Annali di Matematica pura ed applicata (IV), Vol. CLXV (1993), pp. 197-216
2-Spanned Surfaces of Sectional Genus Six(*)...
11 downloads
467 Views
1003KB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
Annali di Matematica pura ed applicata (IV), Vol. CLXV (1993), pp. 197-216
2-Spanned Surfaces of Sectional Genus Six(*). ANTONIO LANTERI
- Un fibrato lineare olomorfo su una superficie algebrica proiettiva complessa non singolare S ~ detto k-spanned se le sue sezioni separano opportuni O-cicli di S di lunghezza k + 1 - - da (k + 1)-uple di punti distinti sino a getti di ordine k - - e forniscono pertanto una immersione proiettiva di S di ordine superiore. Si fornisce un contributo alla classificazione proiettiva delle superfici determinando quelle di genere sezionale 6 immerse da un fibrato 2spanned e quelle di genere sezionale 7 immerse da un fibrato 3-spanned.
Sunto.
Introduction.
The notion of k-spanned line bundle L on a complex projective smooth surface S has been introduced in [BFS] in connection with the concept of higher order embedding. Essentially L is k-spanned if F(L) separates 0-cicles of length k § 1 on S, going from (k + 1)-tuples of distinct points up to jets of order k (for the precise definition see Sect. 0). We say that S is k-spanned to mean that it is embedded via a k-spanned line bundle. Afterwards this notion has been extensively studied in [BS]. In particular, in [BS] k-spanned surfaces (k I> 2) of sectional genus g ~<5 are classified. The aim of this paper is to do the same in case g = 6. The problem looks more complicated than in the previous cases since in some instances to decide whether some admissible surfaces are in fact 2-spanned is not easy and requires non'standard arguments. Here we state the main result of this paper. For the notation we refer to Sect. 0. THEOREM 1. - Let S c P ~ be a complex smooth surface, let L = O~(1) s and assume that g(L) -- 6. The pairs (S, L) such that L is k-spanned, k I> 2, are listed in the Table I except possibly case 8, i n which however L is generically 2-spanned (see (0.6)). Points to blow-up are supposed in general position.
(*) Entrata in Redazione il 14 febbraio 1991, ricevuta versione finale il 20 febbraio 1993. Indirizzo dell'A.: Dipartimento di Matematica ~,F. Enriques~ delrUniversit~, Via C. Saldini 50, 1-20133 Milano, Italy.
198
ANTONIOLANTERI: 2-spanned surfaces of sectional genus s i x TABLE I.
Case
k
S
L
1
5
p2
~ (5)
2
2
elliptic PLbundle with e = - 1
[3~ + f ]
3
3
F0
[3z + 4f]
4
2
elliptic pl-bundle with e = - 1, or 1
[2z + (e + 5) f ]
5
2
Fe, e~<5
[2~+(e+7) f ]
6
2
Del Pezzo surface of degree 5
Ks-2
7
2
Bpl..... p6(Fe), e ~<1
r:*[4a+(2e+5)f]|
+E6] -2
8
2
Bpl .....
pg(Fe), e <~2
zz*[4~r+(2e+6)f]|
... +Eg] -2
9
2
Bp(S'), S' Del Pezzo surface of degree 2
Ks-2 @ 7~rKS~ 1
In particular it turns out there are no 2-spanned surfaces with sectional genus 6 of nonnegative Kodaira dimension. I recall that 2-spanned surfaces are exactly those embedded by a 2-very ample line bundle [BS1]. So Theorem 1 also provides the list of 2-very ample surfaces of sectional genus 6. Here is a sketch of the proof. Since 2-spannedness implies that h~ is large enough, from the Castelnuovo bound for the genus of a space curve we get the inequality d 1> 2g - 2. So, apart from a special case we deal with separately, we have that X is ruled and then adjunction provides a useful tool. Actually, let H = Ks | L; after checking some well known exceptions, we can assume that H is a very ample line bundle. Then we look at the surface S embedded by [HI. Since h ~ = g - q, there are some possibilities according to the values of q, the irregularity of S. The most intricated case is that of rational surfaces in the projective 5-space. However in this case it turns out that the sectional genus g(H) is very low, due to the 2-spannedness Of L. This allows us to keep to a minimum the use of the classification results of surfaces with small sectional genus in finding pairs (X, H), which are candidates for L = H | Ks-1 being 2-spanned, Then the result follows by combining some classification results of LIVORNI[Lil], [Li2], with basic properties of k-spannedness [BS] and some (,ad hoc, 2-spannedness results proven for some special surfaces. I would like to contrast this procedure with the hard job of checking the very long list of surfaces of sectional genus 6 (e.g. see [Li2]). As a consequence of Theorem 1, we get the following
ANTONIO LANTERI: 2-spanned surfaces of sectional genus six
199
COROLLARY. - Let S r be a complex smooth surface of degree d ~< 10, and assume that L = Op~ (1)s is 2-spanned. Then (S, L) is one of the pairs listed in the table II.
TABLE
d
g(L)
4
0
8
9
h o(L)
II.
S
L
6
p2
O~ (2)
1
9
F0
[2~ + 2f]
3
7
Del Pezzo surface of degree 2
Ks-2
5
6
general complete intersection of type (2, 2, 2)
1
10
p2
~2 (3)
The proof is very simple. Actually, apart from the first and the fourth cases in the table above, we can assume that n >I 6 (see (0.4)). So, by the Castelnuovo inequality [GH, p. 252]
g(L) <~[(d - 2)/(n - 2)](d - n + 1 - [(d - n)/(n - 2)]. (n - 2)/2) where [ ] is the greatest integer function, we get the following bound for the sectional genus: g(L) < 6. Since all surfaces in Theorem 1 have degree d/> 12, the assertion follows by checking the list of 2-spanned surfaces of sectional genus ~< 5 in [BS]. The same procedure proving Theorem 1 works also for classifying 3-spanned surfaces of sectional genus 7 and gives THEOREM 2. - There are only two pairs (S, L) where L is a 3-spanned line bundle with g(L) = 7 on a complex projective smooth surface S; they are 1) (F1, [3~ + 6f]) and 2) S = Del Pezzo surface of degree 2, L = Ks-3. In the same way as the above Corollary, it thus follows that (p2, ~p~ (3)) is the unique 3-spanned surface of degree ~< 12. The paper is organized as follows. In Sect. 0 some background material is collected and the k-spannedness of some surfaces occurring in Theorem 1 is proven. Sections 1 and 2 are devoted to the proof of Theorem 1, while Section 3 contains the proof of Theorem 2. I would like to thank M. BELTRAMETTI for many helpful discussions about k-spannedness.
