Reflexive modules on normal Gorenstein Stein surfaces, their deformations and moduli
aa r X i v : . [ m a t h . AG ] A p r REFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES,THEIR DEFORMATIONS AND MODULI
JAVIER FERN ´ANDEZ DE BOBADILLA AND AGUST´IN ROMANO-VEL ´AZQUEZ
Abstract.
In this paper we generalize Artin-Verdier, Esnault and Wunram construction ofMcKay correspondence to arbitrary Gorenstein surface singularities. The key idea is the defi-nition and a systematic use of a degeneracy module, which is an enhancement of the first Chernclass construction via a degeneracy locus. We study also deformation and moduli questions.Among our main result we quote: a full classification of special reflexive MCM modules on nor-mal Gorenstein surface singularities in terms of divisorial valuations centered at the singularity, afirst Chern class determination at an adequate resolution of singularities, construction of modulispaces of special reflexive modules, a complete classification of Gorenstein normal surface singu-larities in representation types, and an study on the deformation theory of MCM modules andits interaction with their pullbacks at resolutions. For the proof of these theorems it is crucial toestablish several isomorphisms between different deformation functors, that we expect that willbe useful in further work as well.
Contents
1. Introduction 22. Description of results 42.1. Setting and notation 42.2. The degeneracy module correspondences 52.3. The structure of reflexive modules via the degeneracy module correspondences 62.4. The classification of reflexive modules 82.5. Deformations and families 82.6. The correspondences as isomorphisms of deformation functors 92.7. Moduli spaces of reflexive modules, Cohen-Macaulay representation types 113. Reflexive modules and full sheaves 123.1. Cohen-Macaulay modules and reflexive modules 123.2. Full sheaves 134. Enhancing the Chern class and the degeneracy modules correspondences 144.1. Degeneracy modules of vector bundles 144.2. Cohen-Macaulay modules of dimension 1 17
Date : 13-3-2019.2010
Mathematics Subject Classification.
Primary: 13C14, 13H10, 14E16, 32S25.The first author is partially supported by IAS and by ERCEA 615655 NMST Consolidator Grant, MINECO bythe project reference MTM2016-76868-C2-1-P (UCM), by the Basque Government through the BERC 2018-2021program and Gobierno Vasco Grant IT1094-16, by the Spanish Ministry of Science, Innovation and Universities:BCAM Severo Ochoa accreditation SEV-2017-0718 and by Bolsa Pesquisador Visitante Especial (PVE) - Ciencias semFronteiras/CNPq Project number: 401947/2013-0. The second author is partially supported by IAS and by ERCEA615655 NMST Consolidator Grant, Bolsa Pesquisador Visitante Especial (PVE) - Ciencias sem Fronteiras/CNPqProject number: 401947/2013-0, CONACYT 253506 and 286447.
Introduction
McKay correspondence [27] is a bijection between the irreducible representations of a finitesubgroup of SL(2 , C ) and the irreducible components of the exceptional divisor of the minimalresolution of the associated quotient surface singularity, such a bijection extends to an isomorphismof the McKay quiver (associated to the structure of representations with respect to direct sumsand tensor products), and the dual graph of the exceptional divisor of the minimal resolution ofsingularities. McKay noticed the correspondence by a case by case study using the classification offinite subgroups of SL(2 , C ). After its discovery by McKay, a conceptual geometric understanding ofthe correspondence was achieved by a series of papers by Gonzalez-Springberg and Verdier [12], byArtin and Verdier [2] for the correspondence at the level of vertices and by Esnault and Kn¨orrer [10]at the level of edges.At the level of vertices the correspondence can be summarized as follows: let π : ˜ X → X be theminimal resolution of the quotient singularity. To an irreducible representation ρ one associates anindecomposable reflexive O X -module M . The module π ∗ M/ Torsion was proved to be locally free,and its first Chern class c ( π ∗ M/ Torsion) is the Poincar´e dual of a curvette hitting transverselyan unique irreducible component of the exceptional divisor. Such a component is the image ofthe representation ρ under McKay correspondence. Artin and Verdier proved that the first Chernclass determines the module M along with the representation ρ . Conversely, for any irreducible EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 3 component of the exceptional divisor there is a representation and a module realizing it. Wedo not spell the correspondence at the level of edges since the main concern of this paper is awide generalization of the results of Artin and Verdier at the level of vertices, and the subsequentcontributions of Esnault, Wunram and Kahn which we describe below.Esnault [9] improved Artin and Verdier correspondence by working on arbitrary rational surfacesingularity. She discovered that already for quotient singularities by finite subgroups of GL(2 , C )not included in SL(2 , C ) the pair given by the first Chern class c ( π ∗ M/ Torsion) and the rank of themodule is not enough to determine the reflexive module. A satisfactory McKay correspondence forarbitrary rational surface singularities was provided by Wunram [37]. For this he defined a reflexivemodule to be special if we have the vanishing R π ∗ ( π ∗ M ) ∨ = 0 (where ( ) ∨ denotes the dual withrespect to the structure sheaf). He proved that the first Chern class construction of Artin andVerdier defines a bijection between: a) the set of special indecomposable reflexive O X -modules andb) the set of irreducible components of the exceptional divisor of the minimal resolution. Moreoverthe first Chern class c ( π ∗ M/ Torsion) determines special reflexive O X -modules. The reader mayconsult the survey of Riemenschneider [33], for a more complete account of results described upto now, a summary of other approaches to McKay correspondence and a nice characterization ofspecial reflexive modules.Beyond the case of rational singularities the only complete study of reflexive modules was pro-vided by Kahn [19] for the case of minimally elliptic singularities. He studied reflexive O X -modules M via the associated locally free O ˜ X -module M := ( π ∗ M ) ∨∨ restricted to a cycle supported at theexceptional divisor. In the simply elliptic case he provides a full classification of reflexive modulesbuilding on Atiyah’s classification of vector bundles on elliptic curves.On normal surface singularities, reflexive modules coincide with Maximal Cohen Macaulay(MCM) ones. A singularity has finite, tame or wild Cohen-Macaulay representation type if themaximal dimension of the families of indecomposable MCM modules is 0, 1 or unbounded respec-tively. A consequence of Kahn’s results is that simply elliptic singularities are of tame representationtype. His results were completed by Drodz, Greuel and Kashuba [6], who proved that cusp sin-gularities, and more generally any log-canonical surface singularity are of tame Cohen-Macaulayrepresentation type, and posed the conjecture that non-canonical surface singularities are of wildrepresentation type (we prove this conjecture in this paper for normal Gorenstein surface singular-ities).Due to a result of Eisenbud [7], in the hypersurface case MCM modules correspond to matrixfactorizations. Therefore any result proven for MCM modules have a translation into matrix fac-torizations. Using this equivalence, Kn¨orrer [24] and Buchweitz, Greuel, Schreyer [4] proved inarbitrary dimension that the isolated hypersurface singularities of finite Cohen-Macaulay represen-tation type are exactly the simple (ADE type) ones.Besides the results described above the knowledge on the classification of MCM modules on othersingularities available in the literature is quite limited.McKay correspondence admits generalizations and extensions in many directions (some of themhinted in Riemenschneider survey [33]). An important one is the study of Auslander and Reitencategories of MCM modules over a singularity. The reader may consult the book of Yoshino [38]for a rather complete account of known results. In this paper we focus in the classification of theobjects of this categories for arbitrary normal Gorenstein singularities, leaving the structure of thecategory ready for later work.Following the line of the first Chern class correspondence described above it is natural to askwhether there is a similar description of indecomposable MCM modules on other singularities, JAVIER FERN´ANDEZ DE BOBADILLA AND AGUST´IN ROMANO-VEL´AZQUEZ and on the other on the characterization of the irreducible components of the exceptional divisorin terms of reflexive modules. Once the singularities are not rational, reflexive modules come infamilies, and deformation and moduli problems are important. We obtain quite complete answersto these questions for the case of Gorenstein normal singularities, including: • a full classification of special reflexive MCM modules in terms of divisorial valuations cen-tered at the singularity, which can be seen as a generalization of McKay correspondence, • a first Chern class determination at an adequate resolution of singularities, • construction of moduli spaces of special reflexive modules, • a complete classification of Gorenstein normal surface singularities in Cohen-Macaulay rep-resentation types, confirming Drodz, Greuel and Kashuba conjecture for this class, • a study on the deformation theory of MCM modules and its interaction with their pullbacksat resolutions,A detailed, non technical, description of the results of this paper is provided in the next section.Although the papers quoted above on McKay correspondence and classification of MCM moduleswork over singularities or complete local rings, we prove many of our results in the more generalsetting of reflexive modules over Stein normal surfaces (usually with Gorenstein singularities). Wefeel that this generalization will have interesting applications. For example it allows to apply ourresults to affine patches of reflexive modules on projective surfaces, opening the way to obtainapplications in the global projective case.It should be expected that our results have applications in other areas. MCM modules onGorenstein singularities have become recently even more important due to the equivalence withmatrix factorizations explained above. Following an idea of Kontsevich, the work of Kasputin andLi (see [23]), and Orlov (see [30], [31]) showed that matrix factorizations have applications to thestudy of Landau-Ginzburg models appearing in string theory, and to the study of Kontsevich’s ho-mological mirror symmetry. By Khovanov and Rozansky [21], [22], MCM modules have interestingapplications to link invariants. Matrix factorizations also have applications to cohomological fieldtheories [32]. Besides these applications, matrix factorizations have connections with representa-tion theory, Hodge theory and other topics; for more information see for example the referencesof the paper by Eisenbud and Peeva [8], where matrix factorizations are generalized to completeintersections. 2. Description of results
In this section we offer a detailed summary of the main results and techniques contained in thispaper. The reader should be able to get a picture of the paper by only reading this section. Alongthis section we cross-reference each of the results so the reader may jump to the appropriate partof the paper for more details. We describe each of the sections of the paper, but first we need toset the terminology.2.1.
Setting and notation.
Throughout this article, unless otherwise stated, we denote by X aStein normal surface. By ( X, x ) we denote either a complex analytic normal surface germ, or thespectrum of a normal complete C -algebra of dimension 2.As usual O X denotes the structure sheaf and O X,y is the local ring at a (non-necessarily closed)point y ∈ X . In this situation X has a dualizing sheaf ω X , and by abuse of notation we denotealso by ω X its stalk at x , which is the dualizing module of the ring O X,x . If X has Gorensteinsingularities then the dualizing sheaf is an invertible sheaf. This means that if ( X, x ) is Gorensteinthen the dualizing module coincides with O X,x . EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 5
Unless otherwise stated, we denote by π : ˜ X → X a resolution of singularities (in a few instancesit will denote a proper modification from a normal origin). The exceptional set is denoted by E := π − ( x ), and its decomposition into irreducible components is E = ∪ i E i . By a curvette wewill mean a (multi)-germ of a curve centered at the exceptional divisor.If ( X, x ) is Gorenstein, there is a Gorenstein 2-form Ω ˜ X which is meromorphic in ˜ X , and hasneither zeros nor poles in ˜ X \ E ; it is called the Gorenstein form . Let div (Ω ˜ X ) = P q i E i be thedivisor associated with the Gorenstein form. If X is a Stein surface with Gorenstein singularitiesthen there exist holomorphic 2-forms Ω ˜ X in ˜ X \ E so that div (Ω ˜ X ) = A + P q i E i , where A isdisjoint from E . The coefficients q i are independent of the form Ω ˜ X having these properties.Next, we extend to Stein normal surfaces some usual notions for singularities and define a newone: Definition 2.1.
Let π : ˜ X → X be a resolution either of a Stein normal surface with Gorensteinsingularities, or of a Gorenstein surface singularity. The canoninal cycle is defined by Z k := P i − q i E i , where the q i are the coefficients defined above.We say that ˜ X is small with respect to the Gorenstein form if Z k is greater than or equal to .The geometric genus of X is defined to be the dimension as a C -vector space of R π ∗ O ˜ X for anyresolution. Remark 2.2.
If a resolution π : ˜ X → X of a Stein normal surface with Gorenstein singularities issmall with respect to the Gorenstein form, then it is an isomorphism over the regular locus of X . We will only consider resolutions which are isomorphisms over the regular locus of X .Given X a normal Stein surface, or ( X, x ) a normal surface singularity germ, we denote by i : U → X the inclusion of U := X \ Sing ( X ) in X .2.2. The degeneracy module correspondences.
We study reflexive modules via resolutions.Let X be a Stein normal surface and π : ˜ X → X be a resolution. Let M be a reflexive O X -module.Its associated full O ˜ X -module is M := ( π ∗ M ) ∨∨ . A basic proposition of Kahn (Proposition 3.4)characterizes full O ˜ X -modules and establishes a bijection between reflexive O X -modules and full O ˜ X -modules.In Section 4 we improve and systematize Artin-Verdier’s first Chern class construction by prov-ing that there is a bijective correspondence between reflexive O X -modules (respectively full O ˜ X -modules) of rank r equipped with a system of r sufficiently generic sections, and Cohen Macaulay O X -modules of dimension 1 together with a system of r generators (respectively dimension 1 Cohen-Macaulay O ˜ X -modules with r global sections generating as a O ˜ X -module). These correspondencesare the main tool in establishing the main results of this paper; we describe them in detail rightbelow.Let π : Y → X denote a proper birational map (which could be the identity or a resolution).Let M be a O Y -module which is locally free of rank r and generated by global sections except ata finite subset of Y . The degeneracy O Y -module A (see Section 4.1) is defined to be the quotientof M by the submodule generated by the sections. When M is the sheaf of sections of a vectorbundle the support of A represents the first Chern class. We prove that A is Cohen-Macaulay ofdimension 1, with support A meeting the exceptional divisor at finitely many points, and such that A y ∼ = O A,y for generic y . We call such a module a generically reduced Cohen-Macaulay modules ofdimension
1. A set of nearly generic sections is a set of r sections so that its associated degeneracymodule is generically reduced Cohen-Macaulay of dimension 1. We prove in Section 4.2 that such JAVIER FERN´ANDEZ DE BOBADILLA AND AGUST´IN ROMANO-VEL´AZQUEZ modules admit a double inclusion O A ⊂ A ⊂ O ˜ A where ˜ A is the normalization of A . Based on this inclusion we associate to A a numerical invariant A called the set of orders to A which consists of a subset of N l and is similar to the semigroup ofa curve with l branches. As in the case of curve semigroups the set A has a minimal conductor element cond ( A ).When Y = ˜ X is a resolution of singularities, for any pair ( A , ( ψ , ..., ψ r )) we define the Contain-ment Condition (see Definition 4.17) which measures the interaction of ( A , ( ψ , ..., ψ r )) with thecanonical sheaf ω ˜ X . The following theorem is a precise statement of the correspondences announcedabove (see Theorems 4.15 and 4.18 in the main body of the paper) Theorem 2.3.
Let π : ˜ X → X be a resolution of a normal Stein surface. • If X has Gorenstein singularities there is a bijection between a) the set of pairs ( M, ( φ , ..., φ r )) of reflexive O X -modules with a set of nearly generic sections and b) the set of pairs ( C , ( ψ , ..., ψ r )) of generically reduced Cohen-Macaulay modules of dimension with r generators. If thesections are generic the module M has free factors if and only if the system of generators ( ψ , ..., ψ r ) is not minimal. • There is a bijection between a) the set of pairs ( M , ( φ , ..., φ r )) of full O ˜ X -modules with aset of nearly generic sections and b) the set of pairs ( A , ( ψ , ..., ψ r )) of generically reducedCohen-Macaulay modules of dimension with r generators as a O ˜ X -module satisfying theContainment Condition. The key to proof of the theorem consists mainly of a cohomological analysis for which the previousstudy on the structure of the degeneracy modules plays a central role. As we will see later, thedegeneracy modules are the crucial new ingredient that allows us to extend McKay correspondencetechniques to general surface singularities.In Propositions 4.34 and 4.36 we investigate the relation between the correspondences at X andat a resolution, and the relation of the correspondences at different resolutions.It is not easy to handle the Containment Condition directly. In Section 4.4.3 we introducea numerical condition for ( A , ( ψ , ..., ψ r )), called the Valuative Condition which requires the setof orders A to be contained in another subset of N l , called the Canonical Set of Orders , whichcodifies the interaction of ( A , ( ψ , ..., ψ r )) with the canonical sheaf ω ˜ X (see Definition 4.26). InProposition 4.31 we show that the Containment Condition implies the Valuative Condition. InProposition 4.33 we prove that the converse is true in sufficiently many cases, that cover all theneeds of this paper.2.3. The structure of reflexive modules via the degeneracy module correspondences.
Each reflexive module over a singular surface has a favorite resolution of singularities (for rationalsingularities it coincides with the minimal resolution). Let M be a reflexive O X -module of rank r over a normal Stein surface X . The minimal adapted resolution of M is the minimal resolutionof singularities π : ˜ X → X such that the full O ˜ X -module M associated with M is generatedby global sections. In Proposition 5.1 we prove its existence and uniqueness. In Proposition 5.5we characterize numerically the minimal adapted resolution of a reflexive module in terms of theminimal conductor of the Canonical Set of Orders of the pair ( A , ( ψ , ..., ψ r )), obtained by applyingthe correspondence of Theorem 2.3 to the pair ( M , ( φ , ..., φ r )) formed by the full O ˜ X -moduleassociated with M and a system of generic sections. EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 7
Similar to Wunram’s generalization of McKay correspondence we obtain the most detailed resultsfor special reflexive modules. The right generalization of special module is provided in Definitions 3.7and 3.8. Let π : ˜ X → X be a resolution of a normal Stein surface, M be a reflexive O X -moduleof rank r and M its associated full O ˜ X -module. We say that M is a special full sheaf if wehave the equality dim C ( R π ∗ (cid:16) M ∨ (cid:17) ) = rp g . The specialty defect of M is the number d ( M ) :=dim C ( R π ∗ (cid:16) M ∨ (cid:17) ) − rp g . We prove that the specialty defect is non-negative. We say M to be a special reflexive module if for any resolution of singularities of X its associated full sheaf is special.We study the behavior of the specialty defect under dominant maps σ : ˜ X → ˜ X betweenresolutions, and various properties of special modules in Sections 5.1 and 5.2. The main results are:(1) the specialty defect of the full O ˜ X -module M associated to M is greater than or equalto the specialty defect of the full O ˜ X -module M associated to M . We have the equalityif ρ is an isomorphism over the locus where M is not generated by global sections (seeProposition 5.7).(2) Given a reflexive O X -module, if at the minimal adapted resolution the specialty defect of thefull sheaf associated to M vanishes, then it vanishes for any resolution (see Theorem 5.9).(3) Assume that the Stein normal surface X has Gorenstein singularities. Let ( C , ( ψ , ..., ψ r ))be the pair associated with ( M, ( φ , ..., φ r )) by the correspondence of Theorem 2.3. If M is a special O X -reflexive module over X and ( φ , ..., φ r ) are generic sections, then C isisomorphic to n ∗ O ˜ C , where n : ˜ C → C is the normalization of the support of C . Theconverse is true (see Proposition 5.13).(4) Assume now that ( X, x ) is a normal Gorenstein surface singularity and that M is specialwithout free factors. Let C be like in the previous property. There is a bijection betweenthe indecomposable direct summands of M and the irreducible components of C .Property (3) gives a geometric understanding of specialty under the degeneracy module correspon-dence, and also allows to produce full sheaves with prescribed Chern classes. If ( X, x ) is a germ, thenormalization map n : ˜ C → C is an arc at the singularity if C is irreducible. Hence Property (3) alsoestablishes a link between arc spaces and reflexive modules. Property (4) is important because itshows how the degeneracy module correspondence recovers the decomposition into indecomposablesin a very geometric way. Properties (3) and (4) are false for non-special modules.A crucial tool in the study of reflexive modules is the computation of the first cohomology of fullsheaves (see Theorem 6.1): Theorem 2.4.
Let X be a Stein normal surface with Gorenstein singularities. Let M be a reflexive O X -module of rank r . Let π : ˜ X → X be a small resolution with respect to the Gorenstein form, let Z k be the canonical cycle at ˜ X . Let M be the full O ˜ X -module associated to M . Let d ( M ) be thespecialty defect of M . Then we have the equality dim C ( R π ∗ M ) = rp g − [ c ( M )] · [ Z k ] + d ( M ) . The proof occupies the whole Section 6. An interesting Corollary is the fact that the minimaladapted resolution π : ˜ X → X of a special reflexive O X -module M over a Stein normal surfacewith Gorenstein singularities is characterized by the fact that dim C ( R π ∗ M ) = rp g , where M isthe associated full O ˜ X -module (see Corollary 6.2). This has the interesting consequence that thecycle representing first Chern class of M does not meet the support of the canonical cycle Z K . Italso shows that dim C ( R π ∗ M ) = rp g can be used as an invariant controlling the minimal adaptedresolution process. JAVIER FERN´ANDEZ DE BOBADILLA AND AGUST´IN ROMANO-VEL´AZQUEZ
The above tools allow us to prove some of the main results of the paper. The first is a determi-nation of special reflexive modules in terms of a first Chern class (Theorem 7.6):
Theorem 2.5.
Let X be a Stein normal surface with Gorenstein singularities. Let M be a special O X -module without free factors. Let π : ˜ X → X be the minimal resolution adapted to M , and M the full O ˜ X -module associated to M . The module M (and equivalently M ) is determined by its firstChern class in Pic ( ˜ X ) . The classification of reflexive modules.
In Section 7.1 we provide a combinatorial classi-fication of reflexive modules. Given a normal surface singularity (
X, x ) and a reflexive O X -module M , we define its associated graph G M to be the dual graph of the minimal good resolution dominat-ing the minimal adapted resolution to M , decorated adding as many arrows to each of its vertices v as the number c ( M )( E v ), where E v is the component of the exceptional divisor correspondingto v and c ( M ) is the first Chern class of the associated full M -module.Let ( A , ( ψ , ..., ψ r )) be the pair associated with ( M , ( φ , ..., φ r )) under the correspondence ofTheorem 2.3, where ( φ , ..., φ r ) are generic sections. Proposition 5.14 shows that the support of A is a disjoint union of as many smooth curvettes as arrows has G M , each of them meeting transverselythe irreducible component of the exceptional divisor corresponding to the vertex where the arrowis attached.In Theorem 7.2 we characterize combinatorially the graphs of special modules over Gorensteinsurface singularities. We prove that these are precisely the graphs such that(1) the graph is numerically Gorenstein.(2) if a vertex has genus 0, self intersection − irreducible divisor over x is the same that adivisorial valuation of the function field of X centered at x . An irreducible divisor over x appearsat a model π : ˜ X → X if its center at ˜ X is a divisor. Theorem 2.6.
Let ( X, x ) be a Gorenstein surface singularity. Then there exists a bijection betweenthe following sets:(1) The set of special indecomposable reflexive O X -modules up to isomorphism.(2) The set of irreducible divisors E over x , such at any resolution of X where E appears, theGorenstein form has neither zeros nor poles along E . This theorem specializes to classical Mckay correspondence in the case of rational double points.By taking direct sums one obtains a full classification of special reflexive modules over Gorensteinsurface singularities.2.5.
Deformations and families.
Having studied reflexive modules as individual objects we turnto deformations and moduli questions for the rest of the paper. Here we work as generally aspossible: we allow deformations of the underlying space when we deform reflexive modules; whenwe deform full sheaves we allow simultaneous deformation of the space and of the resolution.In Section 8.1 we define the relevant deformation functors, morphisms between them and establishthe existence of versal deformations. Let X be a Stein normal surface and M be a reflexive O X -module. EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 9
A deformation of (
X, M ) over a base (
S, s ) consists of a flat deformation X of the X over ( S, s )together with a O X -module which is flat over S and whose fibre over s is isomorphic to M (seeDefinition 8.2). Since reflexivity is an open property in flat families this is an adequate notion ofdeformations of reflexive sheaves. This definition leads to a deformation functor Def X , M . Thefunctor of deformations fixing the base space X is a sub-functor.On the other hand fullness is not an open property in flat families. Therefore the right definitionof the deformation functor of full sheaves needs a cohomological condition: let X and M as before.Let π : ˜ X → X be a resolution and M the full O ˜ X module associated with M . A deformation of( ˜ X, X, M ) is formed by a very weak simultaneous resolution Π : ˜ X → X of a flat deformation X of X over S , and a O ˜ X -module M which is flat over S , whose specialization over s is isomorphicto M and such that R Π ∗ M is flat over S (see Definition 8.7). This leads to a deformationfunctor FullDef ˜X , X , M . One has subfunctors fixing the underlying space and/or the resolution. InProposition 8.8 it is shown that fullness is an open property in families defined as above; the proofuses the flatness of R Π ∗ M in a crucial way. This shows that our definition is the correct notionof deformation within the category of full sheaves.In Proposition 8.10 we show that the push-forward functor Π ∗ defines a natural transformationfrom FullDef ˜X , X , M to Def X , M . Since Kahn’s result (Proposition 3.4) establishes a bijection be-tween reflexive O X -modules and full O ˜ X -modules, and this bijection is via the push-forward functor,one could naively expect that this Π ∗ is an isomorphism of functors. This is not the case as we willsee below. Analyzing the functors FullDef ˜X , X , M and Def X , M , and the transformation Π ∗ directlyseems a difficult task. The extension to deformation functor isomorphisms of the correspondencesof Theorem 2.3 is the crucial tool in our subsequent analysis.2.6. The correspondences as isomorphisms of deformation functors.
First we enrich thefunctors
FullDef ˜X , X , M and Def X , M and define deformation functors FullDef ( φ ,...,φ r ) ˜X , X , M and Def ( φ ,...,φ r ) X , M .Given X and M as above, and ( φ , ..., φ r ) a set of r = rank ( M ) nearly generic sections, the deforma-tion functor Def ( φ ,...,φ r ) X , M associates to ( S, s ) a deformation ( X , M ) in Def X , M ( S, s ) along with a setof sections ( φ , ..., φ r ) extending ( φ , ..., φ r ) (see Definition 8.14). The definition of FullDef ( φ ,...,φ r ) ˜X , X , M is similar (see Definition 8.18). There are obvious forgetful functors from Def ( φ ,...,φ r ) X , M to Def X , M ,and from FullDef ( φ ,...,φ r ) ˜X , X , M to FullDef ˜X , X , M .In order to be able to prove an analog of Theorem 2.3 for deformations we need deformationfunctors of generically reduced 1-dimensional Cohen-Macaulay modules together with sets of gener-ators. The relevant definitions are the following (see Definitions 8.14 and 8.18): let X be as above,let ( C , ( ψ , ..., ψ r )) be a generically reduced 1-dimensional Cohen-Macaulay O X -module, togetherwith a system of generators as a O X -module. A deformation of ( X, C , ( ψ , ..., ψ r )) over a germ( S, s ) consists of a flat deformation X of the space X over ( S, s ), a O X -module C which is flat over S and specializes to C over s , and a set of sections ( ψ , ..., ψ r ) of C which specialize to ( ψ , ..., ψ r )over s . The resulting deformation functor is denoted by Def ( ψ ,...,ψ r ) X , C .Like in the case of deformations of full sheaves the straightforward generalization of this functorto the case of resolutions does not work; we need to add a further condition to the families, that ina certain sense is the analog of the flatness condition of R Π ∗ M in the case of deformations of fullsheaves. Let π : ˜ X → X be a resolution of singularities and A be a 1-dimensional generically reducedCohen-Macaulay O ˜ X -module whose support meets the exceptional divisor at finitely many points.Let ( ψ , ..., ψ r ) be a set of global sections of A generating it as a O ˜ X -module. A specialty defect constant deformation of ( A , ( ψ , ..., ψ r )) over a germ ( S, s ) consists of a very weak simultaneousresolution Π : ˜
X → X of a flat deformation of X over ( S, s ), a O ˜ X -module A which is flat over S and specializes to A over s , and a set of sections ( ψ , ..., ψ r ) which specialize to ( ψ , ..., ψ r ) over s ,and which are so that the cokernel D of the natural mapping Π ∗ O r ˜ X → Π ∗ A induced by the sectionsis flat over S . The resulting deformation functor is denoted by SDCDef ( ψ ,...,ψ r ) ˜X , X , A . The functor Π ∗ defines a natural transformation from SDCDef ( ψ ,...,ψ r ) ˜X , X , A to Def ( ψ ,...,ψ r ) X , C , where C is the submoduleof π ∗ A generated by ( ψ , ..., ψ r ).The following is our main tool in studying deformation and moduli functors (see Theorems 8.17and 8.21 for a more precise version). Theorem 2.7.
