Arxiv:0712.2207V8 [Math.AG] 21 Aug 2008 Hr-Elter Se[R-A H 3 Ch

Home , Chern class

arXiv:0712.2207v8 [math.AG] 21 Aug 2008 hr-elter se[r-a h 3 Ch. [Gri-Ha, (see theory Chern-Weil bandb h xoeta xc eune hs lse r c are classes These sequence. exact exponential the by obtained eRa cohomology Rham de Let HR LSE NDLGECHMLG O OEETANALYTIC COHERENT FOR COHOMOLOGY DELIGNE IN CLASSES CHERN ..Pouto the on Product 7.2. ..Analytic 7.4. Functoriality 7.3. References ..Dfiiino h analytic the of Definition 7.1. ..Cmaiiiyo hr lse n h R formula GRR Analytic the and Appendix. classes Chern of 7. formula Compatibility GRR the morp of 6.2. projective Proof for theorem Grothendieck-Riemann-Roch 6.1. formula proje The Whitney of sheaves the torsion of 6. coherent Proof for theorem structure 5.3. A 5.2. .Dlgechmlg n hr lse o oal resevs7 where case the to Reduction case general formula the Whitney 5.1. in The classes rank the positive of 5. of Construction sheaves of case 4.3. The sheaves torsion for 4.2. Construction classes Chern bundles of 4.1. vector Construction holomorphic for classes 4. sheaves Chern free locally for cohomology 3.2. classes Deligne Chern and cohomology 3.1. Deligne conventions and 3. Notations 2. Introduction 1. X easot ieetal aiodand manifold differentiable smooth a be Abstract. t rpryudrplbcs h hte oml n h Gr t manif the compact that and complex prove formula smooth We Whitney between morphisms the manifold. projective pullbacks, compact under complex property smooth ity a on sheaves K nti ril,w osrc hr lse nrtoa Deli rational in classes Chern construct we article, this In ter ihspoti iio ihsml omlcosn 45 crossing normal simple with divisor a in support with -theory H K 2 i ter ihsupport with -theory ( K X, ter ihsupport with -theory R ,wihgnrlz h rtCencaso iebnl in bundle line a of class Chern first the generalize which ), K ter ihsupport with -theory F § and ] osrcsclasses constructs 3]) 1. UINGRIVAUX JULIEN G Introduction SHEAVES Contents r oal reand free locally are E 1 eacmlxvco udeo rank of bundle vector complex a be olds. tedekReanRc hoe for theorem othendieck-Riemann-Roch c mail ihplbcsudrsmooth under pullbacks with ompatible i ( tv ieso n 34 one dimension ctive eecassstsytefunctorial- the satisfy classes hese E H ) top im 41 hisms satrinsef32 sheaf torsion a is n ooooyfrcoherent for cohomology gne 1 , ≤ i ≤ r ihvle nthe in values with , r on H X 2 ( The . X, Z 43 44 44 47 43 42 41 36 32 28 25 14 14 13 7 6 1 ) 2 JULIEN GRIVAUX morphisms and verify the Whitney sum formula c (E F )top = c (E)top c (F )top. k ⊕ i j i+Xj=k top 2i There exist more refined ways of defining ci(E) in H (X, Z). The first method is due to Chow (see the introduction of [Gro1]). The idea is to define explicitly the Chern classes of the universal bundles of the grassmannians and to write any complex vector bundle as a quotient of a trivial vector bundle. Of course, computations have to be done on the grassmannians to check the compatibilities. Note that in the holomorphic or in the algebraic context, a vector bundle is not in general a quotient of a trivial vector bundle. Nevertheless, if X is projective, this is true after tensorising by a sufficiently high power of an ample line bundle and the construction can be adapted (see [Br]). A more intrinsic construction is the splitting method, introduced by Grothendieck in [Gro1]. Let us briefly recall how it works. By the Leray-Hirsh theorem, we know that H ∗(P(E), Z) is a free module over H ∗(X, Z) with basis 1, α,...,αr−1, where α is the opposite of the first Chern class of the relative Hopf bundle on P(E). Now the Chern classes of E are uniquely defined by the relation αr + p ∗c (E)top αr−1 + + p ∗ctop (E) α + p ∗c (E)top =0. 1 ··· r−1 r (see Grothendieck [Gro1], Voisin [Vo1, Ch.11 2], and Zucker [Zu, 1]). § § The splitting method works amazingly well in various contexts, provided that we have – the definition of the first Chern class of a line bundle, – a structure theorem for the cohomology ring of a projective bundle considered as a module over the cohomology ring of the base. Let us now examine the algebraic case. Let X be a smooth algebraic variety over a field k of characteristic alg zero, and E be an algebraic bundle on X. Then the splitting principle allows to define ci(E) – in the Chow ring CHi(X) if X is quasi-projective, 2i – in the algebraic de Rham cohomology group HDR(X/k). Suppose now that k = C. Then Grothendieck’s comparison theorem (see [Gro3]) says that we have a 2i 2i an canonical isomorphism between HDR(X/C ) and H (X , C ). It is important to notice that the class c (E)alg is mapped to (2π√ 1)ic (E)top by this morphism. i − i Next, we consider the problem in the abstract analytic setting. Let X be a smooth complex analytic p,q manifold and E be a holomorphic vector bundle on X. We denote by C (X) the space of complex p,q A differential forms of type (p, q) on X and we put C(X) = p,q C (X). The Hodge filtration on p,q A A C(X) is defined by F i C(X) = (X). It induces a filtration F iHk(X, C ) on Hk(X, C ). A A p≥i,q AC L For a detailed exposition see [Vo1,L Ch. 7 and 8]. Let ΩX• be the holomorphic de Rham complex on X. This is a complex of locally free sheaves. We can consider the analytic de Rham cohomology k+i >i H (X, ΩX• ) which is the hypercohomology of the truncated de Rham complex. The maps of complexes Ω >i / Ω and Ω >i / Ω i [ i] give two maps Hk+i(X, Ω >i) / F iHk(X, C ) and X• X• X• X − X• k+i >i k i H (X, ΩX• ) / H (X, ΩX ). In the compact Kähler case, the first map is an isomorphism, but it is no longer true in the general case. We will denote by Hp,q(X) the cohomology classes in Hp+q(X, C ) which admit a representative in p,q(X). A If E is endowed with the Chern connection associated to a hermitian metric, the de Rham representative top i 2i−1 of ci(E) obtained by Chern-Weil theory is of type (i,i) and is unique modulo d(F X ). This allows 2i >i i,i i i A to define ci(E) in H (X, ΩX• ), and then in H (X) and H (X, ΩX ). The notations for these three an hodge dolb classes will be ci(E) , ci(E) and ci(E) . CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 3

top If we forget the holomorphic structure of E, we can consider its topological Chern classes ci(E) in the 2i top 2i hodge top Betti cohomology groups H (X, Z). The image of ci(E) in H (X, C ) is ci(E) . Thus ci(E) is an integral cohomology class whose image in H2i(X, C ) lies in F iH2i(X, C ). Such classes are called Hodge classes of weight 2i. 2i i 2i 2i >i i i The Chern classes of E in H (X, Z), F H (X, C ), H X, ΩX• and H X, ΩX appear in the following diagram: H2i(X, Z) c (E)top ∋ i

c (E)an H2i(X, Ω >i) /F iH2i(X, C ) c (E)hodge i ∈ X• ∋ i

c (E)dolb Hi(X, Ω i ) i ∈ X This means in particular that these different classes are compatible with the Hodge decomposition in the compact Kähler case, and in general via the Hodge / de Rham spectral sequence. Furthermore, the an dolb hodge knowledge of ci(E) allows to obtain the two other classes ci(E) and ci(E) , but the converse is an not true. Thus ci(E) contains more information (except torsion) than the other classes in the diagram. p Recall now the Deligne cohomology groups HD(X, Z(q)) (see [Es-Vi, 1] and Section 3.1). We will 2i § 2i be mainly interested in the cohomology groups HD (X, Z(i)). They admit natural maps to H (X, Z) 2i >i and to H (X, ΩX• ), which are compatible with the diagram above. Furthermore, there is an exact sequence 0 / H2i−1(X, Ω 6i−1) / / H2i(X, Z(i)) / H2i(X, Z), (see [Vo1] and X• H2i−1(X, Z) D Proposition 3.3 (i)). Thus a Deligne class is a strong refinement of any of the above mentioned classes. The splitting method works for the construction of c (E) in H2i(X, Z(i)), as explained in [Zu, 4], and i D § [Es-Vi, 8]. These Chern classes, as all the others constructed above, satisfy the following properties: § – they are functorial with respect to pullbacks. – if 0 / E / F / G / 0 is an exact sequence of vector bundles, then for all i

ci(F )= cp(E) cq(G). p+q=i X The last property means that the total Chern class c = 1+c + +c is defined on the Grothendieck group 1 ··· n K(X) of holomorphic vector bundles on X and satisfies the additivity property c(x + x′)= c(x)c(x′). Now, what happens if we work with coherent sheaves instead of locally free ones? If X is quasi-projective and is an algebraic coherent sheaf on X, there exists a locally free resolution F 0 / E / / E / / 0. 1 ··· N F (This is still true under the weaker assumption that X is a regular separated scheme over C by Kleiman’s lemma; see [SGA 6, II, 2.2.7.1]). The total Chern class of is defined by F c( )= c(E ) c(E )−1c(E ) ... F N N−1 N−2 The class c( ) does not depend on the locally free resolution (see [Bo-Se, 4 and 6]). More formally, if F § G(X) is the Grothendieck group of coherent sheaves on X, the canonical map ι : K(X) / G(X) is an isomorphism. The inverse is given by [ ] / [E ] [E ] + [E ] ... F N − N−1 N−2 − In what follows, we consider the complex analytic case. The problem of the existence of global locally free resolutions in the analytic case has been opened for a long time. For smooth complex surfaces, such resolutions always exist by [Sch]. More recently, Schröer and Vezzosi proved in [ScVe] the same result for 4 JULIEN GRIVAUX singular separated surfaces. Nevertheless, for varieties of dimension at least 3, a negative answer to the question is provided by the following counterexample of Voisin: Theorem [Vo2] On any generic complex torus of dimension greater than 3, the ideal sheaf of a point does not admit a global locally free resolution.

Worse than that, even if admits a globally free resolution E •, the method of Borel and Serre [Bo-Se] F −1 does not prove that c(EN ) c(EN−1) c(EN−2) ... is independent of E •. In fact, the crucial point in their argument is that every coherent sheaf should have a resolution. Nevertheless Borel-Serre’s method applies in a weaker context if we consider Chern classes in H ∗(X, Z). ω ω Indeed, if is a coherent sheaf on X and is the sheaf of real-analytic functions on X, then O F CX F⊗ X CX admits a locally free real-analytic resolution by the Grauert vanishing theorem [Gra]. We obtain by this method topological Chern classes c ( )top in H2i(X, Z). i F It is natural to require that ci should take its values in more refined rings depending on the holomorphic structure of and X. Such a construction has been carried out by Atiyah for the Dolbeault cohomology F ring in [At]. Let us briefly describe his method: the exact sequence 0 / Ω1 / ℘1 ( ) / / 0 F ⊗ X X F F of principal parts of of order one (see [EGA IV, 16.7]) gives an extension class (the Atiyah class) a( ) 1 1 F § F in ExtO ( , ΩX ). Then cp( ) is the trace of the p-th Yoneda product of a( ). These classes are used X F F⊗ F F by O’Brian, Toledo and Tong in [OB-To-To1] and [OB-To-To2] to prove the Grothendieck-Riemann-Roch theorem on abstract manifolds in the Hodge ring. The Atiyah class has been constructed by Grothendieck and Illusie for perfect complexes (see [Ill, Ch. 5]). Nevertheless, if X is not a Kähler manifold, there is p p 2p >p no good relation between H (X, ΩX ) and H (X, ΩX• ), as the Frölicher spectral sequence may not degenerate at E1 for example. In this context, the most satisfactory construction was obtained by Green in his unpublished thesis (see [Gre] and [To-To]). He proved the following theorem Gr Theorem 1 [Gre], [To-To] Let be a coherent sheaf on X. Then there exist Chern classes ci( ) in F 2i >i F the analytic de Rham cohomology groups H (X, ΩX• ) which are compatible with Atiyah Chern classes and topological Chern classes. In order to avoid the problem of nonexistence of locally free resolutions, he introduced the notion of a simplicial resolution by simplicial vector bundles with respect to a given covering. Green’s basic result is the following: Theorem 2 [Gre], [To-To] Any coherent sheaf on a smooth complex compact manifold admits a finite simplicial resolution by simplicial holomorphic vector bundles. The next step in order to obtain Theorem 1 above, is to define the Chern classes of a simplicial vector bundle. For this, Green uses Bott’s construction (see [Bott]) which can be adapted to the simplicial context. Though, it is not clear how to extend Green’s method to Deligne cohomology. Let us now state the main result of this article: Theorem 1.1. Let X be a complex compact manifold. For every coherent sheaf on X, we can define 2p F classes c ( ) and ch ( ) in H (X, Q(p)) such that: p F p F D (i) For every exact sequence 0 / / / / 0 of coherent sheaves on X, we have F G H c( )= c( )c( ) and ch( ) = ch( ) + ch( ). The total Chern class c: G(X) / H ∗ (X, Q)× G F H G F H D ∗ is a group morphism and the Chern character ch: G(X) / HD(X, Q) is a ring morphism. CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 5

(ii) If f : X / Y is holomorphic and y is an element of G(Y ), then c(f !y) = f ∗c(y), where f ! : G(Y ) / G(X) is the pullback in analytic K-theory. (iii) If is a locally free sheaf, then c( ) is the usual Chern class in rational Deligne cohomology. E E (iv) If Z is a smooth closed submanifold of X and a coherent sheaf on Z, the Grothendieck-Riemann- F Roch (GRR) theorem is valid for i , , namely Z F −1 ch i = i ch( ) td N . Z∗F Z∗ F Z/X (v) If f : X / Y is a projective morphism between smooth complex compact manifolds, for every coherent sheaf on X, we have the Grothendieck-Riemann-Roch theorem F ch f [ ] td(Y )= f [ch( )] td(X). ! F ∗ F Our approach is completely different from [Bo-Se]. Indeed, Voisin’s result prevents from using locally free resolutions. Our geometric starting point, which will be used instead of locally free resolutions, is the following (Theorem 4.11): Theorem Let be a coherent sheaf on X of generic rank r. Then there exists a bimeromorphic morphism F π: X / X and a locally free sheaf on X of rank r, together with a surjective map π ∗ / . Q F Q It follows that up to torsion sheaves, π ![ ] is locally free. This will allow us to define our Chern classes e F e by induction on the dimension of the base. Of course, we will need to show that our Chern classes satisfy the Whitney formula and are independent of the bimeromorphic model X. The theorem above is a particular case of Hironaka’s flattening theorem (see [Hiro2] and in the algebraic e case [Gr-Ra]). Indeed, if we apply Hironaka’s result to the couple ( , id), there exists a bimeromorphic F map σ : X / X such that σ ∗ σ ∗ is flat with respect to the identity morphism, and thus F F tor locally free. For the sake of completeness, we include an elementary proof of Theorem 4.11. e Property (iv) of Theorem 1.1 is noteworthy. The lack of global resolutions (see [Vo2]) prevents from using the proofs of Borel, Serre and of Baum, Fulton and McPherson (see [Fu, Ch. 15 2]). The equivalent § formula in the topological setting is proved in [At-Hi]. In the holomorphic context, O’Brian, Toledo and Tong ([OB-To-To3]) prove this formula for the Atiyah Chern classes when there exists a retraction from X to Z, then they establish (GRR) for a projection and they deduce that (GRR) is valid for any holomorphic map, so a posteriori for an immersion (see [OB-To-To2]). Nevertheless, our result does not give a new proof of (GRR) formula for an immersion in the case of the Atiyah Chern classes. Indeed, the compatibility between our construction and the Atiyah Chern classes is a consequence of the (GRR) theorem for an immersion in both theories, as explained further. Property (v) is an immediate consequence of (iv), as originally noticed in [Bo-Se], since the natural map N N from G(X) Z G(P ) to G(X P ) is surjective (see [SGA 6, ExposéVI] and [Bei]). Yet, we do not ⊗ × obtain the (GRR) theorem for a general holomorphic map between smooth complex compact manifolds. Remark that c ( ) is only constructed in the rational Deligne cohomology group H2p(X, Q(p)). The p F D reason is that we make full use of the Chern character, which has denominators, and thus determines the total Chern class only up to torsion classes. We think that it could be possible to define c ( ) in p F H2p(X, Z(p)) following our approach, but with huge computations. For p = 1, c ( ) can be easily con- D 1 F structed in H2 (X, Z(1)). Indeed, it suffices to define c ( )= c (det ), where det is the determinant D 1 F 1 F F line bundle of (see [Kn-Mu]). F It is interesting to compare the classes of Theorem 1.1 with other existing theories. We adopt a more general setting by using Theorem 4.11. We prove that, for any cohomology ring satisfying reasonable properties, a theory of Chern classes can be completely determined if we suppose that the GRR formula is valid for immersions. More precisely, our statement is the following (Theorem 6.3): 6 JULIEN GRIVAUX

