Minimal resolutions of geometric \(\mathcal D\)-modules (Q2268589)

Let \(\mathcal{D}\) be the ring of linear differential operators with analytic coefficients. The author studies minimal free resolutions over \(\mathcal{D}\) as defined by \textit{M. Granger} and \textit{T. Oaku} [J. Pure Appl. Algebra 191, No.~1--2, 157--180 (2004; Zbl 1064.16024)]. Bifiltrations on \(\mathcal{D}\) are defined by an admissible weight vector (i.e., by a pair \((u,v)=(u_{1}, \dots,u_{n}, v_{1},\dots, v_{n})\in\mathbb{Z}^{2n}\) with \(u_{i}+v_{i}\geq0\) and \(u_{i}\leq0\) for all \(i\)); if \(P=\sum a_{\alpha,\beta} x^{\alpha}\partial^{\beta}\), ord\(_{(u,v)}P\) is defined as \(\max\left\{\sum u_{i}\alpha_{i}+\sum v_{i}\beta_{i}\mid a_{\alpha,\beta}\neq0\right\}\). The author considers first the filtration given by \(F_{k}^{(u,v)}(\mathcal{D})=\left\{ P\in\mathcal{D},\;\text{ord}_{(u,v)}P\leq k\right\}\) for \(k\in\mathbb{Z}\). (In the particular case \((u,v)=(0,1)\), \(F_{d}^{(0,1)}(\mathcal{D})\) is denoted by \(F_{d}(\mathcal{D})\) and the filtration is the classical \(V\)-filtration). For an admissible weight vector, one can define a bifiltration on \(\mathcal{D}\) by \(F_{d,k}(\mathcal{D})=F_{d}^{(0,1)}(\mathcal{D})\cap F_{k}^{(u,v)}(\mathcal{D})\), \(d\in\mathbb{N}\), \(k\in\mathbb{Z}\), and the bigraded ring bigr\((\mathcal{D})= \bigoplus\)bigr\(_{d,k}(\mathcal{D})=\bigoplus\dfrac{F_{d,k}(\mathcal{D})}{F_{d,k-1}(\mathcal{D})+F_{d-1,k}(\mathcal{D})}\) is a commutative ring. For a \(\mathcal{D}\)-graded module \(M\), the construction of a minimal bifiltered resolution is adapted to the bifiltration of \(\mathcal{D}\). If \(M\) is a filtered \(\mathcal{D}\)-module, the filtration is good if there exist \(f_{1},\dots,f_{r}\in M\) and \(n_{1},\dots,n_{r}\in\mathbb{Z}\) such that, for all \(d\in\mathbb{Z}\), \(F_{d}(M)=F_{d-n_{1}}(\mathcal{D})f_{1}+\cdots+F_{d-n_{r}}(\mathcal{D})f_{r}\). A bifiltration of \(M\) is a collection \(\left\{ F_{d,k}(M)\right\}_{d,k\in\mathbb{Z}}\) of \(\mathbb{C}\left\{ x\right\}\) submodules such that \(F_{d,k+1}(M)\supset F_{d,k}(M)\), \(\bigcup F_{d,k}(M)=M\) and \((F_{d^{'}}(\mathcal{D})\cap V_{k^{'}}(\mathcal{D}))F_{d,k}(M)\subset F_{d+d^{'},k+k^{'}}(M)\) when \(V_{k}(\mathcal{D})=F_{k}^{(u,v)}(\mathcal{D})\) where \((u,v)\neq(0,1)\) is an admissible weight vector. There is a natural definition of a good bifiltration, and, if \(M\) is endowed with a good bifiltration, then \(M\) has a good \(F\)-filtration (where \(F_{d}(M)=\mathop{\bigcup}\limits_{k\in\mathbb{Z}}F_{d,k}(M)\)) and a \(V\)-filtration. A filtered free resolution of \(M\) is a resolution \dots\(\rightarrow\mathcal{L}_{1}\overset{\phi_{1}}{\rightarrow}\mathcal{L}_{0}\overset{\phi_{0}}{\rightarrow}M\rightarrow0\), with \(\mathcal{L}_{i}=\mathcal{D}^{r_{i}}[\mathbf{n}^{(i)}]\) such that, for any \(d\in\mathbb{Z}\), one has an exact sequence \dots\(\rightarrow F_{d}(\mathcal{L}_{1})\overset{\phi_{1}}{\rightarrow}F_{d}(\mathcal{L}_{0})\overset{\phi_{0}}{\rightarrow}F_{d}(M)\rightarrow0\). A minimal filtered free resolution is a filtered free resolution which induces a minimal graded free resolution of gr\(^{F}(M)\):\(\cdots\rightarrow\)gr\(^{F}\mathcal{L}_{1}\rightarrow\)gr\(^{F}\mathcal{L}_{0}\rightarrow\)gr\(^{F}M\rightarrow0\); equivalently, these can be defined via