Arxiv:2104.06941V2 [Math.AG] 26 May 2021

RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES

LEI WU

Abstract. We establish a Riemann-Hilbert correspondence for Alexander complexes (also known as Sabbah specialization complexes) by using relative holonomic D-modules in an equivariant way, which particularly gives a “global” approach to the correspondence for Deligne’s nearby cycles. Using the correspondence and zero loci of Bernstein-Sato ideals, we obtain a formula for the relative support of the Alexander complexes.

Contents 1. Introduction 1 2. Sheaﬁﬁcation of sheaves of modules over commutative rings 8 3. Relative maximal and minimal extensions along F 17 4. Alexander complex of Sabbah 34 5. Comparison 35 References 40

1. Introduction Let f be a holomorphic function on a complex manifold X with D the divisor of f and let Ux be a small open neighborhood of x ∈ D. Then we consider the ﬁber product diagram ∗ ∗ ∗ Uex = (Ux \ D) ×C Ce Ce exp

f ∗ Ux \ D C ∗ ∗ ∗ arXiv:2104.06941v3 [math.AG] 1 Sep 2021 where exp: Ce → C is the universal cover of the punctured complex plane C . ∗ The deck transformation induces a C[π1(C )]-module structure on the compactly i supported cohomology group Hc(Uex, C), which is called the i-th (local) Alexander module of f. Taking Ux sufficiently small, the Alexander modules contain the information of the cohomology groups of the Milnor fibers around x together with their monodromy action. As x varies along D, all the local Alexander modules ∗ give a constructible complex of sheaves of C[π1(C )]-modules, which recovers the Deligne nearby cycle of the constant sheaf along f (see [Bry86]). Sabbah [Sab90] made generalizations for a finite union of holomorphic functions and obtained what he called the Alexander complexes (see §4 for construction).

2010 Mathematics Subject Classiﬁcation. 14F10; 13N10; 32C38; 32S60; 32S55; 14F43; 14L30. 1 2 LEI WU

In this article, we establish a “global” Riemann-Hilbert correspondence for the Alexander complexes in a functorial way (Theorem 1.1). A “local” correspondence for complete specializations of Alexander complexes was constructed by Sabbah [Sab90, Theorem 5.1.2]; see also [BG12] for the algebraic “local” case along a single regular function following the approach of Beilinson-Bernstein [Bei87]. Riemann- Hilbert correspondence between nearby cycles of regular holonomic D-modules and Deligne nearby cycles of complex perverse sheaves was constructed by Kashiwara [Kas83] and Malgrange [Mal83] by using V -filtrations along a single holomorphic function. However, there is no well-defined notion of “normalized” multi indexed V -filtrations along a finite union of holomorphic functions in general (based on the work of Sabbah [Sab87a, Sab87b]; see also [Wu20, Remark 4.22]). Thus, one would not be able to generalize the method of Kashiwara and Malgrange to construct the correspondence for Alexander complexes in general. Roughly speaking, our approach to the “global” correspondence can be seen as a hybrid of the method of Sabbah [Sab87a, Sab87b] and that of Beilinson-Bernstein [Bei87]. To be more precise, the construction of the correspondence relies heavily on the work of Maisonobe [Mai16a] in the study of Bernstei-Sato ideals by using relative holonomic D-modules over algebraic affine spaces and its development in [WZ21, BVWZ21a, BVWZ21b], and on the theory of analytic relative holonomic D- modules and relative constructible complexes developed in a series of fundamental papers [MFS13, MFS19, FF18, FFS21]. The new input we need to establish the correspondence is a GAGA-type principle between analytic and algebraic relative D-modules and between algebraic and analytic relative constructible sheaves, which is called in this paper the sheafification of relative modules over commutative rings R (see §2). The other input is the construction of relative holomorphic D-modules that are G-equivariant with respect to the universal covering of (C∗)r (see §3.4). Let us briefly explain the terminology of sheafification. When the base space X is a point, the sheafification of R-modules is just the functor transforming R-modules into quasi-coherent O-modules on Spec R. For general base spaces, it is the relative analogue for complexes of sheaves of R-modules. One then make the base change an onto Spec R to obtain the analytic one for R commutative finite generated C- algebras. See Remark 3.4 for the reason why the analytic sheafification is necessary in proving Theorem 1.1 and Theorem 1.2.

1.1. Riemann-Hilbert correspondence for Alexander complex. Let fi be holomorphic functions on a complex manifold X for i = 1, . . . , r with Di the divisor of fi. P We write F = (f1, . . . , fr), and D = i Di. Suppose that M is a (left) holonomic DX -module. We write by

M(∗D) = M ⊗O OX (∗D) S the algebraic localization of M along D, where OX (∗D) = OX (kD) is the k∈Z sheaf of meromorphic functions with poles along D. To establish the Riemann-Hilbert correspondence for Alexander complexes of Sabbah, we first construct the relative maximal (resp. minimal) extensions of M (r) (r) along F , denoted by M(∗DF ) (resp. M(!DF )), which are both relative holonomic r r r r DX×C /C -modules where DX×C /C is the sheaf of relative holomorphic differential operators with respect to the projection pr : X × Cr → Cr (see Definition 2.4 for (r) relative holonomicity). More precisely, M(∗DF ) is the analytic sheafification of RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 3 the DX [s]-module s r M(∗D) ⊗C C[s] · F , s = (s1, . . . sr) the algebraic coordinates of C (r) (r) r r and M(!DF ) is the DX×C /C -dual of (DM)(∗DF ), where

s Y si F = fi i is a formal symbol and D(M) is the DX -dual of M. See §3.3 for details. It is worth (r) mentioning that when M = OX , OX (∗DF ) gives a non-trivial example of relative Deligne meromorphic extensions of relative local systems (see §4.2 and also [FS19, §2.2] and [NS96, §8]). The key property for the maximal and minimal extensions is that there exists a natural inclusion (Lemma 3.7) (r) (r) (1.1) M(!DF ) ,→ M(∗DF ), which is a “global”, sheafified and higher dimensional generalization of a classical result of Beilinson and Bernstein (see [Gin86, Proposition 3.8.3] and also [BG12]). We then define a relative holonomic D-module on X × Cr, M(∗D(r)) Ψ (M) = F . F (r) M(!DF ) We denote by √ √ r ∗ r Exp: C → (C ) , (α1, . . . , αr) 7→ (exp(−2π −1α1), . . . , exp(−2π −1αr)) the universal covering of (C∗)r and by r ∗ r π = (id, Exp): X × C → X × (C ) ∗ r ∗ r the induced map. We write G = π1((C ) ), the fundamental group of (C ) . The universal covering makes X×Cr a G-space with the quotient G\X×Cr = X×(C∗)r. The operation s s ti(F ) = fiF with ti representing the (counterclockwise) loops around the puncture of each factor ∗ ∗ r (r) (r) C of (C ) , makes M(∗DF ), M(!DF ) and ΨF (M) all G-equivariant sheaves (see Theorem 3.8, Theorem 3.9 and Remark 3.10(1)).

Theorem 1.1. With the notation as above, if M is a regular holonomic DX -module, then we have a natural quasi-isomorphism G r r π∗ DRX×C /C (ΨF (M)) ' ψeF (DRX (M)), G r r where π∗ is the equivariant direct image functor, DRX×C /C is the relative de r r Rham functor on X × C over C and ψeF (DRX (M)) is the analytic sheaﬁﬁcation of the Alexander complex ψF (DRX (M)). G r r The C[G]-module structure of π∗ DRX×C /C (ΨF (M)) is induced by G r ∗ r Exp∗ (OC ) ' O(C ) and the natural inclusion ±1 ±1 ∗ r C[G] ' C[t1 , . . . , tr ] ,→ O(C ) , 4 LEI WU where in the inclusion C[G] is treated as a constant sheaf on (C∗)r. Theorem 1.1 is related to the local comparision for Alexander complexes of Sabbah [Sab90, Theorem 5.1.2]. To be more precise, one can take the (complete) localization of the quasi-isomorphism in Theorem 1.1 at a point λ ∈ (C∗)r to obtain the local G r r comparison for ψF (DRX (M)). Notice that localizing π∗ DRX×C /C (ΨF (M)) at −1 r r λ is equivalent to localizing DRX×C /C (ΨF (M)) at α for every α ∈ Exp (λ), since G acts freely on Cr with G\Cr = (C∗)r. In particular, when r = 1, localizing the quasi-isomorphism in Theorem 1.1 at λ ∈ C∗ gives the comparison between the α-nearby cycle of M and the λ-nearby cycle of DR(M) along f for every α ∈ Exp−1(λ) (see [Kas83, Mal83, BG12] and also [Wu21]). Notice that the G- action in Theorem 1.1 explains why the correspondence between α-nearby cycles of regular holonomic D-modules and λ-nearby cycles of perverse sheaves along a single holomorphic function is Z-to-1. Since the relative de Rham functor is G-equivariant (see §3.5), we have a natural isomorphism

G G ∗ r ∗ r r r DRX×(C ) /(C ) π∗ (ΨF (M)) ' π∗ DRX×C /C (ΨF (M)) .

∗ r ∗ r Then for regular holonomic DX -modules M, DRX×(C ) /(C ) (•) gives a relative Riemann-Hilbert correspondence on X × (C∗)r over (C∗)r

G π∗ (ΨF (M)) ψeF (DRX (M)) with “←−” induced by the Riemann-Hilbert correspondence for regular holonomic D-modules on X. See also [FFS21] for a relative Riemann-Hilbert correspondence over curves and §5.4 for a related open question. Theorem 1.1 can be further reﬁned to a comparision for higher-codimensional Sabbah specialization complexes (see §4.1) as follows by using sheaves of (algebraic) local cohomology.

Theorem 1.2. In the situation of Theorem 1.1, for every subset I ⊆ {1, 2, . . . , r} we have quasi-isomorphisms

G (r) π DR r r RΓ r ( M)(!D ) ∗ X×C /C D [DI ×C ] D F G (r) ' DR ∗ r ∗ r RΓ ∗ r π ( M)(!D ) X×(C ) /(C ) D [DI ×(C ) ] ∗ D F

' ψeDI (DRX (M)), T where ψDI (DRX (M)) is the Sabbah specialization complexes along DI = i∈I Di.

1.2. Relative Support. We now discuss the relative supports of ΨF (M) and the Alexander complex of DR(M) (see Deﬁnition 2.2 and Proposition-Deﬁnition 2.7). Applying Theorem 1.1, we obtain:

Corollary 1.3. Let M be a regular holonomic DX -module on a complex manifold X and let F = (f1, . . . , fr) be an r-tuple of holomorphic functions on X. Then supp ∗ r ψ (DR(M)) = Exp supp r Ψ (M) (C ) F C F and −1 supp r Ψ (M) = Exp supp ∗ r ψ (DR(M)) . C F (C ) F RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 5

The action of G on Cr induces an action of G on its algebraic coordinate ring C[s], i.e. ( sj + 1, if i = j ti · sj = sj, if i 6= j. Using Bernstein-Sato ideals (see §2.4 and §3.6), we obtain a geometric description of the relative support of ΨF (M): Theorem 1.4. For each pair (M,F ) as in Corollary 1.3, locally on a relative compact open subset 1 of X there exist finite sets r S(F, M) ⊆ N≥0 and κ(L) ⊆ C for each L ∈ S(F, M) such that [ [ [ n supp r Ψ (M) = (g · (L · s + α) = 0) ⊆ , C F C g∈G L∈S(F,M) α∈κ(L) where (g · (L · s + α) = 0) denotes the divisor inside Cr defined by g · (L · s + α) = 0. In the theorem above, S(F, M) is the union of all (primitive) slopes of the codimension-one components of the zero locus of a Bernstein-Sato ideal of M along F , and for every L ∈ S(F, M) the finite set κ(L) is the set of all α ∈ C such that L · s + α = 0 defines an irreducible codimension-one component of the zero locus of the Bernstein-Sato ideal. See §3.6 for details. Corollary 1.3 and Theorem 1.4 together give: Corollary 1.5. Locally on a relative compact open subset of X, √ [ [ L ∗ r supp ∗ r ψ (DR(M)) = (t = exp(2π −1α)) ⊆ ( ) , (C ) F C L∈S(F,M) α∈κ(L) √ where tL = Q tli with L = (l , . . . , l ) and (tL = exp(2π −1α)) is the divisor i i √ 1 r L ∗ r defined by t = exp(2π −1α) in ( ) . In particular, supp ∗ r ψ (DR(M))is a C (C ) F finite union of translated codimensional-one subtori inside (C∗)r.

The study of supp ∗ r ψ (DR(M)) was initiated by Sabbah [Sab90], where the (C ) F support is shown to be included in a union of translated subtori. In the case

M = , it is proved in [BLSW17, Theorem 1.3] that supp ∗ r ψ ( ) is a ﬁnite OX (C ) F CX union of torsion translated codimensional-one subtori from a purely topological approach. Corollary 1.5 gives a precise description of supp ∗ r ψ (DR(M)) in (C ) F general. If additionally M is quasi-unipotent along F (for instance M = OX ), then κ(L) ⊆ for every L and hence supp ∗ r ψ (DR(M)) is a ﬁnite union of Q (C ) F torsion translated subtori inside (C∗)r. In consequence, Theorem 1.1, Corollary 1.3, Theorem 1.4 and Corollary 1.5 together give a generalization of [BVWZ21a, Theorem 1.5.1] as well as [BG12, Theorem 2 and Theorem 3].

As mentioned above, supp ∗ r ψ ( ) has a purely topological interpretation by (C ) F CX using (local) cohomology jumping loci [BLSW17]. But the cohomology jumping loci (of quasi-projective varieties or analytic germ complements) might contain torsion- translated subtori of higher codimension [BW15, BW17]. Then one can naturally ask how to give a D-module interpretation of such lower dimensional loci. Along this

1The condition “locally on a relative compact open subset” (here and elsewhere it appears in this article) is to ensure finiteness. Thus, if we are in the algebraic category, then this condition can usually be eliminated. 6 LEI WU line, we propose a lineariality conjecture (Conjecture 3.20) by using the codimension filtration of Gabber-Kashiwara under the relative setting (see §2.8). The lineariality conjecture is interesting because it would imply a conjecture of Budur [Bud15] that the zero locus of the Bernstein-Sato ideal of F is a finite union of translated linear subspaces of Cr defined over Q (see Proposition 3.22).

