On algebraic surfaces and threefolds whose cotangent ...

J. Math. Kyoto Univ. (JMKYAZ)29-1 (1989) 27-43

On algebraic surfaces and threefolds whose cotangentbundles are generated by global sections

By

Hiro-o TOKUNAGA

§ 0 . Introduction

For smooth subvarieties of abelian variety, the following fact is well-known:Let X b e a sm ooth subvariety o f general type of an abelian variety with

dim X = 1 or codim X = 1 . Then the tricanonical map 0 13K xi is embedding.In this paper, we consider a generalization of the above fact in the case of

dim X = 2 or 3 and show that 0 13Kxl is b ira tiona l. Main results are as follows.

Theorem ( A ) Let S be a surface of general type whose cotangent sheaf Q isgenerated by global sections. If h ° ( ,(4 ) . 4, the bicanonical m ap 01 2 K 5 1

o f S isbirational except the case w here S = C, X C 2 , g(C2 ) = 2.

Moreover, If th e A lbanese torus o f S is a simple abelian variety. 012Ksi is

embedding.

Theorem (B ) Let V be a projective threefold of general type whose cotangentsheaf 01, is generated by global sections. If h ° ( Q ) 5 , the tricanonical map013K0 o f V is birational.

Note that if S and V are smooth subvarieties of an abelian variety ,S4 andare generated by global sections.

I w o u ld lik e to th a n k P ro f . K . U eno for his valuable suggestions andencouragement.

Notation and Convention.C , S, V denote a compact non-singular curve, a compact non-singular surface,

a compact non-singular threefold, respectively.g,(X ):= h i (X , (9,) i- th irregularityp g (X ): geometric genusg(C): genus of a curve CK (X ): Kodaira dimension of X

m-th canonical map of Xe x (D — kx): subsheaf of C(D ) with k-fold zero at x.

Received, June 16, 1987

28 Hiro-o Tokunaga

e x : sheaf of germs of holomorphic vector fieldsQ i: sheaf of germs of holomorphic cotangent vector fields

Let D,, D2 be divisors.D, — D2 : linear equivalence of divisorsD, D2 : numerical equivalence of divisorsD, '',;i() D2 : numerical equivalence of Q — divisors

§ 1 . Preliminaries

In this section, we shall prove some lemmas which will be used later to proveTheorems (A) and (B).

Lemma 1.1. Let X be a surface (or a threefold) whose cotangent sheaf Qk isgenerated by global sections. If K(X)> 0, then ImK x i (m e N) is free f rom basepoints.

Pro o f . H ere w e prove th e lemma fo r dim X = 2. T he proof of the casewhere dim X = 3 is similar.

It is enough to show that, for any point p E X , there exist two holomorphic2-forms such that the one has zero a t p and the o ther does not have zero atp. Since q 1 (x) 3 , th e re e x is ts a holomorphic 1-form cop w ith zero a t p, and bythe assumption, there exists a holomorphic 1-form co with cop A w # O . On theother hand, by the assumption, there exist two holomorphic 1-forms co„, co2 whichspan the cotangent space at p. For these 1-forms, take exterior product co, A co,.This is non-trivial holomorphic 2-form which does not have zero at p.

The following theorem proved by Z. Ran is im portan t. F o r a proof, see Ran

Theorem 1.2. Let 0: X --+ Y be an immersion where X is a compact complexmanifold and Y is a com plex torus. Assume that X is of general type. T h en , wehave the following.

( i) Fyry is finite, where Fo is the Gauss map associated with 0.(ii) The canonical bundle K x is ample.(iii) If A c 0 (X ) is an abelian subvariety of dimension k, then:

n — mk <

n — m + 1

where n = dim Y, m = dim X.

By the above theorem, manifolds of general type whose cotangent sheaf aregenerated by global sections have an ample canonical bundle K .

Next, we shall prove some inequalities.

Lemma 1.3. Let S be a surface of general type whose cotangent sheaf 5-4 isgenerated by global sections. T h e n the inequality

A lgebraic surfaces and threefolds 29

ci(S) c2 (S)

holds, where c(S ) is the i-th Chern class of S.In particular, we have

6x(es).

P ro o f L e t W b e the projectivized tangent bundle P(Ts ) where Ts i s theholomorphic tangent bundle of S with a natural projection 7E: W - S. P u t L —

0 ( 1 ) . Since we have

n,e2w(1) f r 4 ,

L is free from base points and fixed components. Therefore, we have

0 < L 3 = c — c 2 .

Hence we obtain the desired inequality,

c

2 >.

Lemma 1.4. Let S (resp. V) be a surface (resp. threefold) of general type whosecotangent sheaf ,W is generated by global sections. If S (resp. V) has a surjectivemorphism f: S (resp. V) —> C to a non-singular curve. Then following inequalitieshold:

g = g(C) q 1 (S) — 2 (resp. q 1 (V) — 3).

