This is proved by exactly the same proof as used for Lemma 150.1. The
only difference is that, instead of pairs, one uses triples (S, q, ) where R S etale,
q is a prime of S lying over m, and : (q) sep is an embedding of extensions
of .

04GQ

Definition 150.3. Let (R, m, ) be a local ring.

(1) The local ring map R Rh constructed in Lemma 150.1 is called the
henselization of R.
(2) Given a separable algebraic closure sep the local ring map R Rsh
constructed in Lemma 150.2 is called the strict henselization of R with
respect to sep .
(3) A local ring map R Rsh is called a strict henselization of R if it is
isomorphic to one of the local ring maps constructed in Lemma 150.2
The maps R Rh Rsh are flat local ring homomorphisms. By Lemma 149.6 the
R-algebras Rh and Rsh are well defined up to unique isomorphism by the conditions
that they are henselian local, filtered colimits of etale R-algebras with residue field
and sep . In the rest of this section we mostly just discuss functoriality of the (strict)
henselizations. We will discuss more intricate results concerning the relationship
between R and its henselization in More on Algebra, Section 36.

0BSL

Remark 150.4. We can also construct Rsh from Rh . Namely, for any finite separable subextension 0 sep there exists a unique (up to unique isomorphism)
finite etale local ring extension Rh Rh (0 ) whose residue field extension extension
reproduces the given extension, see Lemma 148.7. Hence we can set
[
Rsh =
Rh (0 )
0
sep

COMMUTATIVE ALGEBRA

387

The arrows in this system, compatible with the arrows on the level of residue fields,
exist by Lemma 148.7. This will produce a henselian local ring by Lemma 149.7
since each of the rings Rh (0 ) is henselian by Lemma 148.4. By construction the
residue field extension induced by Rh Rsh is the field extension sep . Hence
Rsh so constructed is strictly henselian. By Lemma 149.2 the R-algebra Rsh is a
colimit of etale R-algebras. Hence the uniqueness of Lemma 149.6 shows that Rsh
is the strict henselization.
04GR

Lemma 150.5. Let R S be a local map of local rings. Let S S h be the

henselization. Let R A be an etale ring map and let q be a prime of A lying
over mR such that R/mR
= (q). Then there exists a unique morphism of rings
f : A S h fitting into the commutative diagram
AO

/ Sh
O
/S

R
such that f 1 (mS h ) = q.

Proof. This is a special case of Lemma 148.11.

04GS

Lemma 150.6. Let R S be a local map of local rings. Let R Rh and S S h

be the henselizations. There exists a unique local ring map Rh S h fitting into
the commutative diagram
/ Sh
RO h
O
f

/S

Proof. Follows immediately from Lemma 149.5.

Here is a slightly different construction of the henselization.

04GV

Lemma 150.7. Let R be a ring. Let p R be a prime ideal. Consider the

category of pairs (S, q) where R S is etale and q is a prime lying over p such
that (p) = (q). This category is filtered and
(Rp )h = colim(S,q) S = colim(S,q) Sq
canonically.
Proof. A morphism of pairs (S, q) (S 0 , q0 ) is given by an R-algebra map :
S S 0 such that 1 (q0 ) = q. Let us show that the category of pairs is filtered,
see Categories, Definition 19.1. The category contains the pair (R, p) and hence is
not empty, which proves part (1) of Categories, Definition 19.1. Suppose that (S, q)
and (S 0 , q0 ) are two pairs. Note that q, resp. q0 correspond to primes of the fibre
rings S (p), resp. S 0 (p) with residue fields (p), hence they correspond to
maximal ideals of S (p), resp. S 0 (p). Set S 00 = S R S 0 . By the above there
exists a unique prime q00 S 00 lying over q and over q0 whose residue field is (p).
The ring map R S 00 is etale by Lemma 141.3. This proves part (2) of Categories,
Definition 19.1. Next, suppose that , : (S, q) (S 0 , q0 ) are two morphisms of
pairs. Consider
S 00 = (S 0 ,S, S 0 ) S 0 R S 0 S 0

COMMUTATIVE ALGEBRA

388

Arguing as above (base change of etale maps is etale, composition of etale maps is
etale) we see that S 00 is etale over R. The fibre ring of S 00 over p is
F 00 = (F 0 ,F, F 0 ) F 0 (p) F 0 F 0
where F 0 , F are the fibre rings of S 0 and S. Since and are morphisms of pairs
the map F 0 (p) corresponding to p0 extends to a map F 00 (p) and in turn
corresponds to a prime ideal q00 S 00 whose residue field is (p). The canonical
map S 0 S 00 (using the right most factor for example) is a morphism of pairs
(S 0 , q0 ) (S 00 , q00 ) which equalizes and . This proves part (3) of Categories,
Definition 19.1. Hence we conclude that the category is filtered.
Recall that in the proof of Lemma 150.1 we constructed (Rp )h as the corresponding
colimit but starting with Rp and its maximal ideal pRp . Now, given any pair (S, q)
for (R, p) we obtain a pair (Sp , qSp ) for (Rp , pRp ). Moreover, in this situation
Sp = colimf R,f 6p Sf .
Hence in order to show the equalities of the lemma, it suffices to show that any
pair (Sloc , qloc ) for (Rp , pRp ) is of the form (Sp , qSp ) for some pair (S, q) over (R, p)
(some details omitted). This follows from Lemma 141.3.

08HU

Lemma 150.8. Let R S be a ring map. Let q S be a prime lying over p R.

Let R Rh and S S h be the henselizations of Rp and Sq . The local ring map
Rh S h of Lemma 150.6 identifies S h with the henselization of Rh R S at the
unique prime lying over mh and q.
Proof. By Lemma 150.7 we see that Rh , resp. S h are filtered colimits of etale R,
resp. S-algebras. Hence we see that Rh R S is a filtered colimit of etale S-algebras
Ai (Lemma 141.3). By Lemma 149.4 we see that S h is a filtered colimit of etale
Rh R S-algebras. Since moreover S h is a henselian local ring with residue field
equal to (q), the statement follows from the uniqueness result of Lemma 149.6. 

05WP

Lemma 150.9. Let R S be a ring map. Let q be a prime of S lying over p in

R. Assume R S is quasi-finite at q. The commutative diagram
Rph
O

/ Sqh
O

Rp

/ Sq

of Lemma 150.6 identifies Sqh with the localization of Rph Rp Sq at the prime generated by q.
Proof. Note that Rph R S is quasi-finite over Rph at the prime ideal corresponding
to q, see Lemma 121.6. Hence the localization S 0 of Rph Rp Sq is henselian, see
Lemma 148.4. As a localization S 0 is a filtered colimit of etale Rph Rp Sq -algebras.
By Lemma 150.8 we see that Sqh is the henselization of Rph Rp Sq . Thus S 0 = Sqh
by the uniqueness result of Lemma 149.6.

05WQ

Lemma 150.10. Let R be a local ring with henselization Rh . Let I mR . Then

Rh /IRh is the henselization of R/I.
Proof. This is a special case of Lemma 150.9.

COMMUTATIVE ALGEBRA

04GT

389

Lemma 150.11. Let : R S be a local map of local rings. Let S/mS sep
be a separable algebraic closure. Let S S sh be the strict henselization of S with
respect to S/mS sep . Let R A be an etale ring map and let q be a prime of A
lying over mR . Given any commutative diagram
(q)
O
R/mR

/ sep
O
/ S/mS

there exists a unique morphism of rings f : A S sh fitting into the commutative

diagram
/ S sh
AO
O
f

/S
R
1
such that f (mS h ) = q and the induced map (q) sep is the given one.

Proof. This is a special case of Lemma 148.11.

04GU

Lemma 150.12. Let R S be a local map of local rings. Choose separable

sh
algebraic closures R/mR sep
and S/mS sep
and S S sh be
1
2 . Let R R
the corresponding strict henselizations. Given any commutative diagram
sep
1O
R/mR

/ sep
2O
/ S/mS

There exists a unique local ring map Rsh S sh fitting into the commutative diagram
/ S sh
Rsh
O
O
f
/S

R
and inducing on the residue fields of R

sh

and S sh .

Proof. Follows immediately from Lemma 149.5.

04GW


sep

Lemma 150.13. Let R be a ring. Let p R be a prime ideal. Let (p)

be
a separable algebraic closure. Consider the category of triples (S, q, ) where R S
is etale, q is a prime lying over p, and : (q) sep is a (p)-algebra map. This
category is filtered and
(Rp )sh = colim(S,q,) S = colim(S,q,) Sq
canonically.
Proof. A morphism of triples (S, q, ) (S 0 , q0 , 0 ) is given by an R-algebra map
: S S 0 such that 1 (q0 ) = q and such that 0 = . Let us show that the
category of pairs is filtered, see Categories, Definition 19.1. The category contains
the triple (R, p, (p) sep ) and hence is not empty, which proves part (1) of

COMMUTATIVE ALGEBRA

390

Categories, Definition 19.1. Suppose that (S, q, ) and (S 0 , q0 , 0 ) are two triples.
Note that q, resp. q0 correspond to primes of the fibre rings S (p), resp. S 0 (p)
with residue fields finite separable over (p) and , resp. 0 correspond to maps
into sep . Hence this data corresponds to (p)-algebra maps
0 : S 0 R (p) sep .

: S R (p) sep ,

Set S 00 = S R S 0 . Combining the maps the above we get a unique (p)-algebra

map
00 = 0 : S 00 R (p) sep
whose kernel corresponds to a prime q00 S 00 lying over q and over q0 , and whose
residue field maps via 00 to the compositum of ((q)) and 0 ((q0 )) in sep . The
ring map R S 00 is etale by Lemma 141.3. Hence (S 00 , q00 , 00 ) is a triple dominating both (S, q, ) and (S 0 , q0 , 0 ). This proves part (2) of Categories, Definition
19.1. Next, suppose that , : (S, q, ) (S 0 , q0 , 0 ) are two morphisms of pairs.
Consider
S 00 = (S 0 ,S, S 0 ) S 0 R S 0 S 0
Arguing as above (base change of etale maps is etale, composition of etale maps is
etale) we see that S 00 is etale over R. The fibre ring of S 00 over p is
F 00 = (F 0 ,F, F 0 ) F 0 (p) F 0 F 0
where F 0 , F are the fibre rings of S 0 and S. Since and are morphisms of triples
the map 0 : F 0 sep extends to a map 00 : F 00 sep which in turn corresponds
to a prime ideal q00 S 00 . The canonical map S 0 S 00 (using the right most factor
for example) is a morphism of triples (S 0 , q0 , 0 ) (S 00 , q00 , 00 ) which equalizes
and . This proves part (3) of Categories, Definition 19.1. Hence we conclude that
the category is filtered.
We still have to show that the colimit Rcolim of the system is equal to the strict
henselization of Rp with respect to sep . To see this note that the system of triples
(S, q, ) contains as a subsystem the pairs (S, q) of Lemma 150.7. Hence Rcolim
contains Rph by the result of that lemma. Moreover, it is clear that Rph Rcolim
is a directed colimit of etale ring extensions. It follows that Rcolim is henselian by
Lemmas 148.4 and 149.7. Finally, by Lemma 141.16 we see that the residue field
of Rcolim is equal to sep . Hence we conclude that Rcolim is strictly henselian and
hence equals the strict henselization of Rp as desired. Some details omitted.

08HV

Lemma 150.14. Let R S be a ring map. Let q S be a prime lying over

and (q) sep
p R. Choose separable algebraic closures (p) sep
1
2 . Let
sh
sh
R and S be the corresponding strict henselizations of Rp and Sq . Given any
commutative diagram
/ sep
sep
2O
1O

(p)

/ (q)

The local ring map Rsh S sh of Lemma 150.12 identifies S sh with the strict
henselization of Rsh R S at a prime lying over msh and q.
Proof. The proof is identical to the proof of Lemma 150.8 except that it uses
Lemma 150.13 instead of Lemma 150.7.


COMMUTATIVE ALGEBRA

05WR

391

Lemma 150.15. Let R S be a ring map. Let q be a prime of S lying over p in

R. Let (q) sep be a separable algebraic closure. Assume R S is quasi-finite
at q. The commutative diagram
Rpsh
O

/ Sqsh
O

Rp

/ Sq

of Lemma 150.12 identifies Sqsh with a localization of Rpsh Rp Sq .

Proof. The residue field of Rpsh is the separable algebraic closure of (p) in sep .
Note that Rpsh R S is quasi-finite over Rpsh at the prime ideal corresponding to q,
see Lemma 121.6. Hence the localization S 0 of Rpsh Rp Sq is henselian, see Lemma
148.4. Note that the residue field of S 0 is sep since it contains both the separable
algebraic closure of (p) and (q). Furthermore, as a localization S 0 is a filtered
colimit of etale Rpsh Rp Sq -algebras. By Lemma 150.14 we see that Sqsh is a strict
henselization of Rpsh Rp Sq . Thus S 0 = Sqh by the uniqueness result of Lemma
149.6.

05WS

Lemma 150.16. Let R be a local ring with strict henselization Rsh . Let I mR .
Then Rsh /IRsh is a strict henselization of R/I.
Proof. This is a special case of Lemma 150.15.

151. Serres criterion for normality

031O

We introduce the following properties of Noetherian rings.

031P

Definition 151.1. Let R be a Noetherian ring. Let k 0 be an integer.

(1) We say R has property (Rk ) if for every prime p of height k the local
ring Rp is regular. We also say that R is regular in codimension k.
(2) We say R has property (Sk ) if for every prime p the local ring Rp has depth
at least min{k, dim(Rp )}.
(3) Let M be a finite R-module. We say M has property (Sk ) if for every prime
p the module Mp has depth at least min{k, dim(Supp(Mp ))}.
Any Noetherian ring has property (S0 ) (and so does any finite module over it). Our
convention that dim() = guarantees that the zero module has property (Sk )
for all k.

031Q

Lemma 151.2. Let R be a Noetherian ring. Let M be a finite R-module. The

following are equivalent:
(1) M has no embedded associated prime, and
(2) M has property (S1 ).
Proof. Let p be an embedded associated prime of M . Then there exists another associated prime q of M such that p q. In particular this implies that
dim(Supp(Mp )) 1 (since q is in the support as well). On the other hand pRp is
associated to Mp (Lemma 62.15) and hence depth(Mp ) = 0 (see Lemma 62.18). In
other words (S1 ) does not hold. Conversely, if (S1 ) does not hold then there exists

COMMUTATIVE ALGEBRA

392

a prime p such that dim(Supp(Mp )) 1 and depth(Mp ) = 0. Then we see (arguing backwards using the lemmas cited above) that p is an embedded associated
prime.

