arXiv:math/0505467v1 [math.AC] 23 May 2005 ON THE REGULARITY OF LOCAL COHOMOLOGY OF BIGRADED ALGEBRAS AHAD RAHIMI Abstract. The Hilbert functions and the regularity of the graded components of local cohomology of a bigraded algebra are considered. Explicit bounds for these invariants are obtained for bigraded hypersurface rings. Introduction In this paper we study algebraic properties of the graded components of local cohomology of a bigraded K-algebra. Let P0 be a Noetherian ring, P = P0 [y1 , . . . , yn ] be the polynomial ring over P0 with the standard grading and P+ = (y1 , . . . , yn ) the irrelevant graded ideal of P . Then for any finitely generated graded P -module M, the local cohomology modules HPi + (M) are naturally graded P -module and each graded component HPi + (M)j is a finitely generated P0 -module. In case P0 = K[x1 , . . . , xm ] is a polynomial ring, the K-algebra P is naturally bigraded with deg xi = (1, 0) and deg yi = (0, 1). In this situation, if M is a finitely generated bigraded P -module, then each of the modules HPi + (M)j is a finitely generated graded P0 -module. We are interested in the Hilbert functions and the Castelnuovo-Mumford regularity of these modules. In Section 1 we introduce the basic facts concerning graded and bigraded local cohomology and give a description of the local cohomology of a graded (bigraded) P -module from its graded (bigraded) P -resolution. In Section 2 we use a result of Gruson, Lazarsfeld and Peskine on the regularity of reduced curves, in order to show that the regularity of HPi + (M)j as a function in j is bounded provided that dimP0 M/P+ M ≤ 1. The rest of the paper is devoted to study of the local cohomology of a hypersurface ring R = P/f P where f ∈ P is a bihomogeneous polynomial. In Section 3 we prove that the Hilbert function of the top local cohomology HPn+ (R)j is a nonincreasing function in j. If moreover, the ideal I(f ) generated by all coefficients of f is m-primary where m is the graded maximal ideal of P0 , then by a result of Katzman and Sharp the P0 - module HPi + (R)j is of finite length. In particular, in this case the regularity of HPi + (R)j is also a nonincreasing function in j. In the following section we compute the regularity of HPi + (R)j for a special class of hypersurfaces. For the computation we use in an essential way a result of Stanley 1991 Mathematics Subject Classification. 13D45, 13D40, 13D02, 13P10. 1 and J. Watanabe. They showed that a monomial complete intersection has the strong Lefschetz property. Stanley used the hard Lefschetz theorem, while Watanabe representation theory of Lie algebras to prove this result. Using these facts the regularity and the Hilbert function of HPi + (P/fλr P )j can be computed explicitly. Pn Here r ∈ N and fλ = i=1 λi xi yi with λi ∈ K. As a consequence we are able n−1 r to show that HP+ (P/f P )j has a linear resolution and its Betti numbers can be computed. We use these results in the last section to show that for any bigraded hypersurface ring R = P/f P for which I(f ) is m-primary, the regularity of HPi + (R)j is linearly bounded in j. I would like to thank Professor Jürgen Herzog for many helpful comments and discussions. 1. Basic facts about graded and bigraded local cohomology Let P0 be a Noetherian ring, and let P = P0 [yL 1 , . . . , yn ] be the polynomial ring over P0 in the variables y1 , . . . , yn . We let Pj = |b|=j P0 y b where y b = y1b1 . . . ynbn P for b = (b1 , . . . , bn ), and where |b| = i bi .
Then P be a standard graded P0 -algebra and Pj is a free P0 -module of rank n+j−1 . n−1 In most cases we assume that P0 is either a local ring with residue class field K, or P0 = K[x1 , . . . , xm ] is the polynomial ring over the field K in the variables x1 , . . . , xm . We always assume that all P -modules considered here are finitely generated and graded. In case that P0 is a polynomial ring, then P itself is bigraded, if we assign to each xi the bidegree (1, 0) and to each yj the bidegree (0, 1). In this case we assume that all P -modules are even bigraded. Observe that if M is bigraded, and if we set M Mj = M(i,j) i L Then M = j Mj is a graded P -module and each graded component Mj is a finitely generated graded P0 -module, with grading (Mj )i = M(i,j) for all i and j. Now L let S = K[y1 , . . . , yn ]. Then P = P0 ⊗K K[y1 , . . . , yn ] = P0 ⊗K S. Let P+ := j>0 Pj be the irrelevant graded ideal of the P0 -algebra P . Next we want to compute the graded P -modules HPi + (P ). Observe that there are isomorphisms of graded R-modules i k HPi + (P ) ∼ = lim −→k≥0 ExtP (P/(P+ ) , P ) k i ∼ = lim −→ ExtP0 ⊗K S (P0 ⊗K S/(y) , P0 ⊗K S) k≥0 i k ∼ = P0 ⊗K lim −→k≥0 ExtP (S/(y) , S) i ∼ (S). = P0 ⊗K H(y) 2 Since HSi + (S) = 0 for i 6= n, we get  n P0 ⊗k H(y) (S) for i = n, i HP+ (P ) = 0 for i 6= n. Let M be a graded S-module. We write M ∨ = HomK (M, K) and consider M ∨ a graded S-module as follows: for ϕ ∈ M ∨ and f ∈ S we let f ϕ be the element in M ∨ with f ϕ(m) = ϕ(f m) for all m ∈ M, and define the grading by setting (M ∨ )j := HomK (M−j , K) for all j ∈ Z. Let ωS be the canonical module of S. Note that ωS = S(−n), since S is a polynomial ring in n indeterminates. By the graded version of the local duality theorem, see [1, Example 13.4.6] we have HSn+ (S)∨ = S(−n) and HSi + (S) = 0 for i 6= n. Applying again the functor ( )∨ we obtain HSn+ (S) = HomK (S(−n), K) = HomK (S, K)(n). We can thus conclude that HSn+ (S)j = Homk (S, K)n+j = HomK (S−n−j , K) for all j ∈ Z. Let Sl = L |a|=l Ky a . Then HomK (S−n−j , K) = M Kz a , |a|=−n−j where z ∈ HomK (S−n−j , K) is the K-linear map with  a−b z , if b ≤ a, a b z (y ) = 0, if b 6≤ a. Here we write b ≤ a if bi ≤ ai for i = 1, . . . , n. Therefore HSn+ (S)j = and this implies that (1) n HPn+ (P )j = P0 ⊗K H(y) (S)j = M L |a|=−n−j Kz a , P0 z a . |a|=−n−j