P ro o f Here we shall prove the surface case o n ly . A proof of the threefoldcase is s im ila r . By the assumption, g q ,(S). Hence, it is enough to show thatg 0 q 1 (S) — 1. I f g = q1 (S) — 1, o n a general fibre F there is only one linearlyindependent holomorphic 1-form which is a restriction of a holomorphic 1-formo n S . S o , by the assumption, the general fibre of f is a n elliptic curve. Thiscontradicts the assumption that K(S) = 2.

Lemma 1.5. Let V be a threefold of general type whose cotangent sheaf Q, isgenerated by global sections. Then the inequality

12holds.

P ro o f B y T heorem 1 .2 , K v i s a m p le . S o , b y th e K o d a ira vanishingtheorem,

hi(V, (9v (2K v )) = 0 (i = 1, 2).

By the Serre duality,

IP(V, e v (2K v )) = (9v(— K v )) = O.

Hence, by the Riemann — Roch Theorem for threefold, we have

x(Ov (2K )) = h°(9(2K v )) =11q, — 3x(e)v ) .

30 Hiro-o Tokunaga

Let S be a general member of K . B y L em m a 1.1, S is a non-singular surface ofgeneral type and Q is generated by global sections. Therefore, by Lemma 1.3,

6x(e s ). (1.2)

On the other hand, by a cohomology exact sequence

0 H ° ((9v (K,))—) ((9v (2K v )) —> (S, &5 (K 5 )) —>

—>H1 ((9,(K v )) —> 0

and h i ((.9,(K v )) = h 2 (0 ,) (by the Serre duality), we obtain an equality,

pg (V)— h ° ((9v (2K v )) + p g (S)— q 2 (V )= 0.

By (1.1) and h'((9) = 11 1 WO, we obtain an equality,

pg (V) — (41<,37 — 3x(Cv )) — q2 (V )+ q i (V )+ p g (S)— q,(S)= 0 .

Therefore,

A es) = P g (S) — + I = 1 1<137 — 2 X(Cv) • (1.3)

B y the adjunction formula, K ,= 2K v 15 , hence f q = 4K . Therefore, by (1.2),(1.3), we have

41(13, 6X(es)

= 31q, — 12x((- v ).

Hence, — 12x(G).

By Yau's inequality, if K v is ample, x(Ov ) < 0 . Therefore, K,3, 12.

We will summarize the results of connectedness of the canonical divisors ofsurfaces which will be used in §2 and §3.

Proposition 1.6. Let S be a minimal surface o f general type, and D be anef fective div isor of S such that D m K , ( i n 1 ) . T h e n D is a numerically 2-connected divisor except the case:

K = 1 , m = 2 , D =D 1 +D 2 , D, D 2 K s .

For a proof, see Bombieri [3].

Corollary 1.7. Under the above notation, D is numerically 1-connected.

The following theorems will be used in §2 and §3.

Theorem 1.8. Let S (resp. C) be a surface (resp. curve), and f:S — > C be arelatively minimal .f ibration such that the genus of a general fibre is positive. T h e nwe have

deg (f,,cos i c ) . 0.


Moreover, this degree vanishes ff and only if either f is locally trivial (hence smooth)o r S is an elliptic surface w ith constant moduli and the only singular f ibres aremultiple fibres.

For a proof, see W. Barth, C. Peters, A. Van de Ven [1].

Theorem 1.9. L et S (resp. C) be a Kahler surface (resp. curve), and f : S Cbe a fibration. Then f .co s ic i s the direct sum CV (h = h i (C,f,,cos )) and a locallyfree sheaf which satisfies h i (C , g [K J) = 0, where OP = (9 c C) • • • C) Cc .

h

For a proof, see Fujita [5].

Proposition 1.10. L et S be a minimal surface of general type. Then there areonly finitely many irreducible curves C on S such that the intersection number K s Cis bounded and C 2 < 0.

Theorem 1.11 (Ramanujam's vanishing theorem). L et S be a surface and D aneffective numerically 1-connected divisor on S with D2 > 0. Then h i ((Ps (—D)) = 0.

For a proof of Proposition 1.10 and Theorem 1.11, see Bombieri [3].The following theorem will be used in §3.

Theorem 1.12 (H orikaw a). L e t X an d Y b e com pact com plex manifolds,f: X — > Y a holomorphic map, and let p: —04 be a fam ily o f deformations ofX = p - 1 (o), o e M . L et fi * denote the canonical homomorphism 0 0 — > I l l (X ,f *O y ). A ssume that f i * is surjective f o r i = 1 and is injective for i = 2. Then,there exists an open neighborhood N of o in M, a family q: y N of deformationsof Y = q - 1 (o), and a holomorphic m ap 0 :

X I N — q , ov er N w hich induces f ov er

X = r i (o).

For a proof, see Horikawa [6].

Next lemma is important to prove our main theorems.

Lemma 1.13. (c i) L et S (resp . V ) be a surface (resp. threefold) and co,, co 2 belinearly independent holomorphic 1-forms such that co, A co2 0 . T h e n th e re e x is tsa curve C with g(C) 2 which satisfies the following conditions (1) and (2):

(1) There exists a surjective holomorphic map f: S C ( re s p . V C ) .(2) There exist holomorphic 1-forms (p i (i 1, 2) such that

(0 1 = f* P i C°2= f*(p2.

