Residual, restarting, and Richardson iteration for the matrix exponential (Q2847717)

The matrix exponential residual within Krylov subspace methods is explained. Then, it is shown how the Chebyshev iterations can be modified to adopt the residual control. This is followed by the presentation of some simple residual-based error estimates and the Richardson iteration for the matrix exponential. Numerical tests show that the proposed residual notion provides a reliable stopping criterion for the iterative methods for computing the matrix exponential. This resolves the question of reliable stopping criteria for these methods. Furthermore, the residual concept seems to set up a whole framework for a new class of methods for evaluating the matrix exponential. Some basic methods of this class are proposed.