Blt Azumaya algebras and moduli of maximal orders (Q2290804)

To give a relevant review, we start by recalling some elementary concepts, which is freely referred from [``Unramified division algebras do not always contain Azumaya maximal orders'', Preprint, \url{arXiv.1209.2216}], by \textit{B. Antieau} and \textit{B. Williams}: For a field \(K\) the Artin-Wedderburn Theorem says that every central simple \(K\)-algebra \(A\) is isomorphic to an algebra \(M_n(D)\) of \(n\times n\)- matrices over a finite dimensional central \(K\)-division algebra \(D\). \(M_n(D)\) and \(M_{n'} (D')\) are Brauer-equivalent if \(D\) and \(D'\) are isomorphic over \(K\). The set of Brauer-equivalence classes is a group under tensor product, \(\operatorname{Br}(K),\) the Brauer group of \(K\). The index of an equivalence class \(a = \operatorname{cl}(M_n(D))\in\operatorname{Br}(K)\) is the degree of \(D\). Let \(X\) be a connected scheme. A central simple algebra over \(K\) is generalized to the concept of an Azumaya algebra over \(X\), i.e., an Azumaya algebra \(\mathcal A\) is a locally free sheaf of algebras which étale locally takes the form of a matrix algebra. Thus, there is an étale cover \(\pi :U\rightarrow X\) such that \(\pi^\ast\mathcal A=M_n(\mathcal O_U).\) The degree of A is \(n.\) With this, Brauer equivalence and a contravariant Brauer group functor can be defined. However, over a scheme we don't have Artin-Wedderburn theory and cannot know that a Brauer class \(\alpha\in\operatorname{ Br}(X)\) contains an Azumaya algebra \(\mathcal A\) whose degree divides that of all other Azumaya algebras having class \(\alpha\). The index of \(\alpha\) is defined to be the greatest common divisor of the degrees of Azumaya algebras with Brauer class \(\alpha\). For a regular integral noetherian scheme \(X\) with generic point \(\operatorname{Spec} K\) and a central simple \(K\)-algebra \(A,\) an order in \(A\) over \(X\) is a torsion-free coherent \(\mathcal O_X\)-algebra \(\mathcal A\) , such that \(\mathcal A\otimes_{\mathcal O_X }K\cong A.\) A maximal order in \(A\) over \(X\) is an order which is not a proper subalgebra of any other order in \(A\) over \(X.\) Now we review the present article: The article consider structure theory of maximal orders over algebraic surfaces. There is a satisfying minimal model program, and the moduli of Azumaya orders in a fixed unramified division algebra and related moduli problems are thoroughly studied. This article extends the moduli theory to orders in a ramified Brauer class, and consider phenomena similar to that which occurs in moduli theory of stable projective surfaces, coming from an analogue of Kollár's condition on the compatibility of the reflexive powers of the dualizing sheaf with base change. Working over surfaces means that the global dimension of the orders are \(2\), and simpler than in the general theory of Kollár. The authors arrive at a satisfying moduli space with a natural compactification and a virtual fundamental class. As in the commutative theory, the moduli problem given by fixing the properties of the fibres of a family contains a refined version as a bijective closed substack. The refined moduli problem is described as a moduli problem of Azumaya algebras on stacks rather than orders on varieties, and can also be considered as a moduli theory of parabolic Azumaya algebras. These Azumaya algebras have precise interaction with the ramification divisor coming from the structure of hereditary orders in matrix algebras over discrete valuation rings. This theory was first given by \textit{A. Brumer} [Bull. Am. Math. Soc. 69, 721--724 (1963; Zbl 0113.26002)] who gave the Azumaya algebras under consideration a structure named Brumer log terminal structure, or blt for short. The paper starts out by relating hereditary algebras over complete discrete valuation rings to Azumaya algebras over root construction stacks. This is then globalized, and leads to a simple approach to families of maximal orders. This gives two resulting moduli problems which are compared, using ideas similar to those of Kollár in his theory of hulls and husks, and a local analysis of reflexive Azumaya algebras on families of rational double points. The comparison leads to several nice and important results. The final result is the compactification of the Azumaya problem using algebra-objects of the derived category of a stack which can be seen as parabolic generalized Azumaya algebras. We find room for citing the following definitions, as they are essential for the exposition; verbatim: Definition 2.2. The hereditary site \(\mathcal F\) of \(\operatorname{Spec} R\) is the site whose underlying category consists of faithfully flat quasi-finite étale \(R\)-schemes \(U\rightarrow\operatorname{Spec} R\) with \(U\) of pure dimension \(1,\) with coverings given by collections of \(R\)-maps \(U_i\rightarrow U\) that are jointly surjective. Definition 2.3. Given an object \(U\rightarrow\operatorname{Spec}R\) of \(\mathcal F,\) an Azumaya algebra \(\mathcal A\) on \(\mathcal X_U\) is \(n\)-typed if for each closed point \(u\in U\) the restriction of \(\mathcal A\) to \(\mathcal X\otimes_R\mathcal O_{U,u}\) has type \(m\) for some positive integer \(m\) dividing \(n.\) Definition 2.4. Given an object \(U\rightarrow\operatorname{Spec}R\) of \(\mathcal F\), the stack \(\mathcal A_n\) has as objects over \(U\) the groupoid of \(n\)-typed Azumaya algebras \(\mathcal A\) of degree \(n\) on \(\mathcal X\times_{\operatorname{Spec}R}U.\) The stack \(\mathcal H_n\) has as objects the groupoid of \(n\)-typed hereditary orders on \(U\). Definition 2.5. Suppose \(n\) is invertible in the residue field \(\kappa\) of \(R\). For any object \(\mathcal A\in\mathcal A_n(U),\) the finite \(\mathcal O_U\)-algebra \(\pi_\ast\mathcal A\) lies in \(\mathcal H_n.\) The resulting map of stacks \(\mathcal A_n\rightarrow\mathcal H_n\) is a \(1\)-isomorphism. Definition 2.6. An Azumaya algebra \(\mathcal A\) on \(\mathcal X\) is Brumer log terminal (blt) if for every \(i\) the local Azumaya algebra \(\mathcal A_{\eta_i}\) has type \(e_i\). Under the condition that the reader have the basic knowledge, the article is very well written and easy to follow. The results are important, and comes from comparing different moduli stacks.