Robust Conjugate Directions Method for Solving Linear Systems

摘要

The method of conjugate gradients is the most popular of several conjugate directions methods. In this paper, a second conjugate directions method, known as A-minimal iterations, is examined. It differs from conjugate gradients primarily in that the direction vectors are computed in a manner independent of the observations. When the observations are the dominant noise source in the problem, this difference makes A-minimal iterations the more robust of the two methods. Examples from the area of signal extrapolation support this assertion. (Author)