031R

Lemma 151.3. Let R be a Noetherian ring. The following are equivalent:

(1) R is reduced, and
(2) R has properties (R0 ) and (S1 ).
Proof. Suppose that R is reduced. Then Rp is a field for every minimal prime p
of R, according to Lemma 24.1. Hence we have (R0 ). Let p be a prime of height
1. Then A = Rp is a reduced local ring of dimension 1. Hence its maximal
ideal m is not an associated prime since this would mean there exists a x m with
annihilator m so x2 = 0. Hence the depth of A = Rp is at least one, by Lemma
62.9. This shows that (S1 ) holds.
Conversely, assume that R satisfies (R0 ) and (S1 ). If p is a minimal prime of R,
then Rp is a field by (R0 ), and hence is reduced. If p is not minimal, then we see
that Rp has depth 1 by (S1 ) and we conclude there exists an element t pRp such
that Rp Rp [1/t] is injective. This implies that Rp is a subring of localizations of
R at primes of smaller height. Thus by induction on the height we conclude that
R is reduced.


031S

Lemma 151.4 (Serres criterion for normality). Let R be a Noetherian ring. The
following are equivalent:
(1) R is a normal ring, and
(2) R has properties (R1 ) and (S2 ).
Proof. Proof of (1) (2). Assume R is normal, i.e., all localizations Rp at primes
are normal domains. In particular we see that R has (R0 ) and (S1 ) by Lemma 151.3.
Hence it suffices to show that a local Noetherian normal domain R of dimension d
has depth min(2, d) and is regular if d = 1. The assertion if d = 1 follows from
Lemma 118.7.
Let R be a local Noetherian normal domain with maximal ideal m and dimension
d 2. Apply Lemma 118.2 to R. It is clear that R does not fall into cases (1) or
(2) of the lemma. Let R R0 as in (4) of the lemma. Since R is a domain we have
R R0 . Since m is not an associated prime of R0 there exists an x m which is
a nonzerodivisor on R0 . Then Rx = Rx0 so R and R0 are domains with the same
fraction field. But finiteness of R R0 implies every element of R0 is integral over
R (Lemma 35.3) and we conclude that R = R0 as R is normal. This means (4)
does not happen. Thus we get the remaining possibility (3), i.e., depth(R) 2 as
desired.
Proof of (2) (1). Assume R satisfies (R1 ) and (S2 ). By Lemma 151.3 we conclude
that R is reduced. Hence it suffices to show that if R is a reduced local Noetherian
ring of dimension d satisfying (S2 ) and (R1 ) then R is a normal domain. If d = 0,
the result is clear. If d = 1, then the result follows from Lemma 118.7.
Let R be a reduced local Noetherian ring with maximal ideal m and dimension
d 2 which satisfies (R1 ) and (S2 ). By Lemma 36.15 it suffices to show that R is
integrally closed in its total ring of fractions Q(R). Pick x Q(R) which is integral
over R. Then R0 = R[x] is a finite ring extension of R (Lemma 35.5). Because
dim(Rp ) < d for every nonmaximal prime p R we have Rp = Rp0 by induction.

COMMUTATIVE ALGEBRA

393

Hence the support of R0 /R is {m}. It follows that R0 /R is annihilated by a power of

m (Lemma 61.4). By Lemma 118.2 this contradicts the assumption that the depth
of R is 2 = min(2, d) and the proof is complete.

0567

Lemma 151.5. A regular ring is normal.

Proof. Let R be a regular ring. By Lemma 151.4 it suffices to prove that R is (R1 )
and (S2 ). As a regular local ring is Cohen-Macaulay, see Lemma 105.3, it is clear
that R is (S2 ). Property (R1 ) is immediate.


031T

Lemma 151.6. Let R be a Noetherian normal domain with fraction field K. Then
(1) for any nonzero a R the quotient R/aR has no embedded primes, and all
its associated primes have height 1
(2)
\
R=
Rp
height(p)=1

(3) For any nonzero x K the quotient R/(R xR) has no embedded primes,
and all its associates primes have height 1.
Proof. By Lemma 151.4 we see that R has (S2 ). Hence for any nonzero element
a R we see that R/aR has (S1 ) (use Lemma 71.6 for example) Hence R/aR has
no embedded primes (Lemma 151.2). We conclude the associated primes of R/aR
are exactly the minimal primes p over (a), which have height 1 as a is not zero
(Lemma 59.10). This proves (1).
Thus, given b R we have b aR if and only if b aRp for every minimal prime
p over (a) (see Lemma 62.19). These primes all have height 1 as seen above so
b/a R if and only if b/a Rp for all height 1 primes. Hence (2) holds.
For (3) write x = a/b. Let p1 , . . . , pr be the minimal primes
over (ab). These all
T
have height 1 by the above. Then we see that R xR = i=1,...,r (R xRpi ) by part
L
(2) of the lemma. Hence R/(R xR) is a submodule of
R/(R xRpi ). As Rpi is
a discrete valuation ring (by property (R1 ) for the Noetherian normal domain R,
see Lemma 151.4) we have xRpi = pei i Rpi for some ei Z. Hence the direct sum
L
(e )
is equal to ei >0 R/pi i , see Definition 63.1. By Lemma 63.2 the only associated
prime of the module R/p(n) is p. Hence the set of associate primes of R/(R xR)
is a subset of {pi } and there are no inclusion relations among them. This proves
(3).

152. Formal smoothness of fields
031U

In this section we show that field extensions are formally smooth if and only if
they are separable. However, we first prove finitely generated field extensions are
separable algebraic if and only if they are formally unramified.

090W

Lemma 152.1. Let k K be a finitely generated field extension. The following

are equivalent
(1) K is a finite separable field extension of k,
(2) K/k = 0,
(3) K is formally unramified over k,
(4) K is unramified over k,
(5) K is formally etale over k,
(6) K is etale over k.

COMMUTATIVE ALGEBRA

394

Proof. The equivalence of (2) and (3) is Lemma 144.2. By Lemma 141.4 we see
that (1) is equivalent to (6). Property (6) implies (5) and (4) which both in turn
imply (3) (Lemmas 146.2, 147.3, and 147.2). Thus it suffices to show that (2) implies
(1). Choose a finitely generated k-subalgebra A K such that K is the fraction
field of the domain A. Set S = A \ {0}. Since 0 = K/k = S 1 A/k (Lemma
130.8) and since A/k is finitely generated (Lemma 130.16), we can replace A by
a localization Af to reduce to the case that A/k = 0 (details omitted). Then A
is unramified over k, hence K/k is finite separable for example by Lemma 147.5
applied with q = (0).

031W

Lemma 152.2. Let K be a field of characteristic p > 0. Let a K. Then da = 0

in K/Fp if and only if a is a pth power.
Proof. By Lemma 130.4 we see that there exists a subfield Fp L K such that
Fp L is a finitely generated field extension and such that da is zero in L/Fp .
Hence we may assume that K is a finitely generated field extension of Fp .
Choose a transcendence basis x1 , . . . , xr K such that K is finite separable over
Fp (x1 , . . . , xr ). We remark that the result holds for the purely transcendental
subfield Fp (x1 , . . . , xr ) K. Namely,
Mr
Fp (x1 ,...,xr )/Fp =
Fp (x1 , . . . , xr )dxi
i=1

and any rational function all of whose partial derivatives are zero is a pth power.
Moreover, we also have
Mr
Kdxi
K/Fp =
i=1

since Fp (x1 , . . . , xr ) K is finite separable (computation omitted). Suppose a K

is an element such that da = 0 in the module of differentials. By our choice of xi we
see that the minimal polynomial P (T ) k(x1 , . . . , xr )[T ] of a is separable. Write
Xd
P (T ) = T d +
ai T di
i=1

and hence
0 = dP (a) =

Xd
i=1

adi dai

in K/Fp . By the description of K/Fp above and the fact that P was the minimal
polynomial of a, we see that this implies dai = 0. Hence ai = bpi for each i.
Therefore by Fields, Lemma 27.2 we see that a is a pth power.

07DZ

Lemma 152.3. Let k be a field of characteristic p > 0. Let a1 , . . . , an k be

elements such that da1 , . . . , dan are linearly independent in k/Fp . Then the field
1/p

1/p

extension k(a1 , . . . , an ) has degree pn over k.

Proof. By induction on n. If n = 1 the result is Lemma 152.2. For the induction
1/p
1/p
step, suppose that k(a1 , . . . , an1 ) has degree pn1 over k. We have to show that
1/p
1/p
an does not map to a pth power in k(a1 , . . . , an1 ). If it does then we can write
X
p
in1 /p
i1 /p
an =
I a1 . . . an1
I=(i1 ,...,in1 ), 0ij p1
X
in1
=
pI ai11 . . . an1
I=(i1 ,...,in1 ), 0ij p1

Applying d we see that dan is linearly dependent on dai , i < n. This is a contradiction.


COMMUTATIVE ALGEBRA

031X

395

Lemma 152.4. Let k be a field of characteristic p > 0. The following are equivalent:
(1) the field extension K/k is separable (see Definition 41.1), and
(2) the map K k k/Fp K/Fp is injective.
Proof. Write K as a directed colimit K = colimi Ki of finitely generated field
extensions k Ki . By definition K is separable if and only if each Ki is separable
over k, and by Lemma 130.4 we see that K k k/Fp K/Fp is injective if and
only if each Ki k k/Fp Ki /Fp is injective. Hence we may assume that K/k
is a finitely generated field extension.
Assume k K is a finitely generated field extension which is separable. Choose
x1 , . . . , xr+1 K as in Lemma 41.3. In this case there exists an irreducible polynomial G(X1 , . . . , Xr+1 ) k[X1 , . . . , Xr+1 ] such that G(x1 , . . . , xr+1 ) = 0 and such
that G/Xr+1 is not identically zero. Moreover K is the field of fractions of the
domain. S = K[X1 , . . . , Xr+1 ]/(G). Write
X
ir+1
G=
aI X I , X I = X1i1 . . . Xr+1
.
Using the presentation of S above we see that
L
S k k i=1,...,r+1 SdXi
P
S/Fp = P I
h X daI + G/Xi dXi i
Since K/Fp is the localization of the S-module S/Fp (see Lemma 130.8) we conclude that
L
K k k i=1,...,r+1 KdXi
P
K/Fp = P I
h X daI + G/Xi dXi i
Now, since the polynomial G/Xr+1 is not identically zero we conclude that the
map K k k/Fp S/Fp is injective as desired.
Assume k K is a finitely generated field extension and that K k k/Fp K/Fp
is injective. (This part of the proof is the same as the argument proving Lemma
43.1.) Let x1 , . . . , xr be a transcendence basis of K over k such that the degree
of inseparability of the finite extension k(x1 , . . . , xr ) K is minimal. If K is
separable over k(x1 , . . . , xr ) then we win. Assume this is not the case to get a
contradiction. Then there exists an element K which is not separable over
k(x1 , . . . , xr ). Let P (T ) k(x1 , . . . , xr )[T ] be its minimal polynomial. Because
is not separable actually P is a polynomial in T p . Clear denominators to get an
irreducible polynomial
X
G(X1 , . . . , Xr , T ) =
aI,i X I T i k[X1 , . . . , Xr , T ]
such that G(x1 , . . . , xr , ) = 0 in L. Note that this means k[X1 , . . . , Xr , T ]/(G)
L. We may assume that for some pair (I0 , i0 ) the coefficient aI0 ,i0 = 1. We claim
that dG/dXi is not identically zero for at least one i. Namely, if this is not the
case, then G is actually a polynomial in X1p , . . . , Xrp , T p . Then this means that
X
xI i daI,i
(I,i)6=(I0 ,i0 )

is zero in K/Fp . Note that there is no k-linear relation among the elements
{xI i | aI,i 6= 0 and (I, i) 6= (I0 , i0 )}

COMMUTATIVE ALGEBRA

396

of K. Hence the assumption that K k k/Fp K/Fp is injective this implies

that daI,i = 0 in k/Fp for all (I, i). By Lemma 152.2 we see that each aI,i is a
pth power, which implies that G is a pth power contradicting the irreducibility of
G. Thus, after renumbering, we may assume that dG/dX1 is not zero. Then we
see that x1 is separably algebraic over k(x2 , . . . , xr , ), and that x2 , . . . , xr , is a
transcendence basis of L over k. This means that the degree of inseparability of
the finite extension k(x2 , . . . , xr , ) L is less than the degree of inseparability of
the finite extension k(x1 , . . . , xr ) L, which is a contradiction.

031Y

Lemma 152.5. Let k K be an extension of fields. If K is formally smooth over

k, then K is a separable extension of k.
Proof. Assume K is formally smooth over k. By Lemma 136.9 we see that K k
k/Z K/Z is injective. Hence K is separable over k by Lemma 152.4.


031Z

Lemma 152.6. Let k K be an extension of fields. Then K is formally smooth

over k if and only if H1 (LK/k ) = 0.
Proof. This follows from Proposition 136.8 and the fact that a vector spaces is
free (hence projective).


0320

Lemma 152.7. Let k K be an extension of fields.

(1) If K is purely transcendental over k, then K is formally smooth over k.
(2) If K is separable algebraic over k, then K is formally smooth over k.
(3) If K is separable over k, then K is formally smooth over k.
Proof. For (1) write K = k(xj ; j J). Suppose that A is a k-algebra, and I A
is an ideal of square zero. Let : K A/I be a k-algebra map. Let aj A be an
element such that aj mod I = (xj ). Then it is easy to see that there is a unique
k-algebra map K A which maps xj to aj and which reduces to mod I. Hence
k K is formally smooth.
In case (2) we see that k K is a colimit of etale ring extensions. An etale ring
map is formally etale (Lemma 146.2). Hence this case follows from Lemma 146.3
and the trivial observation that a formally etale ring map is formally smooth.
In case (3), write K = colim Ki as the filtered colimit of its finitely generated
sub k-extensions. By Definition 41.1 each Ki is separable algebraic over a purely
transcendental extension of k. Hence Ki /k is formally smooth by cases (1) and (2)
and Lemma 136.3. Thus H1 (LKi /k ) = 0 by Lemma 152.6. Hence H1 (LK/k ) = 0 by
Lemma 132.9. Hence K/k is formally smooth by Lemma 152.6 again.


0321

Lemma 152.8. Let k be a field.

(1) If the characteristic of k is zero, then any extension field of k is formally
smooth over k.
(2) If the characteristic of k is p > 0, then k K is formally smooth if and
only if it is a separable field extension.
Proof. Combine Lemmas 152.5 and 152.7.

Here we put together all the different characterizations of separable field extensions.
0322

Proposition 152.9. Let k K be a field extension. If the characteristic of k is

zero then

COMMUTATIVE ALGEBRA

(1)
(2)
(3)
(4)
(5)

397

K is separable over k,
K is geometrically reduced over k,
K is formally smooth over k,
H1 (LK/k ) = 0, and
the map K k k/Z K/Z is injective.

If the characteristic of k is p > 0, then the following are equivalent:

(1)
(2)
(3)
(4)
(5)
(6)

K is separable over k,
the ring K k k 1/p is reduced,
K is geometrically reduced over k,
the map K k k/Fp K/Fp is injective,
H1 (LK/k ) = 0, and
K is formally smooth over k.

Proof. This is a combination of Lemmas 43.1, 152.8 152.5, and 152.4.

Here is yet another characterization of finitely generated separable field extensions.

037X

Lemma 152.10. Let k K be a finitely generated field extension. Then K is

separable over k if and only if K is the localization of a smooth k-algebra.
Proof. Choose a finite type k-algebra R which is a domain whose fraction field is
K. Lemma 138.9 says that k R is smooth at (0) if and only if K/k is separable.
This proves the lemma.


