Non-commutative resolutions of quotient singularities for reductive groups (Q1675269)

The authors' introduction is a rather complete exposition of the results, the body of the article containing explicit definitions and proofs of the results. So this review is just a compressed version of the introduction and verbatim expressions occur. The main issue of the article is to generalize standard results about non-commutative resolutions of quotient singularities for finite groups to arbitrary reductive groups. This involves the statement of the standard resolution and the proof of their generalization. Throughout, \(k\) is an algebraically closed field of characteristic \(0\), all rings are defined over \(k\), for an algebraic group \(G\), \(\widehat G\) denotes the set of isomorphism classes of irreducible \(G\)-representations. For a normal noetherian domain \(S\), a non-commutative (crepant) resolution (NC(C)R) is an \(S\)-algebra of finite global dimension of the form \(\Lambda=\text{End}_S(M)\) where \(M\) is a non-zero finitely generated reflexive \(S\)-module. It is called crepant if \(S\) is Gorenstein and \(\Lambda\) is a maximal Cohen-Macaulay \(S\)-module. The author remarks that the HPD-dual of a smooth projective variety, as defined by Kuznetsov, often is a NCR of a singular variety. The NCCRs are the best NCRs as they have good homological properties. There is a connection to ordinary (commutative) crepant resolutions in algebaic geometry, and it is often possible to pass from one to the other by to a NCCR assigning a crepant resolution as a GIT moduli space of representations, and conversely a crepant resolution might give a NCCR obtained as \(R=\text{End}_X(T)\) for a tilting bundle \(T\) on a variety \(X\). Also, under mild conditions, a NC(C)R yields a categorical crepant resolution of singularities: If \(S\) is a finitely generated integrally closed \(k\)-algebra with \(\Lambda=\text{End}_S(M)\) a NCR, then \(\Lambda\) is smooth as a DG-algebra. Also, if \(M\) is a projective \(\Lambda\)-module, the functor \(D(S)\rightarrow D(\Lambda):N\mapsto M\overset {L}\otimes_S N\) is fully faithful and yields a categorical resolution of singularities of \(S\). Moreover, if \(\Lambda\) is a NCCR then the categorical resolution is crepant. In the following, a twisted NC(C)R of index \(u\) behaves as a NC(C)R except that it is generically a central simple algebra of index \(u\) rather than a matrix ring. When a finite group \(G\) acts on a smooth affine variety \(X\) and \(U\) is a finite-dimensional \(G\)-representation, then \(M(U)=(U\otimes k[X])^G\) is called the corresponding \(k[X]^G\)-module of covariants. The abelian category \(\text{mod}(G,k[X])\) of \(G\)-equivariant finitely generated \(k[X]\)-modules has a projective generator \(U\otimes k[X]\) with \(U=\oplus_{V\in\widehat G}V\), and it follows that \(\text{mod}(G,k[X])\) is equivalent to \(\text{mod}(\Lambda)\) where \(\Lambda=\text{End}_{G,k[X]}(U\otimes k[X])=(\text{End}(U)\otimes k[X])^G=M(\text{End}_k(U))\) so that \(\Lambda\) is a \(k[X]^G\) algebra of covariants. Because \(\text{mod}(G,k[X])\) has finite global dimension, \(\text{gl\; dim} \Lambda<\infty\) and so \(\Lambda\) is a non-commutative resolution of \(k[X]^G=k[X\slash G]\) in a weak sense. If no element of \(G\) fixes a divisor then \(k[X]/k[X]^G\) is étale in codimension one so that \(\Lambda=\text{End}_{k[X]^G}(M(U))\), and then \(\Lambda\) is NCR of \(k[X]^G\). For \(V\) to be a NCCR one also needs \(k[X]^G\) to be Gorenstein, and that is obtained when \(X=\text{Spec}SW\) for a representation \(W\) and \(G\subset\text{SL}(W)\). For general reductive groups, especially when \(G\) is not finite, the