Moduli spaces of principal bundles on singular varieties (Q1950276)

Let \((X,{\mathcal O}_X(1))\) be a polarized projective variety over an algebraically closed field \(k\) of characteristic 0, \(\rho : G \rightarrow \text{SL}(V) \subset \text{GL}(V)\) a fixed faithful representation of a reductive group \(G\), and \(P\) a fixed Hilbert polynomial. A \textit{pseudo G-bundle} is a pair \(({\mathcal A},\tau)\), where \(\mathcal A\) is a torsion-free \({\mathcal O}_X\)-module of rank \(=\, \dim V\), and \(\tau\) is a nontrivial morphism of \({\mathcal O}_X\)-algebras \(: {\mathcal S}ym^\bullet({\mathcal A}\otimes V)^G \rightarrow {\mathcal O}_X\). Giving \(\tau\) is equivalent to giving a section \(\sigma : X \rightarrow {\mathbb H}\text{om}({\mathcal A}, V^\vee \times {\mathcal O}_X)/G\). Let \(U_{\mathcal A}\) be the open subset of \(X\) on which \(\mathcal A\) is locally free. \(({\mathcal A},\tau)\) is a \textit{singular principal G-bundle} if there exists a nonempty open subset \(U \subseteq U_{\mathcal A}\) such that \(\sigma(U) \subset {\mathbb I}\text{som}(V\otimes {\mathcal O}_U, {\mathcal A}^\vee \, | \, U)/G\). If \(U = U_{\mathcal A}\) one says that \(({\mathcal A},\tau)\) is a \textit{honest singular principal G-bundle}. The main result of the paper under review asserts that if \(f : {\mathcal X} \rightarrow S\) is a flat, projective morphism of \(k\)-schemes of finite type, with integral geometric fibers, and if \({\mathcal O}_{\mathcal X}(1)\) is an \(f\)-very ample line bundle on \(\mathcal X\) then there exists a projective moduli space \(M^\rho_{{\mathcal X}/S,P} \rightarrow S\) for flat families of \textit{semistable} singular principal \(G\)-bundles on \({\mathcal X}/S\) such that \({\mathcal A}_s\) has Hilbert polynomial \(P\), \(\forall \, s \in S\). Moreover, if the fibers of \(f\) are Gorenstein and there exists a \(G\)-invariant nondegenerate quadratic form on \(V\), then \(M^\rho_{{\mathcal X}/S,P} \rightarrow S\) parametrizes only honest singular principal \(G\)-bundles. This result refines earlier results of \textit{U. N. Bhosle} [Int. Math. Res. Not. 2004, No. 57, 3057--3077 (2004; Zbl 1106.14027)] and of \textit{A. Schmitt} [Int. Math. Res. Not. 2002, No. 23, 1183--1209 (2002; Zbl 1034.14017); Int. Math. Res. Not. 2004, No. 62, 3327--3366 (2004; Zbl 1093.14507); Int. Math. Res. Not. 2005, No. 23, 1427--1437 (2005; Zbl 1084.14034)]. Actually, the author of the paper under review improves some of the steps in the proofs of those results. Firstly, he shows that a certain technical condition on \(X\), introduced by Bhosle and verified by her only when \(X\) is seminormal or an \(S_2\)-variety, holds for an arbitrary variety \(X\). Then the author proves that there exists an explicit constant \(\gamma\), depending only on \((X,{\mathcal O}_X(1))\), such that for any two torsion-free sheaves \(E_1\) and \(E_2\) on \(X\), of ranks \(r_1\), \(r_2\), one has: \[ \mu_{\text{max}}(E_1\otimes E_2/\text{Tors}) \leq \mu_{\text{max}}(E_1) + \mu_{\text{max}}(E_2) + (r_1 + r_2)\gamma \, . \] This is the ``technical heart of the paper''. Finally, he shows that if \(X\) is Gorenstein and there exists a \(G\)-invariant nondegenerate quadratic form \(\varphi\) on \(V\) then every degree 0 singular principal bundle on \(X\) is an honest singular principal bundle. This generalizes a result of Schmitt from his last mentioned paper.