ANTONIO LANTERI: 2-spanned surfaces of sectional genus six
200
0. - B a c k g r o u n d material and 2-spannedness results.
The word surface will always mean complex projective smooth surface; line bundles are holomorphic line bundles. For a line bundle L on a surface we write L ~ instead of L | and L . L to indicate the self-intersection index. All notation and definitions not specified are the standard ones in algebraic geometry. Let X be a surface: ruled means birationaily ruled. If X is a PLbundle over a smooth curve, then ~ and f will denote a fundamental section (i.e. a section attaining the minimal self-intersection) and a fibre of X, and e = - o. z will stand for the invariant [Ha, p. 372]. In particular F~ = (t~l @ Gp1( - e)). Let L be an ample line bundle on X. I recall the names of some classes of pairs (X, L) frequently occurring in the paper; (X, L) is a scroll if X is a P~-bundle over a smooth curve and L .f= 1 for every fibre f of X; assume that (X, L) is not a scroll: then (X, L) is a conic bundle if X is ruled and L . F = 2 for every fibre F of the ruling. (X, L) is a Del Pezzo pair if X is a Del Pezzo surface (i.e. Kx 1 is ample) and L = K:~ 1. By the way recall that Kx'Kx is said the degree of the Del Pezzo surface X. The symbol Bpl ..... p, (X) will denote the surface obtained by blowing-up X at the points Pl, ..., P~. When not specified, = will stand for the blowing-up and E~ for the exceptional curve =-1 (Pi), i = 1, ..., s. In order to insure the ampleness or the very ampleness of some line bundles we consider on the blown-up surface, the general position assumption on the points p~,...p~ is always implicit. A pair (X, L), where X = Bp~..... p~(p2), L = ='Or2 (4) | [El + ... + E~]- 1 and Pl, ..., P~ (s ~< 10) are in general position is said a Bordiga surface. I briefly recall the notion of k-spannedness [BFS], [BS]. Let X be a surface. A 0-cycle on X is called admissible if the Corresponding 0-dimensional subscheme is locally supported on smooth curves of X. A line bundle L on X is said to be k-spanned (k I> 0) if for any admissible 0-cycle ~ of length k + 1 on X, the restriction homomorphism (0.0.1)
F(L) --, F(L | G~)
is surjective. Recall that 0-spanned is equivalent to L being spanned by global sections and 1-spanned is equivalent to being very ample. I also recall that for 2-spanned line bundles (0.0.1) is a surjection for any 0-dimensional subscheme of length 3 regardless any assumption on the locally supporting curves [BFS, (3.1)]. This can be rephrased by saying that 2-spanned line bundles are the same as 2-very ample line bundles (see [BS1]). For the basic properties of k-spannedness I refer to [BS, Sects. 0, 1]. Here I simply recall some facts including a slight improvement with respect to [BS]. Let X be a surface. (0.1) Let L~ be a ki-spanned line bundle on X, i = 1, .... n. Then L1 | ... | L~ is (kl + ... + k.)-spanned. Let L be a line bundle ~on X.
ANTONIO LANTERI: 2-spanned surfaces of sectional genus six
201
(0.2) If L is k-spanned (k I> 2), then L . C 1> k for every smooth rational curve CoX. (0.3) LEMMA. - If L is 2-spanned then L . C ~>4 for every effective divisor C of X
with p~(C) = 1. PROOF. - If C is reducible the assertion follows from (0.2) for k = 2. So we can assume that C is an irreducible plane cubic of (X, L). But then taking a 0-cycle Z cut out on C by a line we see that (0.0.1) is not a surjection, contradicting the 2-spannedness of L. 9 Note that for C a smooth curve, (0.3) is a consequence of[BS, (0.5.2) and (1.4.1)]. The following result proven by BALLmO[Bal], [Ba2] is very important in the sequel. (0.4) If L is k-spanned (k I> 2), then h ~(L) I> k + 5 unless k = 2 and X embedded by ILl is either the Veronese surface or a general complete intersection of three quadrics of p s . k-spannedness results for some special Pezzo surfaces can be found in [BS, Sect. will refer to them throughout the paper, which we need for cases 7, 9 and 8 in the (0.7)].
classes of surfaces like pl-bundles and Del 3 and (0.8)] (see also [BFS, (2.6)]) and we Here we prove two 2-spannedness results Theorem. The first one is inspired by [BS,
(0.5) PROPOSITION. - Let X be one of the following surfaces polarized by the corresponding very ample line bundle H:
i) Bpl ..... p~(Fe), e ~< 1, H = r~*[2z + (e + 3 ) f ] | [ E 1 + ... + E6] - 1 , ii) Bp (Y), where Y is the Del Pezzo surface of degree Ky. Ky = 2 (p a general point), H = rc*Ky 2 | [E] -1 , where, in both cases 7: denotes the blowing up and E = r~-1 (p), Ei = =-1 (Pi). In both cases the line bundle L = H | K~ 1 is 2-spanned. PROOF. - To prove the 2-spannedness of L we use the main result of[BFS]. Assume that L can be written as (0.5.1)
L = Kx | M ,
with M numerically effective and such that M. M/> 13. Then, if L is not 2-spanned X must contain an effective divisor D satisfying the following numerical inequalities: (0.5.2)
M . D - 3 <~D. D < (M.D)/2 < 3.
202
ANTONIO LANTERI: 2 - s p a n n e d s u r f a c e s o f s e c t i o n a l g e n u s s i x
CASE i). Note that X is a Del Pezzo surface. Actually, recalling that F1 = Bq (p2), while Bpl (F 0) = Bql ' q2(p2), we see that X is gotten by blowing up p2 at seven points in general position. Hence K x 1 is ample. Recalling that K x = r : * [ - 2 o - (e + 2 ) f ] | | + ... + E6] we have L = K x e | r:*f. Then (0.5.1) holds .with M = K x s | r:*f. Moreover M is ample, since K x 1 is so and r:*f is spanned; in addition M . M = 9Kx" K x - 6Kx" r:*f = 30.
