2011
年度 修士論文Characterizations of projective spaces and hyperquadrics for varieties with Picard number one
早稲田大学大学院基幹理工学研究科 数学応用数理専攻
5110A026-5
鈴木 拓 指導教員名 楫 元Abstract. The purpose of this paper is to give a proof of Kov´acs conjecture for varieties with Picard number one by applying a method of slope-stabilities of sheaves.
1 Introduction
Characterizations of projective spaces and hyperquadrics have been studied in several ways. The purpose of this paper is to give a different characterization of them. We will prove the following theorem:
Theorem 1.1. Let X be a smooth complex projective variety of dimension n with Picard number one, and assume that there exist an ample vector bundle E of rank r on X, an integer 1 ≤ p ≤ r and an inclusion ∧
pE , → ∧
pT
X. Then either X ∼ = P
nor X ∼ = Q
p, where Q
pdenotes a smooth hyperquadric in P
p+1.
The first important result on characteriztions of them is S. Mori’s proof of the Hartshorne conjecture by using a method of rational curves on varieties in 1979:
Theorem 1.2 ([Mor79]). Let X be a smooth projective variety of dimension n over an algebraic closed field, and assume that the tangent bundle T
Xis ample.
Then X ∼ = P
n.
Generalizing Mori’s result, it was conjectured that X ∼ = P
nif the tangent bundle T
Xcontains an ample vector bundle E for a smooth complex projective variety X . In this direction, J. Wahl obtained the case rk(E ) = 1 in 1983 ([Wah83]), and F. Campana and T. Peternell obtained the case rk(E ) ≥ n − 2 in 1998 ([CP98]). Finally M. Andreatta and J. Wi´sniewski proved the case of arbitrary rank in 2001:
Theorem 1.3 ([AW01]). Let X be a smooth complex projective variety of dimension n, and assume that T
Xcontains an ample vector bundle E . Then (X, E ) ∼ = (P
n, O
Pn(1)
⊕r) or (X, E ) ∼ = (P
n, T
Pn).
In 2006, C. Araujo gave a different proof of Theorem 1.3 using a method of the variety of minimal rational tangents. Moreover a generalization of Theorem 1.3 was conjectured by S.J. Kov´acs:
Conjecture 1.4 (Kov´acs). Let X be a smooth complex projective variety of dimension n, and assume that there exist an ample vector bundle E of rank r on X, an integer 1 ≤ p ≤ r and an inclusion ∧
pE , → ∧
pT
X. Then either X ∼ = P
nor X ∼ = Q
p.
On this conjecture, Araujo, S. Druel and Kov´acs proved the case E = L
⊕rfor some line bundle L in 2008:
Theorem 1.5 ([ADK08]). Let X be a smooth complex projective variety of
dimension n, and assume that there exist an ample line bundle L on X and
an integer p ≥ 1 such that H
0(X, ∧
pT
X⊗ L
−p) 6= 0. Then either (X, L ) ∼ =
(P
n, O
Pn(1)) or (X, L ) ∼ = (Q
p, O
Qp(1)).
Note that our Theorem 1.1 is the case where the Picard number ρ(X ) of X is one in Conjecture 1.4. Now one can easily see that the assumption of Theorem 1.1 implies that X has a minimal dominating family H and degf
∗E for every [f ] ∈ H is equal to r or to r + 1 (see Section 3). Here I should mention that K.
Ross also claimed the same statement as Theorem 1.1 in [Ros10]. But it seems that the existence of a morphism E → T
Xis implicitly assumed in the second case of the proof there. Then we give a different proof of Theorem 1.1. We use a method of slopes, semistabilities and Harder-Narasimhan filtrations of sheaves, which was given by [ADK08].
2 Preliminaries
First we recall about rational curves on varieties.
Definition 2.1. Let X be a smooth complex projective variety. An irreducible component H ⊂ RatCurves
n(X ) is called a minimal dominating family if H- curves dominate X and the subfamily of H -curves passing through a general point x ∈ X is proper, where an H -curve means a curve parametrized by H . Lemma 2.2 ([Kol96, IV.2.4, 2.9 and 2.10]). If X is uniruled, then X has a minimal dominating family. If H is a minimal dominating family and [f ] ∈ H is a general member, then f
∗T
X∼ = O(2) ⊕ O (1)
⊕d⊕ O
⊕n−d−1, where d =
−degf
∗(K
X) − 2.
