Journal of Algebra 380 (2013) Contents lists available at SciVerse ScienceDirect. Journal of Algebra.

Transcription

1 Journal of Algebra ) Contents lists available at SciVerse ScienceDirect Journal of Algebra Quasi-finite modules and asymptotic prime divisors Daniel Katz a, Tony J. Puthenpurakal b, a Department of Mathematics, University of Kansas, Lawrence, KS 6645, United States b Department of Mathematics, Indian Institute of Technology Bombay, Powai, Mumbai 4 76, India article info abstract Article history: Received 3 August 211 Available online 9 February 213 Communicated by Bernd Ulrich Keywords: Quasi-finite modules Multigraded modules Asymptotic prime divisors Let A be a Noetherian ring, J A an ideal and C a finitely generated A-module. In this note we would like to prove the following statement. Let {I n } n be a collection of ideals satisfying: i) I n J n, for all n, ii) J s I s I r+s, for all r, s and iii) I n I m, whenever m n. Then I n C/ J n C) is independent of n, for n sufficiently large. Note that the set of prime ideals n 1 I n C/ J n C) is finite, so the issue at hand is the realization that the primes in I n C/ J n C) do not behave periodically, as one might have expected, say if n I n were a Noetherian A-algebra generated in degrees greater than one. We also give a multigraded version of our results. 213 Elsevier Inc. All rights reserved. 1. Introduction Let A be a Noetherian ring, J A an ideal and C a finitely generated A-module. Then, by a well known theorem of Brodmann see [1]), C/ J n C) is a stable set of prime ideals for n large. Brodmann s result has many applications and has been generalized in various forms. For example, see [5,8,11,7,6,2], among others. In this note we are motivated by the following question. Given an ideal J A, a finitely generated A-module C and a filtration of ideals {I n } n with J n I n for all n, when is the set of associated primes I n C/ J n C) a stable set of prime ideals? It turns out that the desired stability holds under very mild conditions on the filtration {I n } n. It is important to note that the set of prime ideals n 1 I n C/ J n C) is well known to be a finite set and the issue at hand is the realization that the primes in I n C/ J n C) do not behave periodically, as one might have expected, say if n I n were anoetheriana-algebra generated in degrees greater than one see [11]). Not surprisingly, our approach is through graded modules defined over finitely generated A-algebras. To elaborate, let R = n R n be a finitely standard graded A-algebra, i.e., R = A[R 1 ] * Corresponding author. addresses: dlk@math.ku.edu D. Katz), tputhen@math.iitb.ac.in T.J. Puthenpurakal) /$ see front matter 213 Elsevier Inc. All rights reserved.

2 D. Katz, T.J. Puthenpurakal / Journal of Algebra ) and R 1 is a finite A-module. Let M = n Z M n be a graded R-module with M n = foralln sufficiently small. We will assume throughout that each M n is a finitely generated A-module. Note that we do not assume that M is a finitely generated R-module. Throughout, we set L := H R + M) = n Z L n. We say that M is quasi-finite if, in addition, L n = forn sufficiently large. It turns out that for the theory of asymptotic prime divisors, it is the quasi-finite property that is crucial, and not the finite generation of the module M. Indeed, almost all the results we give in the context of graded modules are already known in the finite case see [9,11,2]). Of course, finite modules are quasi-finite. For an example of a quasi-finite module which is not finite, we will see below that A[X]/A[ JX] as an A[ JX]-module, where X is an indeterminate and J A is a proper ideal with positive grade. The notion of quasi-finite modules was introduced in [4]. It turns out that our results are not much harder to come by if we consider multigraded rings, i.e., standard N d -graded Noetherian A-algebras and multigraded modules over them. In section two, we define the types of modules we are interested in and, in particular, we extend the definition of quasi-finite module to the multigraded case. We then prove our basic results concerning asymptotic prime divisors of quasi-finite multigraded modules. In particular, we note that we can achieve the standard stability result known for finite modules see Theorem 2.7). In section three we give some specific examples and applications of the results in section two, especially in the singly graded case. Our problems would be simpler to solve but less interesting if the graded or multigraded) modules under consideration were always quasi-finite. To deal with modules M which are not quasi-finite, we look at the quasi-finite module M/L. In section four, we consider