Lines on Fano 3-folds according to Shokurov

Lines on Fano 3-folds according to Shokurov
Miles Reid
0
Introduction and the main theorem
This paper is a reconstruction of the proof in [Sh] of the following result:
Theorem Let V be a complete 3-fold with ample anticanonical divisor −KV .
Then just one of the following holds:
(i) V contains a line;
(ii) index V ≥ 2;
(iii) V ∼
= P1 × P2 .
In this section, after some definitions, I state the main intermediate steps
Lemma 2.2, Lemmas 2.3–2.4 and Lemma 2.50 on which Shokurov’s proof is
based, and then give his proof using them.
I want to emphasise that not only the overall scheme of the proof, but
also many of the essential details of this paper have been lifted from [Sh] –
although in many cases I have been able to get rid of some of Shokurov’s
extraordinarily involved proofs, and deduce stronger corollaries. I am now
quite sure that [Sh] contains no serious errors or gaps, so that it contains a
complete – if somewhat elephantine – proof.
This paper had its origins in a review of [Sh] for Math Reviews. Conversations with Bob Friedman, Mark Levine, Ron Livné and David Mumford at
the August 1980 session of the Montreal/Nato S.M.S. impressed on me the
desirability of a laundry job on [Sh]. This paper was written during a month’s
stay at the Mittag-Leffler Institute in Sep 1980, whom I would like to thank
for a travel grant. I also thank Fabrizio Catanese for discussions.
Conventions
A Fano 3-fold is a nonsingular projective 3-fold V with ample −KV . The
index of V is that integer r such that 1r KV ∈ Pic V is primitive. The Fano
3-folds of index ≥ 2 are well understood [Isk, Part I, 4.2]; the varieties of
1
index 1 for which |−KV | has a base locus [Isk, Part I, 3.3], the hyperelliptic
ones for which ϕ−KV is not an isomorphism, [Isk, Part I, 7.2], and the trigonal
varieties for which the anticanonical image V2g−2 ⊂ Pg+1 is not an intersection
of quadrics [Isk, Part I, 2.5] are well known. All these either contain pencils
of del Pezzo surfaces, or are weighted complete intersections, and it is not
difficult to show that they have lines.
A line (respectively conic) C ⊂ V is an irreducible curve C for which
−KV · C = 1 (respectively 2). A Veronese surface (respectively scroll Fn;r )
in V is a surface S ∼
= P2 with OS (−KV ) ∼
= OP2 (2) and OS (−S) ∼
= OP2 (1)
∼
1
(O⊕O(−n))
=
F
,
with
O
(−K
)
O
(respectively S ∼
P
((n+r)A+B),
= P
n
S
V =
Fn
where A is the fibre of Fn → P1 , and B the negative section: B 2 = −n,
AB = 1, A2 = 0). We will be concerned especially with F0;2 and F2;1 , and
these will be referred to collectively as quartic scrolls. These surfaces are said
to be linearly normal if H 0 (OV (−KV )) H 0 (OS (−KV )); I will usually not
make this assumption.
I will say that a surface S ⊂ V is a surface of degree n − 1 spanning Pn
(written S = Sn−1 ⊂ Pn ) if deg S = (−KV )2 S = n−1, and H 0 (OV (−KV )) H 0 (OS (−KV )) with h0 (OS (−KV )) = n + 1. Under ϕ−KV , S goes into one of:
P2 , quadric ⊂ P3 , rational normal scroll (including cones),
normally embedded Veronese surface.
I always assume that |−KV | is free. If X ∈ |−KV | is a general element, and
C = X1 ∩X2 with Xi ∈ |−KV | general, then X is a K3 surface and C a curve,
with |−KV ||C the complete canonical system; C ⊂ X ⊂ V will be referred to
as general curve and K3 sections. By [S-D, p. 611], a complete linear system
|D| on X without fixed components has no base point; a simple argument
shows that the same holds for V provided that H 0 (OV (D)) H 0 (OX (D)).
I will use the shorthand (read “maps onto”) to mean that a linear map
is surjective. I will also write
|−KV | 3 D1 + D2 ,
with Di > 0
to mean that |−KV | contains a reducible divisor, and
|−KV | 63 D1 + D2 ,
with Di > 0
to mean that every D ∈ |−KV | is irreducible.
Plan of proof
Shokurov’s proof is based on the following results 2.2–2.5:
2
Lemma 2.2 ([Sh, §8], and Section 1 below) Assume that −KV is ample, and that |−KV | 3 D1 + D2 , with Di > 0. Then one of the following
holds: (i–iii) of the theorem, or
(iv) V contains a Veronese surface.
The following results are similar to those used in the Fano–Iskovskikh
method of projections and multiple projections from suitable loci in V .
Lemma 2.3–2.4 ([Sh, §§6–7], and Section 3 below) Assume that −KV
is very ample, and that V = V2g−2 ⊂ Pg+1 is an intersection of quadrics.
Suppose that |−KV | 63 D1 + D2 with Di > 0.
(A) Let P ∈ V be a point not lying on any line or conic. Construct the
blowup σ : W → V of P ∈ V , and set σ −1 P = S. Then −KW =
σ ∗ (−KV ) − 2S is again very ample.
(B) Let q ⊂ V be a conic not meeting any lines of V , and with NV |q ∼
=
OP1 ⊕ OP1 or OP1 (1) ⊕ OP1 (−1). Construct the blowup σ : W → V of
q ⊂ V , and set S = σ −1 q. Then −KW = σ ∗ (−KV ) − S is again very
ample.
Note that in (A), W ⊃ S = Veronese surface; in (B), W ⊃ S = quartic
scroll F0;2 or F2;1 .
0
Remark 1 The projected variety W = W2g0 −2 ⊂ Pg +1 is a Fano 3-fold with
g 0 = g − 4 or g − 3 in the two cases, and rank Pic W ≥ 2; such varieties do not
exist for g 0 ≤ 5. Hence proving Lemma 2.3–2.4 in the cases g = 6, 7 and 8 is
equivalent to proving that every conic meets a line and that every point is on
a line or conic.
Remark 2 See [Isk, Part II, 4.3] for information on conics; we need to
know that if q is a conic from a covering family then NV |q ∼
= OP1 ⊕ OP1
or OP1 (1) ⊕ OP1 (−1).
Lemma 2.50 ([Sh, §5], and Section 2 below) Let W be a 3-fold having
very ample −KW , and suppose that W ⊃ S = Veronese surface or quartic
scroll. Make the further assumption
|−KV | 63 D1 + D2 + nS,
with Di > 0 and S 6⊂ Di
(irred mod S.)
Then the linear system |E| = |−KW − 2S| is free, with dim ϕE (W ) = 3.
Note that Lemma 2.50 requires no hypothesis on lines of W .
The key result follows at once from Lemma 2.50 :
3
Proposition 2.5 Let W be a 3-fold with very ample −KW , and suppose that
W contains a Veronese surface or a quartic scroll. Then
|−KV | 3 D1 + D2 + nS,
with Di > 0 and S 6⊂ Di .
