The Weak Lefschetz Principle is False for Ample Cones
Brendan Hassett
∗, Hui-Wen Lin
†and Chin-Lung Wang
‡October 26, 2001
1 Introduction
Let X be an (n + 1)-dimensional smooth complex projective variety and let D be a smooth ample divisor of X with inclusion map i : D → X. The well- known Weak Lefschetz Theorem (see [GrHa]) asserts that the restriction map i∗ : Hk(X; Z) → Hk(D; Z) is an isomorphism for k ≤ n − 1 and is compatible with the Hodge decomposition. For n ≥ 3 one deduces from these results that i∗ : Pic(X) → Pic(D) is also an isomorphism. Grothendieck has shown that this statement is true over any algebraically closed field [Hart]. While an ample line bundle on X always restricts to an ample line bundle on D, it is not at all clear whether Amp(D) ⊂ i∗Amp(X), i.e., whether the Weak Lefschetz Principle holds for the ample cone.
In this note we provide two examples showing that the Weak Lefschetz Principle for the ample cone fails in general. One is obtained by blowing up (§2) and the other is a product with a P1 factor (§3). We also provide some partial positive results (§4). However, a complete picture of how the ample cone behaves under the Weak Lefschetz isomorphism remains elusive.
∗Mathematics Department, Rice University, Houston, USA. Email address: has- sett@math.rice.edu
†Department of Mathematics, National Central University, Chungli, Taiwan. Email address: linhw@math.ncu.edu.tw
‡Department of Mathematics, National Tsing-Hua University, Hsinchu, Taiwan. Email address: dragon@math2.math.nthu.edu.tw
Acknowledgments The first author was partially supported by the Insti- tute of Mathematical Sciences of the Chinese University of Hong Kong and NSF Grant 0070537. He would like to thank the National Center for Theo- retical Sciences of Hsinchu, Taiwan, for the invitation that made this collab- oration possible.
2 A blow-up example
2.1 The construction
We construct our first counterexample (X, D). Let φ : X → P4 be the blow- up of P4 at two distinct points p1 and p2. Let l0 be the line spanned by p1
and p2 and D0 a general smooth cubic hypersurface (threefold) containing p1 and p2 but not the line l0. The conditions satisfied by D0 will be made precise in 2.5. We take D to be the proper transform of D0 in X. We will see that D is a very ample divisor in X but i∗Amp(X) 6= Amp(D). More precisely, Amp(D) is strictly larger than i∗Amp(X) if and only if the Mori cone (the closure of the real cone generated by numerically equivalent classes of effective one-cycles) NE(X) is strictly larger than i∗NE(D) (by Kleiman’s criterion for ampleness [Hart]). Let l be the proper transform of l0 in X, which is an effective one-cycle in X. By Weak Lefschetz, l = i∗λ for some one-cycle λ on D. Our main task is to show that λ 6∈ NE(D). In fact we will determine both the ample cones and the Mori cones of X and D.
2.2 Ample and Mori cones of X
We first set up some notation; we shall often use a single letter to denote a subvariety and its homology class. Let E1 and E2 be the exceptional divisors in X, H0 the general hyperplane in P4 containing p1 and p2, and H its proper transform in X. Note that l0 = H03 and H = φ∗H0 − E1 − E2. It is clear that Pic(X) = ZH + ZE1 + ZE2. Since Ei ∼= P3 and −Ei|Ei is its hyperplane class, Ei3 corresponds to the class of a line in Ei. The group of one-cycles (modulo rational or numerical equivalence) is then given by N(X) = Zφ∗H03+ ZE13 + ZE23 and l = φ∗H03− E13− E23. As we blow up at only two points, it is readily seen that NE(X) = R≥0l + R≥0E13+ R≥0E23 which is already a rational closed cone. For later use, we also observe that l.Ei = 1 and Ei4 = (Ei|Ei)3 = −1.
Proposition 2.1 The ample cone Amp(X) is the interior of the cone gen- erated by H + E1 + E2, H + E1 and H + E2. That is, a Q-divisor L = φ∗H0− a1E1− a2E2 is ample if and only if ai > 0 and a1+ a2 < 1. Moreover, all ample divisors are automatically very ample.
