Filtered restriction of geometric D-modules 2 1 R´emi Arcadias 0 2 Depto. de A´lgebra, Fac. Matem´aticas, Campus Reina Mercedes 41012 Sevilla (Spain) n a E-mail: [email protected] J 0 January 23, 2012 2 ] G Introduction A . Let D (resp. D′) be the ring of germs at the origin of linear partial differential h t operators with analytic coefficients on Cn (resp. Cn × Cp). We take a sys- a tem of coordinates (x ,...,x ) on Cn and (x ,...,x ,t ,...,t ) on Cn ×Cp m 1 n 1 n 1 p and we denote by i : Cn → Cn ×Cp the inclusion x → (x,0). Let M be a [ finitely generated D′-module specializable along Y = {t1 = ··· = tp = 0}. The 1 D-module theoretic restriction i∗M of M along Y is defined as the complex v (D′/ t D′)⊗LM,realizedasaKoszulcomplex(M⊗ΛiCp) ,whosecohomol- i i 9 ogy gProups are finitely generated over D. T. Oaku and N. Takayama show in 5 [10] how to realize this complex by a complex composed of finitely generated 2 4 D-modules. Let V(D′) denote the V-filtration of Malgrange-Kashiwara along . Y, and let k be the maximal integral root of the b-function of M along Y. 1 1 Then(M⊗ΛiCp) is quasi-isomorphicto the sub-complex (V (M)⊗ΛiCp) . 0 i k1+i i 2 Then a free resolution of i∗M can be computed as follows: let 1 : ···→L →L →M →0 (1) v 1 0 i X beafreeresolutionofM adaptedtotheV-filtration. Then(V (M)⊗ΛiCp) k1+i i r is isomorphic, in the derived category of D-modules, to the complex a V (L ) V (L ) ···→ k1 1 → k1 0 →0, t V (L ) t V (L ) i k1+1 1 i k1+1 0 P P denoted by (V (L )/( t V (L ))) . k1 i j j k1+1 i+1 i Ontheotherhand,PthemostclassicalfiltrationinD-moduletheoryisthatby theorder,denotedby(F (D)). Onedefinesthenotionofaminimalfilteredfree d resolution of a F-filtered D-module, thus one has the notion of Betti numbers of such a module. We remark that the notion of Betti numbers can be defined as well for complexes of D-modules in the derived category of F-filtered D- modules DF(D). Furthermore, T. Oaku and N. Takayama[11] and M. Granger and T. Oaku[6] define the notion of a minimal bifiltered free resolution of a 1 (F,V)-bifiltered D′-module. With the help of this theory, we establish that the isomorphism (V (M)⊗ΛiCp) ≃(V (L )/( t V (L ))) k1+i i k1 i j k1+1 i+1 i Xj can be seen in the category DF(D) (Proposition 3.1). AfurtherpropertyoncomplexesofF-filteredD-modulesisthatofstrictness, a property any filtered free resolution of a F-filtered module satisfies. In the case where p =1 and t: M →M is injective, we give conditions such that the complex (V (M)⊗ΛiCp) =(0→V (M)→t V (M)→0) k1+i i k1+1 k1 is strict (Proposition 3.3). In that case the complex (V (L )/(tV (L ))) k1 i k1+1 i+1 i becomes a F-filtered free resolution of the module M/tM. This latter fact, suggested by Toshinori Oaku, was the original motivation of this paper. In the last section, we apply our results to the algebraic local cohomology module N = O[1/f]/O, seen as a D-module. Here f is a quasi-homogeneous polynomial with an isolated singularity at the origin. N is endowed with the good F-filtration F (N)= F (D)[1/fj], d d−j−1 1≤Xj≤k which takes into account the order of the pole f, where −k is the least integral root of the Bernstein-Sato polynomial associated with f. By using the restric- tion,wegiveaminimalpresentationassociatedwiththisdata(Proposition4.3). 