Proof The conclusion of Lemma 2.50 is impossible. For by the Grauert–
Riemenschneider vanishing theorem [G–R, p. 263], H i (OW (−E)) = 0 for
i = 1, 2. Hence H 1 (OE ) = 0 for E ∈ |E|, and χ(OE ) > 0. This is in
contradiction with the values
(
−2 if S = Veronese surface,
χ(OE ) =
−1 if S = quartic scroll,
calculated as follows: if H ∈ |−KW |, then χ(OH ) = 2 follows from
0 → OW (KW ) → OW → OH → 0.
Now if Ei ∈ |H − iS| for i = 1, 2, then OS (−Ei+1 ) = OS (KS + iS), and the
exact sequences
0 → OW (KW + (i + 1)S) → OW → OEi+1 → 0
and
0 → OW (KW + iS) → OW (KW + (i + 1)S) OS (KS + iS)
give χ(OE1 ) = 1 and χ(OE2 ) = −2 or −1 in the two cases. Q.E.D.
Remark Lemma 2.50 could perhaps also be used to bring the hypotheses of
Lemma 2.3–2.4 into contradiction with the results of Mori [M], which imply
that V has a large family of rational curves of degree ≤ 4.
Corollary If |−KV | 63 D1 + D2 with Di > 0 then every point of V lies on
a conic or a line, and every conic meets a line.
Proof of the Theorem Compare [Sh, §2]. We get a contradiction by
assuming that rank Pic V is minimal among V with
−KV is ample; and none of (i–iii) hold.
(∗)
First of all, −KV is very ample and V = V2g−2 ⊂ Pg+1 is an intersection of
quadrics, since otherwise V is known to contain a line. Now |−KV | 3 D1 +D2
with Di > 0, for otherwise, according as V is covered by conics or not, one of
the constructions (A) or (B) of Lemma 2.3–2.4 is applicable, to give a blown
up variety W satisfying the hypothesis of Proposition 2.5. Since σ ∗ (−KV ) =
4
−KW + εS (with ε = 2 or 1), the decomposition given by Proposition 2.5 also
gives |−KV | 3 D1 + D2 with Di > 0.
Now use Lemma 2.2; by (∗), V must contain a Veronese surface S. Since
OS (−S) ∼
= OP2 (1), V is the blowup σ : V → V1 of a nonsingular point P ∈ V1 ,
with σ −1 (P ) = S. Now because −KV = σ ∗ (−KV1 ) − 2S, we see that −KV1 is
ample on V1 . Obviously V1 cannot contain lines, V1 ∼
6= P1 × P2 and index V1 =
1, since otherwise the blowup V would contain curves C with −KV C ≤ 1,
or would itself have index ≥ 2. Thus V1 also satisfies (∗), but has smaller
rank Pic V1 . This contradiction proves the theorem.
1
Proof of Lemma 2.2
Assume that |−KV | is free and ample, and that |−KV | 3 D1 +D2 with Di > 0.
The proof of Lemma 2.2 consists of showing that every such V falls under one
of the following cases (a–e), and then reading off (i–iv) of Lemma 2.2 from
properties of these varieties.
(a) V has a free pencil |S| of del Pezzo surfaces.
(b) V contains a Sn−1 ⊂ Pn .
(c) V → P2 is a conic bundle.
(d) index V ≥ 2.
(e) V is the blowup of P3 or of the nonsingular quadric Q ⊂ P4 in a nonsingular curve C.
Remark The thorough study of each of these classes of varieties presents a
challenging problem.
(a) The study of 3-folds with a pencil of del Pezzo surfaces is begun in Step 4
below – for the purpose of proving Lemma 2.2, it is sufficient to deal with
the cases of surfaces of degree 8 and 9, since the others contain lines.
On the other hand, Fano 3-folds with pencils of del Pezzo surfaces of
degrees 1, 2 or 3 have been treated in [Isk], since for them either |−KV |
has a base locus [Isk, Part I, 3.3 and the erratum in the introduction
of the English translation of Part II], or V is one of the hyperelliptic or
trigonal varieties [Isk, Part I, 7.2 and Part II, 2.5]. In each case there
are one or two families other than the product P1 ×(del Pezzo surfaces).
(b) The surfaces Sn−1 ⊂ Pn are always exceptional: the scrolls arise from a
blowup σ : V → V1 , where V1 is another Fano 3-fold, and the centre of
the blowup is a curve C ⊂ V1 not meeting any lines of V1 ; similarly, the
5
Veronese surface arises by blowing up a point not on any line or conic.
P2 and Q ⊂ P3 can also be obtained in this way, where V1 is a “Fano
3-fold with Mori singularities”; for example, a V with planes can be
obtained by blowing up X4 ⊂ P(13 , 22 ), a variety of index 3/2 (compare
Dëmin [D, Theorem 1, 1B]), and Q by blowing up a double point of a
cubic. In each case V contains only lines in the exceptional surface.
Conjecture A Fano 3-fold is covered by conics, and contains an ample
divisor made up of scrolls of lines, provided that V is not one of the following
(or a blowup of one of the following):
• a variety of index ≥ 2;
• P1 × P2 ;
• a variety containing a plane.
There is a breath-taking analogy with del Pezzo surfaces.
(d) These are known [Isk, Part I, 4.2].
(e) Obviously accessible.
(c) This class is huge (see Sarkisov [S1] and [S2]).
Problem 1 Do there exist Fano 3-folds V such that rank Pic V ≥ 2, but
|−KV | 63 D1 + D2 with Di > 0, and V contains no contractible surfaces?
Problem 2 Classify the 3-folds V for which −KV = rH + D with r ≥ 1,
and D > 0; for surfaces we find again our old friends Sn−1 ⊂ Pn .
Step 1 Assume that V has no free pencil. Let H ∈ |−KV | and let Γ be a
component of H of minimal degree. Then |Γ| is free provided that Γ is not
an Sn−1 ⊂ Pn (for some n ≤ g).
Proof Write Γ + D = H ∈ |−KV |. I can obviously assume that D 6= 0.
Let C ⊂ X ⊂ V be general curve and K3 sections. Both Γ and Γ ∩ X are
irreducible, so H 1 (OV (−Γ)) = H 1 (OX (−Γ)) = 0, and
H 0 (OV (D)) H 0 (OX (D)) H 0 (OC (D)).
I claim that unless Γ = Sn−1 ⊂ Pn , |D| has no fixed components, and is hence
free. This follows from the following, which will be good fun for the reader:
6
Exercise ([Sh, 3.23]) Deduce the following from Clifford’s theorem: let
d > 0 be a special divisor on a curve C; then
deg(fixed part of |KC − d|) ≤ deg d,
with equality if and only if
either deg d = g − 1 and h0 (OC (d)) = 1,
or C is hyperelliptic, deg d = n − 1 and h0 (OC (d)) = 1 for some n ≤ g.
Thus |D| can only have a fixed component if Γ = Sn−1 ⊂ Pn , and is free
otherwise. By Bertini’s theorem, and the assertion that |D| is not composed
of a pencil, |D| is irreducible. Hence H 1 (OV (−D)) = 0, and
H 0 (OV (Γ)) H 0 (OX (Γ))
H 0 (OX (−KV )) H 0 (OX∩Γ (−KV )).
The assumption that Γ is fixed implies that Γ ∩ X is a −2-curve on X, hence
∼
= P1 , and is normally embedded by ϕ−KV . Thus Γ = Sn−1 ⊂ Pn . Q.E.D.
