Morita theory for non-commutative noetherian schemes
Extracted source text · arXiv 1911.01242
\documentclass[11pt]{amsart}
\usepackage{amscd,amsxtra,amssymb,mathrsfs, bbm}
%\usepackage[romanian]{babel}
\usepackage{combelow}
\usepackage[new]{old-arrows}
\usepackage{stmaryrd}
%\usepackage[paper=a4paper,left=30mm,right=20mm,top=25mm,bottom=30mm]{geometry}
\usepackage{geometry}
\newenvironment{dedication}
%{\vspace{1ex}\begin{quotation}\begin{center}\begin{em}}
{\begin{quotation}\begin{center}\begin{em}}
{\par\end{em}\end{center}\end{quotation}}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{conjecture}[theorem]{Conjecture}
\newtheorem{question}[theorem]{Question}
\newtheorem{problem}[theorem]{Problem}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{remark}[theorem]{Remark}
\newtheorem{example}[theorem]{Example}
\newtheorem{case}{Case}
%\numberwithin{equation}{section}
\DeclareMathOperator{\Dbcoh}{\mathsf{D^{b}_{coh}}}
\DeclareMathOperator{\Perf}{\mathsf{Perf}}
\DeclareMathOperator{\Hot}{\mathsf{Hot}}
\DeclareMathOperator{\add}{\mathsf{add}}
\DeclareMathOperator{\ann}{\mathsf{Ann}}
\DeclareMathOperator{\ad}{\mathsf{ad}}
\DeclareMathOperator{\wt}{\mathsf{wt}}
\DeclareMathOperator{\rad}{\mathsf{rad}}
\DeclareMathOperator{\FM}{\mathsf{FM}}
\DeclareMathOperator{\SL}{\mathsf{SL}}
\newcommand{\MPnode}{\mathsf{MP}_\mathrm{nd}}
\newcommand{\MPcusp}{\mathsf{MP}_\mathrm{cp}}
\DeclareMathOperator{\rk}{\mathrm{rk}}
\DeclareMathOperator{\cn}{\mathrm{cn}}
\DeclareMathOperator{\sn}{\mathrm{sn}}
\DeclareMathOperator{\dn}{\mathrm{dn}}
\DeclareMathOperator{\coker}{\mathrm{coker}}
\renewcommand{\ker}{\mathsf{Ker}}
\newcommand{\im}{\mathrm{im}}
\newcommand{\conj}{\mathsf{ad}}
\newcommand{\can}{\mathsf{can}}
\newcommand{\pr}{\mathsf{pr}}
\newcommand{\id}{\mathrm{id}}
\newcommand{\Tr}{\mathsf{Tr}}
\newcommand{\tr}{\mathsf{tr}}
\newcommand{\rank}{\mathrm{rank}}
\DeclareMathOperator{\cl}{\mathrm{cl}}
\DeclareMathOperator{\res}{\mathsf{res}}
\DeclareMathOperator{\ev}{\mathsf{ev}}
\DeclareMathOperator{\Sol}{\mathsf{Sol}}
\DeclareMathOperator{\tors}{\mathsf{tors}}
\DeclareMathOperator{\Coh}{\mathsf{Coh}}
\DeclareMathOperator{\Qcoh}{\mathsf{QCoh}}
\DeclareMathOperator{\Tor}{\mathsf{Tor}}
\DeclareMathOperator{\TF}{\mathsf{TF}}
\DeclareMathOperator{\Latt}{\mathsf{Latt}}
\DeclareMathOperator{\BM}{\mathsf{BM}}
\DeclareMathOperator{\Hot+}{\mathsf{Hot^{+,\,b}_{coh}}}
\DeclareMathOperator{\Hotb}{\mathsf{Hot^{b}_{coh}}}
\DeclareMathOperator{\Comb}{\mathsf{Com^{b}_{coh}}}
\DeclareMathOperator{\Com+}{\mathsf{Com^{+,\,b}_{coh}}}
\DeclareMathOperator{\Inj}{\mathsf{Inj}}
\DeclareMathOperator{\VB}{\mathsf{VB}}
\DeclareMathOperator{\St}{\mathsf{St}}
\DeclareMathOperator{\Spl}{\mathsf{Spl}}
\DeclareMathOperator{\Tri}{\mathsf{Tri}}
\DeclareMathOperator{\Bim}{\mathsf{Bim}}
\DeclareMathOperator{\AF}{\mathsf{AF}}
\DeclareMathOperator{\Pic}{\mathsf{Pic}}
\DeclareMathOperator{\Jac}{\mathsf{Jac}}
\DeclareMathOperator{\Mf}{\mathsf{M}}
\DeclareMathOperator{\fM}{\mathsf{M}}
\DeclareMathOperator{\Hilb}{\mathsf{Hilb}}
\DeclareMathOperator{\Hom}{\mathsf{Hom}}
\DeclareMathOperator{\Nat}{\mathsf{Nat}}
\DeclareMathOperator{\RHom}{\mathsf{RHom}}
\DeclareMathOperator{\Mor}{\mathsf{Mor}}
\DeclareMathOperator{\Ext}{\mathsf{Ext}}
\DeclareMathOperator{\Lin}{\mathsf{Lin}}
\DeclareMathOperator{\GL}{\mathsf{GL}}
\DeclareMathOperator{\Aut}{\mathsf{Aut}}
\DeclareMathOperator{\End}{\mathsf{End}}
\DeclareMathOperator{\Mat}{\mathsf{Mat}}
\DeclareMathOperator{\Quot}{\mathsf{Quot}}
\DeclareMathOperator{\Spec}{\mathsf{Spec}}
\DeclareMathOperator{\Sp}{\mathsf{Sp}}
\DeclareMathOperator{\Max}{\mathsf{Max}}
\DeclareMathOperator{\Ans}{\mathsf{Ans}}
\DeclareMathOperator{\Sets}{\mathsf{Sets}}
\DeclareMathOperator{\For}{\mathsf{For}}
\DeclareMathOperator{\Ob}{\mathsf{Ob}}
\DeclareMathOperator{\Rep}{\mathsf{Rep}}
\DeclareMathOperator{\Mod}{\mathsf{Mod}}
\DeclareMathOperator{\slie}{\mathfrak{sl}}
\DeclareMathOperator{\pgl}{\mathfrak{pgl}}
\DeclareMathOperator{\lieg}{\mathfrak{g}}
\DeclareMathOperator{\lieh}{\mathfrak{h}}
\DeclareMathOperator{\lien}{\mathfrak{n}}
\DeclareMathOperator{\liep}{\mathfrak{p}}
\DeclareMathOperator{\liew}{\mathfrak{w}}
\DeclareMathOperator{\liel}{\mathfrak{l}}
\DeclareMathOperator{\lieb}{\mathfrak{b}}
\DeclareMathOperator{\liea}{\mathfrak{a}}
\DeclareMathOperator{\lieD}{\mathfrak{D}}
\DeclareMathOperator{\lieP}{\mathfrak{P}}
\DeclareMathOperator{\lieW}{\mathfrak{W}}
\DeclareMathOperator{\lieO}{\mathfrak{O}}
\DeclareMathOperator{\catD}{\mathfrak{D}}
\DeclareMathOperator{\catE}{\mathfrak{E}}
\DeclareMathOperator{\catF}{\mathfrak{F}}
\DeclareMathOperator{\catA}{\mathfrak{A}}
\DeclareMathOperator{\catB}{\mathfrak{B}}
\DeclareMathOperator{\catC}{\mathfrak{C}}
\DeclareMathOperator{\dX}{\mathfrak{X}}
\DeclareMathOperator{\ddX}{X}
\DeclareMathOperator{\dY}{\mathfrak{Y}}
\DeclareMathOperator{\dF}{F}
\DeclareMathOperator{\dE}{E}
\DeclareMathOperator{\el}{\mathsf{El}}
%%%%%%%%%Part3
\DeclareMathOperator{\idm}{\mathfrak{m}}
\DeclareMathOperator{\idp}{\mathfrak{p}}
\DeclareMathOperator{\idq}{\mathfrak{q}}
\DeclareMathOperator{\liev}{\mathfrak{v}}
\DeclareMathOperator{\lieu}{\mathfrak{u}}
\DeclareMathOperator{\lies}{\mathfrak{s}}
\DeclareMathOperator{\Vect}{\mathsf{Vect}}
\DeclareMathOperator{\Stab}{\mathsf{Stab}}
\newcommand{\dtens}{\stackrel{\boldsymbol{L}}{\otimes}} %derived tensor product
\newcommand{\ARG}{\,\cdot\,}
\newcommand{\bul}{\scriptstyle{\bullet}}
\def\bop{\bigoplus}
\def\str{\stackrel}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\setlength{\oddsidemargin}{-1in} % Linker Rand 30mm vom Papierrand
\addtolength{\oddsidemargin}{30mm}
\setlength{\evensidemargin}{\oddsidemargin}
\setlength{\textwidth}{150mm}
\newcommand{\kk}{\mathbbm{k}}
\newcommand{\CC}{\mathbb{C}}
\newcommand{\NN}{\mathbb{N}}
\newcommand{\ZZ}{\mathbb{Z}}
\newcommand{\LL}{\mathbb{L}}
\newcommand{\WW}{\mathbb{W}}
\newcommand{\fR}{\mathsf{R}}
\newcommand{\bbX}{\mathbbm{X}}
\newcommand{\bbY}{\mathbbm{Y}}
\newcommand{\bbU}{\mathbbm{U}}
\newcommand{\bbV}{\mathbbm{V}}
\newcommand{\bbW}{\mathbbm{W}}
\newcommand{\bbE}{\mathbbm{E}}
\newcommand{\sm}{\mathsf{m}}
%\newcommand{\catD}{\mathsf{D}}
%\newcommand{\catA}{\mathsf{A}}
%\newcommand{\catB}{\mathsf{B}}
%\newcommand{\catI}{\mathsf{E}}
\newcommand{\kEnd}{\mbox{\textit{End}}}
\newcommand{\kHom}{\mbox{\textit{Hom}}}
\newcommand{\kExt}{\mbox{\textit{Ext}}}
\newcommand{\kMat}{\mbox{\textit{Mat}}}
\newcommand{\BB}{\mathbb{B}}
\newcommand{\FF}{\mathbb{F}}
\newcommand{\GG}{\mathbb{G}}
\newcommand{\EE}{\mathbb{E}}
\newcommand{\XX}{\mathbb{X}}
\newcommand{\YY}{\mathbb{Y}}
\renewcommand{\SS}{\mathbb{S}}
\newcommand{\TT}{\mathbb{T}}
\newcommand{\HH}{\mathsf{H}}
\newcommand{\MM}{\mathsf{M}}
\newcommand{\DD}{\mathbb{D}}
\newcommand{\UU}{\mathbb{U}}
\renewcommand{\HH}{\mathbb{H}}
\newcommand{\PP}{\mathbbm{P}}
\newcommand{\VV}{\mathbb{V}}
\newcommand{\sfK}{\mathsf{K}}
\newcommand{\Ad}{\mathsf{Ad}}
\newcommand{\sMat}{\mbox{\emph{Mat}}}
\newcommand{\sHom}{\mbox{\it{Hom}}}
\newcommand{\sExt}{\mbox{\it{Ext}}}
\newcommand{\kA}{\mathcal{A}}
\newcommand{\ksA}{\tiny{\mathcal A}}
\newcommand{\kB}{\mathcal{B}}
\newcommand{\kC}{\mathcal{C}}
\newcommand{\kE}{\mathcal{E}}
\newcommand{\kF}{\mathcal{F}}
\newcommand{\kG}{\mathcal{G}}
\newcommand{\kH}{\mathcal{H}}
\newcommand{\kI}{\mathcal{I}}
\newcommand{\kJ}{\mathcal{J}}
\newcommand{\kO}{\mathcal{O}}
\newcommand{\kL}{\mathcal{L}}
\newcommand{\kP}{\mathcal{P}}
\newcommand{\kQ}{\mathcal{Q}}
\newcommand{\kR}{\mathcal{R}}
\newcommand{\kK}{\mathcal{K}}
\newcommand{\kM}{\mathcal{M}}
\newcommand{\kN}{\mathcal{N}}
\newcommand{\kV}{\mathcal{V}}
\newcommand{\kW}{\mathcal{W}}
\newcommand{\kT}{\mathcal{T}}
\newcommand{\kS}{\mathcal{S}}
\newcommand{\kX}{\mathcal{X}}
\newcommand{\kY}{\mathcal{Y}}
\newcommand{\kZ}{\mathcal{Z}}
\newcommand{\gS}{\mathfrak{S}}
\newcommand{\tX}{\widetilde{X}}
\newcommand{\tZ}{\widetilde{Z}}
\newcommand{\tx}{\tilde{x}}
\newcommand{\tH}{\widetilde{\mathcal{H}}}
\newcommand{\bflambda}{\boldsymbol\lambda}
\def\bfp{\mathbf p}
%\newcommand{\bul}{\scriptstyle{\bullet}}
\newcommand{\lar}{\longrightarrow}
\newcommand{\llbrace}{(\!(}
\newcommand{\rrbrace}{)\!)}
\def\lha{\leftharpoondown}
\def\rha{\rightharpoondown\phantom{I\!}\!}
%%%%%%%%%%%%%%%%%%%%%%%% BIG SF LETTERS %%%%%%%%%%%%%%%%%%%%
\def\sA{\mathsf A} \def\sN{\mathsf N}
\def\sB{\mathsf B} \def\sO{\mathsf O}
\def\sC{\mathsf C} \def\sP{\mathsf P}
\def\sD{\mathsf D} \def\sQ{\mathsf Q}
\def\sE{\mathsf E} \def\sR{\mathsf R}
\def\sF{\mathsf F} \def\sS{\mathsf S}
\def\sG{\mathsf G} \def\sT{\mathsf T}
\def\sH{\mathsf H} \def\sU{\mathsf U}
\def\sI{\mathsf I} \def\sV{\mathsf V}
\def\sJ{\mathsf J} \def\sW{\mathsf W}
\def\sK{\mathsf K} \def\sX{\mathsf X}
\def\sL{\mathsf L} \def\sY{\mathsf Y}
\def\sM{\mathsf M} \def\sZ{\mathsf Z}
\def\sDF{\mathsf DF}
\def\bu{{\scriptscriptstyle\bullet}}
\usepackage{tikz}
\usetikzlibrary{calc, arrows, positioning, shapes, fit, matrix, decorations}
\usetikzlibrary{decorations.shapes, decorations.pathreplacing}
\newcommand\dhrightarrow{%
\mathrel{\ooalign{$\rightarrow$\cr%
$\mkern3.5mu\rightarrow$}}
}
\input xy
\xyoption{all}
\title[Morita theory for non--commutative schemes]{Morita theory for non--commutative noetherian schemes}
\author{Igor Burban}
\address{
Universit\"at Paderborn\\
Institut f\"ur Mathematik \\
Warburger Stra\ss{}e 100 \\
33098 Paderborn \\
Germany
}
\email{burban@math.uni-paderborn.de}
\author{Yuriy Drozd}
\address{
Institute of Mathematics\\
National Academy of Sciences of Ukraine,
Tereschenkivska str. 3,
01004 Kyiv, Ukraine}
\email{drozd@imath.kiev.ua, y.a.drozd@gmail.com}
%\urladdr{www.imath.kiev.ua/$\sim$drozd}
%\subjclass[2010]{Primary 16E35, 16G60, 14A22, 16S38}
\begin{document}
\begin{abstract}
In this paper, we study equivalences between the categories of quasi--coherent sheaves on non--commutative noetherian schemes. In particular, give a new proof of C\u{a}ld\u{a}raru's conjecture about Morita equivalences of Azumaya algebras on noetherian schemes. Moreover, we derive necessary and sufficient condition for two reduced non--commutative curves to be Morita equivalent.
\end{abstract}
\maketitle
\section{Introduction}
\noindent
A classical results of Gabriel (see
\cite[Section VI.3]{Gabriel}) states that the categories of quasi--coherent sheaves $\Qcoh(X)$ and $\Qcoh(Y)$ of two separated noetherian schemes $X$ and $Y$ are equivalent if and only if $X$ and $Y$ are isomorphic. To prove this result (and in particular to show how the scheme $X$ can be reconstructed from the category $\Qcoh(X)$), Gabriel used the full power of methods of homological algebra, developed in his thesis \cite{Gabriel}.
\medskip
\noindent
In this work, we deal with similar types of questions for the so--called non--commutative noetherian schemes. By definition, these are ringed spaces $\bbX = (X, \kA)$, where $X$ is a separated noetherian scheme and $\kA$ is a sheaf of $\kO_X$--algebras, which is coherent viewed as an $\kO_X$--module. A basic question arising in this context is to establish when the categories of quasi--coherent sheaves
$\Qcoh(\bbX)$ and $\Qcoh(\bbY)$ on two such non--commutative noetherian schemes $\bbX$ and $\bbY$ are equivalent.
\medskip
\noindent
We show first that from the categorical perspective, $\bbX$ and $\bbY$ can without loss of generality assumed to be central; see Subsection \ref{SS:CentralizingNCNS} for details. Following Gabriel's approach \cite{Gabriel}, based on a detailed study of indecomposable injective objects of $\Qcoh(\bbX)$, we prove that the central scheme $X$ can be recovered from the category $\Qcoh(\bbX)$; see Theorem \ref{T:MorphismCentralSchemes}. Using this reconstruction result, we prove Morita theorem in the setting of central non--commutative noetherian schemes; see Theorem \ref{T:MoritaTheoremNCNS} and the discussion afterwards.
\medskip
\noindent
As a first application of this result, we get a new proof of C\u{a}ld\u{a}raru's conjecture about Azumaya algebras on noetherian schemes; see \cite[Conjecture 1.3.17]{CaldararuCrelle}. Namely, we show that
if $\bbX = (X, \kA)$ and $\bbY = (Y, \kB)$ are two non--commutative noetherian schemes, such that $\kA$ and $\kB$ are Azumaya algebras on $X$ and $Y$ respectively, then $\Qcoh(\bbX)$ and $\Qcoh(\bbY)$ are equivalent if and only if
there exists an isomorphism $Y \stackrel{f}\lar X$ such that
$f^\ast\bigl([\kA]\bigr) = [\kB] \in \mathsf{Br(Y)}$, where $\mathsf{Br(Y)}$ is the Brauer group of the scheme $Y$. This result was already proven by Antieau
\cite{Antieau} (see also \cite{Perego} and \cite{CanonacoStellari}) by much more complicated methods.
\medskip
\noindent
Our main motivation to develop Morita theory in the setting of non--commutative algebraic geometry comes from the study of reduced non--commutative curves. By definition, these are central non--commutative noetherian schemes $\bbX = (X, \kA)$, for which $X$ is a reduced excellent noetherian scheme of pure dimension one and $\kA$ is a sheaf of $\kO_X$--orders. Our goal was to derive a handleable criterion
to describe the Morita equivalence class of $\bbX$.
\medskip
\noindent
From the historical perspective, the so--called projective hereditary non--commutative curves, i.e.~those $\bbX = (X, \kA)$, for which $X$ is a projective curve over some field $\kk$ and $\kA$ is a sheaf of hereditary orders, were originally of major interest. For $\bbX = \mathbbm{P}^1$, they appeared (in a different form) in the seminal work of Geigle and Lenzing \cite{GeigleLenzing} on weighted projective lines. Tilting theory on these curves had a significant impact on the development of the representation theory of finite dimensional $\kk$--algebras.
For an algebraically closed field $\kk$, projective hereditary non--commutative curves play a central role in the classification of abelian noetherian $\kk$--linear $\Ext$--finite hereditary categories with Serre duality due to Reiten and van den Bergh \cite{ReitenvandenBergh} (see also
\cite{LenzingReiten, Kussin} for the case of arbitrary fields). In the case of a finite field $\kk$, such non--commutative curves appeared as a key technical tool in the work of Laumon, Rapoport and Stuhler \cite{LaumonRapoportStuhler} in the framework of the Langlands programme. The question of a classification of non--commutative hereditary curves up to Morita equivalence was clarified by Spie\ss{} in \cite{Spiess}. In this case, the Morita equivalence class of such a curve $\bbX$ (however, not $\bbX$ itself, viewed as a ringed space!) is determined by a central simple algebra $\Lambda_{\bbX}$ (which is an analogue of the function field of a commutative curve) and the types of non--regular points of $\bbX$; see Corollary \ref{C:HereditaryMorita} for details.
\medskip
\noindent
However, the case of non--hereditary orders happened to be more tricky. It turns out that even the central curve $X$, the class of the algebra $\Lambda_{\bbX}$ in the Brauer group of the function field of $X$ and the isomorphism classes of non--regular points of $\bbX$ are not sufficient to recover $\bbX$ (up to Morita equivalence); see Example \ref{E:ExampleNonHered}. In Theorem \ref{T:MoritaNonCommCurves}, we give necessary and sufficient conditions for two reduced non--commutative curves to be Morita equivalent.
\medskip
\noindent
Non--hereditary reduced non--commutative projective curves naturally arise as categorical resolutions of singularities of usual singular reduced commutative curves; see
\cite{BurbanDrozdGavranIMRN}. From the point of view of representation theory of finite dimensional $\kk$--algebras,
the so--called tame non--commutative projective nodal curves seem to be of particular importance; see
\cite{bd, bdnpdalcurves}. Special classes of such curves appeared in the framework of the homological mirror symmetry (in a different language and under the name stacky chains/cycles of projective lines) in a work of Lekili and Polishchuk \cite{LekiliPolishchuk}
as holomorphic mirrors of compact oriented surfaces with non--empty boundary; see also \cite[Section 7]{bdnpdalcurves}. Getting a precise description of Morita equivalence classes of tame non--commutative nodal curves (see \cite[Section 4]{bdnpdalcurves}) was another motivation to carry out this work.
\smallskip
\noindent
\emph{Acknowledgement}. The work of the first--named author was partially supported by the DFG project Bu--1866/4--1. The results of this paper
were mainly obtained during the stay of the second--named author at the University of Paderborn in September 2018 and September 2019.
\section{Classical Morita theory and the categorical center}
\subsection{Notation for module theory and reminder of the classical Morita theorem}
\smallskip
\noindent
For any ring $A$, we denote by $A^\circ$ the opposite ring, by $Z(A)$ the center of $A$ and
by $A-\Mod$ (respectively,
$\Mod-A$) the category of all left (respectively, right) $A$--modules.
\smallskip
\noindent
For a commutative ring $R$, an $R$--algebra is a pair $(A, \imath)$, where
$A$ is a ring and $R \stackrel{\imath}\lar A$ an injective homomorphism such that $\imath(R) \subseteq Z(A)$. If $A$ is a finitely generated $R$--module then one says that $A$ is a \emph{finite} $R$--algebra.
Next, $(A, \imath)$ is a \emph{central} $R$--algebra if $\imath(R) = Z(A)$. Usually, $R$ will be viewed as a subset of $A$; in this case, the canonical inclusion map $\imath$ will be suppressed from the notation. We denote by $A^e := A \otimes_R A^\circ$ the enveloping $R$--algebra of $A$ and identify the category of $(A-A)$--bimodules with the category $A^e-\mathsf{Mod}$. The following result is well--known:
\begin{lemma}\label{L:CenterofRing}
If $A$ is an $R$--algebra, then the canonical map $Z(A) \lar \End_{A^e}(A)$ is an isomorphism.
Hence, if $R$ is noetherian and $A$ is a finite $R$--algebra, then
\begin{itemize}
\item for any multiplicative subset $\Sigma \subset R$ we have:
$\Sigma^{-1}\bigl(Z(A)\bigr) \cong Z\bigl(\Sigma^{-1} A\bigr)$;
\item for any $\idm \in \Max(R)$ we have: $\widehat{Z(A)}_{\idm} \cong
Z\bigl(\widehat{A}_{\idm}\bigr)$.
