A splitting theorem for rank two vector bundles on projective spaces in positive characteristic
Hideyasu Sumihiro and Shigehiro Tagami
(Received April 21, 2000)
Abstract. We shall prove the following splitting theorem for rank two vector bundles Eon then-dimensional projective spacePn nV4in positive characteristic. LetPbe a 4- or 5-dimensional projective linear subspace ofPnandEEjPthe restriction ofE toP. ThenEsplits into line bundles if and only if the ®rst cohomology of the sheaf of endomorphisms of E vanishes.
0. Introduction
Let E be a rank two vector bundle on the n-dimensional projective space Pkn nV4 de®ned over an algebraically closed ®eld k.
In [4], H. Sumihiro showed the following theorem in the case of char k0.
Theorem 0.1. Let P be a 4- or 5-dimensional projective linear subspace of Pkn and EEjP the restriction of E to P. Then E splits into line bundles if and only if H1 P;End E 0.
The aim of this article is to prove that this theorem holds also true in char kp>0. The proof is almost the same as the one for chark0, namely, it is obtained by studying some geometric structures of the Hilbert scheme of Pkn at determinantal subvarieties. In charkp>0, however, since we cannotuse the Kodaira vanishing theorem and the Le-Potier vanishing theorem (cf. [1], [3]), we have to observe some vanishings of cohomologies appearing in [4]
carefully.
1. Preliminaries
We ®rst recall the de®nition and some properties of determinantal varieties associated to rank two bundles (cf. [4]).
2000 Mathematics Subject Classi®cation. 14C05, 14F05, 14F17.
Key words and phrases. Vector bundles, Hilbert Schemes, Vanishing Theorems.
1.1. De®nition of determinantal varieties. Let E be a rank two vector bundle on Pkn de®ned over an algebraically closed ®eld k with arbitrary characteris- tic, p:P E !Pkn the projective bundle associated to E over Pkn, LE the tautological line bundle on P E and let GGrass H0 E;m1 be the Grassmann variety which parametrizes m1-dimensional linear subspaces of H0 Pkn;E, where n2m (resp. n2m1). We assume that E is very ample, i.e., LE is a very ample line bundle. Then we can take s hs1;s2;. . .;sm1i AG siAH0 Pkn;E satisfying the following condition
1) YsD1VD2V VDm1 is a smooth closed subscheme of P E
of pure codimension m1,
2) W s1VW s2V VW sm1 q,
where Di is the tautological divisor onP Ede®ned by si andW siis the zero locus of si in Pkn 1UiUm1.
Let Xsp Ys. Then we can show that Xs is a closed subscheme of Pkn which is isomorphic to Ys through p with the following de®ning equations:
si5sj0 1UiUjUm1:
Definition 1.1. We call the closed subscheme Xs of Pkn the determinantal variety associated to E de®ned by sAG.
ThoughXsdepends on the choice ofsAG, we call a closed subvarietyXsa determinantal variety associated to E.
As for determinantal varieties, we obtain the following.
Theorem 1.1. Let the notaion be as above.
1) U fsAGjs satis®es the condition g is a Zariski open subset of G.
2) There exists a closed subscheme X of PknU such that the second projection q:XHPknU!U is faithfully ¯at and Xsq 1 s for any sAU. Thus smooth determinantal varieties associated to E form a smooth family over an open subset of G.
When n4 or 5, let IX be the de®ning ideal of a determinantal subvariety X in Pn. Then IX has the following resolution by vector bundles.
Lemma 1.2. In the above notation, there exists an exact sequence
0!E c1 !03 OPn c1 !IX !0;
where c1 is the ®rst Chern number of E and E is the dual bundle of E.
Proof. Let s fs1;s2;s3g be a setof global sections of E which de®nes the determinantal subvariety X. Then we can de®ne homomorphisms
a:03 OPnCei5ej7!si5sjA 52 E 1Ui<jU3;
b:ECf 7!f s3e15e2 f s2e15e3f s1e25e3 A 03 OPn;
where fei5ejg is a basis of 03 OPn. Then itsu½ces to verify locally on Pn that the following sequence is exact:
0!E !b 03 OPn !a IXnO c1 !0:
r 1.2. Tangent bundles and normal bundles of determinantal varieties. In the following subsections, we consider the case n4 or 5, i.e., m2.
