• Aucun résultat trouvé

<span class="mathjax-formula">$L^2$</span>-theory for the <span class="mathjax-formula">$\protect \overline{\partial }$</span>-operator on complex spaces with isolated singularities

N/A
N/A
Protected

Academic year: 2022

Partager "<span class="mathjax-formula">$L^2$</span>-theory for the <span class="mathjax-formula">$\protect \overline{\partial }$</span>-operator on complex spaces with isolated singularities"

Copied!
35
0
0

Texte intégral

(1)

ANNALES

DE LA FACULTÉ DES SCIENCES

Mathématiques

JEANRUPPENTHAL

L2-theory for the∂-operator on complex spaces with isolated singularities

Tome XXVIII, no2 (2019), p. 225-258.

<http://afst.cedram.org/item?id=AFST_2019_6_28_2_225_0>

© Université Paul Sabatier, Toulouse, 2019, tous droits réservés.

L’accès aux articles de la revue « Annales de la faculté des sci- ences de Toulouse Mathématiques » (http://afst.cedram.org/), implique l’accord avec les conditions générales d’utilisation (http://afst.cedram.

org/legal/). Toute reproduction en tout ou partie de cet article sous quelque forme que ce soit pour tout usage autre que l’utilisation à fin strictement personnelle du copiste est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit contenir la présente mention de copyright.

cedram

Article mis en ligne dans le cadre du

Centre de diffusion des revues académiques de mathématiques

(2)

L

2

-theory for the ∂-operator on complex spaces with isolated singularities

(∗)

Jean Ruppenthal(1)

ABSTRACT. — The present paper is a complement to the L2-theory for the

∂-operator on a Hermitian complex spaceXof pure dimensionnwith isolated sin- gularities, presented in [17] and [13]. The general philosophy is to use a resolution of singularitiesπ:MX to obtain a “regular” model of theL2-cohomology.

First, we show how the representation of theL2loc-cohomology ofX on the level of (n, q)-forms in terms of “regular”L2loc-cohomology onM, given in [17], can be made explicit in terms of differential forms, if a certain reasonable extra condition is satisfied. Second, we prove the analogous statement forL2-cohomology, which is a new result. Finally, we use this in combination with duality observations to give a new proof of the main results from [13], where the resolutionπ:M X is used to express theL2-cohomology ofX on the level of (0, q)-forms in terms of “regular”

L2-cohomology onM.

RÉSUMÉ. — Le présent article est un complément à la théorie deL2pour l’opéra- teursur un espace complexe hermitienXde dimension purenavec des singularités isolées, présenté dans [17] et [13]. La philosophie générale est d’utiliser une résolution de singularitésπ:MXpour obtenir un modèle « régulier » de laL2-cohomologie.

Tout d’abord, nous montrons comment la représentation de laL2loc-cohomologie deX au niveau de (n, q)-formes en termes de laL2loc-cohomologie « réguliere » sur M, donnée dans [17], peut être fait explicite en termes de formes différentielles, si une certaine condition supplémentaire raisonnable est remplie. Deuxièmement, nous prouvons la déclaration analogue pour la L2-cohomologie, ce qui est un nouveau résultat. Enfin, nous l’utilisons en combinaison avec des observations de dualité pour donner une nouvelle preuve des principaux résultats de [13], où la résolutionπ:M Xest utilisée pour exprimer laL2-cohomologie deXsur le niveau de (0, q)-formes en termes deL2-cohomologie « régulière » surM.

(*)Reçu le 11 septembre 2015, accepté le 28 avril 2017.

Keywords:Cauchy–Riemann equations,L2-theory, singular complex spaces.

2010Mathematics Subject Classification:32J25, 32C35, 32W05.

(1) Department of Mathematics, University of Wuppertal, Gaußstr. 20, 42119 Wuppertal, Germany — ruppenthal@uni-wuppertal.de

Article proposé par Vincent Guedj.

(3)

1. Introduction

The present paper is another part of an attempt to create a systematic L2-theory for the∂-operator on singular complex spaces. By a refinement and completion of the techniques introduced in [17], we obtain a better picture for (0, q) and (n, q)-forms on a Hermitian complex spaceX which is of pure dimensionn and has only isolated singularities. The general philosophy is to use a resolution of singularities to obtain a regular model of the L2- cohomology. This complements also the insights of Øvrelid and Vassiliadou in [13].

The key element of our theory is a new kind of canonical sheaf on X introduced in [17] which we denote here byKsX. It is the sheaf of germs of holomorphic square-integrable n-forms which satisfy a Dirichlet boundary condition at the singular set SingX. It comes as the kernel of the∂s-operator on square-integrable (n,0)-forms (see (3.5)). The s-operator is a localized version of the L2-closure of the ∂-operator acting on forms with support away from the singular set (see Section 3 for the precise definition). IfX is a Hermitian complex space with only isolated singularities, we showed in [17, Theorem 1.9], that thes-complex

0→ KsX,→ Fn,0−→ Fs n,1−→ Fs n,2−→s . . .−→ Fn,n→0 (1.1) is a fine resolution of KsX, where theFn,q are the sheaves of germs of L2- forms in the domain of the s-operator. This implies that the cohomology ofKsX is represented by theL2-∂s-cohomology.

