Generalized reflexive and anti-reflexive solutions of \(AX=B\) (Q260137): Difference between revisions

From MaRDI portal
ReferenceBot (talk | contribs)
Changed an Item
Normalize DOI.
Property / DOI
 
Property / DOI: 10.1007/s10092-014-0136-6 / rank
Normal rank
 
Property / DOI
 
Property / DOI: 10.1007/S10092-014-0136-6 / rank
 
Normal rank

Revision as of 15:15, 8 December 2024

scientific article
Language Label Description Also known as
English
Generalized reflexive and anti-reflexive solutions of \(AX=B\)
scientific article

    Statements

    Generalized reflexive and anti-reflexive solutions of \(AX=B\) (English)
    0 references
    0 references
    0 references
    0 references
    18 March 2016
    0 references
    All matrices considered are over the complex numbers. A generalized reflection matrix is a square matrix \(P\) such that \(P=P^{\ast}\) and \(P^{2}=I\). If \(P\) and \(Q\) are generalized reflection matrices of sizes \(m\times m\) and \(n\times n\), respectively, then an \(m\times n\) matrix \(A\) is called \((P,Q)\)-generalized reflexive (-generalized antireflexive) if \(A=PAQ\) (\(A=-PAQ\)), respectively. Let \(A\) and \(B\) be arbitrary matrices of sizes \(m\times n\) and \(m\times k\). Then the equation (*) \(AX=B\) has a solution if and only if \(AA\dag B=B\), where \(A^{\dag}\) is the Moore-Penrose inverse of \(A\), and in the latter case all solutions of (*) are given by \(X=A^{\dag}B+(I-A^{\dag}A)V\) as \(V\) runs over the \(n\times k\) matrices. If \(P\) and \(Q\) are generalized reflection matrices, then the authors give necessary and sufficient conditions for (*) to be solvable by a \((P,Q)\)-generalized reflective matrix \(X\) as follows: Define \[ A_{1}:=\bigl[ \begin{matrix} A(I+P)\\A(I-P) \end{matrix} \bigr] \] and \[ B_{1}:=\bigl[ \begin{matrix} B(I+Q)\\B(I-Q) \end{matrix} \bigr]. \] Then (*) has a \((P,Q)\)-generalized reflexive solution if and only if \(A_{1}A_{1}^{\dag}B_{1}=B_{1}\) and if the latter holds, then all \((P,Q)\)-generalized reflexive solutions are given by \(X=A_{1}^\dag B_{1}+(I-A_{1}^{\dag}A_{1})V\) as \(V\) runs over the \((P,Q)\)-generalized reflexive \(n\times k\) matrices. Furthermore, if (*) has a generalized reflexive solution and \(C\) is an arbitrary \(n\times k\) matrix, then the unique solution \(X\) for which the Frobenius norm \(\left\| X-C\right\| _{F}\) is minimized is defined by taking \(V=\frac{1}{2}(C+PCQ)\). Similar results hold for \((P,Q)\)-generalized antireflexive solutions. These theorems give alternative, and perhaps simpler, solutions to the problems considered by \textit{J.-C. Zhang} et al. [Appl. Math. Comput. 209, No. 2, 254--258 (2009; Zbl 1168.15008)].
    0 references
    0 references
    generalized reflexive matrix equation
    0 references
    matrix nearness problem
    0 references
    \((P, Q)\) generalized reflexive solution
    0 references
    \((P, Q)\) generalized anti-reflexive solution
    0 references
    Moore-Penrose inverse
    0 references

    Identifiers