Block linear method for large scale Sylvester equations (Q959471)

The author proposes a new iterative scheme based on fixed point iteration for solving the Sylvester matrix equation \(AX-XB=C\), where \(X\) is of dimension \(n\times p\). If \(n>p\), it is convenient to transpose the Sylvester equation. The scheme requires the solution of a linear system with multiple right-hand sides at each iteration, which can be solved by using block Krylov subspace methods for linear systems, without solving a low-dimensional Sylvester equation at every iteration. Numerical experiments show that the new method works well for large-scale problems and is competitive with other methods. Mathematical results on the convergence of the iterative scheme is provided. A stable and inexpensive way of computing the residual matrix is presented which provides an efficient stopping criterion.