Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Mathematics
Edited by A. Dold and B. Eckmann
1023
Stephen McAdam
Springer-Verlag
Berlin Heidelberg New York Tokyo 1983
Author
Stephen McAdam
Department of Mathematics, University of Texas at Austin
Austin, Texas 78712, USA
This work is subject to copyright.All rights are reserved,whetherthe whole or part of the material
is concerned,specificallythose of translation,reprinting, re-useof illustrations,broadcasting,
reproduction by photocopyingmachineor similar means,and storage in data banks. Under
§ 54 of the GermanCopyright Law where copies are madefor other than private use, a fee is
payableto "VerwertungsgesellschaftWort", Munich.
© by Springer-VerlagBerlin Heidelberg 1983
Printed in Germany
Printing and binding: Beltz Offsetdruck, Hemsbach/Bergstr.
2146/3140-543210
TO MARTHA
ACKNOWLEDGMENTS
Certain people have been particularly helpful to me, as much in stimulating conver-
Dan Katz and Keith Whittington. My special thanks goes, as it does so often, to
Jack Ratliff.
Part of my research was supported by the National Science Foundation, for which
I am grateful.
Nita Goldrick typed the manuscript. Her great skill and patience eased a
difficult task.
TABLE OF CONTENTS
Page
INTRODUCTION VIII
C H A P T E R IV : A Characterization of A (I) 26
C H A P T E R V: Asymptotic Sequences 32
CHAPTER XI : Miscellaneous 89
REFERENCES 113
INDEX Il7
INTRODUCTION
Asymptotic prime divisors represent the interface of two major ideas in the
study of co~utative Noetherian rings. The first, the concept of prime divisors,
is one of the most valued tools in the researcher's arsenal. The second is the
fact that in a Noetherian ring, large powers of an ideal are well behaved, as shown
Although its roots go back further, the recent interest in asymptotic prime
large? He was able to answer a related question, showing that if [ is the integral
P c Ass(R/In), some n, then P e Ass(R/l TM) for infinitely many m.) Meanwhile,
Brodmann answered the original question, proving that ~ss (R/In) also stabilizes
for large n. Since then, the topic of asymptotic prime divisors has been growing
rapidly, the latest development being the advent of asymptotic sequences, a useful
These notes attempt to present the bulk of the present knowledge of asymptotic
prime divisors in a reasonably efficient way, to ease the task of those wishing to
learn of, or contribute to the subject. Modulo some gnashing of teeth, and rending
of garments, it was both educational and satisfying to write them. I hope that
Ass(R/I n ) stabilizes for large n, as does Ass (In'l/In), the respective stable
values of these two sequences are being denoted A (I) and B (I). Also B (I) is
Chapter Two shows that A (I) - B (I) ~ A s s R, and that P e A (I) - B (I) if
Chapter Three shows that Ass(R/~) c_ Ass(R/I 2) c_,.. , and that this sequence
--,k -~¢ *
eventually stabilizes to a set denoted A (I). Furthermore, A (I) c_ A (I). It
also developes several technical results useful for dealing with A (I), the most
important of these being that in a local ring s P e A (I) if and only if there are
complete local domain is locally quasi-unmixed, this result meshes nicely with the
pletion. This is then used to show that for a given ideal I in any Noetherian
ring, all asymptotic sequences maximal with respect to coming from I have the
xi+ 1 ~ U [P e A ((l,x I,..., xi))}. It is shown that in a local ring, all maximal
Chapter Seven proves that in a local ring, the grade of R/I n stabilizes for
Chapter Eight identifies, with one possible exception, all Noetherian rings for
In Chapter Nine, asymptotic prime divisors play a minor role in proving the
In Chapter Ten, we consider a local ring ~,M) and the ideal transform of
M, T(M). Previously it was known that the following two statements are equivalent:
prime divisor of zero. Our main result adds two more equivalent conditions:
(c) M e A*(J) for every regular ideal J (d) There is a regular element x with
M e A * (J) for all J~xR. Here J~l if for some n and m, i n and jm have
the same integral closure. Motivated by statement (d), we then discuss the possi-
chain conditions necessary to read these notes. In the Appendix, I list those
definitions and basic results (with references for the curious reader) which are
In [R3], Ratliff asked about the behavior of the sequence A(l,n) (and
showed that a related sequence stabilized, see Chapter 3). In [BI], Brodmann
showed that both sequences A(l,n) and B(l,n) stabilize for large n, as we now
LEMMA I.I. a) Let En> 0 R be a Noetherian homogeneous graded ring. Then there
n
is an ~ such that for n >__ ~, (0 :RI) N R n = 0 .
n ~ ~, (In+l : I) n I ~ = I n.
the r.l are homogeneous and have r i e RIT. Thus riai = 0, and so x = 0.
Now select ~ as in Lemma i.I and say n > ~° If P c ASSR0 (Rn) write
P c ASSR0 (Rn+l). Thus ASSR0 (Rn) ~ ASSR0 (Rn+l) for n > ~° As we already
Proof: The exact sequence 0 + In/l n+l ~ R/I n+l + R/I n + 0 shows that
A(I,n). Thus A(I,n+l) ~ A(I,n), and the result is clear since A(I,n) is finite.
i < i < n. To see that P @ A(I,s) for s odd, 1 < s < 2n, note that
P 4 A(I,I) since I is prime. Now for i ~ q ~ 2n, the residues of the set
u
Tq = [z } U [z u2
2 ...z 2 ~n lu2+,,,+U2n = q, 0 ~ u l. < i} form a generating set for
lq/l q+l over k[x]. If q is even, there are no relations, and T gives a
q
free basis •
If q is odd, there is the unique relation xz qq e I q+l . Suppose
of the sequences A(I,n) and B(l,n) will be denoted A (I) and B (I),
respectively.
The fact that A (I) and B (I) behave well under localization is straight-
prime divisors. As this result will be used again when discussing the integral
Proof: Let PI'''" Pm be all of the primes of R which lie over P. Select
containing a prime Q with QNR =P and Q minimal over IT. We first reduce
to the case that T = S. Clearly we may assume T = ~, and by going down we may
for m > n.
The following fact about the integral closure of a Noetherian domain appears
finitely generated ideal of the integral closure R. Then the number of primes of
Proof: Let S = R [ b I ,..., bm] and let I = (b I ,..., bm)S. Thus S is Noetherian
Q ~ S e A (I). Since A (I) is finite and since only finitely many primes of
divisor of akR. It is not difficult to now see that P is also a prime divisor
of aR.
height q < n and (aI, ..., an)T ~ q. By induction, grade q n R ~ height q < n,
P is also a prime divisor of (a I,..., an) by [KI, Section 3-1, Exercise 13].
The next three propositions give easy circumstances under which a prime must
be in A (I).
PROPOSITION I°ii. Let I be an ideal in a Noetherian ring, and let the prime P
be minimal over !. Then P e A (I). Also P e B (I) if and only if height P > 0.
(by Nakayama's Le~mna) which must have at least one prime divisor. However P is
Proof: Localize at P and then write P = (0 : c). For n large enough that
Ratliff.
prime divisor of O° Then there is an integer n > 0 such that for any ideal J
=R[ t-l, It] be the Rees ring of R with respect to I. Then P c B * (I) if and
Proof: Let P e B (I). Consider ~ as in Lemma l.lb, and choose n > ~ with
P e Ass(In/I n+l ). Write P = (In+l : c) with c e In . Since ctn e ~R, note that
Q = (t-l~ :dct n-Mn) is prime in ~ and QNR=P. We must show that It ~ Q° Since
Conversely, suppose that Q = (t'l~ : gtk) with g c Ik, that QNR=P, and
P e B (I).
We close the chapter with a question. We have seen that the sequence
main result being that such primes must be prime divisors of zero.
xn e R with P= (In : Xn), and by Lemma 2.1 we have Xn @ I~" To show that P
is a prime divisor of zero, it is sufficient to show this in the case that (R,P)
is complete, w h i c h we now assume. Let V = (I~ : P)/I ~ and for n > ~ let
%eV n write •= d
n
+i
n
with d
n
e (In : P) and i e I b.
n
For m > n
- -
we have
showing that the sequences [in} and [d n] are Cauchy sequences. Let in + i
B (I) if and only if there is an integer k > I such that for all sufficiently
in = P k N qnl N...N qnm" Clearly P e A*(1). Let n ~ k+l and let P = (In : c).
Suppose A (I) = {P'QI ''''' % } and for n ~ ~ let the primary decomposition of
In be q n N qnl N...N qnm with RAD qn = P" We may assume that qn+li ~ qni ' so
that if Jn=qnl N...n qnm' then Jn+l ~ Jn " Claim J n N I ~ = In . One inclusion
Lemma 1.2, we see that for some r e R, (In : rc) is a proper prime ideal, and
the claim that Jn N I ~ = In, Now since Jn+l c-- Jn ' we have Jn c- J~ " Thus
there is a k with (0 : Ik) = (0 : Ik+l) =... . Call this ideal J. The following
LEMM 2.4. Consider I and J = (0 : Ik) as above. Let I' = I modulo J. Then
Proof: Suppose P' e A*(I') -B*(I'). Then by Proposition 2.2, P' = (0 : c') for
some c' e R'. Since I' c p', we have c'l '= 0. Thus cl c j = (0 : Ik), so that
that x e (In+k : I k) n I ~ = I n for all big n. That is, j n i ~ c Q i n" n = 1,2, ....
We next claim that for large n, In/l n+l m l'n/l 'n+l. To see this, map
a + l n+l to a' + I 'n+l and note that the map is injective since In N J c _ In+l '
from above.
P ~ B (I).
as claimed.
We now work modulo J, using primes to denote images. We have just seen
that Q' is a prime divisor or zero in R'. As P' is minimal over Q'+I',
Combining Propositions 2.2 and 1.12, we see that A (1)=B (1) U[P e Ass(R) I
I ~P}. We know very little of what can be said about the overlap, B (I) N [ p
let PI' P2' P3' QI" Q2 ' Q3 he (respectively) the images of PI' P2' P3' ql ' q2' q3
and let I be the image of (Y). Now I ~ P1 ~ P2 ~ P3 and these three primes
However the image of Y modulo qlq2q3 is a regular element, and so we need only
show that p2/qlq2q3 is not a prime divisor of (y) + qlq2q3/qlq2q3 , which is clear.
that for an ideal I, ~ denotes the integral closure of I. We will let 2(I,n) =
Ass(R/In). Our main goals will be to prove that ~(I,I) ~ A(I,2) ~... , that this
sequence eventually stabilizes at a set denoted A (I), and that A (I) ~ A (I).
These results were first proved by Ratliff in [R3] and [R8]. (In [Rs2], Rees out-
lines a different approach to these ideas.) The essence of the arguments needed are
mostly easily seen when R is a domain. We treat that case first, following Rat-
liff's trail.
LEMMA 3.1. Let R be a Noetherian domain with integral closure R. Let (V,N) be
with pNR=P.
Proof: Write P = (anR : c). Since anR = anRn R, P = (anR : c)~N R. Now the Krull
domain ~ has A.C.C. on ideals of the form (b : d). Thus the argument used in
proving Len~a 1.2 shows that (anR: c)~ can be enlarged to a prime divisor p of
Proof: Using Lemma 3.2 we see that ~ contains a height 1 prime q lying over
Qo Now (V,N) = (~q, qq) is a D.V.R. Suppose that It ~ Q . Then R/QNR =~/Q
namely X~[X], and it intersects T[X~ at XT[X]. With T=R and X = t -1, we
see that (QNR, t-I)R[t -I] must be just t-IR[t-II, so that QNR= O.
A(I,I) ~ ~(I,2) ~ .... This sequence stabilizes to a set denoted A (I). Further-
-4¢ *
more A (I)~ B (I).