1.3. Relative characteristic cycle and monodromy zeta function. Our next results are about understanding the irreducible divisor

r (g · (L · s + α) = 0) ⊆ C inside supp r Ψ (M) and their relations with relative characteristic cycles and the C F monodromy zeta functions. We now assume M a regular holonomic DX -module and F = (f1, . . . , fr) an r-tuple of holomorphic functions on a complex manifold X. Theorem 1.6. With assumptions above , we have the relative characteristic cycles

rel (r) rel (r) r ∗ r CC (M(∗DF )) = CC (M(!DF )) = CC(M(∗D)) × C ⊆ T X × C , where CCrel and CC denote the relative characteristic cycle and the absolute one respectively and similarly for Ch. Moreover, locally on a relative compact open subset W of X, the relative characteristic cycle of ΨF (M) is an inﬁnite sum

rel X X X CC (ΨF (M)) = ΛL,α × (g · (L · s + α) = 0) g∈G L∈S(F,M) α∈κ˜(L) where each ΛL,α is a Lagrangian cycle supported on Ch(M(∗D)) (over D ∩ W ), and κ˜(L) is κ(L) modulo L-equivalence (Deﬁnition 3.16 ).

The relative characteristic cycle of ΨF (M) is a symmetric infinite sum because of the G-action on ΨF (M) (if a G-equivariant sheaf is supported on a point, then it is supported on the G-orbit of the point). If Wf is a relative compact open subset of X × Cr, then X X X ΛL,α × (g · (L · s − α) = 0) Wf g∈G L∈S(F,M) α∈κ˜(L) becomes a finite sum since X × (g · (L · s + α) = 0) intersect Wf for only finite many g ∈ G. See Proposition 3.19 for formulas of ΛL,α in terms of the characteristic cycles of very general choices of one-dimensional nearby cycles associated to M. (r) The relative characteristic cycle formula of M(∗DF ) in Theorem 1.6, as well as the irregular case in Theorem 3.3, has an application to the study of constructibility of the logarithmic de Rham complexes of lattices of holonomic modules [Wu20, Theorem 1.1]. We now fix an arbitrary point x ∈ X. By Theorem 1.4 we can focus on S(F, M) andκ ˜(L) (or κ(L)) for L ∈ S(F, M) locally around x. By Theorem 1.6, both S(F, M) andκ ˜(L) (or κ(L)) are constructible as x varies on X and hence so are supp ∗ r ψ (DR(M)) and supp r Ψ (M). More precisely, we fix a finite Whitney (C ) F C F stratification G G X = X such that Ch(M(∗D)) ⊆ T ∗ X, β Xβ β β RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 7 where T ∗ X is the conormal bundle of the smooth strata X . Then S(F, M) and Xβ β κ˜(L) for L ∈ S(F, M) are locally constant with respect to the above stratification. Moreover, for L ∈ S(F, M) and α ∈ κ˜(L) locally around x, we have X Λ = m (L, α)T ∗ X, m (L, α) ∈ . L,α β Xβ β Z≥0 β

Next, we give a formula for the multiplicity mβ(L, α) by using the monodromy zeta function. By construction, ψF (DR(M)) is a C-constructible complex of sheaves of C[G]-modules (see [KS13, §8.5] for the deﬁnition of C-constructibility). Hence, ∗ r for each generic point q of Supp(C ) (ψF (DR(M)))

ψF,q(DR(M)) = ψF (DR(M)) ⊗C[G] C[G]q is a C-constructible complex of sheaves of C[G]q-modules, where C[G]q is the localization of C[G] at the prime ideal q. We deﬁne

X i i χx(ψF (DR(M)), q) = (−1) lg(H ψF,q(DR(M))|x) i where lg denotes the length function, and X ζx(ψF (DR(M))) = χx(ψF (DR(M)), q)¯q q

∗ r where q goes over the set of all the generic points of Supp(C ) (ψF (DR(M))) (since ∗ r ∗ r Supp(C ) (ψF (DR(M))) is an algebraic closed subset of (C ) by Corollary 2.21). By C-constructibility, χx(ψF (DR(M)) is a Z-valued constructible function on X and ζx(ψF (DR(M))) is a constructible function on X valued in the abelian group ∗ r of algebraic subvarieties in (C ) . The constructible function ζx(ψF (DR(M))) is the so-called monodromy zeta function of ψF (DR(M)) (cf. [Sab90, §2.5]).√ We denote by q(L, α) ∈ Spec C[G] the generic point of (tL − exp(2π −1α) = 0) for L and α. With the help of Corollary 1.5, we obtain a precise formula: (1.2) √ Y Y L χx(ψF (DR(M)),q(L,α)) ζx(ψF (DR(M))) = (t − exp(2π −1α)) , L∈S(F,M) α∈κ˜(L) where we make the constructible function valued in rational functions. Using a local index formula of Kashiwara-Dubson-Ginsburg [Gin86, Theorem

8.2], we get a formula for χx(ψF,q(DR(M))) and mβ1 (L, α):

X dim Xβ (1.3) mβ1 (L, α) = (−1) c(Xβ1 ,Xβ)χxβ (ψF (DR(M)), q(L, α))), ¯ Xβ1 ⊆Xβ where ( χtop(B ∩ X ∩ H), if X 6= X xβ1 β β1 β c(Xβ1 ,Xβ) = 1, if Xβ1 = Xβ. with B a small polydisc open neighborhood inside X of some x ∈ X , H a xβ1 β1 β1 top linear subspace of X suﬃciently close to xβ1 of codimension dim Xβ1 + 1 and χ denoting the topological Euler characteristic. 8 LEI WU

1.4. Outline of the paper. In Section 2, we discuss sheaﬁﬁcation of sheaves of modules on topological spaces over commutative rings and discuss the properties of algebraic and analytic relative supports. Section 3 is about the construction of maximal and minimal extensions under the relative setting and their G-equivariance. In Sec- tion 4, we recall the construction of the Alexander complexes of Sabbah. In Section 5, we discuss comparisons in the sense of relative Riemann-Hilbert correspondence and prove Theorem 1.1 and Theorem 1.2. Acknowledgement. The author would like to thank Nero Budur, Linquan Ma, Mircea Musta¸tˇa,Ruijie Yang and Peng Zhou for useful discussions and comments and for answering questions. The author was supported by the grant Methusalem METH/15/026 from KU Leuven.

2. Sheafification of sheaves of modules over commutative rings In this section, we discuss sheafifying sheaves of modules over commutative rings in a standard way, motivated by the study of relative holonomic D-modules over algebraic affine spaces in [Mai16a] and the theory of analytic relative holonomic D-modules developed in [MFS13, MFS19, FF18, FFS21]. Compared to the references above, we mainly and more focus on studying algebraic and analytic relative supports and their differences.

2.1. Pre-sheafification. Let k be a field, and let X be topological spaces with A a sheaf of k-algebra on X for some base field k. For a commutative k-algebra R, we obtain a new sheaf of rings AR = A ⊗k R. 2 We write by Mod(AR) the abelian category of AR-modules on X. For M ∈ Mod(AR), we define a presheaf on X × Spec R (use the product topology) by pre Mf (U × Spec Rf ) = M (U) ⊗R Rf where Rf is the localization of R with respect to f ∈ R. Since Spec Rf give a basis of the Zariski topology of Spec R, the presheaf above gives a sheaf on X × Spec R, denoted by Mfpre. We call such procedure the relative pre-sheafification of M . pre pre Consequently, Mf is an AfR -module on X × Spec R. One can see that when X is a point, the relative pre-sheafification is the ∼-functor in algebraic geometry transforming R-modules into quasi-coherent sheaves on Spec R.

Lemma 2.1. The stalk of Mfpre at (x, m) ∈ X × Spec R satisﬁes pre Mf(x,m) = Mx ⊗R Rm = (M ⊗R Rm)x where Mx is the stalk of M at x and similarly for (M ⊗R Rm)x.

Proof. Since localizations are colimits, we apply [Sta18, Part 1, Lemma 4.14.10]. Let I ⊆ R be an ideal. The relative pullback of M to Spec R/I is deﬁned as pre ˜∗ pre ipre(Mf ) = M ^⊗R R/I

2Throughout this paper, all the modules over non-commutative rings are assumed to be the left ones, unless indicated otherwise. RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 9 where ˜i: X×Spec R/I ,→ X×Spec R is the closed embedding induced by R → R/I. More generally, if h: Spec S → Spec R is a morphism of aﬃne schemes, then one ˜∗ can deﬁne the relative pullback functor hpre similarly. If m ⊆ R is a maximal ideal, the pullback of M at the closed point of m gives a A -module on X. Hence, Mfis also called a relative A -module over Spec R. Let p ⊆ R be a prime ideal. Then the specialization of M at p is

Mp = M ⊗R Rp where Rp is the localization of R at p, which is a ARp -module on X. The complete specialization of M at p is ˆ ˆ Mp = M ⊗R Rp ˆ ˆ ˆ where Rp is the completion of Rp with respect to p. Then Mp is an ARp -module.

Deﬁnition 2.2. For M ∈ Mod(AR), the relative support of M is

suppSpec RM = {maximal ideals m ∈ Spec R | Mm 6= 0}. • b More generally, if M ∈ D (AR), the bounded derived category of Mod(AR), then we define • [ i • suppSpec RM = suppSpec RH (M ). i 2.2. Analytic sheafification. We now assume X a complex manifold, A a sheaf of an C-algebra and R a commutative C-algebra of finite type. We write Spec R the analytic scheme of Spec R. For M ∈ Mod(AR), the analytic sheafification of M is defined as −1 pre −1 Mf= ι (Mf ) ⊗ −1 prX OSpecanR, prfX OSpec R where ι : X × SpecanR → X × Spec R is the natural inclusion map induced by an an an ιR : Spec R → Spec R from GAGA principle, prX : X × Spec R → Spec R is the projection and pre X is the composition of prX and ιR. The output Mfgives a an AfR-module on X × Spec R. If X is an algebraic scheme over k and R a commutative k-algebra, then we can define the algebraic sheafification of M ∈ Mod(AR) by alg −1 pre Mf = ιalg(Mf ) −1 where ιalg : XR → X × Spec R is the natural inclusion map with XR the scheme of X over R.

2.3. Analytic sheafification for DX,R-module and relative holonomicity. Now we assume A = DX , the sheaf of holomorphic differential operators on a complex manifold X, and R a commutative finite generated C-algebra. In this case, we also want the analytic sheafification to be consistent with the analytic structure sheaf of X × SpecanR (not only with that of the second factor). For instance, the sheafification of an OX,R-module should be its GAGA analytification. Thus, for M ∈ Mod(DX,R) (or M ∈ Mod(OX,R)) we define −1 pre Mf= ι (Mf ) ⊗ −1 pre OX×SpecanR. ι (OeX,R)

Therefore, the output Mfin this case gives a DeX,R = DX×SpecanR/SpecanR-module, i.e. a relative analytic D-module. Let us refer to [Sch12, Chapter III. 1.3] and 10 LEI WU

[MFS13, §3] for the general theory of relative analytic D-modules; see also [Wu20, §2]). The relative pullback for Mfin this case is exactly the base change for relative D-modules [Wu20, §2.2]. Remark 2.3. When X is an smooth complex algebraic variety and M is an algebraic DX,R-module, we define the algebraic sheafification by alg −1 pre Mf = ι (Mf ) ⊗ −1 pre OXR , alg ι (OeX,R) which gives an algebraic relative D-module over Spec R. In this case, the functor ∼alg has a quasi-inverse p∗: alg p∗(Mf ) ' M , where p: XR → X is the natural projection. an Definition 2.4. For M ∈ Mod(DX,R), we say M relative coherent over Spec R if Mfis coherent over DeX,R. We say M (or Mf∈ Mod(DeX,R)) relative holonomic if it is relative coherent and locally on a relative compact open subset W of X rel an Ch (Mf|W ×SpecanR) ⊆ Λ × Spec R, where Λ is a (possibly reducible) conic Lagrangian subvariety inside the cotangent bundle T ∗X (over W) and Chrel(Mf) is the relative characteristic variety.

Notice that coherence over DX,R on X implies relative coherence, but not vice versa. The relative holonomicity above is the analytic sheafification of the algebraic relative holonomicity [BVWZ21a, Definition 3.2.3], which in turn follows the analytic relative holonomicity defined in [MFS13, §3.4]. By definition, the category of relative holonomic modules is abelian. The following lemma is well-known; see for instance [Mai16a, Proposition 8], [FF18, Lemma 2.10]3 and also [BVWZ21a, Proposition 3.2.5].

Lemma 2.5. If Mf∈ Mod(DeX,D) is relative holonomic, then for every pair (W, V ) with W a relative compact open subset of X and V a relative compact open subset an of Spec R, there exist a finite number of closed analytic subvarieties Sw ⊆ V such that rel [ Ch (Mf|W ×V ) = Λw × Sw, w ∗ where Λw are irreducible conic Lagrangian subvarieties in T X|W . If moreover an M ∈ Mod(DX,R) is coherent over DX,R and relative holonomic over Spec R, then rel [ Ch (Mf|W ×SpecanR) = Λw × Sw w an with a finite number of closed analytic subvarieties Sw ⊆ Spec R. Proof. For completeness, we present here a proof essentially due to Maisonobe [Mai16a]. We write rel an Ch (Mf|W ×SpecanR) ⊆ Λ × Spec R P with Λ = w Λw a finite union of irreducible conic Lagrangian subvarieties in ∗ rel T X|W . Notice that the inclusion above does not imply that Ch (Mf|W ×SpecanR)

3Lemma 2.10 in [FF18] seems to be not completely correct since locally on X the index set I can be infinite by Theorem 1.6. RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 11 has finite many irreducible components. For every p ∈ V , by the relative Bernstein inequality (see [Mai16a, Proposition 5] and also [Wu20, Theorem 2.2]), we have rel rel −1 X Ch (Mf)p := Ch (Mf) ∩ pr (p) ⊂ Λw, w where pr : T ∗X × SpecanR → SpecanR is the second projection. Then we define rel Sw = {p ∈ V | Λw ⊆ Ch (Mf)p}, from which we obtain rel [ Ch (Mf|W ×V ) = Λw × Sw. w

The set Sw is analytic and closed since rel −1 {λw} × Sw = Ch (Mf|W ×V ) ∩ pr1 (λw) ∗ an ∗ with λw a general point on Λw, where pr1 : T X × Spec R → T X is the ﬁrst projection. If additionally M is coherent over DX,R, we can deﬁne an rel Sw = {p ∈ Spec R | Λw ⊆ Ch (Mf)p}. rel ∗ an Since Ch (Mf|W ×SpecanR) is a closed analytic subvariety of T W × Spec R, we an conclude that Sw ⊆ Spec R is analytic and closed. −1 an For a maximal ideal m ∈ Spec R, we write pm = ιR (m) ∈ Spec R. For an (x, p) ∈ X × Spec R, Mf(x,p) denotes the (analytic) localization of Mfat (x, p). Lemma 2.6. If M is relative coherent over SpecanR, then

suppSpec RM = {m ∈ Spec R | Mf(x,pm) 6= 0 for some x ∈ X}.

