|« Previous: Lecture 15 Summary||Next: Lecture 17 Summary »|
Derive the Arnoldi algorithm. Unlike the book, I start the derivation by considering it to be modified Gram-Schmidt on (q1,Aq1,Aq2,Aq3,...). In the next lecture, we will re-interpret this as a partial Hessenberg factorization.
Discussed what it means to stop Arnoldi after n<m iterations: finding the "best" solution in the Krylov subspace Kn. Discussed the general principle of Rayleigh-Ritz methods for approximately solving the eigenproblem in a subspace: finding the Ritz vectors/values (= eigenvector/value approximations) with a residual perpendicular to the subspace (a special case of a Galerkin method). Also showed that the max/min Ritz values are the maximum/minimum of the Rayleigh quotient in the subspace.
Mentioned another iterative method for eigenvalues of Hermitian matrices: turning into a problem of maximizing or minimizing the Rayleigh quotient. In particular, we will see later in the class how this works beautifully with the nonlinear conjugate-gradient algorithm. More generally, we will see that there is often a deep connection between solving linear equations and optimization problems/algorithms—often, the former can be turned into the latter or vice versa.
Lanczos: Arnoldi for Hermitian matrices. Showed that in this case we get a three-term recurrence for the tridiagonal reduction of A. (Derivation somewhat different than in the book. By using A=A* and the construction of the q vectors, showed explicitly that qj*v=qj*Aqn=0 for j<n-1, and that for j=n gives |v|=βn from the n-th step. Hence Arnoldi reduces to a three-term recurrence, and the Ritz matrix is tridiagonal.)
Noted that Arnoldi requires Θ(mn2) operations and Θ(mn) storage. If we only care about the eigenvalues and not the eigenvectors, Lanczos requires Θ(mn) operations and Θ(m+n) storage. However, this is complicated by rounding problems, as discussed in the next lecture.