Proof: Let P e A(I,n) and write P = (In : C)R for c e R. Then since
In = t'n~N R, P = (t-n~: e)~N R. By Lemma 3.3, no prime divisor of (t'n~ : c)~R can
Thus (t-n~R : c)~= (t-n~ : cdt)~ and this ideal meets R at P. Since the degree 1
component of t-n~ is (In+l N I)t, we see that P = (In+l N I : cd) = (In+l :cd)
We will now show that A(I,n) ~ B (I). As B (I) is finite, the rest of the
enlarge (t-n~ : c)~ to a prime divisor Q of t-n~ with QNR=P. By Lermna 3.3,
We apply Proposition 1.7 to this context, and easily see the following.
14
LE~RIA 3.6. If I c J are ideals, then I reduces J if and only if for each
Proof: One direction is trivial. Thus suppose that ql ''''' qm are all the
minimal primes, and that for each i, I mod qi reduces J mod qi " Then for
jk+l
sufficiently large k we have jk+l ~- i j k + q i' i= l,...,m. Since IJk
does not reduce (I,c). Thus Lenlna 3.6 shows that for some minimal prime q,
I+q/q does not reduce (I,c) +q/q. Therefore c+q @ (I+q/q). We easily see
a minimal prime in R.
I n / I n N p ~ In+p/p, we see that ~/q ~ (R/p)[t "l,(l+p/p)t] =~', the Rees ring of
R/p with respect to I + p/p. This isomorphism takes Q/q to a prime divisor
Q' of t'n~ '. By Lermma 3.3, either (l+p/p)t ~ Q' or Q' n (R/p) =0. If
proof that P e ~(l,n+l) is identical to the first paragraph of the proof of Prop-
osition 3.4 (using Lemma 3.8 instead of Lermua 3.3). Thus ~(I,I) ~ ( 1 , 2 ) ~ ....
see that UA(l,n) is finite. Thus A(I,I) ~ A(I,2) ~... eventually stabilizes.
DEFINITION. A (I) will denote the limit set of the above sequence.
Note: We will soon show that P e A (I) if and only if for some minimal prime q,
P/q e A*(l+q/q).
Remark: We point out that A (I) is well behaved with respect to localization.
We have yet to show that A (I) ~ A (I). We choose to follow a path laid out
PROPOSITION 3.10. Let ~ be the Rees ring of R with respect to I. The follow-
(i) P e A (I)
Since the degree m component of t-n!R is (In+raN Im)t TM, and since pAR=P, we
have p = (In+m N ITM : c) = (In+m : c) since c e ITM. Clearly c ~ I n+m since P#R.
Then I n ~ InV c InV = Inv since InV is principal and V is integrally closed.
In general, say ql'''" qr' q r + l ' ' ' " qs are the min~nal primes of R with
N [(I+qi)/qi ]n (noting that I + q i is proper) which by the domain case shows that
n I n c_ ql N...N qr " Now let x e ql N...N qr " For any n > 0 and for any
m m
such that for any m>n and any ideal J with I cJ~-I , we have PeAss(R/J).
lemma shows that we may choose n large enough to assure that x @ In , as well.
Let m >__ n and let ITM c:::J c ITM. Now p2mk c_ (l+q) 2m c_ I m + q TM, so that
and P ¢ A (I).
17
Proof: Obvious.
can weaken the above hypothesis to just q is a prime divisor of zero. The
~,M) such that every minimal prime has depth 2 and there also exists a prime
over a R + q. However M ~ ~*(aR) since if it were, then by Lemma 3.7 ( and Prop-
is complete, R/p is quasi-unmixed and satisfies the altitude formula (see the
Appendix). Our next lermma thus shows that height M/p = I, contradicting that
depth p = 2.
of R then I-~N R = ~.
I-~= t'l~ ' N T, so that it will suffice to show that t-l~ ' N ~ = t-l~. Therefore,
statement of the lemma in the special case that I = bR is principal. For this,
Proof: By Lemmaa 3.15, we see that any prime in A (I) lifts to a prime in
A (IR). Thus suppose that P e A (IR) and let P=P NR. Also let
* * -i *
~ = R [ t - l , lt] and ~ =R [t ,IR t]. By Proposition 3.10, there is a
and so we easily see that p N is a prime divisor of t-l~. Also, in the proof
i) P e A (I).
, --, , *
v) There is a P e A (IR) with F NR=P.
Proof: We already have (i) <==> (ii) <==> (v). By Lermma 3.7 and Proposition 3.4, we
have (i) ~ (iv). We now show (iv) ~ (i), proceeding in three steps. First, assume
that I is principal and that R/q satisfies the altitude formula. Then I + q/q
is principal, and by Lemma 3.14 we have that height P/q = I (or else I + q/q = 0
over l+q, and so P e A (I) by Corollary 3.13. For step two, we only assume
By (i) ~ (v) and the fact that (R/q)* ~ R /qR , we have that
* * -, , * *
P /qR e A (IR + q R /qR ). Now by (i) ~ (iv) and the fact that minimal primes of
R* /qR* have the form q*/qR* with q* a minimal prime of R*, we see that there
satisfies the altitude formula, by the first step we have that P e A (IR). By
For the final step, we consider the general case. Let N0 be the Rees ring
of R/q with respect to I+q/q. Let q+=qR[t-l,t] AN, which is a minimal prime
of N, and notice that N/q+ ~ ~0 " Since P/q e A (I+q/q), (i) ~ (ii) shows
p e A*(t-IN), and so by (ii) ~ (i) we have P e ~(I). This proves that (iv)~(i).
For (ii) ~ (iii) we observe that t-nN= t-n~N R so that any p e ~*(t-IN) can
~, p' is also a prime divisor of t-l~. Finally, for (iii) ~ (ii), let p' be a
prime divisor of t-l~. Then height p' = I. Let q' be a minimal prime of
Since height p'/q' =I, Proposition 1.7 shows that p/q e ~*(t'l~+q/q). By
-* -I
(iv) ~ (i), p e A (t ~), so that (iii) ~ (ii).
motivated many of the ideas in the subject (such as Proposition 1.7) and so warrants
mention.
PROPOSITION 3.19. JR6] Let (R,M) be a local domain with completion R and in-
(v) There is an n > 0 such that for every ideal 0 # I c_M n, M e Ass(R/l).
Proof: (i) ~ (ii). Let I4 0 and let M be the maximal prime of R . Also
(iii) ~> (i) by Proposition 3.4, we have M e A (aR). By Proposition 3.18 used
R*/ q * satisfies the altitude formula, so Lemma 3.14 shows that depth q * = i.
(i) P e A (!).
(iii) For some ai, the integral closure of R[al/ai,... , an/ai] contains a
Proof: (ii) <~:~(iii) easily follows from Proposition 3.19 after the appropriate
localization.
(i) ~ (ii). Let ~=R[t-l, lt]. By Proposition 3.10, there is a p e ~(t-l~{) with
p'N~R=p. Let (V,N) = (c~p,, pp,), which is a D.V.R. By eermna 3.3, It ~_ p, Thus
pick a.l such that a.tl ~ p" With A = R [ a l / a i,..., an/ai] , we have A c_ V, since
contains a prime q e ~(t-l~) with qNA=Q. By Lemma 3.3, ait ~ q. Now Propo-
sition 3.19 shows that ~ contains a height I prime q' lying over qo Let
ing aj/ai= ajt/ait to ~, and a.tl ~ q" Thus the transcendence degree of @/q
that P e A (I).
Let I = (aI ,..., an, b), and let A=R[al/b,...,an/b]. Clearly A contains a
height i prime lying over P if and only if A[X] contains a height I prime
lying over PR[X]. Thus Proposition 3.20 easily shows that P e A (I) if and only
if PK[X] e A (IR[X]).
the domain case, we must show that Q ~ ~(IR[X])° Suppose, contrarily, that
f(X) a monic polynomial. By Proposition 3.20, we may write I= (aI ,..., an, b)
height ! prime q lying over Q. Let p=qOA. Since qNR =QNR=P, pNR=P
f(X) e Q e q and f(X) is monic so that f(X) ~ pA[X]. This contradiction com-
for some minimal prime q of R[X], by Proposition 3.18. Now q has form
carries IR[X]+q/q to (l+p/p)R/p[X], and since by the domain case just discussed,
the primes in A ((l+p/p)R/p[X]) have the form (P/p)R/p[X] for P/p e A (l+p/p),
we are done.
23
Proof: First suppose that R and T are domains, and let P be prime in R.
We must show that P c A (I) if and only if there is a Q e A (IT) with QNR=P.
R[Ib -I] contains a height i prime p lying over P. Now R[Ib -I] ~ T[ITb -I]
converse of the domain case is similar, except it needs the additional fact that
height i primes.
In the general case, suppose that P c A (I). By Proposition 3.18 (i) ~ (iv),
which must be minimal. Applying the domain case to R/p ~ T/q, we find a
(b) The above result fails for A (I). To see this, let (R,M) be a 2-dimensional
local domain which satisfies the altitude formula, but which is not Cohen-Macaulay,
N @ A (aR).
24
J(bd) -I. Thus R[Ib -I] ~ R[J(bd) -I] ~ V, so that R[Ib -I] ~ R[J(bd) -I] ~ V.
this prime lies over P, and because we already have J ~ P and bd e J, Propo-
COROLLARY 3.25. Let P e A (I). Then there is an n > 0 such that if J is any
and V as in Lermna 3.23. Since V is a D.V.R., for some n we have pnv c_ IV.
Corollary 3.24. The general case follows by P r o p o s i t i o n 3.18 ((i)~>(iv)) and the
(i) P e A (1).
(iii) P e A (Ic) for any element c not contained in any minimal prime.
(iv) There exists an element c not in any minimal prime, with P e A (Ic).
Proof: (ii) ~ (iii) is immediate, as is (iii) => (iv) since dim R > 0. W e will
prove (i) ~ (ii) and (iv) =~ (i) for R a domain. The general case is then straight-
(IJV : IV) = JV ~ (IJ : I)V. Thus (IJ : I ) V = (IJV : IV). As IJV ~ IV, Lemma 3.23
shows that P e A (IJ). Thus (i) :::::>(ii). For (iv) ~ (i), suppose c~0 and
P~A (J).
CHAPTER IV: A Characterization of A ( 1 )
that if I is an ideal in a local ring (R,M), then ~(I), the analytic spread
generators of In , then there is a polynomial P(X) such that P(n) = f(n) for
all large n (the Hilbert polynomial), and ~ (I) = deg P +I. If ~ = R [ t -I, It] and
then ~(I) is the size of a minimal basis of a minimal reduction of I. Recall also
~(I) = ~(IR(X)) (note: R(X) =R[X]pR[X]), by Proposition 3.21 we may assume that
obviously A (1)=A (J), and we may take l=J= (al,... , an) with ~(1)=n. Now
lying over P. Therefore, by Proposition 3.20 we now see that P e A (I). Con-
versely suppose that P e A (I) and that (R,P) is quasi-unmixed, so that it sat-
altitude formula also shows that height PA = height P - n + I. Thus height P = n = ~ (I).
27
This completes the domain case. Before doing the general case, we require two
lermnas.
LEMMA 4.2. Let I be an ideal in a local ring OR,M). For any minimal prime q,
Proof: Since the minimal number of generators of In+q/q does not exceed the min-
of Hilbert polynomials gives the inequality. Now in ~R, the Rees ring of R with
height n. Now if ~' is the Rees ring of R/q with respect to l+q/q, and if
N' = ...+ R/q t ' l + M / q + (l+q/q)t+ (12+q/q)t 2 +... and J' = ...+ R/q t - l + M / q +
(Ml+q/q)t+ (Ml2+q/q)t 2 +... then ~(l+q/q) =height N'/J'. However ~/Q ~ ~', the
isomorphism taking N/Q to N' and J+Q/Q to J'. Thus height N'/J' = n = ~ ( 1 ) .
LEMMA 4.3. (Rees) Let I be an ideal in a local ring 0R,M). Then ~(1)<heightM.