Step 2 I now suppose that |Γ| is irreducible, dim |Γ| ≥ 2, and H 0 (OV (Γ)) H 0 (OX (Γ)); if V has no free pencil and no Sn−1 ⊂ Pn , then every component
of every H ∈ |−KV | has this property.
The restriction |Γ||X is still irreducible, and of dimension ≥ 2. Hence
Γ2 X > 0; also |Γ| is free.
Proposition Suppose that every Γ ∈ |Γ| is irreducible, and dim ϕΓ (V ) = 2.
Then ϕΓ : V → P2 is a conic bundle. If furthermore ϕΓ has no reducible fibres
f3 (the blowup of a point in P3 ), or V contains lines.
then V ∼
= P1 × P2 or P
f3 has index 2.
Note that P
Proof ϕΓ (V ) certainly contains ϕΓ (X), which by [S-D, §5, p. 616] is either
birational to X, or is an Sn−1 ⊂ Pn . Now ϕΓ : V → X is obviously impossible,
so that if ϕΓ (V ) is 2-dimensional, it is Sn−1 ⊂ Pn . Now n = 2, for otherwise
|Γ| would contain reducible divisors: thus ϕΓ : V → P2 . If f is the fibre, then
−KV · f = 2, and f ∼
= P1 ; if every fibre is irreducible then V = P(ξ), with ξ
a rank 2 vector bundle over P2 . The last assertion comes from the following
lemma, using Van de Ven’s result on uniform bundles [VdV].
Lemma Let V = P(ξ) → P2 , with ξ a rank 2 vector bundle. Then if V is a
Fano 3-fold, for every line l ⊂ P2 , ξ |l ∼
= OP1 (a1 ) ⊕ OP1 (a2 ) with |a1 − a2 | ≤ 2;
if V contains no lines then |a1 − a2 | ≤ 1.
7
Proof (See also Dëmin [D]). Restricting over a line l ⊂ P2 , we get
F ⊂ V
↓
↓
l ⊂ P2
F ∼
= Fa with a = |a1 − a2 |.
We have KF = (KV + F )|F and OF (F ) = ϕ∗ Ol (1). If B ⊂ Fa is the negative
section, then B 2 = −a, KF · B = a − 2, and KV · B = a − 3. Q.E.D.
If |a1 − a2 | ≤ 1 for every line then Van de Ven’s result on uniform bundles
implies that ξ = O ⊕ O or O ⊕ O(1), giving the varieties indicated.
Step 3 Now suppose that none of the statements (a–d) at the beginning of
this section hold for V .
Then for every H ∈ |−KV |, and every divisor D with 0 < D < H, |D|
is free with D3 > 0. Let Γ be of minimal degree among such divisors, and
let r be the maximum integer such that H 0 (OV (−KV − rΓ) 6= 0. Then the
hypotheses of the next result are all satisfied.
Theorem Let V be a 3-fold, −KV = rΓ + F , with r ≥ 1, and suppose that
every Γ ∈ |Γ| is irreducible, |Γ| is free with Γ3 > 0 and H 0 (OV (Γ − F )) =
H 0 (OV (F − Γ)) = 0.
Then either r ≤ 2, Γ3 = 1, ϕΓ : V → P3 , or r = 1, Γ3 = 2, ϕΓ : V →
Q ⊂ P4 , with Q a nonsingular quadric; in either case ϕΓ is a small birational
morphism.
Furthermore, if V is a Fano 3-fold, then ϕD is a blowup of P3 or Q in a
nonsingular (possibly disconnected) curve.
Remark 3 A birational morphism ϕ : X → Y is small if dim ϕ−1 ≤ 1 for
every P ∈ Y . Recall that [Danilov] has proved that if X and Y are smooth and
projective, then every small birational morphism is a composite of blowups
in smooth codimension 2 centres.
Proof Let X ∈ |Γ + F | = |−KV − (r − 1)Γ| be a general element; X is a
nonsingular surface and −KX ∼ (r − 1)Γ|X .
Now we have exact sequences
0 = H 0 (OV (Γ − F )) → H 0 (OX (Γ − F )) → H 1 (OV (−2F )) = 0
0 = H 0 (OV (F − Γ)) → H 0 (OX (F − Γ)) → H 1 (OV (−2Γ)) = 0,
where the H 1 vanish because each of |2F | and |2Γ| contain irreducible divisors.
8
Now H 0 (OX (F − Γ)) = 0 certainly implies that H 0 (OX (F − rΓ)) = 0, so
that by Serre duality on X we get H 2 (X, OX (Γ − F )) = 0. Thus Riemann–
Roch on X gives
χ(OX ) + 12 (Γ − F )(rΓ − F )X = χ(OX (Γ − F )) ≤ 0.
(+)
Now let C = Γ1 ∩ Γ2 and D = F1 ∩ F2 be the general curve sections, with
Γi ∈ |Γ| and Fi ∈ |F | general elements; by hypothesis Γ3 > 0 and F 3 > 0, so
that C and D are irreducible curves, with genus p(C) and p(D) given by the
adjunction formula
2p(C) − 2 = Γ2 −F − (r − 2)Γ
2p(D) − 2 = F 2 (F − rΓ).
Substituting in the inequality (+) gives
2p(C) − 2 + 2(r − 1)D3 + 2p(D) − 2 ≤ −χ(OX ).
If r ≥ 2 then χ(OX ) = 1, and p(C), p(D) ≥ 0 force r = 2, Γ3 = 1.
If r = 1 then X ∈ |−KV | is a K3 surface, so that χ(OX ) = 2, and
p(C) = p(D) = 0; but H 1 (OV ) = H 1 (OΓ ) = 0, so that
H 0 (OV (Γ)) H 0 (OΓ (Γ)) H 0 (OC (Γ));
this implies that ϕΓ embeds each of the curves C as a rational normal curve.
Hence ϕΓ is birational, and ϕΓ (V ) is a 3-fold of degree n − 2 in Pn . Now
n = 3 or 4, for otherwise we contradict the hypothesis that every Γ ∈ |Γ| is
irreducible. Thus in either case, we have a birational morphism
ϕ Γ : V → P3
or ϕΓ : V → Q ⊂ P4
with Q a nonsingular quadric.
The morphism ϕΓ does not contract any surface S ⊂ V to a point, for
otherwise |Γ − S| =
6 ∅, contradicting the irreducibility of every Γ ∈ Γ. Hence
ϕΓ is small.
The last assertion is obtained by using Danilov’s theorem (it also follows
σ1
from [Sh, 4.12]): let V1 −→
V2 → · · · → Vn be any sequence of blowups with
nonsingular codimension 2 centres. If the centre of σ1 meets the exceptional
locus of σn ◦ · · · ◦ σ2 , it has to meet some line whose proper transform on V1
is a curve C with CKV1 ≥ 0. This proves the theorem. Q.E.D.