07BV

Lemma 152.11. Let k K be a field extension. Then K is a filtered colimit

of global complete intersection algebras over k. If K/k is separable, then K is a
filtered colimit of smooth algebras over k.
Proof. Suppose that E K is a finite subset. It suffices to show that there exists
a k subalgebra A K which contains E and which is a global complete intersection (resp. smooth) over k. The separable/smooth case follows from Lemma
152.10. In general let L K be the subfield generated by E. Pick a transcendence
basis x1 , . . . , xd L over k. The extension k(x1 , . . . , xd ) L is finite. Say L =
k(x1 , . . . , xd )[y1 , . . . , yr ]. Pick inductively polynomials Pi k(x1 , . . . , xd )[Y1 , . . . , Yr ]
such that Pi = Pi (Y1 , . . . , Yi ) is monic in Yi over k(x1 , . . . , xd )[Y1 , . . . , Yi1 ] and
maps to the minimum polynomial of yi in k(x1 , . . . , xd )[y1 , . . . , yi1 ][Yi ]. Then it is
clear that P1 , . . . , Pr is a regular sequence in k(x1 , . . . , xr )[Y1 , . . . , Yr ] and that L =
k(x1 , . . . , xr )[Y1 , . . . , Yr ]/(P1 , . . . , Pr ). If h k[x1 , . . . , xd ] is a polynomial such that
Pi k[x1 , . . . , xd , 1/h, Y1 , . . . , Yr ], then we see that P1 , . . . , Pr is a regular sequence
in k[x1 , . . . , xd , 1/h, Y1 , . . . , Yr ] and A = k[x1 , . . . , xd , 1/h, Y1 , . . . , Yr ]/(P1 , . . . , Pr )
is a global complete intersection. After adjusting our choice of h we may assume
E A and we win.

153. Constructing flat ring maps

03C2

The following lemma is occasionally useful.

03C3

Lemma 153.1. Let (R, m, k) be a local ring. Let k K be a field extension. There
exists a local ring (R0 , m0 , k 0 ), a flat local ring map R R0 such that m0 = mR0
and such that k k 0 is isomorphic to k K.

COMMUTATIVE ALGEBRA

398

Proof. Suppose that k k 0 = k() is a monogenic extension of fields. Then k 0

is the residue field of a flat local extension R R0 as in the lemma. Namely, if
is transcendental over k, then we let R0 be the localization
at the prime
P of R[x]
di
d

T
,
then we let
mR[x]. If is algebraic
with
minimal
polynomial
T
+
i
P
R0 = R[T ]/(T d + i T di ).
Consider the collection of triples (k 0 , R R0 , ), where k k 0 K is a subfield,
R R0 is a local ring map as in the lemma, and : R0 k 0 induces an isomorphism
R0 /mR0
= k 0 of k-extensions. These form a big category C with morphisms
(k1 , R1 , 1 ) (k2 , R2 , 2 ) given by ring maps : R1 R2 such that
R1

/ k1

/K

/ k2

/K


R2

commutes. This implies that k1 k2 .

Suppose that I is a directed partially ordered set, and ((Ri , ki , i ), ii0 ) is a system
over I, see Categories, Section 21. In this case we can consider
R0 = colimiI Ri
S
This is a local ring with maximal ideal mR0 , and residue field k 0 = iI ki . Moreover, the ring map R R0 is flat as it is a colimit of flat maps (and tensor products
commute with directed colimits). Hence we see that (R0 , k 0 , 0 ) is an upper bound
for the system.
An almost trivial application of Zorns Lemma would finish the proof if C was a set,
but it isnt. (Actually, you can make this work by finding a reasonable bound on the
cardinals of the local rings occurring.) To get around this problem we choose a well
ordering on K. For x K we let K(x) be the subfield of K generated by all elements
of K which are x. By transfinite induction on x K we will produce ring maps
R R(x) as in the lemma with residue field extension k K(x). Moreover, by
construction we will have that R(x) will contain R(y) for all y x. Namely, if x
has a predecessor x0 , then K(x) = K(x0 )[x] and hence we can let R(x0 ) R(x) be
the local ring extension constructed in the first paragraph of the proof. If x does
0
not have a predecessor, then we first set R0 (x) = colimx0 <x R(x
S ) as in the third
paragraph of the proof. The residue field of R0 (x) is K 0 (x) = x0 <x K(x0 ). Since
K(x) = K 0 (x)[x] we see that we can use the construction of the first paragraph of
the proof to produce R0 (x) R(x). This finishes the proof of the lemma.

09E0

Lemma 153.2. Let (R, m, k) be a local ring. If k K is a separable algebraic

extension, then there exists a directed partially ordered set I and a system of finite
etale extensions R Ri , i I of local rings such that R0 = colim Ri has residue
field K (as extension of k).
Proof. Let R R0 be the extension constructed in the proof of Lemma 153.1. By
construction R0 = colimA R where A is a well-ordered set and the transition
maps R R+1 are finite etale and R = colim< R if is not a successor.
We will prove the result by transfinite induction.
Suppose the result holds for R , i.e., R = colim Ri with Ri finite etale over
R. Since R R+1 is finite etale there exists an i and a finite etale extension

COMMUTATIVE ALGEBRA

399

Ri Ri,1 such that R+1 = R Ri Ri,1 . Thus R+1 = colimi0 i Ri0 Ri Ri,1 and
the result holds for + 1. Suppose is not a successor and the result holds for R
for all < . Since every finite subset E R is contained in R for some <
and we see that E is contained in a finite etale subextension by assumption. Thus
the result holds for R .

07NE

Lemma 153.3. Let R be a ring. Let p R be a prime and let (p) L be a

finite extension of fields. Then there exists a finite free ring map R S such that
q = pS is prime and (p) (q) is isomorphic to the given extension (p) L.
Proof. By induction of the degree of (p) L. If the degree is 1, then we take
R = S. In general, if there exists a sub extension (p) L0 L then we win
by induction on the degree (by first constructing R S 0 corresponding to L0 /(p)
and then construction S 0 S corresponding to L/L0 ). Thus weP
may assume that
L (p) is generated by a single element L. Let X d + i<d ai X i be the
minimal polynomial of over (p), so ai (p). We may write ai as the image of
fi /g for some fi , g R and g 6 p. After replacing by g (and correspondingly
replacing ai by g di ai ) we may P
assume that ai is the image of some fi R. Then
we simply take S = R[x]/(xd + fi xi ).

154. The Cohen structure theorem

0323

Here is a fundamental notion in commutative algebra.

0324

Definition 154.1. Let (R, m) be a local ring. We say R is a complete local ring
if the canonical map
R limn R/mn
to the completion of R with respect to m is an isomorphism10.
Note that an Artinian local ring R is a complete local ring because mnR = 0 for
some n > 0. In this section we mostly focus on Noetherian complete local rings.

0325

Lemma 154.2. Let R be a Noetherian complete local ring. Any quotient of R is

also a Noetherian complete local ring. Given a finite ring map R S, then S is a
product of Noetherian complete local rings.
Proof. The ring S is Noetherian by Lemma 30.1. As an R-module S is complete
by Lemma 96.1. Hence S is the product of the completions at its maximal ideals
by Lemma 96.8.


032B

Lemma 154.3. Let (R, m) be a complete local ring. If m is a finitely generated

ideal then R is Noetherian.
Proof. See Lemma 96.5.

0326

Definition 154.4. Let (R, m) be a complete local ring. A subring R is called

a coefficient ring if the following conditions hold:
(1) is a complete local ring with maximal ideal m,
(2) the residue field of maps isomorphically to the residue field of R, and
(3) m = p, where p is the characteristic of the residue field of R.
10This includes the condition that T mn = (0); in some texts this may be indicated by saying

that R is complete and separated. Warning: It can happen that the completion limn R/mn of a
local ring is non-complete, see Examples, Lemma 6.1. This does not happen when m is finitely
generated, see Lemma 95.5 in which case the completion is Noetherian, see Lemma 96.5.

COMMUTATIVE ALGEBRA

400

Let us make some remarks on this definition. We split the discussion into the
following cases:
(1) The local ring R contains a field. This happens if either Q R, or pR = 0
where p is the characteristic of R/m. In this case a coefficient ring is a
field contained in R which maps isomorphically to R/m.
(2) The characteristic of R/m is p > 0 but no power of p is zero in R. In this
case is a complete discrete valuation ring with uniformizer p and residue
field R/m.
(3) The characteristic of R/m is p > 0, and for some n > 1 we have pn1 6= 0,
pn = 0 in R. In this case is an Artinian local ring whose maximal ideal
is generated by p and which has residue field R/m.
The complete discrete valuation rings with uniformizer p above play a special role
and we baptize them as follows.
0327

Definition 154.5. A Cohen ring is a complete discrete valuation ring with uniformizer p a prime number.

0328

Lemma 154.6. Let p be a prime number. Let k be a field of characteristic p.

There exists a Cohen ring with /p
= k.
Proof. First note that the p-adic integers Zp form a Cohen ring for Fp . Let k be
an arbitrary field of characteristic p. Let Zp R be a flat local ring map such that
mR = pR and R/pR = k, see Lemma 153.1. Then clearly R is a discrete valuation
ring. Hence its completion is a Cohen ring for k.


0329

Lemma 154.7. Let p > 0 be a prime. Let be a Cohen ring with residue field of
characteristic p. For every n 1 the ring map
Z/pn Z /pn
is formally smooth.
Proof. If n = 1, this follows from Proposition 152.9. For general n we argue
by induction on n. Namely, if Z/pn Z /pn is formally smooth, then we
can apply Lemma 136.12 to the ring map Z/pn+1 Z /pn+1 and the ideal
I = (pn ) Z/pn+1 Z.


032A

Theorem 154.8 (Cohen structure theorem). Let (R, m) be a complete local ring.
(1) R has a coefficient ring (see Definition 154.4),
(2) if m is a finitely generated ideal, then R is isomorphic to a quotient
[[x1 , . . . , xn ]]/I
where is either a field or a Cohen ring.
Proof. Let us prove a coefficient ring exists. First we prove this in case the characteristic of the residue field is zero. Namely, in this case we will prove by induction
on n > 0 that there exists a section
n : R/mn
to the canonical map R/mn = R/m. This is trivial for n = 1. If n > 1, let
n1 be given. The field extension Q is formally smooth by Proposition 152.9.

COMMUTATIVE ALGEBRA

401

Hence we can find the dotted arrow in the following diagram

R/mn1 o
O

R/mn
9 O

n1

This proves the induction step. Putting these maps together

limn n : R = limn R/mn
gives a map whose image is the desired coefficient ring.
Next, we prove the existence of a coefficient ring in the case where the characteristic
of the residue field is p > 0. Namely, choose a Cohen ring with = /p, see
Lemma 154.6. In this case we will prove by induction on n > 0 that there exists a
map
n : /pn R/mn
whose composition with the reduction map R/mn produces the given isomorphism /p = . This is trivial for n = 1. If n > 1, let n1 be given. The ring
map Z/pn Z /pn is formally smooth by Lemma 154.7. Hence we can find the
dotted arrow in the following diagram
R/mn1 o
O

R/mn
9 O

n1

/pn o

Z/pn Z

This proves the induction step. Putting these maps together

limn n : = limn /pn R = limn R/mn
gives a map whose image is the desired coefficient ring.
The final statement of the theorem is now clear. Namely, if y1 , . . . , yn are generators
of the ideal m, then we can use the map R just constructed to get a map
[[x1 , . . . , xn ]] R,

xi 7 yi .

This map is surjective on each R/m and hence is surjective as R is complete. Some
details omitted.

032C

Remark 154.9. If k is a field then the power series ring k[[X1 , . . . , Xd ]] is a

Noetherian complete local regular ring of dimension d. If is a Cohen ring then
[[X1 , . . . , Xd ]] is a complete local Noetherian regular ring of dimension d+1. Hence
the Cohen structure theorem implies that any Noetherian complete local ring is a
quotient of a regular local ring. In particular we see that a Noetherian complete
local ring is universally catenary, see Lemma 104.8 and Lemma 105.3.

032D

Lemma 154.10. Let (R, m) be a Noetherian complete local domain. Then there
exists a R0 R with the following properties
(1) R0 is a regular complete local ring,
(2) R0 R is finite and induces an isomorphism on residue fields,
(3) R0 is either isomorphic to k[[X1 , . . . , Xd ]] where k is a field or [[X1 , . . . , Xd ]]
where is a Cohen ring.

COMMUTATIVE ALGEBRA

402

Proof. Let be a coefficient ring of R. Since R is a domain we see that either

is a field or is a Cohen ring.
Case I: = k is a field. Let d = dim(R). Choose x1 , . . . , xd m which generate
an ideal of definition I R. (See Section 59.) By Lemma 95.9 we see that R is Iadically complete as well. Consider the map R0 = k[[X1 , . . . , Xd ]] R which maps
Xi to xi . Note that R0 is complete with respect to the ideal I0 = (X1 , . . . , Xd ),
and that R/I0 R
= R/IR is finite over k = R0 /I0 (because dim(R/I) = 0, see
Section 59.) Hence we conclude that R0 R is finite by Lemma 95.12. Since
dim(R) = dim(R0 ) this implies that R0 R is injective (see Lemma 111.3), and
the lemma is proved.
Case II: is a Cohen ring. Let d + 1 = dim(R). Let p > 0 be the characteristic
of the residue field k. As R is a domain we see that p is a nonzerodivisor in R.
Hence dim(R/pR) = d, see Lemma 59.12. Choose x1 , . . . , xd R which generate
an ideal of definition in R/pR. Then I = (p, x1 , . . . , xd ) is an ideal of definition
of R. By Lemma 95.9 we see that R is I-adically complete as well. Consider the
map R0 = [[X1 , . . . , Xd ]] R which maps Xi to xi . Note that R0 is complete
with respect to the ideal I0 = (p, X1 , . . . , Xd ), and that R/I0 R
= R/IR is finite
over k = R0 /I0 (because dim(R/I) = 0, see Section 59.) Hence we conclude that
R0 R is finite by Lemma 95.12. Since dim(R) = dim(R0 ) this implies that
R0 R is injective (see Lemma 111.3), and the lemma is proved.

155. Japanese rings
0BI1

In this section we being to discuss finiteness of integral closure.

032F

Definition 155.1. Let R be a domain with field of fractions K.

(1) We say R is N-1 if the integral closure of R in K is a finite R-module.
(2) We say R is N-2 or Japanese if for any finite extension K L of fields the
integral closure of R in L is finite over R.
The main interest in these notions is for Noetherian rings, but here is a nonNoetherian example.

0350

Example 155.2. Let k be a field. The domain R = k[x1 , x2 , x3 , . . .] is N-2, but not
Noetherian. The reason is the following. Suppose that R L and the field L is a
finite extension of the fraction field of R. Then there exists an integer n such that L
comes from a finite extension k(x1 , . . . , xn ) L0 by adjoining the (transcendental)
elements xn+1 , xn+2 , etc. Let S0 be the integral closure of k[x1 , . . . , xn ] in L0 . By
Proposition 156.16 below it is true that S0 is finite over k[x1 , . . . , xn ]. Moreover,
the integral closure of R in L is S = S0 [xn+1 , xn+2 , . . .] (use Lemma 36.8) and hence
finite over R. The same argument works for R = Z[x1 , x2 , x3 , . . .].