generalization is more involved. However, this is handled by the authors. The main body of the article looks for non-commutative resolutions of quotient singularities given by algebras of covariants. The first main theorem states that when \(G\) is a reductive group acting on a smooth affine variety \(X\), there exists a finite-dimensional \(G\)-representation \(U\) containing the trivial representation such that \(\Lambda=M(\text{End}(U)\otimes k[X])^G\) have \(\text{gl\;dim}<\infty\). One has that \(\Lambda=M(\text{End}(U))\) is a \(k[X]^G\)-algebra, finitely generated as \(k[X]^G\)-module and with \(k[X]^G\subset Z(\Lambda)\) so that \(\Lambda\) is a weak non-commutative resolution of \(k[X]^G=k[X\slash G]\). As \(U\) contains a trivial direct summand, \(M(U)\) is a projective left \(\Lambda\)-module, one has that \(\text{End}_\Lambda(M(U))=k[X]^G\), and so one gets that if \(G\) is a reductive group acting on a smooth affine variety \(X\), \(k[X]^G\) has a categorical resolution of singularities given by an algebra of covariants. Correspondingly, when \(X\) is projective, the authors prove a GIT-version of the main result, that is, for given \(G,\; X,\;\mathcal L\), there exists a finite-dimensional \(G\)-representation \(U\) containing the trivial representation such that the coherent sheaf of algebras \(\Lambda=M^{ss}(\text{End}(U))\) on \(X^{ss}\slash G\) has finite global dimension when restricted to affine opens. A finite group \(G\) is said to be acting generically on a smooth affine variety \(X\) if \(X\) contains a point with closed orbit and stabilizer, and if \(X^s\subset X\) is the locus of points with that property, then \(\text{codim}(X-X^\ast)\geq 2\). Then a \(G\)-representation \(W\) is generic if \(G\) acts generically on \(\text{Spec}SW\cong W^\ast\). In this case, if \(G\) acts generically on \(X\), there exists a finite-dimensional \(G\)-representation \(U\) containing the trivial representation such that \(\Lambda=\text{End}_{k[X]^G}(M(U))\) is a NCR for \(k[X]^G\). Applying the corresponding techniques to abelian reductive groups, the authors prove that for \(S\subset\mathbb Z^n\) a finitely generated commutative positive normal semigroup, for \(n\in\mathbb N,\;n\gg 0\), the \(k[S]\)-module \(M=k[\frac{1}{n}S]\) defines a NCR for for \(k[S]\). This confirms that in characteristic \(p\) Frobenius twists provide canonical NCRs. After the development of this theory, with the deep results indicated above, it is applied to some specific quotient singularities. The first is the determinantal varieties and the result that the variety \(Y_{n,h}\), \(n<h\), of \(h\times h\)-matrices of rank \(\leq n\), \(k[Y_{n,h}]\) has a NCCR. Considering Pfaffian varieties, letting \(Y_{2n,h}^-\), \(2n<h\), be the variety of skew-symmetric \(h\times h\)-matrices of rank \(\leq 2n\), then for odd \(h\), \(k[Y_{2n,h}^-]\) has a NCCR. For determinantal varieties for symmetric varieties, for \(t<h\), let \(Y^+_{t,h}\) be the variety of symmetric \(h\times h\)-matrices of tank \(\leq t\). If \(t\) and \(h\) have opposite parity, then \(k[Y^+_{t,h}]\) has a NCCR. If the parity is equal, then \(k[Y^+_{t,h}]\) has a twisted NCCR of index \(2^{\lfloor h/2\rfloor}\). Finally, the results for non-commutative resolutions for \(\text{SL}_2\)-invariants are stated, and trace rings are proven to be modules of covariants in a series of cases. The introduction, which is rather exhaustive, ends by stating a number of sufficient conditions for the existences of NC(C)Rs, possibly twisted, for quotient singularities. Thus the necessary machinery is included and this article is a complete introduction to noncommuative resolutions, with a lot of nice examples.