. Moreover, if P0 is Hence we see that HPn+ (P )j is free P0 -module of rank −j−1 n−1 graded M M Kxa z a . HPn+ (P )(i,j) = (P0 )i z a = |a|=i |b|=−n−j |a|=−n−j The next theorem describes how the local cohomology of a graded P -module can be computed from its graded free P -resolution Theorem 1.1. Let M be a finitely generated graded P -module. Let F be a graded free P -resolution of M. Then we have graded isomorphisms H n−i (M) ∼ = Hi (H n (F)). P+ P+ 3 Proof. Let Applying the functor We see that HPn+ F : · · · → F2 → F1 → F0 → 0. to F , we obtain the complex HPn+ (F) : · · · → HPn+ (F2 ) → HPn+ (F1 ) → HPn+ (F0 ) → 0. HPn+ (M) = Coker(HPn+ (F1 ) → HPn+ (F0 )) = H0 (HPn+ (F)), since HPi + (N) = 0 for each i > n and all finitely generated P -modules N. We define the functors: F (M) := HPn+ (M) and Fi (M) := HPn−i (M). + The functors Fi are additive, covariant and strongly connected, i.e. for each short exact sequence 0 → U → V → W → 0 one has the long exact sequence 0 · · · → Fi (U) → Fi (V ) → Fi (W ) → Fi−1 (U) → · · · → F0 (V ) → F0 (W ) → 0. Moreover, F0 = F and Fi (F ) = HPn−i (F ) = 0 for all i > 0 and all free P -modules F . + Therefore, the theorem follows from the dual version of [1, Theorem 1.3.5].  Note that if M is a finitely generated bigraded P -module. Then HPn+ (M) with natural grading is also a finitely generated bigraded P -module, and hence in Theorem 1.1 we have bigraded isomorphisms H n−i (M) ∼ = Hi (H n (F)). P+ P+ 2. regularity of the graded components of local cohomology for modules of small dimension Let P0 = K[x1 , . . . , xm ], and M be a finitely generated graded P0 -module. By Hilbert’s syzygy theorem, M has a graded free resolution over P0 of the form Lti 0 → Fk → · · · → F1 → F0 → M → 0, where Fi = j=1 P0 (−aij ) for some integers aij . Then the Castelnuovo-Mumford regularity reg(M) of M is the nonnegative integer reg M ≤ max{aij − i} i,j with equality holding if the resolution is minimal. If M is an Artinian graded P0 module, then reg(M) = max{j : Mj 6= 0}. We also use the following characterization of regularity reg(M) = min{µ : M≥µ has a linear resolution}. Let M be a finitely generated bigraded P -module, thus HPi + (M)j is a finitely generated graded P0 -module. Let fi,M be the numerical function given by fi,M (j) = reg HPi + (M)j 4 for all j. In this section we show that fi,M is bounded provided that M/P+ M has Krull dimension ≤ 1. There are some explicit examples which show that the condition dimP0 M/P+ M ≤ 1 is indispensable. We postpone the example to Section 4. First one has the following Lemma 2.1. Let M be a finitely generated graded P -module. Then dimP0 Mi ≤ dimP0 M/P+ M for all i. Proof. Let r = min{j : Mj 6= 0}. We prove the lemma by induction on i ≥ r. Let i = r. Note that M/P+ M = Mr ⊕ Mr+1 /P1 Mr ⊕ · · · . It follows that Mr is a direct summand of the P0 -module M/P+ M, so that dimP0 Mr ≤ dimP0 M/P+ M. We now assume that i > r and dimP0 Mj ≤ dimP0 M/P+ M, for j = r, . . . , i − 1. We will show that dimP0 Mi ≤ dimP0 M/P+ M. We consider the exact sequence of P0 -modules ϕ 0 → P1 Mi−1 + · · · + Pi−r Mr → Mi → (M/P+ M)i → 0. By the induction hypothesis, one easily deduces that dimP0 i−r X j=1 Pj Mi−j ≤ dimP0 M/P+ M, and since (M/P+ M)i is a direct summand of M/P+ M it also has dimension ≤ dimP0 M/P+ M. Therefore, by the above exact sequence, dim Mi ≤ dimP0 M/P+ M, too.  The following lemma is needed for the proof next proposition. Lemma 2.2. Let M be a finitely generated graded P -module. Then there exists an integer i0 such that AnnP0 Mi = AnnP0 Mi+1 for all i ≥ i0 Proof. Since P1 Mi ⊆ Mi+1 for all i and M is a finitely generated P -module, there exists an integer t such that P1 Mi = Mi+1 for all i ≥ t. This implies that AnnP0 Mt ⊆ AnnP0 Mt+1 ⊆ . . . . Since P0 is noetherian, there exists an integer k such that AnnP0 Mt+k = AnnP0 Mi for all i ≥ t + k = i0 .  Proposition 2.3. Let M be a finitely generated graded P -module. Then dimP0 HPi + (M)j ≤ dimP0 Mj for all i and j ≫ 0. Proof. Let P+ = (y1 , . . . , yn ). Then by [1, Theorem 5.1.19] we have H i (M) ∼ = H i (C(M). ) for all i ≥ 0 P+ where C(M). denote the (extended) Čech complex of M with respect to y1 , . . . , yn defined as follows: C(M). : 0 → C(M)0 → C(M)1 → · · · → C(M)n → 0 5 with C(M)t = M Myi1 ...yit , 1≤i1 <···<it ≤n and where the differentiation C(M)t −→ C(M)t+1 is given on the component Myi1 ...yit −→ Myj1 ...yjt+1 to be the homomorphism (−1)s−1 nat : Myi1 ...yit −→ (Myi1 ...yit )yjs , if {i1 , . . . , it } = {j1 , . . . , ĵs , . . . , jt+1 } and 0 otherwise. We set I = {i1 , . . . , it } and yI = yi1 . . . yit . For m/yIk ∈ MyI , m homogeneous, we set deg m/yIk = deg m−deg yIk . Then we can define a grading on MyI by setting (MyI )j = {m/yIk ∈ MyI : deg m/yIk = j} for all j. In view of Lemma 2.2 there exists an ideal I ⊆ P0 and an integer j0 such that AnnP0 Mj = I for all j ≥ j0 . We now claim that I ⊆ AnnP0 (MyI )j for all j ≥ j0 . Let a ∈ I and m/yIk ∈ (MyI )j for some integer k. We may choose an integer l such that deg m + deg yIl = deg myIl = t ≥ j0 . Thus am/yIk = amyIl /yIk+l = 0, because myIl ∈ Mt . Thus we have dimP0 (MyI )j = dimP0 P0 / Ann(MyI )j ≤ dimP0 P0 /I = dimP0 Mj . Since HPi + (M)j is a subquotient of the j-th graded component of C(M)i , the desired result follows.  