what happens in the case of multigraded modules that are not necessarily quasi-finite. It turns out that we can isolate the precise obstruction to stability of asymptotic prime divisors in the general case see Theorem 4.4). Finally, in section five we present a multi-ideal version of the result alluded to in the abstract. In particular, we prove the following theorem. Theorem 1.1. Let A be a Noetherian ring, C a finitely generated R-module and J 1,..., J d Afinitelymany ideals. Suppose that for each 1 i d, {I i,ni } ni is a filtration of ideals satisfying: i) I i, = A, ii) J n i I i i,ni, for all n i, iii) for all m i n i N,I i,ni I i,mi and iv) J r i I i i,si I i,ri +s i, for all r i and s i. Then there exists k = k 1,...,k d ) N d such that for all n = n 1,...,n d ) k, I1,n1 I d,nd C/ J n 1 1 J n d d C) = I1,k1 I d,kd C/ J k 1 1 J k d d C). 2. Quasi-finite multigraded modules Throughout this section R will denote a standard N d -graded Noetherian, commutative ring with identity, where N denotes the set of non-negative integers. We denote the degree,...,) component of R by A. Here, we use the term standard in the sense of Stanley, i.e., a standard N d -graded ring is one which is generated in total degree one. Rather than use excessive notation, we will simplify our notation and use n N d to indicate d-tuples. We will use subscripts to denote components of d-tuples. Thus n i means the ith component of n = n 1,...,n d ) N d. Superscripts will be used to indicate lists of d-tuples. Given n,m N d,wewillwriten m if n i m i,forall1 i d. Finally, we extend all of this notation in the obvious way to Z d Notation Let R = n N d R d be a Noetherian standard N d -graded ring as above. a) We will write R + for the ideal consisting of all sums of homogeneous elements x n R n such that n i 1, for all 1 i d. In other words, R + denotes the ideal of R generated by R 1,...,1). b) Throughout this paper, by a multigraded R-module we mean a Z d -graded R-module M = n Z d M n such that: i) Each component M n of M is a finitely generated A-module.

3 2 D. Katz, T.J. Puthenpurakal / Journal of Algebra ) ii) There exists a Z d such that M n = forn a Observation In the notation above, suppose that x M and R c x =, for some c N d with c,...,). Then x H R + M). To see this, suppose that t N is the largest component of c, so t >. Then R t c1,...,t c d ) R c x =. In other words, R t,...,t) x =. Since R is standard graded, R t + x =, therefore x H R + M). We now define multigraded quasi-finite modules. Definition 2.3. Let M = n Z d M n as above be a multigraded R-module. We say M is a quasi-finite R-module if there exists b Z d such that H R + M) n = forn b. Remark 2.4. Notice that if M is a Z d -graded R-module as above, then it follows immediately from the definition that M/H R + M) is quasi-finite as a Z d -graded R-module. More generally, the next proposition shows that a wide range of multigraded modules are quasi-finite. Proposition 2.5. Let U V be multigraded R-modules such that U is finitely generated over R and grader +, V )>.ThenM:= U/V is a quasi-finite R-module. Proof. Consider the exact sequence U V M. Taking local cohomology with respect to R + and using that H R + V ) = weget H R + M) H 1 R + U). But, U is a finitely generated R-module, so there exists b Z d such that H 1 R + U) n =, for n N d with n b. It follows that M is quasi-finite. For example, if R is the Rees algebra determined by an ideal of positive grade and V is the polynomial ring containing R, then, with U = R, M := V /U is an infinitely generated quasi-finite R-module. See section three.) The following proposition is well known in the case of finitely generated graded or multigraded modules. Since R is Noetherian, the proofs are the same even if M is not finitely generated over R. Proposition 2.6. Let R be a not necessarily standard N d -graded Noetherian ring and M as above be a multigraded R-module. Then P M) if and only if P M n ) for some n Z d. Moreover, P M n ) for some n Z d if and only if there exists a prime Q Ass R M) with Q A = P. Consequently, the following statements hold: i) n Z d M n ) is finite if and only if M) is finite. ii) If Ass R M) is finite, then n Z d M n ) is finite. Note also that for any multiplicatively closed subset S A, R S is a standard N d -graded Noetherian A S -algebra and M S is a multigraded R S -module with the original gradings preserved since the elements of S have degree zero. It follows easily from this that if P A is disjoint from S then P is the annihilator of an element of degree n in M if and only if P S is the annihilator of an element of degree n in M S. We will use this observation freely throughout this paper.