Definition 2.3. Let H be a fixed minimal dominating family. Two points x, y ∈ X are H-rationally equivalent if they can be connected by a chain of H- curves. By [Kol96, IV.4.16] there exist an open dense subset X
◦⊆ X , a normal variety Y
◦and a proper morphism π
◦: X
◦→ Y
◦whose fibers are H -rationally equivalent class. This π
◦is called the H-rationally connected quotient. X is said to be H -rationally connected if Y
◦is a point.
Lemma 2.4 ([KMM90]). Let H be a minimal dominating family. Then X is H-rationally connected if ρ(X ) = 1.
The following result is the key lemma to prove that varieties are isomorphic to projective space.
Lemma 2.5 ([ADK08, 2.7]). Let X be a smooth complex projective variety, H a minimal dominating family of rational curves on X, and π
◦: X
◦→ Y
◦the H-rationally connected quotient of X . Suppose that T
Xcontains a subsheaf D such that f
∗D is an ample vector bundle for a general member [f ] ∈ H. Then, after shrinking X
◦and Y
◦if necessary, π
◦becomes a projective space bundle.
Next we recall some facts about slopes of sheaves.
Definition 2.6. Let X be a projective variety of dimension n and H an ample line bundle on X. The slope of a non-zero torsion-free sheaf F with respect to H is defined by
µ
H(F ) = c
1(F ) · c
1(H )
n−1rk(F ) .
A torsion-free sheaf F is µ
H-semistable if µ
H(E ) ≤ µ
H(F ) for any subsheaf E of F .
Lemma 2.7 ([HN75] or [HL97, 1.3.4]). Let X and H be as above. Any torsion- free sheaf F on X has a filtration
F = F
0) F
1) · · · ) F
h= 0,
such that Q
i= F
i/F
i+1is µ
H-semistable for 0 ≤ i ≤ h − 1, and µ
H(Q
0) <
· · · < µ
H(Q
h−1). This is called the Harder-Narasimhan filtration of F . Lemma 2.8 ([HL97, 3.1.4 and 3.2.10]). Let X be a normal projective variety over an algebraic closed field of characteristic zero, H an ample line bundle on X . Suppose that torsion-free sheaves F and G are µ
H-semistable. Then F ⊗ G and ∧
mF are also µ
H-semistable if they are torsion-free.
Lemma 2.9 ([MR82] or [HL97, 7.2.1]). Let X be a smooth projective variety of dimension n ≥ 2, H an ample line bundle on X, and O(1) a very ample line bundle on X . If a torsion-sheaf F on X is µ
H-semistable, then F |
His also µ
H|H-semistable for a sufficiently large integer a and a general hypersurface H ∈ |O(a)|.
3 Proof of Theorem 1.1
Proof. The result is clear if n = 1, so we assume n ≥ 2. First it immedi- ately follows from [Miy87] that X is uniruled. Hence X has a minimal dom- inating family H ⊂ RatCurves
n(X ) of rational curves (Lemma 2.2). Let [f ] be a general member of H . Then f
∗T
X∼ = O(2) ⊕ O(1)
⊕d⊕ O
⊕n−d−1where d = −degf
∗(K
X)−2, and f
∗(∧
pE ) ⊆ f
∗(∧
pT
X). Since E is ample, we have pos- itive integers a
1≥ · · · ≥ a
r≥ 1 such that f
∗E ∼ = L
ri=1
O (a
i). Then it follows that a
1+ · · · + a
p≤ p+ 1. Therefore f
∗E ∼ = O(1)
⊕ror f
∗E ∼ = O (2) ⊕O (1)
⊕r−1. Since degf
∗E depends only on H but not on the choice of [f ], this is true for all [f] ∈ H .
Case 1. f
∗E ∼ = O(1)
⊕rfor every [f ] ∈ H.
