Consideration on computing real logarithms of matrices, Hamiltonian logarithms, and skew-symmetric logarithms (Q1923207)

The author considers some old and some new methods for approximating real logarithms of matrices. More attention is paid to Padé approximation, Newton's method, and a series expansion method. The author proves that the diagonal Padé approximants and the series expansion technique are structure-preserving, i.e. recover the skew-symmetric and Hamiltonian logarithms associated with orthogonal and symplectic matrices, respectively. For Newton's method the algorithmic formulation is simplified. Moreover, it is shown that Newton's method can be used for approximating logarithms of a matrix \(\mathcal T\) which are not primary matrix functions of \(\mathcal T\).