MONOMIAL IDEALS, ALMOST COMPLETE INTERSECTIONS AND THE WEAK LEFSCHETZ PROPERTY JUAN C. MIGLIORE∗, ROSA M. MIRO´-ROIG∗∗, UWE NAGEL+ Abstract. Many algebras are expected to have the Weak Lefschetz property though 9 this is often very difficult to establish. We illustrate the subtlety of the problem by 0 studying monomialandsome closelyrelatedideals. Ourresults exemplify the intriguing 0 dependence of the property on the characteristic of the ground field, and on arithmetic 2 properties of the exponent vectors of the monomials. n a J 7 2 1. Introduction ] Let A be a standard graded Artinian algebra over the field K. Then A is said to have C the Weak Lefschetz property (WLP) if there is a linear form L ∈ (A) such that, for all A 1 integers j, the multiplication map . h t ×L : (A) → (A) a j−1 j m has maximal rank, i.e. it is injective or surjective. In this case, the linear form L is called [ a Lefschetz element of A. (We will often abuse notation and say that the corresponding 2 ideal has the WLP.) The Lefschetz elements of A form a Zariski open, possibly empty, v 3 subset of (A)1. Part of the great interest in the WLP stems from the fact that its presence 2 puts severe constraints on the possible Hilbert functions (see [6]), which can appear in 0 various disguises (see, e.g., [12]). Though many algebras are expected to have the WLP, 1 . establishing this property is often rather difficult. For example, it is open whether every 1 1 complete intersection of height four over a field of characteristic zero has the WLP. (This 8 is true if the height is at most 3 by [6].) 0 In some sense, this note presents a case study of the WLP for monomial ideals and : v almostcomplete intersections. OurresultsillustratehowsubtletheWLPis. Inparticular, i X we investigate its dependence on the characteristic of the ground field K. The following r example (Example 7.7) illustrates the surprising effect that the characteristic can have a on the WLP. Consider the ideal I = (x10,y10,z10,x3y3z3) ⊂ R = K[x,y,z]. Our methods show that R/I fails to have the WLP in characteristics 2, 3 and 11, but possesses it in all other characteristics. One starting point of this paper has been Example 3.1 in [4], where Brenner and Kaid show that, over an algebraically closed field of characteristic zero, any ideal of the form (x3,y3,z3,f(x,y,z)), with degf = 3, fails to have the WLP if and only if ∗ PartoftheworkforthispaperwasdonewhilethefirstauthorwassponsoredbytheNationalSecurity Agency under Grant Number H98230-07-1-0036. ∗∗ PartoftheworkforthispaperwasdonewhilethesecondauthorwaspartiallysupportedbyMTM2007- 61104. + PartoftheworkforthispaperwasdonewhilethethirdauthorwassponsoredbytheNationalSecurity Agency under Grant Number H98230-07-1-0065. TheauthorsthankFabrizioZanelloforusefulandenjoyableconversationsrelatedtosomeofthismaterial. They also thank David Cook II for useful comments. 