4 D. Katz, T.J. Puthenpurakal / Journal of Algebra ) Our first theorem shows that the quasi-finite notion is sufficient to guarantee asymptotic stability of prime divisors. Theorem 2.7. Let R be a standard N d -graded Noetherian ring and M as above be a quasi-finite R-module. Then there exists b Z d such that for all n,m Z d with b n m, M n ) M m ).Moreover,if M) is a finite set, then there exists k Z d such that for all n Z d with n k, M n ) = M k ). Proof. Now, let b Z d be such that H R + M) b = foralln Z d with n b. IfM n = foralln Z d with n b, then M n ) = for all such n and the conclusions of the theorem readily follow. Otherwise, M n for some n b. Without loss of generality, we may take n = b by increasing b if necessary) and assume that M b. Note that since M b H R + M), it follows from Observation 2.2 that for all n b, R n b M b. In particular, M n. Now, take c < h N d. We first show that M b+c ) M b+h ).LetP M b+c ).Without loss of generality, we may assume that A is local at P and P = : u), for u M b+c.wenownote that R h c u. Indeed, suppose R h c u =. Then by Observation 2.2 above, u H R + M). Since u, this contradicts our choice of b. Thus,R h c u. Therefore, xu, for some x R h c.since P xu =, P M b+h ),asrequired. Now, suppose M) is finite. Then by Proposition 2.6, n Z d M n ) is a finite set. Now, let {P 1,...,P r } denote the prime ideals n b M n ). We can write each P j = : A u j ), where u j M h j,forh 1,...,h r Z d,witheachh j b. Choose k Z d such that h j k, forall1 j r. Then, by the paragraph above, P j M k ) for all j and hence, P j M n,forallk n. On the other hand, if n k, thenn b. Thus,ifP M n ), P = P j, for some 1 j r. Thus, for all n Z d with n k, M n ) ={P 1,...,P r }= M k ), and this completes the proof of the proposition. 3. First applications Suppose we have a homogeneous inclusion of singly graded Noetherian R-algebras R S. In other words, S = n S n is a Noetherian ring with S = R = A and S n R n for all n. We assume R is standard graded, but do not assume that S is standard graded. Consider the R-module E = S/R. Notice that E = S n R n n= as an A-module. Note that E need not be a finitely generated R-module. We however have the following: Lemma 3.1. E) is a finite set. Proof. We use a technique due to Rees [1]. LetI denote the ideal of S generated by R 1. Then, as graded ideals in S, for each p, I p = n R p S n and I p+1 = n R p+1 S n. Thus, as a graded S-module, I p /I p+1 = n R p S n /R p+1 S n 1.IfwenowwriteG for the associated graded ring of S with respect to I, wethenhave G = R p S n /R p+1 S n 1, n,p and as such, G can be viewed as a finitely generated not necessarily standard) bigraded A-algebra. Thus, Ass G G) is finite, so by Proposition 2.6, G) is finite. The advantage of Rees s technique is due to the following observation. For t 1, consider the filtration of A-modules R t R t 1 S 1 R t 2 S 2 R 1 S t 1 S t.