Let π : ˜ X → X be a resolution of a Stein normal surface with Gorenstein singular-ities. Let M be a reflexive O X -module of rank r and M be the associated full O ˜ X -module.(1) Let ( φ , ..., φ r ) be nearly generic sections of M , let ( C , ( ψ , ..., ψ r )) be the pair associatedwith ( M, ( φ , ..., φ r )) under the correspondence of Theorem 2.3. There is an isomorphismbetween the functors Def X , M ( φ ,...,φ r ) and Def ( ψ ,...,ψ r ) X , C .(2) Let ( φ , ..., φ r ) be nearly generic sections of M , let ( A , ( ψ , ..., ψ r )) be the pair associatedwith ( M , ( φ , ..., φ r )) under the correspondence of Theorem 2.3. There is an isomorphismbetween the functors FullDef ˜X , X , M ( φ ,...,φ r ) and SDCDef ( ψ ,...,ψ r ) ˜X , X , A . The proof of this Theorem gets quite technical and occupies several pages of the paper, but itssubsequent applications makes the effort worthwhile.In Propositions 8.25 and 8.26 we explain the behavior of the correspondences of Theorem 2.7under the functor Π ∗ .An important corollary of this theorem is that the specialty defect remains constant in a defor-mation of FullDef ˜X , X , M (see Corollary 8.24). In Example 10.6 we give an example of deformationof a special reflexive module such that the generic member of the family is not special. As a con-sequence we produce a deformation which does not lift to the minimal resolution. This shows thatΠ ∗ does not induce an isomorphism of functors from FullDef ˜X , X , M to Def X , M .In the previous example, the reason for which the natural transformation of functors Π ∗ : FullDef ˜X , X , M → Def X , M is not an isomorphism is that, in general, deformations in Def X , M ( S, s )do not lift to ˜ X . This motivates Section 10, in which we study systematically the liftability problemfor families using the correspondences of Theorem 2.7.Let ( X , M ) be an element in Def X , M ( S, s ). Let ( φ , ..., φ r ) be r = rank ( M ) sections of M which specialize to nearly generic sections over s . Let ( X , C , ( ψ , ..., ψ r )) be the result of applyingthe correspondence of Theorem 2.7 to ( X , M , ( φ , ..., φ r )). Let C be the support of C . We say that C lifts to ˜ X if the fibre over s of the strict transform of C by Π coincides with the strict transformby π of the fibre of C over s . This notion is introduced at Definition 10.1, where also the notion ofliftability for ( X , C , ( ψ , ..., ψ r )) is defined.Assume that ( S, s ) is a reduced base. In Proposition 10.4 we prove that there is a deformationin
FullDef ˜X , X , M ( S, s ) which transforms under Π ∗ to ( X , M ) if and only if ( X , C , ( ψ , ..., ψ r )) liftsto ˜ X according with Definition 10.1. Moreover this implies that the support C lifts to ˜ X . InExample 10.5 we exhibit a deformation in FullDef ˜X , X , M ( S, s ) that does not lift to ˜ X because thesupport C does not lift. In Example 10.6 we find a deformation in FullDef ˜X , X , M ( S, s ) that doesnot lift to ˜ X because ( X , C , ( ψ , ..., ψ r )) does not lift to ˜ X , even if the support C does lift. In EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 11
Proposition 10.7 we prove that for any deformation over a reduced base (
S, s ) there exists a Zariskidense open subset on (
S, s ) over which the deformation lifts to a full family.For our later applications we need a sufficient condition for lifting of deformations in
FullDef ˜X , X , M ( S, s )that is easier to handle than the liftability of ( X , C , ( ψ , ..., ψ r )) predicted in Proposition 10.4. Thisis worked out in Section 10.1, where it is proved that under certain conditions the liftability of thesupport C is enough.In Definition 10.8 we introduce the notion of simultaneously normalizable deformations of re-flexive modules: let X be a normal Stein surface, let X be a deformation of X over a reduced base( S, s ), let M be a reflexive O X -module of rank r . A deformation ( X , M ) of ( X, M ) over a reducedbase (
S, s ) is said to be simultaneously normalizable if the degeneracy locus C of M for a genericsystem of r sections admits a simultaneous normalization over S . Using this definition we prove(see Theorem 10.10): Theorem 2.8.
Let X be a normal Gorenstein surface singularity. Let X be a deformation of X over a normal base ( S, s ) . Let Π : ˜
X → X be a very weak simultaneous resolution. Let M be areflexive O X -module and ( X , M ) be a simultaneously normalizable deformation of ( X, M ) over thebase ( S, s ) , so that for each s ′ ∈ S the module M | s ′ is special. If the support of the degeneracymodule of M for a generic system of sections is liftable to ˜ X , then the family ( X , M , ι ) lifts to afull family on ˜ X . Moduli spaces of reflexive modules, Cohen-Macaulay representation types.
Nowwe describe two applications of the machinery developed up to now.The first application appears in Section 9 and confirms a conjecture of Drodz, Greuel andKashuba [6] and, together with previous work in [2] and [6] completes the classification of Goren-stein normal surface singularities in Cohen-Macaulay representation types. Let us recall that asurface singularity (
X, x ) is of finite, tame or wild Cohen-Macaulay representation type if thereare at most finite, 1-dimensional or unbounded dimensional families of indecomposable MaximalCohen-Macaulay O X,x -modules respectively. Here we prove (see Theorem 9.3).
Theorem 2.9.
A Gorenstein surface singularity is of finite Cohen-Macaulay representation type ifand only if it is a rational double point. Gorenstein surface singularities of tame Cohen-Macaulayrepresentation type are precisely the log-canonical ones. The remaining Gorenstein surface singu-larities are of wild Cohen-Macaulay representation type.
The second application is the construction of fine moduli spaces of special modules without freefactors of prescribed graph and rank on Gorenstein normal surface singularities. This enhancesthe classification Theorem 2.6. It is provided in Section 11. Let G be the graph of a specialreflexive O X -module on a Gorenstein normal surface singularity X . Ler r be a positive integer.In Definition 11.1 a moduli functor Mod r G is defined in a similar way as the deformation functorsabove. It parametrizes flat families of special reflexive modules without free factors of rank r andgraph G , over normal base spaces. The main result is Theorem 11.8: Theorem 2.10.
The functor
Mod r G is represented by an algebraic variety. Moreover in its proof we see that the variety representing the functor has a very nice geometricdescription: it parametrizes sequences of infinitely near points to x in the singularity X with agiven combinatorial type. Reflexive modules and full sheaves
See [3], [15] and [29] as basic references on dualizing sheaves, modules and normal surface singu-larities.3.1.
Cohen-Macaulay modules and reflexive modules.
Let X be a normal surface along thissection. Let H om O X ( • , • ) and by E xt i O X ( • , • ) the sheaf theoretic Hom and
Ext functors. Thedual of a O X -module M is M ∨ := H om O X ( M, O X ). The ω X -dual is H om O X ( M, ω X ). A module O X -module M is called reflexive if the natural homomorphism from M to M ∨∨ is an isomorphism.It is called ω X -reflexive if the natural map M → H om O X ( H om O X ( M, ω X ) , ω X ) is an isomorphism.A O X -module M is called Cohen-Macaulay if the depth of each of its stalks M y is equal to thedimension of the module. If the depth of M y is equal to the dimension of O X,y , then the module M y is called maximal Cohen-Macaulay ; this definitions extend to sheaves if we ask that they holdfor every stalk. If M x is Cohen-Macaulay then M is Cohen-Macaulay at a neighborhood of x . Amodule is indecomposable if it can not be written as a direct sum of two non trivial submodules.By [38, Proposition 1.5] and [17, Section 1] some basic properties of maximal Cohen-Macaulaymodules are:(1) If O X,y is a regular local ring, then any maximal Cohen-Macaulay module over it is free.(2) If O X,y is a reduced local ring of dimension one, then an O X,y -module M is maximalCohen-Macaulay if and only if it is torsion free, that is, when the natural homomorphism M → M ∨∨ is a monomorphism.(3) If O X,y is normal of dimension two, then an O X,y -module M is maximal Cohen-Macaulayif and only if it is reflexive.(4) A consequence of the previous properties is that reflexive sheaves over regular rings O X,y of dimension at most 2 are free.(5) Let M be a O X -module. Then M x is a reflexive O X,x -module if and only if the adjunctionmorphism M → i ∗ i ∗ M is an isomorphism.(6) If a O X -module is reflexive at x , then it is reflexive at an open neighbourhood of x in X .The canonical module and Cohen-Macaulay modules have the following properties, which are aspecial case of [3, Theorem 3.3.10]. Theorem 3.1.
Let X be a normal surface. For t = 0 , , and all Cohen-Macaulay O X -modules M of dimension t one has(1) E xt − t O X ( M, ω X ) is Cohen-Macaulay of dimension t,(2) E xt i O X ( M, ω X ) = 0 for all i = 2 − t ,(3) there exists an isomorphism M → E xt − t O X (cid:0) E xt − t O X ( M, ω X ) , ω X (cid:1) which in the case t = 2 isjust the natural homomorphism from M into the ω X -bidual of M . A consequence of the previous Theorem and Property (3) above is that ω X -reflexivity is equiv-alent to reflexivity.The following proposition will be useful: Proposition 3.2.
Let X be a normal surface. Let A be a -dimensional O X -module. Then the O X -module E xt O X ( A , ω X ) is Cohen-Macaulay of dimension . Proof.
Consider the exact sequence 0 → B → A → A ′ →
0, where B is the submodule ofelements with support at x . No element of A ′ has support at x , and hence A ′ is Cohen-Macaulayof dimension 1. Applying E xt O X ( • , ω X ) and considering the associated exact sequence we obtain EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 13 the isomorphism E xt O X ( A , ω X ) ∼ = E xt O X ( A ′ , ω X ), which implies the result by the first assertion ofTheorem 3.1. (cid:3) Full sheaves.
We are interested in studying the reflexive modules on Stein normal surfaces,or in normal surface singularities, via a resolution. For this we will use the notion of full sheaves,introduced by Esnault [9] for rational surface singularities and generalized by Kahn [19]. Along thissection let X be a Stein normal surface and π : ˜ X → X be a resolution. All results of this sectionare valid replacing x by ( X, x ), which is either a germ of normal surface singularity or the spectrumof a normal complete C -algebra of dimension 2. Definition 3.3 ([19, Definition 1.1]) . A O ˜ X -module M is called full if there is a reflexive O X -module M such that M ∼ = ( π ∗ M ) ∨∨ . A O ˜ X -module M is generically generated by global sections if it is generated by global sections except in a finite set. Proposition 3.4 ([19, Proposition 1.2]) . A locally free sheaf M on ˜ X is full if and only if(1) M is generically generated by global sections.(2) The natural map H E ( ˜ X, M ) → H ( ˜ X, M ) is injective.If M is the full sheaf associated to M , then π ∗ M = M . Proof.
Kahn’s proof is for singularities. The proof for Stein normal surfaces is the same if oneuses that a O X -module on a Stein space is generated by global sections. The last assertion isimplicit in Kahn’s proof, and it gives us a natural bijection between reflexive O X -modules and full O ˜ X -modules. (cid:3) The following two lemmas that will be used later.
Lemma 3.5. If M is a full O ˜ X -module, then R π ∗ ( M ⊗ ω ˜ X ) = 0 . Proof. If M is generated by global sections, Grauert-Riemenschneider Vanishing Theorem impliesthat R π ∗ ( M ⊗ ω ˜ X ) is equal to zero.If M is almost generated by global sections, consider M ′ the subsheaf of M generated byglobal sections, therefore we get the exact sequence 0 → M ′ → M → G →
0, with Supp( G ) zerodimensional. Applying the functor − ⊗ ω ˜ X to the previous exact sequence, we get the desiredvanishing via the long exact sequence of the the functor π ∗ − . (cid:3) Lemma 3.6. If M is a full sheaf, then π ∗ (cid:16) M ∨ (cid:17) = ( π ∗ M ) ∨ . Proof.
Consider the following cohomology exact sequence0 H E (cid:16) M ∨ (cid:17) H (cid:16) M ∨ (cid:17) H (cid:16) U, M ∨ (cid:17) H E (cid:16) M ∨ (cid:17) H (cid:16) M ∨ (cid:17) H (cid:16) U, M ∨ (cid:17) . . . Since M is locally free we have that H E (cid:16) M ∨ (cid:17) = 0, H E (cid:16) M ∨ (cid:17) ∼ = H ( M ⊗ ω ˜ X ), by Serreduality. By Lemma 3.5 we get H ( M ⊗ ω ˜ X ) = 0. Hence π ∗ ( M ∨ ) = H (cid:16) M ∨ (cid:17) ∼ = H (cid:16) U, M ∨ (cid:17) . Now denote by M := π ∗ M . Since M ∨ is reflexive we get the equalities M ∨ = i ∗ i ∗ (cid:16) M ∨ (cid:17) = i ∗ (cid:16) M ∨ | U (cid:17) = H (cid:16) U, M ∨ (cid:17) . Therefore we have the isomorphism M ∨ ∼ = π ∗ (cid:16) M ∨ (cid:17) . (cid:3) Another notion that will be important in this work is the concept of specialty. PreviouslyWunram [37] and Riemenschneider [33] defined a special full sheaf as a full sheaf which its dualhas the first cohomology group is equal to zero. Using this definition Wunram generalized McKaycorrespondence in the following sense: he proved that in the case of a rational surface singularityand taking the minimal resolution, there is a bijection between isomorphism classes of special fullsheaves and irreducible components of the exceptional divisor.Their specialty notion is adapted to the case of rational singularities. For us the definition ofspecial is as follows.
Definition 3.7.
A full O ˜ X -module M on ˜ X of rank r is called special if dim C ( R π ∗ (cid:16) M ∨ (cid:17) ) = rp g . The defect of specialty of M is the number dim C ( R π ∗ (cid:16) M ∨ (cid:17) ) − rp g . Notice that this definition is a generalization of the concept given by Wunram and Riemenschnei-der and both definitions coincide in the case of a rational singularity.Since the definition of being special depends on the resolution, we have a another related notion.
Definition 3.8.
Let M be a reflexive O X -module. We say that M is a special module if for anyresolution the full sheaf associated to M is special. Later it will become clear the importance of these concepts.4.
Enhancing the Chern class and the degeneracy modules correspondences
In [2] Artin and Verdier interpret geometrically McKay correspondence as follows: given anindecomposable reflexive O X -module, with X a rational double point, they assign to it the firstChern class of the bi-dual of its pull-back to the minimal resolution. It turns out that the Chern classdetermines the module and that it hits precisely the exceptional divisor that McKay correspondenceassociates to the module.If X is not a rational double point the fisrt Chern class does not determine the module [9]. In thissection we refine Artin-Verdier construction in the following sense. The first Chern class may beconstructed as the degeneracy locus of a set of as many generic sections as the rank of the module.Here we, to the same set of sections, we associate a degeneracy module , which is a Cohen-Macaulaymodule of dimension 1 that whose support is the degeneracy locus. This refined correspondence isone of the main tools in our study of reflexive modules.4.1. Degeneracy modules of vector bundles.
In this section we refine the construction of thefirst Chern class of a vector bundle as explained above. In order to avoid introducing more notationwe only work in the generality needed in this paper, but the construction apply to more situations.Let X be a normal Stein surface and π : ˜ X → X a proper birational map from a normal space˜ X ( not necessarily a resolution, for example π may be the identity map ). Let E := π − ( x ). Let M be a reflexive O ˜ X -module that is generically generated by global sections. Let S ⊂ E be the finitesubset which is the union of the singular locus Sing ( ˜ X ) and the locus where M is not generatedby global sections. Denote by M := π ∗ M the O X -module of global sections of M . EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 15
Suppose that rank( M ) = r and take φ , . . . , φ r generic sections. Consider the exact sequencegiven by the sections(1) 0 → O r ˜ X ( φ ,...,φ r ) −→ M → A ′ → . Definition 4.1.
Given M a reflexive O ˜ X -module of rank r as above and ( φ , . . . , φ r ) a set of r sections, the O ˜ X -module A ′ defined by the previous exact sequence is called the degeneracy module of M associated with the given sections. The sections are called weakly generic if the support ofthe degeneracy module is a proper closed subset. Proposition 4.2.
Given M a O ˜ X -module of rank r as above and a weakly generic set of r sections,the associated degeneracy module is Cohen-Macaulay. Proof.
Dualize the exact sequence (1) with respect to a dualizing sheaf ω ˜ X (which exists sincethe surface ˜ X is normal, and hence Cohen-Macaulay) to obtain(2) 0 → H om ˜ X ( M , ω ˜ X ) → ω r ˜ X → E xt O ˜ X ( A ′ , ω ˜ X ) → , and dualizing again with respect to ω ˜ X we get(3)0 → H om ˜ X ( ω ˜ X , ω ˜ X ) r → H om O ˜ X (cid:0) H om O ˜ X ( M , ω ˜ X ) , ω ˜ X (cid:1) → E xt O ˜ X (cid:16) E xt O ˜ X ( A ′ , ω ˜ X ) , ω ˜ X (cid:17) → . Theorem 3.1 applied to O ˜ X yields that the first term is isomorphic to O r ˜ X . Since M is reflexivewe have that the middle term is isomorphic to M . Functoriality of the double H om implies thenthat the first morphism of the last exact sequence coincides with the first morphism of the exactsequence (1). Therefore we get that A ′ is isomorphic to E xt O ˜ X (cid:16) E xt O ˜ X ( A ′ , ω ˜ X ) , ω ˜ X (cid:17) . Finally since A ′ has dimension one we conclude that A ′ is a Cohen-Macaulay O ˜ X -module of dimension one byProposition 3.2. (cid:3) The following Bertini-Type Proposition is a generalization of Lemma 1.2 of [2].
Proposition 4.3.
Let M a reflexive O ˜ X -module of rank r that is generically generated by globalsections as above and ( φ , . . . , φ r ) a set of r generic sections. Let C be the support of the degeneracymodule A ′ with reduced structure.(1) For any point x ∈ ˜ X not contained in S we have the isomorphism A ′ x ∼ = O C,x .(2) The support of C meets E in finitely many points. Let Z ⊂ ˜ X be any finite set disjointwith S , a sufficiently generic choice of the sections ( φ , . . . , φ r ) ensures that C is smoothoutside S , that C does not meet Z , and that E and C meet in a transversal way at thosemeeting points not contained in S . However C contains S and at these points it may meet E in a non-transversal way. Proof.
The restriction M| ˜ X \ S of the sheaf M to ˜ X \ S is locally free and generated by globalsections of M over ˜ X . Let ψ , . . . , ψ k be global sections of M such that they generate it over ˜ X \ S .Denote by E the vector bundle over ˜ X \ S whose sheaf of sections is M| ˜ X \ S and by ( ˜ X \ S ) × C k the trivial vector bundle of rank k over ˜ X \ S . The generating global sections ( ψ , ..., ψ k ) induce asurjective morphism of vector bundlesΨ : ( ˜ X \ S ) × C k → E, defined by Ψ( x, ( c , . . . , c k )) = P kj =1 ψ j ( x ) c j . Since holomorphic vector bundles over Stein spaces are trivial, and ˜ X \ S admits a finite Steincover, there exist a finite trivializing covering for E . Let U be an trivializing open set. Consider thelocal trivialization E | U → U × C r . In the open set U the global sections ψ , . . . , ψ k can be writtenas follows A = a a . . . a k ... ... ... ... a r a . . . a rk , where the entries of the i -th column are the coordinates of ψ i . Notice that the matrix A has entriesin O ˜ X ( U ).In the trivialization over U the restriction of the map Ψ is Ψ U : U × C k → U × C r , whereΨ U ( x, ( c , . . . , c k )) = ( x, A ( x )( c , . . . , c k ) ⊺ ).Now for each matrix B in Mat ( k × r, C ), we get sections φ , . . . , φ r of M by the formula( φ , . . . , φ r ) = ( ψ , . . . , ψ k ) B, and a choice of generic sections amounts to the choice of a generic matrix B .In the trivializing frame over the open set U the coordinates of the sections φ i is the i -th columnof the matrix product AB : ( φ , . . . , φ r ) = AB.
So, in the open set U the exact sequence (1) is0 O rU O rU A ′ | U . O kU ( φ , . . . , φ r ) B A
Then we have that(4) Supp( A ′ ) ∩ U = { x ∈ U | det( AB ) = 0 } . Consider the stratification by rank in the set Mat ( r × r, C ) and denote byMat ( r × r, C ) i := { c ∈ Mat ( r × r, C ) | corank( c ) ≥ i } . We have that codim(Mat ( r × r, C ) i ) = i and dim( ˜ X \ S ) = 2 . Now consider the map Θ : ( U \ S ) × Mat( k × r, C ) → Mat( r × r, C ), given by ( x, B ) A ( x ) B .Since the sections { ψ , . . . , ψ k } generate M over the set U \ S , we get that the map Θ is a submersionand therefore it is transverse to the rank stratification.By the Parametric Transversality Theorem, for almost every B in Mat( k × r, C ), the mapΘ B : U \ S → M at ( r × r, C )defined by x A ( x ) B is transverse to the rank stratification and to the sets Z and E . By thefiniteness of the trivializing cover we can choose a matrix B generic such that in each trivializationthe map Θ B is transverse to the rank stratification and to Z .By Equation (4) and transversality we have that Supp( A ′ ) ∩ U is smooth of dimension 1, disjointto Z and transversal to E . The tranversality of Θ B to the rank stratification also implies that forany x ∈ U we have the isomorphism A ′ x ∼ = O C,x , where C is the support of A ′ . Since the trivializingopen sets cover ˜ X \ S the proposition is proved. (cid:3) EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 17
The previous Proposition motivates the following definition.
Definition 4.4.
Let M be a reflexive O ˜ X -module of rank r . A collection ( φ , . . . , φ r ) of r sections iscalled nearly generic if the following conditions are satisfied. Let C be the support of the degeneracymodule A ′ .(1) For any point x ∈ ˜ X except in a finite collection we have the isomorphism A ′ x ∼ = ( O C ) x .(2) The support of C meets E in finitely many points (notice that this condition is void in thecase X = ˜ X ). Remark 4.5.
Proposition 4.3 states that a generic set of sections is in particular nearly generic.
Cohen-Macaulay modules of dimension 1.
In this section we study the structure ofCohen-Macaulay modules of dimension 1 which are of rank 1 and generically reduced. The concretedescription that we are about to obtain will be very important in our study of reflexive modules.
Definition 4.6.
Let Y be an analytic space and C a O Y -module dimension . Let C be the supportof C , with reduced structure. The module C is rank 1 generically reduced if C is isomorphic to O C except in finitely many points of the support. Remark 4.7.
Definition 4.4 and the first assertion of Proposition 4.3 states that degeneracy mod-ules (see Definition 4.1) for nearly generic sections are rank 1 generically reduced Cohen-Macaulaymodules of dimension . Proposition 4.8.
Let Y be an analytic space, and C be a rank 1 generically reduced Cohen-Macaulay O Y -module of dimension . Let C be the support of C with reduced structure. Denote by n : ˜ C → C the normalization.(1) The sheaf C is a O C -module, that is, the ideal of C is contained in the annihilator of C .(2) Let n : ˜ C → C be the normalization. If C is Stein then there exists an inclusion of C as O C -submodule of n ∗ O ˜ C which contains O C . In other words, we have the chain of inclusions O C ⊂ C ⊂ n ∗ O ˜ C . Proof.
For the first assertion let f be an element of the ideal of C . Assume that there exists asection c of C such that f · c is different from zero. Since C is rank 1 generically reduced the supportof f · c is a finite set of points, but this forces f · c to vanish, since otherwise C would not have depth1. Now we prove the second assertion. Consider the following map h : C → C ⊗ O C O ˜ C / ( T orsion ) ,c c ⊗ . By hypothesis we have that C is isomorphic to O C except in finitely many points. Therefore thesupport of the kernel of h is finite and since C does not have any finitely supported section we getthat the map h is injective.Now notice that n ∗ C / ( T orsion ) =
C ⊗ O C O ˜ C / ( T orsion ) is a torsion-free O ˜ C -module of rank one,and hence isomorphic to O ˜ C (since ˜ C is smooth and Stein). So we have an injection h : C → O ˜ C .Consider the (multi)-germ of ˜ C at the support of O ˜ C /h ( C ). Enumerate the branches of themulti-germ as ( ˜ C j , p j ) for j = 1 , ..., l . Each O ˜ C j ,p j is a discrete valuation ring for any j . Let t j bea uniformizing parameter of O ˜ C j ,p j . Let us denote by h ( c ) j the germ of h ( c ) in O ˜ C j ,p j .For any section c of C we define ord( h ( c )) := (cid:0) . . . , ord t j ( h ( c ) j ) , . . . (cid:1) , where ord t j denotes thevaluation of the ring O ˜ C j . Notice that ord( h ( c )) belongs to the set N l . Now for any generic λ and µ in C and c and c ′ in C we have thatord( h ( λc + µc ′ )) = min { ord( h ( c )) , ord( h ( c ′ )) } , where the minimum is taken componentwise.As a consequence, for a generic section c of C we have that ord( h ( c )) is the absolute minimumof the image of ord . Denote by ( n , . . . , n l ) = ord( h ( c )). By genericity and the fact that h ( C )spans O ˜ C outside { p , ..., p l } we also know that h ( c ) does not vanish outside { p , ..., p l } .Consider the commutative diagram C O ˜ C O ˜ C [ t − , ..., t − l ] hg · h ( c ) where · h ( c ) is multiplication by h ( c ) .By construction the map g is an O C -monomorphism and the image of g is contained in O ˜ C because min { ord( g ( c )) | c is a section of C} = ord( g ( c )) = (0 , . . . , . Moreover we have the equality g ( c ) = 1. This implies that g provides the needed chain of embeddings. (cid:3) The previous structure result allows to define some invariants of rank 1 generically reducedCohen-Macaulay modules of dimension 1 which will be important for us. Consider the notations ofthe previous proposition and its proof. We have defined an embedding of O C -modules ι : C → O ˜ C , such that O C is included in ι ( C ). Therefore the support of O ˜ C /ι ( C ) is contained in the pre-imageby n of the singular set of C . Let now be { p , ..., p l } the pre-image by n of the singular set, anddenote by ( ˜ C j , p j ) the germ of ˜ C at p j . As in the previous proof we have an order function ord : O ˜ C → N l . Lemma 4.9.
The image of the composition ord ◦ ι is independent of the embedding ι : C → O ˜ C aslong as O C is included in ι ( C ) . Proof.
Any two embeddings differ by multiplication by a unit in O ˜ C . (cid:3) Definition 4.10.
The set of orders C of C is the image in N l of the composition ord ◦ ι for anembedding ι : C → O ˜ C of O C -modules such that O C is included in ι ( C ) . Given a subset B ⊂ Z l , and a vector ( d , ..., d l ), we denote by ( d , ..., d l )+ B ⊂ Z l the translationof B in the direction of the vector ( d , ..., d l ). Remark 4.11.
Since C is a O Y -module, the set of orders C is stable by translation in the directiongiven by any vector ord ( f | C ) for f ∈ O Y (where f | C is the restriction of f to C ). It is well known that O C has a conductor ideal (see for example [1, (19.21)]). In our case wedefine the conductor of C as follows (compare with [1, (19.21)]). Definition 4.12.
Let C be an O C -module such that O C ⊂ C ⊂ n ∗ O ˜ C . The O C -submodule { s ∈ C | s · n ∗ O ˜ C ⊂ C} , EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 19 is called the conductor of C . The image of the conductor under the order function is called the conductor set of C . It is a subset of C . Any element of the conductor set of C is called a conductorof C . Remark 4.13.
Since O C is contained in C , we get that the conductor of C is non-empty. Since theconductor of C is closed by taking O X -linear combinations, the conductor set of C has an absoluteminimum, which is denoted by cond ( C ) .The conductor set of C satisfies the following property: if c = ( c , . . . , c l ) is a conductor of C then for any vector w in N l we have that c + w belongs to the set C . This motivates the following
Definition 4.14.
Let C be a subset of Z l , the conductor set of C is the (perhaps empty) set { v ∈ C | v + N l ⊂ C } . The correspondence at the Stein surface.
In this section we let X be a Stein surfacewith Gorenstein singularities; in many of the cases X will be a Milnor representative of a normalGorenstein surface singularity. The Gorenstein assumption and Theorem 3.1 allow us to understandthe relation between reflexive O X -modules with nearly generic sections and rank 1 genericallyreduced Cohen-Macaulay O X -modules with a system of generators. Theorem 4.15.
Let X be a Stein surface with Gorenstein singularities. There is a bijective cor-respondence between the set of pairs ( M, ( φ , ..., φ r )) of rank r reflexive O X -modules with r nearlygeneric sections and the set of pairs ( C , ( ψ , ..., ψ r )) of rank 1 generically reduced Cohen-Macaulay O X -modules with a system of generators of C as O X -module.Under this correspondence, if the system of generators ( ψ , ..., ψ r ) is not minimal then the module M contains free factors. As a partial converse: if M contains free factors and ( φ , ..., φ r ) aregeneric, then the system of generators ( ψ , ..., ψ r ) is not minimal.The correspondence from the first set to the second is called the direct correspondence at X , itsinverse is called the inverse correspondence at X . Proof.
Let M be a reflexive O X -module of rank r and ( φ , ..., φ r ) be r nearly generic sections.We obtain the exact sequence defining the degeneracy module given by the sections(5) 0 → O rX → M → C ′ → . Since the sections are nearly generic the module C ′ is of rank and 1 generically reduced by Re-mark 4.7.Dualizing the exact sequence with respect to O X we get(6) 0 → N → O rX → E xt O X ( C ′ , O X ) → . where N is the dual of M .By Proposition 3.2 the module C := E xt O X ( C ′ , O X ) is Cohen-Macaulay of dimension one. Let C be the support of C ′ , which coincides with the support of C . A direct computation of C showsthat, at a smooth point y ∈ X , if C ′ y is isomorphic to O C,y then C y is isomorphic to O C,y too. Thisshows that C is rank 1 generically reduced . Therefore we associate to the reflexive module M withthe given sections, the module C with the generators induced by the previous exact sequence.Conversely, let ( C , ( ψ , ..., ψ r )) be a rank 1 generically reduced 1-dimensional Cohen-Macaulaymodule with a system of generators. Define N to be the kernel of the morphism O rX → C inducedby the generators. We have the exact sequence(7) 0 → N → O rX → C → . Dualizing the exact sequence we get(8) 0 → O rX → M → E xt O X ( C , O X ) → , where M is the dual of N , and hence it is reflexive.To the pair ( C , ( ψ , ..., ψ r )) we associate the pair ( M, ( φ , ..., φ r )), where the sections are inducedby the second morphism of the previous exact sequence. The sections are nearly generic since themodule E xt O X ( C , O X ) is generically reduced, for being C rank 1 generically reduced .The correspondences are mutually inverse due to part (iii) of Theorem 3.1.The assertion relating free factors with minimality of the system of generators follows [9]. Supposethat the system of generators ( ψ , ..., ψ r ) of C is not minimal. Then an obvious computation showsthat the module of relations N has free factors. Since M is the dual of N then M has free factors.Conversely, suppose that M has a free factors, that is M ∼ = M ⊕ O aX . Then sequence (5) becomes0 → O r − aX ⊕ O aX → M ⊕ O aX → C ′ → . The genericity of the choice of the system of generators ( φ , ..., φ r ) imply that the morphism O aX →O aX obtained by the triple composition of the natural inclusion of O aX into O r − aX ⊕ O aX , the firstmorphism of the sequence, and the canonical projection of M ⊕ O aX to O aX is an isomorphism.Dualizing we obtain 0 → N ⊕ O aX → O r − aX ⊕ O aX → C → . Since the corresponding morphism O aX → O aX is an isomorphism we conclude that the system ofgenerators of C is not minimal. (cid:3) We may also understand the module of relations of the generators of the Cohen-Macaulay module C . Proposition 4.16.
Let C be an Cohen-Macaulay O X -module of dimension one, { φ , . . . , φ n } a setof generators of C as O X -module and consider the exact sequence obtained by the generators (9) 0 → N → O rX → C → . Then the module, N is reflexive. Proof.