The essential idea of ourL2-theory on singular complex spaces is to rep- resent the cohomology of KsX also by L2-∂-cohomology on a resolution of singularities. The key to this lies in the following representation ofKXs:

Theorem 1.1 ([17, Theorem 1.10]). — Let X be a Hermitian complex space of pure dimension with only isolated singularities. Then there exists a resolution of singularitiesπ:MX with at most normal crossings and an effective divisorD>Z− |Z| with support on the exceptional set such that:

KsX ∼=π KM ⊗ O(−D)

, (1.2)

whereKsX is the canonical sheaf for the∂s-operator, KM is the usual canon- ical sheaf onM andZ =π−1(SingX)the unreduced exceptional divisor.

If the exceptional set of the resolution π:MX has only double self- intersections, which is particularly the case ifdimX = 2or ifX is homoge- neous, then one can takeD=Z− |Z| in (1.2).

(4)

By Grauert’s direct image theorem [4], this yields particularly thatKXs is a coherent analytic sheaf. Note that hereπ:MX is not only a resolution ofX, but – in some sense – also a resolution ofKXs, making it locally free.

It is moreover not hard to show that π KM ⊗ O(−Z)

⊂ KsX.

This follows directly from the proof of Lemma 6.1 in [17] and makes it reasonable to conjecture that one can always takeD=Z− |Z|in (1.2) (as we also know that this is possible e.g. if dimX = 2 or if the singularities are homogeneous so that they can be resolved by a single blow-up). In the present paper, we show how the L2-theory from [17] can be refined and complemented if we assume that actuallyD=Z− |Z|.(1) So, assume from now on that this is the case.

We need a few notations to explain our results. IfN is any Hermitian complex manifold, let

cpt:Ap,qcpt(N)→Ap,q+1cpt (N)

be the ∂-operator on smooth forms with compact support in N. Then we denote by

max:Lp,q(N)→Lp,q+1(N) the maximal and by

min:Lp,q(N)→Lp,q+1(N)

the minimal closed Hilbert space extension of the operatorcptas densely de- fined operator fromL2-(p, q)-formsLp,q(N) toL2-(p, q+1)-formsLp,q+1(N).

Let Hmaxp,q (N) be the L2-Dolbeault cohomology on N with respect to the maximal closed extension max, i.e. the ∂-operator in the sense of distri- butions on N, andHminp,q (N) the L2-Dolbeault cohomology with respect to the minimal closed extensionmin. Our first new main result in the present paper is:

Theorem 1.2. — Let (X, h) be a Hermitian complex space of pure di- mensionn>2 with only isolated singularities and π:MX a resolution of singularities with only normal crossings. Assume thatn= dimX = 2or that the singularities ofX are all homogeneous.(2)

(1)We will show e.g. how certain isomorphisms on cohomology from [17] can be realized explicitly by pull-backs or push-forwards of differential forms.

(2)Under this assumption, Theorem 1.1 implies thatKsX=π(KM⊗ O(|Z| −Z)), i.e., that one can useD=Z− |Z|in (1.2). Actually, we prove Theorem 1.2 under the mere assumption thatD=Z− |Z|is a possible choice.

(5)

Let 06p < n,Ω⊂⊂X a relatively compact domain, Ω :=e π−1(Ω) and = Ω−SingX. Provide Ωe with a (regular) Hermitian metric which is equivalent toπhclose to the boundary beΩ.

Let Z := π−1(Ω∩SingX) denote the unreduced exceptional divisor overand let L|Z|−ZM be a Hermitian holomorphic line bundle such that holomorphic sections inL|Z|−Z correspond to holomorphic sections in O(|Z| −Z).

Then the pull-back of formsπinduces a natural injective homomorphism hp:Hminn,p(Ω)−→Hminn,p(eΩ, L|Z|−Z), (1.3) withcokerhp= Γ(Ω, Rpπ(KM⊗ O(|Z| −Z))ifp>1, andh0is an isomor- phism.

Note that singularities in the boundarybΩ of Ω are permitted. Any regu- lar metric onM will do the job if there are no singularities in the boundary of Ω. If Ω =X is compact, then the case p=ncan be included (see Theo- rem 6.11).

The proof of Theorem 1.2 requires a slight variation of Theorem 1.11 in [17], which we give in Section 5 (see Theorem 5.1). We use this oppor- tunity to give a new and more explicit proof of [17, Theorem 1.11], under the hypothesis that D = Z − |Z|. Theorem 1.2 is then deduced by a de- tailed comparison ofmax- withL2loc-cohomology, and ofmin-cohomology with cohomology with compact support (see Section 6). We exploit the fact that the L2- and the L2loc-Dolbeault cohomology are naturally isomorphic on strongly pseudoconvex domains in complex manifolds. Theorem 1.2 is a completely new result that appears neither in [17] nor in [13].