Proof: By the previous lerm~a, for some minimal prime q, ~(I)=~(l+q/q). Since
height M/q <_ height M, we may assume R is a domain. We may also assume that
R/M is infinite, and let (aI , ..., an) be a minimal reduction of I, with
inequality to the primes M and MR[al/a n,,.., an_l/an] , since the transcendence
degree of R]al/a n,..., an_l/a n ] modulo MR[al/a n .... , an.l/an] over R/M is n - I,
~(~). We may assume that R is local at P~ By Lermma 4.2, pick a minimal prime
sition 3.18, for some minimal prime, P/q e A (l+q/q). By the domain case,
Proof: By (i) => (v) and (i) ~ (iv) of-Proposition 3.18, M e A (I) if and only if
the principal class, i.e. for which the height of I equals the minimal number of
generators of I.) We could prove this now, using Corollary 4.4. Instead, we defer
it until Chapter 5, when we will have the machinery to give a very brief proof.
a,b,c be a system of parameters. Let I= (a,c) and J = (b,c). Since ~(I) does
not exceed the minimal number of generators of I, ~(I) ! 2 < height P° By Propo-
sition 4.1, P @ ~*(I). Similarly P @ ~(J). Now IJ = (ab, ac,bc, c2), and since
primes denoting images in R, let P = (X',Z'). We will show that A (P) = {P}.
Proposition 4.1. Since ~(Pp) does not exceed the minimal number of generators
Thus P is not principal. As R is normal [$2, Theorem i], ~(P) > I (since if
since P has two generators. However, the argument used above shows that ~(Pp) = i
LEMMA 4.5. Let (R,M) be a local domain with integral closure R, and let I# 0
TR ~ aR, showing that IR=aR. Conversely, suppose that iR=c~R for some a e R.
Lemma 4.5.)
Proof: If (i) holds, obviously P e A (I). If (ii) holds, use Proposition 3.19.
30
not satisfy the altitude formula (i.e. is not quasi-unmixed). Since is 2-di-
mensional, the only way this can occur is for [ to contain a height i prime
lying over P.
equals A (I).
(i) P e A (I)
(ii) P e A (I)
Proof: (ii) 4=>(iii) is immediate by Proposition 4.6. (ii) => (i) is by Proposition
3.17. Finally, suppose P e A (I), and assume that P is not minimal over I.
Remarks: (a) The class of domains in which A (I) always equals A (I) will be
(b) The equivalence of (i) and (iii) in Corollary 4.7 was first proved by P. Eakin
n>e.
Proof: Let n > e and suppose that M ~ Ass(R/In). Then height I = I and R/I n
Proposition 4.1 shows that M ~ ~(I). This contradiction proves the result.
Remark: We do not know if the hypothesis can be weakened to M e A (I) or the con-
However, it follows easily from IN1, E x e r c i s e 13, p. 1 0 3 ] , that such sequences are
if (x I ,..., Xn) # R and if for each 1 < i < n, x. is not in any prime contained
l
of R}.
liff, and the author. Rees obtained the first significant result [Rs2], which will
(Proposition 5.4) to be given in this chapter, first appeared in Ratliff's work JR9].
However, K a t z independently followed roughly the same trail [Kzl]. W e have borrowed
LEMMA 5.1. Let (R,M) be a local ring with completion R . Let x I ,..., x n be a
(iii) x I +q,...,x n+q is an asymptotic sequence in R/q for every minimal prime
q of R.
then height P / q = n .
Thus height P = n.
cipal class (i.e. height l = m i n i m a l number of generators of I). This does not work
Proof: One direction is immediate from Lermna 5.2. Thus suppose that
sequence.
34
Proof: Since R /q is quasi-unmixed, this is irmnediate from lemmas 5.1 and 5.3.
for each maximal ideal M. The result is now easy using P r o p o s i t i o n 5.4.
notation.
height(x I , . . . , Xn) ~ + q / q = n . Thus depth q >_ n, so that z(P) >_ n. This gives
half of the first equality. Now since x I , ..., x n is a maximal asymptotic sequence
In order to prove the second equality, it will suffice to show that the Q
plete the proof of the present proposition by proving the following lemma.
LEMMA 5.7. Let x I,..., x n be an asymptotic sequence and let Q e A (x I,..., x n).
Remark: The analog of Lermna 5.7 fails for classical R-sequences. That is, if
Proposition 5.6 shows that all asymptotic sequences maximal with respect to
We give several corollaries to Proposition 5.6. The first shows that Cohen-
sequences.
38
Proof: (i) ~ (ii) is trivial. For (ii) ~ (iii), by Proposition 5.6, we see that
gr (M) = z(M) for all maximal M. Thus height M = z(M) so that ~ is quasi-
to show (iii) ~ (i), in a locally quasi-unmixed ring, height P = z(P) for all
COROLLARY 5.9. Let ~,M) be a local ring. The following are equivalent.
(i) R is quasi-unmixed.
Proof: (i) ~ (iii) is irmmediate by Lemma 5.3, and (iii) ~ (ii) is trivial. For
COROLLARY 5.10. Let x I, ..., Xn be an asymptotic sequence and let the prime P
z(P) = Z(Pp) = n. However, by Lermna 5.2, we also see that height P = n. Thus
COROLLARY 5.11. Let the Noetherian ring R have the property that for any ideal I
in the principal class and any prime P, P e A (I) implies height P = ~ ( ~ ) . Then
R is locally quasi-unmixed.
Proof: Assume that R is not locally quasi-unmixed. Then for some prime Q, RQ
contains a minimal prime of depth less than height Q. Thus, using Proposition 5.6,
sequence coming from Q. By Lenmaa 5.2, I = (x I , ..., Xm) is in the principal class.
prime P e A (xI , ..., Xm) with Q ~ P. Now height P >__ height Q > m ~ ~ ( ~ ) ,
since analytic spread does not exceed the minimal number of generators. We have
Proof: Notice that by Proposition 5.6, z(P) = gr (Pp). Call this number n. Let
Pp e A ((x I ,..., Xn)Rp) , so that P e A ((x I,..., Xn)R). By Lemma 5.7, P e A (I).
asymptotic sequence. In p a r t i c u l a r , gr I ~ gr I ~ h e i g h t I .
Proof: It follows easily from [KI, Exercise 13, p. 103] that if P cA (xI ,..., x i)
gr (I) ~ height I.
the Appendix, (R/p) contains a minimal prime of depth n. Now (R/p) = R /pR ,
and minimal primes in R~/pR ~ have form p*/pR* with p a minimal prime of R .
Proof: (~) * contains a minimal prime of depth z(P), and (RQ/PQ)* contains a
minimal prime of depth z(Q/P). Therefore, a result proved in the Appendix shows
that (RQ) contains a minimal prime of depth z(P) +z(Q/P). Thus z(Q) ~ z(P) +
z(Q/P) °
Proof: The first inequality is obvious from the definition of asymptotic grade. We
over (P,a). Thus height Q/P = I. Obviously z(Q/P) = I, and by Proposition 5.15
z(Q) ~ z(P) + I. Therefore Proposition 5.6 shows that gr (I, a) ~ z(Q) ~ z(P) + I =
gr I + i.
Let R be a domain satisfying the altitude formula and having two maximal ideals
Since z(M) = i and z(N) = 3, clearly gr (I) = i while gr (l,a) = 3. This example
also shows that in the next corollary, the local condition is needed.
COROLLARY 5.17. Let I be an ideal in a local ring (R,M), and suppose that
P is as desired.
Proof: By Len=na 5.13, gr I <_ height I, which clearly cannot exceed n. The proof
of the rest of the result is analogous to the proof of [KI, Theorem 125].
The example [KI, Example 7, p. 102] shows that in the above proposition, if
C O R O L L A R Y 5.20. Let I = (aI ,..., an) and J = (b I ,..., h n) be two ideals in the
the p r e c e e d i n g proposition.
gr P = n.
of b is easily seen to not be in any minimal prime of R/aR. Thus that image is
N 2 , with heights 1 and 2 respectively, such that M=N INN 2 and such that
ii) for any finitely generated extension domain T of R, and any height 1
prime Q of ~, height Q N T = i.
Proof: i)~:=> ii) is trivial since a finitely generated extension domain of aNoetherian
ii) ~> i): Suppose that i) fails. Then by Proposition 5.11, there is an ideal
I = (a I , ..., an) of the principal class and a prime P e A (I) with heightP>~(Ip).
height 1 prime Q lying over P. By [D, Corollary 2], we see that PT is a prime
The first published result involving asymptotic sequences was by Rees [Rs2].
R are called an asymptotic sequence over I if (l,x I ,..., Xn) # R and if for all
local ring (R,M), and showed that r < height M - ~ ( I ) , and that equality holds
if R is quasi-unmixed.
As we now have more tools at our command than did Rees, we will go a bit
further, showing that in any local ring, all maximal asymptotic sequences over I
have the same length, which we characterize. This was done by Katz in [Kz2]. We
the Rees ring of R with respect to I, and let x e M with x not in any minimal
Proof: Write ~(I) and ~(l,x) for the Rees rings of R with respect to I and
(!,x) respectively. Also let Y be an indeterminate and write ~(I,Y) for the
Rees ring of R[Y] with respect to (I,Y). Let ~: R[Y] + R via Y + x. This
ker ~ c rad[(t'l,x)~(l,Y)]. However, we will defer the proof of the claim briefly.
of ~(I,Y) to the maximal ideal N ' = ...+ Rt - I + M + (I,x)t +... of ~(I,x). Since
~(I) +i.
We now return to proving our earlier claim. First note that ker ~ = (Y-x)R[Y]
since x e ~. This completes the proof of the claim, and of the proposition.
We now show that in a local ring, all maximal asymptotic sequences over I
PROPOSITION 6.3. [Kz2] Let I be an ideal in the local ring (R,M). Let
Proof: It follows easily from Proposition 3.18 that for any minimal prime q of
(a) R is quasi-unmixed
(c) Equality holds in (i) for all ideals I in the principal class.
(ii) (a) ~ (b). This is easy using Proposition 6.3, Lemma 4.2, and the fact that
sequence from M. Let I = (xI , ..., Xn). Now Lermna 5.2 shows that I is in the
case, r = 0. By (c), height M = ~ ( 1 ) . But ~(I) lies between height I and the
45
start by showing the numbers compare correctly (Proposition 6.6). First, a lermna.
LEMMA 6.5. Let I be an ideal in a local ring (R,M) and let P be a prime min-
PROPOSITION 6.6. Let I be an ideal in a local ring (R,M) and let r be the
Since R*/q is quasi-unmixed, gr* (M/I) = depth p = height (M*/q) - height (p/q)
LEMMA 6.7. Let I be an ideal and let x I ,..., x n be an asymptotic sequence over
Proof: W e may localize, and assume that R is local at P. The hypothesis shows
that x l+I,.,.,x n+I is a maximal asymptotic sequence from P/I. Thus gr (P/I) = n .
By Proposition 6.6, in the ring (R,P) the length of a maximal asymptotic sequence
x l+l,...,x i+l an asymptotic sequence in R/I for i < n. Let P/I be prime
sequence in R/I.
The converse of Proposition 6.8 is easily seen to fail, since A (I) can
PROPOSITION 6.9. Let I be an ideal in a local ring (R,M), and let Yl ''''' Yn
In particular, r+gr I ~ gr M.
we are done. If r > 0, then M @ A~k (I), and so by Lemma 5.7, M ~ A (Yl ''''' Yn )"
over I, then x~ ,..., X'r is a maximal asymptotic sequence over (l,Xl)). Since
use induction.
min{height I R * + q / q ] q i s a m i n i m a l p r i m e i n R , t h e c o m p l e t i o n o f R}.
PROPOSITION 6.11. Let I be an ideal in a local ring (R,M), and let r be the
(i) r+gr I = gr M.
(ii) For some minimal prime q of R*, ~(IR +q/q) = height(IR +q/q) = gr*l
(iii) The equalities in (ii) hold for every minimal prime q of R satisfying
depth q = gr M.