1 Free resolutions of D-modules LetD =C{x ,...,x }[∂ ,...,∂ ]denotetheringofgermsattheoriginoflinear 1 n 1 n partialdifferentialoperatorswithanalyticcoefficientsonCn. Itisendowedwith a filtration by the order, denoted by (Fd(D))d∈Z. LetM beafinitelygeneratedleftmoduleoverD. AnF-filtration(Fd(M))d∈Z onM isanexhaustivesequenceofsubspacessatisfyingFd(D)Fd′(M)⊂Fd+d′(M) for any d,d′. It is called a good F-filtration if moreoverthere exist f ,...,f ∈ 1 r M and a vector shift n=(n ,...,n ) such that for any d, 1 r F (M)= F (D)f . d d−ni i X For example, we denote by Dr[n] the free module Dr with basis e ,...,e en- 1 r dowed with the filtration F (Dr[n])= F (D)e . d d−ni i X 2 Thus a good F-filtration of M is defined by a surjective map Dr[n]→M. One then defines the notion of a F-filtered free resolution: it is an exact sequence ···→L →L →M →0 (2) 1 0 which induces for each d an exact sequence of vector spaces ···→F (L )→F (L )→F (M)→0. d 1 d 0 d Following T. Oaku and N. Takayama[11] and M. Granger and T. Oaku[6], we are able to define the notion of a minimal F-filtered free resolution. To that end we introduce, following F.J. Castro-Jim´enez and L. Narva´ez-Macarro[4], the homogenization ring RD= F (D)Td. It is isomorphic to the ring d L D(h) =C{x}[∂ ,...,∂ ,h] 1 n satisfying for any i, ∂ x −x ∂ = h. The ring D(h) is graded by the order in i i i i ∂,h. The homogenization of a F-filtered module M is defined by RM = F (M)Td. d Md In particular we have an isomorphism R(Dr[n])≃(D(h))r[n] where at the right hand side the vector shift n refers to the grading of the module. The F-filtered free resolution (2) induces a graded free resolution ···→RL →RL →RM →0. (3) 1 0 Thisresolutioniscalledminimal ifalltheentriesofthematricesrepresentingthe maps RL → RL belong to the maximal two-sided ideal of D(h) generated i i−1 by (x ) ,(∂ ) ,h. The resolution (2) is called minimal if it induces a minimal i i i i graded free resolution (3). There exists a minimal F-filtered free resolution of M, unique up to F-filtered isomorphism (see [6]). Let us denote now D = C{x ,...,x ,t ,...,t }[∂ ,...,∂ ,∂ ,...,∂ ] x,t 1 n 1 p 1 n t1 tp the ring of germs of linear partial differential operators with analytic coeffi- cients on Cn×Cp. It is endowed with the F-filtration by the order (F (D )). d x,t Another filtration is the so-called V-filtration of B. Malgrange and M. Kashi- wara,denoted by (Vk(Dx,t))k∈Z, defined as follows: let us define the V-order of a monomial ordV(λxαtµ∂β∂ν)= ν − µ , t i i X X with λ ∈ C\0, then V (D ) denote the set of operators whose development k x,t only contains monomials xαtν∂β∂µ having order at most k. t We have also the notion of a V-filtration of a D -module: that is an ex- x,t haustive sequence of subspaces (Vk(M))k∈Z such that for any k,k′, we have 3 Vk(Dx,t)Vk′(M)⊂Vk+k′(M). As aboveletDxr,t[m]denotethe free moduleDxr,t with basis e ,...,e endowed with the V-filtration 1 r V (Dr [m])= V (D )e . k x,t k−mi x,t i X IfM isendowedwithaV-filtrationdefinedfromasurjectivemapDr [m]→M, x,t then M admits a V-filtered free resolution: that is an exact sequence ···→Dr1 [m(1)]→Dr0[m(0)]→M →0 x,t x,t which induces for each k an exact sequence of vector spaces ···→V (Dr1 [m(1)])→V (Dr0[m(0)])→V (M)→0. k x,t k x,t k A V-filtered free resolution is the object needed in the computation of the re- striction, as we shall see in the next section. Let us describe now how to define the notion of a minimal V-filtered free resolution of M. To that end we introduce the bifiltration F (D )=F (D )∩V (D ) d,k x,t d x,t k x,t for each d,k ∈ Z. A bifiltration of a D -module M is an exhaustive se- x,t quence of subspaces (F (M)) satisfying F (M) ⊂ F (M)∩F (M) d,k d,k d+1,k d,k+1 and Fd,k(Dx,t)Fd′,k′(M) ⊂ Fd+d′,k+k′(M) for any d,d′,k,k′. It is called a good bifiltration if moreover there exist f ,...,f ∈ M and two vector shifts 1 r n=(n ,...,n ) and m=(m ,...,m ) such that for any d,k, 1 r 1 r F (M)= F (D )f . d,k d−ni,k−mi x,t i X For example, we denote by Dr[n][m] the free module Dr with basis e ,...,e x,t 1 r endowed with the good bifiltration F (Dr [n][m])= F (D )e . d,k x,t d−ni,k−mi x,t i X Let us point out that if M is endowed with a good bifiltration (F (M)), then d,k it is endowed with a good F-filtration F (M) = ∪ F (M) and a V-filtration d k d,k V (M)=∪ F (M). A bifiltered free resolution of M is an exact sequence k d d,k ···→Dr1[n(1)][m(1)]→Dr0[n(0)][m(0)]→M →0 x,t x,t which induces for each d,k an exact sequence of vector spaces ···→F (Dr1 [n(1)][m(1)])→F (Dr0 [n(0)][m(0)])→F (M)→0. d,k x,t d,k x,t d,k TheringD(h) =RD isendowedbyaV-filtrationV (D(h))similarlyasabove, x,t x,t k x,t by giving the weight −1 to each t and the weight 1 to each ∂ . Equivalently, i ti we have V (D(h)) = V (D )Td. In the same way we endow RM with the k x,t d k x,t L 4 V-filtration V (RM)= F (M)Td. For example, we have an isomorphism k d d,k of V-filtered graded D(hL)-modules x,t R(D(r)[n][m])≃(D(h))r[n][m] x,t x,t where the vector shift [n] (resp. [m]) refers to the grading (resp. V-filtration). Let us take a bifiltered free resolution ···→L →L →M →0. (4) 1 0 It induces a V-filtered graded free resolution ···→RL →RL →RM →0, 1 0 thus a bigraded free resolution ···→grV(RL )→grV(RL )→grV(RM)→0. (5) 1 0 Since it makes sense to talk about a minimal bigraded free resolution as (4), the authors of [6] adopt the following definition: a bifiltered free resolution (4) is said to be minimal if so is the bigraded free resolution (5). They prove the existence and uniqueness (up to bifiltered isomorphism) of such a resolution. 2 Free resolutions of complexes of D-modules Inthatsectionweintendtodefinethenotionofaminimalfilteredfreeresolution ofacomplexofD-modulesinthecorrespondingderivedcategory. Wewillreview some facts, perhaps well-known to specialists, for the clarity of our text. Let us denote by CF(D) the category of bounded complexes ··· → M → i M →··· where for each i, M is a F-filtered D-module and the differentials i−1 i are F-adapted. Amapα:M →N issaidtobeafilteredquasi-isomorphismiftheinduced • • map gr(α):grFM →grFN is a quasi-isomorphism. • • If α : M → N is a morphism in CF(D), we denote by C(α) the map- • • ping cone of α, defined by C(α) = M ⊕ N with differential (x,y) 7→ n i−1 i (−φ(x),ψ(y)+α(x)). ItisendowedwiththefiltrationF (C(α) )=F (M )⊕ d i d i−1 F (N ). In that way we have natural isomorphisms F (C(α)) ≃ C(F (α)) and d i d d grF(C(α))≃C(grF(α)). Lemma 2.1. Let α : M → N as above. If for any d, F (α) is a quasi- • • d isomorphism, then grF(α) is a quasi-isomorphism. Proof. F (α) is a quasi-isomorphism is equivalent to saying that C(F (α)) is d d exact. ThusF (C(α))isexactanditeasilyimpliesthatgrF(C(α))≃C(grF(α)) d is exact. That gives the result. We have a converse statement: 5 Proposition 2.1. Let α : M → N be a morphism in CF(D), such that for • • each i, M and N are endowed with good filtrations. If grF(α) is a quasi- i i isomorphism, then α is a quasi-isomorphism and induces for each d a quasi- isomorphism F (α). d Proof. We have that grF(C(α)) ≃ C(grF(α)) is exact. Since each C(α) is i endowedwithagoodfiltration,thenbyastatementanalogousto[6],Proposition 2.5,itfollowsthatC(α) is exact,moreoverfor eachd, F (C(α)) is exact,which d establishes the statement. Corollary 2.1. Let α : M → N be a morphism in CF(D), such that for • • each i, M and N are endowed with good filtrations. Then α is a filtered quasi- i i isomorphism if and only if for each d, F (α) is a quasi-isomorphism. Moreover, d if α is a filtered quasi-isomorphism, then it is a quasi-isomorphism. A complex ···→M →φi M →··· i i−1 in CF(D) is said to be strict if for each i and d we have Imφ ∩F (M ) = i d i−1 φ (F (M )). i d i Lemma 2.2. Let M φ2 //M φ1 // M 2 1 0 α1 N(cid:15)(cid:15) ψ2 // N(cid:15)(cid:15) ψ1 // N(cid:15)(cid:15) 2 1 0 be a commutative diagram of filtered D-modules, such that the rows are com- plexes and α induces isomorphisms H (M )) ≃ H (N )) and H (F (M ))) ≃ 1 1 • 1 • 1 d • H (F (N ))). Then α induces an isomorphism 1 d • 1 F (φ (M )) F (ψ (N )) d 2 2 d 2 2 ≃ . φ (F (M )) ψ (F (N )) 2 d 2 2 d 2 Proof. We have a commutative diagram with exact rows Fd(Kerφ1) // Kerφ1 // Kerφ1 // 0 φ2(Fd(M2)) Imφ2 Fd(Kerφ1)+Imφ2 iso iso (cid:15)(cid:15) (cid:15)(cid:15) (cid:15)(cid:15) Fd(Kerψ1) // Kerψ1 // Kerψ1 //0, ψ2(Fd(N2)) Imψ2 Fd(Kerψ1)+Imψ2 then by the five lemma, α induces an isomorphism 1 Kerφ Kerψ 1 1 ≃ . F (Kerφ )+Imφ F (Kerψ )+Imψ d 1 2 d 1 2 6 Similarly, from the diagram with exact rows 0 // Fd(Kerφ1) // Kerφ1 // Kerφ1 //0 Fd(Imφ2) Imφ2 Fd(Kerφ1)+Imφ2 iso iso (cid:15)(cid:15) (cid:15)(cid:15) (cid:15)(cid:15) 0 // Fd(Kerψ1) // Kerψ1 // Kerψ1 // 0, Fd(Imψ2) Imψ2 Fd(Kerψ1)+Imψ2 we deduce that α induces an isomorphism 1 F (Kerφ ) F (Kerψ ) d 1 d 1 ≃ . F (Imφ ) F (Imψ ) d 2 d 2 Similarly, from the diagram with exact rows 0 // Fd(φ2(M2)) // Fd(Kerφ1) // Fd(Kerφ1) //0 φ2(Fd(M2)) φ2(Fd(M2)) Fd(φ2(M2)) iso iso (cid:15)(cid:15) (cid:15)(cid:15) (cid:15)(cid:15) 0 // Fd(ψ2(N2)) // Fd(Kerψ1) // Fd(Kerψ1) // 0, ψ2(Fd(N2)) ψ2(Fd(N2)) Fd(ψ2(N2)) the result follows. Corollary 2.2. Let α : M → N be a morphism in CF(D) which is a quasi- • • isomorphism