Theorem Under the hypotheses (α)-(δ) of page 42 on the cohomology ring, a theory of Chern classes for coherent sheaves on smooth complex compact manifolds which satisfies the GRR theorem for immersions, the Whitney additivity formula and the functoriality formula is completely determined by the first Chern class of holomorphic line bundles. This theorem yields compatibility results (Corollary 6.5): Corollary The classes of Theorem 1.1 are compatible with the rational topological Chern classes and the Atiyah Chern classes. Nevertheless, since GRR for immersions does not seem to be known for the Green Chern classes if X is not Kähler, the theorem above does not give the compatibility in this setting. In fact, the compatibility is equivalent to the GRR theorem for immersions for the Green Chern classes. Let us mention the link of our construction with secondary characteristic classes. We can look at a subring of the ring of Cheeger-Simons characters on X which are the “holomorphic” characters (that is the G-cohomology defined in [Es, 4], or equivalently the restricted differential char- § acters defined in [Br, 2]). This subring can be mapped onto the Deligne cohomology ring, but not in § an injective way in general. When E is a holomorphic vector bundle with a compatible connection, the Cheeger-Simons theory (see [Ch-Si]) produces Chern classes with values in this subring. It is known that these classes are the same as the Deligne classes (see [Br] in the algebraic case and [Zu, 5] for the general § case). When E is topologically trivial, this construction gives the so-called secondary classes with values in the intermediate jacobians of X (see [Na] for a different construction, and [Ber] who proves the link with the generalized Abel-Jacobi map). The intermediate jacobians of X have been constructed in the Kähler case by Griffiths. They are complex tori (see [Vo1, Ch.12 1], and [Es-Vi, 7 and 8]). If X is not § § Kähler, intermediate jacobians can still be defined but they are no longer complex tori. Our result provides similarly refined Chern classes for coherent sheaves, and in particular, secondary invariants for coherent sheaves with trivial topological Chern classes. The organization of the paper is the following. We recall in Part 3 the basic properties of Deligne cohomology and Chern classes for locally free sheaves in Deligne cohomology. The necessary results of analytic K-theory with support are grouped in Appendix 7; they will be used extensively throughout the paper. The rest of the article is devoted to the proof of Theorem 1.1. The construction of the Chern classes is achieved by induction on dim X. In Part 4 we perform the induction step for torsion sheaves using the (GRR) formula for the immersion of smooth divisors; then we prove a dévissage theorem which enables us to break any coherent sheaf into a locally free sheaf and a torsion sheaf on a suitable modification of X; this is the key of the construction of c ( ) when has strictly positive rank. The i F F Whitney formula, which is a part of the induction process, is proved in Part 5. After several reductions, we use a deformation argument which leads to the deformation space of the normal cone of a smooth hypersurface. We establish the (GRR) theorem for the immersion of a smooth hypersurface in Part 4, in Part 6 we recall how to deduce the general (GRR) theorem for an immersion from this particular case, using excess formulae, then we deduce uniqueness results using Theorem 4.11.

Acknowledgement. I wish to thank Claire Voisin for introducing me to this beautiful subject and for many helpful discussions. I also thank Pierre Schapira for enlightening conversations.

2. Notations and conventions All manifolds are complex smooth analytic connected manifolds. All the results are clearly valid for non connected ones, by reasoning on the connected components. Except in Section 3, all manifolds are compact. By submanifold, we always mean a closed submanifold. CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 7

Holomorphic vector bundles The rank of a holomorphic vector bundle is well defined since the manifolds are connected. If E is a holomorphic vector bundle, we denote by the associated locally free sheaf. The letter always denotes E E a locally free sheaf. Coherent sheaves If is a coherent analytic sheaf on a smooth connected manifold X, then is locally free outside a F F proper analytic subset S of X (see [Gra-Re]). Then U = X S is connected. By definition, the generic \ rank of is the rank of the locally free sheaf . If the generic rank of vanishes, is supported in a F F|U F F proper analytic subset Z of X, it is therefore annihilated by the action of a sufficiently high power of the ideal sheaf . Conversely, if is a torsion sheaf, is identically zero outside a proper analytic subset IZ F F of X, so it has generic rank zero. The letter always denotes a torsion sheaf. T Divisors A strict normal crossing divisor D in X is a formal sum m D + + m D , where D , 1 i N, 1 1 ··· N N i ≤ ≤ are smooth transverse hypersurfaces and mi, 1 i N, are nonzero integers. If all the coefficients ≤ ≤ red mi are positive, D is effective. In that case, the associated reduced divisor D is the effective divisor D + + D . A strict simple normal crossing divisor is reduced if for all i, m = 1. We make no 1 ··· N i difference between a reduced divisor and its support. By a normal crossing divisor we always mean a strict normal crossing divisor. If D is an effective simple normal crossing divisor, it defines an ideal sheaf = ( D). The associated quotient sheaf is denoted ID OX − by . OD We use frequently Hironaka’s desingularization theorem [Hiro1] for complex spaces as stated in [An-Ga, Th.7.9 and 7.10]. Tor sheaves

Let f : X / Y be a holomorphic map and be a coherent sheaf on Y . We denote by Tori( ,f) the −1 F F sheaf Torf OY f −1 , . i F OX Grothendieck groups The Grothendieck group of coherent analytic sheaves (resp. of torsion coherent analytic sheaves) on a complex space X is denoted by G(X) (resp. G (X)). If is a coherent analytic sheaf on X,[ ] denotes tors F F its class in G(X). The notation GZ (X) is deﬁned in Appendix 7. In order to avoid subtle confusions, we never use here the Grothendieck group of locally free sheaves.

3. Deligne cohomology and Chern classes for locally free sheaves In this section, we will expose the basics of Deligne cohomology for the reader’s convenience. For a more detailed exposition, see [Es-Vi, 1, 6, 7, 8], [Vo1, Ch.12], and [EZZ]. § 3.1. Deligne cohomology. Deﬁnition 3.1. Let X be a smooth complex manifold and let p be a nonnegative integer. Then

– The Deligne complex ZD,X (p) of X is the following complex of sheaves

p (2iπ) d d 0 / Z / / / Ωp−1, X OX ··· X where ZX is in degree zero. Similarly, the rational Deligne complex QD,X (p) is the same complex as above with ZX replaced by QX . 8 JULIEN GRIVAUX

i – The Deligne cohomology groups HD(X, Z(p)) are the hypercohomology groups defined by i i HD(X, Z(p)) = H (X, ZD,X (p)). The rational Deligne cohomology groups are defined by the same formula as the hypercohomology groups of the rational Deligne complex. – The same definition holds for the Deligne cohomology with support in a closed subset Z: i i HD,Z (X, Z(p)) = HZ (X, ZD,X (p)). ∗ k – The total Deligne cohomology group of X is HD(X) = HD(X, Z(p)). We will denote by k,p ∗ HD(X, Q) the total rational Deligne cohomology group. L Example 3.2. i i – HD(X, Z(0)) is the usual Betti cohomology group H (X, Z). ∗ – ZD,X (1) is quasi-isomorphic to X [ 1] by the exponential exact sequence. Thus we have a group 2 O1 − ∗ isomorphism HD(X, Z(1)) H X, X Pic(X). The first Chern class of a line bundle L in 2 ≃ O ≃ HD(X, Z(1)) is the element of Pic(X) defined by c1(L)= L . 2 { } – HD(X, Z(2)) is the group of flat holomorphic line bundles, i.e. holomorphic line bundles with a holomorphic connection (see [Es-Vi, 1] and [Es]). § For geometric interpretations of higher Deligne cohomology groups, we refer the reader to [Ga]. Some fundamental properties of Deligne cohomology are listed below: Proposition 3.3. (i) We have an exact sequence 0 / Ω 6p−1[ 1] / Z (p) / Z / 0 . X• − D,X X 2p In particular, HD (X, Z(p)) fits into the exact sequence

2p−1 2p−1 6p−1 2p 2p H (X, Z) / H X, ΩX• / HD (X, Z(p)) / H (X, Z) .

(ii) The complex ZD,X (p)[1] is quasi-isomorphic to the cone of the morphism

(2iπ)p, i Z Ω >p / Ω . X ⊕ X• X• Thus we have a long exact sequence:

/ Hk−1(X, C ) / Hk (X, Z(p)) / Hk X, Ω >p Hk(X, Z) / Hk(X, C ) / ··· D X• ⊕ ··· and a similar exact sequence can be written with support in a closed subset Z. (iii) A cup-product

i j i+j H (X, Z(p)) Z H (X, Z(q)) / H (X, Z(p + q)) D ⊗ D D ∗ is deﬁned and endows HD(X) with a ring structure. (iv) If f : X / Y is a holomorphic map between two smooth complex manifolds, we have a pullback ∗ i i morphism f : HD(Y, Z(p)) / HD(X, Z(p)) which is a ring morphism. ∗ (v) If X is smooth, compact, and if E is a holomorphic vector bundle on X of rank r, then HD(P(E)) ∗ r−1 is a free H (X)-module with basis 1, c1( (1)),...,c1( (1)) . D OE O E (vi) For every t in P1, let j be the inclusion X X t / X P1 . Then the pullback mor- t ≃ ×{ } × phism j ∗ : H ∗ (X P1) / H ∗ (X) is independent of t (homotopy principle). t D × D CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 9

The assertions (i) and (ii) are obvious. The cup product in (iii) comes from a morphism of complexes

Z(p) Z Z(q) / Z(p + q) , see [Es-Vi, 1]. This morphism if functorial with respect to pullbacks, ⊗ § which gives (iv). Property (v) is proved by d´evissage using the exact sequence

0 / Ωp−1[ p] / Z (p) / Z (p 1) / 0 X − D,X D,X − and the ﬁve lemma (see [Es-Vi, 8]). Property (vi) is a consequence of (v): if α is a Deligne class in ∗ 1 § ∗ ∗ ∗ ∗ H (X P ), we can write α = pr λ + pr µ.c pr 1 (1) . Thus j α = λ. D × 1 1 1 2 OP t Remark that (vi) is false if we replace P1(C ) byC, in contrast with the algebraic case. Indeed, take an elliptic curve S and choose an isomorphism φ : S / Pic0(S) . There is a universal line bundle on L S S such that for all x in S, φ(x). Let π : C / S be the universal covering map of S. × L|S×x ≃ Consider the class α = c (id, π) ∗ , then for all t in C, j ∗α = c φ(π(t)) . 1 L t 1 We will now consider more reﬁned properties of Deligne cohomology. Proposition 3.4 (see [EZZ, 2]). § (i) If X is a smooth complex manifold and Z is a smooth submanifold of X of codimension d, there exists a cycle class Z in H2d (X, Z(d)) compatible with the Bloch cycle class (see [Bl, 5], { }D D,Z § [Es-Vi, 6]) and the topological cycle class. If Z and Z′ intersect transversally, Z Z′ = § { ∩ }D Z . Z′ . If Z is a smooth hypersurface of X, the image of Z in H2 (X, Z(1)) Pic(X) { }D { }D { }D D ≃ is the class of (Z). OX (ii) More generally, let f : X / Y be a proper holomorphic map between smooth complex manifolds and d = dim Y dim X. Then there exists a Gysin morphism − 2p 2(p+d) f∗ : HD (X, Z(q)) / HD (Y, Z(q + d)) compatible with the usual Gysin morphisms in integer and analytic de Rham cohomology. If Z is

a smooth submanifold of codimension d of X and iZ : Z / X is the canonical inclusion, then i (1) is the image of Z in H2d(X, Z(d)). Z∗ { }D D 2d−1 The point (i) is easy to understand. By Proposition 3.3 (ii), since HZ (X, Z) = 0, we have an exact sequence 0 / H2d (X, Z(d)) / H2d(X, Ω >d) H2d(X, Z) / H2d(X, C ). D,Z Z X• ⊕ Z Z The couple (2iπ)d Z , Z is mapped to 0 in H2d(X, C ) (see [Es-Vi, 7]). Therefore, it deﬁnes { }Bloch { }top Z § a unique element Z in H2d (X, Z(d)). { }D D,Z k For (ii), we introduce the sheaves Z of locally integral currents of degree k as done in [EZZ, 2], and DX, § [Gi-So, 2.2]. These sheaves are, in a way to be properly deﬁned, a completion of the currents induced § by smooth integral chains on X (see [Ki, 2.1] and [Fe, 4.1.24]). Then § § – X,• Z is a soft resolution of ZX . Dk k ∞ k – Z is a subsheaf of stable by push-forward under proper C maps, where is the sheaf DX, DX DX of usual currents of degree k on X. Thus Z (p)[1] is quasi-isomorphic to the cone of the morphism [2iπ]p,i : F p / . We D,X DX,• Z ⊕ DX• DX• k p k k will denote by Z (p) this cone shifted by minus one. Since the sheaves Z, F and are acyclic, D,X DX, DX DX p p Rf Z (p)[1] is quasi-isomorphic to the cone of the morphism [2iπ] ,i : f Z f F / f . ∗ D,X e ∗DX,• ⊕ ∗ DX• ∗DX• The push-forward of currents by f gives an explicit morphism f∗ : f∗ZD,X (p) / ZD,Y (p + d)[2d] and

e e 10 JULIEN GRIVAUX

then a morphism f : Rf Z (p) / Z (p + d)[2d] in the derived category b Mod(Z ) . We get ∗ ∗ D,X D,Y D Y the Gysin morphism by taking the hypercohomology on Y . The compatibility between Z and i (1) is shown in [EZZ]. { }D Z∗ We now state all the properties of the Gysin morphism needed here. The points (vi) and (vii) use Chern classes of vector bundles. They will be deﬁned in the next section. Proposition 3.5.

(i) f∗ is compatible with the composition of maps and satisﬁes the projection formula. ∗ f∗ x.f y = f∗x . y. In particular, if Γ X Y is the graph of f and if X is compact, then for every Deligne class f ⊆ × α in X, f α = p p ∗α . Γ . ∗ 2∗ 1 { f }D (ii) Consider the cartesian diagram

iY ×Z Y Z / X Z × × p q Y / X iY ∗ ∗ Then q iY ∗ = iY ×Z∗ p . (iii) If f : X / Y is proper and generically ﬁnite of degree d, then f f ∗ = d id. ∗ × (iv) Consider the cartesian diagram, where Y and Z are compact and intersect transversally:

iW →Y W _ / Y _

iW →Z iY Z / X iZ Then i ∗ i = i i ∗ . Y Z∗ W / Y ∗ W / Z (v) Let f : X / Y be a surjective map between smooth complex compact manifolds, and let D be a smooth hypersurface of Y such that f −1(D) is a simple normal crossing divisor. Let us write f ∗D = m D + + m D . Let f : D / D be the restriction of f to D . Then 1 1 ··· N N i i i N e e f ∗ i e= m i f ∗. e D∗ i D ∗ i ei i=1 X (vi) Let X be compact, smooth, and let Y be a smooth submanifold of codimension d of X. Let X be the blowup of X along Y , as shown in the following diagram:

j e E / X

q p e Y / X i CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 11

Then we have an isomorphism

d−1 ∗ ∗ ∗ HD(X) HD(Y ) / HD(X) ⊕ i=1 d−1 L ∗ i−1 x, (yi)1≤i≤d−1 / p x + e j∗ yi c1 N ( 1) . O Y /X − i=1 X h i In particular, if α is a Deligne class on X such that j ∗α is the pullback of a Deligne class on Y , then α is the pullback of a unique Deligne class on X.

Moreover, if F is the excess conormal bundle deﬁned by the exact sequence 0 / F / q ∗N ∗ / N ∗ / 0, Y/X E/X e ∗ ∗ ∗ we have the excess formula p i∗ α = j∗ q α cd−1(F ) . (vii) If Y is a smooth compact submanifold of X of codimension d, we have the auto-intersection formula ∗ iY iY ∗α = αcd NZ/X .

