Balancing sparse Hamiltonian eigenproblems (Q2368735)

The paper is devoted to the theory of balancing (construction of suitable similarity transformations) for eigenvalue problems with sparse Hamiltonian matrices. Some such transformations are suggested to use after reduction of the matrix to irreducable form (a decomposition into smaller-order problems is useful). Computation of a symplectic scaling matrix \(D\) is discussed with the hope to decrease the norm of \(D^{-1}HD\). The main attention is paid to algorithms which do not require the knowledge of elements of \(H\) (only matrix-vector products are necessary). Numerical examples are given for ``practically relevant'' problems.