• Aucun résultat trouvé

K ¨AHLER C-SPACES AND QUADRATIC BISECTIONAL CURVATURE Albert Chau & Luen-Fai Tam Abstract

N/A
N/A
Protected

Academic year: 2022

Partager "K ¨AHLER C-SPACES AND QUADRATIC BISECTIONAL CURVATURE Albert Chau & Luen-Fai Tam Abstract"

Copied!
60
0
0

Texte intégral

(1)

94(2013) 409-468

K ¨AHLER C-SPACES AND QUADRATIC BISECTIONAL CURVATURE

Albert Chau & Luen-Fai Tam Abstract

In this article we give necessary and sufficient conditions for an irreducible K¨ahler C-space with b2 = 1 to have nonnegative or positive quadratic bisectional curvature, assuming the space is not Hermitian symmetric. In the cases of the five exceptional Lie groupsE6, E7, E8, F4, G2, the computer package MAPLE is used to assist our calculations. The results are related to two conjectures of Li-Wu-Zheng.

1. Introduction

Let (Mn, g) be a K¨ahler manifold of complex dimension n and let o ∈M. M is said to have nonnegative quadratic orthogonal bisectional curvature at o if for any unitary frame ei at o and real numbers ξi we have

(1.1) X

i,j

Ri¯ij¯ji−ξj)2 ≥0.

Here Ri¯ij¯j =R(ei,¯ei, ej,e¯j). Recall that M is said to have nonnegative bisectional curvature atoif for any X, Y ∈To(1,0)(M),R(X,X, Y,¯ Y¯)≥ 0, andM is said to have nonnegativeorthogonalbisectional curvature at oif R(X,X, Y,¯ Y¯)≥0 for all unitary pairsX, Y ∈To(1,0)(M). Following [16] we abbreviate by QB ≥ 0 for nonnegative quadratic orthogonal bisectional curvature, B≥0 for nonnegative bisectional curvature, and B≥0 for nonnegative orthogonal bisectional curvature. It is obvious that B ≥0 ⇒B ≥0⇒ QB ≥0. Note that in dimension n= 2, the conditionsB≥0 and QB≥0 are the same.

It is well-known that compact manifolds with B≥0 have been com- pletely classified by the works [18, 20,14, 1, 17]. By these works, we know that any compact simply connected irreducible K¨ahler manifold with B ≥ 0 is either biholomorphic to CPn or is isometrically biholo- morphic to an irreducible compact Hermitian symmetric space of rank at least 2. While the condition B ≥0 seems weaker, by the works of Chen [10] (see also [22]) and Gu-Zhang [13] we know that a compact

Received 3/23/2012.

409

(2)

simply connected irreducible K¨ahler manifold withB≥0 is also either biholomorphic toCPnor is isometrically biholomorphic to an irreducible compact Hermitian symmetric space of rank at least 2. In this sense, no new compact complex manifolds are introduced when we weaken the condition B ≥0 to the condition B≥0.

The condition QB ≥ 0 was first considered by Wu-Yau-Zheng [24]

where they proved that on a compact K¨ahler manifold with QB ≥ 0 any class in the boundary of the K¨ahler cone can be represented by a smooth closed (1,1) form which is everywhere nonnegative. There are other interesting properties satisfied by compact K¨ahler manifolds withQB ≥0. A fundamental property of such manifolds, implicit from earlier works [3] (see [8] for additional references) is that all harmonic real (1,1) forms are parallel. Recently it has been proved in [8] that the scalar curvature of such a manifold must be nonnegative, and if the manifold is irreducible, then the first Chern class is positive.

The ultimate goal is to classify K¨ahler manifolds with QB ≥0. For the compact case, a partial classification of the de Rham factors of the universal cover of such a manifold is given in [8]. Hence it remains to study the structure of compact simply connected irreducible K¨ahler manifolds with QB ≥ 0. By the parallelness of real harmonic (1,1) forms mentioned above, such K¨ahler manifolds also have b2 = 1 (see [14]). In view of the above results for B ≥0, one may wonder if any new compact complex manifolds are introduced when we weaken the condition B ≥ 0 to the condition QB ≥ 0. To address this, Li, Wu, and Zheng [16] constructed the first example of a simply connected irreducible compact K¨ahler manifold having QB ≥ 0, which does not support a K¨ahler metric having B ≥ 0. Their example is (B3, α2), a classical K¨ahlerC-space with second Betti numberb2= 1. It was further conjectured that all K¨ahler C-spaces with second Betti number b2 = 1 must have QB≥0, and the following conjectures were raised in [16]:

Conjecture 1.1. (1) Any K¨ahler C-space with b2 = 1 satisfies QB ≥0 everywhere.

(2) A compact simply connected irreducible K¨ahler manifold (Mn, g) with QB≥0 is biholomorphic to a K¨ahler C-space with b2= 1.

(3) In (2), if the manifold is not CPn, then g is a constant multiple of the standard metric.

A K¨ahler C-space is a compact simply connected K¨ahler manifold such that the group of holomorphic isometries acts transitively on the manifold; see [21, 15]. There is a complete classification of K¨ahler C- spaces withb2 = 1, and this is associated with the classification of sim- ple complex Lie algebras which are just An=sln+1, Bn =so2n+1, Cn= sp2n, Dn=so2nand the exceptional casesE6, E7, E8, F4, G2. Motivated by the work [16], we establish the following theorems related to conjec- tures (1) and (3). For the classical types we have the following:

(3)

Theorem 1.1.