>__ (height M /q- ~(IR +q/q))+height(IR +q/q), the inequality holding since height
(i), we also have r+gr I = g r M = depth q. As we began and ended with depth q,
(ii) ~ (i). Let q be as in (ii). We have that gr*M <_ depth q = height M*/q =
COROLLARY 6.12. Let I he an ideal in a local ring (R,M), and let r be the
, , ,
Proof: Assume that r+gr I =helght M. Since r+gr I ~ gr M ~ height M, we have
r +~(I) = height M. The hypothesis and Corollary 5°8 show ~(I) = gr I = height I.
and let Q be a prime minimal over (P,a 1 , ..., an). Then Q e A (I,a 1 ,..., an).
Proof: We may localize at Q, and then using proposition 3.18, we may go to the
completion and the work modulo a minimal prime. That is, we may assume that (R,Q)
height Q since R is eatenary. Using Lermna 4.3, we see that ~(l,a I ,°.., an) =
In Lermma 6.13 it is not enough to assume that Q e A (P,a I ,..., an). To see
this, consider the example and remark preceeding Lemma 4.5. That example gave a
~(P)=2, such that A (P)={P}. Pick O~x e P and let I=ER. Then ~(I)=I.
PROPOSITION 6.14. Let I be an ideal in a local ring (R,M), and let r be the
Proof: Let P e A (I) and suppose that n=little depth P is minimal among
n > i, show that neither Pn-I nor M=P are in A (I). Thus
n
Pn-I ~ U[Q e A*(1)}. Pick a e Pn-I with a an asymptotic sequence over I. Of
6.13, we see that Pie A (i,a). As the length of a maximal asymptotic sequence
equals depth.
let r be the length of a maxi~nal asymptotic sequence over I. The following are
equivalent.
i) r=mln[depth P I P c A (I)]
height P = height M for all primes P. The equivalence of i) and ii) is now
LEMMA 6.16. Let I be an ideal in a quasi-unmixed local ring. Then (I) >
PROPOSITION 6.17. Let I be an ideal in a local ring ~,M) and let x I ,..., x n
(l,x I ,..., xi). This, together with the fact that any permutation is a product of
that x,y is an asymptotic sequence over I, and must show that y,x is an asymp-
of our hypotheses. B y Proposition 4.1 and Corollary 6.2, height p = ~ ( l , y ) =~(I) +I.
LEMMA 6.18. Let I be an ideal in a local ring R, and let I' be the image of
Proof: For all n, I n requires at least as many generators as I 'n. Thus the
4.1 says height P' = ~ ( I ' ) . N o w using the principal ideal theorem, Lemma 6.18, and
Finally L e m m a 4.3 shows height P = ~ (I,y I , ..., y m ) and so P r o p o s i t i o n 4.1 gives the
result.
sition 3.18 implies that those images are an asymptotic sequence over IR*+q/q,
we may return to (R,M) with the added assumption that it is a quasi-unmixed local
ring.
(x I,..., Xn_l) , we claim that x n' is an asymptotic sequence over I' This is
local ring, then ~(I) = ~ ( ( I , x I ,..., Xn)/(x I ,..., Xn)). W e begin with a lermna due
LEMMA6.21. Let I be an ideal in a local ring (R,M) and suppose the images of
52
Xl~'''~ ~ modulo ITM are an R-sequence in R/I m for all large m. Then
large n, Im/Ml TM ~ l'm/M'l 'm, the prime denoting modulo x I • To see this, note
the hypothesis on xI shows that r e ITM, so that rx e MI TM- Thus y e MI TM. Now
this isomorphism shows that ITM and I 'm require an equal number of generators.
forward to verify that R/I m /x I (R/I TM) ~ R'/I 'm. Since the images of x 2 , • . ., x n are
an R-sequence on the left hand side, going to the right hand side and applying the
PROPOSITION 6.22. Let I he an ideal in a local ring (R,M), and let x ! ,..., x n
analytically unramified. Thus for large m, J = ITM is normal (that is, all powers
are integrally closed) [SO]. We easily see that ~(I) =~(J) and A (I) (J) =
A (J). Since x I 4 U[P e A (I)}, by Lermma 6.21 we have ~(J) =~((J,Xl)/(Xl) ). Now
J = ITM implies ((l,Xl)/(Xl))m ~ (J,Xl)/(Xl) c_ ((l, Xl)/(Xl)) TM, from which we see
~((l, Xl)R /XlR ), we may suppose that R is complete. By Lemma 4.2, for some min-
imal prime q, ~ ( I ) = ~(l+q/q). In the complete local domain R/q, we get ~(I) =
~(l+q/q) =~((l,q,xl)/(q,xl)) < ~((l,Xl)/(Xl)) <__ ~(!), by two uses of Len~na 6.18.
Now inductively assume the result holds for Xl,..., Xn_ I • We first treat the
case that (R,M) is quasi-unmixed. In this case, Lemma 6.19 and Proposition 6.20
53
show that x' is an asymptotic sequence over I', the primes denoting modulo
n
(x I ,..., Xn_l). By the case n = i, ~(I') = ~ ( ( I ' , x ~ ) / ( X n ) ' ) = ~ ( ( I , x I ,..., Xn)/
(x I ,..., Xn)). Since our induction shows that ~(I) = ~(I'), we have proved our
since R*/q is quasi-unmixed, the preceding shows that ~(I) = ~ ( ( I , x I ,..., x n, q)R*/
(x I ,''', x n, q)R*) ! ~((I,x I ,''', Xn)R * /(x I ,''', Xn)R * ) = ~ ( ( I , x I ,''', Xn)/
(x I ,''', Xn)) ! ~(I), using Lemma 6.18. This completes the proof.
Let Yl ''''' Ym be an asymptotic sequence over I, and let primes denote modulo
(Yl ''''' Ym )" Then P' e ~ ( I ' ) if and only if P e A (I,y I ,..., ym ).
is an asymptotic sequence, we use Lemma 6.19 for one direction. Thus suppose that
using Corollary 6.2. Now Lemma 5.2 shows that height P' = h e i g h t P -m=~(I) = ~(I'),
--4.~
the last equality by Proposition 6.22. Thus Proposition 4.1 gives P' e A (I').
will have an interesting use for projective equivalence. For now, we present the
following.
Proof: As A (I)=A (J), the statement is true for xI . Assume it true for
A (I,x I ,..., Xn_l) = A (J,x I ,..., Xn_l). Suppose P c A (I,x I,..., Xn.l). We may
~((J,x I ,..., Xn.l)R +q/q), since our inductive assumption shows that
is identical.
CHAPTER VII: Asymptotic Grade
the sequence grade(R/I n ) n = 1,2,... eventually becomes constant, and if this con-
stant is ~, then ~(I) ! height M - ~ . This chapter presents these proofs and
tries (with only limited success) to extend them to grade(R/In). (Earlier work of
Burch [Bu] has shows that ~(I) ~ height M - lim inf grade(R/In), at least when
R/M is infinite.)
that there is an ideal I of R and an integer h > 0 such that for all n > 0
For (ii), consider the exact sequence 0 + An_I/A n + A/A n + A/An_ 1 ~" O, which
For (iii), let (R,M) be a local ring, and suppose that m is large enough
then x is not a zero divisor on A. To see this, suppose that xa = Oo The choice
of x and m shows that a e ~ for all k >__m. However, the hypothesis shows
Also the sequence A ' -- A 0' _ A i ~_... satisfies the hypothesis of the theorem, and
For (iv), we induct on ~, the case ~= 0 being trivial. Now suppose that
grade R(A/Ak) = ~ > 0 for all k >_ n, and let x I ,..., x~ be an R-sequence on
for k >_ n, and since our isomorphism shows that x 2 , ..., x~ is an R-sequence
Remark: The proof of (iii) above is a bit misleading. It does not show that
COROLLARY 7.3. (Brodmann [B2]) Let I be an ideal in a local ring (R,M). Then
COROLLARY 7.4. (Katz) Let (R,M) be an analytically unramified local ring, and
COROLLARY 7.5. (Katz) Let (R,M) be a complete local ring, and let I be an ideal.
Proof: Let Z = rad R. Thus R/Z is analytically unramified. Letting A = R/Z and
for n ~ i, A n = In/Z (noting that Z ~ In), and considering the ideal I +Z/Z of
stabilize?
PROPOSITION 7.6. Let (R,M) be a local ring, and let I he an ideal. Let
Xl,''', X s be elements of R whose images in R/I n are an R-sequence for all large
Proof: We induct. Suppose xlY = 0. Then xlY e I n for all large n, and by the
modulo X l , since R'/I 'n m R/(In,Xl ) ~ (R/In)/Xl(R/In), we see that the images of
and suppose that the images of x I ,..., x s are an R-sequence in R/I n for all
Proof: Since xI is not a zero divisor on R/I n for all large n, we see that
modulo (x I ,..., Xs_l) , Lermna 6.21 gives ~(I') = ~ ( I ) = h e i g h t P - (s-l). Since (R,P)
could have simply invoked Corollary 6.23, but that uses Proposition 6.22, which
Question: If the images of x I ,..., x s are an R-sequence in R/I n for all large
quasi-unmixed?
59
Question: Let (R,M) be a local ring and let I be an ideal. Let r be the
length of a maximal asymptotic sequence over I, and let ~ be the eventual stable
Remark: If the answer to this last question is yes, then our next result would
Proof: We induct on ~, the case ~= 0 following from Lemma 4.3. For ~ > O,
we have M 4 A*(1). Pick x e M-U {Q e A (I)}, and let a prime denote modulo x.
By Ler~na 6.21 we have ~(I) =~(I'). Also, Lermna 7.6 shows that x is not a zero
local domain which is not Cohen-Macaulay and let I be principal. Then ~(I) = i
and ~ = O.
Let (R,M) be a local domain with R/M infinite, and let I be an ideal.
In [Bu], Burch shows that ~(I) ~ height M - lin inf grade(R/In). Propositions 7.2
and 7.8 give a shorter proof, and also show that the lin inf is in fact the
limit ~. However, [Bu] also shows that ~(I) ~ height M - lin inf grade(R/In). As
(such as in Corollaries 7.4 and 7.5), we can combine that fact with Burch's second
PROPOSITION 7.9. Let I be an ideal in a local ring (R,M) with R/M infinite,
height M = 3.
the completion, and use Corollary 7.5. Since height M = height M and ~(i) = ~(IR )
PROPOSITION 7.10. Let (R,M)be a local ring and let I be an ideal. Let
--#¢
= grade(R*/InR *) for all large n. Then $ ~ height M - ~(I).
Question: Note that grade R/I n = grade R*/InR *. How does grade R/I n compare to
R* / InR * ?
CHAPTER VIII: When A = A
A (I) = A (I) for any ideal I. In this chapter, we will attempt to identify all
domains having that property, and will almost succeed. In particular, among domains
satisfying the altitude formula, we will see that Corollary 4.7 tells the whole
story.
The path we will take generally follows that laid out by R a t l i f f in [R6]. How-
domains in @.
Proof: Consider t as in Lemma l.l(b). Now x(I n+l : x) = I n + I N (x), and by the
(13 : 12) ~ (14:13 ) ~... eventually stabilizes, we let ~ = (In+l : I n ) for all
large n.
(iii) I=I.
(a) JC~.
(b) ~=~.
(c) For all large m, ITM = jm
(v) Let J be any ideal and suppose that for some m, jm = ITM and jm+l = im+l.
Then J c ~.
(vi) I c I c I.
Proof: ( i ) : For large m, ~ = (Im+l : Im), so ~m c (Im 2+m : Im2 ) = Im, by Lermna 8. I.
As I ~ ~, (i) is clear.
(ii): Using Lemma 8.1, for large m and any h, (Im(h+l) : Imh) = Im. Thus
ira= iTM.
(iii) : For large m, I = (~m+l : ~m). Using (i), this equals (Im+l : ITM) = ~.