Proof. By Kleiman’s criterion, L is ample if and only if L.l = 1 − a1− a2 > 0 and L.Ei3 = ai > 0.
For the last statement, one observes that X is naturally a smooth toric variety and for smooth toric varieties, ample divisors are automatically very ample (Theorem of Demazure, see e.g. [Oda]). Q.E.D.
Corollary 2.2 The divisor D ∼ 3φ∗H0− E1 − E2 is very ample in X but
−KX ∼ 5φ∗H0− 3E1− 3E2 is not even ample. In particular, X is not Fano.
2.3 The main argument
We recall the following criterion for when the blow-up of a cubic surface at two points is a Del Pezzo surface:
Proposition 2.3 Let S0 be a smooth cubic surface containing distinct points p1 and p2, l0 the line spanned by these points, and S the blow-up of S0 at p1 and p2. Assume that l0 ∩ S0 consists of three distinct points, and no line containing any one of these three points is contained in S0. Then S is a Del Pezzo surface.
Proof. This argument is inspired by some remarks of Naruki [Nar]. We write l0 ∩ S0 = {p1, p2, p3}, τ0 : T = Bl(S0)p1,p2,p3 → S0, and τ : T → S for the induced map. Let Fi ⊂ T denote the exceptional curve over the point pi, so that KT = τ0∗KS0 + F1+ F2 + F3 = τ∗KS+ F3. Projection from the line l0 induces a morphism µ : T → P1; the fibers of µ are intersections of S0 with hyperplanes containing l0. In particular, the general fiber is a smooth plane cubic, i.e., µ is an elliptic fibration. Since S0 contains no lines containing any of the pi, the fibers of µ are irreducible. Furthermore, the hyperplane class of P1 pulls back to −KT = −τ∗KS− F3.
To prove that −KS is ample we apply the Nakai-Moishezon criterion:
−KS is ample provided that KS2 > 0 and −KS.C > 0 for each closed ir- reducible curve C ⊂ S. It suffices to check the proper transform CT of C intersects −τ∗KS positively. If µ contracts CT to a point then (−τ∗KS).CT = F3.CT > 0 because the fibers of µ are irreducible and F3 is a section of µ. On
the other hand, if CT dominates P1 then −KT.CT is positive. Since CT 6= F3,
(−τ∗KS).CT is positive as well. Q.E.D.
Remark 2.4 The classification theory of surfaces implies that a cubic sur- face S0 is the blow-up of P2 at six general points [GrHa]. Precisely, we require that no two of the points coincide, no three are collinear, and no six are con- tained in a plane conic. Implicit in our proof is a precise condition for when the blow-up of P2 at eight general points is a Del Pezzo surface. In addition to the conditions listed above, we require that there exists no plane cubic passing through all eight points and singular at one of the eight. Equiv- alently, the pencil of cubic curves passing through the eight points should have reduced base locus. This criterion was suggested without proof in §26 of [Manin].
Assumptions 2.5 We make the following generality assumptions on D0: 1.The intersection l0∩ D0 contains three distinct points.
2.There are a finite number of lines contained in D0 and meeting l0∩ D0. For a general point of a cubic threefold, there are six lines containing the point and contained in the threefold.
Let S0 = D0∩ H0 be a hyperplane section of D0 containing p1 and p2, a cubic surface in H0 ∼= P3. We choose S0 so that it is smooth and does not contain any lines meeting l0∩ D0 (such an S0 exists by our second assumption above.) Let S ⊂ H be its proper transform, the blow-up of S0 at p1 and p2.
Note that S = (φ∗H0− E1− E2)|D and KD = (−2φ∗H0+ 2E1+ 2E2)|D =
−2S. Also, since S0 is a cubic surface inside H0 ∼= P3, adjunction shows that KS0 = (KH0 + S0)|S0 = (−4H0|H0 + 3H0|H0)|S0 = −H0|S0. By the blow-up formula, KS = φ∗KS0 + E1|S+ E2|S = (−φ∗H0 + E1+ E2)|S = −H|S. Proposition 2.6 Consider an effective one-cycle ˜l := φ∗H03− a1E13− a2E23 in X with ai ≤ 1 and a1+ a2 > 1. Let ˜λ be the one-cycle class in D which corresponds to ˜l. Then ˜λ 6∈ NE(D). In particular λ 6∈ NE(D).
