The matrix nearness problem for symmetric matrices associated with the matrix equation \([A^{T}XA, B^{T}XB] = [C, D]\) (Q855588)

Using the projection theorem in inner product spaces, as well as the generalized singuar value decomposition (GSVD) and the canonical correction decomposition of matrix pairs, an analytic expression of the solution is obtained. In detail, it can be divided into three phases: first, a least-square symmetric solution \(X_0 \) of the matrix equation \[ [A^T XA, B^T XB]=[C,D] \] by using the GSVD is obtained; then utilizing the solution \(X_0\) and the projection theorem, the problem \(\{\exp(-\tau A)\nu\}\), for a symmetric and positive semidefinite \(A\) is considered. The exponential function is a quickly decaying function. Hence the vector \(\exp(-\tau A)\nu\) is mostly determined by the smallest eigenvalues and their corresponding invariant subspaces. The Lanczos method is applied to the inverse of \((I + \gamma A)\) with \(\gamma>0 \), to emphasizes the first eigenvalues and their eigenvectors using an inner-outer iteration. The convergence speed is independent of the norm of the matrix. In addition, a strategy for choosing the tolerances for the errors in the solution of the shifted system is proposed. The numerical results indicate the effectiveness of the method.