(fl) L e t V be a threefold and co,, co,, co, be linearly independent holomorphic1-forms such that ci), A w j # 0 (i < j) and co, A co2 A 0 3 0 . T h e n th e re e x is ts asurface of general type which satisfies the following conditions (1), (2):

32 Hiro-o Tokunaga

(1) There exists a dominant rational map f: V S.(2) There exist holomorphic 1-forms ço , (i = 1, 2, 3) on S such that

0)1 = f * ( p i ,

For a proof, see Ran [8].

0 )2 = f* (p 2 , (1 )3 = f*(p3 •

Lemma 1.14. L et S be a surface for which p g 2 q i — 4 . Then there existtwo linearly independent holomorphic 1-forms such that w, A co2 0 .

For a proof, see Beauville [2], Lemma 10.7.

§ 2 . Proof of Theorem (A)

When we consider whether the bicanonical map 0 12, 51 is birational or not, it isimportant to know whether S has a structure of a fibre space with curve of genus2 as a general fibre. In case S is a surface of general type whose cotangent sheaf54 is generated by global sections, the structure of such surfaces is simple.

Lemma 2.1. L et S be a surface of general type whose cotangent sheaf S -4 isgenerated by global sections. If S has a structure of a f ibre space w ith curve ofgenus 2 as a general f ibre, then S = C, X C 2 and g(C2 ) = 2.

P ro o f . Let p: S — > C be a fibration with curve of genus 2 as a general fibre.By Leray's spectral sequence, we obtain an exact sequence:

0 - + 11 1 (C, (9,) — > 111 (S, (9s ) — H ° (C, R 1 p* (9s ) 0.

By the relative duality,

(R1P* (9s) - Z W o w

Hence, IP(C, R 1 p,(9s ) = h 1 1C,(P* cosic) 0 (t)c) •On the other hand, by Theorem 1.9, we have

N o w 2 C P" 0 g

where h = ( C , w a s ) and g is a locally free sheaf with 111 (C, e[K c ]) = 0, Sincerank (w o w ) 2,

Therefore,

But by Lemma 1.4, we have

Therefore we have

h = IP(C, R i p ( 9 s ) 2.

171 ((95 ) = g + h _ g + 2 .

h i (e s ) g + 2 .

h = 2 .


B y th e above argum ent, w e o b ta in p,,,cos / c Z (9c. C c , d eg (p,wslc) = O. B yTheorem 1.8, p is locally trivial fibration. S o , S is a fibre bundle with curve F, ofgenus 2 as its fibre. A ssu m e that this fibre bundle is not trivial. Since Aut (F0 )always acts on H

°(f4 0 ) non trivially, this contradicts the fact that 10(00= g + 2.

For any surface of general type, the inequality x(es ) 1 alw ays holds. But, inour case, the following stronger inequality holds.

Lemma 2.2. L et S be a surface of general type with the irregularity qi (S).. 4such that its cotangent sheaf (4 generated by global sections. Then x((9s ) 2 holdsexcept the case where S is a product of curves of genus 2.

Pro o f . I t is e n o u g h to show th a t if x(Cs ) = 1 under th e assumption inLemma 2.2, then S is a product of curves of genus 2. First, by Yau— Miyaoka'sinequality, if x((9s ) = 1, then we have

< 9 .

By Noether's inequality for irregular surface of general type, we have

cf 3q, — 4 8 (q 1

. 4)

Hence, we haveci• = 8 or 9 and pg = q, = 4 .

Now, if S does not have any morphism to a curve of general type, by Lemma 1.13,and Lemma 1.14, we have

pg 2 q 1 — 3 > q, = 4 .

Therefore, a surface which satisfies the assumption in Lemma 2.2, and x(es ) = 1has a structure of a fibre space p: S C with g(C) = 2. O n the other hand, wehave

c, 4(g(C)— 1)(g(F) — 1)

= 4(g(F) — 1)

where F is a general fibre and c, is second Chern class of S. Hence, by Noether'sequality, we have

c? = 8 , c2 = 4 , g(F)= 2 .

Therefore by Lemma 2.1 we conclude that S = C x F. 0

R em ark. By using Beauvelle's result (see Beauville [2], Theorem), we simplifythe proof of Lemma 2.2.

Corollary 2.3. L et S be a surface which satisfies the assumptions in Theorem(A). Then we have .1q. 12 except the case where S is a product of curves of genus2.

34 Hiro-o Tokunaga

Pro o f . It follows immediately from Lemma 1.3 and 2.2.

From now on, we assume that S C , X C 2 (g(C ,)= 2, i = 1, 2).

Proof o f Theorem (A). It is enough to show that if 0 12, s1 is not birational, Shas a structure of a fibre space with curve of genus 2 as a general fibre. A ssum ethat 0 12, s1 is n o t a birational m orphism . Then there exists a Zariski open set Uo n S such that for any point x e U, there exists a po in t y e U with A

1 2 KChoosing a smaller U if necessary, by Proposition 1.10, we may =12K(.0. assume0 tha t U S \ UC, where C runs over all irreducible curves C on S with K s Cand C 2 < 0 . I f (6 is the set of such curves, contains all irreducible curves withK C = 0 o r K s C = 1. I n o u r case, since K s i s ample, there is n o curve withK s C = 0.