032G

Lemma 155.3. Let R be a domain. If R is N-1 then so is any localization of R.

Same for N-2.
Proof. These statements hold because taking integral closure commutes with localization, see Lemma 35.9.


032H

Lemma 155.4. Let R be a domain. Let f1 , . . . , fn R generate the unit ideal. If

each domain Rfi is N-1 then so is R. Same for N-2.

COMMUTATIVE ALGEBRA

403

Proof. Assume Rfi is N-2 (or N-1). Let L be a finite extension of the fraction field
of R (equal to the fraction field in the N-1 case). Let S be the integral closure of
R in L. By Lemma 35.9 we see that Sfi is the integral closure of Rfi in L. Hence
Sfi is finite over Rfi by assumption. Thus S is finite over R by Lemma 23.2.

032I

Lemma 155.5. Let R be a domain. Let R S be a quasi-finite extension of

domains (for example finite). Assume R is N-2 and Noetherian. Then S is N-2.
Proof. Let K = f.f.(R) L = f.f.(S). Note that this is a finite field extension
(for example by Lemma 121.2 (2) applied to the fibre S R K, and the definition
of a quasi-finite ring map). Let S 0 be the integral closure of R in S. Then S 0 is
contained in the integral closure of R in L which is finite over R by assumption.
As R is Noetherian this implies S 0 is finite over R. By Lemma 122.15 there exist
elements g1 , . . . , gn S 0 such that Sg0 i
= Sgi and such that g1 , . . . , gn generate the
unit ideal in S. Hence it suffices to show that S 0 is N-2 by Lemmas 155.3 and 155.4.
Thus we have reduced to the case where S is finite over R.
Assume R S with hypotheses as in the lemma and moreover that S is finite over
R. Let M be a finite field extension of the fraction field of S. Then M is also a
finite field extension of f.f (R) and we conclude that the integral closure T of R in
M is finite over R. By Lemma 35.14 we see that T is also the integral closure of S
in M and we win by Lemma 35.13.


032J

Lemma 155.6. Let R be a Noetherian domain. If R[z, z 1 ] is N-1, then so is R.

Proof. Let R0 be the integral closure of R in its field of fractions K. Let S 0 be the
integral closure of R[z, z 1 ] in its field of fractions. Clearly R0 S 0 . Since K[z, z 1 ]
1
0
is a normal domain we see that S 0 K[z,
P z ].j Suppose that f1 , . . . , fn S
0
1
generate S as R[z, z ]-module. Say fi =
aij z (finite sum), with aij K. For
any x R0 we can write
X
x=
hi fi
1
0
with
the finite R-submodule
P hi R[z, z ]. Thus we see that R is contained in
Raij K. Since R is Noetherian we conclude that R0 is a finite R-module. 

032K

Lemma 155.7. Let R be a Noetherian domain, and let R S be a finite extension

of domains. If S is N-1, then so is R. If S is N-2, then so is R.
Proof. Omitted. (Hint: Integral closures of R in extension fields are contained in
integral closures of S in extension fields.)


032L

Lemma 155.8. Let R be a Noetherian normal domain with fraction field K. Let
K L be a finite separable field extension. Then the integral closure of R in L is
finite over R.
Proof. Consider the trace pairing (Fields, Definition 19.6)
L L K,

(x, y) 7 hx, yi := TraceL/K (xy).

Since L/K is separable this is nondegenerate (Fields, Lemma 19.7). Moreover, if

x L is integral over R, then TraceL/K (x) is integral over R also, and since R is
normal we see TraceL/K (x) R. Pick x1 , . . . , xn L which are integral over R
and which form a K-basis of L. Then the integral closure S L is contained in
the R-module
M = {y L | hxi , yi R, i = 1, . . . , n}

COMMUTATIVE ALGEBRA

404

By linear algebra we see that M

= Rn as an R-module. Hence S Rn is a
finitely generated R-module as R is Noetherian.

03B7

Example 155.9. Lemma 155.8 does not work if the ring is not Noetherian. For
example consider the action of G = {+1, 1} on A = C[x1 , x2 , x3 , . . .] where 1
acts by mapping xi to xi . The invariant ring R = AG is the C-algebra generated
by all xi xj . Hence R A is not finite. But R is a normal domain with fraction
field K = LG the G-invariants in the fraction field L of A. And clearly A is the
integral closure of R in L.
The following lemma can sometimes be used as a substitute for Lemma 155.8 in
case of purely inseparable extensions.

0AE0

Lemma 155.10. Let R be a Noetherian normal domain with fraction field K of

characteristic p > 0. Let a K be an element such that there exists a derivation
D : R R with D(a) 6= 0. Then the integral closure of R in L = K[x]/(xp a) is
finite over R.
Proof. After replacing x by f x and a by f p a for some f R we may assume
a R. Hence also D(a) R. We will show by induction on i p 1 that if
y = a0 + a1 x + . . . + ai xi ,

aj K

is integral over R, then D(a) aj R. Thus the integral closure is contained in the
finite R-module with basis D(a)p+1 xj , j = 0, . . . , p 1. Since R is Noetherian
this proves the lemma.
If i = 0, then y = a0 is integral over R if and only if a0 R and the statement is
true. Suppose the statement holds for some i < p 1 and suppose that
y = a0 + a1 x + . . . + ai+1 xi+1 ,

aj K

is integral over R. Then

y p = ap0 + ap1 a + . . . + api+1 ai+1
is an element of R (as it is in K and integral over R). Applying D we obtain
(ap1 + 2ap2 a + . . . + (i + 1)api+1 ai )D(a)
is in R. Hence it follows that
D(a)a1 + 2D(a)a2 x + . . . + (i + 1)D(a)ai+1 xi
is integral over R. By induction we find D(a)i+1 aj R for j = 1, . . . , i + 1. (Here
we use that 1, . . . , i + 1 are
Hence D(a)i+1 a0 is also in R because it
P invertible.)
i+1
is the difference of y and j>0 D(a) aj xj which are integral over R (since x is
integral over R as a R).

032M

Lemma 155.11. A Noetherian domain of characteristic zero is N-1 if and only

if it is N-2 (i.e., Japanese).
Proof. This is clear from Lemma 155.8 since every field extension in characteristic
zero is separable.


032N

Lemma 155.12. Let R be a Noetherian domain with fraction field K of characteristic p > 0. Then R is N-2 if and only if for every finite purely inseparable
extension K L the integral closure of R in L is finite over R.

COMMUTATIVE ALGEBRA

405

Proof. Assume the integral closure of R in every finite purely inseparable field
extension of K is finite. Let K L be any finite extension. We have to show the
integral closure of R in L is finite over R. Choose a finite normal field extension
K M containing L. As R is Noetherian it suffices to show that the integral closure
of R in M is finite over R. By Fields, Lemma 26.3 there exists a subextension
K Minsep M such that Minsep /K is purely inseparable, and M/Minsep is
separable. By assumption the integral closure R0 of R in Minsep is finite over R.
By Lemma 155.8 the integral closure R00 of R0 in M is finite over R0 . Then R00 is
finite over R by Lemma 7.3. Since R00 is also the integral closure of R in M (see
Lemma 35.14) we win.

032O

Lemma 155.13. Let R be a Noetherian domain. If R is N-1 then R[x] is N-1. If

R is N-2 then R[x] is N-2.
Proof. Assume R is N-1. Let R0 be the integral closure of R which is finite over
R. Hence also R0 [x] is finite over R[x]. The ring R0 [x] is normal (see Lemma 36.8),
hence N-1. This proves the first assertion.
For the second assertion, by Lemma 155.7 it suffices to show that R0 [x] is N-2. In
other words we may and do assume that R is a normal N-2 domain. In characteristic
zero we are done by Lemma 155.11. In characteristic p > 0 we have to show that
the integral closure of R[x] is finite in any finite purely inseparable extension of
f.f.(R[x]) = K(x) L with K = f.f.(R). Clearly there exists a finite purely
inseparable field extension K L0 and q = pe such that L L0 (x1/q ). As R[x] is
Noetherian it suffices to show that the integral closure of R[x] in L0 (x1/q ) is finite
over R[x]. And this integral closure is equal to R0 [x1/q ] with R R0 L0 the
integral closure of R in L0 . Since R is N-2 we see that R0 is finite over R and hence
R0 [x1/q ] is finite over R[x].


0332

Lemma 155.14. Let R be a Noetherian domain. If there exists an f R such

that Rf is normal then
U = {p Spec(R) | Rp is normal}
is open in Spec(R).
Proof. It is clear that the standard open D(f ) is contained in U . By Serres
criterion Lemma 151.4 we see that p 6 U implies that for some q p we have
either
(1) Case I: depth(Rq ) < 2 and dim(Rq ) 2, and
(2) Case II: Rq is not regular and dim(Rq ) = 1.
This in particular also means that Rq is not normal, and hence f q. In case I we
see that depth(Rq ) = depth(Rq /f Rq ) + 1. Hence such a prime q is the same thing
as an embedded associated prime of R/f R. In case II q is an associated prime of
R/f R of height 1. Thus there is a finite set E of such primes q (see Lemma 62.5)
and
[
Spec(R) \ U =
V (q)
qE

as desired.
0333

Lemma 155.15. Let R be a Noetherian domain. Assume

(1) there exists a nonzero f R such that Rf is normal, and
(2) for every maximal ideal m R the local ring Rm is N-1.

COMMUTATIVE ALGEBRA

406

Then R is N-1.
Proof. Set K = f.f.(R). Suppose that R R0 K is a finite extension of R
contained in K. Note that Rf = Rf0 since Rf is already normal. Hence by Lemma
155.14 the set of primes p0 Spec(R0 ) with Rp0 0 non-normal is closed in Spec(R0 ).
Since Spec(R0 ) Spec(R) is closed the image of this set is closed in Spec(R). For
such a ring R0 denote ZR0 Spec(R) this image.
0
Pick a maximal ideal m R. Let Rm Rm
be the integral closure of the local ring
in K. By assumption this is a finite ring extension. By Lemma 35.9 we can find
0
finitely many elements r1 , . . . , rn K integral over R such that Rm
is generated by
0
r1 , . . . , rn over Rm . Let R = R[x1 , . . . , xn ] K. With this choice it is clear that
m 6 ZR0 .

As Spec(R) is quasi-compact,
the above shows that we can find a finite collection
T
R Ri0 K such that ZRi0 = . Let R0 be the subring of K generated by all of
these. It is finite over R. Also ZR0 = . Namely, every prime p0 lies over a prime
p0i such that (Ri0 )p0i is normal. This implies that Rp0 0 = (Ri0 )p0i is normal too. Hence
R0 is normal, in other words R0 is the integral closure of R in K.

032P

Lemma 155.16 (Tate). Let R be a ring. Let x R. Assume

(1) R is a normal Noetherian domain,
(2) R/xR is a domain and N-2,
(3) R
= limn R/xn R is complete with respect to x.
Then R is N-2.
Proof. We may assume x 6= 0 since otherwise the lemma is trivial. Let K be the
fraction field of R. If the characteristic of K is zero the lemma follows from (1), see
Lemma 155.11. Hence we may assume that the characteristic of K is p > 0, and we
may apply Lemma 155.12. Thus given K L be a finite purely inseparable field
extension we have to show that the integral closure S of R in L is finite over R.
Let q be a power of p such that Lq K. By enlarging L if necessary we may
assume there exists an element y L such that y q = x. Since R S induces a
homeomorphism of spectra (see Lemma 45.6) there is a unique prime ideal q S
lying over the prime ideal p = xR. It is clear that
q = {f S | f q p} = yS
since y q = x. Hence Rp and Sq are discrete valuation rings, see Lemma 118.7.
By Lemma 118.10 we see that (p) (q) is a finite field extension. Hence the
integral closure S 0 (q) of R/xR is finite over R/xR by assumption (2). Since
S/yS S 0 this implies that S/yS is finite over R. Note that S/y n S has a finite
filtration whose subquotients are the modules y i S/y i+1 S
= S/yS. Hence we see
that each T
S/y n S is finite over R. In particular S/xS is finite over R. Also, it is
clear
xn S = (0) since an element in the intersection has qth power contained
T that
n
in x R = (0) (Lemma 50.4). Thus we may apply Lemma 95.12 to conclude that
S is finite over R, and we win.


032Q

Lemma 155.17. Let R be a ring. If R is Noetherian, a domain, and N-2, then

so is R[[x]].

COMMUTATIVE ALGEBRA

407

Proof. Observe that R[[x]] is Noetherian by Lemma 30.2. Let R0 R be the

integral closure of R in its fraction field. Because R is N-2 this is finite over R.
Hence R0 [[x]] is finite over R[[x]]. By Lemma 36.9 we see that R0 [[x]] is a normal
domain. Apply Lemma 155.16 to the element x R0 [[x]] to see that R0 [[x]] is N-2.
Then Lemma 155.7 shows that R[[x]] is N-2.

156. Nagata rings
032E

Here is the definition.

032R

Definition 156.1. Let R be a ring.

(1) We say R is universally Japanese if for any finite type ring map R S
with S a domain we have that S is N-2 (i.e., Japanese).
(2) We say that R is a Nagata ring if R is Noetherian and for every prime ideal
p the ring R/p is N-2.
It is clear that a Noetherian universally Japanese ring is a Nagata ring. It is our
goal to show that a Nagata ring is universally Japanese. This is not obvious at all,
and requires some work. But first, here is a useful lemma.

03GH

Lemma 156.2. Let R be a Nagata ring. Let R S be essentially of finite type

with S reduced. Then the integral closure of R in S is finite over R.
Proof. As S is essentially of finite type over R it is Noetherian and has finitely
manyQminimal primes q1 , . . . , qm , see Lemma 30.6. Since S is reduced we have
S
Sqi and each Sqi = Ki is a field, see Lemmas 24.4 and 24.1. It suffices
to show that the integral closure Q
A0i of R in each Ki is finite over R. This is true
because R is Noetherian and A A0i . Let pi R be the prime of R corresponding
to qi . As S is essentially of finite type over R we see that Ki = Sqi = (qi ) is a
finitely generated field extension of (pi ). Hence the algebraic closure Li of (pi ) in
Ki is finite over (pi ), see Fields, Lemma 25.10. It is clear that A0i is the integral
closure of R/pi in Li , and hence we win by definition of a Nagata ring.


0351

Lemma 156.3. Let R be a ring. To check that R is universally Japanese it suffices

to show: If R S is of finite type, and S a domain then S is N-1.
Proof. Namely, assume the condition of the lemma. Let R S be a finite type
ring map with S a domain. Let f.f.(S) L be a finite extension of its fraction
field. Then there exists a finite ring extension S S 0 L with f.f.(S 0 ) = L. By
assumption S 0 is N-1, and hence the integral closure S 00 of S 0 in L is finite over
S 0 . Thus S 00 is finite over S (Lemma 7.3) and S 00 is the integral closure of S in L
(Lemma 35.14). We conclude that R is universally Japanese.