and which induces quasi-isomorphisms F (M ) → F (N ). Then d • d • C is strict if and only if D is strict. • • LetKF(D)denotethecategorywhoseobjectsaretheobjectsofCF(D)and themapsaretakenmodulo(F-adapted)homotopies. ThenDF(D)denotesthe localisation of KF(D) with respect to filtered quasi-isomorphisms, as done in [12], 2.1.8 (see also [8]). In other terms, DF(D) is the localization of KF(D) withrespecttothenullsystemcomposedofthecomplexesM suchthatgr(M ) • • is acyclic (see in [7] an introduction to localization of categories). D(D) will denote the derived category of D-modules. Because of Corollary 2.2, the strictness of a complex M , where each M is • i endowed with a good filtration, makes sense in DF(D). As pointedoutin[13],forastrictcomplexM ,foranyi,d,H (F (M ))isa • i d • subspace of H (M ), more precisely we can define a goodfiltrationon H (M ) i • i • by F (H (M))=H (F (M )). d i i d • Let m denotes the maximal graded ideal of grF(D) generated by x ,...,x ,ξ ,...,ξ 1 n 1 n and m(h) the maximal graded two-sided ideal of D(h) generated by x ,...,x ,∂ ,...,∂ ,h. 1 n 1 n We have grF(D)/m≃C and D(h)/m(h) ≃C. 7 Definition 2.1. Let L =(···→L →ψi L →···) a complex of CF(D) with • i i−1 for each i, Li = Dri[n(i)]. The complex L• is said to be minimal if for any i, the matrix representing Rψ as its entries in m(h) (equivalently, for any i, the i matrix representing grψ as its entries in m). i Definition 2.2. Let M be a complex of DF(D). A minimal resolution of M • • is a minimal complex L isomorphic to M in DF(D). • • The numbers r(i) and the shifts n(i) arising in a minimal resolution of M • make sense. In fact, denoting βi,j = dimCTorgirF(D)(grF(M),C)j = dimCTorDi (h)(RF(M),C)j, we have β =card{k,n(i) =j}. In particular r = β . i,j k i j i,j Let us see finally the link between resolutions ofPmodules and resolutions of complexes. Thenotionofastrictfilteredfreeresolutionofacomplexgeneralizes thenotionofafilteredfreeresolutionofamodule: regardingafilteredmoduleM asacomplexinDF(D)concentratedindegree0,astrictfilteredfreeresolution ···→L →L →0ofM (thus H (L )≃M)providesafilteredfreeresolution 1 0 0 • ···→L →L →H (L )→0 in the sense of [11]. 1 0 0 • 3 Filtered restriction LetM be a D -module endowedwith a goodbibiltration(F (M)) suchthat x,t d,k M =D V (M). Weassumethatthereexistsanon-zeropolynomialb(s)∈C[s] x,t 0 suchthatb(t ∂ +···+t ∂ )grV(M)=0,whichisthe caseif M is holonomic. 1 t1 p tp 0 An algorithmto compute such a polynomial can be found in [10]. Let k be an 1 integer such that b(k)6=0 if k>k . 1 Let i : (Cn,0) → (Cn ×Cp,0) denote the embedding x → (x,0) = (x,t). The restriction i∗M of M along Y = {t = ··· = t = 0} is by definition the 1 p complex (D / T D )⊗L M in D(D)-modules. Let Λi = ΛiCp. Then i∗M is reprexs,etntPedi biy txh,te KoDsxz,utl complex over the sequence t1,··· ,tp: 0→M ⊗Λp →δ ···→M ⊗Λ1 →δ M ⊗Λ0 →0. This complex is made of non finitely generated D-modules. By [9], Proposition 5.2 and [10], section 5, the truncation 0→V (M)⊗Λp →δ ···→V (M)⊗Λ1 →δ V (M)⊗Λ0 →0, k1+p k1+1 k1 denoted by V ⊗Λ•, is quasi-isomorphic to the above Koszul