Numerical solution of a quadratic eigenvalue problem (Q5916489): Difference between revisions

The main purpose of the author is to identify a class of quadratic eigenvalue problems (QEP's), for which the use of solvent approach is really successful. The considered special case is \((\lambda ^{2}M+\lambda G+K)x=0\), where \(M=M^{T}\) is positive definite, \(K=K^{T}\) is negative definite and \(G=-G^{T}\). An algorithm is presented which preserves the Hamiltonian structure of the spectrum of the QEP. In the case when \(2(M-K)\) is not ill-conditioned, the algorithm reveals very efficient and exhibits a very good accuracy for all the eigenvalues, except those with very large modulus. Interesting numerical examples enlighten the features of the presented method.