Accelerated subspace iteration with aggressive shift

摘要: The subspace iteration method is a very classical for solving large general eigenvalue problems, and it accepted as one of the reliable methods to solve size problems through 1970-1980s. However, less efficient than Lanczos in terms CPU time, because its parameters procedure were selected today's small medium problems. In last 30 years, researchers have been trying accelerate different ways, such as, power acceleration, relaxation so that can deal with larger arising finite element analysis. Shifting technique recognized an way speed up convergence rate However shifting cost expensive thus makes unacceptable. That why almost all improvements 20 years did not shifts. this paper, aggressive strategy proposed based on computable criterion involving both eigenvector instead only. A wide range numerical tests shows greatly decrease time.