\end{itemize}
\end{lemma}
\smallskip
\noindent
Let $A, B$ be any rings and $P = {}_{B}P_{A}$ be a $(B-A)$--bimodule. Recall that $P$ is called \emph{balanced}, if both structure maps
$$
B \stackrel{\lambda^P}\lar \End_A(P_A), \, b \mapsto \lambda_b^P \;\; \mbox{\rm and} \; \;
A^\circ \stackrel{\rho^P}\lar \End_B({}_{B}P), \, a \mapsto \rho_a^P
$$
are ring isomorphisms, where $\lambda_b^P(x) = b x$ and $\rho_a^P(x) =
x a$ for any $x \in P$, $a \in A$, $b \in B$.
\smallskip
\noindent
For an additive category $\sC$ and $X \in \Ob(\sC)$, we denote by $\add(X)$ the full subcategory of $\sC$, whose objects are direct summands of finite coproducts of $X$.
\smallskip
\noindent
Let $A$ be any ring and $P$ be a finitely generated right $A$--module. Then $P$ is a \emph{progenerator} of $\Mod-A$ (or just right \emph{$A$--progenerator}) if
$\add(P) = \add(A)$. In this case, for any
$M \in \Mod-A$ there exists a set $I$ and an epimorphism $P^{\oplus (I)} \lar M$. Other characterizations of right progenerators can be for instance found in \cite[Section 18B]{Lam}.
\smallskip
\noindent
Note that for any $(B-A)$--bimodules ${}_{B} P_{A}$ and ${}_{B} Q_{A}$, the canonical map
\begin{equation}\label{E:MoritaYoneda}
\Hom_{B-A}\bigl({}_{B} P_{A}, {}_{B} Q_{A}\bigr) \lar
\Hom\bigl({}_{B} P_{A} \otimes_A -, {}_{B} Q_{A} \otimes_A -\bigr)
\end{equation}
is an isomorphism, where $\Hom$ in the right hand side of (\ref{E:MoritaYoneda}) denotes the abelian group of natural transformations between the corresponding additive functors.
\begin{theorem}[Morita theorem for rings]\label{T:MoritaClassical}
Let $A, B$ be any rings and $$A-\Mod \stackrel{\Phi}\lar B-\Mod$$ be an equivalence of categories. Then we have: $\Phi \cong {}_{B} P_{A} \otimes_{A} -$, where $P$ is a balanced $(B-A)$--bimodule, which is a right progenerator of $A$ (in what follows, such bimodule will be called $(B-A)$--Morita bimodule). Moreover, if ${}_{B} Q_{A}$ is another $(B-A)$--Morita bimodule representing $\Phi$ then $P$ and
$Q$ are canonically isomorphic as bimodules.
\end{theorem}
\smallskip
\noindent
A proof of this standard result can be for instance found in \cite[Chapter 18]{Lam}. \qed
\medskip
\noindent
The goal of this work is to generalize Theorem \ref{T:MoritaClassical} to various settings of non--commutative noetherian schemes.
\subsection{Non--commutative noetherian schemes}
\begin{definition}\label{D:NCNS}
A non--commutative noetherian scheme (abbreviated as \emph{ncns}) is a ringed space $\bbX = (X, \kA)$,
where $X$ is a commutative \emph{separated noetherian} scheme and $\kA$ is a sheaf of $\kO$--algebras \emph{coherent} as $\kO$--module (here, $\kO = \kO_X$ denotes the structure sheaf of $X$).
We say that $\XX$ is \emph{central} if $O_x = Z(A_x)$ for any $x \in X$, where $O_x$ (respectively, $A_x$) is the stalk of $\kO$ (respectively, $\kA$) at the point $x$.
\end{definition}
\smallskip
\noindent
For a ncns $\bbX$, we shall denote by $\Qcoh(\bbX)$ the category of quasi--coherent sheaves on $\bbX$, i.e. the category of sheaves of left $\kA$--modules which are quasi--coherent as sheaves of $\kO$--modules.
For an open subset $U \subseteq X$ and $\kF \in \Qcoh(X)$, we shall use both notations
$\Gamma(U, \kF)$ and $\kF(U)$ for the corresponding group of local sections and write
$O(U) = \kO(U)$ and $A(U) = \kA(U)$. Note that $A(U)$ is a finite $O(U)$--algebra. Moreover, for any pair of open affine subsets $V \subseteq U \subseteq X$, the canonical map $O(V) \otimes_{O(U)} A(U) \rightarrow A(V)$ is an isomorphism of
$O(V)$--algebras. Similarly, for any $\kF \in \Qcoh(\bbX)$, the canonical map
\begin{equation}
O(V) \otimes_{O(U)} \Gamma(U, \kF) \lar \Gamma(V, \kF)
\end{equation}
is an isomorphism of $A(V)$--modules.
\smallskip
\noindent
For any open subset $U \subseteq X$, we get a ncns $\bbU := (U, \kA\big|_{U})$. Since $X$ is assumed to be noetherian, it admits a \emph{finite} open covering $X = U_1 \cup \dots \cup U_n$, where
$U_i = \Spec(R_i)$ for some noetherian ring $R_i$. For any $1 \le i \le n$, let $A_i := A(U_i)$. As in
\cite[Chapitre VI]{Gabriel}, one can easily show that $\Qcoh(\bbX)$ is equivalent to an iterated \emph{Gabriel's recollement} of the abelian categories $A_1-\Mod, \dots, A_n-\Mod$ (see also Definition \ref{D:Recollement} below).
Since $\kA$ is a coherent $\kO$--module, all rings
$A_1, \dots, A_n$ are noetherian. As in \cite[Chapitre VI, Th\'eor\`eme 1]{Gabriel}, one concludes that $\Qcoh(\bbX)$ is a \emph{locally noetherian} abelian category, whose subcategory
of \emph{noetherian objects} is the category $\Coh(\bbX)$ of coherent sheaves on $\bbX$; see
\cite[Section II.4]{Gabriel} for the corresponding definitions.
\smallskip
\noindent
Let $\bbX$ and $\bbY$ be two ncns and $\Qcoh(\bbX) \stackrel{\Phi}\lar \Qcoh(\bbY)$ be an equivalence of categories. It is clear that $\Phi$ restricts to an equivalence
$\Coh(\bbX) \stackrel{\Phi_{|}}\lar \Coh(\bbY)$ between the corresponding subcategories of noetherian objects. However, \cite[Section II.4, Th\'eor\`eme 1]{Gabriel} asserts that conversely, any equivalence $\Coh(\bbX) \rightarrow \Coh(\bbY)$ admits a unique (up to an isomorphism of functors) extension to
an equivalence $\Qcoh(\bbX) \rightarrow \Qcoh(\bbY)$ (in \cite{Gabriel}, this result is attributed to Grothendieck and Serre). Hence, even being primarily interested in the study of the category $\Coh(\bbX)$, it is technically more advantageous to work with a larger category
$\Qcoh(\bbX)$. One of the main reasons for this is a good behavior of the set $\Sp(\bbX)$
of the isomorphism classes of indecomposable injective objects of $\Qcoh(\XX)$ (see \cite[Section IV.2]{Gabriel}), for which it is crucial that $\Qcoh(\XX)$ is locally noetherian.
\smallskip
\noindent
Note that if we assume $X$ to be just \emph{locally noetherian} then even the category $\Qcoh(X)$ need not be locally noetherian in general; see \cite[Section II.7]{ResiduesDuality}. Hence, dropping the assumption for a ncns $\bbX$ to be noetherian would lead to significant technical complications.
\subsection{Reminder on the categorical center of an additive category}
\begin{definition}
The \emph{categorical center} $Z(\sA)$ of an additive category $\sA$ is the set of endomorphisms of the identity functor $\mathsf{Id}_{\sA}$, i.e.
\begin{equation*}
Z(\sA) :=
\left\{
\eta = \bigl(\bigl(X \stackrel{\eta_X}\lar X\bigr)_{X \in \mathsf{Ob}(\sA)}\bigr) \left|
\begin{array}{c}
\xymatrix{
X \ar[r]^-{\eta_X} \ar[d]_-{f} & X \ar[d]^-{f} \\
X' \ar[r]^-{\eta_{X'}} & X'
}
\end{array}
\right.
\;
\mbox{\rm is commutative for all} \; X \stackrel{f}\lar X'
\right\}.
\end{equation*}
It is easy to see that $Z(\sA)$ is a commutative ring.
\end{definition}
\smallskip
\noindent
It is well--known (see e.g. \cite[Proposition II.2.1]{Bass}) that for any ring $A$, the map
\begin{equation}\label{E:CenterCenter}
Z(A) \stackrel{\upsilon}\lar Z(\sA), \; r \mapsto (\lambda^M_r)_{M \in \Ob(\sA)}
\end{equation}
is a ring isomorphism, where $\sA = A-\Mod$.
\medskip
\noindent
The following result must be well--known. Its proof reduces to lengthy but completely straightforward verifications and is therefore left to an interested reader.
\begin{proposition}\label{P:Centers}
Let $\sA$ and $\sB$ be additive categories, $\eta \in Z(\sA)$ and $\sA \stackrel{\Phi}\lar
\sB$ be an additive functor satisfying the following conditions:
\begin{itemize}
\item $\Phi$ is essentially surjective.
\item
For any $X_1, X_2 \in \Ob(\sA)$ and $g \in \Hom_{\sB}\bigl(\Phi(X_1), \Phi(X_2)\bigr)$, there exist $X \in \Ob(\sA)$ and morphisms $X_1 \stackrel{t}\longleftarrow X \stackrel{f}\longrightarrow X_2$ in $\sA$ such that $\Phi(t)$ is an isomorphism and $g = \Phi(f) \cdot\bigl(\Phi(t)\bigr)^{-1}$.
\end{itemize}
For any $Y \in \Ob(\sB)$ choose a pair $\bigl(X_Y, \xi_Y\bigr)$, where $X_Y \in \Ob(\sA)$ and
$\Phi(X_Y) \stackrel{\xi_Y}\lar Y$ is an isomorphism.
Then the following statements are true.
\smallskip
\noindent
$\bullet$ The unique endomorphism $\vartheta_Y \in \End_{\sB}(Y)$, which makes the diagram
$$
\xymatrix{
\Phi(X_Y) \ar[r]^-{\xi_Y} \ar[d]_-{\Phi\bigl(\eta_{X_Y} \bigr)} & Y \ar[d]^-{\vartheta_Y}\\
\Phi(X_Y) \ar[r]^-{\xi_Y} & Y
}
$$
commutative, does not depend on the choice of the pair $\bigl(X_Y, \xi_Y\bigr)$.
\smallskip
\noindent
$\bullet$ Let $\vartheta = \bigl(\vartheta_Y\bigr)_{Y \in \Ob(\sB)}$, then we have: $\vartheta
\in Z(\sB)$ (in other words, for any $\eta \in Z(\sA)$, the family of endomorphisms
$\bigl(\Phi(\eta_X)\bigr)_{X \in \Ob(\sA)}$ in the category $\sB$ can be uniquely extended to an element $\vartheta
\in Z(\sB)$). Moreover, the map
$
Z(\sA) \stackrel{\Phi_c}\lar Z(\sB), \; \eta \mapsto \vartheta
$
is a ring homomorphism.
\smallskip
\noindent
$\bullet$ Let $\sA \stackrel{\Psi}\lar
\sB$ be a functor such that $\Phi \cong \Psi$. Then the induced maps of the corresponding categorical centers are equal: $\Phi_c = \Psi_c$.
Finally, if $\sA_1, \sA_2, \sA_3$ are additive categories and $\sA_1 \stackrel{\Phi_1}\lar
\sA_2 \stackrel{\Phi_2}\lar
\sA_3$ additive functors, satisfying the conditions of this proposition then we have: $
\bigl(\Phi_2 \Phi_1\bigr)_c = \bigl(\Phi_2)_c \bigl(\Phi_1)_c.
$
\end{proposition}
\smallskip
\noindent
From the point of view of applications in this paper, the following two classes of functors satisfying the conditions of
Proposition \ref{P:Centers} are of major interest:
\begin{itemize}
\item Equivalences of additive categories.
\item Serre quotient functors $\sA \rightarrow \sA/\sC$, where $\sC$ is a Serre subcategory of an abelian category $\sA$; see for instance \cite[Section 4.3]{Popescu}.
\end{itemize}
\begin{lemma}\label{L:MoritaCentralMap}
Let $A, B$ be any rings and $P$ be a $(B-A)$--Morita bimodule. Then there exists a unique isomorphism of centers $Z(A) \stackrel{\varphi}\lar Z(B)$ making the diagram
\begin{equation}\label{E:CommDiag}
\begin{array}{c}
\xymatrix{
\End_A(P) & B \ar[l]_-{\lambda^P} \\
Z(A) \ar[r]^-{\varphi} \ar@{^{(}->}[u]^-{\rho^P} & Z(B) \ar@{_{(}->}[u]
}
\end{array}
\end{equation}
commutative. In other words, for any $a \in Z(A)$ and $x \in P$ we have:
$\varphi(a) \cdot x = x \cdot a$. Moreover, $\varphi = \Phi_c$, where
$\Phi := P \otimes_A -: \; A-\Mod \lar B-\Mod$.
\end{lemma}
\begin{proof} Since $P$ is a balanced $(B-A)$--bimodule, the map $\lambda^P$ is bijective. This implies the uniqueness of $\varphi$. To show the existence, we prove that the induced map
of centers $Z(A) \stackrel{\Phi_c}\lar Z(B)$
makes the diagram (\ref{E:CommDiag}) commutative. Let $a \in Z(A)$, $b = \Phi_c(a)$ and $\vartheta = \upsilon(b) \in Z(B-\Mod),$ where $\upsilon$ is the map from (\ref{E:CenterCenter}). Then we have: $\vartheta_P = \lambda^P_b$. Let $P \otimes_A A \stackrel{\gamma}\lar P$ be the canonical isomorphism, then
the following diagram
$$
\xymatrix{
P \otimes_A A \ar[d]_-{\gamma} \ar[rr]^-{\mathsf{id}_P \otimes \lambda_a^A} & & P \otimes_A A \ar[d]^-{\gamma} \\
P \ar[rr]^-{\lambda_b^P} & & P
}
$$
is commutative (see
Proposition \ref{P:Centers}), what implies the statement.
\end{proof}
\begin{remark}\label{R:Pseudocommutative}
Let $A$ and $B$ be two rings, $P$ be a $(B-A)$--Morita bimodule and $\Phi = P \otimes_A -$ be the corresponding equivalence of categories. We may regard $\Phi$ as a ``virtual'' ring homomorphism $\xymatrix{A \ar@{.>}[r]^-{\Phi} & B}$. Then the commutativity of the diagram (\ref{E:CommDiag}) can be rephrased by saying that the diagram
\begin{equation}\label{E:CommDiagr}
\begin{array}{c}
\xymatrix{
A \ar@{.>}[r]^-{\Phi} & B \\
Z(A) \ar[r]^-{\varphi} \ar@{^{(}->}[u] & Z(B) \ar@{_{(}->}[u]
}
\end{array}
\end{equation}
is ``commutative''. Assume additionally that $A$ and $B$ are central $R$--algebras.
We call an equivalence $\Phi = P \otimes_A \,-\,$ \emph{central} if the induced map
$R \stackrel{\Phi_c}\lar R$ is the identity. According to Lemma \ref{L:MoritaCentralMap}, $\Phi$ is central if and only if for any $r$ in $R$ and $x \in P$ we have: $r \cdot x = x \cdot r$.
\end{remark}
\begin{definition}\label{D:Recollement}
Let $\sA, \sB, \sD$ be abelian categories and
$
\sA \stackrel{\Phi}\lar \sD \stackrel{\Psi}\longleftarrow \sB
$
be exact functors.
The \emph{Gabriel's recollement} $\sA {\prod \atop \sD} \sB$
is the category, whose objects are triples
$$
\left\{
(X, Y, f) \left| \begin{array}{c}
X \in \Ob(\sA) \\
Y \in \Ob(\sB)
\end{array}
\right.
\Phi(X) \stackrel{f}\lar \Psi(Y) \; \mbox{is an isomorphism in}\; \sD
\right\}
$$
and a morphism $(X, Y, f) \stackrel{(\alpha, \beta)}\lar (X', Y', f')$
is given by morphisms $X \stackrel{\alpha}\lar X'$ and $Y \stackrel{\beta}\lar Y'$ such that
$\Psi(\beta) f = f' \Phi(\alpha)$ see \cite[Section VI.1]{Gabriel}.
\end{definition}
\smallskip
\noindent
It is not difficult to check that
the category $\sC = \sA {\prod \atop \sD} \sB$ is abelian.
Assume additionally, that $\Phi$ and $\Psi$ are localization functors, i.e.~that they induce equivalences of categories
$$
\sA/\mathsf{Ker}(\Phi) \stackrel{\overline\Phi}\lar \sD \stackrel{\overline\Psi}\longleftarrow \sB/\mathsf{Ker}(\Psi),
$$
and admit right adjoint functors
$
\sA \stackrel{\widetilde\Phi}\longleftarrow \sD \stackrel{\widetilde\Psi}\lar \sB
$ (see \cite[Section III.2]{Gabriel}). Then we have a diagram of abelian categories and functors
\begin{equation}\label{E:Recollement}
\begin{array}{c}
\xymatrix{
\sC \ar[r]^-{\Phi^\dagger} \ar[d]_-{\Psi^\dagger} & \sB \ar[d]^-{\Psi}\\
\sA \ar[r]^-{\Phi} & \sD
}
\end{array}
\end{equation}
where $\Psi^\dagger(X, Y, f) = X$ and $\Phi^\dagger(X, Y, f) = Y$. Moreover, $\Phi^\dagger$ and $\Psi^\dagger$ are localization functors and $\Phi \Psi^\dagger \cong
\Psi \Phi^\dagger$.
\smallskip
\noindent
\begin{lemma}\label{P:CentersRecollement}
In the above setting, let $A, B, C, D$ be the centers of the categories $\sA, \sB, \sC$ and $\sD$, respectively. Then (\ref{E:Recollement}) induces a commutative diagram in the category of rings
\begin{equation}\label{E:RecollCenters}
\begin{array}{c}
\xymatrix{
C \ar[r]^-{\Phi^\dagger_c} \ar[d]_-{\Psi^\dagger_c} & B \ar[d]^-{\Psi_c}\\
A \ar[r]^-{\Phi_c} & D,
}
\end{array}
\end{equation}
which is moreover a \emph{pull--back} diagram. In other words, we have:
$$
C \cong A \times_D B := \bigl\{(a, b) \in A \times B \; \bigl| \; \Phi_c(a) = \Psi_c(b) \bigr\}.
$$
\end{lemma}
\smallskip
\noindent
\emph{Comment to the proof}. This statement is a consequence of Proposition \ref{P:Centers}. \qed
\smallskip
\noindent
We conclude this subsection with the following categorical version of the classical Skolem--Noether theorem.
\begin{theorem}\label{T:SkolemNoether}
Let $\kk$ be a field, $\Lambda$ and $\Gamma$ two semi--simple finite dimensional $\kk$--algebras and
$
\xymatrix{
\Lambda-\mathsf{Mod} \ar@/^/[r]^{\Phi} \ar@/_/[r]_{\Psi} & \Gamma-\mathsf{Mod}
}
$
two equivalences of categories such that
$\Phi_c = \Psi_c$. Then we have: $\Phi \cong \Psi$.
\end{theorem}
\begin{proof} Let $K = Z(\Lambda)$, $L = Z(\Gamma)$ and $K \stackrel{\varphi}\lar L$ be
the common isomorphism of centers induced by the equivalences $\Phi$ and $\Psi$ (i.e.
$\Phi_c = \varphi = \Psi_c$). Next, let $P$ and $Q$ be $(\Gamma-\Lambda)$--bimodules such that $\Phi = P \otimes_\Lambda \,-\,$ and $\Psi = Q \otimes_\Lambda \,-\,$. Let $\gamma = \bigl(\lambda_\Gamma^Q\bigr) \circ \bigl(\lambda_\Gamma^P\bigr)^{-1}$. By Lemma
\ref{L:MoritaCentralMap},
the following diagram of $\kk$--algebras and algebra homomorphisms
\begin{equation}\label{E:CDiag}
\begin{array}{c}
\xymatrix{
\Gamma \ar@/_5.5ex/[dd]_-{\id} \ar[rr]^-{\lambda_\Gamma^P} & & \End_\Lambda(P) \ar@/^5.5ex/[dd]^-{\gamma}\\
L \ar@{^{(}->}[u] \ar@{_{(}->}[d] & & \ar[ll]_-{\varphi} K \ar@{_{(}->}[u]_-{\varrho^P_K} \ar@{^{(}->}[d]^-{\varrho^Q_K} \\
\Gamma \ar[rr]^-{\lambda_\Gamma^Q} & & \End_\Lambda(Q) \\
}
\end{array}
\end{equation}
is commutative. In particular, we have: $\gamma \circ \varrho^P_K = \varrho^Q_K$.
\smallskip
\noindent
Since $\Lambda$ is a semi--simple $\kk$--algebra, there exist simple algebras $\Lambda_1, \dots, \Lambda_t$ such that $\Lambda \cong \Lambda_1 \times \dots \times \Lambda_t$. Moreover, for any $1 \le i \le t$ there exists a finite dimensional skew field $F_i$ over $\kk$ such that $\Lambda_i \cong \mathsf{Mat}_{m_i}(F_i)$ for some $m_i \in \NN$. If $K_i := Z(F_i)$ then we have: $K \cong K_1 \times \dots \times K_t$. Let $U_i$ be a finite dimensional simple right $\Lambda_i$--module (which is unique up to an isomorphism). Then we have: $F_i \cong \End_{\Lambda_i}(U_i)$. Moreover, we have direct sum decompositions
$
P \cong U_1^{\oplus p_1} \oplus \dots \oplus U_t^{\oplus p_t}$ and $
Q \cong U_1^{\oplus q_1} \oplus \dots \oplus U_t^{\oplus q_t}.
$
Then we get:
$$
\End_\Lambda(P) \cong \mathsf{Mat}_{p_1}(F_1) \times \dots \times \mathsf{Mat}_{p_t}(F_t)
\quad \mbox{\rm and} \quad
\End_\Lambda(Q) \cong \mathsf{Mat}_{q_1}(F_1) \times \dots \times \mathsf{Mat}_{q_t}(F_t).
$$
It follows from (\ref{E:CDiag}) that there exists an isomorphism of \emph{$K$--algebras} (and not just of \emph{$\kk$--algebras})
$
\mathsf{Mat}_{p_1}(F_1) \times \dots \times \mathsf{Mat}_{p_t}(F_t)
\lar \mathsf{Mat}_{q_1}(F_1) \times \dots \times \mathsf{Mat}_{q_t}(F_t),
$
what implies that $p_i = q_i$ for all $1 \le i \le t$. In particular, $P$ and $Q$ are isomorphic, at least as right $\Lambda$--modules.
\smallskip
\noindent
Let $P \stackrel{h}\lar Q$ be any isomorphism of right $\Lambda$--modules. Then the following diagram
\begin{equation*}
\begin{array}{c}
\xymatrix{
\End_\Lambda(P) \ar[r]^-{\Ad_h}& \End_\Lambda(Q) \\
K \ar@{_{(}->}[u]^-{\rho^P_K} \ar[r]^-{\id} & K \ar@{^{(}->}[u]_-{\rho^Q_K}
}
\end{array}
\end{equation*}
is commutative, i.e.~the isomorphism $\Ad_h$ is central. Indeed, for any $f \in \End_\Lambda(P)$ we have: $\Ad_h(f) h = h f$. For any $\lambda \in K$ consider the endomorphism
$\varrho_\lambda^P \in \End_\Lambda(P)$. Since $h$ is $K$--linear, we have: $h \varrho_\lambda^P = \varrho_\lambda^Q h$. Hence, $\varrho_\lambda^Q = \Ad_h(\varrho_\lambda^P)$ for
any $\lambda \in K$.
\smallskip
\noindent
Consider the map $\delta:= \Ad_h \cdot \gamma^{-1}: \End_\Lambda(Q) \lar \End_\Lambda(Q)$. From what was said above it follows that $\delta$ is an isomorphism of $K$--algebras. Now we can finally apply the classical Skolem--Noether theorem: there exists $\bar{h} \in \End_\Lambda(Q)$ such that $\delta = \Ad_{\bar{h}}$. Consider the isomorphism of right $\Lambda$--modules $g = \bar{h}^{-1} h: P \lar Q$. Then we have:
$
\gamma = \Ad_{\bar{h}}^{-1} \Ad_h = \Ad_{g}.