Note that ϕΓ cannot be an isomorphism – for then index V ≥ 2; and the
exceptional scroll consists of lines
9
Step 4 (preliminary version) Suppose that V = V2g−2 ⊂ Pg+1 is an intersection of quadrics, and that |S| is a free pencil on V , necessarily of del Pezzo
surfaces. Then
(i) every reducible fibre S ∈ |S| contains a component Γ = Sn−1 ⊂ Pn ;
(ii) if every element of |S| is irreducible, and if the general element has no
lines, then V is one of the following 4 varieties:
1. V = P1 × P1 × P1 ,
2. F1 × P1 (with F1 → P1 defining the pencil),
3. P1 × P2 or
4. F(1, 0, 0) (the blowup of P1 ⊂ P3 ).
The final version will contain a proof of the following result, strengthening
(i) while removing its hypotheses:
Theorem Let f : X → C be a morphism of a smooth 3-fold to a curve, and
suppose that −KX is relatively ample for f . Then every component Γ of a
reducible fibre is a relative Sn−1 ⊂ Pn ; that is, deg Γ = (−KX )2 Γ = n − 1,
and f∗ OX (−KX ) H 0 (OΓ (−KX )) with h0 (OΓ (−KX )) = n + 1.
The proof is based on
e for which −Ke is a “big” positive
(a) a study of nonsingular surfaces Γ
Γ
divisor, and
(b) numerical connectedness of the fibres.
Proof deg S ≤ 9, so that every reducible S contains a component of degree
≤ 4. (i) now follows from the fact that V is an intersection of quadrics.
Now suppose that every S ∈ |S| is irreducible, and that the general element has no lines, that is, S ∼
= P1 × P1 or P2 .
Replacing the given embedding by ϕ−KV +S , I can assume that every S ∈
|S| is linearly normal and is a Sn−1 ⊂ Pn embedded by −KS . The cone is
excluded by the nonsingularity of V . P1 × P1 and F1 cannot fit together in a
smooth family1 since they are not diffeomorphic (alternatively, argue on the
deformation of l ⊂ F1 ). Thus every S ∈ |S| is P1 × P1 or P2 .
1
Erratum, added in 1982: The treatment given here does not consider the possibility
that |S| is a pencil whose general member is P1 ×P1 , but which has singular fibres isomorphic
to the quadric of rank 3 Q ⊂ P3 . In this case X is a quadric bundle over P1 ; I have not as
yet checked that every Fano 3-fold having such a pencil satisfies one of (i–iv) of Lemma 2.2.
10
Lemma If f : X → C is a morphism to a curve, with every fibre ∼
= P1 × P1 ,
2
and if H (X, OX ) = 0 then X ∼
= P(ξ1 )×C P(ξ2 ), with ξi rank 2 vector bundles.
If every fibre is Pn then X ∼
= P(ξ), with rank ξ = n + 1.
Proof The Pn case is standard, and the usual proof can easily be modified
to give the P1 × P1 case.
Now one sees easily that the 4 listed varieties are the only Fano 3-folds of
this type. For example, if B1 ⊂ Fa1 and B2 ⊂ Fa2 are the 2 negative sections,
then
l ⊂ Fa1 ×P1 Fa1 , NFa1 ×P1 Fa1 |l ∼
= O(−a1 ) ⊕ O(−a2 ),
and −K · l = 2 − a1 − a2 > 0. Compare [D].
2
Proof of Lemma 2.50
Recall that we are assuming that −KW is very ample, that W contains a
surfaces S = Veronese surface or quartic scroll, and that the following holds:
|−KV | 63 D1 + D2 + nS,
with Di > 0 and S 6⊂ Di .
(irred mod S)
Write |D| = |−KW − S|.
Step 1 dim |D| ≥ 1.
Proof W = W2g−2 ⊂ Pg+1 is not a complete intersection, since S does not
generate Pic W ; hence by [Isk, Part I, 1.3], g ≥ 6. But hSi = Pn with n ≥ 5,
so that hyperplanes through S form at least a pencil.
Step 2 |D| is free and irreducible.
Let C ⊂ X ⊂ W be general curve and K3 sections. Since S and S ∩ X
are irreducible, H 1 (OW (−S)) = H 1 (OX (−S)) = 0, so that
H 0 (OW (D)) H 0 (OX (D)) H 0 (OC (D)).
By (irred mod S), the only possible fixed component of |D| is S itself. But
by the Exercise in Section 1, Step 1, or [Sh, 3.23], S can only be fixed if
4 = deg S = g − 1, which is absurd. Thus |D| has no fixed components, and
by arguing on the hyperplane section X, it is free; it is irreducible by (irred
mod S).
Step 3 Set |Dr | = |−KW − rS|, so that D = D1 ; if |Dr | is free for some
r ≥ 1 then dim ϕDr (W ) = 3.
11
Proof OS (−KW −rS) = OS (−KS −(r −1)S), so that the restriction of |Dr |
to S is a free linear subsystem of a very ample linear system on S. If ϕDr (W )
is a surface, it is necessarily this embedding of S, but then |Dr | decomposes
as a sum of mobile divisors; this contradicts (irred mod S).
Plan of proof
Before proceeding, I now outline the proof of Lemma 2.50 :
X ∈ |−KW | = |H| and Y ∈ |−KW − S| = |D|
will be nonsingular surfaces. In Step 4, I will show that dim |E| ≥ 2, where
|E| = |−KW − 2S|. Note that OY (KY ) = OY (−S), and that OY (E) =
OY (DY + KY ), where DY is the divisor class on Y corresponding to OY (D).
Furthermore, because H 1 (OW (−S)) = 0,
H 0 (OW (E)) H 0 (OY (E));
thus the fact that |E| is free at every point of Y will follow by the standard
Bombieri–Ramanujam method once we know that every element of |DY | is
numerically 2-connected. This occupies Step 5. Finally, we have to show that
a nonsingular Y ∈ |D| can be found passing through every P ∈ W .
Step 4 ϕD is birational and
(
16 if S = Veronese surface
g≥
15 if S = scroll F0;2 .
Proof Since dim hSi ≤ 5, H 0 (OW (D)) ≥ g − 4; thus W = ϕD (W ) is a
3-fold spanning Pn , with n ≥ g − 5, and hence deg W ≥ g − 7.
Write H = −KV ; from the multiplication table
(
1
H = 3 = 2g − 2; H 2 S = 4; HS 2 = 2; S 3 =
0
we get
(
2g − 21 if S = Veronese surface
deg ϕD deg W = D3 = (H − S)3 =
2g − 20 if S = scroll F0;2 .
Thus obviously ϕD = 1 and D3 ≥ g − 7 gives G ≥ 14 or 13 in the two cases.
We will now exclude deg W = g −7 and g −6. This increases the preceding
bound by 2, and proves the assertion. If deg W = g − 7 then W mus be a
12
scroll whose hyperplane section decomposes as a sum involving ≥ 2 mobile
divisors, contradicting (irred mod S). Thus now g ≥ 14 in either case.
If deg W = g − 6 with W spanning Pg−5 , and if P ∈ Sing W , then the projection of W from P is a scroll of degree ≥ 6, so that again W has hyperplane
sections decomposing as a sum of ≥ 2 mobile divisors, contradicting (irred
mod S). Thus W is nonsingular; since H 0 (OW (1)) is complete, it is easy to
check that the same is true of the hyperplane sections of W , and hence that
W is a Fano 3-fold of index 2. According to [Isk, Part I, 4.2, (ii)], these only
exist if deg W ≤ 7; this contradicts the above bound. Q.E.D.
Corollary dim |E| ≥ 1.