Proof. (i) and (ii) hold at the level of complexes. More precisely, the Gysin morphism f∗ is functorial at the complexes level ZD( . ). For the projection formula, we use the complex ZD,X ( . ) for the variable x and the complex ZD,Y ( . ) for the variable y. To prove the last formula of (i), we remark that p2 is proper since X is compact.e Then we write e ∗ ∗ f∗α = p2∗(id,f)∗α = p2∗(id,f)∗ (id,f) p1 α ∗ ∗ = p2∗ p1 α . (id,f)∗(1) = p2∗ p1 α . Γf D . { } For (ii), we can pull-back currents under p and q since these morphisms are submersions. Then p ∗ and ∗ q are defined for the complexes ZD( . ) and (ii) is an equality of complexes morphisms. For (iii), it is enough by the projection formula to prove that f (1) = d, which is well known. e ∗ The formulae (iv) and (v) are of the same type. Let us prove (v) for instance. We will define first some notations: Let Γ be the graph of i : D / Y and Γ be the graph of i : D / X. We define D i D i ei ′ Γ = f , id Γ D X. We call p1 : D Y / D and p2 : D Y / Y the first and second i i ∗ i ⊆ × × × e projections. In the same manner, we define the projections p′ : D X / D, p′ : D X / X, 1 × 2 × p′ : D X / D , and p′ : D X / X. 1,i i × i 2,i i × e e e 12 JULIEN GRIVAUX

N We have (id,f) ∗ Γ = m Γ′ (this can be seen using explicit description of the Bloch cycle class, { }D i { i}D i=1 see [Bl]). Then X f ∗ i α = f ∗ p p ∗α . Γ by (i) D∗ 2∗ 1 { }D = p′ (id,f) ∗ p ∗ α . Γ by (ii) 2∗ 1 { }D N = m p′ p′ ∗α . Γ′ by the projection formula i 2∗ 1 { i}D i=1 X N = m p′ p′ ∗α . (f , id) Γ i 2∗ 1 i ∗{ i}D i=1 X N = m p′ p′ ∗f ∗α . Γ by the projection formula i 2,i∗ 1,i i { i}D i=1 X N ∗ = mi iD ∗ f i α. ei i=1 X Before dealing with (vi), we prove (vii) when Y is a hypersurface. In the case of the étale cohomology, it is possible to assume that α = 1 (see [SGA 5, ExposéVII, 4] and [SGA 4 1 , Cycle 1.2]). Remark that § 2 § this is no longer possible here, for there is no purity theorem in Deligne cohomology. We use the deformation to the normal cone, an idea which goes back to Mumford (see [LMS] and [SGA 5, ExposéVII 9]). The aim was originally to prove the same formula in the Chow groups. Let M be § Y/X 1 ◦ the blowup of X P along Y 0 , X be the blowup of X along Y , and MY/X = MY/X X. Then × ×{ } \ we have an injection F : Y P1 / M 0 over P1 (see [Fu, Ch. 5] 5.1). We denote the inclusions × e Y/X § e ◦ NY/X / MY/X and Y / N by j and i, the projections of Y P1 M ◦ (resp. Y P1, resp. Y P1 N , Y/X 0 × × Y/X × × × Y/X ′ ′ resp. Y NY/X ) on its first and second factor by pr1 and pr2 (resp pr1 and pr2, resp. pr1 and pr2, resp. ′′ × ′′ 1 ◦ ′ pr1 and pr2 ). Besides, Γ Y P MY/X is the graph of F , and Γ Y NY/X is the graph of i. ⊆ × × ⊆ × Finally, Γ′′ = i , id Γ′ Y P1 N , where i : Y 0 e / Y e P1 is the injection of the 0 NY /X ∗ Y/X 0 ′⊆ × ′′ × ×{ } × central fiber. Remark that pr2 and pr2 are proper maps since Y is compact. ′ ′′ ∗ ′′ ∗ We have i0, idN Γ D = Γ D and idY ×P1 , j0 Γ D = Γ D. Let γ = F∗ pr1 α . Then Y /X ∗{ } { } { } { } ∗ ∗ ∗ ∗ ′ ∗ ∗ ∗ j γ = j pr pr pr α . Γ = pr id P1 , j pr pr α . Γ by (ii) 0 0 2∗ 1 1 { }D 2∗ Y × 0 1 1 { }D e ′ ∗ ∗ ∗ h′′ i = pr id P1 , j pr pr α . Γ 2∗ Y × e 0 1 1 { }D e ′ h ∗ i ∗ ∗ ∗ ′ = pr2∗ i0, idN i0, idN idY ×P1 , j0 pr1 pr1 α . Γ D by the projection formula Y /X ∗ e Y /X { } = pr′′ pr′′∗α . Γ′ = i α. 2∗ 1 { }D ∗ e By the homotopy principle (Proposition 3.3 (vi)), the class F ∗γ is independent of t. If t = 0, we |Y ×{t} 6 ∗ ∗ ∗ ∗ ∗ ∗ have clearly F γ|Y ×{t} = iY iY ∗α. For t = 0, F γ|Y ×{0} = i j0 γ = i i∗α. Let π: NY/X / Y be the ∗ ∗ projection of NY/X on Y . Then α = i π α. Thus i ∗i α = i ∗i (i ∗π ∗α)= i ∗ π ∗α . Y = α.i ∗ Y , ∗ ∗ { }D { }D CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 13 where Y is the cycle class of Y in N . { }D Y/X ∗ Now Y D = c1 N (Y ) , so that i Y D = c1 NY/N = c1 NY/X . { } O Y /X { } Y /X We can now prove (vi). Its first part is straightforward using dévissage as in Proposition 3.3 (v) and the analogous result in Dolbeault cohomology and in integer cohomology. d−1 ∗ i−1 If α is a Deligne class on X, we can write α = p x + j∗ yi c1 N ( 1) . Since E is a O Y /X − i=1 X h i hypersurface of X, by the formula proved above j ∗j λ = λc N = λ c ( 1) for any e ∗ 1 E/X 1 N e O Y /X − d−1 ∗ ∗ ∗ i i ∗ ∗ Deligne class λ one E. We obtain j α = q i x + ( 1) yi c1 N (1) . Since j α = q δ, all the − O Y /X i=1 X∗ classes yi vanish by Proposition 3.3 (v). Thus α = p x. By Proposition 3.5 (iii), x = p∗α. ∗ ∗ For the excess formula, let α be a Deligne class on Y . We define β = j∗ q α cd−1(F ) . Then, by Proposition 3.7 (i) and (ii), j ∗β = q ∗α c (F ∗) c N = q ∗ αc N . d−1 1 E/X d Y/X e By the discussion above, β comes from the base, so that ∗ ∗ ∗ ∗ ∗ ∗ ∗ β = p p∗β = p i∗q∗ q αcd−1(F ) = p i∗ α q∗ cd−1(F ) = p i∗α ∗ for q∗ cd−1(F ) = 1 (see [Bo-Se, Lemme 19.b]). — Proof of (vii). The formula is already true for d = 1. We blowup X along Y , use the excess formula (vi) and we obtain: ∗ ∗ ∗ ∗ ∗ ∗ ∗ q iY iY ∗ α = j p iY ∗ α = j j∗ q αcd−1(F ) ∗ ∗ ∗ = q α cd−1 F c1 NE/X = q αcd NY/X . e Since q ∗ is injective, we get the result. 3.2. Chern classes for holomorphic vector bundles. We refer to [Zu, 4] and [Es-Vi, 8] for all § § this section. From now on, we will suppose that X is smooth of dimension n. Let E be a holomorphic bundle on X of rank r, and P(E) be the projective bundle of E endowed with the line bundle (1). OE Let α = c (1) . By property (v) of Proposition 3.3, we can define c (E) by the relation 1 OE i 1≤i≤r αr + p ∗c (E) αr−1 + + p ∗c (E) = 0 in H2r P(E), Z(r) . 1 ··· r D 2i Thus ci(E) is an element of HD (X, Z(i)). The knowledge of the Chern classes ci(E) allows to construct exponential Chern classes ch (E), 0 i n, in H2i(X, Q(i)). These classes are obtained as the values i ≤ ≤ D of certain universal polynomials with rational coefficients on c1(E),...,cr(E) (see [Hirz]). They can also be constructed with the splitting principle using projective towers (see [Gro1]). They are completely characterized by the following facts: – they satisfy the Whitney additivity formula (Proposition 3.7 (i)); – they satisfy the functoriality formula under pullbacks (Proposition 3.7 (ii)); – if L is a line bundle, ch(L)= ec1(L).

∗ Definition 3.6. The total Chern class of E is the element c(E) of HD(X) defined by c(E)=1+ c (E)+ + c (E). 1 ··· r ∗ The Chern character of E is the element ch(E) of HD(X, Q) defined by ch(E) = ch (E)+ + ch (E). 0 ··· n The splitting machinery gives the following proposition: 14 JULIEN GRIVAUX

Proposition 3.7. (i) If 0 / E / F / G / 0 is an exact sequence of vector bundles on X, the Whitney formula holds: c(F )= c(E)c(G) and ch(F ) = ch(E) + ch(G). (ii) If f is a holomorphic map between X and Y and if E is a holomorphic bundle on Y , we have c f ∗E = f ∗c(E) and ch f ∗E = f ∗ ch(E). (iii) If E and F are two holomorphic vector bundles on X, then ch(E F ) = ch(E) ch(F ). ⊗ Notation 3.8. From now on, if is a locally free sheaf and E is the associated holomorphic vector E bundle, we will denote by ch( ) the Chern character ch(E). Thus ch is well deﬁned on a basis of any E dimension but only for locally free sheaves, and we will make use of it in our construction.

4. Construction of Chern classes The construction of exponential Chern classes ch ( ) in H2p(X, Q(p)) for an arbitrary coherent sheaf p F D F on X will be done by induction on dim X. If dim X = 0, X is a point and everything is obvious. Let us now precisely state the induction hypotheses (Hn):

(En) If dim X n and is a coherent analytic sheaf on X, then the Chern classes chp( ) are defined 2p ≤ F F in HD (X, Q(p)). (W ) If dim X n and 0 / / / / 0 is an exact sequence of analytic sheaves n ≤ F G H on X, then ch( ) = ch( ) + ch( ). This means that ch is defined on G(X) and is a group G F H morphism. (F ) If dim X n, dim Y n, and if f : X / Y is a holomorphic map, then for all y in G(Y ), n ≤ ≤ ch(f !y)= f ∗ ch(y). (C ) If dim X n, the Chern classes are compatible with those constructed in Part 3 on locally free n ≤ sheaves, i.e. ch( )= ch( ) for all locally free sheaf . F F F (P ) If dim X n, ch is a ring morphism: if x, y are two elements of G(X), ch(x.y) = ch(x) ch(y) n ≤ and ch(1) = 1. (RR ) If Z is a smooth hypersurface of X, where dim X n, then the (GRR) theorem holds for i : for n ≤ Z every coherent sheaf on Z, ch(i )= i chZ ( ) td(N )−1 . F Z∗F Z∗ F Z/X For the definition of analytic K-theory and related operations we refer to [Bo-Se]. Remark 4.1. To avoid any confusion, for a coherent sheaf on Z ( X, we use the notation chZ ( ), F emphasizing the fact that the class is taken on Z.

From now on, we will suppose that all the properties of the induction hypotheses (Hn−1) above are true.

Theorem 4.2. Assuming hypotheses (Hn−1), we can define a Chern character for analytic coherent sheaves on compact complex manifolds of dimension n. It further satisfies (Pn), (Fn), (RRn), (Wn) and (Cn). Let us briefly explain the organization of the proof of this theorem. In 4.1, we construct the Chern char- § acter for torsion sheaves. In 4.3, we construct the Chern character for arbitrary coherent sheaves, using § the results of 4.2. Properties (RR ) for a smooth hypersurface and (C ) will be obvious consequences § n n of the construction. In 5.3, we prove (W ) and then (F ) and (P ) using the preliminary results of 5.1 § n n n § and 5.2. Finally, we prove (RR ) in 6. § n § 4.1. Construction for torsion sheaves. In this section, we define Chern classes for torsion sheaves by forcing the Grothendieck-Riemann-Roch formula for immersions of smooth hypersurfaces. Let Gtors(X) denote the Grothendieck group of the abelian category of torsion sheaves on X. We will prove the following version of Theorem 4.2 for torsion sheaves: Proposition 4.3. We can define exponential Chern classes for torsion sheaves on any n-dimensional complex manifold such that: CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 15

(i) (W ) If 0 / / / / 0 is an exact sequence of torsion sheaves on X with dim X n F G H ≤ n, then ch( ) = ch( ) + ch( ). This means that ch is a group morphism defined on G (X). G F H tors (ii) (P ) Let be a locally free sheaf and x be an element of G (X). Then ch([ ]. x)= ch( ) . ch(x). n E tors E E (iii) (F ) Let f : X / Y be a holomorphic map where dim X n and dim Y n, and be a coherent n ≤ ≤ F sheaf on Y such that and f ∗ are torsion sheaves. Then ch(f ![ ]) = f ∗ ch( ). F F F F (iv) (RR ) If Z is a smooth hypersurface of X and is coherent on Z, then n F ch i = i chZ ( ) td(N )−1 . Z∗F Z∗ F Z/X We will proceed in three steps. In 4.1.1, we perform the construction for coherent sheaves supported in § a smooth hypersurface. In 4.1.2, we deal with sheaves supported in a simple normal crossing divisor. § In 4.1.3, we study the general case. § 4.1.1. Let Z be a smooth hypersurface of X where dim X n. For coherent on Z, we define ch i ≤ G Z∗G by the GRR formula ch i = i chZ ( ) td(N )−1 , where chZ ( ) is defined by (E ). Z∗G Z∗ G Z/X G n−1 If 0 / ′ / / ′′ / 0 is an exact sequence of coherent sheaves on Z, by (W ), we G G G n−1 have chZ ( ) = chZ ( ′)+chZ ( ′′). Thus ch i = ch i ′ +ch i ′′ . We obtain now a well-defined G G G Z∗G Z∗G Z∗G morphism ∼ G(Z) / GZ (X) iZ∗

chZ ∗ HD(X, Q) Remark that if is a coherent sheaf on X which can be written i , then the hypersurface Z is not G Z∗F necessarily unique. If Z is chosen, is of course unique. This is the reason why we use the notation F ch ( ). We will see in Proposition 4.8 that ch ( ) is in fact independent of Z. Z G Z G The assertions of the following proposition are particular cases of (Cn), (Fn), and (Pn). Proposition 4.4. Let Z be a smooth hypersurface of X. Z ! ! (i) For all x in GZ (X), ch iZ x = iZ chZ (x). (ii) If is a locally free sheaf on X and x is an element of G (X), then E Z ch ([ ]. x)= ch( ) . ch (x). Z E E Z Proof. (i) We have x = x [i ] in G (X), where x is deﬁned in Appendix 7.2. Thus, Z Z∗OZ Z ∗ ∗ Z −1 iZ chZ (x)= iZ iZ∗ ch (x) td NZ/X Z −1 = ch (x) td NZ/X c1 NZ/X by Proposition 3.5 (vii)

−c N = chZ (x) 1 e 1 Z/X − = chZ (x) chZ i ! i by (C ) Z Z∗OZ n−1 Z ! = ch x.i i by (Pn−1) Z Z∗OZ Z ! = ch iZ x . 16 JULIEN GRIVAUX

(ii) We have ch [ ] . x = ch i i ! [ ] . x Z E Z Z∗ Z E −1 = i chZ i ! [ ] . xtd N Z∗ Z E Z/X −1 = i i ∗ ch chZ (x) td N by Proposition 3.7 (ii), (P ) and (C ) Z∗ Z E Z/X n−1 n−1 −1 = ch i chZ (x) td N by the projection formula E Z∗ Z/X = ch ch (x). E Z 4.1.2. Let D be a divisor in X with simple normal crossing. By Proposition 7.8, we have an exact sequence:

GDij (X) / GDi (X) / GD(X) / 0. i

LDi −1 Dij −1 chD iD ∗ = iD ∗ ch i td ND /X = iD ∗ ch ( ) td ND /X i ij F i Dij / Di∗F i ij F ij because of (RRn−1) and the multiplicativity of the Todd class. Thus ch i = ch i , and we get a map ch : G (X) / H ∗ (X, Q) such that the Di Dij ∗F Dj Dij ∗F D D D diagram

GDi (X) / GD(X) / 0 i L MM MM chD L ch MM Di MM i M& ∗ HD(X, Q) is commutative.

Proposition 4.5. The classes chD have the following properties: (i) If is a locally free sheaf on X and x is an element of G (X), then E D ch ([ ] . x)= ch( ). ch (x). D E E D (ii) Let D be an eﬀective simple normal crossing divisor in X such that Dred = D. Then

chD D =1 ch X ( D) . e O e − O − e (iii) (First lemma of functoriality) Let f : X / Y be a surjective map. Let D be a reduced divisor e in Y with simple normal crossing such that f −1(D) is also a divisor with simple normal crossing ! ∗ in X. Then for all y in GD(Y ), chf −1(D) f y = f chD(y). (iv) (Second lemma of functoriality) Let Y be a smooth submanifold of X and D be a reduced divisor Y ! ∗ in X with simple normal crossing. Then, for every x in GD(X), ch iY x = iY chD(x). Proof. We start with two technical lemmas which will be crucial for the proof of (ii) and (iii).

