scientific article; zbMATH DE number 3986540
accurate inner productscomputer arithmeticconjugate gradient algorithmconvergencedirected roudingseigenvaluesharmonic continuationill-conditioned systemsill-posed problemsinterval arithmeticnumerical experimentsQR algorithmresidual correctionsstiff systems
Analytic continuation of functions of one complex variable (30B40) General theory of numerical methods in complex analysis (potential theory, etc.) (65E05) Numerical linear algebra (65Fxx) Interval and finite arithmetic (65G30) Roundoff error (65G50) Numerical methods for initial value problems involving ordinary differential equations (65L05)
[For the entire collection see Zbl 0597.00030.] The paper contains the results of some numerical experiments in which some new kinds of computer arithmetic are compared with conventional arithmetics. The first examples are ill-posed problems, in particular the harmonic continuation of a function, after that some stiff systems of differential equations. Both classes of problems lead to ill-conditioned systems of linear equations. For the solution of these linear systems a so-called E-method - which is an arithmetic containing accurate inner products combined with directed roudings, interval arithmetic, and residual corrections - is shown to be superior to conventional methods. In the further examples the QR algorithm for computing eigenvalues and the conjugate gradient algorithm for the iterative solution of a linear system are treated. It is demonstrated how far the use of accurate inner products can reduce rounding errors and can influence the convergence behaviour of the algorithms.
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3750041)