Superlinear convergence in minimum residual iterations
Igor Kaporin · Numerical Linear Algebra with Applications · 2005
Abstract The superlinear convergence of minimum residual‐type methods for solving systems of linear equations with diagonalizable non‐singular unsymmetric matrix is estimated using a special conditioning measure. For the construction of the latter, the distance from the spectrum of the matrix to the origin and the Frobenius distance between the matrix and the identity is used. Asymptotical exactness of the presented result is discussed theoretically and illustrated numerically. Copyright © 2005 John Wiley & Sons, Ltd.