032S

Lemma 156.4. If R is universally Japanese then any algebra essentially of finite

type over R is universally Japanese.
Proof. The case of an algebra of finite type over R is immediate from the definition.
The general case follows on applying Lemma 155.3.


032T

Lemma 156.5. Let R be a Nagata ring. If R S is a quasi-finite ring map (for

example finite) then S is a Nagata ring also.

COMMUTATIVE ALGEBRA

408

Proof. First note that S is Noetherian as R is Noetherian and a quasi-finite ring

map is of finite type. Let q S be a prime ideal, and set p = R q. Then
R/p S/q is quasi-finite and hence we conclude that S/q is N-2 by Lemma 155.5
as desired.

032U

Lemma 156.6. A localization of a Nagata ring is a Nagata ring.

Proof. Clear from Lemma 155.3.

032V

Lemma 156.7. Let R be a ring. Let f1 , . . . , fn R generate the unit ideal.

(1) If each Rfi is universally Japanese then so is R.
(2) If each Rfi is Nagata then so is R.
Proof. Let : R S be a finite type ring map so that S is a domain. Then
(f1 ), . . . , (fn ) generate the unit ideal in S. Hence if each Sfi = S(fi ) is N-1
then so is S, see Lemma 155.4. This proves (1).
If each Rfi is Nagata, then each Rfi is Noetherian and hence R is Noetherian, see
Lemma 23.2. And if p R is a prime, then we see each Rfi /pRfi = (R/p)fi is N-2
and hence we conclude R/p is N-2 by Lemma 155.4. This proves (2).


032W

Lemma 156.8. A Noetherian complete local ring is a Nagata ring.

Proof. Let R be a complete local Noetherian ring. Let p R be a prime. Then
R/p is also a complete local Noetherian ring, see Lemma 154.2. Hence it suffices
to show that a Noetherian complete local domain R is N-2. By Lemmas 155.5
and 154.10 we reduce to the case R = k[[X1 , . . . , Xd ]] where k is a field or R =
[[X1 , . . . , Xd ]] where is a Cohen ring.
In the case k[[X1 , . . . , Xd ]] we reduce to the statement that a field is N-2 by Lemma
155.17. This is clear. In the case [[X1 , . . . , Xd ]] we reduce to the statement that a
Cohen ring is N-2. Applying Lemma 155.16 once more with x = p we reduce
yet again to the case of a field. Thus we win.


032X

Definition 156.9. Let (R, m) be a Noetherian local ring. We say R is analytically

unramified if its completion R = limn R/mn is reduced. A prime ideal p R is
said to be analytically unramified if R/p is analytically unramified.
At this point we know the following are true for any Noetherian local ring R: The
map R R is a faithfully flat local ring homomorphism (Lemma 96.3). The
completion R is Noetherian (Lemma 96.5) and complete (Lemma 96.4). Hence
the completion R is a Nagata ring (Lemma 156.8). Moreover, we have seen in
Section 154 that R is a quotient of a regular local ring (Theorem 154.8), and
hence universally catenary (Remark 154.9).

032Y

Lemma 156.10. Let (R, m) be a Noetherian local ring.

(1) If R is analytically unramified, then R is reduced.
(2) If R is analytically unramified, then each minimal prime of R is analytically
unramified.
(3) If R is reduced with minimal primes q1 , . . . , qt , and each qi is analytically
unramified, then R is analytically unramified.
(4) If R is analytically unramified, then the integral closure of R in its total
ring of fractions Q(R) is finite over R.
(5) If R is a domain and analytically unramified, then R is N-1.

COMMUTATIVE ALGEBRA

409

Proof. In this proof we will use the remarks immediately following Definition 156.9.
As R R is a faithfully flat local ring homomorphism it is injective and (1)
follows.
Let q be a minimal prime of R, and assume R is analytically unramified. Then q
is an associated prime of R (see Proposition 62.6). Hence there exists an f R
such that {x R | f x = 0} = q. Note that (R/q) = R /q , and that {x R |
f x = 0} = q , because completion is exact (Lemma 96.2). If x R is such that
x2 q , then f x2 = 0 hence (f x)2 = 0 hence f x = 0 hence x q . Thus q is
analytically unramified and (2) holds.
Assume R is reduced with minimal primes q1 , . . . , qt , and each qi is analytically
unramified. Then R R/q1 . . . R/qt is injective. Since completion is exact
(see Lemma 96.2) we see that R (R/q1 ) . . . (R/qt ) . Hence (3) is clear.
Assume R is analytically unramified. Let p1 , . . . , ps be the minimal primes of R .
Then we see that
Q(R ) = Rp1 . . . Rps
with each Rpi a field as R is reduced (see Lemma 24.4). Hence the integral closure
S of R in Q(R ) is equal to S = S1 . . . Ss with Si the integral closure of R /pi
in its fraction field. In particular S is finite over R . Denote R0 the integral closure
of R in Q(R). As R R is flat we see that R0 R R Q(R) R R Q(R ).
Moreover R0 R R is integral over R (Lemma 35.11). Hence R0 R R S is
a R -submodule. As R is Noetherian it is a finite R -module. Thus we may
find f1 , . . . , fn R0 such that R0 R R is generated by the elements fi 1 as a
R -module. By faithful flatness we see that R0 is generated by f1 , . . . , fn as an
R-module. This proves (4).
Part (5) is a special case of part (4).
032Z

Lemma 156.11. Let R be a Noetherian local ring. Let p R be a prime. Assume

(1) Rp is a discrete valuation ring, and
(2) p is analytically unramified.
Then for any associated prime q of R /pR the local ring (R )q is a discrete
valuation ring.
Proof. Assumption (2) says that R /pR is a reduced ring. Hence an associated
prime q R of R /pR is the same thing as a minimal prime over pR . In
particular we see that the maximal ideal of (R )q is p(R )q . Choose x R such
that xRp = pRp . By the above we see that x (R )q generates the maximal ideal.
As R R is faithfully flat we see that x is a nonzerodivisor in (R )q . Hence we
win.


0330

Lemma 156.12. Let (R, m) be a Noetherian local domain. Let x m. Assume

(1) x 6= 0,
(2) R/xR has no embedded primes, and
(3) for each associated prime p R of R/xR we have
(a) the local ring Rp is regular, and
(b) p is analytically unramified.
Then R is analytically unramified.

COMMUTATIVE ALGEBRA

410

Proof. Let p1 , . . . , pt be the associated primes of the R-module R/xR. Since R/xR
has no embedded primes we see that each pi has height 1, and is a minimal prime
over (x). For each i, let qi1 , . . . , qisi be the associated primes of the R -module
R /pi R . By Lemma 156.11 we see that (R )qij is regular. By Lemma 64.3 we
see that
[
AssR (R /xR ) =
AssR (R /pR ) = {qij }.
pAssR (R/xR)

Let y R with y = 0. As (R )qij is regular, and hence a domain (Lemma 105.2)

we see that y maps to zero in (R )qij . Hence y maps to zero in R /xR by Lemma
62.19. Hence y = xy 0 . Since x is a nonzerodivisor
(as R R is flat) we see that
T n
0 2
(y ) = 0. Hence we conclude that y x R = (0) (Lemma 50.4).

0331

Lemma 156.13. Let (R, m) be a local ring. If R is Noetherian, a domain, and

Nagata, then R is analytically unramified.
Proof. By induction on dim(R). The case dim(R) = 0 is trivial. Hence we assume dim(R) = d and that the lemma holds for all Noetherian Nagata domains of
dimension < d.
Let R S be the integral closure of R in the field of fractions of R. By assumption
S is a finite R-module. By Lemma 156.5 we see that S is Nagata. By Lemma 111.4
we see dim(R) = dim(S). Let m1 , . . . , mt be the maximal ideals of S. Each of these
lies over the maximal ideal m of R. Moreover
(m1 . . . mt )n mS
for sufficiently large n as S/mS is Artinian. By Lemma 96.2 R S is an injective
map, and by
Q the Chinese Remainder Lemma 14.3 combined with Lemma 95.9 we
have S = Si where Si is the completion of S with respect to the maximal ideal
mi . Hence it suffices to show that Smi is analytically unramified. In other words,
we have reduced to the case where R is a Noetherian normal Nagata domain.
Assume R is a Noetherian, normal, local Nagata domain. Pick a nonzero x m
in the maximal ideal. We are going to apply Lemma 156.12. We have to check
properties (1), (2), (3)(a) and (3)(b). Property (1) is clear. We have that R/xR
has no embedded primes by Lemma 151.6. Thus property (2) holds. The same
lemma also tells us each associated prime p of R/xR has height 1. Hence Rp is
a 1-dimensional normal domain hence regular (Lemma 118.7). Thus (3)(a) holds.
Finally (3)(b) holds by induction hypothesis, since R/p is Nagata (by Lemma 156.5
or directly from the definition). Thus we conclude R is analytically unramified. 

0BI2

Lemma 156.14. Let (R, m) be a Noetherian local ring. The following are equivalent
(1) R is Nagata,
(2) for R S finite with S a domain and m0 S maximal the local ring Sm0
is analytically unramified,
(3) for (R, m) (S, m0 ) finite local homomorphism with S a domain, then S
is analytically unramified.
Proof. Assume R is Nagata and let R S and m0 S be as in (2). Then S is
Nagata by Lemma 156.5. Hence the local ring Sm0 is Nagata (Lemma 156.6). Thus
it is analytically unramified by Lemma 156.13. It is clear that (2) implies (3).

COMMUTATIVE ALGEBRA

411

Assume (3) holds. Let p R be a prime ideal and let f.f.(R/p) L be a finite
extension of fields. To prove (1) we have to show that the integral closure of R/p is
finite over R/p. Choose x1 , . . . , xn L which generate L over f.f.(R/p). For each
i let Pi (T ) = T di + ai,1 T di 1 + . . . + ai,di be the minimal polynomial for xi over
f.f.(R/p). After replacing xi by fi xi for a suitable fi R, fi 6 p we may assume
ai,j R/p. In fact, after further multiplying by elements of m, we may assume ai,j
m/p R/p for all i, j. Having done this let S = R/p[x1 , . . . , xn ] L. Then S is
finite over R, a domain, and S/mS is a quotient of R/m[T1 , . . . , Tn ]/(T1d1 , . . . , Tndn ).
Hence S is local. By (3) S is analytically unramified and by Lemma 156.10 we find
that its integral closure S 0 in L is finite over S. Since S 0 is also the integral closure
of R/p in L we win.

The following proposition says in particular that an algebra of finite type over a
Nagata ring is a Nagata ring.
0334

Proposition 156.15 (Nagata). Let R be a ring. The following are equivalent:

(1) R is a Nagata ring,
(2) any finite type R-algebra is Nagata, and
(3) R is universally Japanese and Noetherian.
Proof. It is clear that a Noetherian universally Japanese ring is universally Nagata
(i.e., condition (2) holds). Let R be a Nagata ring. We will show that any finitely
generated R-algebra S is Nagata. This will prove the proposition.
Step 1. There exists a sequence of ring maps R = R0 R1 R2 . . . Rn = S
such that each Ri Ri+1 is generated by a single element. Hence by induction it
suffices to prove S is Nagata if S
= R[x]/I.
Step 2. Let q S be a prime of S, and let p R be the corresponding prime
of R. We have to show that S/q is N-2. Hence we have reduced to the proving
the following: (*) Given a Nagata domain R and a monogenic extension R S of
domains then S is N-2.
Step 3. Let R be a Nagata domain and R S a monogenic extension of domains.
Let R R0 be the integral closure of R in its fraction field. Let S 0 be the subring
of f.f.(S) generated by R0 and S. As R0 is finite over R (by the Nagata property)
also S 0 is finite over S. Since S is Noetherian it suffices to prove that S 0 is N-2
(Lemma 155.7). Hence we have reduced to proving the following: (**) Given a
normal Nagata domain R and a monogenic extension R S of domains then S is
N-2.
Step 4: Let R be a normal Nagata domain and let R S be a monogenic extension
of domains. Suppose the extension of fraction fields f.f.(R) f.f.(S) is purely
transcendental. In this case S = R[x]. By Lemma 155.13 we see that S is N2. Hence we have reduced to proving the following: (**) Given a normal Nagata
domain R and a monogenic extension R S of domains inducing a finite extension
of fraction fields then S is N-2.
Step 5. Let R be a normal Nagata domain and let R S be a monogenic extension
of domains inducing a finite extension of fraction fields K = f.f.(R) f.f.(S) = L.
Choose an element x S which generates S as an R-algebra. Let L M be a
finite extension of fields. Let R0 be the integral closure of R in M . Then the
integral closure S 0 of S in M is equal to the integral closure of R0 [x] in M . Also

COMMUTATIVE ALGEBRA

412

f.f.(R0 ) = M , and R R0 is finite (by the Nagata property of R). This implies
that R0 is a Nagata ring (Lemma 156.5). To show that S 0 is finite over S is the
same as showing that S 0 is finite over R0 [x]. Replace R by R0 and S by S 0 to reduce
to the following statement: (***) Given a normal Nagata domain R with fraction
field K, and x K, the ring S K generated by R and x is N-1.
Step 6. Let R be a normal Nagata domain with fraction field K. Let x = b/a K.
We have to show that the ring S K generated by R and x is N-1. Note that
Sa
= Ra is normal. Hence by Lemma 155.15 it suffices to show that Sm is N-1 for
every maximal ideal m of S.
With assumptions as in the preceding paragraph, pick such a maximal ideal and
set n = R m. The residue field extension (n) (m) is finite (Theorem 33.1)
and generated
by the image of x. Hence there exists a monic polynomial f (X) =
P
X d + i=1,...,d ai X di with f (x) m. Let K K 00 be a finite extension of fields
such that f (X) splits completely in K 00 [X]. Let R0 be the integral closure of R in
K 00 . Let S 0 K 0 be the subring generated by R0 and x. As R is Nagata we see
R0 is finite over R and Nagata (Lemma 156.5). Moreover, S 0 is finite over S. If for
0
0
every maximal ideal m0 of S 0 the local ring Sm
0 is N-1, then Sm is N-1 by Lemma
155.15, which in turn implies that Sm is N-1 by Lemma 155.7. After replacing R by
R0 and S by S 0 , and m by any of the maximal ideals m0 lying overQm we reach the
situation where the polynomial f above split completely: f (X) = i=1,...,d (X ai )
with ai R. Since f (x) m we see that x ai m for some i. Finally, after
replacing x by x ai we may assume that x m.
To recapitulate: R is a normal Nagata domain with fraction field K, x K and S
is the subring of K generated by x and R, finally m S is a maximal ideal with
x m. We have to show Sm is N-1.
We will show that Lemma 156.12 applies to the local ring Sm and the element x.
This will imply that Sm is analytically unramified, whereupon we see that it is N-1
by Lemma 156.10.
We have to check properties (1), (2), (3)(a) and (3)(b). Property (1) is trivial. Let
I = Ker(R[X] S) where X 7 x. We claim that I is generated by all linear
forms aX + b such that ax = b in K. Clearly all these linear forms are in I. If
g = ad X d +. . . a1 X +a0 I, then we see that ad x is integral over R (Lemma 122.1)
and hence b := ad x R as R is normal. Then g (ad X b)X d1 I and we win
by induction on the degree. As a consequence we see that
S/xS = R[X]/(X, I) = R/J
where
J = {b R | ax = b for some a R} = xR R
By Lemma 151.6 we see that S/xS = R/J has no embedded primes as an R-module,
hence as an R/J-module, hence as an S/xS-module, hence as an S-module. This
proves property (2). Take such an associated prime q S with the property q m
(so that it is an associated prime of Sm /xSm it does not matter for the arguments).
Then q is minimal over xS and hence has height 1. By the sequence of equalities
above we see that p = R q is an associated prime of R/J, and so has height 1
(see Lemma 151.6). Thus Rp is a discrete valuation ring and therefore Rp Sq is
an equality. This shows that Sq is regular. This proves property (3)(a). Finally,
(S/q)m is a localization of S/q, which is a quotient of S/xS = R/J. Hence (S/q)m

COMMUTATIVE ALGEBRA

413

is a localization of a quotient of the Nagata ring R, hence Nagata (Lemmas 156.5

and 156.6) and hence analytically unramified (Lemma 156.13). This shows (3)(b)
holds and we are done.

