Lecture 18 Summary

« Previous: Lecture 17 Summary Next: Lecture 19 Summary »

New topic: iterative methods for Ax=b linear equations. Started discussing the GMRES method, which is the direct analogue of Arnoldi for linear equations.

Derived the GMRES method as in lecture 35 of Trefethen, as residual minimization in the Krylov space using Arnoldi's orthonormal basis Qn. Like Arnoldi, this is too expensive to run for many steps without restarting. Unlike Arnoldi, there isn't a clear solution (yet) for a good restarting scheme, and in particular there are problems where restarted GMRES fails to converge; in that case, you can try restarting after a different number of steps, try a different algorithm, or find a better preconditioner (a topic for later lectures).

Just as Arnoldi reduces to Lanczos for Hermitian matrices, GMRES reduces to MINRES, which is a cheap recurrence with no requirement for restarting. Briefly discussed MINRES, the fact that it converges but has worse rounding errors.

Began discussing gradient-based iterative solvers for Ax=b linear systems, starting with the case where A is Hermitian positive-definite. Our goal is the conjugate-gradient method, but we start with a simpler technique. First, we cast this as a minimization problem for f(x)=x*Ax-x*b-b*x. Then, we perform 1d line minimizations of f(x+αd) for some direction d. If we choose the directions d to be the steepest-descent directions b-Ax, this gives the steepest-descent method.