(i) The K¨ahler C-space (Bn, αp), n≥3, 1< p < n satisfies QB ≥0 if and only if 5p + 1 ≤ 4n. Moreover, QB > 0 if and only if 5p+ 1<4n.

(ii) The K¨ahler C-space (Cn, αp), n≥3, 1< p < n satisfies QB ≥0 if and only if 5p ≤ 4n+ 3. Moreover, QB > 0 if and only if 5p <4n+ 3.

(iii) The K¨ahler C-space (Dn, αp), n ≥ 4, 1 < p < n−1 satisfies QB ≥0 if and only if5p+ 3≤4n. Moreover, QB >0 if and only if 5p+ 3<4n.

For the exceptional cases, we have the following:

Theorem 1.2.

(i) The K¨ahler C-space (G2, α2) satisfies QB >0.

(ii) The K¨ahler C-space (F4, αp), 1 ≤ p ≤ 4 satisfies QB ≥ 0 iff p= 1,2,4, in which cases QB >0.

(iii) The K¨ahler C-space (E6, αp), 2 ≤ p ≤ 5 satisfies QB ≥ 0 iff p= 2,3,5, in which cases QB >0.

(iv) The K¨ahler C-space (E7, αp), 1 ≤ p ≤ 6 satisfies QB ≥ 0 iff p= 1,2,5, in which cases QB >0.

(v) The K¨ahler C-space (E8, αp), 1 ≤ p ≤ 8 satisfies QB ≥ 0 iff p= 1,2,8, in which cases QB >0.

We only consider K¨ahlerC-spaces that are not Hermitian symmetric.

According to Itoh [15], Theorem 1.1 and 1.2 include all such K¨ahlerC- spaces withb2 = 1. Here QB >0 means that (1.1) is a strict inequality unless allξiare the same. Note that ifQB >0, then a small perturbation of the K¨ahler metric will still satisfy QB > 0; see Lemma 2.6 (and Remark 2.1). Hence conjecture (1) for the classical types is true only under some restrictions mentioned in Theorem 1.1, while conjecture (3) is too strong. Conjecture (2), however, may still be true in general.

Theorems 1.1 and 1.2 give more information on the curvature prop- erties of K¨ahler C-spaces with b2 = 1. It is well-known that CPn has B >0, and Hermitian symmetric spaces with rank at least 2 haveB≥0 but notB >0. All other K¨ahlerC-spaces which are non-Hermitian sym- metric spaces do not have B ≥0 or even B≥ 0. On the other hand, Itoh [15] proved that a K¨ahler C-space with b2 = 1 is a Hermitian symmetric space if and only if its curvature operator has at most two distinct eigenvalues. Our results show that as far as the sign of curva- ture is concerned, K¨ahlerC-spaces withb2 = 1 which are not Hermitian symmetric are further divided into two groups: some of them satisfy QB ≥0 and others do not have such a property.

(4)

We give here an idea of the proof and refer to§2 for details. Consider a Lie algebra as above, and a corresponding system ∆⊂Rnof root vec- tors in Rn where the induced Killing form is induced by the standard Euclidean inner product. Then each associated K¨ahler C-space corre- sponds to a certain subset of m+ ⊂∆ representing a unitary frame in which curvature approximation reduces to taking sums and inner prod- ucts of the vectors inm+⊂∆ (we can calculate exact values in the case of bisectional curvatures). We combine this with symmetry, counting, and eigenvalue estimate arguments to obtain Theorem 1.1. For the five exceptional cases in Theorem 1.2, the computer package MAPLE was used to assist our calculations and details are provided in the appendix.

Theorem 1.1 was proved in an earlier version of this article [9], and while similar in spirit, our proof here is somewhat simpler, eliminating the need for many calculations from the appendix of [9].

The organization of the paper is as follows. In§2 we will state basic properties and formulae for K¨ahlerC-spaces that will be used through- out the paper. We will discuss the conditions QB ≥0 and QB >0 in general, then in relation to the K¨ahlerC-spaces. In§3 we prove Theorem 1.1 for the classical K¨ahlerC-spaces; details for some of the calculations in these sections can be found in the appendices of [9], which is an ear- lier version of this article. In§4 we present the details of our results on Theorem 1.2 for the exceptional K¨ahler C-spaces, with details of our use of MAPLE provided in the appendix.

Acknowledgments.The authors would like to thank F. Zheng for valuable comments and interest in this work.

The first author’s research was partially supported by NSERC grant no. #327637-11. The second author’s research was partially supported by Hong Kong RGC General Research Fund #CUHK 403011.

2. Basic facts

2.1. The K¨ahler C-spaces and curvature formulae.Consider a compact K¨ahler C-space (M, ω) with transitive holomorphic isometry group G, and suppose b2(M) = 1. Then any real (1,1) form ρ on M is given by ρ = cω +√

−1∂∂f for some constant c and function f where ω is the K¨ahler form. Now if ρ is G invariant, then ∆gf is also G invariant and hence constant on M. Thus f is constant on M and ρ =cω. In particular,g is the unique G invariant K¨ahler metric onM and it is K¨ahler Einstein. For more discussions on K¨ahler C-space, see [2,15,21,16].