$
It follows from commutativity of (\ref{E:CDiag}) that the diagram
$$
\xymatrix{
& \Gamma \ar[rd]^-{\lambda_\Gamma^Q} \ar[ld]_-{\lambda_\Gamma^P}& \\
\End_\Lambda(P) \ar[rr]^-{\Ad_g} & & \End_\Lambda(Q)
}
$$
is commutative, too. Hence, $P \stackrel{g}\lar Q$ is also $\Gamma$--linear. Summing up, $g$ is an isomorphism of $(\Gamma-\Lambda)$--bimodules and
$\Phi \cong \Psi$, as asserted.
\end{proof}
\subsection{Centralizing a non--commutative
noetherian schemes}\label{SS:CentralizingNCNS}
The goal of this subsection is to show, that any ncns can be replaced by a Morita equivalent central ncns.
\begin{proposition}\label{P:CentralSheaf}
Let $\bbX = (X, \kA)$ be a ncns. For all open subsets $U \subseteq X$ we put:
\begin{equation}
\Gamma(U, \kZ_\kA):=
\Bigl\{\alpha \in \Gamma(U, \kA) \, \left|\, \alpha\big|_{V} \in
Z\bigl(\Gamma(V, \kA)\bigr) \; \mbox{for all}\; V \subseteq U \;\mbox{open} \right.\Bigr\}.
\end{equation}
Then $\kZ = \kZ_\kA$ is a coherent sheaf on $X$ such that $
\kZ_x \cong Z(A_x)$ for any $x \in X$. Moreover, the canonical map
\begin{equation}\label{E:CatCenterNcns}
\Gamma(X,
\kZ) \stackrel{\upsilon_{\bbX}}\lar
Z\bigl(\Qcoh(\XX)\bigr)
\end{equation} is a ring isomorphism.
\end{proposition}
\begin{proof} It is clear that $\kZ_\kA$ is a presheaf of commutative rings, which is a sub--presheaf of $\kA$. We have to check the sheaf property of $\kZ$. Let $U \subseteq X$ be any open subset, $U = \cup_{i \in I} U_i$ an open covering and
$$
\bigl(\alpha_i \in \Gamma\bigl(U_i, \kZ\bigr)\bigr)_{i \in I} \;
\mbox{\rm be such that}\; \alpha_k\big|_{U_k \cap U_l} = \alpha_l\big|_{U_k \cap U_l} \; \mbox{\rm for all}\; k, l \in I.
$$
Then there exists a unique section $\alpha \in \Gamma(U, \kA)$ such that
$\alpha\big|_{U_i} = \alpha_i$ for all $i \in I$. We have to show that
$\alpha\big|_{V} \in Z\bigl(\Gamma(V, \kA)\bigr)$ for any open subset $V \subseteq U$. Consider any $\beta \in \Gamma(V, \kA)$. We have to prove that
$\bigl[\alpha\big|_{V}, \beta\bigr] = 0$. Indeed, we have:
$V = \cup_{i \in I} V_i$, where $V_i = V \cap U_i$. Since
$\alpha\big|_{V_i} = \alpha_i\big|_{V_i}$ and $\alpha_i \in \Gamma\bigl(U_i, \kZ_\kA\bigr)$, we conclude that $\alpha\big|_{V_i} \in Z\bigl(\Gamma(V_i, \kA)\bigr)$. It implies that
$
\bigl[\alpha\big|_V, \beta\bigr]\Big|_{V_i} = 0
$
for all $i \in I$. Hence, $\bigl[\alpha\big|_V, \beta\bigr] = 0$.
\smallskip
\noindent
Let $\alpha \in \Gamma(X, \kZ)$. Then for any
$\kF \in \Ob\bigl(\Qcoh(\bbX)\bigr)$, we have an endomorphism $\alpha_\kF \in
\End_{\bbX}(\kF)$ given for any open subset $U \subseteq X$ by the rule
$$
\Gamma(U, \kF) \stackrel{\alpha^{U}_\kF}\lar \Gamma(U, \kF), \; f \mapsto \alpha\big|_{U} \cdot \varphi.
$$
It is clear that the collection of endomorphisms $(\alpha_\kF)$ defines an element
of $Z\bigl(\Qcoh(\XX)\bigr)$, which we denote by $\upsilon_{\bbX}(\alpha)$ (it is how the canonical map $\upsilon_{\bbX}$ from (\ref{E:CatCenterNcns}) is actually defined). If $\alpha \ne 0$ then $\alpha_\kA \ne 0$, too. Hence, the map $\upsilon_{\bbX}$ is at least injective.
\smallskip
\noindent
To show the surjectivity of $\upsilon_{\bbX}$, assume first that $X$ is affine. Let $A := \Gamma(X, \kA)$, then the functor of global sections
$
\Qcoh(\bbX) \stackrel{\Gamma}\lar A-\Mod
$
is an equivalence of categories and the induced map of centers
$Z(A) \lar Z\bigl(\Qcoh(\bbX)\bigr)$ is an isomorphism.
In the same way as above one can show that
$\alpha\big|_{V} \in
Z\bigl(\Gamma(V, \kA)\bigr)$ for any open subset $V \subseteq X$ and $\alpha
\in Z(A)$.
\smallskip
\noindent
Next, note that we have a sheaf isomorphism $\kZ\big|_{U}
\cong \kZ_{\kA|_{U}}$ for any open subset $U \subseteq X$. If $U$ is moreover affine,
it follows that $\Gamma(U, \kZ) =
Z\bigl(\Gamma(U, \kA)\bigr)$.
\smallskip
\noindent
Now, we prove by induction on the minimal number of affine open charts of an affine open covering of $X$ that $\upsilon_{\bbX}$ is an isomorphism. The case of an affine scheme $X$ is already established. Assume that this statement is true for any nccs, which can be covered by $n$ affine charts. Suppose that we have an affine open covering
$X = U_1 \cup \dots \cup U_n \cup U_{n+1}$. Let $U = U_1 \cup \dots \cup U_n$ and
$V = U_{n+1}$. Since $X$ is separated, $(U_1 \cap V) \cup \dots \cup (U_n \cap V)$ is an affine open covering of $W:= U \cap V$ and the category
$\Qcoh(\bbX)$ is equivalent to Gabriel's recollement with respect to the diagram
$$
\Qcoh(\bbU) \stackrel{\left(\varrho^{U}_W\right)^\ast}\lar \Qcoh(\bbW)
\stackrel{\left(\varrho^{V}_W\right)^\ast}\longleftarrow \Qcoh(\bbV).
$$
Let $R := Z\bigl(\Qcoh(\bbU)\bigr)$, $S := Z\bigl(\Qcoh(\bbV)\bigr)$,
$C := Z\bigl(\Qcoh(\bbX)\bigr)$ and $T = Z\bigl(\Qcoh(\bbW)\bigr)$. Then Lemma
\ref{P:CentersRecollement} implies that
$C \cong R \times_T S$.
\smallskip
\noindent
On the other hand, we have a commutative diagram of rings and ring homomorphisms
$$
\xymatrix{
\Gamma(U, \kZ) \ar[r] \ar[d]_-{\upsilon_{\bbU}} & \Gamma(W, \kZ)
\ar[d]_-{\upsilon_{\bbW}} & \ar[l] \Gamma(V, \kZ) \ar[d]^-{\upsilon_{\bbV}} \\
R \ar[r] & T & \ar[l] S
}
$$
in which all vertical maps are isomorphisms due to the hypothesis of induction.
The sheaf property of $\kZ$ implies that the map $\upsilon_{\bbX}$ is surjective, hence bijective.
\smallskip
\noindent
The fact that the sheaf $\kZ$ is coherent and that we have isomorphism $
\kZ_x \cong Z(A_x)
$ for any $x \in X$ are now easy consequences of Lemma \ref{L:CenterofRing}.
\end{proof}
\begin{corollary}\label{C:Centers}
Let $\bbX = (X, \kA)$ be a ncns and $V \subseteq U \subseteq X$ be open subsets. Then the following diagram of rings and ring homomorphisms
\begin{equation}\label{E:ReconstrCenter}
\begin{array}{c}
\xymatrix{
\Gamma(U, \kZ) \ar[rr]^{\upsilon_\bbU} \ar[d]_-{\varrho^U_V}& &
Z\bigl(\Qcoh(\bbU)\bigr) \ar[d]^-{\varphi^U_V}\\
\Gamma(V, \kZ) \ar[rr]^{\upsilon_\bbV} & & Z\bigl(\Qcoh(\bbV)\bigr)
}
\end{array}
\end{equation}
is commutative, where $\varphi^U_V$ is the morphism of centers induced by the localization
functor $\Qcoh(\bbU) \rightarrow \Qcoh(\bbV)$. Moreover, the horizontal maps in (\ref{E:ReconstrCenter}) are isomorphisms.
\end{corollary}
\begin{remark}\label{R:CenterSheaf} Let $\bbX = (X, \kA)$ be a ncns. For any open subset
$V \subseteq X$, consider the map
$$
\Gamma(V, \kZ) \lar
\End_{A(V)^e}\bigl(A(V)\bigr), \; \alpha \mapsto \bigl(\beta \mapsto \alpha \cdot \beta \bigr).
$$
These maps define a morphism of sheaves of $\kO$--algebras $\kZ \rightarrow \mathit{End}_{\kA^e}(\kA)$,
where $\kA^e := \kA \otimes_\kO \kA^\circ$. If $\bbX$ is noetherian then
it is an isomorphism. However, we do not know whether the equality
$\Gamma(U, \kZ) =
Z\bigl(A(U)\bigr)$ is true for an \emph{arbitrary} open subset $U \subseteq X$. Nevertheless, Proposition \ref{P:CentralSheaf} implies that $\bbX$ is central if and only if the canonical morphism $\kO \rightarrow \kZ$ is an isomorphism.
\end{remark}
\begin{remark}\label{R:RemarkCentrality} Let $\bbX = (X, \kA)$ be a ncns.
Since $\kZ$ is a coherent sheaf of commutative $\kO$--algebras, there exists
a commutative noetherian scheme $\widetilde{X} = \Spec(\kZ)$ over $X$; see
\cite[Proposition 1.3.1]{EGAII}. Let $\widetilde{X} \stackrel{\phi}\lar X$ be the corresponding structure morphism and $\widetilde\kA : = \phi^{-1} \kA$. Then the non--commutative scheme $\widetilde{\bbX} = (\widetilde{X}, \widetilde\kA)$ is central. Moreover, the functor
$
\Qcoh(\widetilde\bbX) \stackrel{\phi_*}\lar \Qcoh(\bbX)
$
is an equivalence of categories. Thus, we get the following important conclusion:
any ncns $\bbX$ can be replaced by a \emph{Morita equivalent central} ncns $\widetilde{\bbX}$.
\end{remark}
\section{Indecomposable injective quasi--coherent sheaves on non--commutative noetherian schemes}
\noindent
The goal of this section is to clarify the structure of indecomposable injective objects of the category $\Qcoh(\bbX)$, where $\bbX$ is a ncns.
\subsection{Prime ideals in non--commutative rings}
Let $A$ be any ring. Unless explicitly stated otherwise, by an ideal in $A$ we always mean a two--sided ideal.
\smallskip
\noindent
Recall that an ideal $P$ in $A$ is \emph{prime} if for any ideals $I, J$ in $A$ such that
$I J \subseteq P$ holds: $I \subseteq P$ or $J \subseteq P$. Equivalently, for any
$a, b \in A$ such that $a A b \subseteq I$ we have: $a\in I$ or $b \in I$. We refer to \cite[Proposition 10.2]{LamFirstCourse} for other characterizations of prime ideals in non--commutative rings. Note that any maximal ideal is automatically prime.
\smallskip
\noindent
Similarly to the commutative case, $\Max(A)$ (respectively,
$\Spec(A)$) denotes the set of maximal (respectively, prime) ideals in $A$.
\begin{lemma}\label{L:PrimeIdealsSmallStatement} Let $P \in \Spec(A)$ and $I$ be an ideal in $A$ such that $I \not\subseteq P$. Then for any $a \in A \setminus P$ there exists $b \in I$ such that $ba \notin P$.
\end{lemma}
\begin{proof}
Since $I \not\subseteq P$ and $AaA \not\subseteq P$, we conclude that $IaA \not\subseteq P$, hence $Ia \not\subseteq P$. Therefore, there exists $b \in I$ such that $ba \notin P$.
\end{proof}
\begin{lemma}\label{L:EssExtJacobRadical}
Let $P_1, \dots, P_n \in \Spec(A)$ be such that $P_i \not\subseteq P_j$ for all
$1 \le i \ne j \le n$ and $P := P_1 \cap \dots \cap P_n$. Then the canonical ring homomorphism
$
A/P \stackrel{\jmath}\lar A/P_1 \times \dots \times A/P_n
$
is an essential extension of $A$--modules.
\end{lemma}
\begin{proof}
It is sufficient to show that for any
$0 \ne x \in A/P_1 \times \dots \times A/P_n$, there exists $\lambda \in A$ such that $0 \ne \lambda x \in \mathsf{Im}(\jmath)$. Without loss of generality assume that
$x = (\bar{a}_1, \dots, \bar{a}_n)$ and
$a_1 \notin P_1$. Since $P_2 \not\subseteq P_1$, Lemma
\ref{L:PrimeIdealsSmallStatement} implies that there exists $\mu \in P_2$ such that
$\mu a_1 \notin P_1$. Proceeding inductively, we construct $\lambda \in P_2 \cap \dots \cap P_n$ such that $\lambda a_1 \notin P_1$. Then we get:
$
\lambda x = (\overline{\lambda a_1}, 0, \dots, 0) \in \mathsf{Im}(\jmath),
$
implying the statement.
\end{proof}
\begin{proposition}\label{P:PrimeIdealsBascics} Let $R$ be a commutative noetherian ring and $A$ be a finite $R$--algebra.
Then the following statements are true.
\begin{itemize}
\item For any $P \in \Spec(A)$ we have: $P\cap R \in \Spec(R)$.
\item The map $\Spec(A) \stackrel{\varrho}\lar \Spec(R), P \mapsto P\cap R,$ is surjective and has finite fibers.
\item Let $P \in \Spec(A)$ and $\idp = \varrho(P)$. Then we have: $P_{\idp} \in
\Max(A_{\idp})$.
\item Le $\idp, \idq \in \Spec(R)$ and $P \in \Spec(A)$ be such that $\idp \subseteq \idq$ and $\varrho(P) = \idp$. Then there exists $Q \in \Spec(A)$ such that $P \subseteq Q$ and $\varrho(Q) = \idq$.
\item Let $P, Q \in \Spec(A)$ be such that $P \subseteq Q$ and $\varrho(P) = \varrho(Q)$. Then we have: $P = Q$.
\item Let $\idp \in \Spec(R)$, $A \stackrel{\jmath}\lar A_{\idp}$ be the canonical ring homomorphism, $Q \in \Max(A_{\idp})$ and $\widetilde{Q} := \jmath^{-1}(Q)$. Then we have:
$\widetilde{Q} \in \Spec(A)$ and $\widetilde{Q}_{\idp} = Q$.
If $\varrho^{-1}(\idp) =
\left\{P_1, \dots, P_n\right\}$, where $P_i \ne P_j$ for all $1 \le i \ne j \le n$, then we have:
$
\Max(A_{\idp}) = \bigl\{(P_1)_{\idp}, \dots, (P_n)_{\idp}\bigr\}$
and $\bigl(P_i\bigr)_{\idp} \ne \bigl(P_j\bigr)_{\idp}$ for all $1 \le i \ne j \le n$.
\end{itemize}
\end{proposition}
\smallskip
\noindent
Proofs of all these results are analogous to the commutative case; see \cite[Section V.6]{Gabriel}.
\begin{lemma}\label{L:JacobsonRadical}
Let $(R, \idm)$ be a local commutative noetherian ring, $A$ be a finite $R$--algebra,
$J$ be its Jacobson radical and $\Max(A) = \left\{P_1, \dots, P_n\right\}$. Then we have: $J = P_1 \cap \dots \cap P_n$.
\end{lemma}
\begin{proof} Recall that $J = \bigcap_{U} \ann_{A}(U)$, where the intersection is taken over the annihilators of all simple left $A$--modules; see
\cite[Proposition 5.13]{CurtisReiner}. Note that any such $\ann_{A}(U)$ is a prime ideal; see \cite[Proposition 3.15]{GoodearlWarfield}. On the other hand,
$\idm \subseteq J$; see \cite[Proposition 5.22]{CurtisReiner}; hence
$\idm \subseteq \ann_{A}(U)$. By Proposition \ref{P:PrimeIdealsBascics}, $\ann_{A}(U)$
is a maximal ideal in $A$. Conversely, for any $P \in \Max(A)$ there exists
a simple left $A$--module $U$ such that $P = \ann_{A}(U)$; see \cite[Proposition 3.15]{GoodearlWarfield}. This implies the statement.
\end{proof}
\begin{proposition}\label{P:EndomInjHull}
In the notation of Lemma \ref{L:JacobsonRadical}, let $E_i = E_A(A/P_i)$ be the injective envelope
of $A/P_i$ for all $1 \le i \le n$. Then we have:
$
\End_A(E_1 \oplus \dots \oplus E_n) \cong \widehat{A}^\circ,
$
where $\widehat{A}$ is the $\idm$--adic completion of the algebra $A$.
\end{proposition}
\begin{proof} Let $E$ be the injective envelope of the left $A$--module $T:= A/J$. Lemma \ref{L:EssExtJacobRadical} implies that $E \cong E_1 \oplus \dots \oplus E_n$.
Let $\widehat{J}$ be the Jacobson radical of $\widehat{A}$. Then we have:
$\widehat{J} = J \widehat{A}$ and $A/J \cong \widehat{A}/\widehat{J}$.
The Matlis Duality functor $\DD$ (see \cite[Corollary 4.3]{Matlis})
establishes an anti--equivalence between the categories of noetherian right $\widehat{A}$--modules and artinian left $\widehat{A}$--modules. Since $T$
is semi--simple and of finite length, we have: $\DD(T_{\widehat{A}}) \cong {}_{\widehat{A}}T$. Moreover,
$\DD$ maps the projective cover of $T$ (which is just $\widehat{A}_{\widehat{A}}$) to
the injective envelope of $T$. However, the injective envelope of $T$, viewed as
a left $\widehat{A}$--module, can be identified with $E$ and
$\End_{A}(E) \cong \End_{\widehat{A}}(E)$; see e.g. \cite[Theorem 18.6]{Matsumura}
(the proof of \cite{Matsumura} can be literally generalized to the non--commutative setting). Since
$\End_{\widehat{A}}\bigl(\widehat{A}_{\widehat{A}}\bigr) \cong\widehat{A}$, we conclude that $\End_{A}(E) \cong \widehat{A}^\circ$.
\end{proof}
\subsection{Prime ideals and indecomposable injective modules} Recall the following standard results about indecomposable injective modules.
\begin{lemma}\label{L:EndomIndInj}
Let $A$ be any ring, $I$ be an injective $A$--module and $H = \End_{A}(I)$. Then the following statements are true.
\begin{itemize}
\item $I$ is indecomposable if and only if $H$ is local. Moreover, in this case
$f \in H$ is a unit if and only if $\ker(f) = 0$.
\item Assume additionally that $A$ is left noetherian. If $I$ is indecomposable then any $f \in H$ is either a unit or locally nilpotent (i.e.~for any $x \in I$ there exists $n \in \NN$ such that $f^n(x) = 0$).
\end{itemize}
\end{lemma}
\smallskip
\noindent
\emph{Comment to the proof}. For the first statement, see
\cite[Proposition 2.6]{Matlis}. For the second result, see \cite[Lemme 2, page 428]{Gabriel}. \qed
\smallskip
\noindent
From now on in this subsection, we assume that $R$ is a commutative noetherian ring and $A$ is a finite $R$--algebra. We denote by $\Sp(A)$ the set of the isomorphism classes of indecomposable injective $A$--modules.
\begin{proposition}
For any $P \in \Spec(A)$ there exist uniquely determined $I_P \in \Sp(A)$ and $m_P \in \NN$ such
that $E_A(A/P) \cong I_P^{\oplus m_P}$. Moreover, the assignment
$$
\Spec(A) \stackrel{\varepsilon}\lar \Sp(A), \; P \mapsto I_P
$$
is a bijection.
\end{proposition}
\noindent
\emph{Comment to the proof}. This result is proven in \cite[Section V.4]{Gabriel}. In fact, any indecomposable injective $A$--module $I$ has a uniquely determined associated prime ideal $P$; see also \cite[Section 3F]{Lam} for further details. \qed
\smallskip
\noindent
Composing the inverse of $\varepsilon$ with the map $\varrho$ from Proposition \ref{P:PrimeIdealsBascics},
we get a map
$$
\Sp(A) \stackrel{\alpha}\lar \Spec(R).
$$
It turns out, that $\alpha$ has a clear conceptual meaning: it assigns to an indecomposable injective $A$--module its uniquely determined associated prime ideal in $R$.
\begin{proposition}\label{P:Associator}
Let $I \in \Sp(A)$ and $\idp = \alpha(I)$. For
any $r \in R$, let $\lambda_r^I \in \End_A(I)$ be the (left) multiplication map with $r$. Then the following statements are true.
\begin{enumerate}
\item If $r \in \idp$ then $\lambda^I_r$ is locally nilpotent, i.e. for any $x \in I$ there exists $n \in \NN$ such that $r^n x = 0$.
\item If $r \in R\setminus \idp$ then $\lambda_r^I$ is invertible.
\item $\idp$ is the unique associated prime ideal of $I$ viewed as an $R$--module.
\item We have: $\mathsf{Supp}(I) = \overline{\left\{\idp\right\}} \subset \Spec(R)$.
\end{enumerate}
\end{proposition}
\begin{proof} Let $P \in \Spec(A)$ be the associated prime ideal of $I$ and
$E$ be the injective envelope of $A/P$. Then there exists $m \in \NN$ such that
$E \cong I^{\oplus m}$.
For any $r \in R$, we have a commutative diagram of $A$--modules
$$
\xymatrix{
A/P \ar@{^{(}->}[rr] \ar[d]_-{\lambda_r^{A/P}} & & E \ar[d]^-{\lambda_r^E} \\
A/P \ar@{^{(}->}[rr] & & E
}
$$
Note that $\lambda_r^E = \mathsf{diag}(\lambda_r^I, \dots, \lambda_r^I)$.
\smallskip
\noindent
(1) If $r\in \idp$ then $\lambda_r^{A/P} = 0$. Hence, $\ker(\lambda_r^E) \ne 0$ and
$\ker(\lambda_r^I) \ne 0$, too. According to Lemma \ref{L:EndomIndInj}, the endomorphism
$\lambda_r^I$ is locally nilpotent.
\smallskip
\noindent
(2) Let $r\in R\setminus \idp$. Since $P$ is prime, the map $\lambda_r^{A/P}$ is injective. Since the extension $A/P \subset E$ is essential, we have:
$\ker(\lambda_r^E) = 0$. Hence, $\ker(\lambda_r^I) = 0$ and Lemma \ref{L:EndomIndInj} implies that $\lambda_r^I$ is an isomorphism.
\smallskip
\noindent
(3) We have a non--zero map of $R$--modules $R/\idp \stackrel{\beta}\lar I$, obtained as the composition
$$
R/\idp \longhookrightarrow A/P \longhookrightarrow E \dhrightarrow I,
$$
where the last map is an appropriate projection of $E$ onto one of its indecomposable direct summands. It follows from part (2) that $\beta$ is automatically injective, hence
$\idp$ is an associated prime ideal of $R$.
\smallskip
\noindent
Next, assume that $\idq \ne \idp$ is another associated prime ideal of $I$. Then there exists an inclusion of $R$--modules $R/\idq \longhookrightarrow I$. Note that
for any $r \in R$, the following diagram
$$
\xymatrix{
R/\idq \ar@{^{(}->}[rr] \ar[d]_{\lambda_r^{R/\idq}} & & I \ar[d]^-{\lambda_r^I} \\
R/\idq \ar@{^{(}->}[rr] & & I
}
$$
is commutative. If $r \in \idq \setminus \idp$ then $\lambda_r^{R/\idq} = 0$ and $\lambda_r^I$ is invertible (by part (2)). If $r \in \idp \setminus \idq$ then
$\lambda_r^{R/\idq}$ is injective and $\lambda_r^I$ is locally nilpotent (by part (1)). In both cases, we get a contradiction.
