Zero loci of Bernstein-Sato ideals (Q2041856)

Let \(X\) be a complex manifold and let \(x\) be a point of \(X\). Let us denote by \(\mathcal{D}_{X,x}\) the stalk at \(x\) of the sheaf of analytic differential operators and by \(\mathbf{s}\) the tuple of variables \(s_1,\ldots,s_r\). For any germ of a holomorphic map \(F=(f_1,\ldots,f_r):(X,x)\rightarrow(\mathbb{C}^r,0)\), we can consider the (local) Bernstein-Sato ideal of \(F\), denoted by \(B_F\). It is the ideal generated by all polynomials \(b(\mathbf{s})\in\mathbb{C}[\mathbf{s}]\) such that \(b(\mathbf{s})\prod_{i=1}^r f_i^{s_i}\in\mathcal{D}_{X,x}[\mathbf{s}]\cdot \prod_{i=1}^r f_i^{s_i+1}\), where \(\mathbf{f}^\mathbf{s}:=\prod_{i=1}^r f_i^{s_i}\) and \(\mathbf{f}^\mathbf{s+\mathbf{1}}:=\prod_{i=1}^r f_i^{s_i+1}\) are symbols on which differential operators act in the usual way. If \(X\) is a smooth complex affine variety and the \(f_i\) are regular functions, there is a (global) algebraic version of \(B_F\), replacing everything above by their algebraic counterparts. When \(r=1\), the Bernstein-Sato ideal is generated by the well-known Bernstein-Sato polynomial \(b_f(s)\), whose roots are negative rational numbers that determine the monodromy action on the nearby cycles complex of \(f\), for its eigenvalues form the set \(\{\operatorname{exp}(2\pi i \alpha)\,:\,b_f(\alpha)=0\}\). In the general case where \(r\geq1\), Sabbah introduced in [Duke Math. J. 60, No. 3, 729--814 (1990; Zbl 0715.14007)] an analogous notion of the nearby cycles complex to a finite family of regular functions (the specialization complex), that has an associated monodromy action whose support is denoted by \(S(F)\). Let us call \(Z(B_F)\) the vanishing locus of the Bernstein-Sato ideal \(B_F\). There is a generalization of the aforementioned statement when \(r=1\) to this context, which is straightforward to state: \(\operatorname{exp}(2\pi i Z(B_F))=S(F)\). However, its proof is far from being easy. One of the inclusions (the right-hand object lies within the left-hand one) was proved by the first author of the paper under review in [Ann. Inst. Fourier. 65, No. 2, 549--603 (2015; Zbl 1332.32038)]. This very well-written paper closes the circle and proves unconditionally the other inclusion, both in the analytic and algebraic settings. After some preliminaries on \(S(F)\) in section 2, the authors embark on the proof of the main result. They use intensively several results on pure and Cohen-Macaulay modules over a ring and relative holonomic \(\mathcal{D}\)-modules, as well as rely crucially on some results from [\textit{Ph. Maisonobe}, ``Filtration Relative, l'Idéal de Bernstein et ses pentes'', Preprint, \url{arXiv:1610:03354}]. All that is done in order to show, as a key intermediate result, that for any \(\alpha\in Z(B_F)\subseteq \mathbb{C}^r\), the tensor product \(\mathcal{D}_X[\mathbf{s}]\cdot\mathbf{f}^{\mathbf{s}}/ \mathcal{D}_X[\mathbf{s}]\cdot\mathbf{f}^{\mathbf{s}+\mathbf{1}} \otimes_{\mathbb{C}[\underline{s}]}\mathbb{C}_\alpha\) does not vanish, \(\mathbb{C}_\alpha\) being the residue field of \(\alpha\) in \(\mathbb{C}^r\). This implies the desired inclusion thanks to the results on \(S(F)\).