Let us explain briefly how the hypothesis that one can useD=Z− |Z| in (1.2) enters the proof of Theorem 1.2. We require that the homomor- phism (1.3) and, analogously, the injections (5.1), (5.2) in Theorem 5.1 be- low, are induced by pull-back ofL2-forms under the resolutionπ:MX. The crucial point is now that we can show that the pull-back of (n, q)-forms in the domain of thes-operator vanish just to the order ofZ− |Z|on the exceptional set of the resolution, so that we obtain homomorphisms into Cn,qσ (LZ−|Z|), but not intoCσn,q(LD) ifD is of higher order thanZ− |Z|(see Lemma 5.2). The fact that the homomorphisms in (1.3), (5.1), (5.2) are in- duced by pull-back of (n, q)-forms is then essentially used in the commutative diagrams in the proofs of Lemma 6.7 and Theorem 6.10.

The advantage of Theorem 1.2 lies in the fact that it allows to carry over our results to (0, q)-forms by the use of theL2-version of Serre duality.

Besides the L2-Serre duality between the min- and the max-Dolbeault

(6)

cohomology, there is (forp>1) another duality between the higher direct image sheavesRpπKM⊗O(|Z|−Z) on one hand and the flabby cohomology HEn−pofO(Z−|Z|) with support on the exceptional setE=|Z|on the other hand. This is explained in Section 7.1. Combination of these two kinds of duality with Theorem 1.2 leads to the following statement:

Theorem 1.3. — Under the assumptions of Theorem 1.2, let06q6n if Ω =X is compact and 0< q 6n otherwise. Then there exists a natural exact sequence

0→HEq(eΩ,O(Z− |Z|))→Hmax0,q (eΩ, LZ−|Z|)→Hmax0,q (Ω)→0, whereHE is the flabby cohomology with support on the exceptional setE=

|Z|. In case q=n,HEn(eΩ,O(Z− |Z|))has to be replaced by0.

Note again that singularities in the boundarybΩ of Ω are permitted and that any regular metric onM will do the job if there are no singularities in the boundary of Ω. Note also thatHE0(eΩ,O(Z− |Z|)) = 0 by the identity theorem.

The history of Theorem 1.3 is a bit complicated. The idea to identify the kernel of the natural map

Hmax0,q (eΩ, LZ−|Z|)→Hmax0,q (Ω)

as the flabby cohomology ofO(Z− |Z|) with support onEis due to Øvrelid and Vassiliadou [13] who proved Theorem 1.3 recently in the casesq=n−1 and q = n ([13, Theorem 1.4 and Corollary 1.6]). After the present pa- per appeared as a preprint, Øvrelid and Vassiliadou added another proof of our Theorem 1.3 in the case 0 < q < n−1 also to their paper (see [13, Remark 4.5.1]). Their method is quite different from our approach via Theo- rem 1.2 and duality.(3) Concluding, the statement of Theorem 1.3 appeared already in [13], but we give here another, different proof (under the extra condition thatD=Z− |Z|).

For 0< q < n−1, one can show that Hmax0,q (eΩ, LZ−|Z|) HEq(eΩ,O(Z− |Z|))

∼=Hmax0,q Ωe ,

so that we can recover by use of Theorem 1.3 the following result of Øvrelid and Vassiliadou ([13, Theorem 1.3]):

Hmax0,q Ωe∼=Hmax0,q (Ω), 0< q < n−1.

(3)We may remark that Øvrelid and Vassiliadou also use some results from [17].

(7)

The organization of the present paper is as follows. In the Sections 2, 3 and 4, we provide the necessary tools for the proof of Theorem 5.1 in Sec- tion 5. More precisely, in Section 2 we show how to deal with the cohomology of non-exact complexes. Section 3 contains a review of thes-complex as it is introduced in [17]. This comprises the definition of the canonical sheaf of holomorphicn-forms with Dirichlet boundary conditionKsX and the exact- ness of thes-complex (1.1). In Section 4, we recall from [17] howKsX can be represented as the direct image of an invertible sheaf under a resolution of singularities (see Theorem 1.1). In Section 6 we study how the concepts that appear in the context of Theorem 5.1 are related toL2-Dolbeault co- homology and prove Theorem 1.2. Section 7 finally contains the proof of Theorem 1.3.

An outline of the historical development of the topic can be found in the introduction of the previous paper [17] and in [13].

Acknowledgements.The author thanks Nils Øvrelid for many inter- esting and very helpful discussions on the topic. He is also grateful to the unknown referee for many valuable suggestions which helped to improve the exposition of this paper considerably. This research was supported by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation), grant RU 1474/2 within DFG’s Emmy Noether Programme.