Dualizing the exact sequence (9) and denoting by M := N ∨ , we obtain the exact sequence(10) 0 → O rX → M → E xt O X ( C , O X ) → . Since C is Cohen-Macaulay of dimension one then by Theorem 3.1 the module E xt O X ( C , O X ) isCohen-Macaulay of dimension one and C ∼ = E xt O X (cid:0) E xt O X ( C , O X ) , O X (cid:1) . Now dualizing (10) andusing the previous identification we obtain the exact sequence0 → N ∨∨ → O rX → C → , hence we have the identification N = N ∨∨ and N is reflexive. (cid:3) EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 21
The correspondence at the resolution.
In this section we generalize the correspondencegiven by Artin-Verdier [2], Esnault [9] and Wunram [37] at the resolution to the general case, thatis (
X, x ) is any normal surface singularity; in fact we go further and allow X to be a Stein normalsurface with possibly several singularities. We will obtain a bijection as in previous section. One ofthe sets is formed by pairs of modules together with systems of nearly generic sections. The otherset is formed by rank 1 generically reduced 1-dimensional Cohen-Macaulay modules together withsets of sections that satisfy a certain property. Before we state the main result of the section weneed to explain what the property means precisely.4.4.1. The Containment Condition.
Let π : ˜ X → X be a resolution of singularities of a Steinnormal surface X which is an isomorphism at the regular locus of X . Let E denote the exceptionaldivisor and U = ˜ X \ E . Let A be a rank 1 generically reduced 1-dimensional Cohen-Macaulay O ˜ X -module. Let ( ψ , ..., ψ r ) be r global sections spanning A as O ˜ X -module. The set of sections( ψ , ..., ψ r ) define a morphism O r ˜ X → A . Tensoring with the dualizing sheaf ω ˜ X = Λ Ω X and takingsections in U we obtain a morphism(11) δ : H ( U, ω r ˜ X ) → H ( U, A ⊗ ω ˜ X ) . We also have a restriction morphism γ : H ( ˜ X, A ⊗ ω ˜ X ) → H ( U, A ⊗ ω ˜ X ) . Definition 4.17.
The pair ( A , ( ψ , ..., ψ r )) satisfies the Containment Condition if we have theinclusion Im γ ⊂ Im δ . The correspondence at the resolution.
Theorem 4.18.
Let π : ˜ X → X be a resolution of singularities of a Stein normal surface whichis an isomorphism at the regular locus of X . There is a bijective correspondence between the setof pairs ( M , ( φ , ..., φ r )) formed by a locally free O ˜ X -module which is almost generated by globalsections, and a set of r nearly generic sections, and the set of pairs ( A , ( ψ , ..., ψ r )) formed by arank 1 generically reduced -dimensional Cohen-Macaulay O ˜ X -module, whose support meets E infinitely many points, and a set of r global sections spanning A as O ˜ X -module.Moreover M is full if and only if ( A , ( ψ , ..., ψ r )) satisfies the Containment Condition (see Def-inition 4.17). Proof.
We start defining the first bijection.Given ( M , ( φ , ..., φ r )) we consider the exact sequence induced by the sections:(12) 0 → O r ˜ X → M → A ′ → . The degeneracy module A ′ is 1-dimensional Cohen-Macaulay by Proposition 4.2, and is rank 1generically reduced with support intersecting the exceptional divisor E in a finite set, by definitionof nearly-generic sections (Definition 4.4). Dualizing the sequence we obtain(13) 0 → N → O r ˜ X → A → , where A = E xt O ˜ X ( A ′ , O ˜ X ) is a rank 1 generically reduced 1-dimensional Cohen-Macaulay module(same arguments than in the proof of Theorem 4.15), whose support meets E in finitely many points(since it coincides with the support of A ′ ). Let ( ψ , ..., ψ r ) be the generators of A as O ˜ X -modulegiven by the previous exact sequence. We define the direct correspondence as the correspondence sending the pair ( M , ( φ , ..., φ r )) tothe pair ( A , ( ψ , ..., ψ r )).Conversely, given ( A , ( ψ , ..., ψ r )) we consider the exact sequence (13) given by the sections.Dualizing it we obtain the exact sequence (12), and we define the inverse correspondence sending( A , ( ψ , ..., ψ r )) to ( M , ( φ , ..., φ r )), where ( φ , ..., φ r ) are the sections induce by the sequence (12).The direct and inverse correspondences are inverse to each other, for the same reasons appearingin the proof of Theorem 4.15.In order to prove the Theorem we have to show that M is full if and only if ( A , ( ψ , ..., ψ r ))satisfies the Containment Condition. For this we use the characterization of Proposition 3.4.We start showing that the inverse correspondence always gives a O ˜ X -module that is genericallygenerated by global sections. This is stated in a separate lemma. Lemma 4.19. If ( A , ( ψ , ..., ψ r )) is formed by a rank 1 generically reduced -dimensional Cohen-Macaulay O ˜ X -module, whose support meets E in finitely many points, and a set of r global sectionsspanning A as O ˜ X -module, then the O ˜ X -module M obtained by applying inverse correspondence isgenerically generated by global sections. Proof.
Applying the functor π ∗ − to the exact sequence (12) we get0 → O rX → π ∗ M → π ∗ A ′ → R π ∗ O r ˜ X → R π ∗ M → . Denote by G the image of π ∗ M in π ∗ A ′ , so we obtain the following two exact sequences of O X -modules 0 → O rX → π ∗ M → G → , (14) 0 → G → π ∗ A ′ → R π ∗ O r ˜ X → R π ∗ M → . Since the support of A ′ intersects the exceptional divisor in a finite collection of points, then wecan identify π ∗ A ′ with A ′ , viewed as a O X -module. Then G is a sub O X -module of A ′ .Denote by M ′ the subsheaf of M generated by its global sections, and by G ′ the sub- O ˜ X -moduleof A ′ spanned by G . We have the following diagram of coherent O ˜ X -modules0 0 00 O r ˜ X M ′ G ′ O r ˜ X M A ′ F F ′
00 0It is enough to prove that the support of F , which coincides with the support of F ′ , is a finiteset. For this it is enough to show that dim C ( F ′ ) = dim C ( coker ( G ′ → A ′ )) < ∞ . But we have theinequalities dim C ( coker ( G ′ → A ′ )) ≤ dim C ( coker ( G → π ∗ A ′ )) ≤ rp g by the Exact Sequence (14). (cid:3) EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 23
In order to finish the proof of Theorem 4.18 we have to prove that the map H E ( M ) → H ( M )is injective if and only if ( A , ( ψ , ..., ψ r )) satisfies the Containment Condition.By Serre duality the Containment Condition is equivalent to the surjection of the natural map H E (cid:16) M ∨ ⊗ ω ˜ X (cid:17) → H (cid:16) M ∨ ⊗ ω ˜ X (cid:17) , hence we will study this map. As before we denote N := M ∨ .Apply the functor − ⊗ ω ˜ X to the exact sequence (13),(15) 0 → N ⊗ ω ˜ X → ω r ˜ X → A ⊗ ω ˜ X → . Taking the long exact sequence of cohomology and local cohomology for the previous exactsequence we obtain a diagram of exact sequences: H E ( N ⊗ ω ˜ X ) H E (cid:16) ω r ˜ X (cid:17) H E ( A ⊗ ω ˜ X ) H E ( N ⊗ ω ˜ X ) H E (cid:16) ω r ˜ X (cid:17) H ( N ⊗ ω ˜ X ) H (cid:16) ω r ˜ X (cid:17) H ( A ⊗ ω ˜ X ) H ( N ⊗ ω ˜ X ) H (cid:16) ω r ˜ X (cid:17) H ( U ; N ⊗ ω ˜ X ) H (cid:16) U ; ω r ˜ X (cid:17) H ( U ; A ⊗ ω ˜ X ) H ( U ; N ⊗ ω ˜ X ) H (cid:16) U ; ω r ˜ X (cid:17) We have H ( ω ˜ X ) = 0 by Grauert-Riemenschneider Vanishing Theorem and H E ( A ⊗ ω ˜ X ) = 0because A ⊗ ω ˜ X has depth one and its support intersects the exceptional divisor in a finite set.Therefore we have the following diagram of exact sequences(16) 0 H E ( N ⊗ ω ˜ X ) H (cid:16) ω r ˜ X (cid:17) H ( A ⊗ ω ˜ X ) H ( N ⊗ ω ˜ X ) 0 H (cid:16) U ; ω r ˜ X (cid:17) H ( U ; A ⊗ ω ˜ X ) H ( U ; N ⊗ ω ˜ X ) β αδ γ ϕγ θ A diagram chase shows that H E ( N ⊗ ω ˜ X ) θ −→ H ( N ⊗ ω ˜ X ) is an epimorphism if and only ifIm γ ⊂ Im δ, which is precisely the Containment Condition. (cid:3) Remark 4.20.
Working with the Containment Condition is quite difficult. Because of this, in thenext section, we introduce a numerical condition which is implied by the Containment Condition,and that, in sufficiently many cases for our applications, is equivalent to it.
The Valuative Condition.
Let π : ˜ X → X be a resolution of singularities of a Stein normalsurface X which is an isomorphism over the regular locus of X . Let E denote the exceptionaldivisor and U := ˜ X \ E . Let C ⊂ X be a curve and C be its strict transform to ˜ X . Let n : ˜ C → C be the normalization. Let { p , ..., p l } be the preimage by n of E . Let ( ˜ C j , p j ) be the germ of ˜ C at p j . The ring O ˜ C j ,p j is a discrete valuation ring, and its valuation is denoted by ord ˜ C j .Let β be a meromorphic differential 2-form in H ( U, ω ˜ X ). We define a l -uple ord ( β ) in ( Z ∪ { + ∞} ) l as follows. For any j we let q j := n ( p j ) and choose a non-vanishing holomorphic differential 2-formgerm ω q j at q j . Then β = h j ω q j where h j is a meromorphic function. Define(17) ord ( β ) := ( ord ˜ C ( h ) , ..., ord ˜ C l ( h l )) . It is clear that the definition does not depend on the choice of the forms ω q j . This defines an orderfunction(18) ord : H ( U, ω ˜ X ) → ( Z ∪ {∞} ) l . Definition 4.21.
The canonical set of orders of the curve C at the resolution π is the set K π := ord ( H ( U, ω ˜ X )) ⊂ Z l . Given two subsets A , B ⊂ Z l we denote by A + B the subset of sums a + b where a ∈ A and b ∈ B .Let S be the semigroup of orders of O C ; since H ( U, ω ˜ X ) is a O X -module, we have the equality K π + S = K π .Given α, β ∈ H ( U, ω ˜ X ), for generic λ, µ ∈ C we have the equality(19) ord ( λα + µβ ) = min ( ord ( α ) , ord ( β ));hence the canonical set of orders of the curve C has an absolute minimum. Definition 4.22.
We define the canonical vector ( d ( C ) , ..., d l ( C )) of the curve C ∈ ˜ X to be theabsolute minimum of the canonical set of orders. Remark 4.23.
Since we have the equality K π + S = K π , the set ( d ( C ) , ..., d l ( C )) + S is includedin K π . Then K π has a non-empty conductor set, which contains the conductor set of S translatedby the vector ( d ( C ) , ..., d l ( C )) . Since K π is closed by taking minima, the conductor set of K π hasa unique absolute minimun, which we denote by cond ( K π ) . Now assume that X has Gorenstein singularities. Then there exist a holomorphic 2-form Ω ∈ H ( U, ω ˜ X ) whose associated divisor isdiv(Ω) = A + X q i E i , where each q i is a integer, the E i ’s are the irreducible components of the exceptional divisor and A is a divisor disjoint with E . The integers q i are independent on the choice of Ω. We call Ω a Gorenstein form . Remark 4.24. If X has Gorenstein singularities, and adopting the previous notation, the canonicalvector ( d ( C ) , ..., d l ( C )) of the curve C is given by the formulae d i ( C ) := X j q j C i · E j , EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 25 where C i · E j denotes intersection multiplicity. Remark 4.25. If X has Gorenstein singularities then the canonical set of orders of the curve C at the resolution π is equal to ( d ( C ) , ..., d l ( C )) + S , and its conductor set is the conductor set of S translated by the vector ( d ( C ) , ..., d l ( C )) . Let A be a rank 1 generically reduced 1-dimensional Cohen-Macaulay O ˜ X -module whose supportequals C . Let A be its set of orders (see Definition 4.10). Let ( ψ , ..., ψ r ) be r global sectionsspanning A as O ˜ X -module. The O X -module C spanned by ( ψ , ..., ψ r ) is rank 1 generically reduced1-dimensional Cohen-Macaulay. Let C be the set of orders of C , which is the subset of A obtainedfollowing Definition 4.10.The set of sections ( ψ , ..., ψ r ) define a epimorphism O r ˜ X → A . Tensoring with ω ˜ X we obtain anepimorphism ω r ˜ X → A ⊗ ω ˜ X . Taking global sections in U we obtain a morphism(20) δ : H ( U, ω r ˜ X ) → H ( U, A ⊗ ω ˜ X ) . Since ω ˜ X is locally free of rank 1 and A has Stein support, the restriction of ω ˜ X to the support of A is isomorphic to the structure sheaf of the support. Hence there is an isomorphism A ∼ = A ⊗ ω ˜ X .We have fixed an embedding of A into O ˜ C , which naturally embeds H ( U, A ) into the total fractionring K ( ˜ C ). Such total fractions ring maps into the direct sum ring ⊕ li =1 K ( ˜ C i , p i ) (here K ( ˜ C i , p i )is the quotient field of O ˜ C i ,p i ). Definition 4.26.
We define the canonical set of orders of ( A , ( ψ , ..., ψ r )) to be the image K ( A , ( ψ ,...,ψ r )) of the composition (21) H ( U, ω r ˜ X ) δ −→ H ( U, A ⊗ ω ˜ X ) ∼ = H ( U, A ) ⊂ ⊕ li =1 K ( ˜ C i , p i ) ord −→ ( Z ∪ { + ∞} ) l . It is easy to verify that the previous definition does not depend on the choice of the isomorphism
A ∼ = A ⊗ ω ˜ X . Definition 4.27.
With the notations introduced above, the pair ( A , ( ψ , ..., ψ r )) satisfy the Valua-tive Condition if the inclusion A ⊂ K ( A , ( ψ ,...,ψ r )) is satisfied. Remark 4.28.
We have the inclusion C + K π ⊂ K ( A , ( ψ ,...,ψ r )) . If X has Gorenstein singularities and ( d ( C ) , ..., d l ( C )) is the Gorenstein vector at the resolution π , then we have the equalities K ( A , ( ψ ,...,ψ r )) = C + K π = ( d ( C ) , ..., d l ( C )) + C . Therefore, in the Gorenstein case the valuative condition holds for ( A , ψ , ..., ψ r )) if and only ifthe inclusion A ⊂ ( d ( C ) , ..., d l ( C )) + C is satisfied. Proof.
The first inclusion follows easily from the definition of K ( A , ( ψ ,...,ψ r )) . The equalities followbecause in the Gorenstein case any element of H ( U, ω r ˜ X ) is equal to a Gorenstein form multipliedby a global regular function. (cid:3) Remark 4.29.
Since both C and K π have non empty conductor sets and K ( A , ( ψ ,...,ψ r )) is closedby taking minima, the conductor set of K ( A , ( ψ ,...,ψ r )) is non-empty and has an absolute minimumdenoted by cond ( K ( A , ( ψ ,...,ψ r )) ) . Moreover we have the inequalities (22) cond ( K ( A , ( ψ ,...,ψ r )) ) ≤ cond ( C ) + cond ( K π ) ≤ cond ( C ) + ( d ( C ) , ..., d l ( C )) . In the Gorenstein case this inequality becomes the equality (23) cond ( K ( A , ( ψ ,...,ψ r )) ) = cond ( C ) + ( d ( C ) , ..., d l ( C )) . Likewise, the conductor set of A has an absolute minimum denoted by cond ( A ) . Remark 4.30.
The following set of easy observations will have very useful consequences later.(1) If cond ( K ( A , ( ψ ,...,ψ r )) ) ≤ (0 , ..., , then the Valuative Condition is satisfied.(2) If A equals N l , then the valuative condition is equivalent to the inequality cond ( K ( A , ( ψ ,...,ψ r )) ) ≤ (0 , ..., . (3) If X has Gorenstein singularities, then the condition cond ( K ( A , ( ψ ,...,ψ r )) ) ≤ (0 , ..., holdsif and only if cond ( C ) ≤ − ( d ( C ) , ..., d l ( C )) . Proposition 4.31.
The Containment Condition implies the Valuative Condition.
Proof.
We have to translate the Containment Condition Im γ ⊂ Im δ into the inclusion of sets A ⊂ K ( A , ( ψ ,...,ψ r )) . We need the concrete description of the sheaf A obtained in Proposition 4.8: let C be the supportof A , n : ˜ C → C be its normalization. We have a chain of inclusions(24) O C ⊂ A ⊂ n ∗ O ˜ C , which is not necessarily unique. We fix one of such chains of inclusions.Since ω ˜ X is an invertible sheaf and A has support contained in a Stein open subset of ˜ X we havethe isomorphism A ⊗ ω ˜ X ∼ = A . Hence H ( ˜ X, A ⊗ ω ˜ X ) ∼ = H ( ˜ X, A ) , and H ( U, A ⊗ ω ˜ X ) ∼ = H ( U, A ) ⊂ K ( ˜ C ) , where K ( ˜ C ) denotes the total fraction ring of the ring H ( ˜ C, O ˜ C ).The image Im γ is then identified with the inclusion H ( ˜ X, A ) ⊂ H ( ˜ C, O ˜ C ) ⊂ K ( ˜ C ) . As a consequence, if we consider the order function ord : K ( ˜ C ) → Z l , we obtain the equality of sets(25) ord (Im γ ) = A . The morphism δ is induced by the sections { ψ , . . . , ψ r } of A . The definition of K ( A , ( ψ ,...,ψ r )) gives the equality(26) ord (Im δ ) = K ( A , ( ψ ,...,ψ r )) . Now the result follows, because of equalities (25) and (26). (cid:3)
EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 27
Before we extract some useful consequences of Remark 4.30 and Theorem 4.18 we need a furtherlemma:
Lemma 4.32.
Let ( X, x ) be a complex analytic germ of a normal two-dimensional Gorensteinsingularity and π : ˜ X → X be a resolution with exceptional divisor E = S ni =1 E i . Then for anycomponent E j where the Gorenstein form Ω has a pole and for any component E k where the Goren-stein form has a zero, we have that E j ∩ E k = ∅ . Proof.
If the singularity is rational the Gorenstein form does not have poles at any resolution. Theresult follows for that case.If the singularity is non-rational the Gorenstein form have strict poles at any component of theexceptional divisor of the minimal resolution. If p is a point at a resolution ˜ X , E = ∪ ri =1 E i is thedecomposition in irreducible components of the exceptional divisor at ˜ X , and E p is the exceptionaldivisor at the blow up at p , then the order of Ω at E p equals ord E p (Ω) := 1 + r X i =1 ord E i (Ω) mult p ( E i ) . Using induction on the number of blows ups that are necessary in order to obtain the resolution π : ˜ X → X from the minimal resolution the proposition is proved easily in the non-rational case. (cid:3) The consequences announced above are:
Proposition 4.33.
Let π : ˜ X → X be a resolution. Let A be a rank 1 generically reduced -dimensional Cohen-Macaulay O ˜ X -module. Let A be its set of orders (see Definition 4.10). Let C bethe support of A and n : ˜ C → C be the normalization. Let ( d ( C ) , ..., d l ( C )) be the canonical vectorof C and K π be the canonical set of orders of C . Let ( ψ , ..., ψ r ) be r global sections spanning A as O ˜ X -module. The O X -module C spanned by ( ψ , ..., ψ r ) is rank 1 generically reduced -dimensionalCohen-Macaulay. Let C be the set of orders of C . Let K ( A , ( ψ ,...,ψ r )) be the canonical set of ordersof ( A , ( ψ , ..., ψ r )) .Let ( M , ( φ , ..., φ r )) be the pair associated with ( A , ( ψ , ..., ψ r )) in the proof of Theorem 4.18.(1) If cond ( K ( A , ( ψ ,...,ψ r )) ) ≤ , then M is full.(2) For the previous condition it is enough to have the inequality cond ( C ) ≤ − ( d ( C ) , ..., d l ( C )) .(3) Suppose that the curve C is smooth and meets the exceptional divisor transversely at smoothpoints. Then M is full if and only if we have the inequality cond ( K ( A , ( ψ ,...,ψ r )) ) ≤ . Inthe Gorenstein case the inequality becomes cond ( C ) ≤ − ( d ( C ) , ..., d l ( C )) .(4) If there is an index i such that the strict inequality d i ( C ) > holds, then M is not full. Inthe Gorenstein case, if C meets a component of the exceptional divisor where the Gorensteinform has a zero, then M is not full.(5) If X has Gorenstein singularities, π is small with respect to the Gorenstein form and C = A (that is, if the module M is special), then M is full.(6) Suppose that ( d ( C ) , ..., d l ( C )) ≤ and that C = π ∗ n ∗ O ˜ C . Then M is a special full O ˜ X -module. In the Gorenstein case the first inequality holds when C does not meet an excep-tional divisor where the Gorenstein form has a . Proof. If cond ( K ( A , ( ψ ,...,ψ r )) ) ≤ n ∗ O ˜ C is contained in Im δ . Since Im γ is contained in n ∗ O ˜ C the Containment Condition holds and Assertion (1) follows by Theorem 4.18. Assertion (2) is a consequence of Assertion (1) and Inequality (22).In order to prove Assertion (3) notice that if C is smooth and meets the exceptional divisortransversely at smooth points then A is equal to O D = n ∗ O ˜ D and then we have A = N l . In thissituation the Valuative Condition is equivalent to the inequality cond ( K ( A , ( ψ ,...,ψ r )) ) ≤
0. In theGorenstein case Equality (23) transforms the previous inequality into cond ( C ) ≤ − ( d ( C ) , ..., d l ( C )).This proves Assertion (3).In order to prove Assertion (4) notice that the vector (0 , ...,
0) is always contained in A , since O C is contained in A . On the other hand, if there is an index i such that the strict inequality d i ( C ) > , ...,
0) is not contained in K ( A , ( ψ ,...,ψ r )) , and the Valuative Condition does not hold. ByLemma 4.32, in the Gorenstein case there is an index i such that the strict inequality d i ( C ) > C meets a component of the exceptional divisor where the Gorenstein form hasa zero. This proves Assertion (4).For Assertion (6) notice that if C = π ∗ n ∗ O ˜ C then automatically we have the equality C = A ,and M is special. If ( d ( C ) , ..., d l ( C )) ≤
0, it is easy to show using the equality C = π ∗ n ∗ O ˜ C , thatwe have the inequality cond ( K ( A , ( ψ ,...,ψ r )) ) ≤
0. So by Assertion (1), if ( d ( C ) , ..., d l ( C )) ≤ M is full. This proves the assertion, except the addendum about Gorenstein singularities, whichfollows from Lemma 4.32.Assertion (5) is a bit harder. We will proof the Containment Condition directly by a cohomo-logical argument: we have to prove that image of γ is contained in the image of δ in Diagram (16).Since we have a small resolution with respect to the Gorenstein form (see Definition 2.1), we canconsider the exact sequence(27) 0 → ω ˜ X → O ˜ X → O Z K → . Now apply the functor − ⊗ − to the sequences (13) and (27),0 00
N ⊗ ω ˜ X ω r ˜ X A ⊗ ω ˜ X N O r ˜ X A N ⊗ O Z k O rZ k A ⊗ O Z k
00 0 0
EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 29
By the last diagram we get the following commutative diagram H (cid:16) ˜ X, ω r ˜ X (cid:17) H (cid:16) ˜ X, A ⊗ ω ˜ X (cid:17) H (cid:16) ˜ X, O r ˜ X (cid:17) H (cid:16) ˜ X, A (cid:17) H (cid:16) U, ω r ˜ X (cid:17) H ( U, A ⊗ ω ˜ X ) H (cid:16) U, O r ˜ X (cid:17) H ( U, A ) γ βρν δθ δ ′ αγ ′ We need to prove that(28) im( γ ) ⊂ im( δ ) . Notice that the maps α and θ are isomorphisms because the support of O Z K does not intersect U . Since α is injective, the condition (28) is equivalent toIm( αγ ) ⊂ Im( αδ ) . Since the diagram is commutative and θ is onto we getim( αδ ) = im( δ ′ θ ) = im( δ ′ ) . Hence it is enough to prove that the image of ( αγ ) is contained in the image of δ ′ . Using againthat the diagram is commutative and ρ is onto because M is special, we getim( αγ ) = im( γ ′ β ) ⊂ im( γ ′ ) = im( γ ′ ρ ) = im( δ ′ ν ) ⊂ im( δ ′ ) , as we wish. (cid:3) A comparison of correspondences.
In order to compare the two correspondences we needto impose that X is a normal Stein surface with Gorenstein singularities. The following propositionscompare the correspondence at the Stein surface (Theorem 4.15) and the correspondences at variousresolutions (Theorem 4.18). Proposition 4.34.
Let X be a normal Stein surface with Gorenstein singularities. Let M bea reflexive O X -module of rank r . Let π : ˜ X → X be a resolution and ρ : ˜ X → ˜ X bethe blow up at a point p . Denote by π : ˜ X → X the composition π = π ◦ ρ . Denote by M and M the full sheaves associated with M at each of the resolutions. Let ( φ , ..., φ r ) be r generic sections of M . Let ( A , ( ψ , ..., ψ r )) and ( A , ( ψ , ..., ψ r )) be the pairs associated with ( M , ( φ , ..., φ r )) and ( M , ( φ , ..., φ r )) under Theorem 4.18. Let ( C , ( ψ , ..., ψ r )) be the pair asso-ciated with ( M, ( φ , ..., φ r )) under Theorem 4.15.(1) There are inclusions C ⊂ ( π ) ∗ A ⊂ ( π ) ∗ A . Under this inclusion the sections ( ψ i , ..., ψ ir ) are identified for i = 0 , , . The dimension of the quotient ( π i ) ∗ A i / C as a C -vector spaceequals dim C ( R ( π i ) ∗ M ∨ i ) − rp g for i = 1 , .(2) We have the inclusion A ⊂ ρ ∗ A and the dimension of the quotient ρ ∗ A / A as a C -vectorspace equals dim C ( R ρ ∗ M ∨ ) . Proof.
For i = 1 , → O r ˜ X i → M i → A ′ i → , whose first morphism is induced by the sections ( φ , ..., φ r ). Dualizing we obtain(29) 0 → N i → O r ˜ X i → A i → , where the second morphism is induced by the sections ( ψ i , ..., ψ ir ).The sections ( φ , ..., φ r ) also induce the exact sequence0 → O rX → M → C ′ → , and dualizing it we obtain(30) 0 → N → O rX → C → , where the second morphism is induced by ( ψ , ..., ψ r ).Applying R ( π i ) ∗ to (29), we obtain the exact sequence(31) 0 → ( π i ) ∗ N i → O rX → ( π i ) ∗ A i → R ( π i ) ∗ N i → R ( π i ) ∗ O rX → . By Lemma 3.6 and its proof the first morphism of the previous sequence (for i = 1 ,
2) coincideswith the first morphism of sequence (30). This implies that the image of the second morphism ofthe previous sequence (for i = 1 ,
2) coincides with C and that, under this identification the systemsof sections ( ψ ii , ..., ψ ir ) coincide for i = 0 , ,
2. This proves the inclusions
C ⊂ ( π i ) ∗ A i for i = 1 , A i is generated by ( ψ i , ..., ψ ir ) as a O ˜ X i -module and O ˜ X contains O ˜ X , the inclusion ( π ) ∗ A ⊂ ( π ) ∗ A also holds. The equalitydim C ( π i ) ∗ A i / C ) = dim C ( R ( π i ) ∗ M ∨ i ) − rp g follows from Exact Sequence (31). This shows the first assertion.The proof of the second assertion follows similarly, by applying Rρ ∗ to Sequence (29), and takinginto account the identification of the sections ( ψ ii , ..., ψ ir ) for i = 1 , (cid:3) For later use we need to compare the sets K ( A , ( ψ ,...,ψ r )) and K ( A , ( ψ ,...,ψ r )) (see Section 4.4.3for the corresponding definition). Proposition 4.35.
Consider the same situation than in the previous proposition, but allow non-Gorenstein normal singularities. There exist a non-negative integer vector ( d , ..., d l ) such that wehave the equality (32) K ( A , ( ψ ,...,ψ r )) = ( d , ..., d l ) + K ( A , ( ψ ,...,ψ r )) . As a consequence we have also the equality (33) cond ( K ( A , ( ψ ,...,ψ r )) ) = ( d , ..., d l ) + cond ( K ( A , ( ψ ,...,ψ r )) ) . The vector is strictly positive if and only if the blowing up center of ρ meets the support of A . Proof.
Denote by E i the exceptional divisor of π i . Let C be the support of C , let C i be thesupport of A i . We have a birational morphisms ρ | C : C → C and π | C : C → C . Consider thenormalization n : ˜ C → C . Let p j for j = 1 , ..., l be the points of ˜ C which map via n to a pointof the exceptional divisor. Let ( ˜ C j , p j ) be the germ at p j . Denote p j := n ( p j ) and p j := ρ ( p j ) for j = 1 , ..., l .The proposition is trivial if the center of the blowing up ρ does not coincide with p j for any j . By notational convenience we assume that p is the blowing-up center. We denote by E the EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 31 exceptional divisor of ρ . Choose local coordinates ( x ij , y ij ) of ˜ X i around each point p ij . The choiceis made so that if two points p ij coincide for different j , then the corresponding coordinates are alsothe same, and so that, if ρ ( p j ) = p , then ρ expresses in local coordinates as ρ ( x j , y j ) = ( x j , x j y j ).Let β , ..., β n be a system of generators of H ( U, ω ˜ X ) as a O X -module. The differential form β k expresses in each of the local chart around p ij as β = h ik,j dx ij ∧ dy ij , where h ik,j is a germ ofmeromorphic function at p ij . If ρ ( p j ) = p then h k,j = x j ρ ∗ h k, . If ρ ( p j ) = p m for m = 1 then h k,j = ρ ∗ h k,m .In order to compare K ( A , ( ψ ,...,ψ r )) and K ( A , ( ψ ,...,ψ r )) we compare the images of H ( U, ω ˜ X )in the ring L li =1 K ( ˜ C i , p i ) according with Definition 4.26; we denote each of the images by Im i for i = 1 ,
2. Since, by Proposition 4.34 the sections ( ψ , ..., ψ r ) and ( ψ , ..., ψ r ) are identified,each of them define the same l -uple in L li =1 K ( ˜ C i , p i ), which we denote by ( ψ v | ˜ C , ..., ψ v | ˜ C ) for v ∈ { , ..., r } . Then Im i is the O X -module spanned by (cid:8) ( h i | ˜ C ψ v | ˜ C , ..., h il | ˜ C l ψ v | ˜ C l ) : v ∈ { , ..., r } (cid:9) . Enumerate the points p , ..., p l so that those whose image by ρ equals p are p , ..., p l for l ≤ l .We have the equality Im = ( x | C , ..., x l | C l , , ..., Im . As a consequence, if for any w ≤ l we define the intersection multiplicity d w := I p w ( E , C w ) wehave the equality K ( A , ( ψ ,...,ψ r )) = ( d , ..., d l , , ...,
0) + K ( A , ( ψ ,...,ψ r )) . (cid:3) Proposition 4.36.
