Generalized nearby cycles via relative and logarithmic $\mathscr{D}$-modules

1. Problem Statement and Context

The paper addresses the generalization of the classical theory of nearby cycles and vanishing cycles from the setting of a single holomorphic function to the setting of a regular map $F = (f_1, \dots, f_r): X \to \mathbb{A}^r_{\mathbb{C}}$ (where $X$ is a smooth affine variety).

• Classical Setting: For a single function $f$, Kashiwara and Malgrange established the $V$-filtration on holonomic $\mathcal{D}$-modules, leading to the construction of nearby cycles. Under the Riemann-Hilbert correspondence, these correspond to Deligne's topological nearby cycles for perverse sheaves.
• Multi-variable Setting: For a tuple of functions, Sabbah introduced Sabbah specialization complexes (Alexander complexes) using toric modifications and normal deformations. These generalize Deligne's cycles but lack a direct, unified algebraic construction via $\mathcal{D}$-modules that mirrors the single-variable case.
• The Gap: There was no general construction of "generalized nearby-cycle modules" for regular holonomic $\mathcal{D}$-modules in the algebraic setting that directly corresponds to Sabbah's complexes under the relative Riemann-Hilbert correspondence. Furthermore, the geometric structure of the zero loci of Bernstein-Sato ideals (polynomials governing the functional equations of $F^s$) in the multi-variable case was not fully understood topologically.

Core Objective: To construct generalized nearby-cycle modules for regular holonomic $\mathcal{D}$-modules along log strata induced by the graph of $F$, prove their correspondence to Sabbah's specialization complexes, and use this to provide a topological interpretation of the zero loci of Bernstein-Sato ideals.

2. Methodology and Technical Framework

The author employs a synthesis of logarithmic geometry, relative $\mathcal{D}$-module theory, and Hodge module theory.

A. Logarithmic $\mathcal{D}$-Modules and Log Structures

• Log Structure: The paper considers the log structure on $Z = X \times \mathbb{A}^r_{\mathbb{C}}$ induced by the graph of $F$. This is defined via the pre-log structure $\alpha: \mathbb{N}^r \hookrightarrow S[t_1, \dots, t_r]$.
• Logarithmic Differential Operators: The ring of logarithmic differential operators $D_{Z,E}$ is utilized, where $E$ is the normal crossing divisor $(t_1 \cdots t_r = 0)$.
• Lattices: The construction relies on logarithmic $\mathcal{D}$-module lattices (e.g., $N_0 = \mathcal{D}_X[s_1, \dots, s_r]H_0 \cdot F^s$), which serve as algebraic models for the analytic modules.

B. Relative Regular Holonomic $\mathcal{D}$-Modules

• The paper operates within the framework of relative regular holonomic $\mathcal{D}$-modules (developed by Fiorot, Fernandes, and Sabbah). Here, the base space is $\mathbb{A}^r_{\mathbb{C}}$ (or its analytic counterpart), and the modules are defined over the relative ring $\mathcal{D}_{X \times \mathbb{A}^r / \mathbb{A}^r}$.
• Graph-Embedding Type: A crucial technical condition is that modules are of "graph-embedding type" (GE-type), meaning they are supported on the graph of the map $F$. This allows the use of Kashiwara's equivalence to relate them to modules on the base $X$.

C. Key Technical Tools

1. Relative Minimal Extensions: The author constructs relative minimal extensions of lattices along log boundaries. This generalizes the Deligne-Goresky-MacPherson minimal extension for perverse sheaves and the $j_!$-extension of Beilinson-Bernstein.
2. Logarithmic Artin-Rees Theorem: A significant technical hurdle in $\mathcal{D}$-module theory is the failure of the classical Artin-Rees lemma. The author proves a Logarithmic Artin-Rees Theorem (Theorem 5.14), showing that for ideals generated by monoid elements $\alpha(K)$, the Artin-Rees property holds. This ensures that the generalized nearby cycles form a module rather than just a complex.
3. Double Limit Construction: The generalized nearby cycles are defined via a double limit (projective and inductive) involving localizations and minimal extensions, generalizing Beilinson's construction.
4. Logarithmic Characteristic Cycle Formula: Theorem 5.1 provides a formula for the characteristic cycle of lattices, which is essential for controlling the geometry of supports during specialization.

