Comparing the commutative and non-commutative resolutions for determinantal varieties of skew symmetric and symmetric matrices (Q2013997)

Let \(k\) be an algebraically closed field of characteristic \(0\), let \(Y\) be the determinantal variety of \(n\times n\)-matrices of rank \(n\leq r\), and let \(k[Y]\) be its coordinate ring. There exists a construction of a \textit{non-commutative crepant resolution, (NCCR),} \(\Lambda\) for \(k[Y]\) which is a \(k[Y]\)-algebra which in particular has the property that \(\mathcal D(\Lambda)\) is a \textit{strongly crepant categorical resolution} of \(\text{Perf}(Y)\). Here \(\mathcal D(\Lambda)\) denotes the bounded derived category of right \(\Lambda\)-modules with finitely generated cohomology and \(\text{Perf}(Y)\) is the category of perfect complexes on \(Y\). The NCCR is constructed by the tilting bundle on the standard Springer type resolution of singularities \(Z\rightarrow Y\) with \(Z\) a vector bundle over a Grassmannian. This says that the main properties of \(\Lambda\) are derived from the existence of an equivalence between \(\mathcal D(\Lambda)\) and \(\mathcal D(Z)\). In this article, the authors applies this results to determinantal varieties of symmetric and skew symmetric matrices, which are reasonably similar. Because it is cumbersome to treat both at the same time, the present work concentrates on skew symmetric matrices. Put \(n>r>0\) with \(2|r\), and let \(Y\) be the variety of skew symmetric \(n\times n\)-matrices with \(\text{rank}\leq r\). For odd \(n\) the authors constructed an NCCR \(\Lambda\) for \(k[Y]\). The similar construction for \(r\) even works only almost. This construction of the NCCR \(\lambda\) is based on invariant theory and does not use geometry. It is well known that in this case \(Y\) has a canonical (commutative) Springer type resolution of singularities \(Z\rightarrow Y\) and the interest here is the relationship between the resolutions of \(\Lambda\) and \(Z\). A \(k[Y]\)-linear embedding \(\mathcal D(\Lambda)\hookrightarrow\mathcal D(Z)\) is constructed, and for \(n\) odd it was expected to be an embedding because of the conjecture that NCCRs are minimal categorical resolutions. The embedding is however non-trivial, the image of \(\Lambda\) is a coherent sheaf of \(\mathcal O_Z\)-modules, not a vector bundle. The invariant theory used to construct \(\Lambda\) goes as follows: Let \(H,V\) be vector spaces of dimension \(n,r\) with \(V\) being in addition equipped with a symplectic bilinear form \(\langle-,-\rangle\), and let \(\text{Sp}(V)\) be the corresponding symplectic group. If \(\chi\) is a partition with \(l(\chi)\leq r/2,\) then let \(S^{\langle\chi\rangle}V\) be the irreducible representation of \(\text{Sp}(V)\) with highest weight \(\chi\). If \(\chi=(\chi_1,\dots,\chi_r)\in\mathbb Z^r\) is a dominant \(\text{GL}(V)\)-weight, then \(S^\chi V\) denotes the irreducible \(\text{GL}(V)\)-representation with highest weight \(\chi\). For \(X=\text{Hom}(H,V),\) the coordinate ring of \(X\) is \(T=\text{Sym}_k(H\otimes_k V^\vee)\), put \(M(\chi)=(S^{\langle\chi\rangle}V\otimes_k T)^{\text{Sp}(V)}.\) Let \(B_{m,n}\) be the set of partitions contained in a box with \(m\) rows and \(n\) columns, and let \(M=\bigoplus_{\chi\in B_{r/2,\lfloor n/2\rfloor-r/2}}M(\chi),\) and finally \(\Lambda=\text{End}_R(M)\). A result from earlier work states that \(\text{gl}\dim\Lambda<\infty\), and that if \(n\) is odd then \(\Lambda\) is a NCCR of \(R\). The first fundamental theorem for the symplectic group states that \(R\) is a quotient of \(\text{Sym}_k(\wedge^2 H)\), and dually \(\text{Spec}R\hookrightarrow\wedge^2H^\vee\subset\text{Hom}_k(H,H^\vee).