Let π : ˜ X → X be a resolution of a normal Stein surface with Gorensteinsingularities, which is an isomorphism over the regular locus of X . Let ( A , ( ψ , ..., ψ r )) be a pairformed by a rank 1 generically reduced -dimensional Cohen-Macaulay O ˜ X -module, whose supportmeets E in finitely many points, and a set of r global sections spanning A as O ˜ X -module andsatisfying the Containment Condition. Let C be the O X -module spanned by ψ , ..., ψ r . Then C isa rank 1 generically reduced -dimensional Cohen-Macaulay O X -module. Let ( M , ( φ , ..., φ r )) and ( M, ( φ ′ , ..., φ ′ r )) be the results of applying the correspondences of Theorems 4.18 and 4.15 at ˜ X andat the X to ( A , ( ψ , ..., ψ r )) and ( C , ( ψ , ..., ψ r )) respectively. Then we have the equalities π ∗ M = M and ( φ , ..., φ r ) = ( φ ′ , ..., φ ′ r ) . Proof.
According with the proof of Theorem 4.18 and its proof the module N in the sequence0 → N → O r ˜ X → A → M . Pushing down by π ∗ we obtain0 → π ∗ N → O rX → π ∗ A → R π ∗ N → R π ∗ O r ˜ X → , and the image of the map O rX → π ∗ A is the O X -module spanned by ψ , ..., ψ r , that is, the module C . So we obtain the sequence 0 → π ∗ N → O rX → C → . According with Theorem 4.15 and its proof the module π ∗ N is isomorphic to the dual of M . ByLemma 3.6 the module π ∗ N is isomorphic to the dual of π ∗ M . This concludes the proof of theequality π ∗ M = M . Under the equality, the coincidence of the sections is straightforward. (cid:3) The minimal adapted resolution
In this section we show that, given a Stein normal surface X and a reflexive O X -module, thereis a minimal resolution for which the associated full sheaf is generated by global sections. Thisresolution will be crucial later. Proposition 5.1.
Let X be a Stein normal surface. If M is a reflexive O X -module, then there existsa unique minimal resolution ρ : ˜ X ′ → X such that the associated full O ˜ X -module M := ( ρ ∗ M ) ∨∨ isgenerated by global sections. Proof.
Let M be a reflexive O X -module, π : ˜ X → X be the minimal resolution with exceptionaldivisor E and denote by M = ( π ∗ M ) ∨∨ . If M is generated by global sections, then we are done.If M is not generated by global sections, then there exists a finite set of points S = { p , . . . , p n } ⊂ E where M fails to be generated by global sections.Assume that the rank of M is r . Take r generic sections of M and consider the exact sequencegiven by the sections(34) 0 → O r ˜ X → M → A ′ → . By the degeneracy module definition (Definition 4.1) we have the inclusion S ⊂ Supp( A ′ ).Let C be the support of A ′ and ( d ( C ) , ..., d r ( C )) be the associated canonical vector (see Defi-nition 4.22). By Proposition 4.33, (4) we have the inequality(35) ( d ( C ) , ..., d r ( C )) ≤ (0 , ..., . Denote by σ S : ˜ X ′ → X the blow up at the set of centers S . Therefore we have the followingcommutative diagram ˜ X ′ ˜ XXσ S π ′ π Denote by M ′ the full O ˜ X ′ -module associated to M . If M ′ is generated by global sections, thenwe are done, otherwise we repeat the procedure.In order to prove that this process eventually ends we use ( d ( C ) , ..., d r ( C )) as a resolutioninvariant.We take the same generic global sections for M ′ and M (in both cases the set of global sectionsis M ). The support C ′ of the degeneracy module of M ′ for this sections is the strict transform of C by σ S . The normalization ˜ C of C and C ′ is the same. Let { ( ˜ C i , p i ) } li =1 be the branches of ˜ C considered at the beginning of Section 4.4.3. Let β be meromorphic differential form in ˜ X . By thebehavior of the poles of the pullback of a meromorphic differential form by a blow up at a point ina smooth surface, we obtain that the order of σ ∗ S β at the different branches ( ˜ C i , p i ) for i = 1 , ..., l isgreater or equal that the order of β . Moreover the order is strictly greater if the blowing up centermeets the component that we are dealing with. This implies the strict inequality( d ( C ) , ..., d r ( C )) < ( d ( C ′ ) , ..., d r ( C ′ )) . This together with Inequality (35) shows that the process terminates after finitely many steps. (cid:3)
EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 33
Definition 5.2.
Let M be a reflexive O X -module. The minimal resolution ρ : ˜ X → X where theassociated full O ˜ X -module is generated by global sections is called the minimal adapted resolution to M . Remark 5.3.
Let X be a Stein surface with Gorenstein singularities. Let M be a reflexive O X -module. Then the minimal adapted resolution to M is small with respect to the canonical form. Proof.
This is a consequence of Inequality (35) and Proposition 4.33, (4). (cid:3)
At the minimal adapted resolution the degeneracy module of the full sheaf for a generic set ofsections has special properties, and Lemma 1.2 of [2] holds as stated there.
Lemma 5.4.
Let X be a Stein normal surface. Let M be a reflexive O X -module of rank r . Let ρ : ˜ X → X be a resolution which dominates the minimal adapted resolution to M . The degeneracymodule A ′ of a set of r generic global sections is isomorphic to O D , where D ⊂ ˜ X is a smoothcurve meeting the exceptional divisor transversely at its smooth locus. Moreover, by changing thesections the meeting points of D with the exceptional divisor also change. Proof.
It is a simplification of the proof of Proposition 4.3. (cid:3)
Our aim now is to characterize numerically the minimal adapted resolution.
Proposition 5.5.
Let X be a Stein normal surface. Let M be a reflexive O X -module of rank r . Let ( φ , ..., φ r ) be r generic sections. Let π : ˜ X → X be a resolution, denote by M the full O ˜ X -module associated with M . Let ( A , ( ψ , ..., ψ r )) be the pair associated with ( M , ( φ , ..., φ r )) under Theorem 4.18. Genericity of the sections imply that the associated canonical set of orders K ( A , ( ψ ,...,ψ r )) (see Definition 4.26), and its minimal conductor cond ( K ( A , ( ψ ,...,ψ r )) ) are independentof the chosen sections. Then the following properties hold:(1) At the minimal adapted resolution we have the equality cond ( K ( A , ( ψ ,...,ψ r )) ) = (0 , ..., .(2) At any resolution we have the inequality cond ( K ( A , ( ψ ,...,ψ r )) ) ≤ (0 , ..., .(3) A resolution dominates the minimal adapted resolution if and only if we have the equality cond ( K ( A , ( ψ ,...,ψ r )) ) = (0 , ..., . Proof.
Let π : ˜ X → X be the minimal adapted resolution to M and M the full O ˜ X -moduleassociated with M . Let E be the exceptional divisor. Consider the decomposition in irreduciblecomponents E = ∪ mj =1 E ,j .Let ( A , ( ψ , ..., ψ r )) be the pair associated with ( M , ( φ , ..., φ r )) by Theorem 4.18. Let C be the support of A . By Lemma 5.4, C is a smooth curve meeting the exceptional divisor E transversely at smooth points and we have the isomorphism A ∼ = O C .Then, by Proposition 4.33 (3), at the minimal adapted resolution π : ˜ X → X we have theinequality cond ( K ( A , ( ψ ,...,ψ r )) ) ≤ (0 , ..., cond ( K ( A , ( ψ ,...,ψ r )) ) is nega-tive. Since C is smooth it coincides with its normalization. Let p be the point of C ∩ E so that the order at the branch ( C , p ) gives the first coordinate of the order function (see Sec-tion 4.4.3 to recall the corresponding definitions). Let ρ : ˜ X → ˜ X be the blow up at p ; define π := π ◦ ρ . Let C be the strict transform of C by ρ . Define A := O C . Since ρ inducesan isomorphism between A = O C and A = O C the sections ( ψ , ..., ψ r ) of A may also be regarded as sections of A . A computation like in the proof of Proposition 4.35 shows the equal-ity cond ( K ( A , ( ψ ,...,ψ r )) ) = (1 , , ...,
0) + cond ( K ( A , ( ψ ,...,ψ r )) ). Therefore we have the inequality cond ( K ( A , ( ψ ,...,ψ r )) ) ≤ (0 , ..., A , ( ψ , ..., ψ r )) a pair ( M , ( φ , ..., φ r )), where M is a full O ˜ X -module and ( φ , ..., φ r )is a system of nearly generic global sections. An application of Proposition 4.36 shows the equalities( π ) ∗ M = ( π ) ∗ M and ( φ , ..., φ r ) = ( φ , ..., φ r ).By Lemma 5.4, there is a slight perturbation ( φ ′ , ..., φ ′ r ) of the sections ( φ , ..., φ r ) such that ifwe denote by A ′ the degeneracy module of ( M , ( φ ′ , ..., φ ′ r )), then its support ( C ) ′ satisfies • it does not meet the blowing up center C ∩ E ; • we have the equality of intersection numbers ( C ) ′ · E ,j = C · E ,j for any irreduciblecomponent E ,j .The support of the degeneracy module of ( M , ( φ , ..., φ r )) is equal to C . On the other hand,since ( C ) ′ does not meet the blowing up center C ∩ E , the support of the degeneracy moduleof ( M , ( φ ′ , ..., φ ′ r )) is the strict transform of ( C ) ′ to ˜ X . Let F be the exceptional divisor of ρ .Observe that C · F = 1, and by property (2) above ( C ) ′ · F = 0. Since the Poincare dual of thesupport of the degeneracy locus is the first Chern class of the module M , we have two differentChern class representations intersecting differently the cycle F . This is a contradiction which provesAssertion (1).Assertions (2) and (3) are simple Corollaries of Assertion (1) and Proposition 4.35. (cid:3) It is convenient to specialize the previous Proposition to the Gorenstein case:
Corollary 5.6.
With the notation of the previous Proposition, and of Proposition 4.33, if X hasGorenstein singularities we have:(1) At the minimal adapted resolution we have the equality cond ( C ) = ( − d ( C ) , ..., − d l ( C )) .(2) At any resolution we have the inequality cond ( C ) ≤ ( − d ( C ) , ..., − d l ( C )) .(3) A resolution dominates the minimal adapted resolution if and only if we have the equality cond ( C ) = ( − d ( C ) , ..., − d l ( C )) . The behaviour of the speciality defect and minimal adapted resolutions of specialmodules.
We study the behaviour of the specialty defect under blow up, and show that if thespecialty defect vanishes at a given resolution, it vanishes at any resolution. As a corollary weestablish the existence of special reflexive modules and show an interesting property of their genericdegeneracy modules, which links them with arcs in the singularity.The following two propositions control the behavior of the specialty defect under blow up.
Proposition 5.7.
Let X be a Stein normal surface. Let M be a reflexive O X -module. Let π : ˜ X → X be a resolution and p be a point in ˜ X . Denote by σ : ˜ X ′ → ˜ X the blow up of the point p . Wehave the following diagram ˜ X ′ ˜ XXσρ π where ρ := π ◦ σ . EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 35
Denote by M = ( π ∗ M ) ∨∨ and M ′ = ( ρ ∗ M ) ∨∨ . Then the specialty defect of M is less or equalto the specialty defect of M ′ . Moreover if M is generated by global sections at p , then the specialtydefect of M equals the specialty defect of M ′ . Proof.
Denote by N ′ = M ′ ∨ , N = M ∨ . Since ρ = π ◦ σ , in order to compute R ρ ∗ N ′ we use the Leray spectral sequence. In this casethe page E ( p,q )2 of the spectral sequence is given by2 0 0 01 π ∗ (cid:0) R σ ∗ N ′ (cid:1) R π ∗ (cid:0) R σ ∗ N ′ (cid:1) π ∗ ( σ ∗ N ′ ) R π ∗ ( σ ∗ N ′ ) 00 1 2The spectral sequence degenerates, therefore we obtain the following exact sequence(36) 0 → R π ∗ ( σ ∗ N ′ ) → R ρ ∗ N ′ → π ∗ (cid:0) R σ ∗ N ′ (cid:1) → . Now by adjuction we have the following identification R π ∗ ( σ ∗ N ′ ) = R π ∗ (cid:0) σ ∗ H om O ˜ X ′ ( σ ∗ π ∗ M, O ˜ X ′ ) (cid:1) = R π ∗ H om O ˜ X ( π ∗ M, σ ∗ O ˜ X ′ )= R π ∗ H om O ˜ X ( π ∗ M, O ˜ X )= R π ∗ N . (37)By (36) and (37) we get(38) 0 → R π ∗ N → R ρ ∗ N ′ → π ∗ (cid:0) R σ ∗ N ′ (cid:1) → . Therefore the specialty defect of M is less or equal to the specialty defect of M ′ .Assume that M is generated by global sections at p . By the previous exact sequence we onlyneed to prove that π ∗ (cid:0) R σ ∗ N ′ (cid:1) = 0.Now consider the exact sequence given by the natural map from π ∗ M to its double dual0 T π ∗ M M S , where T is the kernel and S is the cokernel. Notice that the support of S is the set S (the pointswhere M fails to be generated by global sections, see Section 4).The last exact sequence can be split as follows0 T π ∗ M π ∗ M/T , π ∗ M/T
M S . Applying the functor σ ∗ − to the last two exact sequences we obtain0 K σ ∗ T ρ ∗ M σ ∗ π ∗ M/T , K σ ∗ π ∗ M/T σ ∗ M σ ∗ S , where K and K are the modules that make the last sequences exact (recall that σ ∗ − is just aright exact functor).Hence we split the previous exact sequences as follows0 K σ ∗ T H , H ρ ∗ M σ ∗ π ∗ M/T , K σ ∗ π ∗ M/T H , H σ ∗ M σ ∗ S . Dualizing the first, second and third exact sequences we get H ∨ ∼ = 0 , because σ ∗ T is supported in the exceptional divisor , ( σ ∗ π ∗ M/T ) ∨ ∼ = ( ρ ∗ M ) ∨ , by the previous identification ,H ∨ ∼ = ( σ ∗ π ∗ M/T ) ∨ , because K is supported in the exceptional divisor . Hence, as we have N ′ = M ′ ∨ ∼ = ( ρ ∗ M ) ∨ we get N ′ ∼ = ( σ ∗ π ∗ M/T ) ∨ ∼ = H ∨ .Finally dualizing the fourth exact sequence and using the previous identifications we get theexact sequence 0 → ( σ ∗ M ) ∨ → N ′ → E xt O ˜ X ′ ( σ ∗ S , O ˜ X ′ ) . Since the point p does not belong to the support of S , we get that the support of σ ∗ S is zerodimensional, therefore E xt O ˜ X ′ ( σ ∗ S , O ˜ X ′ ) is equal to zero. Hence we get R σ ∗ ( σ ∗ M ) ∨ ∼ = R σ ∗ N ′ . Since M is locally free and we obtain ˜ X ′ taking the blow up in the point p we get R σ ∗ (cid:16) ( σ ∗ M ) ∨ (cid:17) = R σ ∗ (cid:18)(cid:0) σ ∗ O r ˜ X (cid:1) ∨ (cid:19) = R σ ∗ O r ˜ X ′ = 0 . Hence R σ ∗ N ′ is equal to zero. (cid:3) Corollary 5.8.
Let X be a Stein normal surface. Let M be a reflexive O X -module. If the full sheafassociated to M at the minimal adapted resolution is special, the the full sheaf associated to M atthe minimal resolution of X is also special. Theorem 5.9.
Let X be a Stein normal surface. Let M be a reflexive O X -module. The module M is special if and only if the full sheaf associated with M at its minimal adapted resolution is special. Proof.
Denote by π : ˜ X → X the minimal resolution adapted to M and by π min : ˜ X min → X bethe minimal resolution of X .We need to prove that for any resolution ρ : ˆ X → X , the full sheaf ˆ M = ( ρ ∗ M ) ∨∨ is special(Definition 3.8). If the minimal resolution coincides with the minimal resolution adapted to M , then EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 37 by Proposition 5.7 we are done. Suppose that the minimal resolution and the minimal resolutionadapted to M do not coincide. By taking a finite succession of blowing ups in different points weobtain a resolution ˘ ρ : ˘ X → X such that it satisfies the following diagram˘ X ˜ X ˆ X ˜ X min oν πρ where ν and o are a composition of blowings up in points and ˘ ρ = ρ ◦ ν .By Proposition 5.7 the full sheaf associated to M in the resolution ˘ X is special. Again byProposition 5.7 the specialty defect of the full sheaf associated to M in the resolution ˆ X is less orequal to zero, and hence equal to 0. Therefore M is special. (cid:3) Definition 5.10.
Let ( X, x ) be a normal surface singularity, π : ˜ X → X a resolution, and F anirreducible component of the exceptional divisor. The minimal canonical order at F is defined to be (39) ord F ( K X ) := min (cid:0) { ord F ( div ( β ) : β ∈ H ( U, ω ˜ X ) } (cid:1) . Remark 5.11.
The following easy observations hold:(1) The minimal canonical order at F does not depend on the resolution where F appears.(2) In the Gorenstein case the minimal canonical order at F is the order at F of the Gorensteinform.(3) In the Gorenstein case the minimal canonical order of a divisor appearing at the minimalresolution is non-positive.(4) If F is a divisor and F ′ is another divisor obtained by blowing up F at a generic point thenwe have ord F ′ ( K X ) = ord F ( K X ) + 1 . Corollary 5.12.
Let ( X, x ) be a normal Gorenstein surface singularity. Then there exist non-freespecial reflexives modules. Proof.
A consequence of the third and fourth assertions of Remark 5.11 is the existence of aresolution π : ˜ X → ( X, x ) such that a component F of the exceptional divisor has minimal canonicalorder equal to 0. Let D be a irreducible smooth curve transverse to the exceptional divisor F ata generic point. We choose A = O D and consider ( ψ , ..., ψ r ) generators of A as a O X -module.In that case K ( A , ( ψ ,...,ψ r )) equals N , and hence we have the equality cond ( K ( A , ( ψ ,...,ψ r )) ) = 0. ByProposition 4.33 (1), if ( M , ( φ , ..., φ r )) is the pair associated with ( A , ( ψ , ..., ψ r )) by Theorem 4.18,the sheaf M is full.The sheaf M is special. Indeed, since the sections ( ψ , ..., ψ r ) generate A as a O X -module, thethird map in the exact sequence(40) 0 → π ∗ N → O rX ( ψ ,...,ψ r ) −→ π ∗ A → R π ∗ N → R π ∗ O r ˜ X → π : ˜ X → ( X, x ) is the minimal adapted resolution to themodule π ∗ M . Finally, by Theorem 5.9 the reflexive O X -module π ∗ M is special.Non-freeness holds because by construction M has non-trivial first Chern class. (cid:3) The next proposition explains the structure of degeneracy modules of special reflexive modulesfor sets of generic sections. .
Proposition 5.13.
Let X be a Stein normal surface with Gorenstein singularities. Let M be aspecial reflexive O X -module of rank r . Let ( φ , ..., φ r ) be r generic sections. Let C be the degeneracymodule of ( M, ( φ , ..., φ r )) , and let C be its support. Let n : ˜ C → C be the normalization. Then wehave the isomorphism C ∼ = n ∗ O ˜ C .Conversely, let C ⊂ X be a reduced curve and n : ˜ C → C be its normalization. Let ( ψ , ..., ψ r ) be a set of generators of n ∗ O ˜ C as a O X -module. Let ( M, ( φ , ..., φ r )) be the pair associated with ( n ∗ O ˜ C , ( ψ , ..., ψ r )) under the correspondence of Theorem 4.15. Then M is a special reflexive mod-ule. Proof.
Let π : ˜ X → X be the minimal adapted resolution to M and M be the associated full sheaf.Let ( A , ( ψ , ..., ψ r )) be the pair associated with ( M , ( φ , ..., φ r )) by Theorem 4.18. By Lemma 5.4we have the isomorphism A ∼ = O D for a smooth curve meeting the exceptional divisor transverselyat smooth points. By specialty and the exact sequence (40) we have the equality C = π ∗ A .For the converse let ( M, ( φ , ..., φ r )) be the pair associated with ( n ∗ O ˜ C , ( ψ , ..., ψ r )) under thecorrespondence of Theorem 4.15. Consider the minimal resolution π : ˜ X → X adapted to M . Let C be the strict transform of C and let A := n ∗ O ˜ C . By Proposition 4.33 (6), if ( M , ( φ , ..., φ r )) isthe pair associated with ( A , ( ψ , ..., ψ r )) under Theorem 4.18, then the O ˜ X -module M is full andspecial. By Proposition 4.36 we have that π ∗ M is isomorphic to M . Since M is special, the module M is special by Theorem 5.9. (cid:3) Let us record an interesting property of minimal adapted resolutions of special reflexive modules:
Proposition 5.14.
Let X be a Stein normal surface with Gorenstein singularities. Let M be aspecial reflexive O X -module of rank r . Let ( φ , ..., φ r ) be r generic sections. Let π : ˜ X → X be theminimal adapted resolution to M and M be the associated full sheaf. Let ( A , ( ψ , ..., ψ r )) be thepair associated with ( M , ( φ , ..., φ r )) by Theorem 4.18. The support of A is a smooth curve meetingthe exceptional divisor transversely at smooth points which are located at divisors where the minimalcanonical order vanishes. Proof.
We only need to prove that D only meets at components where the minimal canoni-cal order vanishes. Since π : ˜ X → X is the minimal adapted resolution we have the equality cond ( K ( A , ( ψ ,...,ψ r )) ) = (0 , ...,
0) by Proposition 5.5. This, together with the equality C = n ∗ O D (which follows from the previous proposition), implies that D only meets at components where theminimal canonical order vanishes. (cid:3) The decomposition in indecomposables and the irreducible components of thedegeneracy locus.
Here we determine the relation between the decomposition of a special reflexivemodule into indecomposables, and the decomposition of its generic degeneracy locus into irreduciblecomponents. Here we work in the local case instead of the more general Stein surface case: weconsider (
X, x ) to be a normal Gorenstein surface singularity.
Proposition 5.15.
Let ( X, x ) be a normal Gorenstein surface singularity. Let M be a special O X module of rank r and ( φ , ..., φ r ) be a set of r generic sections. Let π : ˜ X → X be any resolution.Let M be the full O ˜ X -module associated to M . Let ( C , ( ψ , ..., ψ r )) and ( A , ( ψ , ..., ψ r )) be theresults of applying Theorems 4.15 and 4.18 to ( M, ( φ , ..., φ r )) and ( M , ( φ , ..., φ r )) respectively.Suppose that M has no free direct factors. Then are natural bijections between the following sets(1) The indecomposable direct summands of M .(2) The indecomposable direct summands of M .(3) The irreducible components of the support of C . EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 39 (4) The irreducible components of the support of A . Proof.
The first and second sets are in a bijection via π ∗ . The third and fourth sets are in bijectionvia π . Now we show a bijection between the first and third set. For this is convenient to choose π : ˜ X → X to be the minimal resolution adapted to M (or at least dominating it).Let C be the support of C . By Lemma 5.4 the support C of A is the strict transform of C by π , and decomposes as a disjoint union C = ` kj =1 C j of k smooth curves meeting the exceptionaldivisor transversely at smooth points. By Proposition 5.13, we have the isomorphism C ∼ = π ∗ O ˜ C .We have the isomorphism O C = ⊕ kj =1 O C j . For each j let ( ψ j, , ..., ψ j,r j ) be a minimal system ofgenerators of n ∗ O ˜ C j as a O X -module. Applying Theorem 4.15 to the pair ( n ∗ O C , ( ψ , , ..., ψ k,r k )) weobtain a reflexive O X -module M ′ , which has no free factors by the minimality of the set of generatorsof n ∗ O C . If we denote by M ′ j the reflexive O X -module obtained by applying Theorem 4.15 to thepair ( n ∗ O C j , ( ψ j, , ..., ψ j,r j )) then we have the direct sum decomposition M ′ = ⊕ kj =1 M ′ j .By the proof of Theorem 4.15 we have that each of M and M ′ are isomorphic to the dual of themodule of relations of a minimal set of generators of n ∗ O C as O X -module. Hence M and M ′ areisomorphic.In order to finish the proof we have to show that each M ′ j is indecomposable. Let M ′ j = ⊕ m j v =1 M ′ j,v be the decomposition in indecomposable reflexive modules. Since M does not have freefactors no one of the factors is free. For each v = 1 , ..., m j choose a generic system of sections( φ j,v, , ..., φ j,v,n j,v ). Since the minimal resolution adapted to a reflexive module dominates theminimal resolution adapted to each of its direct factors we conclude that π : ˜ X → X dominates theminimal resolution adapted to M j . Denote by M ′ j,v be the full O ˜ X -module associated to M ′ j,v . Let( A j,v , ( ψ j,v, , ..., ψ j,v,n j,v )) by the pair associated with ( M ′ j,v , ( φ j,v, , ..., φ j,v,n j,v )) by Theorem 4.18.By Lemma 5.4, the support of A j,v is a disjoint union of smooth curves meeting the exceptionaldivisor transversely at smooth points. Such collection of curves is non-empty since the O ˜ X -module M ′ j,v is non-free.The full O ˜ X -module M associated to M is the direct sum of the modules M ′ j,v when j and v vary. Taking the union of the system of sections ( φ j,v, , ..., φ j,v,n j,v ) letting j and v vary we find asystem of sections of M , whose degeneracy locus is a disjoint union of smooth curves in ˜ X meetingthe exceptional divisor transversely at smooth points. Moreover there is at least a meting point foreach pair of indexes ( j, v ), since the full sheaf M ′ j,v is not free. Consequently, if at least one M ′ j isnot indecomposable the number of meeting points is strictly larger than k .On the other hand we know that the degeneracy locus of M for a system of generic sections isa disjoint union of smooth curves in ˜ X meeting the exceptional divisor transversely at k smoothpoints. The system of sections obtained as the union of ( φ j,v, , ..., φ j,v,m v ) letting j and v vary, maybe deformed continuously into a system of generic sections, and the corresponding degeneracy locideform flatly. Since it is impossible to deform a curve meeting transversely the exceptional divisorin strictly more than k points into a curve meeting transversely the exceptional divisor in precisely k points, we deduce that each M ′ j is indecomposable. (cid:3) The cohomology of full sheaves on normal Stein surfaces with Gorensteinsingularities
The main objective of this section is to prove to following theorem.
Theorem 6.1.
Let X be a Stein normal surface with Gorenstein singularities. Let M be a reflexive O X -module of rank r . Let π : ˜ X → X be a small resolution with respect to the Gorenstein form, let Z k be the canocical cycle at ˜ X , (see Defintion 2.1). Let M be the full O ˜ X -module associated to M . Let d be the specialty defect of M . Then we have the equality dim C ( R π ∗ M ) = rp g − [ c ( M )] · [ Z k ] + d. This theorem will be very important in the following section. It will allow us to prove that a fullspecial sheaf on a Gorenstein surface is determined by its first Chern class in the minimal adaptedresolution. An inmediate Corollary of the theorem and Proposition 5.14 shows how to use thecohomology of full sheaves as a resolution invariant for reflexive modules.
Corollary 6.2.
Let X be a Stein normal surface with Gorenstein singularities. Let M be a specialreflexive O X -module of rank r . Let π : ˜ X → X be a small resolution with respect to the Gorensteinform. Let M be the full O ˜ X -module associated to M . Then the resolution is the minimal adaptedresolution if and only if we have the equality dim C ( R π ∗ M ) = rp g . Proof. [Proof of Theorem 6.1] The Theorem is local in the base X . Therefore we may assumethat ( X, x ) is a normal Gorenstein germ. Then we have dualizing modules ω X = O X and ω ˜ X =Λ Ω X = π ! O X .The proof occupies the rest of the section, including some intermediate results, which we willsingle out as separate Lemmata and Propositions. Let us start with some preliminary work.Let M , π : ˜ X → X , Z k , M , r and d be as in the statement of the Theorem. Take r genericsections and consider the exact sequence obtained by the sections(41) 0 → O r ˜ X → M → A ′ → , and its dual(42) 0 → N → O r ˜ X → A → , where A = E xt O ˜ X ( A ′ , O ˜ X ).Since dualizing the first morphism of (42) we recover the first morphism of (41) back, we deducethat(43) A ′ ∼ = E xt O ˜ X ( A , O ˜ X ) . Applying the functor π ∗ − to the exact sequence (42) we get0 N O rX π ∗ A R π ∗ N R π ∗ O r ˜ X , where N is π ∗ N and, by Lemma 3.6, the module N is equal to the module M ∨ .The last exact sequence can be split as follows0 N O rX C , C π ∗ A D , D R π ∗ N R π ∗ O r ˜ X , where lenght( D ) = d .Dualizing the first and second exact sequence we obtain(44) 0 O rX M Ext O X ( C , O X ) 0 ,h EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 41 (45) 0 Ext O X ( π ∗ A , O X ) Ext O X ( C , O X ) Ext O X ( D , O X ) 0 .i Applying the functor π ∗ − to the exact sequence (41), using the identification (43) and comparingwith the exact sequence (44) we obtain the diagram(46) 0 O rX M Ext O X ( C , O X ) 00 O rX M π ∗ E xt O ˜ X ( A , O ˜ X ) R π ∗ O r ˜ X R π ∗ M hhId Id θ where Id is the identity and θ is the map that makes the diagram commute.Since ( X, x ) is Gorenstein and π : ˜ X → X is small with respect to the Gorenstein form we havethe exact sequence(47) 0 ω ˜ X O ˜ X O Z K .c Applying the functor π (cid:0) E xt O ˜ X ( A , − ) (cid:1) to the map c we obtain the map(48) π (cid:0) E xt O ˜ X ( A , − ) (cid:1) ( c ) : π (cid:0) E xt O ˜ X ( A , ω ˜ X ) (cid:1) → π (cid:0) E xt O ˜ X ( A , O ˜ X ) (cid:1) . Abusing notation, let us denote the previous map by c .Since the singularity ( X, x ) is Gorenstein, the ring O X is the dualizing module for the singularity[3, Theorem 3.3.7]. In this case the Grothendieck duality for the map π [15, Ch. VII] establish theisomorphism Rπ ∗ R H om ( − , ω ˜ X ) ∼ = R Hom O X ( Rπ ∗ − , O X ) . Applying this for A , and using Grothendieck spectral sequence for the composition of two functorswe obtain an isomorphism(49) g : π ∗ E xt O ˜ X ( A , ω ˜ X ) ∼ = Ext O X ( π ∗ A , O X ) . Now using (45), (46), (48) and (49), we have the diagram(50) π ∗ E xt O ˜ X ( A , ω ˜ X ) Ext O X ( π ∗ A , O X ) Ext O X ( C , O X ) π ∗ E xt O ˜ X ( A , O ˜ X ) π ∗ E xt O ˜ X ( A , ω ˜ X ) π ∗ E xt O ˜ X ( A , O ˜ X ) g i θId Idc Lemma 6.3.
