A geometric approach to the Weak Lefschetz Property for height four complete intersections Special Session on Commutative Algebra and Its Interactions with Algebraic Geometry Dalhousie University October 18–19, 2014 Juan C. Migliore University of Notre Dame Juan C. Migliore (Notre Dame) Geometric approach to WLP 1 / 24 Goal: Describe part of a long-standing project with Mats Boij, Rosa Marı́a Miró-Roig and Uwe Nagel (started in May, 2011). The Weak Lefschetz Property (WLP) is a seemingly very algebraic notion which, nevertheless, has many connections to geometry. These manifest themselves in different forms, some surprising. We’ll look at one such connection, involving hyperplane sections of certain smooth curves. Juan C. Migliore (Notre Dame) Geometric approach to WLP 2 / 24 Background Let k be an infinite field of characteristic zero. R = k [x1 , . . . , xr ], graded polynomial ring. Let A = R/I = n M Ai i=0 be a standard graded artinian k -algebra. Note that A is finite dimensional over k. Juan C. Migliore (Notre Dame) Geometric approach to WLP 3 / 24 Let ` be a linear form. For each i, multiplication by ` induces a homomorphism ×` Ai −→ Ai+1 . For a general choice of `, we “expect” maximal rank: rk (×`) = min{dim Ai , dim Ai+1 }. (It’s enough to find just one such `.) Philosophy: something interesting should be going on if this is not the case. In our situation, it will have to do with the question of whether a certain set of points imposes the “expected” number of conditions on forms of a certain degree. Juan C. Migliore (Notre Dame) Geometric approach to WLP 4 / 24 Definition A standard graded artinian k-algebra A has the Weak Lefschetz Property (WLP) if, for a general linear form `, the homomorphism (×`) : Ai → Ai+1 has maximal rank for all i (i.e. for each i it is either injective or surjective). Any such linear form is a Lefschetz element. We say that A has the Strong Lefschetz Property (SLP) if (×`d ) : Ai → Ai+d has maximal rank for all i, d (i.e. is either injective or surjective). Juan C. Migliore (Notre Dame) Geometric approach to WLP 5 / 24 This result got the whole area started: Theorem (Stanley (1980), J. Watanabe (1987), Reid-Roberts-Roitman (1991)) Let R = k[x1 , . . . , xr ], where k has characteristic zero. Let I be an artinian monomial complete intersection, i.e. I = hx1a1 , . . . , xrar i. Then R/I has the SLP. I In particular, R/I has the WLP. Juan C. Migliore (Notre Dame) Geometric approach to WLP 6 / 24 Note: Monomial complete intersections are a (very) special kind of complete intersection. As a consequence, if we fix a1 , . . . , ar then the general complete intersection of type (a1 , . . . , ar ) has the WLP. Big Question: Do all graded artinian complete intersections in characteristic zero have the WLP? Here, for the most part, is what is known. Juan C. Migliore (Notre Dame) Geometric approach to WLP 7 / 24 Fact 1: (HMNW 2003) In 2 variables, not only do all complete intersections have the WLP, but in fact everything has the WLP (at least in characteristic zero). Fact 2: (HMNW 2003) Let R = k[x, y , z], where char(k ) = 0. Let I = (F1 , F2 , F3 ) be a complete intersection. Then R/I has the WLP. The proof of Fact 2 uses the Grauert-Mülich theorem for rank two vector bundles on P2 , which we apply to the syzygy bundle of I. Juan C. Migliore (Notre Dame) Geometric approach to WLP 8 / 24 Remarks. 1 After [HMNW], other authors have extended the use of the syzygy bundle and Grauert-Mülich to study the WLP for other kids of algebras, most notably Brenner and Kaid. However, it does not seem to help much with complete intersections of height ≥ 4. 