C laim 2.4. L e t x , y b e the points as above and n: — > S be obtained bycomposition of blowing-ups of S at x an d y . Then In*K s — 2Ex — 2Ey 1 0, w hereEx and E y are the exceptional curves of the first kind appearing in the blowing-ups atx and y, respectively.

Proof of Claim 2.4. Since Qk is generated by global sections and 11°W D 4,there exist at least two linearly independent holomorphic 1-forms coa , a),, such that

w (x) = cob (x) = 0.

Moreover, these forms have also zero a t y. In fact, if coa (y) 0, then sinceis generated by global sections, there exists a holom orphic 1-form such that(wa A w ) ( y ) 0. The holomorphic 2-form coa A co is not identically zero, and haszero a t x . This contradicts the assum ptions. So, coa (y) = co,,(y) = 0. I f co,,,satisfy Wa A W b # 0, then this proves Claim 2.4. Now, we can find 1-forms (Da , co,such that co a A W b * 0, as follows.

C ase 1 : For any linearly independent, oh, co 2 e H°(S2), co, A (D , 0.

In this case, we have C O , A Wb # 0.

C ase 2 : There exist linearly independent oh, co2 e H°(,(2,) with ah A co, 0.In th is case, by Lemma 1.13, there exists a surjective morphism p: S —> C,

g(C) 2 and w 1 , co2 are the pullback of 1-forms on C . Moreover, by Lemma 1.4,there is a relation g, g + 2 between the irregularity g , of S , and the genus g ofC . Therefore, we take coa in such a way that it is contained in p*H °(W ), and wetake co b in such a way that it is not contained in p* H°( Q ) . F or these 1-formswa, wb, we have COa A Wb 0 .

By Corollary 2.3, a n d S C , x C 2 ( g ( C i ) = 2, i = 1 ,2), w e have lq > 12.Hence,

(n* K s — 2E„ — 2Ey )2 4 > 0 .

Therefore, there does not exist an effective divisor D e In*K s — 2E„ — 2Ey 1 whichis numerically 1-connected. In fact, if D is a numerically 1-connected divisor, byTheorem 1.11, we have

(S, (Os ( — (7t* K s — 2Ex — 2E,,))) = 111(S, D)) = 0 .

Algebraic surfaces and threefolds 35

So, in the cohomology exact sequence,

•• • —› H 0((9,(27c* Ks )) —> 1-10(CE ) CD II° W O

—> H° (Cs (2n* Ks — E — E y ) —> • • • .

we can show,

•• H IV s ( 27E* Ks )) H ° (( E ) '0 11° 1(90 (exact) .

T his contradicts th e assum ption that 012,icsi(x) — 6012Ksi(Y). B y using th e sameargum ent as th a t in B om bieri [4], Theorem 5 , w e can show tha t, if D is notnumerically 1-connected, S has the structure of a fibre space of curve of genus 2 asa general fibre. Therefore, by Lemma 2.1, S = C , x C 2 g(C2 ) = 2. This provesthe first statement in (A).

The next propotion proves the second statement in (A).

Proposition 2.6. Let S be a surface of general type with condition (A), and theAlbanese torus Alb (S) of S be a simple abelian variety. Then, 012K51 is embedding.

Proo f . Notation is the same as a b o v e . A proof of Proposition 2.6 consistsof two steps.

Step 1: 0 12„si is a one to one mapping.L e t x , y b e p o in ts s u c h th a t 6 (x)12Ksi , — 2ics i(.1). T h e n w e c a n show

lit* K s — 2E, — 2E3,1 0 Ø b y th e sam e m ethod that o f C laim 2.4. Hence, theproblem is reduced to show connectedness of members of In* K s — 2E x — 2Ey l. Ifthere exists a connected member of In*K s — 2E x — 2E3,1, by Theorem 1.11, we canprove 6 t x ) 0 6 tri2x s1,-, • ri2K ,H,v) T h i s contradicts t h e assum ption 0 12 ,, (x) _ -0-12Ksi(Y).Hence, any member o f n* K s — 2E„ — 2Ey 1 is not connected. Let C be a membero f 1Ks — 2x — 2k a n d C = C , + C 2 be decom potion in to tw o positive divisors.By using th e sam e argum ent as th a t in B o m b ie ri [4 ] Lem m a 1 , if K , C , 3(i = 1, 2), then C is numerically 1-connected. So, for at least one C1, say C ,, theinequality K s C, m u s t h o ld . B y our assum ption , K s i s a m p le . Hence,K s C, = 1 or K s C , = 2:(i) K s C , = 1By the algebraic index theorem, we have Cf < O . S in c e K s C , + CI

. is even, C f is

odd. Therefore, C f = — 1 . Since Ks i s ample, C , is irreducible. Since there isno rational curve in S , C , is a non-singular elliptic curve. This elliptic curve isimmersed in Alb (S) by the Albanese mapping of S. This contradicts the assump-tion that Alb(S) is simple.(ii) K ,C , = 2In this case, C? = O. If C , is reducible, C , = C ' + C " . K s C' = K s C" = 1. By(i), such curve does not exist. Hence, C , is an irreducible curve with virtual genus2. Possible cases of C , are as follows:a) an elliptic curve with one node or cusp,b) a non-singular curve of genus 2.