The diagram commutes.
Proof.
Denote by c the map given by the composition θ ◦ i ◦ g . Consider the map f := ( c − c ).Since the map π : ˜ X → X is an isomorphism outside the exceptional divisor, we have that forany section s of π ∗ E xt O ˜ X ( A , ω ˜ X ), the section f ( s ) is supported in the exceptional divisor, hence f ( s ) ∈ H E (cid:16) E xt O ˜ X ( A , O ˜ X ) (cid:17) but this cohomology group is zero.Therefore for any section s of π ∗ E xt O ˜ X ( A , ω ˜ X ) we have that f ( s ) = 0 which it is equivalent tosay that the maps c and c coincide. (cid:3) Proposition 6.4.
We have the equality dim C ( R π ∗ M ) = rp g − dim C ( π ∗ E xt O ˜ X ( A , O Z k )) + dim C ( Ext O X ( D , O X )) . Proof.
Applying the functor H om ( − , − ) to the exact sequences (42) and (47) we get0 0 00 ω r ˜ X M ⊗ ω ˜ X E xt O ˜ X ( A , ω ˜ X ) 00 O r ˜ X M E xt O ˜ X ( A , O ˜ X ) 00 O rZ k M ⊗ O Z k E xt O ˜ X ( A , O Z k ) 00 0 0 c Applying the functor π ∗ − to the last commutative diagram we get0 0 00 π ∗ ω r ˜ X π ∗ ( M ⊗ ω ˜ X ) π ∗ E xt O ˜ X ( A , ω ˜ X ) 0 0 00 O rX M π ∗ E xt O ˜ X ( A , O ˜ X ) R π ∗ O r ˜ X R π ∗ M π ∗ O rZ k π ∗ ( M ⊗ O Z k ) π ∗ E xt O ˜ X ( A , O Z k ) 0 0 h αc By this diagram we getdim C ( R π ∗ M ) = rp g − dim C (Im( α )) , dim C ( π ∗ E xt O ˜ X ( A , O Z k )) = dim C ( π ∗ E xt O ˜ X ( A , O ˜ X ) /π ∗ E xt O ˜ X ( A , ω ˜ X )) , (51)and(52) dim C (Im α ) = dim C ( π ∗ E xt O ˜ X ( A , O ˜ X ) / ker α ) = dim C ( π ∗ E xt O ˜ X ( A , O ˜ X ) / Im h ) . Now by (46) we have Im h = Ext O X ( C , O X ). EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 43
Hence by the previous equality, (52), (45) and (49) we getdim C (Im α ) = dim C ( π ∗ E xt O ˜ X ( A , O ˜ X ) / Im h )= dim C ( π ∗ E xt O ˜ X ( A , O ˜ X ) / Ext O X ( C , O X ))= dim C ( π ∗ E xt O ˜ X ( A , O ˜ X ) / Ext O X ( π ∗ A , O X )) − dim C (Ext O X ( D , O X ))= dim C ( π ∗ E xt O ˜ X ( A , O ˜ X ) /π ∗ E xt O ˜ X ( A , ω ˜ X )) − dim C (Ext O X ( D , O X )) . (53)Let c and c be the morphisms given in the diagram (50). Now by (53) and by Lemma 6.3 wehave dim C (Im α ) = dim C ( π ∗ E xt O ˜ X ( A , O ˜ X ) /π ∗ E xt O ˜ X ( A , ω ˜ X )) − dim C (Ext O X ( D , O X ))= dim C ( π ∗ E xt O ˜ X ( A , O ˜ X ) / Im c ) − dim C (Ext O X ( D , O X ))= dim C ( π ∗ E xt O ˜ X ( A , O ˜ X ) / Im c ) − dim C (Ext O X ( D , O X ))= dim C ( π ∗ E xt O ˜ X ( A , O ˜ X ) /π ∗ E xt O ˜ X ( A , ω ˜ X )) − dim C (Ext O X ( D , O X )) . (54)Therefore by (51) and (54) we get the desired equality. (cid:3) By the Proposition 6.4 to prove Theorem 6.1 it is enough to prove the equalitiesdim C ( π ∗ E xt O ˜ X ( A , O Z k )) = [ c ( M )] · [ Z k ] . dim C (Ext O X ( D , O X )) = d. The following lemma gives us the first equality.
Lemma 6.5.
Let A and A be two Cohen-Macaulay O ˜ X -modules of dimension one such that A is contained in A , the quotient A / A is finitely supported and the support of each sheaf intersectsthe exceptional divisor in finitely many points. Then we have the equality dim C ( π ∗ E xt O ˜ X ( A , O Z K )) = dim C ( π ∗ E xt O ˜ X ( A , O Z K )) . Proof.
Let A and A as in the statement. We have the exact sequence0 → A → A → A / A → . Applying the functor H om O ˜ X ( − , O Z K ) to the last exact sequence we get(55) 0 H om O ˜ X ( A / A , O Z K ) H om O ˜ X ( A , O Z K ) H om O ˜ X ( A , O Z K ) E xt O ˜ X ( A / A , O Z K ) E xt O ˜ X ( A , O Z K ) E xt O ˜ X ( A , O Z K ) E xt O ˜ X ( A / A , O Z K ) E xt O ˜ X ( A , O Z K ) E xt O ˜ X ( A , O Z K ) 0Since A and A are Cohen-Macaulay sheaves of dimension one and the support of each sheafintersects the exceptional divisor finitely we have H om O ˜ X ( A , O Z K ) = H om O ˜ X ( A , O Z K ) = 0 , E xt O ˜ X ( A , O Z K ) = E xt O ˜ X ( A , O Z K ) = 0 . (56)The second equality uses Auslander-Buchbaum formula and the fact that each A i has depth 1. Since all the sheaves in (55) are supported in a finite set we can work locally, therefore we assumethat O ˜ X is C [ x, y ] and Z K is O ˜ X / ( f ) for some function f .Now by (55) and (56) we just need to prove the following equality(57) dim C ( E xt O ˜ X ( A / A , O Z K )) = dim C ( E xt O ˜ X ( A / A , O Z K )) . Consider the following resolution of O Z K (58) 0 O ˜ X O ˜ X O Z K . · f Applying the functor H om O ˜ X ( A / A , − ) to the last exact sequence we get(59) 0 H om O ˜ X ( A / A , O ˜ X ) H om O ˜ X ( A / A , O ˜ X ) H om O ˜ X ( A / A , O Z K ) E xt O ˜ X ( A / A , O ˜ X ) E xt O ˜ X ( A / A , O ˜ X ) E xt O ˜ X ( A / A , O Z K ) E xt O ˜ X ( A / A , O ˜ X ) E xt O ˜ X ( A / A , O ˜ X ) E xt O ˜ X ( A / A , O Z K ) 0Now since the support of A / A is zero dimensional, we have by Theorem 3.1 E xt O ˜ X ( A / A , O ˜ X ) = 0 . By the previous equality and the exact sequence (59) we get0 → E xt O ˜ X ( A / A , O Z K ) → E xt O ˜ X ( A / A , O ˜ X ) → E xt O ˜ X ( A / A , O ˜ X ) → E xt O ˜ X ( A / A , O Z K ) → . Taking C -dimensions we immediately obtain:dim C ( E xt O ˜ X ( A / A , O Z K )) = dim C ( E xt O ˜ X ( A / A , O Z K )) . (cid:3) Proposition 6.6.
The equality dim C ( π ∗ E xt O ˜ X ( A , O Z k )) = [ c ( M )] · [ Z k ] holds. Proof.
By the previous Lemma it is enough to assume that A is isomorphic to O C , where C isthe support of A . A direct computation of E xt O ˜ X ( O C , O Z k ) gives the result. (cid:3) Proposition 6.7.
The equality dim C ( Ext O X ( D , O X )) = dim C ( D ) = d holds. Proof.
Using Theorem 3.1 one easily reduces by induction on dim C ( D ) to the case D = C p , where C p is the skyscraper sheaf at a point p with stalk C . A direct computation shows that case. (cid:3) The proof of Theorem 6.1 is complete now. (cid:3)
EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 45 The classification and structure of special reflexive modules
The combinatorial classification.
Let (
X, x ) be a normal surface singularity. Lemma 5.4allows to define the resolution graph of a reflexive module. Denote by M a reflexive O X -module, π : ( ˜ X, E ) → ( X, x ) the minimal adapted resolution to M , M the full sheaf associated to M and r the rank of M . Take r generic sections of M and consider the exact sequence given by the sections0 → O r ˜ X → M → A ′ → . By Lemma 5.4 the sheaf A ′ is isomorphic to O D , where D is a smooth curve meeting theexceptional divisor transversely at smooth points. We construct a graph as follows:(1) Let G oM be the dual graph of ˜ X of the minimal good resolution that dominates π , weightedwith the self-intersection and the genus of each component (see [29]).(2) In each vertex v i , add as many arrows as the first Chern class of M intersects the exceptionaldivisor E i . Call the resulting decorated graph G M . Definition 7.1.
The resolution graph G M of the module M is the graph described in the previousconstruction. In the next theorem we characterize combinatorially resolution graphs of special modules overGorenstein surface singularities. By a negative plumbing graph we mean the dual graph of a goodresolution of a surface singularity. The property of being numerically Gorenstein only depends onthe plumbing graph. In this case there is a canonical cycle with integral coefficients (see [29]).
Theorem 7.2.
Let G be a negative definite plumbing graph, such that to some of its verticesthere are a finite number of arrows attached. There is a Gorenstein surface singularity ( X, x ) and aspecial reflexive module whose resolution graph is isomorphic to G if and only if each of the followingproperties is satisfied:(1) the graph is numerically Gorenstein.(2) if a vertex has genus , self intersection − and has at most two neighboring vertices, thenit supports at least arrow.(3) if a vertex supports arrows then its coefficient in the canonical cycle equals . Proof.
Property (1) is necessary because Gorenstein implies numerically Gorenstein. Property(2) holds by the minimality of the good resolution dominating the minimal adapted resolution.Property (3) is a direct consequence of Proposition 5.14.Conversely, let G be a graph satisfying all the properties. By [28] there is a Gorenstein normalsurface singularity ( X, x ) who has a resolution with plumbing graph equal to the result of deletingthe arrows of G . Let π : ˜ X → X be such a resolution. Let D ⊂ ˜ X be a smooth curvette meeting theexceptional divisor E transversely at smooth points, and so that for each vertex v of V the numberof components of D meeting the irreducible component E v of E corresponding to v , is exactly thenumber of arrows attached to v .Define A := O D and let ψ , ..., ψ r be a set of generators of π ∗ A as a O X -module. Since A is equalto O D and the canonical order at the components E v met by D is 0, choosing D generic enoughwe conclude that cond ( K ( A , ( ψ ,...,ψ r )) ) = (0 , ..., M , ( φ , ..., φ r )) is the result of applying the bijection of Theorem 4.18 to ( A , ( ψ , ..., ψ r )) then M is full. It is also special since ψ , ..., ψ r generate A as a O X -module (same argument than at theproof of Corollary 5.12). By Proposition 5.5 the resolution π : ˜ X → X is the minimal resolution adapted to π ∗ M , and byTheorem 5.9 the module π ∗ M is special. It is clear by construction that the resolution graph of π ∗ M equals G . (cid:3) A consequence of the previous Theorem and Proposition 5.15 is the following corollary
Corollary 7.3.
Let G be a negative definite plumbing graph, such that to some of its vertices thereare a finite number of arrows attached. There is a Gorenstein surface singularity ( X, x ) and anindecomposable special reflexive module whose resolution graph is isomorphic to G if and only if theconditions of the previous Theorem hold and in addition G has only one arrow. The first Chern class of a module at its minimal adapted resolution.
Let X be a Steinnormal surface with Gorenstein singularities. Here we study the relation of a reflexive O X -moduleand its first Chern class in the Picard group of its minimal adapted resolution. We show that if themodule is special then the first Chern class determines the module, providing a vast generalizationof the corresponding result of Artin and Verdier for rational double points [2].Let M be a reflexive O X -module of rank r . Let π : ˜ X → X be the minimal resolution adaptedto M , denote by E the exceptional divisor. Let M be the full O ˜ X -module associated to M . Thefirst Chern class of M in P ic ( ˜ X ) is the class determined by the determinant bundle L := det ( M ). Proposition 7.4.
The full O ˜ X -module M is an extension of the determinant line bundle L by O r − X . Proof.
Take r generic sections ( φ , ..., φ r ) of M and consider the exact sequence given by thesections(60) 0 → O r ˜ X → M → A ′ → . Since the resolution is the minimal adapted resolution we have that A ′ is isomorphic to O D ,where D is a smooth curve meeting the exceptional divisor E transversely at smooth points byLemma 5.4.Locally in a trivializing open subset U of the locally free sheaf M we have that the sections canbe written as follows Q = q q . . . q r ... ... ... ... q r q . . . q rr where each q ij is an element of O ˜ X ( U ).Therefore p belongs to D if and only if the determinant of Q ( p ) is equal to zero.Since D is smooth, the matrix Q must have at least r − r − φ , ..., φ r ) we may assume that these sections are ( φ , ..., φ r − ). These sections give us the exactsequence(61) 0 → O r − X → M → L → , where L is the line bundle det( M ). (cid:3) Now we assume that M is special and ( X, x ) is Gorenstein and prove stronger properties.
Lemma 7.5.
The dimension of R π ∗ L is p g . EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 47
Proof.
By Proposition 5.14 the resolution π : ˜ X → X is small with respect to the Gorensteinform (hence the canonical cycle Z K is non-negative), and moreover D does not meet the supportof Z K . Therefore we have Tor O ˜ X ( O D , O Z K ) = 0 , O D ⊗ O Z K = 0 . By these equalities, applying − ⊗ O Z K the exact sequence(62) 0 → O ˜ X → L → O D → , we get(63) O Z K ∼ = L ⊗ O Z K . Now applying the functor π ∗ − to the exact sequence (62) and using the last isomorphism weobtain0 π ∗ O ˜ X π ∗ L π ∗ O D R π ∗ O ˜ X R π ∗ L R π ∗ O Z K R π ∗ L ⊗ O Z K R π ∗ O ˜ X and R π ∗ O Z K are isomorphic we conclude that R π ∗ O ˜ X and R π ∗ L have the same dimension. (cid:3) Theorem 7.6.
Let X be a Stein normal surface with Gorenstein singularities. Let M be a special O X -module without free factors. Let π : ˜ X → X be the minimal resolution adapted to M , and M the full O ˜ X -module associated to M . The module M (and equivalently M ) is determined by its firstChern class in Pic ( ˜ X ) . Proof.
Applying the functor π ∗ − to the exact sequence (61) we get(64)0 π ∗ O r − X π ∗ M π ∗ L R π ∗ O r − X R π ∗ M R π ∗ L . Since dim C ( R π ∗ M ) = rp g by Corollary 6.2 anddim C ( R π ∗ L ) = dim C ( R π ∗ O ˜ X ) = p g by Lemma 7.5 , we get that the exact sequence (64) split as follows(65) 0 π ∗ O r − X π ∗ M π ∗ L . Therefore π ∗ M ∈
Ext O X (cid:0) π ∗ L , O r − X (cid:1) . Since the module π ∗ M is reflexive and without freefactors we conclude the proof by Lemma 1.9.ii in [2] (this Lemma globalizes to the Stein surfacesituation that we are considering here). (cid:3) The classification of special reflexive modules on Gorenstein surface singularities.
Before we state and prove the classification theorem we need the following lemma.
Lemma 7.7.
Let ( X, x ) be a normal Gorenstein surface singularity and π : ˜ X → X be a resolutionwhich is small with respect to the canonical form such that for some irreducible component E i ofthe exceptional divisor E we have E i Supp ( Z K ) . If D and D are two irreducible curvettes,each one transverse to E i at regular points of E , then we have an isomorphism of line bundles O ˜ X ( − D ) ∼ = O ˜ X ( − D ) . Proof.
We want to prove that O ˜ X ( − D + D ) is isomorphic to O ˜ X .Consider the exponential exact sequence0 Z O ˜ X O ∗ ˜ X . expApplying the functor π ∗ to the previous exact sequence we get(66) . . . H ( ˜ X, Z ) H ( ˜ X, O ˜ X ) H ( ˜ X, O ∗ ˜ X ) H ( ˜ X, Z ) 0 . exp δ We know that the Picard group of ˜ X is H ( ˜ X, O ∗ ˜ X ) and the morphism δ is given by taking thefirst Chern class in cohomology.By hypothesis we know that δ ( O ˜ X ( − D + D )) = 0. By the exact sequence (66) we get thatthere exist an element f in H ( ˜ X, O ˜ X ) such that the line bundle given by exp( f ) is isomorphic to O ˜ X ( − D + D ).Denote by E the exceptional divisor of π . By the location of the curvettes D and D , an easyˇCech cohomology computation shows that there exists a finite Stein cover U = { U i } i ∈ I so that f ∈ H ( ˜ X, O ˜ X ) = ˇ H ( U , O ˜ X ) is represented by a 1-cocycle [ f i,j ] with f i,j ∈ O ˜ X ( U i ∩ U j ) so that f i,j = 0 unless U i ∩ U j ∩ E is included in E i \ Sing ( E ).Since the resolution is small with respect to the Gorenstein form, we have the exact sequence(27) 0 → ω ˜ X → O ˜ X → O Z K → . Applying the functor π ∗ − and by Grauert-Riemenschneider Vanishing Theorem we get that thehomomorphism H ( ˜ X, O ˜ X ) → H ( ˜ X, O Z K )is an isomorphism. Since the image of the ˇCech cocycle [ f i,j ] under this isomorphism is obvi-ously zero for having empty support ( Z k does not have support in E i ), we deduce that f = 0 in H ( ˜ X, O ˜ X ). This implies that O ˜ X ( − D + D ) is the trivial line bundle. (cid:3) Definition 7.8.
Let ( X, x ) be a normal surface singularity and π : ˜ X → X be a resolution. Anyirreducible component E i of the exceptional divisor is called a divisor over X . Remark 7.9.
Let E , . . . , E n be a collection of divisors over X . Then there exists a unique minimalresolution π : ˜ X → X such that E , . . . , E n are irreducible components of the exceptional divisor. Now we present the classification theorem:
Theorem 7.10.
Let ( X, x ) be a Gorenstein surface singularity. Then there exists a bijectionbetween the following sets:(1) The set of special O X -modules without free factors up to isomorphism. EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 49 (2) The set of finite pairs ( E , n ) , . . . , ( E l , n l ) where each E i is a divisor over X and n i is apositive integer, such the minimal resolution given by Remark 7.9 is small with respect tothe Gorenstein form and the Gorenstein form does not have any pole in the components E , . . . , E l . Proof.
Let M be a special O X -module and π : ˜ X → X be the minimal resolution adapted to M with exceptional divisor E = S li =1 E i . Denote by M the full sheaf associated to M and by n j = c ( M ) · E j for j = 1 , . . . , l . We associate to the module M the pairs ( E , n ) , . . . , ( E k , n k )such that n j is different form zero.In order to prove the surjectivity of the previous assignment consider ( E , n ) , . . . , ( E l , n l ) whereeach E i is a divisor over X and n i is a positive integer and denote by π : ˜ X → X be the resolutiongiven by Remark 7.9. The divisors are so that π : ˜ X → X is small with respect to the Gorensteinform, and the coefficient of the canonical cycle at each of them vanishes. For each positive integer n j take a smooth curvette D j with n j irreducible components such that D j intersects only theirreducible component E j and the intersection is transverse. Denote by D = D ` · · · ` D l . Let( ψ , ..., ψ r ) be a minimal set of generators of π ∗ O D as a O X -module. Since the E i ’s are not at thesupport of the canonical cycle we have that the minimal conductor cond ( K ( O D , ( ψ ,...,ψ r )) ) equals 0.Let ( M , ( φ , ..., φ r )) be pair associated to ( O D , ( ψ , ..., ψ r )) by the correspondence of Theorem 4.18.By Proposition 4.33, (1) the module M is full. Since ( ψ , ..., ψ r ) generate of π ∗ O D as a O X -module,we have that M is special. By Theorem 5.9 the module M := π ∗ M is special. The equality cond ( K ( O D , ( ψ ,...,ψ r )) ) = 0 implies that π : ˜ X → X is the minimal resolution adapted to M byProposition 5.5. By construction, the previous assignment applied to M gives ( E , n ) , . . . , ( E l , n l ).In order to prove surjectivity we need that M does not have free factors. If M has free factors wewrite M = M ⊕ O aX , where M is without free factors. Then the previous assignment applied to M also gives ( E , n ) , . . . , ( E l , n l ) and surjectivity is proven.The injectivity follows from Theorem 7.6 and Lemma 7.7. (cid:3) Remark 7.11.
If the union of curvettes D in the previous proof is chosen generic then the obtainedmodule M does not have free factors. Proof.
By the previous Theorem there is a unique reflexive O X -module M without free factorsassociated with the set of pairs ( E , n ) , . . . , ( E l , n l ). Let r be its rank and ( φ , ..., φ r ) be a set ofgeneric sections. Let ( C , ( ψ , ..., ψ r )) be the pair associated with ( M , ( φ , ..., φ r )) by Theorem 4.15.Since M is without free factors and the sections are generic ( ψ , ..., ψ r ) is a minimal set of gen-erators of C . Taking π : ˜ X → X the minimal adapted resolution to M and using specialty andProposition 4.36, we have that C = π ∗ O D for a curve D as in the previous proof.Let D ′ be a union of curvettes as in the proof of the previous Theorem. Let ( ψ , ..., ψ r ) be aminimal set of generators of π ∗ O D ′ as a O X -module. Let ( M ′ , ( φ , ..., φ r )) be the pair associatedwith ( π ∗ O D ′ , ( ψ , ..., ψ r )) by Theorem 4.15. According with Proposition 4.36, if M is the moduleassociated with D ′ by the previous proof, then we have the equality M = M ′ .The minimal number of generators of π ∗ O D ′ as a O X -module is upper semi-continuous underdeformation of D ′ . Then the minimal number of generators among all choices of D ′ as in theprevious proof is r : if it were smaller the module M would have rank smaller than r .Now, if D ′ is chosen generic the module M associated with D ′ by the previous proof has rank r and contains M as a direct factor, hence it is equal to M . (cid:3) Corollary 7.12.
Let ( X, x ) be a Gorenstein surface singularity. Then there exists a bijectionbetween the following sets: (1) The set of special, indecomposable O X -modules up to isomorphism.(2) The set of irreducible divisors E over x , such at any resolution of X where E appears, theGorenstein form has not either zeros or poles along E . Proof.
It follows immediately from Theorem 7.10 and Proposition 5.15. (cid:3)
Notice that if (
X, x ) is a rational double point, then the previous Corollary is the McKay corre-spondence given by Artin and Verdier [2].
Corollary 7.13 ([2]) . Let ( X, x ) be a rational double point and denote by π : ˜ X → X the minimalresolution with exceptional divisor E = S li =1 E i . Then there exists a bijection between the followingsets:(1) The set of reflexive, indecomposable O X -modules up to isomorphism.(2) The set of irreducible divisors E i where E i is an irreducible component of the exceptionaldivisor E . Proof.
Since the singularity is a rational double point the following two sets are the same:(1) The set of irreducible divisors E i where E i is an irreducible component of the exceptionaldivisor E .(2) The set of irreducible divisors E ′ i where E ′ i is a divisor over X , such that the minimal reso-lution given by Lemma 7.9 is small with respect to the Gorenstein form and the Gorensteinform does not have any pole in the components E ′ i .Now the Corollary follows immediately from Corollary 7.12 and from the fact that any reflexivemodule on a rational double point singularity is special. (cid:3) Deformations of reflexive modules and full sheaves
In the next sections we study deformations of reflexive modules. We treat simultaneously de-formations over complex spaces and over complex algebroid germs (spectra of noetherian complete C -algebras).8.1. The deformation functors.
We assume basic knowledge on Deformation Theory. We fol-low [18] as a basic reference. In order to fix terminology we recall some known definitions.
Notation 8.1.
Let
Y → S be flat morphism of two complex spaces, and y, s be points in eachof them. Let M be a O Y -module. We will use the notation M | s := M ⊗ O S ( O S / m s ) , where m s denotes the maximal ideal at s . Clearly M | s is a O Y s module, where Y s is the fibre of Y over s .Furthermore ( M | s ) y will denote the stalk of M | s at y . Definition 8.2.
Let Y be a either complex space or an algebroid germ and M be a O Y -module.(1) A deformation of ( Y, M ) over a germ of complex space (or an algebroid germ) ( S, s ) is atriple ( Y , M , ι ) where Y is a flat deformation of Y over S , M is a O Y -module which is flatover S , and ι is an isomorphism from M to the fibre M | s .(2) A deformation fixing Y of M over a germ of complex space ( S, s ) is a deformation of ( Y, M ) over ( S, s ) such that Y is the trivial deformation Y × S .(3) Given a flat morphism Y → S , a flat family of modules on Y is a O Y -module M which isflat over S .(4) A flat family of O Y -modules fixing Y is a flat family of modules on Y × S . EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 51
Deformations of a pair (
Y, M ) form a contravariant functor
Def Y , M from the category of germsof complex spaces to the category of sets in the usual way: morphisms of germs are transformedinto mappings of set via the pull-back of deformation. Likewise deformations of M fixing the baseform a contravariant functor Def M . If we restrict to the category of spectra of Artinian C -algebras,we can view the functor as a covariant functor from Artinian C -algebras to sets. It is easy to checkthat Schlessinger conditions ( H − ( H
2) of Theorem 16.2 of [18] are satisfied for these two functors.Let us remark that reflexiveness is an open property (see the next Lemma), and hence thedeformation notion of Definition 8.2 is adequate as a deformation notion of reflexive sheaves.
Lemma 8.3.
Let σ : Y → S be a flat family of normal surfaces. Let M be a family of moduleson Y . Let y ∈ Y be a point. Suppose that the O Y σ ( y ) ,y -module ( M | σ ( y ) ) y is reflexive. There existsan open neighborhood U of y in Y such that for any y ′ ∈ U the O Y σ ( y ′ ) ,y ′ -module ( M | σ ( y ′ ) ) y ′ isreflexive. Proof.
Reflexiveness is equivalent to being Cohen-Macaulay of dimension 2. An straightforwardadaptation of EGA IV [14, § M | s ′ ) x ′ is Cohen-Macaulay ofdimension 2 is open. (cid:3) Remark 8.4.
Since being Gorenstein is equivalent to ask that the dualizing sheaf is an invertiblesheaf, it is easy to prove that being Gorenstein is also an open property. In our work we do not usethat property, so we omit the proof.
On the other hand, if we work at the resolution, fullness is not an open property, so one needsto restrict the deformations of Definition 8.2 in order to get a good notion of deformations and flatfamilies of full sheaves.
Definition 8.5 (Laufer [26]) . Let
X → S be a flat family of normal Stein surfaces. A very weaksimultaneous resolution of X → S is a proper birational morphism Π : ˜
X → X satisfying(1) ˜ X is flat over S ,(2) For any closed point s ∈ S the morphism Π | s : ˜ X s → X s is a resolution of singularities.We will use the following notation: for any s ∈ S denote the restriction Π | ˜ X s by Π s : ˜ X s → X s . Lemma 8.6.
Let
X → S be a flat family of normal Stein surfaces and Π : ˜
X → X a very weaksimultaneous resolution. Let M be a O ˜ X -module which is flat over O S . Then the first 2 of thefollowing 3 conditions are equivalent and imply the third:(1) R Π ∗ M is flat as O S -module in a neighborhood of a point s ∈ S ;(2) the natural map (Π ∗ M ) | s → (Π s ) ∗ ( M| s ) is an isomorphism;(3) Π ∗ M is flat as O S -module in a neighborhood of a point s ∈ S .If S is the spectrum of an artinian algebra then the third condition is equivalent to the first twoconditions.For any morphism φ : S ′ → S let Π ′ : ˜ X × S S ′ → X × S S ′ , ˜ ψ : ˜ X × S S ′ → ˜ X and ψ : X × S S ′ → X be the natural maps. If the previous conditions are satisfied for any s ∈ S then the natural map (67) ψ ∗ ( R Π ∗ M ) → R (Π ′ ) ∗ (cid:16) ˜ ψ ∗ M (cid:17) , is an isomorphism. Proof.