(iv): (a) ~ (c): Since I c J c ~, (i) shows that Im=j m for all large m.
(c) ~ (b): For large m, ~ = (Im+l : iTM) = (jm+l : jm) = ~. (b) =3 (d) : Combine (b)
m-I _m m
with (i). (d) ~ (a): Suppose im = j m Then Jl ~_J = I , so J c (ITM : Im-l) c~.
Im=j m is not enough. Let R=K[X,Y] modulo (X2 - Y 4 ) . Let I = (X) and j = (y2)..
Hence x e ~.
LEMMA 8.4. Let (R,M) be a local domain which is not normal. Then for some
ac~ @ aR. Thus (a) # (a). However we clearly have (a) = (a). Therefore (a) ~_ (a)
LE~A 8.5. Let a,b be a pair of analytically independent elements in the local
Proof: Let I = (a,b) and J = (a2,b 2) c 12. Since a,b are analytically independ-
A (J)= {P prime I (a,b) ~ P}. In particular, only finitely many primes of R con-
rain (a, b) .
Proof: If dim R > 3, there is a height 2 prime P with depth P > I. Pick
infinitely many primes contain P, and hence contain (a,b). This contradicts
Len~na 8.6.
Proof: If dim R = I, then for any I~ 0, A (I) = [M} = A (I). If dim R = 2 and
2-dimensional and normal. Since dim R > I, choose a,b analytically independent
PROPOSITION 8.9. Let ~,M) be a local domain which is not quasi-unmixed (so
(i) If dim R = 2, R e
Proof: (i) Since the only way a 2-dimensional local domain can fail to be quasi-
M e A (I) for all I#0. As A (I) c A (I), we now easily see that A (1)=A (I).
over P.
For the converse, assume that R has the given properties. We will show that
R e @. Now Proposition 3.19 shows that M e A (I) ~ A (I) for all I # 0. Also
One case eludes our efforts. If (R,M) is a 3-dimensional normal local domain
which is not quasi-unmixed, then we do not know if R must be, might be, or cannot
be in @. (Only recently has it been shown that normal Noetherian domains which
are not quasi-unmixed exist. See [0] or [H2].) The next proposition might help
PROPOSITION 8.10. Let (R,M) be a 3-dimensional normal local domain which is not
Proof: Note that b e @ for any prime P of height 1 or 2, the first case
being obvious and the second using Corollary 4.7. Thus for any I, the primes of
we see that the only way to have A * (I) # A-~ (I) is to have M e A*(1) ~(I). Now
(a) = (a) so that I = (a), and hence M @ A*(I) since R is a Krull domain and
height M > I. That is, A (I) = A (1) whenever ~(1) = i. Now suppose ~(I) = 3.
By Proposition 4.1, M e A (I). Thus A (I)=A (I) whenever ~(I) =3. Since
(I) <__ height M = 3, the only remaining case is ~ (I)= 2. Let J = (a,b) be a
66
that for any analytically independent pair a,b, we had M e A (a,b). Then
(a2,b 2) ~ (a,b) 2 c_ (a2,b 2) so that clearly A*(a2,b 2) = A-* ( ( a , b )2) = A -* (a,b). Thus
M e A (a,b) as desired.
Having investigated when A (I)= A (I) for all i, we now look at two results,
PROPOSITION 8. Ii. Let (R,M) be a 2-dimensional local domain. Then the following
are equivalent.
(i) R is a U.F.D.
(ii) For any ideal l of R, A(I,I) =A(I, 2) ..... A (I)=~(I,i) =A(I,2) .... = A (I).
Proof: (i) ~ (ii): Any prime minimal over I is in each set in (ii). Thus we
then it is in all of them. To prove the claim, the first equivalence is aneasy
exercise using primary decomposition and the fact that R is a U.F.D. For the
second equivalence, we have (using that R is normal) J not principal if and only
if ~(J) > I if and only if ~(~) > i if and only if ~ is not principal. This
(iii) ~ (i): We first claim that R is normal. If not, by Lemma 8.4, for some
of ((a) +M(a)) n. By (iii) and the fact that (a)n ~ ((a)+M(a))n ~ (a) n c_ (an),
we see that M e A ((a)). Proposition 3.19 now shows that M e A (I) for any ideal
impossible since we may let I=P be a height i prime. This contradiction shows
i) R is Cohen-Macaulay.
Proof: (i) ~ (ii): Let I be in the principal class. The argument used in
[KI, Theorem 125] shows that i can be generated by elements x I , -.., x n with
an R-sequence. Using [KI, Exercise 13, page 103] we easily see that A (I)=
(ii) ~ (i) : Suppose that R is not Cohen-Macaulay. Let P be a maximal prime with
I = (xI ,..., Xn). Again using [KI, Exercise 13, page 103], we have P c A (I).
However since ~(Ip) does not exceed the number of generators of ~, we have
_ ~(~).
height P > n > By Proposition 4.1, P ~ A (I). As x I,..., x n being an
for any integral extension domain T and any prime q of T lying over Q, there
to hold for P ~ Q if R is normal, one easily sees that in the above definition
fails going down. Then for any ideal I with p c I c Q, we have Q e A (I).
contained in ql can lie over P. Thus p c ql ~ R fails going down. By the min-
COROLLARY 9.3. Let P ~ P' be primes in a Noetherian domain which fail going down.
Proof: Shrink P' to a prime Q minimal with the property that p c Q fails
set of primes of R, each of which contains P, and such that p c p' fails
Proof: For P' e W, use Corollary 9.3 to find Q e A (P) with Q ~ P' and
-,
p c Q failing going down. Obviously P~Q. Thus N~P' c W] contains n[Q e A (P)]
a special situation.
above P.
going down for all but finitely many of the primes Q directly above P°
we wish to give a result about going down whose proof has much in common with
an older result about chain conditions. In order to exhibit the similarities be-
DEFINITION• Let Q be a prime ideal and let W be an infinite set of prime ideals
LEMMA 9.6. Let I be an ideal in a Noetherian ring and let W be an infinite set
of primes each of which contains I. Then there is a conforming pair (Q,W) with
For primes P ~ Q, we will consider a relation which may or may not hold be-
is a conforming pair, then P *Q if and only if P * Q' for all but finitely many
Q' ~ W.
Proof: Suppose that P~Q and that (Q,W) is a conforming pair. Let W' = [Q' s w I
P*Q']. We must show that W' is finite. If W' is infinite, then (Q,W') is
clearly a conforming pair, and since P *Q' for all Q' e W', we must have P*Q
Similar reasoning shows that if P ~ Q' for all but finitely many Q' e W, then
P~Q.
LEMMA 9.8. Let R be a Noetherian ring and let * be a conforming relation. Let
Proof: Let W' consist of all the minimal members of W. If W' is infinite,
then Lemma 9.6 shows that there is a conforming pair (Q,W") with I c_ Q and
W" c W'. Since P*Q' for all Q' ~ W", we must have P*Q. Thus Q e W. This
contradicts that Q is properly contained in the members of W", since W" con-
ing relation with P*P. Then there is a chain of ideals p = I 0 c I 1 c...c In with
Proof: We let I0 = P and inductively construct the chain. Suppose that Im has
been constructed, and assume m is even (the case that m is odd being syn~etric).
Let W = [Q' e spec R I Im ~ Q' and P ~ Q ' ] . By Lermmas 9.7 and 9.8, W has only
stops. Suppose now that I is the largest ideal in our chain which is contained
m
in Q, (and still assume m even). If P ~ Q, Then Q e W and so by construction
we would have Im+ I ~ Q, a contradiction. Thus our chain has the stated property.
PROPOSITION 9.10. Let R be a Noetherian ring, and for each P e spec R suppose
Proof: Let P ~ Q and let (Q,W) be a conforming pair. Let !. be the largest
J
ideal in the chain associated with P such that I.j c-- Q. By the definition of
conforming pair, at most finitely many members of W can contain Ij+ 1 (lest
Ij+ 1 ~ Q). Thus Ij is the largest ideal of our chain contained in Q' for all
Remark: Propositions 9.9 and 9.10 together characterize conforming relations with
the property that P*P for all P e spec R. To accommodate P~P, we notice
LEMMA 9.11. Let (Q,W) be a conforming pair in a Noetherian ring. Then Q c Q'
Proof: Let height Q = n and let Q be minimal over (x I ,..., Xn). Let
Q = P0' PI' "''' Pm be all of the primes minimal over (x I ,..., Xn). Suppose that
W '= [Q' e W I Q c Q' fails catanicity}o Then for Q' e W', since height Q' >
72
height Q ' / Q + h e i g h t Q = h e i g h t Q'/Q +n, the principal ideal theorem shows that
height Q'/(x I ,°.., x n) > height Q'/Q. Thus Q' must contain one of PI'"" Pm °
This is true for all Q' e W'. We have P I N ... N Pm ~ N~Q' e W'}. Since
PlN'''N Pm ~ P 0 = Q' the fact that (Q,W) is a conforming pair shows that W'
must be finite.
Proof: Let P ~ Q and let ~,W) be a conforming pair. Letting a prime denote
pair in R'. Let W = [q e W I either Q c~ q fails catanicity or Q' c q' fails cata-
nicity}. By Lemma 9.11 applied to both (Q,W) and (Q',W'), we see that W is
finite.
Now let q e W - W. Then we have both height q = height q/Q + height Q and
height q = height q/Q +height Q ~ height q/Q + height Q/P +height P = height q/P +
We now show that going down is a conforming relation. The next lermma, a
LEMMA 9.13. Let (Q,W) be a conforming pair in a Noetherian domain. Then Q ~ Q'
Proof; Let W ' = [Q' e w I Q ~ Q' fails going down}. By Corollary 9.4, N[Q' e W'}
Proof: Let P ~ Q be primes and let ~,W) be a conforming pair. Suppose that
P ~ Q satisfies going down. If Q c Q' also satisfies going down, then obviously
p c Q' satisfies going down, Therefore Lemma 9.13 shows that P ~Q' satisfies
Now suppose that P ~ Q' satisfies going do~n for all but finitely many
Q' e W. we may ignore the finitely many exceptions and assume P ~ Q' satisfies
going down for all Q' e W. Our task is to show that P ~ Q satisfies going down.
Therefore, if q is a prime of R lying over Q, and if Pl ' "''' Pn are all the
primes of R lying over P, we must show that q contains one of Pl '''°' Pn "
By going up, for each 0' e W, there is a q' prime in R with q' N R = Q ' and
q C q'. Since P ~ Q' satisfies going down, each such q' contains one of
Pl ''''' Pn " Therefore some Pi is contained in infinitely many such q'. Let
U = [q'I Pi ~ q'}' We claim that ~[q' e U} =q. This will show that Pi ~ q' com-
pleting the proof. To prove the claim, let x e f](q' e U}. Since x is integral
and with ~ minimal among such. Since this expression is in q ~ q', and also
Recall that for R Noetherian, the prime Q is a G-ideal exactly when either
Q is maximal or dim(R/Q) = i and R/Q has only finitely many prime ideals [KI,
Theorem 146]. Recall also that the domain R satisfies going down if p c Q sat-
going down whenever Q is a G-ideal and height Q/P = i. Then R satisiies going
down.
74
Proof: Suppose some p c q does not satisfy going down. We may take a counter-
example with q maximal among all such. We claim that q is a G-ideal. Let
q' e W, p c q' has going down. By Proposition 9.14, p c q must have going down,
Fixing q, we now may assume that p has been chosen to make height(q/p)
minimal. If height(q/p) > I then consider the infinite set U = [p' e spec R I
p c p' c q and height (p '/p) = 17. By Corollary 9.5, there is a p' e U with pCp'
satisfying going down. Clearly p' c q cannot satisfy going down. However
In [MI] it is shown that if p c Q has going down for any Q and any height 1
prime P, then the Noetherian domain R has going down. This can be combined
with our previous ideas. To illustrate this in an easy setting, we will consider a
PROPOSITION 9.16. Let R be a catenary Noetherian domain and let n be less than
the height of any G-ideal of R. Suppose that p c Q has going down whenever Q
Proof: We first inductively reduce to the case that n = I. For this, suppose the
hypothesis holds for n and let p be a prime of height n-l. Furthermore let
going down.