5 22 D. Katz, T.J. Puthenpurakal / Journal of Algebra ) By breaking this filtration into short exact sequences, it follows that t S t /R t ) R t j S j /R t j+1 S j 1 ) G). j= It follows that t 1 R t /S t ) is finite, which gives what we want. The next proposition is just a variation on Proposition 2.4 and indicates when E as above is quasifinite. Proposition 3.2. For R, S and E as above, the following are equivalent 1. E = S/R is a quasi-finite R-module. 2. S is a quasi-finite R-module. In particular if grader +, S)> then S/R is a quasi-finite module. Proof. This follows from taking local cohomology with respect to R + ) of the short exact sequence R S S/R and noting that for each i wehaveh i R + R) j = forall j. The following corollary is an immediate consequence of Theorem 2.7, Lemma 3.1, and Proposition 3.2. It recovers from our perspective a special case of Theorem 1.1 from [2], thoughinourcase, we do not need to assume that S is standard graded. Corollary 3.3. Let R S be as in the previous proposition. If grader +, S)>,then S n /R n ) is independent of n, for n sufficiently large. Remark 3.4. It should be noted that the conclusion of Corollary 3.3 can fail if grader +, S) =. Indeed, in [4], Example 3.4, one has S n /R n = forn odd and S n /R n forn even. Thus, for this example, S n /R n ) is not stable for n large. In the example below, we use a result due to Herzog, Hibi and Trung from their paper [3]. Example 3.5. Let A = K [X 1,...,X d ] be a polynomial ring over a field K.LetI 1,...,I r be monomial ideals in A. Then there exists k N such that for all n k, r i=1 In i r i=1 I i) n = r i=1 Ik i r i=1 I i) k. Proof. Consider the algebra S = r n i=1 In i.by[3, Corollary 1.3] we get that S is a finitely generated A-algebra. Notice that R = n r i=1 I i) n is a standard graded A-subalgebra of S. SinceS is a domain we have that grader +, S)>. The result now follows by Corollary 3.3. We continue with our applications in the singly graded case, by way of illustrating the strength of the quasi-finite condition. We begin by letting J A be an ideal and {I n } n a filtration of ideals in A satisfying the following properties: I = A, J n I n for all n and J r I s I r+s,forallr and s. We set R := n J n, the Rees algebra of A with respect to J. For a finitely generated A-module C,

6 D. Katz, T.J. Puthenpurakal / Journal of Algebra ) we write U := n I nc, a not necessarily finite R-module. Let V := n J n C be the Rees module of C with respect to J.WesetM C := U/V. Notice that since n 1 C/ J n C) is a finite set, it follows that M C is a finite set. The following application is a special case of our main result in section five. Proposition 3.6. Maintain the notation established in the paragraph above. Suppose grade J, C) >. Then M C is quasi-finite. In particular, I n C/ J n C) is stable for all n. Proof. If grade J, C) >, then grader +, V )>, so M C is a quasi-finite R-module, by Proposition 2.5. For the second statement, M C ) is a finite set, by our comment in the previous paragraph, so the result now follows from Theorem The non-quasi-finite case We now return to the notation of section two. That is, R denotes a Noetherian standard N d -graded A-algebra and M is a not necessarily finitely generated) multigraded R-module satisfying our standard hypotheses. Throughout we set L := H R + M). Notice that M/L is quasi-finite, since H R + M/L) =. We begin with the following proposition. Proposition 4.1. Let R be a Noetherian standard N d -graded A-algebra and M a multigraded R-module. Then, M) = L) M/L). In particular, for any q Z d,ifm q /L q, there exists s N d such that M q /L q ) M q+s ).Moreover, if M is a finite set, then there exists k N d such that M n /L n ) = M k /L k ) for all n N d with n k. Proof. Because M) L) M/L), for the first statement it suffices to prove that M/L) is contained in M). But for this, the second statement suffices. Let P M q /L q ). Then we may write P = L q : A u), foru M q \L q.thus,p u L q,sothereexistst > such that R t + Pu =. In particular, P R su =, for s = t,...,t). Thus, P : A R s u). Suppose a A and a R s u =. Then, R t + au =, so au L q.thus,a P. It follows that P = : A R s u). SinceR s u is a finite A-module, it follows that P annihilates an element of R s u, and hence P M q+s ).Since we may choose the same t for all P M q /L q ) since M q /L q is a finite A-module), it follows that M q /L q ) M q+s ). This proves our second assertion. For the final statement, if M) is a finite set, it follows from the first statement of this proposition that M/L) is a finite set. However, M/L is quasi-finite, so our last assertion follows from Theorem 2.7. Remark 4.2. In the preceding proof, note that since u / L, R t + u L, so that if P AssM/L), P is the annihilator of an element of M\L. We now set L := : M R + ). Clearly L ) L). The following proposition supplies the reverse containment. Proposition 4.3. If P L n ) for some n N d,thenp L n+s ) for some s Nd.Inparticular, L ) = L). Proof. Only the first statement requires proof. Let P L n ). By localizing at P we may assume that A is local and P = : u) where u L n. Now, for some t 1, R t + u =. If t = 1, then u L n and P AssL n ), as required. Otherwise, we choose t > 1 least so that Rt 1 + u. Notice P Rt 1 + u =.

