where the range of the quadrature is [slo , shi ], the expansion has a relative precision of ϵ over the range [rlo , 1], and h is the spacing of the quadrature points. The parameter T is empirically determined to be 5, 10, 14, 18, and 24, respectively for accuracies 1e-2, 1e-4, 1e-6, 1e-8, and 1e-10. However, because of the super-exponential decay, the number of quadrature points is only weakly dependent upon T , so using the optimal value of T saves at most only 1 or 2 points. The quadrature may be modified to provide compact representations of the Poisson kernel (?=0). To illustrate the automatic procedure, fits were generated for
exp(−30r)/r over [1-e8,1] for accuracies 1e-10, 1e-8, 1e-6, 1e-4 and 1e-2. The errors are plotted in figure 1.