K¨ahler C-spaces with second Betti number b2 = 1 are obtained as follows (see [4,5,6,15,16,21]). LetGbe a simply connected, complex Lie group, and let g be its Lie algebra with Cartan subalgebra h and corresponding root system ∆⊂h. Theng=h⊕L

αCEα, whereEα is a root vector of α. Letl= dimChand fix a fundamental root system

(5)

α1, . . . , αl ⊂∆. This gives an ordering of roots in ∆. Let ∆+and ∆be the set of positive and negative roots, respectively. LetK be the Killing form for g. Then we may choose root vectors {Eα}, α∈∆ such that

K(Eα, Eα) =−1, α∈∆+; [Eα, Eβ] =nα,βEα+β

such that nα,β = nα,β ∈ R with nα,β = 0 if α+β is not a root.

Together with a suitable basis in h, they form a Weyl canonical basis forg. Now for any 1≤r≤l, let

+r(k) ={X

i

niαi ∈∆+|nr=k}, ∆+r = [

k>0

+r(k).

Let P be the subgroup whose Lie algebra is h⊕L

α\+r CEα. Then G/P is a complex homogeneous space having b2 = 1. Now let

m+k = M

α+r(k)

CEα, mk = M

αr(k)

CEα, t=h⊕ M

α+(0)

(CEα⊕CEα).

Then m+=L

k>0m+k can be identified with the tangent space ofG/P. As given in [4, 15, 16], the G-invariant K¨ahler form on G/P is given by:

Lemma 2.1.

(i) In a Weyl canonical basis, letωαα¯ be the dual of Eα andE¯α :=

−Eα, α∈∆+r. The Ginvariant K¨ahler form on G/P is g= 2X

k>0

k X

α+r(k)

ωα·ωα¯ =X

k>0

(−kK)|m+

k×mk.

(ii) [t,m±k]⊂m±k, [m±k,m±l ]⊂m±k+l, [m+k,mk]⊂ t. If k > l > 0, then [m+k,ml ]⊂m+kl, [mk,m+l ]⊂mkl.

The K¨ahler C space thus obtained is denoted as (g, αr). Conversely, every K¨ahler C space with b2= 1 can be obtained by the construction.

Thus the set {eα := 1/√

kEα}; α ∈ ∆+k, k ≥ 1 forms a unitary basis for the tangent space of (g, αr) in the metricg. We call this basis as a Weyl frame. To compute the curvature tensor in this frame, we have the following from Li-Wu-Zheng [16, Proposition 2.1], using the method in [15]. For the sake of completeness, we give a proof.

Proposition 2.1. [Li-Wu-Zheng] Let Xi ∈ m+i , Yj ∈ m+j , Zk ∈ m+k, Wl ∈m+l . Suppose i+k=j+l. Then

R(Xi,Y¯j,Zk,W¯l) =

(k−j)ξkj− kl i+k

K([Xi, Zk],[ ¯Yj,W¯l])

+ (−(k−j)ξkj+kξij+lξji+lδijδkl)K([Xi,Y¯j],[Zk,W¯l]).

(2.1)

R(Xi,Y¯j, Zk,W¯l) = 0 if i+k6=j+l.

(6)

Here ξq= 1 if q >0 andξq= 0 if q≤0.

Proof. Note that g(U,V¯) =−kK(U,V¯), onm+k ×mk etc.

Then [15, p. 43]

R(Xi,Y¯j, Zk,W¯l) =g(R(Xi,Y¯j)Zk,W¯l)

=g([Λ(Xi),Λ( ¯Yj)]Zk,W¯l)−g(Λ([Xi,Y¯j]m)Zk,W¯l)

−g([[Xi,Y¯j]t, Zk],W¯l), (2.2)

where Λ(U)V =n/(n+n)[U, V]m+ if U ∈m+n, V ∈m+n, and Λ( ¯U)V = [ ¯U , V]m+,for all U, V ∈ m+; see [15, p. 45]. Here [U, V]m+ is the com- ponent of [U, V] in m+. Now ifi+k=j+l,

[Λ(Xi),Λ( ¯Yj)]Zk= Λ(Xi)Λ( ¯Yj)−Λ( ¯Yj)Λ(Xi) Zk

=Λ(Xi)([ ¯Yj, Zk]m+)− k

i+kΛ( ¯Yj)([Xi, Zk])

=(k−j)

l ξkj[Xi,[ ¯Yj, Zk]]− k

i+k[ ¯Yj,[Xi, Zk]]m+. (2.3)

Here each term is inm+l . Hence g([Λ(Xi),Λ( ¯Yj)]Zk,W¯l)

=−(k−j)ξkjK([Xi,[ ¯Yj, Zk]],W¯l) + kl

i+kK([ ¯Yj,[Xi, Zk]],W¯l)

=−(k−j)ξkjK(Xi,[[ ¯Yj, Zk],W¯l]) + kl

i+kK([ ¯Yj,[Xi, Zk]],W¯l)

= (k−j)ξkjK(Xi,[[ ¯Wl,Y¯j], Zk] + [[Zk,W¯l]],Y¯j) + kl

i+kK([ ¯Wl,Y¯j],[Xi, Zk]])

=−(k−j)ξkjK([Xi,Y¯j],[Zk,W¯l]) +

(k−j)ξkj− kl i+k

K([Xi, Zk],[ ¯Yj,W¯l]).

(2.4)

Now [Xi,Y¯j]m is inm+ij if i > j, andmji ifj > i, and is 0 ifi=j. So g(Λ([Xi,Y¯j]m)Zk,W¯l) =−(kξij+lξji)K

[Xi,Y¯j],[Zk,W¯l] . Also, [Xi,Y¯j]t = 0 unless i = j. If i = j, then [[Xi,Y¯j]t, Zk] ∈ m+k. Hence

g([[Xi,Y¯j]t, Zk],W¯l) =δijδklg([[Xi,Y¯j], Zk],W¯l)

= −lδijδklK

[Xi,Y¯j],[Zk,W¯l] .