complex, thus k1+• still represents i∗M. The bifiltration allows us to endow V ⊗ Λ• with a k1+• filtration F (V ⊗Λ•): d k1+• 0→F (M)⊗Λp →δ ···→F (M)⊗Λ1 →δ F (M)⊗Λ0 →0. d,k1+p d,k1+1 d,k1 8 LetusremarkthatsincethebifiltrationofM isgood,thenthereexistsf ,...,f 1 r such that F (M) = F (D)f . If k is chosen so that for any i, d,k i d−ni,k−mi i 1 k1 ≥mi, then each fi bPelongs to Vk1(M) and for each i=0,...,p, Fd,k1+i(M) is a good filtration of V (M). k1+i Let ···→L →ψ1 L →ψ0 M →0 1 0 be any bifiltered free resolution of M, with L =Dr(i)[n(i)][m(i)]. In particular i x,t it is a V-filtered free resolution. By [10], Theorem 5.3, the complex V (L ) V (L ) ···→ k1 1 → k1 0 →0, t V (L ) t V (L ) i i k1+1 1 i i k1+1 0 P P denoted by V (L )/ t V (L ) is a free complex isomorphic to i∗M in k1 • i i k1+1 • D(D). We endow it wPith a F-filtration by setting V (L ) F (L ) F k1 i = d,k1 i d (cid:18)tV (L )(cid:19) F (L )∩tV (L ) k1+1 i d,k1 i k1+1 i F (L ) ≃ d,k1 i . tF (L ) d,k1+1 i NotethatthefilteredD-moduleV (L )/ t V (L )isnaturallyisomorphic k1 i i i k1+1 i to some Dr[n]. P Proposition 3.1. The complexes V (M)⊗Λ• and V (L )/ t V (L ) k1+• k1 • i i k1+1 • are isomorphic in DF(D). Thus Vk1(L•)/ itiVk1+1(L•) is a fiPltered free res- olution of i∗M. P Proof. Since t ,...,t is a regular sequence in O , we have that the Koszul 1 p x,t complex R (D ) 0→R (D )⊗Λp →···→R (D )⊗Λ0 → F,V x,t →0 F,V x,t F,V x,t t R (D ) i i F,V x,t P isexact. WemayreplaceD bythebifilteredfreemoduleL ,thenthecomplex x,t i R (L ) 0→R (L )⊗Λp →δ ···→R (L )⊗Λ0 → F,V i →0 F,V i F,V i t R (L ) i i F,V i P is exact. Moreoverit is bigraded, thus the complex F (L ) 0→F (L )⊗Λp →δ ···→F (L )⊗Λ0 → d,k1 i →0 d,k1+p i d,k1 i t F (L ) i i d,k1+1 i P 9 is exact. We have a commutative diagram: 0 0 OO OO ··· //F (M)⊗Λ1 // F (M)⊗Λ0 // 0 d,k1+1 OO d,k1 OO OO ··· // Fd,k1+1(LOO 0)⊗Λ1 //Fd,k1(LOO0)⊗Λ0 // PitFidF,dk,1kOO(1L+01)(L0) // 0 ··· // Fd,k1+1(LOO 1)⊗Λ1 //Fd,k1(LOO1)⊗Λ0 // PitFidF,dk,1kOO(1L+11)(L1) // 0 . . . . . . . . . Wededucethatthecomplexes(F (M)⊗Λi) and(V (L )/ t V (L )) d,k1+i i k1 i i i k1+1 i i are both isomorphic to the complex associated with the doublePcomplex (F (L )⊗Λj) . d,k1+j i i,j Corollary 3.1. The complex V (L )/ t V (L ) is strict if so is the com- k1 • i i k1+1 • plex Vk1+•(M)⊗Λ•. P Letusconsiderfromnowonaspecialcase: p=1. ThecomplexV (M)⊗ k1+• Λ• is V (M) →t V (M). Let us assume furthermore that t : M → M is k1+1 k1 injective. Therestrictioni∗M isconcentratedindegree0withH i∗M ≃M/tM. 0 Thus the complex V (L ) V (L ) M ···→ k1 1 → k1 0 → →0 (6) tV (L ) tV (L ) tM k1+1 1 k1+1 0 is a free resolution of the module M/tM = (V (M)+tM)/tM. That module k1 is naturally endowed with the filtration F (M) F (M/tM):= d,k1 . d F (M)∩tM d,k1 and from Proposition 3.1 and Corollary 3.1 we obtain: Proposition 3.2. Assume that t : M → M is injective and that for any d, F (M)∩tV (M) = tF (M). Then the complex (6) is an F-filtered d,k1 k1+1 d,k1+1 free resolution of M/tM, i.e. for any d the complex F (L ) F (L ) M ···→ d,k1 1 → d,k1 0 →F →0 (7) d tF (L ) tF (L ) (cid:18)tM(cid:19) d,k1+1 1 d,k1+1 0 10