a Rees functor \({\mathbf R}\). Now, \({\mathbf R}(M)=\oplus F_{d}(M)T^{d}\) has a natural \(V\)-filtration and the graded gr\(^{V}(\mathcal{D}^{(h)})_{V}\)-module associated with \(M\) is gr\(^{V}({\mathbf R}(M))=\mathop{\bigoplus}\limits_{k}\)gr\(_{k}^{V}({\mathbf R}(M))\). Finally, a bifiltered free resolution \(\dots\rightarrow\mathcal{L}_{1}\rightarrow\mathcal{L}_{0}\rightarrow M\rightarrow0\) with \(\mathcal{L}_{i}=\mathcal{D}^{r_{i}}[\mathbf{n}^{(i)}][\mathbf{m}^{(i)}]\) is called minimal if the induced bigraded exact sequence \dots\(\rightarrow\)gr\(^{V}({\mathbf R}(\mathcal{L}_{1}))\rightarrow\) gr\(^{V}({\mathbf R}(\mathcal{L}_{0}))\rightarrow\) gr\(^{V}({\mathbf R}(M))\rightarrow0\) is a minimal bigraded resolution of gr\(^{V}({\mathbf R}(M))\). Such a resolution exists and is unique up to a bifiltered isomorphism [Granger-Oaku, loc. cit.]; thus, the \(r_{i}\) and the shifts \(\mathbf{n}^{(i)}, \mathbf{m}^{(i)}\) are invariants of the bifiltered free resolution of \(M\). The main difficulty is here the non-commutativity. The main result is given by Theorem 1.1 which asserts that, if \(M\) is a \(\mathcal{D}\)-module endowed with a good bifiltration such that, for any \(d, k\), \(F_{d,k}(M)=F_{d}(M)\cap V_{k}(M)\) then bigr\(M\) is of finite type over bigr\,\(\mathcal{D}\) and there is a unique bifiltered free resolution which induces a minimal bigraded free resolution \dots\(\rightarrow\)bigr\((\mathcal{L}_{1})\rightarrow\)bigr\((\mathcal{L}_{0})\rightarrow\)bigr\((M)\rightarrow0\). Moreover, the minimal free resolution, i.e., the induced bigraded exact sequence \dots\(\rightarrow\)gr\(^{V}({\mathbf R}(\mathcal{L}_{1}))\rightarrow\) gr\(^{V}({\mathbf R}(\mathcal{L}_{0}))\) \(\rightarrow\)gr\(^{V}({\mathbf R}(M))\rightarrow0\) is a minimal one of gr\(^{V}({\mathbf R}(M))\). The author calls this the reduction to a commutative algebra problem. This is applied to the following situation: Let \(D_{x,t}f\) be the module associated with germs of functions \(f_{1},\dots,f_{p}\) and endowed with the \(V\)-filtration along \(f_{1}=f_{2}=\dots=f_{p}\). Given the application \(f=(f_{1},\dots,f_{p}):X\subset(\mathbb{C}^{n},0)\rightarrow(\mathbb{C}^{p},0)\) with \(X\) open. The Betti numbers of the minimal resolution associated with the corresponding minimal resolution lead to analytic invariants. Let \(i_{f}:X\rightarrow X\times\mathbb{C}^{p}\) be a graph embedding, \(\mathcal{O}_{X}\) the sheaf of analytic functions on X, \(\mathcal{D}\) the ring of germs at \(0\) of the differential operator on \(\mathbb{C}^{n}\times\mathbb{C}^{p}\). The author considers the \(\mathcal{D}\) module \(N_{f}=(i_{f+}(\mathcal{O}_{X}))_0\) endowed with a good bifiltration which defines a \(V\) filtration along \(\left\{ t_{1}=\dots=t_{p}=0\right\}=X\times 0\). The interesting thing is that the algebraic local cohomology \({\mathbf R}\Gamma_{| f=0|}(\mathcal{O}_{X})\) is equal to the restriction of \(i_{f+}(\mathcal{O}_{X})\) along \(\left\{ t_{1}=\dots=t_{p}=0\right\}\) and the restriction can be obtained from a free resolution adapted to the \(V\)-filtration [\textit{T. Oaku} and \textit{N. Takayama}, J. Pure Appl. Algebra 139, No.