The Euler-Maclaurin expansion for the Cauchy principal value integral (Q800687): Difference between revisions

Let \(Q^{(m)}f\) be the m-copy of the quadrature rule approximation Qf or \(Q^{(1)}f\) to the finite integral If having integrand f and interval [a,b]. For most rules Q and integrand functions f, \(Q^{(m)}f\) converges to If as m becomes infinite. For some Q and f, asymptotic expansions in m exist for the error functional \(Q^{(m)}f\)-If. The classical example is the Euler-Maclaurin expansion valid when Q is the trapezoidal rule and \(f\in C^{(p)}[a,b]\). This expansion, in inverse power of m, is the familiar one on which the classical version of Romberg integration is based. Several other expansions of this type have been reported since 1960. In particular, when f(x) has algebraic or logarithmic singularities at the endpoints a,b but is \(C^{(p)}(a,b)\), a significantly different expansion is known. In this paper, the theory is extended once more to cover functions like this, which may in addition have simple poles on the integration interval. In this case If is the Cauchy principal value integral. One result is that, when Q is the trapezoidal rule, \(f(x)=(x- c)^{-1}\phi (x)\), \(0=a<c<b=1\), then the expansion is the same as it would have been if this singularity at c were absent, except that a single term \(S^{[m,1]}f\) should be added to the expansion. This term is either \(\pi\) \(\phi(c)\cot \pi mc\) or \(-\phi'(c)/m\) according as \(mc\) is not or is an integer. A more general result is established and similar expansions for the ''CPV'' Fourier coefficients \(\int f(x)e^{-ikx}dx\) in inverse powers of k are derived.