Weak CMP inverses (Q6917810)

Several types of generalized inverses have been studied by \textit{J. Chen} et al. [Quaest. Math. 43, No. 9, 1331--1348 (2020; Zbl 1492.47002)]: \N\begin{itemize}\N\N\item The \(A^{(2)}_{E,F}\) inverse. For a matrix \(A \in \mathbb{C}^{m \times n}\) and subspaces \(E \subseteq \mathbb{C}^n\), \(F \subseteq \mathbb{C}^m\) with \(\dim(E) = s \leq \operatorname{rank}(A)\) and \(\dim(F) = m - s\), if there exists \(X\) such that\N\begin{align*}\NXAX &= X, \\ \mathcal{R}(X) &= E, \\ \mathcal{N}(X) &= F,\N\end{align*}\Nthen \(X\) is unique and denoted by \(A^{(2)}_{E,F}\). The notation \(A^{(1,2)}_{E,F}\) denotes an \(A^{(2)}_{E,F}\) inverse that also belongs to \(A\{1\}\) (i.e., it also satisfies \(AXA = A\)).\N\N\item Drazin and group inverses. For \(A \in \mathbb{C}^{n \times n}\) with index \(\operatorname{ind}(A) = k\) (the smallest integer such that \(\operatorname{rank}(A^k) = \operatorname{rank}(A^{k+1})\)), the {Drazin inverse} \(A^D = X\) is the unique matrix satisfying:\N\[\NXAX = X, \quad AX = XA, \quad A^{k+1}X = A^k.\N\]\NWhen \(\operatorname{ind}(A) = 1\), \(A^D\) reduces to the {group inverse} \(A^{\#}\). \NApplications of such inverses include Markov chains, differential and difference equations, and cryptography.\N\N\N\item Composite Drazin-Penrose inverses. Several inverses combine Drazin and Moore-Penrose (\(A^\dagger\)) properties:\N\N\begin{itemize}\N\item {DMP inverse}. For \(A \in \mathbb{C}^{n \times n}\) with \(\operatorname{ind}(A)=k\),\N\[\NA^{D,\dagger} = A^D A A^\dagger.\N\]\NIt uniquely satisfies:\N\[\NXAX = X, \quad XA = A A^D, \quad A^k X = A^k A^\dagger.\N\]\N\item {MPD inverse}:\N\[\NA^{\dagger,D} = A^\dagger A A^D.\N\]\NIt uniquely satisfies:\N\[\NXAX = X, \quad AX = A^D A, \quad XA^k = A^\dagger A^k .\N\tag{1}\N\]\N\item {Core-EP inverse}:\N\[\NA^{\dagger \copyright} = A^D A^k (A^k)^\dagger.\N\]\N\N\item {*Core-EP inverse}:\N\[\NA^{\copyright \dagger} = (A^k)^\dagger A^k A^D.\N\]\N\N\item {MPCEP inverse}:\N\[\NA^{\dagger,\dagger \copyright} = A^\dagger A A^{\dagger \copyright}.\N\]\N\N\item {*CEPMP inverse}:\N\[\NA^{\copyright \dagger, \dagger} = A^{\copyright \dagger} A A^\dagger.\N\]\N\end{itemize}\N\N\item CMP Inverse. For \(A \in \mathbb{C}^{n \times n}\) with \(\operatorname{ind}(A)=k\), the {CMP inverse} is defined as:\N\[\NA^{c,\dagger} = A^\dagger A A^D A A^\dagger.\N\]\NIt is the unique solution to either:\N\begin{align*}\NYAY &= Y, \quad AY = A A^D A A^\dagger, \quad Y A^k = A^\dagger A^k, \tag{2} \\\N\text{or} \quad YAY &= Y, \quad YA = A^\dagger A A^D A, \quad A^k Y = A^k A^\dagger. \tag{3}\N\end{align*}\NThis inverse has been extended to rectangular matrices, Hilbert space operators, and tensors.\N\N\item Weak Drazin Inverse. A {weak Drazin inverse} of \(A \in \mathbb{C}^{n \times n}\) with \(\operatorname{ind}(A)=k\) is any matrix \(X\) satisfying:\N\[\NX A^{k+1} = A^k. \N\]\NA {minimal rank weak Drazin inverse} additionally satisfies:\N\[\N\operatorname{rank}(X) = \operatorname{rank}(A^D).\N\]\NThese are not unique but are easier to compute than \(A^D\) and are useful in Markov chains, differential equations, etc. The true Drazin inverse \(A^D\) is the unique minimal rank weak Drazin inverse that commutes with \(A\). Dually, a {right weak Drazin inverse} satisfies \(A^{k+1} X = A^k\), and a {minimal rank right weak Drazin inverse} also satisfies \(\operatorname{rank}(X) = \operatorname{rank}(A^D)\).\N\N\item Previous ``Weak'' generalizations. Replacing \(A^D\) with a fixed minimal rank weak Drazin inverse \(X\) in system (1) leads to the {weak MPD inverse}:\N\[\NA^{w,\dagger,D} = A^\dagger X A,\N\]\Nwhich uniquely solves:\N\[\NYAY = Y, \quad AY = XA, \quad Y A^k = A^\dagger A^k.\N\]\NSimilarly, using a minimal rank {right} weak Drazin inverse \(Z\) gives the {weak DMP inverse}:\N\[\NA^{w,D,\dagger} = A Z A^\dagger.\N\]\NThe classical DMP and MPD inverses are special cases when \(X = A^D\) or \(Z = A^D\).\N\N\end{itemize}\N\N\NThe primary aim of the paper is to extend the CMP inverse by weakening its defining systems (2) and (3). Below is the strategy used by the author:\N\begin{itemize}\N\item Start with the defining systems for the CMP inverse;\N\item Replace the Drazin inverse \(A^D\) appearing in these systems with a {minimal rank weak Drazin inverse} \(X\), and/or a {minimal rank right weak Drazin inverse} \(Z\).\N\item Study the solvability of these new, less restrictive systems.\N\end{itemize}\N\NThe solutions to these extended systems will define new extended inverses, such as weak CMP inverses and right weak CMP inverses.\NThese new classes will contain the standard {CMP}, {MPCEP}, and {*CEPMP} inverses as special cases (when \(X = A^D\) or \(Z = A^D\)).\N and introduce {entirely new types} of generalized inverses not previously studied.\N\NTo summarize, the paper aims to:\N\begin{itemize}\N\item Develop characterizations and representations for weak CMP inverses;\N\item Prove perturbation formulae and bounds;\N\item Present applications for solving linear equations;\N\item Recovering classical Moore-Penrose inverse results as special cases.\N\end{itemize}