Now we can state the main result of this section as follows Theorem 2.4. Let M be a finitely generated bigraded P -module such that dimP0 M/P+ M ≤ 1. Then for all i the functions fi,M (j) = reg HPi + (M)j are bounded. In a first step we prove the following Proposition 2.5. Let M be a finitely generated bigraded P -module with dimP0 M/P+ M ≤ 1. Then the function fn,M (j) = reg HPn+ (M)j is bounded above. Proof. By the bigraded version of Hilbert’s syzygy theorem, M has a bigraded free resolution of the form Lti F : 0 → Fk → · · · → F1 → F0 → M → 0 where Fi = k=1 P (−aik , −bik ). Applying the functor HPn+ (−)j to this resolution yields a graded complex of free P0 - modules HPn+ (F)j : 0 → HPn+ (Fk )j → · · · → HPn+ (F1 )j → HPn+ (F0 )j → HPn+ (M)j → 0. 6 Theorem 1.1, together with Proposition 2.3, Lemma 2.1 and our assumption imply that for j ≫ 0 we have dimP0 Hi (HPn+ (F)j ) = dimP0 HPn−i (M)j ≤ dimP0 M/P+ M ≤ 1 ≤ i for all i ≥ 1. + Moreover we know that HPn+ (M) = H0 (HPn+ (F)). Then by a theorem of Lazardsfeld [6, Lemma 1.6], see also [4, Theorem 12.1], one has reg HPn+ (M)j = reg H0 (HPn+ (F))j ≤ max{bi (HPn+ (F)j ) − i for all i ≥ 0} where bi (HPn+ (F)j ) is the maximal degree of the generators of HPn+ (Fi )j . Note that HPn+ (Fi )j = ti M M P0 (−aik )z a . k=1 |a|=−n−j+bik Thus we conclude that reg HPn+ (M)j ≤ max{aik − i} = c for j ≫ 0, i,k as desired.  Next we want to give a lower bound for the functions fi,M . We first prove Proposition 2.6. Let dp d G : 0 → Gp → Gp−1 → · · · → G1 →1 G0 → 0, L be a complex of free P0 -modules, where Gi = j P0 (−aij ) for all i ≥ 0. Let mi = minj {aij }. Then reg Hi (G) ≥ mi . Proof. Since Hi (G) = Ker di / Im di+1 and Ker di ⊆ Gi for all i ≥ 0, it follows that reg Hi (G) ≥ ≥ ≥ ≥ = largest degree of generators of lowest degree of generators of lowest degree of generators of lowest degree of generators of mi , as desired. Hi (G) Hi (G) Ker di Gi  Corollary 2.7. Let M be a finitely generated bigraded P -module. Then for each i, the function fi,M is bounded below. Proof. Let G be the complex HPn+ (F)j in the proof of Proposition 2.5, then the assertion follows from Proposition 2.6.  7 Proof of Theorem 2.4. Because of Corollary 2.7 it suffices to show that for each i, fi,M is bounded above. ϕ There exists an exact sequence 0 → U → F → M → 0 of finitely generated bigraded P -modules where F is free. This exact sequence yields the exact sequence of P0 -modules ϕ 0 → HPn−1 (M)j → HPn+ (U)j → HPn+ (F )j → HPn+ (M)j → 0. + Let Kj := Ker ϕ. We consider the exact sequences 0 → Kj → HPn+ (F )j → HPn+ (M)j → 0 Thus we have 0 → HPn−1 (M)j → HPn+ (U)j → Kj → 0. + (2) reg Kj ≤ max{reg HPn+ (F )j , reg HPn+ (M)j + 1} (3) reg HPn−1 (M)j ≤ max{reg HPn+ (U)j , reg Kj + 1}. + Let F = Lk i=1 P (−ai , −bi ), then HPn+ (F )j = k M M P0 (−ai )z a . i=1 |a|=−n−j+bi Therefore, reg HPn+ (F )j = maxi {ai }. By Proposition 2.7, the functions fn,M and fn,U are bounded above, so that, by the inequalities (2) and (3), fn−1,M is bounded above. To complete our proof, for i > 1 we see that HPn−i (M)j ∼ (U)j . Thus = HPn−i+1 + + fn−i,M = fn−i+1,U for i > 1. By induction on i > 1 all fi,M are bounded above, as required. 3. The Hilbert function of the components of the top local cohomology of a hypersurface ring Let R be a hypersurface ring. In this section we want to show that the Hilbert function P0 -module HPn+ (R)j is a nonincreasing function in j. Let f ∈ P be a P bihomogeneous of degree (a, b). Write f = |α|=a cαβ xα y β where cαβ ∈ K. We may |β|=b P also write f = |β|=b fβ y β where fβ ∈ P0 with deg fβ = a. The monomials y β which |β| = b are ordered lexicographically induced by y1 > y2 > · · · > yn . We consider the hypersurface ring R = P/f P . From the exact sequence f 0 → P (−a, −b) → P → P/f P → 0, we get an exact sequence of P0 -modules M M f P0 (−a)z c → P0 z c → HPn+ (R)j → 0. |c|=−n−j+b |c|=−n−j c We also order the bases elements z lexicographically P induced by z1 > z2 > · · · > zn . c Applying f to the bases elements we obtain f z = |β|=b fβ z β−c , where z β−c = 0 if β 6≤ c componentwise. With respect to these bases the map of free P0 -modules 8 
−j+b−1