0335

Proposition 156.16. The following types of rings are Nagata and in particular
universally Japanese:
(1) fields,
(2) Noetherian complete local rings,
(3) Z,
(4) Dedekind domains with fraction field of characteristic zero,
(5) finite type ring extensions of any of the above.
Proof. The Noetherian complete local ring case is Lemma 156.8. In the other
cases you just check if R/p is N-2 for every prime ideal p of the ring. This is clear
whenever R/p is a field, i.e., p is maximal. Hence for the Dedekind ring case we
only need to check it when p = (0). But since we assume the fraction field has
characteristic zero Lemma 155.11 kicks in.


09E1

Example 156.17. A discrete valuation ring is Nagata if and only if it is N-2 (this
follows immediately from the definition). The discrete valuation ring A of Example
118.5 is not Nagata, i.e., it is notPN-2. Namely, the finite extensionPA R = A[f ]
is not N-1. To see this say f =
ai xi . For every n 1 set gn = i<n ai xi A.
n
Then hn = (f gn )/x is an element of the fraction field of R and hpn k p [[x]] A.
Hence the integral closure R0 of R contains h1 , h2 , h3 , . . .. Now, if R0 were finite over
R and hence A, then f = xn hn + gn would be contained in the submodule A + xn R0
for all n. By Artin-Rees this would imply f A (Lemma 50.4), a contradiction.

09E2

Lemma 156.18. Let (A, m) be a Noetherian local domain which is Nagata and
has fraction field of characteristic p. If a A has a pth root in A , then a is has
a pth root in A.
Proof. Consider the ring extension A B = A[x]/(xp a). If a does not have a
pth root in A, then B is a domain whose completion isnt reduced. This contradicts
our earlier results, as B is a Nagata (Proposition 156.15) and hence analytically
unramified by Lemma 156.13.

157. Ascending properties

0336

In this section we start proving some algebraic facts concerning the ascent of
properties of rings. To do this for depth of rings one uses the following result on
ascending depth of modules, see [DG67, IV, Proposition 6.3.1].

0338

Lemma 157.1. We have

depth(M R N ) = depth(M ) + depth(N/mR N )
where R S is a local homomorphism of local Noetherian rings, M is a finite
R-module, and N is a finite S-module flat over R.
Proof. In the statement and in the proof below, we take the depth of M as an
R-module, the depth of M R N as an S-module, and the depth of N/MR N as
an S/mR S-module. Denote n the right hand side. First assume that n is zero.
Then both depth(M ) = 0 and depth(N/mR N ) = 0. This means there is a z M
whose annihilator is mR and a y N/mR N whose annihilator is mS /mR S. Let

[DG67, IV,
Proposition 6.3.1]

COMMUTATIVE ALGEBRA

414

y N be a lift of y. Since N is flat over R the map z : R/mR M produces an

injective map N/mR N M R N . Hence the annihilator of z y is mS . Thus
depth(M R N ) = 0 as well.
Assume n > 0. If depth(N/mR N ) > 0, then choose an f mS which maps to an
N/mR N -regular element f S/mR S. Then depth(N/mR N ) = depth(N/(f, mR )N )+
1 by Lemma 71.7. According to Lemma 98.1 the element f S is a N -regular element and N/f N is flat over R. Hence by induction on n we have
depth(M R N/f N ) = depth(M ) + depth(N/(f, mR )N ).
Because N/f N is flat over R the sequence
0 M R N M R N M R N/f N 0
is exact where the first map is multiplication by f (Lemma 38.12). Hence by Lemma
71.7 we find that depth(M R N ) = depth(M R N/f N ) + 1 and we conclude that
equality holds in the formula of the lemma.
If n > 0, but depth(N/mR N ) = 0, then we can choose an M -regular element
f mR . As N is flat over R it is also the case that f is M R N -regular. By
induction on n again we have
depth(M/f M R N ) = depth(M/f M ) + depth(N/mR N ).
In this case depth(M R N ) = depth(M/f M R N )+1 and depth(M ) = depth(M/f M )+
1 by Lemma 71.7 and we conclude that equality holds in the formula of the
lemma.

0337

Lemma 157.2. Suppose that R S is a flat and local ring homomorphism of

Noetherian local rings. Then
depth(S) = depth(R) + depth(S/mR S).
Proof. This is a special case of Lemma 157.1.

045J

Lemma 157.3. Let R S be a local homomorphism of local Noetherian rings.

Assume
(1) S/mR S is Cohen-Macaulay, and
(2) R S is flat.
Then S is Cohen-Macaulay if and only if R is Cohen-Macaulay.
Proof. This follows from the definitions combined with Lemmas 157.2 and 111.7.


0339

Lemma 157.4. Let : R S be a ring map. Assume

(1)
(2)
(3)
(4)
(5)

R is Noetherian,
S is Noetherian,
is flat,
the fibre rings S R (p) are (Sk ), and
R has property (Sk ).

Then S has property (Sk ).

COMMUTATIVE ALGEBRA

415

Proof. Let q be a prime of S lying over a prime p of R. By Lemma 157.2 we have

depth(Sq ) = depth(Sq /pSq ) + depth(Rp ).
On the other hand, we have
dim(Rp ) + dim(Sq /pSq ) dim(Sq )
by Lemma 111.6. (Actually equality holds, by Lemma 111.7 but strictly speaking
we do not need this.) Finally, as the fibre rings of the map are assumed (Sk ) we
see that depth(Sq /pSq ) min(k, dim(Sq /pSq )). Thus the lemma follows by the
following string of inequalities
depth(Sq )

depth(Sq /pSq ) + depth(Rp )

min(k, dim(Sq /pSq )) + min(k, dim(Rp ))

=

min(2k, dim(Sq /pSq ) + k, k + dim(Rp ), dim(Sq /pSq ) + dim(Rp ))

min(k, dim(Sq ))
as desired.
033A

Lemma 157.5. Let : R S be a ring map. Assume

(1) R is Noetherian,
(2) S is Noetherian
(3) is flat,
(4) the fibre rings S R (p) are regular, and
(5) R has property (Rk ).
Then S has property (Rk ).
Proof. Let q be a prime of S lying over a prime p of R. Assume that dim(Sq ) k.
Since dim(Sq ) = dim(Rp ) + dim(Sq /pSq ) by Lemma 111.7 we see that dim(Rp )
k. Hence Rp is regular by assumption. It follows that Sq is regular by Lemma
111.8.


033B

Lemma 157.6. Let : R S be a ring map. Assume

(1) is smooth,
(2) R is reduced.
Then S is reduced.
Proof. First assume R is Noetherian. In this case being reduced is the same as
having properties (S1 ) and (R0 ), see Lemma 151.3. Note that S is Noetherian, and
R S is flat with regular fibres (see the list of results on smooth ring maps in
Section 140). Hence we may apply Lemmas 157.4 and 157.5 and we see that S is
(S1 ) and (R0 ), in other words reduced by Lemma 151.3 again.
In the general case we may find a finitely generated Z-subalgebra R0 R and a
smooth ring map R0 S0 such that S
= R R0 S0 , see remark (10) in Section 140.
Now, if x S is an element with x2 = 0, then we can enlarge R0 and assume that
x comes from an element x0 S0 . After enlarging R0 once more we may assume
that x20 = 0 in S0 . However, since R0 R is reduced we see that S0 is reduced and
hence x0 = 0 as desired.


033C

Lemma 157.7. Let : R S be a ring map. Assume

(1) is smooth,
(2) R is normal.

COMMUTATIVE ALGEBRA

416

Then S is normal.
Proof. First assume R is Noetherian. In this case being normal is the same as
having properties (S2 ) and (R1 ), see Lemma 151.4. Note that S is Noetherian, and
R S is flat with regular fibres (see the list of results on smooth ring maps in
Section 140). Hence we may apply Lemmas 157.4 and 157.5 and we see that S is
(S2 ) and (R1 ), in other words normal by Lemma 151.4 again.
The general case. First note that R is reduced and hence S is reduced by Lemma
157.6. Let q be a prime of S and let p be the corresponding prime of R. Note that
Rp is a normal domain. We have to show that Sq is a normal domain. To do this
we may replace R by Rp and S by Sp . Hence we may assume that R is a normal
domain.
Assume R S smooth, and R a normal domain. We may find a finitely generated
Z-subalgebra R0 R and a smooth ring map R0 S0 such that S
= R R0 S0 , see
remark (10) in Section 140. As R0 is a Nagata domain (see Proposition 156.16) we
see that its integral closure R00 is finite over R0 . Moreover, as R is a normal domain
it is clear that R00 R. Hence we may replace R0 by R00 and S0 by R00 R0 S0
and assume that R0 is a normal Noetherian domain. By the first paragraph of the
proof we conclude that S0 isSa normal ring (it need not be a domain of course).
In this way we see that R = R is the union of normal Noetherian domains and
correspondingly S = colim R R0 S0 is the colimit of normal rings. This implies
that S is a normal ring. Some details omitted.

07NF

Lemma 157.8. Let : R S be a ring map. Assume

(1) is smooth,
(2) R is a regular ring.
Then S is regular.
Proof. This follows from Lemma 157.5 applied for all (Rk ) using Lemma 138.3 to
see that the hypotheses are satisfied.

158. Descending properties

033D

In this section we start proving some algebraic facts concerning the descent of
properties of rings. It turns out that it is often easier to descend properties than
it is to ascend them. In other words, the assumption on the ring map R S are
often weaker than the assumptions in the corresponding lemma of the preceding
section. However, we warn the reader that the results on descent are often useless
unless the corresponding ascent can also be shown! Here is a typical result which
illustrates this phenomenon.

033E

Lemma 158.1. Let R S be a ring map. Assume that

(1) R S is faithfully flat, and
(2) S is Noetherian.
Then R is Noetherian.
Proof. Let I0 I1 I2 . . . be a growing sequence of ideals of R. By assumption
we have In S = In+1 S = In+2 S = . . . for some n. Since R S is flat we have Ik S =
Ik R S. Hence, as R S is faithfully flat we see that In S = In+1 S = In+2 S = . . .
implies that In = In+1 = In+2 = . . . as desired.


COMMUTATIVE ALGEBRA

033F

Lemma 158.2. Let R S be a ring map. Assume that

(1) R S is faithfully flat, and
(2) S is reduced.
Then R is reduced.
Proof. This is clear as R S is injective.

033G

417

Lemma 158.3. Let R S be a ring map. Assume that

(1) R S is faithfully flat, and
(2) S is a normal ring.
Then R is a normal ring.
Proof. Since S is reduced it follows that R is reduced. Let p be a prime of R. We
have to show that Rp is a normal domain. Since Sp is faithfully over Rp too we may
assume that R is local with maximal ideal m. Let q be a prime of S lying over m.
Then we see that R Sq is faithfully flat (Lemma 38.17). Hence we may assume
S is local as well. In particular S is a normal domain. Since R S is faithfully
flat and S is a normal domain we see that R is a domain. Next, suppose that a/b
is integral over R with a, b R. Then a/b S as S is normal. Hence a bS.
This means that a : R R/bR becomes the zero map after base change to S. By
faithful flatness we see that a bR, so a/b R. Hence R is normal.


07NG

Lemma 158.4. Let R S be a ring map. Assume that

(1) R S is faithfully flat, and
(2) S is a regular ring.
Then R is a regular ring.
Proof. We see that R is Noetherian by Lemma 158.1. Let p R be a prime.
Choose a prime q S lying over p. Then Lemma 109.9 applies to Rp Sq and
we conclude that Rp is regular. Since p was arbitrary we see R is regular.


0352

Lemma 158.5. Let R S be a ring map. Assume that

(1) R S is faithfully flat, and
(2) S is Noetherian and has property (Sk ).
Then R is Noetherian and has property (Sk ).
Proof. We have already seen that (1) and (2) imply that R is Noetherian, see
Lemma 158.1. Let p R be a prime ideal. Choose a prime q S lying over p which
corresponds to a minimal prime of the fibre ring SR (p). Then A = Rp Sq = B
is a flat local ring homomorphism of Noetherian local rings with mA B an ideal of
definition of B. Hence dim(A) = dim(B) (Lemma 111.7) and depth(A) = depth(B)
(Lemma 157.2). Hence since B has (Sk ) we see that A has (Sk ).


0353

Lemma 158.6. Let R S be a ring map. Assume that

(1) R S is faithfully flat, and
(2) S is Noetherian and has property (Rk ).
Then R is Noetherian and has property (Rk ).
Proof. We have already seen that (1) and (2) imply that R is Noetherian, see
Lemma 158.1. Let p R be a prime ideal and assume dim(Rp ) k. Choose a
prime q S lying over p which corresponds to a minimal prime of the fibre ring

COMMUTATIVE ALGEBRA

418

SR (p). Then A = Rp Sq = B is a flat local ring homomorphism of Noetherian

local rings with mA B an ideal of definition of B. Hence dim(A) = dim(B) (Lemma
111.7). As S has (Rk ) we conclude that B is a regular local ring. By Lemma 109.9
we conclude that A is regular.