Let E be a very ample rank two bundle on Pkn and X a determinantal variety associated to E which is isomorphic through p to the complete in- tersection Y in P E of the tautological divisors fDiji1;2;3g.
Let H be the restriction of a hyperplane of Pn to X and Dthe restriction of a tautological divisor of P E to X through the isomorphism p.
Then we have the following commutative diagram of exact sequences:
0 0
??
?y
??
?y TP E=P???y njY !@ OX 2D c1H
??
?ya
0 ! TY ! TP EjY ! NY=P E ! 0
???y
??
?y
0 ! TX ! TPn???yjX ! NX=Pn ! 0;
??
?y
0 0
where a is the injection induced by the snake lemma. Since NY=P EF 03 OX D, we obtain the following.
Proposition 1.3. There exists an exact sequence
0!OX 2D c1H !03 OX D !NX=Pn!0:
1.3. Hilbert Schemes. LetHilbbe the Hilbert scheme ofPn. Let j:UCs7!
XsAHilb be the morphism induced by Theorem 1.1. Let Aut E be the automorphism group of E. Then Aut E is a reduced connected linear algebraic group of dimension dimH0 End E.
For every element gAAut E and shs1;s2;s3i AG, we de®ne gshg s1;g s2;g s3i;
where g si is the composite ofsi with g. Then itde®nes an action of Aut E
on G and we have
gsi5gsjdet gsi5sj 1UiUjU3;
where det:Aut ECg7!det gAkknf0g is the determinant character.
Hence XgsXs. Therefore Aut E acts on U and j is an orbitmorphism, i.e., j is constant on any orbit O s fgsjgAAut Eg.
Then we have the following.
Lemma 1.4. The stabilizerStab sof sAU coincides with the multiplicative group k.
As a trivial corollary of the above lemma, we observe that every orbit has the same dimension dim Aut E=k, i.e., dimO s dimH0 End E 1 sAU. Hence the action of Aut E on Uis closed, i.e., every orbitis closed in U.
2. Proof of the theorem
2.1. Since itis well-known thatE splits into line bundles if and only if EEjP splits into line bundles, where P is a 4- or 5-dimensional linear subspaces of Pn, we may assume that E is a rank two vector bundle onPn (n being either 4 or 5) (cf. [2]). In addition after multiplying E by a suitable ample line bundle, we may assume that E is a very ample vector bundle enjoying Hi EnKPn 0 1UiU4, where KPn is the canonical line bundle of Pn.
By Proposition 1.3, we have the following exact sequence 0!H0 OX 2D c1H !03 H0 OX D !H0 NX=Pn
!H1 OX 2D c1H !03 H1 OX D:
Now we recall Y D1VD2VD3. Consider the canonical exact sequence 0!OP E D c1H !OP E 2D c1H !OD1 2D c1H !0;
1
from which we obtain the following exact sequence:
0!H0 OP E D c1H !H0 OP E 2D c1H !H0 OD1 2D c1H
!H1 OP E D c1H !H1 OP E 2D c1H !H1 OD1 2D c1H
!H2 OP E D c1H:
Since Hi OP E D c1H Hi E 0UiU4and we can show thatH0 E
0 and Hi E Hn i EnKPn 0 i1;2 by our assumption, it turns outthatHi OP E 2D c1HFHi OD1 2D c1H i0;1.
In addition considering the following exact sequences similarly 0!OD1 D c1H !OD1 2D c1H !OD1VD2 2D c1H !0;
0!OP E c1H !OP E D c1H !OD1 D c1H !0;
2
0!OD1VD2 D c1H !OD1VD2 2D c1H !OY 2D c1H !0;
0!OD1 c1H !OD1 D c1H !OD1VD2 D c1H !0;
0!OP E D c1H !OP E c1H !OD1 c1H !0;
3
we obtain isomorphisms Hi OD1 2D c1HFHi OD1VD2 2D c1H
and Hi OD1VD2 2D c1HFHi OY 2D c1H i0;1 because Hi OP E D c1H 0 0UiU4. Summing up the above, we con- clude that Hi OX 2D c1HFHi OP E 2D c1HFHi Pn;S2 E c1
i0;1.