Lemma 4.6. Let D = m1D1 + + mN DN be an eﬀective simple normal crossing divisor in X, and µ ∗ ··· be the element of HD(X, Q) deﬁned by ( 1)k−1 k−1 µ = − m D + + m D . k! 1{ 1}D ··· N { N }D kX≥1 ∗ Then there exist ui in G (X), 1 i N, and ζij in H (Dij ), 1 i, j N, i = j, such that Di ≤ ≤ D ≤ ≤ 6 CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 17

(a) u + + u = in G red (X). 1 ··· N OD D (b) ζij = ζji, 1 i, j N, i = j. − ≤ ≤ 6 N −1 ∗ (c) ch(ui) td ND /X mi iDi µ = i ζij , 1 i N. i − Dij / Di∗ ≤ ≤ j=1 Xj6=i Proof. We proceed by induction on the number N of branches of Dred. −1 ∗ If N = 1, we must prove that ch(u1) td N = m1 i µ, where u1 = . In G (X) we have D1/X D1 Om1D1 D1 m1−1 m1−1 ∗⊗q ∗⊗q = i N , thus u1 = N . Therefore Om1D1 D1∗ D1/X D1/X q=0 q=0 X X m1−1 c1 N −1 1 e D1/X qc1 ND1/X − ch(u1) td ND1/X = e− − c1 N q=0 ! D1/X X − m1c1 ND1/X 1 e ∗ = − = m1 i µ. c N D1 1 D1/X

′ ′ Suppose that the lemma holds for divisors D such that Dred has N 1 branches. Let D = m1D1 + + ′ − ′ ··· m D and D = m1D1 + + m D . By induction, there exist u in G (X), 1 i N 1, N N ··· N−1 N−1 i Di ≤ ≤ − and ζ′ in H ∗ (D ), 1 i, j N 1, i = j, satisfying properties (a), (b), and (c) of Lemma 4.6. For ij D ij ≤ ≤ − 6 0 k m , we introduce the divisors Z = m D + + m D + kD . We have exact sequences ≤ ≤ N k 1 1 ··· N−1 N−1 N ∗ 0 / iD X ( Zk) / Z / Z / 0. N O − O k+1 O k

Thus, in GDred (X), we have

mN−1 mN−1 ∗ ∗ ′ D = D′ + iD ∗ iD X ( Zq) = D′ + iD ∗iD X ( D ) X qDN . O O N  N O −  O N N O − O −  q=0 q=0 X X     mN−1 ∗ ′ uN = iD ∗iD X ( D ) X qDN We choose  N N O − O − q=0  h X i u = u′ for 1 i N 1. i i ≤ ≤ −  Let i be such that 1 i N 1. Then ≤ ≤ − ∗ ′ ∗ ′ ∗ ′ ch(ui) td N mi i µ = ch(u ) td N mi i µ + mi i (µ µ) Di/X − Di i Di/X − Di Di − N−1 ∞ k k−1 ′ ∗ ( 1) k 1 = i ζij + mi iD − − m1 D1 D + Dij / Di∗ i k! j { } ··· j=1 " k=1 j=1 Xj6=i X X

k−1−j j + mN−1 DN−1 D mN DN D by induction ··· { } { } # N−1 ∞ k−1 ( 1)k k 1 = i ζ′ + m i i ∗ − − m D + D D ∗ ij i DiN 1 1 D ij / i DiN / Di∗ k! j { } ··· j=1 " k=1 j=1 Xj6=i X X

k−1−j j j−1 + mN−1 DN−1 D mN DN D . ··· { } { } # 18 JULIEN GRIVAUX

For the last equality, we have used that ∗ ∗ ∗ iDi α DN D = iDi α DiN D = i iD α { } { } DiN / Di∗ iN where D is the cycle class of D in D . { iN }D iN i Let us deﬁne ζ = ζ′ if 1 i, j N 1, i = j ij ij ≤ ≤ − 6 ∞ k k−1  ∗ ( 1) k 1 ζiN = mi i − − m1 D1 +  DiN D  " k! j { } ···  k=1 j=1  X X  k−1−j j j−1  + mN−1 DN−1 D mN DN D if 1 i N 1 ··· { } { } # ≤ ≤ −  ζ = ζ if 1 j N 1.  Nj − jN ≤ ≤ −  Properties (a) and (b) of Lemma 4.6 hold, and property (c) of the same lemma hold for 1 i N 1.  ≤ ≤ − For i = N, let us now compute both members of (c). We have

N−1 N−1 ∞ k−1 k−1 ∗ ( 1) k 1 i ζNl = ml iD − − m1 D1 D + DNl / DN ∗ N k! j { } ··· l=1 l=1 "k=1 j=1 X X X X k−1−j j j−1 + mN−1 DN−1 mN DN D Dl ··· { } { } { }# ∞ k−1 k−1 ∗ ( 1) k 1 k−j j j−1 ( ) = iD − − m1 D1 D + + mN−1 DN−1 D m DN . ∗ N k! j { } ··· { } N { }D "k=1 j=1 # X X In the ﬁrst equality, we have used ∗ ∗ ∗ i iD α = iD α DlN D = iD α Dl D DNl / DN ∗ lN N { } N { } where D is the cycle class of D in D . { lN }D lN N Now, −1 ∗ ch(uN ) td ND /X mN iD µ N − N mN−1 1 e− DN ∗ m1 D1 mN−1 DN−1 q DN { } ∗ = iD e− { }−···− { } e − mN iD µ N   − { } D  − N q=0 N X { }     by (Cn−1) and Proposition 3.7 (ii) and (iii)

1 e−mN DN ∗ m1 D1 mN−1 DN−1 { } = iD e− { }−···− { } − mN µ N D − " { N } # ∞ ∞ r+q−1 ∗ ( 1) r q−1 = iD mN − m1 D1 + + mN−1 DN−1 mN DN N r! q! { } ··· { } { } " r=0 q=1 X X ∞ k−1 k−1 ( 1) k 1 k−1−j j m − − m D + + m D m D . − N k! j 1{ 1} ··· N−1{ N−1} N { N }  k=1 j=0 X X  CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 19

In the ﬁrst term, we put k = q + r, p = q 1 and we obtain − ∞ k−1 k−1 ∗ ( 1) k k 1 mN iD − − m1 D1 + N k! p +1 − p { } ··· "k=1 p=0 ! X X k−1−p p ( ) +mN−1 DN−1 mN DN . ∗∗ ··· { } { } # k k 1 k 1 Now − is equal to − for p k 2 and to zero for p = k 1. It suﬃces to put p +1 − p p +1 ≤ − − j = p+ 1 in the sum to obtain the equality of ( ) and ( ). ∗ ∗∗ ∗ Lemma 4.7. Using the same notations as in Lemma 4.6, let αi in HD(Di), 1 i N, be such that ∗ ∗ ≤ ≤ i α = i α . Then there exist u in G (X), satisfying u + +u = in G red (X), i j i Di 1 N D D Dij / Di Dij / Dj ··· O such that N N −1 iDi∗ αi ch(ui) td NDi/X = mi iDi∗(αi) µ. i=1 i=1 ! X X Proof. We pick u1,...,uN given by Lemma 4.6. Then N N −1 i αi ch(ui) td N mi i (αi) µ Di∗ Di/X − Di∗ i=1 i=1 ! X X N −1 ∗ = i αi ch(ui) td N mi i µ by the projection formula Di∗ Di/X − Di i=1 X h i N N

= iDi∗ αi i ζij Dij / Di∗ i=1 j=1 X Xj6=i h i N N ∗ = iDij ∗ i αi ζij by the projection formula. Dij / Di i=1 j=1 X Xj6=i Grouping the terms (i, j) and (j, i), we get 0, since ζ = ζ . ij − ji We now prove Proposition 4.5.

Proof. (i) We write x = x1 + + x in G red (X), where xi is an element of G (X). Then ··· N D Di N N ch ([ ] . x)= ch ([ ] . x )= ch( ) . ch (x ) by Proposition 4.4 (ii) D E Di E i E Di i i=1 i=1 X X = ch( ) . ch (x) by the very deﬁnition of ch (x). E D D

(ii) We choose u1,...,uN such that Lemma 4.6 holds. Then N N −1 ch = ch(u )= i ch(u ) td N D i Di∗ i D /X O e ei i=1 i=1 X X N = m D µ =1 e− m1 D1 D + + mN DN D =1 ch ( D) . i{ i}D − { } ··· { } − OX − i=1 ! X e e e e 20 JULIEN GRIVAUX

(iii) By d´evissage we can suppose that D is a smooth hypersurface of Y . Let f i be deﬁned by the diagram

Di / X

f i f f D / Y ∗ D and let y be an element of G(D). We put αi = f i ch (y). By the functoriality property (Fn−1) we have ∗ ∗ i αi = i αj . We choose again u1,...,uN such that Lemma 4.7 holds. By Proposition D D D D eij / ei eij / ei N ! 7.7 of Appendix 7 we can write f !i y = f y u . Thus D∗ i D i ei i=1 X N D ! D −1 ! ei ei chD f iD∗y = iD ∗ ch f i y ch (ui) td ND /X by (Pn−1) e ei ei i=1 X N D −1 = i α ch ei (u ) td N by (F ) D ∗ i i D /X n−1 ei ei i=1 X N

= mi iD ∗(αi) µ by Lemma 4.7 ei i=1 ! X N − ∗ 1 e D D = m i f chD(y) f ∗ { } i D ∗ i − ei D "i=1 # { }D ! X 1 e− D D = f ∗ i chD(y) − { } by Proposition 3.5 (v) D∗ · D " { }D # ∗ D −1 = f iD∗ ch (y) td ND/Y by the projection formula ∗ = f chD iD∗y .

(iv) We will ﬁrst prove it under the assumption that, for all i, either Y and Di intersect transversally, or Y = Di. By d´evissage, we can suppose that D has only one branch and that Y and D intersect transversally, or Y = D. We deal with both cases separately. – If Y and D intersect transversally, then i ! i = i . Thus, by Proposition 7.5 Y D∗OD Y ∩D / Y ∗ OY ∩D of Appendix 7, i ! x = i ! x i = i ! x i = i i ! x Y Y D D∗OD Y ∩D / D Y ∩D Y ∩D / Y ∗OY ∩D Y ∩D / Y ∗ Y ∩D / D and we obtain Y ! Y ∩D ! −1 ch i x = i ch i x td N by (RRn−1) Y Y ∩D / Y ∗ Y ∩D / D Y ∩D/Y ∗ D ∗ −1 = i i ch (x) i td N by (Fn−1) Y ∩D / Y ∗ Y ∩D / D Y ∩D / D D/X ∗ D −1 = iY iD∗ ch (x) td ND/X by Proposition 3.5 (iv) ∗ = iY chD(x). CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 21

– If Y = D, i ! i = N ∗ . Thus i ! x = x x . N ∗ and Y D∗OD OY − Y/X Y − Y/X chY i ! x = chY (x) chY (x)ch N ∗ by (P ) and (C ) Y − Y/X n−1 n−1 −c N = chY (x) 1 e 1 Y/X − Y −1 = ch (x) td NY/X c1 NY/X ∗ Y −1 = iY iY ∗ ch (x) tdNY/X by Proposition 3.5 (vii) ∗ = iY chY (x). We examine now the general case. By Hironaka’s desingularization theorem, we can desingularize Y D ∪ by a succession τ of k blowups with smooth centers such that τ −1(Y D) is a divisor with simple ∪ normal crossing. By ﬁrst blowing up X along Y , we can suppose that τ −1(Y ) = D˘ is a subdivisor of D = τ −1(Y D). We have the following diagram: ∪ i ˘ Dj e D˘ j / X

q j eτ Y / X iY Then ˘ ∗ Y ! Dj ! ! qj ch iY x = ch qj iY x by (Fn−1) ˘ = chDj i ! τ !x D˘ j ∗ ! ˘ ˘ = i ˘ chD τ x since Dj and Di intersect transversally, or Dj = Di Dj e ∗ ∗ = i ˘ τ chD(x) by the ﬁrst lemma of functoriality 4.5 (iii) Dj e e ∗ ∗ = qj iY chD(x). We can now write q as δ µ , where E is the exceptional divisor of the blowup of X along Y , δ: E / Y j ◦ j is the canonical projection and µ : D˘ / E is the restriction of the last k 1 blowups to D˘ . Write j j − j τ = τ τ τ where τ are the blowups. Let us deﬁne a sequence of divisors E by k ◦ k−1 ◦···◦ 1 i i 0≤i≤k induction: E = E, and E is the strict transform of E under τ . Since the E are smooth divisors, 0 i+1 i i+1 i ˘ all the maps τi+1 : Ei+1 / Ei are isomorphisms. There exists j such that Ek = Dj. We deduce that µj = τ : D˘ j / D is an isomorphism. Since δ is the projection of the projective bundle |D˘ j ∗ ∗ ∗ ∗ Y ! ∗ P NY/X / Y , δ is injective. Thus qj = µj δ is injective and we get ch iY x = iY chD(x).

Now, we can clear up the problem of the dependence with respect to D of ch ( ). D F Proposition 4.8. If D and D are two divisors of X with simple normal crossing such that supp D 1 2 F ⊆ 1 and supp D , then ch ( ) = ch ( ). F ⊆ 2 D1 F D2 F Proof. This property is clear if D D . We will reduce the general situation to this case. By Hironaka’s 1 ⊆ 2 −1 theorem, there exists τ : X / X such that τ (D1 D2) is a divisor with simple normal crossing. −1 −1 ∪ Let D1 = τ D1 and D2 = τ D2. By the ﬁrst functoriality lemma 4.5 (iii), since D1 D, we have ∗ ! e ! ⊆ τ chD1 ( ) = chD τ [ ] = chD τ [ ] . The same property holds for D2. The map τ is a succession F e1 F e F of blowups,e thus τ ∗ is injectivee and we get ch ( ) = ch ( ). e e D1 F D2 F 22 JULIEN GRIVAUX

Definition 4.9. If supp( ) D where D is a normal simple crossing divisor, ch( ) is defined as ch ( ). F ⊆ F D F By Proposition 4.8, this definition makes sense. 4.1.3. We can now define ch( ) for an arbitrary torsion sheaf. F Let be a torsion sheaf. We say that a succession of blowups with smooth centers τ : X / X F is a desingularization of if there exists a divisor with simple normal crossing D in X such that F τ −1 supp( ) D. By Hironaka’s theorem applied to supp( ), there always exists suche a τ. We F ⊆ F say that can be desingularized in d steps if there exists a desingularization τ of consistinge of at most F F d blowups. In that case, ch(τ ![ ]) is defined by Definition 4.9. F Proposition 4.10. There exists a class ch( ) uniquely determined by such that F F (i) If τ is a desingularization of , then τ ∗ ch( ) = ch(τ ![ ]). F F F (ii) If Y is a smooth submanifold of X, then chY i ! [ ] = i ∗ ch( ). Y F Y F Proof. Let d be the number of blowups necessary to desingularize . Both assertions will be proved at F the same time by induction on d. If d = 0, supp( ) is a subset of a divisor with simple normal crossing D. The properties (i) and (ii) are F immediate consequences of the two lemmas of functoriality 4.5 (iii) and (iv). Suppose now that Proposition 4.10 is proved for torsion sheaves which can be desingularized in d 1 − steps. Let be a torsion sheaf which can be desingularized with at most d blowups. Let (X, τ) be such F a desingularization. We write τ as τ τ , where τ is the first blowup in τ with E as exceptional divisor, ◦ 1 as shown in the following diagram: e e e X

eτ1 iE E / X1

q τ e e Y / X iY

Then τ1 consists of at most d 1 blowups and is a desingularization of the sheaves Torj ( , τ), 0 j n. − ∗ F ≤ ≤ By induction, we can consider the following expression in HD(X1, Q): n e γ X , = ( 1)j ch Tor (e , τ) . 1 F − j F j=0 X We have e e n i ∗ γ X , = ( 1)j chE i ! Tor ( , τ) by induction, property (ii) E 1 F − E j F j=0 X e = chE i ! τ ![ ] e by (W ) E F n−1 E ! ! ∗ Y ! = ch q iY [ ] = q ch iY [ ] by (Fn−1). e F F CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 23

By Proposition 3.5 (vi), there exists a unique class ch( , τ) on X such that γ X , = τ ∗ ch , τ . F 1 F F Now e τ ∗ ch( , τ)= τ ∗γ X , e F 1 1 F n = ( e1)j ch τ ! Tor ( , τ) by induction, property (i) − 1 j F j=0 X = ch τ !τ ![ ] = ch τ ![ ] .e 1 F F Suppose that we have two resolutions. We dominate them by a third one, according to the diagram: e W A µ }} AA µ˘ }} AA }} AA }~ } A X1 δ X2 A AA }} AA }} e τ AA }} τ˘ e A }~ } X Then δ ∗ ch( , τ)= µ ∗ ch τ ![ ] F F ! ! = ch µ τ [ ] by the first lemma of functoriality 4.5 (iii) F = chδ ![ ] = δ ∗ ch( , τ˘) by symmetry. F F The map δ ∗ being injective, ch( , τ) = ch( , τ˘), and we can therefore define ch( ) by ch( ) = ch( , τ) F F F F F for any desingularization τ of with at most d blowups. F We have shown that (i) is true when τ consists of at most d blowups. In the general case, let τ : X1 / X be an arbitrary desingularization of andτ ˘ be a desingularization of with at most d blowups. We F F can find W , µ andµ ˘ as before. Then e µ ∗ ch τ ![ ] = ch δ ![ ] by the first functoriality lemma 4.5 (iii) F F ∗ ! =µ ˘ ch τ˘ [ ] by the first functoriality lemma 4.5 (iii) F =µ ˘ ∗τ˘ ∗ ch( ) sinceτ ˘ consists of at most d blowups F = µ ∗τ ∗ ch( ). F It remains to show (ii). For this, we desingularize supp( ) Y exactly as in the proof of the second F ∪ lemma of functoriality 4.5 (iv). We have a diagram

D˘ i / X i ˘ Di qi τ e Y / X iY ∗ where qi is injective for at least one i. Then ∗ ∗ ∗ ∗ ∗ ! qi iY ch( ) = i ˘ τ ch( )= i ˘ ch(τ [ ]) by (i) F Di F Di F D˘ i ! ! = ch i ˘ τ [ ] by the second lemma of functoriality 4.5 (iv) Di F ˘ = chDi q ! i ! [ ] = q ∗ chY i ! [ ] by (F ). i Y F i Y F n−1 Thus i ∗ ch( ) = chY i ! [ ] , which proves (ii). Y F Y F 24 JULIEN GRIVAUX