\smallskip
\noindent
(4) The inclusion $\overline{\{\idp\}} \subseteq \mathsf{Supp}(I)$ follows from part (3). If $\idq \in \Spec(R)$ is such that $\idp \not\subseteq \idq$ then there exists
$r \in \idp \setminus \idq$. By part (1), for any $x \in I$ there exists $n \in \NN$ such that $r^n x = 0$. This implies that $I_{\idq} = 0$. \end{proof}
\begin{lemma}\label{L:mapsindinject} Let $I, J \in \Sp(A)$ be such that $\Hom_A(I, J) \ne 0$. Then we have:
$\idp \subseteq \idq$, where $\idp = \alpha(I)$ and $\idq = \alpha(J)$.
\smallskip
\noindent
Conversely, let $\idp, \idq \in \Spec(R)$ be such that $\idp \subseteq \idq$. Then there exist $I, J \in \Sp(A)$ such that $\Hom_A(I, J) \ne 0$, $\idp = \alpha(I)$ and
$\idq = \alpha(J)$.
\end{lemma}
\begin{proof}
Let $I \stackrel{f}\lar J$ be a non--zero homomorphism of $A$--modules and
$x \in I$ be such that $y := f(x) \ne 0$. Assume that there exists $r \in \idp \setminus \idq$. Then $\lambda_r^I$ is locally nilpotent, so we can find $n \in \NN$ such that
$r^n x = 0$. Hence, $r^n y = 0$, too. On the other hand, the $\lambda_r^J$ is invertible. Contradiction.
\smallskip
\noindent
To prove the second part, take any $P, Q \in \Spec(A)$ such that $P \subseteq Q$,
$P \cap R = \idp$ and $Q \cap R = \idq$ (such $P$ and $Q$ exist by Proposition \ref{P:PrimeIdealsBascics}). Then we have a non--zero homomorphism of $A$--modules
$A/P \stackrel{g}\lar E_A(A/Q)$, defined as the composition
$
A/P \dhrightarrow A/Q \longhookrightarrow E_A(A/Q),
$
where $E_A(A/Q)$ is the injective hull of $A/Q$. By injectivity of $E_A(A/Q)$, there exists a non--zero morphism $E_A(A/P) \stackrel{\tilde{g}}\lar E_A(A/Q)$ extending $g$.
Since $E_A(A/P) \cong I_P^{\oplus m_P}$ and $E_A(A/Q) \cong I_Q^{\oplus m_Q}$ for some $m_P, m_Q \in \NN$, we conclude that $\Hom_A(I_P, I_Q) \ne 0$.
\end{proof}
\begin{corollary}
For any $\idp \in \Spec(R)$ we put:
$
I(\idp) := \bigoplus\limits_{\substack{I \in \Sp(A) \\ \alpha(I) = \idp}} I.
$
Then for any $\idp, \idq \in \Spec(R)$ we have:
$
\Hom_A\bigl(I(\idp), I(\idq)\bigr) \ne 0$ if and only if
$\idp \subseteq \idq.
$
\end{corollary}
\begin{lemma}
Let $P \in \Spec(A)$, $\idp := P \cap R \in \Spec(R)$ and $E$ be the injective hull of the $A_{\idp}$--module $A_{\idp}/P_{\idp}$. Then we have an isomorphism of
$A$--modules $E \cong E_A(A/P)$ and $\End_{A_{\idp}}(E) \cong \End_A(E)$.
\end{lemma}
\begin{proof}
The forgetful functor $A_{\idp}-\Mod \stackrel{\Phi_{\idp}}\lar A-\Mod$ admits an exact left adjoint functor $A-\Mod \lar A_{\idp}-\Mod$ given by the localization with respect to $\idp$. It is easy to see that the corresponding adjunction counit is an isomorphism. This implies that $\Phi_{\idp}$ is fully faithful and maps injective objects to injective objects. Hence, $E$ is an injective $A$--module and $\End_{A_{\idp}}(E) \cong \End_A(E)$. Next, it is not difficult to see that both inclusions
$
A/P \longhookrightarrow (A/P)_{\idp} \longhookrightarrow E
$
are essential extensions of $A$--modules. Hence, $E$ can be identified with the injective hull of $A/P$, implying the result.
\end{proof}
\begin{corollary}\label{C:IndInjModules} For any $\idp \in \Spec(R)$, let
$\overline{\Sp}(A_{\idp})$ be the set of the isomorphism classes of
indecomposable injective \emph{artinian} $A_{\idp}$--modules.
Then we have the following description of indecomposable injective $A$--modules:
$$
\Sp(A) = \bigsqcup\limits_{\idp \in \Spec(R)} \overline{\Sp}(A_{\idp}) = \bigsqcup\limits_{\idp \in \Spec(R)} \Max(A_{\idp}),
$$
where we view $I \in \overline{\Sp}(A_{\idp})$ as an element of $\Sp(A)$ via the forgetful functor $\Phi_{\idp}$.
\end{corollary}
\smallskip
\noindent
The following two results play the key role in the proof of the Morita theorem for ncns.
\begin{proposition}\label{P:NoetherianEndom}
For any $\idp \in \Spec(R)$, the ring $\End_A\bigl(I(\idp)\bigr)$ is a finite
$\widehat{R}_{\idp}$--module. In particular, $\End_A\bigl(I(\idp)\bigr)$ is noetherian.
\end{proposition}
\begin{proof} In the notation of Proposition \ref{P:PrimeIdealsBascics}, let
$\varrho^{-1}(\idp) =
\bigl\{P_1, \dots, P_n\bigr\}$, where $P_i \ne P_j$ for all $1 \le i \ne j \le n$.
Then $J := \bigl(P_1\bigr)_{\idp}\cap \dots \cap \bigl(P_n\bigr)_{\idp}$ is the Jacobson radical of $A_{\idp}$. Let $E_i := E_A(A/P_i)$ and
$E := E_{A_{\idp}}\bigl(A_{\idp}/J\bigr)$. By Proposition \ref{P:EndomInjHull} we have: $\End_{A_{\idp}}(E) \cong {\widehat{A}_{\idp}}^\circ$.
In particular, $\End_{A_{\idp}}(E)$ is a finite $\widehat{R}_{\idp}$--algebra.
On the other hand, we have
an isomorphism of $A$--modules $E \cong E_1 \oplus \dots \oplus E_n$. Since $\Phi_{\idp}$ is fully faithful, we get a ring isomorphism
$$
\End_{A_{\idp}}(E) \cong \End_{A}(E_1 \oplus \dots \oplus E_n).
$$
For any $1 \le i \le n$ there exists $m_i \in \NN$ such that $E_i
\cong I_{P_i}^{\oplus m_i}$. Therefore, $\End_A\bigl(I(\idp)\bigr)$
and ${\widehat{A}_{\idp}}^\circ$ are Morita--equivalent, what implies the statement.
\end{proof}
\begin{lemma}\label{L:KeyLemma}
Let $I, J \in \Sp(A)$ be such that $\Hom_A(I, J) \ne 0$. Assume that $\alpha(I) \ne \alpha(J)$. Then $\Hom_A(I, J)$ is not noetherian viewed as a left $\End_A(J)$--module.
\end{lemma}
\begin{proof}
Let $\idp = \alpha(I)$ and $\idq = \alpha(J)$. By Lemma \ref{L:mapsindinject} we have: $\idp \subseteq \idq$. Assume that $\Hom_A(I, J)$ is noetherian viewed as a left $\End_A(J)$--module. By Proposition \ref{P:NoetherianEndom}, there exists a finite map of rings $\widehat{R}_{\idq} \stackrel{\vartheta}\lar \End_A(J)$. Hence, $\Hom_A(I, J)$ is noetherian viewed as an $\widehat{R}_{\idq}$--module, too. Note that
for any $r \in \idq$, the corresponding element $\vartheta(r) \in \End_A(J)$ acts
on $\Hom_A(I, J)$ by the rule $f \mapsto \lambda_r^J \cdot f = f \cdot \lambda_r^I$.
Suppose now that there exists $r \in \idq \setminus \idp$. Then
$\lambda_r^I \in \End_A(I)$ is a unit. Hence, $r \cdot \Hom_A(I, J) =
\Hom_A(I, J)$. On the other hand, $r \in \idq \widehat{R}_{\idq}$. By Nakayama's Lemma, we get a contradiction.
\end{proof}
\subsection{Indecomposable injective objects of $\Qcoh(\bbX)$}
In this subsection, let $\bbX = (X, \kA)$ be a ncns.
First note the following standard result.
\begin{lemma}\label{L:LocalToGlobal} Let $U \stackrel{\imath}\longhookrightarrow X$ be an open subset.
Then the following statements are true.
\begin{itemize}
\item The direct image functor $\Phi_U = \imath_\ast: \Qcoh(\bbU) \lar
\Qcoh(\bbX)$ is fully faithful and maps (indecomposable) injective objects into (indecomposable) injective objects.
\item Assume that $U$ is affine and $x \in U$. Let $R = \Gamma(U, \kO)$, $A = \Gamma(U, \kA)$, $\idp \in \Spec(R)$ be the prime ideal corresponding to $x$ and
$A_x = A_{\idp}$. Then the functor $$A_x-\Mod \stackrel{\Phi_x}\lar \Qcoh(\bbX)$$ defined as the composition
$
A_x-\Mod \rightarrow A-\Mod \stackrel{\Phi_U}\lar \Qcoh(\bbX),
$
is fully faithful and maps (indecomposable) injective objects into (indecomposable) injective objects.
\item The functor $\Qcoh(\bbX) \lar A_x-\Mod$, assigning to a quasi--coherent $\kA$--module $\kF$ its stalk at the point $x$, is left adjoint
to $\Phi_x$. In particular, the functor $\Phi_x$ does not depend on the choice of an open affine neighbourhood of $x$.
\end{itemize}
\end{lemma}
\smallskip
\noindent
Results from
\cite[Section VI.2]{Gabriel} on Gabriel's recollement of locally noetherian abelian categories, combined with Corollary \ref{C:IndInjModules}, imply the following statement.
\begin{corollary}\label{P:IndInjGlobal}
Let $\Sp(\bbX)$ be the set of the isomorphism classes of indecomposable injective objects of $\Qcoh(\bbX)$. Then we have:
$$
\Sp(\bbX) = \bigsqcup\limits_{x \in X} \overline{\Sp}(A_x) = \bigsqcup\limits_{x \in X} \Max(A_x),
$$
where we view $I \in \overline{\Sp}(A_{x})$ as an element of $\Sp(\bbX)$ via the functor $\Phi_{x}$. In particular, we have a surjective map with finite fibers
\begin{equation}\label{E:Associator}
\Sp(\bbX) \stackrel{\alpha}\lar X,
\end{equation}
assigning to $\kI \in \Sp(\bbX)$ the point $x \in X$ such that
$\kI \cong \Phi_x(I)$ for some $I \in \overline{\Sp}(A_x)$.
\end{corollary}
\begin{lemma}\label{L:SuppInjSheaf} Let $\kI \in \Sp(\bbX)$ and $x = \alpha(\kI) \in X$. Then we have: $\mathsf{Supp}(\kI) = \overline{\{x\}}$.
\end{lemma}
\begin{proof} First note the following topological fact: a point $y \in X$ belongs to
$\overline{\{x\}}$ if and only if for any open neighbourhood $y \in U \subseteq X$ we have: $x \in U$.
\smallskip
\noindent
Let $y \in \overline{\{x\}}$. Consider any open affine neighbourhood $x,y \in V$ and put $R := \Gamma(V, \kO)$, $A := \Gamma(V, \kA)$. Let $\idp, \idq \in \Spec(R)$ be the prime ideals corresponding to $x$ and $y$, respectively.
Note that $I:= \Gamma(V, \kI)$ is an indecomposable injective $A$--module, whose associated prime ideal is $\idp$. It follows that $\idp \subseteq \idq$, hence
$I_{\idq} \ne 0$. Therefore, we have: $y \in \mathsf{Supp}(\kI)$.
\smallskip
\noindent
Assume now that $y \notin \overline{\{x\}}$. Then there exists an open affine subset
$V \subseteq X$ such that $y \in V$ and $x \notin V$.
Proposition \ref{P:Associator} implies that
for any open affine neighbourhood $x \in U$ we have:
$\Gamma(V, \kI) = \Gamma(V\cap U, \kI) = 0$, hence $\kI_y = 0$.
\end{proof}
\begin{proposition}\label{P:HomVanishingNonvanishing}
Let $\kI, \kJ \in \Sp(\bbX)$ be such that $\Hom_{\bbX}(\kI, \kJ) \ne 0$, Then we have: $y \in \overline{\{x\}}$, where $x = \alpha(\kI)$ and $y = \alpha(\kJ)$.
\smallskip
\noindent
Conversely, let $x, y \in X$ be such that $y \in \overline{\{x\}}$. Then there exist
$\kI, \kJ \in \Sp(\bbX)$ such that
$x = \alpha(\kI)$, $y = \alpha(\kJ)$ and $\Hom_{\bbX}(\kI, \kJ) \ne 0$.
\end{proposition}
\begin{proof}
Let $J \in \Sp(A_y)$ be such that $\kJ \cong \Phi_y(J)$. Then $
\Hom_{A_y}\bigl(\kI_y, J) \cong \Hom_{\bbX}(\kI, \kJ) \ne 0,
$
implying that $\kI_y \ne 0$. The first statement is proven.
\smallskip
\noindent
To show the second part, take any common open affine neighbourhood $x, y\in V$. Let
$R := \Gamma(V, \kO)$ and $A := \Gamma(V, \kA)$. Let $\idp, \idq \in \Spec(R)$ be the prime ideals corresponding to the points $x, y \in V$. Then we have:
$\idp \subseteq \idq$. According to Lemma \ref{L:mapsindinject}, there exist
$I, J \in \Sp(A)$ such that $\Hom_A(I, J) \ne 0$ and $\alpha(I) = \idp, \alpha(J) = \idq$. Let $\kI := \Phi_V(I)$ and $\kJ = \Phi_V(J)$. Then we have:
$\kI, \kJ \in \Sp(\bbX)$ and $\alpha(\kI) = x, \alpha(\kJ) = y$.
Moreover, since $\Phi_V$ is fully faithful, we have: $\Hom_{\bbX}(\kI, \kJ) \ne 0$.
\end{proof}
\begin{corollary}
For any $x \in X$ we put:
$
\kI(x) := \bigoplus\limits_{\substack{\kI \in \Sp(\bbX) \\ \alpha(\kI) = x}} \kI.
$
Then for any $x, y \in X$ we have:
$
\Hom_{\bbX}\bigl(\kI(x), \kI(y)\bigr) \ne 0$ if and only if
$y \in \overline{\{x\}}$.
\end{corollary}
\section{Proof of the Morita theorem for non--commutative noetherian schemes}
\noindent
Let $\bbX = (X, \kA)$ be a ncns. Since we focus on the study of the category $\Qcoh(\bbX)$, we additionally assume that $\bbX$ is
\emph{central}; see Remark \ref{R:RemarkCentrality}.
\subsection{Reconstruction of the central scheme} In this subsection we explain, how
the commutative scheme $(X, \kO)$ can be recovered from the category $\Qcoh(\bbX)$.
\begin{lemma}\label{L:KeyNoetherianLemma}
Let $\Lambda$ be a left noetherian ring, $e \in \Lambda$ an idempotent, $\Gamma = e \Lambda e$ and $F = e \Lambda f$, where $f = 1-e$. Then $F$ is noetherian viewed as a left $\Gamma$--module.
\end{lemma}
\begin{proof}
Let $\widetilde{\Gamma} = f \Lambda f$ and $\widetilde{F} = f \Lambda e$, then we have the Peirce decomposition
$
\Lambda =
\left(
\begin{array}{cc}
\widetilde{\Gamma} & \widetilde{F} \\
F & \Gamma
\end{array}
\right).
$
Assume that $F$ is not noetherian. Then there exists an infinite chain of left $\Gamma$--modules
$
F_1 \subsetneqq F_2 \subsetneqq \dots \subsetneqq F.
$
For any $n \in \NN$, put:
$
J_n :=
\left(
\begin{array}{cc}
\widetilde{F} F_n & \widetilde{F} \\
F_n & \Gamma
\end{array}
\right).
$
Then $J_n$ is a left ideal in $\Lambda$ and we get an infinite chain
$
J_1 \subsetneqq J_2 \subsetneqq \dots \subsetneqq \Lambda.
$
Contradiction.
\end{proof}
\begin{proposition}\label{P:PropositionReconstructionI}
For any non--empty finite subset $\Omega \subset \Sp(\bbX)$ we put:
$$
\kI(\Omega) := \bigoplus\limits_{\kI \in \Omega} \kI \quad \mbox{\rm and} \quad
A(\Omega):= \End_{\bbX}\bigl(\kI(\Omega)\bigr).
$$
Let $\Sp(\bbX) \stackrel{\alpha}\lar X$ be the map assigning to an indecomposable injective object of $\Qcoh(\bbX)$ its uniquely determined associated point of $X$ (see Corollary \ref{P:IndInjGlobal}).
Then the following statements are true.
\begin{enumerate}
\item If $\Omega = \alpha^{-1}(x)$ for some $x \in X$, then $A(\Omega)$ is noetherian and connected.
\item Conversely, if $A(\Omega)$ is noetherian and connected then we have:
$\big|\alpha(\Omega)\big| = 1$.
\item Let $\Omega$ be such that $A(\Omega)$ is noetherian and connected, but for any finite $\Omega \subsetneqq \widetilde\Omega$, the algebra
$A(\widetilde\Omega)$ does not have this property. Then $\Omega = \alpha^{-1}(x)$ for some $x \in X$.
\end{enumerate}
\end{proposition}
\begin{proof}
(1) Let $x \in X$ and $\Omega = \alpha^{-1}(x)$. By Proposition \ref{P:NoetherianEndom} and Lemma \ref{L:LocalToGlobal}, the algebra $A(\Omega)$ is a finite $\widehat{O}_x$--module, hence it is noetherian. Moreover, it is Morita--equivalent to the algebra
$\widehat{A}_x^\circ$, hence $\widehat{O}_x = Z(\widehat{A}_x^\circ) =
Z\bigl(A(\Omega)\bigr)$ (at this place we use that $\bbX$ is central). Since the center of a disconnected algebra can not be local, this imples that
$A(\Omega)$ is connected. The first statement is proven.
\smallskip
\noindent
(2) Now, let $\Omega \subset \Sp(\bbX)$ be a finite subset such that $\big|\alpha(\Omega)\big| \ge 2$. Choose any $x \in \Omega$ such that $x \notin \overline{\{y\}}$ for all $y \in \Omega \setminus\{x\}$. For any $\kI, \kJ \in \Omega$ such that $\alpha(\kI) = x$ and $\alpha(\kJ) \ne x$ we have: $\Hom_{\bbX}(\kJ, \kI) = 0$; see Proposition \ref{P:HomVanishingNonvanishing}. If the algebra $A(\Omega)$ is connected then there exist $\kI, \kJ \in \Omega$ such that $\alpha(\kI) = x$, $\alpha(\kJ) = y \ne x$ and $\Hom_{\bbX}(\kI, \kJ) \ne 0$.
Proposition \ref{P:HomVanishingNonvanishing} implies that
$y \in \overline{\{x\}}$.
\smallskip
\noindent
Next, there exists an idempotent $e \in A(\Omega)$ such that $e A(\Omega) e \cong
\End_{\bbX}(\kJ)$. Let $f = 1 -e$.
Then $\Hom_{\bbX}(\kI, \kJ)$ is a direct summand of
$e A(\Omega) f$ viewed as a left $\End_{\bbX}(\kJ)$--module.
Suppose now that $A(\Omega)$ is noetherian. Then Lemma \ref{L:KeyNoetherianLemma} implies that $\Hom_{\bbX}(\kI, \kJ)$ is a noetherian left $\End_{\bbX}(\kJ)$--module.
Let $V \subseteq X$ be an open affine subset such that $x, y \in V$
$R := \Gamma(V, \kO)$ and $A := \Gamma(V, \kA)$. Then there exist
$I, J \in \Sp(A)$ such that $\kI \cong \Phi_V(I)$ and $\kJ \cong \Phi_V(\kJ)$.
Moreover, $\Phi_V$ identifies the left $\End_A(J)$--module $\Hom_A(I, J)$ with
the left $\End_{\bbX}(\kJ)$--module $\Hom_{\bbX}(\kI, \kJ)$. However, since $x \ne y$, the associated prime ideals of $I$ and $J$ in the ring $R$ are different. Lemma \ref{L:KeyLemma} implies that
$\Hom_A(I, J)$
is not noetherian as a left $\End_A(J)$--module. Contradiction.
\smallskip
\noindent
(3) This statement is a consequence of the first two.
\end{proof}
\medskip
\noindent
Proposition \ref{P:PropositionReconstructionI} implies that the scheme $X$, viewed as a topological space, can be recovered from the category $\Qcoh(\bbX)$.
Our next goal is to explain the reconstruction of the structure sheaf of $X$.
For any closed subset $Z \subseteq X$ we put:
\begin{equation}
\Qcoh_{Z}(\bbX) = \left\{\kF \in \mathsf{Ob}\bigl(\Qcoh(\bbX)\bigr) \; \big|\;
\mathsf{Supp}(\kF) \subseteq Z\right\}.
\end{equation}
It is clear that $\Qcoh_{Z}(\bbX)$ is a Serre subcategory of $\Qcoh(\XX)$.
Let $U:= X\setminus Z \stackrel{\imath}\longhookrightarrow X$,
then the restriction functor
$\Qcoh(\bbX) \stackrel{\imath^\ast}\lar \Qcoh(\bbU)$ induces an equivalence of categories
$
\Qcoh(\bbX)/\Qcoh_Z(\bbX) \lar \Qcoh(\bbU).
$
Since $\imath^\ast$ admits a right adjoint functor $\imath_\ast$, $\Qcoh_{Z}(\bbX)$ is a localizing subcategory of $\Qcoh(\bbX)$.
\smallskip
\noindent
Recall that the localizing subcategories of an arbitrary locally noetherian abelian category $\sA$ stand in bijection with the subsets of the set $\Sp(\sA)$ of indecomposable injective objects of $\sA$; see \cite[Section III.4]{Gabriel}. Our next goal is to characterize in these terms the localizing subcategories $\Qcoh_{Z}(\bbX)$ for $Z \subseteq X$ closed.
\begin{proposition}\label{P:LocalizingSubcatViaIndInj}
Let $Z \subseteq X$ be a closed subset. Then we have:
$$
\Qcoh_{Z}(\bbX) = \left\{\kF \in \mathsf{Ob}\bigl(\Qcoh(\bbX)\bigr) \; \big|\;
\Hom_{\bbX}(\kF, \kI) = 0 \; \mbox{\rm for all} \; \kI \in \Sp(\bbX): \alpha(\kI) \in X \setminus Z\right\}.
$$
Let $Z = \overline{\left\{x_1, \dots, x_n\right\}}$, where $x_i \notin \overline{\left\{x_j\right\}}$ for all $1 \le i \ne j \le n$.
Then $\Qcoh_{Z}(\bbX)$ is the smallest localizing subcategory of $\Qcoh(\bbX)$ containing $\kE := \kI(x_1) \oplus \dots \oplus \kI(x_n)$.
\end{proposition}
\begin{proof} If $\kI \in \Sp(\bbX)$ is such that
$x := \alpha(\kI) \in X \setminus Z$ then $\kI \cong \Phi_x(I)$ for some $I \in
\Sp(A_x)$ and $\Hom_{\bbX}(\kF, \kI) \cong \Hom_{A_x}(\kF_x, I) = 0$ for any $\kF \in \Qcoh_{Z}(\bbX)$. Conversely, let
$\kI \in \Sp(\bbX)$ be such that $x := \alpha(\kI) \in Z$. Then $\kI \cong \Phi_x(I)$ for some $I \in \Sp(A_x)$. Let $P \in \Spec(A_x)$ be the associated prime ideal and
$\kF := \Phi_x(A_x/P)$. Then $\kF \in \Qcoh_{Z}(\bbX)$ and
$
\Hom_{\bbX}(\kF, \kI) \cong \Hom_{A_x}(A_x/P, I) \ne 0.