Assume that L is not 2-spanned; then we get from the last inequality in (0.5.2) 6 > M.D
= -3Kx'D
+ r : * f . D >I - 3 K x ' D
in view of the spannedness of r:*f. Hence - K x ' D < 2 and so, the ampleness of K~/1 implies - K x ' D = 1. F r o m the first equality in (0.5.2) we have (0.5.3)
D . D >I M . D - 3 = - 3Kx" D + r~*f. D - 3 = rc*f" D >t 0 ;
furthermore, by the genus formula we see that (0.5.4)
D - D = 2pa (D) - 2 - Kx" D = 2pa (D) - 1
is odd. Putting together (0.5.3) and (0.5.4) we thus get D - D I> 1. On the other hand, since M . D ~< 5, by (0.5.2), the Hodge index theorem implies that 25 i> (M. D) ~ I> (M. M)(D-D) = 30D-D, giving D .D < 1, a contradiction. CASE ii). As K x = ~ v * K y | with
we have L = K x 2 |
1. Thus (0.5.1) holds
M = K x 8 | r:* K y 1 . Note that r:*Ky 1 is numerically effective and that K x 1 is ample, since p is general; so that M is ample; in addition M . M = 9Kx" K x + 6Kx" r:* K y + K y . K y = 23.
Assume that L is not 2-spanned; then the last inequality in (0.5.2) gives 6 > M.D
= - 3 K x ' D - r : * K y . D >I - 3 K x ' D ,
in view of the numerical effectivity of r:*K:71 . Therefore - K x ' D < 2 and then -Kx'D = 1, as before, since K x 1 is ample. The same argument as in case i) thus shows that D . D I> 1. Then, by the Hodge index theorem, 1 = (Kx" D) 2 >I (Kx" K x ) ( D " D) >I Kx" K x = 1,
so we get [D] - K x 1 (in fact [D] = K~/1, since Pic (X) is torsion free). Then, recalling
ANTONIO LANTERI: 2 - s p a n n e d s u r f a c e s o f s e c t i o n a l g e n u s s i x
203
the first inequality in (0,5.2) we have 1 = K x ' K x = D . D >t M . D
- 3 = 3Kx'D - 3 + r~*K~'.D = =*Ky1.D.
On the other hand r:* K ~ I ' D
a contradiction.
= r~* K y ' K x =
r~* Ky.7~* K y = 2,
[]
Now consider the conic bundle (X, H), where (0.6.0) X = Bpl ..... pg(F~),
e~<2 and H = r~*[2~ + (e + 4) f ] Q [EI + ... + Eg] 1.
The line bundle L =H|
I = =*[4~ + (2e + 6 ) f ] |
[El + ... + E g ] -2
is very ample [Lil, p. 169]. Unfortunately the same argument used in the proof of the above lemma does not work in this case. In fact I have not been able to prove that L is 2-spanned, but only that it is generically 2-spanned, in the sense explained below. (0.6) DEFINITION. Let 2 be a very ample line bundle on a surface S and let p e S. Consider the surface S~ = Bp (S) and let 2~ = o* 2 Q [E]- 1, where ~: S' ~ S denotes the blowing-up and E is the corresponding exceptional curve. 2 is said generically 2-spanned if for every p e S, s is ample and spanned and the morphism associated to F ( ~ ) is birational. -
(0.7) LEMMA. Let things be as in (0.6). If (S, 2) does not contain lines through p, then 2~ is ample and spanned. -
PROOF. First of all 2~ is spanned; this follows from the very ampleness of 2, in view of the bijection between 12~ I and 12 - P l. Actually, were q' e S~ a base point of 12~ I, then letting q = a(q') we would have 12 - P l = 12 - P - q l (q infinitely near to p, ff q' ~ E), contradicting the very ampleness of 2. To prove the ampleness we use the Nakai-Moishezon criterion. First of all note that 2~.2~ = 2" 2 - 1 > 0; otherwise it would be (S, 2) = (p2, Op~ (1)), but this possibility is ruled out by our assumption on (S, 2). Now let C be any irreducible curve in S~ ; then either C = E or C = ~* F - m E , F being an irreducible curve on S, with multiplicity m = F. E / > 0 at p. In the former case 2~. C = 2p" E = 1, while in the latter one we have 2~-C = 2. F - m I> 0, 2~ being numerically effective. Assume that 2~. C-- 0; then F would be an irreducible curve of degree m in (S, 2) having a point of multiplicity m. Then 0 ~
ANTONIO LANTE~I: 2-spanned surfaces of sectional genus six
204
Let things be again as in (0.6) and assume that 2 is 2-spanned. Then (S, 2) cannot contain lines, by (0.2), and so 2~ is ample and spanned for every p e S, in view of Lemma (0.7). Assume that there exists a 0-cycle Z' of length 2 on S~ such that
is not surjective and let Z denote the 0-cycle of length 3 on S induced b y Z' and p. Then, in view of the bijection between 12 - E I and 12~ - Z'I we have that for Z the surjectivity of the homomorphism (0.0.1) fails. This proves that (0.7.1)
any 2-spanned line bundle is generically 2-spanned.
(0.8) PROPOSITION. - Let (X, L) be as in (0.6.0). The line bundle L is generically
2-spanned. PROOF. Assume that L is not 2 spanned and let p be a pont in the support of a 0-cycle of length 3 on X for which the surjectivity of the homomorphism (0.0.1) fails, Consider X~ = Bp (X) and L~, as in (0.6) and for simplicity let (X', L ' ) = (X~, L~). Note that (X,L) does not contain lines, since L E 2 Pic(X); hence L' is ample and spanned in view of Lemma (0.7). Let r be the finite morphism associated to F(L') and set Z = r Since L " L ' = L .L - 1 = 11 and h ~ ') = h~ - 1 = 7 (e.g. see ILl2, table in Sect. 2]), from -
l l = degr degZ >I degr176 we conclude that degr = 1.
') - 2)
9
I conclude this section with the following Lemma, which will play a relevant role in Sects. 2, 3. (0.9) LEMMA.- Let L be a 2-spanned line bundle of genus g on a rational surface X with K x ' K x >I O. Let H = Kx | L and set d' = H . H . Then d ' ~< 2g - 6 + 16/(2g + 2), equality implying both that H and L are linearly dependent in Num (X) and L . Kx = = - 4 . In addition Kx. Kx >~ d' + 6 - 2g.