We have now completed the existence part of Theorem 4.2 for torsion sheaves. We turn to the proof of Proposition 4.3. So doing, we establish almost all the properties listed in the induction hypotheses for torsion sheaves. Proof. (i) Let (X, τ) be a desingularization of supp( ) supp( ) and D be the associated simple normal F ∪ H crossing divisor. Then τ ! , τ ! and τ ! belong to G (X) and τ ! + τ ! = τ ! in G (X). Thus, by F G H D F H G D Proposition 4.10e (i), e τ ∗ ch( ) + ch( ) = ch τ ![ ] + ch τ ![ ] = ch τ ![ ] = τ ∗ ch( ). F H F H G G The map τ ∗ being injective, we get the Whitney formula for torsion sheaves. (ii) The method is the same: let x = [ ] and let τ be a desingularization of . Then, by Proposition 4.10 G G (i) and Proposition 4.5 (i), τ ∗ ch([ ] . [ ]) = ch τ ![ ] . τ ![ ] = ch τ ![ ] . ch τ ![ ] = τ ∗ ch( ) . ch( ) . E G E G E G E G (iii) This property is known when f if the immersion of a smooth submanifold and whenf is a bimeromorphic morphism by Proposition 4.10. Let us consider now the general case. By Grauert’s direct image theorem, f(X) is an irreducible analytic subset of Y . We desingularize f(X) as an abstract complex space. We get a connected smooth manifold W and a bimeromorphic morphism τ : W / f(X) obtained as a succession of blowups with smooth centers in f(X). We perform a similar sequence of blowups, starting from Y1 = Y and blowing up at each step in Yi the smooth center blown up at the i-th step of the desingularization of f(X). Let πY : Y / Y be this morphism. The strict transform of f(X) is W . −1 ∼ The map τ : τ f(X)reg / f(X)reg is an isomorphism. So we get a morphism f(X)reg / W e which is in fact a meromorphic map from f(X) to W , and ﬁnally, after composition on the left by f, from X to W . We desingularize this morphism:

X @ @@ f π @@e X e @@ @ X / W and we get the following global diagram, where πX is a bimeromorphic map:

f i X e / W W / Y

π τ π X e e Y X / f(X) / Y f

Now f πX = πY iW f , and we know the functoriality formula for πX , πY and iW by Proposition ◦ ∗ ◦ ◦ 4.10. Since πX is injective, it is enough to show the functoriality formula for f. So we will assume that f is onto. e e Let (τ, Y ) be a desingularization of . We have the diagram F X Y / Y e ×Y e eτ f X / Y CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 25

where τ −1(supp ) = D Y is a divisor with simple normal crossing and the map X Y / X F ⊆ ×Y is a bimeromorphic morphism. We have a meromorphic map X / X Y, and we desingularize e e ×Y e it by a morphism T / X Y. Therefore, we obtain the following commutative diagram, where ×Y e π: T / X is a bimeromorphic map: e f T e / Y

π τ e f X / Y Therefore we can assume that supp( ) is included in a divisor with simple normal crossing D. We F desingularize f −1(D) so that we are led to the case supp( ) D, where D and f −1(D) are divisors with F ⊆ simple normal crossing in Y and X respectively. In this case, we can use the first lemma of functoriality 4.5 (iii). 4.2. The case of sheaves of positive rank. In this section, we consider the case of sheaves of arbitrary rank. We are going to introduce the main tool of the construction, namely a dévissage theorem for coherent analytic sheaves. Let X be a complex compact manifold and an analytic coherent sheaf on F X. We have seen in section 4.1 how to define ch( ) when is a torsion sheaf. F F Suppose that has strictly positive generic rank. When admits a global locally free resolution, we F F could try to define ch( ) the usual way. As explained in the introduction, this condition on is not F F necessarily fulfilled. Even if such a resolution exists, the definition of ch( ) depends a priori on this F resolution. A good substitute for a locally free resolution is a locally free quotient with maximal rank, since the kernel is then a torsion sheaf. Let be the maximal torsion subsheaf of . Then Ftor ⊆ F F admits a locally free quotient of maximal rank if and only if is locally free. In this case, F E F Ftor = . E F Ftor Unfortunately, the existence of such a quotient is not assured (for instance, take a torsion-free sheaf which is not locally free), but we will show that it exists up to a bimeromorphic morphism. Theorem 4.11. Let X be a complex compact manifold and a coherent analytic sheaf on X. There F exists a bimeromorphic morphism σ : X / X, which is a finite composition of blowups with smooth centers, such that σ ∗ admits a locally free quotient of maximal rank on X. Such quotients are unique, F up to a unique isomorphism. e e Proof. Let r be the rank of . We define a universal set X by X = Gr ∗ r, where Gr ∗ r, F x∈X F|x F|x is the dual grassmannian of quotients of |x with rank r. The set X is the disjoint union of all the F e e ` quotients of rank r of all fibers of . The canonical map σ: X / X is an isomorphism on σ−1 . F e Freg We will now endow X with the structure of a reduced complex space. e M Let us first argue locally. Let p / q / / 0 be a presentation of on an open set e O|U O|U F|U F U. Here, M is an element of M . Then, for every x in U, we get the exact sequence q,p OU

M (x) πx Cp / Cq / / 0. F|x We have an inclusion ∼ Gr ∗ r, / Gr ∗ r, Cq / Gr(q r, q) F|x − 26 JULIEN GRIVAUX given by (q,Q) / (q π ,Q) / ker(q π ), ◦ x ◦ x where q and Q appear in the following diagram:

M(x) π Cp / Cq x / / 0 F|x q q e # Q

0 Therefore, we have a ﬁbered inclusion σ−1(U) / U Gr(q r, q) . We know that × − σ−1(U)= (x, E) U Gr(q r, q) such that Im M(x) E . ∈ × − ⊆ q ∗ ∗ Let (e1,...,eq) be the canonical basis of C , and e1 ,...,eq its dual basis. We can suppose that e ∗,...,e ∗ are linearly independent on E. We parametrize Gr(q r, q) in the neighborhood of x by 1 q−r − a matrix A = a M (C ). The associated vector space will be spanned by the columns of the i,j ∈ r,q−r id q−r i matrix . Writing M = Mj 1≤j≤q , Im(x) is a subspace of E if and only if for all i, 1 i p, A 1≤i≤p ≤ ≤ q−r M i(x)e + + M i(x)e e + a e + + a e =0 1 1 ··· q q ∧ l 1,l q−r+1 ··· r,l q l=1 ^ q−r+1 q −1 in C . This is clearly an analytic condition in the variables x and aij , thus σ (U) is an analytic subset of U Gr(q r, q). We endow σ−1(U) with the associated reduced structure. V × − We must check carefully that the structure deﬁned above does not depend on the chosen presentation. Let us consider two resolutions of on U F M π p / q / / 0 O|U O|U F|U ′ M ′ ′ π′ p / q / / 0 O|U O|U F|U

′ and a (q′, q) matrix α: q / q such that the diagram O|U O|U M π p / q / / 0 O|U O|U F|U

α id ′ M ′ ′ π′ p / q / / 0 O|U O|U F|U

id commutes. The morphism σ−1(U) / σ−1(U) _ _

U Gr(q r, q) / U Gr(q′ r, q′) × − × − CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 27

′ −1 is given by (x, E) / x, πx πx(E) according to the following diagram:

M(x) π Cp / Cq x / / 0 F|x α(x) id

′ ′ Cp Cq ′ / ′ / |x / 0 M (x) πx F ′ ′ −1 ′ −1 ′ Since πx α(x)(E) = πx(E), we have πx πx(E)= πx πx α(x)(E) . We can write this ′ −1 ′ ′ πx πx(E)= α(x)(E) +ker πx = α(x)(E)+Im M (x). −1 ′ ′ ′ If (x0, E0) is an element of σ (U) Gr(q r, q), then α(x0)(E0)+Im M (x0) belongs to Gr(q r, q ). ∗ × ∗ − − We can suppose as above that e1 ,...,eq−r are linearly independent on E0. Therefore, E0 is spanned by T q r vectors A1,...,Aq−r where Ai = 0,..., 1,...,a1,i,...,aq−r,i , the first “1” having index i. Then − ′ ′j ′j α(x0)(E0)+Im M (x0) is spanned by the vectors α(x0)(Ai)+ M (x0) 1≤i≤q−r , where the M are the 1≤j≤p′ ′ ′ columns of M . We can find q r independent vectors in this family, and this property holds also in a − ′jk neighborhood of (x0, E0). We denote these vectors by α(x)(Aik )+ M (x) 1≤k≤q′−r. Let us define

′jk f A1,...,Aq−r, x = span α(x)(A )+ M (x) . ik 1≤k≤q′−r ′ ′ This is a holomorphic map defined in a neighborhood of (x0, E0) with values in U Gr(q r, q ). On the × − same pattern, we can define another map g on a neighborhood of f x , E with values in U Gr(q r, q). 0 0 × − The couple (f,g) defines an isomorphism of complex spaces. This proves that X is endowed with the structure of an intrinsic reduced complex analytic space (not generally smooth). e We define a subsheaf of σ ∗ by N F (V )= s σ ∗ (V ) such that x, q,Q V, s ker q . N ∈ F ∀ { } ∈ x ∈ Remark that is supported in the singular locus of σ ∗ . N F Lemma 4.12. satisfies the following properties: N (i) is a coherent subsheaf of σ ∗ . N F (ii) σ ∗ / is locally free and rank σ ∗ / = rank( ). F N F N F p M q π Proof. We take a local presentation / / / 0 of . Then σ ∗ has the presen- O|U O|U F|U F F tation p M◦ σ q ∗ 1 / 1 / σ −1 / 0. O|σ− (U) O|σ− (U) F|σ (U) Let (x, E) / f (x, E),...,f (x, E) be a section of σ ∗ on V σ−1(U). Then s is a section of 1 q F ⊆ if and only if for every (x, E) in V , f1(x, E),...,fq(x, E) is an element of E. Let Uq−r,q be the N q universal bundle on Gr(q r, q). Then U is a subbundle of and is the image of U − q−r,q |V O|V N|V q−r,q |V q π ∗ by the morphism U / 1 / σ −1 . So is coherent. q−r,q |V O|σ− (U) F|σ (U) N (ii) Let us define by = σ ∗ / . For all (x, E) in V , we have an exact sequence E E F N / / / 0 N|(x,E) F|x E|(x,E) 28 JULIEN GRIVAUX and a commutative diagram E / Cq

πx / N|(x,E) F|x The first vertical arrow is the morphism U / restricted at (x, E), so it is onto. Thus π (E) q−r,q |U N x is the image of the morphism / . Since we have an exact sequence N|(x,E) F|x q 0 / π (E) / / Q / 0 x F|x where (x, E) = (x, Q), then dim π (E) = dim r. Since dim = dim π (E) + dim , we get x F|x − F|x x E|(x,E) dim = r. We can see that is a torsion sheaf, for is locally free of rank r. E|(x,E) N E We can now finish the proof of Theorem 4.11. Using Hironaka’s theorem, we desingularize the complex space X. We get a succession of blowups with smooth centers σ: X′ / X where X′ is smooth. By the Hironaka-Chow lemma (see [An-Ga, Th. 7.8]), we can suppose that τ = σ σ is a succession of blowups ◦ with smoothe centers. Since is locally free, the following sequencee e is exact:e e E 0 / σ ∗ / τ ∗ / σ ∗ / 0 . e N F E Therefore τ ∗ admits a locally free quotient of maximal rank. The unicity is clear. This finishes the F e e proof. 4.3. Construction of the classes in the general case. Let X be a complex compact manifold of dimension n. 4.3.1. Let be a coherent sheaf on X which has a locally free quotient of maximal rank. We have an F exact sequence 0 / / / / 0 T F E where is a torsion sheaf and is locally free. Then we define ch( ) by ch( ) = ch( )+ ch( ), where T E F F T E ch( ) has been constructed in part 4.1. Remark that ch( ) depends only on , the exact sequence T F F 0 / / / / 0 being unique up to (a unique) isomorphism. T F E We state now the Whitney formulae which apply to the Chern characters we have defined above. Proposition 4.13. Let 0 / / / / 0 be an exact sequence of coherent analytic F G H sheaves on X. Then ch( ), ch( ) and ch( ) are well defined and verify ch( ) = ch( ) + ch( ) under F G H G F H any of the following hypotheses: (i) , , are locally free sheaves on X. F G H (ii) , , are torsion sheaves. F G H (iii) admits a locally free quotient of maximal rank and is a torsion sheaf. G F Proof. (i) If , , are locally free sheaves on X, then ch( )= ch( ), ch( )= ch( ), ch( )= ch( ) F G H F F G G H H and we use Proposition 3.7 (i). (ii) This is Proposition 4.3 (i). (iii) Let be the locally free quotient of maximal rank of . We have an exact sequence E G 0 / / / / 0 T G E CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 29 where is a torsion sheaf. Since is a torsion sheaf, the morphism / / is identically T F F G E zero. Let us define ′ by the exact sequence T 0 / ′ / / / 0. T H E Then ′ is a torsion sheaf and we have the exact sequence of torsion sheaves T 0 / / / ′ / 0. F T T Thus admits a locally free quotient of maximal rank, so that ch( ) is defined and H H ch( )= ch( ) + ch( ′)= ch( ) + ch( ) ch( ) by (ii) H E T E T − F = ch( ) ch( ). G − F

Let us now look at the functoriality properties with respect to pullbacks. Proposition 4.14. Let f : X / Y be a holomorphic map. We assume that – dim Y = n and dim X n, ≤ – if dim X = n, f is surjective. Then for every coherent sheaf on Y which admits a locally free quotient of maximal rank, the following properties hold: (i) The Chern characters ch Tor ( ,f) are well deﬁned. i F (ii) f ∗ ch( )= ( 1)i ch Tor ( ,f) . F − i F i≥0 X Proof. (i) If dim X

Proof. The proof of (i) will use Lemmas 4.17, 4.18, and 4.19. We will prove the result by induction on the number d of blowups in σ. If d = 0, admits a locally free quotient of maximal rank and we can use F Proposition 4.14. Suppose now that (i) and (ii) hold at step d 1. As usual, we look at the ﬁrst blowup in σ − X

eσ1 iE E / X1 σ

q σ e e Ô Y / X. iY The sheaves Tor ( , σ) are torsion sheaves for j 1 and σ ∗ Tor ( , σ) = σ ∗ admits a locally free j F ≥ 1 0 F F quotient of maximal rank. Since σ1 consists of d 1 blowups, we can deﬁne by induction a class γ( ) in ∗ − F HD(X1, Q) as follows: e j γ( )= ( 1) ch Torj ( , σ) . e F − F Xj≥0 e Lemma 4.17. σ ∗γ( )= ( 1)i ch Tor ( , σ) . 1 F − i F i≥0 X Proof. We have by induction σ ∗γ( )= ( 1)p+q ch Tor (Tor ( , σ), σ ) = ( 1)p+q ch(E p,q) 1 F − p q F 1 − 2 p,qX≥0 h i p,qX≥0 where the Tor spectral sequence satisﬁes e E p,q = Tor (Tor ( , σ), σ ) 2 p q F 1 E p,q = Grp Tor ( , σ). ∞ p+q F p,q e 0,0 p,q All the Er , 2 r , are torsion sheaves except perhaps Er . Remark that no arrow dr starts or 0,0 ≤ ≤ ∞ arrives at Er . Thus we have ( 1)p+q[E p,q]= ( 1)p+q[E p,q]= ( 1)i Tor ( , σ) − 2 − ∞ − i F p,q p,q i≥1 pX+q≥1 pX+q≥1 X in Gtors(X). Using Proposition 4.13 (ii), we get σ ∗γ( ) = ch E 0,0 + ch ( 1)i Tor ( , σ) 1 F 2 − i F i≥1 X = ( 1)i ch Tor ( , σ) . − i F Xi≥0

Lemma 4.18. There exists a unique class ch( , σ) on X such that γ( )= σ ∗ ch( , σ). F F F e CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 31

Proof. We have i ∗ γ( )= i ∗ ( 1)j ch Tor ( , σ) E F E − j F j≥0 X = ( 1)j chE i ! Tor ( , σ) by induction property (ii) − E j F j≥0 X = chE i ! σ ![ ] = chE q !i ! [ ] = q ∗ chY i ! [ ] by (F ). E F Y F Y F n−1 By Proposition 3.5 (vi), there exists a unique class ch( , σ) on X such that γ( )= σ ∗ ch( , σ). e F F F Putting Lemma 4.17 and Lemma 4.18 together e σ ∗ ch( , σ)= σ ∗γ( )= ( 1)i ch Tor ( , σ) . F 1 F − i F i≥0 X Lemma 4.19. The class ch( , σ) does not depend on σ. F Proof. As usual, if we have two resolutions, we dominate them by a third one, as shown in the diagram

X ? ?? σ′ σ1 ?? 1 e ?? ? ′ X1 σ X @ 1 @@ ~ @ ~~ @ ~ ′ e σ @@ ~ σ e e @ ~~ e X Now σ ∗ ch( , σ)= σ ∗γ( ) byLemma4.18 F 1 F = ( 1)i ch Tor ( , σ) by Lemma 4.17. − i F e i≥0 X By symmetry σ ∗ ch( , σ)= σ ∗ ch( , σ′) and we get the result. F F We proved the existence statement and part (i) of Theorem 4.15 if σ consists of at most d blowups. The general case follows usinge the diagrame above.

