A Gauss-Jacobi kernel compression scheme for fractional differential equations (Q2000034)

In this paper, the author proposes a scheme based on the application of a composite Gauss-Jacobi quadrature to \[ \omega (t)= \frac{t^{-1+\alpha}}{\Gamma (\alpha)}=\frac{1}{\Gamma(\alpha)\Gamma(1-\alpha)}\int_{0}^{\infty}s^{-\alpha}e^{-ts}ds, \tag{1} \] where \(\Gamma(\cdot)\) denotes the gamma function. The integration interval \((0,\infty)\) in (1) is divided into \(K+1\) finite intervals and an infinite interval. The integral over the infinite interval is neglected and the integral over each of the remaining \(K+1\) finite intervals is approximated by an appropriate Gauss-Jacobi quadrature with \(J\) quadrature nodes. An approximation of \(\omega_{\delta}\) in \([0,T-\delta]\), where \(\omega_{\delta}(t)=\omega(t+\delta)\), is obtained through \[ S(t)=\sum^{P}_{p=1}b_{p}e^{-a_{p}t}, \tag{2} \] where \(a_{p}\) and \(b_{p}\) are real numbers for \(p=1, 2, \dots , P\) and \(P=(K+1)J\). The application of Gaussian quadratures for approximating (1) seems natural due to the rapid convergence they provide for analytic functions and since they yield real parameters \(a_{p}\) and \(b_{p}\), for \(p=1,2,\dots ,P\). The proposed scheme retains the properties of the earlier works, while it benefits from the rapid convergence of a Gauss-Jacobi quadrature. The main result of this paper provides an estimate of the relative point-wise error of the approximation. Some numerical results are provided to illustrate the results.