Linear Solvers



next up previous
Next: Local optimization methods Up: Optimization Modules Previous: Optimization Modules

Linear Solvers

All of the linear solvers in COOOLare iterative. Descriptions of the algorithms can be found in in the book by Scales and Smith listed at the end. Presently COOOL implements two flavors of row action methods: ART, which stands for Algebraic Reconstruction Technique, and SIRT, which stands for Simultaneous Iterative Reconstruction Technique. These methods are especially attractive when problems are too big to fit into memory. When the matrix elements will fit into memory, Conjugate Gradient Least Squares (CGLS) is very attractive. Finally, we implement a version of the IRLS algorithm (Iteratively Re-weighted Least-Squares) to efficiently solve linear systems in the norm. This algorithm takes advantage of certain approximations to achieve nearly the speed of conventional least squares methods while being able to robustly handle long-tailed noise distributions.




Sun Feb 25 12:08:00 MST 1996