36 Hiro-o Tokunaga

Both cases, by the same argument as that in (i), we can show the existence of anabelian subvariety of A lb (S). This contradicts the assumption.

S tep 2: 0 12„,1 is immersion.Let x be an arbitrary point on S, (9s , x b e its local ring a t x , and X x b e the

maximal ideal of es , x . As well-known, there exists the isomorphism:

es,././111 C Tx* (S)

where Tx* (S ) is co tan g en t sp ace a t x . N ow , consider the exac t sequence ofsheaves,

0 —> 05 (2.1(5 — 2x) —> C5 (21(0 r * 6 9 5,x/X 1 0 Cs (21<5 ) —0 O.

From this exact sequence, we obtain the cohomology exact sequence:

• • H ° V95 (21<s )) * ce Tx*(S) —oH 1(0,(2K , — 2x)) — > • • •

and the problem is reduced to show surjectivity of r*. A proof of surjectivity ofr* consists of two parts.

C ase 1 : The 1-forms coa , co, have simple zero at x.L et co,, co2 be holomorphic 1-forms which d o n o t have zero a t x . S in ce

Alb (S) is simple and by L em m a 1.14. Wa A W 2 , (Ob A (0 2 , and co, A c o , are allnon-trivial holom orphic 2-form s a n d th e y a r e linearly independent. There-fore, (coa A w 2 )(co1 A w2 ), (Wb A W 2)(W 1 A ( 0 2 ) a n d (co, A w2 )2 a r e linearly in-dependent 2-ple 2-forms, and have a t m ost simple zero a t x . This shows r* issurjective.

C ase 2 : One of the 1-forms coa , cob has always 2-fold zero at x.Consider the holomorphic 2-form coa A W . It has 3-fold zero a t x . Now, let

7tx : - - + S be blowing-up at x and Ex b e the corresponding exceptional divisor ofthe first kind. Then In* K s — 3Ex l 0 0 , and, since K s

2 1 2 , w e have

( * K — 3 E ) 2 3 > O .

So, if there exists a numerically 1-connected member of In*K s — 3Ex 1, by Theorem1.11,

hi (Cs ( — (it* K s — 3 E ) ) ) = ( (9 s (27r* K s — 2E )) = O.

Hence r* is surjective, and S tep 2 fo llow s. F o r connectedness o f a member ofrc*Ks — 3Ex 1, we have the following claim.

Claim 2.7. Under the assumption of Proposition 2.6, there exists a numerically1-connected member in 17E* K s — 3Ex l.

Proof o f Claim 2.7. L et s be a n element o f H°(es (K s — 3x)) and C = (s).

Then

(ir x )* C = D + 3Ex

Algebraic surfaces and threefolds 37

where (irx )*C is the total transformation of C . Assume that D = D, + D, whereDi (i = 1, 2) is a positive divisor and put

zl, = Di + (D,Ex )E„.

Then there exist effective divisors C 1, C 2 on S such that (r)*C , = zl i . For theseC 1 , C 2 we have

D, D2 = C1C2 — (Di Ex )(D2 Ex )1D,E x + D2 E„ = 3 .

Hence, if C, > 0 , C 2 > 0 , then by Step 1 and the assumption, we have C, C 2 3.Hence D i D2 1. If one of Ci (i = 1, 2) say C , is a zero divisor, we have

Di D 2 = —(D, Ex )(D2 Ex ).

Since Di i s a positive divisor, D 1E < 0, So, D1 D2 3 . Therefore, Claim 2.7 isproved.

By Step 1 and Step 2, Proposition 2.6 is proved. 0

§ 3 Proof of Theorem (B)

Our proof of Theorem (B) consists of two steps:

Step 1: If 013141 is not birational, its restriction 013K 5 to a general member Sof IK„1 is not also birational.

Step 2: Let D be K v I s where S is a general member of 1Kv l.Then, chi K o .DI is birational.I n o u r case , K y i s am ple, s o b y th e Kodaira vanishing theorem , the

homomorphism

H°((9y(3Ky)) r > H° (S, (95 (K + D))

is surjective. Hence ch,131411s = OlK s + D I, therefore, Theorem (B) follows.Step 1 is proved by Matsuki [7].

Lemma 3.1 (Matsuki [7]). L et V be a threshold which satisfies the assumptiono f Theorem (B) and S be a general member of K . I f 013 ,4 1 is not birational, thenOoKols = Oixs+0 is not also birational.

For a proof, see Matsuki [7].Step 2: If 013„, i is not birational, there exists a Zariski open set V ' which has

the following property:For any point x on V ', there exists a point y on V ' such that

013141(X) — 013K,I(Y) •

The next lemma is crucial.