In this case, the result is proved in [Ros10]. By the induction on r we can reduce to the result of [ADK08] (Theorem 1.5).
Case 2. f
∗E ∼ = O(2) ⊕ O(1)
⊕r−1for every [f ] ∈ H.
In this case, we prove that X is isomorphic to projective space. First we prove the following claim.
Claim 3.1. Fix a very ample vector bundle H on X . Then T
Xcontains a
subsheaf D (possibly not locally-free) such that µ
H(D) ≥ µ
H(E ).
Proof. We write µ simply instead of µ
H. Now we take the Harder-Narasimhan filtration of T
X(Lemma 2.7):
T
X= D
0) D
1) · · · ) D
h−1) D
h= 0,
with µ-semistable Q
i= D
i/D
i+1and µ(Q
0) < · · · < µ(Q
h−1). We would like to show that D = D
h−1= Q
h−1satisfies the condition. Let a be a sufficiently large integer, and let C be an intersection curve of n − 1 general hypersurfaces in |H
⊗a|. Let E
0= E |
C, D
i0= D
i|
C, Q
i0= Q
i|
C, T
X0= T
X|
C, and µ
0= µ
H|C. By Bertini’s theorem, we may assume that C is smooth, and that D
i0and Q
i0are locally-free. Moreover by Lemma 2.9 we may assume that Q
0iis µ
0-semistable.
By the property of exterior products, the short exact sequence 0 → D
i+10→ D
i0→ Q
i0→ 0
yields a filtration
∧
tD
i0= F
(t,i,0)⊇ F
(t,i,1)⊇ · · · ⊇ F
(t,i,t+1)= 0
for any non-negative integer t, such that F
(t,i,j)/F
(t,i,j+1)∼ = ∧
t−jQ
0i⊗ ∧
jD
i+10. For any non-negative integers s, t, i, j and vector bundle A on C, again the short exact sequence
0 → A ⊗ F
(t,i,j+1)→ A ⊗ F
(t,i,j)→ A ⊗ ∧
t−jQ
0i⊗ ∧
jD
i+10→ 0 yields a filtration
∧
s(A ⊗ F
(t,i,j)) = G
(s,t,i,j,A,0)⊇ G
(s,t,i,j,A,1)⊇ · · · ⊇ G
(s,t,i,j,A,s+1)= 0 such that G
(s,t,i,j,A,k)/G
(s,t,i,j,A,k+1)∼ = ∧
s−k(A ⊗ ∧
t−jQ
i0⊗ ∧
jD
i+10) ⊗ ∧
k(A ⊗ F
(t,i,j+1)).
Now let q = rk(∧
pE ) = ¡
rp
¢ . The assumption of Theorem 1.1 implies ∧
q∧
pE ⊆ ∧
q∧
pT
X. By the generality of C we may assume that ∧
q∧
pE
0⊆ ∧
q∧
pT
X0. Note that ∧
q∧
pT
X0= G
(q,p,0,0,O,0). First from the filtration G
(q,p,0,0,O,∗), we have an integer k
0such that ∧
q∧
pE
0⊆ G
(q,p,0,0,O,k0)and ∧
q∧
pE
06⊆ G
(q,p,0,0,O,k0+1). Then an inclusion
∧
q∧
pE
0, → ∧
q−k0(∧
pQ
00⊗ ∧
0D
10) ⊗ ∧
k0F
(p,0,1)is induced. Next from the filtration ∧
q−k0(∧
pQ
00⊗ ∧
0D
10) ⊗ G
(k0,p,0,1,O,∗), we have an integer k
1such that ∧
q∧
pE
0⊆ ∧
q−k0(∧
pQ
00⊗ ∧
0D
10) ⊗ G
(k0,p,0,1,O,k1)and ∧
q∧
pE
06⊆ ∧
q−k0(∧
pQ
00⊗ ∧
0D
10) ⊗ G
(k0,p,0,1,O,k1+1). Then an inclusion
∧
q∧
pE
0, → ∧
q−k0(∧
pQ
00⊗ ∧
0D
10) ⊗ ∧
k0−k1(∧
p−1Q
00⊗ ∧
1D
10) ⊗ ∧
k1F
(p,0,2)is induced. After a similar procedure of p steps, we have non-negative integers b
0, . . . , b
psuch that b
0+ · · · + b
p= q and an inclusion
∧
q∧
pE
0, → ∧
b0(∧
pQ
00⊗∧
0D
10)⊗∧
b1(∧
p−1Q
00⊗∧
1D
10)⊗· · ·⊗∧
bp(∧
0Q
00⊗∧
pD
10).