Proof. We first assume R = C[y] for y = (y1, . . . , yl) for some integral l > 0. Since completion of local rings is faithfully flat, we have ˆ Mm 6= 0 ⇔ Mm 6= 0. By definition, we have ˆ ˆ Mm 6= 0 ⇔ (Mm)x for some x ∈ X, ˆ ˆ ˆ where (Mm)x is the localization of Mm at x because Mm is a sheaf on X. We write by C{x} the stalk of OX at x, and by C[[x]] the ring of formal power series. Since C[[x]] is faithfully flat over C{x}, we know ˆ ˆ (Mm)x 6= 0 ⇔ (Mm)x ⊗C{x} C[[x]] 6= 0. We also have ˆ (Mm)x ⊗ {x} [[x]] ' Mx ⊗ {x}⊗ R [[x]][[y]]. C C C C C Thanks to the fact that the completion of local rings is faithfully flat again, we have 6= 0 ⇔ ⊗ [[x, y]] 6= 0. Mf(x,pm) Mf(x,pm) C{x,y} C By Lemma 2.1, we have

Mx ⊗ {x}⊗ R [[x]][[y]] ' M(x,p ) ⊗ {x,y} [[x, y]]. C C C f m C C Therefore, we conclude

Mm 6= 0 ⇔ Mf(x,pm) 6= 0 for some x ∈ X. 12 LEI WU

In general, considering a surjection

C[y] R, we are done by similar arguments.

an Proposition-Deﬁnition 2.7. Let M ∈ Mod(DX,R) be relative coherent over Spec R. Then −1 rel suppSpecanRM := ιR (suppSpec RM ) = suppSpecanRMf:= pr(Ch (Mf)), where pr : T ∗X × SpecanR → SpecanR is the projection. Proof. We focus ourselves on the characteristic variety

Chrel(Mf) ⊆ T ∗X × SpecanR

rel locally over W × V by picking a good ﬁltration F•Mf. Then Ch (Mf) is the F ∗ an rel support of gre • Mfinside T X ×Spec V , which means Ch (Mf) is a closed analytic subvariety over W × V . We observe by coherence

F ∗ Mf(x,p) 6= 0 ⇔ (gre • Mf)(x,ξ,p) 6= 0 for some ξ ∈ Tx X. Thus, we have

rel ∗ Mf(x,p) 6= 0 ⇔ (x, ξ, p) ∈ Ch (Mf) for some ξ ∈ Tx X. By Lemma 2.6, we hence conclude

−1 rel ιR (suppSpec RM ) = pr(Ch (Mf)) locally over W × V .

2.4. Bernstein-Sato ideal. If M ∈ Mod(DX,R) is coherent over DX,R, then we deﬁne

B(M ) = AnnR(M ) ⊆ R, calling it the Bernstein-Sato ideal of M . We write by Z(B(M )) ⊆ Spec R the zero locus of B(M ). A priori, since M is not ﬁnite generated over R in general, we only have

suppSpec RM ⊆ Z(B(M )), but we do not know if suppSpec RM is an algebraic subvariety.

Proposition 2.8. If M ∈ Mod(DX,R) is coherent over DX,R and relative holonomic over SpecanR, then locally on a relative compact open subset W ⊆ X we have

suppSpec RM |W = Z(B(M |W )). Proof. For simplicity, we assume X = W . By [Mai16a, Proposition 9] (see also [BVWZ21a, Lemma 3.4.1]), we have

Z(B(M )) = pr(Chrel(Mf)).

Then the required equality follows from Proposition-Deﬁnition 2.7. By the proposition above and Proposition 2.5, we have: RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 13

Corollary 2.9. If M ∈ Mod(DX,R) is coherent over DX,R and relative holonomic over SpecanR, then over a relative compact open subset W ⊆ X

rel [ Ch (Mf|W ×SpecanR) = Λw × Sw w with a ﬁnite number of closed algebraic subvarieties Sw ⊆ Spec R. Applying Proposition-Deﬁnition 2.7, Lemma 2.5 and Proposition 2.8, we have:

an Corollary 2.10. Let M ∈ Mod(DX,R) be relative holonomic over Spec R. Then an (1) suppSpecanRM is analytically closed locally over both X and Spec R, i.e. over each open subset W × V satisfying W ⊆ X a relative compact open subset and V ⊆ SpecanR a relative compact open subset,

an suppSpecanR(M |W ) ∩ V ⊆ Spec R is a closed analytic subvariety. (2) if M is also coherent over DX,R, then suppSpec RM is algebraically closed locally over X, i.e. over a relative compact open subset W ⊆ X

suppSpec R(M |W ) ⊆ Spec R is a closed algebraic subvariety. 2.5. Sheafification for derived category. By construction, the analytic sheafification is a faithful and exact functor from Mod(AR) to Mod(AfR) for A in general (similarly for the algebraic sheafification). We define Mod(g AR) as the subabelian category of Mod(AfR) as the image of Mod(AfR) under the analytic sheafification. b b We write by Df(AR) the derived category of Mod(g AR) and by D (AfR) the derived ∗ category of Mod(AfR). We can define the analytic relative pullback functor eh for an affine morphism h : Spec S → Spec R in an obvious way. We denote by Leh∗ its derived functor. By slightly abuse of notations, we use eh∗ to denote the functor for ∗ both Mod(g AR) and Mod(AfR) and similarly for Leh in the derived case. an For DeX,R-module, if µ: T → Spec R is a morphism between analytic schemes, we use u∗ (resp. Lu∗) to denote the (resp. derived) pullback functor of the natural e e an morphism µe: X ×T → X ×Spec R induced by base-change (see also [Wu20, §2.2]). 2.6. Sheaves of algebraic local cohomology. Let X be a complex manifold and let R be a commutative C-algebra of finite type. For Z ⊆ X an analytic subvariety, we use RΓ[Z] to denote the right derived functor of sheaves of algebraic local cohomology along Z (cf. [Bj93, Chapter II.5]). We denote by Ze = Z × SpecanR. For M ∈ Mod(DX,R), by construction we have a natural isomorphism RΓ ( ) ' RΓ ( ). f [Z] M [Ze] Mf The following lemma is standard and we leave its proof for interested readers.

Lemma 2.11. Let h: Spec S → Spec R be an aﬃne morphism. Then for Mf ∈ Mod(DeX,R) we have a natural isomorphism

∗ ∗ Lh RΓ ( ) ' RΓ an (Lh ). e [Ze] Mf [Z×Spec S] e Mf 14 LEI WU

2.7. Duality for relative coherent D-module. We keep notation as in §2.3. We assume dimC X = n. For M ∈ Mod(DX,R), if M is coherent over DX,R, then we deﬁne −1 b DR(M ) = RhomDX,R (M , DX,R) ⊗O ωX [n] ∈ D (DX,R), where ωX is the sheaf of holomorphic n-forms on X. If Mfis relative coherent over SpecanR, then we deﬁne

−1 b SpecanR(M ) = Rhom (M , DX,R) ⊗ ω ^⊗ R[n] ∈ D (DX,R). D f DeX,R f e Oe X C e Notice that by construction, −^1 ∗ −1 ωX ⊗C R ' prX ωX , an where prX : X × Spec R → X is the projection. When R = C, DC(•) is just an the usual duality functor for coherent DX -modules. When R (resp. Spec R) is obvious from the context, we use D to denote DR (resp. DSpecanR) for short. By construction, we immediately obtain the following lemma, with its proof skipped. Lemma 2.12. Let h : Spec S → Spec R be a ﬁnite type morphism of aﬃne schemes. For M ∈ Mod(DX,R), we have:

(1) if M is coherent over DX,R, then L L DS(M ⊗R S) ' DR(M ) ⊗R S, (2) if M is relative coherent over SpecanR, then ∗ ∗ DSpecanS(Leh Mf) ' Leh DSpecanR(Mf). The following definition is the analytification of [BVWZ21a, Definition 3.3.1 and 4.3.4].

an Deﬁnition 2.13. For Mf∈ Mod(DeX,R) relative coherent over Spec R, we say Mf is j-Cohen-Macaulay for some j ∈ Z≥0 if Mf6= 0 and k E xt (Mf, DeX,R) = 0 for every k 6= j. DeX,R

If Mfis n-Cohen-Macaulay, then for simplicity we also use D(Mf) to denote n −^1 E xt (Mf, DeX,R) ⊗O ω ⊗ R. DeX,R X C 2.8. The relative codimension ﬁltration of Gabber-Kashiwara. In this subsection, we let R be a regular commutative ﬁnitely generated C-algebra integral domain. The following lemma is well-known (see for instance [FF18, Lemma 2.9] and [BVWZ21a, §3.6]).

an Lemma 2.14. Let Mfbe relative coherent over Spec R. If Mf(x,p) 6= 0 for some point (x, p) ∈ X × SpecanR, then

rel −1 j(Mf(x,p)) + dimC(Ch (Mf) ∩ πR (Wf)) = 2n + dim R, ∗ an an where πR : T X × Spec R → X × Spec R is the natural projection, Wf is a small open neighborhood of (x, p) and j(•) denotes the graded number for (DX,R)(x,p)- modules (see for instance [BVWZ21a, Deﬁnition 4.3.1]). RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 15

Following the recipe in [Gin86, Gabber-Kashiwara theorem] and [Kas03, §2.4], we now construct the Gabber-Kashiwara codimension filtration under the relative setting. For M a coherent DX,R-module and k ∈ Z≥0, we define a submodule rel Tk(M ) = {m ∈ M | codimCCh (DeX,R · m) ≥ k} ⊆ M , where DeX,R · m ⊆ Mfis the submodule generated by m. Then we have a finite decreasing filtration, called the relative codimension filtration,

· · · ⊆ Tk(M ) ⊆ Tk−1(M ) · · · ⊆ T1(M ) ⊆ T0(M ) = M . In fact, by Theorem 2.16 and the relative Bernstein inequality [Wu20, Theorem 2.2], the relative codimension filtration always stops at k = n + dim R. If M is only relative coherent over SpecanR (or more generally Mfis a coherent DeX,R-module), we set rel Tk(Mf) = {m ∈ Mf| codimCCh (DeX,R · m) ≥ k} ⊆ Mf. One can easily check that the relative codimension filtration is compatible with analytic sheafification, that is, if M is coherent over DX,R, then

T^k(M ) = Tk(Mf).

Deﬁnition 2.15. For a non-zero coherent DX,R-moudle M , we say M pure of codimension l if

(1) Tl(M ) = M and (2) Tk(M ) = 0 if k < l.

For a coherent DeX,R-module Mf, we deﬁne its purity similarly. By deﬁnition, a coherent submodule of a pure module is always pure of the same codimension. We have the following relative Gabber-Kashiwara theorem:

Theorem 2.16. Let Mf(or M ) be a coherent DeX,R-module (resp. DX,R-module). Then we have:

(1) Tk(Mf)(or Tk(M )) is coherent over DeX,R (resp. DX,R). (2) Chrel Tk(Mf) is purely k-codimensional (if not empty), i.e. every irreducible Tk+1(Mf) component is of codimension k. Tk(Mf) Tk(M ) (3) if 6= 0 (or T ( ) 6= 0), then it is pure of codimension k. Tk+1(Mf) k+1 M Proof. We follow the strategy in [Kas03, 2.4]. We use Lemma 2.14 to replace the role of [Kas03, Theorem 2.19]. Then the proof of Theorem 2.18 and Theorem 2.24 in loc. cit. give us all the required statements. If M is coherent over DX,R, then we run the arguments in loc. cit. on X replacing DX by DX,R. If Mfis coherent an over DeX,R, we run the same arguments but replacing X by X × Spec R and DX by DeX,R. By Lemma 2.14 and Theorem 2.16, we immediately obtain:

Corollary 2.17. If Mf is pure of codimension l, then for (x, p) ∈ X × SpecanR if Mf(x,p) 6= 0 then Mf(x,p) is l-pure over (DeX,R)(x,p) in the sense of [BVWZ21a, Deﬁnition 4.3.4]. 16 LEI WU

If M is relative coherent over SpecanR, we set

Tk(Mf) Sk(M ) = suppSpecanR . Tk+1(Mf) Then by Corollary 2.10 and Theorem 2.16 (2), we have: Corollary 2.18. We have: (1) If M is relative holonomic over SpecanR, then over each open subset W ×V satisfying W ⊆ X a relative compact open subset and V ⊆ SpecanR a relative compact open subset, Sk(M ) ⊆ V is a closed analytic subvariety and if Sk(M ) 6= ∅ then it is purely (k − n)-codimensional. an (2) If M is coherent over DX,R and relative holonomic over Spec R, then over each relative compact open subset W ⊆ X, Sk(M ) ⊆ Spec R is a closed algebraic subvariety and if Sk(M ) 6= ∅ then it is purely (k − n)- codimensional. The following proposition links Cohen-Macaulay modules to pure modules.

Proposition 2.19. If Mf(or M ) is relative holonomic over SpecanR and (n + j)- Cohen-Macaulay for some j, then every non-zero coherent submodule of Mfis pure of codimension j. In particular, locally on every open subset W × V satisfying W a relative compact open subset of X and V a relative compact open subset of SpecanR,

suppSpecanRMf|W ×V ⊆ V is purely j-codimensional. If moreover M is coherent over DX,R, then locally over every relative compact open subset W ⊆ X, suppSpec RM ⊆ Spec R is purely j- codimensional.