$
The first description of $\Qcoh_{Z}(\bbX)$ follows now from the correspondence between the localizing subcategories
of $\Qcoh(\bbX)$ and the subsets of $\Sp(\bbX)$; see \cite[page 377]{Gabriel}.
To prove the second statement, note that $\mathsf{Supp}(\kE) = Z$; see Lemma \ref{L:SuppInjSheaf}. Let $\sC(\kE)$ be the smallest localizing subcategory of
$\Qcoh(\bbX)$ containing $\kE$. Then it is a subcategory of $\Qcoh_Z(\bbX)$.
According to \cite[Section III.4]{Gabriel}, $\sC(\kE)$ corresponds to a certain subset $\Sigma$
of $\Sp(\bbX)$ containing $\Sigma_Z:= \left\{\kI \in \Sp(\bbX)\; \big|\; \alpha(\kI) \in X \setminus Z\right\}$. Let $\kJ \in \Sp(\bbX) \setminus \Sigma_Z$, i.e.
$\alpha(\kJ) \in Z$. Then Proposition \ref{P:HomVanishingNonvanishing} implies that
$\Hom_{\bbX}(\kE, \kJ) \ne 0$. This shows that $\Sigma = \Sigma_Z$, hence
$\sC(\kE) = \Qcoh_Z(\bbX)$, as asserted.
\end{proof}
\begin{theorem}\label{T:MorphismCentralSchemes} Let $\Qcoh(\bbX) \stackrel{\Phi}\lar \Qcoh(\bbY)$ be an equivalence of categories, where $\bbX = (X, \kA)$ and $\bbY = (Y, \kB)$ are two central ncns. Then there exists a unique isomorphism of schemes
$Y \stackrel{\Phi_c}\lar X$ such that the following diagram of sets
\begin{equation}\label{E:MapCentralSchemes}
\begin{array}{c}
\xymatrix{
\Sp(\bbX) \ar[rr]^-{\widetilde{\Phi}} \ar[d]_-{\alpha_X} & & \Sp(\bbY) \ar[d]^-{\alpha_Y} \\
X & & \ar[ll]_-{\Phi_c} Y
}
\end{array}
\end{equation}
is commutative, where $\widetilde{\Phi}$ is the bijection induced by $\Phi$.
\end{theorem}
\begin{proof} First note that $\alpha_X$ and $\alpha_Y$ are surjective. Hence, $\Phi_c$ is unique (even as a map of sets), provided it exists.
According to Proposition \ref{P:PropositionReconstructionI}, points of $X$ stand
in bijection with maximal finite subsets $\Omega \subset \Sp(\bbX)$, for which
the algebra $A(\Omega)$ is connected and noetherian (of course, a similar statement
is true for $\bbY$, too). This shows that there exists a unique bijection
$Y \stackrel{\Phi_c}\lar X$ making the diagram (\ref{E:MapCentralSchemes}) commutative.
Let $x \in X$ and $y = \Phi_c^{-1}(x)$. Proposition \ref{P:HomVanishingNonvanishing} implies that $\Phi_c^{-1}\bigl(\overline{\left\{x\right\}}\bigr) = \overline{\left\{y\right\}}$. Hence, the map $\Phi_c$ is continuous.
\smallskip
\noindent
Let $Z \subseteq X$ be any closed subset and $W:= \varphi^{-1}(Z)$. We put:
$U:= U \setminus Z$ and $V:= Y \setminus W$. Then we have a commutative diagram of categories and functors
$$
\xymatrix{
\Qcoh_Z(\bbX) \ar@{_{(}->}[d] \ar[rr]^-{\Phi_{\mid}} & & \Qcoh_W(\bbY) \ar@{^{(}->}[d] \\
\Qcoh(\bbX) \ar@{->>}[d] \ar[rr]^-{\Phi} & & \Qcoh(\bbY) \ar@{->>}[d] \\
\Qcoh(\bbU) \ar[rr]^-{\Phi_U} & & \Qcoh(\bbV)
}
$$
where $\Phi_{\mid}$ and $\Phi_U$ denote the restricted and induced equivalences of the corresponding categories. Let
$Z\bigl(\Qcoh(\bbU)\bigr) \stackrel{\psi_U}\lar Z\bigl(\Qcoh(\bbV)\bigr)$ be the map of centers induced by $\Phi_U$. The characterization of the subcategory
$\Qcoh_Z(\bbX)$ in the terms of indecomposable injective objects (see
Proposition \ref{P:LocalizingSubcatViaIndInj}) combined with Corollary \ref{C:Centers} imply that the collection of ring isomorphisms $(\psi_U)_{U \subseteq X}$ defines
a sheaf isomorphism $\kO_X \rightarrow (\Phi_c)_\ast\kO_Y$. Hence, $Y \stackrel{\Phi_c}\lar X$ is an isomorphism of schemes, as asserted.
\end{proof}
\smallskip
\noindent
\textbf{Summary} (Reconstruction of the central scheme). Let $\bbX = (X, \kA)$ be a central ncns.
\begin{itemize}
\item Consider the set $S = S(\bbX)$, whose elements are maximal finite subsets $\Omega \subset
\Sp(\bbX)$ such that the algebra $A(\Omega)$ is noetherian and connected.
\item Define the topology on $S$ by the following rules:
\begin{itemize}
\item For any $\Omega', \Omega'' \in S(\bbX)$ we say that
$$\Omega'' \in \overline{\{\Omega'\}} \quad \mbox{\rm if and only if}\quad
\Hom_{\bbX}\bigl(\kI(\Omega'), \kI(\Omega'')\bigr) \ne 0.
$$
\item By definition, any non--trivial closed subset of $S$ has the form
$$
Z(\Omega_1, \dots, \Omega_n) := \bigcup\limits_{i = 1}^n \overline{\{\Omega_i\}},
$$
where $\Omega_1, \dots, \Omega_n \in S$ are such that $\Omega_i \notin
\overline{\{\Omega_j\}}$ for all $1 \le i \ne j \le n$.
\end{itemize}
\item For any open subset $U = U(\Omega_1, \dots, \Omega_n) := S \setminus Z(\Omega_1, \dots, \Omega_n)$, let $\sC(Z)$ be the smallest localizing subcategory of $\Qcoh(\bbX)$ containing $\kI(\Omega_1) \oplus \dots \oplus \kI(\Omega_n)$ and
$
\Gamma(U, \kR) := Z\bigl(\Qcoh(\bbX)/\sC(Z)\bigr)
$
\item Finally, let $\widetilde{U} \subseteq U$ be a pair of open subsets, $Z := X \setminus U$, $\widetilde{Z} := X \setminus \widetilde{U}$ and
$\Gamma(U, \kR) \rightarrow \Gamma(\widetilde{U}, \kR)$ be the map of categorical centers
induced by the localization functor $\Qcoh(\bbX)/\sC(Z) \rightarrow \Qcoh(\bbX)/\sC(\widetilde{Z}).$
\end{itemize}
Then $S$ is a topological space and $\kR$ is a sheaf of commutative rings on
$S$. Moreover, the ringed spaces $(X, \kO)$ and
$(S, \kR)$ are isomorphic. The corresponding isomorphism of topological spaces $X \rightarrow S(\bbX)$ is given by the rule $x \mapsto \alpha^{-1}(x)$, where $\alpha$ is the map given by (\ref{E:Associator}). In other words, the underlying commutative scheme $X$ of a central non--commutative noetherian scheme $\bbX$ can be recovered from the category $\Qcoh(\bbX)$. This provides a generalization of the classical reconstruction result of Gabriel \cite[Section VI.3]{Gabriel} on the non--commutative setting.
\subsection{Proof of Morita theorem}
For a ncns $\bbX = (X, \kA)$ we put: $\bbX^\circ := (X, \kA^\circ)$. Next,
$\VB(\bbX)$ (respectively, $\VB(\bbX^\circ)$) will denote the category of coherent sheaves on $\bbX$ (respectively, $\bbX^\circ$) which are locally projective over $\kA$ (respectively, over $\kA^\circ$).
\begin{definition}
Let $\bbX$ be a ncns. Then $\kP \in \VB(\bbX^\circ)$ is a \emph{local right progenerator} of $\bbX$ if $\add(\kP_x) = \add(A_x)$ for all $x \in X$.
\end{definition}
\begin{theorem}\label{T:MoritaTheoremNCNS}
Let $\Qcoh(\bbX) \stackrel{\Phi}\lar \Qcoh(\bbY)$ be an equivalence of categories, where $\bbX = (X, \kA)$ and $\bbY = (Y, \kB)$ are central ncns. Then there exist
a pair $(\kP, \vartheta)$, where
\begin{itemize}
\item $\kP \in \VB(\bbX^\circ)$ is a local right progenerator
\item $\kB \stackrel{\vartheta}\lar
\varphi^\ast\bigl(\mathit{End}_{\kA}(\kP)\bigr)$ is an isomorphism of $\kO_Y$--algebras
\end{itemize}
such that $\Phi \cong \Phi_{\kP, \vartheta, \varphi} :=
\vartheta^{\sharp}\cdot \varphi^\ast \cdot \bigl(\kP \otimes_\kA \,-\, \bigr)$, where $\varphi = \Phi_c: Y \lar X$ is the scheme isomorphism induced by the equivalence $\Phi$ (see Theorem \ref{T:MorphismCentralSchemes}) and $\vartheta^{\sharp}$ is the equivalence of categories induced by $\vartheta$.
\smallskip
\noindent
If $(\kP', \vartheta')$ is another pair representing $\Phi$ (i.e $\Phi \cong \Phi_{\kP', \vartheta', \varphi}$)
then there exists a unique isomorphism $\kP \stackrel{f}\lar \kP'$ in $\VB(\bbX^\circ)$ such that the diagram
$$
\xymatrix{
& \kB \ar[ld]_-{\vartheta} \ar[rd]^-{\vartheta'} & \\
\varphi^\ast\bigl(\mathit{End}_{\kA}(\kP)\bigr) \ar[rr]^-{\varphi^\ast(Ad_f)} & & \varphi^\ast\bigl(\mathit{End}_{\kA}(\kP')\bigr)
}
$$
is commutative.
\smallskip
\noindent
Conversely, if $Y \stackrel{\varphi}\lar X$ is an isomorphism of schemes, $\kP \in \VB(\bbX^\circ)$ is a local right progenerator and $\kB \stackrel{\vartheta}\lar
\varphi^\ast\bigl(\mathit{End}_{\kA}(\kP)\bigr)$ is an isomorphism of $\kO_Y$--algebras
then $\Phi:= \Phi_{\kP, \vartheta, \varphi}$ is an equivalence of categories such that $\Phi_c = \varphi$.
\end{theorem}
\begin{proof} The last part of the theorem is obvious. Hence, it is sufficient to prove the following
\smallskip
\noindent
\underline{Statement}. Let $\Qcoh(\bbX) \stackrel{\Phi}\lar \Qcoh(\widetilde\bbX)$ be a central equivalence of categories, where $\bbX = (X, \kA)$ and $\widetilde\bbX = (X, \widetilde\kA)$ are two central ncns with the same underlying commutative scheme $X$. Then $\Phi \cong \kP \otimes_\kA \,-$, where $\kP \in \VB(\bbX^\circ)$
is a balanced central $(\widetilde\kA-\kA)$--bimodule which is a local right progenerator of $\bbX$.
\smallskip
\noindent
\underline{Claim}.
For any open subset
$U \stackrel{\imath}\longhookrightarrow X$ put: $
\Phi_U:= \imath^\ast \Phi \imath_\ast: \Qcoh(\bbU) \lar
\Qcoh(\widetilde\bbU).$
Then $\Phi_U$ is an equivalence of categories and in the following diagram of categories and functors
\begin{equation}\label{E:inducedEq}
\begin{array}{c}
\xymatrix{
\Qcoh(\bbX) \ar[rr]^-{\imath^\ast} \ar[d]_-{\Phi} & & \Qcoh(\bbU)
\ar[d]^-{\Phi_U} \\
\Qcoh(\widetilde{\bbX}) \ar[rr]^-{\imath^\ast} & & \Qcoh(\widetilde\bbU)
}
\end{array}
\end{equation}
both compositions of functors are isomorphic.
\smallskip
\noindent
Indeed, let $Z:= X \setminus U$. Then
$\Phi$ restricts to an equivalence of the categories
$\Qcoh_Z(\bbX) \rightarrow \Qcoh_Z(\widetilde{\bbX})$ (at this place, we use \emph{centrality} of $\Phi$). The universal property of the Serre quotient category implies that there exists an equivalence of categories $\Qcoh(\bbU) \stackrel{\Psi_U}\lar \Qcoh(\widetilde\bbU)$ such that $\Psi_U \imath^\ast \cong \imath^\ast \Phi$.
Since $\imath^\ast \imath_\ast = \mathsf{Id}_{\bbU}$, we conclude that $\Phi_U \cong \Psi_U$, hence $\Phi_U$ is an equivalence of categories. One can check that the natural transformation
$
\imath^\ast \Phi \stackrel{\zeta_U}\lar \Phi_U \imath^\ast,
$
induced by the adjunction unit $\mathsf{Id}_{\bbX} \rightarrow \imath_\ast \imath^\ast$, is an isomorphism. This proves the claim.
\smallskip
\noindent
Let $V \stackrel{\varepsilon}\longhookrightarrow U$ be an open subset and
$\jmath:= \imath \varepsilon$. Since $\imath_\ast \varepsilon_\ast = \jmath_\ast$ and
$\varepsilon^\ast \imath^\ast = \jmath^\ast$, we conclude:
\begin{equation}\label{E:EquivRestricted}
\Phi_V = \varepsilon^\ast \Phi_U \varepsilon_\ast.
\end{equation}
\smallskip
\noindent
Assume now that $U$ is affine. Then there exists a central $(\widetilde\kA_U-\kA_U)$--Morita bimodule $\kP^U \in \VB(\bbU^\circ)$ and an isomorphism of functors
$
\Phi_U \stackrel{\xi_U}\lar \kP^U \otimes_{\kA_U} \,-\,.
$
Then for any $\kG \in \Qcoh(\bbV)$, we get natural isomorphisms
\begin{equation}\label{E:SomeNatTransf}
\varepsilon^\ast \Phi_U \varepsilon_\ast(\kG) \lar
\varepsilon^\ast\bigl(\kP^U \otimes_{\kA_U} \varepsilon_\ast(\kG)\bigr) \lar
\kP^U\Big|_{V} \otimes_{\kA_V} \kG,
\end{equation}
where we use that $\varepsilon^\ast \varepsilon_\ast = \mathsf{Id}_{\bbV}$.
\smallskip
\noindent
Let $\kP^U\big|_{V} \otimes_{\kA_V} \,-\, \xrightarrow{\widetilde{\sigma}^U_V}
\kP^V \otimes_{\kA_V} \,-\,$ be the unique isomorphism of functors making the following diagram of functors and natural transformations
\begin{equation}\label{E:MoritaCompat}
\begin{array}{c}
\xymatrix{
\varepsilon^\ast \Phi_U \varepsilon_\ast \ar[rr]^-{\xi_U\big|_V} \ar[d]_{=} & & \kP^U\big|_{V} \otimes_{\kA_V} \,-\, \ar[d]^-{\widetilde{\sigma}^U_V} \\
\Phi_V \ar[rr]^-{\xi_V} & & \kP^V \otimes_{\kA_V} \,-\,
}
\end{array}
\end{equation}
commutative, where $\xi_U\big|_V$ is the isomorphism of functors defined by (\ref{E:SomeNatTransf}). According to Theorem \ref{T:MoritaClassical} (classical Morita theorem for rings), there exists a uniquely determined isomorphism of $(\widetilde\kA_V-\kA_V)$--bimodules
$
\kP^U\big|_{V} \stackrel{\sigma^U_V}\lar \kP^V,
$
which induces the natural transformation $\widetilde{\sigma}^U_V$. It follows from
(\ref{E:EquivRestricted}) and (\ref{E:MoritaCompat}) that for any triple
$W \subseteq V \subseteq U$ of open affine subsets of $X$, the following diagram
$$
\xymatrix{
\bigl(\kP^U\big|_V\bigr)\Big|_{W} \ar[rr]^-{(\sigma^U_V)\big|_{W}} \ar[d]_= & &
\kP^V\big|_{W} \ar[d]^-{\sigma^V_W} \\
\kP^U\big|_W \ar[rr]^-{\sigma^U_W} & & \kP^W
}
$$
is commutative. Hence, there exists an $(\widetilde\kA-\kA)$--bimodule $\kP$
and a family of isomorphisms of $(\widetilde\kA_U-\kA_U)$--bimodules $\kP^U \stackrel{\varphi_U}\lar \kP\big|_U$ (for any $U \subseteq X$ open and affine) such that
$$
\xymatrix{
\kP^U\big|_V \ar[rr]^-{(\varphi_U)\big|_{V}} \ar[d]_-{\sigma^U_V} & &
\bigl(\kP\big|_{U}\bigr)\big|_{V} \ar[d]^-{=}\\
\kP^V \ar[rr]^-{\varphi_V} & & \kP\big|_{V}
}
$$
is commutative for any pair of open affine subsets $V \subseteq U$ of $X$. It is clear that
$\kP$ is a central balanced $(\widetilde\kA-\kA)$--bimodule such that
$\add(\kP_x) = \add(A_x)$ for any $x \in X$. Moreover, the datum
$\bigl(\kP, (\varphi_U)_{U \subseteq X}\bigr)$ is unique up to a unique isomorphism.
\smallskip
\noindent
Finally, for any $\kF \in \Qcoh(\bbX)$ and $U \subseteq X$ open and affine, we have an isomorphism $\Phi(\kF)\big|_{U} \lar \bigl(\kP \otimes_\kA \kF\bigr)\big|_U$ defined as the composition
$$
\Phi(\kF)\big|_{U} \stackrel{\zeta^\kF_U}\lar \Phi_U\bigl(\kF\big|_U\bigr)
\xrightarrow{\xi_U^{\kF\big|_U}} \kP^U \otimes_{\kA_U} \kF\big|_U
\xrightarrow{\varphi_U \otimes \mathsf{id}} \kP\big|_U \otimes_{\kA_U} \kF\big|_U
\stackrel{\mathsf{can}}\lar \bigl(\kP \otimes_\kA \kF\bigr)\big|_U.
$$
It follows that these isomorphisms are compatible with restrictions on open affine subsets and define a global isomorphism of left $\widetilde\kA$--modules
$\Phi(\kF) \stackrel{\vartheta^\kF}\lar \kP \otimes_\kA \kF$, which is natural in $\kF$. Hence, we have constructed an isomorphism of functors $\Phi \cong \kP \otimes_\kA \,-$ we were looking for. The uniqueness of $\kP$ follows from the corresponding result in the affine case.
\end{proof}
\begin{remark}
In the case when $\bbX = (X, \kA)$ and $\bbY = (Y, \kB)$ are ncns with $X$ and $Y$
being locally of finite type over a field $\kk$, some related results about equivalences between the categories $\Coh(\bbX)$ and $\Coh(\bbY)$ can also be found in \cite[Section 6]{ArtinZhang}.
\end{remark}
\section{C\u{a}ld\u{a}raru's conjecture on Azumaya algebras on noetherian schemes}
\noindent
Let $X$ be a noetherian scheme and $\kA$ be a sheaf of $\kO$--algebras, which is a locally free coherent sheaf of finite rank on $X$. Then we have a canonical morphism of $\kO$--algebras
$
\kA \otimes_\kO \kA^\circ \stackrel{\mu}\lar \mathit{End}_{\kO}(\kA),
$
given on the level of local sections by the rule $a \otimes b \mapsto (c \mapsto a cb)$. Recall that $\kA$ is an \emph{Azumaya algebra} on $X$ if $\mu$ is an isomorphism. This is equivalent to the condition that $\kA\big|_x := A_x \otimes_{O_x} \bigl(O_x/\idm_x\bigr)$ is a central simple $O_x/\idm_x$--algebra for any point $x \in X$; see \cite[Proposition IV.2.1]{Milne}. It follows that an Azumaya algebra on $X$ is automatically central. Moreover, for any pair of Azumaya algebras $\kA_1$ and
$\kA_2$ on $X$, their tensor product $\kA_1 \otimes_\kO \kA_2$ is again an Azumaya
algebra.
\smallskip
\noindent
Let $\kA$ be an Azumaya algebra on $X$, $\bbX = (X, \kA)$ and $\kP \in \VB(\bbX^\circ)$ be a local right progenerator of $\bbX$. Then $\widetilde\kA := \mathit{End}_\kA(\kP)$ is again an Azumaya algebra on $X$ (since $\widetilde{\kA}\big|_x$ is again a central simple $O_x/\idm_x$--algebra algebra for any $x \in X$). If
$\widetilde{\bbX} = (X, \widetilde\kA)$ then
$$
\Qcoh(\bbX) \xrightarrow{\kP \otimes_\kA \,-\,} \Qcoh(\widetilde\bbX)
$$
is a central equivalence of categories.
\smallskip
\noindent
Let $\kA$ and $\kB$ be two Azumaya algebras on $X$. We put: $\bbX = (X, \kA)$ and $
\widetilde\bbX = (X, \kB)$.
\begin{itemize}
\item $\kA$ and $\kB$ are centrally Morita equivalent (denoted $\kA \approx \kB$) if there exists a central equivalence of categories
$\Qcoh(\bbX) \rightarrow \Qcoh(\widetilde\bbX)$
\item $\kA$ and $\kB$ are \emph{equivalent} (denoted $\kA \sim \kB$) provided there exist $\kF, \kG \in \VB(X)$ such that
$\mathit{End}_{\kO}(\kF) \otimes_\kO \kA$ and $\mathit{End}_{\kO}(\kG)
\otimes_\kO \kB$ are isomorphic as $\kO$--algebras.
\end{itemize}
The second relation is indeed an equivalence relation. The set $\mathsf{Br}(X)$ of equivalence classes of Azumaya algebras on $X$, endowed with the operation
$$
[\kA_1] + [\kA_2] := \bigl[\kA_1 \otimes_\kO \kA_2\bigr]
$$
is a commutative group (called \emph{Brauer group} of the scheme $X$). Recall that
the class $[\kO]$ is the neutral element of $\mathsf{Br}(X)$, whereas
$-[\kA] = [\kA^\circ]$; see \cite[Section IV.2]{Milne}.
\begin{lemma}\label{L:LemmaAzumayaEq}
Let $\kA, \kB$ be Azumaya algebras on $X$ such that $\kA \approx \kB$. Then for any Azumaya algebra $\kC$ on $X$ we have:
$\kA \otimes_\kO \kC \approx \kB \otimes_\kO \kC$.
\end{lemma}
\begin{proof}
Since $\kA \approx \kB$, there exists a local right progenerator $\kP$ for $\kA$ and an isomorphism of $\kO$--algebras $\kB \stackrel{\vartheta}\lar \mathit{End}_\kA(\kP)$. Then we get an induced isomorphism of $\kO$--algebras $\kB \otimes_\kO \kC
\stackrel{\tilde\vartheta}\lar \mathit{End}_{\kA\otimes_\kO \kC}(\kP\otimes_\kO \kC)$ given as the composition
$$
\kB \otimes_\kO \kC \xrightarrow{\vartheta \otimes \mathsf{id}} \mathit{End}_\kA(\kP)
\otimes_\kO \kC \lar \mathit{End}_{\kA\otimes_\kO \kC}(\kP\otimes_\kO \kC).
$$
Note that $\kP\otimes_\kO \kC$ is a local right progenerator for $\kA\otimes_\kO \kC$. Moreover, $\kO$ is the center of $\kA \otimes_\kO \kC$ and
$\kB \otimes_\kO \kC$ (since they both are Azumaya algebras), implying the statement.