The proof is an adaptation of the methods of Section III. 12 of [16]. The main differenceis that in our case the morphism Π is not projective, and that Π ∗ M is not coherent over S . Now we explain the changes needed in each of the results from Hartshorne book that we will use; wenumerate the results as Hartshorne does. Our base S in Hartshorne’s setting is the spectrumof a ring A . Easy adaptations of the proofs allow to modify Hartshorne statements as follows:Proposition 12.1 is true without modification. The complex L • of Proposition 12.2 is boundedabove, L i is finitely generated and free over A if i >
0, and only flat over A if i ≤
0. Proposition12.4 is true asking W i to be flat over A instead of projective, if i = 0, and not asking Q to befinitely generated if i = 0. Proposition 12.5 is true as stated. Corollary 12.6 is true if one asks T ( A ) to be flat instead of projective. Proposition 12.10 works as stated; the only point of the proofof Proposition 12.10 that needs some care is the following: the Theorem on Formal Functions ([16],Chapter III, Theorem 11.1) is used only for the 0-th cohomology. This theorem assumes projectivityfor the morphism Π, but for 0-th cohomology the theorem works without this hypothesis.Now let us proceed to the proof using Hartshorne language. There are only two functors, T and T . Thus T is left exact and T right exact. Condition (1) translates in the flatness of T ( A ),which by the adapted Corollary 12.6 of Hartshorne is equivalent to the exactness of T . This isequivalent to the exactness of T , and by the adapted Corollary 12.6 this implies the flatness of T ( A ), which is exactly Condition (3).Condition (2) is a particular case of the isomorphism (67). If the first condition hold then T isexact and hence T is right exact. Proposition 12.5 of Hartshorne implies the isomorphism (67).Then a direct application of Proposition 12.10 of [16]) gives that Condition (2) implies the rightexactness of T . This implies the exactness of T and, by the adapted Corollary 12.6, Condition(1) holds.Suppose that S is the spectrum of an artinian algebra. The Artinian Principle of Exchangeof [36] gives the equivalence between the first and third conditions. (cid:3) Definition 8.7.
Let X be a normal Stein surface, π : ˜ X → X be a resolution and M be a full O ˜ X -module.(1) A deformation of ( ˜ X, X, M ) over a germ of complex space ( S, s ) is a cuadruple ( ˜ X , X , M , ι ) ,where X is a flat deformation of X over ( S, s ) , there is a proper birational morphism Π : ˜
X → X which is a very weak simultaneous resolution, M is O ˜ X -module which is flatover S , and ι is an isomorphism from M to M| s . A deformation fixing X of ( ˜ X, X, M ) is a deformation ( ˜ X , X , M , ι ) where X is the trivial deformation. A deformation fixing( ˜ X, X ) of ( ˜ X, X, M ) is a deformation where X and ˜ X are trivial deformations.(2) A full deformation of ( ˜ X, X, M ) is a deformation ( ˜ X , X , M , ι ) such that R Π ∗ M is flat as O S -module.(3) Given a morphism Π : ˜
X → X as above, a family of full modules on ˜ X is a triple ( ˜ X , X , M ) where M is a O ˜ X -module which is flat over S , the O S -module R Π ∗ M is flat, and the O ˜ X s -module M| s is full for all s ∈ S . A flat family fixing X and/or ( ˜ X, X ) is defined in theobvious way. The reader may have noticed that while in the definition of full family we ask that the O ˜ X s -module M| s is asked to be full for all s ∈ S , we do not ask the same property for full deformationsover a germ ( S, s ). The reason is the following proposition, which shows that fullness is an openproperty in full deformations.
Proposition 8.8.
Let X be a normal Stein surface, π : ˜ X → X be a resolution and M be a full O ˜ X -module. Let ( ˜ X , X , M , ι ) be a full deformation of ( ˜ X, X, M ) over ( S, s ) . There exists an openneighborhood W of s ∈ S such that for any s ′ ∈ U the O ˜ X s -module M| s is full. EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 53
Proof.
We will use the characterization of Proposition 3.4.Since M is locally free, by flatness we have that M is also locally free. Then M| s ′ is locally freefor any s ′ ∈ S .Since R Π ∗ M is flat over S , by Lemma 8.6 we have that Π ∗ M is flat over S and also the equality(68) (Π ∗ M ) | s ′ = (Π | ˜ X s ′ ) ∗ M| s ′ . This implies that Π ∗ M is a flat deformation of the reflexive module (Π | ˜ X s ) ∗ M (here we use that M is full). By openness of reflexivity (Lemma 8.3), for any s ′ is a neighborhood of s in S we havethat (Π | ˜ X s ′ ) ∗ M| s ′ is a reflexive O X s ′ -module.Let E s ′ be the exceptional divisor at ˜ X s ′ . We have the local cohomology exact sequence0 → H E s ′ ( M| s ′ ) → H ( ˜ X s ′ , M| s ′ ) → H ( ˜ X s ′ \ E s ′ , M| s ′ ) → H E s ′ ( M| s ′ ) → H ( ˜ X s ′ , M| s ′ ) . The morphism H ( ˜ X s ′ , M| s ′ ) → H ( ˜ X s ′ \ E s ′ , M| s ′ ) is surjective since H ( ˜ X s ′ , M| s ′ ) is a reflexive O X s ′ -module and ˜ X s ′ \ E s ′ is identified with X s ′ minus a finite set of points (where the modificationtakes place). Hence H E s ′ ( M| s ′ ) → H ( ˜ X s ′ , M| s ′ ) is injective as needed.Let ( φ , ..., φ m ) be a collection of global sections of M , which almost generate it except at afinite set Z ⊂ ˜ X . By Equality (68) there exist ( φ , ..., φ m ), global sections of M which specializeto ( φ , ..., φ m ) at the fibre ˜ X over s . Let Z ⊂ ˜ X denote the locus where the sections ( φ , ..., φ m )do not generate M . Then we have the equality Z ∩ ˜ X = Z , and as a consequence there is an openneighborhood U of s in S such that Z ∩ ˜ X s ′ is finite for any s ′ ∈ S . Over U we have that M| s ′ isgenerically generated by global sections. (cid:3) At the following proposition we introduce the relevant deformation functors.
Proposition 8.9.
Deformations of ( ˜
X, X, M ) form a contravariant functor from the category ofgerms of complex spaces to the category of sets. We denote it by Def ˜X , X , M . The functors ofdeformations fixing X and ( ˜ X, X ) are denoted respectively by Def ˜X , M and Def M .Full deformations of ( ˜ X, X, M ) form a contravariant functor denoted by FullDef ˜X , X , M . Thefunctors of full deformations fixing X and ( ˜ X, X ) are denoted respectively by FullDef ˜X , M and FullDef M .These functors, restricted to the artinian basis, may be seen as a covariant functor from thecategory of artinian algebras to sets. Proof.
The only non-trivial point is to prove that pullback of full deformations are full deforma-tions, but this follows from the isomorphism (67). (cid:3)
It is again easy to check that Schlessinger conditions ( H − ( H
2) are satisfied for the functorsdefined at the previous proposition.
Proposition 8.10.
Let π : ˜ X → X be a resolution of a normal Stein surface, let M be a full O ˜ X -module and M = π ∗ M be its associated reflexive module. The push forward operation Π ∗ alongthe resolution map defines a natural transformation from FullDef ˜X , X , M to Def X , M .Analogous statements holds for the deformation functors fixing X and/or ˜ X , and for families offull modules. Proof.
For the assertion about deformations let ( ˜ X , X , M , ι ) be a full deformation of ( ˜ X, X, M ).Flatness of the push down Π ∗ M holds by Lemma 8.6. The isomorphism from (Π ∗ M ) | s to M isobtained composing the natural isomorphism (Lemma 8.6) with the isomorphism Π ∗ ι . The remaining assertions are proved similarly. (cid:3)
Remark 8.11.
Although there is a bijection between full sheaves and reflexive sheaves [19] , as wewill see below the transformation
FullDef M → Def M is not an isomorphism of functors. Proposition 8.12.
Let π : ˜ X → X be a resolution of a normal Stein surface. Let M be a full O ˜ X -module and M = π ∗ M be its associated reflexive O X -module. Then the deformation functors Def ˜X , X , M , Def X , M , Def M , FullDef ˜X , X , M , FullDef X , M , FullDef M , Def X , M and Def M haveminiversal deformations.Let ( ˜ X , X , M , ι ) be the miniversal deformation of Def ˜X , X , M , then the miniversal deformationof FullDef ˜X , X , M is the stratum of the flattening stratification of R Π ∗ M containing the origin.Analogous statements hold for Def X , M / FullDef X , M and Def M / FullDef M . Proof.
By Theorem 16.2 of [18] in order to prove the existence of miniversal deformation we onlyhave to prove Schlessinger ( H
3) condition.For
Def M this amounts to prove that Ext O X ( M, M ) is finite dimensional (see for example [35])Since M is reflexive, it is locally free at X \ Sing ( X ). Therefore Ext O X ( M, M ) is finite dimensionalas needed. The functor
Def X , M fibres over the functor of deformations of X , with fibre the functor Def M . Since both the base and fibre functors satisfy ( H Def X , M does it.Since M is locally free the local to global spectral sequence shows that Ext O ˜ X ( M , M ) is isomor-phic to R π ∗ H om O ˜ X ( M , M ), which is finite dimensional. Hence the Schlessinger condition ( H Def M . Since the functor FullDef M is a sub-functor of Def M , the con-dition ( H
3) also holds for it. For the functors
Def ˜X , X , M , Def X , M , FullDef ˜X , X , M , FullDef X , M we use fibration of functors arguments as before.The flattening stratification statement is by versality and definition of the functors. (cid:3) Let us give a basic proposition that we will use later.
Proposition 8.13.
Let π : ˜ X → X be a resolution of a normal Stein surface, let ( ˜ X , X , M , ι ) be afull deformation of ( ˜ X, X, M ) over ( S, s ) . Then Π ∗ (cid:18) M ∨ (cid:19) = (cid:0) Π ∗ M (cid:1) ∨ . Proof.
The proof is an adaptation of Lemma 3.5 and Lemma 3.6. Let ω ˜ X| S be relative thecanonical sheaf over ˜ X . The sheaf M⊗ ω ˜ X| S is locally free, hence it is flat over S and ( M⊗ ω ˜ X | S ) | s ∼ = M ⊗ ω ˜ X , then by Lemma 3.5 and the Cohomology and Base Change Theorem ([16], Chapter III,Theorem 12.11) we have R Π ∗ ( M ⊗ ω ˜ X ) = 0.Now the proof is parallel to the proof of Lemma 3.6. (cid:3) The correspondence for deformations at families of normal Stein surfaces withGorenstein singularities.Definition 8.14.
Let X be a normal Stein surface, let C be a rank generically reduced Cohen-Macaulay O X -module of dimension , and a system of generators ( ψ , ..., ψ r ) of C as O X module.Let M be a reflexive O X -module of rank r and ( φ , ..., φ r ) be r sections.(1) A deformation of ( X, C , ( ψ , ..., ψ r )) over a germ ( S, s ) is a cuadruple ( X , C , ( ψ , ..., ψ r ) , ρ ) ,where ( X , C , ρ ) is a deformation of ( X, C ) and ( ψ , ..., ψ r ) are sections of C , which via theisomorphism ρ restrict to ( ψ , ..., ψ r ) over s . A deformation fixing X is a deformation suchthat X is the trivial deformation of X . EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 55 (2) A deformation of (
X, M, ( φ , ..., φ r )) over a germ ( S, s ) is a cuadruple ( X , M , ( φ , ..., φ r ) , ι ) ,where ( X , M , ι ) is a deformation of M and ( φ , ..., φ r ) are sections of M , which via theisomorphism ι restrict to ( φ , ..., φ r ) over s . A deformation fixing X is a deformation suchthat X is the trivial deformation of X .The deformations defined above, together with the pullback operation, form two contravariant func-tors Def ( ψ ,...,ψ r ) X , C and Def ( φ ,...,φ r ) X , M from the category of germs of complex spaces to the categoryof sets. Deformations fixing the base form two contravariant functors called Def ( ψ ,...,ψ r ) C and Def ( φ ,...,φ r ) M . Now we extend the correspondences at the Stein space (Theorem 4.15) to get an isomorphismof functors. First we need the following lemma, which can be found in Proposition 0.1 of [5], andalso in [34], Proposition 2.2.
Lemma 8.15.
Let Y and ( S, s ) be a complex space and a germ of complex space. Let Y be a flatdeformation of Y over ( S, s ) . Let F be a O Y -module. Let m s be the maximal ideal at s . For any O S -module L and any index i there is a natural morphism (69) φ i ( L ) : E xt i O Y ( F , O Y ) ⊗ O S L → E xt i O Y ( F , O Y ⊗ O S L ) . The following assertions hold:(1) If φ i ( O S,s / m s ) is surjective, then φ i ( L ) is an isomorphism for any L .(2) Assume that φ i ( O S,s / m s ) is surjective. Then φ i − ( O S,s / m s ) is surjective if and only if E xt i O Y ( F , O Y ) is flat over S .(3) If E xt i O Y ( F , O Y ⊗ O S,s / m s ) = 0 then E xt i O Y ( F , O Y ⊗ O S L ) = 0 for all L . Lemma 8.16.
In the situation of the preceeding Lemma, if F is flat over S then we have theisomorphism (70) E xt i O Y ( F , O Y ⊗ O S,s / m s ) ∼ = E xt i O Y ( F ⊗ O
S,s / m s , O Y ⊗ O S,s / m s ) . Proof.
It is straightforward if one compute E xt using a free resolution of F . (cid:3) Now we are able to extend the correspondence given by Theorem 4.15. From now and for thisand the following section we will always assume that X is a normal Stein surface with Gorensteinsingularities. Theorem 8.17.
Let X be a normal Stein surface with Gorenstein singularities. Let ( M, ( φ , .., φ r )) be a reflexive O X -module of rank r and a collection r nearly generic sections. Let ( C , ( ψ , ..., ψ r )) be the rank 1 generically reduced Cohen-Macaulay O X -module of dimension with the collection ofgenerators obtained from ( M, ( φ , .., φ r )) by the direct correspondence at the X (see Theorem 4.15).The correspondence defined at Theorem 4.15 extends to define an isomorphism between the functors Def ( φ ,...,φ r ) X , M and Def ( ψ ,...,ψ r ) X , C . The isomorphism restricts to an isomorphism between Def ( φ ,...,φ r ) M and Def ( ψ ,...,ψ r ) C . Proof.
Let (
S, s ) be a germ of complex space. Let ( X , M , ( φ , ..., φ r ) , ι ) be a deformation of( X, M, ( φ , ..., φ r )) over ( S, s ). Let O r X → M be the morphism induced by the sections, denote itscokernel by ( C ) ′ . Then we have the exact sequence:(71) 0 → O r X → M → ( C ) ′ → . The flatness of M over S , and the fact that the first mapping specializes over s to an injection,implies the flatness of ( C ) ′ over S , by using the Local Criterion of Flatness. The specialization ofthe sequence to the fibre over s is the exact sequence of O X -modules(72) 0 → O rX → M | s → ( C ) ′ | s → , and ι induces an isomorphism between this sequence and the exact sequence(73) 0 → O rX → M → C ′ → , induced by the sections ( φ , ..., φ r ).The dual of the last sequence is the sequence(74) 0 → N → O rX → C → , where the last morphism of the sequence gives rise to the generators ( ψ , ..., ψ r ) of C (see the proofof Theorem 4.15).Dualize the sequence (71) with respect to O X and obtain the exact sequence0 → H om O X ( N , O X ) → O r X → E xt O X (( C ) ′ , O X ) → E xt O X ( M , O X ) . Define N := H om O X ( M , O X ). We claim that E xt O X ( M , O X ) vanishes. Indeed, since M isCohen-Macaulay of dimension 2 we have the vanishing E xt O X ( M, O X ) = 0 by Theorem 3.1. ByLemma 8.16 we have the isomorphism E xt O X ( M , O X ) ∼ = E xt O X ( M, O X ) = 0. This vanishing,together with Lemma 8.15 proves the claim.As a consequence of the claim we have the exact sequence(75) 0 → N → O r X → E xt O X (( C ) ′ , O X ) → . We define C := E xt O X (( C ) ′ , O X ). We claim that the following assertions hold:(1) the O X -module C is flat over S .(2) The specialization N | s → O rX of the first morphism of the sequence coincides with N → O rX .Assume the claim. The second assertion induces an identification of O rX → C with O rX → C| s .Let ρ denote the isomorphism C → C| s . The second morphism of Sequence (75) induces a collectionof sections ( ¯ ψ , ..., ¯ ψ r ). The first assertion shows that ( X , C , ( ψ , ..., ψ r ) , ρ ) is a deformation of( X, C , ( ψ , ..., ψ r )) over ( S, s ).Let us prove the claim. Since C ′ is Cohen-Macaulay of dimension 1 and X has Gorensteinsingularities, Theorem 3.1 implies the vanishing E xt i O X ( C ′ , O X ) = 0 if i ≥
2. Then, since ( C ) ′ isflat over S , using Lemma 8.16 we obtain the vanishing E xt i O X (( C ) ′ , O X ) = 0 if i ≥
2. Now we willapply Lemma 8.15 repeatedly for F = ( C ) ′ : by the vanishing E xt O X (( C ) ′ , O X ) = 0 we deduce that φ ( O S,s / m s ) is surjective; the Lemma 8.15 (1) shows that E xt O X (( C ) ′ , O X ) vanishes (hence it is flatover S ). By Lemma 8.15, (2) we have that φ ( O S,s / m s ) is surjective. Lemma 8.15, (2) shows nowthat C = E xt O X (( C ) ′ , O X ) is flat over S if and only if φ ( O S,s / m s ) is surjective. This surjectivityholds since the target of this map vanishes because ( C ) ′ has proper support. This shows Assertion(1) of the claim.For Assertion (2) we apply Lemma 8.15 for F = M . We need to show the isomorphism N | s ∼ = N ,but this follows from Lemma 8.15, (1), if we show that φ ( O S,s / m s ) is surjective. By Lemma 8.15,(2), this is reduced to prove the vanishing of E xt O X ( M , O X ⊗ O S O S,s / m s ) and the flatness over S of E xt O X ( M , O X ). The vanishing holds because E xt O X ( M , O X ⊗ O S O S,s / m s ) is isomorphic to E xt O X ( M, O X ) by flatness of M and Lemma 8.16, and the second module vanishes by Theorem 3.1. EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 57
The flatness of E xt O X ( M , O X ) holds because this module also vanishes (apply Lemma 8.16 andLemma 8.15, (3) and (1) as before, starting from the vanishing of E xt O X ( M, O X )).In order to have a natural transformation from Def ( φ ,...,φ r ) X , M to Def ( ψ ,...,ψ r ) X , C we have to showthat the construction commutes with pullbacks. This follows from Lemma 8.15, (1), if we show theisomorphism C| s ∼ = E xt O X (( C ) ′ , O X ⊗ O S O S,s / m s ). By the flatness of ( C ) ′ over S and Lemma 8.16the second module is isomorphic to C , and then the desired isomorphism becomes the already provenidentity C| s ∼ = C .Now we define the inverse natural transformation from Def ( ψ ,...,ψ r ) X , C to Def ( φ ,...,φ r ) X , M .Let ( X , C , ( ψ , ..., ψ r ) , ρ ) be a deformation of ( X, C , ( ψ , ..., ψ r )) over ( S, s ). Consider the exactsequence induced by the sections:(76) 0 → N → O r X → C → . The flatness of C over S implies the flatness of N over S . The specialization of the sequence to thefibre over s is the exact sequence of O X -modules(77) 0 → N | s → O rX → C| s → , and ρ induces an isomorphism between this sequence and the exact sequence(78) 0 → N → O rX → C → , induced by the generators ( ψ , ..., ψ r ). The dual of the last sequence is the sequence(79) 0 → O rX → M → E xt O X ( C , O X ) → , where the sections ( φ , ..., φ r ) are induced by the first map of the sequence (see the proof of Theo-rem 4.15).Dualize the sequence (76) with respect to O X and obtain the exact sequence(80) 0 → O r X → M → E xt O X ( C , O X ) → , where we define M := H om O X ( N , O X ).We claim that the following assertions hold:(1) the O X -module M is flat over S .(2) The specialization O rX → M | s of the first morphism of the sequence is isomorphic to O rX → M .Assume the claim. The second assertion induces an isomorphism from O rX → M to O rX → M | s .Let ι denote the isomorphism M → M | s . The first morphism of Sequence (80) induces a collectionof sections ( ¯ φ , ..., ¯ φ r ). The first assertion shows that ( X , M , ( φ , ..., φ r ) , ι ) is a deformation of( X, C , ( φ , ..., φ r )) over ( S, s ).The proof of the claim is an application of Lemmata 8.16 and 8.15 competely analogous to theproof of the previous two claim in this proof. We omit it here.In order to have a natural transformation from
Def ( ψ ,...,ψ r ) X , C to Def ( φ ,...,φ r ) X , M we have to show thatthe construction commutes with pullbacks, but this is an application of Lemmata 8.16 and 8.15,similar to the proof that the transformation from Def ( φ ,...,φ r ) X , M to Def ( ψ ,...,ψ r ) X , C commutes withpullbacks.Finally we need to check that the correspondences that we have just defined are inverse to eachother, but this is clear by construction.It is obvious that the isomorphisms that we have defined restrict to an isomorphism between Def ( φ ,...,φ r ) M and Def ( ψ ,...,ψ r ) C . (cid:3) The correspondence for deformations at simultaneous resolutions of families ofnormal Stein Surfaces with Gorenstein singularities.Definition 8.18.
Let X be a normal Stein surface and π : ˜ X → X be a resolution. Let ( A , ( ψ , ..., ψ r )) be a rank generically reduced Cohen-Macaulay O ˜ X -module of dimension , and a system of gen-erators as O ˜ X -module satisfying the Containment Condition. Let M be a full O ˜ X -module of rank r and ( φ , ..., φ r ) be r nearly generic sections.(1) A deformation of ( ˜ X, X, A , ( ψ , ..., ψ r )) over a germ ( S, s ) is a quintuple ( ˜ X , X , A , ( ψ , ..., ψ r ) , ρ ) ,where ( ˜ X , X , A , ρ ) is a deformation of ( ˜ X, X, A ) , and ( ψ , ..., ψ r ) are sections of A , whichvia the isomorphism ρ restrict to ( ψ , ..., ψ r ) over s .(2) A specialty defect constant deformation of ( A , ( ψ , ..., ψ r )) over a germ ( S, s ) is a defor-mation ( ˜ X , X , A , ( ψ , ..., ψ r ) , ρ ) such that the cokernel D of the natural mapping Π ∗ O r ˜ X → Π ∗ A induced by the sections ( ψ , ..., ψ r ) is flat over S .(3) A (full) deformation of ( ˜ X, X, M , ( φ , ..., φ r )) over a germ ( S, s ) is a quintuple ( ˜ X , X , M , ( φ , ..., φ r ) , ι ) , where ( ˜ X , X , M , ι ) is a (full) deformation of ( ˜ X, X, M ) and ( φ , ..., φ r ) are sections of M ,which via the isomorphism ι restrict to ( φ , ..., φ r ) over s .The deformations defined above give rise to functors Def ( ψ ,...,ψ r ) ˜X , X , A , SDCDef ( ψ ,...,ψ r ) ˜X , X , A , Def ( φ ,...,φ r ) ˜X , X , M and FullDef ( φ ,...,φ r ) ˜X , X , M . The obvious restricted deformation functors fixing ˜ X or ˜ X and X are denotedby Def ( ψ ,...,ψ r ) X , A , SDCDef ( ψ ,...,ψ r ) X , A , Def ( φ ,...,φ r ) X , M , FullDef ( φ ,...,φ r ) X , M , Def ( ψ ,...,ψ r ) A , SDCDef ( ψ ,...,ψ r ) A , Def ( φ ,...,φ r ) M and FullDef ( φ ,...,φ r ) M . Lemma 8.19.
The assignements defined in the previous definition are in fact contravariant func-tors.
Proof.
The only non-trivial assertion is for
SDCDef ( ψ ,...,ψ r ) ˜X , X , A : we need to prove the preservationof the flatness of the cokernel D under pullback. This holds because the formation of cokernelscommutes with pullbacks and flatness is preserved by pullbacks. (cid:3) Lemma 8.20.
In the setting of the previous definition, suppose that ( ψ , ..., ψ r ) generate A as a O X -module (that is, they generate π ∗ A ). Then the functors Def ( ψ ,...,ψ r ) ˜X , X , A and SDCDef ( ψ ,...,ψ r ) ˜X , X , A coincide. Proof.
The specialization of the cokernel of Π ∗ : O r ˜ X → π ∗ X to the fibre over s vanishes, andhence the cokernel vanishes as well by Nakayama Lemma. (cid:3) Theorem 8.21.
Let X be a normal Stein surface with Gorenstein singularities. Let ( A , ( ψ , ..., ψ r )) be a Cohen-Macaulay O ˜ X -module of dimension , and a system of generators as O ˜ X -module satis-fying the Containment Condition, let ( M , ( φ , ..., φ r )) be a full O ˜ X -module of rank r with r nearlygeneric sections. Suppose that the pairs ( A , ( ψ , ..., ψ r )) and ( M , ( φ , ..., φ r )) are related by thecorrespondence of Theorem 4.18. There is an isomorphism of functors between FullDef ( φ ,...,φ r ) ˜X , X , M and SDCDef ( ψ ,...,ψ r ) ˜X , X , A . The isomorphism restricts to isomorphisms between FullDef ( φ ,...,φ r ) X , M and SDCDef ( ψ ,...,ψ r ) X , A , and between FullDef ( φ ,...,φ r ) M and SDCDef ( ψ ,...,ψ r ) A . EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 59
Proof.