PROOF.- Since X is rational we have h 2 ( K ~ 1) = h~ theorem gives h~
= 0. So the Riemann-Roch
~) >1 z(g:~ t) = t + K x ' K x .
Therefore there exists an effective divisor D e IKxI]. Note that the arithmetic genus
ANTONIO LANTERI: 2 - s p a n n e d surfaces o f sectional g e n u s six
205
of D is 1. So by L e m m a (0.3), we have that (0.9.1)
4 <<.D . L = - K x ' L = d - (2g - 2),
where d = L . L . gives
Thus d i> 2g + 2. L e t d ' = H . H ;
then the Hodge index theorem
(2g - 2) 2 = ( H . L ) 2 >I d d ' >1 d ' ( 2 g + 2) and this proves the former inequality. Moreover, i f equality holds, then both b, for some a , b ~ Z and equality holds in (0.9.1), i . e . L . K x = - 4 . Now,
L~-H as
d ' = Kx" K x + 2(2g - 2) - d,
we get from (0.9.1) 4 <<.D . L = d - (2g - 2) = K x . K x - d ' + 2g - 2,
which gives the latter inequality.
9
1. - P r o o f o f T h e o r e m 1: first part.
We are assuming that L is a k-spanned (k t> 2) line bundle on a surface X and that g = g ( L ) = 6. By (0.4), since g = 6, we have
(1.0.1)
h ~ (L) I> 7.
L e t N = h e (L) - 1; by the Castelnuovo inequality [GH, p. 252], we have I N ~ 2 + (2(N - 2)g + 1/4) 1/2 , d >t I N ~ 2 + (2(N - 2)g)1/2,
if N - 4 is odd, if N - 4 is even,
(e.g. see [BS, (0.4)]). Hence, by (1.0.1) we get (1.0.2)
d ~> 10.
(1.1) LEMMA. - E i t h e r X is ruled, or d = 10 and X is a K3 surface embedded by ILl in P~. PROOF. F r o m (1.0.2) and the genus formula we have L . K x = 10 - d ~< 0. So, if d I> 11, no positive multiple of K x can be effective and then X is ruled in view of the Enriques criterion. On the other hand, if d = 10 and X is not ruled, then K x is numerically trivial, hence X is a minimal surface of Kodaira dimension zero. In this case L| 1 is ample and then, h i ( L ) = h i ( K x | 1 7 4 1) = 0 for i = 1,2, by the Kodaira vanishing theorem. Thus the Riemann-Roch theorem gives -
h ~ (L) = z ( L ) = X(Ox) + d / 2 = Z(Ox) + 5,
206
ANTONIO LANTERI: 2-spanned surfaces of sectional genus si~
hence, by (1.0.1) and recalling the Enriques-Kodaira classification we conclude that Z(t~x) = 2, i.e. X is a K3 surface and h~ = 7. 9 Now let H = K x | have
and put d' = H . H .
F r o m the Hodge index theorem we
(1.2.1)
100 = (2g - 2) 2 = ( L ' H ) 2 >i d d ' .
(1.2) LEMMA. - We have d' ~< 10, and if equality holds, then X is a K3 surface embedded by ILl in p6. PROOF. - The inequality d' ~< 10 comes from (1.2.1) and (1.0.2). Now assume that d' = 10; then d ~< 10, which combined with (1.02) shows that d = 10. So (1.2.1) is an equality and then by the Hodge index theorem there exist integers a, b such that L a= ( K x | b. So, in N u m ( X ) | we have (1.2.2)
Kx = L (~ - b)/b .
Note also that from 0 = d ' - d = (Kx | L). (Kx | L) - L . L we get Kx" K x = - 2Kx" L. Recalling (1.2.2), this immediately gives a = -+ b. If a = b, we have that K x is numerically trivial and then the same argument proving (1.1) shows that X is a K3 surface and h~ = 7. Now assume that a = - b; in this case K x | 2 is numerically trivial. This would mean that X is a Del Pezzo surface of index two; in other words it would be X = P ~ • and d = K x ' K x / 4 - - 2 , a contradiction. This proves the Lemma. 9 The proof will proceed with a case by case analysis. CASE A: pg > 0. - In this case it must be L . Kx >I O. By the genus formula, recalling (1.0.2) this gives d = 10. Since X cannot be ruled, L e m m a (1.I) says that X embedded by ILl is a K3 surface of degree 10 in p6. We have the following (1.3) PROPOSITION. not 2-spanned.
-
Let S r p6 be a K3 surface of degree d = 10. Then Op (1)s is
PROOF. We have pg (S) = 1 and one immediately checks that this is the maximum geometric genus for a surface of degree 10 in p6. So S is a Castelnuovo variety in the sense of Harris and it follows from [H, p. 65] that it is either -
i) a divisor inside a rational normal scroll T of P~, or ii) the complete intersection of the cone over the Veronese surface of p5 with a hypersurface not containing the vertex of the cone.
