The generalized bisymmetric solutions of the matrix equation \(A_{1} X_{1} B_{1} + A_{2} X_{2} B_{2} + \dots + A_{l} X_{l} B_{l} = C\) and its optimal approximation (Q1014360)

The authors discuss a very popular problem being motivated with a number of applications in theoretical and experimental mechanics. The need to solve a special type of matrix equation follows from many problems of identification based on experimental data sets or direct parameter fitting. Special sub-classes of the system investigated can be used for efficient solution of large systems produced by a Kronecker algebra instead of their direct solution. The paper itself deals with a formulation and detailed mathematical analysis of an iterative method is constructed to find the generalized bisymmetric solutions of the matrix equation \(A_{1} X_{1} B_{1} + A_{2} X_{2} B_{2} + \dots + A_{l} X_{l} B_{l} = C\) where \([X_{1},X_{2}, \dots ,X_{l}]\) is a group of real matrices. A matrix \(X\) is considered to be generalized bisymmetric in the meaning: \(X = X^{T} = PXP\), where \(P^{T} = P\), \(P^{2} = I\). Two problems are discussed. Problem I regards the solution of the problem itself. Particular steps can be characterized as follows: by iterative method, the solvability of the matrix equation can be judged automatically. When the matrix equation is consistent, for any initial generalized bisymmetric matrix group \(\left[X_1^{(0)},X_2^{(0)},\cdots,X_l^{(0)}\right]\), a generalized bisymmetric solution group can be obtained within finite iteration steps in the absence of roundoff errors, and the least norm generalized bisymmetric solution group can be obtained by choosing a special kind of the initial generalized bisymmetric matrix group. The problem II is focused to the optimal approximation of the generalized bisymmetric solution group with respect to a given generalized bisymmetric matrix group \(\left[\overline{X}_1,\overline{X}_2,\cdots,\overline{X}_l\right]\) in Frobenius norm. It can be obtained by finding the least norm generalized bisymmetric solution group of the new matrix equation \(A_1\widetilde{X}_1B_1+A_2\widetilde{X}_2B_2+\cdots+A_l\widetilde{X}_lB_l=\widetilde{C}\), where \(\widetilde{C}=C-A_1\overline{X}_1B_1-A_2\overline{X}_2B_2-\cdots-A_l\overline{X}_lB_l\). A numerical example is given demonstrating the efficiency of the algorithm developed. Solution of both problem I and problem II are outlined together with accuracy estimations. Convergence observation and numerical stability are estimated. The list of references is rather limited being oriented exclusively to the problem itself without any link to other areas.