Proof ϕD (W ) = W spans Pn , with n ≥ g −4. On the other hand, OS (D) =
OS (−KS ), so that
(
9 if S = Veronese surface,
dim hϕD (S)i ≤
8 if S = scroll F0;2 ;
it follows that hyperplanes through ϕD (S) form at least a pencil. Q.E.D.
Step 5 By (irred mod S), the fixed part of |E| can be at worst a multiple
of S, so that |E − kS| = |−KW − (k + 2)S| has no fixed components for some
k ≥ 0. Now choose Y ∈ |D| to be nonsingular, and not to contain any of the
base curves of |E − kS|.
Claim OW (E) is generated by its global sections H 0 (OW (E)) at every point
P ∈Y.
Proof As above,
H 0 (OW (E)) H 0 (OY (E)) = H 0 (OY (DY + KY )).
Now by Bombieri [B, ??], to check that OY (D + KY ) is generated by its
sections at P ∈ Y , it is sufficient to have
(i) h0 (OY (D)) ≥ 3; (this is obvious).
(ii) Every divisor in |D| is numerically 2-connected.
I will drop the suffix Y on the divisor classes HY , SY , etc. duirng the rest
of this step, which is devoted exclusively to proving (ii).
Note first the following:
(1) H is very ample on V ;
13
(2) |D| = |H − S| is free, and D2 ≥ 8 (since ϕD is birational and because
of the bounds in Step 4);
(3) |H − (k + 2)S| has no fixed components for some k ≥ 0 (this follows by
choice of Y );
(4) KY = −S;
(5) S 3 = −3, HS = 3.
Now on any surface Y , |D| is free and D2 > 0 always implies that D is at
least 1-connnected: if D = E1 + E2 with Ei > 0 and E1 E2 ≤ 0, then DEi ≥ 0
gives
E12 ≥ −E1 E2 ≥ 0
E22 ≥ −E1 E2 ≥ 0.
But by the index theorem,
2
E1 E1 E2 ≤ 0,
det E1 E2 E22 with equality only if E1 ≡ qE2 (with q ∈ Q, q > 0; here ≡ denotes numerical
equivalence). This would contradict D2 > 0.
Now suppose that by some evil chance E1 E2 = 1, and (say) E12 ≤ E22 ; then
by the index theorem, and D2 ≥ 5, we get E12 ≤ 0. Since DE1 = E12 +E1 E2 ≥
0, there are 2 cases:
either: E12 = −1, DE1 = 0; this is impossible, because HE1 > 0 and
(H − S)E1 = 0 implies (H − (k + 2)S)E1 < 0, contradicting (3) above,
or: E12 = 0, DE1 = 1; then
(H − S)E1 = 1 and (H − (k + 2)S)E1 ≥ 0
give
(k + 1)SE1 ≤ 1 and HE1 = SE1 + 1 ≤ 2.
Now if S ≤ E1 , HE1 = H(E1 − S) + HS ≥ 3; hence S 6⊂ E1 , and so
SE1 ≥ 0.
Now because E12 = 0, KY · E1 = −SE1 ≡ 0 mod 2, and so necessarily
SE1 = 0; this gives HE1 = 1, and because H is very ample, E1 must be
an irreducible line. On the other hand, 2pa (E1 ) − 2 = E12 + KY E1 = 0,
which is absurd. This proves that every D ∈ |D| is 2-connected.
14
Step 6 |D| = |−KW − S|, |E|, etc., now revert to being linear systems on
W ; let DX , EX and so on denote their restrictions to X. If X ∈ |−KW | then
as usual H 0 (OW (D)) H 0 (OX (D)).
Now we know from Step 5 that |E| has no fixed components, so that the
complete linear system |EX | is free.2 On the other hand, |EX + 2SX | = |HX |
is ample, and it follows that the free system |DX | = |HX − SX | is also ample
(if Γ ⊂ X has HX · Γ > 0, DX · Γ = 0 then Ex · Γ < 0, contradicting the fact
that |EX | is free).
From Step 4 we know that ϕD |X = ϕDX is birational on X, and it follows
from [S-D, §6, p. 623] that |DX | is very ample on X; it is elementary to see
that |D| is then very ample on W . We can then choose a nonsingular Y ∈ |D|
through any point P , so that |E| is free on W . This completes the proof of
Lemma 2.50 .
3
Proof of Lemma 2.3–2.4
I assume throughout that V = V2g−2 ⊂ Pg+1 is a 3-fold embedded by −KV ,
and is an intersection of quadrics. I can assume that g ≥ 6, since the lines
and conics on the complete intersections are well understood.
Fix the following assumptions and notation:
(A) P ∈ V is a point not on any line; write σ : W → v for the blowup
of P ∈ V . Then S = σ −1 P = Veronese surface, −KW = σ ∗ (−KV ) − 2S,
(−KW )3 = 2g 0 − 2, where g 0 = g − 4 ≥ 2.
(B) q ⊂ V is a conic with NV |q ∼
= OP1 ⊕ OP1 or OP1 (1) ⊕ OP1 (−1); it
follows [Isk, Part II, 4.3] that q is not contained in a plane or quadric surface
of V . Write σ : W → V for the blowup of q ⊂ V . Then S = σ −1 q = quartic
scroll F0;2 or F2;1 , −KW = σ ∗ (−KV ) − S, (−KW )3 = 2g 0 − 2 where g 0 = g − 3.
2
Addendum, dating from 1982. The argument given here is over-hasty. The following is
more complete: since S is the only possible fixed component of |E|, and S ∩ Y = SY 6= 0,
we know from Step 5 that |E| has no fixed components on W . I thus choose X ∈ |−KW |
such that
(a) P ∈ X;
(b) X is nonsingular;
(c) X does not contain any base curve of |E|, and
(d) ϕ−KW −S |X is birational.
Now |EX | ⊃ |E|X has no fixed components, and because |EX + 2SX | is ample, it follows
that also |EX +SX | is ample (because for Γ ⊂ X, 2(EX +SX )·Γ = EX Γ+(EX +2SX )Γ > 0).
But then because of (d) and [S-D, §6, p. 623], DX is very ample on X, and we can find
Y ∈ |D| such that Y is nonsingular and passes through P . This shows that |E| is free, and
completes the proof of Lemma 2.50 .
15
For C ⊂ W a curve not in S, write σ(C) = Γ ⊂ V .
Lemma 2.3–2.4 will be a consequence of the following more precise result of Fano–Iskovskikh type; compare [Isk, Part II, §5]. The deduction of
Lemma 2.3–2.4 requires some slightly tricky arguments in the cases g = 6
or 7 (Steps 6 ffoll. below).
Theorem |−KW | is free and defines a generically finite morphism ϕ−KW ;
π
write ϕ : W → W for the Stein factorisation, ϕ : W → W −→ ϕ−KW (W ) ⊂
0
Pg +1 . Then ϕ is a small birational morphism, W is Gorenstein with ample
−KW and ϕ∗ (−KW ) = −KW . We have H i (OW ) = 0 for i = 1, 2 and 3.
Suppose in addition that |−KV | 63 D1 +D2 with Di > 0. Then for C ⊂ W
a curve not in S and with C · (−KW ) = 0, we have
(A) if g ≥ 8 then Γ is a conic through P ;
(B) if g ≥ 7 then Γ is a line meeting q.