ANTONIO LANTERI: 2-spanned surfaces of sectional genus six
207
Case ii) cannot occur, since the degree 10 is not divisible by 4. In case i) we have that S - 32c - 2~, where ~ and # denote a hyperplane section and a fibre of T respectively [H, p. 56]. Therefore S is residual to two planes in the complete intersection of T with a cubic hy: persurface of p6 (case (i) in [H, p. 65]). In particular
13 c 1 = Iop2(3)1 and this shows that the fibres of T cut out on S a pencil of plane cubics, whose general element C is a smooth curve of genus 1, by Bertini's theorem. Then the assertion follows from L e m m a (0.3). " CASE B: pg -- 0, H not v e r y ample. - The exceptions to the v e r y ampleness of H for a v e r y ample line bundle L are known after [SV]. Recalling that g = 6 and that (X, L) cannot contain lines due to (0.2), such exceptions reduce to the following one (see also [BS, (4.1)]): (X, L) is a conic bundle and X is a pl-bundle. Recall that conic bundles are characterized by the condition d' = 0. Then, by the genus formula, recalling also that Kx" Kx = 8(1 - q), we immediately get (1.4.1)
d = 28 - 8q,
and taking into accouiit (1.0.2), this implies that q ~< 2. Now let a and f be a minimal section and a fibre of X and let e denote the invariant, i.e. e = - a. ~; since (X, L) is a conic bundle we have L - [2~ + bf], for some integer b and then (1.4.1) gives b = e + + 7 - 2q. As a first thing assume that q = 2. In this case z is a smooth curve of genus two and so, since L is 2-spanned it must be 5 < d e g L : = 3 - e; this, combined with Nagata's inequality e I > - q, gives e = - 2, and L N [2~ + f ] . In fact this case cannot occur. To prove it, note that L can be written as L = Kx | M, where M - [4a - 3f] is ample, by [Ha, p. 382]. Thus the Kodaira vanishing and the Riemann-Roch theorems
give h~
= h~174
M) = z(L) = 6,
contradicting (1.0.1). When q = 0 or 1, we have numerical conditions for the kspannedness of a line bundle [BS, Sect. 3]. It turns out that the line bundle [2~ + (e + + 7 - 2q) f ] is 2-spanned exactly in the following cases: q=0,
with
0~<e~<5
q=l,
with
-1~<e~<1
[BS, (3.1)]
and [BS, (3.3), (3.4)].
Moreover L cannot be 3-spanned, by [BS, (1.1.1)]. So we have
208
ANTONIO LANTERI: 2-spanned surfaces o f sectional genus six (1.4) pg -- 0 and H not v e r y ample give rise to cases 4, 5 in Theorem 1.
In view of the above we can assume f r o m now on that pg = 0 and H is v e r y ample. Note that from the Kodaira vanishing and the Riemann-Roch theorems we get (1.5.1)
h ~ (H) = 6 - q.
Moreover, by L e m m a (1.1) we know that X is a ruled surface. Since surfaces in p3 are regular, in view of (1.5.1) we have to consider only the following possibilities: i) h ~
= 5; X embedded by IHI is an elliptic ruled surface of p4,
ii) h ~
= 6; X embedded by IHI is a rational surface of p s .
We will consider the f o r m e r possibility in this section, while the latter one will be dealt with in the next section. CASE C. - H is v e r y ample and X embedded by IHI is an elliptic ruled surface of p4. As a first thing, in view of L e m m a (1.2) we can assume that d' ~< 9. Moreover we can also assume that d' i> 5, since there are no irregular surfaces of degree <~ 4. Note that H ' K x = H ' ( H | L -1) = d ' - H . L = d ' - (2g - 2) = d ' -
10.
In addition we have X(Ox)= 0; so the double point formula for surfaces in p4 [Ha, p. 434] gives (1.6.1)
d'2 _ 1 5 d ' + 50 = 2Kx" K x .
Note also that d'-
K x ' K x = (H | K x 1 ) . ( H | K x ) = L . ( K ~ |
L) = 2 L ' K x + d = 20 - d;
hence, recalling (1.0.2), we get (1.6.2)
K x . K x ~ d ' - 10.
By combining (1.6.1) with (1.6.2) we thus see that d' ~< 7. So (1.6.3)
4 ~< d ' ~< 7.
Now assume that (X,H) is not a scroll. Then, from the inequality ( K x | 9(Kx | H ) >I 0 [So, (2.1)], we get (1.6.4)
Kx" K x >! 20 - 3 d ' .
This, combined with (1.6.1), provides the inequality d' I> 9, which contradicts (1.6.3). It thus follows that (X, H) is a scroll. On the other hand there is a unique elliptic scroll in p4 (e.g. see [La]): X is the pl-bundle of invariant e = - 1 over an elliptic curve and H N [~ + 2f]. It turns out that L N [3~ + f ] , which is in fact 2-spanned, as one can
ANTONIO LANTERI: 2-spanned surfaces of sectional genus six
209
prove by using the same argument of[BS, (3.4)]. However, since d e g L : = 4, L cannot be 3-spanned, according to [BS, (1.4.1)]. So we have proved (1.6) For pg = 0, q = 1 and H very ample we get case 2 in Theorem 1.
2. - P r o o f of Theorem 1: second part. In this section we deal with the last and most intricated CASE D. - H is very ample and X embedded by IHI is a rational surface of pS. In view of (1.2) we know that d' ~< 9. Now let g' = g(H). F r o m (2.0.0)
g' = g + Kx" (Kx | L) = Kx" K x + 3g - 2 - d,
(2.0.1)
d'= (Kx| L).(Kx|
L) = K x ' K x + 4g - 4 - d,
we immediately get (2.0.2)
d' -- g' + 4.
Hence (X,/4) can only have the following characters: (d', g ' ) = (4, 0), (5, 1), (6. 2), (7, 3), (8, 4), (9, 5). Now we look at the various possibilities by using the classification of rational surfaces with small sectional genus [Lil], [Li2]. As a first thing, let (d',g') = (4, 0). In this case (X,H) is either (p2, Op2(2)) or a scroll of degree 4 in P~, i.e. (X, H ) = (F0, [z + 2f]) or (F2, [~ + 3f]). Since L = H | K / 1 , this gives
(p2, Op2 (5)), !
(Z, L) = i(F0, [3z + 4f]), [(F2, [3~ + 7f]). In the first case L is 5 spanned by (0.1). In the second case L = [a + f ] 2 | [~ + 2f], hence it is 3 spanned in view of (0.1). This gives cases 1 and 3 of Theorem 1. In the third case, note that d e g L : = 1, so L cannot be 2-spanned. Let (d', g') = (5, 1). In this case X is the Del Pezzo surface of degree 5, i.e. X = = Bpl ..... p4(p2) is the blow-up of p2 at 4 points in general position and H = K;/1 . Then L = K x e is 2-spanned by [BS, (0.8)]: this gives cases 6 of Theorem 1. Let (d',g') = (6,2). In this case X = B p l ..... p6(Fe), is the blow-up of Fe (e ~<2) at 6 points on distinct fibres and H = r:*[2a+ (e + 3 ) f ] Q [ E 1 +... +E6] -1, where 7: stands for the blowing-up and Ei = =-l(p~) [Lil]. Thus L = 7:*[4~ + (2e + 5) f ] | [El + . . . + E6] -2. L e t z' = r: -1 (~) and note that degL~, ~< 5 - 2e, hence L cannot be 2-spanned for e = 2. On
210
ANTONIO LANTERI: 2-spanned surfaces of sectional genus six
the other hand, for e = 0, i we have L = K~/2 @ =*[f] and then the 2-spannedness of L follows from Proposition (0.5), i). This gives case 7 of Theorem 1. From now on we can assume that g' i> 3. From the Hodge index theorem, recalling (2.0.2), we have (2.0.3)
(g' - 6) 2 = (2g' - 2 - d ,)2 = (H. Kx) 2 >t d' gz" Kx = (g' + 4) gx" g x .