Proof. If ˜λ ∈ NE(D) then ˜λ is the limit of a sequence {Ci} of effective one- cycles on D with rational coefficients. By assumption, ˜λ.S = 1 − a1− a2 < 0 which implies that Ci.S < 0 for i large enough. In particular this shows that Ci ⊂ S for i large enough.
On the other hand, the sequence Ci.KS has limit ˜l.(−H) = −1+a1+a2 >
0. This imples that Ci.KS > 0 for i large enough, contradicting the fact that
−KS is an ample divisor in S. Q.E.D.
Remark 2.7 To get counterexamples to the Weak Lefschetz Principle, it is essential that the cubic threefold D0 does not contain l0. In fact, if D0 contains l0 then S0 will also contain l0. Thus S is not a blow-up of P2 at eight
‘general’ points, −KS is not ample, and no contradiction arises.
2.4 Ample and Mori cones of D
We retain the assumptions of section 2.3. Proposition 2.6 exhibits effective classes in X that are not represented by effective classes in D. We would like to show here that all other effective classes in X lie in the image i∗NE(D).
Theorem 2.8 The image of the Mori cone i∗NE(D) is a closed rational polyhedral cone generated by l + E13, l + E23, E13 and E23. The image of the ample cone i∗Amp(D) is the interior of the rational polyhedral cone generated by H, H + E1, H + E2 and H + E1+ E2 (the nef cone).
l
l+ E
1 E3
2
E
3 1
2 3
3
E l+
K C=0 K C=0
D X
Figure 1: Mori cones of X and D
Proof. Recall that the class l + E13 = φ∗H03− E23 is represented by the proper transform of a line in P4 which contains p2 and is distinct from `. Since D0 is a cubic threefold, through every point of D0 one may find lines [Harris].
The proper transform in D of one such line will have the class l + E13. The same argument applies to l + E23. Since the cone generated by l + E13, l + E23, E13 and E23 is precisely NE(X) with the classes considered in Proposition 2.6 removed, we conclude that it is i∗NE(D).
The statement on the ample cone is an application of Kleiman’s criterion to L := H + a1E1+ a2E2; intersecting L with l + E13, l + E23, E13 and E23, we
get 0 < a1, a2 < 1. Q.E.D.
3 A product example with P
1factors
Much simpler examples can be found if X is a product. Take X = P1 × Pd (for d ≥ 3) and let D be a divisor of type (d, b) with b ∈ N. The divisor D is very ample and thus a generic member of |D| is smooth. It has defining equation
xdf0+ xd−1yf1+ ... + fdyd = 0,
where x, y are coordinates of P1 and the fi’s are polynomials of degree b in Pd. The projection p : D → Pd has positive dimensional fibers exactly when f0 = f1 = ... = fd= 0. This has no nontrivial solutions for general fi’s since there are more equations than variables. However, if p is a finite morphism then each ample divisor L on Pd pulls back to an ample divisor on D (this follows from either the Nakai-Moishezon criterion or Kleiman’s criterion), yet this pull-back divisor is the restriction of the divisor P1× L on X = P1× Pd, which is evidently not ample. This gives another (easier) counterexample to the Weak Lefschetz Principle.
The presence of a one-dimensional factor here is crucial (cf. Theorem 4.1).
4 Positive results
4.1 Products
It is trivial that the Weak Lefschetz Principle holds if X has N´eron-Severi rank one and dim X ≥ 4. One may generalize this in a straightforward manner to obtain:
Theorem 4.1 Let i : D → X = Q
Xj be a smooth ample divisor in a finite product of smooth projective varieties, each with dimension ≥ 2 and N´eron-Severi rank equal to one. Assume that dim X ≥ 4 and Pic(X) = Lp∗jPic(Xj), where pj : X → Xj is the projection map. Then i∗Amp(X) = Amp(D).