On the other hand, since there exists an exact sequence 0!OPn!End E !S2 E c1 !0;
we have a canonical isomorphism H1 S2 E c1FH1 End E and dimH0 S2 E c1 dimH0 End E 1.
Moreover we easily see that dimH0 OX D dimH0 E 3.
Summarizing the above, we get the following proposition.
Proposition 2.1. With the above assumption, if H1 End E 0, then dimH0 NX=Pn 3 dimH0 E 3 dimH0 End E 1:
Remark 2.1. When char k0, we get Hi EFHn i EnKPkn 0 for 0UiUn 2 by the Le-Potier vanishing theorem. So we do not need the assumption Hi EnKPn 0 1UiU4in Proposition 2.1. Also the proof itself becomes slightly simpler because we can use the vanishing theorems.
2.2. Let Hilb0 be an irreducible componentof Hilb containing the closure j U of j U in Hilb and TXs;Hilb the Zariski tangent space of Hilb at Xs. Then itis known thatTXs;HilbFH0 NXs=Pn. So we have the following proposition.
Proposition 2.2. Under the same assumptions in Proposition 2.1, if H1 End E 0 then
1) Hilb0 coincides with j U.
2) Hilb0 is smooth at the determinantal subvarieties associated to E.
Proof. It is su½cient to prove that dimj U dimH0 NXs=Pn for any determinantal surfaceXs. Using the exact sequence in Proposition 1.3, we see that j 1 j s sAU consists of ®nitely many orbits. Hence
dimj U dimU dimO s
dim Grass H0 E;3 dimH0 End E 1
3 dimH0 E 3 dimH0 End E 1:
So our assertion follows by Proposition 2.1. r
2.3. LetPGL n1;k be the automorphism group of Pn and let Ts:PnC x7!sxAPn be the transformation of Pn de®ned by sAPGL n1;k.
Suppose thatH1 End E 0. Then it follows from Proposition 2.2 that sj U j U for every element sAPGL n1;k. Since j U is a con- structible set, there exist two elements s;tAU satisfying Xs sXt, where Xs s is the determinantal subvariey associated to Ts E de®ned by s s hTs s1;Ts s2;Ts s3i. Consider the resolutions of the de®ning ideal sheaves IXt of Xt and IXs s of Xs s respectively (cf. Lemma 1.2):
0 ! E ! 03 OPn ! IXtnO c1 ! 0
c
??
?y F
??
?y
0 ! Ts E ! 03 OPn ! IXs snO c1 ! 0:
Then it is observed that there exists an isomorphism c:03 OPn!03 OPn such that c makes the diagram in commutative and so we see that Ts E is isomorphic to E, i.e., E is a homogeneous vector bundle. Since every homo- geneous bundle on Pn of rank r<n is a directsum of line bundles even if char kp>0 (cf. [2]), we can complete the proof of Theorem 0.1.
References
[ 1 ] J. Le-Potier, Annulation de la cohomologie aÁ valeurs dans un ®bre vectoriel holomorphe positif de rang quelconque, Math. Ann. 218 (1975), 35±53.
[ 2 ] C. Okonek, M. Schneider and H. Spindler, Vector Bundles on Complex Projective Spaces, Progress in Math. 3, BirkhaÈuser (1980).
[ 3 ] B. Shi¨man and A. Sommese, Vanishing Theorem on Complex Manifolds, Progress in Math. 56, BirkhaÈuser (1985).
[ 4 ] H. Sumihiro, Determinantal varieties associated to rank two vector bundles on projective spaces and splitting theorems, Hiroshima Math. J. 29 (1999), 371±434.
Department of Mathematics Graduate School of Science
Hiroshima University Higashi-Hiroshima 739-8526, Japan H. Sumihiro: [email protected]
S. Tagami: [email protected]