\) The second fundamental theorem for the symplectic group then gives \(\text{Spec}R=\{\psi|\psi\in\text{Hom}_k(H,H^\vee),\psi+\psi^\vee=0,\;\text{rk}\psi\leq r\}\) implying \(\text{Spec}R\cong Y\) so that \(k[Y]\) can be identified with \(R\). Let \(F\) be the Grassmannian of \(r\)-dimensional quotients \(H\twoheadrightarrow Q\) of \(H\), \(F=\text{Gr}(r,H)\), and put \(Z=\{(\phi,Q)|Q\in F,\;\phi\in\text{Hom}_k(Q,Q^\vee),\;\phi+\phi^\vee=0\}\). The Springer resolution \(p:Z\rightarrow Y\hookrightarrow\text{Hom}_k(H,H^\vee)\) of \(Y\) is defined by sending \((\phi,Q)\) to the composition \(H\twoheadrightarrow Q\rightarrow Q^\vee\hookrightarrow H^\vee.\) It then follows from the fundamental theorems that \(\text{Sym}_k(Q\otimes_k V^\vee)^{\text{Sp}(V)}\cong\text{Sym}_k(\wedge^2 Q)\). For a partition \(\chi\) with \(l(\chi)\leq r/2\) put \(M_Q(\chi)=(\det Q)^{\otimes r-n}\otimes_k(S^{\langle\chi\rangle}V\otimes_k\text{Sym}_k(Q\otimes_k V^\vee))^{\text{Sp}(V)}\) where \(M_Q(\chi)\) is considered as a \(\text{GL}(Q)\)-equivariant \(\text{Sym}_k(\bigwedge^2 Q)\)-module. Choose a specific \((H\twoheadrightarrow Q)\in F\). Then \(F=\text{GL}(H)/P_Q\), \(P_Q\) the parabolic subgroup of \(\text{GL}(H)\) stabilizing the kernel \(H\twoheadrightarrow Q.\) The \(\text{GL}(Q)\)-equivariant objects are regarded as \(P_Q\)-equivariant objects through the canonical morphism \(P_Q\twoheadrightarrow\text{GL}(Q)\), and taking the fibre in \(Q\) defines an equivalence between \(\text{coh}(\text{GL}(H),Z)\) and \(\text{mod}(P_Q,\mathcal Z_Q)\), where \(\mathcal Z_Q=\text{Sym}_k(\bigwedge^2Q).\) Define \(\mathcal{M}_Z(\chi)=\widetilde{M_A(\chi)}\in\text{coh}(\text{GL}(H),Z).\) The fundamental result in the article is stating that for \(\mu,\lambda\in B_{r/2,n-r}\): (1) For \(i>0\): \(\text{Ext}^i_Z(\mathcal M_Z(\lambda),\mathcal M_Z(\mu))=0\). (2) There are isomorphisms as \(R\)-modules \(R\Gamma(Z,\mathcal M_Z(\lambda))\simeq M(\lambda).\) (3) Applying \(p_\ast\) gives an isomorphism \(\text{Hom}(\mathcal M_Z(\lambda),\mathcal M_Z(\mu))\cong\text{Hom}(p_\ast\mathcal M_Z(\lambda),p_\ast\mathcal M_Z(\mu))\) \(\cong\text{Hom}(\Gamma(Z,\mathcal M_Z(\lambda)),\Gamma(Z,\mathcal M_Z(\mu)))\cong\text{Hom}(M(\lambda),M(\mu))\) From this fundamental result it follows in particular that \(\mathcal M_Z:=\bigoplus_{\chi\in B_{r/2},\lfloor n/2\rfloor-r/2}\mathcal M_Z\) satisfies \(\text{Ext}^i_Z(\mathcal M_z,\mathcal M_Z)=\begin{cases}\lambda\text{ if }i=0\\0\text{ if }i>0\end{cases},\) so that the embedding \(\mathcal D(\lambda)\hookrightarrow\mathcal D(Z)\) can be constructed explicitly: There is a full exact embedding \(-\otimes^L_{\Lambda}\mathcal M_Z:\mathcal D(\Lambda)\hookrightarrow\mathcal D(Z).\) The article includes a study of some of the properties of the \(\text{Sym}_k(\wedge^2Q)\)-modules \(M_Q(\lambda)\) and gives a \(\text{GL}(Q)\)-equivariant free resolution of these. Also, the present study uses the technique of \textit{a spitting functor}: A functor \(\Phi:\mathcal B\rightarrow\mathcal A\) is right splitting if \(\text{ker}\Phi\) is right admissible in \(\mathcal B\), \(\Phi\) restricted to \((\text{ker}\Phi)^{\perp}\) is fully faithful, and \(\text{im}\Phi=\Phi(\text{ker}\Phi^\perp)\) is right admissible in \(\mathcal A\). The application is that if \(\Phi\) is right splitting then \(\Phi\) and \(\Phi^!\) induce inverse equivalences between \(\text{im}\Phi\subset\mathcal A\) and \(\text{im}\Phi^!\subset\mathcal B\). Here \(\Phi^!\) is right adjoint of \(\Phi\) that is a left splitting functor. This is some of the main ingredients to obtain the main results of the article. The article ends with a section explaining the way of obtaining the results for symmetric matrices. The article is nice and illustrates the importance of the noncommutative theory and derived geometry. It adds new techniques to the study of noncommutative resoulutions, in this particular case of the determinantal varieties.