We must now prove Theorem 4.15 (ii). Let Y be a smooth submanifold of X. We choose σ: X / X such that σ ∗ admits a locally free quotient of maximal rank and σ−1(Y ) is a simple normal crossing F ∗ divisor with branches Dj. We choose as usual j such that qj is injective, qj being deﬁned by thee diagram

i Dj Dj / X

q j eσ i Y Y / X We have q ∗ chY i ! [ ] = chDj q !i ! [ ] = chDj i ! σ ![ ] j Y F j Y F Dj F = ( 1)i i ∗ ch Tor ( , σ) by Proposition 4.14 (ii). − Dj i F i≥0 X Now, by the point (i), we have ( 1)i ch Tor ( , σ) = σ ∗ ch( ). Thus we get i≥0 − i F F q ∗PchY (i ! [ ]) =i ∗ σ ∗ ch( )= q ∗ i ∗ ch( ) . j Y F Dj F j Y F 32 JULIEN GRIVAUX

Therefore ch i ! [ ] = i ∗ ch( ) and the proof is complete. Y F Y F 5. The Whitney formula In the previous section, we achieved an important step in the induction process by deﬁning the classes ch( ) when is any coherent sheaf on a n-dimensional manifold. To conclude the proof of Theorem 4.2, F F it remains to check properties (Wn), (Fn) and (Pn). The crux of the proof is in fact property (Wn). The main result of this section is Theorem 5.1. The other induction hypotheses will be proved in Theorem 5.14.

Theorem 5.1. (Wn) holds.

To prove Theorem 5.1, we need several reduction steps. 5.1. Reduction to the case where and are locally free and is a torsion sheaf. F G H Proposition 5.2. Suppose that (W ) holds when and are locally free sheaves and is a torsion n F G H sheaf. Then (Wn) holds for arbitrary sheaves. Proof. We proceed by successive reductions. Lemma 5.3. It is suﬃcient to prove (W ) when , , admit a locally free quotient of maximal rank. n F G H Proof. We take a general exact sequence 0 / / / / 0. Let σ : X / X be a F G H bimeromorphic morphism such that σ ∗ , σ ∗ and σ ∗ admit locally free quotients of maximal rank F G H (we know that such a σ exists by Theorem 4.11). We have an exact sequence deﬁning eand : Q T1

0 0

Q Tor ( , σ) Tor ( , σ) σ ∗ σ ∗ σ ∗ 0 ··· 1 G 1 H F G H

T1 0 0

Remark that is a torsion sheaf. By Proposition 4.13 (iii), admits a locally free quotient of maximal T1 Q rank and ch(σ ∗ ) = ch( ) + ch( ). Furthermore, F T1 Q [ ] Tor ( , σ) + Tor ( , σ) = 0 in G (X). T1 − 1 H 1 G − · · · tors Then by Proposition 4.10 (i) and Proposition 4.13 (ii), e σ ∗ ch( ) + ch( ) ch( ) = ( 1)i ch Tor ( , σ) + ch Tor ( , σ) ch Tor ( , σ) F H − G − i F i H − i G i≥0 X h i = ch σ ∗ + ch σ ∗ ch σ ∗ ch( ) F H − G − T1 = ch( ) + ch σ ∗ ch σ ∗ . Q H − G Since σ ∗ is injective, Lemma 5.3 is proved. Lemma 5.4. It is suﬃcient to prove (W ) when , admit a locally free quotient of maximal rank and n F G is a torsion sheaf. H CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 33

Proof. By Lemma 5.3, we can assume that , , admit a locally free quotient of maximal rank. In F G H the sequel, the letter “ ” will denote a torsion sheaf and the letter “ ” a locally free sheaf. Let be T E E1 the locally free quotient of maximal rank of , so we have an exact sequence H 0 / / / / 0. T1 H E1 We deﬁne by the exact sequence F1 0 / / / / 0. F1 G E1 Then we get a third exact sequence

0 / / / / 0. F F1 T1 We have by definition ch( )= ch( ) + ch( ). Thus, H E1 T1 ch( ) + ch( ) ch( )= ch( ) + ch( ) ch( ) + ch( )+ ch( ) ch( ) F H − G F T1 − F1 F1 E1 − G ch( 1)+ ch( 1) ch( ) − T E − H = ch( ) + ch( 1) ch( 1) + ch( 1)+ ch( 1) ch( ) . F T − F F E − G Let 2 be the locally free quotient of maximal rank of . We define 2 by the exact sequence E G T 0 / / / / 0. T2 G E2 The morphism from to (via ) induces a morphism / which remains of course surjective. G E1 H E2 E1 Let be the kernel of this morphism, then is a locally free sheaf. We get an exact sequence E3 E3 0 / / / / 0. T2 F1 E3 Therefore admits a locally free quotient of maximal rank and ch( ) = ch( )+ ch( ). On the F1 F1 T2 E3 other hand, by Proposition 4.13 (i), ch + ch = ch , and we obtain ch + ch ch( )= E1 E3 E2 F1 E1 − G ch( )+ ch( ) + ch( ) ch( ) ch( )+ ch( ) = 0. Therefore, T2 E3 E2 − E3 − T2 E2 ch( ) + ch( ) ch( ) = ch( ) + ch( 1) ch( 1). F H − G F T − F Since is a torsion sheaf, we are done. T1 We can now conclude the proof of Proposition 5.2. By Lemma 5.4, we can suppose that , admit locally free quotients of maximal rank and is a torsion F G H sheaf. Let and be the locally free quotients of maximal rank of and . We define and by E1 E2 F G T1 T2 the two exact sequences 0 / / / / 0 T1 F E1 0 / / / / 0. T2 G E2 The morphism / induces a morphism / . We get a morphism / with F G T1 T2 E1 E2 torsion kernel and cokernel. Since is a locally free sheaf, this morphism is injective. In the following E1 34 JULIEN GRIVAUX diagram, we introduce the cokernels and : T3 T4 0 0 0

0 / / / / 0 T1 F E1

0 / / / / 0 T2 G E2

T3 H T4

0 0 0 By the nine lemma, 0 / / / / 0 is an exact sequence of torsion sheaves. Then T3 H T4 by Proposition 4.3 (i), ch( ) + ch( ) ch( ) = ch + ch + ch + ch ch ch F H − G T1 E1 T3 T4 − T2 − E2 = ch 1 + ch 4 ch 2. E T − E This finishes the proof. 5.2. A structure theorem for coherent torsion sheaves of projective dimension one. In section 5.1 we have reduced the Whitney formula to the particular case where and are locally free sheaves F G and is a torsion sheaf. We are now going to prove that it is sufficient to suppose that is the H H push-forward of a locally free sheaf on a smooth hypersurface of X. The main tool of this section is the following proposition: Proposition 5.5. Let be a torsion sheaf which admits a global locally free resolution of length two. H Then there exist a bimeromorphic morphism σ : X / X obtained by a finite number of blowups with smooth centers, a simple normal crossing divisor D, D X, and an increasing sequence Di 1≤i≤r of ⊆ r ∗ e subdivisors of D such that σ is everywhere locally isomorphic to X / . H i=1 O e IDi L Proof. Let 0 / / / / 0 be a locally free resolution of , such that rank = E1 E2 H H E1 rank = r. Recall that the kth Fitting ideal of is the coherent ideal sheaf generated by the E2 H determinants of all the k k minors of M when M is any local matrix realization in coordinates of the × morphism / (for a general presentation of the Fitting ideals, see [Ei]). We have E1 E2 Fitt ( ) Fitt ( ) Fitt ( ) ) 0 . 1 H ⊇ 2 H ⊇···⊇ r H { } These ideals have good functoriality properties. Indeed, if σ: X / X is a bimeromorphic morphism, ∗ ∗ ∗ ∗ ∗ the sequence 0 / σ 1 / σ 2 / σ / 0 is exact and Fittj σ = σ Fittj ( ) (by ∗ E E H e H H σ Fittj ( ), we mean of course its image in X ). By the Hironaka theorem, we can suppose, after taking H O e a finite number of pullbacks under blowups with smooth centers, that all the Fitting ideals Fittk( ) are ′ F ideal sheaves associated with effective normal crossing divisors Dk. Now, take an element x of X. Consider an exact sequence r / r / / 0 OU M OU H|U CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 35 in a neighbourhood of x. The matrix M is a r r matrix of holomorphic functions on U. Let φ =0 × 1 be an equation of D around x. Then we can write M = φ M and the coefficients of M generate . 1 1 1 1 Ox Thus, at least one of these coefficients does not vanish at x. We can suppose that it is the upper-left one. By Gauss elimination process, we can write 1 0 0 ··· 0 M = φ1  .  . . M2    0      Then, since Fitt ( ) = Fitt (M), we get Fitt (M)= φ2 Fitt (M ). Since Fitt (M) is principal, so is k F |U k 2 1 1 2 2 Fitt1(M2) and we write Fitt1(M2)= φ2 . Then, by the same argument as above, we can write

φ1 0 0 ······ 0 φ1φ2 0 0  . ···  M = . 0 .    . .   . . M3     0 0    By this algorithm, we get   φ 0 0 1 ··· 0 φ1φ2 M =  . .  . ..    0 φ φ   1 r  ···  and then |U X X . Thus, if D1,...,Dr are the divisors of F ≃ O φ1 X |U ⊕···⊕ O φ1 ...φr X |U O ′ ′ O φ1, φ1 φ2,...,φ1 φ2 ...φr, we have Dk = D D , which shows that the divisors Dk are intrinsically k − k−1 defined by . F From now on, we will say that a torsion sheaf is principal if it is everywhere locally isomorphic to r H a fixed sheaf X / where the Di are (non necessarily reduced) effective normal crossing divisors i=1 O IDi and D1 D2L Dr. We will denote by ν( ) the number of branches of D, counted with their ≤ ≤ ···≤ H multiplicities. Proposition 5.6. It suffices to prove the Whitney formula when and are locally free sheaves and F G H is the push-forward of a locally free sheaf on a smooth hypersuface. Proof. We proceed in several steps. Lemma 5.7. Consider an exact sequence 0 / / /i /0 where Y is a smooth hy- F G Y ∗E persurface of X, is a locally free sheaf on X and is a locally free sheaf on Y . Then is locally free G E F on X.

Proof. Let mx be the maximal ideal of the local ring x. By Nakayama’s lemma, it suﬃces to show that O O for every x in X, Tor x , /m = 0. Since Y is a hypersurface, i admits a locally free resolution 1 Fx Ox x Y ∗E of length two. Thus TorOx , /m TorOx (i ) , /m = 0. 1 Fx Ox x ≃ 2 Y ∗E x Ox x Lemma 5.8. It suﬃces to prove (W )when , are locally free sheaves and is principal. n F G H 36 JULIEN GRIVAUX

Proof. By Proposition 5.2, it is enough to prove the Whitney formula when , are locally free sheaves F G and is a torsion sheaf. So we suppose that , and verify these hypotheses. By Proposition 5.5, H F G H there exists a bimeromorphic morphism σ : X / X such that σ ∗ is principal. We have an exact H sequence 0 / Tor ( , σ) e/ σ ∗ / σ ∗ / σ ∗ / 0. 1 H F G H But Tor ( , σ) is a torsion sheaf and σ ∗ is locally free, so we get an exact sequence 1 H F 0 / σ ∗ / σ ∗ / σ ∗ / 0. F G H and we have Tor ( , σ) =0for i 1. By Proposition 3.7 (ii), we obtain the equalities ch(σ ∗ )= σ ∗ch( ) i H ≥ F F and ch(σ ∗ )= σ ∗ch( ), and by Proposition 4.3 (iii) we get ch(σ ∗ )= σ ∗ ch( ). Thus G G H H σ ∗ ch( ) + ch( ) ch( ) = ch(σ ∗ ) + ch(σ ∗ ) ch(σ ∗ ). F H − G F H − G Lemma 5.9. Suppose that (W ) holds if , are locally free sheaves and is the push-forward of a n F G H locally free sheaf on a smooth hypersurface. Then (W ) holds when , are locally free sheaves and n F G H is principal. Proof. We argue by induction on ν( ). If ν( ) = 0, = 0 and . If ν( ) = 1, is the H H H F ≃ G H H push-forward of a locally free sheaf on a smooth hypersurface and there is nothing to prove. In the general case, let Y be a branch of D . Since Y D for every i with 1 i r, we can see that 1 ≤ i ≤ ≤ = is locally free on Y . Besides, if we define by the exact sequence E H|Y H 0 / / e / i / 0, H H Y ∗E r is everywhere locally isomorphic to e X / . Thus is principal and ν( ) = ν( ) 1. We H i=1 O IDi−Y H H H − define by the exact sequence: L e E e e 0 / / / i / 0. e E G Y ∗E By Lemma 5.7, is locally free. Furthermore, we have an exact sequence E e 0 / / / / 0. e F E H By induction, ch = ch( ) + ch and by our hypothesis ch( )= ch + ch i . Since , and e e Y ∗ E F H G E E H H iY ∗ are torsion sheaves, ch( ) = ch + ch iY ∗ and we get ch( )= ch( ) + ch( ). E e H e H E G e F H e Putting the two lemmas together, we obtain Proposition 5.6. e 5.3. Proof of the Whitney formula. We are now ready to prove Theorem 5.1. In the sections 5.1 and 5.2, we have made successive reductions in order to prove the Whitney formula in a tractable context, so that we are reduced to the case where and are locally free sheaves and F G = i , where Y is a smooth hypersurface of X and is a locally free sheaf on Y . Our working H Y ∗E E hypotheses will be these. Let us briefly explain the sketch of the argument. We consider the sheaf on X P1 obtained by G × deformation of the second extension class of the exact sequence 0 / / / / 0. F G H Then and for t = 0. It will turn out that admitse a locally free quotient G|X×{0} ≃F⊕H G|X×{t} ≃ G 6 G of maximal rank on the blowup of X P1 along Y 0 , and the associated kernel will be the Q × ×{ } N push-forwarde of a locally freee sheaf on the exceptional divisor E, say e= i . Then we consider the N E∗L class α = ch( )+i ch( ) td(N )−1 on the blowup. After explicit computations, it will appear that Q E∗ L E/X CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 37

α is the pullback of a form β on the base X P1. By the P1-homotopy invariance of Deligne cohomology × (Proposition 3.3 (vi)), β|X×{t} does not depend on t. This will give the desired result. Let us first introduce some notations. The morphism / will be denoted by γ. Let s be a global F G 1 section of 1 (1) which vanishes exactly at 0 . Let pr : X P / X be the projection on the first OP { } 1 × factor. The relative (1), namely ⊠ 1 (1), will still be denoted by (1). We define a sheaf on O OX OP O G X P1 by the exact sequence × e 0 / pr ∗ / pr ∗ (1) pr ∗ / / 0. 1 F (id ⊗s, γ) 1 F ⊕ 1 G G Remark that and if t = 0. e G0 ≃F⊕H Gt ≃ G 6 Lemma 5.10. There exist two exact sequences e e (i) 0 / pr ∗ (1) / / pr ∗ / 0 1 F G 1 H ∗ (ii) 0 / / pr1 (1)e / i / 0. G G X0∗H Remark 5.11. (i) implies that is flate over P1. G Proof. (i) The morphism pr ∗ (1) pr ∗ / / pr ∗ / / pr ∗ induces a morphism / / pr ∗ . 1 F e ⊕ 1 G 1 G 1 H G 1 H If is the kernel of this morphism, we obtain an exact sequence K e 0 / pr ∗ / pr ∗ (1) pr ∗ / / 0. 1 F (id ⊗s, id) 1 F ⊕ 1 F K Thus = pr ∗ (1). K 1 F (ii) We consider the morphism pr ∗ (1) pr ∗ / / pr ∗ (1) 1 F ⊕ 1 G 1 G f + g / γ(f) g s. − ⊗ It induces a morphism φ: / pr ∗ (1). The last morphism of (ii) is just the composition G 1 G pr ∗ (1) / / i / / i . e 1 G X0∗G X0∗H The cokernel of this morphism has support in X 0 . Besides, the action of t on this cokernel is ×{ } zero. The restriction of φ to the fiber X = X 0 is the morphism / , thus the sequence 0 ×{ } F ⊕ H G / pr ∗ (1) / i / 0 is exact. The kernel of φ, as its cokernel, is an -module. G 1 G X0∗H OX0 1 Thus we can find such that ker φ = iX0∗ . Since X0 is a hypersurface of X P , for every coherent e 1Z Z × sheaf on X P , we have Tor2( ,iX0 ) = 0. Applying this to = / and using Remark 5.11, we L × L L G iX0∗ get Z e Tor1(iX0∗ ,iX0 ) Tor1( ,iX0 )= 0 . ∗ Z ⊆ G { } 1 But Tor1(iX0∗ ,iX0 ) NX /X×P , so = 0 . Z ≃Z⊗ 0 ≃ Z Z { } e Recall now that = i where Y is a smooth hypersurface of X and is a locally free sheaf on Y . We H Y ∗E E consider the space MY/X of the deformation of the normal cone of Y in X (see [Fu]). Basically, MY/X is the blowup of X P1 along Y 0 . Let σ : M / X P1 be the canonical morphism. Then × ×{ } Y/X × σ ∗X is a Cartier divisor in M with two simple branches: E = P N and D = Bl X X, 0 Y/X Y/X ⊕ OY Y ≃ which intersect at P N Y . The projection of the blowup from E to Y 0 will be denoted by q, Y/X ≃ ×{ } and the canonical isomorphism from D to X 0 will be denoted by µ. ×{ } We now show: 38 JULIEN GRIVAUX

