\(SR\) and \(SZ\) algorithms for the symplectic (butterfly) eigenproblem (Q1301293)

Transformation algorithms to compute eigenvalues and eigenvectors of symplectic (\(J\)-orthogonal) matrices are studied. First the matrix is transformed into a butterfly form, a two by two block matrix consisting of two tridiagonal and two diagonal blocks. An \(SR\) algorithm, built up around symplectic triangular factorization of such butterfly matrices is described. Laurent polynomials are used to accelerate convergence, and explicit and implicit application is described. Departure from symplectic structure caused by rounding errors is corrected in a simple way each iteration. A factorization may replace the butterfly matrix with a symplectic pencil, where one matrix is upper triangular with diagonal blocks and the other contains one tridiagonal and two unit matrix blocks. The \(SZ\) algorithm works on this factorization and keeps the special symplectic structure throughout its application. Numerical results on random matrices and a set of benchmark cases are reported.