is given by a −j−1 matrix which we denote by Uj . This matrix also den−1 n−1 scribes the image of this map as submodule of the free module Fj where Fj = L c n n |c|=−n−j P0 z , so that HP+ (R)j is just Coker f = Fj /Uj . Note that HP+ (R)j = 0 for all j > −n. Let BdP denote the set of all monomials of degree d in the indeterminates z1 , . . . , zn . Let h = v∈B−n−j hv v ∈ Uj where hv ∈ P0 for all v. Then hu u is called the initial term of h if hu 6= 0 and hv = 0 for all v > u, and we set in(h) = hu u. The polynomial hu ∈ P0 is called the initial coefficient and the monomial u is called the initial monomial of h. Now for a monomial u ∈ B−n−j we denote Uj,u the set of elements in Uj whose initial monomial is u, and we denote by Ij,u the ideal generated by the initial coefficients of the elements in Uj,u . Note that [ Uj \ {0} = Uj,u . u∈B−n−j We fix the lexicographical order introduced above, and let in(Uj ) be the submodule generated by {in(h) : h ∈ Uj }. Then M (4) in(Uj ) = Ij,u u. u∈B−n−j Proposition 3.1. With the above notation we have Ij,u = Ij−1,z1 u for all j ≤ −n and u ∈ B−n−j . Proof. Let h0 ∈ Ij,u . Then there exists h ∈ Uj such that h = h0 u+lower terms. We set k = −n − j + b, for short. Since h P is in the image of f , we may P also write h = P c c c−β c+e1 f f z where f ∈ P and f z = f z . We define g = c 0 |c|=k c β≤c β |c|=k fc f z P where f z c+e1 = β≤c+e1 fβ z c+e1 −β and e1 = (1, 0, . . . , 0). We see that g ∈ Uj−1 . We may write X X X X fβ z c+e1 −β . fβ z c+e1 −β + fc g= fc |c|=k β≤c |c|=k β6≤c β≤c+e1 Thus we conclude that g = z1 h + h1 where X X h1 = fc fβ z c+e1 −β . |c|=k β6≤c β≤c+e1 We now claim that h1 does not contain z1 as a factor. For each α ∈ Nn we denote by α(i) the i-th component of α. Assume that (c + e1 − β)(1) > 0 for some β appearing in the sum of h1 . Then c(1) ≥ β(1). Moreover, if i > 1, then (c + e1 − β)(i) ≥ 0 implies that c(i) ≥ β(i). Hence c(i) ≥ β(i) for all i, a contradiction. It follows that in(g) = in(h)z1 . Therefore hu ∈ Ij−1,z1u . Conversely, suppose h0 ∈ Ij−1,z1 u . Then P there exists g ∈ Uj−1 such that g = h0 z1 u+ lower terms. We may write g = |c|=k fc′ f z c+e1 where fc′ ∈ P0 and f z c+e1 = 9 P β≤c+e1 fβ z c+e1 −β . Thus X X X X fβ z c+e1 −β . fβ z c+e1 −β + fc′ g= fc′ |c|=k β≤c |c|=k β6≤c β≤c+e1 P As above we see that g = z1 f ′ + lower terms, where f ′ = |c|=k fc′ f z c . We see that f ′ ∈ Uj , and that in(f ′ )z1 = in(g) = h0 z1 u. Therefore, in(f ′ ) = h0 u, and hence h0 ∈ Ij,u .  P Let M and N be graded P0 -modules. We denote by Hilb(M) = i∈Z dimK Mi ti the Hilbert-series of M. We write Hilb(M) ≤ Hilb(N) when dimK Mi ≤ dimK Ni for all i. Let F be a free P0 -module with basis β = P {u1 , . . . , ur }. Let U be a graded r submodule of F . For f ∈ U, we write f = i=1 fi ui where fi ∈ P0 . We set in(f ) = fj uj where fj 6= 0 and fi = 0 for all i < j. We also set in(U) be the submodule of F generated by all in(f ) such that f ∈ U. Let I be a homogeneous ideal of P0 . We say that set of homogeneous elements of P0 forms a K-basis for P0 /I if its image forms a K-basis for P0 /I. Now we can state the following version of Macaulay’s theorem, whose proof we include for the convenience of the reader. Lemma 3.2. With notation as above we have Hilb(F/U) = Hilb(F/ in(U)). L Proof. As in (4) we have in(U) = ri=1 Iui ui where Iui is the ideal generated by all fi ∈ P0 such that there exists f ∈ F with in(f ) = fi ui . Thus we have F/ in(U) = L r i=1 P0 /Iui . For each j let βj be a set of homogeneous elements hij ∈ P0 which forms a K-basis of P0 /Iuj . Then β = {β1 u1 , . . . , βr ur } is a homogeneous K-basis of F/ in(U). To complete our proof we will show that β is also a K-basis of F/U. We first show thatP the elements of β in F/U are linearlyPindependent. Suppose that in P F/U, we have i,j aij hij uj = 0 with aij ∈ K. Thus rj=1 ( i aij hij )uj ∈ U. We set P hj = i aij hij , so that h1 u1 + · · · + hr ur ∈ U. If all hj = 0, then aij = 0 for all i and j, as required. Assume that hj 6= 0 for some j, and let k be the smallest integer such that P hk 6= 0. It follows that hk uk + hk+1 uk+1 + · · · ∈ U, so that hk ∈ Ik , and hence i aik hik = 0 modulo Ik . Since hik are part of a K-basis of P0 /Ik , it follows that aik = 0 for all i, and hence hk = 0, a contradiction. Now we want to show that each element in F/U can be written as a K-linear combination of elements P of β. Let f + U ∈ F/U where f ∈ F . Thus there exists ∈ P0 /Iu1 , there exists λi1 ∈ K such fi ∈ P0 such thatPf = ri=1 fi ui . Since f1 + Iu1 P that f1 + Iu1 = i λi1 (hi1 + Iu1 ), so that f1 = i λi1 hi1 + hu1 for some hu1 ∈ Iu1 . Hence r X X fi ui . f= λi1 hi1 u1 + hu1 u1 + i=2 i We set ′ f =f− X λi1 hi1 u1 = hu1 u1 + i r X i=2 10 fi ui . Since hu1 ∈ Iu1 , P there exist g2 , . . . , gr ∈ P0 such that hu1 u1 + fore, hu1 u1 = − ri=2 gi ui modulo U. Hence it follow that ′ f =− r X gi u i + r X fi ui = i=2 i=2 r X fi′ ui Pr i=2 gi ui ∈ U. There- modulo U. i=2 Here fi′ = −gi + fi for P i = 2, . . . , r. By induction on the number of summands, we may assume that ri=2 fi′ ui is a linear combination of elements of β modulo U. Since f differs from f ′ only by a linear combination of elements of β, the assertion follows.  