Proof. We only need to prove Tk(Mf) = 0 for every k > n + j. We assume on the an contrary, Tk(Mf)(x,p) 6= 0 for some k > n + j and some (x, p) ∈ X × Spec R. For a (DeX,R)(x,p)-module M, for simplicity we write l l E (M) = Ext (M, (DeX,R)(x,p)). (DeX,R)(x,p) By Lemma 2.14, the graded number

j(Tk(Mf)(x,p)) ≥ k > n + j. n+j n+j n+j Then E (Tk(Mf)(x,p)) = 0 and hence E (E (Tk(Mf)(x,p))) = 0. Taking double-dual of the inclusion

Tk(Mf)(x,p) ,→ Mf(x,p) by Cohen-Macaulayness and naturality we obtain a commutative diagram

Tk(Mf)(x,p) Mf(x,p)

n+j n+j n+j n+j E (E (Tk(Mf)(x,p))) E (E (Mf(x,p))).

n+j n+j We then end up with a contradiction since E (E (Tk(Mf)(x,p))) = 0.

By purity, Sn+j(Mf) = suppSpecanr(Mf). The other required statements follow from Corollary 2.18. RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 17

2.9. A relative Nakayama lemma. In this subsection, our goal is to prove a relative Nakayama lemma for C-constructible sheaves of R-modules (cf. [KS13, Definition 8.5.6]). We assume X a complex manifold, A = CX the constant sheaf on X, and R a noetherian commutative C-algebra. We write the derived category of C- b constructible (or constructible for short) sheaves of C-vector spaces by Dc(CX ), b by Dc(RX ) the derived category of C-constructible sheaves of R-modules and b by Dec(RX ) the derived category of the analytic sheafification of C-constructible sheaves of R-modules. We say a (bounded) complex is 0 in the derived category if it is quasi-isomorphic to the 0 complex in the derived category. One can check that when R is a commutative C-algebra of finite type, if • b F ∈ Dc(RX ), an then Fe• is a relative constructible complex over Cl supported on X × Spec R in the sense of [MFS13, Definition 2.19.(2)], where we fix a closed embedding l Spec R,→ C for some l ∈ Z>0. The following lemma follows directly from constructibility and Nakayama Lemma. We leave its proof for interested readers.

Lemma 2.20. Let F • be a C-constructible complex of R-modules on X. Then the following are equivalent: • b (1) F = 0 in Dc(RX ) • L b (2) F ⊗R R/m = 0 in Dc(CX ) for every maximal ideal m ∈ Spec R • b (3) Fe = 0 in Dfc(RX ). The lemma above immediately implies: • b • Corollary 2.21. If F ∈ Dc(RX ), then suppSpec RF is algebraically closed locally over X, i.e. locally on a relative compact open subset W of X, • suppSpec RF |W ⊆ Spec R is a closed algebraic subvariety.

3. Relative maximal and minimal extensions along F In this section, we construct the maximal and minimal extensions for D-modules p p under the relative setting analogous to the j! and j∗ extensions of aﬃne (or Stein) open inclusions j for perverse sheaves [BBDG18]. 3.1. Notations. We introduce notation for this whole section. Let X be a complex manifold with dimC X = n. For some ﬁxed r ∈ Z>0, we consider the morphism r ∗ r π = (id, Exp): X × C → X × (C ) where √ √ r ∗ r Exp: C → (C ) , (α1, . . . , αr) 7→ (exp(−2π −1α1), . . . , exp(−2π −1αr)) ∗ r ∗ r is the universal cover of (C ) . We write G = π1((C ) ), the fundamental group of (C∗)r. The deck transformation of Exp makes X × Cr a free G-space (we always use the discrete topology of G). We also write • r ∨ R = Sym (C ) ' C[s] 18 LEI WU

• the algebraic coordinate ring of Cr, where Sym (Cr)∨ is the symmetric algebra of the dual vector space (Cr)∨ and the isomorphism is induced from using the standard basis of Cr. Then G acts on R induced by the G-action on (Cr)∨. In algebraic coordinates, the G-action on R is given by ( sj + 1, if i = j (3.1) ti · sj = sj, if i 6= j. where ti represents the (counterclockwise) loops around the puncture of each factor ∗ ∗ r r C of (C ) and such ti induce an isomorphism G ' Z . Let F = (f1, . . . , fr) be an r-tuple of holomorphic functions on X with Di the P divisor of fi and let M be a holonomic DX -module. We write by D = i Di and by

M(∗D) = M ⊗O OX (∗D) S the algebraic localization of M along D, where OX (∗D) = OX (kD) is the k∈Z sheaf of meromorphic functions with poles along D. We write by s M = M(∗D) ⊗C C[s] · F . We have natural actions r X siθ(fi) (3.2) θ · (m · F s) = (θ · m) · F s + m · · F s f i=1 i for every section m of M(∗D) and for holomorphic vector ﬁelds θ on Y , which makes M a DX,R-module. Since M(∗D) is holonomic, by [Mal04, Theorem 3.1] one can assume

M(∗D) = DX ·M0 for M0 a coherent OX -submodule of M(∗D). We then ﬁx such an M0 throughout the remaining of this paper and deﬁne the DX,R-submodules s−k Nk = DX,R(M0 · F ) ⊆ M s−k generated by M0 · F with k = (k, k, . . . , k) for k ∈ Z. By construction, s DX,R(M0 · F ) is coherent over DX,R but M might not be so, and

Nk ⊆ Nk for k1 ≤ k2 and lim Nk = M . 1 2 k→∞ 3.2. G-action and translation on M . The G-action on R induces a G-action on M . In algebraic coordinates, the G-action on M is given by

s s+ei ti · P (s)(m · F ) = P (s + ei)(m · F ) r for P (s) ∈ DX,R, where ei ∈ Z is the i-th unit vector. The G-action on M induces an operation g · Nk ⊂ M , i.e.

s−k+ei ti · Nk = DX [s] · F . Since Cr is homogeneous, for every α ∈ Cr the translation of α induces an isomorphism r r τα : X × C → X × C (x, s) 7→ (x, s + α). Substituting s by s + α gives an DX -linear isomorphism (but not C[s]-linear) s+α−k (3.3) Nk ' DX [s + α](M0 · F ) RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 19

−1 which further induces an prX DX -linear isomorphism −1 (3.4) τα Nfk ' Nfk, r r where prX : X × C → X is the projection. The standard basis on C induces an isomorphism r X Z ' G, a = (a1, . . . , ar) 7→ ga = aiti. i Then for a ∈ Zr, by construction we have −1 (3.5) τa Nfk = g^a · Nk for every k.

3.3. Relative maximal and minimal extension. We now are ready to construct the relative maximal and minimal extensions. We ﬁrst recall a result of Sabbah [Sab87b] about the existence of generalized b-functions.

Theorem 3.1 (Sabbah). Let M be a holonomic DX -module (with a fixed M0). Then, locally on a relative compact open subset W ⊆ X, there exists a polynomial b(s) ∈ C[s] such that the following two conditions are satisfied: Q Q nL,α (1) b(s) = r (L · s + α) over some finite sets of L and α, L∈Z≥0 α∈C (2) b(s) · N0 = 0. N−1

Proposition 3.2. Let M be a holonomic DX -module. For every relative compact open subset W ⊆ X and for every relative compact open subset V ⊆ Cr, there exists an integer l = lW,V > 0 such that

(1) Mf|W ×V = Nfk|W ×V for all k ≥ l. (2) Nf−k|W ×V = Nf−l|W ×V for all k ≥ l Proof. By the construction of the analytic sheaﬁﬁcation, we have

b(s) · N^0/N−1 = 0 where b(s) is the generalized b-function in Theorem 3.1. By translation, for every k ∈ Z with k = (k, k, . . . , k), we have

(3.6) b(gk · s) · N−k^/N−k+1 = 0,

Choose l > 0 large enough (since V is relative compact) such that b(gk · s)|V are invertible for all |k| > l. Then by (3.6), we have

Nfk = Nfk−1 for all |k| > l. Part (2) then follows. Since lim Nk = M , part (1) also follows. k→∞ The following theorem is a natural generalization of [Mai16a, R´esultat] and [BVWZ21b, Theorem 4.3.4].

Theorem 3.3. Let M be a holonomic DX -module. Then M and Nk are relative holonomic over the complex manifold Cr for all k ∈ Z. Moreover, rel rel r CC (Mf) = CC (Nfk) = CC(M(∗D)) × C . 20 LEI WU

Proof. By Proposition 3.2 (1), M is relative coherent over Cn. Now we construct the relative characteristic cycles for Nfk over a relative compact open subset W ⊆ X. For simplicity we assume X = W . Then we define s−k Fp(Nk) = (FpDX ⊗C C[s]) · (M0 · F ) which gives a relative good filtration of Nk over C[s] (cf. [BVWZ21a, §3.1]). Then r F^p(Nk) gives a relative good filtration for Nfk over C (cf. [Wu20, §2.1]). Using the −1 −1 isomorphism (3.4), τα F^p(Nk) gives a relative good filtration of τα Nfk such that F −1 −1 F gr• (τα Nfk) ' τα (gr• Nk), rel which implies CC (Nfk) is invariant under translation by α. Since α is arbitrary, we thus have rel rel r CC (Nfk) = CC (Nfk)|X×{a} × C . r We then pick a = (a, a, . . . , a) ∈ Z with a 0. On a small neighborhood Va of a, by Proposition 3.2(1), we know

Nfk|W ×Va = Mf|W ×Va . Since M is free over C[s], one can repeatedly apply [Wu20, Proposition 2.7] (by choosing general hyperplanes H1,H2,...,Hr passing a) and conclude rel rel CC (Nfk)|X×{a} = CC (Mf)|X×{a} = CC(Mf|X×{a}). But since a ∈ Cn is a Z-point,

Mf|X×{a} ' M ⊗C[s] Ca = M(∗D) n where Ca is the residue field of a ∈ C . We therefore obtain rel r CC (Nfk) = CC(M(∗D)) × C for every k. Thanks to Proposition 3.2 again, we have rel CC (Mf)|X×Va = CC(M(∗D)) × Va, since fixing k and making a 0 is equivalent to fixing Va and making k 0. Thus, globally rel r CC (Mf) = CC(M(∗D)) × C . Remark 3.4. From the proof Proposition 3.2, one can see that it is necessary to use the analytic sheafification of M . To be more precise, even if M is an algebraic alg holonomic DX -module on a smooth complex algebraic variety X, Mf (cf. Remark 2.3) is not relative coherent over Spec R, since every Zariski open subset of Spec R intersects the divisors (b(gk · s) = 0) for all but a finite number of k ∈ Z, where b(s) is the generalized b-function as in Theorem 3.1. We now write (r) Mf= M(∗DF ), calling it the relative maximal extension of M along F . Next we construct the minimal one. (r) Proposition 3.5. If M is holonomic, then Mf= M(∗DF ) is n-Cohen-Macaulay. RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 21

r Proof. We ﬁx a point (x, p) ∈ X × C . First, we know (DeX,R)(x,p) is Auslander regular (by for instance [BVWZ21a, Theorem 4.3.2]). Then by Auslander regularity, the graded number

k j(E xt (Nf0, (DeX,R)(x,p))) ≥ k (DeX,R)(x,p) for each k. Also, for all k as right coherent DeX,R-modules

k E xt (Nf0, DeX,R) DeX,R are relative holonomic, as so is Nfby Theorem 3.3. By Lemma 2.14, we then know k E xt (Nf0, DeX,R) = 0 for k < n DeX,R and rel k dim (Ch (E xt (Nf0, DeX,R))) ≤ n + r − k, for n < k ≤ n + r. C DeX,R

By Lemma 2.5, we conclude that Nf0 is n-Cohen-Macaulay away from W × V for V (Cr some closed algebraic subvariety, where W is a small neighborhood of r x ∈ X. We now translate N0 by τ−a for some a ∈ Z≥0 with each ai 0 so that −1 (x, p) 6∈ τa(W × V ). Thus, τ−a Nf0 is Cohen-Macaulay around (x, p). But from the −1 proof of Proposition 3.2, such τ−a Nf0 is just Mfaround (x, p). Remark 3.6. (1) After applying Theorem 3.3, Proposition 3.5 also follows from [MFS19, Proposition 2]

(2) In general Nfk are not n-Cohen-Macaulay. But since Nfk ⊆ Mfis a submodule of a n-Cohen-Macaulay module, Nfk is always pure of codimension n for every k (by Proposition 2.19).

−s We write by MfDM the analytic sheafification of (DM)(∗D) ⊗C C[s] · F . We apply Proposition 3.5 for DM, and thus see that MfDM is n-Cohen-Macaulay and relative holonomic over Cr. We then define (r) M(!DF ) = D(MfDM) which is a relative holonomic D-module over Cr, thanks to Cohen-Macaulayness. In particular, if r = 0, then we have

(0) M(!DF ) = M(!D) = D((DM)(∗D)).

(r) −s s Notice that in the deﬁnition of M(!DF ) we use the symbol F instead of F because D naturally maps F −s back to F s (see [BVWZ21b, Lemma 5.3.1]). (r) (r) r r Since M(!DF )|(X\D)×C = M(∗DF )|(X\D)×C , the morphism OX ,→ OX (∗D) induces a natural morphism

(r) (r) M(!DF ) −→ M(∗DF ).