2 SLP is still wide open for complete intersections in three variables (always assuming characteristic zero). It’s one natural direction from the previous theorem. 3 In positive characteristic p, many authors have studied WLP and SLP, and their failure, especially for monomial complete intersections. Juan C. Migliore (Notre Dame) Geometric approach to WLP 9 / 24 There are two main parts of our paper. In one, we introduce the non-Lefschetz locus of an artinian graded algebra, complete with a scheme structure. We study it in the case of general complete intersections, and monomial complete intersections. If there’s time, I’ll say more about it at the end of this talk. In the other part we begin the study of WLP for arbitrary complete intersections in k[x1 , x2 , x3 , x4 ]. This is the topic of this talk. Juan C. Migliore (Notre Dame) Geometric approach to WLP 10 / 24 Overview: We consider height 4 artinian complete intersections with generators of the same degree, in characteristic zero. We translate the WLP question into a question involving general hyperplane sections of a certain smooth curve. The first cohomology of the ideal sheaf of this curve (i.e. its Hartshorne-Rao module) is the desired complete intersection. We apply methods from Castelnuovo theory and Davis’s theorem to the general hyperplane section of this curve, and get partial results about whether all such complete intersections have the WLP. Juan C. Migliore (Notre Dame) Geometric approach to WLP 11 / 24 Theorem Let R = k[x1 , x2 , x3 , x4 ] where k has characteristic zero. Let A = R/I = R/hF1 , F2 , F3 , F4 i where I is a complete intersection and deg Fi = d for all i. (a) If d = 2 or 3 then A has the WLP. (b) Let ` be a general linear form. Then the multiplication maps ×` : [A]t−1 → [A]t 3d + 1 are injective for t < . 2 Juan C. Migliore (Notre Dame) Geometric approach to WLP 12 / 24 Notation and first comments: R = k [x1 , x2 , x3 , x4 ] (char k = 0). I = hF1 , F2 , F3 , F4 i is a complete intersection with deg Fi = d ≥ 2 for i = 1, 2, 3, 4. A = R/I. We want to know if A has the WLP. The socle degree of R/I is 4d − 4, an even integer. Thus the Hilbert function of R/I has one “peak,” in degree 2d − 2. Juan C. Migliore (Notre Dame) Geometric approach to WLP 13 / 24 For example, if d = 4 then 2d − 2 = 6 and the Hilbert function is deg 0 1 1 4 2 10 3 20 4 31 5 40 6 44 7 40 8 31 9 20 10 10 11 4 12 1 Easy fact: R/I has the WLP if and only if ×` [R/I]2d−3 −→ [R/I]2d−2 is injective. (You don’t have to worry about any other degree.) Remark: Our main result gives injectivity up to degree ∼ 32 d, while we need degree ∼ 2d. Juan C. Migliore (Notre Dame) Geometric approach to WLP 14 / 24 Construction: After a general choice of basis, set IC1 = hF1 , F2 i and IC2 = hF3 , F4 i, so I = hF1 , F2 , F3 , F4 i = IC1 + IC2 . C1 and C2 are smooth, irred. disjoint curves of degree d 2 by Bertini. ⇒ C = C1 ∪ C2 is a smooth curve in P3 of degree 2d 2 . C1 and C2 are complete intersection (hence ACM) curves in P3 . Since C1 and C2 are ACM curves in P3 , IC = IC1 · IC2 (Serre). So IC has four minimal generators, all of degree 2d. Q: What’s the deficiency (= Rao) module of C, denoted M(C)? Juan C. Migliore (Notre Dame) Geometric approach to WLP 15 / 24 A: From the sequence 0 → IC1 ∩ IC2 → IC1 ⊕ IC2 → IC1 + IC2 → 0 we sheafify and take cohomology, summing over all twists. Since I = IC1 + IC2 is artinian and C1 , C2 are ACM, we obtain 0 IC IC1 ⊕ IC2 R H∗1 (IC ) 0 I 0 and hence Juan C. Migliore (Notre Dame) 0 M(C) := H∗1 (IC ) ∼ = R/I. Geometric approach to WLP 16 / 24 Some ingredients of the proof of the main theorem of this talk: The general hyperplane section of C is the union, Z = Z1 ∪ Z2 , of two sets of d 2 points, each with the uniform position property (UPP). “Symmetry:” The