Linear response eigenvalue problem solved by extended locally optimal preconditioned conjugate gradient methods (Q341304)

Deflation techniques are often used to remove already computed eigenpairs from being recomputed and a common procedure is some kind of low-rank modification to the original matrix so as to displace the computed eigenvalues while keeping the unknown eigenvalues unchanged, or shifted in explicitely known ways. The authors propose improvements to their earlier LOBP4dCG algorithm for the linear response eigenvalue problem using a shifting deflation technique and extending the search space. Numerical examples demonstrate the efficiency of the deflation technique and show dramatic increase in the rate of convergence.