Lemma 3.7. For M a holonomic DX -module, if r ≥ 1, then the natural morphism (r) (r) M(!DF ) ,→ M(∗DF ) is injective. 22 LEI WU

Proof. We ﬁx a small neighborhood W × V of a point (x, p) ∈ X × Cr. By con- (r) r struction, M(∗DF ) has no non-zero coherent submodule supported on D × C . (r) By duality and Lemma 2.14, M(!DF ) has no non-zero coherent quotient module r (r) (r) supported on D × C . Then the image of M(!DF ) → M(∗DF ), denoted by (r) (r) (r) M(!∗DF ), is the minimal extension. Since the cokernel of M(!DF ) → M(∗DF ) is supported on D × Cr, by analytic nullstellensatz,

s+a (r) (M^0 · F )|W ×V ⊆ M(! ∗ DF )|W ×V

r for some a ∈ Z≥0 with each ai 0, and thus

−1 (r) (τa Nf0)|W ×V ⊆ M(! ∗ DF )|W ×V By minimality,

(r) ∗ (3.7) M(! ∗ DF )|W ×V = (τa Nf0)|W ×V

(r) M(!∗DF )|W ×V (r) (since otherwise −1 is a non-zero coherent quotient module of M(!DF ) (τa Nf0)|W ×V r −1 supported on D × C ). Similar to the proof of Proposition 3.5, (τa Nf0)|W ×V is (r) n-Cohen-Macaulay and hence so is M(! ∗ DF )|W ×V . We use K to denote the kernel on W × V . If K 6= 0, then we have a short exact sequence (r) (r) 0 → K → M(!DF )|W ×V → M(! ∗ DF )|W ×V → 0. Since K is relative holonomic, by Proposition 2.19 and Proposition 2.5,

rel r Ch (K) = ΛK × C

∗ for some Lagrangian subvariety ΛK ⊆ T X|W . Taking the dual, we have

(r) (r) 0 → (DM)(! ∗ DF )|W ×V → (DM)(∗DF )|W ×V → D(K) → 0.

By (3.7) (replacing M by DM) and Theorem 3.3,

(r) (r) CC((DM)(! ∗ DF )|W ×V ) = CC((DM)(∗DF )|W ×V ).

rel r Since Ch (D(K)) = ΛK ×C , we get a contradiction by counting multiplicities.

(r) An immediate consequence of the inclusion in the lemma above is that M(!DF ) is the minimal extension (analogous to the Deligne-Goresky-MacPherson extensions (r) for perverse sheaves [HTT08, §8.2.1]). Therefore, when M is holonomic, M(!DF ) satisﬁes the property that it has no non-trivial submodule (or quotient module) r supported on D × C in the category of relative coherent DeX,R-modules. By this property, in the situation of Proposition 3.2, we have

(r) (3.8) M(!DF )|W ×V = Nf−k|W ×V for all k ≥ lW ×V . RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 23

3.4. Gluing data and G-equivariance. Now we discuss how to “glue” the local data in Proposition 3.2 to construct the maximal and minimal extensions alternatively. We keep assuming M a holonomic DX -module and W a relative compact open subset of X. We pick a locally ﬁnite covering

r [ C = Vβ β such that each Vβ is relatively compact. For instance, we can pick a ﬁnite covering for the unit polydisc in Cr and then translate the covering by Zr to get such a r (r) covering of C . We can tautologically “glue” M(!DF ) with the help of Proposition 3.2 (2) as follows:

(i) For every Vβ0 , we have (r) M(!D )|W ×Vβ = Nf−lW,V |W ×Vβ F 0 β0 0 (ii) We set k = max{l | V ∩ V 6= ∅}. 0 W,Vβi β0 βi

(iii) If Vβ0 ∩ Vβi 6= ∅, we can “glue” Nf−lW,V |W ×Vβ and Nf−lW,V |W ×Vβ since β0 0 βi i

Nf−lW,V |W ×(Vβ ∩Vβ ) = Nf−k0 |W ×(Vβ ∩Vβ ) = Nf−lW,V |W ×(Vβ ∩Vβ ). β0 0 i 0 i βi 0 i Notice that the cocycle condition of “gluing” is guaranteed by Condition (ii). Using (r) Proposition 3.2 (1), we can “glue” M(∗DF ) similarly. (r) (r) Theorem 3.8. Let M be a holonomic DX -module. Then both M(∗DF ) and M(!DF ) are G-equivariant sheaves of abelian groups on X × Cr and the inclusion (r) (r) M(!DF ) ,→ M(∗DF ) is a G-equivariant morphism. Proof. By (3.4) and (3.5), we have for every g ∈ G

g^· Nk ' Nfk In particular,

(3.9) g^· | ' | . Nk W ×Vβ0 Nfk W ×Vβ0 The isomorphism above is obviously compatible with the “gluing” data for both (r) (r) r M(∗DF ) and M(!DF ). Since G acts on X × C , for every g ∈ G we have an isomorphism r r g : X × C → X × C . The “gluing” data and (3.9) together give isomorphisms

−1 (r) (r) −1 (r) (r) (3.10) g (M(∗DF )) 'M(∗DF ) and g (M(!DF )) 'M(!DF ). (r) Since we use the discrete topology of G, the isomorphisms above make M(∗DF ) (r) and M(!DF ) G-equivariant sheaves of abelian groups (one can easily check the cocycle condition). Here we use the deﬁnition of G-equivariant sheaves in [BL94, Part I.0.2]. 24 LEI WU

(r) (r) The inclusion M(!DF ) ,→ M(∗DF ) is G-equivariant because of the following commutative diagram ' Nl g · Nl

' Nk g · Nk) for all l < k.

For M a holonomic DX -module, we now deﬁne M(∗D(r)) Ψ (M) = F . F (r) M(!DF )

By Lemma 3.7, another interpretation of ΨF (M) is

q.i. (r) r ΨF (M)[−1] ' RΓ[D×C ](M(!DF )).

Theorem 3.9. Let M be a holonomic DX -module. Then ΨF (M) is G-equivariant r and relative holonomic over C . If moreover, ΨF (M) 6= 0, then it is (n+1)-Cohen- Macaulay. (r) Proof. As a quotient of M(∗DF ), ΨF (M) is relative holonomic. Since the category of G-equivariant sheaves is abelian and the inclusion (r) (r) M(!DF ) ,→ M(∗DF ) is G-equivariant by Theorem 3.8, ΨF (M) is G-equivariant. To prove Cohen-Macaulayness, we consider the short exact sequence (r) (r) 0 → M(!DF ) → M(∗DF ) → ΨF (M) → 0. By Theorem 3.3 and (3.8), we know rel (r) rel (r) CC (M(!DF )) = CC (M(∗DF )).

If ΨF (M) 6= 0, by counting multiplicity we have rel dimC Ch (ΨF (M)) ≤ n + r − 1. By Lemma 2.14, k E xt (ΨF (M)) = 0 for k ≤ n. DeX,R (r) (r) Taking the dual of the short exact sequence, since both M(!DF ) and M(∗DF ) are n-Cohen-Macaulay, ΨF (M) is (n + 1)-Cohen-Macaulay. (r) (r) Remark 3.10. (1) By Theorem 3.8 and Theorem 3.9, M(∗DF ), M(!DF ) and ΨF (M) are G-equivariant sheaves of abelian groups. Since the isomorphisms in (r) (r) r (3.10) are not OX×C -linear (as (3.3) is not C[s]-linear), M(∗DF ), M(!DF ) and ΨF (M) are not G-equivariant DeX,R-modules. However, since π is covering, the sheaf morphism −1 ∗ r ∗ r π (DX×(C ) /(C ) ) → DeX,R is an isomorphism locally and hence also a global isomorphism. In particular, DeX,R is G-equivariant (see [BL94, Part I.0.3]) and its G-equivariance is induced from the deck transformation of the universal covering of (C∗)r. Equivalently, the G-equivariance of DeX,R is induced from the G-action on C[s] given by (3.1). RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 25

(r) (r) Moreover, the G-equivariance of M(∗DF ), M(!DF ) and ΨF (M) is also induced (r) from the G-action on C[s]. Then, for g ∈ G, Q ∈ π∗(DeX,R) and m ∈ π∗(M(∗DF )) (r) (similarly for M(!DF ) and ΨF (M)), we have g · (Q · m) = (g · Q) · (g · m).

G (r) G (r) G Therefore, the G-invariant parts π∗ (M(∗DF )) π∗ (M(!DF )) and π∗ (ΨF (M)) are G ∗ r ∗ r π∗ (DeX,R) ' DX×(C ) /(C ) -modules. Since π is covering, they are automatically ∗ r ∗ r relative holonomic DX×(C ) /(C ) -modules. (2) Since (r) (r) (r) (r) (3.11) D(M(!DF )) ' (DM)(∗DF ) and D(M(∗DF )) ' (DM)(!DF ), (r) (r) −1 G D(M(!DF )) (resp. D(M(∗DF ))) is also G-equivariant. Since π ◦ π∗ ' id (see [BL94, Part I.0.3]), we have G (r) G (r) (3.12) π∗ D M(!DF ) ' Dπ∗ M(!DF ) and G (r) G (r) (3.13) π∗ D M(∗DF ) ' Dπ∗ M(∗DF ) . Theorem 3.9 and (3.11) together imply:

Corollary 3.11. For M a holonomic DX -module, we have a quasi-isomorphism q.i. D(ΨF (M)) ' ΨF (DM)[−1]. Now, we discuss relative intermediate extensions and use them to study local (r) cohomology sheaves for M(!DF ). For every subset I ⊆ {1, 2, . . . , r}, we denote \ I [ DI = Di and D = Di i∈I i∈I and r I I r DeI = DI × C and De = D × C . Then we deﬁne (r) I (r) I M(!DF )(∗De ) := lim M(!DF ) ⊗O OX× r (kDe ). k→+∞ C If J ⊆ I, then similar to Theorem 3.8 we have a G-equivariant inclusion (r) J (r) I (3.14) M(!DF )(∗De ) ,→ M(!DF )(∗De ). In fact, the inclusion in Theorem 3.8 factor through (3.14) when J is empty. Similar to Proposition 3.5 (see also [BVWZ21b, §5.3] for the algebraic case), (r) I M(!DF )(∗De ) is n-Cohen-Macaulay. We thus deﬁne (r) I (r) I M(∗DF )(!De ) := D (DM)(!DF )(∗De ) . If J ⊆ I, then similar to (3.14) we have a G-equivariant inclusion (r) I (r) J (3.15) M(∗DF )(!De ) ,→ M(∗DF )(!De ). Furthermore, similar to (3.12) in Remark 3.10 (2), we have G (r) J G (r) J (3.16) π∗ D M(!DF )(∗De ) ' Dπ∗ M(!DF )(∗De ) . 26 LEI WU

(r) I (r) I By (3.14) and (3.15), for every subset I, M(!DF )(∗De ) and M(∗DF )(!De ) are (r) (r) relative intermediate extensions lying in between M(!DF ) and M(∗DF ).

Theorem 3.12. Let M be a holonomic DX -module. Then for every I ⊆ {1, 2, . . . , r}, the complexes RΓ M(!D(r)) and RΓ ( M)(!D(r)) are G-equivariant. [DeI ] F D [DeI ] D F Moreover, G (r) G (r) π RΓ M(!D ) ' RΓ ∗ r π M(!D ) ∗ [DeI ] F [DI ×(C ) ] ∗ F and G (r) G (r) π RΓ ( M)(!D ) ' RΓ ∗ r π ( M)(!D ) ∗ D [DeI ] D F D [DI ×(C ) ] ∗ D F

∗ r ∗ r are complexes of relative holonomic DX×(C ) /(C ) -modules. Proof. By the deﬁnition of RΓ , RΓ M(!D(r)) is represented by the Koszul- [DeI ] [DeI ] F type complex

(r) M (r) J 0 → M(!DF ) → · · · → M(!DF )(∗De ) → J⊆I,|J|=i M (r) J (r) I → M(!DF )(∗De ) → · · · M(!DF )(∗De ) → 0. J⊆I,|J|=i+1 with cohomogy degrees 0, 1, 2,... |I|, and taking duality of the above complex for M, RΓ ( M)(!D(r)) is the complex D D [DeI ] D F

(r) I M (r) J 0 → M(∗DF )(!De ) → · · · → M(∗DF )(!De ) → J⊆I,|J|=i M (r) J (r) → M(∗DF )(!De ) → · · · M(∗DF ) → 0. J⊆I,|J|=i−1 with cohomogy degrees −|I|, −|I| + 1,..., −1, 0. By using (3.14) and (3.15), the two complexes as above are G-equivariant. The second required claim follows from Remark 3.10 and (3.16). 3.5. G-equivariance of relative de Rham complex. Let E be G-equivariant sheaf r r and a DX×C /C -module (see Remark 3.10). For a diﬀerential operator P ∈ DX , since the G-action on E is induced from the G-action on Cr, we have g∗(P · E ) ' P · (g∗E ), which means that the P -action on E is G-equivariant.

Now we ﬁx a local coordinates (x1, . . . , xn) of X and denote by ∂xi the vector ﬁeld of xi. Then locally the relative de Rham complex of E satisfying

r r (3.17) DRX×C /C (E ) ' Kos(E ; ∂x1 , . . . , ∂xr ), where Kos(E ; ∂x1 , . . . , ∂xr ) is the Koszul complex of the ∂xi -action on E . We thus obtain:

r r Lemma 3.13. Let E be a sheaf (or complexes) of DX×C /C )-modules and G-equivariant. r r Then the de Rham complex DRX×C /C (E ) is G-equivariant and hence we have a natural isomorphism G G r r ∗ r ∗ r π∗ DRX×C /C (E ) ' DRX×(C ) /(C ) (π∗ E ). RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 27

r 3.6. Bernstein-Sato ideal along F . For every a = (a1, . . . , ar) ∈ Z≥0, the Bernstein- Sato ideal of N0 and a along F is the ideal N0 BF (N0, a) = B( ) ⊆ C[s]. ga · N0 r For simplicity, we write BF (N0) = BF (N0, 1r) with 1r = (1, 1,..., 1) ∈ Z . When M0 = OX , we further denote BF (N0) by BF for short. Notice that BF is the major Bernstein-Sato ideal studied in the literature; see for instance [Bud15, BVWZ21a, Mai16b]. Bernstein-Sato ideals along F are natural generalization of the Bernstein-Sato polynomial for a single holomorphic function (see [Kas77] and [Bud15, §4.1]). By Theorem 3.1 (replacing N−1 by ga · N0), we have that, if M is holonomic, then locally on a relative compact open subset W ⊆ X r BF (N0, a) 6= 0 for all a ∈ Z≥0 and a 6= 0. The following theorem is essentially due to Maisonobe [Mai16a]; here we present 4 it in a more generalized form. See also [BVWZ21a] for M0 = OX under the algebraic setting.

Theorem 3.14. Let M be a regular holonomic DX -module. Locally on a relative compact open subset W ⊆ X, if N0/N−1 is not zero on W (⇔ W ∩ D 6= ∅), then r Z(BF (N0)) ⊆ C is of codimension-one (but not purely in general ).