(7)

Also, R(Xi,Y¯j, Zk,W¯l) = 0 if i+k6=j+l. q.e.d.

Lemma 2.2. Same notations as in Proposition2.1. AssumeX, Z, W are canonical Weyl basis vectors and Z 6=W; then

R(X,X, Z,¯ W¯) = 0.

Proof. Since i = j, the lemma is true if k 6= l by Proposition 2.1.

Hence we assumek=l. We first assume thatk≤i. Then R(X,X, Z,¯ W¯) =− k2

i+kK([X, Z],[ ¯X,W¯]) +kK([X,X],¯ [Z,W¯]).

Now letX=Eα,Z =Eβ,W =Eγ withβ 6=γ. Note that ¯Eα=−Eα. Then [X, Z] =nα,βEα+β, [X, W] =nα,γEα+γ. Hence

K([X, Z],[ ¯X,W¯]) =−nα,βnα,γK(Eα+β, Eαβ).

If α +β or α +γ is not a root, then nα,β = 0 or nα,γ = 0 and K([X, Z],[ ¯X,W¯]) = 0. Otherwise, both Eα+β and Eα+γ are canonical Weyl basis vectors and are inm+i+k by Lemma 2.1. Since β6=γ, and K is proportional togonm+i+k×mi+k, we also haveK([X, Z],[ ¯X,W¯]) = 0.

On the other hand, by the fact that K([x, y], z) = K(z,[y, z]), we have

(2.5) K([X,X],¯ [Z,W¯]) =K(X,[ ¯X,[Z,W¯]]).

Now [Z,W¯] =nβ,γEβγ, [ ¯X,[Z,W¯]] =nα,βγnβ,γEα+βγ. Ifβ−γ or −α+β−γ is not a root, then as before we haveK(X,[ ¯X,[Z,W¯]]) = 0. Otherwise, by Lemma 2.1, [Z,W¯] ∈ t and [ ¯X,[Z,W¯]] ∈ mi . Since

−α+β−γ6=−α, so as before

K([X,X],¯ [Z,W¯]) =K([X,[ ¯X,[Z,W¯]) = 0.

Hence the lemma is true when k≤i.

Supposei < k. Then it is equivalent to proveR(X,Y , Z,¯ Z¯) = 0, but assumingi > k and X6=Y. In this case,

R(X,Y , Z,¯ Z) =¯ − k2

i+kK([X, Z],[ ¯Y ,Z]) +¯ kK([X,Y¯],[Z,Z¯]).

The previous argument implies the lemma is true in this case as well.

q.e.d.

To use the formula in Proposition 2.1, we need to compute the Lie bracket and Killing form in the given Weyl basis. Now the Killing form K is negative definite onhand thus induces a positive definite bilinear form, denoted also byK, on the dualh. We can then identifyhwithRl (or a subspace of some Rn) so that K becomes the standard Euclidean inner product and the root system is represented by a subset ∆ ⊂Rl. It turns out that a corresponding Weyl basis{Eα}α+ exists in which the Lie bracket and Killing form are computed in terms of Euclidean

(8)

inner products and addition of the vectors α. We describe this in more detail below.

Let g be a semi simple Lie algebra, and let ∆ = {α, β, . . .} ⊂ Rn be a corresponding root system with standard inner product (·,·) cor- responding to the induced Killing form K. To the positive roots there corresponds a Chevalley basis {Xα, Xα, Hα}α+ where for each α, Xα, Xα are root vectors for α,−α respectively, Hα ∈ h, and the fol- lowing relations are satisfied (see [19, p. 51]):

[Xα, Xα] =Hα, [Xα, Xβ] =

Nα,βXα+β, if α+β is a root, Nα,β =−Nα,β ; 0, if α+β 6= 0 is not a root.

Nα,β =±(p+ 1), pis the largest integer so thatβ−pαis a root.

[Hα, Xβ] =β(Hα)Xβ. (2.6)

We also have:

(a) β(Hα) = 2(β,α)(α,α), [11, p. 337];

(b) K(Hα, Hα) = 2K(Xα, Xα), [11, p. 207].

By these and [11, p. 207–208], we have the formulas K(Hα, Hα) = 4

(α, α), K(Xα, Xα) = 2

(α, α), K(Hα, Hβ) = 4(α, β)

|α|2|β|2, (2.7)

where in the last equation we have used the first equation of (2.6), (2.5), (a), and the second formula in (2.7).

Lemma 2.3. For positive roots α, let (2.8) Eα = √|α|

2Xα, Eα =−√|α| 2Xα. Then for positive roots α, β

K(Eα, Eα) =−1, [Eα, Eβ] =nα,βEα+β, [Eα, Eβ] =nα,βEαβ,

[Eα, Eβ] =nα,βEαβ,if α−β 6= 0, (2.9)

where

nα,β =nα,β =

( |α||β|

2|α+β|Nα,β, if α+β is a root, 0, if α+β is not a root,

(9)

and

nα,β =





|α||β|

2|αβ|Nα,β, if α−β is a positive root,

|2α|α||β|β|Nα,β, if α−β is a negative root, 0, if α−β is not a root.

Hence {Eα} form a Weyl canonical basis.