2. Cohomology of non-exact complexes

LetX be a paracompact Hausdorff space. In this section, we consider a complex of sheaves of abelian groups

0→ A,→ A0−→ Aa0 1−→ Aa1 2−→ Aa2 3−→. . . (2.1) overX which is exact atAandA0such thatA ∼= kera0, but not necessarily exact at Ap, p > 1. We denote by Kp = kerap the kernel of ap and by Ip = Imap the image ofap. Note thatK0 ∼=A. We will now represent the q-th (flabby) cohomology group Hq(X,A) of A over X by use of the q-th cohomology group of the complex (2.1). Since (2.1) is not a resolution ofA, this also involves the quotient sheaves

Rp:=Kp/Ip−1, p>1, (2.2) which are well-defined sinceapap−1 = 0. To compute the (flabby) coho- mology ofA, we first require that the sheavesAp are acyclic:

(8)

Lemma 2.1. — Assume that the sheavesAp,p>0, in the complex(2.1) are acyclic. Then there are natural isomorphisms

δ0: Γ(X,Ip) ap(Γ(X,Ap))

=

−→H1(X,Kp), (2.3)

δq :Hq(X,Ip)−→= Hq+1(X,Kp) (2.4) for allp>0,q>1.

In this lemma and throughout the whole section, we may as well consider global sections Γcpt and cohomologyHcpt with compact support.

Proof. — We shortly repeat the proof which is standard. For anyp>0, we consider the short exact sequence

0→ Kp,→ Ap−→ Iap p→0. (2.5) UsingHr(X,Ap) = 0 forr>1, we obtain the exact cohomology sequences

0→Γ(X,Kp),→Γ(X,Ap)−→ap Γ(X,Ip) δ

0

−→H1(X,Kp)→0, 0→Hq(X,Ip) δ

q

−→Hq+1(X,Kp)→0, q>1, which yield the statement of the lemma.

The assertion that the isomorphisms are natural means the following. Let 0→ B,→ B0−→ Bb0 1−→ Bb1 2−→ Bb2 3−→. . . (2.6) be another such complex and

0 //A //

f

A

g

0 //B //B

a morphism of complexes (where we abbreviate the complex (2.1) by 0→ A → Aand the complex (2.6) by 0→ B → B). If we denote byLp= kerbp the kernel ofbp and byJp = Imbp the image ofbp, then

Γ(X,Ip) ap(Γ(X,Ap))

δ0 //

gp+10

H1(X,Kp)

g00p

Γ(X,Jp) bp(Γ(X,Bp))

δ0 //H1(X,Lp)

(9)

is commutative, whereg0p+1andgp00are the maps induced by the commutative diagram

0 //Kp //

gp

Ap ap //

gp

Ip //

gp+1

0

0 //Lp //Bp bp //Jp //0.

The statement that the isomorphismsδq are natural follows analogously.

To go on, we need some assumptions onRp:

Lemma 2.2. — Forp>1, assume that the quotient sheafRpof the com- plex(2.1)defined in(2.2)is acyclic and that the natural mappingΓ(X,Kp)→ Γ(X,Rp)is surjective. Then there is a natural isomorphism

Hq(X,Ip−1)−→= Hq(X,Kp) for allq>1 (induced by the inclusionIp−1,→ Kp).

Proof. — Under the assumptions, the proof follows directly from the long exact cohomology sequence that is obtained from the short exact sequence

0→ Ip−1,→ Kp−→ Rp=Kp/Ip−1→0. (2.7) If 0 → B → B is another such complex, then we obtain commutative diagrams as in the proof of Lemma 2.1 showing that the isomorphism is

natural.

Note that the assumptions of Lemma 2.2 are trivially fulfilled if the com- plex (2.1) is exact such that Rp = 0 for all p > 1. From Lemma 2.1 and Lemma 2.2, we deduce by induction:

Lemma 2.3. — Under the assumptions of Lemma 2.1 and Lemma 2.2, there are natural isomorphisms

γp: Γ(X,Ip−1) ap−1(Γ(X,Ap−1))

=

−→Hp(X,A)

for all p>1. Here, natural means the following. If 0→ B → B is another such complex as in(2.6) and

0 //A //

f

A

g

0 //B //B

(10)

a morphism of complexes as in the proof of Lemma 2.1, then we obtain a commutative diagram

Γ(X,Ip−1) ap−1(Γ(X,Ap−1))

γp //

gp0

Hp(X,A)

fp

Γ(X,Jp−1) bp−1(Γ(X,Bp−1))

γp //Hp(X,B),

whereJp−1 = Imbp−1,fp is the map on cohomology induced by f :A → B andgp0 is the map induced by gp:Ap→ Bp. Iff is an isomorphism, thengp0 is an isomorphism as well.

We can now make the connection to the p-th cohomology group of the complex (2.1) by use of:

Lemma 2.4. — Forp>1, assume that the natural mappingΓ(X,Kp)→ Γ(X,Rp)is surjective. Then there is a natural injective homomorphism

ip: Γ(X,Ip−1)

ap−1(Γ(X,Ap−1)) −→Hp(Γ(X,A)) = Γ(X,Kp) ap−1(Γ(X,Ap−1)) withcokerip= Γ(X,Rp). More precisely, the natural sequence

0→ Γ(X,Ip−1) ap−1(Γ(X,Ap−1))

ip

−→Hp(Γ(X,A))−→Γ(X,Rp)→0 is exact.