1 2 J. MIGLIORE, R.MIRO´-ROIG,U.NAGEL f ∈ (x3,y3,z3,xyz). In particular, the latter ideal is the only such monomial ideal that fails to have the WLP. This paper continues the study of this question. The example of Brenner and Kaid satisfies several interesting properties. In this paper we isolate several of these properties andexamine the question ofwhether or not the WLP holds for such algebras, and we see to what extent we can generalize these properties and stillgetmeaningfulresults. Someofourresults holdover afieldofarbitrarycharacteristic, while others show different ways in which the characteristic plays a central role in the WLP question. (Almost none are characteristic zero results.) Most of our results concern monomial ideals, although in Section 5 and Section 8 we show that even minor deviations from this property can have drastic effects on the WLP. Most of our results deal with almost complete intersections in three or more variables, but we also study ideals with more generators (generalizing that of Brenner and Kaid in a different way). More specifically, we begin in Section 2 with some simplifying tools for studying the WLP. These are applied throughout the paper. We also recall the construction of basic double linkage. In Section 3 we consider the class of monomial ideals in K[x ,...,x ] of the form 1 r (xk,xk,...,xk)+(all squarefree monomials of degree d). 1 2 r Note that the example of Brenner and Kaid is of this form. Our main result in this section (Theorem 3.3) says that when d = 2 we always have the WLP, but if d = 3 and k ≥ 2 then we have two cases: if K has characteristic 2 then we never have the WLP, but if the characteristic is not 2 then we have the WLP if and only if k is even. InSection 4, we consider almost complete intersections of theform (xr,...,xr,x ···x ) 1 r 1 r with r ≥ 3 (note that the result of Brenner and Kaid dealt with the case r = 3 in characteristic zero). Our main result for these algebras is that they always fail to have the WLP, regardless of the characteristic. The proof is surprisingly difficult. In Section 5 we explicitly illustrate the fact that even a minuscule change in the ideal can affect the WLP. Specifically, we consider the ideals of the form (xr,...,xr,x ···x ·(x +x )). 1 r 1 r−1 1 r Weshow thatthishasthesameHilbert functionasthecorresponding idealintheprevious section, but the WLP behavior is very different. For example, the two ideals (x4,...,x4,x x x x ) and (x4,...,x4,x x x (x +x )) 1 4 1 2 3 4 1 4 1 2 3 1 4 have the same Hilbert function, but the former never has the WLP while the latter has the WLP if and only if the characteristic of K is not two or five. In Section 6 we turn to monomial almost complete intersections in three variables, generalizing the Brenner-Kaid example in a different direction. To facilitate this study, we assume that the algebra is also level (as is the case for Brenner and Kaid’s example). We give a number of results in this section, which depend on the exponent vectors of the monomials. We end with a conjectured classification of the level Artinian monomial ideals in three variables that fail to have the WLP (Conjecture 6.8). The work in Sections 6 and 7 proves most of this conjecture. We end the paper in Section 8 with some suggestive computations and natural questions coming from our work. Contents 1. Introduction 1 2. Tools for studying the WLP 3 3. A class of monomial ideals 5 MONOMIAL IDEALS AND THE WEAK LEFSCHETZ PROPERTY 3 4. Monomial almost complete intersections in any codimension 8 5. An almost monomial almost complete intersection 14 6. Monomial almost complete intersections in three variables 17 7. A proof of half of Conjecture 6.8 20 8. Final Comments 25 References 26 2. Tools for studying the WLP In this section we establish various