On the other hand, by (2.0.1) we have s d ' = K x ' K x + 2 0 - d; recalling (1.0.2) and (2.0.2) this gives (2.0.4)
Kx" Kx >1g' - 6.
By combining (2.0.3), (2.0.4) we get (2.1.1)
-3<.Kx.Kx<.I
for g ' = 3 ,
(2.1.2)
-2
for g ' = 4 ,
(2.1.3)
-I<<.Kx.Kx<.O
for g ' = 5.
Now note that since g' ~> 3, (X, H) can be neither a scroll, nor a Del Pezzo pair; this implies that (Kx | H ) . (Kx @ H ) >i 0, equality holding iff (X, H) is a conic bundle [So, (2.1)]. Recalling (2.0.2) this yields (2.1.4)
K x ' K x >I 8 - 3g'
with = iff (X, H ) is a conic bundle.
So (2.1.1) can be replaced with the sharper inequality (2.1.1')
-I<<.Kx.Kx~
for g ' = 3 , w i t h = o n
the left iff (X,H) is a conic
First of all we deal with the following case: (X,H) is a conic bundle with ( d ' , g ' ) = (7,3). In this case it follows from[Lil] that X = Bpl ..... p9(Fe) is the blow-up of Fe (e ~< 3) at 9 points lying on distinct fibres and H = ~*[2o + (e + 4 ) f ] | + ... + E g ] -1, where r: stands for the blowing-up and E~ = =-1 (Pi). Moreover the 9 points must satisfy some more restrictions on their position in order H to be very ample[Li2, p. 169]. Thus L = r:*[4o+ (2e + + 6) f ] | [El + ... + Eg] -2 9Let o' = =-1 (0) ad note that degL:, ~< 6 - 2e, hence L is not even a m p l e for e = 3. For e ~< 2, the generic 2-spannedness of L follows from Proposition (0.8) and this gives case 8 of Theorem 1. Now note that due to (2.1.2), (2.1.3), (2.1.4), (X, H) cannot be a conic bundle for g ' = 4, 5. So, from now on, we can assume that (X, H) is not a conic bundle. Then from [LP, (5.1)] (see also [So, (2.6.1)]), (2.1.4)can be sharpened giving (2.2.1)
Kx" Kx I> 6 - 2g' cubics.
with = fff (X, H) is either a Bordiga surface or ruled by
ANTONIO LANTERI: 2-spanned surfaces of sectional genus six
211
This implies the following facts: as a first thing (2.1.1') can be further sharpened, giving (2.2.2)
0 <~Kx' Kx ~< 1
with = on the left iff (X, H) is a Bordiga surface or ruled
by cubics. As a second thing, recalling also that Bordiga surfaces have sectional genus three, (2.1.2) can be rewritten as (2.2.3)
- 2 ~< Kx" Kx <<-0
with = on the left iff (X, H) is ruled by cubics.
Finally note that when Kx'Kx >>-0 it can only be d' = 7 and Kx'Kx = 1, due to L e m m a (0.9) and (2.2.2). This fact, recalling (2.2.2), (2.2.3), (2.1.3) shows that the only possible cases to consider are those listed in the table III. TABLE III.
Subcase
g'
Kx" Kx
a
3
1
b
4
-2
c
4
-1
5
-1
d
:
Comments
(X, H) ruled by cubics
Now we look at Livorni's classification [Lil] for a detailed study of the above subcases. In particular look in table at p. 96 at surfaces with h~ = 6 and the corresponding value of Kx" Kx. X is always described as Bpl,. p~(X A), the blow-up of another surface X A at some points Pl, ..., P~ in general position; = will denote the blowing up and Ei = 7~-1 (Pi). For the proof of the very ampleness of H and of the corresponding L, see [Li2]. - L e t X A be the Del Pozzo surface with KxA'KxA = 2. Then and H = r ~ * K ~ | -1. It follows that L = 7 ~ * K ~ | 1 7 4 | I = K:~ 2 | =* Kj/2 and then L is in fact 2-spanned in view of Proposition (0.5),ii). This gives case 9 of Theorem 1. SUBCASE
a).
X = B p ( X A)
SUBCASE b). - We have (X,H) = (Bpl, ..pl0(XA), rv*HA~)[E1 + ... + E l o ] - l ) , where ( X A , H A ) = ( Q , Op~(3)| Q being a smooth quadric of pS. L e t h e IOw(1)| Then we have L = [ r ~ * 5 h - 2 E 1 - . . . - 2 E l o ] and so h~ = = h ~(5h) - 30 = 6, contradicting the 2-spannedness of L, in view of (0.4). So this case does not occur.
212
ANTONIO LANTERI: 2-spanned surfaces of sectional genus six
SUBCASE C). - We have (X, H ) = (Bpl ' . p4(XA), r~*H A | [El + ... + E4]-I), where (X A , H A) = (S, Op~ (2) | Os), S being a smooth cubic o f P ~ . Let h e 10~(1) | ~sl and note that L = [=* 3h - 2E1 - ... - 2E4]. L e t F be the proper transform via = of the hyperplane section of S with the plane spanned by the three points Pl, P2, P~. Then Pa([')=l
and
rel~*h-E1-...-E~l,
so t h a t / ' . L = 3h. h - 6 = 3, contradicting the 2-spannedness of L in view of Lemma (0.3). So Subcase c) does not occur as well. SUBCASE d ) . - (Unfortunately the table in [Lil] omits this case which we need to consider; see[Lie, case 2) in the table at p. 161]). We have ( X , H ) = = (Bpl, ..pl0(p2), r~* Op~ (7) | [El + ... + Ei0]-2). Let h 9 I Op2 (1) 1; then L = [7:* 10h - 3 E 1 - . . . - 3E10] and so we have dim ILl = dim]10h I - 6 0 = 5, contradicting the 2-spannedness of L, in view of (0.4). So Subcase d) does not occur.
