Explicit formulae for the greatest least-squares and minimum norm \(g\)-inverses and the unique group inverse of matrices over commutative residuated dioids (Q309947)

This paper deals with some properties of matrices \(A\) over any commutative residuated dioids. A dioid \({\mathcal D}(\oplus,\otimes)\) is residuated if {\parindent=0.6cm\begin{itemize} \item[1.] for all \(a, b \in {\mathcal D}\), \(\{x \in {\mathcal D} : ax \leq b \}\) admits a maximal element denoted by \(a \bar{\varnothing} b\), \item[2.] \(\{x \in {\mathcal D} : xa \leq b \}\) admits a maximal element denoted by \(b \varnothing a\), \item[3.] \(({\mathcal D}, \leq)\) is a lattice. \end{itemize}} Consider the following matrix equations for a given matrix \(A\) {\parindent=0.8cm\begin{itemize}\item[(1)] \(AXA=A\), \item[(2)] \(XAX=X\), \item[(3)] \((AX)^T=AX\), \item[(4)] \((XA)^T=XA\). \end{itemize}} Let \(A\{i,j,\dots,k\}\) denote the set of matrices \(X\) which satisfy equations (\(i\)), (\(j\)), \dots, (\(k\)). Any matrix in \(A\{1,2 \}\) is called a reflexive \(g\)-inverse or a Thierrin-Vagner inverse of \(A\). Any matrix in \(A\{1,3 \}\) is called a least squares \(g\)-inverse and any matrix in \(A\{1,4 \}\) is called a minimum norm \(g\)-inverse. In this paper, the authors propose explicit formulae for the greatest least-squares and minimum norm \(g\)-inverses of a matrix \(A\) over a commutative residuated dioid. Specifically, they prove that if \({\mathcal D}\) is a commutative residuated dioid, then equation \(A^TAX=A^T\) has a solution if and only if \((A^TA) \bar{\varnothing} A^T\) is its solution, in which case \((A^TA) \bar{\varnothing} A^T\) is the greatest least-squares \(g\)-inverse of \(A\). On the other hand, the equation \(XAA^T=A^T\) has a solution if and only if \(A^T \varnothing (AA^T)\) is its solution, in which case \(A^T \varnothing (AA^T)\) is the greatest minimum norm \(g\)-inverse of \(A\). In addition, the authors prove that the unique group inverse of an \(n \times n\) matrix with entries from a residuated dioid whenever existent is \[ A^{\sharp}=(A \varnothing A^2) A (A^2 \bar{\varnothing} A). \] In particular, they propose formulae for the aforementioned generalized inverses of \(A\) over Boolean algebras, max-plus semirings and a class of complete and completely distributive lattices. The main results remain valid in the context of residuated semigroups.