Matrices, moments and quadrature. II: How to compute the norm of the error iterative methods (Q1371672): Difference between revisions

[For part I see Pitman Res. Notes Math. Ser. 303, 105-156 (1994; Zbl 0795.65019).] Consider the equation \(Ax= b\) and let \(r= b-A\widetilde x\) be the residue for an approximate solution \(\widetilde x\). The energy norm of the error or equivalently the quadratic form \(r^T A^{-1}r\) is to be determined. More generally \(u^Tf(A)u\) is expressed as a Stieltjes integral (with the spectral decomposition of \(A\)) and computed via a quadrature formula. The latter is constructed via orthogonal polynomials, more precisely via a Lanczos procedure which in turn is equivalent to a conjugate gradient method. Reviewer's remark. It would be interesting to see whether (say) \(k\) steps of the computation of the error are related to the improvement of the approximate solution \(\widetilde x\) by \(k\) steps of the conjugate gradient algorithm.