7 24 D. Katz, T.J. Puthenpurakal / Journal of Algebra ) Set s := t 1,...,t 1). Then there is x R s such that x u. We have P xu =. Also R + xu) =. Thus, xu L n+s. It follows that P L n+s ). Here is the main result of this section. It illustrates an obstruction to the asymptotic stability of M n ) in case M is not quasi-finite. Theorem 4.4. Let R as above be a Noetherian standard N d -graded A-algebra and M a multigraded R-module as above. Assume that M) is finite. Consider the following statements: i) There exists k Z d such that L n ) = L k ), for all n Zd with n k. ii) There exists l Z d such that L n ) = L l ), for all n Z d with n l. iii) There exists h Z d such that M n ) = M h ), for all n Z d with n h. Then i) implies ii) and ii) implies iii). Proof. We first note that if there exists t Z d with L n = foralln t, thenl n =, for all n t. Indeed, if some L n forn t, then there exists P L n ). But by Proposition 4.3, P L n+s ), for some s N d, and therefore, L n+s, a contradiction. Thus, if L n ) =, for all n k, then the same holds for L n ). Now suppose L n = L k ), for all n k. Fix n Nd with n k. Take P L k ). Then, by Proposition 4.3, there exists s N s with P L k+s ). Thus, P L k ),bychoiceofk. Thus,P L n ) L n ).Conversely,ifP L n ),then P L n+r ), for some r Nd. Thus, P L k ) L k ). Thus, L n ) = L k ), for all n N d with n k. Therefore i) implies ii). Now suppose that ii) holds. We have two cases. If L n ) = for all n l, thenl n =, for all n l, and therefore M is quasi-finite. Thus, the conclusion of iii) follows from Theorem 2.7. If L n ) for all n l, thenl n, for all n l, and thus, M n, for all n l. To continue, first note that since M) is finite, Proposition 4.1 implies that M/L) is also finite. By Theorem 2.7, thereexistsq Z d such that M/L) n = M/L) q for all n Z d with n q, sincem/l is quasi-finite. If the stable value of M n /L n ) =,thenm n = L n for all n q, and thus M n ) = M l ),foralln N d with n h, foranyh Zd with h q and h l, which gives what we want. Otherwise, if M q /L q, then by Proposition 4.1, thereexistsq Z d with q q such that M q /L q ) M q ). Take h Z d such that h l and h q.letp M h ).WewillshowP M n ) for all n h, and for this, we may assume that A is local at P.IfP L h ), then by the choice of l, P L n ) for all n Z d with n h. Thus P M n ) for all n h. If P / L h ), then P = : A u) for u M h \L h. Thus, by Observation 2.2 above, R n h u, for all n Z d with n h. Noten h. Since P annihilates R n h u and R n h u M n,wehavep M n ),foralln h. Now suppose n Z d and n h. TakeP M n ).IfP L n ),thenp L h ),byour choice of h and thus, P M h ).Otherwise,P M n /L n ).Thus,P M q /L q ),bythedefinition of q. Therefore, by the definition of q, P M q ), and by Remark 4.2 above, we may assume that P is the annihilator of an element of M q \L q. Thus, as in the second paragraph of this proof, we may move P forward so that P M h ),sinceh q.wenowhave M h ) = M n ),forall n h, which gives iii). Remark 4.5. a) Clearly, in light of Proposition 2.6, even in the singly graded case we cannot expect stability of M n ) without a finiteness condition on M). Moreover,inthepresenceof this condition, we cannot do better than Theorem 4.4. Indeed, notice that L is just a direct sum of A-modules. Thus, for example, let γ be an irrational number with a binary expansion γ := a 1 a 2 a 3 and take two distinct primes P 1 and P 2 in A. IfwesetR = A and M := n M n, where M n := A/P 1, for a n = and M n := A/P 2,fora n = 1, then L = M and M) is finite, but M n ) is neither stable nor periodic.

