“…Thus, the elimination of the eigenvalues with very small values as well as the corresponding eigenvectors can retain the principal information of this matrix, and as a result an approximated matrix with reduced rank can be obtained, which is only slightly different from the original one. A similar idea is also widely used in principal component analysis
23 or lossy compression of images
24 . Note that for the elemental inelastic state matrix
, if it is approximated based on the above idea, only a small error will be introduced into the perturbation expansion expression of the structural tangent stiffness and the results of the Woodbury formula.…”