Proof. — From (2.7) we obtain by use of the assumption the exact se- quence

0→Γ(X,Ip−1)−→Γ(X,Kp)−→Γ(X,Rp)→0, and this induces the natural exact sequence

0→ Γ(X,Ip−1) ap−1(Γ(X,Ap−1))

ip

−→ Γ(X,Kp)

ap−1(Γ(X,Ap−1))−→Γ(X,Rp)→0 since

ap−1(Γ(X,Ap−1))⊂Γ(X,Ip−1)⊂Γ(X,Kp).

Combining Lemma 2.3 and Lemma 2.4, we conclude finally:

Theorem 2.5. — Under the assumptions of Lemma 2.1 and Lemma 2.2, there is for allp>1 a natural injective homomorphism

ip◦(γp)−1:Hp(X,A)−→Hp(Γ(X,A))

with cokerip◦(γp)−1 = Γ(X,Rp). More precisely, there is a natural exact sequence

0→Hp(X,A)i

p◦(γp)−1

−→ Hp(Γ(X,A))−→Γ(X,Rp)→0.

(11)

Here, natural means the following. If0 → B → B is another such complex as in (2.6)and

0 //A //

f

A

g

0 //B //B

a morphism of complexes, then we obtain the commutative diagram Γ(X,Rp)

g000p

Hp(Γ(X,A))

oo

[gp]

Γ(X,Ip−1) ap−1(Γ(X,Ap−1)) ip

oo γp //

g0p

Hp(X,A)

fp

Γ(X,Sp)oo Hp(Γ(X,B)) b Γ(X,Jp−1)

p−1(Γ(X,Bp−1)) ip

oo γp //Hp(X,B),

wherefp is the map on cohomology induced by f :A → B and[gp], g0p,gp000 are the maps induced by gp:Ap→ Bp. The quotient sheavesSp are defined for the complex0→ B → B analogously to the sheavesRp for the complex 0→ A → A. The maps γp are bijective and the mapsip are injective.

In the present paper, we need the following consequence of Theorem 2.5.

Here, we make use of our general assumption thatX is a paracompact Haus- dorff space, because this implies that fine sheaves are acyclic.

Theorem 2.6. — Let X, M be paracompact Hausdorff spaces and π : MX a continuous map. LetCbe a sheaf (of abelian groups) overM and 0→ C,→ C0−→ Cc0 1−→ Cc1 2−→ Cc2 3−→. . . (2.8) a fine resolution. Let A ∼= πC be a sheaf on X, isomorphic to the direct image ofC, and0→ A → A a fine resolution ofAoverX.

Let B := πC be the direct image of C and B =πC the direct image complex (which is again fine but not necessarily exact). Since(2.8)is a fine resolution, the non-exactness of 0 → B → B is measured as above by the higher direct image sheavesSp:=RpπC,p>1. Let

0 //A //

= f

A

g

0 //B //B

be a morphism of complexes, and assume that the complex 0 → B → B satisfies the assumption of Lemma 2.2, i.e. that the direct image sheavesSp are acyclic and that the mapsΓ(X,kerbp)→Γ(X,Sp)are surjective for all p>1.

(12)

Theng induces for allp>1 a natural injective homomorphism Hp(Γ(X,A))−→[gp] Hp(Γ(X,B))

withcoker[gp] = Γ(X,Sp).

More precisely, there is a natural exact sequence

0→Hp(Γ(X,A))−→[gp] Hp(Γ(X,B))−→Γ(X,Sp)→0.

In this sequence, one can replaceHp(Γ(X,B))byHp(Γ(M,C))because Γ(X,Bq) = Γ(π−1(X),Cq) = Γ(M,Cq),

bq(Γ(X,Bq)) =cq(Γ(M,Cq)) by definition for allq>0.

Proof. — The proof follows directly from Theorem 2.5 which we apply to the morphism of complexes

(f, g) : (A,A)→(B,B).

Note that the direct image sheavesBq =πCq, q>0, are still fine sheaves for one can push forward a partition of unity under the continuous mapπ.

Consider the big commutative diagram in Theorem 2.5. Since 0→ A → A is a fine resolution, the quotient sheavesRp,p>1, do vanish such that the mapipin the upper line is an isomorphism. By assumption, the induced map on cohomologyfpis an isomorphism, and sogp0 must also be isomorphic (for the mapsγp are isomorphisms, as well). But then

0→Hp(Γ(X,A))−→[gp] Hp(Γ(X,B))−→Γ(X,Sp)→0

is an exact sequence.

3. Review of the s-complex

3.1. Two ∂-complexes on singular spaces

Let us recall some of the essential constructions from [17]. LetXbe always a (singular) Hermitian complex space of pure dimensionn and UX an open subset. On a singular space, it is most fruitful to consider forms that are square-integrable up to the singular set. Hence, we will use the following concept of locally square-integrable forms:

Lp,qloc(U) :={f ∈Lp,qloc(U−SingX) :f|KLp,q(K−SingX)K⊂⊂U}.