\end{proof}
\smallskip
\noindent
The proof of the following result is basically a replica of
\cite[Theorem 1.3.15]{CaldararuThesis}.
\begin{proposition}\label{P:Azumaya}
Let $\kA$ and $\kB$ be two Azumaya algebras on $X$. Then we have:
$$
\kA \sim \kB \quad \mbox{\rm if and only if} \quad \kA \approx \kB.
$$
\end{proposition}
\begin{proof} Let $\bbX = (X, \kA)$ and $\bbY = (X, \kB)$
If $\kA \sim \kB$ then there exist $\kF, \kG \in \VB(X)$ and an isomorphism of $\kO$--algebras
$
\mathit{End}_{\kO}(\kF) \otimes_\kO \kA \stackrel{\vartheta}\lar \mathit{End}_{\kO}(\kG)
\otimes_\kO \kB.
$
Let $\kP := \kF \otimes_\kO \kA$ and $\kQ := \kG \otimes_\kO \kB$. Then $\kP$ is a local right progenerator for $\kA$ and $\kQ$ is a local right progenerator for $\kB$.
Let $\widetilde\kA:= \mathit{End}_{\kA}(\kP)$ and $\widetilde\kB:= \mathit{End}_{\kB}(\kQ)$. Then we have isomorphisms of $\kO$--algebras
$\widetilde\kA \cong \mathit{End}_{\kO}(\kF) \otimes_\kO \kA$ and
$\widetilde\kB \cong \mathit{End}_{\kO}(\kG) \otimes_\kO \kB$ as well as central equivalences of categories
$$
\Qcoh(\bbX) \xrightarrow{\kP \otimes_\kA \,-\,} \Qcoh(\widetilde\bbX)
\xleftarrow{\vartheta^\sharp} \Qcoh(\widetilde\bbY)
\xleftarrow{\kQ \otimes_\kB \,-\,}
\Qcoh(\bbY),
$$
where $\widetilde\bbX = (X, \widetilde\kA)$ and
$\widetilde\bbY = (X, \widetilde\kB)$. Hence, $\kA \approx \kB$.
\smallskip
\noindent
Conversely, assume that $\kA \approx \kB$. Then we have:
$$
\kO \approx \mathit{End}_{\kO}(\kA) \cong \kA \otimes_\kO \kA^\circ \approx \kB \otimes_\kO \kA^\circ,
$$
where the last central equivalence exists by Lemma \ref{L:LemmaAzumayaEq}. Hence, there exists $\kF \in \VB(X)$ and an isomorphism of $\kO$--algebras
$\mathit{End}_{\kO}(\kF) \stackrel{\vartheta}\lar \kB \otimes_\kO \kA^\circ$. Then we get the following induced isomorphism of $\kO$--algebras:
$$
\mathit{End}_{\kO}(\kF) \otimes_\kO \kA \xrightarrow{\vartheta \otimes \mathsf{id}}
\kB \otimes_\kO \kA^\circ \otimes_\kO \kA \stackrel{\mathsf{can}}\lar
\kA \otimes_\kO \kA^\circ \otimes_\kO \kB \xrightarrow{\mu \otimes \mathsf{id}} \mathit{End}_{\kO}(\kA) \otimes_\kO \kB.
$$
Hence, $\kA \sim \kB$, as asserted.
\end{proof}
\begin{theorem}\label{T:CaldararuConjecture}
Let $X$ and $Y$ be two separated noetherian schemes, $\kA$ be an Azumaya algebra on $X$, $\kB$ be an Azumaya algebra on $Y$, $\bbX = (X, \kA)$ and $\bbY = (Y, \kB)$.
Then the categories $\Qcoh(\bbX)$ and $\Qcoh(\bbY)$ are equivalent if and only if there exists an isomorphism of schemes $Y \stackrel{f}\lar X$ such that
$f^\ast\bigl([\kA]\bigr) = [\kB] \in \mathsf{Br}(Y)$.
\end{theorem}
\begin{proof}
If $Y \stackrel{f}\lar X$ is such that
$f^\ast\bigl([\kA]\bigr) = [\kB] \in \mathsf{Br}(X)$ then equivalence of $\Qcoh(\bbX)$ and $\Qcoh(\bbY)$ is a consequence of Proposition \ref{P:Azumaya}.
\smallskip
\noindent
Conversely, let $\Qcoh(\bbX) \stackrel{\Phi}\lar \Qcoh(\bbY)$ be an equivalence of categories. Since both ncns $\bbX$ and $\bbY$ are central, Theorem \ref{T:MorphismCentralSchemes} yields an induced isomorphism of schemes
$Y \stackrel{f}\lar X$, where $f = \Phi_c$. It follows that $\kB \approx f^\ast(\kA)$. By Proposition \ref{P:Azumaya} we get: $\kB \sim f^\ast(\kA)$.
\end{proof}
\begin{remark}
Theorem \ref{T:CaldararuConjecture} was conjectured by C\u{a}ld\u{a}raru in
\cite[Conjecture 1.3.17]{CaldararuCrelle}. In the case of smooth projective varieties over a field, it was proved by Canonaco and Stellari \cite[Corollary 5.3]{CanonacoStellari}. In the full generality, C\u{a}ld\u{a}raru's conjecture was proven by Antieau \cite[Theorem 1.1]{Antieau}, based on a previous work of Perego \cite{Perego} and the theory of derived Azumaya algebras of To\"en \cite{Toen}. In our opinion, the given proof of C\u{a}ld\u{a}raru's conjecture (in which Theorem \ref{T:MorphismCentralSchemes} plays a key role) is significantly simpler.
\end{remark}
\section{Local modification theorem}
\noindent
In this section, let $R$ be a connected reduced excellent commutative ring of Krull dimension one and $K = \Quot(R)$ be its total ring of fractions (which is isomorphic to a finite product of fields). For any $\idm \in \Max(R)$, we have a multiplicatively closed set $S_{(\idm)} := R \setminus \{\idm\} \subset R$. Then we have: $K_{\idm} := S_{(\idm)}^{-1} K \cong \Quot(R_{\idm})$. Next, we have a commutative diagram of rings and canonical ring homomorphisms
$$
\xymatrix{
R \ar[r] \ar@{_{(}->}[d] & R_{\idm} \ar[r] \ar@{_{(}->}[d] & \widehat{R}_{\idm} \ar@{^{(}->}[d] \\
K \ar[r] & K_{\idm} \ar[r] & \widehat{K}_{\idm}
}
$$
where $\widehat{K}_{\idm} = \Quot(\widehat{R}_{\idm})$. Since $R$ is excellent, the completion $\widehat{R}_{\idm}$ is a reduced ring (see e.g. \cite{Dieudonne}) and $\Quot(\widehat{R}_{\idm})$ is isomorphic to a finite product of fields. Note that the canonical ring homomorphism $K \otimes_R \widehat{R}_{\idm}
\rightarrow \widehat{K}_{\idm}$ is an isomorphism.
\smallskip
\noindent
Let $U$ be a finitely generated $K$--module and $L \subset U$ a finitely generated $R$--submodule such that $K \cdot L = U$. Then $L$ is automatically a torsion free $R$--module and the canonical map $K \otimes_R L \rightarrow U$ is an isomorphism. We shall also say that $L$ is an \emph{$R$--lattice} in (its \emph{rational envelope})
$U$.
\smallskip
\noindent
For any $\idm \in \Max(R)$ we put:
$U_{\idm} = K_{\idm} \otimes_K U$ and $\widehat{U}_{\idm} := \widehat{K}_{\idm} \otimes_K U$ as well as
$L_{\idm} := R_{\idm} \otimes_R L$ and
$\widehat{L}_{\idm} := \widehat{R}_{\idm} \otimes_R L$. Using the canonical
ring homomorphisms $K \otimes_R R_{\idm}
\rightarrow K_{\idm}$ and $K \otimes_R \widehat{R}_{\idm}
\rightarrow \widehat{K}_{\idm}$,
we can view $L_{\idm}$ as an $R_{\idm}$--lattice in $U_{\idm}$ and
$\widehat{L}_{\idm}$ as an $\widehat{R}_{\idm}$--lattice in
$\widehat{U}_{\idm}$. Next, we have the following commutative diagram:
\begin{equation}
\begin{array}{c}
\xymatrix{
L \ar[r] \ar@{_{(}->}[d] &
L_{\idm} \ar@{^{(}->}[r] \ar@{_{(}->}[d]^-{\varepsilon_{\idm}} &
\widehat{L}_{\idm} \ar@{^{(}->}[d]^-{\hat{\varepsilon}_{\idm}} \\
U \ar[r]^-{\vartheta_{\idm}} \ar@/_5.5ex/[rr]_-{\hat{\vartheta}_{\idm}} &
U_{\idm} \ar@{^{(}->}[r] & \widehat{U}_{\idm}
}
\end{array}
\end{equation}
in which all maps are the canonical ones.
\begin{lemma}\label{L:LatticeRationalLocal} In the above notation we have:
\begin{equation}\label{E:LatticeRationalLocal}
L = \left\{
x \in U \, \big| \, \hat{\vartheta}_{\idm}(x) \in \mathsf{Im}(\hat{\varepsilon}_{\idm})
\; \mbox{\rm for all} \; \idm \in \Max(R)
\right\}.
\end{equation}
\end{lemma}
\begin{proof} For any $\idm \in \Max(R)$ we have: $L_{\idm} = \widehat{L}_{\idm} \cap U_{\idm}$, where the intersection is taken inside $\widehat{U}_{\idm}$; see for instance \cite[Theorem 5.2]{ReinerMO}. Hence, it is sufficient to show that
$$
L = \widetilde{L} := \left\{
x \in U \, \big| \, \vartheta_{\idm}(x) \in \mathsf{Im}(\varepsilon_{\idm})
\; \mbox{\rm for all} \; \idm \in \Max(R)
\right\}.
$$
It is clear that $L \subseteq \widetilde{L}$, hence we only need to prove the opposite inclusion. Let $x \in \widetilde{L}$ and
$
I:= \left\{
a \in R \, \big| \, a x \in L
\right\}.
$
By definition of $\widetilde{L}$, for any $\idm \in \Max(R)$ there exists
$t \in S_{(\idm)}$ such that $t x \in L$. Since $t \in I \setminus \idm$, we conclude that $I \not\subset \idm$ for any $\idm \in \Max(R)$. As a consequence,
$I = A$ and $x \in L$, as asserted.
\end{proof}
\begin{theorem}[Local modification theorem]\label{T:LocalModification} Let $U$ be a finitely generated $K$--module, $L \subset U$ an $R$--lattice and $\Omega \subset \Max(R)$ a finite subset such that for any $\idm \in \Omega$ we are given an
$\widehat{R}_{\idm}$--lattice $N(\idm) \subset \widehat{U}_{\idm}$. Then there exists a unique lattice $N \subset U$ (\emph{local modification} of $L$) such that for
any $\idm \in \Max(R)$ we have:
\begin{equation}\label{E:prescribedcompletions}
\widehat{N}_{\idm} =
\left\{
\begin{array}{cc}
\widehat{L}_{\idm} & \mbox{\rm if} \; \idm \notin \Omega \\
N(\idm) & \mbox{\rm if} \; \idm \in \Omega,
\end{array}
\right.
\end{equation}
where $\widehat{N}_{\idm}$ is viewed as a subset of $\widehat{U}_{\idm}$.
\end{theorem}
\begin{proof} In order to prove the existence of $N$, we first consider the following special
\noindent
\underline{Case 1}. Suppose that $N(\idm) \subseteq \widehat{L}_{\idm}$ for all $\idm \in \Omega$. Since the $\widehat{R}_{\idm}$--modules $N(\idm)$ and $\widehat{L}_{\idm}$ have the same rational envelope $\widehat{U}_{\idm}$, the $\widehat{R}_{\idm}$--module $T(\idm):= \widehat{L}_{\idm}/N(\idm)$ has finite length. Consequently, $T(\idm)$ has finite length viewed as an $R$--module and $\mathsf{Supp}\bigl(T(\idm)\bigr) = \{\idm\}$. We have a surjective homomorphism of $R$--modules $L \stackrel{c(\idm)}\lar
T(\idm)$ given as the composition
$
L \rightarrow \widehat{L}_{\idm} \dhrightarrow T(\idm).
$
Let $T = \oplus_{\idm \in \Omega} T(\idm)$. Then we get an $R$--module homomorphism $L \stackrel{c}\lar T$, whose components are the maps $c(\idm)$ defined above. By construction, the map $\widehat{L}_{\idm} \stackrel{\hat{c}_{\idm}}\lar
\widehat{T}_{\idm}$ is surjective for all $\idm \in \Max(R)$. As consequence, the map $c$ is surjective, too. Let $N$ be the kernel of $c$. Then $N$ is a noetherian $R$--module, $K \cdot N = K\cdot L = U$ and completions
$\widehat{N}_{\idm} \subset \widehat{U}_{\idm}$
are given by the formula (\ref{E:prescribedcompletions}) for any $\idm \in \Max(R)$.
\smallskip
\noindent
\underline{Case 2}. Consider now the general case, where $N(\idm) \subset \widehat{U}_{\idm}$ is an arbitrary $\widehat{R}_{\idm}$--lattice for $\idm \in \Omega$. Since $\widehat{K}_{\idm} \cdot \bigl(\widehat{L}_{\idm} + N(\idm)\bigr) =
\widehat{K}_{\idm} \cdot \widehat{L}_{\idm} = \widehat{U}_{\idm}$, the
$\widehat{R}_{\idm}$--module
$X(\idm):= \bigl(\widehat{L}_{\idm} + N(\idm)\bigr)/\widehat{L}_{\idm}$ has finite length. It follows that $X(\idm)$ has also finite length viewed as $R$--module and $\mathsf{Supp}\bigl(X(\idm)\bigr) = \{\idm\}$. Hence, there exists
$l \in \NN$ such that $\idm^l \cdot X(\idm) = 0$.
\smallskip
\noindent
Let $\bigl\{\idp_1, \dots, \idp_r\}$ be the set of minimal prime ideals of $R$.
Then $D:= \idp_1 \cup \dots \cup \idp_r$ is the set of zero divisors of $R$. By prime avoidance, $\idm^l \not\subseteq D$, hence there exists $a^{(\idm)} \in \idm^l \setminus D$ such that $a^{(\idm)} X(\idm) = 0$, i.e.
$a^{(\idm)} N(\idm) \subseteq \widehat{L}_{\idm}$. Let $a := \prod_{\idm \in \Omega} a^{(\idm)}$. Then $a$ is a regular element in the ring $R$ and $a N(\idm) \subseteq \widehat{L}_{\idm}$ for any $\idm \in \Omega$.
\smallskip
\noindent
Now we put $M:= \frac{1}{a} L \subset U$. Then $M$ is a lattice in $U$, $L \subseteq M$ and $N(\idm) \subseteq \widehat{L}_{\idm}$ for any $\idm \in \Omega$.
Let
$
\Sigma :=
\bigl\{
\idm \in \Max(R) \, \big| \, a \in \idm\bigr\} = \mathsf{Supp}\bigl(R/(a)\bigr).
$
Then according to Case 1, there exists a sublattice $N \subseteq M$ such that
\begin{equation*}
\widehat{N}_{\idm} =
\left\{
\begin{array}{ll}
\widehat{M}_{\idm} & \mbox{\rm if} \; \idm \notin \Sigma \cup \Omega \\
\widehat{L}_{\idm} & \mbox{\rm if} \; \idm \in \Sigma \setminus \Omega \\
N(\idm) & \mbox{\rm if} \; \idm \in \Omega.
\end{array}
\right.
\end{equation*}
This proves the existence of an $R$--lattice $N \subset U$ with the prescribed completions (\ref{E:prescribedcompletions}).
The uniqueness of $N$ is a consequence of Lemma \ref{L:LatticeRationalLocal}.
\end{proof}
\begin{remark}
The statement of Theorem \ref{T:LocalModification} must be well--known to the experts.
In the case when $R$ is an integral domain, it can be for instance found in \cite[Th\'eor\`eme VII.4.3]{Bourbaki}. However, we were not able to find a proof of this result in the full generality in the known literature. Since it plays a crucial role in our study of non--commutative curves, we decided to include a detailed proof for the sake of completeness and reader's convenience.
\end{remark}
\medskip
\noindent
Let $\Lambda$ be a semi--simple $K$--algebra. Recall that a subring $A \subset \Lambda$ is an \emph{$R$--order} if $R\cdot A = A$, $A$ is finitely generated $R$--module and $K\cdot A = \Lambda$. Note that for any $\idm \in \Max(R)$ we have:
$$
\widehat{A}_{\idm} := \widehat{R}_{\idm} \otimes_R A \quad \mbox{\rm and} \quad
\widehat{\Lambda}_{\idm} := \widehat{K}_{\idm} \otimes_K \Lambda \cong \widehat{R}_{\idm} \otimes_R \Lambda \cong \widehat{K}_{\idm} \otimes_R A.
$$
In particular,
$\widehat{A}_{\idm}$ is an $\widehat{R}_{\idm}$--order in the semi--simple
$\widehat{K}_{\idm}$--algebra
$\widehat{\Lambda}_{\idm}$.
\begin{proposition}\label{P:OrderPrescribedCompl} Let $\Lambda$ be a semi--simple
$K$--algebra and $A \subset \Lambda$ be an $R$--order.
Let $\Omega \subset \Max(R)$ be a finite subset such that for any $\idm \in \Omega$ we are given an $\widehat{R}_{\idm}$--order $B(\idm) \subset \widehat{\Lambda}_{\idm}$. Then there exists a unique $R$--order $B \subset \Lambda$ such that
\begin{equation}\label{E:prescribedcompletionsorder}
\widehat{B}_{\idm} =
\left\{
\begin{array}{cc}
\widehat{A}_{\idm} & \mbox{\rm if} \; \idm \notin \Omega \\
B(\idm) & \mbox{\rm if} \; \idm \in \Omega.
\end{array}
\right.
\end{equation}
\end{proposition}
\begin{proof}
According to Theorem \ref{T:LocalModification} there exists a uniquely determined $R$--lattice $B \subset \Lambda$ with completions given by (\ref{E:prescribedcompletionsorder}). We have to show that $B$ us actually a subring. For any $\idm \in \Max(R)\setminus \Omega$ we put: $B(\idm) = \widehat{A}_{\idm}$. By Lemma \ref{L:LatticeRationalLocal} we have:
$$
B = \left\{
b \in \Lambda \, \big| \, \hat{\vartheta}_{\idm}(b) \in B(\idm)
\; \mbox{\rm for all} \; \idm \in \Max(R)
\right\}.
$$
It follows that $b_1 + b_2, b_1 \cdot b_2 \in B$ for all $b_1, b_2 \in B$.
\end{proof}
\begin{remark} In the notations of Theorem \ref{T:LocalModification} one can prove in the same way that if $U$ is a Lie algebra over $K$, $L \subset U$ a Lie subalgebra over $R$
and $N(\idm) \subset \widehat{U}_{\idm}$ is a Lie subalgebra over
$\widehat{R}_{\idm}$ for any $\idm \in \Omega$ then the $R$--lattice $N \subset U$ is a Lie subalgebra, too.
\end{remark}
\medskip
\noindent
Assume now that $U$ is a finitely generated left $\Lambda$--module. An $A$--submodule $L \subset U$, which is also an $R$--lattice, is called \emph{$A$--lattice}. For any $\idm \in \Max(R)$ we have isomorphisms
$$
\widehat{L}_{\idm} := \widehat{A}_{\idm} \otimes_A L \cong \widehat{R}_{\idm} \otimes_R L \quad \mbox{\rm and} \quad
\widehat{U}_{\idm} := \widehat{\Lambda}_{\idm} \otimes_\Lambda U \cong \widehat{K}_{\idm} \otimes_K U \cong \widehat{R}_{\idm} \otimes_R U.
$$
It follows that $\widehat{L}_{\idm}$ is an $\widehat{A}_{\idm}$--lattice in
$\widehat{U}_{\idm}$.
\begin{proposition}\label{P:OrderPrescribedCompllattice} Let $U$ be a finitely generated $\Lambda$--module and $L\subset U$ an $A$--lattice.
Let $\Omega \subset \Max(R)$ be a finite subset such that for any $\idm \in \Omega$ we are given an $\widehat{A}_{\idm}$--lattice $N(\idm) \subset \widehat{U}_{\idm}$. Then there exists a unique $A$--lattice $N \subset U$ such that
\begin{equation}\label{E:prescribedcompletionslattice}
\widehat{N}_{\idm} =
\left\{
\begin{array}{cc}
\widehat{L}_{\idm} & \mbox{\rm if} \; \idm \notin \Omega \\
N(\idm) & \mbox{\rm if} \; \idm \in \Omega.
\end{array}
\right.
\end{equation}
\end{proposition}
\noindent
The proof of this result is the same as of Proposition \ref{P:OrderPrescribedCompl}.
\section{Morita theorem for one--dimensional orders}
\noindent
As in the previous section, let $R$ be a reduced excellent ring of Krull dimension one, whose total ring of fractions is $K$.
Recall that one can also define the notion of an $R$--order without fixing its rational envelope first. Namely, a finite $R$--algebra $A$ is an $R$--order if and only if it is torsion free, viewed as as $R$--module and the ring $\Lambda:= K \otimes_R A$ (the rational envelope of $A$) is semi--simple. Note that $A$ is an $R$--order if and only if $A$ is an $Z(A)$--order.
If $R = Z(A)$ then $A$ is a \emph{central} $R$--order. If we have a ring extension $A \subseteq A'$ such that $A'$ is an $R$--order and $K \otimes_R A \rightarrow K \otimes_R A'$ is an isomorphism then $A'$ is called \emph{overorder} of $A$. An order without proper overorders is called \emph{maximal}.
\smallskip
\noindent
Analogously, a finitely generated (left) $A$--module $L$ is a (left) $A$--lattice if $L$ is torsion free viewed as an $R$--module. In this case, the $\Lambda$--module $V := K \otimes_R L$ is the rational envelope of $L$. If we have an extension of $A$--modules $L \subseteq N$ such that $L, N$ are both $A$--lattices and the induced map
$K \otimes_R L \rightarrow K \otimes_R N$ is an isomorphism, then $L$ and $N$ are \emph{rationally equivalent} and $N$ is an \emph{overlattice} of $L$ and $L$ is a \emph{sublattice} of $N$, respectively.
\subsection{Categorical characterization of the non--regular locus of an order} Let $\Lambda$ be a semi--simple $K$--algebra and $A \subset \Lambda$ be an $R$--order. Then the set
\begin{equation}
\gS_{A} := \bigl\{\idm \in \Max(R) \, \big|\, A_{\idm} \subset \Lambda_{\idm} \; \mbox{is not a maximal order}\bigr\} \subset \Spec(R)
\end{equation}
is the locus of \emph{non--regular points} of $A$.
\begin{lemma}\label{L:NonRegLocus} The set $\gS_{A}$ is finite.
\end{lemma}
\begin{proof}
Let $\bigl\{\idp_1, \dots \idp_r\bigr\}$ be the set of minimal prime ideals in $R$,
$R_i := R/{\idp}_i$ and $K_i := \Quot(R_i)$ for $1 \le i \le r$. Then we have injective ring homomorphisms:
$$
R \longhookrightarrow R':= R_1 \times \dots \times R_r \longhookrightarrow
K_1 \times \dots \times K_r \cong K.
$$
In these terms, we get a decomposition: $\Lambda \cong \Lambda_1 \times \dots \times \Lambda_r$, where $\Lambda_i$ is a finite dimensional simple $K_i$--algebra for all $1 \le i \le r$. Let $A' := R' \cdot A \subset \Lambda$. Then $A'$ is an overorder of $A$ and we have a decomposition $A' \cong A'_1 \times \dots \times A'_r$, where $A'_i$ is an order in the simple algebra $\Lambda_i$.