Proof. If W ∩ D = ∅, then obviously N0 = N−1. We now assume W ∩ D 6= 0 and then prove N0/N−1 6= 0 on W . To this purpose, we need to treat Nk as a logarithmic D-module. Using the graph embedding (see [Wu20, §5]), the inclusion

ιF,∗(N0) ,→ ιF,∗(M ) ' ιF,+(M(∗D)) makes ιF,∗(N0) a lattice of the regular holonomic DY -module ιF,+(M(∗D)), where r ιF : X,→ Y = X × Cu, x 7→ (x, f1(x), . . . , fr(x)) is the graph embedding with u = (u1, . . . , ur) the coordinates and ιF,+ denotes the D-module pushforward functor of ιF . Since ιF is a closed embedding, we have

N0 ιF,∗(N0) 6= 0 ⇔ 6= 0. N−1 ιF,∗(N−1)

By the construction of ιF,+, we have r Y ui · ιF,∗(N0) = ιF,∗(N−1). i=1 Therefore, we have a short exact sequence Qr i=1 ui ιF,∗(N0) 0 → ιF,∗(N0) −−−−−→ ιF,∗(N0) −→ → 0. ιF,∗(N−1) For simplicity, we write r ιF,∗(N0) Y G = ιF,∗(N0), H = and v = ui. ι ( ) F,∗ N−1 i=1

4 s In loc. cit. the author only considered the cyclic modules DX [s]m · F for sections m of M (instead of N0). 28 LEI WU

Qr Since G is a lattice along the normal crossing divisor DY = ( i=1 ui = 0), by [Wu20, Theorem 1.6] the logarithmic characteristic variety satisﬁes log Ch (G )) = eιF (Ch(M|X\D)), where eιF is the induced embedding ∗ ∗ ∗ ∗ r eιF : T (X \ D) ,→ T (Y \ DY ) ' T (X \ D) × T Cu, (x, ξ) 7→ (x, ξ, F (x), dF (x)) ∗ and the closure is taken inside the logarithmic cotangent bundle T (Y,DY ). In particular, Chlog(G ) has no component over (v = 0).

We now take a good filtration F•G over F•DY,DY (cf. [Wu20, §3]). We thus have a filtered morphism v F•G −→ F•G with the associated graded morphism F v F ϕ: gr• G −→ gr• G . Since the filtration is bounded from below, the filtered morphism gives a convergent spectral sequence. By convergence (see for instance [Lau83, Lemma 3.5.13 (iii)]), we have F [gr• H ] = [Cokerϕ] − [ker ϕ] F in the Grothendieck group K0 generated by gr• DY,DY -modules with support of dimension ≤ n + r − 1, where F•H is the induced filtration on the quotient. But one can check in K0 F [Cokerϕ] − [ker ϕ] = [(gr• G /T )|(v=0)] F F where T ⊆ gr• G is the v-torsion subsheaf. Since gr• G /T has no v-torsion, we have F F supp F ((gr G /T )|(v=0)) = supp F (gr G /T )|(v=0). gr• DY,DY • gr• DY,DY • Since log F Ch (G ) = supp F gr G gr• DY,DY • and Chlog(G ) does not have component over (v = 0), we have F F supp F gr G = supp F (gr G /T ). gr• DY,DY • gr• DY,DY • By [Gin86, Proposition 1.5.1], we conclude log log Chn+r−1(H ) = Ch (G )|v=0 log where Chn+r−1(H ) is the purely (n + r − 1)-dimensional part of the characteristic variety. But by construction, log log Ch (H ) ⊆ Ch (G )|v=0 and hence log log Ch (H ) = Ch (G )|v=0. In fact, the above argument proves more generally log log CC (H ) = CC (G )|v=0.

In particular, H 6= 0 and hence so is N0/N−1. Then Nk/N−k 6= 0 for all k > 0. r Since for k 0 Nfk/Nf−k is just ΨF (M) analytically locally over C , we now conclude that ΨF (M) is (n + 1)-Cohen-Macaulay by Theorem 3.9. By Lemma 2.14, this also implies rel dimC(Ch (Nfk/Nf−k)) = n + r − 1. RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 29

Thus, by [Wu20, Lemma 2.3] rel dimC(Ch (Nfl/Nfl−1)) = n + r − 1 for some −k < l ≤ k. But by translation,

Nl/Nl−1 ' N0/N−1 and hence rel dimC(Ch (Nf0/Nf−1)) = n + r − 1. The proof is then accomplished by Lemma 2.5 and Proposition 2.8. The theorem above and Theorem 3.1 together implies:

Corollary 3.15. Let M be a regular holonomic DX -module. Locally on a relative compact open subset W ⊆ X intersecting D, Zr−1(BF (N0)) is a ﬁnite union of r translated hyperplanes of the form (L · s + α = 0) with L ∈ Z≥0 and α ∈ C, where Zr−1(BF (N0)) denotes the pure (r − 1)-dimensional part of Z(BF (N0)). The corollary above enables us to make the following deﬁnition:

Deﬁnition 3.16. For M a regular holonomic DX -module, locally on a relative compact open subset W ⊆ X, we deﬁne r S(F, M) = {primitive L ∈ Z | (L · s + α = 0) ⊆ Zr−1(BF (N0)), for some α ∈ C} and that for L ∈ S(F, M)

κ(L) = {α ∈ C | (L · s + α = 0) ⊆ Zr−1(BF (N0))}. We set κe(L) = κ(L)/ ∼L where ∼L denotes the equivalence r α1 ∼L α2 ⇐⇒ α2 = L · a + α1 for some a ∈ Z .

Since N0 is depending on M0, BF (N0) is depending on M0. However, one can easily check that

S(F, M) is independent of the choices of M0. See [Mai16a, R´esultat 6] for a logarithmic interpretation of S(F, M) and also [BVWZ21b, Proposition 4.4.4].

Proof of Theorem 1.4. For p ∈ Zr−1(BF (N0)) and for every g ∈ G, by Proposition 2.8 we know g · N0 ( )mg·p 6= 0, g · N−1 where mg·p is the maximal ideal of p. We take k 0 such that g · N0/g · N−1 is a subquotient of Nk/N−k. Then

Nk ( )mg·p 6= 0. N−k By Lemma 2.6, we then have for some x ∈ X

Nfk ( )(x,g·p) 6= 0. Nf−k 30 LEI WU

But by Proposition 3.2 (1) and (3.8), Nfk/Nf−k is ΨF (M) locally around (x, g · p). We thus have [ [ [ (g · (L · s + α) = 0) ⊆ supp r Ψ (M). C F g∈G L∈S(F,M) α∈κ(L) r Conversely, we assume ΦF (M)(x,p) 6= 0 for some x ∈ X and for p ∈ C away from [ [ [ (g · (L · s + α) = 0). g∈G L∈S(F,M) α∈κ(L) Then for some k 0

Nfk ΦF (M)|W ×V = |W ×V 6= 0 Nf−k with W a small neighborhood of x and V a small neighborhood of p. Since ΦF (M) is (n + 1)-Cohen-Macaulay, by Lemma 2.14 and Proposition 2.19, we know

supp r Φ (M)| C F W ×V is purely of codimension one. But by Theorem 3.1 and Proposition 2.8, we know

Nk supp r C N−k is contained in a union of translated linear hyperplanes in Cr of forms (L·s+α = 0) r with L ∈ Z≥0. Therefore, we can assume (L · s + α = 0) ∩ V 3 p is a component of supp r Φ (M)| . Then (L·s+α = 0) must be a component of supp r / C F W ×V C Nl Nl−1 for some −k < l ≤ k. But this means [ [ [ p ∈ (g · (L · s + α) = 0), g∈G L∈S(F,M) α∈κ(L) which is a contradiction. (r) 3.7. Evaluating ΨF (M). In this subsection, we discuss “evaluating” M(∗DF ), (r) r M(!DF ) and ΨF (M) at points in C and applying them to prove Theorem 1.6. We ﬁrst introduce some notations. r For α = (α1, . . . , αr) ∈ C and for i ∈ {1, . . . , r}, we write

αˆ i = (α1, . . . , αi−1, 0, αi+1, . . . , αr). We also write by r iα : {α} ,→ C . the closed embedding and set r δαˆ i : C ,→ C , x 7→ (α1, . . . , αi−1, x, αi+1, . . . , αr). r α For α = (α1, . . . , αr) ∈ C , the multivalued holomorphic function F determines a rank one local system Lλ on X \ D, that is, the local system has local monodromy along each Di given by multiplication by λi with √ λi = exp(−2π −1αi).

By Riemann-Hilbert correspondence, Rj∗Lα determines a regular holonomic DX - module, denoted by Vα, where j : X \ D,→ X the open embedding. We write

Mα = M ⊗O Vα. RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 31

Proposition 3.17. Let M be a holonomic DX -module. With notations as above, for each α ∈ Cr and for each i, we have

∗ (r) q.i. ∗ (r) q.i. Leiα(M(∗DF )) 'Mα(∗D) and Leiα(M(!DF )) 'Mα(!D) and for each 0 < i ≤ r q.i. q.i. Lδ∗ (M(∗D(r))) 'M (∗D(1)) and Lδ∗ (M(!D(r))) 'M (!D(1)). eαˆ i F αˆ i fi eαˆ i F αˆ i fi Proof. By construction, M is ﬂat over R ' C[s]. Then

∗ (r) q.i. ∗ (r) Leiα(M(∗DF )) ' eiα(M(∗DF )) = M ⊗R Cα = Mα(∗D), r where Cα is the residue ﬁeld of α ∈ C . By using Lemma 2.12 and Cohen- Macaulayness, ∗ (r) q.i. ∗ (r) Leiα(M(!DF )) ' eiα(M(!DF )) = Mα(!D) follows similarly.

The other required statements for iαˆ i can be checked in a similar way. We leave details for interested readers. The proposition above has an immediate corollary:

r Corollary 3.18. Let M be a holonomic DX -module. Then for each α ∈ C and for each i, we have q.i. Lδ∗ Ψ (M) ' Ψ (M ). eαˆ i F fi αˆ i (r) Proof of Theorem 1.6. The relative characteristic cycle formula for M(∗DF ) has been obtained in Theorem 3.3 even if M is only holonomic. By (3.8), we also get (r) the relative characteristic cycle of M(!DF ). We now prove the relative characteristic cycle formula for ΨF (M). For L ∈ S(F, M) and α ∈ κ(L), we ﬁx a relative compact open subset V ⊆ Cr intersecting (L · s + α = 0). By Theorem 1.4 and Proposition 2.5, we can assume that

ΛL,α × (L · s + α = 0)|V rel is the part of CC (ΨF (M))|W ×V over (L · s + α = 0), where ΛL,α is a conic Lagrangian cycle supported on Ch(M(∗D)) over D ∩ W . Using the “gluing” date (r) (r) of M(∗DF ) and M(!DF ) in §3.4, we can move ΛL,α × (L · s + α = 0)|V along (L·s+α = 0) as we translate V along (L·s+α = 0) such that ΛL,α ×(L·s+α = 0)|V rel is always the part of CC (ΨF (M))|W ×V over (L · s + α = 0). Therefore,

ΛL,α × (L · s + α = 0) rel r is the part of CC (ΨF (M))|W ×C over (L · s + α = 0). For every g ∈ G, using the isomorphism r G ' Z r we can assume g = ga for some a ∈ Z . Since for every k > 0 ∗ τa (N^k/N−k) ' N^k/N−k, rel r the part of CC (ΨF (M))|W ×C over g · (L · s + α = 0) is

ΛL,α × (L · s + α = 0). 32 LEI WU

Therefore,

rel X X X CC (ΨF (M)) = ΛL,α × (g · (L · s + α) = 0). g∈G L∈S(F,M) α∈κ˜(L)

To avoid counting components repeatedly, the summation above is taken over κe(L) (instead of κ(L)).

We now give a formula for ΛL,α. Proposition 3.19. In the situation of Theorem 1.6, for each L ∈ S(F, M) and for each α ∈ κe(L), then after picking a very general point

α = (α1, . . . , αr) ∈ (L · s + α = 0) for each i ∈ {1, . . . , r} we have Λ = CC(Ψ (M )| ) L,α fi αˆ i W ×Vαi where Vαi is a suﬃciently small neighborhood of αi ∈ C. Proof. Let us ﬁrst explain the notation CC(Ψ (M )| ). fi αˆ i W ×Vαi

In the case r = 1, Ψfi (Mαˆ i ) is a relative holonomic D-module over the complex line C. In this case, S(fi, Mαˆ i ) = {1}. By Theorem 1.4, we then know Ψfi (Mαˆ i ) is supported over an infinite discrete subset of C, where the discrete subset containing α (by Corollary 3.18). With V small enough, Ψ (M )| is just supported i αi fi αˆ i W ×Vαi over α . More explicitly, by construction Ψ (M )| is just i fi αˆ i W ×Vαi ( αˆ i ) Nk mαi ( αˆ i ) N−k mαi l for k 0 (since both of them are killed by (s + αi) for some l > 0, sheafification does nothing), where αˆ i is the -module for M as in §3.1, and m Nk DX,C[si] Nk αˆ i αi is the maximal ideal of αi ∈ C. But the latter module is particularly a DX -module. Thus, by Lemma 2.5 (3.18) CCrel(Ψ (M )| ) = CC(Ψ (M )| ) × {α }. fi αˆ i W ×Vαi fi αˆ i W ×Vαi i As a consequence, Ψ (M )| is a holonomic -module (in fact, it is also fi αˆ i W ×Vαi DX regular if so is M). Moreover, if we use the terminology in [Kas83, Mal83, Wu21], Ψ (M )| is the α -nearby cycle of M . fi αˆ i W ×Vαi i αˆ i We now prove the required identity. By the proof of Theorem 1.6, to obtain ΛL,α it suffices to consider a small neighborhood V intersecting (L · s + α = 0). By the definition of characteristic cycles, we only need to consider a small neighborhood of a general point α ∈ V ∩ (L · s + α = 0), which is equivalent to globally considering a small neighborhood Vα of a very general point α ∈ (L · s + α = 0). By Corollary 3.18, δ∗ (Ψ (M)) = Ψ (M ). eαˆ i F fi αˆ i By Cohen-Macaulayness and Proposition 2.19, we can inductively apply [Wu20, Proposition 2.7] (by picking r − 1 general hyperplanes passing α) and conclude (3.19) δ∗ (CCrel(Ψ (M))) = CCrel(δ∗ (Ψ (M))) = CCrel(Ψ (M )). eαˆ i F eαˆ i F fi αˆ i RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 33