In cases g ≤ 7 we may also have a finite number of curves C ⊂ W of the
following types:
(A) g ≤ 7: C ∼
= P1 , with deg Γ = 4, multP Γ = 2, hΓi = P3 , pa Γ = 1.
g = 6: C ∼
= P1 , with deg Γ = 6, multP Γ = 3, hΓi = P4 , pa Γ = 2.
(B) g = 6: Γ is a conic such that hΓ ∪ qi = P3 , pa (Γ ∪ q) = 1.
Remark 1 there are only finitely many conics through P if g ≥ 9 by [Isk,
Part II, 4.4, (iii)], and similarly, finitely many lines meeting q if g ≥ 8.
Remark 2 The fact that general projections contract finitely many rational
curves to double points occurs throughout Fano’s work. In special cases one is
allowed to assume that NW |C ∼
= OP1 (−1) ⊕ OP1 (−1) for the contracted curves
C, which ensures that W has ordinary double points (xy = zt); singularities
obtained by contracting curves C with NW |C ∼
= OP1 ⊕ OP1 (−2) include xy =
z 2 + t2n for all n ≥ 2. The significance of these singularities in providing
elementary transformations in the birational geometry of 3-folds is beginning
to be appreciated.
Step 1 |−KW | is free and ϕ−KW is generically finite; the general element
X ∈ |−KW | is a K3 surface, and H 0 (OW (−KW )) H 0 (OX (−KW )). There
is no surface F ⊂ W for which ϕ−KW (F ) = pt.
Proof I consider (A) only. (B) requires only trivial modifications.
Write TP V = P3 for the tangent space to V at P ; then ϕ−KW will be
interpreted as the linear projection of V from TP V . We have σ∗ OW (−KW ) =
m2P · OV (−KV ), where mP is the maximal ideal of P , so that
|−KW | = σ ∗ div(s) − 2S s ∈ H 0 (m2P · OV (−KV )) .
16
By definition of the blowup mrP · OW = OW (−rS) for r ≥ 1, so that |−KW |
is free3 if and only if m2P · OV (−KV ) is generated by H 0 (m2P · OV (−KV ));
geometrically, this is the assertion that TP ∩ V = Z, where Z ⊂ V is the
subscheme defined by m2P ; this in turn follows from the fact that V is an
intersection of quadrics, and that P is not on any line: indeed, TP V ∩ V is an
intersection in TP V = P3 of quadrics singular at P . This proves that |−KW |
is free; (−KW )3 > 0, so that ϕ−KW has 3-dimensional image. If F ⊂ W is
a surface contracted to a point by ϕ−KW , F 6= s, then hσ(F ) ∪ TP V i = P4 ;
but then σ(F ) ∩ Tp V would have dimension ≥ 1, contradicting TP V ∩ V = Z.
The remaining assertions are standard.
Step 2 I now discuss −KW .
By construction of ϕ, W is normal, and OW (−KW ) = ϕ∗ OW (1) for some
0
0
'
ample sheaf OW (1) on W . Now write ϕ : W 0 −→ W for an open set W ⊂ W
0
with codim(W \ W ) ≥ 2; then Ω3W ∼
= Ω3W when restricted to this open.
But if follows from the description of OW (KW ) as a divisorial sheaf (see [R,
Appendix to §1] that OW (KW ) ∼
= OW (−1).
∗
∼
Now since ϕ OW (KW ) = OW (KW ), it follows from the projection formula
and the vanishing Ri ϕ∗ OW (KW ) = 0 of [G–R] that Ri ϕ∗ OW = 0 for i = 1, 2
and 3. Thus W has rational singularities, and is hence Gorenstein. H i (OW ) =
0 follows from the Leray spectral sequence.
Step 3 Let C ⊂ W be any curve not in S and contracted by ϕ−KW , and
suppose CS ≥ 2. Then the intersection C ∩ S is part of the singular locus of
ϕ−KW |S .
Proof Z = C ∩ S is a subscheme of S with deg Z = length(OZ ) = CS ≥ 2,
and such that
H 0 (OW (−KW )) → H 0 (OZ (−KW ))
3
Erratum dating from 1982: G. Horrocks has kindly pointed out that this section contains two errors, that fortunately cancel out. A correct argument is as follows:
From the assumption that V is an intersection of quadrics, and that P is not on any line
of V , we see that TP V ∩ V = {P }; indeed, this is an intersection if TP V = P3 of quadrics
singular at P . Now in affine coordinates at P ∈ Pg+1 , and quadric Q with V ⊂ Pg+1 .
Now in affine coordinates at P ∈ Pg+1 , any quadic Q with V ⊂ Q ⊂ Pg+1 has
equation q = q1 + q2 with qi homogeneous of degree i; q1 defines TP Q, and because
TP V = bigcapTP Q, these q1 give the whole of H 0 (m2P · OV (−KV )).
Choosing a local generator of OV (−KV ), and s ∈ H 0 (m2P · OV (−KV )) can be considered
as an element of m2P , with leading term s2 = s ∈ m2 /m3 . Obviously |−KW | is free if and
only if the s2 define a collection of quadric cones in TP V = A3 whose intersection if {P },
correspodning to the empty set in P(TP V ) = S. But this is clear: if s = q1 then s2 = q2 ,
and the s2 define TP V ∩ V = {q1 = q2 = 0 for all V ⊂ Q ⊂ Pg+1 } = {P }
17
has rank 1. Q.E.D.
Step 4 Assume that |−KV | 63 D1 + D2 with Di > 0. Then the restriction
ρ : H 0 (OW (−KW )) → H 0 (OS (−KW )) has maximal rank.4
Proof I suppose that ρ is not injective, and D ∈ |−KW − S|; it is enough
to show that H 1 (OW (D)) = 0. By the hypothesis on |−KV |, D = rS + E for
some r ≥ 0, with E 6= 0 irreducible.
Now let X ∈ |−KW | be a K3 surface meeting E and S in irreducible
curves EX , SX . The exact commutative diagram
k
T
=
H 0 (OW (−KW )) →
k
T
H 0 (OS (−KW ))
↓
→
H 1 (OW (D))
→ 0
↓
H 0 (OX (−KW )) → H 0 (OSX (−KW )) → H 1 (OX (DX )) → 0
shows that H 1 (OW (D)) ∼
= H 1 (OX (DX )). But DX is obviously numerically
connected:
2
DX = EX + rSX , HX · SX = 4, SX
= −2,
so that for b ≥ 1,
(EX + (r − b)SX )bSX = b(HX − (b + 1)SX )SX > 0.
Thus h0 (ODX ) = 1, H 1 (OX (−DX )) = 0, and by Serre duality we have
H 1 (OX (DX )) = 0. Q.E.D.
Step 5 To prove the theorem, I now analyse the curves C ⊂ W with
C · (−KW ) = 0; I start with case (A).
If g ≥ 8, then h0 (OW (−KW )) = g − 2 ≥ 6 = h0 (OS (−KW )). By Step 4,
ϕ−KW embeds S in P5 as the Veronese surface. According to Step 3 the only
curves C ⊂ W contracted by ϕ−KW have CS ≤ 1. Then
0 = C · (−KW ) = C · (σ ∗ (−KV ) − 2S) = deg Γ − 2CS,
(†)
where CS = multP Γ; this is only possible for Γ a conic through P .