general results that help to study the WLP and that are used throughout the remainder of this paper. Throughout this paper we set R = K[x ,...,x ], where K is a field. Sometimes we will have specific values of r (usually 1 r 3) and sometimes we will have further restrictions on the field K. Our first results singles out the crucial maps to be studied if we consider the WLP of a level algebra. Recall that an Artinian algebra is called level if its socle is concentrated in one degree. Proposition 2.1. Let R/I be an Artinian standard graded algebra and let L be a general linear form. Consider the homomorphisms φ : (R/I) → (R/I) defined by multiplica- d d d+1 tion by L, for d ≥ 0. (a) If φ is surjective for some d then φ is surjective for all d ≥ d . d0 0 d 0 (b) If R/I is level and φ is injective for some d ≥ 0 then φ is injective for all d0 0 d d ≤ d . 0 (c) In particular, if R/I is level and dim(R/I) = dim(R/I) for some d then d0 d0+1 0 R/I has the WLP if and only if φ is injective (and hence is an isomorphism). d0 Proof. Consider the exact sequence [I : L] ×L 0 → → R/I −→ (R/I)(1) → (R/(I,L))(1) → 0 I where×Lindegreedisjustφ . Thisshows thatthecokernel ofφ isjust(R/(I,L)) for d d d+1 any d. If φ is surjective, then (R/(I,L)) = 0, and the same necessarily holds for all d0 d0+1 subsequent twists since R/I is a standard graded algebra. Then (a) follows immediately. For (b), recall that the K-dualof the finite length module R/I is a shift of the canonical module of R/I, which we will denote simply by M. Since R/I is level, M is generated in the first degree. But now if we consider the graded homomorphism of M to itself induced by multiplication by L, a similar analysis (recalling that M is generated in the first degree) gives that once this multiplication is surjective in some degree, it is surjective thereafter. The result on R/I follows by duality. (cid:3) Part (c) follows immediately from (a) and (b). If the field is infinite and the K-algebra satisfies the WLP for some linear form, then it does for a general linear form. For monomial ideals there is no need to consider a general linear form. Proposition 2.2. Let I ⊂ R be an Artinian monomial ideal and assume that the field K is infinite. Then R/I has the WLP if and only if x +···+x is a Lefschetz element for 1 r R/I. Proof. Set A = R/I and let L = a x +···+a x be a general linear form in R. Thus, 1 1 r r we may assume that each coefficient a is not zero and, in particular, a = 1. Let i r 4 J. MIGLIORE, R.MIRO´-ROIG,U.NAGEL J ⊂ S := K[x ,...,x ] be the ideal that is generated by elements that are obtained 1 r−1 from the minimal generators of I after substituting a x + ···+ a x for x . Then 1 1 r−1 r−1 r ∼ A/LA = S/J. Each minimal generator of J is of the form xj1···xjr−1(a x +···+a x )jr. Replac- 1 r−1 1 1 r−1 r−1 ing it by (a x )j1···(a x )jr−1(−a x +···−a x )jr does not change the ideal J 1 1 r−1 r−1 1 1 r−1 r−1 because a ···a 6= 0. Using the isomorphism K[y ,...,y ] → S, y 7→ a x , we see 1 r−1 1 r−1 i i i that A/LA and A/(x +···+x )A have the same Hilbert function. Since we can decide 1 r whether L is a Lefschetz element for A by solely looking at the Hilbert function of A/LA, (cid:3) the claim follows. If A is an Artinian K-algebra with the WLP and E is an extension field of K, then also A⊗ E has the WLP. However, the converse is not