By the very general choice of α, we have rel CC (ΨF (M))|W ×Vα = ΛL,α × (L · s + α = 0), Taking V = δ−1(V ), we ﬁnish the proof by (3.18) and (3.19). αi αˆ i α

3.8. A lineariality conjecture. Let M be a regular holonomic DX -module. By (r) Proposition 3.5, the maximal extension M(∗DF ) is n-Cohen-Macaulay. On the contrary, for some k (and hence for all k by applying τ-translation) Nfk are not n-Cohen-Macaulay in general. Thus, we might have non-zero l E xt (Nfk, DeX,R) for n ≤ l ≤ n + r DeX,R and hence non-zero right DX,R-modules xtl ( , ) for n ≤ l ≤ n + r. E DX,R Nk DX,R Motivated by Corollary 3.15 and the relative condimension ﬁltration in §2.8, we conjecture:

Conjecture 3.20. Let M be a regular holonomic DX -module. If locally on a relative compact open subset W ⊆ X xtl ( , ) 6= 0 for some n ≤ l ≤ n + r, E DX,R Nk DX,R then we have: (i) If S ( xtl ( , )) 6= ∅, then it is a ﬁnite union of translated (j−n)- j E DX,R Nk DX,R codimensional linear subspaces of Cr. (ii) If M underlies a Q-mixed Hodge module (for example M = OX ), then all S ( xtl ( , )) are deﬁned over . j E DX,R Nk DX,R Q By Auslander regularity, Lemma 2.14 and Theorem 3.3, we have rel n rel r Ch (E xt (Nfk, DeX,R)) = Ch (Nfk) = Ch(M(∗D)) × DeX,R C and that xtn ( , ) is pure of codimension n. Thus, the conjecture above E DX,R Nk DX,R holds for l = n. When M0 = OX , Budur made a conjecture about the structure of generators of the Bernstein-Sato ideal BF [Bud15, Conjecture 1.1]. The following is a weak version of the conjecture.

Conjecture 3.21 (Budur). Each irreducible component of Z(BF ) is a translated linear subspace of Cr deﬁned over Q. By Sabbah [Sab87b] (or one can apply Corollary 3.15) and [Gyo93], codimension- r one irreducible components of Z(BF ) are translated linear subspaces of C deﬁned over Q. Proposition 3.22. Conjecture 3.20 =⇒ Conjecture 3.21.

Proof. We assume E an irrducible component of Z(BF ) of codimension 2 ≤ l ≤ r. We ﬁx a point (x, p) such that (Nf0/Nf−1)(x,p) 6= 0 (by Proposition 2.8 and

Nf0/Nf−1 = N^0/N−1) and p is a general point in E. By Corollary 2.9 and the dimension of E, we can choose a small neighborhood W of x ∈ X and a Zariski open neighborhood V ⊆ Cr of p such that the codimension ﬁltration satisﬁes

Tj(Nf0/Nf−1) = Nf0/Nf−1 34 LEI WU for j < n + l over W × V . By Lemma 2.14 and Corollary 2.17, the Auslander regularity tells

rel n+j codim Ch (E xt (Nf0/Nf−1, DeX,R)|W ×V ) > l for j > l. C DeX,R By shrinking V in the Zariski topology (pick a more general p ∈ E ∩V if necessary), we can assume that on W × V

Tn+l(Nf0/Nf−1) = Nf0/Nf−1 and Nf0/Nf−1 is (n + l)-Cohen-Macaulay. Now, we consider the short exact sequence (on W )

N0 0 → N−1 → N0 → → 0. N−1 Taking analytic sheaﬁﬁcation and dual, we have a long exact sequence

n+l−1 n+l Nf0 n+l · · · → E xt (Nf−1, DeX,R) → E xt ( , DeX,R) → E xt (Nf0, DeX,R) → · · · . DeX,R DeX,R DeX,R Nf−1

Since Nf0 and Nf−1 are n-pure (since they are submodules of a n-Cohen-Macaulay module Mfby Proposition 3.5 and Proposition 2.19), by [Bj93, A.IV.2.6], Corollary 2.17 and Lemma 2.14 we have

n+l n+l Tn+l+1(E xt (Nf0, DeX,R)) = E xt (Nf0, DeX,R) DeX,R DeX,R and n+l−1 n+l−1 Tn+l(E xt (Nf0, DeX,R)) = E xt (Nf0, DeX,R). DeX,R DeX,R

Since Nf0/Nf−1 is (n + l)-Cohen-Macaulay on W × V , we know

rel n+l−1 Nf0 rel Nf0 Ch (E xt ( , DeX,R)) = Ch ( ) DeX,R Nf−1 Nf−1 on W × V . Therefore,

n+l−1 Tn+l(E xt (Nf0, DeX,R)) rel n+l−1 Nf0 rel DeX,R Ch (E xt ( , DeX,R)) ⊆ Ch ( ). DeX,R n+l−1 Nf−1 Tn+l+1(E xt (Nf0, DeX,R)) DeX,R By Proposition 2.8 and Proposition-Deﬁnition 2.7, we have

n+l−1 E ⊆ Sn+l(E xt (Nf0, DeX,R)), DeX,R which proves the required implication.

4. Alexander complex of Sabbah In this section, we recall the construction of the Alexander complexes in [Sab90]. RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 35

4.1. Alexander complex of Sabbah. We keep using notations in §3.1. The universal Alexander sheaf is deﬁned by univ r ψ = Exp!(CC )

r where Exp! is the proper direct image functor of Exp (cf. [KS13, 2.5]) and CC is the C-constant sheaf on Cr. Since Exp gives the universal covering of (C∗)r, ψuniv is a locally free sheaf of C[G]-modules. We use the isomorphism ± ± C[G] ' C[t1 , . . . , tr ] with ti representing the (counterclockwise) loops around the puncture of each factor ∗ r univ of (C ) . Then the monodromy of ψ around each loop is induced by the ti- ± ± • b multiplication on C[t1 , . . . , tr ]. For F ∈ Dc(CX ), its Alexander complex along F (also known as Sabbah specialization complex) is defined as • −1 • −1 univ ψF (F ) = iD,∗iD Rj∗(F |X\D ⊗C F ψ ), where j : X \ D,→ X is the open embedding and iD : D,→ X is the closed embedding. More generally, for every subset I ⊆ {1, 2, . . . , r}, the Sabbah specialization T complex along DI = i∈I Di is define as ψ (F •) = i i−1Rj (F •| ⊗ F −1ψuniv), DI DI ,∗ DI ∗ X\D C univ where iDI : DI ,→ X is the closed embedding. Since ψ is locally free over C[G], • b ψDI (F ) ∈ Dc(C[G]). −1 ∗ r 4.2. The universal relative local system. We define a locally free pr O(C ) -module on the complex manifold (C∗)r × (C∗)r with pr :(C∗)r × (C∗)r → (C∗)r the second projection, denoted by Luniv, by requiring the following two conditions: univ −1 ∗ r ∗ r ∗ r (i) L |W ×(C ) ' pr O(C ) |W ×(C ) for all simply connected open subsets W ⊂ (C∗)r ∗ r ∗ univ (ii) for every λ = (λ1, . . . , λr) ∈ (C ) , eiλL is the local system on the first ∗ r ∗ r factor of (C ) ×(C ) satisfying that the monodromy along each ti is given ∗ r by multiplication by λi, where iλ : {λ} ,→ (C ) is the closed embedding into the second factor of (C∗)r × (C∗)r. It is easy to check that the two conditions above uniquely determine Luniv. Lemma 4.1. Luniv ' ψeuniv. univ univ Proof. Obviously, ψe satisfies the two conditions of L . 5. Comparison r r 5.1. The universal relative flat bundle. On Cx × C with x = (x1, . . . , xr) the coordinates, we write r Y r Dx = ( xi = 0) and U = C \ D. i=1 (r) Then using the recipe from §3, we have the maximal extension r (∗D ). We set OCx x univ (r) = r (∗D )| r , O OCx x U×C calling it the universal relative flat bundle on U × Cr because ∗ (r) (r) (5.1) F ( r (∗D )) = (∗D ) e OCx x OX F 36 LEI WU in the situation of §3.1, with Fe = (F, id): X × Cr → Cr × Cr the associated morphism, or equivalently ∗ s s F ( r (∗D ) ⊗ [s] · x ) = (∗D ) ⊗ [s] · F . OCx x C C OX F C C r Remark 5.1. Since x is the complex coordinates of Cx, it is a direct computation to get the relative characteristic cycle: rel (r) X ∗ r r CC r (∗D ) = ( T ) × , OCx x X¯β Cx C β F r where β Xβ = Cx is the stratification naturally given by the coordinates x. In (r) particular, r (∗D ) is relative holonomic. In fact, it is regular relative holonomic OCx x (see [MFS13, §2.1] for the definition of relative regularity). Therefore, if M is (r) a regular holonomic DX -module, then the relative holonomicity of M(∗DF ) in Theorem 3.3 can also be deduced from [FFS21, Theorem 2]. To avoid confusion, we set the morphism r ∗ r πx = (id, Exp): U × C → U × (C ) . Lemma 5.2. We have G univ q.i. univ −1 univ univ π (DRU× r / r (O )) 'L [n] and OU× r ⊗pr−1 r π (L ) ' O . x,∗ C C C OC x G univ univ r r Proof. Obviously, πx,∗(DRU×C /C (O )) satisfies the two conditions of L (up to the n-shifts). The quasi-isomorphism particularly gives us a morphism −1 univ univ OU× r ⊗pr−1 r π (L ) → O . C OC x r We pullback the morphism above by eiα for every α ∈ C and get an isomorphism. Thus by [FS19, Proposition 1.9.(1)], the morphism is isomorphic. 5.2. Proof of Theorem 1.2. We first prove the following theorem.

Theorem 5.3. Let M be a regular holonomic DX -module. Then naturally

G (r) q.i. −1 univ r r π∗ (DRX×C /C (M(∗DF ))) ' Rjg∗(DR(M)|X\D ⊗C F ψ ). Proof. By using the second isomorphism in Lemma 5.2 and (5.1), we get

(r) −1 −1 −1 univ r r r (5.2) DRX×C /C (M(∗DF ))|(X\D)×C ' p1 (DR(M|X\D))⊗C Fe (πx (L )), r −1 −1 univ where p1 : X × C → X is the ﬁrst projection. Since F (πx (L )) is G- equivariant and G acts trivially on objects originating from X (for instance M and −1 p1 (DR(M|X\D))), (5.2) is a G-equivariant isomorphism. Therefore, by Lemma 4.1

G (r) −1 univ r r ∗ r ^ π∗ (DRX×C /C (M(∗DF )))|(X\D)×(C ) ' DR(M)|X\D ⊗C F ψ . By adjunction, the isomorphism above induces a morphism G (r) −1 univ r r (5.3) π∗ (DRX×C /C (M(∗DF ))) → Rjg∗(DR(M)|X\D ⊗C F ψ ). ∗ r We then make derived pullback of the morphism above by eiλ for every λ ∈ (C ) . But by definition ∗ G (r) ∗ (r) r r r r Leiλπ∗ (DRX×C /C (M(∗DF ))) ' LeiαDRX×C /C (M(∗DF )) RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 37 for any α ∈ Exp−1(λ). Since M is regular, by Proposition 3.17 ∗ (r) ∗ (r) r r Leiα(DRX×C /C (M(∗DF ))) ' DR(Leiα(M(∗DF )) ' Rj∗(DR(M)|X\D ⊗ Lλ). Moreover, by construction Li∗ Rj (DR(M)| ⊗ F −1ψuniv) ' Rj (DR(M)| ⊗ F −1ψuniv) ⊗L , eλ g∗ X\D C ∗ X\D C C[G] Cλ ∗ r univ where Cλ is the residue field of λ ∈ (C ) . Since ψ is locally free over C[G] and hence over C, −1 univ −1 univ Rj∗(DR(M)|X\D⊗CF ψ )⊗C[G]Cλ ' Rj∗(DR(M)|X\D⊗CF ψ ⊗C[G]Cλ). By the definition of ψuniv, −1 univ F ψ ⊗C[G] Cλ ' Lλ. ∗ Therefore, the Leiλ pullback of the morphism (5.3) is an isomorphism for each λ. By [MFS13, Proposition 2.2 and Theorem 3.7], the morphism (5.3) is isomorphic. Although Theorem 1.1 is a special case of Theorem 1.2 (by Corollary 3.11), we first give a direct proof of Theorem 1.1 as a warm-up. We keep using notation in the proof of Theorem 5.3. By Lemma 3.7,

q.i. (r) (r) ΨF (M) ' cone(M(!DF ) → M(∗DF )). By Proposition 3.17,

∗ (r) (r) q.i. Leiαcone(M(!DF ) → M(!DF )) ' cone(M(!D) → M(∗D)). Taking the relative de Rham functor, we then have that ∗ (r) (r) r r DRX×C /C Leiαcone(M(!DF ) → M(!DF )) is quasi-isomorphic to • B := cone(j!(DR(M)|X\D ⊗ Lλ) → Rj∗(DR(M)|X\D ⊗ Lλ)). But • q.i. −1 B ' iD,∗iD Rj∗(DR(M)|X\D ⊗ Lλ). Thanks to [MFS13, Proposition 2.2], we have a natural isomorphism

q.i. −1 (r) r r r r DRX×C /C (ΨF (M)) ' eiD,∗eiD DRX×C /C (M(∗DF )) r r G where eiD = (iD, id): D × C ,→ X × C is the associated morphism. Since π∗ and −1 eiD,∗eiD commute, the proof of Theorem 1.1 is accomplished by Theorem 5.3 and functoriality of sheaﬁﬁcation. Proof of Theorem 1.2. By Lemma 3.13 and Theorem 3.12, we have G (r) π DR r r RΓ r ( M)(!D ) ∗ X×C /C D [DI ×C ] D F G (r) ' DR ∗ r ∗ r RΓ ∗ r π ( M)(!D ) . X×(C ) /(C ) D [DI ×(C ) ] ∗ D F By [MFS13, Theorem 3.11], we have G (r) DR ∗ r ∗ r RΓ ∗ r π ( M)(!D ) X×(C ) /(C ) D [DI ×(C ) ] ∗ D F G (r) ' DDR ∗ r ∗ r RΓ ∗ r π ( M)(!D ) , X×(C ) /(C ) [DI ×(C ) ] ∗ D F where D denotes the duality functor for relative constructible complexes (cf. [MFS13, (r) §2.6]). By Proposition 3.17, (DM)(!DF ) is relative regular holonomic (cf. [MFS19, 38 LEI WU