Now if C is a contracted curve we have deg Γ = 2 multP Γ from (†); and
from the description of ϕ−KW in terms of projections of V from TP V , we must
have hΓ ∪ TP V i = P4 .
Now V ∩ P4 is an intersection of quadrics, and all its components are of
dimension ≤ 1. There are only the following possibilities for dim Γ = 1:
4
Mori says that this part is wrong.
18
hΓi = P2 , Γ a conic;
hΓi = P3 , deg Γ = 4, Γ an intersection of 2 quadrics;
hΓi = P4 , deg Γ = 6, Γ a component of an intersection of quadrics;
hΓi = P4 , deg Γ = 8, and Γ the complete intersection of 3 quadrics.
The final case is impossible; for on the one hand by the adjunction formula
in P4 , ωΓ = OΓ (1); but on the other hand, Γ is a component of a scheme Z
which is a curve section of V , and also has ωZ = OZ (1). Since g ≥ 6,
deg Z = 2g − 2 > 8, and Z has another component meeting Γ. This is
impossible.
It is easy to see that in each case the blowup C of P ∈ Γ is rational and
nonsingular; the fact that any curve Γ with hΓi = P4 , deg Γ = 6, multP Γ = 3
has pa Γ = 2 will not be used, and is left as a treat for the reader.
The final assertion is the finiteness of the number of contracted curves.
For this we use Step 3, and the geometry of ϕ−KW |S : S → S # ⊂ Pg−3 ; this
is the famous theory of projections of the Veronese [S–R, p. 128]. If g = 7
then either S → S # ⊂ P4 is an embedding, or S # has a line of singularities,
corresponding to a projection from a vertex lying in the plane of some conic.
If g = 6 then S → S # ⊂ P3 is the Steiner surface, with 3 double lines meeting
in a triple point.
By Step 3, any of the exceptional contracted curves map to singularities
of S # ; the fact that there are now only finitely many of these now follows: if
F ⊂ W is a surface contracting to a line under ϕ−KW then |−KW − F | =
6 0
contradicting |−KV | 63 D1 + D2 with Di > 0.
(B) is exactly similar; for the final finiteness assertion I need to know that
a 1-dimensional component of the singular locus of ϕ−KW : S → S # ⊂ P4 is
a line. This follows for example by arguing on the arithmetic genus of the
general section of S # .
This complete the proof of the theorem of this section. I now embark on
the proof of Lemma 2.3–2.4.
Step 6 If ϕ−KW is finite then W = W , and −KW is ample; W cannot be a
hyperelliptic Fano 3-fold, since then by [Isk, Part I, 4.2] we would have either
Pic W ∼
= Z or |−KW | 3 D1 + D2 with Di > 0.
In (A) when g ≥ 8, the obstruction to the finiteness of ϕ−KW consists of
the conics through P ; in (B) when g ≥ 7, it consists of the lines meeting q.
This proves Lemma 2.3–2.4 in these cases.
I now outline the proof of Lemma 2.3–2.4 in the cases g = 6 or 7. In the
diagram
we have the following information:
19
(i) There are finitely many curves Ci ⊂ W with ϕ(Ci ) = Pi ∈ W , and ϕ is
'
an isomorphism W \ {Ci } −→ W \ {Pi }. Every point Pi is a singular
point of S and of W :
{Pi } = Sing S = Sing W .
(ii) ϕ−KW is either an embedding of W as a quartic in P4 or a finite double
cover of a nonsingular quadric Q ⊂ P4 or P3 ; it is easy to see that the
proof of [Isk] carries over to singular W once we know tht |−KW | is
free. The fact that the singular quadric cannot turn up follows from
the irreducibility assumption on |−KV |.
0
(iii) The composite π : S → S # ⊂ Pg +1 is the linear projection of S ⊂ P5 ,
where S is the Veronese surface or the scroll F0;2 .
After dividing into cases, I will in each case either prove that ϕ is finite, or
deduce a contradiction directly. It is convenient to state first the properties
of the projections to P4 of F0;2 ⊂ P5 which I will need. Recall that F0;2 is
P1 × P1 embedded by |2A + B|; a ∈ |A| is a line of the ruling, and b ∈ |B| a
conic.
Proposition Let S = F0;2 ⊂ P5 , and consider the projection π : S → S # ⊂
P4 from a point O ∈ P5 not in S. Then either (i) or (ii) below holds:
(i) O lies in the plane hbi = P2 of some conic b ∈ |B|; π : S → S # maps b
two-to-one onto a double line l ⊂ S # , l = Sing S # .
(ii) O lies on a unique secant or tangent line of S. The surface S # has one
double point P ∈ S # , and the projectivised tangent cone Z = P(TP S # )
is either a pair of skew lines, or a double line not lying in any plane P2 .
In particular Z is not contained in any quadric of rank 3.
Proof Every O ∈ P5 lies on at least one secant line l of S; l ∩ S contains a
finite scheme L ⊂ l ∩ S of degree 2 (in fact L = l ∩ S), and L is not contained
in any a ∈ |A|, since then l = a ⊂ S. There are thus two cases (see Figure 1)
(i) L ⊂ b for some b ∈ |B|;
Figure 1: (i) L ⊂ b or (ii) L 6⊂ b.
20
(ii) L 6⊂ b.
There is just one of each of these cases up to projective equivalent, and it is
elementary to give the parametrisation P1 × P1 → S # and to verify the above
assertions in each case. The final exotic case goes as follows: in coordinates
(x, y) × (s, t) on P1 × P1 , let L be the tangent vector at P = (1, 0) × (1, 0)
to ys = xt. The parametrisation of S # is (xs2 , xst − ys2 , xt2 , yst, yt2 ). In
inhomogeneous coordinates at P this can be written (1, t − y, t2 , yt, yt2 ), and
the equations of S # can be written
u1 u2 u3 u4 u2 u21 u4 u1 u3 = 0
2
(that is, set to zero the 2 × 2 minors), where
u1 = t2 ,
u2 = t3 ,
u3 = t − y,
u4 = t2 − ty.
The tangent cone Z is given by
u 1 u4 = u2 u3
and u22 = u2 u4 = u24 = 0,
which is a (2, 0)-divisor on the nonsingular quadric Q ⊂ P3 given by u1 u4 =
u2 u3 (see Figure 2). Hence Z is not contained in a plane. Finally, every line
Figure 2: The subscheme Z in the quadric Q : u1 u4 = u2 u3 .
pair or double line in a quadric of rank 3 of P3 is contained in a plane. Q.E.D.
0
Step 7 I now deduce a contradiction assuming that ϕ−KW : W → Pg +1 is
a finite double cover of P3 or of a nonsingular quadric Q ⊂ P4 .
In either case S # is a complete intersection, and necessarily has a double
line l. I show that the double cover splits over l; this contradicts the fact that
in S → S # the double line unzips into an irreducible conic.
Note that W , and hence the branch locus B, has only finitely many singularities; B is thus irreducible, and does not contain S # . Since S → S #
has degree 1, the double cover splits over S # , and so B lifts back to twice
an effective divisor on S. B cannot contain l, because then it would have to
contain the tangent cone to S # at a general point of l, and thus be singular
along l. Hence B touches l.
21
Step 8 I now assume that ϕ−KW : W ,→ P4 .