Let $\widetilde{R}_i$ be the integral closure of $R_i$ in $K_i$. Since $R$ is \emph{excellent} of Krull dimension one, the ring $\widetilde{R}_i$ is regular and the ring extension $R_i \subseteq \widetilde{R}_i$ is finite. It follows that $A''_i :=
\widetilde{R}_i \cdot A'_i$ is an $\widetilde{R}_i$--order in the simple $K_i$--algebra
$\Lambda_i$. According to \cite[Corollary 10.4]{ReinerMO}, $A''_i$ is contained in a maximal order $\widetilde{A}_i$. Let $\widetilde{A} := \widetilde{A}_1 \times \dots \times \widetilde{A}_r$. Then $\widetilde{A}$ is a maximal order in $\Lambda$, which is an overorder of $A$. It follows from results of \cite[Section 11]{ReinerMO} that $\mathfrak{S}_A$ is the support of the finite length $R$--module $\widetilde{A}/A$. Hence, $\mathfrak{S}_A$ is a finite set.
\end{proof}
\begin{remark}
Let $\idm \in \Max(R)$ be a \emph{regular} point of $A$, i.e.~$A_{\idm} \subset \Lambda_{\idm}$ is a maximal order.
According to
\cite[Lemma 2.3]{Harada}, its center $Z(A_{\idm})$ is a Dedekind ring.
Moreover, $A_{\idm}$ itself is hereditary, too; see \cite[Theorem 18.1]{ReinerMO}.
Note also that $A_{\idm}$ is a maximal order if and only if $\widehat{A}_{\idm}$ is a maximal order; see
\cite[Theorem 11.5]{ReinerMO}.
\end{remark}
\begin{lemma}\label{L:OverorderTechnical}
Let $A$ be an order in $\Lambda$ and $B$ its overorder such that $A\cong B$, viewed as left $A$--modules. Then we have: $A = B$.
\end{lemma}
\begin{proof}
First note that the canonical morphism
$$
\bigl\{\lambda \in A \,|\, B \lambda \subseteq A\bigr\} \lar \Hom_A(B, A), \;
\lambda \mapsto (b \stackrel{\rho_\lambda}\mapsto b \lambda)
$$
is an isomorphism. Next, the following diagram is commutative:
$$
\xymatrix{
A^\circ \ar[rr]^-{\can} \ar@{_{(}->}[dd] & & \End_A(A) \\
& & \End_A(B) \ar[u]_-{\Ad_{\lambda}} \\
B^\circ \ar[rr]^-{\can} & & \End_B(B), \ar[u]_-{\can}
}
$$
where $\lambda \in A$ is such that $B \lambda = A$ and all canonical arrows are isomorphisms. It follows that $A = B$, as asserted.
\end{proof}
\smallskip
\noindent
It turns out that the set $\gS_{A}$ admits the following characterization.
\begin{theorem}\label{T:MaximalityCategorical} Let $A$ be a central $R$--order in the semi--simple $K$--algebra
$\Lambda$.
Let $S$ be a simple $A$--module, $F := \End_A(S)$ the corresponding skew field and $\mathsf{Supp}_R(S) = \{\idm\}$. Then $A_{\idm}$ is a maximal order if and only if the following conditions are satisfied:
\begin{itemize}
\item The length of the left $F$--module $\Ext_A^1(S, S)$ is one.
\item For any simple $A$--module $T \not\cong S$ we have: $\Ext^1_A(S, T) = 0$.
\end{itemize}
\end{theorem}
\begin{proof} If $A_{\idm}$ is a maximal order then there exists a unique simple
$A$--module $S$ supported at $\idm$; see for instance \cite[Theorem 18.7]{ReinerMO}. Moreover, we have an isomorphism of left $F$--modules
$\Ext_A^1(S, S) \cong F$. If $T$ is a simple $A$--module such that $T \not\cong S$ then $\mathsf{Supp}(T) \ne \mathsf{Supp}(S)$ and $\Ext^1_A(S, T) = 0$.
\medskip
\noindent
To prove the converse direction, we may without loss of generality assume $R$ to be local and complete. Consider the short exact sequence in $A-\mathsf{mod}$:
$$
0 \lar Q \lar P \stackrel{\pi}\lar S \lar 0,
$$
where $P$ is a projective cover of $S$ and $Q = \mathsf{rad}(P)$ its radical. Then $P$ is indecomposable and the following sequence of left $F$--modules is exact:
$$
0 \lar \End_A(S) \stackrel{\pi^\ast}\lar \Hom_A(P, S) \lar \Hom_A(Q, S) \lar \Ext^1_A(S, S) \lar 0.
$$
Let $Q' = \mathsf{rad}(Q)$. Since $\pi^\ast$ is an isomorphism, we get:
$
\Hom_A\bigl(Q/Q', S\bigr) \cong \Hom_A(Q, S) \cong F.
$
Let $T \not\cong S$ be a simple $A$--module. Then we get an exact sequence
$$
0 \lar \Hom_A(S, T) \lar \Hom_A(P, T) \lar \Hom_A(Q, T) \lar \Ext^1_A(S, T) \lar 0.
$$
Since $P$ is a projective cover of a simple module $S$, we have: $\Hom_A(P, T) = 0$. By assumption,
$\Ext^1_A(S, T) = 0$, hence $\Hom_A(Q, T) = 0$, too. It follows that $Q/Q' \cong S$. Hence, there exists a surjective homomorphism of $A$--modules
$P \stackrel{\nu}\lar Q$. Next, $\bar{P} = \ker(\nu)$ is an $A$--lattice. Since $Q$ is a sublattice of $P$, we conclude that $K \otimes_R Q \cong K \otimes P$ and $K \otimes_R \bar{P} = 0$. Hence, $\bar{P} = 0$ and $\nu$ is an isomorphism.
\smallskip
\noindent
Using induction on the length, one can show now that
for any sublattice $P' \subseteq P$ we have: $P' \cong P$.
Moreover, we claim that
$\widetilde{P}:= \Lambda \otimes_A P \cong K \otimes_R P$ is an indecomposable $\Lambda$--module. Indeed, if $\widetilde{P} \cong \widetilde{P}_1 \oplus \widetilde{P}_2$ then there exist $A$--sublattices $P_i \subset \widetilde{P}_i$ and $P_1 \oplus P_2$ is a sublattice of $P$. From what was proven above it follows that $P_1 \oplus P_2 \cong P$. However, $P$ is indecomposable, hence $P_1 = 0$ or $P_2 = 0$. Thus, $\widetilde{P}_1 = 0$ or $\widetilde{P}_2 = 0$, implying the claim.
\smallskip
\noindent
Since $\widetilde{P}$ is an indecomposable projective $\Lambda$--module, for any $A$--submodule $0 \ne X \subseteq P$ holds: $P/X$ has finite length and $X \cong P$. It implies that for any indecomposable projective $A$--module $U \not\cong P$ we have: $\Hom_A(U, P) = 0$. Since $Z(A) = R$ is local, the algebra $A$ is connected. Since its rational envelope $\Lambda$ is semi--simple, we conclude that there exists an isomorphism of left $A$--modules $A \cong P^{\oplus n}$ for some $n \in \NN$.
\smallskip
\noindent
Let $L$ be an indecomposable $A$--lattice. Then there exists an injective homomorphism of $A$--modules $L \longhookrightarrow A^{\oplus m} \cong P^{\oplus mn}$ for some $m \in \NN$, hence $\Hom_A(L, P) \ne 0$. From what was proven above
it follows that $L \cong P$.
\smallskip
\noindent
Assume now that $A \subseteq A' \subset \Lambda$ is an overorder. Then $A'$ is an $A$--lattice, rationally equivalent to $A$. Hence, $A$ and $A'$ are isomorphic as left $A$--modules. Lemma \ref{L:OverorderTechnical} implies that $A' = A$. Hence, the order $A$ is maximal, as asserted.
\end{proof}
\subsection{Morita equivalences of central orders}
We developed all necessary tools to prove the following result.
\begin{proposition}\label{P:CentralMoritaOrders}
Let $A$ and $B$ be two central $R$--orders, whose rational envelopes are semi--simple central $K$--algebras $\Lambda$ and $\Gamma$, respectively. Then $A$ and $B$ are centrally Morita equivalent if and only if the following conditions are satisfied:
\begin{itemize}
\item $\Lambda$ and $\Gamma$ are centrally Morita equivalent;
\item we have: $\mathfrak{S}_A = \mathfrak{S}_B$;
\item for any $\idm \in \mathfrak{S}_A$, the $\widehat{R}_{\idm}$--orders $\widehat{A}_{\idm}$ and $\widehat{B}_{\idm}$ are centrally Morita equivalent.
\end{itemize}
\end{proposition}
\begin{proof} Assume that $A$ and $B$ are centrally Morita equivalent. Theorem \ref{T:MaximalityCategorical} implies that $\mathfrak{S}_A = \mathfrak{S}_B$.
Let $P$ be a Morita $(B-A)$--bimodule, for which the left and right actions of $R$ coincide. Then the following diagram of rings and ring homomorphisms
\begin{equation}\label{E:CommDiag2}
\begin{array}{c}
\xymatrix{
\End_A(P) & & B \ar[ll]_-{\lambda^P} \\
& R \ar@{_{(}->}[lu]^-{\rho^P} \ar@{^{(}->}[ru] &
}
\end{array}
\end{equation}
is commutative. Passing in (\ref{E:CommDiag2}) to localizations and completions, we conclude that
$\widehat{A}_{\idm}$ and $\widehat{B}_{\idm}$ are centrally Morita equivalent for all $\idm \in \Max(R)$. In the same way, $\Lambda$ and $\Gamma$ are centrally Morita equivalent.
\smallskip
\noindent
Proof of the converse direction is more involved. Let $V$ be a Morita $(\Gamma-\Lambda)$--bimodule inducing a central equivalence of categories $\Lambda-\mathsf{mod} \lar
\Gamma-\mathsf{mod}$. For any $\idm \in \gS := \gS_{A}$, we get
a Morita $(\widehat{\Gamma}_{\idm}-\widehat{\Lambda}_{\idm})$--bimodule $\widehat{V}_{\idm}$, which induces a central equivalence of categories $\widehat{\Lambda}_{\idm}-\mathsf{mod} \lar \widehat{\Gamma}_{\idm}
-\mathsf{mod}$. Let $P(\idm)$ be a Morita $(\widehat{B}_{\idm}-\widehat{A}_{\idm})$--bimodule inducing a central equivalence of categories $\widehat{A}_{\idm}-\mathsf{mod} \lar
\widehat{B}_{\idm}-\mathsf{mod}$. Then $$\widetilde{P}(\idm) := \widehat{K}_{\idm}
\otimes_{\widehat{R}_{\idm}} P(\idm) \cong \widehat{\Gamma}_{\idm}
\otimes_{\widehat{B}_{\idm}} P(\idm) \cong P(\idm)
\otimes_{\widehat{A}_{\idm}} \widehat{\Lambda}_{\idm}
$$
is a Morita $(\widehat{\Gamma}_{\idm}-\widehat{\Lambda}_{\idm})$--bimodule, which induces a central equivalence of categories $\widehat{\Lambda}_{\idm}-\mathsf{mod} \rightarrow
\widehat{\Gamma}_{\idm}-\mathsf{mod}$
as well.
Since $\widehat{\Lambda}_{\idm}$ and $\widehat{\Gamma}_{\idm}$ are semi--simple rings,
Theorem \ref{T:SkolemNoether} implies that $\widehat{V}_{\idm}$ and $\widetilde{P}(\idm)$ are isomorphic as $(\widehat{\Gamma}_{\idm}-\widehat{\Lambda}_{\idm})$--bimodules. Therefore, we can without loss of generality assume that
$P(\idm) \subset \widehat{V}_{\idm}$ and the left action of
$\widehat{B}_{\idm}$ as well as the right action of
$\widehat{A}_{\idm}$ on $P(\idm)$ and
$\widehat{V}_{\idm}$
match for all $\idm \in \gS$. Our goal now is to construct a $(B-A)$--subbimodule
$P \subset V$, which induces a central equivalence of categories
$A-\mathsf{mod} \rightarrow B-\mathsf{mod}$.
\smallskip
\noindent
We start with an arbitrary $R$--lattice $L \subset V$ and put:
$Q := B \cdot L \cdot A \subset V$. Then $Q$ is a finitely generated $R$--module and $K \cdot Q = V$, i.e. $Q$ is an $R$--overlattice of $L$ with the same rational envelope $V$. Moreover, $Q$ is a $(B-A)$--bimodule with central action of $R$. Note that the following diagram of rings and ring homomorphisms
$$
\xymatrix{
B \ar@{_{(}->}[d] \ar[rr]^-{\lambda_B^Q} & & \End_A(Q)
\ar@{^{(}->}[d] \\
\Gamma \ar[rr]^-{\lambda_\Gamma^V} & & \End_\Lambda(V)
}
$$
is commutative. Since $\mu := \lambda_\Gamma^V$ is an isomorphism, the map $\lambda := \lambda_B^Q$ is injective. Hence, $
\widehat{B}_{\idm} \stackrel{\widehat{\lambda}_{\idm}}\lar
\End_{\widehat{A}_{\idm}}(\widehat{Q}_{\idm})$ is injective for any $\idm \in \Max(R)$.
\smallskip
\noindent
We claim that $\widehat{\lambda}_{\idm}$ is an isomorphism for all
$\idm \in \Max(R) \setminus \gS$. Indeed, $
\widehat{\Gamma}_{\idm} \stackrel{\widehat{\mu}_{\idm}}\lar
\End_{\widehat{\Lambda}_{\idm}}(\widehat{V}_{\idm})$ is an isomorphism and
$
\widehat{B}_{\idm}$ is a maximal order in the semi--simple algebra $
\widehat{\Gamma}_{\idm}$. Hence,
$\widehat{\lambda}_{\idm}\bigl(\widehat{B}_{\idm} \bigr)$ is a maximal order in the semi--simple algebra $\End_{\widehat{\Lambda}_{\idm}}(\widehat{V}_{\idm})$ and as a consequence, we get:
$\widehat{\lambda}_{\idm}\bigl(\widehat{B}_{\idm}\bigr) = \End_{\widehat{A}_{\idm}}(\widehat{Q}_{\idm})$.
\smallskip
\noindent
According with Lemma \ref{L:NonRegLocus}, the set $\gS$ is finite. By Proposition
\ref{P:OrderPrescribedCompllattice}, there exists a unique right $A$--lattice $P \subset V$ such that
$
\widehat{P}_{\idm} =
\left\{
\begin{array}{cl}
\widehat{Q}_{\idm} & \mbox{\rm if} \; \idm \notin \gS \\
P(\idm) & \mbox{\rm if} \; \idm \in \gS.
\end{array}
\right.
$
\noindent
It follows from Lemma \ref{L:LatticeRationalLocal} that $B \cdot P = P$, i.e.~ $P$ is an $(B-A)$--bimodule with central action of $R$. Since
$\widehat{P}_{\idm}$ is a right $\widehat{A}_{\idm}$--progenerator
for any $\idm \in \Max(R)$, $P$ is a right $A$--progenerator.
The ring homomorphism $B \stackrel{\lambda_B^P}\lar \End_A(P)$ is an isomorphism, since so are its completions for all $\idm \in \Max(R)$. Hence, $P$ is a Morita
$(B-A)$--bimodule, which induces a central equivalence of categories $A-\mathsf{mod} \lar
B-\mathsf{mod}$ we are looking for.
\end{proof}
\begin{lemma}\label{L:ConjugateMorita}
Let $\Lambda$ be a semi--simple central $K$--algebra, $A \subset \Lambda$ be a central $R$--order, $\upsilon \in \Lambda^\ast$ and $B:= \upsilon A \upsilon^{-1}$. Then $A$ and $B$ are centrally Morita equivalent.
\end{lemma}
\begin{proof}
Let $P:= \upsilon A$. Obviously, $P \cong A$ as right $A$--modules. In particular, $P$ is a right $A$--progenerator. Note that $B \cdot P =
\bigl(\upsilon A \upsilon^{-1}\bigr)\cdot (\upsilon A) = \upsilon A = P$. It follows that $P$ is an $(B-A)$--bimodule with central $R$--action.
\noindent
Since $\upsilon \in \Lambda^\ast$, the ring homomorphism
$B \stackrel{\lambda^P_B}\lar \End_A(P)$ is injective. Let $f \in \End_A(P)$.
As $P$ is a right $A$--lattice, whose rational envelope is $\Lambda_{\Lambda}$, there exists $b \in \Lambda$ such that $f = \lambda^P_b$. Since
$b \upsilon A \subseteq \upsilon A$, there exists $a \in A$ such that $b \upsilon = \upsilon a$. It follows that $b = \upsilon a \upsilon^{-1} \in B$. Hence, $\lambda^P_B$ is an isomorphism and the $(B-A)$--bimodule $P$ induces a central equivalence we are looking for.
\end{proof}
\subsection{Morita equivalences of non--commutative curves}
\noindent
A \emph{reduced non--commutative curve} (abbreviated as \emph{rncc}) is a ringed space $\bbX = (X, \kA)$, where $X$ is a reduced excellent noetherian scheme of pure dimension one and $\kA$ is a sheaf of $\kO$--orders (i.e $A(U)$ is an $O(U)$--order for any affine open subset $U \subseteq X$). Following Definition \ref{D:NCNS}, we say that $\bbX$ is \emph{central} if
$O_x = Z(A_x)$ for all $x \in X$.
\smallskip
\noindent
From now on, let $\bbX$ be a central rncc. If $\kK$ is the sheaf of rational functions on $X$ then $K := \Gamma(X, \kK)$ is a semi--simple ring. Moreover, $\Lambda_{\bbX}:= \Gamma(X, \kK \otimes_\kO \kA)$ is a central semi--simple $K$--algebra. This algebra can be viewed as the ``ring of rational functions'' on $\bbX$. If $\Coh_0(\bbX)$ denotes the abelian category of
objects of finite length in $\Coh(\bbX)$ then the functor
$\Gamma(X, \kK \otimes_\kO - )$ induces an equivalence of categories
$
\Coh(\bbX)/\Coh_0(\bbX) \simeq \Lambda_{\bbX}-\mathsf{mod}.
$
\smallskip
\noindent
For any $x \in X$, we have a central $\widehat{O}_x$--order $\widehat{A}_x$, whose rational envelope can be canonically identified with the semi--simple ring
$\widehat{\Lambda}_x = \widehat{K}_x \otimes_K \Lambda_{\bbX}$. Let
\begin{equation}
\gS_{\bbX} := \{x \in X \, \big|\, \widehat{A}_x
\; \mbox{is not a maximal order in} \;
\widehat{\Lambda}_x\}
\end{equation}
be the locus of \emph{non--regular} points of $\bbX$. According to Lemma \ref{L:NonRegLocus}, $\gS_{\bbX}$ is a finite set. Let
$x \in X$ be a \emph{regular} point. By \cite[Lemma 2.3]{Harada},
$\widehat{O}_x = Z\bigl(\widehat{\Lambda}_x\bigr)$ is a discrete valuation ring.
It follows that $\widehat{K}_x$ is a field and $\widehat{\Lambda}_x$ is a central simple
$\widehat{K}_x$--algebra. Hence, there exists a skew field $F_x \supseteq \widehat{K}_x$ (such that $Z(F_x) = \widehat{K}_x$) and $n = n(x) \in \NN$ such that $\widehat{\Lambda}_x \cong \Mat_n(F_x)$.
Moreover, one can explicitly describe the order
$\widehat{A}_x$ in this case.
Namely, there exists a uniquely determined maximal $\widehat{R}_x$--order $T_x \subset F_x$; see \cite[Theorem 12.8]{ReinerMO}, and $\Mat_n(T_x)$ can be identified with a maximal order in $\widehat{\Lambda}_x$. Moreover, any two maximal orders $O_x', O_x'' \subset \widehat{\Lambda}_x$ are conjugate, i.e.~there exists $\upsilon \in \widehat{\Lambda}_x^\ast$ such that $O_x'= \upsilon
O_x'' \upsilon^{-1}$;
see \cite[Theorem 17.3]{ReinerMO}. In particular, the orders $O_x'$ and $O_x''$ are centrally Morita equivalent (see Lemma \ref{L:ConjugateMorita}) and
$\widehat{A}_x \cong \Mat_n(T_x)$.
\begin{proposition}\label{P:CentralMoritaEqNCC}
Let $\bbX = (X, \kA)$ and $\bbY = (X, \kB)$ be two central rncc with the same central curve
$X$. Then the categories $\Qcoh(\bbX)$ and
$\Qcoh(\bbY)$ are \emph{centrally} equivalent if any only if the following conditions are satisfied:
\begin{itemize}
\item the semi--simple $K$--algebras $\Lambda_{\bbX}$ and $\Lambda_{\bbY}$ are centrally Morita equivalent;
\item we have: $\mathfrak{S}_{\bbX} = \mathfrak{S}_{\bbY}$;
\item for any $x \in \mathfrak{S}_{\bbX}$, the $\widehat{O}_{x}$--orders $\widehat{A}_{x}$ and $\widehat{B}_{x}$ are centrally Morita equivalent.
\end{itemize}
\end{proposition}
\begin{proof} This result is just a global version of Proposition \ref{P:CentralMoritaOrders}
and the proof below is basically a ``sheafified'' version of the arguments from the affine case.
\smallskip
\noindent
Let $\Qcoh(\bbX) \stackrel{\Phi}\lar
\Qcoh(\bbY)$ be a central equivalence of categories.
According to Theorem \ref{T:MoritaNonCommCurves}, we have: $\Phi \cong \kP \otimes_{\kA} \,-\,$, where
$\kP$ is a sheaf of $(\kB-\kA)$--bimodules such that
$\kP \in \VB(\bbX^\circ)$ is a local right progenerator and
$\kB \stackrel{\lambda_\kB^\kP}\lar \bigl(\mathit{End}_{\kA}(\kP)\bigr)$ an isomorphism of $\kO$--algebras. Let $V:= \Gamma(X, \kK \otimes \kP)$. Then $V$ is a Morita
$(\Lambda_{\bbY}-\Lambda_{\bbX})$--bimodule inducing a central Morita equivalence
$\xymatrix{\Lambda_{\bbX} \ar@{.>}[r] & \Lambda_{\bbY}}$.
Similarly, for any $x \in X$ we get a central Morita $\bigl(\widehat{B}_x - \widehat{A}_{x}\bigr)$--bimodule
$\widehat{P}_{x}$. Finally,
Theorem \ref{T:MaximalityCategorical} implies that $\mathfrak{S}_{\bbX} = \mathfrak{S}_{\bbY}$.
\smallskip
\noindent
Conversely, assume that rncc $\bbX$ and $\bbY$ satisfy three conditions above. Let $V$ be a Morita
$(\Lambda_{\bbY}-\Lambda_{\bbX})$--bimodule with central action of $K$.
Let $\widetilde\kA = \kK \otimes_\kO \kA$ and $\widetilde\kB = \kK \otimes_\kO \kB$.
Passing to sheaves, we get
a balanced $(\widetilde{\kB}-\widetilde{\kA})$--bimodule $\kV$ with central action of $\kK$. Let $\kL \subset \kV$ be any sheaf of $\kO$--lattices, i.e.~a coherent submodule of $\kV$ such that the canonical morphism $\kK \otimes_\kO \kL \rightarrow \kV$ is an isomorphism.
Then $\kQ := \kB \cdot \kL \cdot \kA \subset \kV$ is a sheaf of lattices with rational envelope $\kV$. Moreover, $\kQ$ is a $({\kB}-{\kA})$--bimodule having a central action of $\kO$. As in the proof of Proposition \ref{P:CentralMoritaOrders}, one can show that for any $x \in \mathfrak{S} := \mathfrak{S}_{\bbX}$ there exists a Morita $(\widehat{B}_x-\widehat{A}_x)$--bimodule $P(x) \subset \widehat{V}_x$, having a central action of $\widehat{O}_x$. By Theorem \ref{T:LocalModification}, there exists a unique $\kP \in \VB(\bbX^\circ)$ such that
$\kP \subset \kV$ is a right $\kA$--lattice and
\begin{equation*}
\widehat{P}_{x} =
\left\{
\begin{array}{cl}
\widehat{Q}_{x} & \mbox{\rm if} \; \idm \notin \gS \\
P(x) & \mbox{\rm if} \; x \in \gS.
\end{array}
\right.