(13)

It is easy to check that the presheaves given as Lp,q(U) :=Lp,qloc(U)

are already sheavesLp,qX. OnLp,qloc(U), we denote by

w:Lp,qloc(U)→Lp,q+1loc (U)

the∂-operator in the sense of distributions on U −SingX which is closed and densely defined. The subscript refers to w as an operator in a weak sense. We can define the presheaves of germs of forms in the domain ofw,

Cp,q :=Lp,q−1w Lp,q+1,

given byCp,q(U) =Lp,q(U)∩Domw(U). It is not hard to check that these are actually already sheaves.

Moreover, it is easy to see that the sheavesCp,q admit partitions of unity, and so we obtain fine sequences

Cp,0−→ Cw p,1−→ Cw p,2−→w . . . (3.1) We will see later, when we deal with resolution of singularities, that

KX:= kerw⊂ Cn,0

is just the canonical sheaf of Grauert and Riemenschneider since the L2- property of (n,0)-forms remains invariant under modifications of the metric.

TheL2,loc-Dolbeault cohomology with respect to thew-operator on an open setUX is the cohomology of the complex (3.1) which is denoted by Hq(Γ(U,Cp,∗)).

Secondly, we need a suitable local realization of a minimal version of the

∂-operator. This is the ∂-operator with a Dirichlet boundary condition at the singular set SingX ofX. Let

s:Lp,qloc(U)→Lp,q+1loc (U)

be defined as follows. We say that f ∈ Domw is in the domain of s if there exists a sequence of forms{fj}j ⊂DomwLp,qloc(U) with essential support away from the singular set,

suppfj∩SingX =∅, such that

fjf inLp,q(K−SingX), (3.2)

wfjwf inLp,q+1(K−SingX) (3.3) for each compact subsetK⊂⊂U. The subscript refers tosas an extension in a strong sense. Note that we can assume without loss of generality (by

(14)

use of cut-off functions and smoothing with Dirac sequences) that the forms fj are smooth with compact support in U−SingX.

We define the presheaves of germs of forms in the domain ofs, Fp,q :=Lp,q−1s Lp,q+1,

given byFp,q(U) =Lp,q(U)∩Doms(U). It is not hard to see that theFp,q are already sheaves (see [17, Section 6.1]).

As forCp,q, it is clear that the sheavesFp,q are fine, and we obtain fine sequences

Fp,0−→ Fs p,1−→ Fs p,2−→s . . . (3.4) We can now introduce the sheaf

KsX:= kers⊂ Fn,0 (3.5)

which we may call the canonical sheaf of holomorphicn-forms with Dirichlet boundary condition. One of the main objectives of the present paper is to compare different representations of the cohomology of KsX. One of them will be theL2,loc-Dolbeault cohomology with respect to the s-operator on open setsUX, i.e. the cohomology of the complex (3.4) which is denoted byHq(Γ(U,Fp,∗)).

3.2. LocalL2-solvability for (n, q)-forms

It is clearly interesting to study wether the sequences (3.1) and (3.4) are exact, which is well-known to be the case in regular points ofX where the

w- and the s-operator coincide. In singular points, the situation is quite complicated for forms of arbitrary degree and not completely understood.

However, thew-equation is locally solvable in theL2-sense at arbitrary sin- gularities for forms of degree (n, q), q > 0 (see [14, Proposition 2.1]), and for forms of degree (p, q), p+q > n, at isolated singularities (see [3, The- orem 1.2]). Since we are concerned with canonical sheaves, we may restrict our attention to the case of (n, q)-forms and conclude:

Theorem 3.1. — Let X be a Hermitian complex space of pure dimen- sionn. Then

0→ KX,→ Cn,0−→ Cw n,1−→ Cw n,2−→w . . .−→ Cn,n→0 (3.6) is a fine resolution. For an open setUX, it follows that

Hq(U,KX)∼=Hq(Γ(U,Cn,∗)), Hcptq (U,KX)∼=Hqcpt(U,Cn,∗)).

(15)

Concerning thes-equation, localL2-solvability for forms of degree (n, q) is known to hold on spaces with isolated singularities (see [17, Theorem 1.9]), but the problem is open at arbitrary singularities. If X has only isolated singularities, then the s-equation is locally exact on (n, q)-forms for 1 6 q6n by [17, Lemma 5.5]. The statement was deduced from the results of Fornæss, Øvrelid and Vassiliadou [3].

4. Resolution of(X,KsX) 4.1. Desingularization and comparison of metrics

Let π : MX be a resolution of singularities (which exists due to Hironaka [8]), i.e. a proper holomorphic surjection such that

π|M−E:MEX−SingX

is biholomorphic, whereE =|π−1(SingX)|is the exceptional set. We may assume that E is a divisor with only normal crossings, i.e. the irreducible components of E are regular and meet complex transversely. Let Z :=

π−1(SingX) be the unreduced exceptional divisor. For the topic of desingu- larization, we refer to [1], [2] and [6]. Let γ :=πh be the pullback of the Hermitian metrichofX toM. γis positive semidefinite (a pseudo-metric) with degeneracy locusE.

