Iterative Solution of Indefinite Symmetric Linear Systems by Methods Using Orthogonal Polynomials over Two Disjoint Intervals (Q3339175)

From MaRDI portal

!

This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:

scientific article; zbMATH DE number 3872563
Language Label Description Also known as
default for all languages
No label defined
    English
    Iterative Solution of Indefinite Symmetric Linear Systems by Methods Using Orthogonal Polynomials over Two Disjoint Intervals
    scientific article; zbMATH DE number 3872563

      Statements

      Iterative Solution of Indefinite Symmetric Linear Systems by Methods Using Orthogonal Polynomials over Two Disjoint Intervals (English)
      0 references
      0 references
      1983
      0 references
      orthogonal polynomials
      0 references
      adaptive parameter estimation
      0 references
      indefinite symmetric linear systems
      0 references
      large invertible symmetric matrix
      0 references
      Chebyshev semi-iterative method
      0 references
      conjugate gradient
      0 references
      conjugate residual methods
      0 references
      Chebyshev polynomials
      0 references
      convergence
      0 references
      Lanczos algorithm
      0 references
      Rayleigh-Ritz type estimates
      0 references
      Numerical experiments
      0 references
      comparison
      0 references
      SYMMLQ algorithm
      0 references
      As the title indicates, the author is concerned with the iterative solution of indefinite symmetric linear systems. In this case, if the linear system is written as \(Ax=f\), A is a large invertible symmetric matrix with both positive and negative eigenvalues. The author's approach is to regard the spectrum of A as contained in the union of two disjoint intervals, [a,b]\(\cup [c,d]\) with \(b<0<c\). A generalized Chebyshev semi- iterative method can then be defined by generating the set of polynomials orthogonal with respect to the weighted integral inner product given by \(\int^{b}_{a}p_ 1(\lambda)p_ 2(\lambda)w_ 1(\lambda)d\lambda +\int^{d}_{c}p_ 1(\lambda)p_ 2(\lambda)w_ 2(\lambda)d\lambda\), where \(w_ 1(\lambda)\) and \(w_ 2(\lambda)\) represent the usual Chebyshev weight functions for [a,b] and [c,d] respectively. Similarities to conjugate gradient and conjugate residual methods are noted, and such algorithms are also discussed. The paper is largely self-contained. The theory of polynomials orthogonal over two disjoint intervals is developed and algorithms are presented reflecting the theory. A priori error bounds are established in terms of the Chebyshev polynomials for \(A^ 2\) over [a,d] in order to prove convergence of the method. Estimates of the uniform norm of the orthogonal polynomials are also presented. An application to interior eigenvalue problems is also mentioned. The practical problem of estimating the parameters a,b,c,d in an adaptive fashion is attacked. The extreme values a and d can be found using well- known methods such as the Lanczos algorithm. The interior parameters b and c can be estimated by iterating with a polynomial with fixed degree and then using Rayleigh-Ritz type estimates. Numerical experiments and comparison with the SYMMLQ algorithm of \textit{C. C. Paige} and \textit{M. A. Saunders} [ibid. 12, 617-629 (1975; Zbl 0319.65025)] are presented.
      0 references

      Identifiers

      0 references
      0 references
      0 references
      0 references
      0 references
      0 references