|Algorithm Design for Fast Linear Solvers|
|Allison Baker - Department of Applied Mathematics, University of Colorado|
John Dennis - Department of Computer Science, University of Colorado
Elizabeth Jessup - Department of Computer Science, University of Colorado
We describe a new technique for solving a sparse linear system $Ax = b$ as a block system $AX = B$, where multiple starting vectors and righthand sides are chosen so as to accelerate convergence. Efficiency is gained by reusing the matrix $A$ in block operations with $X$ and $B$. Techniques for reducing the cost of the extra matrix-vector operations are presented.