An efficient algorithm for the generalized \((P,Q)\)-reflexive solution to a quaternion matrix equation and its optimal approximation (Q2511459): Difference between revisions

A reflection matrix satisfies \(P=P^H\) and \(P^2=I\). A matrix \(A\) is generalized \((P,Q)\)-reflexive if \(A=PAQ\). The problem is to solve \(\sum_{l=1}^n A_lXB_l+\sum_{s=1}^v C_s\overline{X}D_s=F\) for a generalized \((P,Q)\)-reflexive matrix \(X\) where all the items in the equation are matrices over the quaternions \(\mathbb{H}\). The unknown \(X\in\mathbb{H}^{n\times n}\) (and \(\overline{X}\) is its conjugate), \(F\in \mathbb{H}^{p\times q}\) and the other matrices have corresponding dimensions. An iterative algorithm is given and it is shown that if the equation is consistent, then a solution is computed in a finite number of steps if exact arithmetic is used. By choosing a particular form for the initial condition, a solution is computed with minimal Frobenius norm. The problem can be transformed such that a solution of the new problem gives a solution closest to a given matrix in Frobenius norm. The proof runs in an inner product space of \(\mathbb{H}^{m\times n}\) over the field \(\mathbb{R}\) with inner product \(\langle A,B\rangle=\mathrm{Re}[\mathrm{tr}(A^HB)]\) and associated Frobenius norm.