38 Hiro-o Tokunaga

Lemma 3.2. Let notation be the same as ab o v e . The inequality

dim c H°(6,(1(,, — 2x — 2y)) 2

holds.

Pro o f . By the assumption q 1 (V) 5, there exists at least two holomorphic1-forms with w a (x) = cob (x) = 0, wa (y) = w,,(y) = O. The proof of lemma consistsof three parts.

C ase 1 : For any three linearly independent holomorphic 1-forms co,, w2 , w 3 ,co, A W2 A W 3 * O.

In this case, the proof is easy. Let co, c7) be holomorphic 1-forms such that co,65, w „, oh, are lienarly independent. Then, coa A cob A W # 0, w a A Wb A 6)- # O.Moreover these 3-forms are linearly independent. Hence we have the desiredinequality.

C ase 2 : There exist two linearly independent holomorphic 1-forms 0) 1 , w 2

such that w, A w 2 0 .In this case, by Lemma 1.14, there exists a surjective morphism from V to C,

a curve of genus 2 , and co,, 0)2 are the pullback of holomorphic 1-forms q),, (p2

on C . By Lemma 1.4, we have q 1 (V) g + 3, hence we can choose wa , o4 insuch a w a y th a t wa i s the pullback o f a 1-form on C and c o , is n o t thepullback. By local calculation, we obtain the following claim.

Claim 3.3. For w„, o4 as above, Wa A W b * O.

Claim 3.4. Let w a , cab be as above. Then there exist holomorphic 1-forms w,6) on V with following three conditions.(i) wa , 04, to, C13 are linearly independent.(ii) wa A W b A 0.) * 0, W a A Wb A 6.-.) * O.

(iii) w a A w b A W, wa A oh, A 6 are linearly independent holomorphic 3-forms.

Proo f o f Claim 3.4. S in c e coa A W b * 0 , and Q1, is genera ted by globalsections, the re ex ists a t least one holomorphic 1-form with W a A W b A W O.Assume Wa A W b A (1) 0 for any other holomorphic 1-forms cu. I n this case, bylocal calculation, the restriction of co, A to F, a general fibre of f: V — >C, isidentically zero. Let cob , co,(= w), w 2 . . . . . W k b e a part of a basis of H

°(11,)

which is not in f* H°(,(4 ) . Since g21, is generated by cob , c o , if Wa A W b A

W i * 0 (i = 2, ..., k ), F becomes an elliptic surface. This contradicts the assump-tion K(V) = 3. Therefore, the conditions (i) and (ii) fo llow . For the condition(iii), a proof is similar to that of the condition (ii).

By Claim 3.3, 3.4, we obtain desired inequality for Case 2.C ase 3: For any two linearly independent holom orphic 1-form s co, co',

co A ol O . But there exist three linearly independent holomorphic 1-forms w1 ,0 25 (03 with (01 A (0 2 A W 3 * O.

B y Lem m a 1.14, there exist a ra tiona l m ap ço: V --+S o f V t o a surface ofgeneral type and holom orphic 1-forms 1,, ce3 on S with co, = cp*a i . Consider


the following commutative diagram:

V ------ —>S

where E is a composition of blowing-ups at non-singular centers such thates'o =g 0 9 is a morphism.

L et E b e a n exceptional locus of e a n d E' = e * E . Since irreducible com-ponents of E are all ruled surfaces or 132 and S is a surface of general type, 9 (E) isa curve. Therefore the following commutative diagram holds:

V\E S\O(E).(ply\E

This shows that, generically, 9 and 9 are the sam e fibrations. In the follow-ing argument, we assume tha t the Zariski open set in Lemma 3.2 satisfies V' cV\E', and at least one of 1-forms oh, and W contained in 9*H

°(f2k).

i) The case in which both 1-forms oh, and oh, are contained in 9*Fr(flk).B y th e assumption, w a A co, 0 a n d these a re linearly independent. Let

w , Co" be linearly independent holomorphic 1-forms which is not contained in9*1

-1°(Q , ) . It is c lea r tha t co a A co, A w # 0 and co a A cob A 6) O . S o , it is

enough to show tha t w e can take co, (7) in such a w ay tha t COa A Wb A co andwa A wb A di are linearly independent. Now, fix co(e 9* H

°( 4 ) ) . Let (75 be anyother holomorphic 1-form which is not contained in co*I-1° (Q ) . I f wa A 04 A CO

a n d (Oa A W b A (7) a re linearly dependent fo r any 1-form 6 , th is shows th a t ageneral fib re of 9 i s a n elliptic curve. T his contradicts th e fac t tha t K(17

) =K(v) = 3.

ii) The case in which wa is contained in co*I-P(S2) but co„ is not.By the assumption, we have co„ A oh, # O.ii-a) The case in which there exists a holomorphic 1-form w such that wa ,

cob , w are linearly independent and oh, A 0 4 A CO —= O.By taking another fibration, this case is reduced to the case i).ii-b) There exist no holomorphic 1-forms as those in ii-a).This case is clear.By i) and ii), w e obtain the desired inequality in c a s e 3 . By Case 1, 2, 3,

Lemma 3.2 follows.

