Continuation of invariant subspaces

From MaRDI portal





The basic problem of computing a smooth similarity transformation to block triangular form of a matrix valued real function is formulated. The continuation of invariant subspaces (CIS) algorithm generates a smooth path of orthogonal matrix factorizations. It can be viewed as a continuous analog of standard linear algebra techniques. In the context of smooth block Schur factorization, the authors use the underlying differential equations governing the evolution of the singular value decomposition factors in order to obtain initial approximations of the exact factors. An adaptive Newton iteration is then used to refine these approximations. The paper contains a proof of the existence of a \(C^k\) orthogonal parameter dependent function \(Q(t)\), \(t \in \langle 0,1 \rangle ,\) by which the block triangularization of a given \(C^k\) matrix valued real function \(A(t)\), \(t \in \langle 0,1 \rangle ,\) is achieved. The CIS algorithm is described and its implementation aspects are discussed. Numerical experiments of both ``easy problems (well-separated blocks and slowly varying eigenvalues and subspaces) and ``harder problems (weakly separated blocks and/or rapidly varying eigenvalues or subspaces) are analyzed.




Cited in
(19)








This page was built for publication: Continuation of invariant subspaces

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4828882)