The technique in the first paragraph of the proof of Proposition 9.15 allows us to
conclude that p c Q has going down for any height I P and any Q. The result
which P ~ Pl does not satisfy going down, while p c P2 does satisfy it.
has n > 2.
in the integral closure R, at least two primes, Q1 and Q2' lie over PI "
Let c e R with c e QI-Q2 and with c in every prime of R which lies over
We claim going down fails here. This is obvious, since (Q2' X)R[X] lies over
(PI ' X)R[X] and q ~ (Q2 ' X)R[X] since c ~ Q2 ° On the other hand,
P c (P2' X)R[X] does satisfy going down, since every prime of R[X] lying over
(P2 ' X)R[X] has form (Q,X)R[X], with Q a prime of R lying over P2 " By
(P2 ' X)R[X] with P ~ (PI' X)R[X] failing going down, while P ~ (P2 ' X)R[X]
l~J implies that A (I) = A (J), so that by Proposition 3.17 we see that A (I)
EXAMPLE. Let (R,M) be a 2-dimensional local domain which is quasi-unmixed but not
unmixed. That is, every minimal prime of R has depth 2, but there is a prime
for all J ~ I. On the other hand, Proposition 3.19 shows that M @ A (I) since R
this chapter, and see that it has connections to known results concerning ideal
transforms.
DEFINITION. Let R be a Noetherian ring with total quotient ring Q(R). Let I
T (IS).
77
vi) I does not consist of zero divisors modulo xT(1) n R for any regular x e I.
Proof: These are fairly simple. We prove only (v) and (vi). For (v), since ly
is finitely generated, for some large n we have In+ly c xR. As y/x e Q(R), and
In+l(y/x) ~ R, we have y/x e T(1). This proves (v), and (vi) follows easily.
at the integral case. The next lemma simplifies matters. Recall that z(P) is
a prime divisor of this ideal, then for some c e R we have P = (bR: ac)= (b-~: ac)
since bRNR=b-~. Thus z(P)=l. As (R : a/b) ~ (~: a/b) c p, we have a/b ~ R [I].
PROPOSITION 10.3 Let (R,M) be a local ring with M regular. The following are
equivalent.
i) T (M) c R.
(iii) ~ (ii): Clearly r~) = [ye Q~)I M~rad(R : y)}. By (iii), z~)#l, and
i) ~> iii): We will prove the contrapositive. Thus assume that (iii) fails. Since
M2nk ~ ((x)+Q)2n ~ (xn) + Q n Thus M2nks ~ (xn). Therefore s/x n e T(M), but
are equivalent:
i) T (I) c R.
(iv) ~=> (iii). We treat the eontrapositive. Thus suppose that I ~_ P and z(P) <__ I.
we see that P e A (I). As z(P) # i, (iv) fails. This shows that (iv) =# (iii).
(iii) ~> (ii): Since T(1) = {y e Q(R) I I e rad(R :y)}, this is obvious from the
definition of R [I].
COROLLARY 10.5. Let R be a Noetherian ring Then R [I] = UT(1) over all regular
if P is a prime containing I, then z(P) > I. Now the preceding corollary tells
COROLLARY 10.6. Let (R,M) be a local ring and let n = min{depth ql q is a minimal
Proof: By Corollary 10.4 we must show that z(P) > i for every prime P contain-
ing I. Suppose to the contrary that z(P) = i (z(P) = 0 being impossible since I
is regular). Now ~p) has a depth I minimal prime. Also (R/P) obviously
has a minimal prime of depth equal to depth P. Thus by a result proved in the
We remark that for a Noetherian domain R, R [I] = N ~ , z(P) =I. Thus R [I]
interesting that for (R,M) local, these two extensions differ by only a finite
PROPOSITION i0.7. Let (R,M) be a local ring. The set {P e spec R I z(P) = I
and height P > I. As this is true for each of the finitely many minimal q , we
are done.
following was known((ll)<==> (iii) due to Nishimura, (i) <==> (iii) due to Ratliff
JR4]).
80
THEOREM. Let (R,M) be a local ring with M regular~ The following are equivalent.
Our contribution (this being work done with Katz) will be to add the following two
regular element. Then T(I) is a finite R-module if and only if T(Ip) is a finite
from which it follows that xmT(1) N R ~ xR. We claim T(1) ~ Rx "m. Since x e I,
Proof: (i) -~- (ii): This is trivial since I c p implies TOP) c__T(1).
(iii) ~ (i): We make use of Lemma i0.8. Therefore let P e Ass(R/xR) with I c_ p.
By (iii) we have that TOPp) is a finite R~-module, and our goal is to show that
T(~) is a finite ~ - m o d u l e . Suppose this is false for some such P. We may assume
a finite A-module. In order to see this, we again use Len~a 10.8. Therefore let
that qp#Pp . Now Lermna 10.1(iv) shows that AQ= (~)qp=Rq, so that T(IAQ) =
Lermna 10.8, our claim that T(IA) is a finite A-module is now proved. As (iii) says
Therefore if In = (aI , ..., am) then for k >_ i, Xn+ k - x n = (Xn+k - x n + k _ I) +...+
a (limZr .). We note that lim Zrij exists in R since rij e I j c M j. Thus
m k+~ mj k+~
i) T(Pp) is an infinite ~ - m o d u l e .
82
iii) There is an m > i such that for any regular ideal J c p(m), p e Ass ~/J).
Proof: (i) ~ (ii): For this, we may assume that R is local at P. Letting R
Artin-Rees Lemma gives us that for some k and large n, (xn : y) , c__ (0 : y) , +
R R
x n - ~ *. As y e x n T(P * ), we see that (xn : y) , is P * primary. Thus (0 : y) , +
R R
xn-~ * is P* primary. Since y # 0, this obviously shows that R* has a depth I
(ii) ~ (iii): For this we again may localize at P (so that P(n) = pn) and pass
(~) ~ (i): Suppose that (v) holds but that T(Pp) is a finite ~-module. We will
total quotient ring). This easily implies that T(Pp) = (T(P)nR)p . Now for any
(iv) --~ (vi): Suppose (iv) holds. If x e P is a regular element and if xkT(P)N
x -k ~ T(Pp) so that for some m, Ppm c- XkRp . This shows that height P = i, prov-
ing (vi).
(vi) ~ (i): If height P = I then clearly (iv) holds, so (i) holds. Suppose that
k*
P e A*(xkT(p)n R) for large k. Then Pp e A (x T(Pp)N Rp) and the argument em-
k.
Question: Is P e A ~ (x~), x e P regular, k large, equivalent to the conditions
We may now add another equivalent statement to the list in Proposition 10.9,
are equivalent.
Proof: (i) ~ (ii) is by Proposition 10.9. Suppose now that (i) fails. Then by
(so Q c p and Q e Ass R). Now ( ~ , ) * = (~/Qp) * = (Rp) * /Qp(Rp) and Q /Qp(~)
Proof: For the first assertion, we have that P satisfies (iv) of Proposition I0.ii
For the second assertion, we let S = R -P and note that (R)s is a faithfully
PROPOSITION 10.15. Let P be a regular prime in a local ring (R,M) with comple-
T ( Pp~'~ ) an infinite R .' -modu i e. By Proposition I0.II~ P ' e A (J) for every
~ e ~
ideal J~0 of R ' (since R ' is a domain). Let x~0 be any element of P '
(jn=jn all n). As P ' e A (J), for large n, P ' is a prime divisor of jn=jn
the altitude formula so that Lemma 3.14 shows that height P ' = i. Thus
COROLLARY 10.16. Let P be a regular prime in a local ring (R,M). Then (~)
contains a depth I prime divisor of zero if and only if there exist primes
* , * * , * p*/Q*
Q c p of R with P nR=P, Q e Ass(R ) and height = i.
we do not know of a direct proof of the preceding. For the analogous result
for minimal primes, we do not need ideal transforms (hence we do not need P regular)
as we now show.
(Rp) contains a
depth 1 minimal prime if and o n l y i f t h e r e exist primes Q c p
* * * * *
in R with P NR=P, Q minimal, and height P /Q =I.
COROLLARY 10.18. Let P be a regular prime in a local ring (R,M). Then T(Pp) ~_
-- * * * * *
if and only if there are primes Q ~ P in R with P N R = P , Q minimal
and h e i g h t P /Q = 1.
Proof: (a): Since n = i, z0M) = I. Proposition 5.6 and Corollary 5.12 show
-4¢
M e A (I). Corollary 10.4 shows T(1) ~ R.
use Propositions 10.15 and 10.9 to see that T(1) is an infinite R-module.
COROLLARY 10.20. Let (R,M) be a local ring and let n = m i n [ d e p t h Q*IQ* e Ass(R*)}.
R-module.
Proof: If false, then by Propositions 10.9 and 10.15 there are primes Q c P in
* * = *A
R with Q e Ass(R*), height P*/Q* I and I c p R. As R*/Q* is catenary,
nition of n.
R [I] was defined in terms of primes with z(P) = i, that is, with ( ~ ) * having
R <I>= {y e QOR) I(R: y) ~ e for all primes P such that ( ~ ) * contains adepth Iprime
divisor of zero}.
T(1) c R (I> if and only if T(1) is a finite R-module. Also R <I>=UT(1) over
prime divisor of zero. As T(1) = [y e Q(R) II ~ rad(R: y)}, clearly T(1) ~ R <I>
For the converse, suppose that T(1) is an infinite R-module. Then Proposition 10.9
P e A (xR). Thus for some integer n and c e R, P = (xn: c). Now y = c/x n e Q ( R ) ,
Clearly UT(1), T(1) finite, is contained in R~IIL~. For the reverse inclusion
say z e R <I> and let I = (R: z). As z e Q(R), I is regular. Since z c R <I),
Propositions I0.9 and i0.II now show that T(1) is a finite R-module. Thus
Remark: In [B3], Brodmann shows that in a local ring (R,M), if the characteristic
stronger than our Corollary 10.20. We ask a question, which if true, is stronger
still.
Remark: In [Sc], the following result is reported. Let (R,M) be a local ring.
The following are equivalent: (a) There is a Q c Ass(R ) with depth Q < n,
88
not know if Proposition i0.II ((ii) <==>(v)) can also be extended. We now present a
sequence in (R,M). Suppose also that Q e Ass(R ) and height((Xl, x2)R +Q*/Q*)
height P /Q = 1, a contradiction.
CHAPTER XI: Miscellaneous
Rees' Valuations
are the ones mentioned in Proposition 3.18(iii), and this material will produce a
PROPOSITION II.I. Let I be an ideal in any ring R and let x e R. Then Vl(X ) =
Vl(Xn)
lim - - exists (possibly being i n f i n i t e ) .
n
VI (xn)
Proof: Let ~ = lira sup - - Let ~ < ~ and ~ < i. We will show that for all
n
Vl(Xn)
sufficiently large n, ~ ~. Since ~ can be arbitrarily close to l, and
Vl(Xn)
Vl(xkm ) ~ k V l ( X m ) ~ km~. This, together with n < (k+l)m, gives n
- - > m (k+l)
k
k+i $ ~ 8~° As this holds for all n ~ ~m, we are done.
Noetherian ring, it follows that x e Ik, First, however, we show that in a non-
Noetherian ring this need not by the case. For this, consider K(X,Y), with K a
Let W he the valuation ring of w and let I=YW. Since w(X) < w(Y), X 4 I.
Since for all n > I, (n-l,n-l) < (n, 0) < (n,n), we have X n e in-I - in. Thus
closed, so ~ = I, and X ~ ~.