8 D. Katz, T.J. Puthenpurakal / Journal of Algebra ) b) Neither of the implications in Theorem 4.4 can be reversed. To see this, let R = k[x] denote the polynomial ring in one variable over the field k. LetT be the graded R-module R/X 2 R.Thus, T = T T 1,withT and T 1 one-dimensional vector spaces over k. Moreover,X 2 T =, X T and X T 1 =. Let M = n M n, where M n = T if n is even and M n = T 1 if n is odd. Then M = L := H R + M) and Ass k M n ) = ), foralln, so L n ) is stable for all n. On the other hand, setting L := : R R + ), it follows that L n = forn even and L n, for n odd. Thus, Ass kl n ) is not stable for n large. Therefore, statement ii) in Theorem 4.4 does not imply statement i). A similar example can be constructed to show that iii) does not imply ii) in the statement of Theorem Second applications Let C be a finitely generated A-module, J 1,..., J d be a family of ideals and for each 1 i d let {I i,ni } ni be a filtration of ideals satisfying: i) I i, = A, ii) J n i I i i,ni,foralln i, iii) I i,ni I i,mi, whenever n i m i, and iv) J r i i I i,si I i,ri +s i,forallr i and s i. An application of the main result of this section shows that if C is a finitely generated A-module, there exists k = k 1,...,k d ) N d such that for all n = n 1,...,n d ) k, I1,n1 I d,nd C/ J n 1 1 J n d d C) = I1,k1 I d,kd C/ J k 1 1 J k d d C). Not surprisingly, we accomplish our goal by recasting the given data in terms of multigraded rings and modules. Throughout the rest of this section, we fix ideals J 1,..., J d A. Keeping the notational conventions from the previous section, we will write J n for J n 1 1 J n d for all n = n d 1,...,n d ) N d. We will need the following definition. Definition 5.1. For J 1,..., J d as above, we call a collection of ideals {I n } n N d respect to J 1,..., J d if the following conditions hold: i) I,...,) = A. ii) J n I n,foralln N d. iii) For all n m N d, I m I n. iv) For all n, h N d, J n I h I n+h. a multi-filtration with Given a multi-filtration with respect to J 1,..., J d, it is now a simple matter to create a multigraded set-up to which we can apply the results of the previous section Notation Let {I n } n N d be a multi-filtration with respect to J 1,..., J d.letc be a finitely generated A-module. We set R := n N d J n, the multigraded Rees ring of A with respect to J 1,..., J d.wealsosetu:= n N d I n C, V := n N d J n C and M := U/V. Note that R is a standard N d -graded Noetherian ring, and U, V and M are multigraded R-modules satisfying the standard hypotheses from the previous section. Our goal now is to show that, with the notation just established, there exists k N d such that for all n N d with n k, M n ) = M k ).Wefirstnotethat M) is finite, since on the one hand, M n N d C/ J n C while on the other hand, it is well known that n N d C/ J n C) is finite in fact, ultimately stable), and therefore, n N d C/ J n C) is finite. We need the following lemma which follows from [8, Lemma 1.3] see also [5, Proposition 1.4]). Lemma 5.3. Let J 1,..., J d A be as above and suppose that C is a finitely generated A-module. Suppose that each J i contains a nonzero divisor on C. Then there exists k N d such that for all n Nd with n k, J n+r C : C J r ) = J n C, for all r N d.

9 26 D. Katz, T.J. Puthenpurakal / Journal of Algebra ) We are now ready for the principal result of this section, which immediately yields the main result of this paper. Theorem 5.4. Let A be a Noetherian ring, C a finitely generated A-module and J 1,..., J d Afinitelymany ideals. Suppose that {I n } n N d is a multi-filtration with respect to J 1,..., J d. Then there exists k N d such that for all n N d with n k, I n C/ J n C) = I k C/ J k C). Proof. Set R := n Nd J n and M := n N d I n C/ J n C, so that M is a not necessarily finitely generated multigraded R-module. Following the notation of the previous section, we set L := H R + M). Since M) is finite, by Theorem 4.4 it suffices to show that there exists l N d such that L n ) = L l ),foralln N d with n l. We now calculate an expression for L n,forn N d. Suppose u L n. Then there exists t N such that R t + u = inl, forallu L n since L n is a finitely generated A-module). It follows that if we set r = t,...,t) N d,then J r u J n+r C in C. Thus,L n = J n+r C : C J r ) I n / J n.note,r depends upon n. Now,setT := H J C), where J = J 1 J d.thus,each J j contains a non-zerodivisor on C/T. It follows from Lemma 5.3 that there exists k N d such that for all n Nd with n k and all r Nd, J n+r C : C J r ) J n C + T. Now, suppose n N d and n k. We may choose r as in the previous paragraph so that T = : C J r) and L n = J n+r C : C J r) I n / J n. In particular, J n+r C : C J r ) = J n C + T. Increasing k if necessary, we may further assume that T J n C =, by using the multigraded form of the Artin Rees lemma. Under these conditions L n = J n C + ) T I n / J n C = J n C + I n C ) T / J n C = I n C T. The path to the end of the proof is now clear. If we let P 1,...,P g denote L n ) = I n C T ), n k n k then each P i is an associated prime of some L h i with h i N d, h i k, 1 i r. Wenowjustusethe fact that for all m,n N d with m n k, L m ) L n.now,ifthereexistsc N d such that L n = foralln c, then we take l = c and note that L n ) =,foralln c. Otherwise, we order the set of primes P 1,...,P g so that for each P 1,...,P s,ifp i is among these primes, there exists a i N d with P / L a i ) and no such a i exists if s + 1 i g. It follows that if l N d and l ai for 1 i s, then for all n l, P i / L n ),if1 i s and P j L n ),fors + 1 j g. In particular, L n ) = L l ),foralln l, and the proof of the theorem is complete. We now provide the result stated at the beginning of this section. Corollary 5.5. Let A be a Noetherian ring, C a finitely generated R-module and J 1,..., J d Afinitelymany ideals. Suppose that for each 1 i d, {I i,ni } ni is a filtration of ideals satisfying: i) I i, = A. ii) J n i I i i,ni, for all n i. iii) I i,ni I i,mi, for all m i n i N. iv) J r i I i i,si I i,ri +s i, for all r i and s i.

10 D. Katz, T.J. Puthenpurakal / Journal of Algebra ) Then there exists k = k 1,...,k d ) N d such that for all n = n 1,...,n d ) k, I1,n1 I d,nd C/ J n 1 1 J n d d C) = I1,k1 I d,kd C/ J k 1 1 J k d d C). Proof. For n = n 1,...,n d ) in N d,seti n := I 1,n1 I d,nd. The result follows immediately from Theorem 5.4 since the collection of ideals {I n } n N d is a multigraded filtration with respect to J 1,..., J d. Remark 5.6. It is important to note that in the preceding theorem and corollary, we do not need to make any assumption that yields quasi-finiteness. This should be contrasted with Corollary 3.3 and Remark 3.4. Indeed, if we take d = 1 in the preceding corollary, and set R := n J n and S := n I n,thenwehave S n /R n ) is stable for large n, evenifgrader +, S) =. We now give two examples illustrating certain aspects of the theorem. Example 5.7. Let J 1,..., J d A be finitely many ideals, let K A be any ideal and set I n := J n : K, for all n N d.then{ J n : K } is a multi-filtration with respect to J 1,..., J d,sobytheorem5.4, there exists k N d such that J n : K )/ J n is stable for all n k. Note, that in general, n N d J n : K ) need not be a Noetherian A-algebra. Note also that, as the proof of Theorem 5.4 shows, the stability of AssM n ) = Ass J n : K )/ J n depends upon the stability of AssL n ),forl := H R + M). Wedetermine this latter set of primes. It follows from the proof of Theorem 5.4 that there exists c N d such that L n = J n : K ) H J A), foralln Nd with n c. Recall J = J 1 J d.) We now note that J n : K ) H J A) = H K A) H J A), for n sufficiently large in N d ). To see this, clearly H K A) H J A) J n : K ) H J A). Forthereverse inclusion, let u J n : K ) H J A). ThenuKr J n for some r 1. Notice uk r H J A). SouKr H J A) J n = for n large in N d, by the multigraded version of the Artin Rees lemma. Thus u H K A), as required. It follows that in the present case, AssH K A) H J A)) is the stable value of AssL n ). Finally, suppose H J A) H K A) and P AssH J A) H K A)). Then P AssA) and P must contain both J and K,elseH K A) P H J A) P =. Conversely, suppose P AssA) and P contains J and K.ThenP = : a) and thus J a =, so a H J A). Similarly, a H K A). It