This concludes the proof of Theorem 1.
3. - 3 - s p a n n e d s u r f a c e s o f s e c t i o n a l g e n u s s e v e n .
The same procedure as in Sections 1, 2, works also for proving Theorem 2. Let L be a 3-spanned line bundle on a surface X and assume that g = g(L) = 7. In this case we have h~ >I 8, by (0.4), and then the Castelnuovo inequality gives (3.0.1)
d >I 12,
equality implying h ~( L ) = 8. The same argument proving Lemma (1.1), now gives (3.1) Either X is ruled or d = 12 and X is a K3 surface embedded by ILl in pT. Letting a g a i n H = K x | and d ' = H . H , the Hodge index theorem gives 144 ~> dd', and in the same way as in Sect. 1 we get (3.2) d' <~ 12, equality implying that X is a K3 surface embedded by ILl in pT. Now, following the same procedure as in Sect. 1, we deal as a first thing with CASE A': p~ > 0. - By (3.1) we thus see that X is a K3 surface embedded by ILl in p7 and the same argument as in Proposition (1.3) rules out this case. CASE B': p~ = 0 and H not very ample. - Since L is 3-spanned (X, L) can contain neither lines nor conics, by (0.2). Hence, as g = 7, it follows from [SV] that H is very ample. So this case does not occur.
ANTONIO LANTERI: 2 - s p a n n e d surfaces o f sectional g e n u s s i x
213
In view of the above we can assume that pg = 0 and H is very ample. Then by reasoning as in Sect. 1 we get the following possibilities: C') X is a ruled surface with q = 2 embedded by IHI in p4; C") X is a ruled surface with q = 1 embedded by IHI in p s ; D') X is a rational surface embedded by IHI in p6. We analyze the above possibilities. CASE C'. - We have H ' K x = d ' - 12, Z(~gx) = - 1, K x ' K x <- 8(1 - q) = - 8. So the double point formula for surfaces in p4 gives the inequality d '~ - 15d' + 48 = 2 K x ' K x <<- - 16, which can never be satisfied. So this case does not happen. CASE C". - Let g' = g(H); then from (2.0.0), (2.0.1) we get (3.3.1)
d' = g ' + 5.
Moreover, in view of (3.2) we have d' ~< 11, so that (3.3.2)
g' ~< 6.
(3.4) REMARK. If L is 3-spanned and (X, H) is a conic bundle, then X is a pl-bundle. Actually all the fibres of the ruling of X must be irreducible, in view of (O.2). -
Now, looking at the classification of irregular ruled surfaces of sectional genus ~< 6 [Li2, table in Sect. 3] and taking into account (3.3.1), (3.3.2), (3.4) and the equality h ~ --- 6, we see that X can only be the elliptic PLbundle of invariant e = - 1 and H is numerically equivalent to either [~ + 2f] or [3z]. However, in both cases we have degL~ = 4, contradicting the 3-spannedness of L in view of[BS, (1.4.1)]. It remains to consider CASE D'. characters:
By (3.2) and (3.3.1), (X,H)
can only we have the following
(d', g ' ) - (5, 0), (6, 1), (7, 2), (8, 3), (9, 4), (10, 5), (11, 6). Since g' ~< 6, we can look at the various possibilities by using the classification of rational surfaces with small sectional genus [LU, [Lie]. As a first thing, let (d', g') = (5, 0). In this case (X, H) is a scroll of degree 4 in p s , i.e. (X, H) = = (F~, [~ + 3f]) or (Fs, [~ + 4f]). Note that in the latter case L = [3~ + 9f] is not even ample since degL~ = 0. The former case leads to the pair (X, L) = (F1, [3~ + 6f])
214
ANTONIO LANTERI: 2-spanned surfaces of sectional genus six
and L is in fact 3-spanned in view of (0.1), since L = [~ + 2f] 3 . This gives case 1 in Theorem 2. (d',g') = (6, 1). In this case X is the Del Pezzo surface of degree 6, i.e. X = = Bpl ..... p3(p2) is the blow-up of p2 at 3 points in general position and H = K ~ 1 . Then L = K~ 2 is 2-spanned by (0.1), but not 3-spanned, since L ' E i = 2, for i = 1,2,3. (d',g') = (7,2). In this case (X,H) is a conic bundle, with X = B p I ..... ps(Fe), e ~< 2. So this case cannot occur due to Remark (3.4). More generally, since X is never an Fe, Remark (3.4) allows us to exclude conic bundles from our consideration. So, from now on we can assume that (X, H) is not a conic bundle. Then, recalling (3.3.1), from the inequality ( K x | 1 7 4 >t g ' - 2 and the characterization of the corresponding equality[LP, Th. 5.1], we get (3.5.1)
Kx" K x >I 7 - 2 g ' , with = iff (X, H) is either ruled by cubics, a Bordiga surface (g' = 3) or some special surface with g ' = 6.
On the other hand, as in Sect. 2, by the Hodge index theorem, we get the inequality (3.5.2)
(g' - 7)2 = (2g' - 2 - d ' ) 2 = ( H . K x ) 2 >I d ' K x ' K x = (g' + 5 ) K x ' K x .
Now we can deal with case (d',g') = (8,3). In view of (3.5.1), (3.5.2), when g' = 3 we have 1 ~< Kx.Kx<~ 2, with = on the left iff (X, H) is either a Bordiga surface or ruled by cubics. Now look at [Li2, tables at pages 161, 169]. Due to (3.5.1), an inspection of the invariant ]'0 = = d' - K x " K x shows t h a t (X, H) cannot be ruled by cubics. Moreover ~(Bpl ..... ps (P2), 7c* Op2 (4) ~ [E 1 + ... + Es] -1)
if K x ' K x = 1,
(X, H ) = [(Bp~ ..... pT(P2), 7:* Op2(6) | [El + ... + ET] -2)
if K x ' K x = 2.
In the former case we have L "Ei = 2 for i = 1, ..., 8, hence L cannot be 3-spanned in view of (0.2). In the latter case X is a Del Pezzo surface of degree 2 and L-= Kx~; hence L is 3-spanned by [BS, (0.8.3)]. This gives case 2 in Theorem 2. To deal w i t h the remaining cases, we first prove the following (3.6) LEMMA.