In this case S = S # , so that S # has only finitely many singularities. In
case (A), S # is then necessarily the nonsingular projection of the Veronese,
'
so that by Step 3 ϕ : W −→ W , and we are home.
There remains to eliminate the case that S # is one of the projections
of F0;2 of (ii) of the Proposition in Step 6. I will show (following an idea
suggested by Fabrizio Catanese) that in this case the only singularity of W is
an ordinary double point. We will then get a contradiction by the next easy
result.
Lemma 2.50 Let X ⊂ P4 be a hypersurface of degree ≥ 3 having an ordinary
double point P ∈ X as its only singularity. Then every surface S ⊂ X is the
intersection of X with a hypersurface.
Proof Let σ : F → P4 be the blowup of P ∈ Sing X; Then σ ∗ X = σ 0 X +2B,
where B is the exceptional component, and σ 0 X is a smooth very ample
divisor on F . By the Lefschetz theorem, Pic σ 0 X ∼
= Pic F . Every irreducible
0
0
divisor on s X is then the intersection of σ X with an irreducible divisor on
F in the class B of σ ∗ (aH − bB) with a ≥ b, as follows from the fact that
H 1 (F, L) for all L ∈ Pic F .
The singularity P ∈ W is also a singular point of S, and the tangent
cont to W at P contains the tanget cone to S at P ; thus it is certainly not
a quadric of rank 3! It will therefore be enouth to prove that the general
hyperplane section through P ∈ W is a surface X having an ordinary double
point.
Claim Let q and Γ be as in the exception case for (B), g = 6 in the theorem,
that is, hΓ ∪ qi = P3 , pa (Γ ∪ q) = 1. then the general hyperplane section
X ∈ |−KV | passing through q ∪ Γ is nonsingular.
This will clearly do what we want. When we blow up Q, σ : W → V , we
get σ ∗ X = S + σ 0 X, where σ 0 X is isomorphic to X, and maps to X under
ϕ; the singularity of X can thus be resolved by introducing just one curve
C∼
= P1 , and this is well known to characterise ordinary double points.
The claim is proved by an easy dimension count: hyperplanes of Pg+1
containing hΓ ∪ qi = P3 form an ∞g−3 , and one checks that the hyperplanes
containing the tangent space to V at one of the sngularities of Γ∪q,a nd those
containing the tangent space to V at a general point of Γ or q are ∞g−4 .
This complete the proof of Lemma 2.3–2.4.
Remark There is another beautiful idea in [Sh]: if V is a Fano 3-fold, then
the splitting of the curve sections of V also has strong consequences for V .
22
For example, if g ≥ 5 and a curve section splits as Γ ∪ Γ0 with deg Γ = 2 or
3, and pa Γ = 1, then V is necessarily hyperelliptic or trigonal.
Another example: if g = 6 then curve sections of V10 ⊂ P7 might split as
Γ ∪ Γ0 , where
deg Γ = 4, hΓi = P3 , pa Γ = 1
deg Γ0 = 6, hΓi = P4 , pa Γ = 2.
Now if there exists one such splitting there will exist many – in fact Γ will
move in an elliptic pencil, and Γ0 in a 2-dimensional linear system on every
K3 section of V . This is a proerty that V inherits from the geometric form
of Riemann–Roch on its general curve section.
Shokurov uses this idea [Sh, 6.15 and 7.8] to show that if the contracted
curves Γ in the theorem of this section exist, then either they meet lines or
conics of V , or the projection V 99K P3 from a general such curve will be a
finite double cover, leading to a contradiction as usual.
One might be tempted to think (in analogy with [R2]) that then necessarily V has a pencil of surfaces cutting out Γ on its K3 sections. This is not
the case, since one can obviously construct a nonsingular V = Q ∩ H1 ∩ H2 ∩
Grass(2, 5) having Pic V = Z, and mpose some elliptic curve of degree 4 in
Grass(2, 5) on V . This incidentally shows that there exist V with Pic V = Z,
but for which every K3 section X has rank Pic V ≥ 2. I have some reason
for believing that this happens for every V in Iskovskikh’s terminal family of
Fano 3-folds with g = 12 [Isk, Part II, §6].
References
[Sh]
V.V. Shokurov, The existence of lines on Fano 3-folds, Izv. Akad. Nauk
SSSR, Ser. Mat. 43 (1979) 922–964
[Sh2] V.V. Shokurov, The smoothness of the general anticanonical divisor on
a Fano variety, Izv. Akad. Nauk SSSR Ser. Mat. 43 (1979) 430–441 =
Math USSS Izvestiya 14 (1980)
[Isk] V.A. Iskovskikh, Fano threefolds I, II and III, Izv. Akad. Nauk SSSR
Ser. Mat. 41 (1977) 516–562; 42 (1978) 506–549; Double projection
from a line onto Fano threefolds of the first kind, Mat. Sb. 180 (1989)
260–278 = Math. USSR-Sb. 66 (1990) 265–284
[Isk2] V.A. Iskovskikh, Anticanonical models of algebraic 3-folds, Contemporary problems in Math, Itogi nauki i tekhniki. Translated in J. Soviet
Math., Consultants Bureau, N.Y.
23
[S-D] B. Saint-Donat, Projective models of K3 surfaces, Amer. J. Math 96
(1974) 602–639
[G–R] H. Grauert and O. Riemenschneider, Verschwindungssätze für analytischen Kohomologiegruppen auf komplexen Räumen, Invent. Math
11 (1970) 263–292
[S1]
V.G. Sarkisov, Birational automorphisms of conic bundles, Izv. Akad.
Nauk SSSR, Ser. Mat. 44 (1980) 918–945 = Math. USSR Izv. 17 (1981)
355–390
[S2]
V.G. Sarkisov, On the structure of conic bundles, Izv. AN SSSR, Ser.
Math. 46 (1982) 371–408 = Math USSR Izvestiya = Math. USSR Izv.
20 (1982) 355–390
[D]
I. V. Dëmin, Fano threefolds that can be represented as P1 -bundles
over surfaces, Izv. Akad. Nauk SSSR, Ser. Mat. 44 (1980) 963–971
[VdV] A. Van de Ven, On uniform vector bundles, Math. Ann. 195 (1972)
245–248
[M]
S. Mori, Threefolds whose canonical bundles are not numerically effective, Ann. of Math. 116 (1982) 133–176
[S–R] J. Semple and L. Roth, Introduction to algebraic geometry, Oxford,
1949
[R]
M. Reid, Canonical 3-folds, in Journées de géométrie algébrique d’Angers, A. Beauville (ed.), Sijthoff and Noordhoff, Alphen (1980) 273–310
[L]
Henry B. Laufer, On CP1 as an exceptional set, in Recent developments
in several complex variables (Princeton, 1979), Ann. of Math. Stud.
100, Princeton Univ. Press, Princeton, N.J., 1981, pp. 261–275
[R2] M. Reid, Special linear systems on curves lying on a K3 surface, J.
London Math Soc. 13 (1976) 454–458
[Danilov] V.I. Danilov, Decomposition of certain birational morphisms, Izv.
Akad. Nauk SSSR, Ser. Mat. 44 (1980) 465–477
[B]
E. Bombieri, Canonical models of surfaces of general type, Publ. Math
IHES 42 (1973) 171–219
24