The bisymmetric solutions of the matrix equation \(A_{1}X_{1}B_{1}+A_{2}X_{2}B_{2}+\cdots+A_{l}X_{l}B_{l}=C\) and its optimal approximation (Q996312)

The authors propose an iterative method for finding the bisymmetric solutions of the matrix equation \(A_1X_1B_1+A_2X_2B_2+\cdots +A_lX_lB_l=C~(*)\) where \([X_1,\dots ,X_l]\) is a real matrix group. (A square matrix is bisymmetric if it is symmetric w.r.t. both the diagonal and the antidiagonal.) The method gives also the answer to the question whether equation \((*)\) is consistent or not. When it is, then for any initial bisymmetric matrix group \([X_1^{(0)},\ldots ,X_l^{(0)}]\) a solution is obtained within finitely many iteration steps in the absence of roundoff errors, and the choice of a special initial matrix group allows one to find the least norm bisymmetric solution group (LNBSG). The optimal approximation bisymmetric solution group to a given group \([\bar{X}_1,\dots ,\bar{X}_l]\) in the Frobenius norm is obtained by finding the LNBSG of equation \((*)\) with right-hand side \(C-A_1\bar{X}_1B_1-\cdots -A_l\bar{X}_lB_l\).