~1--3, 201--233 (1999; Zbl 0960.14008)]. The author gives an explicit computation of the Betti numbers of the graded \(\mathcal{D}_{x,t}^{(h)}\)-module where \(\mathcal{D}^{(h)}\) is the homogenization of \(\mathcal{D}\). If \(i_{++}(\mathcal{O})\) is the stalk at zero of \(i_{++}(\mathcal{O}_{X})\), then \(N_{f}=\mathcal{D}_{x,t}f^{s}\) is isomorphic to \(i_{++}(\mathcal{O})\) and there is an isomorphism between \({\mathbf R}(N_{f})\) and \(\frac{\mathcal{D}_{x,t}^{(h)}}{\left((t_{j}-f_{j})_{j},(\partial_{x_{i}}+\sum_{j}\frac{\partial f_{j}}{\partial x_{i}\partial t_j})_{i}\right)}\) and the Betti numbers of \({\mathbf R}(N_{f})\) are \(\left(\begin{matrix} n+p\\ i \end{matrix}\right)\) for \(i=0,\dots,n+p\). The Betti numbers and the shifts of \(N_{f}\) are invariants for the ideal \(I=(f_{1},\dots,f_{p})\) if \(p\) is minimal, i.e., \(p=\)dim\(_{\mathbb{C}}(I/\mathbf{m}I)\). The author defines an invariant for the germ of a complex space (i.e., a local algebra \(\mathbb{C}[x]/J\)). If \(n_{0}\) is the embedding dimension of \(V\) at 0 and \(p_{0}\) the minimal number of a generating set defining the ideal defining \(V\) in \(\mathbb{C}\left\{ x_{1},\dots,x_{n_{0}}\right\}\) and \(f_{1},\dots,f_{p}\) a generating set of \(I\), \(c_{0}=p_{0}+n_{0}\) is an invariant of \(V\), and the series \(\beta_{V}(T)=\frac{\beta_{f}(T)}{(1+T)^{n+p-c_{0}}}\) (which turns out to be a polynomial in \(T\)) is an invariant for the germ of \(V\). Here, \(\beta_{f}(T)=\mathop{\sum}_{i\in\mathbb{Z}}\beta_{i}(f)T^{j}\), where the \(\beta_{j}(f)\) are the Betti numbers of \(N_{f}\). In the case of hypersurfaces (\(p=1\), Section 2.2), the author gives a representation of gr\(^{V}({\mathbf R}(N_{f}))\) using ann\(_{\mathcal{D[s]}}f^{s}\) and a condition under which one can apply the reduction to commutative algebra of \(N_{f}\). This condition is based on the notion of \(h\)-saturation of a bigraded gr\(^{V}(\mathcal{D}^{(h)})\) module \(N\), which means that the homogenization \(h:N\rightarrow N\) is one-to-one, and more precisely that \(\text{gr}^{F} (\text{ann}_{\mathcal{D}[s]} f^{s})=\text{ker}\, \varphi_{f}\) that ensures that \(\text{gr}^{V} ({\mathbf R}(\mathcal{D}_{x} f^{s}))\) is \(h\)-saturated (here, \(\varphi_{f}:\text{gr}^{F} (\mathcal{D}[s])\cong\mathcal{O} [s,\xi_{1},\dots,\xi_{n}]\rightarrow\bigoplus_{d\geq0}(\mathcal{O}_{f}+I(f))^{d}T^{d}\) and \({\mathbf R} (\mathcal{D}_{x}f^{s})\) is the Rees ring associated to \(\mathcal{D}[s]\) endowed with the \(F\)-filtration). Finally, in Section 3, the author uses the results obtained above to study isolated singularities and quasi-homogenity. For an isolated singularity \(f:(\mathbb{C}^{n},0)\rightarrow(\mathbb{C},0)\), the author studies the Betti numbers of the bifiltered module \(N=\mathcal{D}_{x,t}f^{s}\). Supposing that \(f\) is quasi-homogeneous, the author gives expressions for the Betti numbers of \(\mathcal{D}[s]f^{s}\) and of those of \(M=\frac{\mathcal{D}[s]f^{s}}{\mathcal{D}[s]f^{s+1}}\), and finally for those of \(N\). The author obtains also a characterization of quasi-homogeneity: \(f\) is \(q\)-homogeneous if and only if reg\(_{f}N=0\) where the definition of reg\(_{f}N\) is related to a minimal free resolution of \(N\) (more specifically, to the shifts that appear there). In the case of reduced curves, the first Betti number is sufficient: if \(f\) is quasi-homogeneous, then \(\beta_{i}=5\), otherwise \(\beta_{i}\geq 6\).