clear. We pose this as a problem. K Problem 2.3. Is it true that A has the WLP if and only if A⊗ E has the WLP? K Proposition 2.2 shows that the answer is affirmative in the case of monomial ideals. Corollary 2.4. Let E be an extension field of the infinite field K. If I ⊂ R is an Artinian monomial ideal, then R/I has the WLP if and only if (R/I)⊗ E does. K The following result applies if we can hope that the multiplication by a linear form is surjective. Proposition 2.5. Let I ⊂ R = K[x ,...,x ], where K is a field and A = R/I is 1 r Artinian. Let d be any degree such that h (d−1) ≥ h (d) > 0. Let L be a linear form, A A ¯ ¯ ¯ ¯ ¯ ¯ let R = R/(L) and let I be the image of I in R. Denote by A the quotient R/I. Consider the minimal free R¯-resolution of A¯: pr−1 p1 0 → R¯(−b ) → ··· → R¯(−a ) → R¯ → A¯ → 0 i j Mi=1 Mj=1 where a ≤ ··· ≤ a and b ≤ ··· ≤ b . Then the following are equivalent: 1 p1 1 pr−1 (a) the multiplication by L from A to A fails to be surjective; d−1 d (b) A¯ 6= 0; d (c) b ≥ d+r −1; pr−1 ¯ (d) Let G ,...,G be a regular sequence in I of degrees c ,...,c respectively that 1 r−1 1 r−1 extends to a minimal generating set for I¯. Then there exists a form F ∈ R¯ of degree ≤ c +···+c −(d+r −1), non-zero modulo (G ,...,G ), such that 1 r−1 1 r−1 F ·I¯⊂ (G ,...,G ). 1 r−1 Proof. From the exact sequence ×L ··· → A −→ A → (R/(I,L)) → 0 d−1 d d it follows that the multiplication fails to be surjective if and only if A¯ = (R/(I,L)) 6= 0. d d The latter holds if and only if d ≤ socle degree of A¯ = b −(r −1), pr−1 from which the equivalence of (a), (b) and (c) follows. To show the equivalence of (c) and (d) we invoke liaison theory. Let J = (G ,...,G ) : I¯. 1 r−1 MONOMIAL IDEALS AND THE WEAK LEFSCHETZ PROPERTY 5 A free resolution for J can be obtained from that of I¯and (G,G′) by a standard mapping cone argument (see for instance [8]), as follows. We have the following commutative diagram (where the second one is the Koszul resolution for (G ,...,G )): 1 r−1 0 → pr−1R¯(−b ) → ··· → p1 R¯(−a ) → I¯ → 0 i=1 i j=1 j L ↑ L ↑ ↑ 0 → R¯(−c −···−c ) → ··· → r−1R¯(−c ) → (G,G′) → 0 1 r−1 k=1 k where the rightmost vertical arrow is an inclusiLon. This yields a free resolution for J (after splitting r−1 R¯(−c ) and re-numbering the a , and setting c := c +···+c ): k=1 k j 1 r−1 L p1−(r−1) pr−1 r−1 0 → R¯(a −c) → ··· → R¯(b −c)⊕ R¯(−c ) → J → 0. j i k Mj=1 Mi=1 Mk=1 Clearly b ≥ d + r − 1 if and only if J has a minimal generator, F, of degree ≤ pr−1 c−(d+r−1). The result then follows from the definition of J as an ideal quotient. (cid:3) We conclude this section by recalling a concept from liaison theory, which we do not state in the greatest generality. Let J ⊂ I ⊂ R = K[x ,...,x ]behomogeneous idealssuch thatcodimJ = codimI−1. 1 r Let ℓ ∈ R be a linear form such that J : ℓ = J. Then the ideal I′ := ℓ·I +J is called a basic double link of I. The name stems from the fact that I′ can be Gorenstein linked to I in two steps if I is unmixed and R/J is Cohen-Macaulay and generically Gorenstein ([7], Proposition 5.10). However, here we only need the relation among the Hilbert functions. Lemma 2.6. For each integer j, dim (R/I′) = dim (R/I) +dim (R/J) −dim (R/J) . K j K j−1 K j K j−1 Proof. This follows from the exact sequence (see [7], Lemma 4.8) 0 → J(−1) → J ⊕I(−1) → I′ → 0. (cid:3) 3. A class of monomial ideals We now begin our study of a certain class of Artinian monomial ideals. Let I be r,k,d the monomial ideal defined by (3.1) (xk,xk,...,xk)+(all squarefree monomials of degree d). 