\end{equation*}
Then $\kP$ is a sheaf of $(\kB-\kA)$--bimodules with central action of $\kO$. Moreover, $\kP \in \VB(\bbX^\circ)$ is a local right progenerator such that
$\kB \stackrel{\lambda_\kB^\kP}\lar \mathit{End}_{\kA}(\kP)$ is an isomorphism of $\kO$--algebras. Hence, $\Phi:= \kP \otimes_\kA \,-\,$ is a central equivalence of categories we are looking for.
\end{proof}
\smallskip
\noindent
After all preparations we can now prove the main result of this section.
\begin{theorem}\label{T:MoritaNonCommCurves}
Let $\bbX = (X, \kA)$ and $\bbY = (Y, \kB)$ be two central rncc. Then the categories $\Qcoh(\bbX)$ and
$\Qcoh(\bbY)$ are equivalent if any only if there exists a scheme isomorphism
$Y \stackrel{\varphi}\lar X$ satisfying the following conditions.
\begin{itemize}
\item There exists a Morita equivalence $\xymatrix{\Lambda_{\bbX}
\ar@{.>}[r]^-{\widetilde\Phi} & \Lambda_{\bbY}}$ such that the diagram
\begin{equation}\label{E:MoritaBirational}
\begin{array}{c}
\xymatrix{
\Lambda_{\bbX} \ar@{.>}[r]^-{\widetilde\Phi} & \Lambda_{\bbY} \\
K_X \ar[r]^-{\varphi^\ast} \ar@{^{(}->}[u] & K_Y \ar@{_{(}->}[u]
}
\end{array}
\end{equation}
is ``commutative'' (here we follow the notation of Remark \ref{R:Pseudocommutative}).
\item We have: $\varphi\bigl(\gS_{\bbY}\bigr) = \gS_{\bbX}$ and for any $y \in \gS_{\bbY}$, there exists a Morita equivalence
$\xymatrix{\widehat{A}_{\varphi(y)} \ar@{.>}[r]^-{\Phi_y} & \widehat{B}_{y}}$
such that the diagram
\begin{equation}\label{E:MoritaComplete}
\begin{array}{c}
\xymatrix{
\widehat{A}_{\varphi(y)} \ar@{.>}[r]^-{\Phi_y} & \widehat{B}_{y} \\
\widehat{O}_{\varphi(y)} \ar[r]^-{\varphi_y^\ast} \ar@{^{(}->}[u] &
\widehat{O}_{y} \ar@{_{(}->}[u]
}
\end{array}
\end{equation}
is ``commutative''.
\end{itemize}
\end{theorem}
\begin{proof} According to Theorem \ref{T:MoritaNonCommCurves}, any equivalence of categories
$\Qcoh(\bbX) \stackrel{\Phi}\lar
\Qcoh(\bbY)$ is isomorphic to a functor of the form $\Phi_{\kP, \vartheta, \varphi}$, where
$Y \stackrel{\varphi}\lar X$ is a scheme isomorphism,
$\kP \in \VB(\bbX^\circ)$ a local right progenerator and
$\kB \stackrel{\vartheta}\lar
\varphi^\ast\bigl(\mathit{End}_{\kA}(\kP)\bigr)$ an isomorphism of $\kO_Y$--algebras. Let $V:= \Gamma(X, \kK_X \otimes \kP)$. Then $V$ is a Morita
$(\Lambda_{\bbY}-\Lambda_{\bbX})$--bimodule inducing an equivalence
$\xymatrix{\Lambda_{\bbX} \ar@{.>}[r]^-{\widetilde\Phi} & \Lambda_{\bbY}}$ such that the diagram (\ref{E:MoritaBirational}) is ``commutative''. Similarly, for any $y \in Y$ we get a Morita $\bigl(\widehat{B}_y - \widehat{A}_{\varphi(y)}\bigr)$--bimodule
$\widehat{P}_{\varphi(y)}$, which induces an equivalence $\xymatrix{\widehat{A}_{\varphi(y)} \ar@{.>}[r]^-{\Phi_y} & \widehat{B}_{y}}$
such that the diagram (\ref{E:MoritaComplete}) is ``commutative''. Finally,
Theorem \ref{T:MaximalityCategorical} implies that $y \in \gS_{\bbY}$ if and only if $\varphi(y) \in \gS_{\bbX}$.
\smallskip
\noindent
Conversely, assume that we are given a scheme isomorphism $Y \stackrel{\varphi}\lar X$ as well as Morita equivalences $\xymatrix{\Lambda_{\bbX} \ar@{.>}[r]^-{\widetilde\Phi} & \Lambda_{\bbY}}$ and $\bigl(\xymatrix{\widehat{A}_{\varphi(y)} \ar@{.>}[r]^-{\Phi_y} & \widehat{B}_{y}}\bigr)_{y \in \gS_{\bbY}}$ satisfying the compatibility constraints (\ref{E:MoritaBirational}) and (\ref{E:MoritaComplete}). Let $\bbY' = \bigl(Y, \varphi^\ast(\kA)\bigr)$. Then
$\bbY'$ is a central rncc and $\gS_{\bbY} = \gS_{\bbY'}$. By Proposition \ref{P:CentralMoritaEqNCC},
$\bbY$ and $\bbY'$ are centrally Morita equivalent rncc, implying the result.
\end{proof}
\subsection{Morita equivalences of hereditary non--commutative curves}
Recall that a central rncc $\XX = (X, \kA)$ is \emph{hereditary} if
$\widehat{A}_x$ is a hereditary order for all $x \in X$. In this case, the central curve $X$ is automatically smooth; see \cite[Theorem 2.6]{Harada}. We may without loss of generality assume $X$ to be connected, hence $\Lambda_{\bbX}$ is a central simple $K_X$--algebra, which defines an element in the Brauer group of the function field $K_X$.
\smallskip
\noindent
Following the notation of the previous subsection, for any $x \in X$ there exists
a skew field $F_x$ (whose center is $\widehat{K}_x$) and $n = n(x) \in \NN$ such that $\widehat{\Lambda}_x \cong \Mat_n(F_x)$. Any maximal order in $\widehat{\Lambda}_x$ is conjugate to $\Mat_n(T_x)$, where $T_x$ is the unique maximal order in $F_x$.
\smallskip
\noindent
Let $x \in \gS_{\XX}$ and $\mathfrak{t}_x$ be the Jacobson radical of $T_x$. Viewing $\widehat{H}_x$ as an order in the simple algebra $\Mat_n(F_x)$, we have the following result:
$\widehat{A}_x$ is \emph{conjugate} to the order
\begin{equation}\label{E:StandardHereditary}
H\bigl(T_x, (n_1, \dots, n_t)\bigr) :=
\left[
\begin{array}{cccc}
T_x & \mathfrak{t}_x & \dots & \mathfrak{t}_x\\
T_x & T_x & \dots & \mathfrak{t}_x\\
\vdots & \vdots & \ddots & \vdots \\
T_x & \mathfrak{t}_x & \dots & T_x\\
\end{array}
\right]^{\underline{(n_1, \dots, n_t)}} \subseteq \Mat_n(T_x)
\end{equation}
where $(n_1, \dots, n_t) \in \NN^t$ is such that
$n = n_1 + \dots + n_t$. The elements of $H\bigl(T_x, (n_1, \dots, n_t)\bigr)$ are matrices, such that for any $1 \le i, j \le t$, the $(i,j)$-th entry is itself an arbitrary matrix of size $(n_i \times n_j)$ with coefficients in $T_x$ for $i \ge j$ and in $\mathfrak{t}_x$ for $i < j$. The length of this tuple $t = t(x)$ (called \emph{type} of the hereditary order $\widehat{H}_x$) is equal to the number of non--isomorphic simple $\widehat{H}_x$--modules. Moreover, $H\bigl(T_x, (n_1, \dots, n_t)\bigr) \cong H\bigl(T_x, (n'_1, \dots, n'_t)\bigr)$ if and only if
$(n_1, \dots, n_t)$ and $(n'_1, \dots, n'_t)$ are obtained from each other by a cyclic permutation.
We refer to \cite[Theorem 39.14 and Corollary 39.24]{ReinerMO} for proofs of all these fundamental facts.
\smallskip
\noindent
A point $x \in X$ is a regular point of $\XX$ if and only if $t(x) = 1$.
It is easy to see that $H\bigl(T_x, (n_1, \dots, n_t)\bigr)$ is centrally Morira equivalent to the basic order $H\bigl(T_x, \underbrace{(1, \dots, 1)\bigr)}_{\scriptstyle t \; \mathrm{times}}$.
It follows from Lemma \ref{L:ConjugateMorita}
that any two hereditary orders of the same type in the simple algebra $\widehat{\Lambda}_x$ are \emph{centrally} Morita equivalent.
Theorem \ref{T:MoritaNonCommCurves} implies the following result, which was proven for the first time by Spie\ss{}; see \cite[Proposition 2.9]{Spiess}.
\begin{corollary}\label{C:HereditaryMorita} Let $\bbX = (X, \kA)$ and $\bbY = (Y, \kB)$ be two hereditary rnnc and $\Lambda_{\bbX}$ and $\Lambda_{\bbY}$ the corresponding simple algebras over the function fields $K_X$ and $K_Y$, respectively. Then the categories $\Qcoh(\bbX)$ and $\Qcoh(\bbY)$
are equivalent if and only if there exists an isomorphism $Y \stackrel{\varphi}\lar X$ such that $t(\varphi(y)) = t(y)$ for any $y \in Y$ and
$\bigl[\Lambda_{\bbY}\bigr] = \bigl[\varphi^*(\Lambda_{\bbX})\bigr] \in \mathsf{Br}(K_Y)$, where $\mathsf{Br}(K_Y)$ is the Brauer group of $K_Y$.
\end{corollary}
\begin{remark} In the setting of Corollary \ref{C:HereditaryMorita},
assume additionally that $X$ and $Y$ are quasi--projective curves over an
algebraically closed field $\kk$. By Tsen's theorem, we have:
$\mathsf{Br}(K_X) = 0 = \mathsf{Br}(K_Y)$; see \cite[Proposition 6.2.3 and Theorem 6.2.8]{GilleSzamuely}. It follows that $\bbX$ and $\bbY$ are Morita equivalent if and only if there exists an isomorphism $Y \stackrel{\varphi}\lar X$ such that $t\bigl(\varphi(y)\bigr) = t(y)$ for any $y \in Y$.
\end{remark}
\begin{remark}
A hereditary rncc $\bbX = (X, \kA)$ is called \emph{regular} if
$\gS_{\bbX} = \emptyset$. Assume additionally that $X$ is a \emph{projective} curve over a field $\kk$ (as already mentioned, the central curve $X$ is automatically smooth in this case). Then $\Coh(\bbX)$ is a noetherian hereditary category with finite dimensional $\Hom$-- and
$\Ext$--spaces, admitting an Auslander--Reiten translation functor $\Coh(\bbX)\stackrel{\tau}\lar \Coh(\bbX)$ such that $\tau(\kF) \cong \kF$ for any object $\kF$ of $\Coh_0(\bbX)$. Various properties of the category $\Coh(\bbX)$ were studied in detail (from a slightly different perspective) by Kussin in \cite{Kussin}.
\smallskip
\noindent
Let $\Lambda = \Lambda_{\bbX}$ be the algebra of ``rational functions'' on $\bbX$ and $K = K_X$ be its center. There exists a unique (up an isomorphism) smooth projective curve over $\kk$ (namely, $X$ itself), whose field of rational functions is isomorphic to $K$; see for instance \cite[Proposition 7.3.13]{Liu}. Let $\kB$ be any sheaf of \emph{maximal} orders on $X$ such that $\Gamma(X, \kK \otimes_\kO \kB) \cong \Lambda$. It is well--known that the ringed spaces $\bbX = (X, \kA)$ and $\bbX' = (X, \kB)$ need not be in general isomorphic (see for instance \cite{Deuring} or \cite[Corollary]{DrozdTurchin} for examples of non--isomorphic maximal orders in the same central simple algebra). However, Corollary \ref{C:HereditaryMorita} implies that the
categories $\Qcoh(\bbX)$ and $\Qcoh(\bbX')$ are equivalent. Hence, a regular rncc $\bbX$ with central projective curve $X$ is up to a Morita equivalence determined by an element in the Brauer group
$\mathsf{Br}(K_X)$. \qed
\end{remark}
\smallskip
\noindent
The following example shows that the compatibility constraints (\ref{E:MoritaBirational}) and (\ref{E:MoritaComplete}) are necessary to end up with a global Morita equivalence.
\begin{example}\label{E:ExampleNonHered}
Let $\kk$ be an infinite field $S = \kk[x]$, $J = \bigl((x-\lambda')(x-\lambda'')(x^2-1)\bigr)$ for $\lambda' \ne \lambda'' \in \kk \setminus \{1, -1\}$ and
$
H = \left(
\begin{array}{ccc}
S & J & J \\
S & S & J \\
S & S & S
\end{array}
\right).
$
Next, we put:
$$
A_+ = \left\{p \in H \left| \begin{array}{l}
p_{11}(1) = p_{11}(-1) \\
p_{22}(1) = p_{33}(1)
\end{array}
\right.
\right\} \quad \mbox{\rm and} \quad
A_- = \left\{p \in H \left| \begin{array}{l}
p_{11}(1) = p_{11}(-1) \\
p_{22}(-1) = p_{33}(-1)
\end{array}
\right.
\right\}.
$$
Then we have: $R := Z\bigl(A_{\pm}\bigr) = \kk\bigl[x^2-1, x(x^2-1)\bigr] \cong
\kk[u, v]/(f),
$
where $f = v^2-u^3-u^2$. It is clear that $A_\pm$ are $R$--orders with common rational envelope $\Lambda = \Mat_3\bigl(\kk(x)\bigr)$. Passing to the corresponding sheaves of orders, we get a pair of rncc $\bbE_\pm := (E, \kA_\pm)$, where $E = V(f) \subset \mathbbm{A}^2$ is a plane nodal cubic. These curves have the same locus of non--regular points $\gS = \{s, q', q''\}$, where $s = (0, 0)$ is the singular point of $E$, whereas $q' = \bigl(\lambda'^2-1,
\lambda' (\lambda'^2-1)\bigr)$ and $q'' = \bigl(\lambda''^2-1,
\lambda'' (\lambda''^2-1)\bigr)$ are two distinct smooth points of $E$.
We have:
$$
\bigl(\widehat{A}_+\bigr)_{q'} = \bigl(\widehat{A}_-\bigr)_{q'} = \left(
\begin{array}{ccc}
O' & \idm' & \idm' \\
O' & O' & \idm' \\
O' & O' & O'
\end{array}
\right),
$$
where $O' = \widehat{O}_{q'}$ is the completed local ring of $E$ at the point $q'$ and $\idm'$ its maximal ideal. Of course, the analogous statement holds for the second point $q''$, too.
\smallskip
\noindent
Both orders $\bigl(\widehat{A}_+\bigr)_s$ and $\bigl(\widehat{A}_-\bigr)_s$
have the common center $D = \kk\llbracket w_+, w_-\rrbracket/(w_+ w_-)$, where
$w_\pm = v \pm u(1 - u + u^2 - \dots) \in \kk\llbracket u, v\rrbracket$. It is clear that $\bigl(\widehat{A}_+\bigr)_s$ and $\bigl(\widehat{A}_-\bigr)_s$ are isomorphic as rings.
However, although $\bbE_+$ and $\bbE_-$ have the same central curve $E$, the common algebra of rational functions $\Lambda$ and the same singularity types, they are \emph{not Morita equivalent}!
\smallskip
\noindent
Indeed, any equivalence of categories $\Qcoh(\bbE_+) \stackrel{\Phi}\lar \Qcoh(\bbE_-)$ induces an automorphism $E \stackrel{\varphi}\lar E$; see
Theorem \ref{T:MorphismCentralSchemes}. It follows that $\varphi(s) = s$ and
$\varphi\bigl(\left\{q', q''\right\}\bigr) = \left\{q', q''\right\}$.
In particular, we get
the restricted Morita equivalence
$\xymatrix{\bigl(\widehat{A}_+\bigr)_s \ar@{.>}[r]^{\Phi_s} & \bigl(\widehat{A}_-\bigr)_s}$.
However, any such equivalelnce induces an automorphism of $D$ swapping both branches $(w_+)$ and $(w_-)$. If $\lambda', \lambda''$ were chosen sufficiently general, such an automorphism $\varphi$ does not exist. Hence,
$\Qcoh(\bbE_+)$ and $\Qcoh(\bbE_-)$ are not equivalent, as asserted. \qed
\end{example}
\begin{thebibliography}{99}
\bibitem{Antieau}
B.~Antieau, \emph{A reconstruction theorem for abelian categories of twisted sheaves}, J. Reine Angew. Math. \textbf{712} (2016), 175--188.
\bibitem{ArtinZhang}
M.~Artin \& J.~Zhang, \emph{Noncommutative projective schemes},
Adv. Math. \textbf{109} (1994), no. 2, 228--287.
\bibitem{Bass}
H.~Bass, \emph{Algebraic K--theory}, W. A. Benjamin, Inc., New York--Amsterdam 1968.
\bibitem{Bourbaki}
N.~Bourbaki, \emph{\'El\'ements de math\'ematique. Alg\`ebre commutative.
Chapitres 5 \`a 7}, Masson, Paris, 1985. 351 pp.
\bibitem{bd}
I.~Burban \& Yu.~Drozd,
\emph{Tilting on non--commutative rational projective curves},
Math.~Ann.~\textbf{351}, no.~3 (2011), 665--709.
\bibitem{bdnpdalcurves} I.~Burban \& Yu.~Drozd,
\emph{Non--commutative nodal curves and derived tame algebras},
\texttt{arXiv:1805.05174}.
\bibitem{BurbanDrozdGavranIMRN}
I.~Burban, Yu.~Drozd \& V.~Gavran, \emph{Singular curves and quasi--hereditary algebras}, Int.~Math.~Res.~Notices \textbf{2017}, no.~3, 895--920 (2017).
\bibitem{CaldararuThesis}
A.~C\u{a}ld\u{a}raru, \emph{Derived categories of twisted sheaves on Calabi--Yau manifolds}, Thesis (Ph.D.)–Cornell University. 2000. 196 pp
\bibitem{CaldararuCrelle}
A.~C\u{a}ld\u{a}raru, \emph{Derived categories of twisted sheaves on elliptic threefolds}, J. Reine Angew. Math. \textbf{544} (2002), 161--179.
\bibitem{CanonacoStellari}
A.~Canonaco \& P.~Stellari, \emph{Twisted Fourier--Mukai functors}, Adv. Math.
\textbf{212}(2007), no. 2, 484--503.
\bibitem{CurtisReiner} Ch.~Curtis \& I.~Reiner, \emph{Methods of representation theory. Vol. I. With applications to finite groups and orders}, Pure and Applied Mathematics, A Wiley--Interscience Publication, 1981.
\bibitem{Deuring}
M.~Deuring, \emph{Die Anzahl der Typen von Maximalordnungen in einer Quaternionenalgebra von primer Grundzahl}, Nachr. Akad. Wiss. G\"ottingen. Math.-Phys. Kl. Math.-Phys.-Chem. Abt. (1945) 48--50.
\bibitem{Dieudonne}
J.~Dieudonn\'e, \emph{Topics in Local Algebra}, Notre Dame: University of Notre Dame Press, 1967.
\bibitem{DrozdTurchin}
Yu.~Drozd \& V.~Turchin, \emph{Locally conjugate orders},
Mat. Zametki \textbf{24} (1978), no. 6, 871--878, 895.
\bibitem{Gabriel}
P.~Gabriel, \emph{Des cat\'egories ab\'eliennes},
{Bull. Soc. Math. France} \textbf{90} (1962), 323--448.
\bibitem{GeigleLenzing}
W.~Geigle \& H.~Lenzing,
A class of weighted projective curves arising in representation theory
of finite-dimensional algebras,
{\it Singularities, representation of algebras, and vector bundles}, 265--297,
Lecture Notes in Mathematics \textbf{1273}, Springer (1987).
\bibitem{GilleSzamuely}
Ph.~Gille \& T.~Szamuely, \emph{Central Simple Algebras and Galois Cohomology},
Cambridge Studies in Advanced Mathematics \textbf{101}, Cambridge University Press (2006).
\bibitem{GoodearlWarfield}
K.~Goodearl \& R.~Warfield, \emph{An introduction to noncommutative Noetherian rings}, London Mathematical Society Student Texts \textbf{61}, Cambridge University Press (2004).
\bibitem{EGAII}
A.~Grothendieck, \emph{\'El\'ements de g\'eom\'etrie alg\`ebrique II. \'Etude globale \'el\'ementaire de quelques classes de morphismes}, Inst. Hautes \'Etudes Sci. Publ. Math. \textbf{8} (1961), 222 pp.
\bibitem{Harada}
M.~Harada, \emph{Hereditary orders}, Trans. Amer. Math. Soc. \textbf{107} (1963) 273--290.
\bibitem{ResiduesDuality}
R.~Hartshorne, \emph{
Residues and duality}, Lecture Notes in Mathematics \textbf{20}, Springer,
Berlin--New York (1966).
\bibitem{Kussin}
D.~Kussin, \emph{Weighted noncommutative regular projective curves}, J. Noncommut. Geom. \textbf{10} (2016), no. 4, 1465--1540.
\bibitem{Lam}
T.~Lam,
\emph{Lectures on modules and rings},
Graduate Texts in Mathematics \textbf{189}, Springer--Verlag, New York, 1999.
\bibitem{LamFirstCourse}
T.~Lam, \emph{A first course in noncommutative rings},
Second edition. Graduate Texts in Mathematics \textbf{131}. Springer--Verlag, New York, 2001.
\bibitem{LaumonRapoportStuhler}
G.~Laumon, M.~Rapoport \& U.~Stuhler, \emph{D--elliptic sheaves and the Langlands correspondence}, Invent. Math. \textbf{113} (1993), no. 2, 217--338.
\bibitem{LekiliPolishchuk}
Y.~Lekili \& A.~Polishchuk, \emph{Auslander orders over nodal stacky curves and partially wrapped Fukaya categories}, J. Topol. \textbf{11} (2018), no. 3,
615--644.
\bibitem{LenzingReiten}
H.~Lenzing \& I.~Reiten, \emph{Hereditary Noetherian categories of positive Euler characteristic}, Math. Z. \textbf{254} (2006), no. 1, 133--171.
\bibitem{Liu}
Q.~Liu, \emph{Algebraic geometry and arithmetic curves}, Oxford Graduate Texts in Mathematics \textbf{6}. Oxford University Press, Oxford (2002).
\bibitem{Matsumura} H.~Matsumura, \emph{Commutative ring theory},
Cambridge Studies in Advanced Mathematics \textbf{8},
Cambridge University Press, Cambridge (1989).
\bibitem{Matlis}
E.~Matlis, \emph{Injective modules over Noetherian rings}, Pacific J. Math.
\textbf{8} (1958), 511--528.
\bibitem{Milne}
J.~Milne, \emph{\'Etale cohomology},
Princeton Mathematical Series \textbf{33}, Princeton University Press (1980).
\bibitem{Perego}
A.~Perego, \emph{A Gabriel theorem for coherent twisted sheaves}, Math. Z. \textbf{262} (2009), no. 3, 571--583.
\bibitem{Popescu} N.~Popescu, \emph{Abelian categories with applications to rings and
modules}, London Mathematical Society Monographs, no. \textbf{3},
Academic Press (1973).
\bibitem{Spiess}
M.~Spie\ss{}, \emph{Twists of Drinfeld--Stuhler modular varieties}, Doc. Math. 2010, Extra vol.: Andrei A. Suslin sixtieth birthday, 595--654.
\bibitem{ReinerMO}
I.~Reiner, \emph{Maximal orders}, London Mathematical Society Monographs, New Series \textbf{28}. The Clarendon Press, Oxford University Press, Oxford, 2003.
\bibitem{ReitenvandenBergh}
I.~Reiten \& M.~Van den Bergh, \emph{Noetherian hereditary abelian categories satisfying Serre duality}, J.~Amer.~Math.~Soc.~\textbf{15} (2002), no. 2, 295--366.
\bibitem{Toen}
B.~To\"en, \emph{Derived Azumaya algebras and generators for twisted derived categories},
Invent. Math. \textbf{189} (2012), no. 3, 581--652.
\end{thebibliography}
\end{document}