Subspace iteration repeatedly applies a matrix to several independent vectors and orthonormalizes them, converging to a dominant invariant subspace when a spectral gap separates the selected eigenvalue cluster.
A dominant invariant subspace is spanned by eigenvectors whose eigenvalue moduli exceed those outside the subspace.
Suppose a finite-dimensional matrix is diagonalizable and its eigenvalues of largest modulus form a cluster, separated in modulus from the remaining eigenvalues. If the initial vector has a nonzero projection onto the sum of their eigenspaces, normalized power method iterates approach in distance, but need not converge to one vector.
For example, and initial vector give alternating limiting directions in the first two coordinates. If the initial vector has no component in , it never discovers that cluster. Without diagonalizability, Jordan blocks introduce polynomial growth factors, so a statement about the dominant subspace requires additional care.
Articles by others on the same topic
There are currently no matching articles.