Now we are able to prove that the Hilbert-series of the P0 -module HPn+ (R)j is a nonincreasing function in j. Theorem 3.3. Let R = P/f P be a hypersurface ring. Then Hilb(HPn+ (R)j−1 ) ≥ Hilb(HPn+ (R)j ) for all j ≤ −n. L P Proof. Let Fj = u = z1a1 . . . znan with ni=1 ai = −n − j. In u∈B−n−j P0 u where L view of (4) we have Fj / in(Uj ) = u∈B−n−j P0 /Ij,u . By Lemma 3.2 we know that Fj /Uj and Fj / in(Uj ) have the same Hilbert function. Thus Proposition 3.1 implies that for all j ≤ −n we have X M Hilb(HPn+ (R)j ) = Hilb(Fj /Uj ) = dimK ( P0 /Ij,u)i ti i = X X i = = ≤ = dimK (P0 /Ij−1,z1u )i ti u∈B−n−j X dimK (P0 /Ij−1,v )i ti i X v∈B−n−j+1 a1 >0 X X dimK (P0 /Ij−1,v )i ti i X i as desired. dimK (P0 /Ij,u )i ti u∈B−n−j X X i u∈B−n−j v∈B−n−j+1 dimK ( M P0 /Ij−1,v )i ti = Hilb(HPn+ (R)j−1 ), v∈B−n−j+1  Corollary 3.4. Let R be the hypersurface ring P/f P such that the P0 -module HPn+ (R)j has finite length for all j. Then reg HPn+ (R)j−1 ≥ reg HPn+ (R)j for all j ≤ −n. Proof. The assertion follows from the fact that reg HPn+ (R)j = deg Hilb(HPn+ (R)j ).  11 Now one could ask when P0 -module HPn + (R)j is of finite length. To answer this question we need some preparation. Let A be a Noetherian ring and M be a finitely generated A-module with presentation ϕ Am → An → M → 0. Let U be the corresponding matrix of the map ϕ and In−i (U) for i = 0, . . . , n − 1 be the ideal generated by the (n − i)-minors of matrix U. Then Fitti (M) := In−i (U) is called the i-th Fitting ideal of M. We use the convention that Fitti (M) = 0 if n−i > min{n, m}, and Fitti (M) = A if i ≥ n. In particular, we obtain Fittr (M) = 0 if r < 0, Fitt0 (M) is generated by the n-minors of U, and Fittn−1 (M) is generated by all entries of U. Note that Fitti (M) is an invariant on M, i.e. independent of the presentation. By [5, Proposition 20.7] we have Fitt0 (M) ⊆ Ann M and if M can r be generated by p √ r element, then (Ann M) ⊆ Fitt0 (M). Thus we can conclude that Fitt0 (M) = Ann M . Therefore (5) dim M = dim A/ Ann M = dim A/In (U). Now we can state the following Proposition 3.5. Let R be the hypersurface ring P/f P , and I(f ) the ideal of generated by all the coefficients of f . Then dimP0 HPn + (R)j ≤ dim P0 /I(f ). In particular, if I(f ) is m-primary where m = (x1 , . . . , xn ). Then P0 -module HPn + (R)j is of finite length for j ≤ −n. Proof. As we have already seen, HPn + (R)j has P0 -presentation ϕ P0n1 (−a) → P0n0 → HPn + (R)j → 0,

where n0 = −j−1 and n1 = −j+b−1 . In view of (5) we have dimP0 HPn + (R)j = n−1 n−1 dim P0 /In0 (Ujp ) where Ujpis the corresponding matrix of the map ϕ. By [8, Lemma 1.4] we have I(f ) ⊆ In0 (Uj ). It follows that dimP0 HPn + (R)j ≤ dim P0 /I(f ). Since I(f ) is m-primary it follows that dim P0 /I(f ) = 0. Therefore dimP0 HPn + (R)j = 0, and hence HPn + (R)j has finite length, as required.  4. The regularity of the graded components of local cohomology for a special class of hypersurfaces Ln Let A = i=0 Ai be a standard graded Artinian K-algebra, where K is a field of characteristic 0. We say that A has the weak Lefschetz property if there is a linear l form l of degree 1 such that the multiplication map Ai −→ Ai+1 has maximal rank for all i. This means the corresponding matrix has maximal rank, i.e., l is either injective or surjective. Such an element l is called a weak Lefschetz element on A. We also say that A has the strong Lefschetz property if there is a linear form l of lk degree 1 such that the multiplication map Ai −→ Ai+k has maximal rank for all i and k. Such an element l is called a strong Lefschetz element on A. Note that the set of all weak Lefschetz elements on A is a Zariski-open subset of the affine space A1 , and the same holds for the set of all strong Lefschetz elements on A. For an algebra A as above, we say that A has the strong Stanley property(SSP) if there 12 exists l ∈ A1 such that ln−2i : Ai → An−i is bijective for i = 0, 1, . . . , [n/2]. Note that the Hilbert function of standard graded K-algebra satisfying the weak Lefschetz property is unimodal. Stanley [9] and Watanabe [10] proved the following result: Let a1 , . . . , an be the integers such that ai ≥ 1 and assume as always in this section that char K = 0. Then A = K[x1 , . . . , xn ]/(xa11 , . . . , xann ) has the strong Lefschetz property. P Theorem 4.1. Let r ∈ N and fλ = ni=1 λi xi yi with λi ∈ K and n ≥ 2, and assume that char K = 0. Then there exists a Zariski open subset V ⊂ K n such that for all λ = (λ1 , · · · , λn ) ∈ V one has reg HPn + (P/fλr P )j = −n − j + r − 1. Proof. We first prove the theorem in the case that f = f(1,...,1) = fr Pn i=1 xi yi , and set R = P/f r P . ¿From the exact sequence 0 → P (−r, −r) → P → R → 0, we get an exact sequence of P0 -modules, M M fr P0 (−r)z b → (6) P0 z b → HPn+ (R)j → 0. |b|=−n−j+r |b|=−n−j Note that HPn + (R)j is generated by elements of degree 0 and the ideal generated by the coefficients of f is m-primary. By Proposition 3.5, we need only to show that (a) [HPn + (R)j ]−n−j+r−1 6= 0, and (b) [HPn + (R)j ]−n−j+r = 0. Let k = −n − j for short. For the proof of (a), we take the (k + r − 1)-th component of the exact sequence (6), and obtain the exact sequence of K-vector spaces M M fr Kxa z b → Kxa z b → [HPn+ (R)j ]k+r−1 → 0. |a|=k−1 |b|=k+r |a|=k+r−1 |b|=k
n+k+r−1
Kxa z b . Hence one has dimK Vk−1,k+r = n+k−2 k−1 k+r