Proof. Let Hj be an ample class on Xj and let hj = p∗jHj. Notice the following fact: If nj = dim Xj, then Q
hmj j is an effective cycle if and only if mj ≤ nj for all j and p = Q
hnjj is a positive integer. It is then easy to see that P
ajhj is an ample class on X if and only if that aj > 0 for all j:
simply intersect it with hnjj−1.Q
k6=jhnkk to get ajp > 0.
Now let D be given by the ample classP
djhjwith dj > 0. Since dim X ≥ 4, a divisor on D takes the form L|D = P
ajhj|D. We need to show that L|D is ample implies that aj > 0 for all j. Indeed, since nj ≥ 2, we simply intersect L|D with the effective cycle hj|nDj−2.Q
k6=jhk|nDk on D to get 0 < hnjj−2.Y
k6=jhnkk.X
ajhj.X
djhj = ajdjp.
It follows that aj > 0. Q.E.D.
Remark 4.2 The condition Pic(X) = L
p∗jPic(Xj) holds if all but one of the factors satisfy h1(Xj, OXj) = 0, or more generally, if the Jacobians of the factors admit no nontrivial endomorphisms η : Jac(Xi) → Jac(Xj), i 6= j.
4.2 A partial theorem on Mori cones
The counterexample of section 2 suggests the following equality on the neg- ative part of Mori cones of D and X (compare Proposition 2.1 and Theorem 2.8). The proof relies on Mori’s theory of extremal rays [Mori] and is essen- tially contained in [Wi´s] and [Koll´ar].
Theorem 4.3 Let i : D → X be a smooth ample divisor in a smooth variety X with dim X ≥ 4. Then i∗NE(D)KD≤0 = NE(X)KD≤0.
Proof. Since KD = KX|D + D|D, which is numerically strictly more positive than KX, we know by Mori’s theory that NE(X)KD≤0 is a finite polyhedral cone generated by extremal rays. Let R = RC with C ∼= P1 be such a ray and φ : X → Y the corresponding contraction. We want to show that the class of C is also an effective class in D. If φ has a fiber F of dimension at least two then D ∩ F has positive dimension and contains a curve with class in R. If all fibers are one-dimensional, Wi´sniewski’s theorem shows that Y is smooth and either φ is a blow-up of Y along a smooth codimension-two subvariety Z or φ is in fact a conic bundle. These cases are ruled out when dim X ≥ 4 by an argument of Koll´ar (the Lemma of [Koll´ar]). Q.E.D.
References
[GrHa] P. Griffiths and J. Harris; Principles of Algebraic Geometry, Wiley- Interscience, New York, 1978.
[Harris] J. Harris; Algebraic Geometry: A First Course, Graduate Texts in Math.
133, Springer-Verlag, New York, 1992.
[Hart] R. Hartshorne; Ample Subvarieties of Algebraic Varieties, Lecture Notes in Math. 156, Springer-Verlag, Berlin-New York, 1970.
[Koll´ar] J. Koll´ar; Appendix to “C. Borcea; Homogeneous Vector bundles and fam- ilies of Calabi-Yau Threefolds, II”, Proc. Symp. Pure Math. 52 (1991).
[Manin] Y.I. Manin; Cubic Forms: Algebra, Geometry, Arithmetic, North- Holland, Amsterdam, 1974.
[Mori] S. Mori; Threefolds whose canonical bundles are not numerically effective, Ann. of Math. 116 (1982), 133-176.
[Nar] I. Naruki; Cross ratios as moduli of del Pezzo surfaces of degree one, Proc.
Japan Acad. Ser. A Mat. Sci. 56 (1980), no. 7, 328-331.
[Oda] T. Oda; Convex Bodies and Algebraic Geometry, Springer-Verlag, Berlin, 1988.
[Wi´s] J.A. Wi´sniewski; On contractions of extremal rays of Fano manifolds, J.
reine angew. Math. 417 (1991), 141-157.