1 2 r Our first observation follows immediately be determining the socle of R/I . It shows r,k,d that we may apply Proposition 2.1. Proposition 3.1. The inverse system for I is generated by the module generated by r,k,d all monomials of the form xk−1···xk−1. i1 id−1 Corollary 3.2. The algebra R/I is level of socle degree (k −1)(d−1) and socle type r,k,d r . d−1 (cid:0) (cid:1) Concerning the WLP we have: Theorem 3.3. Consider the ring R/I . r,k,d (a) If d = 2, then it has the WLP. (b) Let d = 3 and k ≥ 2. Then: (i) If K has characteristic two, then R/I does not have the WLP. r,k,d 6 J. MIGLIORE, R.MIRO´-ROIG,U.NAGEL (ii) If the characteristic of K is not two, then R/I has the WLP if and only r,k,d if k is even. Proof. For simplicity, write I = I and A = R/I . r,k,d r,k,d Claim (a) follows easily from the observation that A has socle degree k−1 and that up to degree k−1 the ideal I is radical, so multiplication by a general linear form is injective in degree ≤ k −1. To show claim (b) we first describe bases of (A) and (A) , respectively. We choose k−1 k the residue classes of the elements in the following two sets. B = {xjxk−1−j | 1 ≤ i < m ≤ r, 1 ≤ j ≤ k −2}∪{xk−1 | 1 ≤ i ≤ r} k−1 i m i B = {xjxk−j | 1 ≤ i < m ≤ r, 1 ≤ j ≤ k −1}. k i m Counting we get r r (3.2) h (k −1) = (k −2) +r ≤ (k −1) , A (cid:18)2(cid:19) (cid:18)2(cid:19) where the inequality follows from r ≥ d = 3. Now we assume that k is odd. In this case we claim that A does not have the WLP. Because of Inequality (3.2), this follows once we have shown that, for each linear form ×L L ∈ R, the multiplication map φ : (A) −→ (A) is not injective. k k−1 k To show the latter assertion we exhibit a non-trivial element in its kernel. Write L = a x +...+a x for some a ,...,a ∈ K. We define the polynomial f ∈ R as 1 1 r r 1 r f = (−1)max{ji,jm}(a x )ji(a x )jm. i i m m X ji+jm=k−1 Note that f is not in I. We claim that L·f is in I. Indeed, since all monomials involving three distinct variables are in I, a typical monomial in L·f mod I is of the form (a x )ji(a x )k−ji. i i m m It arises in exactly two ways in Lf, namely as (a x )·(a x )ji−1(a x )k−ji and as (a x )· i i i i m m m m (a x )ji(a x )k−1−ji. Using that k−1 is even, it is easy to see that these two monomials i i m m occurinf withdifferentsigns. Itfollowsthattheabovemultiplicationmapisnotinjective. If k is even, but charK = 2, then the same analysis again shows that φ is not injective. k Hence, for the remainder of the proof we may assume that the characteristic of K is not two. Assume k is even. Then we claim that L = x + ···+ x is a Lefschetz element. To 1 r ×L this end we first show that the multiplication map φ : (A) −→ (A) is injective. k k−1 k Let f be any element in the vector space generated by B . Pick three of the variables k−1 x ,...,x and call them x,y,z. Below we explicitly list all the terms in f that involve 1 r only the variables x,y,z: f = a xk−1 +a xk−2y +···+a xyk−2+ 0 1 k−2 b xk−2z +···+b xzk−2 +b zk−1 + 1 k−2 k−1 c yk−1+c yk−2z +···+c yzk−2 0 1 k−2 +··· . As above, we see that each monomial in L·f arises from exactly two of the monomials in f. Hence the condition L · f ∈ I