Corollary 3.5. Let S, x, y be the same as abov e . Let 7E: S —> S be a composi-tion of blowing-ups at x and y. Then, In*D — 2Ex — 2Ey 1 0 where D = K v1 5 ,and Ex , Ey are the exceptional divisors of the f irst kind appearing in the blowing-ups,respectively.

40 Hiro-o Tokunaga

Pro o f . Consider the exact sequence of sheaves,

0 —> —> & ( K ) C s (D) 0 .

Taking cohomology sequence, and by Lemma 3.2, Corollary 3.5 follows.

Proposition 3.6. 0 1„,,D1 is birational.

Pro o f . Assume 0 1„5 , D1 is n o t b ira tio n a l. S in c e c 6rimcvils = 013141 isno t a lso b ira tiona l. L e t x , y b e points with O IK , „ 1(x) = O lK s .„0 (y). Then, byCorollary 3.5, In*D — 2Ex — 2Ey 1 0 Ø . W e m a y assume that a Zariski open setS n V ' is contained in SVC w here is defined as follows:

= {C is an irreducible curve on S with DC < 2D 2 , C 2 < 0}

By Proposition 1.10, is a finite set. By Lemma 1.5, we have

(n*D — 2Ex — 2Ey )2 12 — 8 = 4 > 0 .

Hence the problem is reduced to show connectedness o f a member of In*D —2Ex — B y the assumption, a numerically connected member does not exist.Let C be an effective divisor linearly equivalent to D and a defining section s of Cis contained in H

°(e s (D — 2x — 2y)). Put

1r* C = A + 2E x + E y( n * C is a total transform)

and let A = A , + A2 be a decomposition into positive d iv iso rs . Put

A i = A i + (A i Ex )Ex + (A i Ey )Ey( i = 1, 2)

Then there exist effective divisors Ci (i = 1, 2) on S with n*Ci = A , C 1 + C2 = C.Therefore, we have

, A 2 = C1 C2 — (A 1 Ex )(A 2 E,c ) — (A 1 Ey )(A 2 Ey )tA ,E x + A 2 Ex = 2, A,Ey + A 2 Ey = 2 .

Therefore, the inequality

A 1 A 2 C 1 C2 —2

holds and the equality holds if and only if

A i Ex = A i E,= 1 (i = 1, 2) .

Moreover, i f C, = 0 , then, since A , i s a positive divisor, we have A ,E x 0 ,A 2 Ex 0 a n d at least one of the inequalities is strictly inequality, hence, A 1 A2

3. Hence both C , and C2 are positive divisors. Moreover, since we can showC1 C2 1, A is not numerically connected if and only if one o f the followingsholds:

(1) CI C2 = 1 A ,Ex = A ,E y = 1 , A ,E x = 2 , A 2 Ey = 0

A ,E„ = A ,E y = 1 , A 2 E„ = 0 , A 2 Ey = 2

LII


A 2 E„ = A 2 Ey = 1 , A ,E „= 2 , A ,E y = 0

A2 Ex = A 2 E y = 1 , A ,E x = 0 , A ,E y = 2

A,E„ = A i Ey = 1 (i = 1, 2)

(2) C, C2 = 2 A,Ex = A i Ey = 1 (i = 1, 2)

Case ( 1 ) : The case in which C1 C2 =1, A 1 Ex =A 1 Ey =1, A 2 E = 2, A 2 Ey =0.This shows that X , y are smooth points on C ,, and x is a double point on

C2. H e n c e C 1 C2 2 , and th is is a contrad ic tion . Similarly, we can show thatother cases in (1) do not occur.

Case ( 2 ) : If D C , 3 (i = 1, 2), we can prove C 1 C23 . H e n c e for at leastone of Ci , say C,, DC, 2.

Case (a): DC, = 1Since we have C , + C2 D, w e have C, C2 = 2 , C,(C, + C 2 ) = 1 and C = 0 ,

and the virtual genus of C, is integer, so this case does not occur.Case (fl): DC, = 2By a similar calculation in the case (a), we have Cf = 0, and the virtual genus

of C , is 3. M oreover, w e can prove that C 1 is irreducible. Therefore, S has astructure of a fibre space with curve of genus 3 as a general fibre.

Claim 3.7. A general member o f 1K ,I does not hav e a structure of a fibrespace with curve of genus 3 as a general fibre.

Proof of Claim 3.7. Assume a general member S o f 1K , has a structure of afibre space p: S —■ C with curve of genus 3 as a general fibre. By Leray's spectralsequence, we have an exact sequence:

0 —+ I-11 (C, 111(S, e s ) —> H° (C, R 1 p,,,(9,) —> 0 .

Hence, 111 ((95 ) = g + 2 o r g + 3 where g is the genus of C . If h i (e s ) = g + 3, wecan show that S = C x F, g(F)= 3. Assume that the following Claim 3.8 holds.