n+k+r−2 n+k−1 r which is less than dimK Vk+r−1,k = k+r−1 for n ≥ 2. Thus f is not k surjective, so (a) follows. For the proof of (b), we take the (k + r)-th component of the exact sequence (6), and obtain the exact sequence of K-vector spaces M M fr Kxa z b → [HPn+ (R)j ]k+r → 0. Kxa z b → We set Vα,β := L |a|=α |b|=β |a|=k+r |b|=k |a|=k |b|=k+r Note that dimK Vk,k+r = dimK Vk+r,k . We will show that f r is an isomorphism, then we are done. We fix c ∈ Nn0 such that c = (c1 , . . . , cn ) where ci ≥ 0. We set M M c := Vα,β Kxa z b and Aci := Kxa . |a|=α |b|=β a+b=c |a|=i a≤c c We define ϕ : Vk,k+r −→ Ack by setting ϕ(xa z b ) = xa . Note that ϕ is an isomorphism L c c of K-vector spaces. Let Ac = |c| i=0 Ai . We can define an algebra structure on A . 13 For xs , xt ∈ Ac we define s t xx =  xs+t if s + t ≤ c, 0 if s + t 6≤ c. A K-basis of Ac is given by all monomials xa with a ≤ c. It follws that Ac = K[x1 , . . . , xn ]/(xc11 +1 , . . . , xncn +1 ). Now we see that the map Vk,k+r = M |c|=2k+r fr c Vk,k+r → M c Vk+r,k = Vk+r,k |c|=2k+r c c c : Vk,k+r −→ Vk+r,k is is an isomorphism if and only if restriction map f ′ := f r |Vk,k+r an isomorphism for all c with |c| = 2k + r. For each such c we have a commutative diagram f′ c c Vk,k+r −→ Vk+r,k ↓ ↓ r l Ack −→ Ack+r , lr with l = x1 + x2 + . . . + xn ∈ Ac1 and where Ack −→ Ack+r is multiplication by lr in the K-algebra Ac . Since the socle degree of Ac equals s = 2k + r, we have k + r = s − k. Therefore the multiplication map lr : Ak → As−k with r = s − 2k is an isomorphism by the strong Stanley property of the algebra Ac , see [10, Corollary 3.5] Now if we replace f by fλ , then the corresponding linear form in the above commutative diagram is the form lλ = λ1 x1 + λ2 x2 + · · · + λn xn . It is known that the property of lλ to be a weak Lefschetz element is an open condition, that is, there exists a Zariski open set V ⊂ K n such that lλ is a weak Lefschetz element. This open set is not empty since λ = (1, . . . , 1) ∈ V . Since any weak Lefschetz element satisfies (SSP), we can replace in the above proof f by fλ for each λ ∈ V , and obtain the same conclusion.  Remark 4.2. It is now the time that to show Theorem 2.4 may fail without the assumption that dimP0 M/P+ M ≤ 1. In case of Theorem 4.1 we have M = R = P/fλr P , and so M/P+ M = P0 . Therefore in that case dimP0 M/P+ M = dimP0 P0 = n ≥ 2, and in fact fn,R is not bounded. Now in the Theorem 4.1, we want to compute the Hilbert function of the P0 module HPn+ (R)j . Corollary 4.3. With the assumption of Theorem 4.1, we have (

n+i−1 −j−1 , if i ≤ r, n i
−n−j


dimK (HP+ (R)j )i = n+i−1 −j−1 n+i−r−1 −j+r−1 − , if r ≤ i ≤ −n − j + r − 1. i −n−j i−r −n−j+r 14 Proof. We set −n − j = k, for short. We take i-th component of exact sequence (6), and obtain the exact sequence of K-vector space M M fr Kxa z b → [HPn+ (R)j ]i → 0. Kxa z b → |a|=i |b|=k |a|=i−r |b|=k+r If i ≤ r, from the above exact sequence we see that    −j − 1 n+i−1 n . dimK (HP+ (R)j )i = dimK Vi,k = −n − j i Now let r ≤ i ≤ −n − j + r − 1. First one has dimK Vi−r,k+r < dimK Vi,k . We claim that f r is injective, then we are done. We see that the map M M fr c c Vi−r,k+r = Vi−r,k+r → Vi,k = Vi,k |c|=i+k |c|=i+k c : ⊂ is injective if and only if restriction map f ′ := f r |Vi−r,k+r injective for all c with |c| = i + k. For each such c we have a commutative diagram r c where f (Vi−r,k+r ) c c Vi−r,k+r −→ Vi,k is c Vi,k f′ c c Vi−r,k+r −→ Vi,k ↓ ↓ lr c Ai−r −→ Aci , with l = x1 + x2 + . . . + xn ∈ Ac1 . Since i < −n − j + r, then i < |c| − (i − r) and by the weak Lefschetz property the algebra Ac is unimodal. Therefore dimK Aci−r ≤ dimK Aci . The strong Lefschetz property implies that the map lr is injective, and hence f ′ is injective, as required.  Corollary 4.4. With the assumption of Theorem 4.1, we have r reg HPn−1 + (P/fλ P )j = −n − j + r + 1. Proof. We consider the exact sequence of P0 -modules M M fr P0 (−r)z b → (7) 0 → HPn−1 P0 z b → HPn + (R)j → 0, + (R)j → |b|=−n−j+r |b|=−n−j where R = P/fλr P . It follows that HPn−1 + (R)j is the second syzygy module of n HP + (R)j . Let ··· → t2 M j=1 P0 (−a1j ) → t1 M j=1 P0 (−a0j ) → HPn−1 + (R)j → 0 be the minimal graded free resolution of HPn−1 + (R)j . We combine two above resolutions, and obtain a graded free resolution for HPn + (R)j of the form ··· → t1 M j=1 d P0 (−a0j ) →0 M |b|=−n−j+r fr P0 (−r)z b → 15 M |b|=−n−j P0 z b → HPn + (R)j → 0. We choose a basis element h ∈ Lt1 P0 (−a0j ) of degree a0j . Thus X d0 (h) = hb z b j=1 |b|=−n−j+r where hb ∈ P0 with deg hb = a0j − r. Because the free resolution is minimal, at least one hb 6= 0, so that r < a0j and hence r − 1 ≤ a0j − 2. Thus we have reg HPn + (R)j = max{0, r − 1, aij − i − 2} = max{aij − i − 2}. i,j i,j Theorem 4.1 implies that reg HPn−1 + (R)j = max{aij − i} = −n − j + r + 1. i,j  r Corollary 4.5. With the assumption of the Theorem 4.1 the P0 -module HPn−1 + (P/fλ P )j has a linear resolution. Proof. Taking the k- th component of the exact sequence (7), we obtain the exact sequence of K-vector spaces M M r a b f Kx z Kxa z b → [HPn+ (R)j ]k → 0. 0 → [HPn−1 (R) ] → → j k + |a|=k−r |b|=−n−j+r |a|=k |b|=−n−j For k we distinguish several cases. Let k = −n − j + r + 1. One has dimK Vk−r,−n−j+r > dimK Vk,−n−j . This implies that [HPn−1 (R)j ]k 6= 0 for all k ≥ −n − j + r + 1, + since HPn−1 (R)j is torsion-free. + Let k = −n−j+r. Then dimK Vk−r,−n−j+r = dimK Vk,−n−j , so that [HPn−1 (R)j ]k = +

n+k−r−1 −j+r−1 0. Finally let k < −n − j + r. We claim that dimK Vk−r,−n−j+r = k−r −n−j+r
−j−1