0354

Lemma 158.7. Let R S be a ring map. Assume that

(1) R S is smooth and surjective on spectra, and
(2) S is a Nagata ring.
Then R is a Nagata ring.
Proof. Recall that a Nagata ring is the same thing as a Noetherian universally
Japanese ring (Proposition 156.15). We have already seen that R is Noetherian in
Lemma 158.1. Let R A be a finite type ring map into a domain. According to
Lemma 156.3 it suffices to check that A is N-1. It is clear that B = AR S is a finite
type S-algebra and hence Nagata (Proposition 156.15). Since A B is smooth
(Lemma 135.4) we see that B is reduced (Lemma 157.6). Since B is Noetherian it
has only a finite number of minimal primes q1 , . . . , qt (see Lemma 30.6). As A B
is flat each of these lies over (0) A (by going down, see Lemma 38.18) The total
ring of fractions Q(B) is the product of the Li = f.f.(B/qi ) (Lemmas 24.4 and
24.1). Moreover, the integral closure B 0 of B in Q(B) is the product of the integral
closures Bi0 of the B/qi in the factors Li (compare with Lemma 36.15). Since B is
0
universally
Q 0 Japanese the ring extensions B/qi Bi are finite and we conclude that
0
B = Bi is finite over B. Since A B is flat we see that any nonzerodivisor on
A maps to a nonzerodivisor on B. The corresponding map
Q(A) A B = (A \ {0})1 A A B = (A \ {0})1 B Q(B)
is injective (we used Lemma 11.15). Via this map A0 maps into B 0 . This induces a
map
A0 A B B 0
which is injective (by the above and the flatness of A B). Since B 0 is a finite
B-module and B is Noetherian we see that A0 A B is a finite B-module. Hence
there exist finitely many elements xi A0 such that the elements xi 1 generate
A0 A B as a B-module. Finally, by faithful flatness of A B we conclude that
the xi also generated A0 as an A-module, and we win.


0355

Remark 158.8. The property of being universally catenary does not descend;
not even along etale ring maps. In Examples, Section 16 there is a construction
of a finite ring map A B with A local Noetherian and not universally catenary,
B semi-local with two maximal ideals m, n with Bm and Bn regular of dimension
2 and 1 respectively, and the same residue fields as that of A. Moreover, mA
generates the maximal ideal in both Bm and Bn (so A B is unramified as well
as finite). By Lemma 147.10 there exists a local etale ring map A A0 such that
B A A0 = B1 B2 decomposes with A0 Bi surjective. This shows that A0 has
two minimal primes qi with A0 /qi
= Bi . Since Bi is regular local (since it is etale
over either Bm or Bn ) we conclude that A0 is universally catenary.
159. Geometrically normal algebras

037Y

In this section we put some applications of ascent and descent of properties of rings.

COMMUTATIVE ALGEBRA

037Z

419

Lemma 159.1. Let k be a field. Let A be a k-algebra. The following properties

of A are equivalent:
(1) k 0 k A is a normal ring for every field extension k k 0 ,
(2) k 0 k A is a normal ring for every finitely generated field extension k k 0 ,
and
(3) k 0 k A is a normal ring for every finite purely inseparable extension k k 0 .
where normal ring is as defined in Definition 36.11.
Proof. It is clear that (1) (2) (3).
Assume (2) and let k k 0 be any field extension. Then we can write k 0 = colimi ki
as a directed colimit of finitely generated field extensions. Hence we see that k 0 k
A = colimi ki k A is a directed colimit of normal rings. Thus we see that k 0 k A
is a normal ring by Lemma 36.16. Hence (1) holds.
Assume (3) and let k K be a finitely generated field extension. By Lemma 44.3
we can find a diagram
/ K0
KO
O
/ k0
k
where k k 0 , K K 0 are finite purely inseparable field extensions such that
k 0 K 0 is separable. By Lemma 152.10 there exists a smooth k 0 -algebra B such
that K 0 is the fraction field of B. Now we can argue as follows: Step 1: k 0 k A is
a normal ring because we assumed (3). Step 2: B k0 k 0 k A is a normal ring as
k 0 k A B k0 k 0 k A is smooth (Lemma 135.4) and ascent of normality along
smooth maps (Lemma 157.7). Step 3. K 0 k0 k 0 k A = K 0 k A is a normal ring as it
is a localization of a normal ring (Lemma 36.13). Step 4. Finally K k A is a normal
ring by descent of normality along the faithfully flat ring map K k A K 0 k A
(Lemma 158.3). This proves the lemma.


0380

Definition 159.2. Let k be a field. A k-algebra R is called geometrically normal

over k if the equivalent conditions of Lemma 159.1 hold.

06DE

Lemma 159.3. Let k be a field. A localization of a geometrically normal k-algebra

is geometrically normal.
Proof. This is clear as being a normal ring is checked at the localizations at prime
ideals.


06DF

Lemma 159.4. Let k be a field. Let A, B be k-algebras. Assume A is geometrically

normal over k and B is a normal ring. Then A k B is a normal ring.
Proof. Let r be a prime ideal of Ak B. Denote p, resp. q the corresponding prime
of A, resp. B. Then (A k B)r is a localization of Ap k Bq . Hence it suffices to
prove the result for the ring Ap k Bq , see Lemma 36.13 and Lemma 159.3. Thus
we may assume A and B are domains.
Assume that A and B are domains with fractions fields K and L. Note that B is
the filtered colimit of its finite type normal k-sub algebras (as k is a Nagata ring,
see Proposition 156.16, and hence the integral closure of a finite type k-sub algebra
is still a finite type k-sub algebra by Proposition 156.15). By Lemma 36.16 we
reduce to the case that B is of finite type over k.

COMMUTATIVE ALGEBRA

420

Assume that A and B are domains with fractions fields K and L and B of finite
type over k. In this case the ring K k B is of finite type over K, hence Noetherian
(Lemma 30.1). In particular K k B has finitely many minimal primes (Lemma
30.6). Since A A k B is flat, this implies that A k B has finitely many minimal
primes (by going down for flat ring maps Lemma 38.18 these primes all lie over
(0) A). Thus it suffices to prove that A k B is integrally closed in its total ring
of fractions (Lemma 36.15).
We claim that K k B and A k L are both normal rings. If this is true then
any element x of Q(A k B) which is integral over A k B is (by Lemma 36.12)
contained in K k B A k L = A k B and were done. Since A K L is a normal
ring by assumption, it suffices to prove that K k B is normal.
As A is geometrically normal over k we see K is geometrically normal
S over k
(Lemma 159.3) hence K is geometrically reduced over k. Hence K = Ki is the
union of finitely generated field extensions of k which are geometrically reduced
(Lemma 42.2). Each Ki is the localization of a smooth k-algebra (Lemma 152.10).
So Ki k B is the localization of a smooth B-algebra hence normal (Lemma 157.7).
Thus K k B is a normal ring (Lemma 36.16) and we win.

160. Geometrically regular algebras
045K

Let k be a field. Let A be a Noetherian k-algebra. Let k K be a finitely generated

field extension. Then the ring K k A is Noetherian as well, see Lemma 30.7. Thus
the following lemma makes sense.

0381

Lemma 160.1. Let k be a field. Let A be a k-algebra. Assume A is Noetherian.

The following properties of A are equivalent:
(1) k 0 k A is regular for every finitely generated field extension k k 0 , and
(2) k 0 k A is regular for every finite purely inseparable extension k k 0 .
Here regular ring is as in Definition 109.7.
Proof. The lemma makes sense by the remarks preceding the lemma. It is clear
that (1) (2).
Assume (2) and let k K be a finitely generated field extension. By Lemma 44.3
we can find a diagram
/ K0
KO
O
k

/ k0

where k k 0 , K K 0 are finite purely inseparable field extensions such that

k 0 K 0 is separable. By Lemma 152.10 there exists a smooth k 0 -algebra B such
that K 0 is the fraction field of B. Now we can argue as follows: Step 1: k 0 k A is
a regular ring because we assumed (2). Step 2: B k0 k 0 k A is a regular ring as
k 0 k A B k0 k 0 k A is smooth (Lemma 135.4) and ascent of regularity along
smooth maps (Lemma 157.8). Step 3. K 0 k0 k 0 k A = K 0 k A is a regular ring as
it is a localization of a regular ring (immediate from the definition). Step 4. Finally
K k A is a regular ring by descent of regularity along the faithfully flat ring map
K k A K 0 k A (Lemma 158.4). This proves the lemma.


COMMUTATIVE ALGEBRA

0382

421

Definition 160.2. Let k be a field. Let R be a Noetherian k-algebra. The

k-algebra R is called geometrically regular over k if the equivalent conditions of
Lemma 160.1 hold.
It is clear from the definition that K k R is a geometrically regular algebra over
K for any finitely generated field extension K of k. We will see later (More on
Algebra, Proposition 27.1) that it suffices to check R k k 0 is regular whenever
k k 0 k 1/p (finite).

07NH

Lemma 160.3. Let k be a field. Let A B be a faithfully flat k-algebra map. If

B is geometrically regular over k, so is A.
Proof. Assume B is geometrically regular over k. Let k k 0 be a finite, purely
inseparable extension. Then A k k 0 B k k 0 is faithfully flat as a base change
of A B (by Lemmas 29.3 and 38.7) and B k k 0 is regular by our assumption on
B over k. Then A k k 0 is regular by Lemma 158.4.


07QF

Lemma 160.4. Let k be a field. Let A B be a smooth ring map of k-algebras.

If A is geometrically regular over k, then B is geometrically regular over k.
Proof. Let k k 0 be a finitely generated field extension. Then A k k 0 B k k 0
is a smooth ring map (Lemma 135.4) and A k k 0 is regular. Hence B k k 0 is
regular by Lemma 157.8.


07QG

Lemma 160.5. Let k be a field. Let A be an algebra over k. Let k = colim ki be

a directed colimit of subfields. If A is geometrically regular over each ki , then A is
geometrically regular over k.
Proof. Let k k 0 be a finite purely inseparable field extension. We can get k 0
by adjoining finitely many variables to k and imposing finitely many polynomial
relations. Hence we see that there exists an i and a finite purely inseparable field
extension ki ki0 such that ki = k ki ki0 . Thus A k k 0 = A ki ki0 and the lemma
is clear.


07QH

Lemma 160.6. Let k k 0 be a separable algebraic field extension. Let A be

an algebra over k 0 . Then A is geometrically regular over k if and only if it is
geometrically regular over k 0 .
Proof. Let k L be a finite purely inseparable field extension. Then L0 = k 0 k L
is a field (see material in Fields, Section 27) and A k L = A k0 L0 . Hence if A is
geometrically regular over k 0 , then A is geometrically regular over k.
Assume A is geometrically regular over k. Since k 0 is the filtered colimit of finite extensions of k we may assume by Lemma 160.5 that k 0 /k is finite separable.
Consider the ring maps
k 0 A k k 0 A.
Note that A k k 0 is geometrically regular over k 0 as a base change of A to k 0 . Note
that A k k 0 A is the base change of k 0 k k 0 k 0 by the map k 0 A. Since
k 0 /k is an etale extension of rings, we see that k 0 k k 0 k 0 is etale (Lemma 141.3).
Hence A is geometrically regular over k 0 by Lemma 160.4.


COMMUTATIVE ALGEBRA

422

161. Geometrically Cohen-Macaulay algebras

045L

This section is a bit of a misnomer, since Cohen-Macaulay algebras are automatically geometrically Cohen-Macaulay. Namely, see Lemma 129.6 and Lemma 161.2
below.

045M

Lemma 161.1. Let k be a field and let k K and k L be two field extensions
such that one of them is a field extension of finite type. Then K k L is a Noetherian
Cohen-Macaulay ring.
Proof. The ring Kk L is Noetherian by Lemma 30.7. Say K is a finite extension of
the purely transcendental extension k(t1 , . . . , tr ). Then k(t1 , . . . , tr )k L K k L
is a finite free ring map. By Lemma 111.9 it suffices to show that k(t1 , . . . , tr ) k L
is Cohen-Macaulay. This is clear because it is a localization of the polynomial ring
L[t1 , . . . , tr ]. (See for example Lemma 103.7 for the fact that a polynomial ring is
Cohen-Macaulay.)


045N

Lemma 161.2. Let k be a field. Let S be a Noetherian k-algebra. Let k K be

a finitely generated field extension, and set SK = K k S. Let q S be a prime
of S. Let qK SK be a prime of SK lying over q. Then Sq is Cohen-Macaulay if
and only if (SK )qK is Cohen-Macaulay.
Proof. By Lemma 30.7 the ring SK is Noetherian. Hence Sq (SK )qK is a flat
local homomorphism of Noetherian local rings. Note that the fibre
(SK )q /q(SK )q
= ((q) k K)q0
K

is the localization of the Cohen-Macaulay (Lemma 161.1) ring (q) k K at a

suitable prime ideal q0 . Hence the lemma follows from Lemma 157.3.

162. Colimits and maps of finite presentation, II
07RF

This section is a continuation of Section 126.

We start with an application of the openness of flatness. It says that we can
approximate flat modules by flat modules which is useful.

02JO

Lemma 162.1. Let R S be a ring map. Let M be an S-module. Assume that

(1) R S is of finite presentation,
(2) M is a finitely presented S-module, and
(3) M is flat over R.
In this case we have the following:
(1) There exists a finite type Z-algebra R0 and a finite type ring map R0 S0
and a finite S0 -module M0 such that M0 is flat over R0 , together with a
ring maps R0 R and S0 S and an S0 -module map M0 M such that
S
= R R0 S0 and M = S S0 M0 .
(2) If R = colim R is written as a directed colimit, then there exists a
and a ring map R S of finite presentation, and an S -module M of
finite presentation such that M is flat over R and such that S = RR S
and M = S S M .
(3) If
(R S, M ) = colim (R S , M )
is written as a directed colimit such that
(a) R R S S and S S M M are isomorphisms for ,

COMMUTATIVE ALGEBRA

423

(b) R S is of finite presentation,

(c) M is a finitely presented S -module,
then for all sufficiently large the module M is flat over R .
Proof. We first write (R S, M ) as the directed colimit of a system (R
S , M ) as in as in Lemma 126.15. Let q S be a prime. Let p R, q S , and
p R the corresponding primes. As seen in the proof of Theorem 128.4
((R )p , (S )q , (M )q )
is a system as in Lemma 126.11, and hence by Lemma 127.3 we see that for some
q for all q the module M is flat over R at the prime q .
By Theorem 128.4 we get an open subset U Spec(S ) such that M flat over R
at all the primes of U . Denote V Spec(S) the inverse image of U under the
map Spec(S) Spec(S ). The argument above shows that for every q Spec(S)
there exists a q such that q V for all q . Since Spec(S) is quasi-compact
we see this implies there exists a single 0 such that V0 = Spec(S).
The complement Spec(S0 ) \ U0 is V (I) for some ideal I S0 . As V0P
= Spec(S)
we see that IS = S. Choose f1 , . . . , fr I and s1 , . . . , sn S such that fi si = 1.
SinceP
colim S = S, after increasing 0 we may assume there exist si,0 S0 such
that
fi si,0 = 1. Hence for this 0 we have U0 = Spec(S0 ). This proves (1).
Proof of (2). Let (R0 S0 , M0 ) be as in (1) and suppose that R = colim R .
Since R0 is a finite type Z algebra, there exists a and a map R0 R such that
R0 R R is the given map R0 R (see Lemma 126.2). Then, part (2) follows
by taking S = R R0 S0 and M = S S0 M0 .
Finally, we come to the proof of (3). Let (R S , M ) be as in (3). Choose
(R0 S0 , M0 ) and R0 R as in (1). As in the proof of (2), there exists a 0
and a ring map R0 R0 such that R0 R0 R is the given map R0 R.
Since S0 is of finite presentation over R0 and since S = colim S we see that for
some 1 0 we get an R0 -algebra map S0 S1 such that the composition
S0 S1 S is the given map S0 S (see Lemma 126.2). For all 1 this
gives maps
: R R0 S0 R R1 S1
= S
the last isomorphism by assumption. By construction colim is an isomorphism.
Hence is an isomorphism for all large enough by Lemma 126.6. In the same
vein, there exists a 2 1 and an S0 -module map M0 M2 such that M0
M2 M is the given map M0 M (see Lemma 126.3). For 2 there is an
induced map
S S0 M0 S S2 M2
= M
and for large enough this map is an isomorphism by Lemma 126.4. This implies
(3) because M0 is flat over R0 .