G (r) Deﬁnition 2.1]) and thus so is π∗ (DM)(!DF ) . By regularity and Lemma 2.11, DR and RΓ commute and thus we have G (r) DDR ∗ r ∗ r RΓ ∗ r π ( M)(!D ) X×(C ) /(C ) [DI ×(C ) ] ∗ D F G (r) ' DRΓ ∗ r DR ∗ r ∗ r π ( M)(!D ) . DI ×(C ) X×(C ) /(C ) ∗ D F (r) (r) By (3.11), we know (DM)(!DF ) ' D(M(∗DF )) and thus G (r) DRΓ ∗ r DR ∗ r ∗ r π ( M)(!D ) DI ×(C ) X×(C ) /(C ) ∗ D F G (r) ' DRΓ ∗ r DR ∗ r ∗ r π M(∗D ) DI ×(C ) X×(C ) /(C ) ∗ D F G (r) ' DRΓ ∗ r DDR ∗ r ∗ r π M(∗D ) , DI ×(C ) X×(C ) /(C ) ∗ F where in the second isomorphism above follows from (3.13) in Remark 3.10(2). By [KS13, Proposition 3.1.11] and the construction of D, we have

! DRΓ ∗ r D ' i D˜i D DI ×(C ) eDI ,∗ DI r r where eiDI = (iDI , id): DI × C ,→ X × C is the associated morphism. More- over, by [KS13, Proposition 3.1.13] (see also [MFS13, Remark 2.24]), for relative constructible complexes we have a functorial isomorphism D˜i! D ' i−1. DI eDI Therefore, G (r) DRΓ ∗ r DDR ∗ r ∗ r π M(∗D ) DI ×(C ) X×(C ) /(C ) ∗ F −1 G (r) ' i i DR ∗ r ∗ r π M(∗D ) . eDI ,∗eDI X×(C ) /(C ) ∗ F By Theorem 5.3 we ﬁnally have

−1 G (r) i i DR ∗ r ∗ r π M(∗D ) ' ψ (DR (M)). eDI ,∗eDI X×(C ) /(C ) ∗ F eDI X Proof of Corollary 1.3. By Lemma 2.20, we have

∗ r ∗ supp ∗ r ψ (DRM) = {λ ∈ ( ) | Li (ψ (DRM)) 6= 0}. (C ) F C eλ eF ∗ Since Leiα and DR commute, by the Riemann-Hilbert correspondence for regular holonomic DX -modules we have r ∗ r ∗ r r {α ∈ C | Leiα(DRX×C /C ΨF (M)) 6= 0} = {α ∈ C | Leiα(ΨF (M)) 6= 0} By Theorem 1.1,

r ∗ ∗ r ∗ r r G\{α ∈ C | Leiα(DRX×C /C ΨF (M)) 6= 0} = {λ ∈ (C ) | Leiλ(ψeF (DRM)) 6= 0}. Thus, it is enough to prove

r ∗ supp r Ψ (M) = {α ∈ | Li (Ψ (M)) 6= 0}. C F C eα F But the equality above is the analytiﬁcation of [BVWZ21a, Proposition 3.4.3] as ΨF (M) is (n+1)-Cohen-Macaulay. We leave details for interested readers; see also [BVWZ21a, §3.6]. RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 39

5.3. Proof of the local index formula. In this subsection, we prove the local index formula (1.3) in §1.3. We keep notations as in §1.3 and §3. We focus on a small neighborhood W ⊆ X around x. We write F • = DR(M) for simplicity and let q be the generic point of an • irreducible component of supp ∗ r ψ F . By Corollary 1.5,q ¯ is always a translated (C ) F subtori of codimension-one. We pick a general point λ ∈ q¯ and a point α ∈ Exp−1(λ). Then for i ∈ {1, . . . , r} we consider the closed embedding δ : ∗ ,→ ( ∗)r, t 7→ (λ , . . . , λ , t, λ , . . . , λ ). λˆ i C C 1 i−1 i+1 r By pullback the quasi-isomorphism in Theorem 1.1 and Corollary 3.18 we obtain: ∗ G (5.4) δe π DRX× / Ψfi (Mαˆ i ) ' ψefi (Fαˆ i ), λˆ i ∗ C C with Fαˆ i = DR(Mαˆ i ). By the deﬁnition of Mαˆ i , (5.4) is independent of the choices of α ∈ Exp−1(λ). r We pick a suﬃcient small neighborhood Vα of α ∈ C such that Exp(Vα) ' Vα (since Exp is the universal covering) and write V = δ−1(V ). αi αˆ i α We then take restriction of (5.4) on W × Exp(V ). Since Ψ (M )| is only αi fi αˆ i W ×Vαi supported on W × {α }, we can treat Ψ (M )| as a regular holonomic i fi αˆ i W ×Vαi DX -module (over W) and hence we get (5.5) DR (Ψ (M )| ) ' ψ (F ) X fi αˆ i W ×Vαi fi αˆ i mαi where mαi is the maximal ideal of αi ∈ C. Lemma 5.4. With notations as above, we have χ (ψ (F •), q) = χ (ψ (F • ), m ). x F x F αˆ i αi Proof. Since the length function is additive with respect to short exact sequence, the required identity follows from the general choice of α inq ¯.

We now apply [Gin86, Index Theorem 8.2] to the regular holonomic DX -module Ψ (M )| . In consequence, the formula (1.3) follows from (5.5), Lemma 5.4 fi αˆ i Vαi and Proposition 3.19.

5.4. A question about an algebraic relative Riemann-Hilbert correspondence. Both Theorem 1.1 and Theorem 5.3 can be seen as some special (regular) relative Riemann- Hilbert correspondence over arbitrary dimensional base spaces. See [FFS21] for the general regular relative Riemann-Hilbert correspondence over curves. Let us wrap up this article by asking the following question, motivated by our method in approaching Theorem 1.1: Question 5.5. For F • a constructible complex of sheaves of R-modules on a complex manifold X with R a commutative C-algebra of ﬁnite type, is it possible to construct r r a (regular) holonomic DX×C /C -module Mf(or a complex with (regular) relative holonomic cohomology sheaves) supported on X × SpecanR such that naturally

q.i. • r r DRX×C /C (Mf) ' Fe by ﬁxing an embedding Spec R,→ Cr? 40 LEI WU

References [BBDG18] Alexander Beilinson, Joseph Bernstein, Pierre Deligne, and Ofer Gabber, Faisceaux pervers, Sociétémathématique de France, 2018. [Bei87] Alexander Beilinson, How to glue perverse sheaves, K-theory, arithmetic and geometry (Moscow, 1984–1986), Lecture Notes in Math., vol. 1289, Springer, Berlin, 1987, pp. 42–51. [BG12] Alexander Beilinson and Dennis Gaitsgory, A corollary of the b-function lemma, Se- lecta Mathematica 18 (2012), no. 2, 319–327. [Bj93] Jan-Erik Björk, Analytic D-modules and applications, Mathematics and its Applica- tions, vol. 247, Kluwer Academic Publishers Group, Dordrecht, 1993. [BL94] Joseph Bernstein and Valery Lunts, Equivariant sheaves and functors, Lecture Notes in Mathematics, vol. 1578, Springer-Verlag, Berlin, 1994. [Bry86] Jean-Luc Brylinski, Transformations canoniques, dualitéprojective, théorie de Lef- schetz, transformations de Fourier et sommes trigonométriques, Astérisque 140 (1986), no. 141, 3–134. [Bud15] Nero Budur, Bernstein-sato ideals and local systems, Annales de l’Institut Fourier 65 (2015), no. 2, 549–603 (en). [BLSW17] Nero Budur, Yongqiang Liu, Luis Saumell, and Botong Wang, Cohomology support loci of local systems, Michigan Mathematical Journal 66 (2017), 295–307. [BVWZ21a] Nero Budur, Robin van der Veer, Lei Wu, and Peng Zhou, Zero loci of Bernstein-Sato ideals, Inventiones mathematicae 225 (2021), no. 1, 45–72. [BVWZ21b] , Zero loci of Bernstein-Sato ideals II, Sel. Math. New Ser. 27, 32 (2021), no. 3, 1022–1824. [BW15] Nero Budur and Botong Wang, Cohomology jump loci of quasi-projective varieties, Ann. Sci. Ec.´ Norm. Supér.(4) 48 (2015), no. 1, 227–236. [BW17] , Local systems on analytic germ complements, Adv. Math. 306 (2017), 905– 928. [FF18] Luisa Fiorot and Teresa Monteiro Fernandes, t-structures for relative D-modules and t-exactness of the de Rham functor, J. Algebra 509(2018), 419–444. [FFS21] Luisa Fiorot, Teresa Monteiro Fernandes, and Claude Sabbah, Relative regular Riemann-Hilbert correspondence, Proceedings of the London Mathematical Society 122 (2021), no. 3, 434–457. [FS19] Teresa Monteiro Fernandes and Claude Sabbah, Riemann-Hilbert correspondence for mixed twistor-modules, Journal of the Institute of Mathematics of Jussieu 18 (2019), no. 3, 629–672. [Gin86] Victor Ginsburg, Characteristic varieties and vanishing cycles, Inventiones Mathe- maticae 84 (1986), no. 2, 327–402. [Gyo93] Akihiko Gyoja, Bernstein-sato’s polynomial for several analytic functions, Journal of Mathematics of Kyoto University 33 (1993), no. 2, 399–411. [HTT08] Ryoshi Hotta, Kiyoshi Takeuchi, and Toshiyuki Tanisaki, D-modules, perverse sheaves, and representation theory, Progress in Mathematics, vol. 236, Birkhäuser Boston, Inc., Boston, MA, 2008, Translated from the 1995 Japanese edition by Takeuchi. [Kas83] Masaki Kashiwara, Vanishing cycle sheaves and holonomic systems of differential equations, Algebraic geometry (Tokyo/Kyoto, 1982), Lecture Notes in Math., vol. 1016, Springer, Berlin, 1983, pp. 134–142. [Kas03] , D-modules and microlocal calculus, Translations of Mathematical Mono- graphs, vol. 217, American Mathematical Society, Providence, RI, 2003, Translated from the 2000 Japanese original by Mutsumi Saito, Iwanami Series in Modern Math- ematics. MR 1943036 [Kas77] , B-functions and holonomic systems. Rationality of roots of b-functions, In- vent. Math. 38 (1976/77), no. 1, 33–53. MR 0430304 [KS13] Masaki Kashiwara and Pierre Schapira, Sheaves on manifolds: With a short his- tory.«Les débuts de la théorie des faisceaux», by Christian Houzel, vol. 292, Springer Science & Business Media, 2013. RIEMANN-HILBERT CORRESPONDENCE FOR ALEXANDER COMPLEXES 41

[Lau83] Gérard Laumon, Sur la catégorie dérivée des D-modules filtrés, Algebraic geometry (Tokyo/Kyoto, 1982), Lecture Notes in Math., vol. 1016, Springer, Berlin, 1983, pp. 151–237. MR 726427 [Mai16a] Philippe Maisonobe, Filtration Relative, l’Idéal de Bernstein et ses pentes, arXiv preprint arXiv:1610.03354 (2016). [Mai16b] , L’Idéal de Bernstein d’un arrangement libre d’hyperplans linéaires, preprint arXiv:1610.03356 (2016). [Mal83] B. Malgrange, Polynômes de Bernstein-Sato et cohomologie évanescente, Analysis and topology on singular spaces, II, III (Luminy, 1981), Astérisque, vol. 101, Soc. Math. France, Paris, 1983, pp. 243–267. [Mal04] B. Malgrange, On irregular holonomic D-modules, in Eléments de la théorie des sys- temes différentiels géométriques, Séminaires & Congrés,vol. 8 (2004), 391-410. [MFS13] Teresa Monteiro Fernandes and Claude Sabbah, On the de Rham complex of mixed twistor D-modules, Int. Math. Res. Not. IMRN (2013), no. 21, 4961–4984. [MFS19] Teresa Monteiro Fernandes and Claude Sabbah, Riemann-Hilbert correspondence for mixed twistor D-Modules, Journal of the Institute of Mathematics of Jussieu 18 (2013), no. 3, 629–672. [NS96] Nitin Nitsure and Claude Sabbah, Moduli of pre-D-modules, perverse sheaves and the Riemann-Hilbert morphism-I, Math. Annalen 306 (1996), 47–73. [Sab87a] Claude Sabbah, Proximitéévanescente. I. La structure polaire d’un D-module, Com- positio Math. 62 (1987), no. 3, 283–328. [Sab87b] , Proximitéévanescente. II. Equations´ fonctionnelles pour plusieurs fonctions analytiques, Compositio Math. 64 (1987), no. 2, 213–241. [Sab90] , Modules d’Alexander et D-modules, Duke Mathematical Journal 60 (1990), no. 3, 729–814. [Sch12] Pierre Schapira, Microdifferential systems in the complex domain, vol. 269, Springer Science & Business Media, 2012. [Sta18] The Stacks Project Authors, Stacks Project, https://stacks.math.columbia.edu, 2018. [Wu20] Lei Wu, Characteristic cycles associated to holonomic D-modules, arXiv preprint arXiv:2011.14527 (2020). [Wu21] , On the comparison of nearby cycles via b-functions, J. Singul. 23 (2021), 92-106. [WZ21] Lei Wu and Peng Zhou, Log D-modules and index theorems, Forum of Mathematics, Sigma 9 (2021).

Lei Wu, KU Leuven, Celestijnenlaan 200B, B-3001 Leuven, Belgium Email address: [email protected]