is less than dimK Vk,−n−j = n+k−1 . Indeed, k −n−j dimK Vk−r,−n−j+r = r Y i=1 −j+r−i −n−j+r−i+1 −j + r − i −n − j + r − i + 1 and dimK Vk,−n−j = r Y n+k−i i=1 k−i+1 . < n+k−i for all i = 1, . . . , r if and only if k(n − 1) < (−n − j + Since k−i+1 r)(n − 1), the claim is clear. Thus the regularity of HPn−1 + (R)j is equal to the least n−1 integer k such that [HP+ (R)j ]k 6= 0. This means that P0 -module HPn−1 + (R)j has a linear resolution, and its resolution is the form · · · → P0β3 (n + j − r − 2) → P0β2 (n + j − r − 1) → P0β1 (−r) → P0β0 → HPn+ (R)j → 0.  16 
−j−1 In the above resolution we know already the Betti numbers β0 = −n−j and
−j+r−1 β1 = −n−j+r . Next we are going to compute the remaining Betti numbers and also the multiplicity of HPn+ (R)j . For this we need to prove the following extension of the formula of Herzog and Kühl [2]. Proposition 4.6. Let M be finitely generated graded Cohen-Macaulay P0 -module. Let 0 → P0βs (−ds ) → · · · → P0β1 (−d1 ) → P0β0 → M → 0, be the minimal graded free resolution of M where s = codim(M). Then Y dj βi = (−1)i+1 β0 (dj − di ) j6=i Proof. We consider the square matrix A of size s and the following s × 1 matrices of X and Y :  1 1 ··· 1  d2 · · · ds  d A =  .1 .. .. ..  .. . . . s−1 s−1 s−1 · · · ds d1 d2    ,     X =  −β1 β2 .. . (−1)s βs         and Y =   −β0 0 .. . 0    .  With similar arguments as in the proof of Lemma 1.1 in [7] one has  s X 0 for 1 ≤ k < s, i k (−1) βi di = (−1)s s!e(M) for k = s, i=1 Ps Note that i=1 (−1)i βi = β0 . Thus we can conclude that AX = Y . Now we can apply Cramer’s rule for the computation of βi . We replace the i-th column of A by Y , then we expand the determinant |A| of A along to the Y , we get βi = −β0 |A′ | / |A| where A′ is the matrix   d1 · · · di−1 di+1 · · · ds  2   d1 · · · d2i−1 d2i+1 · · · d2s  ,  . .. .. .. .. ..   .. . . . . .  s−1 s−1 d1s−1 · · · di−1 di+1 · · · dss−1 Q of size s − 1. A is a Vandermonde matrix whose determinant is 1≤j<i≤s (di − dj ). We also note that Y Y (dk − dt ), dj |A′ | = j6=i 1≤t<k≤s t6=i so the desired formula follows. Proposition 4.7. With the assumption of Proposition 4.6, we have s β0 Y di . e(M) = s! i=1 17 Proof. We consider the square matrix   β1 d1 β2 d2 · · · βs−1 ds−1 βs ds    β1 d21 β2 d22 · · · βs−1 d2s−1 βs d2s  (8) M = . .. .. .. ..   .. . . . .  s s s β1 d1 β2 d2 · · · βs−1 ds−1 βs dss of size s. We will compute the determinant |M| of M in two different ways. First we replace the last column of M by the alternating sum of all columns of M. The resulting matrix will be denoted by M ′ . It is clear that |M| = (−1)s |M ′ |. Moreover, due to [7, Lemma 1.1], the last column of M ′ is the transpose of the vector (0, . . . , 0, (−1)s se(M)). Thus if we expand M ′ with respect to the last column we get |M| = (−1)s |M ′ | = s!e(M)|N| where N is the matrix β1 d1  β2 d2 · · · βs−1 ds−1  β2 d22 · · · βs−1 d2s−1  β d2 N =  1. 1 .. .. .. ..  . . . s−1 β1 d1s−1 β2 d2s−1 · · · βs−1 ds−1 of size s − 1. Thus (9) |M| = s!e(M) s−1 Y i=1 βi s−1 Y i=1      di |V (d1 , . . . , ds−1)| where Q V (d1 , . . . , ds−1) is the Vandermonde matrix of size s − 1 whose determinant is 1≤j<i≤s−1 (di − dj ). On the other hand, directly from (8) we get (10) |M| = s Y i=1 βi s Y i=1 di |V (d1 , . . . , ds )| where V (d1 , . . . , ds ) is the Vandermonde matrix of size s whose determinant is Q  1≤j<i≤s (di − dj ). In view of (9) and (10) we get the desired formula. Now we are able to compute all Betti numbers and the multiplicity of HPn + (R)j . We recall that its resolution is the form β 0 → P0βn (j − r + 1) → P0 n−1 (j − r + 2) → · · · → P0β3 (n + j − r − 2) → P0β2 (n + j − r − 1) → P0β1 (−r) → P0β0 → HPn+ (R)j → 0,

−j−1 −j+r−1 where β0 = −n−j and β1 = −n−j+r . Corollary 4.8. With the above notation we have βi = (−1)i r(n − 1)!β0 β1 (i − 2)!(n − i)!(−n − j + r + i − 1)(n + j − i + 1) 18 for all i ≥ 2, and e(HPn+ (R)j ) = r(−j + r − 1)!β0 . n!(−n − j + r)! Proof. The assertion follows from Proposition 4.7 and Proposition 4.6.  5. linear bounds for the regularity of the graded components of local cohomology for hypersurface In this section for a bihomogenous polynomial f ∈ P we want to give a linear bound for the function fi,R (j) = reg HPi + (R)j where R = P/f P . First we prove the following P Proposition 5.1. Let R be the hypersurface ring P/f P where f = ni=1 fi yi with fi ∈ P0 . Suppose that deg fi = d and that I(f ) is the m-primary. Then there exists an integer q such that for j ≪ 0 we have (a) reg HPn+ (R)j ≤ (−n − j + 1)d + q, and (b) reg HPn−1 (R)j ≤ (−n − j + 1)d + q + 2. + f Proof. (a) From the exact sequence 0 → P (−d, −1) → P → R → 0, we get exact sequence P0 -modules M (11) |b|=−n−j+1 f P0 (−d)z b → M |b|=−n−j P0 z b → HPn+ (R)j → 0. We first assume that fi = xi . Theorem 4.1 implies that reg HPn+ (R)j = −n − j. We set k = −n − j. Thus we can get the surjective map of K-vector spaces M M Kxa z b . Kxa z b −→ |a|=k+1 |b|=k |a|=k |b|=k+1 Replacing xi by fi , we therefore get a surjective map M (I(f )k )dk z b = |b|=k+1 f M Kf1a1 . . . fnan z b −→ M Kf1a1 . . . fnan z b = |a|=k |b|=k+1 M (I(f )k+1)d(k+1) z b . |b|=k |a|=k+1 |b|=k Since I(f ) is m-primary by [3, Theorem 2.4] there exists an integer q such that reg(P0 /I(f )k+1) = (k + 1)d + q We set l = (k + 1)d + q. Then for l ≫ 0 we have for k ≫ 0. (P0 )l+1 = (I(f )k+1)l+1 . We take the (l + 1)-th component of the exact sequence (11) and consider the following diagram 19 L |b|=k+1 (P0 )l−d+1 z L x   |b|=k+1 (I(f ) k b −−−→ )l−d+1 z b −−−→ L |b|=k (P0 )l+1 z L |b|=k (I(f ) k+1 b −−−→ [HPn+ (R)j ]l+1 −−−→ 0 )l+1 z b −−−→ 0, in which left-hand vertical homomorphism is inclusion. Thus we conclude that [HPn+ (R)j ]l+1 = 0, so that reg HPn+ (R)j ≤ l = (k + 1)d + q, as required. For the proof (b), we notice that the exact sequence of P0 -modules of (11) breaks into two short exact sequence of P0 -modules M 0 → Kj → P0 z b → HPn+ (R)j → 0, |b|=k 0 → HPn−1 (R)j → + M |b|=k+1 P0 (−d)z b → Kj → 0, where Kj = Im f . We see from the first of these sequences that reg Kj ≤ reg HPn+ (R)j + 1. The second short exact sequence, together with part (a) this theorem and the fact that d ≤ reg Kj implies that reg HPn−1 (R)j ≤ max{d, reg Kj + 1} = reg Kj + 1 ≤ (−n − j + 1)d + q + 2, + as desired.  Proposition 5.2. Let Nnd = {β ∈ Nn : |β| = d}, P0 = K[{xβ }β∈Nnd ] and P = P P0 [y1 , . . . , yn ]. Let R = P/f P where f = |β|=d xβ y β . Then reg HPn+ (R)j ≤ (−n − j + 1)d − 1 Proof. We set P+ = (y1 , . . . , yn ) and P0 = K[x1 , . . . , xm ] where m = f n+d−1 d
, as useual . ¿From the exact sequence 0 → P (−1, −d) → P → R → 0, we get the exact sequence of P0 -modules M M f P0 (−1)(y b )∗ → P0 (y b )∗ → HPn+ (R)j → 0, |b|=−n−j+d |b|=−n−j whose i-th graded component is M M f Kxa (y b)∗ → (12) Kxa (y b)∗ → HPn+ (R)(i,j) → 0. |a|=i−1 |b|=−n−j+d |a|=i |b|=−n−j Here (y b)∗ = z b in the of Section 1. Now we exchange the role of x and y: P notation β We may write f = |β|=d y xβ and set Q+ = (x1 , . . . , xm ) and Q0 = K[y1 , . . . , yn ]. f ¿From the exact sequence 0 → P (−d, −1) → P → R → 0, we get the exact sequence of P0 -modules M M f Q0 (−d)(xb )∗ → Q0 (xb )∗ → HQm+ (R)t → 0. |b|=−m−t+1 |b|=−m−t 20 whose s-th graded component is M f Ky a (xb )∗ → |a|=s−d |b|=−m−t+1 M |a|=s |b|=−m−t Ky a (xb )∗ → HQm+ (R)(s,t) → 0. Applying the functor HomK (−, K) to the above exact sequence and due to the exact sequence (12)we have M f 0 → HQm+ (R)∗(s,t) → K(y a )∗ xb → |a|=s |b|=−m−t M |a|=s−d |b|=−m−t+1 Therefore K(y a )∗ xb → HPn+ (R)(−m−t+1,−n−s+d) → 0. HQm+ (R)∗(s,t) ∼ (R)(−m−t+1,−n−s+d) . = HPn−1 + Thus we have M 0 → (HPn−1 (R)−n−s+d )−m−t+1 → + M |a|=s |b|=−m−t K(y a )∗ xb f K(y a )∗ xb → → (HPn+ (R)−n−s+d)−m−t+1 → 0. |a|=s−d |b|=−m−t+1 We set j = −n − s + d. Proposition 5.1 implies that reg HPn+ (R)j ≤ (−n − j + 1)d + q for some q. Since I(f ) = (y1 , . . . , yn )d , thus reg(P0 /I(f )k+1) = (k + 1)d − 1. Hence in Proposition 5.1 we have q = −1.  Now the main result of this section is the following Theorem 5.3. Let P = K[x1 , . . . , xm , y1 , . . . , yn ], and f ∈ P be a bihomogenous polynomial such that I(f ) is m-primary. Let R = P/f P . Then the regularity of HPn+ (R)j is linearly bounded. P Proof. We may write f = |β|=d fβ y β and let deg fβ = c. ¿From the exact sequence f 0 → P (−c, −d) → P → R → 0, we get the exact sequence of P0 -modules M M f P0 (−c)z b → P0 z b → HPn+ (R)j → 0, |b|=−n−j+d |b|=−n−j We first assume that fβ = xβ . Proposition 5.2 implies that reg HPn+ (R)j ≤ (−n − j + 1)d − 1. We set k = (−n − j + 1)d. Thus we get the surjective map of K-vector spaces M M Kxa z b . Kxa z b −→ |a|=k |b|=−n−j |a|=k−1 |b|=−n−j+d We proceed as in the proof of Proposition 5.1, and we get [HPn+ (R)j ]kd+q′ +1 = 0 for some q ′ . Therefore reg HPn+ (R)j ≤ (−n − j + 1)d2 + q ′ .  21 Corollary 5.4. With the assumption of Theorem 5.3, we have 2 ′ reg HPn−1 + (R)j ≤ (−n − j + 1)d + q + 2. Proof. For the proof one use the same argument as in the proof of Proposition 5.1(b).  References [1] M. Brodmann and R.Y. Sharp, ”Local cohomology”: an algebraic introduction with geometric applications, Cambbridge Studies in Advanced Mathematics 60, Cambridge University Press, 1998. [2] W. Bruns and J. Herzog, ”Cohen-Macaulay rings” (Revised edition), Cambridge Studies in Advanced Mathematics 39, Cambridge University Press, 1998. [3] S. D. Cutkosky, J. Herzog, N. V. Trung, Asymptotic behaviour of the Castelnovo-Mumford regularity. Compositio Math. 118(1999), no. 3, 243-261. [4] M. Chardin, some results and questions on Castelnuovo-Mumford regularity. Preprint [5] D. Eisenbud, Commutative Algebra with a view to Algebraic geometry, Springer Verlag, 1995. [6] L. Gruson, R. Lazarsfeld, C. Peskine, On a theorem of Castelnuovo, and the equations defining space curves. Inventions mathematicae. 72(1983), 491-506 [7] J. Herzog and H. Srinivasan, Bounds for multiplicities. Trans. Am. Math. Soc. 350(1998), no. 7, 2879-2902. [8] M. Katzman and R.Y. Sharp, Some properties of top local cohomology modules, Journal of Algebra 259(2003), 599–612. [9] R. Stanley, Weyl groups, the hard Lefschetz theorem, and the Sperner property, SIAM J. Algebraic Discrete Methods 1 (1980) 168-184. [10] J. Watanabe, The Dilworth number of Artinian rings and finite posets with rank function, in: Commutative Algebra and Combinatorics, in: Advanced Studies in Pure Mathematics. Vol. 11, Kinokuniya, North-Holland, Amesterdam, 1987, pp. 303-312. Ahad Rahimi, Fachbereich Mathematik und Informatik, Universität DuisburgEssen, Campus Essen, 45117 Essen, Germany E-mail address: ahad.rahimi@uni-essen.de 22