Identity of the conjugate gradient and Lanczos algorithms for matrix inversion in lattice fermion calculations
Abstract
Two of the methods that are widely used in lattice gauge theory calculations requiring inversion of the fermion matrix are the Lanczos and the conjugate gradient algorithms. Those algorithms are already known to be closely related. In fact for matrix inversion, in exact arithmetic, they give identical results at each iteration and are just alternative formulations of a single algorithm. This equivalence survives rounding errors. We give the identities between the coefficients of the two formulations, enabling many of the best features of them to be combined.
- Publication:
-
Physics Letters B
- Pub Date:
- April 1988
- DOI:
- 10.1016/0370-2693(88)90401-7
- Bibcode:
- 1988PhLB..205...69B