Figure 8From: A fast multipole method for stellar dynamics Acceleration errors for improved FMM. Same as Figure 7 but for the multipole-acceptance criterion (16a) with ϵ=2× 10 − 7 (left) or (16b) with ϵ= 10 − 7 (right). The values for a and f are either taken from the direct-summation run (black), or obtained by low-oder FMM (red, see Section 5.4.2). In all cases the computational effort is similar to that of the FMM run shown in Figure 7 (since a≤f criterion (16a) gives a tighter constraint than (16b) and hence requires larger ϵ for the same computational effort).Back to article page