We giveM the structure of a Hermitian manifold with a freely chosen (positive definite) metricσ. Then γ.σ and γσ on compact subsets of ME. For an open set UM, we denote by Lp,qγ (U) and Lp,qσ (U) the spaces of square-integrable (p, q)-forms with respect to the (pseudo-)metrics γandσ, respectively.

Sinceσ is positive definite andγ is positive semi-definite, a straightfor- ward comparison of the volume forms (see [17, Section 2.2]) leads to the inclusions

Ln,qγ (U)⊂Ln,qσ (U), (4.1) L0,qσ (U)⊂L0,qγ (U). (4.2) for relatively compact open setsU ⊂⊂M and all 06q6n.

For an open set Ω ⊂ X, Ω = Ω−SingX, Ω :=e π−1(Ω), pullback of forms underπgives the isometry

π:Lp,q(Ω)−→Lp,qγ (eΩ−E)∼=Lp,qγ (eΩ), (4.3) where the last identification is by trivial extension of forms over the thin exceptional setE.

(16)

4.2. Representation ofKXs under desingularization

By use of (4.3), both complexes, the resolution (Cn,∗, ∂w) ofKX and the resolution (Fn,∗, ∂s) ofKsX, can be studied as well on the complex manifold M. This is the point of view that was taken in [17] where we considered the sheavesLp,qγ onM given by

Lp,qγ (U) :=Lp,qγ,loc(U) and

Cγ,Ep,q :=Lp,qγ−1w,ELp,q+1γ , (4.4) where w,E is the ∂-operator in the sense of distributions with respect to compact subsets ofM−E. (4.4) is given by the presheafCγ,Ep,q(U) =Lp,qγ (U)∩

Domw,E(U). It follows from (4.3) that (Cp,∗, ∂w) can be canonically iden- tified with the direct image complex (πCγ,Ep,∗, πw,E). SinceLn,0γ =Ln,0σ for the regular metric σ on M by use of (4.1) and (4.2), we can use the fact that the∂-equation in the sense of distributions forL2σ-forms extends over exceptional sets (see e.g. [16, Lemma 2.1]) to conclude that

KM := kerw,E ⊂ Ln,0γ =Ln,0σ

is just the usual canonical sheaf on the complex manifoldM, and thatKX∼= πKM is in fact the canonical sheaf of Grauert–Riemenschneider.

Analogously, we consider now thes-complex. Lets,Ebe the∂-operator acting onLp,qγ -forms, defined as thes-operator onX above, but with the exceptional setE in place of the singular set SingX. Let

Fγ,Ep,q :=Lp,qγ−1s,ELp,q+1γ .

Then it follows from (4.3) that (Fp,∗, ∂s) can be canonically identified with the direct image complex (πFγ,Ep,∗, πs,E).

It remains to study kers,E inLn,0γ =Ln,0σ becauseπ(Fγ,En,0∩kers,E) = KsX. This was a central point [17] (see [17, Section 6.3]): There exists a resolution of singularities π : MX with only normal crossings and an effective divisor D > Z− |Z| with support on the exceptional set, where Z=π−1(SingX) is the unreduced exceptional divisor, such that

KsX =π KM ⊗ O(−D)

. (4.5)

In many situations, e.g., if dimX = 2 or ifX has only homogeneous singu- larities, then one can takeD=Z− |Z|. In the present paper, we will assume from now on that this is actually the case, and show how the L2-theory from [17] can be improved under this assumption.

So, for the rest of the paper, we assume that (4.5) holds withD=Z−|Z|.

(17)

5. Resolution ofL2loc-cohomology of (n, q)-forms

In this Section, we prove a slight variation of Theorem 1.11 in [17]:

Theorem 5.1. — Let X be a Hermitian complex space of pure dimen- sionn >2 with only isolated singularities and π :MX a resolution of singularities with only normal crossings such that

KsX∼=π KM⊗ O(|Z| −Z) ,

whereKsX is the canonical sheaf for the∂s-operator (i.e. the canonical sheaf of holomorphic (n,0)-forms with Dirichlet boundary condition), KM is the usual canonical sheaf onM andZ =π−1(SingX)the unreduced exceptional divisor.

Then the pull-back of forms under π induces for p > 1 natural exact sequences

0→Hp(X,KsX)

p]

−→Hp(M,KM⊗ O(|Z| −Z))−→Γ(X,Rp)→0, (5.1) 0→Hcptp (X,KsX)

p]

−→Hcptp (M,KM ⊗ O(|Z| −Z))−→Γ(X,Rp)→0, (5.2) whereRp is the higher direct image sheafRpπ(KM⊗ O(|Z| −Z)).