3. Key Contributions and Results

I. Construction of Generalized Nearby Cycles

The paper defines the generalized nearby-cycle module $\Psi_K(M)$ (or $\tilde{i}^!_K N_0$) for a regular holonomic $\mathcal{D}$-module $M$ along a monoid ideal $K \subseteq \mathbb{N}^r$.

• Theorem 5.17 & Definition 5.26: These modules are constructed as relative regular holonomic modules. They are shown to be independent of the choice of generators for $K$.
• Comparison Theorem (Theorem 5.25): Under the relative regular Riemann-Hilbert correspondence, the de Rham complex of the generalized nearby-cycle module is isomorphic to the Sabbah specialization complex $\psi_K(\text{DR}(M))$. This generalizes the Kashiwara-Malgrange comparison theorem to the multi-variable case.

II. Geometry of Supports (Theorems C and D)

The paper analyzes the geometric structure of the relative supports of these modules.

• Linear Structure: The support of the generalized nearby cycles in the parameter space $\mathbb{C}^r$ is an infinite (countable) union of translated linear subvarieties.
• Exponential Map: Under the universal covering map $\text{Exp}: \mathbb{C}^r \to (\mathbb{C}^*)^r$, these supports map to a finite union of translated subtori in the torus.
• Hodge Module Refinement: If $M$ underlies a $\mathbb{Q}$-mixed Hodge module, the supports are unions of $\mathbb{Q}$-translated linear subvarieties, mapping to torsion translated subtori.

III. Logarithmic Interpretation of Bernstein-Sato Ideals

The paper provides a logarithmic interpretation of Bernstein-Sato ideals $B_K(N_0)$, defined as the annihilator of the module $N_0 / \alpha(K)N_0$.

• Theorem 6.5: Establishes the fundamental equality:
  $$\text{Exp}\left( Z(B_K(N_0)) \right) = \text{supp}\left( \psi_K(\text{DR}(M)) \right)$$
  where $Z(B_K(N_0))$ is the zero locus of the Bernstein-Sato ideal.
• Corollary 6.6: Consequently, the zero locus of the Bernstein-Sato ideal, when exponentiated, is a finite union of translated subtori (and torsion subtori in the Hodge case).

IV. Answering the Budur-Shi-Zuo Question

The results provide a complete topological interpretation of the zero loci of Bernstein-Sato ideals for a regular map $F$.

• Corollary B: For $M = \mathcal{O}_X$, $\text{Exp}(Z(B_K^F)) = \text{supp}(\psi_K(\mathbb{C}_X))$.
• This confirms that the zero loci are geometrically "linear" (in the log sense) and answers a question posed by Budur, Shi, and Zuo regarding the topological nature of these ideals.

4. Significance

1. Unification of Frameworks: The paper successfully unifies the algebraic theory of $\mathcal{D}$-modules (via relative regular holonomic modules) with the topological theory of Sabbah specialization complexes. It provides a robust algebraic construction for objects previously defined via topological or toric methods.
2. Resolution of Open Problems: It settles the question of the geometric structure of Bernstein-Sato ideals for multi-variable maps, proving they are unions of translated subtori. This validates and generalizes previous conjectures by Budur.
3. New Technical Machinery: The development of the Logarithmic Artin-Rees Theorem and the theory of relative minimal extensions for $\mathcal{D}$-modules provides new tools for the study of singularities and D-modules in logarithmic and relative settings.
4. Hodge Theory Connection: By refining the results for $\mathbb{Q}$-mixed Hodge modules, the paper connects the arithmetic properties of Bernstein-Sato ideals (torsion points) with the underlying Hodge-theoretic structure of the variety.

In summary, Lei Wu's work extends the classical Kashiwara-Malgrange theory to the multi-variable logarithmic setting, offering a powerful algebraic framework to study the topology of singularities and the arithmetic of Bernstein-Sato ideals.