Next for 0 ≤ m ≤ p, let B
m= N
t6=m
∧
bt(∧
p−tQ
00⊗ ∧
tD
10). Then
∧
q∧
pE
0, → B
m⊗ ∧
bm(∧
p−mQ
00⊗ ∧
mD
10).
Note that ∧
bm(∧
p−mQ
00⊗ ∧
mD
10) = G
(bm,m,1,0,∧p−mQ00,0). From the filtration B
m⊗ G
(bm,m,1,0,∧p−mQ00,∗), we have an integer l
0and an inclusion
∧
q∧
pE
0, → B
m⊗ ∧
bm−l0(∧
p−mQ
00⊗ ∧
mQ
10⊗ ∧
0D
20) ⊗ ∧
l0(∧
p−mQ
00⊗F
(m,1,1)).
Next from the filtration B
m⊗∧
bm−l0(∧
p−mQ
00⊗∧
mQ
01⊗∧
0D
20)⊗G
(l0,m,1,1,∧p−mQ00,∗), we have an integer l
1and an inclusion
∧
q∧
pE
0, → B
m⊗ ∧
bm−l0(∧
p−mQ
00⊗ ∧
mQ
10⊗ ∧
0D
20)
⊗ ∧
l0−l1(∧
p−mQ
00⊗ ∧
m−1Q
10⊗ ∧
1D
20)
⊗ ∧
l1(∧
p−mQ
00⊗ F
(m,1,2)).
After a similar procedure of m steps, we have non-negative integers a
m0, . . . , a
mmsuch that a
m0+ · · · + a
mm= b
mand an inclusion
∧
q∧
pE
0, → B
m⊗ ∧
am0(∧
p−mQ
00⊗ ∧
mQ
01⊗ ∧
0D
20)
⊗ ∧
am1(∧
p−mQ
00⊗ ∧
m−1Q
10⊗ ∧
1D
20)
· · ·
⊗ ∧
amm(∧
p−mQ
00⊗ ∧
0Q
10⊗ ∧
mD
20).
Repeating a similar procedure for m = 0, 1, . . . , p, we have integers {a
αβγ}
α,β,γ≥0,α+β+γ=psuch that P
a
αβγ= q and an inclusion
∧
q∧
pE
0, → O
α,β,γ
∧
aαβγ(∧
αQ
00⊗ ∧
βQ
10⊗ ∧
γD
20).
After a similar procedure of h−1 steps, finally we have integers {a
α0...αh−1}
αi≥0,Pαi=psuch that P
a
α0...αh−1= q and an inclusion
∧
q∧
pE
0, → O
α0,...,αh−1
∧
aα0...αh−1(∧
α0Q
00⊗ ∧
α1Q
01⊗ · · · ⊗ ∧
αh−1Q
h−10) =: Q
0. Now since each Q
0iis µ
0-semistable, so is Q
0(Lemma 2.8). Therefore µ
0(∧
q∧
pE
0) ≤ µ
0(Q
0). On the other hand,
µ
0(∧
q∧
pE
0) = qpa
n−1µ(E ), µ
0(Q
0) = X
α0,...,αh−1
a
α0...αh−1{α
0µ
0(Q
00) + · · · + α
h−1µ
0(Q
h−10)}
≤ X
α0,...,αh−1
a
α0...αh−1pµ
0(Q
0h−1)
= qpa
n−1µ(Q
h−1).
Thus µ(E ) ≤ µ(Q
h−1).
Now let [f ] ∈ H be a general member. Then we have integers a
1≥ . . . ≥ a
ssuch that f
∗D ∼ = L
si=1