In [17], we stated Theorem 5.1 under the additional assumption thatX is compact (see [17, Theorem 1.11]), but the proof in [17] actually gives also the statement of Theorem 5.1 above. New in the present paper is the proof of Theorem 5.1. Whereas in [17], we just gave an abstract proof by use of the Leray spectral sequence, we give here an explicit realization of all the mappings in (5.1) and (5.2) in terms of differential forms. Clearly, statement (5.2) is also new because it is contained in (5.1) ifX is compact.

The proof of Theorem 5.1 is based on the following observation. IfCn,∗

is a fine resolution ofKM⊗ O(|Z| −Z), then the non-exactness of the direct image complexπCn,∗ can be expressed by the higher direct image sheaves Rp =RpπKM⊗ O(|Z| −Z), p>1. These are skyscraper sheaves because X has only isolated singularities. So, they are acyclic. On the other hand, global sections inRp can be expressed globally by L2-forms with compact support. These two properties allow to express the cohomology of the canon- ical sheafKXs in terms of the cohomology of the direct image complexπCn,∗

modulo global sections in Rp. Another difficulty is to show that the exact sequences in Theorem 5.1 are actually induced by the pull-back ofL2-forms (see Lemma 5.2). We will also see that the surjections in (5.1) and (5.2) are simply induced by taking germs of differential forms. So, we obtain an explicit and localized proof of Theorem 1.11 in [17].

(18)

5.1. Proof of Theorem 5.1

We can use Theorem 2.6 to represent the cohomology groups Hq(X,KsX)∼=Hq(Γ(X,Fn,∗)),

Hcptq (X,KsX)∼=Hqcpt(X,Fn,∗))

in terms of cohomology groups on the resolution π : MX, where we letπ : MX be a resolution of singularities as in Theorem 1.1 so that D = Z − |Z|. Recall also that X is a Hermitian complex space of pure dimensionnwith only isolated singularities.

Clearly, we intend to use Theorem 2.6 withA=KsX and (A, a) = (Fn,∗, ∂s),

so that 0→ A → A is a fine resolution ofAoverX by [17, Theorem 1.9].

By assumption,

A=KsX∼=π KM⊗ O(|Z| −Z) , so that we can choose

C=KM ⊗ O(|Z| −Z)

for the application of Theorem 2.6. It remains to choose a suitable fine res- olution 0 → C → C. Since (M, σ) is an ordinary Hermitian manifold, we can use the usualL2σ-complex of forms with values inO(|Z| −Z). To realize that, we can adopt two different points of view. First, letL|Z|−ZM be the holomorphic line bundle associated to the divisor|Z| −Zsuch that holo- morphic sections of L|Z|−Z correspond to sections of O(|Z| −Z), and give L|Z|−Z the structure of a Hermitian line bundle by choosing an arbitrary positive definite Hermitian metric. Then, denote by

Lp,qσ (U, L|Z|−Z), Lp,qσ,loc(U, L|Z|−Z)

the spaces of (locally) square-integrable (p, q)-forms with values in L|Z|−Z (with respect to the metricσon M and the chosen metric on L|Z|−Z). We can then define the sheaves of germs of square-integrable (p, q)-forms with values inL|Z|−Z,Lp,qσ (L|Z|−Z), by the assignment

Lp,qσ (L|Z|−Z)(U) =Lp,qσ,loc(U, L|Z|−Z).

The second point of view is to use the sheavesLp,qσ ⊗ O(|Z| −Z) which are canonically isomorphic to the sheavesLp,qσ (L|Z|−Z). Let us keep both points of view in mind. As in (4.4), let

Cσ,Ep,q(L|Z|−Z) :=Lp,qσ (L|Z|−Z)∩−1w,ELp,q+1σ (L|Z|−Z), (5.3) where w,E is the ∂-operator in the sense of distributions with respect to compact subsets of ME (and for forms with values inL|Z|−Z). Since σ is positive definite, the∂-equation in the sense of distributions forL2σ-forms

Références

Documents relatifs

Soit (n, V) une représentation irréductible de G ayant un caractère central, et de dimension infinie.. Les fonctions de Kirillov

Toute utilisation commerciale ou impression systématique est constitutive d’une infrac- tion pénale.. Toute copie ou impression de ce fichier doit contenir la présente mention

L’accès aux archives de la revue « Nouvelles annales de mathématiques » implique l’accord avec les conditions générales d’utilisation ( http://www.numdam.org/conditions )..

L’accès aux archives de la revue « Nouvelles annales de mathématiques » implique l’accord avec les conditions générales d’utilisation ( http://www.numdam.org/conditions )..

— Dans la formule qui doni^e 2 cos ma, on a divisé les deux membres par 2cos#, pour donner la même physionomie à toutes

L’accès aux archives de la revue « Nouvelles annales de mathématiques » implique l’accord avec les conditions générales d’utilisation ( http://www.numdam.org/conditions )..

Mais, comme les formules qu'il s'agit de développer sont fort simples, il vaut mieux éviter l'emploi d'une formule plus compliquée; cela se fait facilement à Faide de la formule tout

In this section, some useful performance measures of the proposed model like expected number of customers in the queue E(Q), expected length of idle period E(I), expected length of