Application of a modified FFT to product type integration (Q1184119): Difference between revisions

An automatic integration scheme is proposed for evaluating the integral \(Q=Q(K,f)=\int^ y_ xK(t)f(t)dt\), \(-1\leq x,y\leq 1\), where \(f(t)\) is assumed to be a smooth function. For the singular or badly-behaved function \(K(t)\) the cases \(\ln| t-c|\), \(| t-c|^ \alpha\), \(\alpha > -1\), Cauchy principal value \(1/(t-c)\) and \(e^{i\omega t}\), \(|\omega|\gg 1\) are discussed. The function \(f\) is approximated by a truncated Chebyshev series \(p_ N(t)\) of degree \(N\), whose coefficients are efficiently computed using the FFT. The approximation \(Q_ N(K,f)\) to the integral \(Q\) is given by \(Q(K,p_ N)\). The sequence \(\{p_ N\}\) is recursively generated until the required tolerance for the integral is satisfied. The recursion is numerically stable. For the automatic quadrature the degree \(N\) is increased more slowly than doubling. The evaluations of \(Q_ N(K,f)=Q(K,p_ N)\) for a set \(\{(x,y,c)\}\) can be efficiently made by using recurrence relations for the singular kernels \(K\) above. Numerical examples for \(K(t)=| t-c|^ \alpha\), \(\alpha=-0.7\) of the present method are compared with those of QAWS (and QAGP) in QUADPACK.