LEMMA 11.3. Let R be a Krull domain and let I =uR be a principal ideal. Let
Proof: Let ~=min{ e ~ [ i= l,...,r} and let VI(X ) =k. We claim that ~-i < k
I
< ~. Since x e Ik = u ~ , we have v.(x) > vi(uk) = k e i, showing ~ > k. Now let
LEMMA 11.4. Let I be an ideal in any domain R, and let the domain T be an
Proof: Since In ~ j n clearly V--l(X) ! V--j(x). For the reverse, we first claim
_ jn
that jnNR c In . For this, let y e nR, writing y = alt l + . . . + a m t m with
in
over R. Since ywj e InTl , ywj = bjlW I +...+bjkW k with bj~ e A determinant
Consider ~ < ~j(x) with ~ rational. There are large integers n with
~l(X n)
~l(X n) ~ n~. Thus ~i(x)= n ~ ~" Letting ~ + VL(x) , we see that V--I(x)
be the Rees ring of R with respect to I. Let u = t-I . Let PI'''" Pr be the
Proof: Since um~NR=l TM, for all m >__ 0, clearly Vl(Xn ) =Vu~(X n ) for all n>__0.
Thus for x e R, Vl(X ) =Vu~(X). Now by Lerm~a 11.4 we see that Vl(X ) =Vu~(X ) . We
Proof: According to Proposition 11.2 we must only show that if Vl(X) = k, then
PROPOSITION 11.7. Let R be a Noetherian ring with minimal primes ql ' "" "' qs "
Proof: Let ~ <_ min{Vli(Xi)} with B rational, and let n be a positive integer
-- n
with n~ integral. Now n~ <_ min[Vl.(X i)}. Since R/qi is a domain, Corollary 11.6
1 n~ n
shows that x n e InD'l Thus I n~l reduces (I i ,xi) . By Lermaa 3.6, Inp reduces
(In~,xn), showing that x n e InD . By Proposition 11.2 ~l(X n) >__n~. Thus V--l(x)=
92
~I(X n)
> ~. This gives ~l(X) > min[~l.(Xi) I i = I,.. sl.
i
Now suppose y < VI(X ) with y rational. For infinitely many positive inte-
gers m, VI(X m) > m y and my is integral. Thus xm e Imy so that for i= l,...,s,
xim e I.
mY,l giving Vl.l(x~) ~ m y . As m+~, we have Vii(xi) ~ y , which shows that
Proof: This follows easily from Corollary 11.6 and Lemma 3.6.
Proof: i) Suppose Vl(X) =VJ(x) for all x e R. Using Corollary 11.8 we have
Conversly, suppose that I=~. Then for all k ~ I, Ik=l-k=J-k=Jk. Suppose now
-
that for some x, (x) > Vj(x), We will derive a contradiction. For sufficiently
large n, there will be an integer k such that ~l(X n) =n~l(X) ~ k > n~j(x)=
let $ < V--l(X) with ~ rational. There are infinitely many m with m~ integral
and with Vl(xnm ) > nm~. Thus x nm e Inm~ so that V in(X nm) ~__m~. As m-~oo, we
similarly.
similarly.
93
A Question of Krull
and he pointed out that the construction of a Noetherian example would probably re-
found (see [H2]) and shortly thereafter Ratliff used it to construct a Noetherian
counte~example to Krull's question. Although the construction does not require the
use of asymptotic sequences, using them gives a short path to the example.
primes in R.
Two lemmas will proceed our example. The first shifts attention from integral
Proof: Let p contain the monic polynomial f(Y). In some splitting field, let
f(Y) = (Y-Ul) .°. (Y-Un) , and let S = R [ u I ,..., Un]. Thus R ~ S is an integral ex-
tension. In S[Y], let p' be a prime lying over p. Since f(Y) c p', for some
easily seen to be the kernel of the map R[Y] +R[ui]. Since this kernel is contained
in p, the images under this map of p c q are two adjacent primes of R[ui],
LEMMA ii.II. Let ~,M) be a local domain of altitude at least 3. Let a,b be
prime M' lying over M. Then with X an indeterminate, R[X] is not a G.B.-ring.
Proof: Let x = a/b and pick y e M' with y not in any other prime of R[x]
which lies over M' NR[x]. Then N = M ' NR[x,y] is a height 1 prime of R[x,y].
p = p N R[X] and Q = q n R[X], to invoke Lemma ii. I0, we must show that heightQ/P > i,
Now P is the kernel of the map R[X] +R[x] and this map clearly sends Q to
By [D, Corollary 2], MR[x] is a prime with height equal to dim R - I ~ 2. Thus
EXAMPLE. Let (R,M) be a local 3-dimensional normal domain not satisfying the
z~) > I (Proposition 3.19). Therefore z(M) = 2. Using Proposition 5.6, let a,b
prime lying over M. By Len~aa II.II, R[X] is a normal Noetherian domain which is
not a G.B.-ring.
Remark: R[X I ,..., X n] not being a G.B.-ring is thoroughly studied in [R5] and
[B4].
Proof~ Suppose that z(P) > n - i. Since z(P) = gr Pp , let a I,..., a n be ele-
all but finitely many Q e W, the following is True: If p e UA ((a I,..., ai)),
i =0,1,...,n, and p e Q, then p ~ P. This is easily seen since UA ((a I,..., ai))
By deleting the finitely many exceptions in the above claim, we add the assump-
p _c p. We now claim that for all Q e W, the images of a I,..., an are an asymp-
totic sequence in RQ. If not, then for some Q e W and i=O,l,...,n-I we have
a prime pQ e A ((a I ,..., ai)RQ) with the image of ai+ I in pQ . Note that
pp e A ((a I ,..., ai)Rp) and the image of ai+ I is in pp . This contradicts that
second claim.
Finally, we claim that P e A ((a I ,..., an)). If not, pick an+ 1 e P with
an+ 1 not in any prime p satisfying p e A (a I ,..., an) and p c p, Our assump-
tion on W easily shows that the images of a I ,..., an+ 1 form an asymptotic
claim. We now have P e A (a I,..,, an) so that the images of a I ,..., an are a
pair, and Proposition II.12 shows that z(P) ~ n-l, a contradiction. Thus W' is
finite. Now let W ''= [Q e W 1 height Q > n + l ] . By Lemma 9.11, W" is finite.
Since we always have z(Q) ~ height Q, for any Q e w - (W' UW") we have n+l
Noetherian domain with primes p c Q such that height P = 2, height Q/P = i and
little height Q = 2. As another corollary of the above, we show that for a fixed
many such Q. (If R is not normal, there can be infinitely many such Q.)
Proof: Since z(Q) ~ little height Q (see the Appendix), we have z(Q) ~ 2 for
Proposition 11.12, z(P) = i (since P # 0 implies z(P) # 0). Thus (~) contains a
following definition.
are equivalent.
97
i) I is prenormal.
v) for all n.
Proof: (i) ~ (ii), and (iii) =~ (i) are obvious. For (ii) ~ (iii) we use Lermna 8.1
and have (In+~ : I~) = In for all large n and any ~. Since Inl ~ ~ In+9, and
s i n c e by (ii) we may choose ~ such that In+~ = In+~, we see that In c (In+~ : I~) =
In , proving (iii). For (iii) ~ (v), for any n, select m large enough that
DEFINITION. For I a regular ideal in a Noetherian ring, let C (I) denote the
Proof: One direction is trivial. Thus suppose C (I) = [PI' "''' Pm ] and that
is large enough that Ass(l-k/lk) =C*(1) and r~i = --~ for all i and
98
__ m
for all k > n. We claim that I k = Ik for all k > n. If not, pick x e ik - ik.
Then (Ik : x) is a proper ideal. By Lemma 1.2, for some d c R, (Ik : xd) is a
Proof: IQ is prenormal if and only if IQ=~ for all big n if and only if
A (I)UC (I).
Proof: We have ~ for all big n. By Proposition 8.2(ii), ~=~ for all
PROPOSITION 11.21. Let I= (aI ,..., an) he a regular ideal in a Noetherian ring
OOROLLARY 11.23. Let (R,M) be a local domain with R/M infinite. Let I be a
Remark: We do not know if this works for a prime other than the maximal ideal of a
local ring, the problem being that ~ a reduction of ~, does not imply that
Throughout this section, R will always be Noetherian, and all ideals will be
LEMMA 11.24. Suppose l~J, with In= Jm , n and m positive integers. Then if
all x e R. In order to conclude that n/m = k/~, we need only exhibit an x with
Corollary 11.8 also gives Vj(x) < ~. (Note: finding this x is the purpose of
LEMMA 11.25. Let l~J and let P be a prime containing I. Then ~ ~ Jp and
~ ( ~ , Jp) =y(I,J).
Then ~=~=J~ and ~ = N 2N = J N " Thus ~~JM and ~~JN" Yet l~J since
if l~J, then Lemma 11.25 would give y ( ~ , JM) =y(l,J) = ¥ ( ~ , JN ). However, ob-
The preceding example illustrates the main problem in going from local to glo-
(i) i ~J
(iii) ~Jp for all P e A (I)UA (J), and [y(IQ, JQ) IQ is minimal over I}
has size I.
Proof: (i) ~ (ii);~> (iii). Both follow from Lermna 11.25. Suppose (iii) holds,
with [y (IQ , JQ) IQ is minimal over I] = In/m}. We will show that in = jm, proving
This shows that rad ~ = rad Jp , so there is a prime Q ~ P with Q minimal over
both I and J. We already have y(IQ , JQ) =n/m, and so by Lemma 11.25,
~(l~n) = B(l,n)
it is identical to ~(l,n). The key is a lemma due to Katz, which will also give an
LEMMA 11.27. Let I be an ideal in a Noetherian ring R. Then for all n > m > i
Proof: Since In-m ~ (In : I TM) ~ (l--n: Im), it is enough to show (I-~: I TM) ~ In'm.
= i n-m
this holds for all such V, and as In'm (N In-mv) N R, we have x e . Now
for an arhitrary Noetherian ring R, we let q be any minimal prime ideal, and let
primes denote modulo q. Then (In : Im) ' ~ (l'n : I 'm ) ~ l'n-mo As this holds for
any minimal prime q, Lermma 3°6 shows that (l--n: I TM) ~ I ~'~m.
Proof: Clearly B(l,n) c_ ~(l,n) o Let P e ~(l,n) and write P = (In : c), e e R.
~(l,n) =B(l,n) • Furthermore, clearly p c (In+l : c~) while if x e (In+l : c~) then
We wish to give an example to show that in Proposition I0oli ((ii) ¢=> (iv))
and in Proposition 3.19 ((i) <==> (ii)) it is not enough to only consider prime ideals.
Thus we construct an n-dimensional local domain (R,M) such that M e A (P) for
all primes P ~ 0 (in fact M e A (I) for any ideal I containing a nonzero prime)
and M e Ass(R/P TM) for all m > 2, but such that R has no depth i prime di-
ular element. Suppose that RN R is a finite R-module. Then T(I) c R [I] if and
x
only if T(I) c R (I>.
Proof: Since R <I> ~ R [I], one direction is obvious. Thus suppose that T(1) c R [I]
the Altitude Formula with exactly two maximal ideals N1 and N 2 , both of height
process (See [DL]) there is a local domain ~,M) satisfying the Altitude Formula
with R=T, MT c_R (so that [ = T is a finite R-module), M = N 1 N N 2, and for any
with Q c__ q' and q' N R = p . Now q' ~ N 2 since Q ~ N2 . Since q c_N 2, q4q'
and these are two distinct primes lying over p. By the nature of R c T, we must
have p=M, so that q' = N 2 . This shows that N2 is minimal over PT as desired.
Therefore, by Proposition 3.5, M e A (P). (Note: The same argument works for any
We now wish to show that M ~ Ass (R/PTM) for m > 2. For this, choose
Since we already have N2 minimal over PT, for any m >_. 2 we have an se T - N 2
and a k >__ I with sNk c_pmT. Using that b ~ NI~N2=M and MT ~ R , weseethat
m>2.
Finally we want that R does not contain a depth I prime divisor of zero.