Proof. It is easy to see that K(Eα, Eα) = −1. If α+β is a root, then

[Eα, Eβ] =|α||β|

2 [Xα, Xβ]

=Nα,β|α||β| 2 Xα+β

= |α||β|

√2|α+β|Nα,βEα+β =nα,βEα+β

[Eα, Eβ] =|α||β|

2 [Xα, Xβ]

=Nα,β|α||β| 2 Xαβ

=Nα,β√|α||β|

2|α+β|Eαβ =nα,βEαβ where nα,β = |α||β|

2|α+β|Nα,β. Here we have used the fact that Nα,β =

−Nα,β and Xαβ = −|α+β2|Eαβ. If α +β is not a root, then [E,a, Eβ] = 0.

If α−β 6= 0 and is a positive root, then [Eα, Eβ] = |α||β|

2 [Xα, Xβ] =Nα,β|α||β| 2 Xαβ

= |α||β|

√2|α−β|Nα,βEαβ. If α−β 6= 0 and is a negative root, then

[Eα, Eβ] = |α||β|

2 [Xα, Xβ] =Nα,β|α||β| 2 Xαβ

=−√|α||β|

2|α−β|Nα,βEαβ.

q.e.d.

Now let η ∈ ∆+, and consider the K¨ahler C-space (g, η) with cor- responding Weyl frame (unitary frame for (g, η)) eα = 1

kEα for α ∈

(10)

+(k). For any positive roots α, β, define (2.10)

( Neα,β= ||α+βα||β||Nα,β;

Neα,β = ||αα||ββ||Nα,β, if α−β 6= 0.

We can now combine Lemma 2.2 and Proposition 2.1 with Lemma 2.3, (2.6), and (2.7) to obtain the following from [15, Proposition 2.4].

In the above setting, we denote the curvature tensor R(eα,¯eβ, eγ,e¯δ) also by R(α,β, γ,¯ δ) or¯ Rα,β,γ,¯ ¯δ.

Lemma 2.4. Letα∈∆+(i),β ∈∆+(j), withi≤j, and letRα¯αββ¯= R(eα,e¯α, eβ,e¯β). Then

(2.11) Rα¯αββ¯= 1 j

(α, β) +1 2

i i+jNeα,β2

.

Next let us consider R(α,β, γ,¯ δ) =¯ R(eα,e¯β, eγ,¯eδ) with α − β, γ−δ6= 0.

Lemma 2.5. Let eα ∈ ∆+(i), eβ ∈ ∆+(j), eγ ∈ ∆+(k), and eδ

+(l).

1) If α−β6=δ−γ, then R(α,β, γ,¯ δ) = 0.¯ 2) If α−β=δ−γ 6= 0, then

R(α,β, γ,¯ δ¯) =− 1 2√

ijkl

(k−j)ξkj− kl i+k

Neα,γNeβ,δ

+ 1

2√ h ijkl

(−(k−j)ξkj+kξij+lξji+lδijδkl)Neα,βNeγ,δ.i

=:R1(α,β, γ,¯ δ) +¯ R2(α,β, γ,¯ δ).¯ (2.12)

Proof. (1) follows from Lemma 2.2, and the fact thatK(Eα, Eβ) = 0 unless α+β = 0.

(2) Note that ¯eα =−eα, etc. First assume thatα+γ and α−β are both roots. By Lemma 2.3,

[eα, eγ] = 1

√ik[Eα, Eγ]

=nα,γ 1

√ikEα+γ

=Nα,γ|α||γ|

√2ikEα+γ. Similarly,

[eβ, eδ] =Nβ,δ|√γ||δ|

2jlEβδ,

(11)

We may assume that α−β is a positive root, thenγ−δ is a negative root.

[eα, eβ] =Nα,β|√α||β| 2ijEαβ, [eγ, eδ] =−Nγ,δ|γ||δ|

√2klEγδ.

Since α+γ =β+δ and K(Eσ, Eσ) =−1, we see that (2) is true by Proposition 2.1. The cases where α+γ or α−β is not a root can be proved similarly.

q.e.d.

2.2. The condition QB ≥ 0.We first discuss the conditionQB ≥ 0 on a K¨ahler manifold (M, ω) with K¨ahler formω. We will also consider the conditionQB >0 atp, which we define as:QB ≥0 atpwith strict inequality in (1.1) provided not all ξis are the same. Now define the following bilinear forms on the space Ω1,1R (M) of real (1,1) forms on M:

F(η, σ) = X

i,j,k,l

Ri¯jk¯lρi¯lσk¯j = X

i,j,k,l

Ri¯lk¯jρi¯lσk¯j

G(η, σ) = 1

2(Ri¯jgk¯l+Rk¯lgi¯ji¯lσk¯j

whereρi¯l, σk¯j are the local components ofρ, σwith indices raised. Clearly, G and F are well defined real symmetric bilinear forms on Ω1,1R (p) for any p. Now let θA be a unitary frame at any p with co-frame ηA and let aA be real numbers. Take X = P

A

√−1aAηA∧ηA1,1R (p). Then a simple calculation gives

G(X, X)−F(X, X) =X

A

RAA¯a2A−X

A,B

RAAB¯ B¯aAaB

= 1 2

X

A,B

RAAB¯ B¯(aA−aB)2. (2.13)

The following was observed by Yau [26].

Lemma 2.6. At any pointp we have (a) QB ≥0 if and only if G−F ≥0.

(b) QB >0 if and only if G−F >0 on Ω1,1R (p)\Rω(p).