Claim 3.8. L et <99 = {S,}, c 4 , 4 = { t e CIIt < e l be a fam ily o f general mem-ber o f IK,,1 with So = S. By a theorem of Lefshetz, holomorphic 1-forms of So areobtained by restriction of holomorphic 1-forms on V to So . L et coi ls ° , . ,Wg+ 2150 w 9 + 31s0 (o r (0 1 iso , . . . , wslso W 9+1150 , W g + 2 0 be holomorphic 1-forms and, 1

W11S0' • • • , Woo b e the pullback o f holomorphic 1-forms o n C = Co . Then, coils ,(i = 1, g ) are also the pullback of holomorphic 1-forms on C, where C, is curve ofgenus g.

By Claim 3.8, we have co, A coi l s , 0 ( i,j = 1, g). Since 0 1„ 1 i s a finitemorphism, w A co; 0 o n V. By Lemma 1.13, there exists a morphism f : V Cwhere C is a curve o f genus g. Therefore, by Lemma 1.4, we have h i (e ,)>g + 2. I n c a s e h1 V 9) = g + 3, a general S is the product C x F of curves C andF with g(F) = 3. S o the similar facts in C laim 3.8 hold for wa + ,, 0 g + 2 , cog + 3 .Hence, we have w i A co; 0 (i,j = I, ..., g) a n d coo , A w g + k 0 (h, k = 1, 2, 3).Since Q1, is generated by global sections, this is a contradiction.

42 Hiro-o Tokunaga

Proof o f Claim 3.8. F irs t , w e sh o w th a t co n d itio n o f Theorem 1.12is s a t is f ie d . F o r p: S C, H2(C, 0c ) —> H2 (S, p*Coc ) is injective. becauseH 2 (C, Oc ) = O. N e x t, w e sh a ll p ro v e pf: 11 1(C, Oc ) (S, p*Oc ) is surjective.By Leray's spectral sequence, we have an exact sequence:

I-11 (C, 111(S, p*Oc ) —> W(C, R 1 p p*Oc ) _ o .

Therefore, it is enough to show H ° (C, 12 1 p* p* 9c ) = 0. By the projection formula,R 1 p* p* Oc Z R 1 p ( 9 0 O c . In case h

°(C, R 1 p &5) = 2, by Theorem 1.9, we have

R 1 N e s 0 Oc Z(13*(0 51 )'1 O c Z (Cc Ci Cc CD S v ) 0 e c

where e is an invertible sheaf with 111(C, e [Kc ]) = O. S in c e g(C) 2, we haveW (C , p* Os e c ) = O.

In case h° (C, R 1 p & ) = 3, by Theorem 1.9, we have

R 1 p (9 ® Oc Z Oc 0 ec 0 Oc

Hence, we have H ° (C, p * (9 s 0 O c ) = O.Therefore, pr is surjective. By Theorem 1.12, there exists a family of curves4 ' = f t e C t < e' < el w ith Co = C, and a holomorphic map with Olso =

p which satisfies the following diagram:

ois. P

4'

F or families 9°, and (6, there exist families of Albanese tori dey (91 and f (e)such that the following diagram is commutative:

„glee G.99 )\

I)44 0)

/1' A'

Let S, be a general member of 1K, 1, so, we have

we (9°) = Alb (S0 ) x 4 ' = Alb (V ) x 4 '.

Since there exists no continuous family of abelian subvarieties in a fixed abelianvariety, we have / 0 ) = i(co ) x S . This proves Claim 3.8.

By the argum ent as above, there exists a numerically 1-connected member inD — 2E„ — 2Ey l. Therefore, Proposition 3.5 follows.

DEPARTMENT OF MATHEMATICSKYOTO UNIVERSITY


References

[ 1 W. Barth, C. Peters, A. Van de Ven, Compact Complex Surfaces, Springer, 1984.[ 2 ]

A. Beauville, L'inégalité 2q — 4 pour les surfaces de type général, Bull. Soc. Math. France,110 (1982), 343-346.

[ 3 A. Beauville, Complex Algebraic Surfaces, London Math. Soc. L.N.S. 68.[ 4 ]

E. Bombieri, Canonical models of surfaces of general type, Publ. Math. Inst. Hautes Etud. Sci.,42 (1973), 171-219.

5 ]

T. Fujita, On Kdhler fibre spaces over curves, J. Math. Soc. Japan, 30 (1978), 779-794.[ 6 ]

E . H orikaw a, On Deformations of Holomorphic Maps III, Math. ann., 222 (1976), 275-282.

[ 7 ]

K . M atsuki, O n pluricanonical maps for threefolds of general type, J . Math. Soc. Japan, 38(1986), 339-359.

[ 8 ] Z . Ran, On subvarieties of Abelian Varieties, Inv. math., 62 (1981), 459-479.[ 9 ] Z . Ran, The structure of Gauss-like maps, Compo. Math., 52 (1984), 171-177.[ 1 0 ] K . U e n o , Classification Theory of Algebraic Varieties and Compact Complex Spaces, Springer

L.N.M. 439 (1975)

Author's present addressDepartment of MathematicsFaculty of ScienceKochi UniversityKochi, 780 JAPAN

On algebraic surfaces and threefolds whose cotangent ...

Documents