follows that H J A) H K A)) = AssA) V J + K ). Therefore, for all n sufficiently large in Nd,AssL n) = AssA) V J + K ). Example 5.8. Let J A be an ideal and take I A any ideal containing J and set I n := I n. Then, by Theorem 5.4, I n / J n ) is stable for all n. Let M = n 1 In / J n. Then by the proof of Theorem 5.4, L n = H J A) In,foralln, for L = H R + M). As in the previous example, we calculate the stable value of L n ). We now make two claims. Claim 1. H J A) In ) ={P I n P J}. To see this, note that if P J then H J A)) P =, so P / H J A) In ).Thus, H J A) In ) is contained in {P I n P J}. On the other hand, if P J and P I n ), then we localize at P and assume A is local at P and J A. SayP = : x) for x I n.thenp x =. So x : P) : J ).Thusx H J A) In. Therefore P H J A) In ). This proves the first claim. Now, set Z = nil-radical of A. Clearly we have the following exact sequence Z I n I n /Z )

11 28 D. Katz, T.J. Puthenpurakal / Journal of Algebra ) We make a second claim, Claim 2. For n sufficiently large, I n ) = Z)) {P P MinA) and I P}. Note, here we regard ZasanA-module.) To see, we first note that I n /Z) ={P P MinA) and I P}, forn large. Clearly any minimal prime P not containing I belongs to I n /Z) just localize at P ). On the other hand, if P I n /Z), thenp A/Z), sop is a minimal prime. But if P contains I, theni P is nilpotent. Thus, I n ) P = Z p for n large, so I n P /Z P =. Thus, P / I n /Z), a contradiction. Therefore, for n sufficiently large, It follows immediately from this and 5.8.1) that In /Z ) = { P P MinA) and I P }. I n ) Z) ) { P P MinA) and I P }, for all large n. For the reverse containment, clearly Z) I n ).Moreover,ifP MinA) and P I then by localizing at P one sees that P I n ). Thus, Claim 2 holds. Now, as before, setting L := H R + M), wehavel n = H J A) In, for all large n and by Claims 1 and 2, [ AssA Z) ) { P P MinA) and I P }] V J), is the stable value of L n ),forn sufficiently large. Remark 5.9. Regarding Claim 2, we have an analogue for powers of an ideal. We note that for n sufficiently large, I n ) ={P P and P I}. Indeed, if P A) and P I then I n P = R P.SoP I n ) for all n 1. On the other hand suppose P I n ) and P I, sayp = : c n ) where c n I n.soforn sufficiently large we have c n : P) I n = I n n I n : P) ) =, a contradiction. Thus, for sufficiently large n, ifp I n ), P A), and I P. Finally, we note the following, which should be clear from our results above. Let C be a finitely generated A-module and J 1,..., J d finitely many ideals. Assuming that n N d C/ J n C is finite, asymptotic stability of C/ J n C), n N d follows from Theorem 5.4 by taking the multi-filtration with respect to J 1,..., J d to be I n = A for all n N d. The finiteness of n N d C/ J n C follows fairly readily from basic properties of prime divisors of regular elements on a module by using extended Rees algebras. For example, see [8]. References [1] M. Brodmann, Asymptotic stability of AssM/I n M), Proc. Amer. Math. Soc ) [2] F. Hayaska, Asymptotic stability of primes associated to homogeneous components of multigraded modules, J. Algebra 36 26) [3] J. Herzog, T. Hibi, N.V. Trung, Symbolic powers of monomial ideals and vertex cover algebras, Adv. Math )

12 D. Katz, T.J. Puthenpurakal / Journal of Algebra ) [4] J. Herzog, T.J. Puthenpurakal, J.K. Verma, Hilbert polynomials and powers of ideals, Math. Proc. Cambridge Philos. Soc ) [5] D. Katz, S. McAdam, L.J. Ratliff Jr., Prime divisors and divisorial ideals, J. Pure Appl. Algebra ) [6] D. Katz, G. Rice, Asymptotic prime divisors of torsion-free symmetric powers of modules, J. Algebra ) [7] D. Katz, E. West, A linear function associated to asymptotic primes, Proc. Amer. Math. Soc ) 24) [8] A.K. Kingsbury, R.Y. Sharp, Asymptotic behaviour of certain sets of prime ideals, Proc. Amer. Math. Soc ) [9] S. McAdam, Asymptotic Prime Divisors, Lecture Notes in Math., vol. 123, Springer-Verlag, New York, [1] D. Rees, A-transforms of local rings and a theorem on multiplicities of ideals, Proc. Cambridge Philos Soc. 2) ) [11] E. West, Primes associated to multigraded modules, J. Algebra 271 2) 24)