Lemma 5.12. The sheaf σ ∗ admits a locally free quotient with maximal rank on M , and the G Y/X associated kernel is the push-forward of a locally free sheaf on E. More explicitly, if F is the excess N conormal bundle of q, = i eq ∗ F . N E∗ E ⊗ Proof. We start from the exact sequence 0 / / pr ∗ (1) / i / 0. We deﬁne G 1 G X0∗H Q by the exact sequence 0 / / σ ∗ pr ∗ (1) / σ ∗i / 0. Since σ ∗i is the push- Q 1 G e X0∗H X0∗H forward of a locally free sheaf on E, by Lemma 5.7, the sheaf is locally free on M . The sequence Q Y/X 0 / Tor (i , σ) / σ ∗ / / 0 1 X0∗H G Q is exact. The ﬁrst sheaf being a torsion sheaf, is a locally free quotient of with maximal rank. Q e G Besides, using the notations given in the following diagram e iE E / MY/X

q σ Y 0 / X P1 ×{ } iY ×{0} × we have Tor (i , σ) = i q ∗ F where F is the excess conormal bundle of q (see [Bo-Se, 15]. 1 X0∗H E∗ E ⊗ § Be aware of the fact that what we note F is F ∗ in [Bo-Se]). This finishes the proof. We consider now the exact sequence 0 / / σ ∗ / / 0 where is locally free on ∗ N G Q ∗ Q MY/X and = iE∗ q F = iE∗ . We would like to introduce the form ch(σ ), but it is not defined N E⊗ L e G since M is of dimension n + 1. However, σ ∗ fits in a short exact sequence where the Chern classes Y/X G of the two other sheaves can be defined. Remark that we need Lemma 5.12 to performe this trick, since it cannot be done on X P1. e × −1 Lemma 5.13. Let α be the Deligne class on MY/X defined by α = ch( )+ iE∗ ch( ) td NE/M . Q L Y /X (i) The class α is the pullback of a Deligne class on X P1. × (ii) We have i ∗ α = µ ∗ chX0 . D G0 Proof. We compute explicitly i ∗α. E e ∗ −1 −1 iE iE∗ ch( ) td NE/M = ch( ) td NE/M c1 NE/M by Proposition 3.5 (vii) L Y /X L Y /X Y /X −c1 NE/M = ch( ) 1 e Y /X L − ∗ = ch( ) ch NE/M by Proposition 3.7 (iii) L − L⊗ Y /X ∗ ∗ = ch iE ch NE/M . N − L⊗ Y /X CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 39

From the exact sequence 0 / / σ ∗ / / 0, we get the exact sequence of locally N G Q free sheaves on E: 0 / i ∗ / i ∗ σ ∗ / i ∗ / 0. Since i ∗ ch( )= ch i ∗ , we obtain EN E G e EQ E Q EQ ∗ ∗ ∗ ∗ iEα = ch iE + ch iE ch NE/M Q N − L⊗e Y /X ∗ ∗ ∗ = chiEσ ch NE/M by Proposition 3.7 (i) G − L⊗ Y /X ∗ ∗ ∗ ∗ ∗ = chq iY + chq iY ch NE/M Fe H − L⊗ Y /X ∗ ∗ ∗ ∗ ∗ = q ch iY + q ch( ) ch q F NE/M by Proposition 3.7 (ii). F E − E⊗ ⊗ Y /X Recall that the conormal excess bundle F is the line bundle deﬁned by the exact sequence ∗ ∗ ∗ 0 / F / q N 1 / N / 0. Y/X×P E/MY /X

∗ ∗ ∗ ∗ ∗ ∗ ∗ Thus, we have det q NY/X×P1 = F NE/M . Since det q NY/X×P1 = q det NY/X×P1 , we get by ⊗ Y /X Proposition 3.7 (ii) again ∗ ∗ ∗ ∗ i α = q ch i + ch( ) ch det N P1 . E Y F E − E ⊗ Y/X× This proves (i). (ii) The divisors E and D meet transversally. Then, by Proposition 3.5 (iv),

∗ ∗ ∗ −1 iDα = iDch( )+ iDiE∗ ch( ) td NE/M Q L Y /X ∗ ∗ −1 = ch iD + i i ch( ) td NE/M Q E∩D / D∗ E∩D / E L Y /X − 1 = ch i ∗ + i ch i ∗ td N . DQ E∩D / D∗ E∩D / E L E∩D/D We remark now that i ∗ = i ∗ . Since dim( E D)= n 1, we obtain E∩D / E L D∩E N ∩ − i ∗ α = ch i ∗ + chD i i ∗ = ch i ∗ + chD i ∗ . D DQ E∩D / D∗ E∩DN DQ DN We have the exact sequence on D: 0 / i ∗ / i ∗ σ ∗ / i ∗ / 0. Therefore i ∗ σ ∗ DN D G DQ D G admits a locally free quotient of maximal rank and, µ being an isomorphism, e e ch i ∗ + chD(i ∗ ) = chD i ∗ σ ∗ = chD µ ∗ = µ ∗ chX0 . DQ DN D G G0 G0 e e e We are now ready to use the homotopy property for Deligne cohomology (Proposition 3.3 (vi)). Let α be the form deﬁned in Lemma 5.13. Using (i) of this lemma and Proposition 3.5 (vi), we can write α = σ ∗β. Thus i ∗ α = i ∗ σ ∗β = µ ∗i ∗ β. By (ii) of the same lemma, i ∗ α = µ ∗ chX0 and we get D D X0 D G0 ∗ X0 1 i β = ch 0 . If t P 0 , we have clearly β|X = ch( ). Since β = β we obtain X0 t |Xt |X0 G ∈ \{ } G e ch( ) = chX0 = ch( ) + ch( ). e G G0 F H ✷ e We can now establish the remaining induction properties. Theorem 5.14. The following assertions are valid:

(i) Property (Fn) holds. (ii) Property (Pn) holds. 40 JULIEN GRIVAUX

Proof. (i) We take y = [ ]. Let us ﬁrst suppose that f is a bimeromorphic map. Then there exists a F bimeromorphic map σ : X / X such that (f σ) ∗ admits a locally free quotient of maximal rank. ◦ F Then by Theorem 4.15 (i), e σ ∗ ch f ![ ] = ch σ ! f ![ ] = (f σ) ∗ ch = σ ∗ f ∗ ch( ) . F F ◦ F F Suppose now that f is a surjective map. Then there exist two bimeromorphic maps πX : X / X, π : Y / Y and a surjective map f : X / Y such that: Y e f e– the diagram X e / Y eise commutative.e

π π X e Y e X / Y f – the sheaf π ∗ admits a locally free quotient of maximal rank. Y F E We can write π ! [ ] = [ ]+y in G(Y ), where y is in the image of the natural map ι: G (Y ) / G(Y ). Y F E tors The functoriality property being known for bimeromorphic maps, it holds for πX and πY . The result is now a consequence of Propositione 3.7e (ii) ande Proposition 4.3 (iii). e e In the general case, we consider the diagram used in the proof of Proposition 4.3 (iii)

f i X e / W W / Y

π τ π X e e Y X / f(X) / Y f where f is surjective. Then the functoriality property holds for f by the argument above and for iW by Theorem 4.15 (ii). This ﬁnishes the proof. e e (ii) We can suppose that x = [ ], y = [ ] and that and admit locally free quotients , of maximal F G F G E1 E2 rank. Let and be the associated kernels. We can also suppose that supp( ) lies in a simple normal T1 T2 T1 crossing divisor. Then ch([ ].[ ]) = ch([ ].[ ]) + ch([ ].[ ]) + ch([ ].[ ]) + ch([ ].[ ]) F G E1 E2 E1 T2 E2 T1 T1 T2 = ch( )ch( )+ ch( ) ch( )+ ch( ) ch( ) + ch([ ].[ ]) E1 E2 E1 T2 E2 T1 T1 T2 by (W ), Proposition 4.3 (ii) and Proposition 3.7 (iii). By d´evissage, we can suppose that is a n T1 -module, where Z is a smooth hypersurface of X. We write [ ] = i u and [ ] = v. Then OZ T1 Z! T2 [ ] . [ ]= i u.i ! v . So T1 T2 Z! Z −1 ch [ ].[ ] = i chZ (u.i ! v) td N T1 T2 Z∗ Z Z/X Z ∗ −1 = iZ∗ ch (u) iZ ch(v) td NZ/X by (Pn−1) and Proposition 4.10 (ii) Z −1 = iZ∗ ch (u) td NZ/X ch(v) by the projection formula = ch i u ch(v) = ch [ ] ch [ ] Z! T1 T2

The proof of Theorem 4.2 is now concluded with the exception of property (RRn). CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 41

6. The Grothendieck-Riemann-Roch theorem for projective morphisms 6.1. Proof of the GRR formula. We have already obtained the (GRR) formula for the immersion of a smooth divisor. We reduce now the general case to the divisor case by a blowup. This construction is classical (see [Bo-Se]). Theorem 6.1. Let Y be a smooth submanifold of X. Then, for all y in G(Y ), we have −1 ch iY !y = iY ∗ ch(y) td(NZ/X ) . Proof. We blow up Y along X as shown below, where E is the exceptional divisor.

i E E / X

q p e Y / X iY Let F be the excess conormal bundle of q deﬁned by the exact sequence 0 / F / q ∗N ∗ / N ∗ / 0. Y/X E/X e If d is the codimension of Y in X, then rank(F )= d 1. Recall the following formulae: − ! ! (a) Excess formula in K-theory (Proposition 7.6 (ii)): for all y in G(Y ), p i!y = iE! q y . λ−1F . (b) Excess formula in Deligne cohomology (Proposition 3.5 (vi)): for each Deligne class β on Y , ∗ ∗ ∗ p iY ∗ β = iE∗ q β cd−1(F ) . ∗ ∗ −1 (c) If G is a vector bundle of rank r, then ch λ−1[G] = cr(G ) td(G ) ([Bo-Se, Lemme 18]). We compute now ∗ ! ! p ch iY !y = ch p iY !y = ch iE!(q y . λ−1[F ]) by (Fn) and (a) E ! −1 = iE∗ ch q y . λ−1[F ] td NE/X by (GRR) for iE e ∗ Y ∗ −1 ∗ = iE∗ q ch (y) ch λ−1[F ] q td NY/X td F by (Fn), (Pn) and Proposition 3.7 (i) ∗ Y −1 ∗ = iE∗ q ch (y) td NY/X cd−1(F ) by (c) ∗ Y −1 = p iY ∗ ch (y) td NY/X by (b). Y −1 Thus ch iY !y = iY ∗ ch (y) td NY/X . Now we can prove a more general Grothendieck-Riemann-Roch theorem: Theorem 6.2. The GRR theorem holds in rational Deligne cohomology for projective morphisms between smooth complex compact manifolds. Proof. Let f : X / Y be a projective morphism. Then we can write f as the composition of an immersion i: X / Y PN and the second projection p: Y PN / Y . By Theorem 6.1, GRR is × × N N true for i. Now the arguments in [Bei] show that the canonical map from G(Y ) Z G(P ) to G(Y P ) ⊗ × is surjective. Therefore, it is enough to prove GRR for p with elements of the form y . w, where y belongs to G(Y ) and w belongs to G(PN ). By the product formula for the Chern character, we are led to the Hirzebruch-Riemann-Roch formula for PN , which is well known. 42 JULIEN GRIVAUX

6.2. Compatibility of Chern classes and the GRR formula. We will show that the GRR formula for immersions combined with some basic properties can be suﬃcient to characterize completely a theory of Chern classes. This will give various compatibility theorems. We assume to be given for each smooth complex compact manifold X a graded commutative cohomology ring A(X)= dim X Ai(X) which is an algebra over Q, Q A0(X), with the following properties: ⊕i=0 ⊂ (α) For each holomorphic map f : X / Y , there is a pull-back morphism f ∗ : A(Y ) / A(X) which is functorial and compatible with the products and the gradings. (β) If σ is the blowup of a smooth complex compact manifold along a smooth submanifold, then σ ∗ is injective. (γ) If E is a holomorphic vector bundle on X and π: P(E) / X is the projection of the associated projective bundle, then π ∗ is injective. (δ) If X is a smooth complex compact manifold and Y is a smooth submanifold of codimension d, ∗ ∗+d then there is a Gysin morphism i∗ : A (Y ) / A (X). The main examples are: i 2i 2i >i i i i 2i 2i 2i A (X)= HD (X, Q(i)), H (X, ΩX• ), H (X, ΩX ), F H (X, C), H (X, Q) and H (X, C). Then we have the following theorem:

′ Theorem 6.3. Suppose that we have two theories of Chern classes ci and ci for coherent sheaves on any i ′ smooth complex compact manifold X with values in A (X) such that c0 = c0 =1 and (i) The Whitney formula holds for the total classes c and c′. (ii) The functoriality formula holds for c and c′. (iii) If L is a holomorphic line bundle, then ′ ′ c(L)=1+ c1(L)= c (L)=1+ c1(L). (iv) In both theories, the GRR theorem holds for immersions. Then for every coherent sheaf , c( ) and c′( ) are equal. F F F Remark 6.4. 1. The same conclusion holds for cohomology algebras over Z if we assume GRR without denominators. 2. If X is projective, (i) and (iii) are suﬃcient to imply the equality of c and c′ because of the existence of global locally free resolutions. 3. In (iv), the Todd classes of the normal bundle are deﬁned in A(X) since A(X) is a Q-algebra. Proof. We start by proving that for any holomorphic vector bundle E, c(E) and c′(E) are equal. Actually, this proof is a variant of the splitting principle. We argue by induction on the rank of E. Let π : P(E) / X be the projective bundle of E. Then we have the exact sequence

0 / ( 1) / π ∗E / F / 0 OE − on P(E), where F is a holomorphic vector bundle on P(E) whose rank is the rank of E minus one. By induction, c(F ) = c′(F ) and by (iii), c( ( 1)) = c′( ( 1)). By (i), c(π ∗E) = c′(π ∗E) and by (ii), OE − OE − π ∗[c(E) c′(E)]=0. By (γ), c(E)= c′(E). − We can now prove Theorem 6.3. As usual, we deal with exponential Chern classes. The proof proceeds by induction on the dimension of the base manifold X.

Let be a coherent sheaf on X. By Theorem 4.11 there exists a bimeromorphic morphism σ: X / X F which is a finite composition of blowups with smooth centers and a locally free sheaf on X which E is a quotient of maximal rank of σ ∗ . Furthermore, by Hironaka’s theorem, we can supposee that the F e CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 43 exceptional locus of σ and the kernel of the morphism σ ∗ / are both contained in a simple F E normal crossing divisor D of X. Thus n n ! i i σ [ ]=e ( 1) Tori( , σ) = [ ]+ ( 1) Tori( , σ) F − F E − F i=0 i=1 X X and then σ ![ ] belongs to [ ]+ G (X). Now there is a surjective morphism F E D N e GDi (X) / GD(X) . i=1 L e e ′ Moreover, by Proposition 7.2, G(Di) is isomorphic to G (X). Remark that td N = td N . Di D /X D /X i e i e ′ By the GRR formulae (iv) and the induction hypothesis, ch and ch are equal on each G (X). By the e Di first part of the proof, ch( ) = ch′( ). Thus ch σ ![ ] = ch′ σ ![ ] . By (ii), σ ∗ ch( ) ch′( ) = 0. E E F F F − F Since σ ∗ is injective by (β), ch( ) = ch′( ). e F F Corollary 6.5. Let be a coherent analytic sheaf on X. Then: F (i) The classes c ( ) in H2i(X, Q(i)) and c ( )top in H2i(X, Z) have the same image in H2i(X, Q). i F D i F (ii) The image of c ( ) via the natural morphism from H2i(X, Q(i)) to Hi(X, Ωi ) is the ith Atiyah i F D X Chern class of F Proof. If L is a holomorphic line bundle, then (i) and (ii) hold for L. Indeed, using the isomorphism 2 between Pic(X) and HD(X, Z(1)), the topological and Atiyah first Chern classes are obtained by the two morphisms of complexes 2iπ 2iπ Z / and Z / O O d/2iπ Z 1 ΩX On the other hand, GRR for immersions holds for topological Chern classes by [At-Hi] and for Atiyah Chern classes by [OB-To-To3]. Remark 6.6. If X is a Kähler complex manifold, the Green Chern classes are the same as the Atiyah Chern classes and the complex topological Chern classes. If X is non Kähler, GRR does not seem to be known for the Green Chern classes, except for a constant morphism (see [To-To]). If this were true Gr 2i for immersions, it would imply the compatibility of ci( ) and ci( ) , via the map from HD (X, Q(i)) 2i >i F F to H (X, ΩX• ). On the other hand, if this compatibility holds, it implies GRR for immersions for the Green Chern classes.

