End-point corrected trapezoidal quadrature rules for singular functions (Q753403)

From MaRDI portal
scientific article
Language Label Description Also known as
English
End-point corrected trapezoidal quadrature rules for singular functions
scientific article

    Statements

    End-point corrected trapezoidal quadrature rules for singular functions (English)
    0 references
    1990
    0 references
    A natural quadrature problem is to obtain rules of the form \(Qf=\sum w_ if(x_ i),\) for integration over [0,1], which are exact for \(f(x)=s(x)g_ 1(x)+g_ 2(x),\) where s(x) is a specified weight function and \(g_ 1\) and \(g_ 2\) are polynomials of degree k-1 or less. The author considers in particular \(s(x)=\ln x\) and \(s(x)=x^{\gamma}\) and obtains Gregory-type rules. The generalization leaves the n-panel trapezoidal rule component and the standard k-term right-hand correction intact but replaces the left-hand correction by one which depends on s(x) but not on n, and uses 2k points, equally spaced in the first interval [0,1/n]. The n-panel rule with \(k=6\) has the form \[ Qf=1/n(\sum^{6}_{j=1}\lambda^ 3_ jf(j/6n)+\sum '{}^{n}_{i=1}f_ i-1/24(f_{n-2}-4f_{n-1}+3f_ n)), \] where \(f_ j=f(j/n)\) and \(\lambda^ k_ j\), essentially determined from solving 2k linear moment equations, are listed for \(s(x)=\ln x\) and \(s(x)=x^{\gamma}\), \(\gamma =\pm 1/2,\pm 1/3,-9/10.\) It is refreshing to note that the author makes no particular claim for the numerical accuracy of these rules. He demonstrates numerically that successively increasing n (keeping \(k=3)\) results in the expected fourth- order convergence, the numerical accuracy being comparable to that of the standard Gregory's formula applied to a corresponding regular function. The author discusses problems associated with constructing such rules for other more sophisticated singularities s(x). He treats these problems in depth, particularly those relating to the existence of solutions, the condition error of the resulting formula, and the question of whether or not \(\lambda^ k_ j\) above depends also on n. This reviewer notes with regret that the author has limited the scope of his paper so as to exclude Gaussian-type rules for this problem. He has also excluded numerical comparison with extrapolation methods, which are, of course, a natural and simple choice for this sort of problem.
    0 references
    end-point corrected trapezoidal quadrature rules
    0 references
    Gregory quadrature rule
    0 references
    singular integrand
    0 references
    fourth-order convergence
    0 references
    0 references
    0 references

    Identifiers