general hyperplane section can’t have any property that “favors” Z1 over Z2 or vice versa. Let ` be a general linear form defining a plane H, and let t be any integer. We have a standard exact sequence of sheaves ×` 0 → IC (t − 1) −→ IC (t) → IC∩H|H (t) → 0 leading to a long exact sequence in cohomology: Juan C. Migliore (Notre Dame) Geometric approach to WLP 17 / 24 ×` ×` 0 → [IC ]t−1 −→ [IC ]t → [IC∩H ]t → [R/I]t−1 −→ [R/I]t → H 1 (IC∩H (t)) → H 2 (IC (t − 1)) → H 2 (IC (t)) → 0 where IC∩H is the ideal of C ∩ H viewed in H ∼ = P2 . One can show I dim[IC ]2d−1 = 0, dim[IC ]2d = 4 (as already noted). I h2 (IC (t)) = 0 for t ≥ 2d − 3. Then setting t = 2d − 2 above, we get ×` 0 → [IC∩H ]2d−2 → [R/I]2d−3 −→ [R/I]2d−2 → H 1 (IC∩H (2d − 2)) → 0. So the WLP depends entirely on the Hilbert function of C ∩ H. Juan C. Migliore (Notre Dame) Geometric approach to WLP 18 / 24 WLP ⇔ dim[IC∩H ]2d−2 = 0. Specifically: What can we say about the h-vector (= first difference of the Hilbert function) for C ∩ H ⊂ P2 ? Fact: The value of the h-vector in degree 2d − 1 is always d, whether WLP holds or not. This translates to the “expected” h-vector for C ∩ H: degree 0 1 1 2 2 3 ... ... 2d − 3 2d − 2 2d − 2 2d − 1 ↑ − 2d − 1 d ↑ 2d 0 ↑ + The expected value in degree 2d is then forced to be 0 since deg C = 2d 2 . (Simple calculation.) One shows there’s a balanced see-saw effect if WLP fails. Juan C. Migliore (Notre Dame) Geometric approach to WLP 19 / 24 Using the UPP property for the two “halves” of C ∩ H for general H, the mentioned symmetry property, the famous theorem of Davis, one can show: The h-vector of C ∩ H is of decreasing type (and a bit more). The main theorem follows from a very careful study of the possibilities remaining after these are taken into account. Juan C. Migliore (Notre Dame) Geometric approach to WLP 20 / 24 Example Say d = 2. The expected h-vector of C ∩ H is (1, 2, 3, 2). The “2” in degree 3 is forced. The only other possible h-vector would be (1, 2, 2, 2, 1). This can be shown to be impossible. Hence R/I has the WLP. Example If d = 3, we also show WLP holds. If d = 4 then injectivity is forced by our theorem except in the very middle. Juan C. Migliore (Notre Dame) Geometric approach to WLP 21 / 24 The other part of the paper For an artinian complete intersection I ⊂ R = k[x1 , . . . , xn ], define the non-Lefschetz locus, LI ⊂ Pn−1 , to be the locus of linear forms that fail to be Lefschetz elements. Note WLP ⇔ codim LI ≥ 1. Assume the socle degree of R/I is e. Then R/I has the WLP if and only if, for a general linear form `, ×` : [R/I]b e−1 c → [R/I]b e+1 c is injective. 2 2 The part of LI coming from any two consecutive components of R/I has a scheme structure (not necessarily reduced) and in fact is determinantal. As a scheme, LI turns out to be determined by the single ×` mentioned above. (This comes from the Gorenstein property.) Juan C. Migliore (Notre Dame) Geometric approach to WLP 22 / 24 The codimension of LI is a finer measure of how “general” linear forms behave than simply the question of whether WLP holds. There is an expected codimension for LI depending on the Hilbert function of R/I. We show that in several situations (in terms of number of variables), a general complete intersection has non-Lefschetz locus of expected codimension. In contrast, we show that a monomial complete intersection almost always has non-Lefschetz locus of codimension 1 (and we describe it). This highlights a key difference between the monomial case and the general case, despite the fact that both have WLP and SLP. Juan C. Migliore (Notre Dame) Geometric approach to WLP 23 / 24 Thank you. Juan C. Migliore (Notre Dame) Geometric approach to WLP 24 / 24