7. Appendix. Analytic K-theory with support Our aim in this appendix is to answer the following question: if f : X / Y is a morphism, a torsion F sheaf on Y with support W , and Z = f −1(W ), can we express the sheaves Tor ( ,f) in terms of the i F sheaves Tor ( ,f )? i F|W |Z 7.1. Deﬁnition of the analytic K-theory with support. Let X be a smooth compact complex manifold and Z be an analytic subset of X. We will denote by cohZ (X) the abelian category of coherent sheaves on X with support in Z. Then, X being compact, coh (X)= , coh(X) such that n = 0 for n 0 . Z F F ∈ IZ F ≫ 44 JULIEN GRIVAUX

Definition 7.1. We define G (X) as the quotient of Z coh (X) by the relations: + = if there Z Z F H G exists an exact sequence 0 / / / / 0 with , and elements of coh (X). In F G H F G H Z other words, G (X) is the Grothendieck group of coh (X). The image of an element of coh (X) in Z Z F Z G (X) will be denoted by [ ]. Z F We first prove

Proposition 7.2. The map iZ∗ : G(Z) / GZ (X) is a group isomorphism.

Proof. We consider the inclusion coh(Z) cohZ (X). If is in cohZ (X), we can filter by the formula i i ⊆ F i F F = . This is a finite filtration and the associated quotients F / i+1 are in coh(Z). Therefore, F IZ F F F we can use the dévissage theorem for the Grothendieck group, which is in fact validF for all K-theory groups (see [Qui, 5, Theorem 4]). § 7.2. Product on the K-theory with support. Let Z be a smooth submanifold of X. For any x in

GZ (X), by Proposition 7.2, there exists a unique x in G(Z) such that iZ∗x = x. Deﬁnition 7.3. We deﬁne a product G(Z) G (X) G (X) by x y = i (x. y). Z Z / Z Z Z∗ ⊗ Z In other words, the following diagram is commutative:

G(Z) Z GZ (X) / GZ (X) ⊗ Z ≃ ≃ G(Z) Z G(Z) / G(Z) ⊗ Remark 7.4. Deﬁnition 7.3 has some obvious consequences: 1. For any x in G(Z), i x = x [i ]. Z∗ Z Z∗OZ 2. More generally, i (x. y)= x i y. Z∗ Z Z∗ 3. The product endows G (X) with the structure of a G(Z)-module. Z Z 7.3. Functoriality. Let f : X / Y be a holomorphic map and W be a smooth submanifold of Y such −1 ! that Z = f (W ) is smooth. We can deﬁne a morphism f : GW (Y ) / GZ (X) by the usual formula f ∗[ ]= ( 1)i Tor ( ,f) . Let f be the restriction of f to Z, as shown in the following diagram: K i≥0 − i K h i P i Z Z / X

f f W / Y iW Then we have the following proposition: Proposition 7.5. For all x in G(W ) and for all y in G (Y ), we have f ! x y = f !x f !y. W W Z

Recall that if E is a holomorphic vector bundle on X, λ−1[E] is the element of G(X ) deﬁned by 2 3 λ [E]=1 [E]+ E E + −1 − − ··· Proposition 7.6. (In the algebraic case, see [Bo-Se,^ Proposition ^ 12 and Lemme 19 c.]) ! ∗ (i) If Z is a smooth submanifold of X, then for all x in G(Z), iZ iZ∗x = x . λ−1[NZ/X ]. CHERN CLASSES IN DELIGNE COHOMOLOGY FOR COHERENT ANALYTIC SHEAVES 45

(ii) Consider the blowup X of X along a smooth submanifold Y, and let E be the exceptional divisor as shown in the following diagram:

e j E / X

q p e Y / X i Let F be the excess conormal bundle on E deﬁned by the exact sequence 0 / F / q ∗N ∗ / N ∗ / 0. Y/X E/X e ! ! Then for all x in G(Y ), p i !x = j ! q x . λ−1F . The assertion (ii) of Proposition 7.6 is the excess formula in K-theory. Proof. (i) We write i ! i x = i ! x [i ] = x.i ! [i ]. Now Tor (i ,i )= i N ∗ ([Bo-Se, Z Z∗ Z Z Z∗OZ Z Z∗OZ i Z∗OZ Z Z/X Proposition 12]) and we are done. V (ii) The equality p ! i = j (λ [F ]) is proved in [Bo-Se, Lemme 19.c]. Thus ∗OY ! −1 p !i x = p ! x [i ∗ ] = q !x p ![i ] = q !x j λ [F ] = j q !x . λ [F ] . ! Y Y Y ∗ Y Y ! −1 ∗ −1 O e O e We will now need a statement similar to Proposition 7.5, with the hypothesis that Z is a divisor with simple normal crossing. Let f : X / Y be a surjective holomorphic map, D be a smooth hypersurface of Y , and D = f ∗(D). We suppose that D is a divisor of X with simple normal crossing. Let D , 1 k N, be the branches k ≤ ≤ red e of D . We denote by f the restricted map f : Dk / D. e k e Proposition 7.7. For all elements u in G (X) such that = u + + u and for all y in G(D), k D D 1 N e ek e O e ··· N f ! i y = f ! (y) u D! k D k ek k=1 X 7.4. Analytic K-theory with support in a divisor with simple normal crossing. Let X be a smooth complex compact manifold and D be a reduced divisor with simple normal crossing. The branches of D will be denoted by D1,...,DN and Dij = Di Dj . By a d´evissage argument, the n ∩ canonical map from i=1 GDi (X) to GD(X) is surjective. If x belongs to GDij (X), then the element (x, x) of G (X) G (X) is in the kernel of this map. We will show that this kernel is generated by − Di ⊕LDj these elements: Proposition 7.8. There is an exact sequence:

where the first map is given by α / (α, α) . Let us verify that this complex is exact. Consider the − ′ ′ ∗ map ψ : Z[ coh(D)] / G(D ) G(D ) G(D D ) defined by ψ( ) = [i ′ ] + [ ′ ]. Remark ⊕ N ∩ N F D F ID F ′ that ′ is a sheaf of D -modules, namely the sheaf of ideals of D D in D extended to D by zero. ID O N ∩ N N Let us show that ψ can be defined in K-theory. We consider an exact sequence 0 / / / /0 of coherent sheaves on D. Let us define F G H the sheaf by the exact sequence N ∗ ∗ ∗ 0 / / i ′ / i ′ / i ′ / 0. N D F D G D H ′ ∗ ∗ ∗ It is clear that is a sheaf of ′ -modules with support in D D , and [i ′ ] [i ′ ] [i ′ ]= [ ] N OD ∩ N D G − D F − D H − N in G(D′). Let us consider the following exact sequence of complexes: ∗ 0 / ′ / / i ′ / 0 ID F F D F

∗ 0 / ′ / / i ′ / 0 ID G G D G

∗ 0 / ′ / / i ′ / 0 ID H H D H Let be the ﬁrst complex, that is the ﬁrst column of the diagram above. It is a complex of D -modules. C O N If we denote by k( ), 0 k 2, the cohomology sheaves of , we have the long exact sequence H C ≤ ≤ C 0 / 0( ) / 0 / / 1( ) / 0 / 0 / 2( ) / 0. H C N H C H C 1 Since ( ) is a sheaf of D -modules, is also a sheaf of D -modules. Therefore, is a sheaf of H C O N N O N N D′∩D -modules. O N 0 1 2 In G(D ) we have [ ′ ] [ ′ ] + [ ′ ] = [ ( )] [ ( )] + [ ( )] = [ ]. Thus ψ( ) ψ( )+ N ID F − ID G ID H H C − H C H C − N F − G ψ( ) = ([ ], [ ]) = 0 in the quotient. H N − N ∗ If belongs to G(D), then [ ] = [i ′ ] + [ ′ ] in G(D). This means that π ψ = id. We consider F F D F ID F ◦ now in G(D′) and in G(D ). Then H K N ∗ ∗ ψ(π( , )) = ([iD′ ] + [iD′ ]) ([ D′ ] + [ D′ ]) = ([ ] + [ |D′∩D ]) [ D′∩D ]. H K H K ⊕ I H I K H K N ⊕ I N K

Remark that [ D′∩D ] = [ ] [ |D′∩D ] in G DN . Thus ([ ] + [ |D′∩D ]) ([ ] [ |D′∩D ]) = I N K K − K N H K N ⊕ K − K N [ ] [ ] modulo G(D′ D ), so that ψ π = id. This proves that ( ) is exact. H ⊕ K ∩ N ◦ ∗ We can now use the induction hypothesis with D′. We obtain the following diagram, where the columns as well as the ﬁrst line are exact: 0 0 O O

′ r ′ u G(D DN ) / G(D ) G(DN ) / G(D) / 0 ∩O ⊕O q p

s i

π The map i G(Di) / G(D) is clearly onto. Let α be an element of i G(Di) such that π(α) = 0. Then u(p(α)) = 0, so that there exists β such that r(β)= p(α). There exists γ such that q(γ)= β. Then L L p(α s(γ)) = p(α) r(q(γ)) = 0. So there exists δ such that α = s(γ)+ t(δ). It follows that α is in the − − image of i

References

[An-Ga] V. Ancona, B. Gaveau, Differential Forms on Singular Varieties, Chapman and Hall/CRC, (2006). [At] M. F. Atiyah, Complex analytic connections in fiber bundles, Trans. Amer. Math. Soc., 85, (1957), 181–207. [At-Hi] M. F. Atiyah, F. Hirzebruch, The Riemann-Roch theorem for Analytic embeddings, Topology, 1, (1962), 151–166. [Bei] A. A. Beilinson, Coherent sheaves on Pn and problems in linear algebra, Funkt. Analiz. Appl., 12(3), (1978), 68–69. [Ber] A. Berthomieu, Proof of Nadel’s conjecture and direct image for relative K-theory, Bull. Soc. math. France, 130-2, (2002), 253–307. [Bl] S. Bloch, Semi-regularity and de Rham cohomology, Invent. Math., 17, (1972), 51–66. [Bo-Se] A. Borel, J.-P. Serre, Le théorème de Riemann-Roch (d’après Grothendieck), Bull. Soc. Math. France, 86, (1958), 97–136. [Bott] R. Bott, Lectures on characteristic classes and foliations, in Lectures on Algebraic and Differential Topology, Lectures Notes in Math, 279, Springer, (1972). [Br] J.-L. Brylinski, Comparison of the Beilinson-Chern classes with the Chern-Cheeger-Simons classes, in Ad- vances in Geometry, Prog. Nat., 172, Birkhäuser, Boston, (1999), 95–105. [Ch-Si] J. Cheeger, J. Simons, Differential characters and geometric invariants, in Geometry and Topology (College Park 83-84), Lecture Notes in Math., 1167, Springer, (1985), 50–80. [Ei] D. Eisenbud, Commutative Algebra, Springer, (1995). [Es] H. Esnault, Characteristic classes of flat bundles, Topology, 27-3, (1988), 323–352. [Es-Vi] H. Esnault, E. Viehweg, Deligne-Beilinson Cohomology, in Beilinson’s conjecture on special values of L- functions (Rapoport, Schappacher & Schneider, eds.) Perspec. Math., Vol. 4, Academic Press, (1988), 43–91. [EZZ] F. El Zein, S. Zucker, Extendability of normal functions associated to algebraic cycles, in Topics in Tran- scendental Algebraic Geometry, Annals of Math. Studies, 106, Princeton University Press, (1984), 269–288. [Fe] H. Federer, Geometric Measure Theory, Springer, (1969). [Fu] W. Fulton, Intersection Theory, Springer, (1984). [Ga] P. Gajer, Geometry of Deligne cohomology, Inv. Math., Springer 127, (1997), 155–207. [Gi-So] H. Gillet, C. Soule,´ Arithmetic Chow groups and differential characters, in Algebraic K-theory: Connections with Geometry and Topology, Ed. J. F. Jardine, V. P. Snaith, Kluwer, (1989), 29–68. [Gra] H. Grauert, On Levi’s problem and the imbedding of real-analytic manifolds, Ann. Math., Princeton, 68, (1958), 460–472. [Gra-Re] H. Grauert, R. Remmert, Coherent Analytic Sheaves, Springer, (1984). [Gre] H. I. Green, Chern classes for coherent sheaves, Ph.D. Thesis, University of Warwick, January 1980. [Gri-Ha] P. Griffiths, J. Harris Principles of Algebraic Geometry, John Wiley and Sons, (1978). [Gro1] A. Grothendieck, La théorie des classes de Chern, Bull. Soc. Math. France, 86, (1958), 137–154. [Gro2] A. Grothendieck, Eléments´ de géométrie algébrique III, étude cohomologique des faisceaux cohérents, Pub- lications Mathématiques de l’I.H.E.S.´ , 11, (1961). [Gro3] A. Grothendieck, On the de Rham cohomology of algebraic varieties, Publications Mathématiques de l’I.H.E.S.´ , 29, (1966) 95–103. [EGA IV] A. Grothendieck, J. Dieudonne,´ Etude´ locale des schémas et des morphismes de schémas, Publications Mathématiques de l’I.H.E.S.,´ (1967). [Gr-Ra] L. Gruson, M. Raynaud, Critères de platitude et de projectivité, Inv. Math., 13, (1971), 1–89. [Hiro1] H. Hironaka, Resolution of singularities of an algebraic variety over a field of characteristic zero, Ann. of Math., 79, (1964), 109–326. [Hiro2] H. Hironaka, Flattening Theorem in Complex-Analytic Geometry, Amer. J. of Math., 97-2, (1975), 503–547. [Hirz] F. Hirzebruch, Topological methods in algebraic geometry, Springer, (1966). [Ill] L. Illusie, Complexe cotangent et déformations I, Lecture Notes in Math., Springer, 239, (1971). 48 JULIEN GRIVAUX

[Ki] J. King, The currents defined by analytic varieties, Acta Math., 127, (1971), 185–220. [Kn-Mu] F. F. Knudsen, D. Mumford, The projectivity of moduli spaces of stable curves I: Preliminaries on ‘det’ and ‘Div’, Math. Scand., 39, (1976), 19–55. [LMS] A. T. Lascu, D. Mumford, D. B. Scott, The self-intersection formula and the “formule-clef”, Proc. Cam- bridge Phil. Soc., 78, (1975), 117–123. [Na] A. Nadel, Invariants for holomorphic vector bundles, Math. Ann., 309, (1997), 37–52. [OB-To-To1] N. R. O’Brian, D. Toledo, Y. L. L. Tong, A trace map and characteristic classes for coherent sheaves, Amer. J. Math., 103, (1981), 253–271. [OB-To-To2] N. R. O’Brian, D. Toledo, Y. L. L. Tong, Grothendieck-Riemann-Roch for complex manifolds, Bull. Amer. Math. Soc., 5-2, (1981), 182–184. [OB-To-To3] N. R. O’Brian, D. Toledo, Y. L. L. Tong, A Grothendieck-Riemann-Roch formula for maps of complex manifolds, Math. Ann., 271, (1985), 493–526. [Qui] D. Quillen, Higher algebraic K-theory I, in Higher algebraic K-theory I, Lecture Notes in Math., 431, Springer, (1972). [ScVe] S. Schroer,¨ G. Vezzosi Existence of vector bundles and global resolutions for singular surfaces, Compos. Math., 140-3, (2004), 717–728. [Sch] H. W. Schuster, Locally free resolutions of coherent sheaves on surfaces, J. Reine Angew. Math., 337, (1982), 159–165. 1 1 [SGA 4 2 ] P. Deligne, Séminaire de Géométrie Algébrique du Bois Marie 1964/65 SGA 4 2 , Cohomologie étale, Lecture Notes in Math., 569, Springer, (1976). [SGA 5] A. Grothendieck, Séminaire de Géométrie Algébrique du Bois Marie 1965/66, SGA 5, Cohomologie l-adique et fonctions L, Lecture Notes in Math., 589, Springer, (1977). [SGA 6] P. Berthelot, A Grothendieck, L. Illusie, Séminaire de Géométrie Algébrique du Bois Marie 1966/67, SGA 6, Théorie des Intersections et Théorème de Riemann-Roch, Lecture Notes in Math., 225, Springer, (1971). [To-To] D. Toledo, Y. L. L. Tong, Green’theory of Chern classes and the Riemann-Roch formula, Contemp. Math., 58-I, (1986), 261–275. [Vo1] C. Voisin, Théorie de Hodge et géométrie algébrique complexe, SociétéMathématique de France, (2002). [Vo2] C. Voisin, A counterexample to the Hodge conjecture extended to Kähler varieties, Int. Math. Res. Notices, 20, (2002), 1057–1075. [Zu] S. Zucker, The Cheeger-Simons invariant as a Chern class, in Algebraic Analysis, Geometry and Number Theory (Baltimore 1988), Johns Hopkins Univ. Press, Baltimore, (1989), 397–417.

Institut de Mathematiques´ de Jussieu, UMR 7586, Case 247, Universite´ Pierre et Marie Curie, 4, place Jussieu, F-75252 Paris Cedex 05, France E-mail address: [email protected]