034Y

Lemma 162.2. Let R A B be ring maps. Assume A B faithfully flat of

finite presentation. Then there exists a commutative diagram
R

/ A0

/ B0


/A


/B

COMMUTATIVE ALGEBRA

424

with R A0 of finite presentation, A0 B0 faithfully flat of finite presentation

and B = A A0 B0 .
Proof. We first prove the lemma with R replaced Z. By Lemma 162.1 there exists
a diagram
/A
AO 0
O
B0

/B

where A0 is of finite type over Z, B0 is flat of finite presentation over A0 such that
B = A A0 B0 . As A0 B0 is flat of finite presentation we see that the image
of Spec(B0 ) Spec(A0 ) is open, see Proposition 40.8. Hence the complement of
the image is V (I0 ) for some ideal I0 A0 . As A B is faithfully flat the map
Spec(B) Spec(A) is surjective, see Lemma 38.16. Now we use that the base
change of
Pthe image is the image of the base change. Hence I0 A = A. Pick a
relation
fi ri = 1, with ri A, fi I0 . Then after enlarging A0 to contain the
elements ri (and correspondingly enlarging B0 ) we see that A0 B0 is surjective
on spectra also, i.e., faithfully flat.
Thus the lemma holds in case R = Z. In the general case, take the solution A00 B00
just obtained and set A0 = A00 Z R, B0 = B00 Z R.

07RG

Lemma 162.3. Let A = colimiI Ai be a directed colimit of rings. Let 0 I and

0 : B0 C0 a map of A0 -algebras. Assume
(1) A A0 B0 A A0 C0 is finite,
(2) C0 is of finite type over B0 .
Then there exists an i 0 such that the map Ai A0 B0 Ai A0 C0 is finite.
Proof. Let x1 , . . . , xm be generators for C0 over B0 . Pick monic polynomials
Pj A A0 B0 [T ] such that Pj (1 xj ) = 0 in A A0 C0 . For some i 0 we can find
Pj,i Ai A0 B0 [T ] mapping to Pj . Since commutes with colimits we see that
Pj,i (1 xj ) is zero in Ai A0 C0 after possibly increasing i. Then this i works. 

07RH

Lemma 162.4. Let A = colimiI Ai be a directed colimit of rings. Let 0 I and

0 : B0 C0 a map of A0 -algebras. Assume
(1) A A0 B0 A A0 C0 is surjective,
(2) C0 is of finite type over B0 .
Then for some i 0 the map Ai A0 B0 Ai A0 C0 is surjective.
Proof. Let x1 , . . . , xm be generators for C0 over B0 . Pick bj A A0 B0 mapping
to 1 xj in A A0 C0 . For some i 0 we can find bj,i Ai A0 B0 mapping to bj .
Then this i works.


07RI

Lemma 162.5. Let A = colimiI Ai be a directed colimit of rings. Let 0 I and

0 : B0 C0 a map of A0 -algebras. Assume
(1) A A0 B0 A A0 C0 is etale,
(2) B0 C0 is of finite presentation.
Then for some i 0 the map Ai A0 B0 Ai A0 C0 is etale.

COMMUTATIVE ALGEBRA

425

Proof. Note that B0 C0 is of finite presentation, see Lemma 6.2. Write

C0 = B0 [x1 , . . . , xn ]/(f1,0 , . . . , fm,0 ). Write Bi = Ai A0 B0 and Ci = Ai A0 C0 .
Note that Ci = Bi [x1 , . . . , xn ]/(f1,i , . . . , fm,i ) where fj,i is the image of fj,0 in the
polynomial ring over Bi . Write B = A A0 B0 and C = A A0 C0 . Note that
C = B[x1 , . . . , xn ]/(f1 , . . . , fm ) where fj is the image of fj,0 in the polynomial ring
over B. The assumption is that the map
M
d : (f1 , . . . , fm )/(f1 , . . . , fm )2
Cdxk
is an isomorphism. Thus for sufficiently large i we can find elements
k,i (f1,i , . . . , fm,i )/(f1,i , . . . , fm,i )2
L
with dk,i = dxk in
Ci dxk . Moreover, on increasing i if necessary, we see that
P
(fj,i /xk )k,i = fj,i mod (f1,i , . . . , fm,i )2 since this is true in the limit. Then
this i works.

The following lemma is an application of the results above which doesnt seem to
fit well anywhere else.
034Z

Lemma 162.6. Let R S be a faithfully flat ring map of finite presentation.

Then there exists a commutative diagram
/ S0
>

S_
R
0

where R S is quasi-finite, faithfully flat and of finite presentation.

Proof. As a first step we reduce this lemma to the case where R is of finite type
over Z. By Lemma 162.2 there exists a diagram
/S
SO 0
O
R0

/R

where R0 is of finite type over Z, and S0 is faithfully flat of finite presentation over
R0 such that S = R R0 S0 . If we prove the lemma for the ring map R0 S0 , then
the lemma follows for R S by base change, as the base change of a quasi-finite
ring map is quasi-finite, see Lemma 121.8. (Of course we also use that base changes
of flat maps are flat and base changes of maps of finite presentation are of finite
presentation.)
Assume R S is a faithfully flat ring map of finite presentation and that R is
Noetherian (which we may assume by the preceding paragraph). Let W Spec(S)
be the open set of Lemma 129.4. As R S is faithfully flat the map Spec(S)
Spec(R) is surjective, see Lemma 38.16. By Lemma 129.5 the map W Spec(R)
is also surjective. Hence by replacing S with a product Sg1 . . . Sgm we may
assume W = Spec(S); here we use that Spec(R) is quasi-compact (Lemma 16.10),
and that the map Spec(S) Spec(R) is open (Proposition 40.8). Suppose that
p R is a prime. Choose a prime q S lying over p which corresponds to a
maximal ideal of the fibre ring S R (p). The Noetherian local ring S q = Sq /pSq
is Cohen-Macaulay, say of dimension d. We may choose f1 , . . . , fd in the maximal

COMMUTATIVE ALGEBRA

426

ideal of Sq which map to a regular sequence in S q . Choose a common denominator

g S, g 6 q of f1 , . . . , fd , and consider the R-algebra
S 0 = Sg /(f1 , . . . , fd ).
By construction there is a prime ideal q0 S 0 lying over p and corresponding to q
(via Sg Sg0 ). Also by construction the ring map R S 0 is quasi-finite at q as
the local ring
Sq0 0 /pSq0 0 = Sq /(f1 , . . . , fd ) + pSq = S q /(f 1 , . . . , f d )
has dimension zero, see Lemma 121.2. Also by construction R S 0 is of finite
presentation. Finally, by Lemma 98.3 the local ring map Rp Sq0 0 is flat (this
is where we use that R is Noetherian). Hence, by openness of flatness (Theorem
128.4), and openness of quasi-finiteness (Lemma 122.14) we may after replacing
g by gg 0 for a suitable g 0 S, g 0 6 q assume that R S 0 is flat and quasifinite. The image Spec(S 0 ) Spec(R) is open and contains p. In other words we
have shown a ring S 0 as in the statement of the lemma exists (except possibly the
faithfulness part) whose image contains any given prime. Using one more time the
quasi-compactness of Spec(R) we see that a finite product of such rings does the
job.

163. Other chapters
Preliminaries
(1) Introduction
(2) Conventions
(3) Set Theory
(4) Categories
(5) Topology
(6) Sheaves on Spaces
(7) Sites and Sheaves
(8) Stacks
(9) Fields
(10) Commutative Algebra
(11) Brauer Groups
(12) Homological Algebra
(13) Derived Categories
(14) Simplicial Methods
(15) More on Algebra
(16) Smoothing Ring Maps
(17) Sheaves of Modules
(18) Modules on Sites
(19) Injectives
(20) Cohomology of Sheaves
(21) Cohomology on Sites
(22) Differential Graded Algebra
(23) Divided Power Algebra
(24) Hypercoverings
Schemes
(25) Schemes

(26) Constructions of Schemes

(27) Properties of Schemes
(28) Morphisms of Schemes
(29) Cohomology of Schemes
(30) Divisors
(31) Limits of Schemes
(32) Varieties
(33) Topologies on Schemes
(34) Descent
(35) Derived Categories of Schemes
(36) More on Morphisms
(37) More on Flatness
(38) Groupoid Schemes
(39) More on Groupoid Schemes

(40) Etale
Morphisms of Schemes
Topics in Scheme Theory
(41) Chow Homology
(42) Intersection Theory
(43) Picard Schemes of Curves
(44) Adequate Modules
(45) Dualizing Complexes
(46) Algebraic Curves
(47) Resolution of Surfaces
(48) Fundamental Groups of Schemes

(49) Etale
Cohomology
(50) Crystalline Cohomology
(51) Pro-etale Cohomology

COMMUTATIVE ALGEBRA

427

Algebraic Spaces
(74) Deformation Theory
(75) The Cotangent Complex
(52) Algebraic Spaces
(53) Properties of Algebraic Spaces
Algebraic Stacks
(54) Morphisms of Algebraic Spaces
(76) Algebraic Stacks
(55) Decent Algebraic Spaces
(77) Examples of Stacks
(56) Cohomology of Algebraic Spaces
(78) Sheaves on Algebraic Stacks
(57) Limits of Algebraic Spaces
(79) Criteria for Representability
(58) Divisors on Algebraic Spaces
(80) Artins Axioms
(59) Algebraic Spaces over Fields
(81) Quot and Hilbert Spaces
(60) Topologies on Algebraic Spaces
(82) Properties of Algebraic Stacks
(61) Descent and Algebraic Spaces
(83) Morphisms of Algebraic Stacks
(62) Derived Categories of Spaces
(84) Cohomology of Algebraic Stacks
(63) More on Morphisms of Spaces
(85) Derived Categories of Stacks
(64) Pushouts of Algebraic Spaces
(86) Introducing Algebraic Stacks
(65) Groupoids in Algebraic Spaces
(87) More on Morphisms of Stacks
(66) More on Groupoids in Spaces
Miscellany
(67) Bootstrap
(88) Examples
Topics in Geometry
(89) Exercises
(68) Quotients of Groupoids
(90) Guide to Literature
(69) Simplicial Spaces
(91) Desirables
(70) Formal Algebraic Spaces
(92) Coding Style
(71) Restricted Power Series
(93) Obsolete
(72) Resolution of Surfaces Revisited
(94) GNU Free Documentation LiDeformation Theory
cense
(73) Formal Deformation Theory
(95) Auto Generated Index
References
[AB57]
[Alp14]
[Aus55]
[BE73]
[Ben73]
[Bko70]
[CE56]
[Cha60]
[DG67]
[DM83]
[FR70]
[GR71]

Maurice Auslander and David A. Buchsbaum, Homological dimension in local rings,

Trans. Amer. Math. Soc. 85 (1957), 390405.
Jarod Alper, Adequate moduli spaces and geometrically reductive group schemes, Algebr.
Geom. 1 (2014), no. 4, 489531.
Maurice Auslander, On the dimension of modules and algebras. III. Global dimension,
Nagoya Math. J. 9 (1955), 6777.
David Alvin Buchsbaum and David Eisenbud, What makes a complex exact?, J. Algebra
25 (1973), 259268.
Bruce Bennett, On the structure of non-excellent curve singularities in characteristic p,

Inst. Hautes Etudes

Sci. Publ. Math. (1973), no. 42, 129170.
Rudolphe Bkouche, Puret
e, mollesse et paracompacit
e, C. R. Acad. Sci. Paris S
er. A-B
270 (1970).
Henri Cartan and Samuel Eilenberg, Homological algebra, Princeton University Press,
Princeton, N. J., 1956.
Stephen U. Chase, Direct products of modules, Trans. Amer. Math. Soc. 97 (1960),
457473.
ements de g
Jean Dieudonn
e and Alexander Grothendieck, El
eom
etrie alg
ebrique, Inst.

Hautes Etudes
Sci. Publ. Math. 4, 8, 11, 17, 20, 24, 28, 32 (19611967).
Guiseppe De Marco, Projectivity of pure ideals, Rend. Sem. Mat. Univ. Padova 69 (1983),
289304.
Daniel Ferrand and Michel Raynaud, Fibres formelles dun anneau local noeth
erien,

Ann. Sci. Ecole

Norm. Sup. (4) 3 (1970), 295311.
Laurent Gruson and Michel Raynaud, Crit`
eres de platitude et de projectivit
e, Invent.
math. 13 (1971), 189.

COMMUTATIVE ALGEBRA

428

[Gru73] Laurent Gruson, Dimension homologique des modules plats sur un anneau commutatif
noeth
erien, Symposia Mathematica, Vol. XI (Convegno di Algebra Commutativa, INDAM, Rome, 1971), Academic Press, London, 1973, pp. 243254.
[Kab71] Thomas Kabele, Regularity conditions in nonnoetherian rings, Trans. Amer. Math. Soc.
155 (1971), 363374.
[Kap58] Irving Kaplansky, Projective modules, Ann. of Math (2) 68 (1958), 372377.
[Kun83] Kenneth Kunen, Set theory, Elsevier Science, 1983.
[Lam99] Tsit Yuen Lam, Lectures on modules and rings, Graduate Texts in Mathematics, vol.
189, Springer-Verlag, New York, 1999.
[Laz67] Daniel Lazard, Disconnexit
es des spectres danneaux et des pr
esch
emas, Bull. Soc. Math.
France 95 (1967), 95108.
, Autour de la platitude, Bull. Soc. Math. France 97 (1969), 81128.
[Laz69]
[LR08] Tsit Yuen Lam and Manuel Lionel Reyes, A prime ideal principle in commutative algebra,
Journal of Algebra 319 (2008), no. 7, 30063027.
[Mat70] Hideyuki Matsumura, Commutative algebra, W. A. Benjamin, Inc., 1970.
[Mat78] Eben Matlis, The higher properties of R-sequences, J. Algebra 50 (1978), no. 1, 77112.
[Maz68] Pierre Mazet, G
en
erateurs, relations et
epimorphismes danneaux, C. R. Acad. Sci. Paris
S
er. A-B 266 (1968).
[Nag57] Masayoshi Nagata, A remark on the unique factorization theorem, J. Math. Soc. Japan
9 (1957), 143145.
[Pes66] Christian Peskine, Une g
en
eralisation du main theorem de Zariski, Bull. Sci. Math.
(2) 90 (1966), 119127.
[Ray70] Michel Raynaud, Anneaux locaux hens
eliens, Lecture Notes in Mathematics, vol. 169,
Spinger-Verlag, 1970.