leads to the following three systems of equations. MONOMIAL IDEALS AND THE WEAK LEFSCHETZ PROPERTY 7 Focussing only on the variables x,y we get: a +a = 0 0 1 a +a = 0 1 2 . . . a +a = 0 k−3 k−2 a +c = 0. k−2 0 It follows that a = (−1)ia and i 0 (3.3) c = (−1)k−2a = a 0 0 0 because k is even. Considering the variables x,z we obtain: a +b = 0 0 1 b +b = 0 1 2 . . . b +b = 0, k−2 k−1 hence (3.4) b = (−1)ia . i 0 Finally, using the variables y,z we get: c +c = 0 0 1 . . . c +c = 0 k−1 k−2 c +b = 0. k−2 k−1 Combining this, it follows that −a = b = −c = a . 0 k−1 0 0 Since we assumed that the characteristic of K is not two, we conclude that the three linear systems above have only the trivial solution. Since the variables x,y,z were chosen arbitrarily, we see that the map φ is injective, as claimed. k ×L AccordingtoLemma2.1itremainstoshowthatthemultiplicationmapφ : (A) −→ k+1 k (A) is surjective. Note that the residue classes of the elements of the form xjxk+1−j k+1 i m with 2 ≤ j ≤ k − 1, 1 ≤ i < m ≤ r form a basis of (A) . Setting for simplicity k+1 x := x ,y = x it is enough to show that, for each j = 2,...,k −1, the residue class of i m xjyk+1−j is in the image of φ . k+1 We induct on j ≥ 2. If j = 2, then we get modulo I that L · xyk−1 ≡ x2yk−1, thus x2yk−1 ∈ imφ , as claimed. Let 3 ≤ j ≤ k −1, then, modulo I, we get L·xj−1yk−j ≡ k+1 xjyk−j + xj−1yk−j+1. Since by induction xj−1yk−j+1 ∈ imφ , we also obtain xjyk−j ∈ k+1 imφ . This completes the proof. (cid:3) k+1 The above result and our computer experiments suggest that the larger d becomes, the rarer it is that R/I has the WLP. Based on computer experiments we expect the r,k,d following to be true. Conjecture 3.4. Consider the algebra R/I . Then r,k,d (a) If d = 4, then it has the WLP if and only if kmod 4 is 2 or 3. (b) If d = 5, then the WLP fails. 8 J. MIGLIORE, R.MIRO´-ROIG,U.NAGEL (c) If d = 6, then the WLP fails. We summarize our results in case k = d = 3. Example 3.5. Consider the ideal I = (x3,x3,...,x3,(all squarefree monomials of degree 3)). r,3,3 1 2 r Then the corresponding inverse system is (x2x2, x2x2, ..., x2 x2). Furthermore, the 1 2 1 3 r−1 r Hilbert function of R/I is r,3,3 r +1 r 1 r r(r −1) 0 (cid:18) 2 (cid:19) (cid:18)2(cid:19) and R/I fails to have the WLP because the map from degree 2 to degree 3 by a general r,3,3 linear form is not injective. Remark 3.6. By truncating, we get a compressed level algebra with socle degree 3 that fails to have the WLP. We expect that there are compressed level algebras with larger socle degree that fail to have the WLP. However, we do not know such an example. 4. Monomial almost complete intersections in any codimension In the paper [9] the first and second authors asked the following question (Question 4.2, page 95): For any integer n ≥ 3, find the minimum number A(n) (if it exists) such that every Artinian ideal I ⊂ K[x ,...,x ] with number of generators µ(I) ≤ A(n) has the 1 n WLP. In Example 7.10 below, we show that A(n) does not exist in positive characteristic. In any case, in [4] it was shown for n = 3 and characteristic zero that A(3) = 3 (also using a result of [6]), as noted in the introduction. A consequence of the main result of this section, below, is that in any number of variables and any characteristic there is an almost complete intersection that fails to have the