The restriction statements are obvious after finding an isomorphism between
FullDef ( φ ,...,φ r ) ˜X , X , M and SDCDef ( ψ ,...,ψ r ) ˜X , X , A .The proof of this isomorphism has two parts. In the first we find an isomorphism between Def ( φ ,...,φ r ) ˜X , X , M and Def ( ψ ,...,ψ r ) ˜X , X , A . In the second part we prove that, under the defined isomorphismfull deformations correspond to specialty defect constant deformations. Part 1 . The first part runs parallel to the proof of Theorem 8.17, so we skip many details andinclude only what is needed to define the isomorphism and to set up the notation for the proof ofPart 2.Let (
S, s ) be a germ of complex space. Let ( ˜ X , X , M , ( φ , ..., φ r ) , ι ) be an element of Def ( φ ,...,φ r ) ˜X , X , M ( S, s ).Consider the exact sequence induced by the sections:(81) 0 → O r ˜ X → M → ( A ) ′ → . The flatness of M over S , and the fact that the first mapping specializes over s to an injection,implies the flatness of ( A ) ′ over S , by using the local criterion of flatness. The specialization of thesequence to the fibre over s is the exact sequence of O ˜ X -modules(82) 0 → O r ˜ X → M| s → ( A ) ′ | s → , and ι induces an isomorphism between this sequence and the exact sequence(83) 0 → O r ˜ X → M → ( A ) ′ → , induced by the sections ( φ , ..., φ r ).The dual of this last sequence is the sequence(84) 0 → N → O r ˜ X → A → , where the last morphism of the sequence gives rise to the generators ( ψ , ..., ψ r ) of A as a O ˜ X -module(see the proof of Theorem 4.18).Dualize the sequence (81) with respect to O ˜ X and obtain the exact sequence(85) 0 → N → O r ˜ X → A → , where N := H om O ˜ X ( M , O ˜ X ) and A := E xt O ˜ X (( A ) ′ , O ˜ X ). The following two assertions are provedlike the corresponding ones in the proof of Theorem 8.17.(1) the O ˜ X -module A is flat over S .(2) The specialization N | s → O rX of the first morphism of the sequence is isomorphic to N →O rX .The second assertion induces an isomorphism from O r ˜ X → A to O r ˜ X → A| s . Let ρ denote theisomorphism A → A| s . The second morphism of Sequence (85) induces a collection of sections( ¯ ψ , ..., ¯ ψ r ). We have that ( ˜ X , X , A , ( ψ , ..., ψ r ) , ρ ) is a deformation of ( ˜ X, X, A , ( ψ , ..., ψ r )) over( S, s ) by the first assertion. So, we have defined map
Def ( φ ,...,φ r ) ˜X , X , M ( S, s ) → Def ( ψ ,...,ψ r ) ˜X , X , A ( S, s ). Inorder to have a natural transformation of functors from
Def ( φ ,...,φ r ) ˜X , X , M to Def ( ψ ,...,ψ r ) ˜X , X , A we need toshow that the transformation which we have defined commutes with pullbacks. This is analogousto the corresponding statement in the proof of Theorem 8.17. Now we define the inverse natural transformation from
Def ( ψ ,...,ψ r ) ˜X , X , A to Def ( φ ,...,φ r ) ˜X , X , M . Let( ˜ X , X , A , ( ψ , ..., ψ r ) , ρ ) be a deformation of ( ˜ X, X, A , ( ψ , ..., ψ r )) over ( S, s ). Consider the ex-act sequence induced by the sections:(86) 0 → N → O r ˜ X → A → . The flatness of A over S implies the flatness of N over S . The specialization of the sequence to thefibre over s is the exact sequence of O ˜ X -modules(87) 0 → N | s → O r ˜ X → A| s → , and ρ induces an isomorphism between this sequence and the exact sequence(88) 0 → N → O rX → A → . induced by the generators ( ψ , ..., ψ r ). The dual of this last sequence is the sequence(89) 0 → O r ˜ X → M → E xt O ˜ X ( A , O ˜ X ) → , where the sections ( φ , ..., φ r ) are induced by the first map of the sequence (see the proof of Theo-rem 4.18).Dualize the sequence (86) with respect to O ˜ X and obtain the exact sequence(90) 0 → O r ˜ X → M → E xt O ˜ X ( A , O ˜ X ) → , where M := H om O ˜ X ( N , O ˜ X ).The following two assertions are proved like the corresponding ones in the proof of Theorem 8.17:(1) the O ˜ X -module M is flat over S .(2) The specialization O r ˜ X → M| s of the first morphism of the sequence is isomorphic to O r ˜ X → M .The second assertion induces an isomorphism from O r ˜ X → M to O r ˜ X → M| s . Let ι de-note the isomorphism M → M| s . The first morphism of Sequence (90) induces a collection ofsections ( ¯ φ , ..., ¯ φ r ). The first assertion shows that ( ˜ X , X , M , ( φ , ..., φ r ) , ι ) is a deformation of( ˜ X, X, M , ( φ , ..., φ r )) over ( S, s ). Part 2 . Let ( ˜ X , X , M , ( φ , ..., φ r ) , ι ) be an element of FullDef ( φ ,...,φ r ) ˜X , X , M ( S, s ). We have toshow that the quintuple ( ˜ X , X , A , ( ψ , ..., ψ r ) , ρ ) associated to it in Part 1 is a specialty constantdeformation. In this part the Gorenstein condition plays an important role.For this we have to show that the cokernel D of the map Π ∗ O r ˜ X → Π ∗ A is flat over S . Denoteby C the image of the same map. We have the exact sequences(91) 0 → Π ∗ N → O r X → C → , (92) 0 → C → Π ∗ A → D → . Now we specialize Sequence (92) at s . If we denote the maximal ideal of s by m s , and use theflatness of A over S , we have the exact sequence0 → T or O S ( D , O S / m s ) → ( C ) | s → (Π ∗ A ) | s → ( D ) | s → . We claim that the second morphism of the sequence can be identified with the injective morphism
C → π ∗ A , where C is the O X -module spanned by ( ψ , ..., ψ r ). If the claim is true we deduce thevanishing of T or O S ( D , O S / m s ). This implies the flatness of D using the local criterion of flatness. EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 61
In order to prove the claim we produce natural identifications (Π ∗ A ) | s ∼ = π ∗ A and ( C ) | s ∼ = C ,which show that the morphism ( C ) | s → (Π ∗ A ) | s is injective, yielding the desired vanishing.For the first identification notice that Π ∗ A coincides with A with the O X -module structureobtained by restriction of scalars; similarly π ∗ A coincides with A with the O X -module structureobtained by restriction of scalars. Then the needed identification is the isomorphism ρ definedabove.The second identification is a bit more involved (at the moment we do not even know theflatness of C over S ). We proceed like in Section 6 to obtain a diagram similar to (46). Dualizingsequence (91) we obtain the sequence0 → O r X → H om O X (Π ∗ N , O X ) → E xt O X (cid:0) C , O X (cid:1) → . Applying Π ∗ to Sequence (81) and using the isomorphism ( A ) ′ ∼ = E xt O ˜ X ( A , O ˜ X ), which comes fromthe fact that the natural transformations are inverse to each other, we obtain the exact sequence0 → O r X → Π ∗ M → Π ∗ E xt O ˜ X (cid:0) A , O ˜ X (cid:1) → R Π ∗ O r ˜ X → R Π ∗ M → . We have the chain of equalities H om O X (Π ∗ N , O X ) = H om O X (Π ∗ H om O ˜ X ( M , O ˜ X ) , O X ) == H om O X ( H om O X (Π ∗ M , O X ) , O X ) = Π ∗ M . The first is because H om O ˜ X ( M , O ˜ X ) ∼ = N , the second is a consequence of Proposition 8.13, and thethird is a consequence of Proposition 8.10 and the fact that flat deformations of reflexive modulesare reflexive.The last two exact sequences, together with the previous identifications yields the followingcommutative diagram:(93)0 O r X H om O X (Π ∗ N , O X ) E xt O X (cid:0) C , O X (cid:1) O r X Π ∗ M Π ∗ E xt O ˜ X (cid:0) A , O ˜ X (cid:1) R Π ∗ O r ˜ X R Π ∗ M hhId Id θ where θ is the only map making the diagram commutative.Since we have the isomorphism H om O ˜ X ( N , O ˜ X ) ∼ = M , dualizing the sequence (85), and usingthe exact sequence (81) we obtain that E xt O ˜ X (cid:0) A , O ˜ X (cid:1) is isomorphic to ( A ′ ), and, as O S -module,Π ∗ ( A ′ ) is equal to ( A ′ ), which is flat over S . Then, the exactness of the lower row of the diagramand the flatness of R Π ∗ M imply that E xt O X (cid:0) C , O X (cid:1) is flat over S .As a consequence, specializing the first row of diagram (93) over s we obtain the exact sequence0 → O rX → M → (cid:0) E xt O X (cid:0) C , O X (cid:1)(cid:1) | s → . Comparing with the first row of diagram (46) we obtain the isomorphism(94) (cid:0) E xt O X (cid:0) C , O X (cid:1)(cid:1) | s ∼ = E xt O X ( C , O X ) . We have the chain of isomorphisms: E xt O X ( E xt O X (cid:0) C , O X (cid:1) , O X ) ∼ = E xt O X ( (cid:0) E xt O X (cid:0) C , O X (cid:1)(cid:1) | s =0 , O X ) ∼ = ∼ = E xt O X ( E xt O X ( C , O X ) , O X ) = 0 . The first isomorphism is due to the flatness of E xt O X (cid:0) C , O X (cid:1) and Lemma 8.16, the second isomor-phism is because of Equation (94), and the vanishing is due to the fact that C is Cohen-Macaulayof dimension 1, Theorem 3.1 and the Gorenstein condition.The last vanishing shows, applying Lemma 8.15 (3), (2), and (1) for F = E xt O X (cid:0) C , O X (cid:1) , theisomorphism(95) E xt O X ( E xt O X (cid:0) C , O X (cid:1) , O X ) | s ∼ = E xt O X ( E xt O X (cid:0) C , O X (cid:1) , O X ) . Dualizing the first row of diagram (93) we obtain the exact sequence0 → H om O X ( H om O X (Π ∗ N , O X ) , O X ) → O r X → E xt O X ( E xt O X (cid:0) C , O X (cid:1) , O X ) →→ E xt O X ( H om O X (Π ∗ N , O X ) , O X ) . We have the chain of equalities H om O X ( H om O X (Π ∗ N , O X ) , O X ) == H om O X ( H om O X ( H om O X (Π ∗ M , O X ) , O X ) , O X ) == H om O X (Π ∗ M , O X ) = Π ∗ N . The first and third equalities are applications of Proposition 8.13 and the second is because a tripledual coincides with a single dual. We also have E xt O X ( H om O X (Π ∗ N , O X ) , O X ) = E xt O X (Π ∗ M , O X ) = 0 . The first equality has been shown in the chain of equalities prior to diagram (93), and the vanishingfollows by an application of Lemmata 8.16 and 8.15, the fact that Π ∗ M is a flat deformation ofa reflexive O X -module and the Gorenstein condition. After these identifications the last exactsequence becomes: 0 → Π ∗ N → O r X → E xt O X ( E xt O X (cid:0) C , O X (cid:1) , O X ) → , and this gives, by comparison with Exact Sequence (91) the isomorphism(96) C ∼ = E xt O X ( E xt O X (cid:0) C , O X (cid:1) , O X ) . The following chain of isomorphisms gives the needed identification:( C ) | s ∼ = E xt O X ( E xt O X (cid:0) C , O X (cid:1) , O X ) | s ∼ = E xt O X ( E xt O X (cid:0) C , O X (cid:1) , O X ) ∼ = ∼ = E xt O X (( E xt O X (cid:0) C , O X (cid:1) ) | s , O X ) ∼ = E xt O X ( E xt O X ( C , O X ) , O X ) ∼ = C . The first isomorphism is by (96); the second by (95); the third by flatness of E xt O X (cid:0) C , O X (cid:1) andLemma 8.16; the fourth by (94); the fifth is by Theorem 3.1, using that C is Cohen-Macaulay ofdimension 1.We have proven that the pair ( ˜ X , X , A , ( ψ , ..., ψ r ) , ρ ) is a specialty defect constant deformationof the pair ( ˜ X, X, A , ( ψ , ..., ψ r )) over ( S, s ).In order to finish Part 2 of the proof we let ( ˜ X , X , A , ( ψ , ..., ψ r ) , ρ ) be a specialty constant defor-mation of ( ˜ X, X, A , ( ψ , ..., ψ r )) over ( S, s ), and consider ( ˜ X , X , M , ( φ , ..., φ r ) , ι ), the deformationassigned by the isomorphism of functors from Def ( ψ ,...,ψ r ) ˜X , X , A to Def ( φ ,...,φ r ) ˜X , X , M . We have to prove that( ˜ X , X , M , ( φ , ..., φ r ) , ι ) is an element of FullDef ( φ ,...,φ r ) ˜X , X , M ( S, s ). For this we need to show that R Π ∗ M is flat over S . EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 63
As before denote by C the image of Π ∗ O X → Π ∗ A and consider the exact sequences (91) and (92).Denote by D the image of π ∗ O X → π ∗ A , and consider the exact sequences(97) 0 → π ∗ N → O rX → C → , (98) 0 → C → π ∗ A → D → . Sequence (92), and the flatness of Π ∗ A and D implies the flatness of C over ( S, s ). Observe thatΠ ∗ A and D specialize over s to π ∗ A and D . Consequently, specializing Sequence (92) over s andcomparing with Sequence (98) we conclude the isomorphism(99) C| s ∼ = C . We have the chain of isomorphisms E xt O X ( C , O X ) ∼ = E xt O X ( C| s , O X ) ∼ = E xt O X ( C , O X ) = 0 . The first isomorphism is by flatness of C and Lemma 8.16, the second follows from (99), and thevanishing by Theorem 3.1, the fact that C is Cohen-Macaulay of dimension 1 and the Gorensteincondition. Then, by Lemma 8.15 we have the isomorphism(100) E xt O X (cid:0) C , O X (cid:1) | s ∼ = E xt O X ( C , O X ) . Observe also the vanishing H om O X ( C , O X ) = 0 . Using the last isomorphism and the vanishing,Lemma 8.15 (2) implies that E xt O X (cid:0) C , O X (cid:1) is flat over S .Specializing the first row of Diagram (93) over s , we obtain the exact sequence0 → O rX → Π ∗ M| s → E xt O X (cid:0) C , O X (cid:1) | s → . Using the identification (100) and comparing with the sequence0 → O rX → π ∗ M → E xt O X ( C , O X ) → , obtained by dualizing Sequence (97), we deduce the isomorphism Π ∗ M| s ∼ = π ∗ M . Hence Condition2 of Lemma 8.6 holds and this concludes the proof. (cid:3) The previous theorem, together with Theorem 4.18 gives the following set of corollaries:
Corollary 8.22.
Assume that ( A , ( ψ , ..., ψ r )) is a Cohen-Macaulay O ˜ X -module of dimension with a collection of generators as a O ˜ X -module satisfying the Containment Condition. Let ( ˜ X , X , A , ( ψ , ..., ψ r ) , ρ ) be a specialty defect constant deformation of ( A , ( ψ , ..., ψ r )) over a germ ( S, s ) . Then for any s ′ ∈ S the Cohen-Macaulay O ˜ X s ′ -module with generators ( A| s ′ , ( ψ | s ′ , ..., ψ r | s ′ )) satisfies the Containment Condition. Remark 8.23.
Assume that ( A , ( ψ , ..., ψ r )) is a Cohen-Macaulay O ˜ X -module of dimension witha collection of generators as a O ˜ X -module satisfying the Containment Condition. Let ( ˜ X , X , A , ( ψ , ..., ψ r ) , ρ ) be a specialty defect constant deformation of ( A , ( ψ , ..., ψ r )) over a germ ( S, s ) . Applying the corre-spondence defined in Theorem 8.21 we obtain a full deformation of full O ˜ X -modules whose specialtydefect is constant. Corollary 8.24.
The specialty defect is constant in a full deformation of a full sheaf.
Proof.
Given a full sheaf M , let ( φ , ..., φ r ) be generic sections. For any deformation ( M , ι ) of M over ( S, s ) let ( φ , ..., φ r ) be an extension of the sections over ( S, s ). Applying the correspondenceof Theorem 8.21 to ( M , ( φ , ..., φ r ) , ι ) we obtain a speciality defect constant deformation. Apply the inverse correspondence to get back ( M , ( φ , ..., φ r ) , ι ) and use the pre-vious Remark. (cid:3) Finally we need to compare the isomorphism between the functors
Def ( φ ,...,φ r ) X , M and Def ( ψ ,...,ψ r ) X , C with the isomorphism between the functors FullDef ( φ ,...,φ r ) ˜X , X , M and SDCDef ( ψ ,...,ψ r ) ˜X , X , A .The results we need are the following: Proposition 8.25.
Let X be a normal Stein surface with Gorenstein singularities, let ( M, ( φ , ..., φ r )) be a reflexive O X -module of rank r together with r generic sections, π : ˜ X → X be a resolutionand M be the associated full O ˜ X -module. Let ( A , ( ψ , ..., ψ r )) be the result of applying the cor-respondence of Theorem 4.18 to ( M , ( φ , ..., φ r )) , and ( C , ( ψ ′ , ..., ψ ′ r )) the result of applying thecorrespondence of Theorem 4.15 to ( M, ( φ , ..., φ r )) .Consider a full deformation ( ˜ X , X , M , ( φ , ..., φ r ) , ι ) of ( ˜ X, X, M , ( φ , ..., φ r )) over a base S . Let ( X , M , ( φ , ..., φ r ) , ι ) be the deformation of ( X, M, ( φ , ..., φ r )) obtained applying Π ∗ (see Proposi-tion 8.10).Let ( ˜ X , X , A , ( ψ , ..., ψ r ) , ρ ) be the result of applying the correspondence of Theorem 8.21 to ( ˜ X , X , M , ( φ , ..., φ r ) , ι ) , and let ( X , C , ( ψ ′ , ..., ψ ′ r ) , ρ ′ ) be the result of applying the correspondenceof Theorem 8.17 to ( X , M , ( φ , ..., φ r ) , ι ) .Then there is a natural inclusion C ⊂ Π ∗ A which extends the inclusion of C into π ∗ A predictedin Proposition 4.34. Under this inclusion the sections ( ψ , ..., ψ r ) are identified with ( ψ ′ , ..., ψ ′ r ) . Proof.
The proof is a straightforward adaptation of the proof of Proposition 4.34 in which oneshould quote Proposition 8.13 instead of Lemma 3.6. (cid:3)
Proposition 8.26.
Let π : ˜ X → X be a resolution of a normal Stein surface with Gorensteinsingularities. Let ( A , ( ψ , ..., ψ r )) be a pair formed by a rank 1 generically reduced -dimensionalCohen-Macaulay O ˜ X -module, whose support meets the exceptional divisor E in finitely many points,and a set of r global sections spanning A as O ˜ X -module and satisfying the Containment Condition.Let C be the O X -module spanned by ψ , ..., ψ r (then C is a rank 1 generically reduced -dimensionalCohen-Macaulay O X -module). Let ( ˜ X , X , A , ( ψ , ..., ψ r ) , ρ ) be a specialty constant deformation of ( ˜ X, X, A , ( ψ , ..., ψ r )) , with Π : ˜
X → X the simultaneous resolution. Denote by C the O X -submoduleof Π ∗ A spanned by ( ψ , ..., ψ r ) . Then ( X , C , ( ψ , ..., ψ r ) , ρ ) is a deformation of ( X, C , ( ψ , ..., ψ r )) .Moreover, let ( ˜ X , X , M , ( φ , ..., φ r ) , ρ ) be the result of applying the correspondence of Theo-rem 8.21 to ( ˜ X , X , A , ( ψ , ..., ψ r ) , ρ ) and ( X , M , ( φ ′ , ..., φ ′ r ) , ρ ′ ) be the result of applying the cor-respondence of Theorem 8.17 to ( X , C , ( ψ , ..., ψ r ) , ρ ) . Then we have the equality of deformations ( X , Π ∗ M , ( φ , ..., φ r ) , ρ ) = ( X , M , ( φ ′ , ..., φ ′ r ) , ρ ′ ) . Proof.
In order to prove that ( X , C , ( ψ , ..., ψ r ) , ι ) is a deformation of ( X, C , ( ψ , ..., ψ r )) we onlyneed to prove the flatness of C over the base S of the deformation. This follows because we havethe exact sequence 0 → C → Π ∗ A → D → , the module D is flat over S because ( ˜ X , X , A , ( ψ , ..., ψ r ) , ρ ) is a specialty constant deformationand the module Π ∗ A is also flat over S because it coincides with A as a O S -module.The remaining assertion runs parallel to the proof of Proposition 4.36: according with the proofof Theorem 8.21 and its proof the module N in the sequence 0 → N → O r ˜ X → A → EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 65 of M . Pushing down by Π ∗ we obtain0 → Π ∗ N → O r X → Π ∗ A → R Π ∗ N → R Π ∗ O r ˜ X → , and the image of the map O r X → Π ∗ A is the O X -module spanned by ψ , ..., ψ r , that is, the module C . So we obtain the sequence 0 → Π ∗ N → O r X → C → . According with Theorem 8.17 and its proof the module Π ∗ N is isomorphic to the dual of M . ByProposition 8.13 the module Π ∗ N is isomorphic to the dual of Π ∗ M . This concludes the proof ofthe equality Π ∗ M = M . (cid:3) Classification of Gorenstein normal surface singularities in Cohen-Macaulayrepresentation types
In this section we prove that non log-canonical Gorenstein surface singularities are of wild Cohen-Macaulay representation type. This confirms a conjecture by Drodz, Greuel and Kashuba [6], andcompletes the classification in Cohen-Macaulay representation types of Gorenstein normal surfacesingularities. The reader may consult [6] and the references quoted there for a full definition offinite, tame and wild Cohen-Macaulay representation types. For our purposes it is enough to knowthat if a singularity admits essential families of indecomposable reflexive modules then it is of wildCohen-Macaulay representation type.
Definition 9.1.
Let ( X, x ) be a normal surface singularity. A family M of O X -modules over avariety S is called essential for any reflexive module M , if the set of points s ∈ S such that M | s isisomorphic to M is a -dimensional analytic subvariety. Proposition 9.2.
Let ( X, x ) be a normal Gorenstein surface singularity. If the minimal canonicalorder of prime divisors over x (see Definition 5.10) is unbounded from below, then there are essentialfamilies of indecomposable special reflexive O X -modules with arbitrarily high dimensional base. Thus ( X, x ) is of wild Cohen-Macaulay representation type. Proof.
Let π : ˜ X → X be a resolution which contains an irreducible component F of the ex-ceptional divisor E of π such that the minimal canonical order at F equals − d . We are going toconstruct a d -dimensional essential family of indecomposable special reflexive O X -modules.Let V be the set of sequences ( x , ..., x d ) of infinitely near points to x having the followinginductive properties: the point x is a smooth point of the exceptional divisor E belonging to F .For 2 ≤ i ≤ d − π x ,...,x i − i − : ˜ X i − → X be the composition of the blow-ups of ˜ X at x , ..., x i − i ,and the map π , let E i − be the exceptional divisor of π x ,...,x i − i − and F i − be the exceptional divisorof the blow up at x i − ; the point x i is a smooth point of the exceptional divisor E i − belongingto F i − . According with the methods of [11], Chapter 3, there is a variety V parametrizing suchsequences of infinitely near points, and a universal sequence of proper birational maps Z d − d − −→ Z d − → ... → Z −→ ˜ X × V Π −→ X × V, so that Π is the map ( π, Id V ) and for any ( x , ..., x d ) ∈ V and any k ≤ d −
1, the fibre morphismΠ x ,...,x d ) ◦ ... ◦ Π k ( x ,...,x d ) : Z k ( x ,...,x d ) → X × { ( x , ..., x d ) } , is equal to π ◦ π x .... ◦ π x ,...,x k k . We denote the composition Π ◦ ... ◦ Π d − by ρ .Let D ⊂ Z d − be a divisor such that its fibre over ( x , ..., x d ) ∈ V is a smooth curvette meetingthe exceptional divisor E d − transversely through x d . It is clear that D exists at least over a Zariski open subset of V . Let r ( x , ..., x d ) be the minimal number of generators of ( ρ ∗ O D ) ( x ,...,x d ) as a O X -module. It is easy to prove that the function r ( x , ..., x d ) is upper semi-continuous in V . Byshrinking V to a Zariski open subset we may assume the existence of sections ( ψ , ...., ψ r ) of ρ ∗ O D that specialized over each point ( x , ..., x d ) ∈ V gives a minimal set of generators of ( ρ ∗ O D ) ( x ,...,x d ) .Applying the correspondence of Theorem 8.17 to (( ρ ∗ O D ) ( x ,...,x d ) , ( ψ , ...., ψ r )) we obtain pair( M , ( φ , ..., φ r )), where M is a family of reflexive O X -modules. Since R ρ ∗ O D vanishes, Lemma 8.6implies the equality ( ρ ∗ O D ) ( x ,...,x d ) = ( ρ ( x ,...,x d ) ) ∗ O D | ( x ,...,xd ) for any ( x , ..., x d ) ∈ V . Conse-quently the module M | ( x ,...,x d ) is the result of applying the correspodence of Theorem 4.15 tothe pair (( ρ ( x ,...,x d ) ) ∗ O D | ( x ,...,xd ) , ( ψ | ( x ,...,x d ) , ...., ψ r | ( x ,...,x d ) )). Then, having chosen D generic,by Proposition 5.13, Remark 4.23 and Proposition 5.15 we conclude that M | ( x ,...,x d ) is a specialindecomposable module.The dimension of V equals d , since each of the infinitely near points is free. Therefore, in orderto finish the proof we only have to show that the family is essential. We claim that ( ρ ( x ,...,x d ) ) ∗ is the minimal resolution adapted to M | ( x ,...,x d ) . If the claim is true the modules M | ( x ,...,x d ) arepairwise non-isomorphic by Theorem 7.10 and we are done.The claim follows from Proposition 5.5, noticing the facts that cond (cid:18) K (( ρ ( x ,...,xd ) ) ∗ O D | ( x ,...,xd ) , ( ψ | ( x ,...,xd ) ,....,ψ r | ( x ,...,xd ) )) (cid:19) = 0 , and that the minimal canonical order of F i vanishes if and only if i = d . (cid:3) Theorem 9.3.
A Gorenstein surface singularity is of finite Cohen-Macaulay representation type ifand only if it is a rational double point. Gorenstein surface singularities of tame Cohen-Macaulayrepresentation type are precisely the log-canonical ones. The remaining Gorenstein surface singu-larities are of wild Cohen-Macaulay representation type.
Proof.
In [9] it is proved that a normal surface singularity has finitely many indecomposablereflexive modules if and only if it is a quotient singularity. This implies that the Gorenstein surfacesingularities of finite Cohen-Macaulay representation type are exactly the rational double points.In [6] it is proved that log-canonical surface singularities are of tame Cohen-Macaulay representationtype. By the classification of Example 3.27 of [25], if X is a Gorenstein singularity which is not log-canonical, and P q i E i is the divisor associated with its Gorenstein form at the minimal resolution,then either there is a q i < −
1, or the exceptional divisor have singularities of Milnor number atleast 3. In this case it is possible to obtain resolutions which are small with respect to the canonicalcicle with arbitrarily negative coefficients for the divisor of the Gorenstein form. Proposition 9.2shows now that non log-canonical singularities are of wild Cohen-Macaulay representation type. (cid:3)
Lifting deformations
Let X be a normal Stein surface. Let π : ˜ X → X a resolution with exceptional divisor E . Let M be a reflexive O X -module of rank r and M be the associated full O ˜ X -module. We study whendeformations of M lift to full deformations of M . Definition 10.1.
Let X be a normal Stein surface. Let X be a deformation of X over a base ( S, s ) . Let Π : ˜
X → X be a very weak simultaneous resolution with exceptional divisor E . The strict EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 67 transform ˆ A of a subscheme A ⊂ X is the scheme theoretic closure of (Π | ˜ X \E ) − ( A \ Π( E ) in ˜ X . Asubscheme A ⊂ X lifs to ˜ X if the strict transform ˆ A to ˜ X is finite over A .Let C be a rank generically reduced -dimensional Cohen-Macaulay O X -module, ( ψ , ..., ψ r ) bea system of generators. A deformation ( X , C , ( ψ , ..., ψ r ) , ι ) of ( C , ( ψ , ..., ψ r )) lifs in a specialtydefect constant way to ˜ X if • there is a rank generically reduced -dimensional Cohen-Macaulay O ˜ X -module A meetingthe exceptional divisor in finitely many points, and a set of generators ( ψ ′ , ..., ψ ′ r ) as a O ˜ X module such that the O X -submodule of π ∗ A generated by the sections ( ψ ′ , ..., ψ ′ r ) is isomor-phic to C . Then, by abuse of notation we denote the sections ( ψ ′ , ..., ψ ′ r ) by ( ψ , ..., ψ r ) . • There is a specialty defect constant deformation ( ˜ X , X , A , ( ψ , ..., ψ r ) , ι ′ ) of ( A , ( ψ , ..., ψ r )) whose image by Π ∗ equals ( X , C , ( ψ , ..., ψ r ) , ι ) . Remark 10.2.
In the setting of the previous definition, it is clear that A lifts to ˜ X if and only ifthe fibre over s of the strict transform of A coincides with the strict transform of the fibre of A over s . Lemma 10.3.
Let X be a normal Stein surface. Let X be a deformation of X over a reducedbase ( S, s ) . Let Π : ˜
X → X be a very weak simultaneous resolution with exceptional divisor E . Let A ⊂ X be a closed subscheme such that the fibre A s of A over s is of dimension , and such thatthe Zariski open subset A ∩ ( X \ Π( E )) is flat over S . Then A is liftable if and only if for any forany arc γ : Spec ( C [[ t ]]) → ( S, s ) the subscheme A × S Spec ( C [[ t ]]) ⊂ X × S Spec ( C [[ t ]]) is liftable forthe very weak simultaneous resolution obtained by pullback. Proof. If A is liftable, then, by the previous Remark it is obvious that for any arc γ : Spec ( C [[ t ]]) → ( S, s ) the subscheme A × S Spec ( C [[ t ]]) ⊂ X × S Spec ( C [[ t ]]) is liftable.It is clear that a subscheme A is liftable if and only if each of the irreducible components of thecorresponding reduced subscheme A red are liftable. Hence we may assume A to be reduced andirreducible.Conversely, assume that A is not liftable. Then, by the previous Remark the strict transform ofthe fibre A s to ˜ X s is strictly contained in the fibre ˆ A | s over s of the strict transform of A to ˜ X . Let a ∈ ˆ A | s a point not contained in the strict transform of A s . On the other hand, there is a Zariskiopen subset W of S such that for any s ′ ∈ W the fibre ˆ A | s ′ coincides with the strict transform ofthe fibre over s ′ .Since A is irreducible, so it is its strict transform ˆ A . Then the strict transform ˆ A | W is Zariski-dense in ˆ A , and consequently a is at the closure of ˆ A | W . By Curve Selection Lemma there existsan arc ˆ γ : Spec ( C [[ t ]]) → ˆ A such that ˆ γ (0) = a and such that the generic point of Spec ( C [[ t ]]) ismapped to ˆ A | W . Let γ : Spec ( C [[ t ]]) → S be the composition of ˆ γ with Π and the projection to S . Then γ is an arc so that A × S Spec ( C [[ t ]]) ⊂ X × S Spec ( C [[ t ]]) is not liftable for the very weaksimultaneous resolution obtained by pullback. (cid:3) Proposition 10.4.
Let X be a normal Stein surface with Gorenstein singularities. Let X be adeformation of X over a reduced base ( S, s ) . Let Π : ˜
X → X be a very weak simultaneous resolutionwith exceptional divisor E . Denote by π : ˜ X → X the fibre of Π over s . Let M be a reflexive O X -module of rank r . Let ( X , M , ι ) be a deformation of M over ( S, s ) . Let M be the full O ˜ X -moduleassociated to M . The first 3 of the following conditions are equivalent and imply the fourth andfifth.(1) There is a deformation ( ˜ X , X , ( M ) , ι ′ ) of M which transforms under Π ∗ to ( X , M , ι ) . (2) For any collection ( φ , ...., φ r ) of nearly generic global sections of M and any extension ( φ , ..., φ r ) as sections of M , the deformation ( X , C , ( ψ , ..., ψ r ) , ρ ) obtained applying thecorrespondence of Theorem 8.17 to ( X , M , ( φ , ..., φ r ) , ι ) lifts in a specialty defect constantway to ˜ X .(3) There exists a collection ( φ , ...., φ r ) of nearly generic global sections of M and an extension ( φ , ..., φ r ) as sections of M , such that the deformation ( X , C , ( ψ , ..., ψ r ) , ρ ) obtained ap-plying the correspondence of Theorem 8.17 to ( X , M , ( φ , ..., φ r ) , ι ) lifts in a specialty defectconstant way to ˜ X .(4) For any collection ( φ , ...., φ r ) of nearly generic global sections of M and any extension ( φ , ..., φ r ) as sections of M , the support of the degeneracy module of ( M , ( φ , ..., φ r )) isliftable.(5) There exists a collection ( φ , ...., φ r ) of nearly generic global sections of M and an ex-tension ( φ , ..., φ r ) as sections of M , such that the support of the degeneracy module of ( M , ( φ , ..., φ r )) is liftable. Proof.
Suppose Condition (1) holds. Consider a collection ( φ , ...., φ r ) of nearly generic globalsections of M and an extension ( φ , ..., φ r ) as sections of M . Applying the correspondence ofTheorem 8.21 to ( ˜ X , X , M , ( φ , ..., φ r ) , ι ′ ) we obtain the desired lifting. This proves Condition (2).Condition (2) implies Condition (3) trivially.If Condition (3) holds let ( ˜ X , X , A , ( ψ , ..., ψ r ) , ρ ) be the specialty defect constant lifting. Ap-plying the correspondence of Theorem 8.21 to it we obtain a deformation ( ˜ X , X , M , ( φ , ..., φ r ) , ι ′ ).The cuadruple ( ˜ X , X , M ) , ι ′ ) obtained by forgetting the sections is the deformation that we needto obtain to show that Condition (1) holds.If Condition (2) holds let ( ˜ X , X , A , ( ψ , ..., ψ r ) , ρ ) be the specialty defect constant lifting. Thesupport of A is the strict transform of the support of the degeneracy module of ( M , ( φ , ..., φ r )).Since for any s ′ ∈ S the support of A| s ′ meets the exceptional divisor at finitely many points,Condition (4) holds.Condition (4) implies condition (5) obviously. (cid:3) In the next example we use the non-liftability of the support of the degeneracy module to provethat a deformation of a reflexive module does not lift to a full deformation.
Example 10.5.
Let X = V ( xz − y ) ⊂ C . Let S = Spec ( C [[ s ]]) . Define α : Spec ( C [[ t, s ]]) → X × S by ( t, s ) → ( t , t + st, ( t + s ) ) . Define C := α ∗ C [[ t, s ]] , and let ( ψ , ..., ψ r ) be a system of generatorsof C as O X × S -module. The correspondence of Theorem 8.21 defines a deformation M of reflexivemodules which does not lift to a full deformation of full sheaves, since the support of the degeneracymodule is not liftable. Here we show an example where the support of the degeneracy module is liftable, but there isno full deformation.
Example 10.6.