HereΩ1,1R (p)\Rω(p)are the real(1,1)forms atpwhich are not multiples of the K¨ahler form.

Proof. We first prove (a). The fact that G−F ≥ 0 implies QB ≥ 0 follows immediately from (2.13) and the fact that θA and aA are arbitrary. Conversely, supposeQB ≥0 and letXbe any real (1,1) form at p. Then we can always diagonalizeX. Namely, there exists a unitary frame eA with co-frame ηA such that X = P

A

√−1aAηA∧ηA. Now

(12)

(2.13), and the assumption QB ≥ 0, immediately implies G(X, X)− F(X, X)≥0.

Now we prove (b). The proof is basically the same in part (a) once we observe that X ∈ Rω(p) if and only if: for every unitary frame eA at p with co-frame ηA we have X = cP

A

√−1ηA∧ηA for some real constant c. The fact thatG−F >0 on Ω1,1R (p)\Rω(p) impliesQB >0 now follows immediately from (2.13) and the fact that θA and aA are arbitrary. Conversely, suppose QB > 0 and let X ∈ Ω1,1R (p)\Rω(p).

Then there exists a unitary frame eA with co-frame ηA such that X = P

A

√−1aAηA ∧ηA with aAs not all the same. Now (2.13) and the assumption QB >0 immediately impliesG(X, X)−F(X, X)>0.

This concludes the proof of the lemma. q.e.d.

Remark 2.1. Thus QB > 0 if and only if G−F is positive in the orthogonal complement of Rω. In particular, if (M, g) is a compact K¨ahler manifold with QB > 0, then a K¨ahler metric which is a small perturbation of g will also satisfy QB >0.

Remark 2.2. Viewed as an endomorphism on Ω1,1R (M),G−F is in fact the curvature term in the Weitzenb¨ock identity for real (1,1) forms:

g−∆ is given byG−F up to a positive constant multiple where ∆g is the Bochner Laplacian with respect togand ∆ is the Laplace-Beltrami operator. The standard Bochner technique and Lemma 2.6 then give:all real harmonic (1,1) forms onM are parallel providedQB≥0; moreover, dim(HR1,1(M)) = 1providedQB >0 whereHR1,1(M) is the space of real harmonic (1,1) forms on M. See §1 for a reference to these facts and their implicit appearance in earlier works.

By Lemma 2.6, to check whetherQB ≥0 (orQB >0), it is sufficient to consider G−F ≥0 in a unitary frame of our choice. In the case of K¨ahler C-spaces, the natural choice is a Weyl frame. By Lemmas 2.2 and 2.6, we have:

Corollary 2.1. On a K¨ahler C-space, let Ric = µg and let eA be a Weyl frame. Then QB ≥ 0 if and only if the largest eigenvalues of the quadratic forms P

A,BRAAB¯ B¯xAxB, with xA’s real, and P

A,B,C,D;

A6=B,C6=D

RABC¯ D¯xABxCD, with xAB = xBA, are at most µ. QB >0 if QB ≥ 0 and the eigenvalue µ of P

A,BRAAB¯ B¯xAxB is simple and the largest eigenvalue of P

A,B,C,D;

A6=B,C6=D RABC¯ D¯xABxCD is less thanµ.

The following simple fact will be used throughout the paper to esti- mate the largest eigenvalue of a quadratic form.

Lemma 2.7. [row sums] Letx1, . . . , xn, a1, . . . , an, and λbe real or complex numbers. Suppose |xk|= max{|xi| 1≤i≤n}>0 and

λxk= Xn

i=1

aixi.

(13)

Then

|λ| ≤ Xn

j=1

|ai|.

In particular, if λ is an eigenvalue of ann×n matrix (aij), then

|λ| ≤max

i

 Xn

j=1

|aij|

.

We also note the following modification of Lemma 2.7 which will only be needed in a few exceptional cases.

Lemma 2.8. [weighted row sums] Letλbe an eigenvalue of ann×n matrix A = (aij) such that |λ| > 0. Let µ > 0 be a positive number.

Define bsj inductively: b(0)j = 1, and b(s+1)j = min(1,X

l

|ajl|b(s)l µ ).

Then for all s≥0,

(2.14) |λ| ≤max{max

i

 Xn

j=1

|aij|b(s)j

, µ}. In particular, if for some s≥0,

(2.15) max

i

 Xn

j=1

|aij|b(s)j

< µ, then |λ|< µ.

Proof. First we show that b(s+1)j ≤ b(s)j for all j. Note that by defi- nition 1 ≥ b(s)j ≥ 0. It is obviously true that b(1)j ≤ 1 = b(0)j . Suppose b(s+1)j ≤b(s)j for allj; then

b(s+2)j = min(1,X

l

|ajl|b(s+1)l /µ)≤min(1,X

l

|ajl|b(s)l /µ) =b(s+1)j . To prove the lemma: If |λ| ≤ µ, then the lemma is true. Suppose

|λ|> µ. Letxibe the components of an eigenvector ofAwith eigenvalue λ. Suppose, without loss of generality, that maxi|xi|= 1. We claim that for all s≥0,

|xi| ≤b(s)i for all i≥1. For s= 1, then, for anyj,

λxj =X

l

ajlxl

(14)

and

|xj| ≤ 1

|λ| X

l

|ajlxl| ≤ 1

|λ| X

l

|ajl|.

So |xj| ≤b(1)j because|λ|> µ and |xj| ≤1. Now suppose|xj| ≤b(s)j for all j. Then as before,

|λ||xj| ≤X

l

|ajl||xl| ≤X

l

|ajl|b(s)l and

|xj| ≤X

l

|ajl|b(s)l µ . Hence |xj| ≤b(s+1)j . Hence the claim is true.