WLP. Hence the main open question that remains is whether, in characteristic zero, all complete intersections have the WLP (as was shown for n = 3 in [6]), i.e. whether A(n) = n in characteristic zero. We begin by considering ideals of the form (4.1) I = (xk,...,xk,x ...x ) ⊂ K[x ,...,x ]. r,k 1 r 1 r 1 r Notice that this is a special case of the class of ideals described in Section 3. It is not too difficult to determine the graded Betti numbers. Proposition 4.1. The minimal free resolution of I has the form r,k ( r ) (r) R(−(r−1)k)) r−1 R(−3k) 3 0→ R(−r+(r−1)(1−k))(r−r1) → ⊕ →···→ ⊕ → ( r ) (r) R(−r+(r−2)(1−k)) r−2 R(−r+2(1−k)) 2 R(−2k)(r2) R(−k)r ⊕ → ⊕ →Ir,k →0. R(−r+(1−k))r R(−r) Proof. Since I is an almost complete intersection, we can link it using the complete r,k intersection a = (xk,...,xk) to an Artinian Gorenstein ideal, J. However, since both I 1 r r,k and a are monomial, so is J. But it was first shown by Beintema [2] that any monomial Artinian Gorenstein ideal is a complete intersection. Hence we get by direct computation that (xk,...,xk) : x x ···x = (xk−1,...,xk−1). Then use the mapping cone and observe 1 r 1 2 r 1 r (cid:3) that there is no splitting. MONOMIAL IDEALS AND THE WEAK LEFSCHETZ PROPERTY 9 Before we come to the main result of this section, we prove a preliminary result about the Hilbert function of complete intersections that will allow us to apply Proposition 2.5. Lemma 4.2. Let R = K[x ,...,x ] with s ≥ 2, and let 1 s I = (xs,...,xs) and J = (xs+1,xs+1,xs,...,xs). s 1 s s 1 2 3 s Note that the midpoint of the Hilbert function of R/I is s and that of R/J is s +1. s 2 s 2 Then (cid:0) (cid:1) (cid:0) (cid:1) h s −h s −1 ≤ h s +1 −h s +2 . R/Is 2 R/Is 2 R/Js 2 R/Js 2 (cid:0)(cid:0) (cid:1)(cid:1) (cid:0)(cid:0) (cid:1) (cid:1) (cid:0)(cid:0) (cid:1) (cid:1) (cid:0)(cid:0) (cid:1) (cid:1) Proof. The lemma is trivial to verify when s = 2 or s = 3, so we assume s ≥ 4 for this proof. Observe that both quantities are positive, but one involves a difference to the left of the midpoint of the Hilbert function, while the other involves a difference to the right. We will use this formulation, although there exists others thanks to the symmetry of the Hilbert function of an Artinian complete intersection. Our approach will be via basic double linkage. We will use the formula in Lemma 2.6 without comment. In fact, J is obtained from I by a sequence of two basic double links: s s I x ·I +(xs,...,xs) := G = (xs+1,xs,...,xs) s 1 s 2 s 1 2 s x ·G+(xs+1,xs,...,xs) = J . 2 1 3 s s Note that G is a complete intersection of codimension s and that the ideals C := 1 (xs,...,xs) and C := (xs+1,xs,...,xs) are complete intersections of codimension s−1. 2 s 2 1 3 s The midpoints of the h-vectors of R/C and R/C are (s−1)2 and (s−1)2+1 respectively. 1 2 2 2 We now compute Hilbert functions (and notice the shift, and that the lines for R/C and 1 R/C are the first difference of those Hilbert functions, i.e. the h-vectors): 2 R/I 1 s ... h s −1 h s ... s R/Is 2 R/Is 2 R/C 1 s−1 ... ... ∆h(cid:0)(cid:0) (cid:1)s (cid:1) ∆h (cid:0)s(cid:0) +(cid:1)(cid:1)1 ... 1 C1 2 C1 2 R/G 1 s ... ... A(cid:0)(cid:0) (cid:1)(cid:1) (cid:0)B(cid:0) (cid:1) (cid:1) ... where h s = A = h s −1 +∆h s , R/G 2 R/Is 2 R/C1 2 (4.2) (cid:0)(cid:0) (cid:1)(cid:1) (cid:0)(cid:0) (cid:1) (cid:1) (cid:0)(cid:0) (cid:1)(cid:1) h s +1 = B = h s +∆h s +1 , R/G 2 R/Is 2 R/C1 2 (cid:0)(cid:0) (cid:1) (cid:1) (cid:0)(cid:0) (cid:1)(cid:1) (cid:0)(cid:0) (cid:1) (cid:1) and R/G 1 s ... A B ... R/C 1 s−1 ... ... ∆h s +1 ∆h s +2 ... 