If it did, then by Proposition i0.ii and 10.21 we would have T(M) ~ R (I>. As
is a finite R-module, Lemma 11.29 shows that T(M) ~ R [I]. Thus byPropositionl0.3,
cept of strong asymptotic sequences which would stand in relation to prime divisors
of zero, as asymptotic sequences stand to minimal primes. This section will discuss
such a concept. As it was developed too late for major treatment in the text, this
postscript will point out the path, a full exposition appearing elsewhere. I do not
know if the definition of strong asymptotic sequence given here coincides with that
(Thanks to P. Schenzel for the nice notation, and for illustrating the significance
of this ideal.)
LEMMA ii.30. Let I be an ideal in a local ring (R,M). The following are equiv-
alent.
Proof: i) ~ i i i ) . If Rad I = M , then Im: < M > = R for all m > 0 and it is
trivial. Thus assume Rad I # M . Then ITM c] ITM : <M>#R, and so by (i), M is
0 with Rad ql =Q" Select 0 # x e (q2 N...N qn ) - ql " Let ITM : <M) ~ P ¢M, P
primary ideal. Since M is clearly not a prime divisor of ITM : <M>, primary de-
In case R is complete, the equivalence of (iii) is easy using [N, 30.1] and
ii) There is an integer k > 0 such that P e Ass R/J for every ideal J
iii) There is an integer k > 0 such that for all m > O, Im : <p> ~_ p(k)
im%)* : <(Pp) * > ~ (Pp) *k are all equivalent. Thus the equivalence of (i) and (iii)
q~
(i) ~ (ii). Choose k to equal the n of Lermma 1.13 applied to ~ (Pp) . If
P e Ass R/J.
(ii) ~ (iii). If P is not minimal over I, then Rad I =Rad(Im : <P)) and since
P ~ Ass R/(Im : <P>), (iii) follows easily from (ii). If P is minimal over I,
I c p and there is a k > 0 such that P e Ass R/J for all ideals J satisfying
q N R ? =Qp)o
Rad I and J ~ (P'N R) (k), then Rad JT = Rad IT and JT ~ p,(k) so that
The converse of Lermna 11.32 is true, but requires the following clever result.
Proof: Suppose for some k > 0 we have ITM : <j> ~_ jk for all m > 0. Increasing
with the property that this non-containment is true for some k and all m. We
claim for some q* e Ass (Rp)* , height l ( ~ ) ~'+ q ~ /' q ~~ =height J(Rp) * +q*/q*. To ease
106
notation, we will assume R = %)*. That is, we will assume 0R,M) is a complete
local ring, that M e A * (J), that ITM : <J> ~ jk for all m > O, and that if
p e A (J) - [M} then for all k there is an m with I~ : <Jp> c_ jkp . Now let
= iTM
Em : <J> and note that E 1 r~ E 2 ~ E 3 ~ .... Fix a k large enough that
Ass R/J k has stabilized to A*(J). Since A*(J) - {M) is finite, our assumption
E +jk/jk is M. The annihilator of any element of our module therefore has radical
m
equal to M. Since our module is finitely generated, its annihilator contains a
power of M. Thus for large m, E m + J k / J k has finite length, and as our modules
decrease as m increases, we find that for a fixed large k the ideals E +jk
m
stabilize for large m. Now R is complete in the M-adic topology, hence also in
the J-adic topology. The argument used to prove IN, 30.1] can be used to show that
co
if A E = 0 then for all k > 0 there is an m with E c j k However we are
m m --
m=l
assuming this containment fails for some k. Thus n E # 0. Let x @ 0 be in that
m
m=l
intersection. A well known corollary to the Artin-Rees Lemma shows that for large
contradicted.
k > 0 such that for all m > O, Im : (J) ~_ jk if and only if there is a
P c A,(I) with J c_ p.
Proof: Suppose such P exists. By Proposition ii.31, there is a k > 0 such that
for all m > O, Im : <e> ~ pOk) . As ITM : (p> c_ Im : <J>, and jk c_ p(k), we have
im : <J) ~ j k Conversely, suppose ITM : <J> ~_ jk for all m > O. By
Proof: Lenmaa 11.32 gives one direction, Thus suppose P e A.(1)° Let S = R - P,
so ~S : <PsTs > ~ P~Ts " By Lermma 11.34, there is a P' e Spec T such that
PS' e A.(IsT S) and PsTs ~ PS' " Of course P' e A.(IT), and since P c P' and
strong asymptotic sequence if (xI ,..., Xn) # R and for each i= l,...,n
(d) R-sequences are strong asymptotic sequences (using A.(1) ~ A (I) and the fact
that A (xI ,..., Xn.l) = A s s R/(x I ,..., Xn_l) if x I, ..., Xn_ I is an R-sequence).
asymptotic, a sequence must behave well modulo any minimal prime in (~)* for
any prime containing it. However strong asymptotic sequences behave well modulo
(f) If R c T
_ is a faithfully flat extension of Noetherian rings, x I,o.o, x n in
(Proposition 11.35.)
(g) Repeating, we do not know if this definition of strong asymptotic sequence co-
P e A,(I)}.
Thus n is equal to or less than the minimum of the first set above, which in turn
is clearly equal to or less than the minimum of the second set. Finally, the maxi-
DEFINITION. For I an ideal in a Noetherian ring, let the strong asymptotic grade
coming from I.
We now state some results without proof. They are not hard, using the forgoing
machinery.
PROPOS ITION ii. 38, If I is an ideal in a local ring (R,M), then gr.l =
q e Ass(~) ).
If (R,M) is a local ring and P c Spec R, with S = R -P, then (R)S and
Q e Ass R , and P n R = P } .
The analogue of Proposition II.40, which talks about minimal primes rather than
,
associated primes of zero is also true. Its proof uses gr PS '
APPENDIX: Chain Conditions
This brief appendix states facts referred to in the preceding text, with enough
references that the interested reader may pursue the matter. One new result is pre-
sented here.
for i = 0,1,...,n-l.
lii) The ring R is catenary if any two saturated chains of primes with
iv) The local ring (R,M) is quasi-unmixed if every minimal prime in the
*
completion R has depth equal to dim R.
the subdomain A.
Remark: It is well known that the Altitude Inequality holds for any Noetherian
It Was long known that affine rings are catenary. As a consequence of his
structure theorem for complete local rings, Cohen showed that complete local domains
not catenary. The following pair of results are fundamental (see [RI ] or [MD]).
THEOREM A2. Let R be a Noetherian domain. Then R satisfies the Altitude Formula
THEOREM A3. If the local ring (R,M) is quasi-unmixed, so is every finitely gen-
THEOREM A4. Let (R,M) be a local domain and let X be an indeterminate. Let
N e spec R[X] with NnR=M but N#MR[X]. Then the following three sets of in-
height 1 maximal then by going up to T and then going down to R, we see that
of Theorem A4.
The next result is referred to in Chapter 5 and Chapter i0. As it does not
9¢
THEOREM A5. Let P be a prime in a local ring (R,M). Suppose that (R/P) has
localized at (M/P,X). Under the natural map R[X] ~" (R/P)[X], suppose the inverse
image of that chain is PR[X] c PI c...c p c ~,X). Now this chain is saturated,
n
and by [HM, Corollary 1.5] we may assume that PINR=P (but PI#PR[X]).
q~[X], we easily find a saturated chain of primes of length m + l having the form
qR[X] c QI c . . . C Q m c PI "
[BI] M. Brodmann, "Asymptotic stability of Ass~/InM) '', proc. Am. Math. Soc.,
74(1979), 16-18.
[BR] M. Brodmann and C~ Rotthaus, "Local domains with bad sets of formal prime
divisors", J. Algebra, 75(1982), 386-394~
[Bn] L. Burch, "Codimension and analytic spread", Proc. Camb. Phil. Soc.,
72(1972), 369-373.
[El E.G. Evans, "A generalization of Zariski's Main Theorem", Proc. Am. Math.
Soc., 26(1970), 45-48.
[FR] D. Ferrand and M. Raynaud, "Fibres formelles d'un anneau local Noetherian",
Ann. Sci. Ecole Norm Sup., 3(1970), 295-311.
[Kz2] , "A note on asymptotic prime sequences", Proc. Am. Math. Soc.,
(to appear).
[~D] S. McAdam and E. Davis, "Prime divisors and saturated chains", Indiana
Univ. Math. J., 26(1977), 653-662.
[ME] S. McAdam and P. Eakin, "The asymptotic ass", J. Algebra, 61(1979), 71-81.
[Ni] J. Nishimura, "On ideal transforms of Noetherian rings II", J. Math. Kyoto
Univ., 20(1980), 149-154.
JR1] L.J. Ratliff, Jr., "On quasi-unmixed local domains, the altitude formula,
and the chain condition for prime ideals (I)", Amer. J. Math.~ 91(1969),
508-528.
[R2] , "On quasi-unmixed local domains, the altitude formula, and the
chain condition for prime ideals (II)", Amer. J. Math., 92(1970), 99-144.
n
[R3] , "On prime divisors of I , n large", Michisan Math. J.,
23(1976), 337-352.
[RM] L.J. Ratliff, Jr. and S. McAdam, "Maximal chains of prime ideals in in-
tegral extension domains I", Trans. Am. Math. Soc., 224(1976), 103-116.
[RR] L.J. Ratliff, Jr. and D. Rush, "Two notes on reductions of ideals",
Indiana Univ. Math~. J., 27(1978), 929-934.
[Rsl] D. Rees, "Valuations associated with ideals II", J. London Math. Soc.,
36(1956), 221-228.
[w] K. Whittington, "Prime divisors and the altitude formula", Ph.D. Disser-
tation, University of Texas at Austin, 1980.
[zs] O. Zariski and P. Samuel, Corm~utative Algebra, vol. II, D. Van Nostrand,
1980.
LIST OF NOTATION
Pase
A(l,n) ASS ~ / I n) 3
lim Vl (xn)/n 39
n->o~
z (P) rain[depth q*Iq* is a minimal prime of (~)*}
A I
Altitude Formula 17,41,61,]I0 l-Adic Topology 81
Altitude Inequality II0 Ideal Transform 76
Analytically Independent 63,65 Integral Closure
Analytic Spread 26 of an ideal 3
Artin-Rees Lermma 80,82 of a domain 4,29
Asymptotic Sequence 32
Asymptotic Sequence Over An Ideal 42,58
J
Jacobson Radical 38, 39
B
Brodmann, M. 1,2,51,55,57,87
Burch, L. 55,59 K
Katz, D. 15,32,42,55,57,100
Kaplansky, I. 93
C Krull~ W° 93
Catenary II0
Catenicity 71,72
Chain Conjecture 93,96 L
Cohen-MacAuley 30,35,67 Little Depth 48
Conforming Relation 70 Locally Quasi-Unmixed 28,36,37,53,
Conforming Pair 69,94 58,111
D M
Directly Above 69 Maximal Chain 110,II!
D.V.R. 12,24 Minimal Reduction 26
Monotone Sequence 2,13,15
Multiplicity 30
E
Eakin, P. 9, 30
Evans, E.G. 12 N
Nagata, M. 1 |0
Nishimura, J. 77,79,80
F Normal Domain 30, 63
Flat Extension 76, 84
Faithfully Flat Extension 17,84
0
O gom~s Example 65
G
m
index (continued)
R
Ratliff, L.J. Jr. 1,6,7,12,20,32,48,61,79,93
Reduction 14
Rees, D. 12,27,30,32,42,89
Rees Ring 7,15,42
R-Sequence 5,32,35,37,40,52,55,57
Rush, D. 61
S
Samuel, P. 53
Saturated Chain l]O
Sathaye, A. 2
Strong Asymptotic Sequence 88
System of Parameters 28,36
T
Transcendence Degree 12, I ]0
U
UFD 66
Unmixed Local Ring 76
Unmixedness Theorem 40
V
Valuation 90, 91
Vector Space 8
W
Whittington, K. 22,41
Z
n