A block Chebyshev-Davidson method for linear response eigenvalue problems (Q503482)

The authors propose a Chebyshev-Davidson-type method for computing a few smallest positive eigenvalues and the corresponding eigenvectors of linear response eigenvalue problems. The coefficient matrix for these problems is a two-by-two block matrix whose principal diagonal blocks are zero and whose off-diagonal blocks are Hermitian, such that one of them is positive definite while the other one is indefinite. The method presented applies to general problems where some purely imaginary eigenvalues may exist, thereby extending the existing method. In the presence of a certain condition, a tight upper bound is obtained for the Chebyshev filter. For the situation when this condition is not applicable, an adaptive strategy for updating the estimated upper bound is employed. An estimate for the rate of convergence of the Ritz values is obtained. Numerical examples are presented to illustrate the effectiveness of the proposed method.