Now we may assume without loss of generality that |x1|= 1. (2.14) is true fors= 0 by the previous lemma. Fors≥1,

|λ|=|λx1| ≤X

l

|a1l|xl| ≤X

l

|a1lb(s)l . Hence (2.14) is also true in this case.

If (2.15) is true, then it is still true if µis replaced byµ−ǫforǫ >0 small enough. Then by (2.14),|λ| ≤µ−ǫ < µ.

q.e.d.

3. K¨ahler C-spaces of classical type

According to [15], the K¨ahler C-spaces withb2 = 1 of classical type which are not Hermitian symmetric spaces are (Bn, αp), with n ≥ 3, 1 < p < n, (Cn, αp), withn≥3, 1< p < n, and (Dn, αp), withn≥4, 1 < p < n−1. For each Lie algebra Bn, Cn, Dn below, we assume an identification has been made between h, the dual Cartan subalgebra, and V = Rn so that the induced Killing form corresponds to the Eu- clidean inner product (·,·). We will then present the corresponding root system ∆ as a set of vectors in V =Rn. We refer to [7] for details.

3.1. The spaces (Bn, αp).We first consider the space (Bn, αp), with n ≥3, 1< p < n. Let V =Rn and let εi be the standard basis on V. The root system for Bnis

(3.1) ∆ ={±εi±εj|1≤i, j≤n, i6=j} ∪ {±εi|1≤i≤n} Simple positive roots are

(3.2) α11−ε2, α22−ε3, . . . , αn1n1−εn, αnn. Positive roots are

(3.3) ∆+={εij}i<j ∪ {εi−εj}i<j ∪ {εi}.

(15)

In terms of the αi’s, the positive roots are εii+· · ·+αn

εiji+· · ·+αj1+ 2αj+· · ·+ 2αn, i < j εi−εji+· · ·+αj1, i < j.

(3.4)

Let 1< p < n. Recall that

+p(k) =



α∈∆+|α=kαp+X

i6=p

miαi, mi ≥0, mi ∈Z



. By (3.3) and (3.4), we have

+p(1) ={εa|1≤a≤p}[

ai|1≤a≤p, p+ 1≤i≤n} [{εa−εi|1≤a≤p, p+ 1≤i≤n},

(3.5)

+p(2) ={εab|1≤a < b≤p}. (3.6)

(3.7) ∆+p(k) =∅

fork≥3. The dimension of (Bm, αp) is 12p(4n−3p+ 1). We denote the elements of the ∆+p(k)’s by: Xai = εa−εi, Yai = εai, 1 ≤ a ≤ p, p+ 1≤ i≤ n; Ua = εa, Wabab, 1 ≤ a, b≤ p. In the following a, b, . . . will range from 1 to p and i, j, . . . will range from p+ 1 to n.

Thus

+p(1) ={Xai}1ap;p+1in

[{Yai}1ap;p+1in

[{Ua}1ap,

+p(2) ={Wab}1a<bp.

Now recall that Nα,β = ±(p+ 1) where p is the largest integer, so that β−pαis a root, and also the definition of Neα,β in (2.10).

Lemma 3.1. Let α, β be positive roots in (Bn, αp); then Neα,β =

√2 sgn(Nα,β). If α−β 6= 0, then Neα,β =√

2 sgn(Nα,β).

Proof. Note that if σ is a root, then either |σ|2 = 1 or |σ|2 = 2. We begin by proving the first part of the lemma. Letα, β be positive roots.

We may assume α+β is a root; otherwise the first part of the lemma is obviously true.

Suppose|α|2=|β|2= 1 and suppose|α+β|2 = 1; then (α, β) =−12, and this is impossible because one can see that (α, β) is an integer.

Hence |α+β|2 = 2 and (α, β) = 0. So α−β is also a root [11, p. 324].

α−2β is not a root because|α−2β|2 = 5. HenceNα,β=±2. Therefore, by the definition ofNeα,β in (2.10), Neα,β =√

2 sgn(Nα,β).

Suppose|α|2 = 1 and|β|2 = 2. As before, one can prove that (α, β) =

−1 andNα,β =±1. Hence Neα,β =√

2 sgn(Nα,β).

(16)

Suppose |α|2 =|β|2 = 2. As before, one can prove that (α, β) =−1, Nα,β =±1, and henceNeα,β=√

2 sgn(Nα,β).

The case forNeα,β can be proved similarly. q.e.d.

By Lemmas 2.4, 2.5, and 3.1 and the fact that R(α,β, γ,¯ δ) =¯ R(α,δ, γ,¯ β), we have:¯

Corollary 3.1. Let α ∈ ∆+(i), β ∈ ∆+(j), γ ∈ ∆+(k), and δ ∈

+(l).

1)

Rα¯αββ¯=



(α, β) + 12(sgn(Nα,β)2), i=j = 1,

1

2(α, β), i= 1, j = 2;

1

2(α, β), i=j = 2,

2) If α−β6=δ−γ, then R(α,β, γ,¯ δ) = 0.¯

3) If α−β=δ−γ 6= 0, then for(i, j, k, l) = (1,1,1,1), R(α,β, γ,¯ δ) =¯



1

2sgn(Nα,γ)sgn(Nβ,δ), if α−β is not a root, sgn(Nα,β)sgn(Nγ,δ), if α+γ is not a root,

12sgn(Nα,γ)sgn(Nδ,β), if β−γ 6= 0 is not a root.