Let X = V ( x + y + z ) . This is a minimally elliptic singularity as studiedin [19] . The blowing up at the origin π : ˜ X → X produces its minimal resolution. Its exceptionaldivisor is a smooth elliptic curve E . Let ˜ C be a smooth curvette embedded in ˜ X , which meets E at a single point p with intersection multiplicity equal to . Let t be a uniformizing parameter forthe germ ( ˜ C, p ) . The curve C := π ( ˜ C ) has an ordinary cusp singularity at the origin of C (that is C = Spec C [[ t , t ]] ). Let S := Spec C [[ s ]] . We have the isomorphism O ˜ C × S ∼ = C [[ t, s ]] ; this endows C [[ t, s ]] with structures of O X × S -module and of O ˜ X × S -module. Denote by C the O X × S -submodule EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 69 of C [[ t, s ]] spanned by ( s + t, t ) . We let D be its cokernel as O X × S module. It is easy to check that D is C [[ s ]] -flat, and hence O C is C [[ s ]] -flat as well. Apply the correspondence of Theorem 8.17 to ( C , ( s + t, t )) and obtain ( M , ( φ , φ )) , where M is a family of reflexive O X -modules.The support of C equals C × S , and its strict transform equals ˜ C × S . So C × S lifts. On theother hand, if there is a family M of full ˜ X -modules lifting M , then applying the correspondence ofTheorem 8.21 to ( M , ( φ , φ )) we would obtain ( A , ( s + t, t )) , where A is the O ˜ C × S module spannedby ( s + t, t ) . Since this module is not Cohen-Macaulay of dimension (it is not a free module over C [[ t, s ]] ), it can not be a flat family over C [[ s ]] of -dimensional Cohen-Macaulay O ˜ X -modules. Thisis a contradiction which implies that the there is no family of full ˜ X -modules lifting M .Another way of proving the non-existence of lifting is observing that the specialty defect of thefull O ˜ X -module lifting M | is zero, that the specialty defect of the full O ˜ X -module lifting M | s is notzero if s = 0 , and using Corollary 8.24. Proposition 10.7.
Let X be a normal Stein surface with Gorenstein singularities. Let X be adeformation of X over a reduced base ( S, s ) . Let Π : ˜
X → X be a very weak simultaneous resolution.Let M be a reflexive O X -module and ( X , M , ι ) be a deformation of ( X, M ) over ( S, s ) . Then thereexists a dense Zariski open subset of S where the deformation lifts as a full family of O ˜ X -modules. Proof.
Denote by N := (cid:0) Π ∗ M (cid:1) ∨ , M := (cid:0) Π ∗ M (cid:1) ∨∨ , and also denote by M s ′ := (cid:0) Π ∗ M | s ′ (cid:1) ∨∨ the full sheaf associated to M | s ′ and N s ′ := (cid:0) Π ∗ M | s ′ (cid:1) ∨ =( M ′ s ) ∨ for any s ′ ∈ S .By genericity of flatness over a reduced base, there exists a Zariski dense open subset U ⊂ S such that the following two properties hold:(1) The sheaves Π ∗ M , N , E xt O ˜ X (cid:0) Π ∗ M , O ˜ X (cid:1) and E xt O ˜ X (cid:0) Π ∗ M , O ˜ X (cid:1) are flat over U .(2) The sheaf R Π ∗ M is flat over U .For any s ′ ∈ S we know that ˜ X | s ′ is a smooth surface of dimension two, hence the Auslander-Buchsbaum Formula [3, Theorem 1.3.3] implies that E xt j O ˜ X | s (cid:0) Π ∗ M | s ′ , O ˜ X | s ′ (cid:1) = 0 for any j ≥ ∗ M implies the isomorphism N | s ′ ∼ = N s ′ for any s ′ ∈ U . So we havethat N | U is locally free, and dualizing we obtain that M| U is flat over U (locally free on ˜ X ) andthe isomorphism M| s ′ ∼ = M s ′ for any s ′ ∈ U . Finally Assertion (2) of the previous list implies that( ˜ X | U , X | U , M| U ) is a family of full modules over U .In order to finish the proof we need to verify that the natural morphism of coherent O X -modules M | U → (Π | U ) ∗ M| U obtained as the composition M | U → (Π | U ) ∗ (Π | U ) ∗ M | U → (Π | U ) ∗ ((Π | U ) ∗ M | U ) ∨∨ = (Π | U ) ∗ M| U is an isomorphism. By Lemma 8.6 the sheaf (Π | U ) ∗ M| U is flat over U . Using this and Nakayama’s Lemma we arereduced to prove that and for any s ′ ∈ U we have the specialization M | s ′ → ((Π | U ) ∗ M| U ) | s ′ is an isomorphism. Lemma 8.6 implies that this morphism coincides with M | s ′ → (Π | s ′ ) ∗ ( M| s ′ ) , but the second sheaf in the module has been proved to be isomorphic to (Π | s ′ ) ∗ M s ′ , where M s ′ isthe full O ˜ X s ′ -module associated with M | s ′ . Then the morphism is an isomorphism as needed. (cid:3) Sufficient conditions for liftability to full deformations.
In this section we show suf-ficient conditions ensuring the liftability of a family of reflexive sheaves to a full family on a veryweak simultaneous resolution. The result we prove probably is not the best that one can hope for,but is more than sufficient for the applications we have in mind. In particular, at some point wesimplify things by working on a normal surface singularity rather than on a normal Stein surface.
Definition 10.8.
Let X be a normal Stein surface. Let X be a deformation of X over a reducedbase ( S, s ) . Let M be a reflexive O X -module of rank r . A deformation ( X , M , ι ) of ( X, M ) overa reduced base ( S, s ) is said to be simultaneously normalizable if the degeneracy locus C of M fora generic system of r sections admits a simultaneous normalization over S . That is, there existsa smooth family of curves D over S , and a morphism n : D → C such that for any s ′ ∈ S therestriction n | D s ′ : D s ′ → C reds ′ is the normalization. Remark 10.9. (1) The existence of simultaneous normalization of C in the previous definitiondoes not depend on the choice of the system of generic sections.(2) The reader may consult [13] and [20] for recent accounts on simultaneous normalization.(3) There is a way to define a functor of simultaneously normalizable deformations of reflexivemodules, considering also non-reduced bases ( S, s ) . The definition has some subtlety sincethe family of supports C need not be a flat family of reduced curves. Since the applicationspresented in this paper do not need such a definition we avoid it.(4) If at the previous definition we assume X to be a normal surface singularity (that is, agerm), we may assume that the normalization of C s is a disjoint union of discs ` i D i . Asa consequence if the family is simultaneously normalizable, then the predicted simultaneousnormalization D is equal to ` i D i × S . Theorem 10.10.
Let X be a normal Gorenstein surface singularity. Let X be a deformation of X over a normal base ( S, s ) . Let Π : ˜
X → X be a very weak simultaneous resolution. Let M bea reflexive O X -module and ( X , M , ι ) be a simultaneously normalizable deformation of ( X, M ) overthe base ( S, s ) , so that for each s ′ ∈ S the module M | s ′ is special. If the support of the degeneracymodule of M for a generic system of sections is liftable to ˜ X , then the family ( X , M , ι ) lifts to afull family on ˜ X . Proof.
Let ( φ , ..., φ r ) be a system of generic global sections of M . Let ( X , C , ( ψ , ..., ψ r ) , ρ ) bethe result of applying the correspondence of Theorem 8.17 to ( X , M , ( φ , ..., φ r ) , ι ). Let C be thesupport of C . Let n : D → C be the simultaneous normalization that exists because ( X , M , ι ) is simultaneously normalizable. ByRemark 10.9 we may assume that D is the product of S with a disjoint union of discs. EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 71
Let K ( C ) be the total fraction ring of C . If C = ∪ mi =1 C i is the decomposition in irreduciblecomponents then the total fraction ring equals the direct product K ( C ) = Q mi =1 K ( C i ) of thefunction fields of the components. Denote by T the set of non-zero divisors of O C . The localization T − C is a K ( C )-module, which expresses as T − C = Q mi =1 C i , where C i is a K ( C i )-vector space.Since C is a flat family of 1-dimensional rank 1 generically reduced Cohen-Macaulay modules overthe normal base S , the natural map to the localization C → T − C is injective, and each C i is a1-dimensional vector space. Hence the localization T − C is isomorphic to Q mi =1 K ( C i ) = K ( C ). Wehave found a O C -module monomorphism ι : C ֒ → K ( C ) . Noticing that O D is a sub-ring of K ( C ),it makes sense to define B to be the O D -submodule of K ( C ) spanned by C . Since C is generatedas O C module by ( ψ , ..., ψ r ), multiplying by the common denominator of ι ( ψ ) , ..., ι ( ψ r ) we mayassume that the image of ι lies in O D . We have got the chain of inclusions of O C -modules(101) C ֒ → B ֒ → O D . The second inclusion is an inclusion of O D -modules; thus B is an ideal in O D . Suppose thatthe zero set V ( B ) contains an irreducible component Z that is dominant over S by the naturalprojection map. Such a Z is contained in a unique connected component of D , and each of theseconnected components is isomorphic to the product of S times a disc. Let t be a coordinate of thedisc. By the normality of ( S, s ) there is a Weierstrass polynomial P in O S [[ t ]] whose zero locusdefines Z . Then, dividing by the appropriate Weierstrass polynomials we may assume that theembedding ι is so that the zero set of the ideal B does not dominate S . Hence there exists a Zariskidense open subset U in S where, under the embedding ι we have the equality(102) B| U = O D | U . For any s ′ ∈ S , by specialty of M | s ′ and Proposition 5.13 we have the isomorphism C| s ′ ∼ = O D s ′ .This implies that B| s ′ is a monic ideal in O D s ′ and the equality C| s ′ = B| s ′ . Then Equation (102)implies the equality(103) C| U = O D | U . Define D := O D / C . Denote by m S the maximal ideal of O S,s . Applying (cid:5) ⊗ O S O S / m S to theexact sequence of O C -modules 0 → C → O D → D → , we obtain the exact sequence0 → T or O S ( D , O S / m S ) → C ⊗ O S O S / m S → O D ⊗ O S O S / m S → D ⊗ O S O S / m S → . The second group in the sequence is isomorphic to C| s , the third group is O D | s and the morphismconnecting them is injective at the generic points of the support of C| s . Then, since C| s is a rank1 generically reduced Cohen-Macaulay O X -module of dimension 1 the morphism C ⊗ O S O S / m S → O D ⊗ O S O S / m S is injective and then T or O S ( D , O S / m S ) vanishes. Then D is flat over S by the Local Criterionof Flatness. Since by Equality (103) the O S -module D has proper support we conclude that D vanishes. This proves that the inclusion (101) becomes the equality(104) C = O D . Since C has been assumed to liftable to ˜ X the restriction of Π to the strict transform ˆ C of C to ˜ X is finite and birational over C . This implies that D dominates ˆ C , and then we have a ring monomorphism O ˆ C ֒ → O D . As a consequence O D inherits a structure of O ˜ X -module. Summingup we have shown that ( ˜ X , X , O D , ( ψ , ..., ψ r ) , Id | O Ds ) is a specialty defect constant deformationof ( ˜ X, X, O D s , ( ψ | s , ..., ψ r | s )).Applying the correspondence of Theorem 8.21 to ( ˜ X , X , O D , ( ψ , ..., ψ r ) , Id | O Ds ) we obtain a fulldeformation ( ˜ X , X , M , ( φ , ..., φ r ) , ι ). An application of Proposition 8.26 concludes the proof. (cid:3) Moduli spaces of special reflexive sheaves on Gorenstein surface singularities
Let (
X, x ) be a Gorenstein normal surface singularity. In this section we consider deformationsand families of O X -modules fixing the space X . Our aim is to construct moduli spaces of specialreflexive modules with prescribed combinatorial type. Definition 11.1.
Let ( X.x ) be a Gorenstein surface singularity. Let G be the graph of a specialreflexive O X -module and r a positive integer. A family of special modules with graph G and rank r over a complex space S is a O X × S -module M which is flat over S and such that for any s ∈ S ,the module M | s is a special reflexive module of rank r and graph G .Define a moduli functor Mod r G from the category of Normal Complex Spaces to the categoryof Sets, assigning to a normal complex space S the set of families of special modules without freefactors with graph G and rank r , and to a morphisms of complex spaces the corresponding pullbackof families. In this section we prove that the previous functors are representable by a complex algebraicvariety.
Remark 11.2.
Our moduli functor is somewhat restricted: we only consider families over normalcomplex spaces. It is an open problem to show that our moduli spaces represent the usual modulifunctors. One should notice that even the definition of the moduli functor has some subtleties: re-stricting Example 10.6 to the base
Spec ( C [[ s ]] / ( s )) one sees a flat deformation of a special reflexive O X -module, such that over each point of the base the corresponding reflexive module is special, butthat should not be considered as a family of special modules. Let G be the graph of a special reflexive module (the possible graphs are classified in Theorem 7.2).Deleting the arrows of G we obtain a resolution graph G o of the singularity X . Two differentresolutions with the same resolution graph only differ in the positions of the infinitely near pointswhich are the centers of the blow ups occuring after the minimal resolution.Let M ′G o be the set of pairs ( π, ϕ ), where π : ˜ X → X is a resolution of singularities and ϕ isa bijection from the vertices of G o to the irreducible components of the exceptional divisor E of π inducing an isomorphism from G o to the dual graph of the resolution. The group Aut ( G o ) ofautomorphims of the graph G o acts on M ′G o by composition on the left at the second coordinate.The group Aut ( G ) is a subgroup of Aut ( G o ). Define M G o and M G to be the quotient of M ′ G o by Aut ( G o ) and Aut ( G ) respectively. The points of M G o and M G are equivalence classes which will bedenoted by ( π, ϕ ) for simplicity. Lemma 11.3.
The sets M G and M G o have a natural structure of algebraic variety, and the naturalmap from the first to the second is an etale covering.The variety M G is a moduli space of resolutions of X , and has a universal family in the followingsense: there is a birational morphism Π : ˜
X → X × M G , EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 73 such that for any ( π, ϕ ) ∈ M G the pullback morphism Π | [ ˜ X ( π,ϕ ) : ˜ X ( π,ϕ ) → X × { ( π, ϕ ) } coincides with π , where ˜ X ( π,ϕ ) denotes the fibre of ˜ X over ( π, ϕ ) by the composition of Π with theprojection to the second factor. Proof.
See [11], Chapter 3. (cid:3)
Notation 11.4.
Let U be an open subset of M G , the restriction of the universal family over U isdenoted by Π | U : ˜ X | U → X × U. Now we associate a special reflexive module to each point of M G . Let ( π, ϕ ) ∈ M G .A curve D ⊂ ˜ X | ( π,ϕ ) is ( π, ϕ )- appropriate if(1) it is a disjoint union of smooth curvettes which meet the exceptional divisor of π transversely,and for any vertex of M G the number of curvettes meeting the divisor that ϕ assigns tothis vertex is exactly the number of arrows attached to this vertex,(2) the minimal number of generators of π ∗ O D as a O X module is minimal among the curveswith the previous property.In Remark 7.11 and its proof it is shown that a generic curve having the first property also satisfiesthe second.Let D be a ( π, ϕ )-appropriate curve. Let ( ψ , ..., ψ r ) be a minimal set of generators of the O X -module π ∗ O D . Applying the correspondence of Theorem 4.15 to ( π ∗ O D , ( ψ , ..., ψ r )) we obtain( M ( π,ϕ ) , ( φ , ..., φ r )), where M ( π,ϕ ) is special reflexive O X -module of rank r (specialty is by Proposi-tion 5.13). In Sections 7.2 and 7.3 it is proved that M ( π,φ ) does not depend on the ( π, ϕ )-appropriatecurve.However, there is no reason for which the rank r should be independent on the combinatorialtype. In fact the rank of M ( π,ϕ ) induces a stratification in M G because of the following lemma. Lemma 11.5.
The function rank ( M ( π,ϕ ) ) is upper-semicontinuous in M G for the Zariski topology. Proof.
The assertion is local. Consider ( π, ϕ ) ∈ M G . Given a small neighborhood U of ( π, ϕ ) itis easy to construct a subscheme D ⊂ ˜ X | U such that for any ( π ′ , ϕ ′ ) ∈ U we have that the fibre D | ( π ′ ,ϕ ′ ) is a ( π ′ , ϕ ′ )-appropriate curve and the O ˜ X | U -module O D is flat over U (this is easy bythe genericity statement given in Remark 7.11). Since R Π ∗ O D vanishes, Lemma 8.6 implies thatΠ ∗ O D is flat over U and that (Π ∗ O D ) | ( π ′ ,ϕ ′ ) equals ( π ′ ) ∗ ( O D | ( π ′ ,ϕ ′ ) ). Then, the minimal numberof generators of (Π ∗ O D ) | ( π ′ ,ϕ ′ ) as O X -module is upper-semicontinuous in the Zariski topology, andcoincides with the rank of M ( π ′ ,ϕ ′ ) . (cid:3) Denote by M r G to be the locally closed subset corresponding to modules of rank r . By Theo-rem 7.10, the closed points of M r G are in a bijection with the set of reflexive modules without freefactors of graph G and rank r . Our next step is to construct a universal family over each M r G . Lemma 11.6.
Let U be an open subset of M r G . For i = 1 , let D i ⊂ ˜ X | U such that for any ( π ′ , ϕ ′ ) ∈ U we have that the fibre D i | ( π ′ ,ϕ ′ ) is a ( π ′ , ϕ ′ ) -appropriate curve and the O ˜ X | U -module O D i is flatover U . Let ( ψ i , ..., ψ ir ) be a system of generators of O D i as a O X × S -module. Let ( M i , ( φ i , ..., φ ir )) be the result of applying the correspondence of Theorem 8.17 to (Π ∗ O D i , ( ψ i , ..., ψ ir )) . Then we havethe equality M = M . Proof.
Let ( M i , ( φ i , ..., φ ir )) be the result of applying the correspondence of Theorem 8.21 to( O D i , ( ψ i , ..., ψ ir )). By Proposition 8.26 we have the equality Π ∗ M i = M i . Repeating in familythe arguments of Section 7.2 we obtain that M i is determined by the line bundle O ˜ X | U ( − D i ).An argument like in Lemma 7.7 show the isomorphism O ˜ X | U ( − D ) ∼ = O ˜ X | U ( − D ). (cid:3) Lemma 11.7.
There exists a unique family of special modules M r G with graph G and rank r over M r G such that for any ( π, ϕ ) ∈ M r G we have the isomorphism M r G | ( π,ϕ ) = M ( π,ϕ ) . Proof.
The previous Lemma shows how to construct the family locally, and it shows, using The-orem 8.21 for suitable generic sections, that the result is unique up to the choices made. So theprocedure glues well to a global universal family. (cid:3)
Theorem 11.8.
The variety M r G represents the functor Mod r G . Proof.
Let M be a family of special modules without free factor with graph G and rank r over anormal base S . Consider the mapping θ : S → M r G sending s to the unique point of ( π, ϕ ) ∈ M r G such that we have the isomorphism M | s ∼ = M r G | ( π,ϕ ) .We have to prove that the map is a complex analytic morphism, and that the pullback of the theuniversal family over M r G gives back the family M .The infinitely near points that one need to blow up to get π from the minimal resolution in X arepartially ordered as follows: the first generation points is the set of points appearing in the minimalresolution, the second generation set of points are those appearing in the resolution obtained byblow up the first generation points, and succesively. Let k be the maximal generation order of theinfinitely near points needed to obtain π . Let G oi be the dual graph of the result of blowing up the i -th generation points. We have a natural sequence of morphisms(105) M r G ֒ → M G → M G o = M G ok → M G ok − → ... → M G o , where the first morphism is a locally closed inclusion and the second morphism is an etale covering.We will prove by induction that the composition θ i : S → M G oi is a complex analytic morphism for any i .The initial step of the induction runs as follows. Let π : ˜ X min → X be the minimal resolution.Letand Π : ˜ X min × S → X × S be the product of the map π with the identity at S . Denote by E Π the exceptional divisor of Π. For any s ∈ S denote by M s the full O ˜ X min -module associated with M | s . By Proposition 10.7 there is a Zariski open subset V in S and a flat O ˜ X min × S -module M V such that R Π ∗ M V is flat, for any s ∈ V the restriction M V | s is isomorphic to M s , and we haveΠ ∗ M V = M | V .Fix s ∈ S . Choose a neighborhood U of s in S and sections ( φ , ..., φ r ) of M | U so that thesections ( φ | s , ..., φ r | s ) are nearly generic for • the O ˜ X min -module M s for any s in U ∩ V or s = s , • the O X -module M s for any s ∈ U . EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 75
Choosing ( φ , ..., φ r ) generic and perhaps shrinking U suffices to guarantee the properties above.Since the assertion we want to prove is local in S we may assume U = S and V ⊂ U to save somenotation.Applying the correspondence of Theorem 8.17 to ( M , ( φ , ..., φ r )) we obtain ( C , ( ψ , ..., ψ r )).Since Theorem 8.17 establishes an isomorphism of deformation functors, applying the correspon-dence of Theorem 4.15 to ( M | s , ( φ | s , ..., φ r | s )), we obtain the pair ( C| s , ( ψ | s , ..., ψ r ) | s )).The correspondence of Theorem 4.18 applied to ( M s , ( φ | s , ..., φ r | s )) gives a pair ( A s , ( ψ ,s , ..., ψ r,s )),where A s is 1-dimensional Cohen-Macaulay O ˜ X min -module generated by the sections ( ψ ,s , ..., ψ r,s ).By Proposition 4.34 and specialty we have that π ∗ A s equals C| s , and we have the identification ofsections ( ψ ,s , ..., ψ r,s ) = ( ψ | s , ..., ψ r | s ).The correspondence of Theorem 8.21 applied to ( M V , ( φ , ..., φ r )) gives a pair ( A V , ( ψ ,V , ..., ψ r,V )).By Proposition 8.25 and specialty we have that Π ∗ ( A V ) equals C| V , and the sections ( ψ ,V , ..., ψ r,V )are identified with the restriction ( ψ | V , ..., ψ r | V ) of the sections ( ψ , ..., ψ r ) over V . Since Theo-rem 8.21 establishes an isomorphism of deformation functors, for any s ∈ V we have the equalityof pairs ( A s , ( ψ ,s , ..., ψ r,s )) = ( A V | s , ( ψ | s , ..., ψ r ) | s ) . For any s ∈ V , let A s be the support of A s . By the construction of the minimal adaptedresolution (see the proof of Proposition 5.1), the intersection of A s ∩ E Π is a finite set containingthe set of infinitely near points of first generation of Π | s . If A V is the support of A V we have thatthe fibre A V | s of A V over s equals A s . This implies that the graph of the restriction θ | V : V → M G o is a complex analytic subvariety of V × G o . Using Zariski’s Main Theorem this implies that θ | V isa complex analytic morphism since V is normal. Claim I . Let C ⊂ X × S be the support of C . Then C lifts to ˜ X min × S .Assume that the claim is true. Denote by ˜ C the strict transform ˜ C of C . Then, for any s ∈ S the intersection of ˜ C | s ∩ E Π is a finite set and moreover the fibre ˜ C | s is the support of A s for s ∈ S .Consequently ˜ C | s ∩ E Π contains the set of infinitely near points of first generation of Π | s . Thisimplies that the graph of θ is a complex analytic subvariety of U × G o , and using Zariski’s MainTheorem this implies that θ | U is a complex analytic morphism since U is normal.Let us prove the claim. By Lemma 10.3 we may assume that ( S, s ) is a germ of smooth curve.The fibre of ˆ C s over a generic point of S equals A s . The fibre ˆ C | s equals A s + F where F is a divisorin ˜ X min with support in E , which is non-negative. Now, since the modules M | s have the samecombinatorial type for any s ∈ S , we have the equality of intersection numbers E i (cid:5) A s = E i (cid:5) A s for any component E i of E . This implies the vanishing E i (cid:5) F = 0, and hence F = 0 by thenon-degeneracy of the intersection form. This shows that F vanishes, and shows the liftability.This sets up the initial step of the induction. The inductive step is completely similar. So, wehave shown that θ k is an analytic morphism.Now, since the first morphism of the sequence (105) is a locally closed inclusion, and the secondis an etale map, we conclude that θ is a complex analytic morphism.Observe that our induction procedure shows that the support C of C lifts to the universal familyof minimal adapted resolutions ˜ X . Then an application of Theorem 10.10 shows that M lifts to afull family M on ˜ X . In order to prove that M coincides with the pullback by θ of the universalfamily it is enough to show it locally, but this is a consequence of Lemma 11.6 and the procedurein which M is constructed in the proof of Theorem 10.10. (cid:3) References [1] S. Abhyankar.
Local Analytic Geometry . Pure and applied mathematics . New York-London Academic Press,xvi, (1964).[2] M. Artin, J.L. Verdier Reflexive modules over rational double points.
Mathematische Annalen , 79-82, (1985).[3] W. Bruns, H. J. Herzog.
Cohen-Macaulay Rings . Cambridge Studies in Advanced Mathematics . CambridgeUniversity Press, (1993).[4] R. O. Buchweitz, G. M. Greuel, F. O. Schreyer. Cohen-Macaulay modules on hypersurface singularities II ,Invent. Math. , 165182, (1987).[5] T. de Jong, D. van Straten. Deformations of Non-Isolated Singularities . Chapter III of T. de Jong Ph. D. Thesis.Katholieke Universiteit te Nijmegen, (1988).[6] Y. Drozd, G.M. Greuel, I. Kashuba.
On Cohen-Macaulay modules on surface singularities . Moscow Math. J. no.2, 397-418, (2003).[7] D. Eisenbud. Homological algebra on a complete intersection, with an application to group representations .Trans. Amer. Math. Soc. , 35-64, (1980).[8] D. Eisenbud, I. Peeva.
Matrix Factorizations for Complete Intersections and Minimal Free Resolutions. arXiv:1306.2615, (2013).[9] H. Esnault.
Reflexive modules on quotient surface singularities . J. Reine Angew. Math.
Reflexive modules over rational double points . Math. Ann. , 545-548, (1985).[11] J. Fern´andez de Bobadilla.
Moduli spaces of polynomials in two variables . Mem. Amer. Math. Soc. ,(2005).[12] G. Gonz´alez-Sprinberg, J.L. Verdier.
Construction geometrique de la correspondance de McKay.
Ann. Sci. Ec.Norm. Super. , 409-449, (1983).[13] G. M. Greuel. Equisingular and Equinormalizable Deformations of Isolated Non-Normal Singularities .arXiv:1702.05505, to appear in “Second Special Issue: International Conference on Singularity Theory –inHonor of Henry Laufer’s 70th Birthday, Methods and Applications of Analysis (MAA)”[14] A. Grothendieck. ´El´ements de g´eom´etrie alg´ebrique. IV: ´Etude locale des sch´emas et des morphismes desch´emas. (S´econde partie) . Publ. Mat. IHES. (1965).[15] R. Hartshorne. Residues and duality . Lecture Notes in Mathematics . Springer, (1966).[16] R. Hastshorne. Algebraic Geometry . Graduate Texts in Mathematics . Springer, (1977).[17] R. Hartshorne. Stable reflexive sheaves . Math. Ann. , 121-176, (1980).[18] R. Hartshorne.
Deformation Theory . Graduate Texts in Mathematics . Springer, (2010).[19] C. Kahn.
Reflexive modules on minimally elliptic singularities . Math. Ann. (1), 141-160, (1989).[20] J. Koll´ar.
Simultaneous normalization and algebra husks . Asian J. Math. , no. 3, 437-450, (2011).[21] M. Khovanov and L. Rozansky. Matrix factorizations and link homology . Fund. Math. , 191, (2008).[22] M. Khovanov and L. Rozansky.
Matrix factorizations and link homology II . Geometry and Topology 12,13871425, (2008).[23] A. Kapustin, Y. Li.
D-branes in Landau-Ginzburg models and algebraic geometry . J. High Energy Phys. 12, 143,(2003).[24] H. Kn¨orrer.
Cohen-Macaulay modules on hypersurface singularities . Invent. Math. , 153164 (1987).[25] J. Koll´ar. Singularities of the minimal model program. Cambridge Tracts in Mathematics , CambridgeUniversity Press, (2013).[26] H. Laufer. Weak simultaneous resolution for deformations of Gorenstein surface singularities. in: Singularities(P. Orlik, ed.), Proc. Sympos. Pure Math., vol. 40, Part 2, Amer.Math. Soc., pp. 130, (1983).[27] J. McKay.
Graphs, singularities and finite groups . In: Finite groups, Santa Cruz. Proc. Symp. Pure Math. ,183-186, (1980).[28] P. Popescu-Pampu. Numerically Gorenstein surface singularities are homeomorphic to Gorenstein ones.
DukeMath. Journal , No. 3, 539-559, (2011).[29] A. Nemethi.
Five Lectures on Normal Surface Singularities . Proceedings of the Summer School, Bolyai SocietyMathematical Studies, , Low Dimensional Topology, 269-351, (1999).[30] D. Orlov. Triangulated categories of singularities and D-branes in Landau-Ginzburg models . Tr. Mat. Inst.Steklova , (2004), Algebr. Geom. Metody, Svyazi i Prilozh, ; translation in Proc. Steklov Inst. Math. , 227248 (2004).[31] D. Orlov.
Landau-Ginzburg models, D-Branbes, and mirror symmetry . ArXiv 1111.2962, (2011).
EFLEXIVE MODULES ON NORMAL GORENSTEIN STEIN SURFACES 77 [32] A. Polishchuck, A. Vaintrob.
Matrix factorizations and cohomological field theories . J. Reine Angew. Math. ,1-122 (2016).[33] O. Riemenschneider.
Special representations and the two-dimensional McKay correspondence.
Hokkaido Math.J. no 2, 317-333, (2003).[34] E. Schlesinger. Ext and base change ∼ enrsch/EXT.pdf.[35] G. Trautmann. Deformations of sheaves and bundles . In: Variet´es analytiques compactes, Colloque, Nice 1977,Springer Lecture Notes in Mathematics
Equisingular deformations of normal surface singularities I . Annals of Math. no. 2, 325-356,(1976).[37] J. Wunram.
Reflexive modules on quotient surface singularities.
Math. Ann. (4), 583-598, (1988).[38] Y. Yoshino.
Cohen-Macaulay modules over Cohen-Macaulay rings . London Mathematical Society Lecture NotesSeries . Cambridge University Press, (1990).
Javier Fern´andez de Bobadilla: (1) IKERBASQUE, Basque Foundation for Science, Maria Diaz deHaro 3, 48013, Bilbao, Bizkaia, Spain (2) BCAM Basque Center for Applied Mathematics, Mazarredo 14,E48009 Bilbao, Basque Country, Spain (3) Academic Colaborator at UPV/EHU
E-mail address : [email protected] Agust´ın Romano-Vel´azquez: (1) CIMAT, Jalisco S/N, Mineral de Valenciana, 36023, Guanajuato,M´exico
E-mail address ::