-
If d' >/9, then Kx" K x < O.
PROOF. Assume, by contradiction, that Kx.Kx>~ O. Then' d ' < 9, by L e m m a (0.9), equality implying that L a - ( K x | b for some a,b e Z and L ' K x = - 4 . L e t t = (a - b)/b; then -
(3.6.1)
td = - 4.
ANTONIO LANTERI: 2-spanned surfaces o f sectional genus six
215
On the other hand, from (3.6.2)
9 = d ' = H . H = K x ' K x + 2(2g - 2) - d = K x ' K x + 24 - g,
we have (t 2 - 1)d = - 15.
(3.6.3)
Now (3.6.1), (3.6.3) give t = - 1/4 or 4. Of course, it cannot be t = 4, otherwise K x would be ample, contradicting the fact that X is a rational surface. Hence L - K~ 4 . In particular X is a Del Pezzo surface, L = Ks/4, since Pic (X) is torsion free, and (3.6.2) gives Kx" K x = 1. Recalling the general properties of Del Pezzo surfaces we thus conclude that IK~/~l contains a smooth curve F of genus 1. But F. L = 4Kx" K x = 4,
which contradicts the 3-spannedness of L in view of[BS, (1.4.1)]. So case d' = 9 cannot occur. 9 Now let (d',g') = (9,4). In this case Lemma (3.6) and (3.5.1) imply that K x . K x = - 1 and (X, H) is ruled by cubic. F r o m [Lie, p. 169] we see that X is not a Pl-bundle and so this case cannot occur in view of the following (3.7) REMARK. If L is 3-spanned and (X, H) is ruled by cubics, then X must be a PLbundle. Actually, were F = F1 + F2 a reducible fibre of the ruling of X, then for the irreducible component F~ satisfying H-F~ = 1 we would get L . F i = H ' F i - K x " 9Fi ~< 1 + 1 = 2, since Fi or another irreducible component of F is an exceptional curve. But this contradicts the 3-spannedness of L, by (0.2). -
Now, taking into account (3.0.1) and (3.3.1), we get from (2.0.1) Kx. K x = d ' + d - 24 ~ d' - 1 2 = g ' - 7.
L e m m a (3.6) and the above inequality show that (3.8)
If ( d ' , g ' ) = (10, 5) then K x ' K x = - 1
or - 2 , while if ( d ' , g ' ) = (11,6) then
gx.gx = -1.
Now let (d', g') = (10, 5). By (3.8) we have to look in [Li2, pp. 161, 169] only at surfaces with •o = 11 and 12 and we thus see that (X,H) alwasy contains a ( - 1)-line E. Then L . E = H . E - K x ' E = 2, which contradicts the 3-spannedness of L, by (0.2). Finally let (d', g') = (11, 6). By (3.8) we have to look in [Li2, pp. 161, 169] only at surfaces with ~'0 = 12. Then X = Bp, ..... pl0 (p2), with H = 7~* Op2 (9) | [3E1 + ... + 3E6 + 2Es +
216
ANTONIO LANTERI: 2-spanned surfaces of sectional genus six
+ ... + 2E10] -1 . However, for the corresponding line bundle L we have h~ = 7 [Li2, Case 19 in the table at p. 161], and this contradicts the 3-spannedness of L, by (O.4). This concludes the discussion of case D' and the proof of Theorem 2.
A d d e n d u m ( F e b r u a r y 1993). After the submission of this paper and its circulation as a preprint further progress on the subject was made. I r e f e r to the paper by M. ANDREATTA, Surface8 of sectional genus <~8 with no trisecant lines, Arch. Math., 60 (1993), pp. 85-95.
REFERENCES BALLICO E., On k-spanned projective surfaces, Algebraic Geometry, Proc. L'Aquila, 1988, Lect. Notes Math., 1417, Springer (1989), p. 23. [Bae] BALLICO E., A characterization of the Veronese surface, Proc. Amer. Math. Soc., 105 (1989), pp. 531-534. [BS] BELTRAMETTIM. - SOMMESEn. g., On k-spannedness for projective surfaces, Algebraic Geometry, Proc. L'Aquila, 1988, Lect. Notes Math., 1417, Springer (1989), pp. 24-51. Essll BELTRAMETTIM.- SOMMESEA. J., O-cycles and k-th order embeddings of smooth projective surfaces, Problems in the Theory of Surfaces, Proc. Cortona, 1988, to appear. [BFS] BELTRi~METTIM. - FRANCIA P. - SOMMESEt . J., On Reider's method and higher order embeddings, Duke Math. J., 58 (1989), pp. 425-439. [GH] GRIFFITHS PH. - HARRIS J., Principles of Algebraic Geometry, J. Wiley & Sons, New York (1978). HARRIS J., A bound on the geometric genus of projective varieties, Ann. Sc. Norm. Sup. [H] Pisa,~(4), 8 (1981), pp. 35-68. [Ha] HARTSHORNE R., Algebraic Geometry, Springer-Verlag, Berlin, New York (1977). LANTERI t., On the existence of scrolls in p4, Acc. Naz. Lincei, Rend. C1. Sc. Fis. Mat. [La] Nat. (8), 69 (1980), pp. 223-227. [LP] LANTERIA.- PALLESCHIM., On the adjoint system to a very ample divisor on a surface and connected inequalities, II, Acc. Naz. Lincei, Rend. C1. Sc. Fis. Mat. Nat. (8), 71 (~981), pp. 166-175. [Lil] LIVORNI L. E., Classification of algebraic surfaces with sectional genus less than or equal to six. - I: Rational surfaces, Pac. J. Math., 113 (1984), pp. 93-114. [Li2] LIVORNI L. E., On the existencv of some surfaces, Algebraic Geometry, Proc. L'Aquila, 1988, Lect. Notes Math., 1417, Springer (1989), pp. 155-179. SOMMESEA. J., Hyperplane sections of projective surfaces. - I: The adjunction mapping, [S] Duke Math. J., 46 (1979), pp. 377-401. [SV] SOMMESE Ao J. - VAN DE VEN A., On the adjunction mapping, Math. Ann., 278 (1987), pp. ~593-603. [Bal]