For other cases, R(α,β, γ,¯ δ) =¯



1

2sgn(Nα,β)sgn(Nγ,δ), if (i, j, k, l) = (1,1,2,2),

1

2sgn(Nα,β)sgn(Nγ,δ), if (i, j, k, l) = (2,2,2,2),

1

2sgn(Nα,β)sgn(Nγ,δ) if (i, j, k, l) = (1,2,2,1).

To compute the Ricci curvature, we know that Ric =µgand thus µ= Ric(W12,W¯12)

=X

a,i

R(W12,W¯12, Xai,X¯ai) +R(W12,W¯12, Yai,Y¯ai)

+X

a

R(W12,W¯12, Ua,U¯a) +X

a<b

R(W12,W¯12, Wab,W¯ab)

=1

2[2(n−p) + 2(n−p)] + 1 + 1

2(p+ (p−2))

= 2n−p.

Lemma 3.2. Let λbe the largest eigenvalue of the quadratic form X

A,B

RAAB¯ B¯xAxB in the Weyl frame, where xA are real.

(a) λ≤2n−p if and only if 5p+ 1≤4n.

(b) If 5p+ 1<4n, thenλ= (2n−p) iff the corresponding eigenvector satisfies xA=xB for all A, B.

(c) If5p+1 = 4n, then there is an eigenvector with eigenvalue(2n−p) such that xA6=xB for some A6=B.

(17)

Proof. We begin with the proof of (a). Letv= (xA) be an eigenvector corresponding to the largest eigenvalueλfor the quadratic form. Assume the components satisfy maxA|xA| = 1. Let us denote the components xA more specifically byxai, yai, a≤p < i;ua, a≤p;tab, a < b≤p, and let us denote R(Xai,X¯ai, Xbj,X¯bj) by R(Xai, Xbj) etc. Then P(v) = P

A,BRAAB¯ B¯xAxB is equal to:

P(v) = X

a,bp<i,j

R(Xai, Xbj)xaixbj+ X

a,bp<i,j

R(Yai, Ybj)yaiybj

+ X

a,bp<i,j

R(Xai, Ybj)xaiybj+ X

a,bp<i,j

R(Yai, Xbj)yaixbj

+ 2 X

a,cp<i

R(Xai, Uc)xaiuc+ 2 X

a,cp<i

R(Yai, Uc)yaiuc

+ 2 X

a<b,cp<i

R(wab, Xci)tabxci+ 2 X

a<b,cp<i

R(wab, Yci)tabyci

+ X

a,bp

R(Ua, Ub)uaub+ 2 X

a<b,cp

R(wab, Uc)tabuc

+ X

a<bp,c<dp

R(Wab, Wcd)tabtcd. (3.8)

From Corollary 3.1, it is easy to see that R(Xai, Xbj) =R(Yai, Ybj), R(Xai, Ybj) = R(Yai, Xbj), R(Xai, Ub) = R(Yai, Ub), R(Xai, Wbc) = R(Yai, Wbc). We see that if we interchangexaiandyaifor alla, iand ob- tain a vector w, thenP(v) =P(w) and |v|=|w|. We may then assume that eitherxai=yaifor alla, i, or by consideringv−w, thatxai=−yai and ua=tab= 0 for all a, b.

Suppose|ua|= 1 for somea. We may assume thatua= 1. By Corol- lary 3.1, R(Ua,U¯a, x,x)¯ ≥0 because (Ua, x)≥0 for all x∈∆+p(k), k= 1,2.

λua=X

bp

R(Ua, Ub)ub+ X

bp<i

R(Xbi, Ua)xbi+ X

bp<i

R(Ybi, Ua)ybi

+ X

c<dp

R(wcd, Ua)tcd. (3.9)

Notice that the coefficients are all nonnegative and the sum is just Ric(Ua,U¯a) = 2n−p. Hence λ ≤ 2n−p. Moreover, if λ = 2n−p, then we must in fact have

(3.10) xa,i=ya,i=ub =tcd= 1

Références

Documents relatifs

Therefore, in this way, we can find all solutions (which are infinitely many) to equation (1.2) in the bal- anced class of a given balanced metric on X of nonnegative

Zhu, A Uniformization Theorem Of Complete Noncompact K¨ ahler Surfaces With Positive Bisectional Curvature, J.. Zhu, On complete noncompact K¨ ahler manifolds with positive

We show how two topologically distinct spaces - the K¨ ahler K3 × T 2 and the non- K¨ahler T 2 bundle over K 3 - can be smoothly connected in heterotic string theory.. The

Kobayashi, S., Ochiai, T.: Three-dimensional compact K~ihler manifolds with positive holo- morphic bisectional curvature... Kobayashi, S., Ochiai, T.: Characterizations

Calabi flow, Geodesic rays, and uniqueness of constant scalar curvature K¨ ahler metrics.. Xiuxiong Chen and

Tian [Tia9] proved that there is a K¨ ahler-Einstein metric if and only if the K-energy is proper on the space of all K¨ ahler metrics in c 1 (X).. So the problem is how to test

It is worth to point out that the same proof as in the K¨ ahler case (following Mok) fails for the Chern-Ricci flow since the evolution of the Riemann curvature tensor under

Motivated by the recent work of Wu and Yau on the ampleness of canonical line bundle for compact K¨ ahler manifolds with negative holomorphic sectional curvature, we introduce a