Discrete QMR and BCG in the numerical solution of linear systems of ODEs (Q1298534): Difference between revisions

Three Krylov subspace methods in Hilbert space for an approximate solution of the abstract equation of the following form \[ Ax=b \] are compared, namely the quasi-minimal residuals method (QMR), the biorthogonal Lanczos process (BCG), and GMRES. The goal is to apply BCG or QMR rather than GMRES, since GMRES, if not restarted, produces a triangular matrix of the dimension of the Krylov subspace. The dimension of the Krylov subspace determines the level of approximation of the problem, hence it is assumed to be large enough. Two other methods deal with the tridiagonal matrix only. However one has to pay for this advantage with weaker convergence. The first result of the paper compares the rate of convergence to zero of the residuals of BCG and QMR with this one of GMRES. Then the author proves the convergence of the considered algorithms in the context of the (finite-dimensional) approximation of the original problem (now \(A=I-K\)), using the technique of \textit{collectively compact operator approximation} of P. M. Anselone. The last section is devoted to an application of the discussed methods to systems of linear ordinary differential equations (ODEs). The paper contains numerical examples.