2 C2 2 C2 2 R/J 1 s ... ... (cid:0)C(cid:0) (cid:1) (cid:1) (cid:0)D(cid:0) (cid:1) (cid:1) ... s where h s +1 = C = h s −1 +∆h s +∆h s +1 , R/Js 2 R/Is 2 R/C1 2 R/C2 2 (cid:0)(cid:0) (cid:1) (cid:1) (cid:0)(cid:0) (cid:1) (cid:1) (cid:0)(cid:0) (cid:1)(cid:1) (cid:0)(cid:0) (cid:1) (cid:1) h s +2 = D = h s +∆h s +1 +∆h s +2 . R/Js 2 R/Is 2 R/C1 2 R/C2 2 (cid:0)(cid:0) (cid:1) (cid:1) (cid:0)(cid:0) (cid:1)(cid:1) (cid:0)(cid:0) (cid:1) (cid:1) (cid:0)(cid:0) (cid:1) (cid:1) Now observe that the complete intersection G has odd socle degree s(s − 1) + 1; hence A = B. Then it follows from (4.2) that (4.3) h s −h s −1 = ∆h s −∆h s +1 . R/Is 2 R/Is 2 R/C1 2 R/C1 2 (cid:0)(cid:0) (cid:1)(cid:1) (cid:0)(cid:0) (cid:1) (cid:1) (cid:0)(cid:0) (cid:1)(cid:1) (cid:0)(cid:0) (cid:1) (cid:1) 10 J. MIGLIORE, R.MIRO´-ROIG,U.NAGEL Thus we obtain h ( s +1)−h ( s +2) = h ( s −1)−h ( s ) R/Js 2 R/Js 2 R/Is 2 R/Is 2 (cid:0) (cid:1) (cid:0) (cid:1) (cid:0) (cid:1) (cid:0) (cid:1) +∆h ( s )−∆h ( s +1) R/C1 2 R/C1 2 (4.4) (cid:0) (cid:1) (cid:0) (cid:1) +∆h ( s +1)−∆h ( s +2) R/C2 2 R/C2 2 (cid:0) (cid:1) (cid:0) (cid:1) = ∆h ( s +1)−∆h ( s +2). R/C2 2 R/C2 2 (cid:0) (cid:1) (cid:0) (cid:1) Combining (4.3) and (4.4), we see that it remains to show that (4.5) ∆h s −∆h s +1 ≤ ∆h ( s +1)−∆h ( s +2). R/C1 2 R/C1 2 R/C2 2 R/C2 2 (cid:0)(cid:0) (cid:1)(cid:1) (cid:0)(cid:0) (cid:1) (cid:1) (cid:0) (cid:1) (cid:0) (cid:1) By the symmetry of the h-vectors of R/C and R/C we see that this is equivalent to 1 2 showing (4.6) ∆h s−1 −∆h s−1 −1 ≤ ∆h ( s−1 )−∆h ( s−1 −1). R/C1 2 R/C1 2 R/C2 2 R/C2 2 (cid:0)(cid:0) (cid:1)(cid:1) (cid:0)(cid:0) (cid:1) (cid:1) (cid:0) (cid:1) (cid:0) (cid:1) Now, ∆h (i = 1,2) is the Hilbert function of an Artinian monomial complete inter- R/Ci section in R, namely R/C′, where C′ is obtained from C by adding the missing variable. i i i Furthermore, if wereplace C′ by D = (x ,xs+1,xs,...,xs), we have thatR/C′ andR/D 2 2 1 2 3 s 2 2 have the same Hilbert function, and D ⊂ C′. 2 1 But such ideals have the Weak Lefschetz property ([11], [13]). In particular, if L is a general linear form, then the left-hand side of (4.6) is the Hilbert function of R/(C′+(L)) 1 in degree s−1 and the right-hand side is the Hilbert function of R/(D + (L)) in the 2 2 same degr(cid:0)ee. (cid:1)Because of the inclusion of the ideals, (4.6) follows and so the proof is (cid:3) complete. We now come to the main result of this section. The case r = 3 was proven by Brenner and Kaid [4]. Note that when r ≤ 2, all quotients of R have the WLP by a result of [6]. Theorem 4.3. Let R = K[x ,...,x ], with r ≥ 3, and consider 1 r I = (xr,...,xr,x x ···x ). r,r 1 r 1 2 r Then the level Artinian algebra R/I fails to have the WLP. r,r Proof. Specifically, we will check that the multiplication on R/I by a general linear form r,r fails surjectivity from degree r −1 to degree r , even though the value of the Hilbert 2 2 function is non-increasing bet(cid:0)we(cid:1)en these two de(cid:0)gr(cid:1)ees. The proof is in two steps. Step 1. We first prove this latter fact, namely that h (d−1) ≥ h (d) for d = r . R/Ir,r R/Ir,r 2 (cid:0) (cid:1)